aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2009-09-24 11:57:29 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2009-09-24 11:57:29 -0400
commitdc2af6a6bcf3abdf44ac545759a6547dfe12070e (patch)
tree068ea2c5f5df55f72167ab3b51e001a98b7300e4
parent6c5daf012c9155aafd2c7973e4278766c30dfad0 (diff)
parent54bcf382daf08c1396edb8b81e650b58930ccaef (diff)
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable: (42 commits) Btrfs: hash the btree inode during fill_super Btrfs: relocate file extents in clusters Btrfs: don't rename file into dummy directory Btrfs: check size of inode backref before adding hardlink Btrfs: fix releasepage to avoid unlocking extents we haven't locked Btrfs: Fix test_range_bit for whole file extents Btrfs: fix errors handling cached state in set/clear_extent_bit Btrfs: fix early enospc during balancing Btrfs: deal with NULL space info Btrfs: account for space used by the super mirrors Btrfs: fix extent entry threshold calculation Btrfs: remove dead code Btrfs: fix bitmap size tracking Btrfs: don't keep retrying a block group if we fail to allocate a cluster Btrfs: make balance code choose more wisely when relocating Btrfs: fix arithmetic error in clone ioctl Btrfs: add snapshot/subvolume destroy ioctl Btrfs: change how subvolumes are organized Btrfs: do not reuse objectid of deleted snapshot/subvol Btrfs: speed up snapshot dropping ...
Diffstat
-rw-r--r--fs/btrfs/async-thread.c254
-rw-r--r--fs/btrfs/async-thread.h12
-rw-r--r--fs/btrfs/btrfs_inode.h1
-rw-r--r--fs/btrfs/compression.c8
-rw-r--r--fs/btrfs/ctree.c6
-rw-r--r--fs/btrfs/ctree.h78
-rw-r--r--fs/btrfs/dir-item.c47
-rw-r--r--fs/btrfs/disk-io.c235
-rw-r--r--fs/btrfs/export.c133
-rw-r--r--fs/btrfs/extent-tree.c1662
-rw-r--r--fs/btrfs/extent_io.c330
-rw-r--r--fs/btrfs/extent_io.h16
-rw-r--r--fs/btrfs/extent_map.c103
-rw-r--r--fs/btrfs/extent_map.h5
-rw-r--r--fs/btrfs/file.c35
-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.c663
-rw-r--r--fs/btrfs/ioctl.c339
-rw-r--r--fs/btrfs/ioctl.h3
-rw-r--r--fs/btrfs/ordered-data.c33
-rw-r--r--fs/btrfs/ordered-data.h3
-rw-r--r--fs/btrfs/orphan.c20
-rw-r--r--fs/btrfs/relocation.c280
-rw-r--r--fs/btrfs/root-tree.c138
-rw-r--r--fs/btrfs/super.c1
-rw-r--r--fs/btrfs/transaction.c38
-rw-r--r--fs/btrfs/tree-log.c25
-rw-r--r--fs/btrfs/volumes.c117
-rw-r--r--fs/btrfs/volumes.h3
31 files changed, 2740 insertions, 1981 deletions
diff --git a/fs/btrfs/async-thread.c b/fs/btrfs/async-thread.c
index 019e8af449ab..282ca085c2fb 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. */
@@ -71,7 +74,12 @@ static void check_idle_worker(struct btrfs_worker_thread *worker)
71 unsigned long flags; 74 unsigned long flags;
72 spin_lock_irqsave(&worker->workers->lock, flags); 75 spin_lock_irqsave(&worker->workers->lock, flags);
73 worker->idle = 1; 76 worker->idle = 1;
74 list_move(&worker->worker_list, &worker->workers->idle_list); 77
78 /* the list may be empty if the worker is just starting */
79 if (!list_empty(&worker->worker_list)) {
80 list_move(&worker->worker_list,
81 &worker->workers->idle_list);
82 }
75 spin_unlock_irqrestore(&worker->workers->lock, flags); 83 spin_unlock_irqrestore(&worker->workers->lock, flags);
76 } 84 }
77} 85}
@@ -87,23 +95,49 @@ static void check_busy_worker(struct btrfs_worker_thread *worker)
87 unsigned long flags; 95 unsigned long flags;
88 spin_lock_irqsave(&worker->workers->lock, flags); 96 spin_lock_irqsave(&worker->workers->lock, flags);
89 worker->idle = 0; 97 worker->idle = 0;
90 list_move_tail(&worker->worker_list, 98
91 &worker->workers->worker_list); 99 if (!list_empty(&worker->worker_list)) {
100 list_move_tail(&worker->worker_list,
101 &worker->workers->worker_list);
102 }
92 spin_unlock_irqrestore(&worker->workers->lock, flags); 103 spin_unlock_irqrestore(&worker->workers->lock, flags);
93 } 104 }
94} 105}
95 106
96static noinline int run_ordered_completions(struct btrfs_workers *workers, 107static void check_pending_worker_creates(struct btrfs_worker_thread *worker)
97 struct btrfs_work *work)
98{ 108{
109 struct btrfs_workers *workers = worker->workers;
99 unsigned long flags; 110 unsigned long flags;
100 111
112 rmb();
113 if (!workers->atomic_start_pending)
114 return;
115
116 spin_lock_irqsave(&workers->lock, flags);
117 if (!workers->atomic_start_pending)
118 goto out;
119
120 workers->atomic_start_pending = 0;
121 if (workers->num_workers >= workers->max_workers)
122 goto out;
123
124 spin_unlock_irqrestore(&workers->lock, flags);
125 btrfs_start_workers(workers, 1);
126 return;
127
128out:
129 spin_unlock_irqrestore(&workers->lock, flags);
130}
131
132static noinline int run_ordered_completions(struct btrfs_workers *workers,
133 struct btrfs_work *work)
134{
101 if (!workers->ordered) 135 if (!workers->ordered)
102 return 0; 136 return 0;
103 137
104 set_bit(WORK_DONE_BIT, &work->flags); 138 set_bit(WORK_DONE_BIT, &work->flags);
105 139
106 spin_lock_irqsave(&workers->lock, flags); 140 spin_lock(&workers->order_lock);
107 141
108 while (1) { 142 while (1) {
109 if (!list_empty(&workers->prio_order_list)) { 143 if (!list_empty(&workers->prio_order_list)) {
@@ -126,45 +160,118 @@ static noinline int run_ordered_completions(struct btrfs_workers *workers,
