aboutsummaryrefslogtreecommitdiffstats
path: root/fs/btrfs
diff options
context:
space:
mode:
Diffstat (limited to 'fs/btrfs')
-rw-r--r--fs/btrfs/acl.c6
-rw-r--r--fs/btrfs/async-thread.c329
-rw-r--r--fs/btrfs/async-thread.h18
-rw-r--r--fs/btrfs/btrfs_inode.h11
-rw-r--r--fs/btrfs/compression.c8
-rw-r--r--fs/btrfs/ctree.c6
-rw-r--r--fs/btrfs/ctree.h112
-rw-r--r--fs/btrfs/dir-item.c47
-rw-r--r--fs/btrfs/disk-io.c283
-rw-r--r--fs/btrfs/export.c133
-rw-r--r--fs/btrfs/extent-tree.c2193
-rw-r--r--fs/btrfs/extent_io.c446
-rw-r--r--fs/btrfs/extent_io.h43
-rw-r--r--fs/btrfs/extent_map.c103
-rw-r--r--fs/btrfs/extent_map.h5
-rw-r--r--fs/btrfs/file.c73
-rw-r--r--fs/btrfs/free-space-cache.c36
-rw-r--r--fs/btrfs/inode-item.c4
-rw-r--r--fs/btrfs/inode-map.c93
-rw-r--r--fs/btrfs/inode.c986
-rw-r--r--fs/btrfs/ioctl.c404
-rw-r--r--fs/btrfs/ioctl.h3
-rw-r--r--fs/btrfs/ordered-data.c133
-rw-r--r--fs/btrfs/ordered-data.h7
-rw-r--r--fs/btrfs/orphan.c20
-rw-r--r--fs/btrfs/relocation.c284
-rw-r--r--fs/btrfs/root-tree.c138
-rw-r--r--fs/btrfs/super.c7
-rw-r--r--fs/btrfs/transaction.c48
-rw-r--r--fs/btrfs/tree-log.c39
-rw-r--r--fs/btrfs/volumes.c125
-rw-r--r--fs/btrfs/volumes.h3
-rw-r--r--fs/btrfs/xattr.c2
33 files changed, 3850 insertions, 2298 deletions
diff --git a/fs/btrfs/acl.c b/fs/btrfs/acl.c
index f128427b995b..69b355ae7f49 100644
--- a/fs/btrfs/acl.c
+++ b/fs/btrfs/acl.c
@@ -27,7 +27,7 @@
27#include "btrfs_inode.h" 27#include "btrfs_inode.h"
28#include "xattr.h" 28#include "xattr.h"
29 29
30#ifdef CONFIG_FS_POSIX_ACL 30#ifdef CONFIG_BTRFS_POSIX_ACL
31 31
32static struct posix_acl *btrfs_get_acl(struct inode *inode, int type) 32static struct posix_acl *btrfs_get_acl(struct inode *inode, int type)
33{ 33{
@@ -313,7 +313,7 @@ struct xattr_handler btrfs_xattr_acl_access_handler = {
313 .set = btrfs_xattr_acl_access_set, 313 .set = btrfs_xattr_acl_access_set,
314}; 314};
315 315
316#else /* CONFIG_FS_POSIX_ACL */ 316#else /* CONFIG_BTRFS_POSIX_ACL */
317 317
318int btrfs_acl_chmod(struct inode *inode) 318int btrfs_acl_chmod(struct inode *inode)
319{ 319{
@@ -325,4 +325,4 @@ int btrfs_init_acl(struct inode *inode, struct inode *dir)
325 return 0; 325 return 0;
326} 326}
327 327
328#endif /* CONFIG_FS_POSIX_ACL */ 328#endif /* CONFIG_BTRFS_POSIX_ACL */
diff --git a/fs/btrfs/async-thread.c b/fs/btrfs/async-thread.c
index 019e8af449ab..c0861e781cdb 100644
--- a/fs/btrfs/async-thread.c
+++ b/fs/btrfs/async-thread.c
@@ -48,6 +48,9 @@ struct btrfs_worker_thread {
48 /* number of things on the pending list */ 48 /* number of things on the pending list */
49 atomic_t num_pending; 49 atomic_t num_pending;
50 50
51 /* reference counter for this struct */
52 atomic_t refs;
53
51 unsigned long sequence; 54 unsigned long sequence;
52 55
53 /* protects the pending list. */ 56 /* protects the pending list. */
@@ -61,6 +64,51 @@ struct btrfs_worker_thread {
61}; 64};
62 65
63/* 66/*
67 * btrfs_start_workers uses kthread_run, which can block waiting for memory
68 * for a very long time. It will actually throttle on page writeback,
69 * and so it may not make progress until after our btrfs worker threads
70 * process all of the pending work structs in their queue
71 *
72 * This means we can't use btrfs_start_workers from inside a btrfs worker
73 * thread that is used as part of cleaning dirty memory, which pretty much
74 * involves all of the worker threads.
75 *
76 * Instead we have a helper queue who never has more than one thread
77 * where we scheduler thread start operations. This worker_start struct
78 * is used to contain the work and hold a pointer to the queue that needs
79 * another worker.
80 */
81struct worker_start {
82 struct btrfs_work work;
83 struct btrfs_workers *queue;
84};
85
86static void start_new_worker_func(struct btrfs_work *work)
87{
88 struct worker_start *start;
89 start = container_of(work, struct worker_start, work);
90 btrfs_start_workers(start->queue, 1);
91 kfree(start);
92}
93
94static int start_new_worker(struct btrfs_workers *queue)
95{
96 struct worker_start *start;
97 int ret;
98
99 start = kzalloc(sizeof(*start), GFP_NOFS);
100 if (!start)
101 return -ENOMEM;
102
103 start->work.func = start_new_worker_func;
104 start->queue = queue;
105 ret = btrfs_queue_worker(queue->atomic_worker_start, &start->work);
106 if (ret)
107 kfree(start);
108 return ret;
109}
110
111/*
64 * helper function to move a thread onto the idle list after it 112 * helper function to move a thread onto the idle list after it
65 * has finished some requests. 113 * has finished some requests.
66 */ 114 */
@@ -71,7 +119,12 @@ static void check_idle_worker(struct btrfs_worker_thread *worker)
71 unsigned long flags; 119 unsigned long flags;
72 spin_lock_irqsave(&worker->workers->lock, flags); 120 spin_lock_irqsave(&worker->workers->lock, flags);
73 worker->idle = 1; 121 worker->idle = 1;
74 list_move(&worker->worker_list, &worker->workers->idle_list); 122
123 /* the list may be empty if the worker is just starting */
124 if (!list_empty(&worker->worker_list)) {
125 list_move(&worker->worker_list,
126 &worker->workers->idle_list);
127 }
75 spin_unlock_irqrestore(&worker->workers->lock, flags); 128 spin_unlock_irqrestore(&worker->workers->lock, flags);
76 } 129 }
77} 130}
@@ -87,23 +140,51 @@ static void check_busy_worker(struct btrfs_worker_thread *worker)
87 unsigned long flags; 140 unsigned long flags;
88 spin_lock_irqsave(&worker->workers->lock, flags); 141 spin_lock_irqsave(&worker->workers->lock, flags);
89 worker->idle = 0; 142 worker->idle = 0;
90 list_move_tail(&worker->worker_list, 143
91 &worker->workers->worker_list); 144 if (!list_empty(&worker->worker_list)) {
145 list_move_tail(&worker->worker_list,
146 &worker->workers->worker_list);
147 }
92 spin_unlock_irqrestore(&worker->workers->lock, flags); 148 spin_unlock_irqrestore(&worker->workers->lock, flags);
93 } 149 }
94} 150}
95 151
96static noinline int run_ordered_completions(struct btrfs_workers *workers, 152static void check_pending_worker_creates(struct btrfs_worker_thread *worker)
97 struct btrfs_work *work)
98{ 153{
154 struct btrfs_workers *workers = worker->workers;
99 unsigned long flags; 155 unsigned long flags;
100 156
157 rmb();
158 if (!workers->atomic_start_pending)
159 return;
160
161 spin_lock_irqsave(&workers->lock, flags);
162 if (!workers->atomic_start_pending)
163 goto out;
164
165 workers->atomic_start_pending = 0;
166 if (workers->num_workers + workers->num_workers_starting >=
167 workers->max_workers)
168 goto out;
169
170 workers->num_workers_starting += 1;
171 spin_unlock_irqrestore(&workers->lock, flags);
172 start_new_worker(workers);
173 return;
174
175out:
176 spin_unlock_irqrestore(&workers->lock, flags);
177}
178
179static noinline int run_ordered_completions(struct btrfs_workers *workers,
180 struct btrfs_work *work)
181{
101 if (!workers->ordered) 182 if (!workers->ordered)
102 return 0; 183 return 0;
103 184
104 set_bit(WORK_DONE_BIT, &work->flags); 185 set_bit(WORK_DONE_BIT, &work->flags);
105 186
106 spin_lock_irqsave(&workers->lock, flags); 187 spin_lock(&workers->order_lock);
107 188
108 while (1) { 189 while (1) {
109 if (!list_empty(&workers->prio_order_list)) { 190 if (!list_empty(&workers->prio_order_list)) {
@@ -126,45 +207,118 @@ static noinline int run_ordered_completions(struct btrfs_workers *workers,
126 if (test_and_set_bit(WORK_ORDER_DONE_BIT, &work->flags)) 207 if (test_and_set_bit(WORK_ORDER_DONE_BIT, &work->flags))
127 break; 208 break;
128 209
129 spin_unlock_irqrestore(&workers->lock, flags); 210 spin_unlock(&workers->order_lock);
130 211
131 work->ordered_func(work); 212 work->ordered_func(work);
132 213
133 /* now take the lock again and call the freeing code */ 214 /* now take the lock again and call the freeing code */
134 spin_lock_irqsave(&workers->lock, flags); 215 spin_lock(&workers->order_lock);
135 list_del(&work->order_list); 216 list_del(&work->order_list);
136 work->ordered_free(work); 217 work->ordered_free(work);
137 } 218 }
138 219
139 spin_unlock_irqrestore(&workers->lock, flags); 220 spin_unlock(&workers->order_lock);
140 return 0; 221 return 0;
141} 222}
142 223
224static void put_worker(struct btrfs_worker_thread *worker)
225{
226 if (atomic_dec_and_test(&worker->refs))
227 kfree(worker);
228}
229
230static int try_worker_shutdown(struct btrfs_worker_thread *worker)
231{
232 int freeit = 0;
233
234 spin_lock_irq(&worker->lock);
235 spin_lock(&worker->workers->lock);
236 if (worker->workers->num_workers > 1 &&
237 worker->idle &&
238 !worker->working &&
239 !list_empty(&worker->worker_list) &&
240 list_empty(&worker->prio_pending) &&
241 list_empty(&worker->pending) &&
242 atomic_read(&worker->num_pending) == 0) {
243 freeit = 1;
244 list_del_init(&worker->worker_list);
245 worker->workers->num_workers--;
246 }
247 spin_unlock(&worker->workers->lock);
248 spin_unlock_irq(&worker->lock);
249
250 if (freeit)
251 put_worker(worker);
252 return freeit;
253}
254
255static struct btrfs_work *get_next_work(struct btrfs_worker_thread *worker,
256 struct list_head *prio_head,
257 struct list_head *head)
258{
259 struct btrfs_work *work = NULL;
260 struct list_head *cur = NULL;
261
262 if(!list_empty(prio_head))
263 cur = prio_head->next;
264
265 smp_mb();
266 if (!list_empty(&worker->prio_pending))
267 goto refill;
268
269 if (!list_empty(head))
270 cur = head->next;
271
272 if (cur)
273 goto out;
274
275refill:
276 spin_lock_irq(&worker->lock);
277 list_splice_tail_init(&worker->prio_pending, prio_head);
278 list_splice_tail_init(&worker->pending, head);
279
280 if (!list_empty(prio_head))
281 cur = prio_head->next;
282 else if (!list_empty(head))
283 cur = head->next;
284 spin_unlock_irq(&worker->lock);
285
286 if (!cur)
287 goto out_fail;
288
289out:
290 work = list_entry(cur, struct btrfs_work, list);
291
292out_fail:
293 return work;
294}
295
143/* 296/*
144 * main loop for servicing work items 297 * main loop for servicing work items
145 */ 298 */
146static int worker_loop(void *arg) 299static int worker_loop(void *arg)
147{ 300{
148 struct btrfs_worker_thread *worker = arg; 301 struct btrfs_worker_thread *worker = arg;
149 struct list_head *cur; 302 struct list_head head;
303 struct list_head prio_head;
150 struct btrfs_work *work; 304 struct btrfs_work *work;
305
306 INIT_LIST_HEAD(&head);
307 INIT_LIST_HEAD(&prio_head);
308
151 do { 309 do {
152 spin_lock_irq(&worker->lock); 310again:
153again_locked:
154 while (1) { 311 while (1) {
155 if (!list_empty(&worker->prio_pending)) 312
156 cur = worker->prio_pending.next; 313
157 else if (!list_empty(&worker->pending)) 314 work = get_next_work(worker, &prio_head, &head);
158 cur = worker->pending.next; 315 if (!work)
159 else
160 break; 316 break;
161 317
162 work = list_entry(cur, struct btrfs_work, list);
163 list_del(&work->list); 318 list_del(&work->list);
164 clear_bit(WORK_QUEUED_BIT, &work->flags); 319 clear_bit(WORK_QUEUED_BIT, &work->flags);
165 320
166 work->worker = worker; 321 work->worker = worker;
167 spin_unlock_irq(&worker->lock);
168 322
169 work->func(work); 323 work->func(work);
170 324
@@ -175,9 +329,13 @@ again_locked:
175 */ 329 */
176 run_ordered_completions(worker->workers, work); 330 run_ordered_completions(worker->workers, work);
177 331
178 spin_lock_irq(&worker->lock); 332 check_pending_worker_creates(worker);
179 check_idle_worker(worker); 333
180 } 334 }
335
336 spin_lock_irq(&worker->lock);
337 check_idle_worker(worker);
338
181 if (freezing(current)) { 339 if (freezing(current)) {
182 worker->working = 0; 340 worker->working = 0;
183 spin_unlock_irq(&worker->lock); 341 spin_unlock_irq(&worker->lock);
@@ -216,8 +374,10 @@ again_locked:
216 spin_lock_irq(&worker->lock); 374 spin_lock_irq(&worker->lock);
217 set_current_state(TASK_INTERRUPTIBLE); 375 set_current_state(TASK_INTERRUPTIBLE);
218 if (!list_empty(&worker->pending) || 376 if (!list_empty(&worker->pending) ||
219 !list_empty(&worker->prio_pending)) 377 !list_empty(&worker->prio_pending)) {
220 goto again_locked; 378 spin_unlock_irq(&worker->lock);
379 goto again;
380 }
221 381
222 /* 382 /*
223 * this makes sure we get a wakeup when someone 383 * this makes sure we get a wakeup when someone
@@ -226,8 +386,13 @@ again_locked:
226 worker->working = 0; 386 worker->working = 0;
227 spin_unlock_irq(&worker->lock); 387 spin_unlock_irq(&worker->lock);
228 388
229 if (!kthread_should_stop()) 389 if (!kthread_should_stop()) {
230 schedule(); 390 schedule_timeout(HZ * 120);
391 if (!worker->working &&
392 try_worker_shutdown(worker)) {
393 return 0;
394 }
395 }
231 } 396 }
232 __set_current_state(TASK_RUNNING); 397 __set_current_state(TASK_RUNNING);
233 } 398 }
@@ -242,41 +407,61 @@ int btrfs_stop_workers(struct btrfs_workers *workers)
242{ 407{
243 struct list_head *cur; 408 struct list_head *cur;
244 struct btrfs_worker_thread *worker; 409 struct btrfs_worker_thread *worker;
410 int can_stop;
245 411
412 spin_lock_irq(&workers->lock);
246 list_splice_init(&workers->idle_list, &workers->worker_list); 413 list_splice_init(&workers->idle_list, &workers->worker_list);
247 while (!list_empty(&workers->worker_list)) { 414 while (!list_empty(&workers->worker_list)) {
248 cur = workers->worker_list.next; 415 cur = workers->worker_list.next;
249 worker = list_entry(cur, struct btrfs_worker_thread, 416 worker = list_entry(cur, struct btrfs_worker_thread,
250 worker_list); 417 worker_list);
251 kthread_stop(worker->task); 418
252 list_del(&worker->worker_list); 419 atomic_inc(&worker->refs);
253 kfree(worker); 420 workers->num_workers -= 1;
421 if (!list_empty(&worker->worker_list)) {
422 list_del_init(&worker->worker_list);
423 put_worker(worker);
424 can_stop = 1;
425 } else
426 can_stop = 0;
427 spin_unlock_irq(&workers->lock);
428 if (can_stop)
429 kthread_stop(worker->task);
430 spin_lock_irq(&workers->lock);
431 put_worker(worker);
254 } 432 }
433 spin_unlock_irq(&workers->lock);
255 return 0; 434 return 0;
256} 435}
257 436
258/* 437/*
259 * simple init on struct btrfs_workers 438 * simple init on struct btrfs_workers
260 */ 439 */
261void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max) 440void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max,
441 struct btrfs_workers *async_helper)
262{ 442{
263 workers->num_workers = 0; 443 workers->num_workers = 0;
444 workers->num_workers_starting = 0;
264 INIT_LIST_HEAD(&workers->worker_list); 445 INIT_LIST_HEAD(&workers->worker_list);
265 INIT_LIST_HEAD(&workers->idle_list); 446 INIT_LIST_HEAD(&workers->idle_list);
266 INIT_LIST_HEAD(&workers->order_list); 447 INIT_LIST_HEAD(&workers->order_list);
267 INIT_LIST_HEAD(&workers->prio_order_list); 448 INIT_LIST_HEAD(&workers->prio_order_list);
268 spin_lock_init(&workers->lock); 449 spin_lock_init(&workers->lock);
450 spin_lock_init(&workers->order_lock);
269 workers->max_workers = max; 451 workers->max_workers = max;
270 workers->idle_thresh = 32; 452 workers->idle_thresh = 32;
271 workers->name = name; 453 workers->name = name;
272 workers->ordered = 0; 454 workers->ordered = 0;
455 workers->atomic_start_pending = 0;
456 workers->atomic_worker_start = async_helper;
273} 457}
274 458
275/* 459/*
276 * starts new worker threads. This does not enforce the max worker 460 * starts new worker threads. This does not enforce the max worker
277 * count in case you need to temporarily go past it. 461 * count in case you need to temporarily go past it.
278 */ 462 */
279int btrfs_start_workers(struct btrfs_workers *workers, int num_workers) 463static int __btrfs_start_workers(struct btrfs_workers *workers,
464 int num_workers)
280{ 465{
281 struct btrfs_worker_thread *worker; 466 struct btrfs_worker_thread *worker;
282 int ret = 0; 467 int ret = 0;
@@ -293,7 +478,9 @@ int btrfs_start_workers(struct btrfs_workers *workers, int num_workers)
293 INIT_LIST_HEAD(&worker->prio_pending); 478 INIT_LIST_HEAD(&worker->prio_pending);
294 INIT_LIST_HEAD(&worker->worker_list); 479 INIT_LIST_HEAD(&worker->worker_list);
295 spin_lock_init(&worker->lock); 480 spin_lock_init(&worker->lock);
481
296 atomic_set(&worker->num_pending, 0); 482 atomic_set(&worker->num_pending, 0);
483 atomic_set(&worker->refs, 1);
297 worker->workers = workers; 484 worker->workers = workers;
298 worker->task = kthread_run(worker_loop, worker, 485 worker->task = kthread_run(worker_loop, worker,
299 "btrfs-%s-%d", workers->name, 486 "btrfs-%s-%d", workers->name,
@@ -303,11 +490,12 @@ int btrfs_start_workers(struct btrfs_workers *workers, int num_workers)
303 kfree(worker); 490 kfree(worker);
304 goto fail; 491 goto fail;
305 } 492 }
306
307 spin_lock_irq(&workers->lock); 493 spin_lock_irq(&workers->lock);
308 list_add_tail(&worker->worker_list, &workers->idle_list); 494 list_add_tail(&worker->worker_list, &workers->idle_list);
309 worker->idle = 1; 495 worker->idle = 1;
310 workers->num_workers++; 496 workers->num_workers++;
497 workers->num_workers_starting--;
498 WARN_ON(workers->num_workers_starting < 0);
311 spin_unlock_irq(&workers->lock); 499 spin_unlock_irq(&workers->lock);
312 } 500 }
313 return 0; 501 return 0;
@@ -316,6 +504,14 @@ fail:
316 return ret; 504 return ret;
317} 505}
318 506
507int btrfs_start_workers(struct btrfs_workers *workers, int num_workers)
508{
509 spin_lock_irq(&workers->lock);
510 workers->num_workers_starting += num_workers;
511 spin_unlock_irq(&workers->lock);
512 return __btrfs_start_workers(workers, num_workers);
513}
514
319/* 515/*
320 * run through the list and find a worker thread that doesn't have a lot 516 * run through the list and find a worker thread that doesn't have a lot
321 * to do right now. This can return null if we aren't yet at the thread 517 * to do right now. This can return null if we aren't yet at the thread
@@ -325,7 +521,10 @@ static struct btrfs_worker_thread *next_worker(struct btrfs_workers *workers)
325{ 521{
326 struct btrfs_worker_thread *worker; 522 struct btrfs_worker_thread *worker;
327 struct list_head *next; 523 struct list_head *next;
328 int enforce_min = workers->num_workers < workers->max_workers; 524 int enforce_min;
525
526 enforce_min = (workers->num_workers + workers->num_workers_starting) <
527 workers->max_workers;
329 528
330 /* 529 /*
331 * if we find an idle thread, don't move it to the end of the 530 * if we find an idle thread, don't move it to the end of the
@@ -350,7 +549,6 @@ static struct btrfs_worker_thread *next_worker(struct btrfs_workers *workers)
350 */ 549 */
351 next = workers->worker_list.next; 550 next = workers->worker_list.next;
352 worker = list_entry(next, struct btrfs_worker_thread, worker_list); 551 worker = list_entry(next, struct btrfs_worker_thread, worker_list);
353 atomic_inc(&worker->num_pending);
354 worker->sequence++; 552 worker->sequence++;
355 553
356 if (worker->sequence % workers->idle_thresh == 0) 554 if (worker->sequence % workers->idle_thresh == 0)
@@ -367,35 +565,49 @@ static struct btrfs_worker_thread *find_worker(struct btrfs_workers *workers)
367{ 565{
368 struct btrfs_worker_thread *worker; 566 struct btrfs_worker_thread *worker;
369 unsigned long flags; 567 unsigned long flags;
568 struct list_head *fallback;
370 569
371again: 570again:
372 spin_lock_irqsave(&workers->lock, flags); 571 spin_lock_irqsave(&workers->lock, flags);
373 worker = next_worker(workers); 572 worker = next_worker(workers);
374 spin_unlock_irqrestore(&workers->lock, flags);
375 573
376 if (!worker) { 574 if (!worker) {
377 spin_lock_irqsave(&workers->lock, flags); 575 if (workers->num_workers + workers->num_workers_starting >=
378 if (workers->num_workers >= workers->max_workers) { 576 workers->max_workers) {
379 struct list_head *fallback = NULL; 577 goto fallback;
380 /* 578 } else if (workers->atomic_worker_start) {
381 * we have failed to find any workers, just 579 workers->atomic_start_pending = 1;
382 * return the force one 580 goto fallback;
383 */
384 if (!list_empty(&workers->worker_list))
385 fallback = workers->worker_list.next;
386 if (!list_empty(&workers->idle_list))
387 fallback = workers->idle_list.next;
388 BUG_ON(!fallback);
389 worker = list_entry(fallback,
390 struct btrfs_worker_thread, worker_list);
391 spin_unlock_irqrestore(&workers->lock, flags);
392 } else { 581 } else {
582 workers->num_workers_starting++;
393 spin_unlock_irqrestore(&workers->lock, flags); 583 spin_unlock_irqrestore(&workers->lock, flags);
394 /* we're below the limit, start another worker */ 584 /* we're below the limit, start another worker */
395 btrfs_start_workers(workers, 1); 585 __btrfs_start_workers(workers, 1);
396 goto again; 586 goto again;
397 } 587 }
398 } 588 }
589 goto found;
590
591fallback:
592 fallback = NULL;
593 /*
594 * we have failed to find any workers, just
595 * return the first one we can find.
596 */
597 if (!list_empty(&workers->worker_list))
598 fallback = workers->worker_list.next;
599 if (!list_empty(&workers->idle_list))
600 fallback = workers->idle_list.next;
601 BUG_ON(!fallback);
602 worker = list_entry(fallback,
603 struct btrfs_worker_thread, worker_list);
604found:
605 /*
606 * this makes sure the worker doesn't exit before it is placed
607 * onto a busy/idle list
608 */
609 atomic_inc(&worker->num_pending);
610 spin_unlock_irqrestore(&workers->lock, flags);
399 return worker; 611 return worker;
400} 612}
401 613
@@ -427,7 +639,7 @@ int btrfs_requeue_work(struct btrfs_work *work)
427 spin_lock(&worker->workers->lock); 639 spin_lock(&worker->workers->lock);
428 worker->idle = 0; 640 worker->idle = 0;
429 list_move_tail(&worker->worker_list, 641 list_move_tail(&worker->worker_list,
430 &worker->workers->worker_list); 642 &worker->workers->worker_list);
431 spin_unlock(&worker->workers->lock); 643 spin_unlock(&worker->workers->lock);
432 } 644 }
433 if (!worker->working) { 645 if (!worker->working) {
@@ -435,9 +647,9 @@ int btrfs_requeue_work(struct btrfs_work *work)
435 worker->working = 1; 647 worker->working = 1;
436 } 648 }
437 649
438 spin_unlock_irqrestore(&worker->lock, flags);
439 if (wake) 650 if (wake)
440 wake_up_process(worker->task); 651 wake_up_process(worker->task);
652 spin_unlock_irqrestore(&worker->lock, flags);
441out: 653out:
442 654
443 return 0; 655 return 0;
@@ -463,14 +675,18 @@ int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
463 675
464 worker = find_worker(workers); 676 worker = find_worker(workers);
465 if (workers->ordered) { 677 if (workers->ordered) {
466 spin_lock_irqsave(&workers->lock, flags); 678 /*
679 * you're not allowed to do ordered queues from an
680 * interrupt handler
681 */
682 spin_lock(&workers->order_lock);
467 if (test_bit(WORK_HIGH_PRIO_BIT, &work->flags)) { 683 if (test_bit(WORK_HIGH_PRIO_BIT, &work->flags)) {
468 list_add_tail(&work->order_list, 684 list_add_tail(&work->order_list,
469 &workers->prio_order_list); 685 &workers->prio_order_list);
470 } else { 686 } else {
471 list_add_tail(&work->order_list, &workers->order_list); 687 list_add_tail(&work->order_list, &workers->order_list);
472 } 688 }
473 spin_unlock_irqrestore(&workers->lock, flags); 689 spin_unlock(&workers->order_lock);
474 } else { 690 } else {
475 INIT_LIST_HEAD(&work->order_list); 691 INIT_LIST_HEAD(&work->order_list);
476 } 692 }
@@ -481,7 +697,6 @@ int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
481 list_add_tail(&work->list, &worker->prio_pending); 697 list_add_tail(&work->list, &worker->prio_pending);
482 else 698 else
483 list_add_tail(&work->list, &worker->pending); 699 list_add_tail(&work->list, &worker->pending);
484 atomic_inc(&worker->num_pending);
485 check_busy_worker(worker); 700 check_busy_worker(worker);
486 701
487 /* 702 /*
@@ -492,10 +707,10 @@ int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
492 wake = 1; 707 wake = 1;
493 worker->working = 1; 708 worker->working = 1;
494 709
495 spin_unlock_irqrestore(&worker->lock, flags);
496
497 if (wake) 710 if (wake)
498 wake_up_process(worker->task); 711 wake_up_process(worker->task);
712 spin_unlock_irqrestore(&worker->lock, flags);
713
499out: 714out:
500 return 0; 715 return 0;
501} 716}
diff --git a/fs/btrfs/async-thread.h b/fs/btrfs/async-thread.h
index 1b511c109db6..5077746cf85e 100644
--- a/fs/btrfs/async-thread.h
+++ b/fs/btrfs/async-thread.h
@@ -64,6 +64,8 @@ struct btrfs_workers {
64 /* current number of running workers */ 64 /* current number of running workers */
65 int num_workers; 65 int num_workers;
66 66
67 int num_workers_starting;
68
67 /* max number of workers allowed. changed by btrfs_start_workers */ 69 /* max number of workers allowed. changed by btrfs_start_workers */
68 int max_workers; 70 int max_workers;
69 71
@@ -73,6 +75,16 @@ struct btrfs_workers {
73 /* force completions in the order they were queued */ 75 /* force completions in the order they were queued */
74 int ordered; 76 int ordered;
75 77
78 /* more workers required, but in an interrupt handler */
79 int atomic_start_pending;
80
81 /*
82 * are we allowed to sleep while starting workers or are we required
83 * to start them at a later time? If we can't sleep, this indicates
84 * which queue we need to use to schedule thread creation.
85 */
86 struct btrfs_workers *atomic_worker_start;
87
76 /* list with all the work threads. The workers on the idle thread 88 /* list with all the work threads. The workers on the idle thread
77 * may be actively servicing jobs, but they haven't yet hit the 89 * may be actively servicing jobs, but they haven't yet hit the
78 * idle thresh limit above. 90 * idle thresh limit above.
@@ -90,6 +102,9 @@ struct btrfs_workers {
90 /* lock for finding the next worker thread to queue on */ 102 /* lock for finding the next worker thread to queue on */
91 spinlock_t lock; 103 spinlock_t lock;
92 104
105 /* lock for the ordered lists */
106 spinlock_t order_lock;
107
93 /* extra name for this worker, used for current->name */ 108 /* extra name for this worker, used for current->name */
94 char *name; 109 char *name;
95}; 110};
@@ -97,7 +112,8 @@ struct btrfs_workers {
97int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work); 112int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work);
98int btrfs_start_workers(struct btrfs_workers *workers, int num_workers); 113int btrfs_start_workers(struct btrfs_workers *workers, int num_workers);
99int btrfs_stop_workers(struct btrfs_workers *workers); 114int btrfs_stop_workers(struct btrfs_workers *workers);
100void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max); 115void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max,
116 struct btrfs_workers *async_starter);
101int btrfs_requeue_work(struct btrfs_work *work); 117int btrfs_requeue_work(struct btrfs_work *work);
102void btrfs_set_work_high_prio(struct btrfs_work *work); 118void btrfs_set_work_high_prio(struct btrfs_work *work);
103#endif 119#endif
diff --git a/fs/btrfs/btrfs_inode.h b/fs/btrfs/btrfs_inode.h
index ea1ea0af8c0e..c71abec0ab90 100644
--- a/fs/btrfs/btrfs_inode.h
+++ b/fs/btrfs/btrfs_inode.h
@@ -128,6 +128,16 @@ struct btrfs_inode {
128 u64 last_unlink_trans; 128 u64 last_unlink_trans;
129 129
130 /* 130 /*
131 * Counters to keep track of the number of extent item's we may use due
132 * to delalloc and such. outstanding_extents is the number of extent
133 * items we think we'll end up using, and reserved_extents is the number
134 * of extent items we've reserved metadata for.
135 */
136 spinlock_t accounting_lock;
137 int reserved_extents;
138 int outstanding_extents;
139
140 /*
131 * ordered_data_close is set by truncate when a file that used 141 * ordered_data_close is set by truncate when a file that used
132 * to have good data has been truncated to zero. When it is set 142 * to have good data has been truncated to zero. When it is set
133 * the btrfs file release call will add this inode to the 143 * the btrfs file release call will add this inode to the
@@ -138,6 +148,7 @@ struct btrfs_inode {
138 * of these. 148 * of these.
139 */ 149 */
140 unsigned ordered_data_close:1; 150 unsigned ordered_data_close:1;
151 unsigned dummy_inode:1;
141 152
142 struct inode vfs_inode; 153 struct inode vfs_inode;
143}; 154};
diff --git a/fs/btrfs/compression.c b/fs/btrfs/compression.c
index 9d8ba4d54a37..a11a32058b50 100644
--- a/fs/btrfs/compression.c
+++ b/fs/btrfs/compression.c
@@ -506,10 +506,10 @@ static noinline int add_ra_bio_pages(struct inode *inode,
506 */ 506 */
507 set_page_extent_mapped(page); 507 set_page_extent_mapped(page);
508 lock_extent(tree, last_offset, end, GFP_NOFS); 508 lock_extent(tree, last_offset, end, GFP_NOFS);
509 spin_lock(&em_tree->lock); 509 read_lock(&em_tree->lock);
510 em = lookup_extent_mapping(em_tree, last_offset, 510 em = lookup_extent_mapping(em_tree, last_offset,
511 PAGE_CACHE_SIZE); 511 PAGE_CACHE_SIZE);
512 spin_unlock(&em_tree->lock); 512 read_unlock(&em_tree->lock);
513 513
514 if (!em || last_offset < em->start || 514 if (!em || last_offset < em->start ||
515 (last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) || 515 (last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) ||
@@ -593,11 +593,11 @@ int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
593 em_tree = &BTRFS_I(inode)->extent_tree; 593 em_tree = &BTRFS_I(inode)->extent_tree;
594 594
595 /* we need the actual starting offset of this extent in the file */ 595 /* we need the actual starting offset of this extent in the file */
596 spin_lock(&em_tree->lock); 596 read_lock(&em_tree->lock);
597 em = lookup_extent_mapping(em_tree, 597 em = lookup_extent_mapping(em_tree,
598 page_offset(bio->bi_io_vec->bv_page), 598 page_offset(bio->bi_io_vec->bv_page),
599 PAGE_CACHE_SIZE); 599 PAGE_CACHE_SIZE);
600 spin_unlock(&em_tree->lock); 600 read_unlock(&em_tree->lock);
601 601
602 compressed_len = em->block_len; 602 compressed_len = em->block_len;
603 cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS); 603 cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index 3fdcc0512d3a..ec96f3a6d536 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -2853,6 +2853,12 @@ static noinline int split_leaf(struct btrfs_trans_handle *trans,
2853 int split; 2853 int split;
2854 int num_doubles = 0; 2854 int num_doubles = 0;
2855 2855
2856 l = path->nodes[0];
2857 slot = path->slots[0];
2858 if (extend && data_size + btrfs_item_size_nr(l, slot) +
2859 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
2860 return -EOVERFLOW;
2861
2856 /* first try to make some room by pushing left and right */ 2862 /* first try to make some room by pushing left and right */
2857 if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) { 2863 if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
2858 wret = push_leaf_right(trans, root, path, data_size, 0); 2864 wret = push_leaf_right(trans, root, path, data_size, 0);
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 837435ce84ca..1bb897ecdeeb 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -114,6 +114,10 @@ struct btrfs_ordered_sum;
114 */ 114 */
115#define BTRFS_DEV_ITEMS_OBJECTID 1ULL 115#define BTRFS_DEV_ITEMS_OBJECTID 1ULL
116 116
117#define BTRFS_BTREE_INODE_OBJECTID 1
118
119#define BTRFS_EMPTY_SUBVOL_DIR_OBJECTID 2
120
117/* 121/*
118 * we can actually store much bigger names, but lets not confuse the rest 122 * we can actually store much bigger names, but lets not confuse the rest
119 * of linux 123 * of linux
@@ -670,21 +674,29 @@ struct btrfs_space_info {
670 u64 bytes_reserved; /* total bytes the allocator has reserved for 674 u64 bytes_reserved; /* total bytes the allocator has reserved for
671 current allocations */ 675 current allocations */
672 u64 bytes_readonly; /* total bytes that are read only */ 676 u64 bytes_readonly; /* total bytes that are read only */
673 677 u64 bytes_super; /* total bytes reserved for the super blocks */
674 /* delalloc accounting */ 678 u64 bytes_root; /* the number of bytes needed to commit a
675 u64 bytes_delalloc; /* number of bytes reserved for allocation, 679 transaction */
676 this space is not necessarily reserved yet
677 by the allocator */
678 u64 bytes_may_use; /* number of bytes that may be used for 680 u64 bytes_may_use; /* number of bytes that may be used for
679 delalloc */ 681 delalloc/allocations */
682 u64 bytes_delalloc; /* number of bytes currently reserved for
683 delayed allocation */
680 684
681 int full; /* indicates that we cannot allocate any more 685 int full; /* indicates that we cannot allocate any more
682 chunks for this space */ 686 chunks for this space */
683 int force_alloc; /* set if we need to force a chunk alloc for 687 int force_alloc; /* set if we need to force a chunk alloc for
684 this space */ 688 this space */
689 int force_delalloc; /* make people start doing filemap_flush until
690 we're under a threshold */
685 691
686 struct list_head list; 692 struct list_head list;
687 693
694 /* for controlling how we free up space for allocations */
695 wait_queue_head_t allocate_wait;
696 wait_queue_head_t flush_wait;
697 int allocating_chunk;
698 int flushing;
699
688 /* for block groups in our same type */ 700 /* for block groups in our same type */
689 struct list_head block_groups; 701 struct list_head block_groups;
690 spinlock_t lock; 702 spinlock_t lock;
@@ -726,6 +738,15 @@ enum btrfs_caching_type {
726 BTRFS_CACHE_FINISHED = 2, 738 BTRFS_CACHE_FINISHED = 2,
727}; 739};
728 740
741struct btrfs_caching_control {
742 struct list_head list;
743 struct mutex mutex;
744 wait_queue_head_t wait;
745 struct btrfs_block_group_cache *block_group;
746 u64 progress;
747 atomic_t count;
748};
749
729struct btrfs_block_group_cache { 750struct btrfs_block_group_cache {
730 struct btrfs_key key; 751 struct btrfs_key key;
731 struct btrfs_block_group_item item; 752 struct btrfs_block_group_item item;
@@ -733,6 +754,7 @@ struct btrfs_block_group_cache {
733 spinlock_t lock; 754 spinlock_t lock;
734 u64 pinned; 755 u64 pinned;
735 u64 reserved; 756 u64 reserved;
757 u64 bytes_super;
736 u64 flags; 758 u64 flags;
737 u64 sectorsize; 759 u64 sectorsize;
738 int extents_thresh; 760 int extents_thresh;
@@ -742,8 +764,9 @@ struct btrfs_block_group_cache {
742 int dirty; 764 int dirty;
743 765
744 /* cache tracking stuff */ 766 /* cache tracking stuff */
745 wait_queue_head_t caching_q;
746 int cached; 767 int cached;
768 struct btrfs_caching_control *caching_ctl;
769 u64 last_byte_to_unpin;
747 770
748 struct btrfs_space_info *space_info; 771 struct btrfs_space_info *space_info;
749 772
@@ -782,13 +805,16 @@ struct btrfs_fs_info {
782 805
783 /* the log root tree is a directory of all the other log roots */ 806 /* the log root tree is a directory of all the other log roots */
784 struct btrfs_root *log_root_tree; 807 struct btrfs_root *log_root_tree;
808
809 spinlock_t fs_roots_radix_lock;
785 struct radix_tree_root fs_roots_radix; 810 struct radix_tree_root fs_roots_radix;
786 811
787 /* block group cache stuff */ 812 /* block group cache stuff */
788 spinlock_t block_group_cache_lock; 813 spinlock_t block_group_cache_lock;
789 struct rb_root block_group_cache_tree; 814 struct rb_root block_group_cache_tree;
790 815
791 struct extent_io_tree pinned_extents; 816 struct extent_io_tree freed_extents[2];
817 struct extent_io_tree *pinned_extents;
792 818
793 /* logical->physical extent mapping */ 819 /* logical->physical extent mapping */
794 struct btrfs_mapping_tree mapping_tree; 820 struct btrfs_mapping_tree mapping_tree;
@@ -822,11 +848,7 @@ struct btrfs_fs_info {
822 struct mutex transaction_kthread_mutex; 848 struct mutex transaction_kthread_mutex;
823 struct mutex cleaner_mutex; 849 struct mutex cleaner_mutex;
824 struct mutex chunk_mutex; 850 struct mutex chunk_mutex;
825 struct mutex drop_mutex;
826 struct mutex volume_mutex; 851 struct mutex volume_mutex;
827 struct mutex tree_reloc_mutex;
828 struct rw_semaphore extent_commit_sem;
829
830 /* 852 /*
831 * this protects the ordered operations list only while we are 853 * this protects the ordered operations list only while we are
832 * processing all of the entries on it. This way we make 854 * processing all of the entries on it. This way we make
@@ -835,10 +857,16 @@ struct btrfs_fs_info {
835 * before jumping into the main commit. 857 * before jumping into the main commit.
836 */ 858 */
837 struct mutex ordered_operations_mutex; 859 struct mutex ordered_operations_mutex;
860 struct rw_semaphore extent_commit_sem;
861
862 struct rw_semaphore subvol_sem;
863
864 struct srcu_struct subvol_srcu;
838 865
839 struct list_head trans_list; 866 struct list_head trans_list;
840 struct list_head hashers; 867 struct list_head hashers;
841 struct list_head dead_roots; 868 struct list_head dead_roots;
869 struct list_head caching_block_groups;
842 870
843 atomic_t nr_async_submits; 871 atomic_t nr_async_submits;
844 atomic_t async_submit_draining; 872 atomic_t async_submit_draining;
@@ -882,6 +910,7 @@ struct btrfs_fs_info {
882 * A third pool does submit_bio to avoid deadlocking with the other 910 * A third pool does submit_bio to avoid deadlocking with the other
883 * two 911 * two
884 */ 912 */
913 struct btrfs_workers generic_worker;
885 struct btrfs_workers workers; 914 struct btrfs_workers workers;
886 struct btrfs_workers delalloc_workers; 915 struct btrfs_workers delalloc_workers;
887 struct btrfs_workers endio_workers; 916 struct btrfs_workers endio_workers;
@@ -889,6 +918,7 @@ struct btrfs_fs_info {
889 struct btrfs_workers endio_meta_write_workers; 918 struct btrfs_workers endio_meta_write_workers;
890 struct btrfs_workers endio_write_workers; 919 struct btrfs_workers endio_write_workers;
891 struct btrfs_workers submit_workers; 920 struct btrfs_workers submit_workers;
921 struct btrfs_workers enospc_workers;
892 /* 922 /*
893 * fixup workers take dirty pages that didn't properly go through 923 * fixup workers take dirty pages that didn't properly go through
894 * the cow mechanism and make them safe to write. It happens 924 * the cow mechanism and make them safe to write. It happens
@@ -980,6 +1010,8 @@ struct btrfs_root {
980 atomic_t log_commit[2]; 1010 atomic_t log_commit[2];
981 unsigned long log_transid; 1011 unsigned long log_transid;
982 unsigned long log_batch; 1012 unsigned long log_batch;
1013 pid_t log_start_pid;
1014 bool log_multiple_pids;
983 1015
984 u64 objectid; 1016 u64 objectid;
985 u64 last_trans; 1017 u64 last_trans;
@@ -996,10 +1028,12 @@ struct btrfs_root {
996 u32 stripesize; 1028 u32 stripesize;
997 1029
998 u32 type; 1030 u32 type;
999 u64 highest_inode; 1031
1000 u64 last_inode_alloc; 1032 u64 highest_objectid;
1001 int ref_cows; 1033 int ref_cows;
1002 int track_dirty; 1034 int track_dirty;
1035 int in_radix;
1036
1003 u64 defrag_trans_start; 1037 u64 defrag_trans_start;
1004 struct btrfs_key defrag_progress; 1038 struct btrfs_key defrag_progress;
1005 struct btrfs_key defrag_max; 1039 struct btrfs_key defrag_max;
@@ -1920,8 +1954,8 @@ void btrfs_put_block_group(struct btrfs_block_group_cache *cache);
1920int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, 1954int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1921 struct btrfs_root *root, unsigned long count); 1955 struct btrfs_root *root, unsigned long count);
1922int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len); 1956int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len);
1923int btrfs_update_pinned_extents(struct btrfs_root *root, 1957int btrfs_pin_extent(struct btrfs_root *root,
1924 u64 bytenr, u64 num, int pin); 1958 u64 bytenr, u64 num, int reserved);
1925int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans, 1959int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
1926 struct btrfs_root *root, struct extent_buffer *leaf); 1960 struct btrfs_root *root, struct extent_buffer *leaf);
1927int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, 1961int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
@@ -1971,9 +2005,10 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans,
1971 u64 root_objectid, u64 owner, u64 offset); 2005 u64 root_objectid, u64 owner, u64 offset);
1972 2006
1973int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len); 2007int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len);
2008int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
2009 struct btrfs_root *root);
1974int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, 2010int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1975 struct btrfs_root *root, 2011 struct btrfs_root *root);
1976 struct extent_io_tree *unpin);
1977int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, 2012int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1978 struct btrfs_root *root, 2013 struct btrfs_root *root,
1979 u64 bytenr, u64 num_bytes, u64 parent, 2014 u64 bytenr, u64 num_bytes, u64 parent,
@@ -1984,6 +2019,7 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1984int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr); 2019int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr);
1985int btrfs_free_block_groups(struct btrfs_fs_info *info); 2020int btrfs_free_block_groups(struct btrfs_fs_info *info);
1986int btrfs_read_block_groups(struct btrfs_root *root); 2021int btrfs_read_block_groups(struct btrfs_root *root);
2022int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr);
1987int btrfs_make_block_group(struct btrfs_trans_handle *trans, 2023int btrfs_make_block_group(struct btrfs_trans_handle *trans,
1988 struct btrfs_root *root, u64 bytes_used, 2024 struct btrfs_root *root, u64 bytes_used,
1989 u64 type, u64 chunk_objectid, u64 chunk_offset, 2025 u64 type, u64 chunk_objectid, u64 chunk_offset,
@@ -1997,7 +2033,12 @@ u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags);
1997void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *ionde); 2033void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *ionde);
1998void btrfs_clear_space_info_full(struct btrfs_fs_info *info); 2034void btrfs_clear_space_info_full(struct btrfs_fs_info *info);
1999 2035
2000int btrfs_check_metadata_free_space(struct btrfs_root *root); 2036int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items);
2037int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items);
2038int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
2039 struct inode *inode, int num_items);
2040int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
2041 struct inode *inode, int num_items);
2001int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode, 2042int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
2002 u64 bytes); 2043 u64 bytes);
2003void btrfs_free_reserved_data_space(struct btrfs_root *root, 2044void btrfs_free_reserved_data_space(struct btrfs_root *root,
@@ -2006,7 +2047,6 @@ void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
2006 u64 bytes); 2047 u64 bytes);
2007void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode, 2048void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
2008 u64 bytes); 2049 u64 bytes);
2009void btrfs_free_pinned_extents(struct btrfs_fs_info *info);
2010/* ctree.c */ 2050/* ctree.c */
2011int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key, 2051int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
2012 int level, int *slot); 2052 int level, int *slot);
@@ -2100,12 +2140,15 @@ int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
2100 struct extent_buffer *parent); 2140 struct extent_buffer *parent);
2101/* root-item.c */ 2141/* root-item.c */
2102int btrfs_find_root_ref(struct btrfs_root *tree_root, 2142int btrfs_find_root_ref(struct btrfs_root *tree_root,
2103 struct btrfs_path *path, 2143 struct btrfs_path *path,
2104 u64 root_id, u64 ref_id); 2144 u64 root_id, u64 ref_id);
2105int btrfs_add_root_ref(struct btrfs_trans_handle *trans, 2145int btrfs_add_root_ref(struct btrfs_trans_handle *trans,
2106 struct btrfs_root *tree_root, 2146 struct btrfs_root *tree_root,
2107 u64 root_id, u8 type, u64 ref_id, 2147 u64 root_id, u64 ref_id, u64 dirid, u64 sequence,
2108 u64 dirid, u64 sequence, 2148 const char *name, int name_len);
2149int btrfs_del_root_ref(struct btrfs_trans_handle *trans,
2150 struct btrfs_root *tree_root,
2151 u64 root_id, u64 ref_id, u64 dirid, u64 *sequence,
2109 const char *name, int name_len); 2152 const char *name, int name_len);
2110int btrfs_del_root(struct btrfs_trans_handle *trans, struct btrfs_root *root, 2153int btrfs_del_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2111 struct btrfs_key *key); 2154 struct btrfs_key *key);
@@ -2120,6 +2163,7 @@ int btrfs_find_last_root(struct btrfs_root *root, u64 objectid, struct
2120int btrfs_search_root(struct btrfs_root *root, u64 search_start, 2163int btrfs_search_root(struct btrfs_root *root, u64 search_start,
2121 u64 *found_objectid); 2164 u64 *found_objectid);
2122int btrfs_find_dead_roots(struct btrfs_root *root, u64 objectid); 2165int btrfs_find_dead_roots(struct btrfs_root *root, u64 objectid);
2166int btrfs_find_orphan_roots(struct btrfs_root *tree_root);
2123int btrfs_set_root_node(struct btrfs_root_item *item, 2167int btrfs_set_root_node(struct btrfs_root_item *item,
2124 struct extent_buffer *node); 2168 struct extent_buffer *node);
2125/* dir-item.c */ 2169/* dir-item.c */
@@ -2138,6 +2182,10 @@ btrfs_lookup_dir_index_item(struct btrfs_trans_handle *trans,
2138 struct btrfs_path *path, u64 dir, 2182 struct btrfs_path *path, u64 dir,
2139 u64 objectid, const char *name, int name_len, 2183 u64 objectid, const char *name, int name_len,
2140 int mod); 2184 int mod);
2185struct btrfs_dir_item *
2186btrfs_search_dir_index_item(struct btrfs_root *root,
2187 struct btrfs_path *path, u64 dirid,
2188 const char *name, int name_len);
2141struct btrfs_dir_item *btrfs_match_dir_item_name(struct btrfs_root *root, 2189struct btrfs_dir_item *btrfs_match_dir_item_name(struct btrfs_root *root,
2142 struct btrfs_path *path, 2190 struct btrfs_path *path,
2143 const char *name, int name_len); 2191 const char *name, int name_len);
@@ -2160,6 +2208,7 @@ int btrfs_insert_orphan_item(struct btrfs_trans_handle *trans,
2160 struct btrfs_root *root, u64 offset); 2208 struct btrfs_root *root, u64 offset);
2161int btrfs_del_orphan_item(struct btrfs_trans_handle *trans, 2209int btrfs_del_orphan_item(struct btrfs_trans_handle *trans,
2162 struct btrfs_root *root, u64 offset); 2210 struct btrfs_root *root, u64 offset);
2211int btrfs_find_orphan_item(struct btrfs_root *root, u64 offset);
2163 2212
2164/* inode-map.c */ 2213/* inode-map.c */
2165int btrfs_find_free_objectid(struct btrfs_trans_handle *trans, 2214int btrfs_find_free_objectid(struct btrfs_trans_handle *trans,
@@ -2232,6 +2281,10 @@ int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2232int btrfs_add_link(struct btrfs_trans_handle *trans, 2281int btrfs_add_link(struct btrfs_trans_handle *trans,
2233 struct inode *parent_inode, struct inode *inode, 2282 struct inode *parent_inode, struct inode *inode,
2234 const char *name, int name_len, int add_backref, u64 index); 2283 const char *name, int name_len, int add_backref, u64 index);
2284int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2285 struct btrfs_root *root,
2286 struct inode *dir, u64 objectid,
2287 const char *name, int name_len);
2235int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, 2288int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2236 struct btrfs_root *root, 2289 struct btrfs_root *root,
2237 struct inode *inode, u64 new_size, 2290 struct inode *inode, u64 new_size,
@@ -2242,7 +2295,7 @@ int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end);
2242int btrfs_writepages(struct address_space *mapping, 2295int btrfs_writepages(struct address_space *mapping,
2243 struct writeback_control *wbc); 2296 struct writeback_control *wbc);
2244int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, 2297int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
2245 struct btrfs_root *new_root, struct dentry *dentry, 2298 struct btrfs_root *new_root,
2246 u64 new_dirid, u64 alloc_hint); 2299 u64 new_dirid, u64 alloc_hint);
2247int btrfs_merge_bio_hook(struct page *page, unsigned long offset, 2300int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
2248 size_t size, struct bio *bio, unsigned long bio_flags); 2301 size_t size, struct bio *bio, unsigned long bio_flags);
@@ -2258,6 +2311,7 @@ int btrfs_write_inode(struct inode *inode, int wait);
2258void btrfs_dirty_inode(struct inode *inode); 2311void btrfs_dirty_inode(struct inode *inode);
2259struct inode *btrfs_alloc_inode(struct super_block *sb); 2312struct inode *btrfs_alloc_inode(struct super_block *sb);
2260void btrfs_destroy_inode(struct inode *inode); 2313void btrfs_destroy_inode(struct inode *inode);
2314void btrfs_drop_inode(struct inode *inode);
2261int btrfs_init_cachep(void); 2315int btrfs_init_cachep(void);
2262void btrfs_destroy_cachep(void); 2316void btrfs_destroy_cachep(void);
2263long btrfs_ioctl_trans_end(struct file *file); 2317long btrfs_ioctl_trans_end(struct file *file);
@@ -2275,6 +2329,8 @@ int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode);
2275int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode); 2329int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode);
2276void btrfs_orphan_cleanup(struct btrfs_root *root); 2330void btrfs_orphan_cleanup(struct btrfs_root *root);
2277int btrfs_cont_expand(struct inode *inode, loff_t size); 2331int btrfs_cont_expand(struct inode *inode, loff_t size);
2332int btrfs_invalidate_inodes(struct btrfs_root *root);
2333extern const struct dentry_operations btrfs_dentry_operations;
2278 2334
2279/* ioctl.c */ 2335/* ioctl.c */
2280long btrfs_ioctl(struct file *file, unsigned int cmd, unsigned long arg); 2336long btrfs_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
@@ -2286,11 +2342,11 @@ int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync);
2286int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end, 2342int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
2287 int skip_pinned); 2343 int skip_pinned);
2288int btrfs_check_file(struct btrfs_root *root, struct inode *inode); 2344int btrfs_check_file(struct btrfs_root *root, struct inode *inode);
2289extern struct file_operations btrfs_file_operations; 2345extern const struct file_operations btrfs_file_operations;
2290int btrfs_drop_extents(struct btrfs_trans_handle *trans, 2346int btrfs_drop_extents(struct btrfs_trans_handle *trans,
2291 struct btrfs_root *root, struct inode *inode, 2347 struct btrfs_root *root, struct inode *inode,
2292 u64 start, u64 end, u64 locked_end, 2348 u64 start, u64 end, u64 locked_end,
2293 u64 inline_limit, u64 *hint_block); 2349 u64 inline_limit, u64 *hint_block, int drop_cache);
2294int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, 2350int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
2295 struct btrfs_root *root, 2351 struct btrfs_root *root,
2296 struct inode *inode, u64 start, u64 end); 2352 struct inode *inode, u64 start, u64 end);
@@ -2317,7 +2373,7 @@ int btrfs_parse_options(struct btrfs_root *root, char *options);
2317int btrfs_sync_fs(struct super_block *sb, int wait); 2373int btrfs_sync_fs(struct super_block *sb, int wait);
2318 2374
2319/* acl.c */ 2375/* acl.c */
2320#ifdef CONFIG_FS_POSIX_ACL 2376#ifdef CONFIG_BTRFS_POSIX_ACL
2321int btrfs_check_acl(struct inode *inode, int mask); 2377int btrfs_check_acl(struct inode *inode, int mask);
2322#else 2378#else
2323#define btrfs_check_acl NULL 2379#define btrfs_check_acl NULL
diff --git a/fs/btrfs/dir-item.c b/fs/btrfs/dir-item.c
index 1d70236ba00c..f3a6075519cc 100644
--- a/fs/btrfs/dir-item.c
+++ b/fs/btrfs/dir-item.c
@@ -281,6 +281,53 @@ btrfs_lookup_dir_index_item(struct btrfs_trans_handle *trans,
281 return btrfs_match_dir_item_name(root, path, name, name_len); 281 return btrfs_match_dir_item_name(root, path, name, name_len);
282} 282}
283 283
284struct btrfs_dir_item *
285btrfs_search_dir_index_item(struct btrfs_root *root,
286 struct btrfs_path *path, u64 dirid,
287 const char *name, int name_len)
288{
289 struct extent_buffer *leaf;
290 struct btrfs_dir_item *di;
291 struct btrfs_key key;
292 u32 nritems;
293 int ret;
294
295 key.objectid = dirid;
296 key.type = BTRFS_DIR_INDEX_KEY;
297 key.offset = 0;
298
299 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
300 if (ret < 0)
301 return ERR_PTR(ret);
302
303 leaf = path->nodes[0];
304 nritems = btrfs_header_nritems(leaf);
305
306 while (1) {
307 if (path->slots[0] >= nritems) {
308 ret = btrfs_next_leaf(root, path);
309 if (ret < 0)
310 return ERR_PTR(ret);
311 if (ret > 0)
312 break;
313 leaf = path->nodes[0];
314 nritems = btrfs_header_nritems(leaf);
315 continue;
316 }
317
318 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
319 if (key.objectid != dirid || key.type != BTRFS_DIR_INDEX_KEY)
320 break;
321
322 di = btrfs_match_dir_item_name(root, path, name, name_len);
323 if (di)
324 return di;
325
326 path->slots[0]++;
327 }
328 return NULL;
329}
330
284struct btrfs_dir_item *btrfs_lookup_xattr(struct btrfs_trans_handle *trans, 331struct btrfs_dir_item *btrfs_lookup_xattr(struct btrfs_trans_handle *trans,
285 struct btrfs_root *root, 332 struct btrfs_root *root,
286 struct btrfs_path *path, u64 dir, 333 struct btrfs_path *path, u64 dir,
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index e83be2e4602c..100551a66c46 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -41,6 +41,7 @@
41 41
42static struct extent_io_ops btree_extent_io_ops; 42static struct extent_io_ops btree_extent_io_ops;
43static void end_workqueue_fn(struct btrfs_work *work); 43static void end_workqueue_fn(struct btrfs_work *work);
44static void free_fs_root(struct btrfs_root *root);
44 45
45static atomic_t btrfs_bdi_num = ATOMIC_INIT(0); 46static atomic_t btrfs_bdi_num = ATOMIC_INIT(0);
46 47
@@ -123,15 +124,15 @@ static struct extent_map *btree_get_extent(struct inode *inode,
123 struct extent_map *em; 124 struct extent_map *em;
124 int ret; 125 int ret;
125 126
126 spin_lock(&em_tree->lock); 127 read_lock(&em_tree->lock);
127 em = lookup_extent_mapping(em_tree, start, len); 128 em = lookup_extent_mapping(em_tree, start, len);
128 if (em) { 129 if (em) {
129 em->bdev = 130 em->bdev =
130 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 131 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
131 spin_unlock(&em_tree->lock); 132 read_unlock(&em_tree->lock);
132 goto out; 133 goto out;
133 } 134 }
134 spin_unlock(&em_tree->lock); 135 read_unlock(&em_tree->lock);
135 136
136 em = alloc_extent_map(GFP_NOFS); 137 em = alloc_extent_map(GFP_NOFS);
137 if (!em) { 138 if (!em) {
@@ -144,7 +145,7 @@ static struct extent_map *btree_get_extent(struct inode *inode,
144 em->block_start = 0; 145 em->block_start = 0;
145 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 146 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
146 147
147 spin_lock(&em_tree->lock); 148 write_lock(&em_tree->lock);
148 ret = add_extent_mapping(em_tree, em); 149 ret = add_extent_mapping(em_tree, em);
149 if (ret == -EEXIST) { 150 if (ret == -EEXIST) {
150 u64 failed_start = em->start; 151 u64 failed_start = em->start;
@@ -163,7 +164,7 @@ static struct extent_map *btree_get_extent(struct inode *inode,
163 free_extent_map(em); 164 free_extent_map(em);
164 em = NULL; 165 em = NULL;
165 } 166 }
166 spin_unlock(&em_tree->lock); 167 write_unlock(&em_tree->lock);
167 168
168 if (ret) 169 if (ret)
169 em = ERR_PTR(ret); 170 em = ERR_PTR(ret);
@@ -772,7 +773,7 @@ static void btree_invalidatepage(struct page *page, unsigned long offset)
772 } 773 }
773} 774}
774 775
775static struct address_space_operations btree_aops = { 776static const struct address_space_operations btree_aops = {
776 .readpage = btree_readpage, 777 .readpage = btree_readpage,
777 .writepage = btree_writepage, 778 .writepage = btree_writepage,
778 .writepages = btree_writepages, 779 .writepages = btree_writepages,
@@ -821,14 +822,14 @@ struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
821 822
822int btrfs_write_tree_block(struct extent_buffer *buf) 823int btrfs_write_tree_block(struct extent_buffer *buf)
823{ 824{
824 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start, 825 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
825 buf->start + buf->len - 1, WB_SYNC_ALL); 826 buf->start + buf->len - 1);
826} 827}
827 828
828int btrfs_wait_tree_block_writeback(struct extent_buffer *buf) 829int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
829{ 830{
830 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping, 831 return filemap_fdatawait_range(buf->first_page->mapping,
831 buf->start, buf->start + buf->len - 1); 832 buf->start, buf->start + buf->len - 1);
832} 833}
833 834
834struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr, 835struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
@@ -895,8 +896,7 @@ static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
895 root->fs_info = fs_info; 896 root->fs_info = fs_info;
896 root->objectid = objectid; 897 root->objectid = objectid;
897 root->last_trans = 0; 898 root->last_trans = 0;
898 root->highest_inode = 0; 899 root->highest_objectid = 0;
899 root->last_inode_alloc = 0;
900 root->name = NULL; 900 root->name = NULL;
901 root->in_sysfs = 0; 901 root->in_sysfs = 0;
902 root->inode_tree.rb_node = NULL; 902 root->inode_tree.rb_node = NULL;
@@ -952,14 +952,16 @@ static int find_and_setup_root(struct btrfs_root *tree_root,
952 root, fs_info, objectid); 952 root, fs_info, objectid);
953 ret = btrfs_find_last_root(tree_root, objectid, 953 ret = btrfs_find_last_root(tree_root, objectid,
954 &root->root_item, &root->root_key); 954 &root->root_item, &root->root_key);
955 if (ret > 0)
956 return -ENOENT;
955 BUG_ON(ret); 957 BUG_ON(ret);
956 958
957 generation = btrfs_root_generation(&root->root_item); 959 generation = btrfs_root_generation(&root->root_item);
958 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item)); 960 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
959 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 961 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
960 blocksize, generation); 962 blocksize, generation);
961 root->commit_root = btrfs_root_node(root);
962 BUG_ON(!root->node); 963 BUG_ON(!root->node);
964 root->commit_root = btrfs_root_node(root);
963 return 0; 965 return 0;
964} 966}
965 967
@@ -1095,7 +1097,6 @@ struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1095 struct btrfs_fs_info *fs_info = tree_root->fs_info; 1097 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1096 struct btrfs_path *path; 1098 struct btrfs_path *path;
1097 struct extent_buffer *l; 1099 struct extent_buffer *l;
1098 u64 highest_inode;
1099 u64 generation; 1100 u64 generation;
1100 u32 blocksize; 1101 u32 blocksize;
1101 int ret = 0; 1102 int ret = 0;
@@ -1110,7 +1111,7 @@ struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1110 kfree(root); 1111 kfree(root);
1111 return ERR_PTR(ret); 1112 return ERR_PTR(ret);
1112 } 1113 }
1113 goto insert; 1114 goto out;
1114 } 1115 }
1115 1116
1116 __setup_root(tree_root->nodesize, tree_root->leafsize, 1117 __setup_root(tree_root->nodesize, tree_root->leafsize,
@@ -1120,39 +1121,30 @@ struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1120 path = btrfs_alloc_path(); 1121 path = btrfs_alloc_path();
1121 BUG_ON(!path); 1122 BUG_ON(!path);
1122 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0); 1123 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1123 if (ret != 0) { 1124 if (ret == 0) {
1124 if (ret > 0) 1125 l = path->nodes[0];
1125 ret = -ENOENT; 1126 read_extent_buffer(l, &root->root_item,
1126 goto out; 1127 btrfs_item_ptr_offset(l, path->slots[0]),
1128 sizeof(root->root_item));
1129 memcpy(&root->root_key, location, sizeof(*location));
1127 } 1130 }
1128 l = path->nodes[0];
1129 read_extent_buffer(l, &root->root_item,
1130 btrfs_item_ptr_offset(l, path->slots[0]),
1131 sizeof(root->root_item));
1132 memcpy(&root->root_key, location, sizeof(*location));
1133 ret = 0;
1134out:
1135 btrfs_release_path(root, path);
1136 btrfs_free_path(path); 1131 btrfs_free_path(path);
1137 if (ret) { 1132 if (ret) {
1138 kfree(root); 1133 if (ret > 0)
1134 ret = -ENOENT;
1139 return ERR_PTR(ret); 1135 return ERR_PTR(ret);
1140 } 1136 }
1137
1141 generation = btrfs_root_generation(&root->root_item); 1138 generation = btrfs_root_generation(&root->root_item);
1142 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item)); 1139 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1143 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 1140 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1144 blocksize, generation); 1141 blocksize, generation);
1145 root->commit_root = btrfs_root_node(root); 1142 root->commit_root = btrfs_root_node(root);
1146 BUG_ON(!root->node); 1143 BUG_ON(!root->node);
1147insert: 1144out:
1148 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) { 1145 if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1149 root->ref_cows = 1; 1146 root->ref_cows = 1;
1150 ret = btrfs_find_highest_inode(root, &highest_inode); 1147
1151 if (ret == 0) {
1152 root->highest_inode = highest_inode;
1153 root->last_inode_alloc = highest_inode;
1154 }
1155 }
1156 return root; 1148 return root;
1157} 1149}
1158 1150
@@ -1187,39 +1179,66 @@ struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1187 return fs_info->dev_root; 1179 return fs_info->dev_root;
1188 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 1180 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1189 return fs_info->csum_root; 1181 return fs_info->csum_root;
1190 1182again:
1183 spin_lock(&fs_info->fs_roots_radix_lock);
1191 root = radix_tree_lookup(&fs_info->fs_roots_radix, 1184 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1192 (unsigned long)location->objectid); 1185 (unsigned long)location->objectid);
1186 spin_unlock(&fs_info->fs_roots_radix_lock);
1193 if (root) 1187 if (root)
1194 return root; 1188 return root;
1195 1189
1190 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1191 if (ret == 0)
1192 ret = -ENOENT;
1193 if (ret < 0)
1194 return ERR_PTR(ret);
1195
1196 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location); 1196 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1197 if (IS_ERR(root)) 1197 if (IS_ERR(root))
1198 return root; 1198 return root;
1199 1199
1200 WARN_ON(btrfs_root_refs(&root->root_item) == 0);
1200 set_anon_super(&root->anon_super, NULL); 1201 set_anon_super(&root->anon_super, NULL);
1201 1202
1203 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1204 if (ret)
1205 goto fail;
1206
1207 spin_lock(&fs_info->fs_roots_radix_lock);
1202 ret = radix_tree_insert(&fs_info->fs_roots_radix, 1208 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1203 (unsigned long)root->root_key.objectid, 1209 (unsigned long)root->root_key.objectid,
1204 root); 1210 root);
1211 if (ret == 0)
1212 root->in_radix = 1;
1213 spin_unlock(&fs_info->fs_roots_radix_lock);
1214 radix_tree_preload_end();
1205 if (ret) { 1215 if (ret) {
1206 free_extent_buffer(root->node); 1216 if (ret == -EEXIST) {
1207 kfree(root); 1217 free_fs_root(root);
1208 return ERR_PTR(ret); 1218 goto again;
1219 }
1220 goto fail;
1209 } 1221 }
1210 if (!(fs_info->sb->s_flags & MS_RDONLY)) { 1222
1211 ret = btrfs_find_dead_roots(fs_info->tree_root, 1223 ret = btrfs_find_dead_roots(fs_info->tree_root,
1212 root->root_key.objectid); 1224 root->root_key.objectid);
1213 BUG_ON(ret); 1225 WARN_ON(ret);
1226
1227 if (!(fs_info->sb->s_flags & MS_RDONLY))
1214 btrfs_orphan_cleanup(root); 1228 btrfs_orphan_cleanup(root);
1215 } 1229
1216 return root; 1230 return root;
1231fail:
1232 free_fs_root(root);
1233 return ERR_PTR(ret);
1217} 1234}
1218 1235
1219struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info, 1236struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1220 struct btrfs_key *location, 1237 struct btrfs_key *location,
1221 const char *name, int namelen) 1238 const char *name, int namelen)
1222{ 1239{
1240 return btrfs_read_fs_root_no_name(fs_info, location);
1241#if 0
1223 struct btrfs_root *root; 1242 struct btrfs_root *root;
1224 int ret; 1243 int ret;
1225 1244
@@ -1236,7 +1255,7 @@ struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1236 kfree(root); 1255 kfree(root);
1237 return ERR_PTR(ret); 1256 return ERR_PTR(ret);
1238 } 1257 }
1239#if 0 1258
1240 ret = btrfs_sysfs_add_root(root); 1259 ret = btrfs_sysfs_add_root(root);
1241 if (ret) { 1260 if (ret) {
1242 free_extent_buffer(root->node); 1261 free_extent_buffer(root->node);
@@ -1244,9 +1263,9 @@ struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1244 kfree(root); 1263 kfree(root);
1245 return ERR_PTR(ret); 1264 return ERR_PTR(ret);
1246 } 1265 }
1247#endif
1248 root->in_sysfs = 1; 1266 root->in_sysfs = 1;
1249 return root; 1267 return root;
1268#endif
1250} 1269}
1251 1270
1252static int btrfs_congested_fn(void *congested_data, int bdi_bits) 1271static int btrfs_congested_fn(void *congested_data, int bdi_bits)
@@ -1325,9 +1344,9 @@ static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1325 offset = page_offset(page); 1344 offset = page_offset(page);
1326 1345
1327 em_tree = &BTRFS_I(inode)->extent_tree; 1346 em_tree = &BTRFS_I(inode)->extent_tree;
1328 spin_lock(&em_tree->lock); 1347 read_lock(&em_tree->lock);
1329 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE); 1348 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1330 spin_unlock(&em_tree->lock); 1349 read_unlock(&em_tree->lock);
1331 if (!em) { 1350 if (!em) {
1332 __unplug_io_fn(bdi, page); 1351 __unplug_io_fn(bdi, page);
1333 return; 1352 return;
@@ -1352,6 +1371,7 @@ static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1352{ 1371{
1353 int err; 1372 int err;
1354 1373
1374 bdi->name = "btrfs";
1355 bdi->capabilities = BDI_CAP_MAP_COPY; 1375 bdi->capabilities = BDI_CAP_MAP_COPY;
1356 err = bdi_init(bdi); 1376 err = bdi_init(bdi);
1357 if (err) 1377 if (err)
@@ -1359,8 +1379,10 @@ static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1359 1379
1360 err = bdi_register(bdi, NULL, "btrfs-%d", 1380 err = bdi_register(bdi, NULL, "btrfs-%d",
1361 atomic_inc_return(&btrfs_bdi_num)); 1381 atomic_inc_return(&btrfs_bdi_num));
1362 if (err) 1382 if (err) {
1383 bdi_destroy(bdi);
1363 return err; 1384 return err;
1385 }
1364 1386
1365 bdi->ra_pages = default_backing_dev_info.ra_pages; 1387 bdi->ra_pages = default_backing_dev_info.ra_pages;
1366 bdi->unplug_io_fn = btrfs_unplug_io_fn; 1388 bdi->unplug_io_fn = btrfs_unplug_io_fn;
@@ -1450,9 +1472,12 @@ static int cleaner_kthread(void *arg)
1450 break; 1472 break;
1451 1473
1452 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE); 1474 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1453 mutex_lock(&root->fs_info->cleaner_mutex); 1475
1454 btrfs_clean_old_snapshots(root); 1476 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1455 mutex_unlock(&root->fs_info->cleaner_mutex); 1477 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1478 btrfs_clean_old_snapshots(root);
1479 mutex_unlock(&root->fs_info->cleaner_mutex);
1480 }
1456 1481
1457 if (freezing(current)) { 1482 if (freezing(current)) {
1458 refrigerator(); 1483 refrigerator();
@@ -1557,15 +1582,36 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1557 err = -ENOMEM; 1582 err = -ENOMEM;
1558 goto fail; 1583 goto fail;
1559 } 1584 }
1560 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS); 1585
1586 ret = init_srcu_struct(&fs_info->subvol_srcu);
1587 if (ret) {
1588 err = ret;
1589 goto fail;
1590 }
1591
1592 ret = setup_bdi(fs_info, &fs_info->bdi);
1593 if (ret) {
1594 err = ret;
1595 goto fail_srcu;
1596 }
1597
1598 fs_info->btree_inode = new_inode(sb);
1599 if (!fs_info->btree_inode) {
1600 err = -ENOMEM;
1601 goto fail_bdi;
1602 }
1603
1604 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1561 INIT_LIST_HEAD(&fs_info->trans_list); 1605 INIT_LIST_HEAD(&fs_info->trans_list);
1562 INIT_LIST_HEAD(&fs_info->dead_roots); 1606 INIT_LIST_HEAD(&fs_info->dead_roots);
1563 INIT_LIST_HEAD(&fs_info->hashers); 1607 INIT_LIST_HEAD(&fs_info->hashers);
1564 INIT_LIST_HEAD(&fs_info->delalloc_inodes); 1608 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1565 INIT_LIST_HEAD(&fs_info->ordered_operations); 1609 INIT_LIST_HEAD(&fs_info->ordered_operations);
1610 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1566 spin_lock_init(&fs_info->delalloc_lock); 1611 spin_lock_init(&fs_info->delalloc_lock);
1567 spin_lock_init(&fs_info->new_trans_lock); 1612 spin_lock_init(&fs_info->new_trans_lock);
1568 spin_lock_init(&fs_info->ref_cache_lock); 1613 spin_lock_init(&fs_info->ref_cache_lock);
1614 spin_lock_init(&fs_info->fs_roots_radix_lock);
1569 1615
1570 init_completion(&fs_info->kobj_unregister); 1616 init_completion(&fs_info->kobj_unregister);
1571 fs_info->tree_root = tree_root; 1617 fs_info->tree_root = tree_root;
@@ -1584,12 +1630,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1584 fs_info->sb = sb; 1630 fs_info->sb = sb;
1585 fs_info->max_extent = (u64)-1; 1631 fs_info->max_extent = (u64)-1;
1586 fs_info->max_inline = 8192 * 1024; 1632 fs_info->max_inline = 8192 * 1024;
1587 if (setup_bdi(fs_info, &fs_info->bdi)) 1633 fs_info->metadata_ratio = 0;
1588 goto fail_bdi;
1589 fs_info->btree_inode = new_inode(sb);
1590 fs_info->btree_inode->i_ino = 1;
1591 fs_info->btree_inode->i_nlink = 1;
1592 fs_info->metadata_ratio = 8;
1593 1634
1594 fs_info->thread_pool_size = min_t(unsigned long, 1635 fs_info->thread_pool_size = min_t(unsigned long,
1595 num_online_cpus() + 2, 8); 1636 num_online_cpus() + 2, 8);
@@ -1599,7 +1640,10 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1599 1640
1600 sb->s_blocksize = 4096; 1641 sb->s_blocksize = 4096;
1601 sb->s_blocksize_bits = blksize_bits(4096); 1642 sb->s_blocksize_bits = blksize_bits(4096);
1643 sb->s_bdi = &fs_info->bdi;
1602 1644
1645 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1646 fs_info->btree_inode->i_nlink = 1;
1603 /* 1647 /*
1604 * we set the i_size on the btree inode to the max possible int. 1648 * we set the i_size on the btree inode to the max possible int.
1605 * the real end of the address space is determined by all of 1649 * the real end of the address space is determined by all of
@@ -1618,28 +1662,32 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1618 1662
1619 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops; 1663 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1620 1664
1665 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1666 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1667 sizeof(struct btrfs_key));
1668 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1669 insert_inode_hash(fs_info->btree_inode);
1670
1621 spin_lock_init(&fs_info->block_group_cache_lock); 1671 spin_lock_init(&fs_info->block_group_cache_lock);
1622 fs_info->block_group_cache_tree.rb_node = NULL; 1672 fs_info->block_group_cache_tree.rb_node = NULL;
1623 1673
1624 extent_io_tree_init(&fs_info->pinned_extents, 1674 extent_io_tree_init(&fs_info->freed_extents[0],
1625 fs_info->btree_inode->i_mapping, GFP_NOFS); 1675 fs_info->btree_inode->i_mapping, GFP_NOFS);
1676 extent_io_tree_init(&fs_info->freed_extents[1],
1677 fs_info->btree_inode->i_mapping, GFP_NOFS);
1678 fs_info->pinned_extents = &fs_info->freed_extents[0];
1626 fs_info->do_barriers = 1; 1679 fs_info->do_barriers = 1;
1627 1680
1628 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1629 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1630 sizeof(struct btrfs_key));
1631 insert_inode_hash(fs_info->btree_inode);
1632 1681
1633 mutex_init(&fs_info->trans_mutex); 1682 mutex_init(&fs_info->trans_mutex);
1634 mutex_init(&fs_info->ordered_operations_mutex); 1683 mutex_init(&fs_info->ordered_operations_mutex);
1635 mutex_init(&fs_info->tree_log_mutex); 1684 mutex_init(&fs_info->tree_log_mutex);
1636 mutex_init(&fs_info->drop_mutex);
1637 mutex_init(&fs_info->chunk_mutex); 1685 mutex_init(&fs_info->chunk_mutex);
1638 mutex_init(&fs_info->transaction_kthread_mutex); 1686 mutex_init(&fs_info->transaction_kthread_mutex);
1639 mutex_init(&fs_info->cleaner_mutex); 1687 mutex_init(&fs_info->cleaner_mutex);
1640 mutex_init(&fs_info->volume_mutex); 1688 mutex_init(&fs_info->volume_mutex);
1641 mutex_init(&fs_info->tree_reloc_mutex);
1642 init_rwsem(&fs_info->extent_commit_sem); 1689 init_rwsem(&fs_info->extent_commit_sem);
1690 init_rwsem(&fs_info->subvol_sem);
1643 1691
1644 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 1692 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1645 btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 1693 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
@@ -1699,20 +1747,24 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1699 goto fail_iput; 1747 goto fail_iput;
1700 } 1748 }
1701 1749
1702 /* 1750 btrfs_init_workers(&fs_info->generic_worker,
1703 * we need to start all the end_io workers up front because the 1751 "genwork", 1, NULL);
1704 * queue work function gets called at interrupt time, and so it 1752
1705 * cannot dynamically grow.
1706 */
1707 btrfs_init_workers(&fs_info->workers, "worker", 1753 btrfs_init_workers(&fs_info->workers, "worker",
1708 fs_info->thread_pool_size); 1754 fs_info->thread_pool_size,
1755 &fs_info->generic_worker);
1709 1756
1710 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc", 1757 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1711 fs_info->thread_pool_size); 1758 fs_info->thread_pool_size,
1759 &fs_info->generic_worker);
1712 1760
1713 btrfs_init_workers(&fs_info->submit_workers, "submit", 1761 btrfs_init_workers(&fs_info->submit_workers, "submit",
1714 min_t(u64, fs_devices->num_devices, 1762 min_t(u64, fs_devices->num_devices,
1715 fs_info->thread_pool_size)); 1763 fs_info->thread_pool_size),
1764 &fs_info->generic_worker);
1765 btrfs_init_workers(&fs_info->enospc_workers, "enospc",
1766 fs_info->thread_pool_size,
1767 &fs_info->generic_worker);
1716 1768
1717 /* a higher idle thresh on the submit workers makes it much more 1769 /* a higher idle thresh on the submit workers makes it much more
1718 * likely that bios will be send down in a sane order to the 1770 * likely that bios will be send down in a sane order to the
@@ -1726,15 +1778,20 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1726 fs_info->delalloc_workers.idle_thresh = 2; 1778 fs_info->delalloc_workers.idle_thresh = 2;
1727 fs_info->delalloc_workers.ordered = 1; 1779 fs_info->delalloc_workers.ordered = 1;
1728 1780
1729 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1); 1781 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1782 &fs_info->generic_worker);
1730 btrfs_init_workers(&fs_info->endio_workers, "endio", 1783 btrfs_init_workers(&fs_info->endio_workers, "endio",
1731 fs_info->thread_pool_size); 1784 fs_info->thread_pool_size,
1785 &fs_info->generic_worker);
1732 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta", 1786 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1733 fs_info->thread_pool_size); 1787 fs_info->thread_pool_size,
1788 &fs_info->generic_worker);
1734 btrfs_init_workers(&fs_info->endio_meta_write_workers, 1789 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1735 "endio-meta-write", fs_info->thread_pool_size); 1790 "endio-meta-write", fs_info->thread_pool_size,
1791 &fs_info->generic_worker);
1736 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write", 1792 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1737 fs_info->thread_pool_size); 1793 fs_info->thread_pool_size,
1794 &fs_info->generic_worker);
1738 1795
1739 /* 1796 /*
1740 * endios are largely parallel and should have a very 1797 * endios are largely parallel and should have a very
@@ -1743,20 +1800,19 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1743 fs_info->endio_workers.idle_thresh = 4; 1800 fs_info->endio_workers.idle_thresh = 4;
1744 fs_info->endio_meta_workers.idle_thresh = 4; 1801 fs_info->endio_meta_workers.idle_thresh = 4;
1745 1802
1746 fs_info->endio_write_workers.idle_thresh = 64; 1803 fs_info->endio_write_workers.idle_thresh = 2;
1747 fs_info->endio_meta_write_workers.idle_thresh = 64; 1804 fs_info->endio_meta_write_workers.idle_thresh = 2;
1748 1805
1749 btrfs_start_workers(&fs_info->workers, 1); 1806 btrfs_start_workers(&fs_info->workers, 1);
1807 btrfs_start_workers(&fs_info->generic_worker, 1);
1750 btrfs_start_workers(&fs_info->submit_workers, 1); 1808 btrfs_start_workers(&fs_info->submit_workers, 1);
1751 btrfs_start_workers(&fs_info->delalloc_workers, 1); 1809 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1752 btrfs_start_workers(&fs_info->fixup_workers, 1); 1810 btrfs_start_workers(&fs_info->fixup_workers, 1);
1753 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size); 1811 btrfs_start_workers(&fs_info->endio_workers, 1);
1754 btrfs_start_workers(&fs_info->endio_meta_workers, 1812 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1755 fs_info->thread_pool_size); 1813 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1756 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1814 btrfs_start_workers(&fs_info->endio_write_workers, 1);
1757 fs_info->thread_pool_size); 1815 btrfs_start_workers(&fs_info->enospc_workers, 1);
1758 btrfs_start_workers(&fs_info->endio_write_workers,
1759 fs_info->thread_pool_size);
1760 1816
1761 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super); 1817 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1762 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages, 1818 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
@@ -1916,6 +1972,9 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1916 } 1972 }
1917 } 1973 }
1918 1974
1975 ret = btrfs_find_orphan_roots(tree_root);
1976 BUG_ON(ret);
1977
1919 if (!(sb->s_flags & MS_RDONLY)) { 1978 if (!(sb->s_flags & MS_RDONLY)) {
1920 ret = btrfs_recover_relocation(tree_root); 1979 ret = btrfs_recover_relocation(tree_root);
1921 BUG_ON(ret); 1980 BUG_ON(ret);
@@ -1959,6 +2018,7 @@ fail_chunk_root:
1959 free_extent_buffer(chunk_root->node); 2018 free_extent_buffer(chunk_root->node);
1960 free_extent_buffer(chunk_root->commit_root); 2019 free_extent_buffer(chunk_root->commit_root);
1961fail_sb_buffer: 2020fail_sb_buffer:
2021 btrfs_stop_workers(&fs_info->generic_worker);
1962 btrfs_stop_workers(&fs_info->fixup_workers); 2022 btrfs_stop_workers(&fs_info->fixup_workers);
1963 btrfs_stop_workers(&fs_info->delalloc_workers); 2023 btrfs_stop_workers(&fs_info->delalloc_workers);
1964 btrfs_stop_workers(&fs_info->workers); 2024 btrfs_stop_workers(&fs_info->workers);
@@ -1967,6 +2027,7 @@ fail_sb_buffer:
1967 btrfs_stop_workers(&fs_info->endio_meta_write_workers); 2027 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1968 btrfs_stop_workers(&fs_info->endio_write_workers); 2028 btrfs_stop_workers(&fs_info->endio_write_workers);
1969 btrfs_stop_workers(&fs_info->submit_workers); 2029 btrfs_stop_workers(&fs_info->submit_workers);
2030 btrfs_stop_workers(&fs_info->enospc_workers);
1970fail_iput: 2031fail_iput:
1971 invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 2032 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1972 iput(fs_info->btree_inode); 2033 iput(fs_info->btree_inode);
@@ -1975,6 +2036,8 @@ fail_iput:
1975 btrfs_mapping_tree_free(&fs_info->mapping_tree); 2036 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1976fail_bdi: 2037fail_bdi:
1977 bdi_destroy(&fs_info->bdi); 2038 bdi_destroy(&fs_info->bdi);
2039fail_srcu:
2040 cleanup_srcu_struct(&fs_info->subvol_srcu);
1978fail: 2041fail:
1979 kfree(extent_root); 2042 kfree(extent_root);
1980 kfree(tree_root); 2043 kfree(tree_root);
@@ -2234,20 +2297,29 @@ int write_ctree_super(struct btrfs_trans_handle *trans,
2234 2297
2235int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root) 2298int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2236{ 2299{
2237 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 2300 spin_lock(&fs_info->fs_roots_radix_lock);
2238 radix_tree_delete(&fs_info->fs_roots_radix, 2301 radix_tree_delete(&fs_info->fs_roots_radix,
2239 (unsigned long)root->root_key.objectid); 2302 (unsigned long)root->root_key.objectid);
2303 spin_unlock(&fs_info->fs_roots_radix_lock);
2304
2305 if (btrfs_root_refs(&root->root_item) == 0)
2306 synchronize_srcu(&fs_info->subvol_srcu);
2307
2308 free_fs_root(root);
2309 return 0;
2310}
2311
2312static void free_fs_root(struct btrfs_root *root)
2313{
2314 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2240 if (root->anon_super.s_dev) { 2315 if (root->anon_super.s_dev) {
2241 down_write(&root->anon_super.s_umount); 2316 down_write(&root->anon_super.s_umount);
2242 kill_anon_super(&root->anon_super); 2317 kill_anon_super(&root->anon_super);
2243 } 2318 }
2244 if (root->node) 2319 free_extent_buffer(root->node);
2245 free_extent_buffer(root->node); 2320 free_extent_buffer(root->commit_root);
2246 if (root->commit_root)
2247 free_extent_buffer(root->commit_root);
2248 kfree(root->name); 2321 kfree(root->name);
2249 kfree(root); 2322 kfree(root);
2250 return 0;
2251} 2323}
2252 2324
2253static int del_fs_roots(struct btrfs_fs_info *fs_info) 2325static int del_fs_roots(struct btrfs_fs_info *fs_info)
@@ -2256,6 +2328,20 @@ static int del_fs_roots(struct btrfs_fs_info *fs_info)
2256 struct btrfs_root *gang[8]; 2328 struct btrfs_root *gang[8];
2257 int i; 2329 int i;
2258 2330
2331 while (!list_empty(&fs_info->dead_roots)) {
2332 gang[0] = list_entry(fs_info->dead_roots.next,
2333 struct btrfs_root, root_list);
2334 list_del(&gang[0]->root_list);
2335
2336 if (gang[0]->in_radix) {
2337 btrfs_free_fs_root(fs_info, gang[0]);
2338 } else {
2339 free_extent_buffer(gang[0]->node);
2340 free_extent_buffer(gang[0]->commit_root);
2341 kfree(gang[0]);
2342 }
2343 }
2344
2259 while (1) { 2345 while (1) {
2260 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2346 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2261 (void **)gang, 0, 2347 (void **)gang, 0,
@@ -2285,9 +2371,6 @@ int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2285 root_objectid = gang[ret - 1]->root_key.objectid + 1; 2371 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2286 for (i = 0; i < ret; i++) { 2372 for (i = 0; i < ret; i++) {
2287 root_objectid = gang[i]->root_key.objectid; 2373 root_objectid = gang[i]->root_key.objectid;
2288 ret = btrfs_find_dead_roots(fs_info->tree_root,
2289 root_objectid);
2290 BUG_ON(ret);
2291 btrfs_orphan_cleanup(gang[i]); 2374 btrfs_orphan_cleanup(gang[i]);
2292 } 2375 }
2293 root_objectid++; 2376 root_objectid++;
@@ -2357,12 +2440,12 @@ int close_ctree(struct btrfs_root *root)
2357 free_extent_buffer(root->fs_info->csum_root->commit_root); 2440 free_extent_buffer(root->fs_info->csum_root->commit_root);
2358 2441
2359 btrfs_free_block_groups(root->fs_info); 2442 btrfs_free_block_groups(root->fs_info);
2360 btrfs_free_pinned_extents(root->fs_info);
2361 2443
2362 del_fs_roots(fs_info); 2444 del_fs_roots(fs_info);
2363 2445
2364 iput(fs_info->btree_inode); 2446 iput(fs_info->btree_inode);
2365 2447
2448 btrfs_stop_workers(&fs_info->generic_worker);
2366 btrfs_stop_workers(&fs_info->fixup_workers); 2449 btrfs_stop_workers(&fs_info->fixup_workers);
2367 btrfs_stop_workers(&fs_info->delalloc_workers); 2450 btrfs_stop_workers(&fs_info->delalloc_workers);
2368 btrfs_stop_workers(&fs_info->workers); 2451 btrfs_stop_workers(&fs_info->workers);
@@ -2371,11 +2454,13 @@ int close_ctree(struct btrfs_root *root)
2371 btrfs_stop_workers(&fs_info->endio_meta_write_workers); 2454 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2372 btrfs_stop_workers(&fs_info->endio_write_workers); 2455 btrfs_stop_workers(&fs_info->endio_write_workers);
2373 btrfs_stop_workers(&fs_info->submit_workers); 2456 btrfs_stop_workers(&fs_info->submit_workers);
2457 btrfs_stop_workers(&fs_info->enospc_workers);
2374 2458
2375 btrfs_close_devices(fs_info->fs_devices); 2459 btrfs_close_devices(fs_info->fs_devices);
2376 btrfs_mapping_tree_free(&fs_info->mapping_tree); 2460 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2377 2461
2378 bdi_destroy(&fs_info->bdi); 2462 bdi_destroy(&fs_info->bdi);
2463 cleanup_srcu_struct(&fs_info->subvol_srcu);
2379 2464
2380 kfree(fs_info->extent_root); 2465 kfree(fs_info->extent_root);
2381 kfree(fs_info->tree_root); 2466 kfree(fs_info->tree_root);
diff --git a/fs/btrfs/export.c b/fs/btrfs/export.c
index 9596b40caa4e..ba5c3fd5ab8c 100644
--- a/fs/btrfs/export.c
+++ b/fs/btrfs/export.c
@@ -28,7 +28,7 @@ static int btrfs_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
28 len = BTRFS_FID_SIZE_NON_CONNECTABLE; 28 len = BTRFS_FID_SIZE_NON_CONNECTABLE;
29 type = FILEID_BTRFS_WITHOUT_PARENT; 29 type = FILEID_BTRFS_WITHOUT_PARENT;
30 30
31 fid->objectid = BTRFS_I(inode)->location.objectid; 31 fid->objectid = inode->i_ino;
32 fid->root_objectid = BTRFS_I(inode)->root->objectid; 32 fid->root_objectid = BTRFS_I(inode)->root->objectid;
33 fid->gen = inode->i_generation; 33 fid->gen = inode->i_generation;
34 34
@@ -60,34 +60,61 @@ static int btrfs_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
60} 60}
61 61
62static struct dentry *btrfs_get_dentry(struct super_block *sb, u64 objectid, 62static struct dentry *btrfs_get_dentry(struct super_block *sb, u64 objectid,
63 u64 root_objectid, u32 generation) 63 u64 root_objectid, u32 generation,
64 int check_generation)
64{ 65{
66 struct btrfs_fs_info *fs_info = btrfs_sb(sb)->fs_info;
65 struct btrfs_root *root; 67 struct btrfs_root *root;
68 struct dentry *dentry;
66 struct inode *inode; 69 struct inode *inode;
67 struct btrfs_key key; 70 struct btrfs_key key;
71 int index;
72 int err = 0;
73
74 if (objectid < BTRFS_FIRST_FREE_OBJECTID)
75 return ERR_PTR(-ESTALE);
68 76
69 key.objectid = root_objectid; 77 key.objectid = root_objectid;
70 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); 78 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
71 key.offset = (u64)-1; 79 key.offset = (u64)-1;
72 80
73 root = btrfs_read_fs_root_no_name(btrfs_sb(sb)->fs_info, &key); 81 index = srcu_read_lock(&fs_info->subvol_srcu);
74 if (IS_ERR(root)) 82
75 return ERR_CAST(root); 83 root = btrfs_read_fs_root_no_name(fs_info, &key);
84 if (IS_ERR(root)) {
85 err = PTR_ERR(root);
86 goto fail;
87 }
88
89 if (btrfs_root_refs(&root->root_item) == 0) {
90 err = -ENOENT;
91 goto fail;
92 }
76 93
77 key.objectid = objectid; 94 key.objectid = objectid;
78 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); 95 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
79 key.offset = 0; 96 key.offset = 0;
80 97
81 inode = btrfs_iget(sb, &key, root); 98 inode = btrfs_iget(sb, &key, root);
82 if (IS_ERR(inode)) 99 if (IS_ERR(inode)) {
83 return (void *)inode; 100 err = PTR_ERR(inode);
101 goto fail;
102 }
103
104 srcu_read_unlock(&fs_info->subvol_srcu, index);
84 105
85 if (generation != inode->i_generation) { 106 if (check_generation && generation != inode->i_generation) {
86 iput(inode); 107 iput(inode);
87 return ERR_PTR(-ESTALE); 108 return ERR_PTR(-ESTALE);
88 } 109 }
89 110
90 return d_obtain_alias(inode); 111 dentry = d_obtain_alias(inode);
112 if (!IS_ERR(dentry))
113 dentry->d_op = &btrfs_dentry_operations;
114 return dentry;
115fail:
116 srcu_read_unlock(&fs_info->subvol_srcu, index);
117 return ERR_PTR(err);
91} 118}
92 119
93static struct dentry *btrfs_fh_to_parent(struct super_block *sb, struct fid *fh, 120static struct dentry *btrfs_fh_to_parent(struct super_block *sb, struct fid *fh,
@@ -111,7 +138,7 @@ static struct dentry *btrfs_fh_to_parent(struct super_block *sb, struct fid *fh,
111 objectid = fid->parent_objectid; 138 objectid = fid->parent_objectid;
112 generation = fid->parent_gen; 139 generation = fid->parent_gen;
113 140
114 return btrfs_get_dentry(sb, objectid, root_objectid, generation); 141 return btrfs_get_dentry(sb, objectid, root_objectid, generation, 1);
115} 142}
116 143
117static struct dentry *btrfs_fh_to_dentry(struct super_block *sb, struct fid *fh, 144static struct dentry *btrfs_fh_to_dentry(struct super_block *sb, struct fid *fh,
@@ -133,66 +160,76 @@ static struct dentry *btrfs_fh_to_dentry(struct super_block *sb, struct fid *fh,
133 root_objectid = fid->root_objectid; 160 root_objectid = fid->root_objectid;
134 generation = fid->gen; 161 generation = fid->gen;
135 162
136 return btrfs_get_dentry(sb, objectid, root_objectid, generation); 163 return btrfs_get_dentry(sb, objectid, root_objectid, generation, 1);
137} 164}
138 165
139static struct dentry *btrfs_get_parent(struct dentry *child) 166static struct dentry *btrfs_get_parent(struct dentry *child)
140{ 167{
141 struct inode *dir = child->d_inode; 168 struct inode *dir = child->d_inode;
169 static struct dentry *dentry;
142 struct btrfs_root *root = BTRFS_I(dir)->root; 170 struct btrfs_root *root = BTRFS_I(dir)->root;
143 struct btrfs_key key;
144 struct btrfs_path *path; 171 struct btrfs_path *path;
145 struct extent_buffer *leaf; 172 struct extent_buffer *leaf;
146 int slot; 173 struct btrfs_root_ref *ref;
147 u64 objectid; 174 struct btrfs_key key;
175 struct btrfs_key found_key;
148 int ret; 176 int ret;
149 177
150 path = btrfs_alloc_path(); 178 path = btrfs_alloc_path();
151 179
152 key.objectid = dir->i_ino; 180 if (dir->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
153 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY); 181 key.objectid = root->root_key.objectid;
154 key.offset = (u64)-1; 182 key.type = BTRFS_ROOT_BACKREF_KEY;
183 key.offset = (u64)-1;
184 root = root->fs_info->tree_root;
185 } else {
186 key.objectid = dir->i_ino;
187 key.type = BTRFS_INODE_REF_KEY;
188 key.offset = (u64)-1;
189 }
155 190
156 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 191 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
157 if (ret < 0) { 192 if (ret < 0)
158 /* Error */ 193 goto fail;
159 btrfs_free_path(path); 194
160 return ERR_PTR(ret); 195 BUG_ON(ret == 0);
196 if (path->slots[0] == 0) {
197 ret = -ENOENT;
198 goto fail;
161 } 199 }
200
201 path->slots[0]--;
162 leaf = path->nodes[0]; 202 leaf = path->nodes[0];
163 slot = path->slots[0]; 203
164 if (ret) { 204 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
165 /* btrfs_search_slot() returns the slot where we'd want to 205 if (found_key.objectid != key.objectid || found_key.type != key.type) {
166 insert a backref for parent inode #0xFFFFFFFFFFFFFFFF. 206 ret = -ENOENT;
167 The _real_ backref, telling us what the parent inode 207 goto fail;
168 _actually_ is, will be in the slot _before_ the one
169 that btrfs_search_slot() returns. */
170 if (!slot) {
171 /* Unless there is _no_ key in the tree before... */
172 btrfs_free_path(path);
173 return ERR_PTR(-EIO);
174 }
175 slot--;
176 } 208 }
177 209
178 btrfs_item_key_to_cpu(leaf, &key, slot); 210 if (found_key.type == BTRFS_ROOT_BACKREF_KEY) {
211 ref = btrfs_item_ptr(leaf, path->slots[0],
212 struct btrfs_root_ref);
213 key.objectid = btrfs_root_ref_dirid(leaf, ref);
214 } else {
215 key.objectid = found_key.offset;
216 }
179 btrfs_free_path(path); 217 btrfs_free_path(path);
180 218
181 if (key.objectid != dir->i_ino || key.type != BTRFS_INODE_REF_KEY) 219 if (found_key.type == BTRFS_ROOT_BACKREF_KEY) {
182 return ERR_PTR(-EINVAL); 220 return btrfs_get_dentry(root->fs_info->sb, key.objectid,
183 221 found_key.offset, 0, 0);
184 objectid = key.offset; 222 }
185
186 /* If we are already at the root of a subvol, return the real root */
187 if (objectid == dir->i_ino)
188 return dget(dir->i_sb->s_root);
189 223
190 /* Build a new key for the inode item */ 224 key.type = BTRFS_INODE_ITEM_KEY;
191 key.objectid = objectid;
192 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
193 key.offset = 0; 225 key.offset = 0;
194 226 dentry = d_obtain_alias(btrfs_iget(root->fs_info->sb, &key, root));
195 return d_obtain_alias(btrfs_iget(root->fs_info->sb, &key, root)); 227 if (!IS_ERR(dentry))
228 dentry->d_op = &btrfs_dentry_operations;
229 return dentry;
230fail:
231 btrfs_free_path(path);
232 return ERR_PTR(ret);
196} 233}
197 234
198const struct export_operations btrfs_export_ops = { 235const struct export_operations btrfs_export_ops = {
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 72a2b9c28e9f..d0c4d584efad 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -32,12 +32,12 @@
32#include "locking.h" 32#include "locking.h"
33#include "free-space-cache.h" 33#include "free-space-cache.h"
34 34
35static int update_reserved_extents(struct btrfs_root *root,
36 u64 bytenr, u64 num, int reserve);
37static int update_block_group(struct btrfs_trans_handle *trans, 35static int update_block_group(struct btrfs_trans_handle *trans,
38 struct btrfs_root *root, 36 struct btrfs_root *root,
39 u64 bytenr, u64 num_bytes, int alloc, 37 u64 bytenr, u64 num_bytes, int alloc,
40 int mark_free); 38 int mark_free);
39static int update_reserved_extents(struct btrfs_block_group_cache *cache,
40 u64 num_bytes, int reserve);
41static int __btrfs_free_extent(struct btrfs_trans_handle *trans, 41static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
42 struct btrfs_root *root, 42 struct btrfs_root *root,
43 u64 bytenr, u64 num_bytes, u64 parent, 43 u64 bytenr, u64 num_bytes, u64 parent,
@@ -57,10 +57,19 @@ static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
57 u64 parent, u64 root_objectid, 57 u64 parent, u64 root_objectid,
58 u64 flags, struct btrfs_disk_key *key, 58 u64 flags, struct btrfs_disk_key *key,
59 int level, struct btrfs_key *ins); 59 int level, struct btrfs_key *ins);
60
61static int do_chunk_alloc(struct btrfs_trans_handle *trans, 60static int do_chunk_alloc(struct btrfs_trans_handle *trans,
62 struct btrfs_root *extent_root, u64 alloc_bytes, 61 struct btrfs_root *extent_root, u64 alloc_bytes,
63 u64 flags, int force); 62 u64 flags, int force);
63static int pin_down_bytes(struct btrfs_trans_handle *trans,
64 struct btrfs_root *root,
65 struct btrfs_path *path,
66 u64 bytenr, u64 num_bytes,
67 int is_data, int reserved,
68 struct extent_buffer **must_clean);
69static int find_next_key(struct btrfs_path *path, int level,
70 struct btrfs_key *key);
71static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
72 int dump_block_groups);
64 73
65static noinline int 74static noinline int
66block_group_cache_done(struct btrfs_block_group_cache *cache) 75block_group_cache_done(struct btrfs_block_group_cache *cache)
@@ -153,34 +162,34 @@ block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
153 return ret; 162 return ret;
154} 163}
155 164
156/* 165static int add_excluded_extent(struct btrfs_root *root,
157 * We always set EXTENT_LOCKED for the super mirror extents so we don't 166 u64 start, u64 num_bytes)
158 * overwrite them, so those bits need to be unset. Also, if we are unmounting
159 * with pinned extents still sitting there because we had a block group caching,
160 * we need to clear those now, since we are done.
161 */
162void btrfs_free_pinned_extents(struct btrfs_fs_info *info)
163{ 167{
164 u64 start, end, last = 0; 168 u64 end = start + num_bytes - 1;
165 int ret; 169 set_extent_bits(&root->fs_info->freed_extents[0],
170 start, end, EXTENT_UPTODATE, GFP_NOFS);
171 set_extent_bits(&root->fs_info->freed_extents[1],
172 start, end, EXTENT_UPTODATE, GFP_NOFS);
173 return 0;
174}
166 175
167 while (1) { 176static void free_excluded_extents(struct btrfs_root *root,
168 ret = find_first_extent_bit(&info->pinned_extents, last, 177 struct btrfs_block_group_cache *cache)
169 &start, &end, 178{
170 EXTENT_LOCKED|EXTENT_DIRTY); 179 u64 start, end;
171 if (ret)
172 break;
173 180
174 clear_extent_bits(&info->pinned_extents, start, end, 181 start = cache->key.objectid;
175 EXTENT_LOCKED|EXTENT_DIRTY, GFP_NOFS); 182 end = start + cache->key.offset - 1;
176 last = end+1; 183
177 } 184 clear_extent_bits(&root->fs_info->freed_extents[0],
185 start, end, EXTENT_UPTODATE, GFP_NOFS);
186 clear_extent_bits(&root->fs_info->freed_extents[1],
187 start, end, EXTENT_UPTODATE, GFP_NOFS);
178} 188}
179 189
180static int remove_sb_from_cache(struct btrfs_root *root, 190static int exclude_super_stripes(struct btrfs_root *root,
181 struct btrfs_block_group_cache *cache) 191 struct btrfs_block_group_cache *cache)
182{ 192{
183 struct btrfs_fs_info *fs_info = root->fs_info;
184 u64 bytenr; 193 u64 bytenr;
185 u64 *logical; 194 u64 *logical;
186 int stripe_len; 195 int stripe_len;
@@ -192,17 +201,42 @@ static int remove_sb_from_cache(struct btrfs_root *root,
192 cache->key.objectid, bytenr, 201 cache->key.objectid, bytenr,
193 0, &logical, &nr, &stripe_len); 202 0, &logical, &nr, &stripe_len);
194 BUG_ON(ret); 203 BUG_ON(ret);
204
195 while (nr--) { 205 while (nr--) {
196 try_lock_extent(&fs_info->pinned_extents, 206 cache->bytes_super += stripe_len;
197 logical[nr], 207 ret = add_excluded_extent(root, logical[nr],
198 logical[nr] + stripe_len - 1, GFP_NOFS); 208 stripe_len);
209 BUG_ON(ret);
199 } 210 }
211
200 kfree(logical); 212 kfree(logical);
201 } 213 }
202
203 return 0; 214 return 0;
204} 215}
205 216
217static struct btrfs_caching_control *
218get_caching_control(struct btrfs_block_group_cache *cache)
219{
220 struct btrfs_caching_control *ctl;
221
222 spin_lock(&cache->lock);
223 if (cache->cached != BTRFS_CACHE_STARTED) {
224 spin_unlock(&cache->lock);
225 return NULL;
226 }
227
228 ctl = cache->caching_ctl;
229 atomic_inc(&ctl->count);
230 spin_unlock(&cache->lock);
231 return ctl;
232}
233
234static void put_caching_control(struct btrfs_caching_control *ctl)
235{
236 if (atomic_dec_and_test(&ctl->count))
237 kfree(ctl);
238}
239
206/* 240/*
207 * this is only called by cache_block_group, since we could have freed extents 241 * this is only called by cache_block_group, since we could have freed extents
208 * we need to check the pinned_extents for any extents that can't be used yet 242 * we need to check the pinned_extents for any extents that can't be used yet
@@ -215,9 +249,9 @@ static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
215 int ret; 249 int ret;
216 250
217 while (start < end) { 251 while (start < end) {
218 ret = find_first_extent_bit(&info->pinned_extents, start, 252 ret = find_first_extent_bit(info->pinned_extents, start,
219 &extent_start, &extent_end, 253 &extent_start, &extent_end,
220 EXTENT_DIRTY|EXTENT_LOCKED); 254 EXTENT_DIRTY | EXTENT_UPTODATE);
221 if (ret) 255 if (ret)
222 break; 256 break;
223 257
@@ -249,22 +283,27 @@ static int caching_kthread(void *data)
249{ 283{
250 struct btrfs_block_group_cache *block_group = data; 284 struct btrfs_block_group_cache *block_group = data;
251 struct btrfs_fs_info *fs_info = block_group->fs_info; 285 struct btrfs_fs_info *fs_info = block_group->fs_info;
252 u64 last = 0; 286 struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
287 struct btrfs_root *extent_root = fs_info->extent_root;
253 struct btrfs_path *path; 288 struct btrfs_path *path;
254 int ret = 0;
255 struct btrfs_key key;
256 struct extent_buffer *leaf; 289 struct extent_buffer *leaf;
257 int slot; 290 struct btrfs_key key;
258 u64 total_found = 0; 291 u64 total_found = 0;
259 292 u64 last = 0;
260 BUG_ON(!fs_info); 293 u32 nritems;
294 int ret = 0;
261 295
262 path = btrfs_alloc_path(); 296 path = btrfs_alloc_path();
263 if (!path) 297 if (!path)
264 return -ENOMEM; 298 return -ENOMEM;
265 299
266 atomic_inc(&block_group->space_info->caching_threads); 300 exclude_super_stripes(extent_root, block_group);
301 spin_lock(&block_group->space_info->lock);
302 block_group->space_info->bytes_super += block_group->bytes_super;
303 spin_unlock(&block_group->space_info->lock);
304
267 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); 305 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
306
268 /* 307 /*
269 * We don't want to deadlock with somebody trying to allocate a new 308 * We don't want to deadlock with somebody trying to allocate a new
270 * extent for the extent root while also trying to search the extent 309 * extent for the extent root while also trying to search the extent
@@ -277,74 +316,64 @@ static int caching_kthread(void *data)
277 316
278 key.objectid = last; 317 key.objectid = last;
279 key.offset = 0; 318 key.offset = 0;
280 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); 319 key.type = BTRFS_EXTENT_ITEM_KEY;
281again: 320again:
321 mutex_lock(&caching_ctl->mutex);
282 /* need to make sure the commit_root doesn't disappear */ 322 /* need to make sure the commit_root doesn't disappear */
283 down_read(&fs_info->extent_commit_sem); 323 down_read(&fs_info->extent_commit_sem);
284 324
285 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); 325 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
286 if (ret < 0) 326 if (ret < 0)
287 goto err; 327 goto err;
288 328
329 leaf = path->nodes[0];
330 nritems = btrfs_header_nritems(leaf);
331
289 while (1) { 332 while (1) {
290 smp_mb(); 333 smp_mb();
291 if (block_group->fs_info->closing > 1) { 334 if (fs_info->closing > 1) {
292 last = (u64)-1; 335 last = (u64)-1;
293 break; 336 break;
294 } 337 }
295 338
296 leaf = path->nodes[0]; 339 if (path->slots[0] < nritems) {
297 slot = path->slots[0]; 340 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
298 if (slot >= btrfs_header_nritems(leaf)) { 341 } else {
299 ret = btrfs_next_leaf(fs_info->extent_root, path); 342 ret = find_next_key(path, 0, &key);
300 if (ret < 0) 343 if (ret)
301 goto err;
302 else if (ret)
303 break; 344 break;
304 345
305 if (need_resched() || 346 caching_ctl->progress = last;
306 btrfs_transaction_in_commit(fs_info)) { 347 btrfs_release_path(extent_root, path);
307 leaf = path->nodes[0]; 348 up_read(&fs_info->extent_commit_sem);
308 349 mutex_unlock(&caching_ctl->mutex);
309 /* this shouldn't happen, but if the 350 if (btrfs_transaction_in_commit(fs_info))
310 * leaf is empty just move on.
311 */
312 if (btrfs_header_nritems(leaf) == 0)
313 break;
314 /*
315 * we need to copy the key out so that
316 * we are sure the next search advances
317 * us forward in the btree.
318 */
319 btrfs_item_key_to_cpu(leaf, &key, 0);
320 btrfs_release_path(fs_info->extent_root, path);
321 up_read(&fs_info->extent_commit_sem);
322 schedule_timeout(1); 351 schedule_timeout(1);
323 goto again; 352 else
324 } 353 cond_resched();
354 goto again;
355 }
325 356
357 if (key.objectid < block_group->key.objectid) {
358 path->slots[0]++;
326 continue; 359 continue;
327 } 360 }
328 btrfs_item_key_to_cpu(leaf, &key, slot);
329 if (key.objectid < block_group->key.objectid)
330 goto next;
331 361
332 if (key.objectid >= block_group->key.objectid + 362 if (key.objectid >= block_group->key.objectid +
333 block_group->key.offset) 363 block_group->key.offset)
334 break; 364 break;
335 365
336 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) { 366 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
337 total_found += add_new_free_space(block_group, 367 total_found += add_new_free_space(block_group,
338 fs_info, last, 368 fs_info, last,
339 key.objectid); 369 key.objectid);
340 last = key.objectid + key.offset; 370 last = key.objectid + key.offset;
341 }
342 371
343 if (total_found > (1024 * 1024 * 2)) { 372 if (total_found > (1024 * 1024 * 2)) {
344 total_found = 0; 373 total_found = 0;
345 wake_up(&block_group->caching_q); 374 wake_up(&caching_ctl->wait);
375 }
346 } 376 }
347next:
348 path->slots[0]++; 377 path->slots[0]++;
349 } 378 }
350 ret = 0; 379 ret = 0;
@@ -352,33 +381,65 @@ next:
352 total_found += add_new_free_space(block_group, fs_info, last, 381 total_found += add_new_free_space(block_group, fs_info, last,
353 block_group->key.objectid + 382 block_group->key.objectid +
354 block_group->key.offset); 383 block_group->key.offset);
384 caching_ctl->progress = (u64)-1;
355 385
356 spin_lock(&block_group->lock); 386 spin_lock(&block_group->lock);
387 block_group->caching_ctl = NULL;
357 block_group->cached = BTRFS_CACHE_FINISHED; 388 block_group->cached = BTRFS_CACHE_FINISHED;
358 spin_unlock(&block_group->lock); 389 spin_unlock(&block_group->lock);
359 390
360err: 391err:
361 btrfs_free_path(path); 392 btrfs_free_path(path);
362 up_read(&fs_info->extent_commit_sem); 393 up_read(&fs_info->extent_commit_sem);
363 atomic_dec(&block_group->space_info->caching_threads);
364 wake_up(&block_group->caching_q);
365 394
395 free_excluded_extents(extent_root, block_group);
396
397 mutex_unlock(&caching_ctl->mutex);
398 wake_up(&caching_ctl->wait);
399
400 put_caching_control(caching_ctl);
401 atomic_dec(&block_group->space_info->caching_threads);
366 return 0; 402 return 0;
367} 403}
368 404
369static int cache_block_group(struct btrfs_block_group_cache *cache) 405static int cache_block_group(struct btrfs_block_group_cache *cache)
370{ 406{
407 struct btrfs_fs_info *fs_info = cache->fs_info;
408 struct btrfs_caching_control *caching_ctl;
371 struct task_struct *tsk; 409 struct task_struct *tsk;
372 int ret = 0; 410 int ret = 0;
373 411
412 smp_mb();
413 if (cache->cached != BTRFS_CACHE_NO)
414 return 0;
415
416 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
417 BUG_ON(!caching_ctl);
418
419 INIT_LIST_HEAD(&caching_ctl->list);
420 mutex_init(&caching_ctl->mutex);
421 init_waitqueue_head(&caching_ctl->wait);
422 caching_ctl->block_group = cache;
423 caching_ctl->progress = cache->key.objectid;
424 /* one for caching kthread, one for caching block group list */
425 atomic_set(&caching_ctl->count, 2);
426
374 spin_lock(&cache->lock); 427 spin_lock(&cache->lock);
375 if (cache->cached != BTRFS_CACHE_NO) { 428 if (cache->cached != BTRFS_CACHE_NO) {
376 spin_unlock(&cache->lock); 429 spin_unlock(&cache->lock);
377 return ret; 430 kfree(caching_ctl);
431 return 0;
378 } 432 }
433 cache->caching_ctl = caching_ctl;
379 cache->cached = BTRFS_CACHE_STARTED; 434 cache->cached = BTRFS_CACHE_STARTED;
380 spin_unlock(&cache->lock); 435 spin_unlock(&cache->lock);
381 436
437 down_write(&fs_info->extent_commit_sem);
438 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
439 up_write(&fs_info->extent_commit_sem);
440
441 atomic_inc(&cache->space_info->caching_threads);
442
382 tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n", 443 tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
383 cache->key.objectid); 444 cache->key.objectid);
384 if (IS_ERR(tsk)) { 445 if (IS_ERR(tsk)) {
@@ -1511,7 +1572,8 @@ static int remove_extent_backref(struct btrfs_trans_handle *trans,
1511static void btrfs_issue_discard(struct block_device *bdev, 1572static void btrfs_issue_discard(struct block_device *bdev,
1512 u64 start, u64 len) 1573 u64 start, u64 len)
1513{ 1574{
1514 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL); 1575 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
1576 DISCARD_FL_BARRIER);
1515} 1577}
1516#endif 1578#endif
1517 1579
@@ -1656,7 +1718,6 @@ static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1656 parent, ref_root, flags, 1718 parent, ref_root, flags,
1657 ref->objectid, ref->offset, 1719 ref->objectid, ref->offset,
1658 &ins, node->ref_mod); 1720 &ins, node->ref_mod);
1659 update_reserved_extents(root, ins.objectid, ins.offset, 0);
1660 } else if (node->action == BTRFS_ADD_DELAYED_REF) { 1721 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1661 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, 1722 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1662 node->num_bytes, parent, 1723 node->num_bytes, parent,
@@ -1782,7 +1843,6 @@ static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1782 extent_op->flags_to_set, 1843 extent_op->flags_to_set,
1783 &extent_op->key, 1844 &extent_op->key,
1784 ref->level, &ins); 1845 ref->level, &ins);
1785 update_reserved_extents(root, ins.objectid, ins.offset, 0);
1786 } else if (node->action == BTRFS_ADD_DELAYED_REF) { 1846 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1787 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, 1847 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1788 node->num_bytes, parent, ref_root, 1848 node->num_bytes, parent, ref_root,
@@ -1817,16 +1877,32 @@ static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1817 BUG_ON(extent_op); 1877 BUG_ON(extent_op);
1818 head = btrfs_delayed_node_to_head(node); 1878 head = btrfs_delayed_node_to_head(node);
1819 if (insert_reserved) { 1879 if (insert_reserved) {
1880 int mark_free = 0;
1881 struct extent_buffer *must_clean = NULL;
1882
1883 ret = pin_down_bytes(trans, root, NULL,
1884 node->bytenr, node->num_bytes,
1885 head->is_data, 1, &must_clean);
1886 if (ret > 0)
1887 mark_free = 1;
1888
1889 if (must_clean) {
1890 clean_tree_block(NULL, root, must_clean);
1891 btrfs_tree_unlock(must_clean);
1892 free_extent_buffer(must_clean);
1893 }
1820 if (head->is_data) { 1894 if (head->is_data) {
1821 ret = btrfs_del_csums(trans, root, 1895 ret = btrfs_del_csums(trans, root,
1822 node->bytenr, 1896 node->bytenr,
1823 node->num_bytes); 1897 node->num_bytes);
1824 BUG_ON(ret); 1898 BUG_ON(ret);
1825 } 1899 }
1826 btrfs_update_pinned_extents(root, node->bytenr, 1900 if (mark_free) {
1827 node->num_bytes, 1); 1901 ret = btrfs_free_reserved_extent(root,
1828 update_reserved_extents(root, node->bytenr, 1902 node->bytenr,
1829 node->num_bytes, 0); 1903 node->num_bytes);
1904 BUG_ON(ret);
1905 }
1830 } 1906 }
1831 mutex_unlock(&head->mutex); 1907 mutex_unlock(&head->mutex);
1832 return 0; 1908 return 0;
@@ -2691,60 +2767,448 @@ void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2691 alloc_target); 2767 alloc_target);
2692} 2768}
2693 2769
2770static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items)
2771{
2772 u64 num_bytes;
2773 int level;
2774
2775 level = BTRFS_MAX_LEVEL - 2;
2776 /*
2777 * NOTE: these calculations are absolutely the worst possible case.
2778 * This assumes that _every_ item we insert will require a new leaf, and
2779 * that the tree has grown to its maximum level size.
2780 */
2781
2782 /*
2783 * for every item we insert we could insert both an extent item and a
2784 * extent ref item. Then for ever item we insert, we will need to cow
2785 * both the original leaf, plus the leaf to the left and right of it.