126 if (test_and_set_bit(WORK_ORDER_DONE_BIT, &work->flags)) 160 if (test_and_set_bit(WORK_ORDER_DONE_BIT, &work->flags))
127 break; 161 break;
128 162
129 spin_unlock_irqrestore(&workers->lock, flags); 163 spin_unlock(&workers->order_lock);
130 164
131 work->ordered_func(work); 165 work->ordered_func(work);
132 166
133 /* now take the lock again and call the freeing code */ 167 /* now take the lock again and call the freeing code */
134 spin_lock_irqsave(&workers->lock, flags); 168 spin_lock(&workers->order_lock);
135 list_del(&work->order_list); 169 list_del(&work->order_list);
136 work->ordered_free(work); 170 work->ordered_free(work);
137 } 171 }
138 172
139 spin_unlock_irqrestore(&workers->lock, flags); 173 spin_unlock(&workers->order_lock);
140 return 0; 174 return 0;
141} 175}
142 176
177static void put_worker(struct btrfs_worker_thread *worker)
178{
179 if (atomic_dec_and_test(&worker->refs))
180 kfree(worker);
181}
182
183static int try_worker_shutdown(struct btrfs_worker_thread *worker)
184{
185 int freeit = 0;
186
187 spin_lock_irq(&worker->lock);
188 spin_lock(&worker->workers->lock);
189 if (worker->workers->num_workers > 1 &&
190 worker->idle &&
191 !worker->working &&
192 !list_empty(&worker->worker_list) &&
193 list_empty(&worker->prio_pending) &&
194 list_empty(&worker->pending) &&
195 atomic_read(&worker->num_pending) == 0) {
196 freeit = 1;
197 list_del_init(&worker->worker_list);
198 worker->workers->num_workers--;
199 }
200 spin_unlock(&worker->workers->lock);
201 spin_unlock_irq(&worker->lock);
202
203 if (freeit)
204 put_worker(worker);
205 return freeit;
206}
207
208static struct btrfs_work *get_next_work(struct btrfs_worker_thread *worker,
209 struct list_head *prio_head,
210 struct list_head *head)
211{
212 struct btrfs_work *work = NULL;
213 struct list_head *cur = NULL;
214
215 if(!list_empty(prio_head))
216 cur = prio_head->next;
217
218 smp_mb();
219 if (!list_empty(&worker->prio_pending))
220 goto refill;
221
222 if (!list_empty(head))
223 cur = head->next;
224
225 if (cur)
226 goto out;
227
228refill:
229 spin_lock_irq(&worker->lock);
230 list_splice_tail_init(&worker->prio_pending, prio_head);
231 list_splice_tail_init(&worker->pending, head);
232
233 if (!list_empty(prio_head))
234 cur = prio_head->next;
235 else if (!list_empty(head))
236 cur = head->next;
237 spin_unlock_irq(&worker->lock);
238
239 if (!cur)
240 goto out_fail;
241
242out:
243 work = list_entry(cur, struct btrfs_work, list);
244
245out_fail:
246 return work;
247}
248
143/* 249/*
144 * main loop for servicing work items 250 * main loop for servicing work items
145 */ 251 */
146static int worker_loop(void *arg) 252static int worker_loop(void *arg)
147{ 253{
148 struct btrfs_worker_thread *worker = arg; 254 struct btrfs_worker_thread *worker = arg;
149 struct list_head *cur; 255 struct list_head head;
256 struct list_head prio_head;
150 struct btrfs_work *work; 257 struct btrfs_work *work;
258
259 INIT_LIST_HEAD(&head);
260 INIT_LIST_HEAD(&prio_head);
261
151 do { 262 do {
152 spin_lock_irq(&worker->lock); 263again:
153again_locked:
154 while (1) { 264 while (1) {
155 if (!list_empty(&worker->prio_pending)) 265
156 cur = worker->prio_pending.next; 266
157 else if (!list_empty(&worker->pending)) 267 work = get_next_work(worker, &prio_head, &head);
158 cur = worker->pending.next; 268 if (!work)
159 else
160 break; 269 break;
161 270
162 work = list_entry(cur, struct btrfs_work, list);
163 list_del(&work->list); 271 list_del(&work->list);
164 clear_bit(WORK_QUEUED_BIT, &work->flags); 272 clear_bit(WORK_QUEUED_BIT, &work->flags);
165 273
166 work->worker = worker; 274 work->worker = worker;
167 spin_unlock_irq(&worker->lock);
168 275
169 work->func(work); 276 work->func(work);
170 277
@@ -175,9 +282,13 @@ again_locked:
175 */ 282 */
176 run_ordered_completions(worker->workers, work); 283 run_ordered_completions(worker->workers, work);
177 284
178 spin_lock_irq(&worker->lock); 285 check_pending_worker_creates(worker);
179 check_idle_worker(worker); 286
180 } 287 }
288
289 spin_lock_irq(&worker->lock);
290 check_idle_worker(worker);
291
181 if (freezing(current)) { 292 if (freezing(current)) {
182 worker->working = 0; 293 worker->working = 0;
183 spin_unlock_irq(&worker->lock); 294 spin_unlock_irq(&worker->lock);
@@ -216,8 +327,10 @@ again_locked:
216 spin_lock_irq(&worker->lock); 327 spin_lock_irq(&worker->lock);
217 set_current_state(TASK_INTERRUPTIBLE); 328 set_current_state(TASK_INTERRUPTIBLE);
218 if (!list_empty(&worker->pending) || 329 if (!list_empty(&worker->pending) ||
219 !list_empty(&worker->prio_pending)) 330 !list_empty(&worker->prio_pending)) {
220 goto again_locked; 331 spin_unlock_irq(&worker->lock);
332 goto again;
333 }
221 334
222 /* 335 /*
223 * this makes sure we get a wakeup when someone 336 * this makes sure we get a wakeup when someone
@@ -226,8 +339,13 @@ again_locked:
226 worker->working = 0; 339 worker->working = 0;
227 spin_unlock_irq(&worker->lock); 340 spin_unlock_irq(&worker->lock);
228 341
229 if (!kthread_should_stop()) 342 if (!kthread_should_stop()) {
230 schedule(); 343 schedule_timeout(HZ * 120);
344 if (!worker->working &&
345 try_worker_shutdown(worker)) {
346 return 0;
347 }
348 }
231 } 349 }
232 __set_current_state(TASK_RUNNING); 350 __set_current_state(TASK_RUNNING);
233 } 351 }
@@ -242,16 +360,30 @@ int btrfs_stop_workers(struct btrfs_workers *workers)
242{ 360{
243 struct list_head *cur; 361 struct list_head *cur;
244 struct btrfs_worker_thread *worker; 362 struct btrfs_worker_thread *worker;
363 int can_stop;
245 364
365 spin_lock_irq(&workers->lock);
246 list_splice_init(&workers->idle_list, &workers->worker_list); 366 list_splice_init(&workers->idle_list, &workers->worker_list);
247 while (!list_empty(&workers->worker_list)) { 367 while (!list_empty(&workers->worker_list)) {
248 cur = workers->worker_list.next; 368 cur = workers->worker_list.next;
249 worker = list_entry(cur, struct btrfs_worker_thread, 369 worker = list_entry(cur, struct btrfs_worker_thread,
250 worker_list); 370 worker_list);
251 kthread_stop(worker->task); 371
252 list_del(&worker->worker_list); 372 atomic_inc(&worker->refs);
253 kfree(worker); 373 workers->num_workers -= 1;
374 if (!list_empty(&worker->worker_list)) {
375 list_del_init(&worker->worker_list);
376 put_worker(worker);
377 can_stop = 1;
378 } else
379 can_stop = 0;
380 spin_unlock_irq(&workers->lock);
381 if (can_stop)
382 kthread_stop(worker->task);
383 spin_lock_irq(&workers->lock);
384 put_worker(worker);
254 } 385 }
386 spin_unlock_irq(&workers->lock);
255 return 0; 387 return 0;
256} 388}
257 389
@@ -266,10 +398,13 @@ void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max)
266 INIT_LIST_HEAD(&workers->order_list); 398 INIT_LIST_HEAD(&workers->order_list);
267 INIT_LIST_HEAD(&workers->prio_order_list); 399 INIT_LIST_HEAD(&workers->prio_order_list);
268 spin_lock_init(&workers->lock); 400 spin_lock_init(&workers->lock);
401 spin_lock_init(&workers->order_lock);
269 workers->max_workers = max; 402 workers->max_workers = max;
270 workers->idle_thresh = 32; 403 workers->idle_thresh = 32;
271 workers->name = name; 404 workers->name = name;
272 workers->ordered = 0; 405 workers->ordered = 0;
406 workers->atomic_start_pending = 0;
407 workers->atomic_worker_start = 0;
273} 408}
274 409
275/* 410/*
@@ -293,7 +428,9 @@ int btrfs_start_workers(struct btrfs_workers *workers, int num_workers)
293 INIT_LIST_HEAD(&worker->prio_pending); 428 INIT_LIST_HEAD(&worker->prio_pending);
294 INIT_LIST_HEAD(&worker->worker_list); 429 INIT_LIST_HEAD(&worker->worker_list);
295 spin_lock_init(&worker->lock); 430 spin_lock_init(&worker->lock);
431
296 atomic_set(&worker->num_pending, 0); 432 atomic_set(&worker->num_pending, 0);
433 atomic_set(&worker->refs, 1);
297 worker->workers = workers; 434 worker->workers = workers;
298 worker->task = kthread_run(worker_loop, worker, 435 worker->task = kthread_run(worker_loop, worker,
299 "btrfs-%s-%d", workers->name, 436 "btrfs-%s-%d", workers->name,
@@ -303,7 +440,6 @@ int btrfs_start_workers(struct btrfs_workers *workers, int num_workers)
303 kfree(worker); 440 kfree(worker);
304 goto fail; 441 goto fail;
305 } 442 }
306
307 spin_lock_irq(&workers->lock); 443 spin_lock_irq(&workers->lock);
308 list_add_tail(&worker->worker_list, &workers->idle_list); 444 list_add_tail(&worker->worker_list, &workers->idle_list);
309 worker->idle = 1; 445 worker->idle = 1;
@@ -350,7 +486,6 @@ static struct btrfs_worker_thread *next_worker(struct btrfs_workers *workers)
350 */ 486 */
351 next = workers->worker_list.next; 487 next = workers->worker_list.next;
352 worker = list_entry(next, struct btrfs_worker_thread, worker_list); 488 worker = list_entry(next, struct btrfs_worker_thread, worker_list);
353 atomic_inc(&worker->num_pending);
354 worker->sequence++; 489 worker->sequence++;
355 490
356 if (worker->sequence % workers->idle_thresh == 0) 491 if (worker->sequence % workers->idle_thresh == 0)
@@ -367,28 +502,18 @@ static struct btrfs_worker_thread *find_worker(struct btrfs_workers *workers)
367{ 502{
368 struct btrfs_worker_thread *worker; 503 struct btrfs_worker_thread *worker;
369 unsigned long flags; 504 unsigned long flags;
505 struct list_head *fallback;
370 506
371again: 507again:
372 spin_lock_irqsave(&workers->lock, flags); 508 spin_lock_irqsave(&workers->lock, flags);
373 worker = next_worker(workers); 509 worker = next_worker(workers);
374 spin_unlock_irqrestore(&workers->lock, flags);
375 510
376 if (!worker) { 511 if (!worker) {
377 spin_lock_irqsave(&workers->lock, flags);
378 if (workers->num_workers >= workers->max_workers) { 512 if (workers->num_workers >= workers->max_workers) {
379 struct list_head *fallback = NULL; 513 goto fallback;
380 /* 514 } else if (workers->atomic_worker_start) {
381 * we have failed to find any workers, just 515 workers->atomic_start_pending = 1;
382 * return the force one 516 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 { 517 } else {
393 spin_unlock_irqrestore(&workers->lock, flags); 518 spin_unlock_irqrestore(&workers->lock, flags);
394 /* we're below the limit, start another worker */ 519 /* we're below the limit, start another worker */
@@ -396,6 +521,28 @@ again:
396 goto again; 521 goto again;
397 } 522 }
398 } 523 }
524 goto found;
525
526fallback:
527 fallback = NULL;
528 /*
529 * we have failed to find any workers, just
530 * return the first one we can find.