2786 *
2787 * Unless we are talking about the extent root, then we just want the
2788 * number of items * 2, since we just need the extent item plus its ref.
2789 */
2790 if (root == root->fs_info->extent_root)
2791 num_bytes = num_items * 2;
2792 else
2793 num_bytes = (num_items + (2 * num_items)) * 3;
2794
2795 /*
2796 * num_bytes is total number of leaves we could need times the leaf
2797 * size, and then for every leaf we could end up cow'ing 2 nodes per
2798 * level, down to the leaf level.
2799 */
2800 num_bytes = (num_bytes * root->leafsize) +
2801 (num_bytes * (level * 2)) * root->nodesize;
2802
2803 return num_bytes;
2804}
2805
2694/* 2806/*
2695 * for now this just makes sure we have at least 5% of our metadata space free 2807 * Unreserve metadata space for delalloc. If we have less reserved credits than
2696 * for use. 2808 * we have extents, this function does nothing.
2697 */ 2809 */
2698int btrfs_check_metadata_free_space(struct btrfs_root *root) 2810int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
2811 struct inode *inode, int num_items)
2699{ 2812{
2700 struct btrfs_fs_info *info = root->fs_info; 2813 struct btrfs_fs_info *info = root->fs_info;
2701 struct btrfs_space_info *meta_sinfo; 2814 struct btrfs_space_info *meta_sinfo;
2702 u64 alloc_target, thresh; 2815 u64 num_bytes;
2703 int committed = 0, ret; 2816 u64 alloc_target;
2817 bool bug = false;
2704 2818
2705 /* get the space info for where the metadata will live */ 2819 /* get the space info for where the metadata will live */
2706 alloc_target = btrfs_get_alloc_profile(root, 0); 2820 alloc_target = btrfs_get_alloc_profile(root, 0);
2707 meta_sinfo = __find_space_info(info, alloc_target); 2821 meta_sinfo = __find_space_info(info, alloc_target);
2708 2822
2709again: 2823 num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
2824 num_items);
2825
2710 spin_lock(&meta_sinfo->lock); 2826 spin_lock(&meta_sinfo->lock);
2711 if (!meta_sinfo->full) 2827 spin_lock(&BTRFS_I(inode)->accounting_lock);
2712 thresh = meta_sinfo->total_bytes * 80; 2828 if (BTRFS_I(inode)->reserved_extents <=
2713 else 2829 BTRFS_I(inode)->outstanding_extents) {
2714 thresh = meta_sinfo->total_bytes * 95; 2830 spin_unlock(&BTRFS_I(inode)->accounting_lock);
2831 spin_unlock(&meta_sinfo->lock);
2832 return 0;
2833 }
2834 spin_unlock(&BTRFS_I(inode)->accounting_lock);
2835
2836 BTRFS_I(inode)->reserved_extents--;
2837 BUG_ON(BTRFS_I(inode)->reserved_extents < 0);
2838
2839 if (meta_sinfo->bytes_delalloc < num_bytes) {
2840 bug = true;
2841 meta_sinfo->bytes_delalloc = 0;
2842 } else {
2843 meta_sinfo->bytes_delalloc -= num_bytes;
2844 }
2845 spin_unlock(&meta_sinfo->lock);
2846
2847 BUG_ON(bug);
2848
2849 return 0;
2850}
2851
2852static void check_force_delalloc(struct btrfs_space_info *meta_sinfo)
2853{
2854 u64 thresh;
2855
2856 thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2857 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
2858 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
2859 meta_sinfo->bytes_may_use;
2715 2860
2861 thresh = meta_sinfo->total_bytes - thresh;
2862 thresh *= 80;
2716 do_div(thresh, 100); 2863 do_div(thresh, 100);
2864 if (thresh <= meta_sinfo->bytes_delalloc)
2865 meta_sinfo->force_delalloc = 1;
2866 else
2867 meta_sinfo->force_delalloc = 0;
2868}
2717 2869
2718 if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved + 2870struct async_flush {
2719 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) { 2871 struct btrfs_root *root;
2720 struct btrfs_trans_handle *trans; 2872 struct btrfs_space_info *info;
2721 if (!meta_sinfo->full) { 2873 struct btrfs_work work;
2722 meta_sinfo->force_alloc = 1; 2874};
2723 spin_unlock(&meta_sinfo->lock);
2724 2875
2725 trans = btrfs_start_transaction(root, 1); 2876static noinline void flush_delalloc_async(struct btrfs_work *work)
2726 if (!trans) 2877{
2727 return -ENOMEM; 2878 struct async_flush *async;
2879 struct btrfs_root *root;
2880 struct btrfs_space_info *info;
2728 2881
2729 ret = do_chunk_alloc(trans, root->fs_info->extent_root, 2882 async = container_of(work, struct async_flush, work);
2730 2 * 1024 * 1024, alloc_target, 0); 2883 root = async->root;
2731 btrfs_end_transaction(trans, root); 2884 info = async->info;
2885
2886 btrfs_start_delalloc_inodes(root);
2887 wake_up(&info->flush_wait);
2888 btrfs_wait_ordered_extents(root, 0);
2889
2890 spin_lock(&info->lock);
2891 info->flushing = 0;
2892 spin_unlock(&info->lock);
2893 wake_up(&info->flush_wait);
2894
2895 kfree(async);
2896}
2897
2898static void wait_on_flush(struct btrfs_space_info *info)
2899{
2900 DEFINE_WAIT(wait);
2901 u64 used;
2902
2903 while (1) {
2904 prepare_to_wait(&info->flush_wait, &wait,
2905 TASK_UNINTERRUPTIBLE);
2906 spin_lock(&info->lock);
2907 if (!info->flushing) {
2908 spin_unlock(&info->lock);
2909 break;
2910 }
2911
2912 used = info->bytes_used + info->bytes_reserved +
2913 info->bytes_pinned + info->bytes_readonly +
2914 info->bytes_super + info->bytes_root +
2915 info->bytes_may_use + info->bytes_delalloc;
2916 if (used < info->total_bytes) {
2917 spin_unlock(&info->lock);
2918 break;
2919 }
2920 spin_unlock(&info->lock);
2921 schedule();
2922 }
2923 finish_wait(&info->flush_wait, &wait);
2924}
2925
2926static void flush_delalloc(struct btrfs_root *root,
2927 struct btrfs_space_info *info)
2928{
2929 struct async_flush *async;
2930 bool wait = false;
2931
2932 spin_lock(&info->lock);
2933
2934 if (!info->flushing) {
2935 info->flushing = 1;
2936 init_waitqueue_head(&info->flush_wait);
2937 } else {
2938 wait = true;
2939 }
2940
2941 spin_unlock(&info->lock);
2942
2943 if (wait) {
2944 wait_on_flush(info);
2945 return;
2946 }
2947
2948 async = kzalloc(sizeof(*async), GFP_NOFS);
2949 if (!async)
2950 goto flush;
2951
2952 async->root = root;
2953 async->info = info;
2954 async->work.func = flush_delalloc_async;
2955
2956 btrfs_queue_worker(&root->fs_info->enospc_workers,
2957 &async->work);
2958 wait_on_flush(info);
2959 return;
2960
2961flush:
2962 btrfs_start_delalloc_inodes(root);
2963 btrfs_wait_ordered_extents(root, 0);
2964
2965 spin_lock(&info->lock);
2966 info->flushing = 0;
2967 spin_unlock(&info->lock);
2968 wake_up(&info->flush_wait);
2969}
2970
2971static int maybe_allocate_chunk(struct btrfs_root *root,
2972 struct btrfs_space_info *info)
2973{
2974 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
2975 struct btrfs_trans_handle *trans;
2976 bool wait = false;
2977 int ret = 0;
2978 u64 min_metadata;
2979 u64 free_space;
2980
2981 free_space = btrfs_super_total_bytes(disk_super);
2982 /*
2983 * we allow the metadata to grow to a max of either 5gb or 5% of the
2984 * space in the volume.
2985 */
2986 min_metadata = min((u64)5 * 1024 * 1024 * 1024,
2987 div64_u64(free_space * 5, 100));
2988 if (info->total_bytes >= min_metadata) {
2989 spin_unlock(&info->lock);
2990 return 0;
2991 }
2992
2993 if (info->full) {
2994 spin_unlock(&info->lock);
2995 return 0;
2996 }
2997
2998 if (!info->allocating_chunk) {
2999 info->force_alloc = 1;
3000 info->allocating_chunk = 1;
3001 init_waitqueue_head(&info->allocate_wait);
3002 } else {
3003 wait = true;
3004 }
3005
3006 spin_unlock(&info->lock);
3007
3008 if (wait) {
3009 wait_event(info->allocate_wait,
3010 !info->allocating_chunk);
3011 return 1;
3012 }
3013
3014 trans = btrfs_start_transaction(root, 1);
3015 if (!trans) {
3016 ret = -ENOMEM;
3017 goto out;
3018 }
3019
3020 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3021 4096 + 2 * 1024 * 1024,
3022 info->flags, 0);
3023 btrfs_end_transaction(trans, root);
3024 if (ret)
3025 goto out;
3026out:
3027 spin_lock(&info->lock);
3028 info->allocating_chunk = 0;
3029 spin_unlock(&info->lock);
3030 wake_up(&info->allocate_wait);
3031
3032 if (ret)
3033 return 0;
3034 return 1;
3035}
3036
3037/*
3038 * Reserve metadata space for delalloc.
3039 */
3040int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
3041 struct inode *inode, int num_items)
3042{
3043 struct btrfs_fs_info *info = root->fs_info;
3044 struct btrfs_space_info *meta_sinfo;
3045 u64 num_bytes;
3046 u64 used;
3047 u64 alloc_target;
3048 int flushed = 0;
3049 int force_delalloc;
3050
3051 /* get the space info for where the metadata will live */
3052 alloc_target = btrfs_get_alloc_profile(root, 0);
3053 meta_sinfo = __find_space_info(info, alloc_target);
3054
3055 num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
3056 num_items);
3057again:
3058 spin_lock(&meta_sinfo->lock);
3059
3060 force_delalloc = meta_sinfo->force_delalloc;
3061
3062 if (unlikely(!meta_sinfo->bytes_root))
3063 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3064
3065 if (!flushed)
3066 meta_sinfo->bytes_delalloc += num_bytes;
3067
3068 used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3069 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3070 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3071 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3072
3073 if (used > meta_sinfo->total_bytes) {
3074 flushed++;
3075
3076 if (flushed == 1) {
3077 if (maybe_allocate_chunk(root, meta_sinfo))
3078 goto again;
3079 flushed++;
3080 } else {
3081 spin_unlock(&meta_sinfo->lock);
3082 }
3083
3084 if (flushed == 2) {
3085 filemap_flush(inode->i_mapping);
3086 goto again;
3087 } else if (flushed == 3) {
3088 flush_delalloc(root, meta_sinfo);
2732 goto again; 3089 goto again;
2733 } 3090 }
3091 spin_lock(&meta_sinfo->lock);
3092 meta_sinfo->bytes_delalloc -= num_bytes;
2734 spin_unlock(&meta_sinfo->lock); 3093 spin_unlock(&meta_sinfo->lock);
3094 printk(KERN_ERR "enospc, has %d, reserved %d\n",
3095 BTRFS_I(inode)->outstanding_extents,
3096 BTRFS_I(inode)->reserved_extents);
3097 dump_space_info(meta_sinfo, 0, 0);
3098 return -ENOSPC;
3099 }
2735 3100
2736 if (!committed) { 3101 BTRFS_I(inode)->reserved_extents++;
2737 committed = 1; 3102 check_force_delalloc(meta_sinfo);
2738 trans = btrfs_join_transaction(root, 1); 3103 spin_unlock(&meta_sinfo->lock);
2739 if (!trans) 3104
2740 return -ENOMEM; 3105 if (!flushed && force_delalloc)
2741 ret = btrfs_commit_transaction(trans, root); 3106 filemap_flush(inode->i_mapping);
2742 if (ret) 3107
2743 return ret; 3108 return 0;
3109}
3110
3111/*
3112 * unreserve num_items number of items worth of metadata space. This needs to
3113 * be paired with btrfs_reserve_metadata_space.
3114 *
3115 * NOTE: if you have the option, run this _AFTER_ you do a
3116 * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref
3117 * oprations which will result in more used metadata, so we want to make sure we
3118 * can do that without issue.
3119 */
3120int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items)
3121{
3122 struct btrfs_fs_info *info = root->fs_info;
3123 struct btrfs_space_info *meta_sinfo;
3124 u64 num_bytes;
3125 u64 alloc_target;
3126 bool bug = false;
3127
3128 /* get the space info for where the metadata will live */
3129 alloc_target = btrfs_get_alloc_profile(root, 0);
3130 meta_sinfo = __find_space_info(info, alloc_target);
3131
3132 num_bytes = calculate_bytes_needed(root, num_items);
3133
3134 spin_lock(&meta_sinfo->lock);
3135 if (meta_sinfo->bytes_may_use < num_bytes) {
3136 bug = true;
3137 meta_sinfo->bytes_may_use = 0;
3138 } else {
3139 meta_sinfo->bytes_may_use -= num_bytes;
3140 }
3141 spin_unlock(&meta_sinfo->lock);
3142
3143 BUG_ON(bug);
3144
3145 return 0;
3146}
3147
3148/*
3149 * Reserve some metadata space for use. We'll calculate the worste case number
3150 * of bytes that would be needed to modify num_items number of items. If we
3151 * have space, fantastic, if not, you get -ENOSPC. Please call
3152 * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of
3153 * items you reserved, since whatever metadata you needed should have already
3154 * been allocated.
3155 *
3156 * This will commit the transaction to make more space if we don't have enough
3157 * metadata space. THe only time we don't do this is if we're reserving space
3158 * inside of a transaction, then we will just return -ENOSPC and it is the
3159 * callers responsibility to handle it properly.
3160 */
3161int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items)
3162{
3163 struct btrfs_fs_info *info = root->fs_info;
3164 struct btrfs_space_info *meta_sinfo;
3165 u64 num_bytes;
3166 u64 used;
3167 u64 alloc_target;
3168 int retries = 0;
3169
3170 /* get the space info for where the metadata will live */
3171 alloc_target = btrfs_get_alloc_profile(root, 0);
3172 meta_sinfo = __find_space_info(info, alloc_target);
3173
3174 num_bytes = calculate_bytes_needed(root, num_items);
3175again:
3176 spin_lock(&meta_sinfo->lock);
3177
3178 if (unlikely(!meta_sinfo->bytes_root))
3179 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3180
3181 if (!retries)
3182 meta_sinfo->bytes_may_use += num_bytes;
3183
3184 used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3185 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3186 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3187 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3188
3189 if (used > meta_sinfo->total_bytes) {
3190 retries++;
3191 if (retries == 1) {
3192 if (maybe_allocate_chunk(root, meta_sinfo))
3193 goto again;
3194 retries++;
3195 } else {
3196 spin_unlock(&meta_sinfo->lock);
3197 }
3198
3199 if (retries == 2) {
3200 flush_delalloc(root, meta_sinfo);
2744 goto again; 3201 goto again;
2745 } 3202 }
3203 spin_lock(&meta_sinfo->lock);
3204 meta_sinfo->bytes_may_use -= num_bytes;
3205 spin_unlock(&meta_sinfo->lock);
3206
3207 dump_space_info(meta_sinfo, 0, 0);
2746 return -ENOSPC; 3208 return -ENOSPC;
2747 } 3209 }
3210
3211 check_force_delalloc(meta_sinfo);
2748 spin_unlock(&meta_sinfo->lock); 3212 spin_unlock(&meta_sinfo->lock);
2749 3213
2750 return 0; 3214 return 0;
@@ -2764,13 +3228,16 @@ int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
2764 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); 3228 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2765 3229
2766 data_sinfo = BTRFS_I(inode)->space_info; 3230 data_sinfo = BTRFS_I(inode)->space_info;
3231 if (!data_sinfo)
3232 goto alloc;
3233
2767again: 3234again:
2768 /* make sure we have enough space to handle the data first */ 3235 /* make sure we have enough space to handle the data first */
2769 spin_lock(&data_sinfo->lock); 3236 spin_lock(&data_sinfo->lock);
2770 if (data_sinfo->total_bytes - data_sinfo->bytes_used - 3237 if (data_sinfo->total_bytes - data_sinfo->bytes_used -
2771 data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved - 3238 data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
2772 data_sinfo->bytes_pinned - data_sinfo->bytes_readonly - 3239 data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
2773 data_sinfo->bytes_may_use < bytes) { 3240 data_sinfo->bytes_may_use - data_sinfo->bytes_super < bytes) {
2774 struct btrfs_trans_handle *trans; 3241 struct btrfs_trans_handle *trans;
2775 3242
2776 /* 3243 /*
@@ -2782,7 +3249,7 @@ again:
2782 3249
2783 data_sinfo->force_alloc = 1; 3250 data_sinfo->force_alloc = 1;
2784 spin_unlock(&data_sinfo->lock); 3251 spin_unlock(&data_sinfo->lock);
2785 3252alloc:
2786 alloc_target = btrfs_get_alloc_profile(root, 1); 3253 alloc_target = btrfs_get_alloc_profile(root, 1);
2787 trans = btrfs_start_transaction(root, 1); 3254 trans = btrfs_start_transaction(root, 1);
2788 if (!trans) 3255 if (!trans)
@@ -2794,12 +3261,17 @@ again:
2794 btrfs_end_transaction(trans, root); 3261 btrfs_end_transaction(trans, root);
2795 if (ret) 3262 if (ret)
2796 return ret; 3263 return ret;
3264
3265 if (!data_sinfo) {
3266 btrfs_set_inode_space_info(root, inode);
3267 data_sinfo = BTRFS_I(inode)->space_info;
3268 }
2797 goto again; 3269 goto again;
2798 } 3270 }
2799 spin_unlock(&data_sinfo->lock); 3271 spin_unlock(&data_sinfo->lock);
2800 3272
2801 /* commit the current transaction and try again */ 3273 /* commit the current transaction and try again */
2802 if (!committed) { 3274 if (!committed && !root->fs_info->open_ioctl_trans) {
2803 committed = 1; 3275 committed = 1;
2804 trans = btrfs_join_transaction(root, 1); 3276 trans = btrfs_join_transaction(root, 1);
2805 if (!trans) 3277 if (!trans)
@@ -2827,7 +3299,7 @@ again:
2827 BTRFS_I(inode)->reserved_bytes += bytes; 3299 BTRFS_I(inode)->reserved_bytes += bytes;
2828 spin_unlock(&data_sinfo->lock); 3300 spin_unlock(&data_sinfo->lock);
2829 3301
2830 return btrfs_check_metadata_free_space(root); 3302 return 0;
2831} 3303}
2832 3304
2833/* 3305/*
@@ -2926,17 +3398,15 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans,
2926 BUG_ON(!space_info); 3398 BUG_ON(!space_info);
2927 3399
2928 spin_lock(&space_info->lock); 3400 spin_lock(&space_info->lock);
2929 if (space_info->force_alloc) { 3401 if (space_info->force_alloc)
2930 force = 1; 3402 force = 1;
2931 space_info->force_alloc = 0;
2932 }
2933 if (space_info->full) { 3403 if (space_info->full) {
2934 spin_unlock(&space_info->lock); 3404 spin_unlock(&space_info->lock);
2935 goto out; 3405 goto out;
2936 } 3406 }
2937 3407
2938 thresh = space_info->total_bytes - space_info->bytes_readonly; 3408 thresh = space_info->total_bytes - space_info->bytes_readonly;
2939 thresh = div_factor(thresh, 6); 3409 thresh = div_factor(thresh, 8);
2940 if (!force && 3410 if (!force &&
2941 (space_info->bytes_used + space_info->bytes_pinned + 3411 (space_info->bytes_used + space_info->bytes_pinned +
2942 space_info->bytes_reserved + alloc_bytes) < thresh) { 3412 space_info->bytes_reserved + alloc_bytes) < thresh) {
@@ -2950,7 +3420,7 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans,
2950 * we keep a reasonable number of metadata chunks allocated in the 3420 * we keep a reasonable number of metadata chunks allocated in the
2951 * FS as well. 3421 * FS as well.
2952 */ 3422 */
2953 if (flags & BTRFS_BLOCK_GROUP_DATA) { 3423 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
2954 fs_info->data_chunk_allocations++; 3424 fs_info->data_chunk_allocations++;
2955 if (!(fs_info->data_chunk_allocations % 3425 if (!(fs_info->data_chunk_allocations %
2956 fs_info->metadata_ratio)) 3426 fs_info->metadata_ratio))
@@ -2958,8 +3428,11 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans,
2958 } 3428 }
2959 3429
2960 ret = btrfs_alloc_chunk(trans, extent_root, flags); 3430 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3431 spin_lock(&space_info->lock);
2961 if (ret) 3432 if (ret)
2962 space_info->full = 1; 3433 space_info->full = 1;
3434 space_info->force_alloc = 0;
3435 spin_unlock(&space_info->lock);
2963out: 3436out:
2964 mutex_unlock(&extent_root->fs_info->chunk_mutex); 3437 mutex_unlock(&extent_root->fs_info->chunk_mutex);
2965 return ret; 3438 return ret;
@@ -3008,10 +3481,12 @@ static int update_block_group(struct btrfs_trans_handle *trans,
3008 num_bytes = min(total, cache->key.offset - byte_in_group); 3481 num_bytes = min(total, cache->key.offset - byte_in_group);
3009 if (alloc) { 3482 if (alloc) {
3010 old_val += num_bytes; 3483 old_val += num_bytes;
3484 btrfs_set_block_group_used(&cache->item, old_val);
3485 cache->reserved -= num_bytes;
3011 cache->space_info->bytes_used += num_bytes; 3486 cache->space_info->bytes_used += num_bytes;
3487 cache->space_info->bytes_reserved -= num_bytes;
3012 if (cache->ro) 3488 if (cache->ro)
3013 cache->space_info->bytes_readonly -= num_bytes; 3489 cache->space_info->bytes_readonly -= num_bytes;
3014 btrfs_set_block_group_used(&cache->item, old_val);
3015 spin_unlock(&cache->lock); 3490 spin_unlock(&cache->lock);
3016 spin_unlock(&cache->space_info->lock); 3491 spin_unlock(&cache->space_info->lock);
3017 } else { 3492 } else {
@@ -3056,127 +3531,136 @@ static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3056 return bytenr; 3531 return bytenr;
3057} 3532}
3058 3533
3059int btrfs_update_pinned_extents(struct btrfs_root *root, 3534/*
3060 u64 bytenr, u64 num, int pin) 3535 * this function must be called within transaction
3536 */
3537int btrfs_pin_extent(struct btrfs_root *root,
3538 u64 bytenr, u64 num_bytes, int reserved)
3061{ 3539{
3062 u64 len;
3063 struct btrfs_block_group_cache *cache;
3064 struct btrfs_fs_info *fs_info = root->fs_info; 3540 struct btrfs_fs_info *fs_info = root->fs_info;
3541 struct btrfs_block_group_cache *cache;
3065 3542
3066 if (pin) 3543 cache = btrfs_lookup_block_group(fs_info, bytenr);
3067 set_extent_dirty(&fs_info->pinned_extents, 3544 BUG_ON(!cache);
3068 bytenr, bytenr + num - 1, GFP_NOFS);
3069
3070 while (num > 0) {
3071 cache = btrfs_lookup_block_group(fs_info, bytenr);
3072 BUG_ON(!cache);
3073 len = min(num, cache->key.offset -
3074 (bytenr - cache->key.objectid));
3075 if (pin) {
3076 spin_lock(&cache->space_info->lock);
3077 spin_lock(&cache->lock);
3078 cache->pinned += len;
3079 cache->space_info->bytes_pinned += len;
3080 spin_unlock(&cache->lock);
3081 spin_unlock(&cache->space_info->lock);
3082 fs_info->total_pinned += len;
3083 } else {
3084 int unpin = 0;
3085 3545
3086 /* 3546 spin_lock(&cache->space_info->lock);
3087 * in order to not race with the block group caching, we 3547 spin_lock(&cache->lock);
3088 * only want to unpin the extent if we are cached. If 3548 cache->pinned += num_bytes;
3089 * we aren't cached, we want to start async caching this 3549 cache->space_info->bytes_pinned += num_bytes;
3090 * block group so we can free the extent the next time 3550 if (reserved) {
3091 * around. 3551 cache->reserved -= num_bytes;
3092 */ 3552 cache->space_info->bytes_reserved -= num_bytes;
3093 spin_lock(&cache->space_info->lock); 3553 }
3094 spin_lock(&cache->lock); 3554 spin_unlock(&cache->lock);
3095 unpin = (cache->cached == BTRFS_CACHE_FINISHED); 3555 spin_unlock(&cache->space_info->lock);
3096 if (likely(unpin)) {
3097 cache->pinned -= len;
3098 cache->space_info->bytes_pinned -= len;
3099 fs_info->total_pinned -= len;
3100 }
3101 spin_unlock(&cache->lock);
3102 spin_unlock(&cache->space_info->lock);
3103 3556
3104 if (likely(unpin)) 3557 btrfs_put_block_group(cache);
3105 clear_extent_dirty(&fs_info->pinned_extents,
3106 bytenr, bytenr + len -1,
3107 GFP_NOFS);
3108 else
3109 cache_block_group(cache);
3110 3558
3111 if (unpin) 3559 set_extent_dirty(fs_info->pinned_extents,
3112 btrfs_add_free_space(cache, bytenr, len); 3560 bytenr, bytenr + num_bytes - 1, GFP_NOFS);
3113 } 3561 return 0;
3114 btrfs_put_block_group(cache); 3562}
3115 bytenr += len; 3563
3116 num -= len; 3564static int update_reserved_extents(struct btrfs_block_group_cache *cache,
3565 u64 num_bytes, int reserve)
3566{
3567 spin_lock(&cache->space_info->lock);
3568 spin_lock(&cache->lock);
3569 if (reserve) {
3570 cache->reserved += num_bytes;
3571 cache->space_info->bytes_reserved += num_bytes;
3572 } else {
3573 cache->reserved -= num_bytes;
3574 cache->space_info->bytes_reserved -= num_bytes;
3117 } 3575 }
3576 spin_unlock(&cache->lock);
3577 spin_unlock(&cache->space_info->lock);
3118 return 0; 3578 return 0;
3119} 3579}
3120 3580
3121static int update_reserved_extents(struct btrfs_root *root, 3581int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3122 u64 bytenr, u64 num, int reserve) 3582 struct btrfs_root *root)
3123{ 3583{
3124 u64 len;
3125 struct btrfs_block_group_cache *cache;
3126 struct btrfs_fs_info *fs_info = root->fs_info; 3584 struct btrfs_fs_info *fs_info = root->fs_info;
3585 struct btrfs_caching_control *next;
3586 struct btrfs_caching_control *caching_ctl;
3587 struct btrfs_block_group_cache *cache;
3127 3588
3128 while (num > 0) { 3589 down_write(&fs_info->extent_commit_sem);
3129 cache = btrfs_lookup_block_group(fs_info, bytenr);
3130 BUG_ON(!cache);
3131 len = min(num, cache->key.offset -
3132 (bytenr - cache->key.objectid));
3133 3590
3134 spin_lock(&cache->space_info->lock); 3591 list_for_each_entry_safe(caching_ctl, next,
3135 spin_lock(&cache->lock); 3592 &fs_info->caching_block_groups, list) {
3136 if (reserve) { 3593 cache = caching_ctl->block_group;
3137 cache->reserved += len; 3594 if (block_group_cache_done(cache)) {
3138 cache->space_info->bytes_reserved += len; 3595 cache->last_byte_to_unpin = (u64)-1;
3596 list_del_init(&caching_ctl->list);
3597 put_caching_control(caching_ctl);
3139 } else { 3598 } else {
3140 cache->reserved -= len; 3599 cache->last_byte_to_unpin = caching_ctl->progress;
3141 cache->space_info->bytes_reserved -= len;
3142 } 3600 }
3143 spin_unlock(&cache->lock);
3144 spin_unlock(&cache->space_info->lock);
3145 btrfs_put_block_group(cache);
3146 bytenr += len;
3147 num -= len;
3148 } 3601 }
3602
3603 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3604 fs_info->pinned_extents = &fs_info->freed_extents[1];
3605 else
3606 fs_info->pinned_extents = &fs_info->freed_extents[0];
3607
3608 up_write(&fs_info->extent_commit_sem);
3149 return 0; 3609 return 0;
3150} 3610}
3151 3611
3152int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy) 3612static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3153{ 3613{
3154 u64 last = 0; 3614 struct btrfs_fs_info *fs_info = root->fs_info;
3155 u64 start; 3615 struct btrfs_block_group_cache *cache = NULL;
3156 u64 end; 3616 u64 len;
3157 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
3158 int ret;
3159 3617
3160 while (1) { 3618 while (start <= end) {
3161 ret = find_first_extent_bit(pinned_extents, last, 3619 if (!cache ||
3162 &start, &end, EXTENT_DIRTY); 3620 start >= cache->key.objectid + cache->key.offset) {
3163 if (ret) 3621 if (cache)
3164 break; 3622 btrfs_put_block_group(cache);
3623 cache = btrfs_lookup_block_group(fs_info, start);
3624 BUG_ON(!cache);
3625 }
3626
3627 len = cache->key.objectid + cache->key.offset - start;
3628 len = min(len, end + 1 - start);
3165 3629
3166 set_extent_dirty(copy, start, end, GFP_NOFS); 3630 if (start < cache->last_byte_to_unpin) {
3167 last = end + 1; 3631 len = min(len, cache->last_byte_to_unpin - start);
3632 btrfs_add_free_space(cache, start, len);
3633 }
3634
3635 spin_lock(&cache->space_info->lock);
3636 spin_lock(&cache->lock);
3637 cache->pinned -= len;
3638 cache->space_info->bytes_pinned -= len;
3639 spin_unlock(&cache->lock);
3640 spin_unlock(&cache->space_info->lock);
3641
3642 start += len;
3168 } 3643 }
3644
3645 if (cache)
3646 btrfs_put_block_group(cache);
3169 return 0; 3647 return 0;
3170} 3648}
3171 3649
3172int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, 3650int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3173 struct btrfs_root *root, 3651 struct btrfs_root *root)
3174 struct extent_io_tree *unpin)
3175{ 3652{
3653 struct btrfs_fs_info *fs_info = root->fs_info;
3654 struct extent_io_tree *unpin;
3176 u64 start; 3655 u64 start;
3177 u64 end; 3656 u64 end;
3178 int ret; 3657 int ret;
3179 3658
3659 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3660 unpin = &fs_info->freed_extents[1];
3661 else
3662 unpin = &fs_info->freed_extents[0];
3663
3180 while (1) { 3664 while (1) {
3181 ret = find_first_extent_bit(unpin, 0, &start, &end, 3665 ret = find_first_extent_bit(unpin, 0, &start, &end,
3182 EXTENT_DIRTY); 3666 EXTENT_DIRTY);
@@ -3185,10 +3669,8 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3185 3669
3186 ret = btrfs_discard_extent(root, start, end + 1 - start); 3670 ret = btrfs_discard_extent(root, start, end + 1 - start);
3187 3671
3188 /* unlocks the pinned mutex */
3189 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
3190 clear_extent_dirty(unpin, start, end, GFP_NOFS); 3672 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3191 3673 unpin_extent_range(root, start, end);
3192 cond_resched(); 3674 cond_resched();
3193 } 3675 }
3194 3676
@@ -3198,7 +3680,8 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3198static int pin_down_bytes(struct btrfs_trans_handle *trans, 3680static int pin_down_bytes(struct btrfs_trans_handle *trans,
3199 struct btrfs_root *root, 3681 struct btrfs_root *root,
3200 struct btrfs_path *path, 3682 struct btrfs_path *path,
3201 u64 bytenr, u64 num_bytes, int is_data, 3683 u64 bytenr, u64 num_bytes,
3684 int is_data, int reserved,
3202 struct extent_buffer **must_clean) 3685 struct extent_buffer **must_clean)
3203{ 3686{
3204 int err = 0; 3687 int err = 0;
@@ -3230,15 +3713,15 @@ static int pin_down_bytes(struct btrfs_trans_handle *trans,
3230 } 3713 }
3231 free_extent_buffer(buf); 3714 free_extent_buffer(buf);
3232pinit: 3715pinit:
3233 btrfs_set_path_blocking(path); 3716 if (path)
3717 btrfs_set_path_blocking(path);
3234 /* unlocks the pinned mutex */ 3718 /* unlocks the pinned mutex */
3235 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); 3719 btrfs_pin_extent(root, bytenr, num_bytes, reserved);
3236 3720
3237 BUG_ON(err < 0); 3721 BUG_ON(err < 0);
3238 return 0; 3722 return 0;
3239} 3723}
3240 3724
3241
3242static int __btrfs_free_extent(struct btrfs_trans_handle *trans, 3725static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3243 struct btrfs_root *root, 3726 struct btrfs_root *root,
3244 u64 bytenr, u64 num_bytes, u64 parent, 3727 u64 bytenr, u64 num_bytes, u64 parent,
@@ -3412,7 +3895,7 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3412 } 3895 }
3413 3896
3414 ret = pin_down_bytes(trans, root, path, bytenr, 3897 ret = pin_down_bytes(trans, root, path, bytenr,
3415 num_bytes, is_data, &must_clean); 3898 num_bytes, is_data, 0, &must_clean);
3416 if (ret > 0) 3899 if (ret > 0)
3417 mark_free = 1; 3900 mark_free = 1;
3418 BUG_ON(ret < 0); 3901 BUG_ON(ret < 0);
@@ -3543,8 +4026,7 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans,
3543 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { 4026 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
3544 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); 4027 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
3545 /* unlocks the pinned mutex */ 4028 /* unlocks the pinned mutex */
3546 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); 4029 btrfs_pin_extent(root, bytenr, num_bytes, 1);
3547 update_reserved_extents(root, bytenr, num_bytes, 0);
3548 ret = 0; 4030 ret = 0;
3549 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { 4031 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
3550 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, 4032 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
@@ -3584,19 +4066,33 @@ static noinline int
3584wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, 4066wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
3585 u64 num_bytes) 4067 u64 num_bytes)
3586{ 4068{
4069 struct btrfs_caching_control *caching_ctl;
3587 DEFINE_WAIT(wait); 4070 DEFINE_WAIT(wait);
3588 4071
3589 prepare_to_wait(&cache->caching_q, &wait, TASK_UNINTERRUPTIBLE); 4072 caching_ctl = get_caching_control(cache);
3590 4073 if (!caching_ctl)
3591 if (block_group_cache_done(cache)) {
3592 finish_wait(&cache->caching_q, &wait);
3593 return 0; 4074 return 0;
3594 }
3595 schedule();
3596 finish_wait(&cache->caching_q, &wait);
3597 4075
3598 wait_event(cache->caching_q, block_group_cache_done(cache) || 4076 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
3599 (cache->free_space >= num_bytes)); 4077 (cache->free_space >= num_bytes));
4078
4079 put_caching_control(caching_ctl);
4080 return 0;
4081}
4082
4083static noinline int
4084wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4085{
4086 struct btrfs_caching_control *caching_ctl;
4087 DEFINE_WAIT(wait);
4088
4089 caching_ctl = get_caching_control(cache);
4090 if (!caching_ctl)
4091 return 0;
4092
4093 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4094
4095 put_caching_control(caching_ctl);
3600 return 0; 4096 return 0;
3601} 4097}
3602 4098
@@ -3634,6 +4130,8 @@ static noinline int find_free_extent(struct btrfs_trans_handle *trans,
3634 int last_ptr_loop = 0; 4130 int last_ptr_loop = 0;
3635 int loop = 0; 4131 int loop = 0;
3636 bool found_uncached_bg = false; 4132 bool found_uncached_bg = false;
4133 bool failed_cluster_refill = false;
4134 bool failed_alloc = false;
3637 4135
3638 WARN_ON(num_bytes < root->sectorsize); 4136 WARN_ON(num_bytes < root->sectorsize);
3639 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); 4137 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
@@ -3731,7 +4229,16 @@ have_block_group:
3731 if (unlikely(block_group->ro)) 4229 if (unlikely(block_group->ro))
3732 goto loop; 4230 goto loop;
3733 4231
3734 if (last_ptr) { 4232 /*
4233 * Ok we want to try and use the cluster allocator, so lets look
4234 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4235 * have tried the cluster allocator plenty of times at this
4236 * point and not have found anything, so we are likely way too
4237 * fragmented for the clustering stuff to find anything, so lets
4238 * just skip it and let the allocator find whatever block it can
4239 * find
4240 */
4241 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
3735 /* 4242 /*
3736 * the refill lock keeps out other 4243 * the refill lock keeps out other
3737 * people trying to start a new cluster 4244 * people trying to start a new cluster
@@ -3806,9 +4313,11 @@ refill_cluster:
3806 spin_unlock(&last_ptr->refill_lock); 4313 spin_unlock(&last_ptr->refill_lock);
3807 goto checks; 4314 goto checks;
3808 } 4315 }
3809 } else if (!cached && loop > LOOP_CACHING_NOWAIT) { 4316 } else if (!cached && loop > LOOP_CACHING_NOWAIT
4317 && !failed_cluster_refill) {
3810 spin_unlock(&last_ptr->refill_lock); 4318 spin_unlock(&last_ptr->refill_lock);
3811 4319
4320 failed_cluster_refill = true;
3812 wait_block_group_cache_progress(block_group, 4321 wait_block_group_cache_progress(block_group,
3813 num_bytes + empty_cluster + empty_size); 4322 num_bytes + empty_cluster + empty_size);
3814 goto have_block_group; 4323 goto have_block_group;
@@ -3820,25 +4329,30 @@ refill_cluster:
3820 * cluster. Free the cluster we've been trying 4329 * cluster. Free the cluster we've been trying
3821 * to use, and go to the next block group 4330 * to use, and go to the next block group
3822 */ 4331 */
3823 if (loop < LOOP_NO_EMPTY_SIZE) { 4332 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3824 btrfs_return_cluster_to_free_space(NULL,
3825 last_ptr);
3826 spin_unlock(&last_ptr->refill_lock);
3827 goto loop;
3828 }
3829 spin_unlock(&last_ptr->refill_lock); 4333 spin_unlock(&last_ptr->refill_lock);
4334 goto loop;
3830 } 4335 }
3831 4336
3832 offset = btrfs_find_space_for_alloc(block_group, search_start, 4337 offset = btrfs_find_space_for_alloc(block_group, search_start,
3833 num_bytes, empty_size); 4338 num_bytes, empty_size);
3834 if (!offset && (cached || (!cached && 4339 /*
3835 loop == LOOP_CACHING_NOWAIT))) { 4340 * If we didn't find a chunk, and we haven't failed on this
3836 goto loop; 4341 * block group before, and this block group is in the middle of
3837 } else if (!offset && (!cached && 4342 * caching and we are ok with waiting, then go ahead and wait
3838 loop > LOOP_CACHING_NOWAIT)) { 4343 * for progress to be made, and set failed_alloc to true.
4344 *
4345 * If failed_alloc is true then we've already waited on this
4346 * block group once and should move on to the next block group.
4347 */
4348 if (!offset && !failed_alloc && !cached &&
4349 loop > LOOP_CACHING_NOWAIT) {
3839 wait_block_group_cache_progress(block_group, 4350 wait_block_group_cache_progress(block_group,
3840 num_bytes + empty_size); 4351 num_bytes + empty_size);
4352 failed_alloc = true;
3841 goto have_block_group; 4353 goto have_block_group;
4354 } else if (!offset) {
4355 goto loop;
3842 } 4356 }
3843checks: 4357checks:
3844 search_start = stripe_align(root, offset); 4358 search_start = stripe_align(root, offset);
@@ -3880,9 +4394,13 @@ checks:
3880 search_start - offset); 4394 search_start - offset);
3881 BUG_ON(offset > search_start); 4395 BUG_ON(offset > search_start);
3882 4396
4397 update_reserved_extents(block_group, num_bytes, 1);
4398
3883 /* we are all good, lets return */ 4399 /* we are all good, lets return */
3884 break; 4400 break;
3885loop: 4401loop:
4402 failed_cluster_refill = false;
4403 failed_alloc = false;
3886 btrfs_put_block_group(block_group); 4404 btrfs_put_block_group(block_group);
3887 } 4405 }
3888 up_read(&space_info->groups_sem); 4406 up_read(&space_info->groups_sem);
@@ -3940,21 +4458,32 @@ loop:
3940 return ret; 4458 return ret;
3941} 4459}
3942 4460
3943static void dump_space_info(struct btrfs_space_info *info, u64 bytes) 4461static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
4462 int dump_block_groups)
3944{ 4463{
3945 struct btrfs_block_group_cache *cache; 4464 struct btrfs_block_group_cache *cache;
3946 4465
4466 spin_lock(&info->lock);
3947 printk(KERN_INFO "space_info has %llu free, is %sfull\n", 4467 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
3948 (unsigned long long)(info->total_bytes - info->bytes_used - 4468 (unsigned long long)(info->total_bytes - info->bytes_used -
3949 info->bytes_pinned - info->bytes_reserved), 4469 info->bytes_pinned - info->bytes_reserved -
4470 info->bytes_super),
3950 (info->full) ? "" : "not "); 4471 (info->full) ? "" : "not ");
3951 printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu," 4472 printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
3952 " may_use=%llu, used=%llu\n", 4473 " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu"
4474 "\n",
3953 (unsigned long long)info->total_bytes, 4475 (unsigned long long)info->total_bytes,
3954 (unsigned long long)info->bytes_pinned, 4476 (unsigned long long)info->bytes_pinned,
3955 (unsigned long long)info->bytes_delalloc, 4477 (unsigned long long)info->bytes_delalloc,
3956 (unsigned long long)info->bytes_may_use, 4478 (unsigned long long)info->bytes_may_use,
3957 (unsigned long long)info->bytes_used); 4479 (unsigned long long)info->bytes_used,
4480 (unsigned long long)info->bytes_root,
4481 (unsigned long long)info->bytes_super,
4482 (unsigned long long)info->bytes_reserved);
4483 spin_unlock(&info->lock);
4484
4485 if (!dump_block_groups)
4486 return;
3958 4487
3959 down_read(&info->groups_sem); 4488 down_read(&info->groups_sem);
3960 list_for_each_entry(cache, &info->block_groups, list) { 4489 list_for_each_entry(cache, &info->block_groups, list) {
@@ -3972,12 +4501,12 @@ static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3972 up_read(&info->groups_sem); 4501 up_read(&info->groups_sem);
3973} 4502}
3974 4503
3975static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans, 4504int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3976 struct btrfs_root *root, 4505 struct btrfs_root *root,
3977 u64 num_bytes, u64 min_alloc_size, 4506 u64 num_bytes, u64 min_alloc_size,
3978 u64 empty_size, u64 hint_byte, 4507 u64 empty_size, u64 hint_byte,
3979 u64 search_end, struct btrfs_key *ins, 4508 u64 search_end, struct btrfs_key *ins,
3980 u64 data) 4509 u64 data)
3981{ 4510{
3982 int ret; 4511 int ret;
3983 u64 search_start = 0; 4512 u64 search_start = 0;
@@ -4022,7 +4551,7 @@ again:
4022 printk(KERN_ERR "btrfs allocation failed flags %llu, " 4551 printk(KERN_ERR "btrfs allocation failed flags %llu, "
4023 "wanted %llu\n", (unsigned long long)data, 4552 "wanted %llu\n", (unsigned long long)data,
4024 (unsigned long long)num_bytes); 4553 (unsigned long long)num_bytes);
4025 dump_space_info(sinfo, num_bytes); 4554 dump_space_info(sinfo, num_bytes, 1);
4026 } 4555 }
4027 4556
4028 return ret; 4557 return ret;
@@ -4043,25 +4572,8 @@ int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4043 ret = btrfs_discard_extent(root, start, len); 4572 ret = btrfs_discard_extent(root, start, len);
4044 4573
4045 btrfs_add_free_space(cache, start, len); 4574 btrfs_add_free_space(cache, start, len);
4575 update_reserved_extents(cache, len, 0);
4046 btrfs_put_block_group(cache); 4576 btrfs_put_block_group(cache);
4047 update_reserved_extents(root, start, len, 0);
4048
4049 return ret;
4050}
4051
4052int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4053 struct btrfs_root *root,
4054 u64 num_bytes, u64 min_alloc_size,
4055 u64 empty_size, u64 hint_byte,
4056 u64 search_end, struct btrfs_key *ins,
4057 u64 data)
4058{
4059 int ret;
4060 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
4061 empty_size, hint_byte, search_end, ins,
4062 data);
4063 if (!ret)
4064 update_reserved_extents(root, ins->objectid, ins->offset, 1);
4065 4577
4066 return ret; 4578 return ret;
4067} 4579}
@@ -4222,15 +4734,46 @@ int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4222{ 4734{
4223 int ret; 4735 int ret;
4224 struct btrfs_block_group_cache *block_group; 4736 struct btrfs_block_group_cache *block_group;
4737 struct btrfs_caching_control *caching_ctl;
4738 u64 start = ins->objectid;
4739 u64 num_bytes = ins->offset;
4225 4740
4226 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); 4741 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4227 cache_block_group(block_group); 4742 cache_block_group(block_group);
4228 wait_event(block_group->caching_q, 4743 caching_ctl = get_caching_control(block_group);
4229 block_group_cache_done(block_group));
4230 4744
4231 ret = btrfs_remove_free_space(block_group, ins->objectid, 4745 if (!caching_ctl) {
4232 ins->offset); 4746 BUG_ON(!block_group_cache_done(block_group));
4233 BUG_ON(ret); 4747 ret = btrfs_remove_free_space(block_group, start, num_bytes);
4748 BUG_ON(ret);
4749 } else {
4750 mutex_lock(&caching_ctl->mutex);
4751
4752 if (start >= caching_ctl->progress) {
4753 ret = add_excluded_extent(root, start, num_bytes);
4754 BUG_ON(ret);
4755 } else if (start + num_bytes <= caching_ctl->progress) {
4756 ret = btrfs_remove_free_space(block_group,
4757 start, num_bytes);
4758 BUG_ON(ret);
4759 } else {
4760 num_bytes = caching_ctl->progress - start;
4761 ret = btrfs_remove_free_space(block_group,
4762 start, num_bytes);
4763 BUG_ON(ret);
4764
4765 start = caching_ctl->progress;
4766 num_bytes = ins->objectid + ins->offset -
4767 caching_ctl->progress;
4768 ret = add_excluded_extent(root, start, num_bytes);
4769 BUG_ON(ret);
4770 }
4771
4772 mutex_unlock(&caching_ctl->mutex);
4773 put_caching_control(caching_ctl);
4774 }
4775
4776 update_reserved_extents(block_group, ins->offset, 1);
4234 btrfs_put_block_group(block_group); 4777 btrfs_put_block_group(block_group);
4235 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, 4778 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4236 0, owner, offset, ins, 1); 4779 0, owner, offset, ins, 1);
@@ -4254,9 +4797,9 @@ static int alloc_tree_block(struct btrfs_trans_handle *trans,
4254 int ret; 4797 int ret;
4255 u64 flags = 0; 4798 u64 flags = 0;
4256 4799
4257 ret = __btrfs_reserve_extent(trans, root, num_bytes, num_bytes, 4800 ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4258 empty_size, hint_byte, search_end, 4801 empty_size, hint_byte, search_end,
4259 ins, 0); 4802 ins, 0);
4260 if (ret) 4803 if (ret)
4261 return ret; 4804 return ret;
4262 4805
@@ -4267,7 +4810,6 @@ static int alloc_tree_block(struct btrfs_trans_handle *trans,
4267 } else 4810 } else
4268 BUG_ON(parent > 0); 4811 BUG_ON(parent > 0);
4269 4812
4270 update_reserved_extents(root, ins->objectid, ins->offset, 1);
4271 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { 4813 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4272 struct btrfs_delayed_extent_op *extent_op; 4814 struct btrfs_delayed_extent_op *extent_op;
4273 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); 4815 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
@@ -4346,452 +4888,108 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4346 return buf; 4888 return buf;
4347} 4889}
4348 4890
4349#if 0 4891struct walk_control {
4350int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans, 4892 u64 refs[BTRFS_MAX_LEVEL];
4351 struct btrfs_root *root, struct extent_buffer *leaf) 4893 u64 flags[BTRFS_MAX_LEVEL];
4352{ 4894 struct btrfs_key update_progress;
4353 u64 disk_bytenr; 4895 int stage;
4354 u64 num_bytes; 4896 int level;
4355 struct btrfs_key key; 4897 int shared_level;
4356 struct btrfs_file_extent_item *fi; 4898 int update_ref;
4357 u32 nritems; 4899 int keep_locks;
4358 int i; 4900 int reada_slot;
4359 int ret; 4901 int reada_count;
4360 4902};
4361 BUG_ON(!btrfs_is_leaf(leaf));
4362 nritems = btrfs_header_nritems(leaf);
4363
4364 for (i = 0; i < nritems; i++) {
4365 cond_resched();
4366 btrfs_item_key_to_cpu(leaf, &key, i);
4367
4368 /* only extents have references, skip everything else */
4369 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4370 continue;
4371
4372 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4373
4374 /* inline extents live in the btree, they don't have refs */
4375 if (btrfs_file_extent_type(leaf, fi) ==
4376 BTRFS_FILE_EXTENT_INLINE)
4377 continue;
4378
4379 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4380
4381 /* holes don't have refs */
4382 if (disk_bytenr == 0)
4383 continue;
4384
4385 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4386 ret = btrfs_free_extent(trans, root, disk_bytenr, num_bytes,
4387 leaf->start, 0, key.objectid, 0);
4388 BUG_ON(ret);
4389 }
4390 return 0;
4391}
4392
4393static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
4394 struct btrfs_root *root,
4395 struct btrfs_leaf_ref *ref)
4396{
4397 int i;
4398 int ret;
4399 struct btrfs_extent_info *info;
4400 struct refsort *sorted;
4401
4402 if (ref->nritems == 0)
4403 return 0;
4404
4405 sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
4406 for (i = 0; i < ref->nritems; i++) {
4407 sorted[i].bytenr = ref->extents[i].bytenr;
4408 sorted[i].slot = i;
4409 }
4410 sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
4411
4412 /*
4413 * the items in the ref were sorted when the ref was inserted
4414 * into the ref cache, so this is already in order
4415 */
4416 for (i = 0; i < ref->nritems; i++) {
4417 info = ref->extents + sorted[i].slot;
4418 ret = btrfs_free_extent(trans, root, info->bytenr,
4419 info->num_bytes, ref->bytenr,
4420 ref->owner, ref->generation,
4421 info->objectid, 0);
4422
4423 atomic_inc(&root->fs_info->throttle_gen);
4424 wake_up(&root->fs_info->transaction_throttle);
4425 cond_resched();
4426
4427 BUG_ON(ret);
4428 info++;
4429 }
4430
4431 kfree(sorted);
4432 return 0;
4433}
4434
4435
4436static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
4437 struct btrfs_root *root, u64 start,
4438 u64 len, u32 *refs)
4439{
4440 int ret;
4441
4442 ret = btrfs_lookup_extent_refs(trans, root, start, len, refs);
4443 BUG_ON(ret);
4444
4445#if 0 /* some debugging code in case we see problems here */
4446 /* if the refs count is one, it won't get increased again. But
4447 * if the ref count is > 1, someone may be decreasing it at
4448 * the same time we are.
4449 */
4450 if (*refs != 1) {
4451 struct extent_buffer *eb = NULL;
4452 eb = btrfs_find_create_tree_block(root, start, len);
4453 if (eb)
4454 btrfs_tree_lock(eb);
4455
4456 mutex_lock(&root->fs_info->alloc_mutex);
4457 ret = lookup_extent_ref(NULL, root, start, len, refs);
4458 BUG_ON(ret);
4459 mutex_unlock(&root->fs_info->alloc_mutex);
4460
4461 if (eb) {
4462 btrfs_tree_unlock(eb);
4463 free_extent_buffer(eb);
4464 }
4465 if (*refs == 1) {
4466 printk(KERN_ERR "btrfs block %llu went down to one "
4467 "during drop_snap\n", (unsigned long long)start);
4468 }
4469
4470 }
4471#endif
4472
4473 cond_resched();
4474 return ret;
4475}
4476 4903
4904#define DROP_REFERENCE 1
4905#define UPDATE_BACKREF 2
4477 4906
4478/* 4907static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4479 * this is used while deleting old snapshots, and it drops the refs 4908 struct btrfs_root *root,
4480 * on a whole subtree starting from a level 1 node. 4909 struct walk_control *wc,
4481 * 4910 struct btrfs_path *path)
4482 * The idea is to sort all the leaf pointers, and then drop the
4483 * ref on all the leaves in order. Most of the time the leaves
4484 * will have ref cache entries, so no leaf IOs will be required to
4485 * find the extents they have references on.
4486 *
4487 * For each leaf, any references it has are also dropped in order
4488 *
4489 * This ends up dropping the references in something close to optimal
4490 * order for reading and modifying the extent allocation tree.
4491 */
4492static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
4493 struct btrfs_root *root,
4494 struct btrfs_path *path)
4495{ 4911{
4496 u64 bytenr; 4912 u64 bytenr;
4497 u64 root_owner; 4913 u64 generation;
4498 u64 root_gen; 4914 u64 refs;
4499 struct extent_buffer *eb = path->nodes[1]; 4915 u64 flags;
4500 struct extent_buffer *leaf; 4916 u64 last = 0;
4501 struct btrfs_leaf_ref *ref; 4917 u32 nritems;
4502 struct refsort *sorted = NULL; 4918 u32 blocksize;
4503 int nritems = btrfs_header_nritems(eb); 4919 struct btrfs_key key;
4920 struct extent_buffer *eb;
4504 int ret; 4921 int ret;
4505 int i; 4922 int slot;
4506 int refi = 0; 4923 int nread = 0;
4507 int slot = path->slots[1];
4508 u32 blocksize = btrfs_level_size(root, 0);
4509 u32 refs;
4510
4511 if (nritems == 0)
4512 goto out;
4513
4514 root_owner = btrfs_header_owner(eb);
4515 root_gen = btrfs_header_generation(eb);
4516 sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
4517 4924
4518 /* 4925 if (path->slots[wc->level] < wc->reada_slot) {
4519 * step one, sort all the leaf pointers so we don't scribble 4926 wc->reada_count = wc->reada_count * 2 / 3;
4520 * randomly into the extent allocation tree 4927 wc->reada_count = max(wc->reada_count, 2);
4521 */ 4928 } else {
4522 for (i = slot; i < nritems; i++) { 4929 wc->reada_count = wc->reada_count * 3 / 2;
4523 sorted[refi].bytenr = btrfs_node_blockptr(eb, i); 4930 wc->reada_count = min_t(int, wc->reada_count,
4524 sorted[refi].slot = i; 4931 BTRFS_NODEPTRS_PER_BLOCK(root));
4525 refi++;
4526 } 4932 }
4527 4933
4528 /* 4934 eb = path->nodes[wc->level];
4529 * nritems won't be zero, but if we're picking up drop_snapshot 4935 nritems = btrfs_header_nritems(eb);
4530 * after a crash, slot might be > 0, so double check things 4936 blocksize = btrfs_level_size(root, wc->level - 1);
4531 * just in case.
4532 */
4533 if (refi == 0)
4534 goto out;
4535 4937
4536 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL); 4938 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
4939 if (nread >= wc->reada_count)
4940 break;
4537 4941
4538 /* 4942 cond_resched();
4539 * the first loop frees everything the leaves point to 4943 bytenr = btrfs_node_blockptr(eb, slot);
4540 */ 4944 generation = btrfs_node_ptr_generation(eb, slot);
4541 for (i = 0; i < refi; i++) {
4542 u64 ptr_gen;
4543 4945
4544 bytenr = sorted[i].bytenr; 4946 if (slot == path->slots[wc->level])
4947 goto reada;
4545 4948
4546 /* 4949 if (wc->stage == UPDATE_BACKREF &&
4547 * check the reference count on this leaf. If it is > 1 4950 generation <= root->root_key.offset)
4548 * we just decrement it below and don't update any
4549 * of the refs the leaf points to.
4550 */
4551 ret = drop_snap_lookup_refcount(trans, root, bytenr,
4552 blocksize, &refs);
4553 BUG_ON(ret);
4554 if (refs != 1)
4555 continue; 4951 continue;
4556 4952
4557 ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot); 4953 /* We don't lock the tree block, it's OK to be racy here */
4558 4954 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
4559 /* 4955 &refs, &flags);
4560 * the leaf only had one reference, which means the
4561 * only thing pointing to this leaf is the snapshot
4562 * we're deleting. It isn't possible for the reference
4563 * count to increase again later
4564 *
4565 * The reference cache is checked for the leaf,
4566 * and if found we'll be able to drop any refs held by
4567 * the leaf without needing to read it in.
4568 */
4569 ref = btrfs_lookup_leaf_ref(root, bytenr);
4570 if (ref && ref->generation != ptr_gen) {
4571 btrfs_free_leaf_ref(root, ref);
4572 ref = NULL;
4573 }
4574 if (ref) {
4575 ret = cache_drop_leaf_ref(trans, root, ref);
4576 BUG_ON(ret);
4577 btrfs_remove_leaf_ref(root, ref);
4578 btrfs_free_leaf_ref(root, ref);
4579 } else {
4580 /*
4581 * the leaf wasn't in the reference cache, so
4582 * we have to read it.
4583 */
4584 leaf = read_tree_block(root, bytenr, blocksize,
4585 ptr_gen);
4586 ret = btrfs_drop_leaf_ref(trans, root, leaf);
4587 BUG_ON(ret);
4588 free_extent_buffer(leaf);
4589 }
4590 atomic_inc(&root->fs_info->throttle_gen);
4591 wake_up(&root->fs_info->transaction_throttle);
4592 cond_resched();
4593 }
4594
4595 /*
4596 * run through the loop again to free the refs on the leaves.
4597 * This is faster than doing it in the loop above because
4598 * the leaves are likely to be clustered together. We end up
4599 * working in nice chunks on the extent allocation tree.
4600 */
4601 for (i = 0; i < refi; i++) {
4602 bytenr = sorted[i].bytenr;
4603 ret = btrfs_free_extent(trans, root, bytenr,
4604 blocksize, eb->start,
4605 root_owner, root_gen, 0, 1);
4606 BUG_ON(ret); 4956 BUG_ON(ret);
4957 BUG_ON(refs == 0);
4607 4958
4608 atomic_inc(&root->fs_info->throttle_gen); 4959 if (wc->stage == DROP_REFERENCE) {
4609 wake_up(&root->fs_info->transaction_throttle); 4960 if (refs == 1)
4610 cond_resched(); 4961 goto reada;
4611 }
4612out:
4613 kfree(sorted);
4614
4615 /*
4616 * update the path to show we've processed the entire level 1
4617 * node. This will get saved into the root's drop_snapshot_progress
4618 * field so these drops are not repeated again if this transaction
4619 * commits.
4620 */
4621 path->slots[1] = nritems;
4622 return 0;
4623}
4624
4625/*
4626 * helper function for drop_snapshot, this walks down the tree dropping ref
4627 * counts as it goes.