531 */
532 if (!list_empty(&workers->worker_list))
533 fallback = workers->worker_list.next;
534 if (!list_empty(&workers->idle_list))
535 fallback = workers->idle_list.next;
536 BUG_ON(!fallback);
537 worker = list_entry(fallback,
538 struct btrfs_worker_thread, worker_list);
539found:
540 /*
541 * this makes sure the worker doesn't exit before it is placed
542 * onto a busy/idle list
543 */
544 atomic_inc(&worker->num_pending);
545 spin_unlock_irqrestore(&workers->lock, flags);
399 return worker; 546 return worker;
400} 547}
401 548
@@ -427,7 +574,7 @@ int btrfs_requeue_work(struct btrfs_work *work)
427 spin_lock(&worker->workers->lock); 574 spin_lock(&worker->workers->lock);
428 worker->idle = 0; 575 worker->idle = 0;
429 list_move_tail(&worker->worker_list, 576 list_move_tail(&worker->worker_list,
430 &worker->workers->worker_list); 577 &worker->workers->worker_list);
431 spin_unlock(&worker->workers->lock); 578 spin_unlock(&worker->workers->lock);
432 } 579 }
433 if (!worker->working) { 580 if (!worker->working) {
@@ -435,9 +582,9 @@ int btrfs_requeue_work(struct btrfs_work *work)
435 worker->working = 1; 582 worker->working = 1;
436 } 583 }
437 584
438 spin_unlock_irqrestore(&worker->lock, flags);
439 if (wake) 585 if (wake)
440 wake_up_process(worker->task); 586 wake_up_process(worker->task);
587 spin_unlock_irqrestore(&worker->lock, flags);
441out: 588out:
442 589
443 return 0; 590 return 0;
@@ -463,14 +610,18 @@ int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
463 610
464 worker = find_worker(workers); 611 worker = find_worker(workers);
465 if (workers->ordered) { 612 if (workers->ordered) {
466 spin_lock_irqsave(&workers->lock, flags); 613 /*
614 * you're not allowed to do ordered queues from an
615 * interrupt handler
616 */
617 spin_lock(&workers->order_lock);
467 if (test_bit(WORK_HIGH_PRIO_BIT, &work->flags)) { 618 if (test_bit(WORK_HIGH_PRIO_BIT, &work->flags)) {
468 list_add_tail(&work->order_list, 619 list_add_tail(&work->order_list,
469 &workers->prio_order_list); 620 &workers->prio_order_list);
470 } else { 621 } else {
471 list_add_tail(&work->order_list, &workers->order_list); 622 list_add_tail(&work->order_list, &workers->order_list);
472 } 623 }
473 spin_unlock_irqrestore(&workers->lock, flags); 624 spin_unlock(&workers->order_lock);
474 } else { 625 } else {
475 INIT_LIST_HEAD(&work->order_list); 626 INIT_LIST_HEAD(&work->order_list);
476 } 627 }
@@ -481,7 +632,6 @@ int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
481 list_add_tail(&work->list, &worker->prio_pending); 632 list_add_tail(&work->list, &worker->prio_pending);
482 else 633 else
483 list_add_tail(&work->list, &worker->pending); 634 list_add_tail(&work->list, &worker->pending);
484 atomic_inc(&worker->num_pending);
485 check_busy_worker(worker); 635 check_busy_worker(worker);
486 636
487 /* 637 /*
@@ -492,10 +642,10 @@ int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
492 wake = 1; 642 wake = 1;
493 worker->working = 1; 643 worker->working = 1;
494 644
495 spin_unlock_irqrestore(&worker->lock, flags);
496
497 if (wake) 645 if (wake)
498 wake_up_process(worker->task); 646 wake_up_process(worker->task);
647 spin_unlock_irqrestore(&worker->lock, flags);
648
499out: 649out:
500 return 0; 650 return 0;
501} 651}
diff --git a/fs/btrfs/async-thread.h b/fs/btrfs/async-thread.h
index 1b511c109db6..fc089b95ec14 100644
--- a/fs/btrfs/async-thread.h
+++ b/fs/btrfs/async-thread.h
@@ -73,6 +73,15 @@ struct btrfs_workers {
73 /* force completions in the order they were queued */ 73 /* force completions in the order they were queued */
74 int ordered; 74 int ordered;
75 75
76 /* more workers required, but in an interrupt handler */
77 int atomic_start_pending;
78
79 /*
80 * are we allowed to sleep while starting workers or are we required
81 * to start them at a later time?
82 */
83 int atomic_worker_start;
84
76 /* list with all the work threads. The workers on the idle thread 85 /* 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 86 * may be actively servicing jobs, but they haven't yet hit the
78 * idle thresh limit above. 87 * idle thresh limit above.
@@ -90,6 +99,9 @@ struct btrfs_workers {
90 /* lock for finding the next worker thread to queue on */ 99 /* lock for finding the next worker thread to queue on */
91 spinlock_t lock; 100 spinlock_t lock;
92 101
102 /* lock for the ordered lists */
103 spinlock_t order_lock;
104
93 /* extra name for this worker, used for current->name */ 105 /* extra name for this worker, used for current->name */
94 char *name; 106 char *name;
95}; 107};
diff --git a/fs/btrfs/btrfs_inode.h b/fs/btrfs/btrfs_inode.h
index ea1ea0af8c0e..82ee56bba299 100644
--- a/fs/btrfs/btrfs_inode.h
+++ b/fs/btrfs/btrfs_inode.h
@@ -138,6 +138,7 @@ struct btrfs_inode {
138 * of these. 138 * of these.
139 */ 139 */
140 unsigned ordered_data_close:1; 140 unsigned ordered_data_close:1;
141 unsigned dummy_inode:1;
141 142
142 struct inode vfs_inode; 143 struct inode vfs_inode;
143}; 144};
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..80599b4e42bd 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
@@ -670,6 +674,7 @@ 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 */
677 u64 bytes_super; /* total bytes reserved for the super blocks */
673 678
674 /* delalloc accounting */ 679 /* delalloc accounting */
675 u64 bytes_delalloc; /* number of bytes reserved for allocation, 680 u64 bytes_delalloc; /* number of bytes reserved for allocation,
@@ -726,6 +731,15 @@ enum btrfs_caching_type {
726 BTRFS_CACHE_FINISHED = 2, 731 BTRFS_CACHE_FINISHED = 2,
727}; 732};
728 733
734struct btrfs_caching_control {
735 struct list_head list;
736 struct mutex mutex;
737 wait_queue_head_t wait;
738 struct btrfs_block_group_cache *block_group;
739 u64 progress;
740 atomic_t count;
741};
742
729struct btrfs_block_group_cache { 743struct btrfs_block_group_cache {
730 struct btrfs_key key; 744 struct btrfs_key key;
731 struct btrfs_block_group_item item; 745 struct btrfs_block_group_item item;
@@ -733,6 +747,7 @@ struct btrfs_block_group_cache {
733 spinlock_t lock; 747 spinlock_t lock;
734 u64 pinned; 748 u64 pinned;
735 u64 reserved; 749 u64 reserved;
750 u64 bytes_super;
736 u64 flags; 751 u64 flags;
737 u64 sectorsize; 752 u64 sectorsize;
738 int extents_thresh; 753 int extents_thresh;
@@ -742,8 +757,9 @@ struct btrfs_block_group_cache {
742 int dirty; 757 int dirty;
743 758
744 /* cache tracking stuff */ 759 /* cache tracking stuff */
745 wait_queue_head_t caching_q;
746 int cached; 760 int cached;
761 struct btrfs_caching_control *caching_ctl;
762 u64 last_byte_to_unpin;
747 763
748 struct btrfs_space_info *space_info; 764 struct btrfs_space_info *space_info;
749 765
@@ -782,13 +798,16 @@ struct btrfs_fs_info {
782 798
783 /* the log root tree is a directory of all the other log roots */ 799 /* the log root tree is a directory of all the other log roots */
784 struct btrfs_root *log_root_tree; 800 struct btrfs_root *log_root_tree;
801
802 spinlock_t fs_roots_radix_lock;
785 struct radix_tree_root fs_roots_radix; 803 struct radix_tree_root fs_roots_radix;
786 804
787 /* block group cache stuff */ 805 /* block group cache stuff */
788 spinlock_t block_group_cache_lock; 806 spinlock_t block_group_cache_lock;
789 struct rb_root block_group_cache_tree; 807 struct rb_root block_group_cache_tree;
790 808
791 struct extent_io_tree pinned_extents; 809 struct extent_io_tree freed_extents[2];
810 struct extent_io_tree *pinned_extents;
792 811
793 /* logical->physical extent mapping */ 812 /* logical->physical extent mapping */
794 struct btrfs_mapping_tree mapping_tree; 813 struct btrfs_mapping_tree mapping_tree;
@@ -822,11 +841,7 @@ struct btrfs_fs_info {
822 struct mutex transaction_kthread_mutex; 841 struct mutex transaction_kthread_mutex;
823 struct mutex cleaner_mutex; 842 struct mutex cleaner_mutex;
824 struct mutex chunk_mutex; 843 struct mutex chunk_mutex;
825 struct mutex drop_mutex;
826 struct mutex volume_mutex; 844 struct mutex volume_mutex;
827 struct mutex tree_reloc_mutex;
828 struct rw_semaphore extent_commit_sem;
829
830 /* 845 /*
831 * this protects the ordered operations list only while we are 846 * this protects the ordered operations list only while we are
832 * processing all of the entries on it. This way we make 847 * processing all of the entries on it. This way we make
@@ -835,10 +850,16 @@ struct btrfs_fs_info {
835 * before jumping into the main commit. 850 * before jumping into the main commit.
836 */ 851 */
837 struct mutex ordered_operations_mutex; 852 struct mutex ordered_operations_mutex;
853 struct rw_semaphore extent_commit_sem;
854
855 struct rw_semaphore subvol_sem;
856
857 struct srcu_struct subvol_srcu;
838 858
839 struct list_head trans_list; 859 struct list_head trans_list;
840 struct list_head hashers; 860 struct list_head hashers;
841 struct list_head dead_roots; 861 struct list_head dead_roots;
862 struct list_head caching_block_groups;
842 863
843 atomic_t nr_async_submits; 864 atomic_t nr_async_submits;
844 atomic_t async_submit_draining; 865 atomic_t async_submit_draining;
@@ -996,10 +1017,12 @@ struct btrfs_root {
996 u32 stripesize; 1017 u32 stripesize;
997 1018
998 u32 type; 1019 u32 type;
999 u64 highest_inode; 1020
1000 u64 last_inode_alloc; 1021 u64 highest_objectid;
1001 int ref_cows; 1022 int ref_cows;
1002 int track_dirty; 1023 int track_dirty;
1024 int in_radix;
1025
1003 u64 defrag_trans_start; 1026 u64 defrag_trans_start;
1004 struct btrfs_key defrag_progress; 1027 struct btrfs_key defrag_progress;
1005 struct btrfs_key defrag_max; 1028 struct btrfs_key defrag_max;
@@ -1920,8 +1943,8 @@ void btrfs_put_block_group(struct btrfs_block_group_cache *cache);
1920int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, 1943int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1921 struct btrfs_root *root, unsigned long count); 1944 struct btrfs_root *root, unsigned long count);
1922int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len); 1945int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len);
1923int btrfs_update_pinned_extents(struct btrfs_root *root, 1946int btrfs_pin_extent(struct btrfs_root *root,
1924 u64 bytenr, u64 num, int pin); 1947 u64 bytenr, u64 num, int reserved);
1925int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans, 1948int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
1926 struct btrfs_root *root, struct extent_buffer *leaf); 1949 struct btrfs_root *root, struct extent_buffer *leaf);
1927int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, 1950int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
@@ -1971,9 +1994,10 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans,
1971 u64 root_objectid, u64 owner, u64 offset); 1994 u64 root_objectid, u64 owner, u64 offset);
1972 1995
1973int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len); 1996int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len);
1997int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
1998 struct btrfs_root *root);
1974int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, 1999int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1975 struct btrfs_root *root, 2000 struct btrfs_root *root);
1976 struct extent_io_tree *unpin);
1977int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, 2001int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1978 struct btrfs_root *root, 2002 struct btrfs_root *root,
1979 u64 bytenr, u64 num_bytes, u64 parent, 2003 u64 bytenr, u64 num_bytes, u64 parent,
@@ -1984,6 +2008,7 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1984int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr); 2008int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr);
1985int btrfs_free_block_groups(struct btrfs_fs_info *info); 2009int btrfs_free_block_groups(struct btrfs_fs_info *info);
1986int btrfs_read_block_groups(struct btrfs_root *root); 2010int btrfs_read_block_groups(struct btrfs_root *root);
2011int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr);
1987int btrfs_make_block_group(struct btrfs_trans_handle *trans, 2012int btrfs_make_block_group(struct btrfs_trans_handle *trans,
1988 struct btrfs_root *root, u64 bytes_used, 2013 struct btrfs_root *root, u64 bytes_used,
1989 u64 type, u64 chunk_objectid, u64 chunk_offset, 2014 u64 type, u64 chunk_objectid, u64 chunk_offset,
@@ -2006,7 +2031,6 @@ void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
2006 u64 bytes); 2031 u64 bytes);
2007void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode, 2032void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
2008 u64 bytes); 2033 u64 bytes);
2009void btrfs_free_pinned_extents(struct btrfs_fs_info *info);
2010/* ctree.c */ 2034/* ctree.c */
2011int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key, 2035int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
2012 int level, int *slot); 2036 int level, int *slot);
@@ -2100,12 +2124,15 @@ int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
2100 struct extent_buffer *parent); 2124 struct extent_buffer *parent);
2101/* root-item.c */ 2125/* root-item.c */
2102int btrfs_find_root_ref(struct btrfs_root *tree_root, 2126int btrfs_find_root_ref(struct btrfs_root *tree_root,
2103 struct btrfs_path *path, 2127 struct btrfs_path *path,
2104 u64 root_id, u64 ref_id); 2128 u64 root_id, u64 ref_id);
2105int btrfs_add_root_ref(struct btrfs_trans_handle *trans, 2129int btrfs_add_root_ref(struct btrfs_trans_handle *trans,
2106 struct btrfs_root *tree_root, 2130 struct btrfs_root *tree_root,
2107 u64 root_id, u8 type, u64 ref_id, 2131 u64 root_id, u64 ref_id, u64 dirid, u64 sequence,
2108 u64 dirid, u64 sequence, 2132 const char *name, int name_len);
2133int btrfs_del_root_ref(struct btrfs_trans_handle *trans,
2134 struct btrfs_root *tree_root,
2135 u64 root_id, u64 ref_id, u64 dirid, u64 *sequence,
2109 const char *name, int name_len); 2136 const char *name, int name_len);
2110int btrfs_del_root(struct btrfs_trans_handle *trans, struct btrfs_root *root, 2137int btrfs_del_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2111 struct btrfs_key *key); 2138 struct btrfs_key *key);
@@ -2120,6 +2147,7 @@ int btrfs_find_last_root(struct btrfs_root *root, u64 objectid, struct
2120int btrfs_search_root(struct btrfs_root *root, u64 search_start, 2147int btrfs_search_root(struct btrfs_root *root, u64 search_start,
2121 u64 *found_objectid); 2148 u64 *found_objectid);
2122int btrfs_find_dead_roots(struct btrfs_root *root, u64 objectid); 2149int btrfs_find_dead_roots(struct btrfs_root *root, u64 objectid);