4628 */
4629static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4630 struct btrfs_root *root,
4631 struct btrfs_path *path, int *level)
4632{
4633 u64 root_owner;
4634 u64 root_gen;
4635 u64 bytenr;
4636 u64 ptr_gen;
4637 struct extent_buffer *next;
4638 struct extent_buffer *cur;
4639 struct extent_buffer *parent;
4640 u32 blocksize;
4641 int ret;
4642 u32 refs;
4643
4644 WARN_ON(*level < 0);
4645 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4646 ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start,
4647 path->nodes[*level]->len, &refs);
4648 BUG_ON(ret);
4649 if (refs > 1)
4650 goto out;
4651
4652 /*
4653 * walk down to the last node level and free all the leaves
4654 */
4655 while (*level >= 0) {
4656 WARN_ON(*level < 0);
4657 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4658 cur = path->nodes[*level];
4659
4660 if (btrfs_header_level(cur) != *level)
4661 WARN_ON(1);
4662
4663 if (path->slots[*level] >=
4664 btrfs_header_nritems(cur))
4665 break;
4666 4962
4667 /* the new code goes down to level 1 and does all the 4963 if (wc->level == 1 &&
4668 * leaves pointed to that node in bulk. So, this check 4964 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4669 * for level 0 will always be false. 4965 continue;
4670 * 4966 if (!wc->update_ref ||
4671 * But, the disk format allows the drop_snapshot_progress 4967 generation <= root->root_key.offset)
4672 * field in the root to leave things in a state where 4968 continue;
4673 * a leaf will need cleaning up here. If someone crashes 4969 btrfs_node_key_to_cpu(eb, &key, slot);
4674 * with the old code and then boots with the new code, 4970 ret = btrfs_comp_cpu_keys(&key,
4675 * we might find a leaf here. 4971 &wc->update_progress);
4676 */ 4972 if (ret < 0)
4677 if (*level == 0) { 4973 continue;
4678 ret = btrfs_drop_leaf_ref(trans, root, cur); 4974 } else {
4679 BUG_ON(ret); 4975 if (wc->level == 1 &&
4680 break; 4976 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4977 continue;
4681 } 4978 }
4682 4979reada:
4683 /* 4980 ret = readahead_tree_block(root, bytenr, blocksize,
4684 * once we get to level one, process the whole node 4981 generation);
4685 * at once, including everything below it. 4982 if (ret)
4686 */
4687 if (*level == 1) {
4688 ret = drop_level_one_refs(trans, root, path);
4689 BUG_ON(ret);
4690 break; 4983 break;
4691 } 4984 last = bytenr + blocksize;
4692 4985 nread++;
4693 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
4694 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
4695 blocksize = btrfs_level_size(root, *level - 1);
4696
4697 ret = drop_snap_lookup_refcount(trans, root, bytenr,
4698 blocksize, &refs);
4699 BUG_ON(ret);
4700
4701 /*
4702 * if there is more than one reference, we don't need
4703 * to read that node to drop any references it has. We
4704 * just drop the ref we hold on that node and move on to the
4705 * next slot in this level.
4706 */
4707 if (refs != 1) {
4708 parent = path->nodes[*level];
4709 root_owner = btrfs_header_owner(parent);
4710 root_gen = btrfs_header_generation(parent);
4711 path->slots[*level]++;
4712
4713 ret = btrfs_free_extent(trans, root, bytenr,
4714 blocksize, parent->start,
4715 root_owner, root_gen,
4716 *level - 1, 1);
4717 BUG_ON(ret);
4718
4719 atomic_inc(&root->fs_info->throttle_gen);
4720 wake_up(&root->fs_info->transaction_throttle);
4721 cond_resched();
4722
4723 continue;
4724 }
4725
4726 /*
4727 * we need to keep freeing things in the next level down.
4728 * read the block and loop around to process it
4729 */
4730 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
4731 WARN_ON(*level <= 0);
4732 if (path->nodes[*level-1])
4733 free_extent_buffer(path->nodes[*level-1]);
4734 path->nodes[*level-1] = next;
4735 *level = btrfs_header_level(next);
4736 path->slots[*level] = 0;
4737 cond_resched();
4738 }
4739out:
4740 WARN_ON(*level < 0);
4741 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4742
4743 if (path->nodes[*level] == root->node) {
4744 parent = path->nodes[*level];
4745 bytenr = path->nodes[*level]->start;
4746 } else {
4747 parent = path->nodes[*level + 1];
4748 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
4749 } 4986 }
4750 4987 wc->reada_slot = slot;
4751 blocksize = btrfs_level_size(root, *level);
4752 root_owner = btrfs_header_owner(parent);
4753 root_gen = btrfs_header_generation(parent);
4754
4755 /*
4756 * cleanup and free the reference on the last node
4757 * we processed
4758 */
4759 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
4760 parent->start, root_owner, root_gen,
4761 *level, 1);
4762 free_extent_buffer(path->nodes[*level]);
4763 path->nodes[*level] = NULL;
4764
4765 *level += 1;
4766 BUG_ON(ret);
4767
4768 cond_resched();
4769 return 0;
4770} 4988}
4771#endif
4772
4773struct walk_control {
4774 u64 refs[BTRFS_MAX_LEVEL];
4775 u64 flags[BTRFS_MAX_LEVEL];
4776 struct btrfs_key update_progress;
4777 int stage;
4778 int level;
4779 int shared_level;
4780 int update_ref;
4781 int keep_locks;
4782};
4783
4784#define DROP_REFERENCE 1
4785#define UPDATE_BACKREF 2
4786 4989
4787/* 4990/*
4788 * hepler to process tree block while walking down the tree. 4991 * hepler to process tree block while walking down the tree.
4789 * 4992 *
4790 * when wc->stage == DROP_REFERENCE, this function checks
4791 * reference count of the block. if the block is shared and
4792 * we need update back refs for the subtree rooted at the
4793 * block, this function changes wc->stage to UPDATE_BACKREF
4794 *
4795 * when wc->stage == UPDATE_BACKREF, this function updates 4993 * when wc->stage == UPDATE_BACKREF, this function updates
4796 * back refs for pointers in the block. 4994 * back refs for pointers in the block.
4797 * 4995 *
@@ -4800,11 +4998,10 @@ struct walk_control {
4800static noinline int walk_down_proc(struct btrfs_trans_handle *trans, 4998static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4801 struct btrfs_root *root, 4999 struct btrfs_root *root,
4802 struct btrfs_path *path, 5000 struct btrfs_path *path,
4803 struct walk_control *wc) 5001 struct walk_control *wc, int lookup_info)
4804{ 5002{
4805 int level = wc->level; 5003 int level = wc->level;
4806 struct extent_buffer *eb = path->nodes[level]; 5004 struct extent_buffer *eb = path->nodes[level];
4807 struct btrfs_key key;
4808 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; 5005 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4809 int ret; 5006 int ret;
4810 5007
@@ -4816,8 +5013,9 @@ static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4816 * when reference count of tree block is 1, it won't increase 5013 * when reference count of tree block is 1, it won't increase
4817 * again. once full backref flag is set, we never clear it. 5014 * again. once full backref flag is set, we never clear it.
4818 */ 5015 */
4819 if ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || 5016 if (lookup_info &&
4820 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag))) { 5017 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5018 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
4821 BUG_ON(!path->locks[level]); 5019 BUG_ON(!path->locks[level]);
4822 ret = btrfs_lookup_extent_info(trans, root, 5020 ret = btrfs_lookup_extent_info(trans, root,
4823 eb->start, eb->len, 5021 eb->start, eb->len,
@@ -4827,21 +5025,6 @@ static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4827 BUG_ON(wc->refs[level] == 0); 5025 BUG_ON(wc->refs[level] == 0);
4828 } 5026 }
4829 5027
4830 if (wc->stage == DROP_REFERENCE &&
4831 wc->update_ref && wc->refs[level] > 1) {
4832 BUG_ON(eb == root->node);
4833 BUG_ON(path->slots[level] > 0);
4834 if (level == 0)
4835 btrfs_item_key_to_cpu(eb, &key, path->slots[level]);
4836 else
4837 btrfs_node_key_to_cpu(eb, &key, path->slots[level]);
4838 if (btrfs_header_owner(eb) == root->root_key.objectid &&
4839 btrfs_comp_cpu_keys(&key, &wc->update_progress) >= 0) {
4840 wc->stage = UPDATE_BACKREF;
4841 wc->shared_level = level;
4842 }
4843 }
4844
4845 if (wc->stage == DROP_REFERENCE) { 5028 if (wc->stage == DROP_REFERENCE) {
4846 if (wc->refs[level] > 1) 5029 if (wc->refs[level] > 1)
4847 return 1; 5030 return 1;
@@ -4878,6 +5061,136 @@ static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4878} 5061}
4879 5062
4880/* 5063/*
5064 * hepler to process tree block pointer.
5065 *
5066 * when wc->stage == DROP_REFERENCE, this function checks
5067 * reference count of the block pointed to. if the block
5068 * is shared and we need update back refs for the subtree
5069 * rooted at the block, this function changes wc->stage to
5070 * UPDATE_BACKREF. if the block is shared and there is no
5071 * need to update back, this function drops the reference
5072 * to the block.
5073 *
5074 * NOTE: return value 1 means we should stop walking down.
5075 */
5076static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5077 struct btrfs_root *root,
5078 struct btrfs_path *path,
5079 struct walk_control *wc, int *lookup_info)
5080{
5081 u64 bytenr;
5082 u64 generation;
5083 u64 parent;
5084 u32 blocksize;
5085 struct btrfs_key key;
5086 struct extent_buffer *next;
5087 int level = wc->level;
5088 int reada = 0;
5089 int ret = 0;
5090
5091 generation = btrfs_node_ptr_generation(path->nodes[level],
5092 path->slots[level]);
5093 /*
5094 * if the lower level block was created before the snapshot
5095 * was created, we know there is no need to update back refs
5096 * for the subtree
5097 */
5098 if (wc->stage == UPDATE_BACKREF &&
5099 generation <= root->root_key.offset) {
5100 *lookup_info = 1;
5101 return 1;
5102 }
5103
5104 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5105 blocksize = btrfs_level_size(root, level - 1);
5106
5107 next = btrfs_find_tree_block(root, bytenr, blocksize);
5108 if (!next) {
5109 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5110 reada = 1;
5111 }
5112 btrfs_tree_lock(next);
5113 btrfs_set_lock_blocking(next);
5114
5115 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5116 &wc->refs[level - 1],
5117 &wc->flags[level - 1]);
5118 BUG_ON(ret);
5119 BUG_ON(wc->refs[level - 1] == 0);
5120 *lookup_info = 0;
5121
5122 if (wc->stage == DROP_REFERENCE) {
5123 if (wc->refs[level - 1] > 1) {
5124 if (level == 1 &&
5125 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5126 goto skip;
5127
5128 if (!wc->update_ref ||
5129 generation <= root->root_key.offset)
5130 goto skip;
5131
5132 btrfs_node_key_to_cpu(path->nodes[level], &key,
5133 path->slots[level]);
5134 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5135 if (ret < 0)
5136 goto skip;
5137
5138 wc->stage = UPDATE_BACKREF;
5139 wc->shared_level = level - 1;
5140 }
5141 } else {
5142 if (level == 1 &&
5143 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5144 goto skip;
5145 }
5146
5147 if (!btrfs_buffer_uptodate(next, generation)) {
5148 btrfs_tree_unlock(next);
5149 free_extent_buffer(next);
5150 next = NULL;
5151 *lookup_info = 1;
5152 }
5153
5154 if (!next) {
5155 if (reada && level == 1)
5156 reada_walk_down(trans, root, wc, path);
5157 next = read_tree_block(root, bytenr, blocksize, generation);
5158 btrfs_tree_lock(next);
5159 btrfs_set_lock_blocking(next);
5160 }
5161
5162 level--;
5163 BUG_ON(level != btrfs_header_level(next));
5164 path->nodes[level] = next;
5165 path->slots[level] = 0;
5166 path->locks[level] = 1;
5167 wc->level = level;
5168 if (wc->level == 1)
5169 wc->reada_slot = 0;
5170 return 0;
5171skip:
5172 wc->refs[level - 1] = 0;
5173 wc->flags[level - 1] = 0;
5174 if (wc->stage == DROP_REFERENCE) {
5175 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5176 parent = path->nodes[level]->start;
5177 } else {
5178 BUG_ON(root->root_key.objectid !=
5179 btrfs_header_owner(path->nodes[level]));
5180 parent = 0;
5181 }
5182
5183 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5184 root->root_key.objectid, level - 1, 0);
5185 BUG_ON(ret);
5186 }
5187 btrfs_tree_unlock(next);
5188 free_extent_buffer(next);
5189 *lookup_info = 1;
5190 return 1;
5191}
5192
5193/*
4881 * hepler to process tree block while walking up the tree. 5194 * hepler to process tree block while walking up the tree.
4882 * 5195 *
4883 * when wc->stage == DROP_REFERENCE, this function drops 5196 * when wc->stage == DROP_REFERENCE, this function drops
@@ -4904,7 +5217,6 @@ static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
4904 if (level < wc->shared_level) 5217 if (level < wc->shared_level)
4905 goto out; 5218 goto out;
4906 5219
4907 BUG_ON(wc->refs[level] <= 1);
4908 ret = find_next_key(path, level + 1, &wc->update_progress); 5220 ret = find_next_key(path, level + 1, &wc->update_progress);
4909 if (ret > 0) 5221 if (ret > 0)
4910 wc->update_ref = 0; 5222 wc->update_ref = 0;
@@ -4935,8 +5247,6 @@ static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
4935 path->locks[level] = 0; 5247 path->locks[level] = 0;
4936 return 1; 5248 return 1;
4937 } 5249 }
4938 } else {
4939 BUG_ON(level != 0);
4940 } 5250 }
4941 } 5251 }
4942 5252
@@ -4989,39 +5299,28 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4989 struct btrfs_path *path, 5299 struct btrfs_path *path,
4990 struct walk_control *wc) 5300 struct walk_control *wc)
4991{ 5301{
4992 struct extent_buffer *next;
4993 struct extent_buffer *cur;
4994 u64 bytenr;
4995 u64 ptr_gen;
4996 u32 blocksize;
4997 int level = wc->level; 5302 int level = wc->level;
5303 int lookup_info = 1;
4998 int ret; 5304 int ret;
4999 5305
5000 while (level >= 0) { 5306 while (level >= 0) {
5001 cur = path->nodes[level]; 5307 if (path->slots[level] >=
5002 BUG_ON(path->slots[level] >= btrfs_header_nritems(cur)); 5308 btrfs_header_nritems(path->nodes[level]))
5309 break;
5003 5310
5004 ret = walk_down_proc(trans, root, path, wc); 5311 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5005 if (ret > 0) 5312 if (ret > 0)
5006 break; 5313 break;
5007 5314
5008 if (level == 0) 5315 if (level == 0)
5009 break; 5316 break;
5010 5317
5011 bytenr = btrfs_node_blockptr(cur, path->slots[level]); 5318 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5012 blocksize = btrfs_level_size(root, level - 1); 5319 if (ret > 0) {
5013 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[level]); 5320 path->slots[level]++;
5014 5321 continue;
5015 next = read_tree_block(root, bytenr, blocksize, ptr_gen); 5322 }
5016 btrfs_tree_lock(next); 5323 level = wc->level;
5017 btrfs_set_lock_blocking(next);
5018
5019 level--;
5020 BUG_ON(level != btrfs_header_level(next));
5021 path->nodes[level] = next;
5022 path->slots[level] = 0;
5023 path->locks[level] = 1;
5024 wc->level = level;
5025 } 5324 }
5026 return 0; 5325 return 0;
5027} 5326}
@@ -5111,9 +5410,7 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5111 err = ret; 5410 err = ret;
5112 goto out; 5411 goto out;
5113 } 5412 }
5114 btrfs_node_key_to_cpu(path->nodes[level], &key, 5413 WARN_ON(ret > 0);
5115 path->slots[level]);
5116 WARN_ON(memcmp(&key, &wc->update_progress, sizeof(key)));
5117 5414
5118 /* 5415 /*
5119 * unlock our path, this is safe because only this 5416 * unlock our path, this is safe because only this
@@ -5148,6 +5445,7 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5148 wc->stage = DROP_REFERENCE; 5445 wc->stage = DROP_REFERENCE;
5149 wc->update_ref = update_ref; 5446 wc->update_ref = update_ref;
5150 wc->keep_locks = 0; 5447 wc->keep_locks = 0;
5448 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5151 5449
5152 while (1) { 5450 while (1) {
5153 ret = walk_down_tree(trans, root, path, wc); 5451 ret = walk_down_tree(trans, root, path, wc);
@@ -5200,9 +5498,24 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5200 ret = btrfs_del_root(trans, tree_root, &root->root_key); 5498 ret = btrfs_del_root(trans, tree_root, &root->root_key);
5201 BUG_ON(ret); 5499 BUG_ON(ret);
5202 5500
5203 free_extent_buffer(root->node); 5501 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5204 free_extent_buffer(root->commit_root); 5502 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
5205 kfree(root); 5503 NULL, NULL);
5504 BUG_ON(ret < 0);
5505 if (ret > 0) {
5506 ret = btrfs_del_orphan_item(trans, tree_root,
5507 root->root_key.objectid);
5508 BUG_ON(ret);
5509 }
5510 }
5511
5512 if (root->in_radix) {
5513 btrfs_free_fs_root(tree_root->fs_info, root);
5514 } else {
5515 free_extent_buffer(root->node);
5516 free_extent_buffer(root->commit_root);
5517 kfree(root);
5518 }
5206out: 5519out:
5207 btrfs_end_transaction(trans, tree_root); 5520 btrfs_end_transaction(trans, tree_root);
5208 kfree(wc); 5521 kfree(wc);
@@ -5254,6 +5567,7 @@ int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5254 wc->stage = DROP_REFERENCE; 5567 wc->stage = DROP_REFERENCE;
5255 wc->update_ref = 0; 5568 wc->update_ref = 0;
5256 wc->keep_locks = 1; 5569 wc->keep_locks = 1;
5570 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5257 5571
5258 while (1) { 5572 while (1) {
5259 wret = walk_down_tree(trans, root, path, wc); 5573 wret = walk_down_tree(trans, root, path, wc);
@@ -5396,9 +5710,9 @@ static noinline int relocate_data_extent(struct inode *reloc_inode,
5396 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); 5710 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5397 while (1) { 5711 while (1) {
5398 int ret; 5712 int ret;
5399 spin_lock(&em_tree->lock); 5713 write_lock(&em_tree->lock);
5400 ret = add_extent_mapping(em_tree, em); 5714 ret = add_extent_mapping(em_tree, em);
5401 spin_unlock(&em_tree->lock); 5715 write_unlock(&em_tree->lock);
5402 if (ret != -EEXIST) { 5716 if (ret != -EEXIST) {
5403 free_extent_map(em); 5717 free_extent_map(em);
5404 break; 5718 break;
@@ -6841,287 +7155,86 @@ int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
6841 return 0; 7155 return 0;
6842} 7156}
6843 7157
6844#if 0 7158/*
6845static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 7159 * checks to see if its even possible to relocate this block group.
6846 struct btrfs_root *root, 7160 *
6847 u64 objectid, u64 size) 7161 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6848{ 7162 * ok to go ahead and try.
6849 struct btrfs_path *path; 7163 */
6850 struct btrfs_inode_item *item; 7164int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
6851 struct extent_buffer *leaf;
6852 int ret;
6853
6854 path = btrfs_alloc_path();
6855 if (!path)
6856 return -ENOMEM;
6857
6858 path->leave_spinning = 1;
6859 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
6860 if (ret)
6861 goto out;
6862
6863 leaf = path->nodes[0];
6864 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
6865 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
6866 btrfs_set_inode_generation(leaf, item, 1);
6867 btrfs_set_inode_size(leaf, item, size);
6868 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
6869 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
6870 btrfs_mark_buffer_dirty(leaf);
6871 btrfs_release_path(root, path);
6872out:
6873 btrfs_free_path(path);
6874 return ret;
6875}
6876
6877static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
6878 struct btrfs_block_group_cache *group)
6879{ 7165{
6880 struct inode *inode = NULL; 7166 struct btrfs_block_group_cache *block_group;
6881 struct btrfs_trans_handle *trans; 7167 struct btrfs_space_info *space_info;
6882 struct btrfs_root *root; 7168 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6883 struct btrfs_key root_key; 7169 struct btrfs_device *device;
6884 u64 objectid = BTRFS_FIRST_FREE_OBJECTID; 7170 int full = 0;
6885 int err = 0; 7171 int ret = 0;
6886 7172
6887 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID; 7173 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
6888 root_key.type = BTRFS_ROOT_ITEM_KEY;
6889 root_key.offset = (u64)-1;
6890 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
6891 if (IS_ERR(root))
6892 return ERR_CAST(root);
6893 7174
6894 trans = btrfs_start_transaction(root, 1); 7175 /* odd, couldn't find the block group, leave it alone */
6895 BUG_ON(!trans); 7176 if (!block_group)
7177 return -1;
6896 7178
6897 err = btrfs_find_free_objectid(trans, root, objectid, &objectid); 7179 /* no bytes used, we're good */
6898 if (err) 7180 if (!btrfs_block_group_used(&block_group->item))
6899 goto out; 7181 goto out;
6900 7182
6901 err = __insert_orphan_inode(trans, root, objectid, group->key.offset); 7183 space_info = block_group->space_info;
6902 BUG_ON(err); 7184 spin_lock(&space_info->lock);
6903
6904 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
6905 group->key.offset, 0, group->key.offset,
6906 0, 0, 0);
6907 BUG_ON(err);
6908
6909 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
6910 if (inode->i_state & I_NEW) {
6911 BTRFS_I(inode)->root = root;
6912 BTRFS_I(inode)->location.objectid = objectid;
6913 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
6914 BTRFS_I(inode)->location.offset = 0;
6915 btrfs_read_locked_inode(inode);
6916 unlock_new_inode(inode);
6917 BUG_ON(is_bad_inode(inode));
6918 } else {
6919 BUG_ON(1);
6920 }
6921 BTRFS_I(inode)->index_cnt = group->key.objectid;
6922
6923 err = btrfs_orphan_add(trans, inode);
6924out:
6925 btrfs_end_transaction(trans, root);
6926 if (err) {
6927 if (inode)
6928 iput(inode);
6929 inode = ERR_PTR(err);
6930 }
6931 return inode;
6932}
6933
6934int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
6935{
6936
6937 struct btrfs_ordered_sum *sums;
6938 struct btrfs_sector_sum *sector_sum;
6939 struct btrfs_ordered_extent *ordered;
6940 struct btrfs_root *root = BTRFS_I(inode)->root;
6941 struct list_head list;
6942 size_t offset;
6943 int ret;
6944 u64 disk_bytenr;
6945
6946 INIT_LIST_HEAD(&list);
6947
6948 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
6949 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
6950
6951 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
6952 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
6953 disk_bytenr + len - 1, &list);
6954
6955 while (!list_empty(&list)) {
6956 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
6957 list_del_init(&sums->list);
6958
6959 sector_sum = sums->sums;
6960 sums->bytenr = ordered->start;
6961 7185
6962 offset = 0; 7186 full = space_info->full;
6963 while (offset < sums->len) {
6964 sector_sum->bytenr += ordered->start - disk_bytenr;
6965 sector_sum++;
6966 offset += root->sectorsize;
6967 }
6968 7187
6969 btrfs_add_ordered_sum(inode, ordered, sums); 7188 /*
7189 * if this is the last block group we have in this space, we can't
7190 * relocate it unless we're able to allocate a new chunk below.
7191 *
7192 * Otherwise, we need to make sure we have room in the space to handle
7193 * all of the extents from this block group. If we can, we're good
7194 */
7195 if ((space_info->total_bytes != block_group->key.offset) &&
7196 (space_info->bytes_used + space_info->bytes_reserved +
7197 space_info->bytes_pinned + space_info->bytes_readonly +
7198 btrfs_block_group_used(&block_group->item) <
7199 space_info->total_bytes)) {
7200 spin_unlock(&space_info->lock);
7201 goto out;
6970 } 7202 }
6971 btrfs_put_ordered_extent(ordered); 7203 spin_unlock(&space_info->lock);
6972 return 0;
6973}
6974
6975int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
6976{
6977 struct btrfs_trans_handle *trans;
6978 struct btrfs_path *path;
6979 struct btrfs_fs_info *info = root->fs_info;
6980 struct extent_buffer *leaf;
6981 struct inode *reloc_inode;
6982 struct btrfs_block_group_cache *block_group;
6983 struct btrfs_key key;
6984 u64 skipped;
6985 u64 cur_byte;
6986 u64 total_found;
6987 u32 nritems;
6988 int ret;
6989 int progress;
6990 int pass = 0;
6991
6992 root = root->fs_info->extent_root;
6993
6994 block_group = btrfs_lookup_block_group(info, group_start);
6995 BUG_ON(!block_group);
6996
6997 printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
6998 (unsigned long long)block_group->key.objectid,
6999 (unsigned long long)block_group->flags);
7000
7001 path = btrfs_alloc_path();
7002 BUG_ON(!path);
7003
7004 reloc_inode = create_reloc_inode(info, block_group);
7005 BUG_ON(IS_ERR(reloc_inode));
7006
7007 __alloc_chunk_for_shrink(root, block_group, 1);
7008 set_block_group_readonly(block_group);
7009
7010 btrfs_start_delalloc_inodes(info->tree_root);
7011 btrfs_wait_ordered_extents(info->tree_root, 0);
7012again:
7013 skipped = 0;
7014 total_found = 0;
7015 progress = 0;
7016 key.objectid = block_group->key.objectid;
7017 key.offset = 0;
7018 key.type = 0;
7019 cur_byte = key.objectid;
7020
7021 trans = btrfs_start_transaction(info->tree_root, 1);
7022 btrfs_commit_transaction(trans, info->tree_root);
7023 7204
7024 mutex_lock(&root->fs_info->cleaner_mutex); 7205 /*
7025 btrfs_clean_old_snapshots(info->tree_root); 7206 * ok we don't have enough space, but maybe we have free space on our
7026 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1); 7207 * devices to allocate new chunks for relocation, so loop through our
7027 mutex_unlock(&root->fs_info->cleaner_mutex); 7208 * alloc devices and guess if we have enough space. However, if we
7209 * were marked as full, then we know there aren't enough chunks, and we
7210 * can just return.
7211 */
7212 ret = -1;
7213 if (full)
7214 goto out;
7028 7215
7029 trans = btrfs_start_transaction(info->tree_root, 1); 7216 mutex_lock(&root->fs_info->chunk_mutex);
7030 btrfs_commit_transaction(trans, info->tree_root); 7217 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7218 u64 min_free = btrfs_block_group_used(&block_group->item);
7219 u64 dev_offset, max_avail;
7031 7220
7032 while (1) { 7221 /*
7033 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 7222 * check to make sure we can actually find a chunk with enough
7034 if (ret < 0) 7223 * space to fit our block group in.
7035 goto out; 7224 */
7036next: 7225 if (device->total_bytes > device->bytes_used + min_free) {
7037 leaf = path->nodes[0]; 7226 ret = find_free_dev_extent(NULL, device, min_free,
7038 nritems = btrfs_header_nritems(leaf); 7227 &dev_offset, &max_avail);
7039 if (path->slots[0] >= nritems) { 7228 if (!ret)
7040 ret = btrfs_next_leaf(root, path);
7041 if (ret < 0)
7042 goto out;
7043 if (ret == 1) {
7044 ret = 0;
7045 break; 7229 break;
7046 } 7230 ret = -1;
7047 leaf = path->nodes[0];
7048 nritems = btrfs_header_nritems(leaf);
7049 }
7050
7051 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7052
7053 if (key.objectid >= block_group->key.objectid +
7054 block_group->key.offset)
7055 break;
7056
7057 if (progress && need_resched()) {
7058 btrfs_release_path(root, path);
7059 cond_resched();
7060 progress = 0;
7061 continue;
7062 }
7063 progress = 1;
7064
7065 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
7066 key.objectid + key.offset <= cur_byte) {
7067 path->slots[0]++;
7068 goto next;
7069 }
7070
7071 total_found++;
7072 cur_byte = key.objectid + key.offset;
7073 btrfs_release_path(root, path);
7074
7075 __alloc_chunk_for_shrink(root, block_group, 0);
7076 ret = relocate_one_extent(root, path, &key, block_group,
7077 reloc_inode, pass);
7078 BUG_ON(ret < 0);
7079 if (ret > 0)
7080 skipped++;
7081
7082 key.objectid = cur_byte;
7083 key.type = 0;
7084 key.offset = 0;
7085 }
7086
7087 btrfs_release_path(root, path);
7088
7089 if (pass == 0) {
7090 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
7091 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
7092 }
7093
7094 if (total_found > 0) {
7095 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
7096 (unsigned long long)total_found, pass);
7097 pass++;
7098 if (total_found == skipped && pass > 2) {
7099 iput(reloc_inode);
7100 reloc_inode = create_reloc_inode(info, block_group);
7101 pass = 0;
7102 } 7231 }
7103 goto again;
7104 } 7232 }
7105 7233 mutex_unlock(&root->fs_info->chunk_mutex);
7106 /* delete reloc_inode */
7107 iput(reloc_inode);
7108
7109 /* unpin extents in this range */
7110 trans = btrfs_start_transaction(info->tree_root, 1);
7111 btrfs_commit_transaction(trans, info->tree_root);
7112
7113 spin_lock(&block_group->lock);
7114 WARN_ON(block_group->pinned > 0);
7115 WARN_ON(block_group->reserved > 0);
7116 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
7117 spin_unlock(&block_group->lock);
7118 btrfs_put_block_group(block_group);
7119 ret = 0;
7120out: 7234out:
7121 btrfs_free_path(path); 7235 btrfs_put_block_group(block_group);
7122 return ret; 7236 return ret;
7123} 7237}
7124#endif
7125 7238
7126static int find_first_block_group(struct btrfs_root *root, 7239static int find_first_block_group(struct btrfs_root *root,
7127 struct btrfs_path *path, struct btrfs_key *key) 7240 struct btrfs_path *path, struct btrfs_key *key)
@@ -7164,8 +7277,18 @@ int btrfs_free_block_groups(struct btrfs_fs_info *info)
7164{ 7277{
7165 struct btrfs_block_group_cache *block_group; 7278 struct btrfs_block_group_cache *block_group;
7166 struct btrfs_space_info *space_info; 7279 struct btrfs_space_info *space_info;
7280 struct btrfs_caching_control *caching_ctl;
7167 struct rb_node *n; 7281 struct rb_node *n;
7168 7282
7283 down_write(&info->extent_commit_sem);
7284 while (!list_empty(&info->caching_block_groups)) {
7285 caching_ctl = list_entry(info->caching_block_groups.next,
7286 struct btrfs_caching_control, list);
7287 list_del(&caching_ctl->list);
7288 put_caching_control(caching_ctl);
7289 }
7290 up_write(&info->extent_commit_sem);
7291
7169 spin_lock(&info->block_group_cache_lock); 7292 spin_lock(&info->block_group_cache_lock);
7170 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { 7293 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7171 block_group = rb_entry(n, struct btrfs_block_group_cache, 7294 block_group = rb_entry(n, struct btrfs_block_group_cache,
@@ -7179,8 +7302,7 @@ int btrfs_free_block_groups(struct btrfs_fs_info *info)
7179 up_write(&block_group->space_info->groups_sem); 7302 up_write(&block_group->space_info->groups_sem);
7180 7303
7181 if (block_group->cached == BTRFS_CACHE_STARTED) 7304 if (block_group->cached == BTRFS_CACHE_STARTED)
7182 wait_event(block_group->caching_q, 7305 wait_block_group_cache_done(block_group);
7183 block_group_cache_done(block_group));
7184 7306
7185 btrfs_remove_free_space_cache(block_group); 7307 btrfs_remove_free_space_cache(block_group);
7186 7308
@@ -7250,7 +7372,6 @@ int btrfs_read_block_groups(struct btrfs_root *root)
7250 spin_lock_init(&cache->lock); 7372 spin_lock_init(&cache->lock);
7251 spin_lock_init(&cache->tree_lock); 7373 spin_lock_init(&cache->tree_lock);
7252 cache->fs_info = info; 7374 cache->fs_info = info;
7253 init_waitqueue_head(&cache->caching_q);
7254 INIT_LIST_HEAD(&cache->list); 7375 INIT_LIST_HEAD(&cache->list);
7255 INIT_LIST_HEAD(&cache->cluster_list); 7376 INIT_LIST_HEAD(&cache->cluster_list);
7256 7377
@@ -7272,8 +7393,6 @@ int btrfs_read_block_groups(struct btrfs_root *root)
7272 cache->flags = btrfs_block_group_flags(&cache->item); 7393 cache->flags = btrfs_block_group_flags(&cache->item);
7273 cache->sectorsize = root->sectorsize; 7394 cache->sectorsize = root->sectorsize;
7274 7395
7275 remove_sb_from_cache(root, cache);
7276
7277 /* 7396 /*
7278 * check for two cases, either we are full, and therefore 7397 * check for two cases, either we are full, and therefore
7279 * don't need to bother with the caching work since we won't 7398 * don't need to bother with the caching work since we won't
@@ -7282,13 +7401,19 @@ int btrfs_read_block_groups(struct btrfs_root *root)
7282 * time, particularly in the full case. 7401 * time, particularly in the full case.
7283 */ 7402 */
7284 if (found_key.offset == btrfs_block_group_used(&cache->item)) { 7403 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7404 exclude_super_stripes(root, cache);
7405 cache->last_byte_to_unpin = (u64)-1;
7285 cache->cached = BTRFS_CACHE_FINISHED; 7406 cache->cached = BTRFS_CACHE_FINISHED;
7407 free_excluded_extents(root, cache);
7286 } else if (btrfs_block_group_used(&cache->item) == 0) { 7408 } else if (btrfs_block_group_used(&cache->item) == 0) {
7409 exclude_super_stripes(root, cache);
7410 cache->last_byte_to_unpin = (u64)-1;
7287 cache->cached = BTRFS_CACHE_FINISHED; 7411 cache->cached = BTRFS_CACHE_FINISHED;
7288 add_new_free_space(cache, root->fs_info, 7412 add_new_free_space(cache, root->fs_info,
7289 found_key.objectid, 7413 found_key.objectid,
7290 found_key.objectid + 7414 found_key.objectid +
7291 found_key.offset); 7415 found_key.offset);
7416 free_excluded_extents(root, cache);
7292 } 7417 }
7293 7418
7294 ret = update_space_info(info, cache->flags, found_key.offset, 7419 ret = update_space_info(info, cache->flags, found_key.offset,
@@ -7296,6 +7421,10 @@ int btrfs_read_block_groups(struct btrfs_root *root)
7296 &space_info); 7421 &space_info);
7297 BUG_ON(ret); 7422 BUG_ON(ret);
7298 cache->space_info = space_info; 7423 cache->space_info = space_info;
7424 spin_lock(&cache->space_info->lock);
7425 cache->space_info->bytes_super += cache->bytes_super;
7426 spin_unlock(&cache->space_info->lock);
7427
7299 down_write(&space_info->groups_sem); 7428 down_write(&space_info->groups_sem);
7300 list_add_tail(&cache->list, &space_info->block_groups); 7429 list_add_tail(&cache->list, &space_info->block_groups);
7301 up_write(&space_info->groups_sem); 7430 up_write(&space_info->groups_sem);
@@ -7345,7 +7474,6 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7345 atomic_set(&cache->count, 1); 7474 atomic_set(&cache->count, 1);
7346 spin_lock_init(&cache->lock); 7475 spin_lock_init(&cache->lock);
7347 spin_lock_init(&cache->tree_lock); 7476 spin_lock_init(&cache->tree_lock);
7348 init_waitqueue_head(&cache->caching_q);
7349 INIT_LIST_HEAD(&cache->list); 7477 INIT_LIST_HEAD(&cache->list);
7350 INIT_LIST_HEAD(&cache->cluster_list); 7478 INIT_LIST_HEAD(&cache->cluster_list);
7351 7479
@@ -7354,15 +7482,23 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7354 cache->flags = type; 7482 cache->flags = type;
7355 btrfs_set_block_group_flags(&cache->item, type); 7483 btrfs_set_block_group_flags(&cache->item, type);
7356 7484
7485 cache->last_byte_to_unpin = (u64)-1;
7357 cache->cached = BTRFS_CACHE_FINISHED; 7486 cache->cached = BTRFS_CACHE_FINISHED;
7358 remove_sb_from_cache(root, cache); 7487 exclude_super_stripes(root, cache);
7359 7488
7360 add_new_free_space(cache, root->fs_info, chunk_offset, 7489 add_new_free_space(cache, root->fs_info, chunk_offset,
7361 chunk_offset + size); 7490 chunk_offset + size);
7362 7491
7492 free_excluded_extents(root, cache);
7493
7363 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, 7494 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7364 &cache->space_info); 7495 &cache->space_info);
7365 BUG_ON(ret); 7496 BUG_ON(ret);
7497
7498 spin_lock(&cache->space_info->lock);
7499 cache->space_info->bytes_super += cache->bytes_super;
7500 spin_unlock(&cache->space_info->lock);
7501
7366 down_write(&cache->space_info->groups_sem); 7502 down_write(&cache->space_info->groups_sem);
7367 list_add_tail(&cache->list, &cache->space_info->block_groups); 7503 list_add_tail(&cache->list, &cache->space_info->block_groups);
7368 up_write(&cache->space_info->groups_sem); 7504 up_write(&cache->space_info->groups_sem);
@@ -7428,8 +7564,7 @@ int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7428 up_write(&block_group->space_info->groups_sem); 7564 up_write(&block_group->space_info->groups_sem);
7429 7565
7430 if (block_group->cached == BTRFS_CACHE_STARTED) 7566 if (block_group->cached == BTRFS_CACHE_STARTED)
7431 wait_event(block_group->caching_q, 7567 wait_block_group_cache_done(block_group);
7432 block_group_cache_done(block_group));
7433 7568
7434 btrfs_remove_free_space_cache(block_group); 7569 btrfs_remove_free_space_cache(block_group);
7435 7570
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 68260180f587..96577e8bf9fd 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -280,6 +280,14 @@ static struct extent_buffer *buffer_search(struct extent_io_tree *tree,
280 return NULL; 280 return NULL;
281} 281}
282 282
283static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
284 struct extent_state *other)
285{
286 if (tree->ops && tree->ops->merge_extent_hook)
287 tree->ops->merge_extent_hook(tree->mapping->host, new,
288 other);
289}
290
283/* 291/*
284 * utility function to look for merge candidates inside a given range. 292 * utility function to look for merge candidates inside a given range.
285 * Any extents with matching state are merged together into a single 293 * Any extents with matching state are merged together into a single
@@ -303,6 +311,7 @@ static int merge_state(struct extent_io_tree *tree,
303 other = rb_entry(other_node, struct extent_state, rb_node); 311 other = rb_entry(other_node, struct extent_state, rb_node);
304 if (other->end == state->start - 1 && 312 if (other->end == state->start - 1 &&
305 other->state == state->state) { 313 other->state == state->state) {
314 merge_cb(tree, state, other);
306 state->start = other->start; 315 state->start = other->start;
307 other->tree = NULL; 316 other->tree = NULL;
308 rb_erase(&other->rb_node, &tree->state); 317 rb_erase(&other->rb_node, &tree->state);
@@ -314,33 +323,37 @@ static int merge_state(struct extent_io_tree *tree,
314 other = rb_entry(other_node, struct extent_state, rb_node); 323 other = rb_entry(other_node, struct extent_state, rb_node);
315 if (other->start == state->end + 1 && 324 if (other->start == state->end + 1 &&
316 other->state == state->state) { 325 other->state == state->state) {
326 merge_cb(tree, state, other);
317 other->start = state->start; 327 other->start = state->start;
318 state->tree = NULL; 328 state->tree = NULL;
319 rb_erase(&state->rb_node, &tree->state); 329 rb_erase(&state->rb_node, &tree->state);
320 free_extent_state(state); 330 free_extent_state(state);
331 state = NULL;
321 } 332 }
322 } 333 }
334
323 return 0; 335 return 0;
324} 336}
325 337
326static void set_state_cb(struct extent_io_tree *tree, 338static int set_state_cb(struct extent_io_tree *tree,
327 struct extent_state *state, 339 struct extent_state *state,
328 unsigned long bits) 340 unsigned long bits)
329{ 341{
330 if (tree->ops && tree->ops->set_bit_hook) { 342 if (tree->ops && tree->ops->set_bit_hook) {
331 tree->ops->set_bit_hook(tree->mapping->host, state->start, 343 return tree->ops->set_bit_hook(tree->mapping->host,
332 state->end, state->state, bits); 344 state->start, state->end,
345 state->state, bits);
333 } 346 }
347
348 return 0;
334} 349}
335 350
336static void clear_state_cb(struct extent_io_tree *tree, 351static void clear_state_cb(struct extent_io_tree *tree,
337 struct extent_state *state, 352 struct extent_state *state,
338 unsigned long bits) 353 unsigned long bits)
339{ 354{
340 if (tree->ops && tree->ops->clear_bit_hook) { 355 if (tree->ops && tree->ops->clear_bit_hook)
341 tree->ops->clear_bit_hook(tree->mapping->host, state->start, 356 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
342 state->end, state->state, bits);
343 }
344} 357}
345 358
346/* 359/*
@@ -358,6 +371,7 @@ static int insert_state(struct extent_io_tree *tree,
358 int bits) 371 int bits)
359{ 372{
360 struct rb_node *node; 373 struct rb_node *node;
374 int ret;
361 375
362 if (end < start) { 376 if (end < start) {
363 printk(KERN_ERR "btrfs end < start %llu %llu\n", 377 printk(KERN_ERR "btrfs end < start %llu %llu\n",
@@ -365,12 +379,15 @@ static int insert_state(struct extent_io_tree *tree,
365 (unsigned long long)start); 379 (unsigned long long)start);
366 WARN_ON(1); 380 WARN_ON(1);
367 } 381 }
382 state->start = start;
383 state->end = end;
384 ret = set_state_cb(tree, state, bits);
385 if (ret)
386 return ret;
387
368 if (bits & EXTENT_DIRTY) 388 if (bits & EXTENT_DIRTY)
369 tree->dirty_bytes += end - start + 1; 389 tree->dirty_bytes += end - start + 1;
370 set_state_cb(tree, state, bits);
371 state->state |= bits; 390 state->state |= bits;
372 state->start = start;
373 state->end = end;
374 node = tree_insert(&tree->state, end, &state->rb_node); 391 node = tree_insert(&tree->state, end, &state->rb_node);
375 if (node) { 392 if (node) {
376 struct extent_state *found; 393 struct extent_state *found;
@@ -387,6 +404,15 @@ static int insert_state(struct extent_io_tree *tree,
387 return 0; 404 return 0;
388} 405}
389 406
407static int split_cb(struct extent_io_tree *tree, struct extent_state *orig,
408 u64 split)
409{
410 if (tree->ops && tree->ops->split_extent_hook)
411 return tree->ops->split_extent_hook(tree->mapping->host,
412 orig, split);
413 return 0;
414}
415
390/* 416/*
391 * split a given extent state struct in two, inserting the preallocated 417 * split a given extent state struct in two, inserting the preallocated
392 * struct 'prealloc' as the newly created second half. 'split' indicates an 418 * struct 'prealloc' as the newly created second half. 'split' indicates an
@@ -405,6 +431,9 @@ static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
405 struct extent_state *prealloc, u64 split) 431 struct extent_state *prealloc, u64 split)
406{ 432{
407 struct rb_node *node; 433 struct rb_node *node;
434
435 split_cb(tree, orig, split);
436
408 prealloc->start = orig->start; 437 prealloc->start = orig->start;
409 prealloc->end = split - 1; 438 prealloc->end = split - 1;
410 prealloc->state = orig->state; 439 prealloc->state = orig->state;
@@ -431,7 +460,8 @@ static int clear_state_bit(struct extent_io_tree *tree,
431 struct extent_state *state, int bits, int wake, 460 struct extent_state *state, int bits, int wake,
432 int delete) 461 int delete)
433{ 462{
434 int ret = state->state & bits; 463 int bits_to_clear = bits & ~EXTENT_DO_ACCOUNTING;
464 int ret = state->state & bits_to_clear;
435 465
436 if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { 466 if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
437 u64 range = state->end - state->start + 1; 467 u64 range = state->end - state->start + 1;
@@ -439,7 +469,7 @@ static int clear_state_bit(struct extent_io_tree *tree,
439 tree->dirty_bytes -= range; 469 tree->dirty_bytes -= range;
440 } 470 }
441 clear_state_cb(tree, state, bits); 471 clear_state_cb(tree, state, bits);
442 state->state &= ~bits; 472 state->state &= ~bits_to_clear;
443 if (wake) 473 if (wake)
444 wake_up(&state->wq); 474 wake_up(&state->wq);
445 if (delete || state->state == 0) { 475 if (delete || state->state == 0) {
@@ -471,10 +501,14 @@ static int clear_state_bit(struct extent_io_tree *tree,
471 * bits were already set, or zero if none of the bits were already set. 501 * bits were already set, or zero if none of the bits were already set.
472 */ 502 */
473int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, 503int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
474 int bits, int wake, int delete, gfp_t mask) 504 int bits, int wake, int delete,
505 struct extent_state **cached_state,
506 gfp_t mask)
475{ 507{
476 struct extent_state *state; 508 struct extent_state *state;
509 struct extent_state *cached;
477 struct extent_state *prealloc = NULL; 510 struct extent_state *prealloc = NULL;
511 struct rb_node *next_node;
478 struct rb_node *node; 512 struct rb_node *node;
479 u64 last_end; 513 u64 last_end;
480 int err; 514 int err;
@@ -488,6 +522,17 @@ again:
488 } 522 }
489 523
490 spin_lock(&tree->lock); 524 spin_lock(&tree->lock);
525 if (cached_state) {
526 cached = *cached_state;
527 *cached_state = NULL;
528 cached_state = NULL;
529 if (cached && cached->tree && cached->start == start) {
530 atomic_dec(&cached->refs);
531 state = cached;
532 goto hit_next;
533 }
534 free_extent_state(cached);
535 }
491 /* 536 /*
492 * this search will find the extents that end after 537 * this search will find the extents that end after
493 * our range starts 538 * our range starts
@@ -496,6 +541,7 @@ again:
496 if (!node) 541 if (!node)
497 goto out; 542 goto out;
498 state = rb_entry(node, struct extent_state, rb_node); 543 state = rb_entry(node, struct extent_state, rb_node);
544hit_next:
499 if (state->start > end) 545 if (state->start > end)
500 goto out; 546 goto out;
501 WARN_ON(state->end < start); 547 WARN_ON(state->end < start);
@@ -526,13 +572,11 @@ again:
526 if (err) 572 if (err)
527 goto out; 573 goto out;
528 if (state->end <= end) { 574 if (state->end <= end) {
529 set |= clear_state_bit(tree, state, bits, 575 set |= clear_state_bit(tree, state, bits, wake,
530 wake, delete); 576 delete);
531 if (last_end == (u64)-1) 577 if (last_end == (u64)-1)
532 goto out; 578 goto out;
533 start = last_end + 1; 579 start = last_end + 1;
534 } else {
535 start = state->start;
536 } 580 }
537 goto search_again; 581 goto search_again;
538 } 582 }
@@ -547,19 +591,30 @@ again:
547 prealloc = alloc_extent_state(GFP_ATOMIC); 591 prealloc = alloc_extent_state(GFP_ATOMIC);
548 err = split_state(tree, state, prealloc, end + 1); 592 err = split_state(tree, state, prealloc, end + 1);
549 BUG_ON(err == -EEXIST); 593 BUG_ON(err == -EEXIST);
550
551 if (wake) 594 if (wake)
552 wake_up(&state->wq); 595 wake_up(&state->wq);
553 set |= clear_state_bit(tree, prealloc, bits, 596
554 wake, delete); 597 set |= clear_state_bit(tree, prealloc, bits, wake, delete);
598
555 prealloc = NULL; 599 prealloc = NULL;
556 goto out; 600 goto out;
557 } 601 }
558 602
603 if (state->end < end && prealloc && !need_resched())
604 next_node = rb_next(&state->rb_node);
605 else
606 next_node = NULL;
607
559 set |= clear_state_bit(tree, state, bits, wake, delete); 608 set |= clear_state_bit(tree, state, bits, wake, delete);
560 if (last_end == (u64)-1) 609 if (last_end == (u64)-1)
561 goto out; 610 goto out;
562 start = last_end + 1; 611 start = last_end + 1;
612 if (start <= end && next_node) {
613 state = rb_entry(next_node, struct extent_state,
614 rb_node);
615 if (state->start == start)
616 goto hit_next;
617 }
563 goto search_again; 618 goto search_again;
564 619
565out: 620out:
@@ -641,40 +696,59 @@ out:
641 return 0; 696 return 0;
642} 697}
643 698
644static void set_state_bits(struct extent_io_tree *tree, 699static int set_state_bits(struct extent_io_tree *tree,
645 struct extent_state *state, 700 struct extent_state *state,
646 int bits) 701 int bits)
647{ 702{
703 int ret;
704
705 ret = set_state_cb(tree, state, bits);
706 if (ret)
707 return ret;
708
648 if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { 709 if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
649 u64 range = state->end - state->start + 1; 710 u64 range = state->end - state->start + 1;
650 tree->dirty_bytes += range; 711 tree->dirty_bytes += range;
651 } 712 }
652 set_state_cb(tree, state, bits);
653 state->state |= bits; 713 state->state |= bits;
714
715 return 0;
716}
717
718static void cache_state(struct extent_state *state,
719 struct extent_state **cached_ptr)
720{
721 if (cached_ptr && !(*cached_ptr)) {
722 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
723 *cached_ptr = state;
724 atomic_inc(&state->refs);
725 }
726 }
654} 727}
655 728
656/* 729/*
657 * set some bits on a range in the tree. This may require allocations 730 * set some bits on a range in the tree. This may require allocations or
658 * or sleeping, so the gfp mask is used to indicate what is allowed. 731 * sleeping, so the gfp mask is used to indicate what is allowed.
659 * 732 *
660 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the 733 * If any of the exclusive bits are set, this will fail with -EEXIST if some
661 * range already has the desired bits set. The start of the existing 734 * part of the range already has the desired bits set. The start of the
662 * range is returned in failed_start in this case. 735 * existing range is returned in failed_start in this case.
663 * 736 *
664 * [start, end] is inclusive 737 * [start, end] is inclusive This takes the tree lock.
665 * This takes the tree lock.
666 */ 738 */
739
667static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, 740static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
668 int bits, int exclusive, u64 *failed_start, 741 int bits, int exclusive_bits, u64 *failed_start,
742 struct extent_state **cached_state,
669 gfp_t mask) 743 gfp_t mask)
670{ 744{
671 struct extent_state *state; 745 struct extent_state *state;
672 struct extent_state *prealloc = NULL; 746 struct extent_state *prealloc = NULL;
673 struct rb_node *node; 747 struct rb_node *node;
674 int err = 0; 748 int err = 0;
675 int set;
676 u64 last_start; 749 u64 last_start;
677 u64 last_end; 750 u64 last_end;
751
678again: 752again:
679 if (!prealloc && (mask & __GFP_WAIT)) { 753 if (!prealloc && (mask & __GFP_WAIT)) {
680 prealloc = alloc_extent_state(mask); 754 prealloc = alloc_extent_state(mask);
@@ -683,6 +757,13 @@ again:
683 } 757 }
684 758
685 spin_lock(&tree->lock); 759 spin_lock(&tree->lock);
760 if (cached_state && *cached_state) {
761 state = *cached_state;
762 if (state->start == start && state->tree) {
763 node = &state->rb_node;
764 goto hit_next;
765 }
766 }
686 /* 767 /*
687 * this search will find all the extents that end after 768 * this search will find all the extents that end after
688 * our range starts. 769 * our range starts.
@@ -694,8 +775,8 @@ again:
694 BUG_ON(err == -EEXIST); 775 BUG_ON(err == -EEXIST);
695 goto out; 776 goto out;
696 } 777 }
697
698 state = rb_entry(node, struct extent_state, rb_node); 778 state = rb_entry(node, struct extent_state, rb_node);
779hit_next:
699 last_start = state->start; 780 last_start = state->start;
700 last_end = state->end; 781 last_end = state->end;
701 782
@@ -706,17 +787,32 @@ again:
706 * Just lock what we found and keep going 787 * Just lock what we found and keep going
707 */ 788 */
708 if (state->start == start && state->end <= end) { 789 if (state->start == start && state->end <= end) {
709 set = state->state & bits; 790 struct rb_node *next_node;
710 if (set && exclusive) { 791 if (state->state & exclusive_bits) {
711 *failed_start = state->start; 792 *failed_start = state->start;
712 err = -EEXIST; 793 err = -EEXIST;
713 goto out; 794 goto out;
714 } 795 }
715 set_state_bits(tree, state, bits); 796
797 err = set_state_bits(tree, state, bits);
798 if (err)
799 goto out;
800
801 cache_state(state, cached_state);
716 merge_state(tree, state); 802 merge_state(tree, state);
717 if (last_end == (u64)-1) 803 if (last_end == (u64)-1)
718 goto out; 804 goto out;
805
719 start = last_end + 1; 806 start = last_end + 1;
807 if (start < end && prealloc && !need_resched()) {
808 next_node = rb_next(node);
809 if (next_node) {
810 state = rb_entry(next_node, struct extent_state,
811 rb_node);
812 if (state->start == start)
813 goto hit_next;
814 }
815 }
720 goto search_again; 816 goto search_again;
721 } 817 }
722 818
@@ -737,8 +833,7 @@ again:
737 * desired bit on it. 833 * desired bit on it.
738 */ 834 */
739 if (state->start < start) { 835 if (state->start < start) {
740 set = state->state & bits; 836 if (state->state & exclusive_bits) {
741 if (exclusive && set) {
742 *failed_start = start; 837 *failed_start = start;
743 err = -EEXIST; 838 err = -EEXIST;
744 goto out; 839 goto out;
@@ -749,13 +844,14 @@ again:
749 if (err) 844 if (err)
750 goto out; 845 goto out;
751 if (state->end <= end) { 846 if (state->end <= end) {
752 set_state_bits(tree, state, bits); 847 err = set_state_bits(tree, state, bits);
848 if (err)
849 goto out;
850 cache_state(state, cached_state);
753 merge_state(tree, state); 851 merge_state(tree, state);
754 if (last_end == (u64)-1) 852 if (last_end == (u64)-1)
755 goto out; 853 goto out;
756 start = last_end + 1; 854 start = last_end + 1;
757 } else {
758 start = state->start;
759 } 855 }
760 goto search_again; 856 goto search_again;
761 } 857 }
@@ -774,10 +870,13 @@ again:
774 this_end = last_start - 1; 870 this_end = last_start - 1;
775 err = insert_state(tree, prealloc, start, this_end, 871 err = insert_state(tree, prealloc, start, this_end,
776 bits); 872 bits);
777 prealloc = NULL;
778 BUG_ON(err == -EEXIST); 873 BUG_ON(err == -EEXIST);
779 if (err) 874 if (err) {
875 prealloc = NULL;
780 goto out; 876 goto out;
877 }
878 cache_state(prealloc, cached_state);
879 prealloc = NULL;
781 start = this_end + 1; 880 start = this_end + 1;
782 goto search_again; 881 goto search_again;
783 } 882 }
@@ -788,8 +887,7 @@ again:
788 * on the first half 887 * on the first half
789 */ 888 */
790 if (state->start <= end && state->end > end) { 889 if (state->start <= end && state->end > end) {
791 set = state->state & bits; 890 if (state->state & exclusive_bits) {
792 if (exclusive && set) {
793 *failed_start = start; 891 *failed_start = start;
794 err = -EEXIST; 892 err = -EEXIST;
795 goto out; 893 goto out;
@@ -797,7 +895,12 @@ again:
797 err = split_state(tree, state, prealloc, end + 1); 895 err = split_state(tree, state, prealloc, end + 1);
798 BUG_ON(err == -EEXIST); 896 BUG_ON(err == -EEXIST);
799 897
800 set_state_bits(tree, prealloc, bits); 898 err = set_state_bits(tree, prealloc, bits);
899 if (err) {
900 prealloc = NULL;
901 goto out;
902 }
903 cache_state(prealloc, cached_state);
801 merge_state(tree, prealloc); 904 merge_state(tree, prealloc);
802 prealloc = NULL; 905 prealloc = NULL;
803 goto out; 906 goto out;
@@ -826,86 +929,65 @@ int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
826 gfp_t mask) 929 gfp_t mask)
827{ 930{
828 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL, 931 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
829 mask); 932 NULL, mask);
830}
831
832int set_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
833 gfp_t mask)
834{
835 return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, mask);
836} 933}
837 934
838int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, 935int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
839 int bits, gfp_t mask) 936 int bits, gfp_t mask)
840{ 937{
841 return set_extent_bit(tree, start, end, bits, 0, NULL, 938 return set_extent_bit(tree, start, end, bits, 0, NULL,
842 mask); 939 NULL, mask);
843} 940}
844 941
845int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, 942int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
846 int bits, gfp_t mask) 943 int bits, gfp_t mask)
847{ 944{
848 return clear_extent_bit(tree, start, end, bits, 0, 0, mask); 945 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
849} 946}
850 947
851int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, 948int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
852 gfp_t mask) 949 gfp_t mask)
853{ 950{
854 return set_extent_bit(tree, start, end, 951 return set_extent_bit(tree, start, end,
855 EXTENT_DELALLOC | EXTENT_DIRTY, 952 EXTENT_DELALLOC | EXTENT_DIRTY | EXTENT_UPTODATE,
856 0, NULL, mask); 953 0, NULL, NULL, mask);
857} 954}
858 955
859int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, 956int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
860 gfp_t mask) 957 gfp_t mask)
861{ 958{
862 return clear_extent_bit(tree, start, end, 959 return clear_extent_bit(tree, start, end,
863 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask); 960 EXTENT_DIRTY | EXTENT_DELALLOC |
864} 961 EXTENT_DO_ACCOUNTING, 0, 0,
865 962 NULL, mask);
866int clear_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
867 gfp_t mask)
868{
869 return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0, mask);
870} 963}
871 964
872int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, 965int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
873 gfp_t mask) 966 gfp_t mask)
874{ 967{
875 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL, 968 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
876 mask); 969 NULL, mask);
877} 970}
878 971
879static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end, 972static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
880 gfp_t mask) 973 gfp_t mask)
881{ 974{
882 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask); 975 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0,
976 NULL, mask);
883} 977}
884 978
885int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, 979int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
886 gfp_t mask) 980 gfp_t mask)
887{ 981{
888 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL, 982 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
889 mask); 983 NULL, mask);
890} 984}
891 985
892static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, 986static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
893 u64 end, gfp_t mask) 987 u64 end, gfp_t mask)
894{ 988{
895 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask); 989 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
896} 990 NULL, mask);
897
898static int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
899 gfp_t mask)
900{
901 return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
902 0, NULL, mask);
903}
904
905static int clear_extent_writeback(struct extent_io_tree *tree, u64 start,
906 u64 end, gfp_t mask)
907{
908 return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
909} 991}
910 992
911int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end) 993int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
@@ -917,13 +999,15 @@ int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
917 * either insert or lock state struct between start and end use mask to tell 999 * either insert or lock state struct between start and end use mask to tell
918 * us if waiting is desired. 1000 * us if waiting is desired.
919 */ 1001 */
920int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) 1002int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1003 int bits, struct extent_state **cached_state, gfp_t mask)
921{ 1004{
922 int err; 1005 int err;
923 u64 failed_start; 1006 u64 failed_start;
924 while (1) { 1007 while (1) {
925 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 1008 err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
926 &failed_start, mask); 1009 EXTENT_LOCKED, &failed_start,
1010 cached_state, mask);
927 if (err == -EEXIST && (mask & __GFP_WAIT)) { 1011 if (err == -EEXIST && (mask & __GFP_WAIT)) {
928 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); 1012 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
929 start = failed_start; 1013 start = failed_start;
@@ -935,27 +1019,40 @@ int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
935 return err; 1019 return err;
936} 1020}
937 1021
1022int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
1023{
1024 return lock_extent_bits(tree, start, end, 0, NULL, mask);
1025}
1026
938int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end, 1027int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
939 gfp_t mask) 1028 gfp_t mask)
940{ 1029{
941 int err; 1030 int err;
942 u64 failed_start; 1031 u64 failed_start;
943 1032
944 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 1033 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
945 &failed_start, mask); 1034 &failed_start, NULL, mask);
946 if (err == -EEXIST) { 1035 if (err == -EEXIST) {
947 if (failed_start > start) 1036 if (failed_start > start)
948 clear_extent_bit(tree, start, failed_start - 1, 1037 clear_extent_bit(tree, start, failed_start - 1,
949 EXTENT_LOCKED, 1, 0, mask); 1038 EXTENT_LOCKED, 1, 0, NULL, mask);
950 return 0; 1039 return 0;
951 } 1040 }
952 return 1; 1041 return 1;
953} 1042}
954 1043
1044int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1045 struct extent_state **cached, gfp_t mask)
1046{
1047 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1048 mask);
1049}
1050
955int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, 1051int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
956 gfp_t mask) 1052 gfp_t mask)
957{ 1053{
958 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask); 1054 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1055 mask);
959} 1056}
960 1057
961/* 1058/*
@@ -974,7 +1071,6 @@ int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
974 page_cache_release(page); 1071 page_cache_release(page);
975 index++; 1072 index++;
976 } 1073 }
977 set_extent_dirty(tree, start, end, GFP_NOFS);
978 return 0; 1074 return 0;
979} 1075}
980 1076
@@ -994,7 +1090,6 @@ static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
994 page_cache_release(page); 1090 page_cache_release(page);
995 index++; 1091 index++;
996 } 1092 }
997 set_extent_writeback(tree, start, end, GFP_NOFS);
998 return 0; 1093 return 0;
999} 1094}
1000 1095
@@ -1232,6 +1327,7 @@ static noinline u64 find_lock_delalloc_range(struct inode *inode,
1232 u64 delalloc_start; 1327 u64 delalloc_start;
1233 u64 delalloc_end; 1328 u64 delalloc_end;
1234 u64 found; 1329 u64 found;
1330 struct extent_state *cached_state = NULL;
1235 int ret; 1331 int ret;
1236 int loops = 0; 1332 int loops = 0;
1237 1333
@@ -1269,6 +1365,7 @@ again:
1269 /* some of the pages are gone, lets avoid looping by 1365 /* some of the pages are gone, lets avoid looping by
1270 * shortening the size of the delalloc range we're searching 1366 * shortening the size of the delalloc range we're searching
1271 */ 1367 */
1368 free_extent_state(cached_state);
1272 if (!loops) { 1369 if (!loops) {
1273 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1); 1370 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1274 max_bytes = PAGE_CACHE_SIZE - offset; 1371 max_bytes = PAGE_CACHE_SIZE - offset;
@@ -1282,18 +1379,21 @@ again:
1282 BUG_ON(ret); 1379 BUG_ON(ret);
1283 1380
1284 /* step three, lock the state bits for the whole range */ 1381 /* step three, lock the state bits for the whole range */
1285 lock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS); 1382 lock_extent_bits(tree, delalloc_start, delalloc_end,
1383 0, &cached_state, GFP_NOFS);
1286 1384
1287 /* then test to make sure it is all still delalloc */ 1385 /* then test to make sure it is all still delalloc */
1288 ret = test_range_bit(tree, delalloc_start, delalloc_end, 1386 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1289 EXTENT_DELALLOC, 1); 1387 EXTENT_DELALLOC, 1, cached_state);
1290 if (!ret) { 1388 if (!ret) {
1291 unlock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS); 1389 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1390 &cached_state, GFP_NOFS);
1292 __unlock_for_delalloc(inode, locked_page, 1391 __unlock_for_delalloc(inode, locked_page,
1293 delalloc_start, delalloc_end); 1392 delalloc_start, delalloc_end);
1294 cond_resched(); 1393 cond_resched();
1295 goto again; 1394 goto again;
1296 } 1395 }
1396 free_extent_state(cached_state);
1297 *start = delalloc_start; 1397 *start = delalloc_start;
1298 *end = delalloc_end; 1398 *end = delalloc_end;
1299out_failed: 1399out_failed:
@@ -1303,11 +1403,7 @@ out_failed:
1303int extent_clear_unlock_delalloc(struct inode *inode, 1403int extent_clear_unlock_delalloc(struct inode *inode,
1304 struct extent_io_tree *tree, 1404 struct extent_io_tree *tree,
1305 u64 start, u64 end, struct page *locked_page, 1405 u64 start, u64 end, struct page *locked_page,
1306 int unlock_pages, 1406 unsigned long op)
1307 int clear_unlock,
1308 int clear_delalloc, int clear_dirty,
1309 int set_writeback,
1310 int end_writeback)
1311{ 1407{
1312 int ret; 1408 int ret;
1313 struct page *pages[16]; 1409 struct page *pages[16];
@@ -1317,16 +1413,21 @@ int extent_clear_unlock_delalloc(struct inode *inode,
1317 int i; 1413 int i;
1318 int clear_bits = 0; 1414 int clear_bits = 0;
1319 1415
1320 if (clear_unlock) 1416 if (op & EXTENT_CLEAR_UNLOCK)
1321 clear_bits |= EXTENT_LOCKED; 1417 clear_bits |= EXTENT_LOCKED;
1322 if (clear_dirty) 1418 if (op & EXTENT_CLEAR_DIRTY)
1323 clear_bits |= EXTENT_DIRTY; 1419 clear_bits |= EXTENT_DIRTY;
1324 1420
1325 if (clear_delalloc) 1421 if (op & EXTENT_CLEAR_DELALLOC)
1326 clear_bits |= EXTENT_DELALLOC; 1422 clear_bits |= EXTENT_DELALLOC;
1327 1423
1328 clear_extent_bit(tree, start, end, clear_bits, 1, 0, GFP_NOFS); 1424 if (op & EXTENT_CLEAR_ACCOUNTING)
1329 if (!(unlock_pages || clear_dirty || set_writeback || end_writeback)) 1425 clear_bits |= EXTENT_DO_ACCOUNTING;
1426
1427 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1428 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1429 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1430 EXTENT_SET_PRIVATE2)))
1330 return 0; 1431 return 0;
1331 1432
1332 while (nr_pages > 0) { 1433 while (nr_pages > 0) {
@@ -1334,17 +1435,21 @@ int extent_clear_unlock_delalloc(struct inode *inode,
1334 min_t(unsigned long, 1435 min_t(unsigned long,
1335 nr_pages, ARRAY_SIZE(pages)), pages); 1436 nr_pages, ARRAY_SIZE(pages)), pages);
1336 for (i = 0; i < ret; i++) { 1437 for (i = 0; i < ret; i++) {
1438
1439 if (op & EXTENT_SET_PRIVATE2)
1440 SetPagePrivate2(pages[i]);
1441
1337 if (pages[i] == locked_page) { 1442 if (pages[i] == locked_page) {
1338 page_cache_release(pages[i]); 1443 page_cache_release(pages[i]);
1339 continue; 1444 continue;
1340 } 1445 }
1341 if (clear_dirty) 1446 if (op & EXTENT_CLEAR_DIRTY)
1342 clear_page_dirty_for_io(pages[i]); 1447 clear_page_dirty_for_io(pages[i]);
1343 if (set_writeback) 1448 if (op & EXTENT_SET_WRITEBACK)
1344 set_page_writeback(pages[i]); 1449 set_page_writeback(pages[i]);
1345 if (end_writeback) 1450 if (op & EXTENT_END_WRITEBACK)
1346 end_page_writeback(pages[i]); 1451 end_page_writeback(pages[i]);
1347 if (unlock_pages) 1452 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1348 unlock_page(pages[i]); 1453 unlock_page(pages[i]);
1349 page_cache_release(pages[i]); 1454 page_cache_release(pages[i]);
1350 } 1455 }
@@ -1476,14 +1581,17 @@ out:
1476 * range is found set. 1581 * range is found set.
1477 */ 1582 */
1478int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, 1583int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1479 int bits, int filled) 1584 int bits, int filled, struct extent_state *cached)
1480{ 1585{
1481 struct extent_state *state = NULL; 1586 struct extent_state *state = NULL;
1482 struct rb_node *node; 1587 struct rb_node *node;
1483 int bitset = 0; 1588 int bitset = 0;
1484 1589
1485 spin_lock(&tree->lock); 1590 spin_lock(&tree->lock);
1486 node = tree_search(tree, start); 1591 if (cached && cached->tree && cached->start == start)
1592 node = &cached->rb_node;
1593 else
1594 node = tree_search(tree, start);
1487 while (node && start <= end) { 1595 while (node && start <= end) {
1488 state = rb_entry(node, struct extent_state, rb_node); 1596 state = rb_entry(node, struct extent_state, rb_node);
1489 1597
@@ -1503,6 +1611,10 @@ int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1503 bitset = 0; 1611 bitset = 0;
1504 break; 1612 break;
1505 } 1613 }
1614
1615 if (state->end == (u64)-1)
1616 break;
1617
1506 start = state->end + 1; 1618 start = state->end + 1;
1507 if (start > end) 1619 if (start > end)
1508 break; 1620 break;
@@ -1526,7 +1638,7 @@ static int check_page_uptodate(struct extent_io_tree *tree,
1526{ 1638{
1527 u64 start = (u64)page->index << PAGE_CACHE_SHIFT; 1639 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1528 u64 end = start + PAGE_CACHE_SIZE - 1; 1640 u64 end = start + PAGE_CACHE_SIZE - 1;
1529 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1)) 1641 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1530 SetPageUptodate(page); 1642 SetPageUptodate(page);
1531 return 0; 1643 return 0;
1532} 1644}
@@ -1540,7 +1652,7 @@ static int check_page_locked(struct extent_io_tree *tree,
1540{ 1652{
1541 u64 start = (u64)page->index << PAGE_CACHE_SHIFT; 1653 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1542 u64 end = start + PAGE_CACHE_SIZE - 1; 1654 u64 end = start + PAGE_CACHE_SIZE - 1;
1543 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0)) 1655 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1544 unlock_page(page); 1656 unlock_page(page);
1545 return 0; 1657 return 0;
1546} 1658}
@@ -1552,10 +1664,7 @@ static int check_page_locked(struct extent_io_tree *tree,
1552static int check_page_writeback(struct extent_io_tree *tree, 1664static int check_page_writeback(struct extent_io_tree *tree,
1553 struct page *page) 1665 struct page *page)
1554{ 1666{
1555 u64 start = (u64)page->index << PAGE_CACHE_SHIFT; 1667 end_page_writeback(page);
1556 u64 end = start + PAGE_CACHE_SIZE - 1;
1557 if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1558 end_page_writeback(page);
1559 return 0; 1668 return 0;
1560} 1669}
1561 1670
@@ -1613,13 +1722,11 @@ static void end_bio_extent_writepage(struct bio *bio, int err)
1613 } 1722 }
1614 1723
1615 if (!uptodate) { 1724 if (!uptodate) {
1616 clear_extent_uptodate(tree, start, end, GFP_ATOMIC); 1725 clear_extent_uptodate(tree, start, end, GFP_NOFS);
1617 ClearPageUptodate(page); 1726 ClearPageUptodate(page);
1618 SetPageError(page); 1727 SetPageError(page);
1619 } 1728 }
1620 1729
1621 clear_extent_writeback(tree, start, end, GFP_ATOMIC);
1622
1623 if (whole_page) 1730 if (whole_page)
1624 end_page_writeback(page); 1731 end_page_writeback(page);
1625 else 1732 else
@@ -1983,7 +2090,8 @@ static int __extent_read_full_page(struct extent_io_tree *tree,
1983 continue; 2090 continue;
1984 } 2091 }
1985 /* the get_extent function already copied into the page */ 2092 /* the get_extent function already copied into the page */
1986 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) { 2093 if (test_range_bit(tree, cur, cur_end,
2094 EXTENT_UPTODATE, 1, NULL)) {
1987 check_page_uptodate(tree, page); 2095 check_page_uptodate(tree, page);
1988 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); 2096 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1989 cur = cur + iosize; 2097 cur = cur + iosize;
@@ -2078,6 +2186,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2078 u64 iosize; 2186 u64 iosize;
2079 u64 unlock_start; 2187 u64 unlock_start;
2080 sector_t sector; 2188 sector_t sector;
2189 struct extent_state *cached_state = NULL;
2081 struct extent_map *em; 2190 struct extent_map *em;
2082 struct block_device *bdev; 2191 struct block_device *bdev;
2083 int ret; 2192 int ret;
@@ -2124,6 +2233,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2124 delalloc_end = 0; 2233 delalloc_end = 0;
2125 page_started = 0; 2234 page_started = 0;
2126 if (!epd->extent_locked) { 2235 if (!epd->extent_locked) {
2236 u64 delalloc_to_write = 0;
2127 /* 2237 /*
2128 * make sure the wbc mapping index is at least updated 2238 * make sure the wbc mapping index is at least updated
2129 * to this page. 2239 * to this page.
@@ -2143,8 +2253,24 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2143 tree->ops->fill_delalloc(inode, page, delalloc_start, 2253 tree->ops->fill_delalloc(inode, page, delalloc_start,
2144 delalloc_end, &page_started, 2254 delalloc_end, &page_started,
2145 &nr_written); 2255 &nr_written);
2256 /*
2257 * delalloc_end is already one less than the total
2258 * length, so we don't subtract one from
2259 * PAGE_CACHE_SIZE
2260 */
2261 delalloc_to_write += (delalloc_end - delalloc_start +
2262 PAGE_CACHE_SIZE) >>
2263 PAGE_CACHE_SHIFT;
2146 delalloc_start = delalloc_end + 1; 2264 delalloc_start = delalloc_end + 1;
2147 } 2265 }
2266 if (wbc->nr_to_write < delalloc_to_write) {
2267 int thresh = 8192;
2268
2269 if (delalloc_to_write < thresh * 2)
2270 thresh = delalloc_to_write;
2271 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2272 thresh);
2273 }
2148 2274
2149 /* did the fill delalloc function already unlock and start 2275 /* did the fill delalloc function already unlock and start
2150 * the IO? 2276 * the IO?
@@ -2160,15 +2286,10 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2160 goto done_unlocked; 2286 goto done_unlocked;
2161 } 2287 }
2162 } 2288 }
2163 lock_extent(tree, start, page_end, GFP_NOFS);
2164
2165 unlock_start = start;
2166
2167 if (tree->ops && tree->ops->writepage_start_hook) { 2289 if (tree->ops && tree->ops->writepage_start_hook) {
2168 ret = tree->ops->writepage_start_hook(page, start, 2290 ret = tree->ops->writepage_start_hook(page, start,
2169 page_end); 2291 page_end);
2170 if (ret == -EAGAIN) { 2292 if (ret == -EAGAIN) {
2171 unlock_extent(tree, start, page_end, GFP_NOFS);
2172 redirty_page_for_writepage(wbc, page); 2293 redirty_page_for_writepage(wbc, page);
2173 update_nr_written(page, wbc, nr_written); 2294 update_nr_written(page, wbc, nr_written);
2174 unlock_page(page); 2295 unlock_page(page);
@@ -2184,12 +2305,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2184 update_nr_written(page, wbc, nr_written + 1); 2305 update_nr_written(page, wbc, nr_written + 1);
2185 2306
2186 end = page_end; 2307 end = page_end;
2187 if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0))
2188 printk(KERN_ERR "btrfs delalloc bits after lock_extent\n");
2189
2190 if (last_byte <= start) { 2308 if (last_byte <= start) {
2191 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
2192 unlock_extent(tree, start, page_end, GFP_NOFS);
2193 if (tree->ops && tree->ops->writepage_end_io_hook) 2309 if (tree->ops && tree->ops->writepage_end_io_hook)
2194 tree->ops->writepage_end_io_hook(page, start, 2310 tree->ops->writepage_end_io_hook(page, start,
2195 page_end, NULL, 1); 2311 page_end, NULL, 1);
@@ -2197,13 +2313,10 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2197 goto done; 2313 goto done;
2198 } 2314 }
2199 2315
2200 set_extent_uptodate(tree, start, page_end, GFP_NOFS);
2201 blocksize = inode->i_sb->s_blocksize; 2316 blocksize = inode->i_sb->s_blocksize;
2202 2317
2203 while (cur <= end) { 2318 while (cur <= end) {
2204 if (cur >= last_byte) { 2319 if (cur >= last_byte) {
2205 clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
2206 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2207 if (tree->ops && tree->ops->writepage_end_io_hook) 2320 if (tree->ops && tree->ops->writepage_end_io_hook)
2208 tree->ops->writepage_end_io_hook(page, cur, 2321 tree->ops->writepage_end_io_hook(page, cur,
2209 page_end, NULL, 1); 2322 page_end, NULL, 1);
@@ -2235,12 +2348,6 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2235 */ 2348 */
2236 if (compressed || block_start == EXTENT_MAP_HOLE || 2349 if (compressed || block_start == EXTENT_MAP_HOLE ||
2237 block_start == EXTENT_MAP_INLINE) { 2350 block_start == EXTENT_MAP_INLINE) {
2238 clear_extent_dirty(tree, cur,
2239 cur + iosize - 1, GFP_NOFS);
2240
2241 unlock_extent(tree, unlock_start, cur + iosize - 1,
2242 GFP_NOFS);
2243
2244 /* 2351 /*
2245 * end_io notification does not happen here for 2352 * end_io notification does not happen here for
2246 * compressed extents 2353 * compressed extents
@@ -2265,13 +2372,12 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2265 } 2372 }
2266 /* leave this out until we have a page_mkwrite call */ 2373 /* leave this out until we have a page_mkwrite call */
2267 if (0 && !test_range_bit(tree, cur, cur + iosize - 1, 2374 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2268 EXTENT_DIRTY, 0)) { 2375 EXTENT_DIRTY, 0, NULL)) {
2269 cur = cur + iosize; 2376 cur = cur + iosize;
2270 pg_offset += iosize; 2377 pg_offset += iosize;
2271 continue; 2378 continue;
2272 } 2379 }
2273 2380
2274 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
2275 if (tree->ops && tree->ops->writepage_io_hook) { 2381 if (tree->ops && tree->ops->writepage_io_hook) {
2276 ret = tree->ops->writepage_io_hook(page, cur, 2382 ret = tree->ops->writepage_io_hook(page, cur,
2277 cur + iosize - 1); 2383 cur + iosize - 1);
@@ -2309,12 +2415,12 @@ done:
2309 set_page_writeback(page); 2415 set_page_writeback(page);
2310 end_page_writeback(page); 2416 end_page_writeback(page);
2311 } 2417 }
2312 if (unlock_start <= page_end)
2313 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2314 unlock_page(page); 2418 unlock_page(page);
2315 2419
2316done_unlocked: 2420done_unlocked:
2317 2421
2422 /* drop our reference on any cached states */
2423 free_extent_state(cached_state);
2318 return 0; 2424 return 0;
2319} 2425}
2320 2426
@@ -2339,9 +2445,9 @@ static int extent_write_cache_pages(struct extent_io_tree *tree,
2339 writepage_t writepage, void *data, 2445 writepage_t writepage, void *data,
2340 void (*flush_fn)(void *)) 2446 void (*flush_fn)(void *))
2341{ 2447{
2342 struct backing_dev_info *bdi = mapping->backing_dev_info;
2343 int ret = 0; 2448 int ret = 0;
2344 int done = 0; 2449 int done = 0;
2450 int nr_to_write_done = 0;
2345 struct pagevec pvec; 2451 struct pagevec pvec;
2346 int nr_pages; 2452 int nr_pages;
2347 pgoff_t index; 2453 pgoff_t index;
@@ -2361,7 +2467,7 @@ static int extent_write_cache_pages(struct extent_io_tree *tree,
2361 scanned = 1; 2467 scanned = 1;
2362 } 2468 }
2363retry: 2469retry:
2364 while (!done && (index <= end) && 2470 while (!done && !nr_to_write_done && (index <= end) &&
2365 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, 2471 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2366 PAGECACHE_TAG_DIRTY, min(end - index, 2472 PAGECACHE_TAG_DIRTY, min(end - index,
2367 (pgoff_t)PAGEVEC_SIZE-1) + 1))) { 2473 (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
@@ -2412,12 +2518,15 @@ retry:
2412 unlock_page(page); 2518 unlock_page(page);
2413 ret = 0; 2519 ret = 0;
2414 } 2520 }
2415 if (ret || wbc->nr_to_write <= 0) 2521 if (ret)
2416 done = 1;
2417 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2418 wbc->encountered_congestion = 1;
2419 done = 1; 2522 done = 1;
2420 } 2523
2524 /*
2525 * the filesystem may choose to bump up nr_to_write.
2526 * We have to make sure to honor the new nr_to_write
2527 * at any time
2528 */
2529 nr_to_write_done = wbc->nr_to_write <= 0;
2421 } 2530 }
2422 pagevec_release(&pvec); 2531 pagevec_release(&pvec);
2423 cond_resched(); 2532 cond_resched();
@@ -2604,10 +2713,11 @@ int extent_invalidatepage(struct extent_io_tree *tree,
2604 return 0; 2713 return 0;
2605 2714
2606 lock_extent(tree, start, end, GFP_NOFS); 2715 lock_extent(tree, start, end, GFP_NOFS);
2607 wait_on_extent_writeback(tree, start, end); 2716 wait_on_page_writeback(page);
2608 clear_extent_bit(tree, start, end, 2717 clear_extent_bit(tree, start, end,
2609 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC, 2718 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
2610 1, 1, GFP_NOFS); 2719 EXTENT_DO_ACCOUNTING,
2720 1, 1, NULL, GFP_NOFS);
2611 return 0; 2721 return 0;
2612} 2722}
2613 2723
@@ -2687,7 +2797,7 @@ int extent_prepare_write(struct extent_io_tree *tree,
2687 !isnew && !PageUptodate(page) && 2797 !isnew && !PageUptodate(page) &&
2688 (block_off_end > to || block_off_start < from) && 2798 (block_off_end > to || block_off_start < from) &&
2689 !test_range_bit(tree, block_start, cur_end, 2799 !test_range_bit(tree, block_start, cur_end,
2690 EXTENT_UPTODATE, 1)) { 2800 EXTENT_UPTODATE, 1, NULL)) {
2691 u64 sector; 2801 u64 sector;
2692 u64 extent_offset = block_start - em->start; 2802 u64 extent_offset = block_start - em->start;
2693 size_t iosize; 2803 size_t iosize;
@@ -2701,7 +2811,7 @@ int extent_prepare_write(struct extent_io_tree *tree,
2701 */ 2811 */
2702 set_extent_bit(tree, block_start, 2812 set_extent_bit(tree, block_start,
2703 block_start + iosize - 1, 2813 block_start + iosize - 1,
2704 EXTENT_LOCKED, 0, NULL, GFP_NOFS); 2814 EXTENT_LOCKED, 0, NULL, NULL, GFP_NOFS);
2705 ret = submit_extent_page(READ, tree, page, 2815 ret = submit_extent_page(READ, tree, page,
2706 sector, iosize, page_offset, em->bdev, 2816 sector, iosize, page_offset, em->bdev,
2707 NULL, 1, 2817 NULL, 1,
@@ -2742,13 +2852,18 @@ int try_release_extent_state(struct extent_map_tree *map,
2742 int ret = 1; 2852 int ret = 1;
2743 2853
2744 if (test_range_bit(tree, start, end, 2854 if (test_range_bit(tree, start, end,
2745 EXTENT_IOBITS | EXTENT_ORDERED, 0)) 2855 EXTENT_IOBITS, 0, NULL))
2746 ret = 0; 2856 ret = 0;
2747 else { 2857 else {
2748 if ((mask & GFP_NOFS) == GFP_NOFS) 2858 if ((mask & GFP_NOFS) == GFP_NOFS)
2749 mask = GFP_NOFS; 2859 mask = GFP_NOFS;
2750 clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 2860 /*
2751 1, 1, mask); 2861 * at this point we can safely clear everything except the
2862 * locked bit and the nodatasum bit
2863 */
2864 clear_extent_bit(tree, start, end,
2865 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
2866 0, 0, NULL, mask);
2752 } 2867 }
2753 return ret; 2868 return ret;
2754} 2869}
@@ -2771,29 +2886,28 @@ int try_release_extent_mapping(struct extent_map_tree *map,
2771 u64 len; 2886 u64 len;
2772 while (start <= end) { 2887 while (start <= end) {
2773 len = end - start + 1; 2888 len = end - start + 1;
2774 spin_lock(&map->lock); 2889 write_lock(&map->lock);
2775 em = lookup_extent_mapping(map, start, len); 2890 em = lookup_extent_mapping(map, start, len);
2776 if (!em || IS_ERR(em)) { 2891 if (!em || IS_ERR(em)) {
2777 spin_unlock(&map->lock); 2892 write_unlock(&map->lock);
2778 break; 2893 break;
2779 } 2894 }
2780 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || 2895 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2781 em->start != start) { 2896 em->start != start) {
2782 spin_unlock(&map->lock); 2897 write_unlock(&map->lock);
2783 free_extent_map(em); 2898 free_extent_map(em);
2784 break; 2899 break;
2785 } 2900 }
2786 if (!test_range_bit(tree, em->start, 2901 if (!test_range_bit(tree, em->start,
2787 extent_map_end(em) - 1, 2902 extent_map_end(em) - 1,
2788 EXTENT_LOCKED | EXTENT_WRITEBACK | 2903 EXTENT_LOCKED | EXTENT_WRITEBACK,
2789 EXTENT_ORDERED, 2904 0, NULL)) {
2790 0)) {
2791 remove_extent_mapping(map, em); 2905 remove_extent_mapping(map, em);
2792 /* once for the rb tree */ 2906 /* once for the rb tree */
2793 free_extent_map(em); 2907 free_extent_map(em);
2794 } 2908 }
2795 start = extent_map_end(em); 2909 start = extent_map_end(em);
2796 spin_unlock(&map->lock); 2910 write_unlock(&map->lock);
2797 2911
2798 /* once for us */ 2912 /* once for us */
2799 free_extent_map(em); 2913 free_extent_map(em);
@@ -3203,7 +3317,7 @@ int extent_range_uptodate(struct extent_io_tree *tree,
3203 int uptodate; 3317 int uptodate;
3204 unsigned long index; 3318 unsigned long index;
3205 3319
3206 ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1); 3320 ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL);
3207 if (ret) 3321 if (ret)
3208 return 1; 3322 return 1;
3209 while (start <= end) { 3323 while (start <= end) {
@@ -3233,7 +3347,7 @@ int extent_buffer_uptodate(struct extent_io_tree *tree,
3233 return 1; 3347 return 1;
3234 3348
3235 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1, 3349 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3236 EXTENT_UPTODATE, 1); 3350 EXTENT_UPTODATE, 1, NULL);
3237 if (ret) 3351 if (ret)
3238 return ret; 3352 return ret;
3239 3353
@@ -3269,7 +3383,7 @@ int read_extent_buffer_pages(struct extent_io_tree *tree,
3269 return 0; 3383 return 0;
3270 3384
3271 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1, 3385 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3272 EXTENT_UPTODATE, 1)) { 3386 EXTENT_UPTODATE, 1, NULL)) {
3273 return 0; 3387 return 0;
3274 } 3388 }
3275 3389
diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h
index 5bc20abf3f3d..36de250a7b2b 100644
--- a/fs/btrfs/extent_io.h
+++ b/fs/btrfs/extent_io.h
@@ -13,10 +13,9 @@
13#define EXTENT_DEFRAG (1 << 6) 13#define EXTENT_DEFRAG (1 << 6)
14#define EXTENT_DEFRAG_DONE (1 << 7) 14#define EXTENT_DEFRAG_DONE (1 << 7)
15#define EXTENT_BUFFER_FILLED (1 << 8) 15#define EXTENT_BUFFER_FILLED (1 << 8)
16#define EXTENT_ORDERED (1 << 9) 16#define EXTENT_BOUNDARY (1 << 9)
17#define EXTENT_ORDERED_METADATA (1 << 10) 17#define EXTENT_NODATASUM (1 << 10)
18#define EXTENT_BOUNDARY (1 << 11) 18#define EXTENT_DO_ACCOUNTING (1 << 11)
19#define EXTENT_NODATASUM (1 << 12)
20#define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK) 19#define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK)
21 20
22/* flags for bio submission */ 21/* flags for bio submission */
@@ -27,6 +26,16 @@
27#define EXTENT_BUFFER_BLOCKING 1 26#define EXTENT_BUFFER_BLOCKING 1
28#define EXTENT_BUFFER_DIRTY 2 27#define EXTENT_BUFFER_DIRTY 2
29 28
29/* these are flags for extent_clear_unlock_delalloc */
30#define EXTENT_CLEAR_UNLOCK_PAGE 0x1
31#define EXTENT_CLEAR_UNLOCK 0x2
32#define EXTENT_CLEAR_DELALLOC 0x4
33#define EXTENT_CLEAR_DIRTY 0x8
34#define EXTENT_SET_WRITEBACK 0x10
35#define EXTENT_END_WRITEBACK 0x20
36#define EXTENT_SET_PRIVATE2 0x40
37#define EXTENT_CLEAR_ACCOUNTING 0x80
38
30/* 39/*
31 * page->private values. Every page that is controlled by the extent 40 * page->private values. Every page that is controlled by the extent
32 * map has page->private set to one. 41 * map has page->private set to one.
@@ -62,8 +71,13 @@ struct extent_io_ops {
62 struct extent_state *state, int uptodate); 71 struct extent_state *state, int uptodate);
63 int (*set_bit_hook)(struct inode *inode, u64 start, u64 end, 72 int (*set_bit_hook)(struct inode *inode, u64 start, u64 end,
64 unsigned long old, unsigned long bits); 73 unsigned long old, unsigned long bits);
65 int (*clear_bit_hook)(struct inode *inode, u64 start, u64 end, 74 int (*clear_bit_hook)(struct inode *inode, struct extent_state *state,
66 unsigned long old, unsigned long bits); 75 unsigned long bits);
76 int (*merge_extent_hook)(struct inode *inode,
77 struct extent_state *new,
78 struct extent_state *other);
79 int (*split_extent_hook)(struct inode *inode,
80 struct extent_state *orig, u64 split);
67 int (*write_cache_pages_lock_hook)(struct page *page); 81 int (*write_cache_pages_lock_hook)(struct page *page);
68}; 82};
69 83
@@ -81,10 +95,14 @@ struct extent_state {
81 u64 start; 95 u64 start;
82 u64 end; /* inclusive */ 96 u64 end; /* inclusive */
83 struct rb_node rb_node; 97 struct rb_node rb_node;
98
99 /* ADD NEW ELEMENTS AFTER THIS */
84 struct extent_io_tree *tree; 100 struct extent_io_tree *tree;
85 wait_queue_head_t wq; 101 wait_queue_head_t wq;
86 atomic_t refs; 102 atomic_t refs;
87 unsigned long state; 103 unsigned long state;
104 u64 split_start;
105 u64 split_end;
88 106
89 /* for use by the FS */ 107 /* for use by the FS */
90 u64 private; 108 u64 private;
@@ -142,6 +160,8 @@ int try_release_extent_state(struct extent_map_tree *map,
142 struct extent_io_tree *tree, struct page *page, 160 struct extent_io_tree *tree, struct page *page,
143 gfp_t mask); 161 gfp_t mask);
144int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask); 162int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask);
163int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
164 int bits, struct extent_state **cached, gfp_t mask);
145int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask); 165int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask);
146int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end, 166int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
147 gfp_t mask); 167 gfp_t mask);
@@ -155,11 +175,12 @@ u64 count_range_bits(struct extent_io_tree *tree,
155 u64 max_bytes, unsigned long bits); 175 u64 max_bytes, unsigned long bits);
156 176
157int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, 177int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
158 int bits, int filled); 178 int bits, int filled, struct extent_state *cached_state);
159int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, 179int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
160 int bits, gfp_t mask); 180 int bits, gfp_t mask);
161int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, 181int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
162 int bits, int wake, int delete, gfp_t mask); 182 int bits, int wake, int delete, struct extent_state **cached,
183 gfp_t mask);
163int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, 184int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
164 int bits, gfp_t mask); 185 int bits, gfp_t mask);
165int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, 186int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
@@ -278,9 +299,5 @@ int extent_range_uptodate(struct extent_io_tree *tree,
278int extent_clear_unlock_delalloc(struct inode *inode, 299int extent_clear_unlock_delalloc(struct inode *inode,
279 struct extent_io_tree *tree, 300 struct extent_io_tree *tree,
280 u64 start, u64 end, struct page *locked_page, 301 u64 start, u64 end, struct page *locked_page,
281 int unlock_page, 302 unsigned long op);
282 int clear_unlock,
283 int clear_delalloc, int clear_dirty,
284 int set_writeback,
285 int end_writeback);
286#endif 303#endif
diff --git a/fs/btrfs/extent_map.c b/fs/btrfs/extent_map.c
index 30c9365861e6..2c726b7b9faa 100644
--- a/fs/btrfs/extent_map.c
+++ b/fs/btrfs/extent_map.c
@@ -36,7 +36,7 @@ void extent_map_exit(void)
36void extent_map_tree_init(struct extent_map_tree *tree, gfp_t mask) 36void extent_map_tree_init(struct extent_map_tree *tree, gfp_t mask)
37{ 37{
38 tree->map.rb_node = NULL; 38 tree->map.rb_node = NULL;
39 spin_lock_init(&tree->lock); 39 rwlock_init(&tree->lock);
40} 40}
41 41
42/** 42/**
@@ -198,6 +198,56 @@ static int mergable_maps(struct extent_map *prev, struct extent_map *next)
198 return 0; 198 return 0;
199} 199}
200 200
201int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len)
202{
203 int ret = 0;
204 struct extent_map *merge = NULL;
205 struct rb_node *rb;
206 struct extent_map *em;
207
208 write_lock(&tree->lock);
209 em = lookup_extent_mapping(tree, start, len);
210
211 WARN_ON(em->start != start || !em);
212
213 if (!em)
214 goto out;
215
216 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
217
218 if (em->start != 0) {
219 rb = rb_prev(&em->rb_node);
220 if (rb)
221 merge = rb_entry(rb, struct extent_map, rb_node);
222 if (rb && mergable_maps(merge, em)) {
223 em->start = merge->start;
224 em->len += merge->len;
225 em->block_len += merge->block_len;
226 em->block_start = merge->block_start;
227 merge->in_tree = 0;
228 rb_erase(&merge->rb_node, &tree->map);
229 free_extent_map(merge);
230 }
231 }
232
233 rb = rb_next(&em->rb_node);
234 if (rb)
235 merge = rb_entry(rb, struct extent_map, rb_node);
236 if (rb && mergable_maps(em, merge)) {
237 em->len += merge->len;
238 em->block_len += merge->len;
239 rb_erase(&merge->rb_node, &tree->map);
240 merge->in_tree = 0;
241 free_extent_map(merge);
242 }
243
244 free_extent_map(em);
245out:
246 write_unlock(&tree->lock);
247 return ret;
248
249}
250
201/** 251/**
202 * add_extent_mapping - add new extent map to the extent tree 252 * add_extent_mapping - add new extent map to the extent tree
203 * @tree: tree to insert new map in 253 * @tree: tree to insert new map in
@@ -222,7 +272,6 @@ int add_extent_mapping(struct extent_map_tree *tree,
222 ret = -EEXIST; 272 ret = -EEXIST;
223 goto out; 273 goto out;
224 } 274 }
225 assert_spin_locked(&tree->lock);
226 rb = tree_insert(&tree->map, em->start, &em->rb_node); 275 rb = tree_insert(&tree->map, em->start, &em->rb_node);
227 if (rb) { 276 if (rb) {
228 ret = -EEXIST; 277 ret = -EEXIST;
@@ -285,7 +334,6 @@ struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
285 struct rb_node *next = NULL; 334 struct rb_node *next = NULL;
286 u64 end = range_end(start, len); 335 u64 end = range_end(start, len);
287 336
288 assert_spin_locked(&tree->lock);
289 rb_node = __tree_search(&tree->map, start, &prev, &next); 337 rb_node = __tree_search(&tree->map, start, &prev, &next);
290 if (!rb_node && prev) { 338 if (!rb_node && prev) {
291 em = rb_entry(prev, struct extent_map, rb_node); 339 em = rb_entry(prev, struct extent_map, rb_node);
@@ -319,6 +367,54 @@ out:
319} 367}
320 368
321/** 369/**
370 * search_extent_mapping - find a nearby extent map
371 * @tree: tree to lookup in
372 * @start: byte offset to start the search
373 * @len: length of the lookup range
374 *
375 * Find and return the first extent_map struct in @tree that intersects the
376 * [start, len] range.
377 *
378 * If one can't be found, any nearby extent may be returned
379 */
380struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
381 u64 start, u64 len)
382{
383 struct extent_map *em;
384 struct rb_node *rb_node;
385 struct rb_node *prev = NULL;
386 struct rb_node *next = NULL;
387
388 rb_node = __tree_search(&tree->map, start, &prev, &next);
389 if (!rb_node && prev) {
390 em = rb_entry(prev, struct extent_map, rb_node);
391 goto found;
392 }
393 if (!rb_node && next) {
394 em = rb_entry(next, struct extent_map, rb_node);
395 goto found;
396 }
397 if (!rb_node) {
398 em = NULL;
399 goto out;
400 }
401 if (IS_ERR(rb_node)) {
402 em = ERR_PTR(PTR_ERR(rb_node));
403 goto out;
404 }
405 em = rb_entry(rb_node, struct extent_map, rb_node);
406 goto found;
407
408 em = NULL;
409 goto out;
410
411found:
412 atomic_inc(&em->refs);
413out:
414 return em;
415}
416
417/**
322 * remove_extent_mapping - removes an extent_map from the extent tree 418 * remove_extent_mapping - removes an extent_map from the extent tree
323 * @tree: extent tree to remove from 419 * @tree: extent tree to remove from
324 * @em: extent map beeing removed 420 * @em: extent map beeing removed
@@ -331,7 +427,6 @@ int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
331 int ret = 0; 427 int ret = 0;
332 428
333 WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags)); 429 WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
334 assert_spin_locked(&tree->lock);
335 rb_erase(&em->rb_node, &tree->map); 430 rb_erase(&em->rb_node, &tree->map);
336 em->in_tree = 0; 431 em->in_tree = 0;
337 return ret; 432 return ret;
diff --git a/fs/btrfs/extent_map.h b/fs/btrfs/extent_map.h
index fb6eeef06bb0..ab6d74b6e647 100644
--- a/fs/btrfs/extent_map.h
+++ b/fs/btrfs/extent_map.h
@@ -31,7 +31,7 @@ struct extent_map {
31 31
32struct extent_map_tree { 32struct extent_map_tree {
33 struct rb_root map; 33 struct rb_root map;
34 spinlock_t lock; 34 rwlock_t lock;
35}; 35};
36 36
37static inline u64 extent_map_end(struct extent_map *em) 37static inline u64 extent_map_end(struct extent_map *em)
@@ -59,4 +59,7 @@ struct extent_map *alloc_extent_map(gfp_t mask);
59void free_extent_map(struct extent_map *em); 59void free_extent_map(struct extent_map *em);
60int __init extent_map_init(void); 60int __init extent_map_init(void);
61void extent_map_exit(void); 61void extent_map_exit(void);
62int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len);
63struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
64 u64 start, u64 len);
62#endif 65#endif
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index 4b833972273a..2d623aa0625f 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -112,8 +112,6 @@ static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
112 int err = 0; 112 int err = 0;
113 int i; 113 int i;
114 struct inode *inode = fdentry(file)->d_inode; 114 struct inode *inode = fdentry(file)->d_inode;
115 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
116 u64 hint_byte;
117 u64 num_bytes; 115 u64 num_bytes;
118 u64 start_pos; 116 u64 start_pos;
119 u64 end_of_last_block; 117 u64 end_of_last_block;
@@ -125,23 +123,10 @@ static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
125 root->sectorsize - 1) & ~((u64)root->sectorsize - 1); 123 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
126 124
127 end_of_last_block = start_pos + num_bytes - 1; 125 end_of_last_block = start_pos + num_bytes - 1;
126 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
127 if (err)
128 return err;
128 129
129 lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
130 trans = btrfs_join_transaction(root, 1);
131 if (!trans) {
132 err = -ENOMEM;
133 goto out_unlock;
134 }
135 btrfs_set_trans_block_group(trans, inode);
136 hint_byte = 0;
137
138 set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
139
140 /* check for reserved extents on each page, we don't want
141 * to reset the delalloc bit on things that already have
142 * extents reserved.
143 */
144 btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
145 for (i = 0; i < num_pages; i++) { 130 for (i = 0; i < num_pages; i++) {
146 struct page *p = pages[i]; 131 struct page *p = pages[i];
147 SetPageUptodate(p); 132 SetPageUptodate(p);
@@ -155,9 +140,6 @@ static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
155 * at this time. 140 * at this time.
156 */ 141 */
157 } 142 }
158 err = btrfs_end_transaction(trans, root);
159out_unlock:
160 unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
161 return err; 143 return err;
162} 144}
163 145
@@ -189,18 +171,18 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
189 if (!split2) 171 if (!split2)
190 split2 = alloc_extent_map(GFP_NOFS); 172 split2 = alloc_extent_map(GFP_NOFS);
191 173
192 spin_lock(&em_tree->lock); 174 write_lock(&em_tree->lock);
193 em = lookup_extent_mapping(em_tree, start, len); 175 em = lookup_extent_mapping(em_tree, start, len);
194 if (!em) { 176 if (!em) {
195 spin_unlock(&em_tree->lock); 177 write_unlock(&em_tree->lock);
196 break; 178 break;
197 } 179 }
198 flags = em->flags; 180 flags = em->flags;
199 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { 181 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
200 spin_unlock(&em_tree->lock);
201 if (em->start <= start && 182 if (em->start <= start &&
202 (!testend || em->start + em->len >= start + len)) { 183 (!testend || em->start + em->len >= start + len)) {
203 free_extent_map(em); 184 free_extent_map(em);
185 write_unlock(&em_tree->lock);
204 break; 186 break;
205 } 187 }
206 if (start < em->start) { 188 if (start < em->start) {
@@ -210,6 +192,7 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
210 start = em->start + em->len; 192 start = em->start + em->len;
211 } 193 }
212 free_extent_map(em); 194 free_extent_map(em);
195 write_unlock(&em_tree->lock);
213 continue; 196 continue;
214 } 197 }
215 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); 198 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
@@ -260,7 +243,7 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
260 free_extent_map(split); 243 free_extent_map(split);
261 split = NULL; 244 split = NULL;
262 } 245 }
263 spin_unlock(&em_tree->lock); 246 write_unlock(&em_tree->lock);
264 247
265 /* once for us */ 248 /* once for us */
266 free_extent_map(em); 249 free_extent_map(em);
@@ -289,7 +272,7 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
289noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans, 272noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
290 struct btrfs_root *root, struct inode *inode, 273 struct btrfs_root *root, struct inode *inode,
291 u64 start, u64 end, u64 locked_end, 274 u64 start, u64 end, u64 locked_end,
292 u64 inline_limit, u64 *hint_byte) 275 u64 inline_limit, u64 *hint_byte, int drop_cache)
293{ 276{
294 u64 extent_end = 0; 277 u64 extent_end = 0;
295 u64 search_start = start; 278 u64 search_start = start;
@@ -314,7 +297,8 @@ noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
314 int ret; 297 int ret;
315 298
316 inline_limit = 0; 299 inline_limit = 0;
317 btrfs_drop_extent_cache(inode, start, end - 1, 0); 300 if (drop_cache)
301 btrfs_drop_extent_cache(inode, start, end - 1, 0);
318 302
319 path = btrfs_alloc_path(); 303 path = btrfs_alloc_path();
320 if (!path) 304 if (!path)
@@ -894,7 +878,8 @@ again:
894 btrfs_put_ordered_extent(ordered); 878 btrfs_put_ordered_extent(ordered);
895 879
896 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos, 880 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
897 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC, 881 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
882 EXTENT_DO_ACCOUNTING,
898 GFP_NOFS); 883 GFP_NOFS);
899 unlock_extent(&BTRFS_I(inode)->io_tree, 884 unlock_extent(&BTRFS_I(inode)->io_tree,
900 start_pos, last_pos - 1, GFP_NOFS); 885 start_pos, last_pos - 1, GFP_NOFS);
@@ -936,21 +921,35 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
936 start_pos = pos; 921 start_pos = pos;
937 922
938 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); 923 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
924
925 /* do the reserve before the mutex lock in case we have to do some
926 * flushing. We wouldn't deadlock, but this is more polite.
927 */
928 err = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
929 if (err)
930 goto out_nolock;
931
932 mutex_lock(&inode->i_mutex);
933
939 current->backing_dev_info = inode->i_mapping->backing_dev_info; 934 current->backing_dev_info = inode->i_mapping->backing_dev_info;
940 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); 935 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
941 if (err) 936 if (err)
942 goto out_nolock; 937 goto out;
938
943 if (count == 0) 939 if (count == 0)
944 goto out_nolock; 940 goto out;
945 941
946 err = file_remove_suid(file); 942 err = file_remove_suid(file);
947 if (err) 943 if (err)
948 goto out_nolock; 944 goto out;
945
949 file_update_time(file); 946 file_update_time(file);
950 947
951 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL); 948 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
952 949
953 mutex_lock(&inode->i_mutex); 950 /* generic_write_checks can change our pos */
951 start_pos = pos;
952
954 BTRFS_I(inode)->sequence++; 953 BTRFS_I(inode)->sequence++;
955 first_index = pos >> PAGE_CACHE_SHIFT; 954 first_index = pos >> PAGE_CACHE_SHIFT;
956 last_index = (pos + count) >> PAGE_CACHE_SHIFT; 955 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
@@ -1024,9 +1023,8 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
1024 } 1023 }
1025 1024
1026 if (will_write) { 1025 if (will_write) {
1027 btrfs_fdatawrite_range(inode->i_mapping, pos, 1026 filemap_fdatawrite_range(inode->i_mapping, pos,
1028 pos + write_bytes - 1, 1027 pos + write_bytes - 1);
1029 WB_SYNC_ALL);
1030 } else { 1028 } else {
1031 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1029 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1032 num_pages); 1030 num_pages);
@@ -1047,6 +1045,7 @@ out:
1047 mutex_unlock(&inode->i_mutex); 1045 mutex_unlock(&inode->i_mutex);
1048 if (ret) 1046 if (ret)
1049 err = ret; 1047 err = ret;
1048 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
1050 1049
1051out_nolock: 1050out_nolock:
1052 kfree(pages); 1051 kfree(pages);
@@ -1203,7 +1202,7 @@ out:
1203 return ret > 0 ? EIO : ret; 1202 return ret > 0 ? EIO : ret;
1204} 1203}
1205 1204
1206static struct vm_operations_struct btrfs_file_vm_ops = { 1205static const struct vm_operations_struct btrfs_file_vm_ops = {
1207 .fault = filemap_fault, 1206 .fault = filemap_fault,
1208 .page_mkwrite = btrfs_page_mkwrite, 1207 .page_mkwrite = btrfs_page_mkwrite,
1209}; 1208};
@@ -1215,7 +1214,7 @@ static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1215 return 0; 1214 return 0;
1216} 1215}
1217 1216
1218struct file_operations btrfs_file_operations = { 1217const struct file_operations btrfs_file_operations = {
1219 .llseek = generic_file_llseek, 1218 .llseek = generic_file_llseek,
1220 .read = do_sync_read, 1219 .read = do_sync_read,
1221 .aio_read = generic_file_aio_read, 1220 .aio_read = generic_file_aio_read,
diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c
index 5edcee3a617f..5c2caad76212 100644
--- a/fs/btrfs/free-space-cache.c
+++ b/fs/btrfs/free-space-cache.c
@@ -259,7 +259,9 @@ static int link_free_space(struct btrfs_block_group_cache *block_group,
259 259
260static void recalculate_thresholds(struct btrfs_block_group_cache *block_group) 260static void recalculate_thresholds(struct btrfs_block_group_cache *block_group)
261{ 261{
262 u64 max_bytes, possible_bytes; 262 u64 max_bytes;
263 u64 bitmap_bytes;
264 u64 extent_bytes;
263 265
264 /* 266 /*
265 * The goal is to keep the total amount of memory used per 1gb of space 267 * The goal is to keep the total amount of memory used per 1gb of space
@@ -269,22 +271,27 @@ static void recalculate_thresholds(struct btrfs_block_group_cache *block_group)
269 max_bytes = MAX_CACHE_BYTES_PER_GIG * 271 max_bytes = MAX_CACHE_BYTES_PER_GIG *
270 (div64_u64(block_group->key.offset, 1024 * 1024 * 1024)); 272 (div64_u64(block_group->key.offset, 1024 * 1024 * 1024));
271 273
272 possible_bytes = (block_group->total_bitmaps * PAGE_CACHE_SIZE) + 274 /*
273 (sizeof(struct btrfs_free_space) * 275 * we want to account for 1 more bitmap than what we have so we can make
274 block_group->extents_thresh); 276 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
277 * we add more bitmaps.
278 */
279 bitmap_bytes = (block_group->total_bitmaps + 1) * PAGE_CACHE_SIZE;
275 280
276 if (possible_bytes > max_bytes) { 281 if (bitmap_bytes >= max_bytes) {
277 int extent_bytes = max_bytes - 282 block_group->extents_thresh = 0;
278 (block_group->total_bitmaps * PAGE_CACHE_SIZE); 283 return;
284 }
279 285
280 if (extent_bytes <= 0) { 286 /*
281 block_group->extents_thresh = 0; 287 * we want the extent entry threshold to always be at most 1/2 the maxw
282 return; 288 * bytes we can have, or whatever is less than that.
283 } 289 */
290 extent_bytes = max_bytes - bitmap_bytes;
291 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
284 292
285 block_group->extents_thresh = extent_bytes / 293 block_group->extents_thresh =
286 (sizeof(struct btrfs_free_space)); 294 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
287 }
288} 295}
289 296
290static void bitmap_clear_bits(struct btrfs_block_group_cache *block_group, 297static void bitmap_clear_bits(struct btrfs_block_group_cache *block_group,
@@ -403,6 +410,7 @@ static void add_new_bitmap(struct btrfs_block_group_cache *block_group,
403 BUG_ON(block_group->total_bitmaps >= max_bitmaps); 410 BUG_ON(block_group->total_bitmaps >= max_bitmaps);
404 411
405 info->offset = offset_to_bitmap(block_group, offset); 412 info->offset = offset_to_bitmap(block_group, offset);
413 info->bytes = 0;
406 link_free_space(block_group, info); 414 link_free_space(block_group, info);
407 block_group->total_bitmaps++; 415 block_group->total_bitmaps++;
408 416
diff --git a/fs/btrfs/inode-item.c b/fs/btrfs/inode-item.c
index 6b627c611808..72ce3c173d6a 100644
--- a/fs/btrfs/inode-item.c
+++ b/fs/btrfs/inode-item.c
@@ -149,6 +149,8 @@ int btrfs_insert_inode_ref(struct btrfs_trans_handle *trans,
149 ptr = (unsigned long)(ref + 1); 149 ptr = (unsigned long)(ref + 1);
150 ret = 0; 150 ret = 0;
151 } else if (ret < 0) { 151 } else if (ret < 0) {
152 if (ret == -EOVERFLOW)
153 ret = -EMLINK;
152 goto out; 154 goto out;
153 } else { 155 } else {
154 ref = btrfs_item_ptr(path->nodes[0], path->slots[0], 156 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
@@ -177,8 +179,6 @@ int btrfs_insert_empty_inode(struct btrfs_trans_handle *trans,
177 179
178 ret = btrfs_insert_empty_item(trans, root, path, &key, 180 ret = btrfs_insert_empty_item(trans, root, path, &key,
179 sizeof(struct btrfs_inode_item)); 181 sizeof(struct btrfs_inode_item));
180 if (ret == 0 && objectid > root->highest_inode)
181 root->highest_inode = objectid;
182 return ret; 182 return ret;
183} 183}
184 184
diff --git a/fs/btrfs/inode-map.c b/fs/btrfs/inode-map.c
index 9abbced1123d..c56eb5909172 100644
--- a/fs/btrfs/inode-map.c
+++ b/fs/btrfs/inode-map.c
@@ -43,9 +43,10 @@ int btrfs_find_highest_inode(struct btrfs_root *root, u64 *objectid)
43 slot = path->slots[0] - 1; 43 slot = path->slots[0] - 1;
44 l = path->nodes[0]; 44 l = path->nodes[0];
45 btrfs_item_key_to_cpu(l, &found_key, slot); 45 btrfs_item_key_to_cpu(l, &found_key, slot);
46 *objectid = found_key.objectid; 46 *objectid = max_t(u64, found_key.objectid,
47 BTRFS_FIRST_FREE_OBJECTID - 1);
47 } else { 48 } else {
48 *objectid = BTRFS_FIRST_FREE_OBJECTID; 49 *objectid = BTRFS_FIRST_FREE_OBJECTID - 1;
49 } 50 }
50 ret = 0; 51 ret = 0;
51error: 52error:
@@ -53,91 +54,27 @@ error:
53 return ret; 54 return ret;
54} 55}
55 56
56/*
57 * walks the btree of allocated inodes and find a hole.
58 */
59int btrfs_find_free_objectid(struct btrfs_trans_handle *trans, 57int btrfs_find_free_objectid(struct btrfs_trans_handle *trans,
60 struct btrfs_root *root, 58 struct btrfs_root *root,
61 u64 dirid, u64 *objectid) 59 u64 dirid, u64 *objectid)
62{ 60{
63 struct btrfs_path *path;
64 struct btrfs_key key;
65 int ret; 61 int ret;
66 int slot = 0;
67 u64 last_ino = 0;
68 int start_found;
69 struct extent_buffer *l;
70 struct btrfs_key search_key;
71 u64 search_start = dirid;
72
73 mutex_lock(&root->objectid_mutex); 62 mutex_lock(&root->objectid_mutex);
74 if (root->last_inode_alloc >= BTRFS_FIRST_FREE_OBJECTID &&
75 root->last_inode_alloc < BTRFS_LAST_FREE_OBJECTID) {
76 *objectid = ++root->last_inode_alloc;
77 mutex_unlock(&root->objectid_mutex);
78 return 0;
79 }
80 path = btrfs_alloc_path();
81 BUG_ON(!path);
82 search_start = max(search_start, (u64)BTRFS_FIRST_FREE_OBJECTID);
83 search_key.objectid = search_start;
84 search_key.type = 0;
85 search_key.offset = 0;
86
87 start_found = 0;
88 ret = btrfs_search_slot(trans, root, &search_key, path, 0, 0);
89 if (ret < 0)
90 goto error;
91 63
92 while (1) { 64 if (unlikely(root->highest_objectid < BTRFS_FIRST_FREE_OBJECTID)) {
93 l = path->nodes[0]; 65 ret = btrfs_find_highest_inode(root, &root->highest_objectid);
94 slot = path->slots[0]; 66 if (ret)
95 if (slot >= btrfs_header_nritems(l)) { 67 goto out;
96 ret = btrfs_next_leaf(root, path); 68 }
97 if (ret == 0)
98 continue;
99 if (ret < 0)
100 goto error;
101 if (!start_found) {
102 *objectid = search_start;
103 start_found = 1;
104 goto found;
105 }
106 *objectid = last_ino > search_start ?
107 last_ino : search_start;
108 goto found;
109 }
110 btrfs_item_key_to_cpu(l, &key, slot);
111 if (key.objectid >= search_start) {
112 if (start_found) {
113 if (last_ino < search_start)
114 last_ino = search_start;
115 if (key.objectid > last_ino) {
116 *objectid = last_ino;
117 goto found;
118 }
119 } else if (key.objectid > search_start) {
120 *objectid = search_start;
121 goto found;
122 }
123 }
124 if (key.objectid >= BTRFS_LAST_FREE_OBJECTID)
125 break;
126 69
127 start_found = 1; 70 if (unlikely(root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
128 last_ino = key.objectid + 1; 71 ret = -ENOSPC;
129 path->slots[0]++; 72 goto out;
130 } 73 }
131 BUG_ON(1); 74
132found: 75 *objectid = ++root->highest_objectid;
133 btrfs_release_path(root, path); 76 ret = 0;
134 btrfs_free_path(path); 77out:
135 BUG_ON(*objectid < search_start);
136 mutex_unlock(&root->objectid_mutex);
137 return 0;
138error:
139 btrfs_release_path(root, path);
140 btrfs_free_path(path);
141 mutex_unlock(&root->objectid_mutex); 78 mutex_unlock(&root->objectid_mutex);
142 return ret; 79 return ret;
143} 80}
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 59cba180fe83..9e138b793dc7 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -55,14 +55,14 @@ struct btrfs_iget_args {
55 struct btrfs_root *root; 55 struct btrfs_root *root;
56}; 56};
57 57
58static struct inode_operations btrfs_dir_inode_operations; 58static const struct inode_operations btrfs_dir_inode_operations;
59static struct inode_operations btrfs_symlink_inode_operations; 59static const struct inode_operations btrfs_symlink_inode_operations;
60static struct inode_operations btrfs_dir_ro_inode_operations; 60static const struct inode_operations btrfs_dir_ro_inode_operations;
61static struct inode_operations btrfs_special_inode_operations; 61static const struct inode_operations btrfs_special_inode_operations;
62static struct inode_operations btrfs_file_inode_operations; 62static const struct inode_operations btrfs_file_inode_operations;
63static struct address_space_operations btrfs_aops; 63static const struct address_space_operations btrfs_aops;
64static struct address_space_operations btrfs_symlink_aops; 64static const struct address_space_operations btrfs_symlink_aops;
65static struct file_operations btrfs_dir_file_operations; 65static const struct file_operations btrfs_dir_file_operations;
66static struct extent_io_ops btrfs_extent_io_ops; 66static struct extent_io_ops btrfs_extent_io_ops;
67 67
68static struct kmem_cache *btrfs_inode_cachep; 68static struct kmem_cache *btrfs_inode_cachep;
@@ -231,7 +231,8 @@ static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
231 } 231 }
232 232
233 ret = btrfs_drop_extents(trans, root, inode, start, 233 ret = btrfs_drop_extents(trans, root, inode, start,
234 aligned_end, aligned_end, start, &hint_byte); 234 aligned_end, aligned_end, start,
235 &hint_byte, 1);
235 BUG_ON(ret); 236 BUG_ON(ret);
236 237
237 if (isize > actual_end) 238 if (isize > actual_end)
@@ -240,7 +241,7 @@ static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
240 inline_len, compressed_size, 241 inline_len, compressed_size,
241 compressed_pages); 242 compressed_pages);
242 BUG_ON(ret); 243 BUG_ON(ret);
243 btrfs_drop_extent_cache(inode, start, aligned_end, 0); 244 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
244 return 0; 245 return 0;
245} 246}
246 247
@@ -423,9 +424,12 @@ again:
423 * and free up our temp pages. 424 * and free up our temp pages.
424 */ 425 */
425 extent_clear_unlock_delalloc(inode, 426 extent_clear_unlock_delalloc(inode,
426 &BTRFS_I(inode)->io_tree, 427 &BTRFS_I(inode)->io_tree,
427 start, end, NULL, 1, 0, 428 start, end, NULL,
428 0, 1, 1, 1); 429 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
430 EXTENT_CLEAR_DELALLOC |
431 EXTENT_CLEAR_ACCOUNTING |
432 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
429 ret = 0; 433 ret = 0;
430 goto free_pages_out; 434 goto free_pages_out;
431 } 435 }
@@ -611,9 +615,9 @@ static noinline int submit_compressed_extents(struct inode *inode,
611 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); 615 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
612 616
613 while (1) { 617 while (1) {
614 spin_lock(&em_tree->lock); 618 write_lock(&em_tree->lock);
615 ret = add_extent_mapping(em_tree, em); 619 ret = add_extent_mapping(em_tree, em);
616 spin_unlock(&em_tree->lock); 620 write_unlock(&em_tree->lock);
617 if (ret != -EEXIST) { 621 if (ret != -EEXIST) {
618 free_extent_map(em); 622 free_extent_map(em);
619 break; 623 break;
@@ -636,11 +640,14 @@ static noinline int submit_compressed_extents(struct inode *inode,
636 * clear dirty, set writeback and unlock the pages. 640 * clear dirty, set writeback and unlock the pages.