2150int btrfs_find_orphan_roots(struct btrfs_root *tree_root);
2123int btrfs_set_root_node(struct btrfs_root_item *item, 2151int btrfs_set_root_node(struct btrfs_root_item *item,
2124 struct extent_buffer *node); 2152 struct extent_buffer *node);
2125/* dir-item.c */ 2153/* dir-item.c */
@@ -2138,6 +2166,10 @@ btrfs_lookup_dir_index_item(struct btrfs_trans_handle *trans,
2138 struct btrfs_path *path, u64 dir, 2166 struct btrfs_path *path, u64 dir,
2139 u64 objectid, const char *name, int name_len, 2167 u64 objectid, const char *name, int name_len,
2140 int mod); 2168 int mod);
2169struct btrfs_dir_item *
2170btrfs_search_dir_index_item(struct btrfs_root *root,
2171 struct btrfs_path *path, u64 dirid,
2172 const char *name, int name_len);
2141struct btrfs_dir_item *btrfs_match_dir_item_name(struct btrfs_root *root, 2173struct btrfs_dir_item *btrfs_match_dir_item_name(struct btrfs_root *root,
2142 struct btrfs_path *path, 2174 struct btrfs_path *path,
2143 const char *name, int name_len); 2175 const char *name, int name_len);
@@ -2160,6 +2192,7 @@ int btrfs_insert_orphan_item(struct btrfs_trans_handle *trans,
2160 struct btrfs_root *root, u64 offset); 2192 struct btrfs_root *root, u64 offset);
2161int btrfs_del_orphan_item(struct btrfs_trans_handle *trans, 2193int btrfs_del_orphan_item(struct btrfs_trans_handle *trans,
2162 struct btrfs_root *root, u64 offset); 2194 struct btrfs_root *root, u64 offset);
2195int btrfs_find_orphan_item(struct btrfs_root *root, u64 offset);
2163 2196
2164/* inode-map.c */ 2197/* inode-map.c */
2165int btrfs_find_free_objectid(struct btrfs_trans_handle *trans, 2198int btrfs_find_free_objectid(struct btrfs_trans_handle *trans,
@@ -2232,6 +2265,10 @@ int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2232int btrfs_add_link(struct btrfs_trans_handle *trans, 2265int btrfs_add_link(struct btrfs_trans_handle *trans,
2233 struct inode *parent_inode, struct inode *inode, 2266 struct inode *parent_inode, struct inode *inode,
2234 const char *name, int name_len, int add_backref, u64 index); 2267 const char *name, int name_len, int add_backref, u64 index);
2268int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2269 struct btrfs_root *root,
2270 struct inode *dir, u64 objectid,
2271 const char *name, int name_len);
2235int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, 2272int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2236 struct btrfs_root *root, 2273 struct btrfs_root *root,
2237 struct inode *inode, u64 new_size, 2274 struct inode *inode, u64 new_size,
@@ -2242,7 +2279,7 @@ int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end);
2242int btrfs_writepages(struct address_space *mapping, 2279int btrfs_writepages(struct address_space *mapping,
2243 struct writeback_control *wbc); 2280 struct writeback_control *wbc);
2244int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, 2281int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
2245 struct btrfs_root *new_root, struct dentry *dentry, 2282 struct btrfs_root *new_root,
2246 u64 new_dirid, u64 alloc_hint); 2283 u64 new_dirid, u64 alloc_hint);
2247int btrfs_merge_bio_hook(struct page *page, unsigned long offset, 2284int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
2248 size_t size, struct bio *bio, unsigned long bio_flags); 2285 size_t size, struct bio *bio, unsigned long bio_flags);
@@ -2258,6 +2295,7 @@ int btrfs_write_inode(struct inode *inode, int wait);
2258void btrfs_dirty_inode(struct inode *inode); 2295void btrfs_dirty_inode(struct inode *inode);
2259struct inode *btrfs_alloc_inode(struct super_block *sb); 2296struct inode *btrfs_alloc_inode(struct super_block *sb);
2260void btrfs_destroy_inode(struct inode *inode); 2297void btrfs_destroy_inode(struct inode *inode);
2298void btrfs_drop_inode(struct inode *inode);
2261int btrfs_init_cachep(void); 2299int btrfs_init_cachep(void);
2262void btrfs_destroy_cachep(void); 2300void btrfs_destroy_cachep(void);
2263long btrfs_ioctl_trans_end(struct file *file); 2301long btrfs_ioctl_trans_end(struct file *file);
@@ -2275,6 +2313,8 @@ int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode);
2275int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode); 2313int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode);
2276void btrfs_orphan_cleanup(struct btrfs_root *root); 2314void btrfs_orphan_cleanup(struct btrfs_root *root);
2277int btrfs_cont_expand(struct inode *inode, loff_t size); 2315int btrfs_cont_expand(struct inode *inode, loff_t size);
2316int btrfs_invalidate_inodes(struct btrfs_root *root);
2317extern struct dentry_operations btrfs_dentry_operations;
2278 2318
2279/* ioctl.c */ 2319/* ioctl.c */
2280long btrfs_ioctl(struct file *file, unsigned int cmd, unsigned long arg); 2320long btrfs_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
@@ -2290,7 +2330,7 @@ extern struct file_operations btrfs_file_operations;
2290int btrfs_drop_extents(struct btrfs_trans_handle *trans, 2330int btrfs_drop_extents(struct btrfs_trans_handle *trans,
2291 struct btrfs_root *root, struct inode *inode, 2331 struct btrfs_root *root, struct inode *inode,
2292 u64 start, u64 end, u64 locked_end, 2332 u64 start, u64 end, u64 locked_end,
2293 u64 inline_limit, u64 *hint_block); 2333 u64 inline_limit, u64 *hint_block, int drop_cache);
2294int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, 2334int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
2295 struct btrfs_root *root, 2335 struct btrfs_root *root,
2296 struct inode *inode, u64 start, u64 end); 2336 struct inode *inode, u64 start, u64 end);
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 6c4173146bb7..644e796fd643 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);
@@ -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;
@@ -1360,8 +1379,10 @@ static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1360 1379
1361 err = bdi_register(bdi, NULL, "btrfs-%d", 1380 err = bdi_register(bdi, NULL, "btrfs-%d",
1362 atomic_inc_return(&btrfs_bdi_num)); 1381 atomic_inc_return(&btrfs_bdi_num));