637 */ 641 */
638 extent_clear_unlock_delalloc(inode, 642 extent_clear_unlock_delalloc(inode,
639 &BTRFS_I(inode)->io_tree, 643 &BTRFS_I(inode)->io_tree,
640 async_extent->start, 644 async_extent->start,
641 async_extent->start + 645 async_extent->start +
642 async_extent->ram_size - 1, 646 async_extent->ram_size - 1,
643 NULL, 1, 1, 0, 1, 1, 0); 647 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
648 EXTENT_CLEAR_UNLOCK |
649 EXTENT_CLEAR_DELALLOC |
650 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
644 651
645 ret = btrfs_submit_compressed_write(inode, 652 ret = btrfs_submit_compressed_write(inode,
646 async_extent->start, 653 async_extent->start,
@@ -711,9 +718,15 @@ static noinline int cow_file_range(struct inode *inode,
711 start, end, 0, NULL); 718 start, end, 0, NULL);
712 if (ret == 0) { 719 if (ret == 0) {
713 extent_clear_unlock_delalloc(inode, 720 extent_clear_unlock_delalloc(inode,
714 &BTRFS_I(inode)->io_tree, 721 &BTRFS_I(inode)->io_tree,
715 start, end, NULL, 1, 1, 722 start, end, NULL,
716 1, 1, 1, 1); 723 EXTENT_CLEAR_UNLOCK_PAGE |
724 EXTENT_CLEAR_UNLOCK |
725 EXTENT_CLEAR_DELALLOC |
726 EXTENT_CLEAR_ACCOUNTING |
727 EXTENT_CLEAR_DIRTY |
728 EXTENT_SET_WRITEBACK |
729 EXTENT_END_WRITEBACK);
717 *nr_written = *nr_written + 730 *nr_written = *nr_written +
718 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; 731 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
719 *page_started = 1; 732 *page_started = 1;
@@ -725,9 +738,20 @@ static noinline int cow_file_range(struct inode *inode,
725 BUG_ON(disk_num_bytes > 738 BUG_ON(disk_num_bytes >
726 btrfs_super_total_bytes(&root->fs_info->super_copy)); 739 btrfs_super_total_bytes(&root->fs_info->super_copy));
727 740
741
742 read_lock(&BTRFS_I(inode)->extent_tree.lock);
743 em = search_extent_mapping(&BTRFS_I(inode)->extent_tree,
744 start, num_bytes);
745 if (em) {
746 alloc_hint = em->block_start;
747 free_extent_map(em);
748 }
749 read_unlock(&BTRFS_I(inode)->extent_tree.lock);
728 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); 750 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
729 751
730 while (disk_num_bytes > 0) { 752 while (disk_num_bytes > 0) {
753 unsigned long op;
754
731 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent); 755 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
732 ret = btrfs_reserve_extent(trans, root, cur_alloc_size, 756 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
733 root->sectorsize, 0, alloc_hint, 757 root->sectorsize, 0, alloc_hint,
@@ -737,7 +761,6 @@ static noinline int cow_file_range(struct inode *inode,
737 em = alloc_extent_map(GFP_NOFS); 761 em = alloc_extent_map(GFP_NOFS);
738 em->start = start; 762 em->start = start;
739 em->orig_start = em->start; 763 em->orig_start = em->start;
740
741 ram_size = ins.offset; 764 ram_size = ins.offset;
742 em->len = ins.offset; 765 em->len = ins.offset;
743 766
@@ -747,9 +770,9 @@ static noinline int cow_file_range(struct inode *inode,
747 set_bit(EXTENT_FLAG_PINNED, &em->flags); 770 set_bit(EXTENT_FLAG_PINNED, &em->flags);
748 771
749 while (1) { 772 while (1) {
750 spin_lock(&em_tree->lock); 773 write_lock(&em_tree->lock);
751 ret = add_extent_mapping(em_tree, em); 774 ret = add_extent_mapping(em_tree, em);
752 spin_unlock(&em_tree->lock); 775 write_unlock(&em_tree->lock);
753 if (ret != -EEXIST) { 776 if (ret != -EEXIST) {
754 free_extent_map(em); 777 free_extent_map(em);
755 break; 778 break;
@@ -776,11 +799,17 @@ static noinline int cow_file_range(struct inode *inode,
776 /* we're not doing compressed IO, don't unlock the first 799 /* we're not doing compressed IO, don't unlock the first
777 * page (which the caller expects to stay locked), don't 800 * page (which the caller expects to stay locked), don't
778 * clear any dirty bits and don't set any writeback bits 801 * clear any dirty bits and don't set any writeback bits
802 *
803 * Do set the Private2 bit so we know this page was properly
804 * setup for writepage
779 */ 805 */
806 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
807 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
808 EXTENT_SET_PRIVATE2;
809
780 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, 810 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
781 start, start + ram_size - 1, 811 start, start + ram_size - 1,
782 locked_page, unlock, 1, 812 locked_page, op);
783 1, 0, 0, 0);
784 disk_num_bytes -= cur_alloc_size; 813 disk_num_bytes -= cur_alloc_size;
785 num_bytes -= cur_alloc_size; 814 num_bytes -= cur_alloc_size;
786 alloc_hint = ins.objectid + ins.offset; 815 alloc_hint = ins.objectid + ins.offset;
@@ -852,8 +881,8 @@ static int cow_file_range_async(struct inode *inode, struct page *locked_page,
852 u64 cur_end; 881 u64 cur_end;
853 int limit = 10 * 1024 * 1042; 882 int limit = 10 * 1024 * 1042;
854 883
855 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED | 884 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
856 EXTENT_DELALLOC, 1, 0, GFP_NOFS); 885 1, 0, NULL, GFP_NOFS);
857 while (start < end) { 886 while (start < end) {
858 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); 887 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
859 async_cow->inode = inode; 888 async_cow->inode = inode;
@@ -994,6 +1023,7 @@ next_slot:
994 1023
995 if (found_key.offset > cur_offset) { 1024 if (found_key.offset > cur_offset) {
996 extent_end = found_key.offset; 1025 extent_end = found_key.offset;
1026 extent_type = 0;
997 goto out_check; 1027 goto out_check;
998 } 1028 }
999 1029
@@ -1080,9 +1110,9 @@ out_check:
1080 em->bdev = root->fs_info->fs_devices->latest_bdev; 1110 em->bdev = root->fs_info->fs_devices->latest_bdev;
1081 set_bit(EXTENT_FLAG_PINNED, &em->flags); 1111 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1082 while (1) { 1112 while (1) {
1083 spin_lock(&em_tree->lock); 1113 write_lock(&em_tree->lock);
1084 ret = add_extent_mapping(em_tree, em); 1114 ret = add_extent_mapping(em_tree, em);
1085 spin_unlock(&em_tree->lock); 1115 write_unlock(&em_tree->lock);
1086 if (ret != -EEXIST) { 1116 if (ret != -EEXIST) {
1087 free_extent_map(em); 1117 free_extent_map(em);
1088 break; 1118 break;
@@ -1100,8 +1130,10 @@ out_check:
1100 BUG_ON(ret); 1130 BUG_ON(ret);
1101 1131
1102 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, 1132 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1103 cur_offset, cur_offset + num_bytes - 1, 1133 cur_offset, cur_offset + num_bytes - 1,
1104 locked_page, 1, 1, 1, 0, 0, 0); 1134 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1135 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1136 EXTENT_SET_PRIVATE2);
1105 cur_offset = extent_end; 1137 cur_offset = extent_end;
1106 if (cur_offset > end) 1138 if (cur_offset > end)
1107 break; 1139 break;
@@ -1147,6 +1179,89 @@ static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1147 return ret; 1179 return ret;
1148} 1180}
1149 1181
1182static int btrfs_split_extent_hook(struct inode *inode,
1183 struct extent_state *orig, u64 split)
1184{
1185 struct btrfs_root *root = BTRFS_I(inode)->root;
1186 u64 size;
1187
1188 if (!(orig->state & EXTENT_DELALLOC))
1189 return 0;
1190
1191 size = orig->end - orig->start + 1;
1192 if (size > root->fs_info->max_extent) {
1193 u64 num_extents;
1194 u64 new_size;
1195
1196 new_size = orig->end - split + 1;
1197 num_extents = div64_u64(size + root->fs_info->max_extent - 1,
1198 root->fs_info->max_extent);
1199
1200 /*
1201 * if we break a large extent up then leave oustanding_extents
1202 * be, since we've already accounted for the large extent.
1203 */
1204 if (div64_u64(new_size + root->fs_info->max_extent - 1,
1205 root->fs_info->max_extent) < num_extents)
1206 return 0;
1207 }
1208
1209 spin_lock(&BTRFS_I(inode)->accounting_lock);
1210 BTRFS_I(inode)->outstanding_extents++;
1211 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1212
1213 return 0;
1214}
1215
1216/*
1217 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1218 * extents so we can keep track of new extents that are just merged onto old
1219 * extents, such as when we are doing sequential writes, so we can properly
1220 * account for the metadata space we'll need.
1221 */
1222static int btrfs_merge_extent_hook(struct inode *inode,
1223 struct extent_state *new,
1224 struct extent_state *other)
1225{
1226 struct btrfs_root *root = BTRFS_I(inode)->root;
1227 u64 new_size, old_size;
1228 u64 num_extents;
1229
1230 /* not delalloc, ignore it */
1231 if (!(other->state & EXTENT_DELALLOC))
1232 return 0;
1233
1234 old_size = other->end - other->start + 1;
1235 if (new->start < other->start)
1236 new_size = other->end - new->start + 1;
1237 else
1238 new_size = new->end - other->start + 1;
1239
1240 /* we're not bigger than the max, unreserve the space and go */
1241 if (new_size <= root->fs_info->max_extent) {
1242 spin_lock(&BTRFS_I(inode)->accounting_lock);
1243 BTRFS_I(inode)->outstanding_extents--;
1244 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1245 return 0;
1246 }
1247
1248 /*
1249 * If we grew by another max_extent, just return, we want to keep that
1250 * reserved amount.
1251 */
1252 num_extents = div64_u64(old_size + root->fs_info->max_extent - 1,
1253 root->fs_info->max_extent);
1254 if (div64_u64(new_size + root->fs_info->max_extent - 1,
1255 root->fs_info->max_extent) > num_extents)
1256 return 0;
1257
1258 spin_lock(&BTRFS_I(inode)->accounting_lock);
1259 BTRFS_I(inode)->outstanding_extents--;
1260 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1261
1262 return 0;
1263}
1264
1150/* 1265/*
1151 * extent_io.c set_bit_hook, used to track delayed allocation 1266 * extent_io.c set_bit_hook, used to track delayed allocation
1152 * bytes in this file, and to maintain the list of inodes that 1267 * bytes in this file, and to maintain the list of inodes that
@@ -1155,6 +1270,7 @@ static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1155static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end, 1270static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1156 unsigned long old, unsigned long bits) 1271 unsigned long old, unsigned long bits)
1157{ 1272{
1273
1158 /* 1274 /*
1159 * set_bit and clear bit hooks normally require _irqsave/restore 1275 * set_bit and clear bit hooks normally require _irqsave/restore
1160 * but in this case, we are only testeing for the DELALLOC 1276 * but in this case, we are only testeing for the DELALLOC
@@ -1162,6 +1278,10 @@ static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1162 */ 1278 */
1163 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { 1279 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1164 struct btrfs_root *root = BTRFS_I(inode)->root; 1280 struct btrfs_root *root = BTRFS_I(inode)->root;
1281
1282 spin_lock(&BTRFS_I(inode)->accounting_lock);
1283 BTRFS_I(inode)->outstanding_extents++;
1284 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1165 btrfs_delalloc_reserve_space(root, inode, end - start + 1); 1285 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1166 spin_lock(&root->fs_info->delalloc_lock); 1286 spin_lock(&root->fs_info->delalloc_lock);
1167 BTRFS_I(inode)->delalloc_bytes += end - start + 1; 1287 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
@@ -1178,22 +1298,31 @@ static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1178/* 1298/*
1179 * extent_io.c clear_bit_hook, see set_bit_hook for why 1299 * extent_io.c clear_bit_hook, see set_bit_hook for why
1180 */ 1300 */
1181static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end, 1301static int btrfs_clear_bit_hook(struct inode *inode,
1182 unsigned long old, unsigned long bits) 1302 struct extent_state *state, unsigned long bits)
1183{ 1303{
1184 /* 1304 /*
1185 * set_bit and clear bit hooks normally require _irqsave/restore 1305 * set_bit and clear bit hooks normally require _irqsave/restore
1186 * but in this case, we are only testeing for the DELALLOC 1306 * but in this case, we are only testeing for the DELALLOC
1187 * bit, which is only set or cleared with irqs on 1307 * bit, which is only set or cleared with irqs on
1188 */ 1308 */
1189 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { 1309 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1190 struct btrfs_root *root = BTRFS_I(inode)->root; 1310 struct btrfs_root *root = BTRFS_I(inode)->root;
1191 1311
1312 if (bits & EXTENT_DO_ACCOUNTING) {
1313 spin_lock(&BTRFS_I(inode)->accounting_lock);
1314 BTRFS_I(inode)->outstanding_extents--;
1315 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1316 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
1317 }
1318
1192 spin_lock(&root->fs_info->delalloc_lock); 1319 spin_lock(&root->fs_info->delalloc_lock);
1193 if (end - start + 1 > root->fs_info->delalloc_bytes) { 1320 if (state->end - state->start + 1 >
1321 root->fs_info->delalloc_bytes) {
1194 printk(KERN_INFO "btrfs warning: delalloc account " 1322 printk(KERN_INFO "btrfs warning: delalloc account "
1195 "%llu %llu\n", 1323 "%llu %llu\n",
1196 (unsigned long long)end - start + 1, 1324 (unsigned long long)
1325 state->end - state->start + 1,
1197 (unsigned long long) 1326 (unsigned long long)
1198 root->fs_info->delalloc_bytes); 1327 root->fs_info->delalloc_bytes);
1199 btrfs_delalloc_free_space(root, inode, (u64)-1); 1328 btrfs_delalloc_free_space(root, inode, (u64)-1);
@@ -1201,9 +1330,12 @@ static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1201 BTRFS_I(inode)->delalloc_bytes = 0; 1330 BTRFS_I(inode)->delalloc_bytes = 0;
1202 } else { 1331 } else {
1203 btrfs_delalloc_free_space(root, inode, 1332 btrfs_delalloc_free_space(root, inode,
1204 end - start + 1); 1333 state->end -
1205 root->fs_info->delalloc_bytes -= end - start + 1; 1334 state->start + 1);
1206 BTRFS_I(inode)->delalloc_bytes -= end - start + 1; 1335 root->fs_info->delalloc_bytes -= state->end -
1336 state->start + 1;
1337 BTRFS_I(inode)->delalloc_bytes -= state->end -
1338 state->start + 1;
1207 } 1339 }
1208 if (BTRFS_I(inode)->delalloc_bytes == 0 && 1340 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1209 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { 1341 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
@@ -1374,10 +1506,8 @@ again:
1374 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); 1506 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1375 1507
1376 /* already ordered? We're done */ 1508 /* already ordered? We're done */
1377 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, 1509 if (PagePrivate2(page))
1378 EXTENT_ORDERED, 0)) {
1379 goto out; 1510 goto out;
1380 }
1381 1511
1382 ordered = btrfs_lookup_ordered_extent(inode, page_start); 1512 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1383 if (ordered) { 1513 if (ordered) {
@@ -1413,11 +1543,9 @@ static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1413 struct inode *inode = page->mapping->host; 1543 struct inode *inode = page->mapping->host;
1414 struct btrfs_writepage_fixup *fixup; 1544 struct btrfs_writepage_fixup *fixup;
1415 struct btrfs_root *root = BTRFS_I(inode)->root; 1545 struct btrfs_root *root = BTRFS_I(inode)->root;
1416 int ret;
1417 1546
1418 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end, 1547 /* this page is properly in the ordered list */
1419 EXTENT_ORDERED, 0); 1548 if (TestClearPagePrivate2(page))
1420 if (ret)
1421 return 0; 1549 return 0;
1422 1550
1423 if (PageChecked(page)) 1551 if (PageChecked(page))
@@ -1455,9 +1583,19 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1455 BUG_ON(!path); 1583 BUG_ON(!path);
1456 1584
1457 path->leave_spinning = 1; 1585 path->leave_spinning = 1;
1586
1587 /*
1588 * we may be replacing one extent in the tree with another.
1589 * The new extent is pinned in the extent map, and we don't want
1590 * to drop it from the cache until it is completely in the btree.
1591 *
1592 * So, tell btrfs_drop_extents to leave this extent in the cache.
1593 * the caller is expected to unpin it and allow it to be merged
1594 * with the others.
1595 */
1458 ret = btrfs_drop_extents(trans, root, inode, file_pos, 1596 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1459 file_pos + num_bytes, locked_end, 1597 file_pos + num_bytes, locked_end,
1460 file_pos, &hint); 1598 file_pos, &hint, 0);
1461 BUG_ON(ret); 1599 BUG_ON(ret);
1462 1600
1463 ins.objectid = inode->i_ino; 1601 ins.objectid = inode->i_ino;
@@ -1485,7 +1623,6 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1485 btrfs_mark_buffer_dirty(leaf); 1623 btrfs_mark_buffer_dirty(leaf);
1486 1624
1487 inode_add_bytes(inode, num_bytes); 1625 inode_add_bytes(inode, num_bytes);
1488 btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1489 1626
1490 ins.objectid = disk_bytenr; 1627 ins.objectid = disk_bytenr;
1491 ins.offset = disk_num_bytes; 1628 ins.offset = disk_num_bytes;
@@ -1596,6 +1733,9 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1596 ordered_extent->len, 1733 ordered_extent->len,
1597 compressed, 0, 0, 1734 compressed, 0, 0,
1598 BTRFS_FILE_EXTENT_REG); 1735 BTRFS_FILE_EXTENT_REG);
1736 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1737 ordered_extent->file_offset,
1738 ordered_extent->len);
1599 BUG_ON(ret); 1739 BUG_ON(ret);
1600 } 1740 }
1601 unlock_extent(io_tree, ordered_extent->file_offset, 1741 unlock_extent(io_tree, ordered_extent->file_offset,
@@ -1623,6 +1763,7 @@ nocow:
1623static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, 1763static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1624 struct extent_state *state, int uptodate) 1764 struct extent_state *state, int uptodate)
1625{ 1765{
1766 ClearPagePrivate2(page);
1626 return btrfs_finish_ordered_io(page->mapping->host, start, end); 1767 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1627} 1768}
1628 1769
@@ -1669,13 +1810,13 @@ static int btrfs_io_failed_hook(struct bio *failed_bio,
1669 failrec->last_mirror = 0; 1810 failrec->last_mirror = 0;
1670 failrec->bio_flags = 0; 1811 failrec->bio_flags = 0;
1671 1812
1672 spin_lock(&em_tree->lock); 1813 read_lock(&em_tree->lock);
1673 em = lookup_extent_mapping(em_tree, start, failrec->len); 1814 em = lookup_extent_mapping(em_tree, start, failrec->len);
1674 if (em->start > start || em->start + em->len < start) { 1815 if (em->start > start || em->start + em->len < start) {
1675 free_extent_map(em); 1816 free_extent_map(em);
1676 em = NULL; 1817 em = NULL;
1677 } 1818 }
1678 spin_unlock(&em_tree->lock); 1819 read_unlock(&em_tree->lock);
1679 1820
1680 if (!em || IS_ERR(em)) { 1821 if (!em || IS_ERR(em)) {
1681 kfree(failrec); 1822 kfree(failrec);
@@ -1794,7 +1935,7 @@ static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1794 return 0; 1935 return 0;
1795 1936
1796 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && 1937 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1797 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) { 1938 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1798 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, 1939 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1799 GFP_NOFS); 1940 GFP_NOFS);
1800 return 0; 1941 return 0;
@@ -2352,6 +2493,69 @@ static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2352 return ret; 2493 return ret;
2353} 2494}
2354 2495
2496int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2497 struct btrfs_root *root,
2498 struct inode *dir, u64 objectid,
2499 const char *name, int name_len)
2500{
2501 struct btrfs_path *path;
2502 struct extent_buffer *leaf;
2503 struct btrfs_dir_item *di;
2504 struct btrfs_key key;
2505 u64 index;
2506 int ret;
2507
2508 path = btrfs_alloc_path();
2509 if (!path)
2510 return -ENOMEM;
2511
2512 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2513 name, name_len, -1);
2514 BUG_ON(!di || IS_ERR(di));
2515
2516 leaf = path->nodes[0];
2517 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2518 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2519 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2520 BUG_ON(ret);
2521 btrfs_release_path(root, path);
2522
2523 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2524 objectid, root->root_key.objectid,
2525 dir->i_ino, &index, name, name_len);
2526 if (ret < 0) {
2527 BUG_ON(ret != -ENOENT);
2528 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2529 name, name_len);
2530 BUG_ON(!di || IS_ERR(di));
2531
2532 leaf = path->nodes[0];
2533 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2534 btrfs_release_path(root, path);
2535 index = key.offset;
2536 }
2537
2538 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2539 index, name, name_len, -1);
2540 BUG_ON(!di || IS_ERR(di));
2541
2542 leaf = path->nodes[0];
2543 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2544 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2545 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2546 BUG_ON(ret);
2547 btrfs_release_path(root, path);
2548
2549 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2550 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2551 ret = btrfs_update_inode(trans, root, dir);
2552 BUG_ON(ret);
2553 dir->i_sb->s_dirt = 1;
2554
2555 btrfs_free_path(path);
2556 return 0;
2557}
2558
2355static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) 2559static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2356{ 2560{
2357 struct inode *inode = dentry->d_inode; 2561 struct inode *inode = dentry->d_inode;
@@ -2361,29 +2565,31 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2361 struct btrfs_trans_handle *trans; 2565 struct btrfs_trans_handle *trans;
2362 unsigned long nr = 0; 2566 unsigned long nr = 0;
2363 2567
2364 /*
2365 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2366 * the root of a subvolume or snapshot
2367 */
2368 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || 2568 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2369 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) { 2569 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
2370 return -ENOTEMPTY; 2570 return -ENOTEMPTY;
2371 }
2372 2571
2373 trans = btrfs_start_transaction(root, 1); 2572 trans = btrfs_start_transaction(root, 1);
2374 btrfs_set_trans_block_group(trans, dir); 2573 btrfs_set_trans_block_group(trans, dir);
2375 2574
2575 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2576 err = btrfs_unlink_subvol(trans, root, dir,
2577 BTRFS_I(inode)->location.objectid,
2578 dentry->d_name.name,
2579 dentry->d_name.len);
2580 goto out;
2581 }
2582
2376 err = btrfs_orphan_add(trans, inode); 2583 err = btrfs_orphan_add(trans, inode);
2377 if (err) 2584 if (err)
2378 goto fail_trans; 2585 goto out;
2379 2586
2380 /* now the directory is empty */ 2587 /* now the directory is empty */
2381 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, 2588 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2382 dentry->d_name.name, dentry->d_name.len); 2589 dentry->d_name.name, dentry->d_name.len);
2383 if (!err) 2590 if (!err)
2384 btrfs_i_size_write(inode, 0); 2591 btrfs_i_size_write(inode, 0);
2385 2592out:
2386fail_trans:
2387 nr = trans->blocks_used; 2593 nr = trans->blocks_used;
2388 ret = btrfs_end_transaction_throttle(trans, root); 2594 ret = btrfs_end_transaction_throttle(trans, root);
2389 btrfs_btree_balance_dirty(root, nr); 2595 btrfs_btree_balance_dirty(root, nr);
@@ -2864,7 +3070,12 @@ again:
2864 goto again; 3070 goto again;
2865 } 3071 }
2866 3072
2867 btrfs_set_extent_delalloc(inode, page_start, page_end); 3073 ret = btrfs_set_extent_delalloc(inode, page_start, page_end);
3074 if (ret) {
3075 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3076 goto out_unlock;
3077 }
3078
2868 ret = 0; 3079 ret = 0;
2869 if (offset != PAGE_CACHE_SIZE) { 3080 if (offset != PAGE_CACHE_SIZE) {
2870 kaddr = kmap(page); 3081 kaddr = kmap(page);
@@ -2895,15 +3106,11 @@ int btrfs_cont_expand(struct inode *inode, loff_t size)
2895 u64 last_byte; 3106 u64 last_byte;
2896 u64 cur_offset; 3107 u64 cur_offset;
2897 u64 hole_size; 3108 u64 hole_size;
2898 int err; 3109 int err = 0;
2899 3110
2900 if (size <= hole_start) 3111 if (size <= hole_start)
2901 return 0; 3112 return 0;
2902 3113
2903 err = btrfs_check_metadata_free_space(root);
2904 if (err)
2905 return err;
2906
2907 btrfs_truncate_page(inode->i_mapping, inode->i_size); 3114 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2908 3115
2909 while (1) { 3116 while (1) {
@@ -2935,15 +3142,21 @@ int btrfs_cont_expand(struct inode *inode, loff_t size)
2935 cur_offset, 3142 cur_offset,
2936 cur_offset + hole_size, 3143 cur_offset + hole_size,
2937 block_end, 3144 block_end,
2938 cur_offset, &hint_byte); 3145 cur_offset, &hint_byte, 1);
3146 if (err)
3147 break;
3148
3149 err = btrfs_reserve_metadata_space(root, 1);
2939 if (err) 3150 if (err)
2940 break; 3151 break;
3152
2941 err = btrfs_insert_file_extent(trans, root, 3153 err = btrfs_insert_file_extent(trans, root,
2942 inode->i_ino, cur_offset, 0, 3154 inode->i_ino, cur_offset, 0,
2943 0, hole_size, 0, hole_size, 3155 0, hole_size, 0, hole_size,
2944 0, 0, 0); 3156 0, 0, 0);
2945 btrfs_drop_extent_cache(inode, hole_start, 3157 btrfs_drop_extent_cache(inode, hole_start,
2946 last_byte - 1, 0); 3158 last_byte - 1, 0);
3159 btrfs_unreserve_metadata_space(root, 1);
2947 } 3160 }
2948 free_extent_map(em); 3161 free_extent_map(em);
2949 cur_offset = last_byte; 3162 cur_offset = last_byte;
@@ -3003,6 +3216,11 @@ void btrfs_delete_inode(struct inode *inode)
3003 } 3216 }
3004 btrfs_wait_ordered_range(inode, 0, (u64)-1); 3217 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3005 3218
3219 if (inode->i_nlink > 0) {
3220 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3221 goto no_delete;
3222 }
3223
3006 btrfs_i_size_write(inode, 0); 3224 btrfs_i_size_write(inode, 0);
3007 trans = btrfs_join_transaction(root, 1); 3225 trans = btrfs_join_transaction(root, 1);
3008 3226
@@ -3070,29 +3288,67 @@ out_err:
3070 * is kind of like crossing a mount point. 3288 * is kind of like crossing a mount point.
3071 */ 3289 */
3072static int fixup_tree_root_location(struct btrfs_root *root, 3290static int fixup_tree_root_location(struct btrfs_root *root,
3073 struct btrfs_key *location, 3291 struct inode *dir,
3074 struct btrfs_root **sub_root, 3292 struct dentry *dentry,
3075 struct dentry *dentry) 3293 struct btrfs_key *location,
3294 struct btrfs_root **sub_root)
3076{ 3295{
3077 struct btrfs_root_item *ri; 3296 struct btrfs_path *path;
3297 struct btrfs_root *new_root;
3298 struct btrfs_root_ref *ref;
3299 struct extent_buffer *leaf;
3300 int ret;
3301 int err = 0;
3078 3302
3079 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY) 3303 path = btrfs_alloc_path();
3080 return 0; 3304 if (!path) {
3081 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 3305 err = -ENOMEM;
3082 return 0; 3306 goto out;
3307 }
3083 3308
3084 *sub_root = btrfs_read_fs_root(root->fs_info, location, 3309 err = -ENOENT;
3085 dentry->d_name.name, 3310 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3086 dentry->d_name.len); 3311 BTRFS_I(dir)->root->root_key.objectid,
3087 if (IS_ERR(*sub_root)) 3312 location->objectid);
3088 return PTR_ERR(*sub_root); 3313 if (ret) {
3314 if (ret < 0)
3315 err = ret;
3316 goto out;
3317 }
3089 3318
3090 ri = &(*sub_root)->root_item; 3319 leaf = path->nodes[0];
3091 location->objectid = btrfs_root_dirid(ri); 3320 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3092 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); 3321 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3093 location->offset = 0; 3322 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3323 goto out;
3094 3324
3095 return 0; 3325 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3326 (unsigned long)(ref + 1),
3327 dentry->d_name.len);
3328 if (ret)
3329 goto out;
3330
3331 btrfs_release_path(root->fs_info->tree_root, path);
3332
3333 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3334 if (IS_ERR(new_root)) {
3335 err = PTR_ERR(new_root);
3336 goto out;
3337 }
3338
3339 if (btrfs_root_refs(&new_root->root_item) == 0) {
3340 err = -ENOENT;
3341 goto out;
3342 }
3343
3344 *sub_root = new_root;
3345 location->objectid = btrfs_root_dirid(&new_root->root_item);
3346 location->type = BTRFS_INODE_ITEM_KEY;
3347 location->offset = 0;
3348 err = 0;
3349out:
3350 btrfs_free_path(path);
3351 return err;
3096} 3352}
3097 3353
3098static void inode_tree_add(struct inode *inode) 3354static void inode_tree_add(struct inode *inode)
@@ -3101,11 +3357,13 @@ static void inode_tree_add(struct inode *inode)
3101 struct btrfs_inode *entry; 3357 struct btrfs_inode *entry;
3102 struct rb_node **p; 3358 struct rb_node **p;
3103 struct rb_node *parent; 3359 struct rb_node *parent;
3104
3105again: 3360again:
3106 p = &root->inode_tree.rb_node; 3361 p = &root->inode_tree.rb_node;
3107 parent = NULL; 3362 parent = NULL;
3108 3363
3364 if (hlist_unhashed(&inode->i_hash))
3365 return;
3366
3109 spin_lock(&root->inode_lock); 3367 spin_lock(&root->inode_lock);
3110 while (*p) { 3368 while (*p) {
3111 parent = *p; 3369 parent = *p;
@@ -3132,13 +3390,87 @@ again:
3132static void inode_tree_del(struct inode *inode) 3390static void inode_tree_del(struct inode *inode)
3133{ 3391{
3134 struct btrfs_root *root = BTRFS_I(inode)->root; 3392 struct btrfs_root *root = BTRFS_I(inode)->root;
3393 int empty = 0;
3135 3394
3136 spin_lock(&root->inode_lock); 3395 spin_lock(&root->inode_lock);
3137 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { 3396 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3138 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); 3397 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3139 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 3398 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3399 empty = RB_EMPTY_ROOT(&root->inode_tree);
3140 } 3400 }
3141 spin_unlock(&root->inode_lock); 3401 spin_unlock(&root->inode_lock);
3402
3403 if (empty && btrfs_root_refs(&root->root_item) == 0) {
3404 synchronize_srcu(&root->fs_info->subvol_srcu);
3405 spin_lock(&root->inode_lock);
3406 empty = RB_EMPTY_ROOT(&root->inode_tree);
3407 spin_unlock(&root->inode_lock);
3408 if (empty)
3409 btrfs_add_dead_root(root);
3410 }
3411}
3412
3413int btrfs_invalidate_inodes(struct btrfs_root *root)
3414{
3415 struct rb_node *node;
3416 struct rb_node *prev;
3417 struct btrfs_inode *entry;
3418 struct inode *inode;
3419 u64 objectid = 0;
3420
3421 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3422
3423 spin_lock(&root->inode_lock);
3424again:
3425 node = root->inode_tree.rb_node;
3426 prev = NULL;
3427 while (node) {
3428 prev = node;
3429 entry = rb_entry(node, struct btrfs_inode, rb_node);
3430
3431 if (objectid < entry->vfs_inode.i_ino)
3432 node = node->rb_left;
3433 else if (objectid > entry->vfs_inode.i_ino)
3434 node = node->rb_right;
3435 else
3436 break;
3437 }
3438 if (!node) {
3439 while (prev) {
3440 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3441 if (objectid <= entry->vfs_inode.i_ino) {
3442 node = prev;
3443 break;
3444 }
3445 prev = rb_next(prev);
3446 }
3447 }
3448 while (node) {
3449 entry = rb_entry(node, struct btrfs_inode, rb_node);
3450 objectid = entry->vfs_inode.i_ino + 1;
3451 inode = igrab(&entry->vfs_inode);
3452 if (inode) {
3453 spin_unlock(&root->inode_lock);
3454 if (atomic_read(&inode->i_count) > 1)
3455 d_prune_aliases(inode);
3456 /*
3457 * btrfs_drop_inode will remove it from
3458 * the inode cache when its usage count
3459 * hits zero.
3460 */
3461 iput(inode);
3462 cond_resched();
3463 spin_lock(&root->inode_lock);
3464 goto again;
3465 }
3466
3467 if (cond_resched_lock(&root->inode_lock))
3468 goto again;
3469
3470 node = rb_next(node);
3471 }
3472 spin_unlock(&root->inode_lock);
3473 return 0;
3142} 3474}
3143 3475
3144static noinline void init_btrfs_i(struct inode *inode) 3476static noinline void init_btrfs_i(struct inode *inode)
@@ -3225,15 +3557,41 @@ struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3225 return inode; 3557 return inode;
3226} 3558}
3227 3559
3560static struct inode *new_simple_dir(struct super_block *s,
3561 struct btrfs_key *key,
3562 struct btrfs_root *root)
3563{
3564 struct inode *inode = new_inode(s);
3565
3566 if (!inode)
3567 return ERR_PTR(-ENOMEM);
3568
3569 init_btrfs_i(inode);
3570
3571 BTRFS_I(inode)->root = root;
3572 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3573 BTRFS_I(inode)->dummy_inode = 1;
3574
3575 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3576 inode->i_op = &simple_dir_inode_operations;
3577 inode->i_fop = &simple_dir_operations;
3578 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3579 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3580
3581 return inode;
3582}
3583
3228struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) 3584struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3229{ 3585{
3230 struct inode *inode; 3586 struct inode *inode;
3231 struct btrfs_inode *bi = BTRFS_I(dir); 3587 struct btrfs_root *root = BTRFS_I(dir)->root;
3232 struct btrfs_root *root = bi->root;
3233 struct btrfs_root *sub_root = root; 3588 struct btrfs_root *sub_root = root;
3234 struct btrfs_key location; 3589 struct btrfs_key location;
3590 int index;
3235 int ret; 3591 int ret;
3236 3592
3593 dentry->d_op = &btrfs_dentry_operations;
3594
3237 if (dentry->d_name.len > BTRFS_NAME_LEN) 3595 if (dentry->d_name.len > BTRFS_NAME_LEN)
3238 return ERR_PTR(-ENAMETOOLONG); 3596 return ERR_PTR(-ENAMETOOLONG);
3239 3597
@@ -3242,29 +3600,52 @@ struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3242 if (ret < 0) 3600 if (ret < 0)
3243 return ERR_PTR(ret); 3601 return ERR_PTR(ret);
3244 3602
3245 inode = NULL; 3603 if (location.objectid == 0)
3246 if (location.objectid) { 3604 return NULL;
3247 ret = fixup_tree_root_location(root, &location, &sub_root, 3605
3248 dentry); 3606 if (location.type == BTRFS_INODE_ITEM_KEY) {
3249 if (ret < 0) 3607 inode = btrfs_iget(dir->i_sb, &location, root);
3250 return ERR_PTR(ret); 3608 return inode;
3251 if (ret > 0) 3609 }
3252 return ERR_PTR(-ENOENT); 3610
3611 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3612
3613 index = srcu_read_lock(&root->fs_info->subvol_srcu);
3614 ret = fixup_tree_root_location(root, dir, dentry,
3615 &location, &sub_root);
3616 if (ret < 0) {
3617 if (ret != -ENOENT)
3618 inode = ERR_PTR(ret);
3619 else
3620 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3621 } else {
3253 inode = btrfs_iget(dir->i_sb, &location, sub_root); 3622 inode = btrfs_iget(dir->i_sb, &location, sub_root);
3254 if (IS_ERR(inode))
3255 return ERR_CAST(inode);
3256 } 3623 }
3624 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3625
3257 return inode; 3626 return inode;
3258} 3627}
3259 3628
3629static int btrfs_dentry_delete(struct dentry *dentry)
3630{
3631 struct btrfs_root *root;
3632
3633 if (!dentry->d_inode && !IS_ROOT(dentry))
3634 dentry = dentry->d_parent;
3635
3636 if (dentry->d_inode) {
3637 root = BTRFS_I(dentry->d_inode)->root;
3638 if (btrfs_root_refs(&root->root_item) == 0)
3639 return 1;
3640 }
3641 return 0;
3642}
3643
3260static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, 3644static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3261 struct nameidata *nd) 3645 struct nameidata *nd)
3262{ 3646{
3263 struct inode *inode; 3647 struct inode *inode;
3264 3648
3265 if (dentry->d_name.len > BTRFS_NAME_LEN)
3266 return ERR_PTR(-ENAMETOOLONG);
3267
3268 inode = btrfs_lookup_dentry(dir, dentry); 3649 inode = btrfs_lookup_dentry(dir, dentry);
3269 if (IS_ERR(inode)) 3650 if (IS_ERR(inode))
3270 return ERR_CAST(inode); 3651 return ERR_CAST(inode);
@@ -3603,9 +3984,6 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3603 if (ret != 0) 3984 if (ret != 0)
3604 goto fail; 3985 goto fail;
3605 3986
3606 if (objectid > root->highest_inode)
3607 root->highest_inode = objectid;
3608
3609 inode->i_uid = current_fsuid(); 3987 inode->i_uid = current_fsuid();
3610 3988
3611 if (dir && (dir->i_mode & S_ISGID)) { 3989 if (dir && (dir->i_mode & S_ISGID)) {
@@ -3673,26 +4051,35 @@ int btrfs_add_link(struct btrfs_trans_handle *trans,
3673 struct inode *parent_inode, struct inode *inode, 4051 struct inode *parent_inode, struct inode *inode,
3674 const char *name, int name_len, int add_backref, u64 index) 4052 const char *name, int name_len, int add_backref, u64 index)
3675{ 4053{
3676 int ret; 4054 int ret = 0;
3677 struct btrfs_key key; 4055 struct btrfs_key key;
3678 struct btrfs_root *root = BTRFS_I(parent_inode)->root; 4056 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3679 4057
3680 key.objectid = inode->i_ino; 4058 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
3681 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); 4059 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
3682 key.offset = 0; 4060 } else {
4061 key.objectid = inode->i_ino;
4062 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4063 key.offset = 0;
4064 }
4065
4066 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4067 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4068 key.objectid, root->root_key.objectid,
4069 parent_inode->i_ino,
4070 index, name, name_len);
4071 } else if (add_backref) {
4072 ret = btrfs_insert_inode_ref(trans, root,
4073 name, name_len, inode->i_ino,
4074 parent_inode->i_ino, index);
4075 }
3683 4076
3684 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3685 parent_inode->i_ino,
3686 &key, btrfs_inode_type(inode),
3687 index);
3688 if (ret == 0) { 4077 if (ret == 0) {
3689 if (add_backref) { 4078 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3690 ret = btrfs_insert_inode_ref(trans, root, 4079 parent_inode->i_ino, &key,
3691 name, name_len, 4080 btrfs_inode_type(inode), index);
3692 inode->i_ino, 4081 BUG_ON(ret);
3693 parent_inode->i_ino, 4082
3694 index);
3695 }
3696 btrfs_i_size_write(parent_inode, parent_inode->i_size + 4083 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3697 name_len * 2); 4084 name_len * 2);
3698 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; 4085 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
@@ -3732,11 +4119,18 @@ static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3732 if (!new_valid_dev(rdev)) 4119 if (!new_valid_dev(rdev))
3733 return -EINVAL; 4120 return -EINVAL;
3734 4121
3735 err = btrfs_check_metadata_free_space(root); 4122 /*
4123 * 2 for inode item and ref
4124 * 2 for dir items
4125 * 1 for xattr if selinux is on
4126 */
4127 err = btrfs_reserve_metadata_space(root, 5);
3736 if (err) 4128 if (err)
3737 goto fail; 4129 return err;
3738 4130
3739 trans = btrfs_start_transaction(root, 1); 4131 trans = btrfs_start_transaction(root, 1);
4132 if (!trans)
4133 goto fail;
3740 btrfs_set_trans_block_group(trans, dir); 4134 btrfs_set_trans_block_group(trans, dir);
3741 4135
3742 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); 4136 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
@@ -3774,6 +4168,7 @@ out_unlock:
3774 nr = trans->blocks_used; 4168 nr = trans->blocks_used;
3775 btrfs_end_transaction_throttle(trans, root); 4169 btrfs_end_transaction_throttle(trans, root);
3776fail: 4170fail:
4171 btrfs_unreserve_metadata_space(root, 5);
3777 if (drop_inode) { 4172 if (drop_inode) {
3778 inode_dec_link_count(inode); 4173 inode_dec_link_count(inode);
3779 iput(inode); 4174 iput(inode);
@@ -3794,10 +4189,18 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
3794 u64 objectid; 4189 u64 objectid;
3795 u64 index = 0; 4190 u64 index = 0;
3796 4191
3797 err = btrfs_check_metadata_free_space(root); 4192 /*
4193 * 2 for inode item and ref
4194 * 2 for dir items
4195 * 1 for xattr if selinux is on
4196 */
4197 err = btrfs_reserve_metadata_space(root, 5);
3798 if (err) 4198 if (err)
3799 goto fail; 4199 return err;
4200
3800 trans = btrfs_start_transaction(root, 1); 4201 trans = btrfs_start_transaction(root, 1);
4202 if (!trans)
4203 goto fail;
3801 btrfs_set_trans_block_group(trans, dir); 4204 btrfs_set_trans_block_group(trans, dir);
3802 4205
3803 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); 4206 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
@@ -3838,6 +4241,7 @@ out_unlock:
3838 nr = trans->blocks_used; 4241 nr = trans->blocks_used;
3839 btrfs_end_transaction_throttle(trans, root); 4242 btrfs_end_transaction_throttle(trans, root);
3840fail: 4243fail:
4244 btrfs_unreserve_metadata_space(root, 5);
3841 if (drop_inode) { 4245 if (drop_inode) {
3842 inode_dec_link_count(inode); 4246 inode_dec_link_count(inode);
3843 iput(inode); 4247 iput(inode);
@@ -3860,10 +4264,16 @@ static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3860 if (inode->i_nlink == 0) 4264 if (inode->i_nlink == 0)
3861 return -ENOENT; 4265 return -ENOENT;
3862 4266
3863 btrfs_inc_nlink(inode); 4267 /*
3864 err = btrfs_check_metadata_free_space(root); 4268 * 1 item for inode ref
4269 * 2 items for dir items
4270 */
4271 err = btrfs_reserve_metadata_space(root, 3);
3865 if (err) 4272 if (err)
3866 goto fail; 4273 return err;
4274
4275 btrfs_inc_nlink(inode);
4276
3867 err = btrfs_set_inode_index(dir, &index); 4277 err = btrfs_set_inode_index(dir, &index);
3868 if (err) 4278 if (err)
3869 goto fail; 4279 goto fail;
@@ -3875,20 +4285,19 @@ static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3875 4285
3876 err = btrfs_add_nondir(trans, dentry, inode, 1, index); 4286 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3877 4287
3878 if (err) 4288 if (err) {
3879 drop_inode = 1;
3880
3881 btrfs_update_inode_block_group(trans, dir);
3882 err = btrfs_update_inode(trans, root, inode);
3883
3884 if (err)
3885 drop_inode = 1; 4289 drop_inode = 1;
4290 } else {
4291 btrfs_update_inode_block_group(trans, dir);
4292 err = btrfs_update_inode(trans, root, inode);
4293 BUG_ON(err);
4294 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
4295 }
3886 4296
3887 nr = trans->blocks_used; 4297 nr = trans->blocks_used;
3888
3889 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
3890 btrfs_end_transaction_throttle(trans, root); 4298 btrfs_end_transaction_throttle(trans, root);
3891fail: 4299fail:
4300 btrfs_unreserve_metadata_space(root, 3);
3892 if (drop_inode) { 4301 if (drop_inode) {
3893 inode_dec_link_count(inode); 4302 inode_dec_link_count(inode);
3894 iput(inode); 4303 iput(inode);
@@ -3908,17 +4317,21 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3908 u64 index = 0; 4317 u64 index = 0;
3909 unsigned long nr = 1; 4318 unsigned long nr = 1;
3910 4319
3911 err = btrfs_check_metadata_free_space(root); 4320 /*
4321 * 2 items for inode and ref
4322 * 2 items for dir items
4323 * 1 for xattr if selinux is on
4324 */
4325 err = btrfs_reserve_metadata_space(root, 5);
3912 if (err) 4326 if (err)
3913 goto out_unlock; 4327 return err;
3914 4328
3915 trans = btrfs_start_transaction(root, 1); 4329 trans = btrfs_start_transaction(root, 1);
3916 btrfs_set_trans_block_group(trans, dir); 4330 if (!trans) {
3917 4331 err = -ENOMEM;
3918 if (IS_ERR(trans)) {
3919 err = PTR_ERR(trans);
3920 goto out_unlock; 4332 goto out_unlock;
3921 } 4333 }
4334 btrfs_set_trans_block_group(trans, dir);
3922 4335
3923 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); 4336 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3924 if (err) { 4337 if (err) {
@@ -3967,6 +4380,7 @@ out_fail:
3967 btrfs_end_transaction_throttle(trans, root); 4380 btrfs_end_transaction_throttle(trans, root);
3968 4381
3969out_unlock: 4382out_unlock:
4383 btrfs_unreserve_metadata_space(root, 5);
3970 if (drop_on_err) 4384 if (drop_on_err)
3971 iput(inode); 4385 iput(inode);
3972 btrfs_btree_balance_dirty(root, nr); 4386 btrfs_btree_balance_dirty(root, nr);
@@ -4064,11 +4478,11 @@ struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4064 int compressed; 4478 int compressed;
4065 4479
4066again: 4480again:
4067 spin_lock(&em_tree->lock); 4481 read_lock(&em_tree->lock);
4068 em = lookup_extent_mapping(em_tree, start, len); 4482 em = lookup_extent_mapping(em_tree, start, len);
4069 if (em) 4483 if (em)
4070 em->bdev = root->fs_info->fs_devices->latest_bdev; 4484 em->bdev = root->fs_info->fs_devices->latest_bdev;
4071 spin_unlock(&em_tree->lock); 4485 read_unlock(&em_tree->lock);
4072 4486
4073 if (em) { 4487 if (em) {
4074 if (em->start > start || em->start + em->len <= start) 4488 if (em->start > start || em->start + em->len <= start)
@@ -4215,6 +4629,11 @@ again:
4215 map = kmap(page); 4629 map = kmap(page);
4216 read_extent_buffer(leaf, map + pg_offset, ptr, 4630 read_extent_buffer(leaf, map + pg_offset, ptr,
4217 copy_size); 4631 copy_size);
4632 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
4633 memset(map + pg_offset + copy_size, 0,
4634 PAGE_CACHE_SIZE - pg_offset -
4635 copy_size);
4636 }
4218 kunmap(page); 4637 kunmap(page);
4219 } 4638 }
4220 flush_dcache_page(page); 4639 flush_dcache_page(page);
@@ -4259,7 +4678,7 @@ insert:
4259 } 4678 }
4260 4679
4261 err = 0; 4680 err = 0;
4262 spin_lock(&em_tree->lock); 4681 write_lock(&em_tree->lock);
4263 ret = add_extent_mapping(em_tree, em); 4682 ret = add_extent_mapping(em_tree, em);
4264 /* it is possible that someone inserted the extent into the tree 4683 /* it is possible that someone inserted the extent into the tree
4265 * while we had the lock dropped. It is also possible that 4684 * while we had the lock dropped. It is also possible that
@@ -4299,7 +4718,7 @@ insert:
4299 err = 0; 4718 err = 0;
4300 } 4719 }
4301 } 4720 }
4302 spin_unlock(&em_tree->lock); 4721 write_unlock(&em_tree->lock);
4303out: 4722out:
4304 if (path) 4723 if (path)
4305 btrfs_free_path(path); 4724 btrfs_free_path(path);
@@ -4398,13 +4817,21 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4398 u64 page_start = page_offset(page); 4817 u64 page_start = page_offset(page);
4399 u64 page_end = page_start + PAGE_CACHE_SIZE - 1; 4818 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4400 4819
4820
4821 /*
4822 * we have the page locked, so new writeback can't start,
4823 * and the dirty bit won't be cleared while we are here.
4824 *
4825 * Wait for IO on this page so that we can safely clear
4826 * the PagePrivate2 bit and do ordered accounting
4827 */
4401 wait_on_page_writeback(page); 4828 wait_on_page_writeback(page);
4829
4402 tree = &BTRFS_I(page->mapping->host)->io_tree; 4830 tree = &BTRFS_I(page->mapping->host)->io_tree;
4403 if (offset) { 4831 if (offset) {
4404 btrfs_releasepage(page, GFP_NOFS); 4832 btrfs_releasepage(page, GFP_NOFS);
4405 return; 4833 return;
4406 } 4834 }
4407
4408 lock_extent(tree, page_start, page_end, GFP_NOFS); 4835 lock_extent(tree, page_start, page_end, GFP_NOFS);
4409 ordered = btrfs_lookup_ordered_extent(page->mapping->host, 4836 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4410 page_offset(page)); 4837 page_offset(page));
@@ -4415,16 +4842,22 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4415 */ 4842 */
4416 clear_extent_bit(tree, page_start, page_end, 4843 clear_extent_bit(tree, page_start, page_end,
4417 EXTENT_DIRTY | EXTENT_DELALLOC | 4844 EXTENT_DIRTY | EXTENT_DELALLOC |
4418 EXTENT_LOCKED, 1, 0, GFP_NOFS); 4845 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
4419 btrfs_finish_ordered_io(page->mapping->host, 4846 NULL, GFP_NOFS);
4420 page_start, page_end); 4847 /*
4848 * whoever cleared the private bit is responsible
4849 * for the finish_ordered_io
4850 */
4851 if (TestClearPagePrivate2(page)) {
4852 btrfs_finish_ordered_io(page->mapping->host,
4853 page_start, page_end);
4854 }
4421 btrfs_put_ordered_extent(ordered); 4855 btrfs_put_ordered_extent(ordered);
4422 lock_extent(tree, page_start, page_end, GFP_NOFS); 4856 lock_extent(tree, page_start, page_end, GFP_NOFS);
4423 } 4857 }
4424 clear_extent_bit(tree, page_start, page_end, 4858 clear_extent_bit(tree, page_start, page_end,
4425 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | 4859 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4426 EXTENT_ORDERED, 4860 EXTENT_DO_ACCOUNTING, 1, 1, NULL, GFP_NOFS);
4427 1, 1, GFP_NOFS);
4428 __btrfs_releasepage(page, GFP_NOFS); 4861 __btrfs_releasepage(page, GFP_NOFS);
4429 4862
4430 ClearPageChecked(page); 4863 ClearPageChecked(page);
@@ -4473,6 +4906,13 @@ int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4473 goto out; 4906 goto out;
4474 } 4907 }
4475 4908
4909 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
4910 if (ret) {
4911 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4912 ret = VM_FAULT_SIGBUS;
4913 goto out;
4914 }
4915
4476 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ 4916 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
4477again: 4917again:
4478 lock_page(page); 4918 lock_page(page);
@@ -4504,7 +4944,24 @@ again:
4504 goto again; 4944 goto again;
4505 } 4945 }
4506 4946
4507 btrfs_set_extent_delalloc(inode, page_start, page_end); 4947 /*
4948 * XXX - page_mkwrite gets called every time the page is dirtied, even
4949 * if it was already dirty, so for space accounting reasons we need to
4950 * clear any delalloc bits for the range we are fixing to save. There
4951 * is probably a better way to do this, but for now keep consistent with
4952 * prepare_pages in the normal write path.
4953 */
4954 clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
4955 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
4956 GFP_NOFS);
4957
4958 ret = btrfs_set_extent_delalloc(inode, page_start, page_end);
4959 if (ret) {
4960 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4961 ret = VM_FAULT_SIGBUS;
4962 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4963 goto out_unlock;
4964 }
4508 ret = 0; 4965 ret = 0;
4509 4966
4510 /* page is wholly or partially inside EOF */ 4967 /* page is wholly or partially inside EOF */
@@ -4521,11 +4978,15 @@ again:
4521 } 4978 }
4522 ClearPageChecked(page); 4979 ClearPageChecked(page);
4523 set_page_dirty(page); 4980 set_page_dirty(page);
4981 SetPageUptodate(page);
4524 4982
4525 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1; 4983 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
4526 unlock_extent(io_tree, page_start, page_end, GFP_NOFS); 4984 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4527 4985
4528out_unlock: 4986out_unlock:
4987 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
4988 if (!ret)
4989 return VM_FAULT_LOCKED;
4529 unlock_page(page); 4990 unlock_page(page);
4530out: 4991out:
4531 return ret; 4992 return ret;
@@ -4594,11 +5055,11 @@ out:
4594 * create a new subvolume directory/inode (helper for the ioctl). 5055 * create a new subvolume directory/inode (helper for the ioctl).
4595 */ 5056 */
4596int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, 5057int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4597 struct btrfs_root *new_root, struct dentry *dentry, 5058 struct btrfs_root *new_root,
4598 u64 new_dirid, u64 alloc_hint) 5059 u64 new_dirid, u64 alloc_hint)
4599{ 5060{
4600 struct inode *inode; 5061 struct inode *inode;
4601 int error; 5062 int err;
4602 u64 index = 0; 5063 u64 index = 0;
4603 5064
4604 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, 5065 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
@@ -4611,11 +5072,10 @@ int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4611 inode->i_nlink = 1; 5072 inode->i_nlink = 1;
4612 btrfs_i_size_write(inode, 0); 5073 btrfs_i_size_write(inode, 0);
4613 5074
4614 error = btrfs_update_inode(trans, new_root, inode); 5075 err = btrfs_update_inode(trans, new_root, inode);
4615 if (error) 5076 BUG_ON(err);
4616 return error;
4617 5077
4618 d_instantiate(dentry, inode); 5078 iput(inode);
4619 return 0; 5079 return 0;
4620} 5080}
4621 5081
@@ -4641,6 +5101,9 @@ struct inode *btrfs_alloc_inode(struct super_block *sb)
4641 return NULL; 5101 return NULL;
4642 ei->last_trans = 0; 5102 ei->last_trans = 0;
4643 ei->logged_trans = 0; 5103 ei->logged_trans = 0;
5104 ei->outstanding_extents = 0;
5105 ei->reserved_extents = 0;
5106 spin_lock_init(&ei->accounting_lock);
4644 btrfs_ordered_inode_tree_init(&ei->ordered_tree); 5107 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4645 INIT_LIST_HEAD(&ei->i_orphan); 5108 INIT_LIST_HEAD(&ei->i_orphan);
4646 INIT_LIST_HEAD(&ei->ordered_operations); 5109 INIT_LIST_HEAD(&ei->ordered_operations);
@@ -4693,6 +5156,16 @@ void btrfs_destroy_inode(struct inode *inode)
4693 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); 5156 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4694} 5157}
4695 5158
5159void btrfs_drop_inode(struct inode *inode)
5160{
5161 struct btrfs_root *root = BTRFS_I(inode)->root;
5162
5163 if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0)
5164 generic_delete_inode(inode);
5165 else
5166 generic_drop_inode(inode);
5167}
5168
4696static void init_once(void *foo) 5169static void init_once(void *foo)
4697{ 5170{
4698 struct btrfs_inode *ei = (struct btrfs_inode *) foo; 5171 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
@@ -4761,31 +5234,37 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4761{ 5234{
4762 struct btrfs_trans_handle *trans; 5235 struct btrfs_trans_handle *trans;
4763 struct btrfs_root *root = BTRFS_I(old_dir)->root; 5236 struct btrfs_root *root = BTRFS_I(old_dir)->root;
5237 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
4764 struct inode *new_inode = new_dentry->d_inode; 5238 struct inode *new_inode = new_dentry->d_inode;
4765 struct inode *old_inode = old_dentry->d_inode; 5239 struct inode *old_inode = old_dentry->d_inode;
4766 struct timespec ctime = CURRENT_TIME; 5240 struct timespec ctime = CURRENT_TIME;
4767 u64 index = 0; 5241 u64 index = 0;
5242 u64 root_objectid;
4768 int ret; 5243 int ret;
4769 5244
4770 /* we're not allowed to rename between subvolumes */ 5245 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4771 if (BTRFS_I(old_inode)->root->root_key.objectid != 5246 return -EPERM;
4772 BTRFS_I(new_dir)->root->root_key.objectid) 5247
5248 /* we only allow rename subvolume link between subvolumes */
5249 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
4773 return -EXDEV; 5250 return -EXDEV;
4774 5251
5252 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
5253 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
5254 return -ENOTEMPTY;
5255
4775 if (S_ISDIR(old_inode->i_mode) && new_inode && 5256 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4776 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) { 5257 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4777 return -ENOTEMPTY; 5258 return -ENOTEMPTY;
4778 }
4779 5259
4780 /* to rename a snapshot or subvolume, we need to juggle the 5260 /*
4781 * backrefs. This isn't coded yet 5261 * 2 items for dir items
5262 * 1 item for orphan entry
5263 * 1 item for ref
4782 */ 5264 */
4783 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) 5265 ret = btrfs_reserve_metadata_space(root, 4);
4784 return -EXDEV;
4785
4786 ret = btrfs_check_metadata_free_space(root);
4787 if (ret) 5266 if (ret)
4788 goto out_unlock; 5267 return ret;
4789 5268
4790 /* 5269 /*
4791 * we're using rename to replace one file with another. 5270 * we're using rename to replace one file with another.
@@ -4796,8 +5275,40 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4796 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) 5275 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
4797 filemap_flush(old_inode->i_mapping); 5276 filemap_flush(old_inode->i_mapping);
4798 5277
5278 /* close the racy window with snapshot create/destroy ioctl */
5279 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5280 down_read(&root->fs_info->subvol_sem);
5281
4799 trans = btrfs_start_transaction(root, 1); 5282 trans = btrfs_start_transaction(root, 1);
5283 btrfs_set_trans_block_group(trans, new_dir);
5284
5285 if (dest != root)
5286 btrfs_record_root_in_trans(trans, dest);
5287
5288 ret = btrfs_set_inode_index(new_dir, &index);
5289 if (ret)
5290 goto out_fail;
4800 5291
5292 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5293 /* force full log commit if subvolume involved. */
5294 root->fs_info->last_trans_log_full_commit = trans->transid;
5295 } else {
5296 ret = btrfs_insert_inode_ref(trans, dest,
5297 new_dentry->d_name.name,
5298 new_dentry->d_name.len,
5299 old_inode->i_ino,
5300 new_dir->i_ino, index);
5301 if (ret)
5302 goto out_fail;
5303 /*
5304 * this is an ugly little race, but the rename is required
5305 * to make sure that if we crash, the inode is either at the
5306 * old name or the new one. pinning the log transaction lets
5307 * us make sure we don't allow a log commit to come in after
5308 * we unlink the name but before we add the new name back in.