1363 if (err) 1382 if (err) {
1383 bdi_destroy(bdi);
1364 return err; 1384 return err;
1385 }
1365 1386
1366 bdi->ra_pages = default_backing_dev_info.ra_pages; 1387 bdi->ra_pages = default_backing_dev_info.ra_pages;
1367 bdi->unplug_io_fn = btrfs_unplug_io_fn; 1388 bdi->unplug_io_fn = btrfs_unplug_io_fn;
@@ -1451,9 +1472,12 @@ static int cleaner_kthread(void *arg)
1451 break; 1472 break;
1452 1473
1453 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE); 1474 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1454 mutex_lock(&root->fs_info->cleaner_mutex); 1475
1455 btrfs_clean_old_snapshots(root); 1476 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1456 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 }
1457 1481
1458 if (freezing(current)) { 1482 if (freezing(current)) {
1459 refrigerator(); 1483 refrigerator();
@@ -1558,15 +1582,36 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1558 err = -ENOMEM; 1582 err = -ENOMEM;
1559 goto fail; 1583 goto fail;
1560 } 1584 }
1561 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);
1562 INIT_LIST_HEAD(&fs_info->trans_list); 1605 INIT_LIST_HEAD(&fs_info->trans_list);
1563 INIT_LIST_HEAD(&fs_info->dead_roots); 1606 INIT_LIST_HEAD(&fs_info->dead_roots);
1564 INIT_LIST_HEAD(&fs_info->hashers); 1607 INIT_LIST_HEAD(&fs_info->hashers);
1565 INIT_LIST_HEAD(&fs_info->delalloc_inodes); 1608 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1566 INIT_LIST_HEAD(&fs_info->ordered_operations); 1609 INIT_LIST_HEAD(&fs_info->ordered_operations);
1610 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1567 spin_lock_init(&fs_info->delalloc_lock); 1611 spin_lock_init(&fs_info->delalloc_lock);
1568 spin_lock_init(&fs_info->new_trans_lock); 1612 spin_lock_init(&fs_info->new_trans_lock);
1569 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);
1570 1615
1571 init_completion(&fs_info->kobj_unregister); 1616 init_completion(&fs_info->kobj_unregister);
1572 fs_info->tree_root = tree_root; 1617 fs_info->tree_root = tree_root;
@@ -1585,11 +1630,6 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1585 fs_info->sb = sb; 1630 fs_info->sb = sb;
1586 fs_info->max_extent = (u64)-1; 1631 fs_info->max_extent = (u64)-1;
1587 fs_info->max_inline = 8192 * 1024; 1632 fs_info->max_inline = 8192 * 1024;
1588 if (setup_bdi(fs_info, &fs_info->bdi))
1589 goto fail_bdi;
1590 fs_info->btree_inode = new_inode(sb);
1591 fs_info->btree_inode->i_ino = 1;
1592 fs_info->btree_inode->i_nlink = 1;
1593 fs_info->metadata_ratio = 8; 1633 fs_info->metadata_ratio = 8;
1594 1634
1595 fs_info->thread_pool_size = min_t(unsigned long, 1635 fs_info->thread_pool_size = min_t(unsigned long,
@@ -1602,6 +1642,8 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1602 sb->s_blocksize_bits = blksize_bits(4096); 1642 sb->s_blocksize_bits = blksize_bits(4096);
1603 sb->s_bdi = &fs_info->bdi; 1643 sb->s_bdi = &fs_info->bdi;
1604 1644
1645 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1646 fs_info->btree_inode->i_nlink = 1;
1605 /* 1647 /*
1606 * 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.
1607 * 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
@@ -1620,28 +1662,32 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1620 1662
1621 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;
1622 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
1623 spin_lock_init(&fs_info->block_group_cache_lock); 1671 spin_lock_init(&fs_info->block_group_cache_lock);
1624 fs_info->block_group_cache_tree.rb_node = NULL; 1672 fs_info->block_group_cache_tree.rb_node = NULL;
1625 1673
1626 extent_io_tree_init(&fs_info->pinned_extents, 1674 extent_io_tree_init(&fs_info->freed_extents[0],
1627 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];
1628 fs_info->do_barriers = 1; 1679 fs_info->do_barriers = 1;
1629 1680
1630 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1631 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1632 sizeof(struct btrfs_key));
1633 insert_inode_hash(fs_info->btree_inode);
1634 1681
1635 mutex_init(&fs_info->trans_mutex); 1682 mutex_init(&fs_info->trans_mutex);
1636 mutex_init(&fs_info->ordered_operations_mutex); 1683 mutex_init(&fs_info->ordered_operations_mutex);
1637 mutex_init(&fs_info->tree_log_mutex); 1684 mutex_init(&fs_info->tree_log_mutex);
1638 mutex_init(&fs_info->drop_mutex);
1639 mutex_init(&fs_info->chunk_mutex); 1685 mutex_init(&fs_info->chunk_mutex);
1640 mutex_init(&fs_info->transaction_kthread_mutex); 1686 mutex_init(&fs_info->transaction_kthread_mutex);
1641 mutex_init(&fs_info->cleaner_mutex); 1687 mutex_init(&fs_info->cleaner_mutex);
1642 mutex_init(&fs_info->volume_mutex); 1688 mutex_init(&fs_info->volume_mutex);
1643 mutex_init(&fs_info->tree_reloc_mutex);
1644 init_rwsem(&fs_info->extent_commit_sem); 1689 init_rwsem(&fs_info->extent_commit_sem);
1690 init_rwsem(&fs_info->subvol_sem);
1645 1691
1646 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 1692 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1647 btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 1693 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
@@ -1700,7 +1746,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1700 err = -EINVAL; 1746 err = -EINVAL;
1701 goto fail_iput; 1747 goto fail_iput;
1702 } 1748 }
1703 1749printk("thread pool is %d\n", fs_info->thread_pool_size);
1704 /* 1750 /*
1705 * we need to start all the end_io workers up front because the 1751 * we need to start all the end_io workers up front because the
1706 * queue work function gets called at interrupt time, and so it 1752 * queue work function gets called at interrupt time, and so it
@@ -1745,20 +1791,22 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1745 fs_info->endio_workers.idle_thresh = 4; 1791 fs_info->endio_workers.idle_thresh = 4;
1746 fs_info->endio_meta_workers.idle_thresh = 4; 1792 fs_info->endio_meta_workers.idle_thresh = 4;
1747 1793
1748 fs_info->endio_write_workers.idle_thresh = 64; 1794 fs_info->endio_write_workers.idle_thresh = 2;