5309 */
5310 btrfs_pin_log_trans(root);
5311 }
4801 /* 5312 /*
4802 * make sure the inode gets flushed if it is replacing 5313 * make sure the inode gets flushed if it is replacing
4803 * something. 5314 * something.
@@ -4807,18 +5318,6 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4807 btrfs_add_ordered_operation(trans, root, old_inode); 5318 btrfs_add_ordered_operation(trans, root, old_inode);
4808 } 5319 }
4809 5320
4810 /*
4811 * this is an ugly little race, but the rename is required to make
4812 * sure that if we crash, the inode is either at the old name
4813 * or the new one. pinning the log transaction lets us make sure
4814 * we don't allow a log commit to come in after we unlink the
4815 * name but before we add the new name back in.
4816 */
4817 btrfs_pin_log_trans(root);
4818
4819 btrfs_set_trans_block_group(trans, new_dir);
4820
4821 btrfs_inc_nlink(old_dentry->d_inode);
4822 old_dir->i_ctime = old_dir->i_mtime = ctime; 5321 old_dir->i_ctime = old_dir->i_mtime = ctime;
4823 new_dir->i_ctime = new_dir->i_mtime = ctime; 5322 new_dir->i_ctime = new_dir->i_mtime = ctime;
4824 old_inode->i_ctime = ctime; 5323 old_inode->i_ctime = ctime;
@@ -4826,47 +5325,60 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4826 if (old_dentry->d_parent != new_dentry->d_parent) 5325 if (old_dentry->d_parent != new_dentry->d_parent)
4827 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); 5326 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
4828 5327
4829 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode, 5328 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4830 old_dentry->d_name.name, 5329 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
4831 old_dentry->d_name.len); 5330 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
4832 if (ret) 5331 old_dentry->d_name.name,
4833 goto out_fail; 5332 old_dentry->d_name.len);
5333 } else {
5334 btrfs_inc_nlink(old_dentry->d_inode);
5335 ret = btrfs_unlink_inode(trans, root, old_dir,
5336 old_dentry->d_inode,
5337 old_dentry->d_name.name,
5338 old_dentry->d_name.len);
5339 }
5340 BUG_ON(ret);
4834 5341
4835 if (new_inode) { 5342 if (new_inode) {
4836 new_inode->i_ctime = CURRENT_TIME; 5343 new_inode->i_ctime = CURRENT_TIME;
4837 ret = btrfs_unlink_inode(trans, root, new_dir, 5344 if (unlikely(new_inode->i_ino ==
4838 new_dentry->d_inode, 5345 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4839 new_dentry->d_name.name, 5346 root_objectid = BTRFS_I(new_inode)->location.objectid;
4840 new_dentry->d_name.len); 5347 ret = btrfs_unlink_subvol(trans, dest, new_dir,
4841 if (ret) 5348 root_objectid,
4842 goto out_fail; 5349 new_dentry->d_name.name,
5350 new_dentry->d_name.len);
5351 BUG_ON(new_inode->i_nlink == 0);
5352 } else {
5353 ret = btrfs_unlink_inode(trans, dest, new_dir,
5354 new_dentry->d_inode,
5355 new_dentry->d_name.name,
5356 new_dentry->d_name.len);
5357 }
5358 BUG_ON(ret);
4843 if (new_inode->i_nlink == 0) { 5359 if (new_inode->i_nlink == 0) {
4844 ret = btrfs_orphan_add(trans, new_dentry->d_inode); 5360 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4845 if (ret) 5361 BUG_ON(ret);
4846 goto out_fail;
4847 } 5362 }
4848
4849 } 5363 }
4850 ret = btrfs_set_inode_index(new_dir, &index);
4851 if (ret)
4852 goto out_fail;
4853 5364
4854 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode, 5365 ret = btrfs_add_link(trans, new_dir, old_inode,
4855 old_inode, new_dentry->d_name.name, 5366 new_dentry->d_name.name,
4856 new_dentry->d_name.len, 1, index); 5367 new_dentry->d_name.len, 0, index);
4857 if (ret) 5368 BUG_ON(ret);
4858 goto out_fail;
4859 5369
4860 btrfs_log_new_name(trans, old_inode, old_dir, 5370 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
4861 new_dentry->d_parent); 5371 btrfs_log_new_name(trans, old_inode, old_dir,
5372 new_dentry->d_parent);
5373 btrfs_end_log_trans(root);
5374 }
4862out_fail: 5375out_fail:
4863
4864 /* this btrfs_end_log_trans just allows the current
4865 * log-sub transaction to complete
4866 */
4867 btrfs_end_log_trans(root);
4868 btrfs_end_transaction_throttle(trans, root); 5376 btrfs_end_transaction_throttle(trans, root);
4869out_unlock: 5377
5378 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5379 up_read(&root->fs_info->subvol_sem);
5380
5381 btrfs_unreserve_metadata_space(root, 4);
4870 return ret; 5382 return ret;
4871} 5383}
4872 5384
@@ -4938,11 +5450,18 @@ static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4938 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) 5450 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4939 return -ENAMETOOLONG; 5451 return -ENAMETOOLONG;
4940 5452
4941 err = btrfs_check_metadata_free_space(root); 5453 /*
5454 * 2 items for inode item and ref
5455 * 2 items for dir items
5456 * 1 item for xattr if selinux is on
5457 */
5458 err = btrfs_reserve_metadata_space(root, 5);
4942 if (err) 5459 if (err)
4943 goto out_fail; 5460 return err;
4944 5461
4945 trans = btrfs_start_transaction(root, 1); 5462 trans = btrfs_start_transaction(root, 1);
5463 if (!trans)
5464 goto out_fail;
4946 btrfs_set_trans_block_group(trans, dir); 5465 btrfs_set_trans_block_group(trans, dir);
4947 5466
4948 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); 5467 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
@@ -5023,6 +5542,7 @@ out_unlock:
5023 nr = trans->blocks_used; 5542 nr = trans->blocks_used;
5024 btrfs_end_transaction_throttle(trans, root); 5543 btrfs_end_transaction_throttle(trans, root);
5025out_fail: 5544out_fail:
5545 btrfs_unreserve_metadata_space(root, 5);
5026 if (drop_inode) { 5546 if (drop_inode) {
5027 inode_dec_link_count(inode); 5547 inode_dec_link_count(inode);
5028 iput(inode); 5548 iput(inode);
@@ -5044,6 +5564,11 @@ static int prealloc_file_range(struct btrfs_trans_handle *trans,
5044 5564
5045 while (num_bytes > 0) { 5565 while (num_bytes > 0) {
5046 alloc_size = min(num_bytes, root->fs_info->max_extent); 5566 alloc_size = min(num_bytes, root->fs_info->max_extent);
5567
5568 ret = btrfs_reserve_metadata_space(root, 1);
5569 if (ret)
5570 goto out;
5571
5047 ret = btrfs_reserve_extent(trans, root, alloc_size, 5572 ret = btrfs_reserve_extent(trans, root, alloc_size,
5048 root->sectorsize, 0, alloc_hint, 5573 root->sectorsize, 0, alloc_hint,
5049 (u64)-1, &ins, 1); 5574 (u64)-1, &ins, 1);
@@ -5058,9 +5583,12 @@ static int prealloc_file_range(struct btrfs_trans_handle *trans,
5058 0, 0, 0, 5583 0, 0, 0,
5059 BTRFS_FILE_EXTENT_PREALLOC); 5584 BTRFS_FILE_EXTENT_PREALLOC);
5060 BUG_ON(ret); 5585 BUG_ON(ret);
5586 btrfs_drop_extent_cache(inode, cur_offset,
5587 cur_offset + ins.offset -1, 0);
5061 num_bytes -= ins.offset; 5588 num_bytes -= ins.offset;
5062 cur_offset += ins.offset; 5589 cur_offset += ins.offset;
5063 alloc_hint = ins.objectid + ins.offset; 5590 alloc_hint = ins.objectid + ins.offset;
5591 btrfs_unreserve_metadata_space(root, 1);
5064 } 5592 }
5065out: 5593out:
5066 if (cur_offset > start) { 5594 if (cur_offset > start) {
@@ -5201,7 +5729,7 @@ static int btrfs_permission(struct inode *inode, int mask)
5201 return generic_permission(inode, mask, btrfs_check_acl); 5729 return generic_permission(inode, mask, btrfs_check_acl);
5202} 5730}
5203 5731
5204static struct inode_operations btrfs_dir_inode_operations = { 5732static const struct inode_operations btrfs_dir_inode_operations = {
5205 .getattr = btrfs_getattr, 5733 .getattr = btrfs_getattr,
5206 .lookup = btrfs_lookup, 5734 .lookup = btrfs_lookup,
5207 .create = btrfs_create, 5735 .create = btrfs_create,
@@ -5219,11 +5747,12 @@ static struct inode_operations btrfs_dir_inode_operations = {
5219 .removexattr = btrfs_removexattr, 5747 .removexattr = btrfs_removexattr,
5220 .permission = btrfs_permission, 5748 .permission = btrfs_permission,
5221}; 5749};
5222static struct inode_operations btrfs_dir_ro_inode_operations = { 5750static const struct inode_operations btrfs_dir_ro_inode_operations = {
5223 .lookup = btrfs_lookup, 5751 .lookup = btrfs_lookup,
5224 .permission = btrfs_permission, 5752 .permission = btrfs_permission,
5225}; 5753};
5226static struct file_operations btrfs_dir_file_operations = { 5754
5755static const struct file_operations btrfs_dir_file_operations = {
5227 .llseek = generic_file_llseek, 5756 .llseek = generic_file_llseek,
5228 .read = generic_read_dir, 5757 .read = generic_read_dir,
5229 .readdir = btrfs_real_readdir, 5758 .readdir = btrfs_real_readdir,
@@ -5245,6 +5774,8 @@ static struct extent_io_ops btrfs_extent_io_ops = {
5245 .readpage_io_failed_hook = btrfs_io_failed_hook, 5774 .readpage_io_failed_hook = btrfs_io_failed_hook,
5246 .set_bit_hook = btrfs_set_bit_hook, 5775 .set_bit_hook = btrfs_set_bit_hook,
5247 .clear_bit_hook = btrfs_clear_bit_hook, 5776 .clear_bit_hook = btrfs_clear_bit_hook,
5777 .merge_extent_hook = btrfs_merge_extent_hook,
5778 .split_extent_hook = btrfs_split_extent_hook,
5248}; 5779};
5249 5780
5250/* 5781/*
@@ -5259,7 +5790,7 @@ static struct extent_io_ops btrfs_extent_io_ops = {
5259 * 5790 *
5260 * For now we're avoiding this by dropping bmap. 5791 * For now we're avoiding this by dropping bmap.
5261 */ 5792 */
5262static struct address_space_operations btrfs_aops = { 5793static const struct address_space_operations btrfs_aops = {
5263 .readpage = btrfs_readpage, 5794 .readpage = btrfs_readpage,
5264 .writepage = btrfs_writepage, 5795 .writepage = btrfs_writepage,
5265 .writepages = btrfs_writepages, 5796 .writepages = btrfs_writepages,
@@ -5269,16 +5800,17 @@ static struct address_space_operations btrfs_aops = {
5269 .invalidatepage = btrfs_invalidatepage, 5800 .invalidatepage = btrfs_invalidatepage,
5270 .releasepage = btrfs_releasepage, 5801 .releasepage = btrfs_releasepage,
5271 .set_page_dirty = btrfs_set_page_dirty, 5802 .set_page_dirty = btrfs_set_page_dirty,
5803 .error_remove_page = generic_error_remove_page,
5272}; 5804};
5273 5805
5274static struct address_space_operations btrfs_symlink_aops = { 5806static const struct address_space_operations btrfs_symlink_aops = {
5275 .readpage = btrfs_readpage, 5807 .readpage = btrfs_readpage,
5276 .writepage = btrfs_writepage, 5808 .writepage = btrfs_writepage,
5277 .invalidatepage = btrfs_invalidatepage, 5809 .invalidatepage = btrfs_invalidatepage,
5278 .releasepage = btrfs_releasepage, 5810 .releasepage = btrfs_releasepage,
5279}; 5811};
5280 5812
5281static struct inode_operations btrfs_file_inode_operations = { 5813static const struct inode_operations btrfs_file_inode_operations = {
5282 .truncate = btrfs_truncate, 5814 .truncate = btrfs_truncate,
5283 .getattr = btrfs_getattr, 5815 .getattr = btrfs_getattr,
5284 .setattr = btrfs_setattr, 5816 .setattr = btrfs_setattr,
@@ -5290,7 +5822,7 @@ static struct inode_operations btrfs_file_inode_operations = {
5290 .fallocate = btrfs_fallocate, 5822 .fallocate = btrfs_fallocate,
5291 .fiemap = btrfs_fiemap, 5823 .fiemap = btrfs_fiemap,
5292}; 5824};
5293static struct inode_operations btrfs_special_inode_operations = { 5825static const struct inode_operations btrfs_special_inode_operations = {
5294 .getattr = btrfs_getattr, 5826 .getattr = btrfs_getattr,
5295 .setattr = btrfs_setattr, 5827 .setattr = btrfs_setattr,
5296 .permission = btrfs_permission, 5828 .permission = btrfs_permission,
@@ -5299,7 +5831,7 @@ static struct inode_operations btrfs_special_inode_operations = {
5299 .listxattr = btrfs_listxattr, 5831 .listxattr = btrfs_listxattr,
5300 .removexattr = btrfs_removexattr, 5832 .removexattr = btrfs_removexattr,
5301}; 5833};
5302static struct inode_operations btrfs_symlink_inode_operations = { 5834static const struct inode_operations btrfs_symlink_inode_operations = {
5303 .readlink = generic_readlink, 5835 .readlink = generic_readlink,
5304 .follow_link = page_follow_link_light, 5836 .follow_link = page_follow_link_light,
5305 .put_link = page_put_link, 5837 .put_link = page_put_link,
@@ -5309,3 +5841,7 @@ static struct inode_operations btrfs_symlink_inode_operations = {
5309 .listxattr = btrfs_listxattr, 5841 .listxattr = btrfs_listxattr,
5310 .removexattr = btrfs_removexattr, 5842 .removexattr = btrfs_removexattr,
5311}; 5843};
5844
5845const struct dentry_operations btrfs_dentry_operations = {
5846 .d_delete = btrfs_dentry_delete,
5847};
diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index bd88f25889f7..cdbb054102b9 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -230,8 +230,8 @@ static noinline int create_subvol(struct btrfs_root *root,
230 struct btrfs_root_item root_item; 230 struct btrfs_root_item root_item;
231 struct btrfs_inode_item *inode_item; 231 struct btrfs_inode_item *inode_item;
232 struct extent_buffer *leaf; 232 struct extent_buffer *leaf;
233 struct btrfs_root *new_root = root; 233 struct btrfs_root *new_root;
234 struct inode *dir; 234 struct inode *dir = dentry->d_parent->d_inode;
235 int ret; 235 int ret;
236 int err; 236 int err;
237 u64 objectid; 237 u64 objectid;
@@ -239,9 +239,15 @@ static noinline int create_subvol(struct btrfs_root *root,
239 u64 index = 0; 239 u64 index = 0;
240 unsigned long nr = 1; 240 unsigned long nr = 1;
241 241
242 ret = btrfs_check_metadata_free_space(root); 242 /*
243 * 1 - inode item
244 * 2 - refs
245 * 1 - root item
246 * 2 - dir items
247 */
248 ret = btrfs_reserve_metadata_space(root, 6);
243 if (ret) 249 if (ret)
244 goto fail_commit; 250 return ret;
245 251
246 trans = btrfs_start_transaction(root, 1); 252 trans = btrfs_start_transaction(root, 1);
247 BUG_ON(!trans); 253 BUG_ON(!trans);
@@ -304,11 +310,17 @@ static noinline int create_subvol(struct btrfs_root *root,
304 if (ret) 310 if (ret)
305 goto fail; 311 goto fail;
306 312
313 key.offset = (u64)-1;
314 new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
315 BUG_ON(IS_ERR(new_root));
316
317 btrfs_record_root_in_trans(trans, new_root);
318
319 ret = btrfs_create_subvol_root(trans, new_root, new_dirid,
320 BTRFS_I(dir)->block_group);
307 /* 321 /*
308 * insert the directory item 322 * insert the directory item
309 */ 323 */
310 key.offset = (u64)-1;
311 dir = dentry->d_parent->d_inode;
312 ret = btrfs_set_inode_index(dir, &index); 324 ret = btrfs_set_inode_index(dir, &index);
313 BUG_ON(ret); 325 BUG_ON(ret);
314 326
@@ -322,43 +334,20 @@ static noinline int create_subvol(struct btrfs_root *root,
322 ret = btrfs_update_inode(trans, root, dir); 334 ret = btrfs_update_inode(trans, root, dir);
323 BUG_ON(ret); 335 BUG_ON(ret);
324 336
325 /* add the backref first */
326 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, 337 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
327 objectid, BTRFS_ROOT_BACKREF_KEY, 338 objectid, root->root_key.objectid,
328 root->root_key.objectid,
329 dir->i_ino, index, name, namelen); 339 dir->i_ino, index, name, namelen);
330 340
331 BUG_ON(ret); 341 BUG_ON(ret);
332 342
333 /* now add the forward ref */ 343 d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
334 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
335 root->root_key.objectid, BTRFS_ROOT_REF_KEY,
336 objectid,
337 dir->i_ino, index, name, namelen);
338
339 BUG_ON(ret);
340
341 ret = btrfs_commit_transaction(trans, root);
342 if (ret)
343 goto fail_commit;
344
345 new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
346 BUG_ON(!new_root);
347
348 trans = btrfs_start_transaction(new_root, 1);
349 BUG_ON(!trans);
350
351 ret = btrfs_create_subvol_root(trans, new_root, dentry, new_dirid,
352 BTRFS_I(dir)->block_group);
353 if (ret)
354 goto fail;
355
356fail: 344fail:
357 nr = trans->blocks_used; 345 nr = trans->blocks_used;
358 err = btrfs_commit_transaction(trans, new_root); 346 err = btrfs_commit_transaction(trans, root);
359 if (err && !ret) 347 if (err && !ret)
360 ret = err; 348 ret = err;
361fail_commit: 349
350 btrfs_unreserve_metadata_space(root, 6);
362 btrfs_btree_balance_dirty(root, nr); 351 btrfs_btree_balance_dirty(root, nr);
363 return ret; 352 return ret;
364} 353}
@@ -375,19 +364,27 @@ static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
375 if (!root->ref_cows) 364 if (!root->ref_cows)
376 return -EINVAL; 365 return -EINVAL;
377 366
378 ret = btrfs_check_metadata_free_space(root); 367 /*
368 * 1 - inode item
369 * 2 - refs
370 * 1 - root item
371 * 2 - dir items
372 */
373 ret = btrfs_reserve_metadata_space(root, 6);
379 if (ret) 374 if (ret)
380 goto fail_unlock; 375 goto fail_unlock;
381 376
382 pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS); 377 pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
383 if (!pending_snapshot) { 378 if (!pending_snapshot) {
384 ret = -ENOMEM; 379 ret = -ENOMEM;
380 btrfs_unreserve_metadata_space(root, 6);
385 goto fail_unlock; 381 goto fail_unlock;
386 } 382 }
387 pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS); 383 pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
388 if (!pending_snapshot->name) { 384 if (!pending_snapshot->name) {
389 ret = -ENOMEM; 385 ret = -ENOMEM;
390 kfree(pending_snapshot); 386 kfree(pending_snapshot);
387 btrfs_unreserve_metadata_space(root, 6);
391 goto fail_unlock; 388 goto fail_unlock;
392 } 389 }
393 memcpy(pending_snapshot->name, name, namelen); 390 memcpy(pending_snapshot->name, name, namelen);
@@ -420,14 +417,15 @@ static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
420 * sys_mkdirat and vfs_mkdir, but we only do a single component lookup 417 * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
421 * inside this filesystem so it's quite a bit simpler. 418 * inside this filesystem so it's quite a bit simpler.
422 */ 419 */
423static noinline int btrfs_mksubvol(struct path *parent, char *name, 420static noinline int btrfs_mksubvol(struct path *parent,
424 int mode, int namelen, 421 char *name, int namelen,
425 struct btrfs_root *snap_src) 422 struct btrfs_root *snap_src)
426{ 423{
424 struct inode *dir = parent->dentry->d_inode;
427 struct dentry *dentry; 425 struct dentry *dentry;
428 int error; 426 int error;
429 427
430 mutex_lock_nested(&parent->dentry->d_inode->i_mutex, I_MUTEX_PARENT); 428 mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
431 429
432 dentry = lookup_one_len(name, parent->dentry, namelen); 430 dentry = lookup_one_len(name, parent->dentry, namelen);
433 error = PTR_ERR(dentry); 431 error = PTR_ERR(dentry);
@@ -438,99 +436,39 @@ static noinline int btrfs_mksubvol(struct path *parent, char *name,
438 if (dentry->d_inode) 436 if (dentry->d_inode)
439 goto out_dput; 437 goto out_dput;
440 438
441 if (!IS_POSIXACL(parent->dentry->d_inode))
442 mode &= ~current_umask();
443
444 error = mnt_want_write(parent->mnt); 439 error = mnt_want_write(parent->mnt);
445 if (error) 440 if (error)
446 goto out_dput; 441 goto out_dput;
447 442
448 error = btrfs_may_create(parent->dentry->d_inode, dentry); 443 error = btrfs_may_create(dir, dentry);
449 if (error) 444 if (error)
450 goto out_drop_write; 445 goto out_drop_write;
451 446
452 /* 447 down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
453 * Actually perform the low-level subvolume creation after all 448
454 * this VFS fuzz. 449 if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
455 * 450 goto out_up_read;
456 * Eventually we want to pass in an inode under which we create this 451
457 * subvolume, but for now all are under the filesystem root.
458 *
459 * Also we should pass on the mode eventually to allow creating new
460 * subvolume with specific mode bits.
461 */
462 if (snap_src) { 452 if (snap_src) {
463 struct dentry *dir = dentry->d_parent; 453 error = create_snapshot(snap_src, dentry,
464 struct dentry *test = dir->d_parent; 454 name, namelen);
465 struct btrfs_path *path = btrfs_alloc_path();
466 int ret;
467 u64 test_oid;
468 u64 parent_oid = BTRFS_I(dir->d_inode)->root->root_key.objectid;
469
470 test_oid = snap_src->root_key.objectid;
471
472 ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
473 path, parent_oid, test_oid);
474 if (ret == 0)
475 goto create;
476 btrfs_release_path(snap_src->fs_info->tree_root, path);
477
478 /* we need to make sure we aren't creating a directory loop
479 * by taking a snapshot of something that has our current
480 * subvol in its directory tree. So, this loops through
481 * the dentries and checks the forward refs for each subvolume
482 * to see if is references the subvolume where we are
483 * placing this new snapshot.
484 */
485 while (1) {
486 if (!test ||
487 dir == snap_src->fs_info->sb->s_root ||
488 test == snap_src->fs_info->sb->s_root ||
489 test->d_inode->i_sb != snap_src->fs_info->sb) {
490 break;
491 }
492 if (S_ISLNK(test->d_inode->i_mode)) {
493 printk(KERN_INFO "Btrfs symlink in snapshot "
494 "path, failed\n");
495 error = -EMLINK;
496 btrfs_free_path(path);
497 goto out_drop_write;
498 }
499 test_oid =
500 BTRFS_I(test->d_inode)->root->root_key.objectid;
501 ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
502 path, test_oid, parent_oid);
503 if (ret == 0) {
504 printk(KERN_INFO "Btrfs snapshot creation "
505 "failed, looping\n");
506 error = -EMLINK;
507 btrfs_free_path(path);
508 goto out_drop_write;
509 }
510 btrfs_release_path(snap_src->fs_info->tree_root, path);
511 test = test->d_parent;
512 }
513create:
514 btrfs_free_path(path);
515 error = create_snapshot(snap_src, dentry, name, namelen);
516 } else { 455 } else {
517 error = create_subvol(BTRFS_I(parent->dentry->d_inode)->root, 456 error = create_subvol(BTRFS_I(dir)->root, dentry,
518 dentry, name, namelen); 457 name, namelen);
519 } 458 }
520 if (error) 459 if (!error)
521 goto out_drop_write; 460 fsnotify_mkdir(dir, dentry);
522 461out_up_read:
523 fsnotify_mkdir(parent->dentry->d_inode, dentry); 462 up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
524out_drop_write: 463out_drop_write:
525 mnt_drop_write(parent->mnt); 464 mnt_drop_write(parent->mnt);
526out_dput: 465out_dput:
527 dput(dentry); 466 dput(dentry);
528out_unlock: 467out_unlock:
529 mutex_unlock(&parent->dentry->d_inode->i_mutex); 468 mutex_unlock(&dir->i_mutex);
530 return error; 469 return error;
531} 470}
532 471
533
534static int btrfs_defrag_file(struct file *file) 472static int btrfs_defrag_file(struct file *file)
535{ 473{
536 struct inode *inode = fdentry(file)->d_inode; 474 struct inode *inode = fdentry(file)->d_inode;
@@ -596,9 +534,8 @@ again:
596 clear_page_dirty_for_io(page); 534 clear_page_dirty_for_io(page);
597 535
598 btrfs_set_extent_delalloc(inode, page_start, page_end); 536 btrfs_set_extent_delalloc(inode, page_start, page_end);
599
600 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
601 set_page_dirty(page); 537 set_page_dirty(page);
538 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
602 unlock_page(page); 539 unlock_page(page);
603 page_cache_release(page); 540 page_cache_release(page);
604 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1); 541 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
@@ -609,7 +546,8 @@ out_unlock:
609 return 0; 546 return 0;
610} 547}
611 548
612static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg) 549static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
550 void __user *arg)
613{ 551{
614 u64 new_size; 552 u64 new_size;
615 u64 old_size; 553 u64 old_size;
@@ -718,10 +656,7 @@ static noinline int btrfs_ioctl_snap_create(struct file *file,
718{ 656{
719 struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root; 657 struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
720 struct btrfs_ioctl_vol_args *vol_args; 658 struct btrfs_ioctl_vol_args *vol_args;
721 struct btrfs_dir_item *di;
722 struct btrfs_path *path;
723 struct file *src_file; 659 struct file *src_file;
724 u64 root_dirid;
725 int namelen; 660 int namelen;
726 int ret = 0; 661 int ret = 0;
727 662
@@ -739,32 +674,9 @@ static noinline int btrfs_ioctl_snap_create(struct file *file,
739 goto out; 674 goto out;
740 } 675 }
741 676
742 path = btrfs_alloc_path();
743 if (!path) {
744 ret = -ENOMEM;
745 goto out;
746 }
747
748 root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
749 di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
750 path, root_dirid,
751 vol_args->name, namelen, 0);
752 btrfs_free_path(path);
753
754 if (di && !IS_ERR(di)) {
755 ret = -EEXIST;
756 goto out;
757 }
758
759 if (IS_ERR(di)) {
760 ret = PTR_ERR(di);
761 goto out;
762 }
763
764 if (subvol) { 677 if (subvol) {
765 ret = btrfs_mksubvol(&file->f_path, vol_args->name, 678 ret = btrfs_mksubvol(&file->f_path, vol_args->name, namelen,
766 file->f_path.dentry->d_inode->i_mode, 679 NULL);
767 namelen, NULL);
768 } else { 680 } else {
769 struct inode *src_inode; 681 struct inode *src_inode;
770 src_file = fget(vol_args->fd); 682 src_file = fget(vol_args->fd);
@@ -781,17 +693,157 @@ static noinline int btrfs_ioctl_snap_create(struct file *file,
781 fput(src_file); 693 fput(src_file);
782 goto out; 694 goto out;
783 } 695 }
784 ret = btrfs_mksubvol(&file->f_path, vol_args->name, 696 ret = btrfs_mksubvol(&file->f_path, vol_args->name, namelen,
785 file->f_path.dentry->d_inode->i_mode, 697 BTRFS_I(src_inode)->root);
786 namelen, BTRFS_I(src_inode)->root);
787 fput(src_file); 698 fput(src_file);
788 } 699 }
789
790out: 700out:
791 kfree(vol_args); 701 kfree(vol_args);
792 return ret; 702 return ret;
793} 703}
794 704
705/*
706 * helper to check if the subvolume references other subvolumes
707 */
708static noinline int may_destroy_subvol(struct btrfs_root *root)
709{
710 struct btrfs_path *path;
711 struct btrfs_key key;
712 int ret;
713
714 path = btrfs_alloc_path();
715 if (!path)
716 return -ENOMEM;
717
718 key.objectid = root->root_key.objectid;
719 key.type = BTRFS_ROOT_REF_KEY;
720 key.offset = (u64)-1;
721
722 ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
723 &key, path, 0, 0);
724 if (ret < 0)
725 goto out;
726 BUG_ON(ret == 0);
727
728 ret = 0;
729 if (path->slots[0] > 0) {
730 path->slots[0]--;
731 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
732 if (key.objectid == root->root_key.objectid &&
733 key.type == BTRFS_ROOT_REF_KEY)
734 ret = -ENOTEMPTY;
735 }
736out:
737 btrfs_free_path(path);
738 return ret;
739}
740
741static noinline int btrfs_ioctl_snap_destroy(struct file *file,
742 void __user *arg)
743{
744 struct dentry *parent = fdentry(file);
745 struct dentry *dentry;
746 struct inode *dir = parent->d_inode;
747 struct inode *inode;
748 struct btrfs_root *root = BTRFS_I(dir)->root;
749 struct btrfs_root *dest = NULL;
750 struct btrfs_ioctl_vol_args *vol_args;
751 struct btrfs_trans_handle *trans;
752 int namelen;
753 int ret;
754 int err = 0;
755
756 if (!capable(CAP_SYS_ADMIN))
757 return -EPERM;
758
759 vol_args = memdup_user(arg, sizeof(*vol_args));
760 if (IS_ERR(vol_args))
761 return PTR_ERR(vol_args);
762
763 vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
764 namelen = strlen(vol_args->name);
765 if (strchr(vol_args->name, '/') ||
766 strncmp(vol_args->name, "..", namelen) == 0) {
767 err = -EINVAL;
768 goto out;
769 }
770
771 err = mnt_want_write(file->f_path.mnt);
772 if (err)
773 goto out;
774
775 mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
776 dentry = lookup_one_len(vol_args->name, parent, namelen);
777 if (IS_ERR(dentry)) {
778 err = PTR_ERR(dentry);
779 goto out_unlock_dir;
780 }
781
782 if (!dentry->d_inode) {
783 err = -ENOENT;
784 goto out_dput;
785 }
786
787 inode = dentry->d_inode;
788 if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
789 err = -EINVAL;
790 goto out_dput;
791 }
792
793 dest = BTRFS_I(inode)->root;
794
795 mutex_lock(&inode->i_mutex);
796 err = d_invalidate(dentry);
797 if (err)
798 goto out_unlock;
799
800 down_write(&root->fs_info->subvol_sem);
801
802 err = may_destroy_subvol(dest);
803 if (err)
804 goto out_up_write;
805
806 trans = btrfs_start_transaction(root, 1);
807 ret = btrfs_unlink_subvol(trans, root, dir,
808 dest->root_key.objectid,
809 dentry->d_name.name,
810 dentry->d_name.len);
811 BUG_ON(ret);
812
813 btrfs_record_root_in_trans(trans, dest);
814
815 memset(&dest->root_item.drop_progress, 0,
816 sizeof(dest->root_item.drop_progress));
817 dest->root_item.drop_level = 0;
818 btrfs_set_root_refs(&dest->root_item, 0);
819
820 ret = btrfs_insert_orphan_item(trans,
821 root->fs_info->tree_root,
822 dest->root_key.objectid);
823 BUG_ON(ret);
824
825 ret = btrfs_commit_transaction(trans, root);
826 BUG_ON(ret);
827 inode->i_flags |= S_DEAD;
828out_up_write:
829 up_write(&root->fs_info->subvol_sem);
830out_unlock:
831 mutex_unlock(&inode->i_mutex);
832 if (!err) {
833 shrink_dcache_sb(root->fs_info->sb);
834 btrfs_invalidate_inodes(dest);
835 d_delete(dentry);
836 }
837out_dput:
838 dput(dentry);
839out_unlock_dir:
840 mutex_unlock(&dir->i_mutex);
841 mnt_drop_write(file->f_path.mnt);
842out:
843 kfree(vol_args);
844 return err;
845}
846
795static int btrfs_ioctl_defrag(struct file *file) 847static int btrfs_ioctl_defrag(struct file *file)
796{ 848{
797 struct inode *inode = fdentry(file)->d_inode; 849 struct inode *inode = fdentry(file)->d_inode;
@@ -865,8 +917,8 @@ static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
865 return ret; 917 return ret;
866} 918}
867 919
868static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd, 920static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
869 u64 off, u64 olen, u64 destoff) 921 u64 off, u64 olen, u64 destoff)
870{ 922{
871 struct inode *inode = fdentry(file)->d_inode; 923 struct inode *inode = fdentry(file)->d_inode;
872 struct btrfs_root *root = BTRFS_I(inode)->root; 924 struct btrfs_root *root = BTRFS_I(inode)->root;
@@ -976,7 +1028,7 @@ static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
976 1028
977 /* punch hole in destination first */ 1029 /* punch hole in destination first */
978 btrfs_drop_extents(trans, root, inode, off, off + len, 1030 btrfs_drop_extents(trans, root, inode, off, off + len,
979 off + len, 0, &hint_byte); 1031 off + len, 0, &hint_byte, 1);
980 1032
981 /* clone data */ 1033 /* clone data */
982 key.objectid = src->i_ino; 1034 key.objectid = src->i_ino;
@@ -1071,9 +1123,10 @@ static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
1071 datao += off - key.offset; 1123 datao += off - key.offset;
1072 datal -= off - key.offset; 1124 datal -= off - key.offset;
1073 } 1125 }
1074 if (key.offset + datao + datal + key.offset > 1126
1075 off + len) 1127 if (key.offset + datal > off + len)
1076 datal = off + len - key.offset - datao; 1128 datal = off + len - key.offset;
1129
1077 /* disko == 0 means it's a hole */ 1130 /* disko == 0 means it's a hole */
1078 if (!disko) 1131 if (!disko)
1079 datao = 0; 1132 datao = 0;
@@ -1182,15 +1235,15 @@ static long btrfs_ioctl_trans_start(struct file *file)
1182 struct inode *inode = fdentry(file)->d_inode; 1235 struct inode *inode = fdentry(file)->d_inode;
1183 struct btrfs_root *root = BTRFS_I(inode)->root; 1236 struct btrfs_root *root = BTRFS_I(inode)->root;
1184 struct btrfs_trans_handle *trans; 1237 struct btrfs_trans_handle *trans;
1185 int ret = 0; 1238 int ret;
1186 1239
1240 ret = -EPERM;
1187 if (!capable(CAP_SYS_ADMIN)) 1241 if (!capable(CAP_SYS_ADMIN))
1188 return -EPERM; 1242 goto out;
1189 1243
1190 if (file->private_data) { 1244 ret = -EINPROGRESS;
1191 ret = -EINPROGRESS; 1245 if (file->private_data)
1192 goto out; 1246 goto out;
1193 }
1194 1247
1195 ret = mnt_want_write(file->f_path.mnt); 1248 ret = mnt_want_write(file->f_path.mnt);
1196 if (ret) 1249 if (ret)
@@ -1200,12 +1253,19 @@ static long btrfs_ioctl_trans_start(struct file *file)
1200 root->fs_info->open_ioctl_trans++; 1253 root->fs_info->open_ioctl_trans++;
1201 mutex_unlock(&root->fs_info->trans_mutex); 1254 mutex_unlock(&root->fs_info->trans_mutex);
1202 1255
1256 ret = -ENOMEM;
1203 trans = btrfs_start_ioctl_transaction(root, 0); 1257 trans = btrfs_start_ioctl_transaction(root, 0);
1204 if (trans) 1258 if (!trans)
1205 file->private_data = trans; 1259 goto out_drop;
1206 else 1260
1207 ret = -ENOMEM; 1261 file->private_data = trans;
1208 /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/ 1262 return 0;
1263
1264out_drop:
1265 mutex_lock(&root->fs_info->trans_mutex);
1266 root->fs_info->open_ioctl_trans--;
1267 mutex_unlock(&root->fs_info->trans_mutex);
1268 mnt_drop_write(file->f_path.mnt);
1209out: 1269out:
1210 return ret; 1270 return ret;
1211} 1271}
@@ -1221,24 +1281,20 @@ long btrfs_ioctl_trans_end(struct file *file)
1221 struct inode *inode = fdentry(file)->d_inode; 1281 struct inode *inode = fdentry(file)->d_inode;
1222 struct btrfs_root *root = BTRFS_I(inode)->root; 1282 struct btrfs_root *root = BTRFS_I(inode)->root;
1223 struct btrfs_trans_handle *trans; 1283 struct btrfs_trans_handle *trans;
1224 int ret = 0;
1225 1284
1226 trans = file->private_data; 1285 trans = file->private_data;
1227 if (!trans) { 1286 if (!trans)
1228 ret = -EINVAL; 1287 return -EINVAL;
1229 goto out;
1230 }
1231 btrfs_end_transaction(trans, root);
1232 file->private_data = NULL; 1288 file->private_data = NULL;
1233 1289
1290 btrfs_end_transaction(trans, root);
1291
1234 mutex_lock(&root->fs_info->trans_mutex); 1292 mutex_lock(&root->fs_info->trans_mutex);
1235 root->fs_info->open_ioctl_trans--; 1293 root->fs_info->open_ioctl_trans--;
1236 mutex_unlock(&root->fs_info->trans_mutex); 1294 mutex_unlock(&root->fs_info->trans_mutex);
1237 1295
1238 mnt_drop_write(file->f_path.mnt); 1296 mnt_drop_write(file->f_path.mnt);
1239 1297 return 0;
1240out:
1241 return ret;
1242} 1298}
1243 1299
1244long btrfs_ioctl(struct file *file, unsigned int 1300long btrfs_ioctl(struct file *file, unsigned int
@@ -1258,6 +1314,8 @@ long btrfs_ioctl(struct file *file, unsigned int
1258 return btrfs_ioctl_snap_create(file, argp, 0); 1314 return btrfs_ioctl_snap_create(file, argp, 0);
1259 case BTRFS_IOC_SUBVOL_CREATE: 1315 case BTRFS_IOC_SUBVOL_CREATE:
1260 return btrfs_ioctl_snap_create(file, argp, 1); 1316 return btrfs_ioctl_snap_create(file, argp, 1);
1317 case BTRFS_IOC_SNAP_DESTROY:
1318 return btrfs_ioctl_snap_destroy(file, argp);
1261 case BTRFS_IOC_DEFRAG: 1319 case BTRFS_IOC_DEFRAG:
1262 return btrfs_ioctl_defrag(file); 1320 return btrfs_ioctl_defrag(file);
1263 case BTRFS_IOC_RESIZE: 1321 case BTRFS_IOC_RESIZE:
diff --git a/fs/btrfs/ioctl.h b/fs/btrfs/ioctl.h
index b320b103fa13..bc49914475eb 100644
--- a/fs/btrfs/ioctl.h
+++ b/fs/btrfs/ioctl.h
@@ -65,5 +65,6 @@ struct btrfs_ioctl_clone_range_args {
65 65
66#define BTRFS_IOC_SUBVOL_CREATE _IOW(BTRFS_IOCTL_MAGIC, 14, \ 66#define BTRFS_IOC_SUBVOL_CREATE _IOW(BTRFS_IOCTL_MAGIC, 14, \
67 struct btrfs_ioctl_vol_args) 67 struct btrfs_ioctl_vol_args)
68 68#define BTRFS_IOC_SNAP_DESTROY _IOW(BTRFS_IOCTL_MAGIC, 15, \
69 struct btrfs_ioctl_vol_args)
69#endif 70#endif
diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c
index d6f0806c682f..5799bc46a309 100644
--- a/fs/btrfs/ordered-data.c
+++ b/fs/btrfs/ordered-data.c
@@ -159,8 +159,6 @@ static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
159 * 159 *
160 * len is the length of the extent 160 * len is the length of the extent
161 * 161 *
162 * This also sets the EXTENT_ORDERED bit on the range in the inode.
163 *
164 * The tree is given a single reference on the ordered extent that was 162 * The tree is given a single reference on the ordered extent that was
165 * inserted. 163 * inserted.
166 */ 164 */
@@ -181,6 +179,7 @@ int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
181 entry->start = start; 179 entry->start = start;
182 entry->len = len; 180 entry->len = len;
183 entry->disk_len = disk_len; 181 entry->disk_len = disk_len;
182 entry->bytes_left = len;
184 entry->inode = inode; 183 entry->inode = inode;
185 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE) 184 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
186 set_bit(type, &entry->flags); 185 set_bit(type, &entry->flags);
@@ -195,9 +194,6 @@ int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
195 &entry->rb_node); 194 &entry->rb_node);
196 BUG_ON(node); 195 BUG_ON(node);
197 196
198 set_extent_ordered(&BTRFS_I(inode)->io_tree, file_offset,
199 entry_end(entry) - 1, GFP_NOFS);
200
201 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); 197 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
202 list_add_tail(&entry->root_extent_list, 198 list_add_tail(&entry->root_extent_list,
203 &BTRFS_I(inode)->root->fs_info->ordered_extents); 199 &BTRFS_I(inode)->root->fs_info->ordered_extents);
@@ -241,13 +237,10 @@ int btrfs_dec_test_ordered_pending(struct inode *inode,
241 struct btrfs_ordered_inode_tree *tree; 237 struct btrfs_ordered_inode_tree *tree;
242 struct rb_node *node; 238 struct rb_node *node;
243 struct btrfs_ordered_extent *entry; 239 struct btrfs_ordered_extent *entry;
244 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
245 int ret; 240 int ret;
246 241
247 tree = &BTRFS_I(inode)->ordered_tree; 242 tree = &BTRFS_I(inode)->ordered_tree;
248 mutex_lock(&tree->mutex); 243 mutex_lock(&tree->mutex);
249 clear_extent_ordered(io_tree, file_offset, file_offset + io_size - 1,
250 GFP_NOFS);
251 node = tree_search(tree, file_offset); 244 node = tree_search(tree, file_offset);
252 if (!node) { 245 if (!node) {
253 ret = 1; 246 ret = 1;
@@ -260,11 +253,16 @@ int btrfs_dec_test_ordered_pending(struct inode *inode,
260 goto out; 253 goto out;
261 } 254 }
262 255
263 ret = test_range_bit(io_tree, entry->file_offset, 256 if (io_size > entry->bytes_left) {
264 entry->file_offset + entry->len - 1, 257 printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
265 EXTENT_ORDERED, 0); 258 (unsigned long long)entry->bytes_left,
266 if (ret == 0) 259 (unsigned long long)io_size);
260 }
261 entry->bytes_left -= io_size;
262 if (entry->bytes_left == 0)
267 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); 263 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
264 else
265 ret = 1;
268out: 266out:
269 mutex_unlock(&tree->mutex); 267 mutex_unlock(&tree->mutex);
270 return ret == 0; 268 return ret == 0;
@@ -308,6 +306,12 @@ int btrfs_remove_ordered_extent(struct inode *inode,
308 tree->last = NULL; 306 tree->last = NULL;
309 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags); 307 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
310 308
309 spin_lock(&BTRFS_I(inode)->accounting_lock);
310 BTRFS_I(inode)->outstanding_extents--;
311 spin_unlock(&BTRFS_I(inode)->accounting_lock);
312 btrfs_unreserve_metadata_for_delalloc(BTRFS_I(inode)->root,
313 inode, 1);
314
311 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); 315 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
312 list_del_init(&entry->root_extent_list); 316 list_del_init(&entry->root_extent_list);
313 317
@@ -460,7 +464,7 @@ void btrfs_start_ordered_extent(struct inode *inode,
460 * start IO on any dirty ones so the wait doesn't stall waiting 464 * start IO on any dirty ones so the wait doesn't stall waiting
461 * for pdflush to find them 465 * for pdflush to find them
462 */ 466 */
463 btrfs_fdatawrite_range(inode->i_mapping, start, end, WB_SYNC_ALL); 467 filemap_fdatawrite_range(inode->i_mapping, start, end);
464 if (wait) { 468 if (wait) {
465 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE, 469 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
466 &entry->flags)); 470 &entry->flags));
@@ -476,6 +480,7 @@ int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
476 u64 orig_end; 480 u64 orig_end;
477 u64 wait_end; 481 u64 wait_end;
478 struct btrfs_ordered_extent *ordered; 482 struct btrfs_ordered_extent *ordered;
483 int found;
479 484
480 if (start + len < start) { 485 if (start + len < start) {
481 orig_end = INT_LIMIT(loff_t); 486 orig_end = INT_LIMIT(loff_t);
@@ -489,19 +494,18 @@ again:
489 /* start IO across the range first to instantiate any delalloc 494 /* start IO across the range first to instantiate any delalloc
490 * extents 495 * extents
491 */ 496 */
492 btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_ALL); 497 filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
493 498
494 /* The compression code will leave pages locked but return from 499 /* The compression code will leave pages locked but return from
495 * writepage without setting the page writeback. Starting again 500 * writepage without setting the page writeback. Starting again
496 * with WB_SYNC_ALL will end up waiting for the IO to actually start. 501 * with WB_SYNC_ALL will end up waiting for the IO to actually start.
497 */ 502 */
498 btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_ALL); 503 filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
499 504
500 btrfs_wait_on_page_writeback_range(inode->i_mapping, 505 filemap_fdatawait_range(inode->i_mapping, start, orig_end);
501 start >> PAGE_CACHE_SHIFT,
502 orig_end >> PAGE_CACHE_SHIFT);
503 506
504 end = orig_end; 507 end = orig_end;
508 found = 0;
505 while (1) { 509 while (1) {
506 ordered = btrfs_lookup_first_ordered_extent(inode, end); 510 ordered = btrfs_lookup_first_ordered_extent(inode, end);
507 if (!ordered) 511 if (!ordered)
@@ -514,6 +518,7 @@ again:
514 btrfs_put_ordered_extent(ordered); 518 btrfs_put_ordered_extent(ordered);
515 break; 519 break;
516 } 520 }
521 found++;
517 btrfs_start_ordered_extent(inode, ordered, 1); 522 btrfs_start_ordered_extent(inode, ordered, 1);
518 end = ordered->file_offset; 523 end = ordered->file_offset;
519 btrfs_put_ordered_extent(ordered); 524 btrfs_put_ordered_extent(ordered);
@@ -521,8 +526,8 @@ again:
521 break; 526 break;
522 end--; 527 end--;
523 } 528 }
524 if (test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end, 529 if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
525 EXTENT_ORDERED | EXTENT_DELALLOC, 0)) { 530 EXTENT_DELALLOC, 0, NULL)) {
526 schedule_timeout(1); 531 schedule_timeout(1);
527 goto again; 532 goto again;
528 } 533 }
@@ -613,7 +618,7 @@ int btrfs_ordered_update_i_size(struct inode *inode,
613 */ 618 */
614 if (test_range_bit(io_tree, disk_i_size, 619 if (test_range_bit(io_tree, disk_i_size,
615 ordered->file_offset + ordered->len - 1, 620 ordered->file_offset + ordered->len - 1,
616 EXTENT_DELALLOC, 0)) { 621 EXTENT_DELALLOC, 0, NULL)) {
617 goto out; 622 goto out;
618 } 623 }
619 /* 624 /*
@@ -664,7 +669,7 @@ int btrfs_ordered_update_i_size(struct inode *inode,
664 */ 669 */
665 if (i_size_test > entry_end(ordered) && 670 if (i_size_test > entry_end(ordered) &&
666 !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1, 671 !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1,
667 EXTENT_DELALLOC, 0)) { 672 EXTENT_DELALLOC, 0, NULL)) {
668 new_i_size = min_t(u64, i_size_test, i_size_read(inode)); 673 new_i_size = min_t(u64, i_size_test, i_size_read(inode));
669 } 674 }
670 BTRFS_I(inode)->disk_i_size = new_i_size; 675 BTRFS_I(inode)->disk_i_size = new_i_size;
@@ -715,90 +720,6 @@ out:
715} 720}
716 721
717 722
718/**
719 * taken from mm/filemap.c because it isn't exported
720 *
721 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
722 * @mapping: address space structure to write
723 * @start: offset in bytes where the range starts
724 * @end: offset in bytes where the range ends (inclusive)
725 * @sync_mode: enable synchronous operation
726 *
727 * Start writeback against all of a mapping's dirty pages that lie
728 * within the byte offsets <start, end> inclusive.
729 *
730 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
731 * opposed to a regular memory cleansing writeback. The difference between
732 * these two operations is that if a dirty page/buffer is encountered, it must
733 * be waited upon, and not just skipped over.
734 */
735int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start,
736 loff_t end, int sync_mode)
737{
738 struct writeback_control wbc = {
739 .sync_mode = sync_mode,
740 .nr_to_write = mapping->nrpages * 2,
741 .range_start = start,
742 .range_end = end,
743 .for_writepages = 1,
744 };
745 return btrfs_writepages(mapping, &wbc);
746}
747
748/**
749 * taken from mm/filemap.c because it isn't exported
750 *
751 * wait_on_page_writeback_range - wait for writeback to complete
752 * @mapping: target address_space
753 * @start: beginning page index
754 * @end: ending page index
755 *
756 * Wait for writeback to complete against pages indexed by start->end
757 * inclusive
758 */
759int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
760 pgoff_t start, pgoff_t end)
761{
762 struct pagevec pvec;
763 int nr_pages;
764 int ret = 0;
765 pgoff_t index;
766
767 if (end < start)
768 return 0;
769
770 pagevec_init(&pvec, 0);
771 index = start;
772 while ((index <= end) &&
773 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
774 PAGECACHE_TAG_WRITEBACK,
775 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
776 unsigned i;
777
778 for (i = 0; i < nr_pages; i++) {
779 struct page *page = pvec.pages[i];
780
781 /* until radix tree lookup accepts end_index */
782 if (page->index > end)
783 continue;
784
785 wait_on_page_writeback(page);
786 if (PageError(page))
787 ret = -EIO;
788 }
789 pagevec_release(&pvec);
790 cond_resched();
791 }
792
793 /* Check for outstanding write errors */
794 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
795 ret = -ENOSPC;
796 if (test_and_clear_bit(AS_EIO, &mapping->flags))
797 ret = -EIO;
798
799 return ret;
800}
801
802/* 723/*
803 * add a given inode to the list of inodes that must be fully on 724 * add a given inode to the list of inodes that must be fully on
804 * disk before a transaction commit finishes. 725 * disk before a transaction commit finishes.
diff --git a/fs/btrfs/ordered-data.h b/fs/btrfs/ordered-data.h
index 3d31c8827b01..f82e87488ca8 100644
--- a/fs/btrfs/ordered-data.h
+++ b/fs/btrfs/ordered-data.h
@@ -85,6 +85,9 @@ struct btrfs_ordered_extent {
85 /* extent length on disk */ 85 /* extent length on disk */
86 u64 disk_len; 86 u64 disk_len;
87 87
88 /* number of bytes that still need writing */
89 u64 bytes_left;
90
88 /* flags (described above) */ 91 /* flags (described above) */
89 unsigned long flags; 92 unsigned long flags;
90 93
@@ -150,10 +153,6 @@ btrfs_lookup_first_ordered_extent(struct inode * inode, u64 file_offset);
150int btrfs_ordered_update_i_size(struct inode *inode, 153int btrfs_ordered_update_i_size(struct inode *inode,
151 struct btrfs_ordered_extent *ordered); 154 struct btrfs_ordered_extent *ordered);
152int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr, u32 *sum); 155int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr, u32 *sum);
153int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
154 pgoff_t start, pgoff_t end);
155int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start,
156 loff_t end, int sync_mode);
157int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only); 156int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only);
158int btrfs_run_ordered_operations(struct btrfs_root *root, int wait); 157int btrfs_run_ordered_operations(struct btrfs_root *root, int wait);
159int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans, 158int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
diff --git a/fs/btrfs/orphan.c b/fs/btrfs/orphan.c
index 3c0d52af4f80..79cba5fbc28e 100644
--- a/fs/btrfs/orphan.c
+++ b/fs/btrfs/orphan.c
@@ -65,3 +65,23 @@ out:
65 btrfs_free_path(path); 65 btrfs_free_path(path);
66 return ret; 66 return ret;
67} 67}
68
69int btrfs_find_orphan_item(struct btrfs_root *root, u64 offset)
70{
71 struct btrfs_path *path;
72 struct btrfs_key key;
73 int ret;
74
75 key.objectid = BTRFS_ORPHAN_OBJECTID;
76 key.type = BTRFS_ORPHAN_ITEM_KEY;
77 key.offset = offset;
78
79 path = btrfs_alloc_path();
80 if (!path)
81 return -ENOMEM;
82
83 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
84
85 btrfs_free_path(path);
86 return ret;
87}
diff --git a/fs/btrfs/relocation.c b/fs/btrfs/relocation.c
index c04f7f212602..cfcc93c93a7b 100644
--- a/fs/btrfs/relocation.c
+++ b/fs/btrfs/relocation.c
@@ -121,6 +121,15 @@ struct inodevec {
121 int nr; 121 int nr;
122}; 122};
123 123
124#define MAX_EXTENTS 128
125
126struct file_extent_cluster {
127 u64 start;
128 u64 end;
129 u64 boundary[MAX_EXTENTS];
130 unsigned int nr;
131};
132
124struct reloc_control { 133struct reloc_control {
125 /* block group to relocate */ 134 /* block group to relocate */
126 struct btrfs_block_group_cache *block_group; 135 struct btrfs_block_group_cache *block_group;
@@ -2180,7 +2189,7 @@ static int tree_block_processed(u64 bytenr, u32 blocksize,
2180 struct reloc_control *rc) 2189 struct reloc_control *rc)
2181{ 2190{
2182 if (test_range_bit(&rc->processed_blocks, bytenr, 2191 if (test_range_bit(&rc->processed_blocks, bytenr,
2183 bytenr + blocksize - 1, EXTENT_DIRTY, 1)) 2192 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2184 return 1; 2193 return 1;
2185 return 0; 2194 return 0;
2186} 2195}
@@ -2529,56 +2538,94 @@ out:
2529} 2538}
2530 2539
2531static noinline_for_stack 2540static noinline_for_stack
2532int relocate_inode_pages(struct inode *inode, u64 start, u64 len) 2541int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
2542 u64 block_start)
2543{
2544 struct btrfs_root *root = BTRFS_I(inode)->root;
2545 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2546 struct extent_map *em;
2547 int ret = 0;
2548
2549 em = alloc_extent_map(GFP_NOFS);
2550 if (!em)
2551 return -ENOMEM;
2552
2553 em->start = start;
2554 em->len = end + 1 - start;
2555 em->block_len = em->len;
2556 em->block_start = block_start;
2557 em->bdev = root->fs_info->fs_devices->latest_bdev;
2558 set_bit(EXTENT_FLAG_PINNED, &em->flags);
2559
2560 lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
2561 while (1) {
2562 write_lock(&em_tree->lock);
2563 ret = add_extent_mapping(em_tree, em);
2564 write_unlock(&em_tree->lock);
2565 if (ret != -EEXIST) {
2566 free_extent_map(em);
2567 break;
2568 }
2569 btrfs_drop_extent_cache(inode, start, end, 0);
2570 }
2571 unlock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
2572 return ret;
2573}
2574
2575static int relocate_file_extent_cluster(struct inode *inode,
2576 struct file_extent_cluster *cluster)
2533{ 2577{
2534 u64 page_start; 2578 u64 page_start;
2535 u64 page_end; 2579 u64 page_end;
2536 unsigned long i; 2580 u64 offset = BTRFS_I(inode)->index_cnt;
2537 unsigned long first_index; 2581 unsigned long index;
2538 unsigned long last_index; 2582 unsigned long last_index;
2539 unsigned int total_read = 0; 2583 unsigned int dirty_page = 0;
2540 unsigned int total_dirty = 0;
2541 struct page *page; 2584 struct page *page;
2542 struct file_ra_state *ra; 2585 struct file_ra_state *ra;
2543 struct btrfs_ordered_extent *ordered; 2586 int nr = 0;
2544 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2545 int ret = 0; 2587 int ret = 0;
2546 2588
2589 if (!cluster->nr)
2590 return 0;
2591
2547 ra = kzalloc(sizeof(*ra), GFP_NOFS); 2592 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2548 if (!ra) 2593 if (!ra)
2549 return -ENOMEM; 2594 return -ENOMEM;
2550 2595
2596 index = (cluster->start - offset) >> PAGE_CACHE_SHIFT;
2597 last_index = (cluster->end - offset) >> PAGE_CACHE_SHIFT;
2598
2551 mutex_lock(&inode->i_mutex); 2599 mutex_lock(&inode->i_mutex);
2552 first_index = start >> PAGE_CACHE_SHIFT;
2553 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2554 2600
2555 /* make sure the dirty trick played by the caller work */ 2601 i_size_write(inode, cluster->end + 1 - offset);
2556 while (1) { 2602 ret = setup_extent_mapping(inode, cluster->start - offset,
2557 ret = invalidate_inode_pages2_range(inode->i_mapping, 2603 cluster->end - offset, cluster->start);
2558 first_index, last_index);
2559 if (ret != -EBUSY)
2560 break;
2561 schedule_timeout(HZ/10);
2562 }
2563 if (ret) 2604 if (ret)
2564 goto out_unlock; 2605 goto out_unlock;
2565 2606
2566 file_ra_state_init(ra, inode->i_mapping); 2607 file_ra_state_init(ra, inode->i_mapping);
2567 2608
2568 for (i = first_index ; i <= last_index; i++) { 2609 WARN_ON(cluster->start != cluster->boundary[0]);
2569 if (total_read % ra->ra_pages == 0) { 2610 while (index <= last_index) {
2570 btrfs_force_ra(inode->i_mapping, ra, NULL, i, 2611 page = find_lock_page(inode->i_mapping, index);
2571 min(last_index, ra->ra_pages + i - 1));
2572 }
2573 total_read++;
2574again:
2575 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
2576 BUG_ON(1);
2577 page = grab_cache_page(inode->i_mapping, i);
2578 if (!page) { 2612 if (!page) {
2579 ret = -ENOMEM; 2613 page_cache_sync_readahead(inode->i_mapping,
2580 goto out_unlock; 2614 ra, NULL, index,
2615 last_index + 1 - index);
2616 page = grab_cache_page(inode->i_mapping, index);
2617 if (!page) {
2618 ret = -ENOMEM;
2619 goto out_unlock;
2620 }
2621 }
2622
2623 if (PageReadahead(page)) {
2624 page_cache_async_readahead(inode->i_mapping,
2625 ra, NULL, page, index,
2626 last_index + 1 - index);
2581 } 2627 }
2628
2582 if (!PageUptodate(page)) { 2629 if (!PageUptodate(page)) {
2583 btrfs_readpage(NULL, page); 2630 btrfs_readpage(NULL, page);
2584 lock_page(page); 2631 lock_page(page);
@@ -2589,75 +2636,79 @@ again:
2589 goto out_unlock; 2636 goto out_unlock;
2590 } 2637 }
2591 } 2638 }
2592 wait_on_page_writeback(page);
2593 2639
2594 page_start = (u64)page->index << PAGE_CACHE_SHIFT; 2640 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2595 page_end = page_start + PAGE_CACHE_SIZE - 1; 2641 page_end = page_start + PAGE_CACHE_SIZE - 1;
2596 lock_extent(io_tree, page_start, page_end, GFP_NOFS); 2642
2597 2643 lock_extent(&BTRFS_I(inode)->io_tree,
2598 ordered = btrfs_lookup_ordered_extent(inode, page_start); 2644 page_start, page_end, GFP_NOFS);
2599 if (ordered) { 2645
2600 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2601 unlock_page(page);
2602 page_cache_release(page);
2603 btrfs_start_ordered_extent(inode, ordered, 1);
2604 btrfs_put_ordered_extent(ordered);
2605 goto again;
2606 }
2607 set_page_extent_mapped(page); 2646 set_page_extent_mapped(page);
2608 2647
2609 if (i == first_index) 2648 if (nr < cluster->nr &&
2610 set_extent_bits(io_tree, page_start, page_end, 2649 page_start + offset == cluster->boundary[nr]) {
2650 set_extent_bits(&BTRFS_I(inode)->io_tree,
2651 page_start, page_end,
2611 EXTENT_BOUNDARY, GFP_NOFS); 2652 EXTENT_BOUNDARY, GFP_NOFS);
2653 nr++;
2654 }
2612 btrfs_set_extent_delalloc(inode, page_start, page_end); 2655 btrfs_set_extent_delalloc(inode, page_start, page_end);
2613 2656
2614 set_page_dirty(page); 2657 set_page_dirty(page);
2615 total_dirty++; 2658 dirty_page++;
2616 2659
2617 unlock_extent(io_tree, page_start, page_end, GFP_NOFS); 2660 unlock_extent(&BTRFS_I(inode)->io_tree,
2661 page_start, page_end, GFP_NOFS);
2618 unlock_page(page); 2662 unlock_page(page);
2619 page_cache_release(page); 2663 page_cache_release(page);
2664
2665 index++;
2666 if (nr < cluster->nr &&
2667 page_end + 1 + offset == cluster->boundary[nr]) {
2668 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2669 dirty_page);
2670 dirty_page = 0;
2671 }
2672 }
2673 if (dirty_page) {
2674 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2675 dirty_page);
2620 } 2676 }
2677 WARN_ON(nr != cluster->nr);
2621out_unlock: 2678out_unlock:
2622 mutex_unlock(&inode->i_mutex); 2679 mutex_unlock(&inode->i_mutex);
2623 kfree(ra); 2680 kfree(ra);
2624 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
2625 return ret; 2681 return ret;
2626} 2682}
2627 2683
2628static noinline_for_stack 2684static noinline_for_stack
2629int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key) 2685int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
2686 struct file_extent_cluster *cluster)
2630{ 2687{
2631 struct btrfs_root *root = BTRFS_I(inode)->root; 2688 int ret;
2632 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2633 struct extent_map *em;
2634 u64 start = extent_key->objectid - BTRFS_I(inode)->index_cnt;
2635 u64 end = start + extent_key->offset - 1;
2636
2637 em = alloc_extent_map(GFP_NOFS);
2638 em->start = start;
2639 em->len = extent_key->offset;
2640 em->block_len = extent_key->offset;
2641 em->block_start = extent_key->objectid;
2642 em->bdev = root->fs_info->fs_devices->latest_bdev;
2643 set_bit(EXTENT_FLAG_PINNED, &em->flags);
2644 2689
2645 /* setup extent map to cheat btrfs_readpage */ 2690 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
2646 lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS); 2691 ret = relocate_file_extent_cluster(inode, cluster);
2647 while (1) { 2692 if (ret)
2648 int ret; 2693 return ret;
2649 spin_lock(&em_tree->lock); 2694 cluster->nr = 0;
2650 ret = add_extent_mapping(em_tree, em);
2651 spin_unlock(&em_tree->lock);
2652 if (ret != -EEXIST) {
2653 free_extent_map(em);
2654 break;
2655 }
2656 btrfs_drop_extent_cache(inode, start, end, 0);
2657 } 2695 }
2658 unlock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
2659 2696
2660 return relocate_inode_pages(inode, start, extent_key->offset); 2697 if (!cluster->nr)
2698 cluster->start = extent_key->objectid;
2699 else
2700 BUG_ON(cluster->nr >= MAX_EXTENTS);
2701 cluster->end = extent_key->objectid + extent_key->offset - 1;
2702 cluster->boundary[cluster->nr] = extent_key->objectid;
2703 cluster->nr++;
2704
2705 if (cluster->nr >= MAX_EXTENTS) {
2706 ret = relocate_file_extent_cluster(inode, cluster);
2707 if (ret)
2708 return ret;
2709 cluster->nr = 0;
2710 }
2711 return 0;
2661} 2712}
2662 2713
2663#ifdef BTRFS_COMPAT_EXTENT_TREE_V0 2714#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
@@ -3203,10 +3254,12 @@ static int check_extent_flags(u64 flags)
3203 return 0; 3254 return 0;
3204} 3255}
3205 3256
3257
3206static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 3258static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3207{ 3259{
3208 struct rb_root blocks = RB_ROOT; 3260 struct rb_root blocks = RB_ROOT;
3209 struct btrfs_key key; 3261 struct btrfs_key key;
3262 struct file_extent_cluster *cluster;
3210 struct btrfs_trans_handle *trans = NULL; 3263 struct btrfs_trans_handle *trans = NULL;
3211 struct btrfs_path *path; 3264 struct btrfs_path *path;
3212 struct btrfs_extent_item *ei; 3265 struct btrfs_extent_item *ei;
@@ -3216,10 +3269,17 @@ static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3216 int ret; 3269 int ret;
3217 int err = 0; 3270 int err = 0;
3218 3271
3272 cluster = kzalloc(sizeof(*cluster), GFP_NOFS);
3273 if (!cluster)
3274 return -ENOMEM;
3275
3219 path = btrfs_alloc_path(); 3276 path = btrfs_alloc_path();
3220 if (!path) 3277 if (!path)
3221 return -ENOMEM; 3278 return -ENOMEM;
3222 3279
3280 rc->extents_found = 0;
3281 rc->extents_skipped = 0;
3282
3223 rc->search_start = rc->block_group->key.objectid; 3283 rc->search_start = rc->block_group->key.objectid;
3224 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY, 3284 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY,
3225 GFP_NOFS); 3285 GFP_NOFS);
@@ -3306,14 +3366,15 @@ static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3306 } 3366 }
3307 3367
3308 nr = trans->blocks_used; 3368 nr = trans->blocks_used;
3309 btrfs_end_transaction_throttle(trans, rc->extent_root); 3369 btrfs_end_transaction(trans, rc->extent_root);
3310 trans = NULL; 3370 trans = NULL;
3311 btrfs_btree_balance_dirty(rc->extent_root, nr); 3371 btrfs_btree_balance_dirty(rc->extent_root, nr);
3312 3372
3313 if (rc->stage == MOVE_DATA_EXTENTS && 3373 if (rc->stage == MOVE_DATA_EXTENTS &&
3314 (flags & BTRFS_EXTENT_FLAG_DATA)) { 3374 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3315 rc->found_file_extent = 1; 3375 rc->found_file_extent = 1;
3316 ret = relocate_data_extent(rc->data_inode, &key); 3376 ret = relocate_data_extent(rc->data_inode,
3377 &key, cluster);
3317 if (ret < 0) { 3378 if (ret < 0) {
3318 err = ret; 3379 err = ret;
3319 break; 3380 break;
@@ -3328,6 +3389,14 @@ static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3328 btrfs_btree_balance_dirty(rc->extent_root, nr); 3389 btrfs_btree_balance_dirty(rc->extent_root, nr);
3329 } 3390 }
3330 3391
3392 if (!err) {
3393 ret = relocate_file_extent_cluster(rc->data_inode, cluster);
3394 if (ret < 0)
3395 err = ret;
3396 }
3397
3398 kfree(cluster);
3399
3331 rc->create_reloc_root = 0; 3400 rc->create_reloc_root = 0;
3332 smp_mb(); 3401 smp_mb();
3333 3402
@@ -3348,8 +3417,7 @@ static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3348} 3417}
3349 3418
3350static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 3419static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3351 struct btrfs_root *root, 3420 struct btrfs_root *root, u64 objectid)
3352 u64 objectid, u64 size)
3353{ 3421{
3354 struct btrfs_path *path; 3422 struct btrfs_path *path;
3355 struct btrfs_inode_item *item; 3423 struct btrfs_inode_item *item;
@@ -3368,7 +3436,7 @@ static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3368 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 3436 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
3369 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); 3437 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3370 btrfs_set_inode_generation(leaf, item, 1); 3438 btrfs_set_inode_generation(leaf, item, 1);
3371 btrfs_set_inode_size(leaf, item, size); 3439 btrfs_set_inode_size(leaf, item, 0);
3372 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 3440 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
3373 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS); 3441 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
3374 btrfs_mark_buffer_dirty(leaf); 3442 btrfs_mark_buffer_dirty(leaf);
@@ -3404,12 +3472,7 @@ static struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
3404 if (err) 3472 if (err)
3405 goto out; 3473 goto out;
3406 3474
3407 err = __insert_orphan_inode(trans, root, objectid, group->key.offset); 3475 err = __insert_orphan_inode(trans, root, objectid);
3408 BUG_ON(err);
3409
3410 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
3411 group->key.offset, 0, group->key.offset,
3412 0, 0, 0);
3413 BUG_ON(err); 3476 BUG_ON(err);
3414 3477
3415 key.objectid = objectid; 3478 key.objectid = objectid;
@@ -3455,7 +3518,7 @@ int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
3455 BUG_ON(!rc->block_group); 3518 BUG_ON(!rc->block_group);
3456 3519
3457 btrfs_init_workers(&rc->workers, "relocate", 3520 btrfs_init_workers(&rc->workers, "relocate",
3458 fs_info->thread_pool_size); 3521 fs_info->thread_pool_size, NULL);
3459 3522
3460 rc->extent_root = extent_root; 3523 rc->extent_root = extent_root;
3461 btrfs_prepare_block_group_relocation(extent_root, rc->block_group); 3524 btrfs_prepare_block_group_relocation(extent_root, rc->block_group);
@@ -3475,14 +3538,15 @@ int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
3475 btrfs_wait_ordered_extents(fs_info->tree_root, 0); 3538 btrfs_wait_ordered_extents(fs_info->tree_root, 0);
3476 3539
3477 while (1) { 3540 while (1) {
3478 mutex_lock(&fs_info->cleaner_mutex);
3479 btrfs_clean_old_snapshots(fs_info->tree_root);
3480 mutex_unlock(&fs_info->cleaner_mutex);
3481
3482 rc->extents_found = 0; 3541 rc->extents_found = 0;
3483 rc->extents_skipped = 0; 3542 rc->extents_skipped = 0;
3484 3543
3544 mutex_lock(&fs_info->cleaner_mutex);
3545
3546 btrfs_clean_old_snapshots(fs_info->tree_root);
3485 ret = relocate_block_group(rc); 3547 ret = relocate_block_group(rc);
3548
3549 mutex_unlock(&fs_info->cleaner_mutex);
3486 if (ret < 0) { 3550 if (ret < 0) {
3487 err = ret; 3551 err = ret;
3488 break; 3552 break;
@@ -3514,10 +3578,10 @@ int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
3514 } 3578 }
3515 } 3579 }
3516 3580
3517 filemap_fdatawrite_range(fs_info->btree_inode->i_mapping, 3581 filemap_write_and_wait_range(fs_info->btree_inode->i_mapping,
3518 rc->block_group->key.objectid, 3582 rc->block_group->key.objectid,
3519 rc->block_group->key.objectid + 3583 rc->block_group->key.objectid +
3520 rc->block_group->key.offset - 1); 3584 rc->block_group->key.offset - 1);
3521 3585
3522 WARN_ON(rc->block_group->pinned > 0); 3586 WARN_ON(rc->block_group->pinned > 0);
3523 WARN_ON(rc->block_group->reserved > 0); 3587 WARN_ON(rc->block_group->reserved > 0);
@@ -3530,6 +3594,26 @@ out:
3530 return err; 3594 return err;
3531} 3595}
3532 3596
3597static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
3598{
3599 struct btrfs_trans_handle *trans;
3600 int ret;
3601
3602 trans = btrfs_start_transaction(root->fs_info->tree_root, 1);
3603
3604 memset(&root->root_item.drop_progress, 0,
3605 sizeof(root->root_item.drop_progress));
3606 root->root_item.drop_level = 0;
3607 btrfs_set_root_refs(&root->root_item, 0);
3608 ret = btrfs_update_root(trans, root->fs_info->tree_root,
3609 &root->root_key, &root->root_item);
3610 BUG_ON(ret);
3611
3612 ret = btrfs_end_transaction(trans, root->fs_info->tree_root);
3613 BUG_ON(ret);
3614 return 0;
3615}
3616
3533/* 3617/*
3534 * recover relocation interrupted by system crash. 3618 * recover relocation interrupted by system crash.
3535 * 3619 *
@@ -3589,8 +3673,12 @@ int btrfs_recover_relocation(struct btrfs_root *root)
3589 fs_root = read_fs_root(root->fs_info, 3673 fs_root = read_fs_root(root->fs_info,
3590 reloc_root->root_key.offset); 3674 reloc_root->root_key.offset);
3591 if (IS_ERR(fs_root)) { 3675 if (IS_ERR(fs_root)) {
3592 err = PTR_ERR(fs_root); 3676 ret = PTR_ERR(fs_root);
3593 goto out; 3677 if (ret != -ENOENT) {
3678 err = ret;
3679 goto out;
3680 }
3681 mark_garbage_root(reloc_root);
3594 } 3682 }
3595 } 3683 }
3596 3684
@@ -3613,7 +3701,7 @@ int btrfs_recover_relocation(struct btrfs_root *root)
3613 mapping_tree_init(&rc->reloc_root_tree); 3701 mapping_tree_init(&rc->reloc_root_tree);
3614 INIT_LIST_HEAD(&rc->reloc_roots); 3702 INIT_LIST_HEAD(&rc->reloc_roots);
3615 btrfs_init_workers(&rc->workers, "relocate", 3703 btrfs_init_workers(&rc->workers, "relocate",
3616 root->fs_info->thread_pool_size); 3704 root->fs_info->thread_pool_size, NULL);
3617 rc->extent_root = root->fs_info->extent_root; 3705 rc->extent_root = root->fs_info->extent_root;
3618 3706
3619 set_reloc_control(rc); 3707 set_reloc_control(rc);
diff --git a/fs/btrfs/root-tree.c b/fs/btrfs/root-tree.c
index 0ddc6d61c55a..9351428f30e2 100644
--- a/fs/btrfs/root-tree.c
+++ b/fs/btrfs/root-tree.c
@@ -94,17 +94,23 @@ int btrfs_find_last_root(struct btrfs_root *root, u64 objectid,
94 goto out; 94 goto out;
95 95
96 BUG_ON(ret == 0); 96 BUG_ON(ret == 0);
97 if (path->slots[0] == 0) {
98 ret = 1;
99 goto out;
100 }
97 l = path->nodes[0]; 101 l = path->nodes[0];
98 BUG_ON(path->slots[0] == 0);
99 slot = path->slots[0] - 1; 102 slot = path->slots[0] - 1;
100 btrfs_item_key_to_cpu(l, &found_key, slot); 103 btrfs_item_key_to_cpu(l, &found_key, slot);
101 if (found_key.objectid != objectid) { 104 if (found_key.objectid != objectid ||
105 found_key.type != BTRFS_ROOT_ITEM_KEY) {
102 ret = 1; 106 ret = 1;
103 goto out; 107 goto out;
104 } 108 }
105 read_extent_buffer(l, item, btrfs_item_ptr_offset(l, slot), 109 if (item)
106 sizeof(*item)); 110 read_extent_buffer(l, item, btrfs_item_ptr_offset(l, slot),
107 memcpy(key, &found_key, sizeof(found_key)); 111 sizeof(*item));
112 if (key)
113 memcpy(key, &found_key, sizeof(found_key));
108 ret = 0; 114 ret = 0;
109out: 115out:
110 btrfs_free_path(path); 116 btrfs_free_path(path);
@@ -249,6 +255,59 @@ err:
249 return ret; 255 return ret;
250} 256}
251 257
258int btrfs_find_orphan_roots(struct btrfs_root *tree_root)
259{
260 struct extent_buffer *leaf;
261 struct btrfs_path *path;
262 struct btrfs_key key;
263 int err = 0;
264 int ret;
265
266 path = btrfs_alloc_path();
267 if (!path)
268 return -ENOMEM;
269
270 key.objectid = BTRFS_ORPHAN_OBJECTID;
271 key.type = BTRFS_ORPHAN_ITEM_KEY;
272 key.offset = 0;
273
274 while (1) {
275 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
276 if (ret < 0) {
277 err = ret;
278 break;
279 }
280
281 leaf = path->nodes[0];
282 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
283 ret = btrfs_next_leaf(tree_root, path);
284 if (ret < 0)
285 err = ret;
286 if (ret != 0)
287 break;
288 leaf = path->nodes[0];
289 }
290
291 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
292 btrfs_release_path(tree_root, path);
293
294 if (key.objectid != BTRFS_ORPHAN_OBJECTID ||
295 key.type != BTRFS_ORPHAN_ITEM_KEY)
296 break;
297
298 ret = btrfs_find_dead_roots(tree_root, key.offset);
299 if (ret) {
300 err = ret;
301 break;
302 }
303
304 key.offset++;
305 }
306
307 btrfs_free_path(path);
308 return err;
309}
310
252/* drop the root item for 'key' from 'root' */ 311/* drop the root item for 'key' from 'root' */
253int btrfs_del_root(struct btrfs_trans_handle *trans, struct btrfs_root *root, 312int btrfs_del_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
254 struct btrfs_key *key) 313 struct btrfs_key *key)
@@ -278,31 +337,57 @@ out:
278 return ret; 337 return ret;
279} 338}
280 339
281#if 0 /* this will get used when snapshot deletion is implemented */
282int btrfs_del_root_ref(struct btrfs_trans_handle *trans, 340int btrfs_del_root_ref(struct btrfs_trans_handle *trans,
283 struct btrfs_root *tree_root, 341 struct btrfs_root *tree_root,
284 u64 root_id, u8 type, u64 ref_id) 342 u64 root_id, u64 ref_id, u64 dirid, u64 *sequence,
343 const char *name, int name_len)
344
285{ 345{
346 struct btrfs_path *path;
347 struct btrfs_root_ref *ref;
348 struct extent_buffer *leaf;
286 struct btrfs_key key; 349 struct btrfs_key key;
350 unsigned long ptr;
351 int err = 0;
287 int ret; 352 int ret;
288 struct btrfs_path *path;
289 353
290 path = btrfs_alloc_path(); 354 path = btrfs_alloc_path();
355 if (!path)
356 return -ENOMEM;
291 357
292 key.objectid = root_id; 358 key.objectid = root_id;
293 key.type = type; 359 key.type = BTRFS_ROOT_BACKREF_KEY;
294 key.offset = ref_id; 360 key.offset = ref_id;
295 361again:
296 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); 362 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
297 BUG_ON(ret); 363 BUG_ON(ret < 0);
298 364 if (ret == 0) {
299 ret = btrfs_del_item(trans, tree_root, path); 365 leaf = path->nodes[0];
300 BUG_ON(ret); 366 ref = btrfs_item_ptr(leaf, path->slots[0],
367 struct btrfs_root_ref);
368
369 WARN_ON(btrfs_root_ref_dirid(leaf, ref) != dirid);
370 WARN_ON(btrfs_root_ref_name_len(leaf, ref) != name_len);
371 ptr = (unsigned long)(ref + 1);
372 WARN_ON(memcmp_extent_buffer(leaf, name, ptr, name_len));
373 *sequence = btrfs_root_ref_sequence(leaf, ref);
374
375 ret = btrfs_del_item(trans, tree_root, path);
376 BUG_ON(ret);
377 } else
378 err = -ENOENT;
379
380 if (key.type == BTRFS_ROOT_BACKREF_KEY) {
381 btrfs_release_path(tree_root, path);
382 key.objectid = ref_id;
383 key.type = BTRFS_ROOT_REF_KEY;
384 key.offset = root_id;
385 goto again;
386 }
301 387
302 btrfs_free_path(path); 388 btrfs_free_path(path);
303 return ret; 389 return err;
304} 390}
305#endif
306 391
307int btrfs_find_root_ref(struct btrfs_root *tree_root, 392int btrfs_find_root_ref(struct btrfs_root *tree_root,
308 struct btrfs_path *path, 393 struct btrfs_path *path,
@@ -319,7 +404,6 @@ int btrfs_find_root_ref(struct btrfs_root *tree_root,
319 return ret; 404 return ret;
320} 405}
321 406
322
323/* 407/*
324 * add a btrfs_root_ref item. type is either BTRFS_ROOT_REF_KEY 408 * add a btrfs_root_ref item. type is either BTRFS_ROOT_REF_KEY
325 * or BTRFS_ROOT_BACKREF_KEY. 409 * or BTRFS_ROOT_BACKREF_KEY.
@@ -335,8 +419,7 @@ int btrfs_find_root_ref(struct btrfs_root *tree_root,
335 */ 419 */
336int btrfs_add_root_ref(struct btrfs_trans_handle *trans, 420int btrfs_add_root_ref(struct btrfs_trans_handle *trans,
337 struct btrfs_root *tree_root, 421 struct btrfs_root *tree_root,
338 u64 root_id, u8 type, u64 ref_id, 422 u64 root_id, u64 ref_id, u64 dirid, u64 sequence,
339 u64 dirid, u64 sequence,
340 const char *name, int name_len) 423 const char *name, int name_len)
341{ 424{
342 struct btrfs_key key; 425 struct btrfs_key key;
@@ -346,13 +429,14 @@ int btrfs_add_root_ref(struct btrfs_trans_handle *trans,
346 struct extent_buffer *leaf; 429 struct extent_buffer *leaf;
347 unsigned long ptr; 430 unsigned long ptr;
348 431
349
350 path = btrfs_alloc_path(); 432 path = btrfs_alloc_path();
433 if (!path)
434 return -ENOMEM;
351 435
352 key.objectid = root_id; 436 key.objectid = root_id;
353 key.type = type; 437 key.type = BTRFS_ROOT_BACKREF_KEY;
354 key.offset = ref_id; 438 key.offset = ref_id;
355 439again:
356 ret = btrfs_insert_empty_item(trans, tree_root, path, &key, 440 ret = btrfs_insert_empty_item(trans, tree_root, path, &key,
357 sizeof(*ref) + name_len); 441 sizeof(*ref) + name_len);
358 BUG_ON(ret); 442 BUG_ON(ret);
@@ -366,6 +450,14 @@ int btrfs_add_root_ref(struct btrfs_trans_handle *trans,
366 write_extent_buffer(leaf, name, ptr, name_len); 450 write_extent_buffer(leaf, name, ptr, name_len);
367 btrfs_mark_buffer_dirty(leaf); 451 btrfs_mark_buffer_dirty(leaf);
368 452
453 if (key.type == BTRFS_ROOT_BACKREF_KEY) {
454 btrfs_release_path(tree_root, path);
455 key.objectid = ref_id;
456 key.type = BTRFS_ROOT_REF_KEY;
457 key.offset = root_id;
458 goto again;
459 }
460
369 btrfs_free_path(path); 461 btrfs_free_path(path);
370 return ret; 462 return 0;
371} 463}
diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
index 6d6d06cb6dfc..9de9b2236419 100644
--- a/fs/btrfs/super.c
+++ b/fs/btrfs/super.c
@@ -51,7 +51,7 @@
51#include "export.h" 51#include "export.h"
52#include "compression.h" 52#include "compression.h"
53 53
54static struct super_operations btrfs_super_ops; 54static const struct super_operations btrfs_super_ops;
55 55
56static void btrfs_put_super(struct super_block *sb) 56static void btrfs_put_super(struct super_block *sb)
57{ 57{
@@ -344,7 +344,9 @@ static int btrfs_fill_super(struct super_block *sb,
344 sb->s_export_op = &btrfs_export_ops; 344 sb->s_export_op = &btrfs_export_ops;
345 sb->s_xattr = btrfs_xattr_handlers; 345 sb->s_xattr = btrfs_xattr_handlers;
346 sb->s_time_gran = 1; 346 sb->s_time_gran = 1;
347#ifdef CONFIG_BTRFS_POSIX_ACL
347 sb->s_flags |= MS_POSIXACL; 348 sb->s_flags |= MS_POSIXACL;
349#endif
348 350
349 tree_root = open_ctree(sb, fs_devices, (char *)data); 351 tree_root = open_ctree(sb, fs_devices, (char *)data);
350 352
@@ -675,7 +677,8 @@ static int btrfs_unfreeze(struct super_block *sb)
675 return 0; 677 return 0;
676} 678}
677 679
678static struct super_operations btrfs_super_ops = { 680static const struct super_operations btrfs_super_ops = {
681 .drop_inode = btrfs_drop_inode,
679 .delete_inode = btrfs_delete_inode, 682 .delete_inode = btrfs_delete_inode,
680 .put_super = btrfs_put_super, 683 .put_super = btrfs_put_super,
681 .sync_fs = btrfs_sync_fs, 684 .sync_fs = btrfs_sync_fs,
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index cdbb5022da52..0b8f36d4400a 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -104,7 +104,6 @@ static noinline int record_root_in_trans(struct btrfs_trans_handle *trans,
104{ 104{
105 if (root->ref_cows && root->last_trans < trans->transid) { 105 if (root->ref_cows && root->last_trans < trans->transid) {
106 WARN_ON(root == root->fs_info->extent_root); 106 WARN_ON(root == root->fs_info->extent_root);
107 WARN_ON(root->root_item.refs == 0);
108 WARN_ON(root->commit_root != root->node); 107 WARN_ON(root->commit_root != root->node);
109 108
110 radix_tree_tag_set(&root->fs_info->fs_roots_radix, 109 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
@@ -187,6 +186,9 @@ static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
187 h->alloc_exclude_start = 0; 186 h->alloc_exclude_start = 0;
188 h->delayed_ref_updates = 0; 187 h->delayed_ref_updates = 0;
189 188
189 if (!current->journal_info)
190 current->journal_info = h;
191
190 root->fs_info->running_transaction->use_count++; 192 root->fs_info->running_transaction->use_count++;
191 record_root_in_trans(h, root); 193 record_root_in_trans(h, root);
192 mutex_unlock(&root->fs_info->trans_mutex); 194 mutex_unlock(&root->fs_info->trans_mutex);
@@ -318,6 +320,9 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
318 wake_up(&cur_trans->writer_wait); 320 wake_up(&cur_trans->writer_wait);
319 put_transaction(cur_trans); 321 put_transaction(cur_trans);
320 mutex_unlock(&info->trans_mutex); 322 mutex_unlock(&info->trans_mutex);
323
324 if (current->journal_info == trans)
325 current->journal_info = NULL;
321 memset(trans, 0, sizeof(*trans)); 326 memset(trans, 0, sizeof(*trans));
322 kmem_cache_free(btrfs_trans_handle_cachep, trans); 327 kmem_cache_free(btrfs_trans_handle_cachep, trans);
323 328
@@ -720,7 +725,8 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
720 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 725 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
721 726
722 key.objectid = objectid; 727 key.objectid = objectid;
723 key.offset = 0; 728 /* record when the snapshot was created in key.offset */
729 key.offset = trans->transid;
724 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); 730 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
725 731
726 old = btrfs_lock_root_node(root); 732 old = btrfs_lock_root_node(root);
@@ -743,6 +749,7 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
743 memcpy(&pending->root_key, &key, sizeof(key)); 749 memcpy(&pending->root_key, &key, sizeof(key));
744fail: 750fail:
745 kfree(new_root_item); 751 kfree(new_root_item);
752 btrfs_unreserve_metadata_space(root, 6);
746 return ret; 753 return ret;
747} 754}
748 755
@@ -778,24 +785,14 @@ static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info,
778 ret = btrfs_update_inode(trans, parent_root, parent_inode); 785 ret = btrfs_update_inode(trans, parent_root, parent_inode);
779 BUG_ON(ret); 786 BUG_ON(ret);
780 787
781 /* add the backref first */
782 ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root, 788 ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
783 pending->root_key.objectid, 789 pending->root_key.objectid,
784 BTRFS_ROOT_BACKREF_KEY,
785 parent_root->root_key.objectid, 790 parent_root->root_key.objectid,
786 parent_inode->i_ino, index, pending->name, 791 parent_inode->i_ino, index, pending->name,
787 namelen); 792 namelen);
788 793
789 BUG_ON(ret); 794 BUG_ON(ret);
790 795
791 /* now add the forward ref */
792 ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
793 parent_root->root_key.objectid,
794 BTRFS_ROOT_REF_KEY,
795 pending->root_key.objectid,
796 parent_inode->i_ino, index, pending->name,
797 namelen);
798
799 inode = btrfs_lookup_dentry(parent_inode, pending->dentry); 796 inode = btrfs_lookup_dentry(parent_inode, pending->dentry);
800 d_instantiate(pending->dentry, inode); 797 d_instantiate(pending->dentry, inode);
801fail: 798fail:
@@ -874,7 +871,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
874 unsigned long timeout = 1; 871 unsigned long timeout = 1;
875 struct btrfs_transaction *cur_trans; 872 struct btrfs_transaction *cur_trans;
876 struct btrfs_transaction *prev_trans = NULL; 873 struct btrfs_transaction *prev_trans = NULL;
877 struct extent_io_tree *pinned_copy;
878 DEFINE_WAIT(wait); 874 DEFINE_WAIT(wait);
879 int ret; 875 int ret;
880 int should_grow = 0; 876 int should_grow = 0;
@@ -915,13 +911,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
915 return 0; 911 return 0;
916 } 912 }
917 913
918 pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
919 if (!pinned_copy)
920 return -ENOMEM;
921
922 extent_io_tree_init(pinned_copy,
923 root->fs_info->btree_inode->i_mapping, GFP_NOFS);
924
925 trans->transaction->in_commit = 1; 914 trans->transaction->in_commit = 1;
926 trans->transaction->blocked = 1; 915 trans->transaction->blocked = 1;
927 if (cur_trans->list.prev != &root->fs_info->trans_list) { 916 if (cur_trans->list.prev != &root->fs_info->trans_list) {
@@ -1019,6 +1008,8 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1019 ret = commit_cowonly_roots(trans, root); 1008 ret = commit_cowonly_roots(trans, root);
1020 BUG_ON(ret); 1009 BUG_ON(ret);
1021 1010
1011 btrfs_prepare_extent_commit(trans, root);
1012
1022 cur_trans = root->fs_info->running_transaction; 1013 cur_trans = root->fs_info->running_transaction;
1023 spin_lock(&root->fs_info->new_trans_lock); 1014 spin_lock(&root->fs_info->new_trans_lock);
1024 root->fs_info->running_transaction = NULL; 1015 root->fs_info->running_transaction = NULL;
@@ -1042,8 +1033,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1042 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, 1033 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1043 sizeof(root->fs_info->super_copy)); 1034 sizeof(root->fs_info->super_copy));
1044 1035
1045 btrfs_copy_pinned(root, pinned_copy);
1046
1047 trans->transaction->blocked = 0; 1036 trans->transaction->blocked = 0;
1048 1037
1049 wake_up(&root->fs_info->transaction_wait); 1038 wake_up(&root->fs_info->transaction_wait);
@@ -1059,8 +1048,7 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1059 */ 1048 */
1060 mutex_unlock(&root->fs_info->tree_log_mutex); 1049 mutex_unlock(&root->fs_info->tree_log_mutex);
1061 1050
1062 btrfs_finish_extent_commit(trans, root, pinned_copy); 1051 btrfs_finish_extent_commit(trans, root);
1063 kfree(pinned_copy);
1064 1052
1065 /* do the directory inserts of any pending snapshot creations */ 1053 /* do the directory inserts of any pending snapshot creations */
1066 finish_pending_snapshots(trans, root->fs_info); 1054 finish_pending_snapshots(trans, root->fs_info);
@@ -1078,6 +1066,9 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1078 1066
1079 mutex_unlock(&root->fs_info->trans_mutex); 1067 mutex_unlock(&root->fs_info->trans_mutex);
1080 1068
1069 if (current->journal_info == trans)
1070 current->journal_info = NULL;
1071
1081 kmem_cache_free(btrfs_trans_handle_cachep, trans); 1072 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1082 return ret; 1073 return ret;
1083} 1074}
@@ -1096,8 +1087,13 @@ int btrfs_clean_old_snapshots(struct btrfs_root *root)
1096 1087
1097 while (!list_empty(&list)) { 1088 while (!list_empty(&list)) {
1098 root = list_entry(list.next, struct btrfs_root, root_list); 1089 root = list_entry(list.next, struct btrfs_root, root_list);
1099 list_del_init(&root->root_list); 1090 list_del(&root->root_list);
1100 btrfs_drop_snapshot(root, 0); 1091
1092 if (btrfs_header_backref_rev(root->node) <
1093 BTRFS_MIXED_BACKREF_REV)
1094 btrfs_drop_snapshot(root, 0);
1095 else
1096 btrfs_drop_snapshot(root, 1);
1101 } 1097 }
1102 return 0; 1098 return 0;
1103} 1099}
diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c
index d91b0de7c502..4edfdc2acc5f 100644
--- a/fs/btrfs/tree-log.c
+++ b/fs/btrfs/tree-log.c
@@ -137,11 +137,20 @@ static int start_log_trans(struct btrfs_trans_handle *trans,
137 137
138 mutex_lock(&root->log_mutex); 138 mutex_lock(&root->log_mutex);
139 if (root->log_root) { 139 if (root->log_root) {
140 if (!root->log_start_pid) {
141 root->log_start_pid = current->pid;
142 root->log_multiple_pids = false;
143 } else if (root->log_start_pid != current->pid) {
144 root->log_multiple_pids = true;
145 }
146
140 root->log_batch++; 147 root->log_batch++;
141 atomic_inc(&root->log_writers); 148 atomic_inc(&root->log_writers);
142 mutex_unlock(&root->log_mutex); 149 mutex_unlock(&root->log_mutex);
143 return 0; 150 return 0;
144 } 151 }
152 root->log_multiple_pids = false;
153 root->log_start_pid = current->pid;
145 mutex_lock(&root->fs_info->tree_log_mutex); 154 mutex_lock(&root->fs_info->tree_log_mutex);
146 if (!root->fs_info->log_root_tree) { 155 if (!root->fs_info->log_root_tree) {
147 ret = btrfs_init_log_root_tree(trans, root->fs_info); 156 ret = btrfs_init_log_root_tree(trans, root->fs_info);
@@ -263,8 +272,8 @@ static int process_one_buffer(struct btrfs_root *log,
263 struct walk_control *wc, u64 gen) 272 struct walk_control *wc, u64 gen)
264{ 273{
265 if (wc->pin) 274 if (wc->pin)
266 btrfs_update_pinned_extents(log->fs_info->extent_root, 275 btrfs_pin_extent(log->fs_info->extent_root,
267 eb->start, eb->len, 1); 276 eb->start, eb->len, 0);
268 277
269 if (btrfs_buffer_uptodate(eb, gen)) { 278 if (btrfs_buffer_uptodate(eb, gen)) {
270 if (wc->write) 279 if (wc->write)
@@ -534,7 +543,7 @@ static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
534 saved_nbytes = inode_get_bytes(inode); 543 saved_nbytes = inode_get_bytes(inode);
535 /* drop any overlapping extents */ 544 /* drop any overlapping extents */
536 ret = btrfs_drop_extents(trans, root, inode, 545 ret = btrfs_drop_extents(trans, root, inode,
537 start, extent_end, extent_end, start, &alloc_hint); 546 start, extent_end, extent_end, start, &alloc_hint, 1);
538 BUG_ON(ret); 547 BUG_ON(ret);
539 548
540 if (found_type == BTRFS_FILE_EXTENT_REG || 549 if (found_type == BTRFS_FILE_EXTENT_REG ||
@@ -1985,7 +1994,7 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
1985 if (atomic_read(&root->log_commit[(index1 + 1) % 2])) 1994 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
1986 wait_log_commit(trans, root, root->log_transid - 1); 1995 wait_log_commit(trans, root, root->log_transid - 1);
1987 1996
1988 while (1) { 1997 while (root->log_multiple_pids) {
1989 unsigned long batch = root->log_batch; 1998 unsigned long batch = root->log_batch;
1990 mutex_unlock(&root->log_mutex); 1999 mutex_unlock(&root->log_mutex);
1991 schedule_timeout_uninterruptible(1); 2000 schedule_timeout_uninterruptible(1);
@@ -2011,6 +2020,7 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
2011 root->log_batch = 0; 2020 root->log_batch = 0;
2012 root->log_transid++; 2021 root->log_transid++;
2013 log->log_transid = root->log_transid; 2022 log->log_transid = root->log_transid;
2023 root->log_start_pid = 0;
2014 smp_mb(); 2024 smp_mb();
2015 /* 2025 /*
2016 * log tree has been flushed to disk, new modifications of 2026 * log tree has been flushed to disk, new modifications of
@@ -2605,7 +2615,7 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
2605 extent); 2615 extent);
2606 cs = btrfs_file_extent_offset(src, extent); 2616 cs = btrfs_file_extent_offset(src, extent);
2607 cl = btrfs_file_extent_num_bytes(src, 2617 cl = btrfs_file_extent_num_bytes(src,
2608 extent);; 2618 extent);
2609 if (btrfs_file_extent_compression(src, 2619 if (btrfs_file_extent_compression(src,
2610 extent)) { 2620 extent)) {
2611 cs = 0; 2621 cs = 0;
@@ -2841,7 +2851,7 @@ static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
2841 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb) 2851 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2842 break; 2852 break;
2843 2853
2844 if (parent == sb->s_root) 2854 if (IS_ROOT(parent))
2845 break; 2855 break;
2846 2856
2847 parent = parent->d_parent; 2857 parent = parent->d_parent;
@@ -2880,6 +2890,12 @@ int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
2880 goto end_no_trans; 2890 goto end_no_trans;
2881 } 2891 }
2882 2892
2893 if (root != BTRFS_I(inode)->root ||
2894 btrfs_root_refs(&root->root_item) == 0) {
2895 ret = 1;
2896 goto end_no_trans;
2897 }
2898
2883 ret = check_parent_dirs_for_sync(trans, inode, parent, 2899 ret = check_parent_dirs_for_sync(trans, inode, parent,
2884 sb, last_committed); 2900 sb, last_committed);
2885 if (ret) 2901 if (ret)
@@ -2907,12 +2923,15 @@ int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
2907 break; 2923 break;
2908 2924
2909 inode = parent->d_inode; 2925 inode = parent->d_inode;
2926 if (root != BTRFS_I(inode)->root)
2927 break;
2928
2910 if (BTRFS_I(inode)->generation > 2929 if (BTRFS_I(inode)->generation >
2911 root->fs_info->last_trans_committed) { 2930 root->fs_info->last_trans_committed) {
2912 ret = btrfs_log_inode(trans, root, inode, inode_only); 2931 ret = btrfs_log_inode(trans, root, inode, inode_only);
2913 BUG_ON(ret); 2932 BUG_ON(ret);
2914 } 2933 }
2915 if (parent == sb->s_root) 2934 if (IS_ROOT(parent))
2916 break; 2935 break;
2917 2936
2918 parent = parent->d_parent; 2937 parent = parent->d_parent;
@@ -2951,7 +2970,6 @@ int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
2951 struct btrfs_key tmp_key; 2970 struct btrfs_key tmp_key;
2952 struct btrfs_root *log; 2971 struct btrfs_root *log;
2953 struct btrfs_fs_info *fs_info = log_root_tree->fs_info; 2972 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
2954 u64 highest_inode;
2955 struct walk_control wc = { 2973 struct walk_control wc = {
2956 .process_func = process_one_buffer, 2974 .process_func = process_one_buffer,
2957 .stage = 0, 2975 .stage = 0,
@@ -3010,11 +3028,6 @@ again:
3010 path); 3028 path);
3011 BUG_ON(ret); 3029 BUG_ON(ret);
3012 } 3030 }
3013 ret = btrfs_find_highest_inode(wc.replay_dest, &highest_inode);
3014 if (ret == 0) {
3015 wc.replay_dest->highest_inode = highest_inode;
3016 wc.replay_dest->last_inode_alloc = highest_inode;
3017 }
3018 3031
3019 key.offset = found_key.offset - 1; 3032 key.offset = found_key.offset - 1;
3020 wc.replay_dest->log_root = NULL; 3033 wc.replay_dest->log_root = NULL;
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 5dbefd11b4af..7eda483d7b5a 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -260,7 +260,7 @@ loop_lock:
260 num_run++; 260 num_run++;
261 batch_run++; 261 batch_run++;
262 262
263 if (bio_sync(cur)) 263 if (bio_rw_flagged(cur, BIO_RW_SYNCIO))
264 num_sync_run++; 264 num_sync_run++;
265 265
266 if (need_resched()) { 266 if (need_resched()) {
@@ -276,7 +276,7 @@ loop_lock:
276 * is now congested. Back off and let other work structs 276 * is now congested. Back off and let other work structs
277 * run instead 277 * run instead
278 */ 278 */
279 if (pending && bdi_write_congested(bdi) && batch_run > 32 && 279 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
280 fs_info->fs_devices->open_devices > 1) { 280 fs_info->fs_devices->open_devices > 1) {
281 struct io_context *ioc; 281 struct io_context *ioc;
282 282
@@ -446,8 +446,10 @@ static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
446 goto error; 446 goto error;
447 447
448 device->name = kstrdup(orig_dev->name, GFP_NOFS); 448 device->name = kstrdup(orig_dev->name, GFP_NOFS);
449 if (!device->name) 449 if (!device->name) {
450 kfree(device);
450 goto error; 451 goto error;
452 }
451 453
452 device->devid = orig_dev->devid; 454 device->devid = orig_dev->devid;
453 device->work.func = pending_bios_fn; 455 device->work.func = pending_bios_fn;
@@ -719,10 +721,9 @@ error:
719 * called very infrequently and that a given device has a small number 721 * called very infrequently and that a given device has a small number
720 * of extents 722 * of extents
721 */ 723 */
722static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans, 724int find_free_dev_extent(struct btrfs_trans_handle *trans,
723 struct btrfs_device *device, 725 struct btrfs_device *device, u64 num_bytes,
724 u64 num_bytes, u64 *start, 726 u64 *start, u64 *max_avail)
725 u64 *max_avail)
726{ 727{
727 struct btrfs_key key; 728 struct btrfs_key key;
728 struct btrfs_root *root = device->dev_root; 729 struct btrfs_root *root = device->dev_root;
@@ -1736,6 +1737,10 @@ static int btrfs_relocate_chunk(struct btrfs_root *root,
1736 extent_root = root->fs_info->extent_root; 1737 extent_root = root->fs_info->extent_root;
1737 em_tree = &root->fs_info->mapping_tree.map_tree; 1738 em_tree = &root->fs_info->mapping_tree.map_tree;
1738 1739
1740 ret = btrfs_can_relocate(extent_root, chunk_offset);
1741 if (ret)
1742 return -ENOSPC;
1743
1739 /* step one, relocate all the extents inside this chunk */ 1744 /* step one, relocate all the extents inside this chunk */
1740 ret = btrfs_relocate_block_group(extent_root, chunk_offset); 1745 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1741 BUG_ON(ret); 1746 BUG_ON(ret);
@@ -1749,9 +1754,9 @@ static int btrfs_relocate_chunk(struct btrfs_root *root,
1749 * step two, delete the device extents and the 1754 * step two, delete the device extents and the
1750 * chunk tree entries 1755 * chunk tree entries
1751 */ 1756 */
1752 spin_lock(&em_tree->lock); 1757 read_lock(&em_tree->lock);
1753 em = lookup_extent_mapping(em_tree, chunk_offset, 1); 1758 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1754 spin_unlock(&em_tree->lock); 1759 read_unlock(&em_tree->lock);
1755 1760
1756 BUG_ON(em->start > chunk_offset || 1761 BUG_ON(em->start > chunk_offset ||
1757 em->start + em->len < chunk_offset); 1762 em->start + em->len < chunk_offset);
@@ -1780,9 +1785,9 @@ static int btrfs_relocate_chunk(struct btrfs_root *root,
1780 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset); 1785 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1781 BUG_ON(ret); 1786 BUG_ON(ret);
1782 1787
1783 spin_lock(&em_tree->lock); 1788 write_lock(&em_tree->lock);
1784 remove_extent_mapping(em_tree, em); 1789 remove_extent_mapping(em_tree, em);
1785 spin_unlock(&em_tree->lock); 1790 write_unlock(&em_tree->lock);
1786 1791
1787 kfree(map); 1792 kfree(map);
1788 em->bdev = NULL; 1793 em->bdev = NULL;
@@ -1807,12 +1812,15 @@ static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1807 struct btrfs_key found_key; 1812 struct btrfs_key found_key;
1808 u64 chunk_tree = chunk_root->root_key.objectid; 1813 u64 chunk_tree = chunk_root->root_key.objectid;
1809 u64 chunk_type; 1814 u64 chunk_type;
1815 bool retried = false;
1816 int failed = 0;
1810 int ret; 1817 int ret;
1811 1818
1812 path = btrfs_alloc_path(); 1819 path = btrfs_alloc_path();
1813 if (!path) 1820 if (!path)
1814 return -ENOMEM; 1821 return -ENOMEM;
1815 1822
1823again:
1816 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 1824 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1817 key.offset = (u64)-1; 1825 key.offset = (u64)-1;
1818 key.type = BTRFS_CHUNK_ITEM_KEY; 1826 key.type = BTRFS_CHUNK_ITEM_KEY;
@@ -1842,7 +1850,10 @@ static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1842 ret = btrfs_relocate_chunk(chunk_root, chunk_tree, 1850 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1843 found_key.objectid, 1851 found_key.objectid,
1844 found_key.offset); 1852 found_key.offset);
1845 BUG_ON(ret); 1853 if (ret == -ENOSPC)
1854 failed++;
1855 else if (ret)
1856 BUG();
1846 } 1857 }
1847 1858
1848 if (found_key.offset == 0) 1859 if (found_key.offset == 0)
@@ -1850,6 +1861,14 @@ static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1850 key.offset = found_key.offset - 1; 1861 key.offset = found_key.offset - 1;
1851 } 1862 }
1852 ret = 0; 1863 ret = 0;
1864 if (failed && !retried) {
1865 failed = 0;
1866 retried = true;
1867 goto again;
1868 } else if (failed && retried) {
1869 WARN_ON(1);
1870 ret = -ENOSPC;
1871 }
1853error: 1872error:
1854 btrfs_free_path(path); 1873 btrfs_free_path(path);
1855 return ret; 1874 return ret;
@@ -1894,6 +1913,8 @@ int btrfs_balance(struct btrfs_root *dev_root)
1894 continue; 1913 continue;
1895 1914
1896 ret = btrfs_shrink_device(device, old_size - size_to_free); 1915 ret = btrfs_shrink_device(device, old_size - size_to_free);
1916 if (ret == -ENOSPC)
1917 break;
1897 BUG_ON(ret); 1918 BUG_ON(ret);
1898 1919
1899 trans = btrfs_start_transaction(dev_root, 1); 1920 trans = btrfs_start_transaction(dev_root, 1);
@@ -1938,9 +1959,8 @@ int btrfs_balance(struct btrfs_root *dev_root)
1938 chunk = btrfs_item_ptr(path->nodes[0], 1959 chunk = btrfs_item_ptr(path->nodes[0],
1939 path->slots[0], 1960 path->slots[0],
1940 struct btrfs_chunk); 1961 struct btrfs_chunk);
1941 key.offset = found_key.offset;
1942 /* chunk zero is special */ 1962 /* chunk zero is special */
1943 if (key.offset == 0) 1963 if (found_key.offset == 0)
1944 break; 1964 break;
1945 1965
1946 btrfs_release_path(chunk_root, path); 1966 btrfs_release_path(chunk_root, path);
@@ -1948,7 +1968,8 @@ int btrfs_balance(struct btrfs_root *dev_root)
1948 chunk_root->root_key.objectid, 1968 chunk_root->root_key.objectid,
1949 found_key.objectid, 1969 found_key.objectid,
1950 found_key.offset); 1970 found_key.offset);
1951 BUG_ON(ret); 1971 BUG_ON(ret && ret != -ENOSPC);
1972 key.offset = found_key.offset - 1;
1952 } 1973 }
1953 ret = 0; 1974 ret = 0;
1954error: 1975error:
@@ -1974,10 +1995,13 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1974 u64 chunk_offset; 1995 u64 chunk_offset;
1975 int ret; 1996 int ret;
1976 int slot; 1997 int slot;
1998 int failed = 0;
1999 bool retried = false;
1977 struct extent_buffer *l; 2000 struct extent_buffer *l;
1978 struct btrfs_key key; 2001 struct btrfs_key key;
1979 struct btrfs_super_block *super_copy = &root->fs_info->super_copy; 2002 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1980 u64 old_total = btrfs_super_total_bytes(super_copy); 2003 u64 old_total = btrfs_super_total_bytes(super_copy);
2004 u64 old_size = device->total_bytes;
1981 u64 diff = device->total_bytes - new_size; 2005 u64 diff = device->total_bytes - new_size;
1982 2006
1983 if (new_size >= device->total_bytes) 2007 if (new_size >= device->total_bytes)
@@ -1987,12 +2011,6 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1987 if (!path) 2011 if (!path)
1988 return -ENOMEM; 2012 return -ENOMEM;
1989 2013
1990 trans = btrfs_start_transaction(root, 1);
1991 if (!trans) {
1992 ret = -ENOMEM;
1993 goto done;
1994 }
1995
1996 path->reada = 2; 2014 path->reada = 2;
1997 2015
1998 lock_chunks(root); 2016 lock_chunks(root);
@@ -2001,8 +2019,8 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2001 if (device->writeable) 2019 if (device->writeable)
2002 device->fs_devices->total_rw_bytes -= diff; 2020 device->fs_devices->total_rw_bytes -= diff;
2003 unlock_chunks(root); 2021 unlock_chunks(root);
2004 btrfs_end_transaction(trans, root);
2005 2022
2023again:
2006 key.objectid = device->devid; 2024 key.objectid = device->devid;
2007 key.offset = (u64)-1; 2025 key.offset = (u64)-1;
2008 key.type = BTRFS_DEV_EXTENT_KEY; 2026 key.type = BTRFS_DEV_EXTENT_KEY;
@@ -2017,6 +2035,7 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2017 goto done; 2035 goto done;
2018 if (ret) { 2036 if (ret) {
2019 ret = 0; 2037 ret = 0;
2038 btrfs_release_path(root, path);
2020 break; 2039 break;
2021 } 2040 }
2022 2041
@@ -2024,14 +2043,18 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2024 slot = path->slots[0]; 2043 slot = path->slots[0];
2025 btrfs_item_key_to_cpu(l, &key, path->slots[0]); 2044 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2026 2045
2027 if (key.objectid != device->devid) 2046 if (key.objectid != device->devid) {
2047 btrfs_release_path(root, path);
2028 break; 2048 break;
2049 }
2029 2050
2030 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 2051 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2031 length = btrfs_dev_extent_length(l, dev_extent); 2052 length = btrfs_dev_extent_length(l, dev_extent);
2032 2053
2033 if (key.offset + length <= new_size) 2054 if (key.offset + length <= new_size) {
2055 btrfs_release_path(root, path);
2034 break; 2056 break;
2057 }
2035 2058
2036 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); 2059 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2037 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); 2060 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
@@ -2040,8 +2063,26 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2040 2063
2041 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid, 2064 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2042 chunk_offset); 2065 chunk_offset);
2043 if (ret) 2066 if (ret && ret != -ENOSPC)
2044 goto done; 2067 goto done;
2068 if (ret == -ENOSPC)
2069 failed++;
2070 key.offset -= 1;
2071 }
2072
2073 if (failed && !retried) {
2074 failed = 0;
2075 retried = true;
2076 goto again;
2077 } else if (failed && retried) {
2078 ret = -ENOSPC;
2079 lock_chunks(root);
2080
2081 device->total_bytes = old_size;
2082 if (device->writeable)
2083 device->fs_devices->total_rw_bytes += diff;
2084 unlock_chunks(root);
2085 goto done;
2045 } 2086 }
2046 2087
2047 /* Shrinking succeeded, else we would be at "done". */ 2088 /* Shrinking succeeded, else we would be at "done". */
@@ -2294,9 +2335,9 @@ again:
2294 em->block_len = em->len; 2335 em->block_len = em->len;
2295 2336
2296 em_tree = &extent_root->fs_info->mapping_tree.map_tree; 2337 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2297 spin_lock(&em_tree->lock); 2338 write_lock(&em_tree->lock);
2298 ret = add_extent_mapping(em_tree, em); 2339 ret = add_extent_mapping(em_tree, em);
2299 spin_unlock(&em_tree->lock); 2340 write_unlock(&em_tree->lock);
2300 BUG_ON(ret); 2341 BUG_ON(ret);
2301 free_extent_map(em); 2342 free_extent_map(em);
2302 2343
@@ -2491,9 +2532,9 @@ int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2491 int readonly = 0; 2532 int readonly = 0;
2492 int i; 2533 int i;
2493 2534
2494 spin_lock(&map_tree->map_tree.lock); 2535 read_lock(&map_tree->map_tree.lock);
2495 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); 2536 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2496 spin_unlock(&map_tree->map_tree.lock); 2537 read_unlock(&map_tree->map_tree.lock);
2497 if (!em) 2538 if (!em)
2498 return 1; 2539 return 1;
2499 2540
@@ -2518,11 +2559,11 @@ void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2518 struct extent_map *em; 2559 struct extent_map *em;
2519 2560
2520 while (1) { 2561 while (1) {
2521 spin_lock(&tree->map_tree.lock); 2562 write_lock(&tree->map_tree.lock);
2522 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); 2563 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2523 if (em) 2564 if (em)
2524 remove_extent_mapping(&tree->map_tree, em); 2565 remove_extent_mapping(&tree->map_tree, em);
2525 spin_unlock(&tree->map_tree.lock); 2566 write_unlock(&tree->map_tree.lock);
2526 if (!em) 2567 if (!em)
2527 break; 2568 break;
2528 kfree(em->bdev); 2569 kfree(em->bdev);
@@ -2540,9 +2581,9 @@ int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2540 struct extent_map_tree *em_tree = &map_tree->map_tree; 2581 struct extent_map_tree *em_tree = &map_tree->map_tree;
2541 int ret; 2582 int ret;
2542 2583
2543 spin_lock(&em_tree->lock); 2584 read_lock(&em_tree->lock);
2544 em = lookup_extent_mapping(em_tree, logical, len); 2585 em = lookup_extent_mapping(em_tree, logical, len);
2545 spin_unlock(&em_tree->lock); 2586 read_unlock(&em_tree->lock);
2546 BUG_ON(!em); 2587 BUG_ON(!em);
2547 2588
2548 BUG_ON(em->start > logical || em->start + em->len < logical); 2589 BUG_ON(em->start > logical || em->start + em->len < logical);
@@ -2604,9 +2645,9 @@ again:
2604 atomic_set(&multi->error, 0); 2645 atomic_set(&multi->error, 0);
2605 } 2646 }
2606 2647
2607 spin_lock(&em_tree->lock); 2648 read_lock(&em_tree->lock);
2608 em = lookup_extent_mapping(em_tree, logical, *length); 2649 em = lookup_extent_mapping(em_tree, logical, *length);
2609 spin_unlock(&em_tree->lock); 2650 read_unlock(&em_tree->lock);
2610 2651
2611 if (!em && unplug_page) 2652 if (!em && unplug_page)
2612 return 0; 2653 return 0;
@@ -2763,9 +2804,9 @@ int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2763 u64 stripe_nr; 2804 u64 stripe_nr;
2764 int i, j, nr = 0; 2805 int i, j, nr = 0;
2765 2806
2766 spin_lock(&em_tree->lock); 2807 read_lock(&em_tree->lock);
2767 em = lookup_extent_mapping(em_tree, chunk_start, 1); 2808 em = lookup_extent_mapping(em_tree, chunk_start, 1);
2768 spin_unlock(&em_tree->lock); 2809 read_unlock(&em_tree->lock);
2769 2810
2770 BUG_ON(!em || em->start != chunk_start); 2811 BUG_ON(!em || em->start != chunk_start);
2771 map = (struct map_lookup *)em->bdev; 2812 map = (struct map_lookup *)em->bdev;
@@ -2903,7 +2944,7 @@ static noinline int schedule_bio(struct btrfs_root *root,
2903 bio->bi_rw |= rw; 2944 bio->bi_rw |= rw;
2904 2945
2905 spin_lock(&device->io_lock); 2946 spin_lock(&device->io_lock);
2906 if (bio_sync(bio)) 2947 if (bio_rw_flagged(bio, BIO_RW_SYNCIO))
2907 pending_bios = &device->pending_sync_bios; 2948 pending_bios = &device->pending_sync_bios;
2908 else 2949 else
2909 pending_bios = &device->pending_bios; 2950 pending_bios = &device->pending_bios;
@@ -3053,9 +3094,9 @@ static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3053 logical = key->offset; 3094 logical = key->offset;
3054 length = btrfs_chunk_length(leaf, chunk); 3095 length = btrfs_chunk_length(leaf, chunk);
3055 3096
3056 spin_lock(&map_tree->map_tree.lock); 3097 read_lock(&map_tree->map_tree.lock);
3057 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); 3098 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3058 spin_unlock(&map_tree->map_tree.lock); 3099 read_unlock(&map_tree->map_tree.lock);
3059 3100
3060 /* already mapped? */ 3101 /* already mapped? */
3061 if (em && em->start <= logical && em->start + em->len > logical) { 3102 if (em && em->start <= logical && em->start + em->len > logical) {
@@ -3114,9 +3155,9 @@ static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3114 map->stripes[i].dev->in_fs_metadata = 1; 3155 map->stripes[i].dev->in_fs_metadata = 1;
3115 } 3156 }
3116 3157
3117 spin_lock(&map_tree->map_tree.lock); 3158 write_lock(&map_tree->map_tree.lock);
3118 ret = add_extent_mapping(&map_tree->map_tree, em); 3159 ret = add_extent_mapping(&map_tree->map_tree, em);
3119 spin_unlock(&map_tree->map_tree.lock); 3160 write_unlock(&map_tree->map_tree.lock);
3120 BUG_ON(ret); 3161 BUG_ON(ret);
3121 free_extent_map(em); 3162 free_extent_map(em);
3122 3163
diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
index 5139a833f721..31b0fabdd2ea 100644
--- a/fs/btrfs/volumes.h
+++ b/fs/btrfs/volumes.h
@@ -181,4 +181,7 @@ int btrfs_balance(struct btrfs_root *dev_root);
181void btrfs_unlock_volumes(void); 181void btrfs_unlock_volumes(void);
182void btrfs_lock_volumes(void); 182void btrfs_lock_volumes(void);
183int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset); 183int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset);
184int find_free_dev_extent(struct btrfs_trans_handle *trans,
185 struct btrfs_device *device, u64 num_bytes,
186 u64 *start, u64 *max_avail);
184#endif 187#endif
diff --git a/fs/btrfs/xattr.c b/fs/btrfs/xattr.c
index a9d3bf4d2689..b0fc93f95fd0 100644
--- a/fs/btrfs/xattr.c
+++ b/fs/btrfs/xattr.c
@@ -260,7 +260,7 @@ err:
260 * attributes are handled directly. 260 * attributes are handled directly.
261 */ 261 */
262struct xattr_handler *btrfs_xattr_handlers[] = { 262struct xattr_handler *btrfs_xattr_handlers[] = {
263#ifdef CONFIG_FS_POSIX_ACL 263#ifdef CONFIG_BTRFS_POSIX_ACL
264 &btrfs_xattr_acl_access_handler, 264 &btrfs_xattr_acl_access_handler,
265 &btrfs_xattr_acl_default_handler, 265 &btrfs_xattr_acl_default_handler,
266#endif 266#endif
generated by cgit v1.2.2 (git 2.25.0) at 2025-12-31 20:12:03 -0500