1749 fs_info->endio_meta_write_workers.idle_thresh = 64; 1795 fs_info->endio_meta_write_workers.idle_thresh = 2;
1796
1797 fs_info->endio_workers.atomic_worker_start = 1;
1798 fs_info->endio_meta_workers.atomic_worker_start = 1;
1799 fs_info->endio_write_workers.atomic_worker_start = 1;
1800 fs_info->endio_meta_write_workers.atomic_worker_start = 1;
1750 1801
1751 btrfs_start_workers(&fs_info->workers, 1); 1802 btrfs_start_workers(&fs_info->workers, 1);
1752 btrfs_start_workers(&fs_info->submit_workers, 1); 1803 btrfs_start_workers(&fs_info->submit_workers, 1);
1753 btrfs_start_workers(&fs_info->delalloc_workers, 1); 1804 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1754 btrfs_start_workers(&fs_info->fixup_workers, 1); 1805 btrfs_start_workers(&fs_info->fixup_workers, 1);
1755 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size); 1806 btrfs_start_workers(&fs_info->endio_workers, 1);
1756 btrfs_start_workers(&fs_info->endio_meta_workers, 1807 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1757 fs_info->thread_pool_size); 1808 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1758 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1809 btrfs_start_workers(&fs_info->endio_write_workers, 1);
1759 fs_info->thread_pool_size);
1760 btrfs_start_workers(&fs_info->endio_write_workers,
1761 fs_info->thread_pool_size);
1762 1810
1763 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super); 1811 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1764 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages, 1812 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
@@ -1918,6 +1966,9 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1918 } 1966 }
1919 } 1967 }
1920 1968
1969 ret = btrfs_find_orphan_roots(tree_root);
1970 BUG_ON(ret);
1971
1921 if (!(sb->s_flags & MS_RDONLY)) { 1972 if (!(sb->s_flags & MS_RDONLY)) {
1922 ret = btrfs_recover_relocation(tree_root); 1973 ret = btrfs_recover_relocation(tree_root);
1923 BUG_ON(ret); 1974 BUG_ON(ret);
@@ -1977,6 +2028,8 @@ fail_iput:
1977 btrfs_mapping_tree_free(&fs_info->mapping_tree); 2028 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1978fail_bdi: 2029fail_bdi:
1979 bdi_destroy(&fs_info->bdi); 2030 bdi_destroy(&fs_info->bdi);
2031fail_srcu:
2032 cleanup_srcu_struct(&fs_info->subvol_srcu);
1980fail: 2033fail:
1981 kfree(extent_root); 2034 kfree(extent_root);
1982 kfree(tree_root); 2035 kfree(tree_root);
@@ -2236,20 +2289,29 @@ int write_ctree_super(struct btrfs_trans_handle *trans,
2236 2289
2237int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root) 2290int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2238{ 2291{
2239 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 2292 spin_lock(&fs_info->fs_roots_radix_lock);
2240 radix_tree_delete(&fs_info->fs_roots_radix, 2293 radix_tree_delete(&fs_info->fs_roots_radix,
2241 (unsigned long)root->root_key.objectid); 2294 (unsigned long)root->root_key.objectid);
2295 spin_unlock(&fs_info->fs_roots_radix_lock);
2296
2297 if (btrfs_root_refs(&root->root_item) == 0)
2298 synchronize_srcu(&fs_info->subvol_srcu);
2299
2300 free_fs_root(root);
2301 return 0;
2302}
2303
2304static void free_fs_root(struct btrfs_root *root)
2305{
2306 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2242 if (root->anon_super.s_dev) { 2307 if (root->anon_super.s_dev) {
2243 down_write(&root->anon_super.s_umount); 2308 down_write(&root->anon_super.s_umount);
2244 kill_anon_super(&root->anon_super); 2309 kill_anon_super(&root->anon_super);
2245 } 2310 }
2246 if (root->node) 2311 free_extent_buffer(root->node);
2247 free_extent_buffer(root->node); 2312 free_extent_buffer(root->commit_root);
2248 if (root->commit_root)
2249 free_extent_buffer(root->commit_root);
2250 kfree(root->name); 2313 kfree(root->name);
2251 kfree(root); 2314 kfree(root);
2252 return 0;
2253} 2315}
2254 2316
2255static int del_fs_roots(struct btrfs_fs_info *fs_info) 2317static int del_fs_roots(struct btrfs_fs_info *fs_info)
@@ -2258,6 +2320,20 @@ static int del_fs_roots(struct btrfs_fs_info *fs_info)
2258 struct btrfs_root *gang[8]; 2320 struct btrfs_root *gang[8];
2259 int i; 2321 int i;
2260 2322
2323 while (!list_empty(&fs_info->dead_roots)) {
2324 gang[0] = list_entry(fs_info->dead_roots.next,
2325 struct btrfs_root, root_list);
2326 list_del(&gang[0]->root_list);
2327
2328 if (gang[0]->in_radix) {
2329 btrfs_free_fs_root(fs_info, gang[0]);
2330 } else {
2331 free_extent_buffer(gang[0]->node);
2332 free_extent_buffer(gang[0]->commit_root);
2333 kfree(gang[0]);
2334 }
2335 }
2336
2261 while (1) { 2337 while (1) {
2262 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2338 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2263 (void **)gang, 0, 2339 (void **)gang, 0,
@@ -2287,9 +2363,6 @@ int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2287 root_objectid = gang[ret - 1]->root_key.objectid + 1; 2363 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2288 for (i = 0; i < ret; i++) { 2364 for (i = 0; i < ret; i++) {
2289 root_objectid = gang[i]->root_key.objectid; 2365 root_objectid = gang[i]->root_key.objectid;
2290 ret = btrfs_find_dead_roots(fs_info->tree_root,
2291 root_objectid);
2292 BUG_ON(ret);
2293 btrfs_orphan_cleanup(gang[i]); 2366 btrfs_orphan_cleanup(gang[i]);
2294 } 2367 }
2295 root_objectid++; 2368 root_objectid++;
@@ -2359,7 +2432,6 @@ int close_ctree(struct btrfs_root *root)
2359 free_extent_buffer(root->fs_info->csum_root->commit_root); 2432 free_extent_buffer(root->fs_info->csum_root->commit_root);
2360 2433
2361 btrfs_free_block_groups(root->fs_info); 2434 btrfs_free_block_groups(root->fs_info);
2362 btrfs_free_pinned_extents(root->fs_info);
2363 2435
2364 del_fs_roots(fs_info); 2436 del_fs_roots(fs_info);
2365 2437
@@ -2378,6 +2450,7 @@ int close_ctree(struct btrfs_root *root)
2378 btrfs_mapping_tree_free(&fs_info->mapping_tree); 2450 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2379 2451
2380 bdi_destroy(&fs_info->bdi); 2452 bdi_destroy(&fs_info->bdi);
2453 cleanup_srcu_struct(&fs_info->subvol_srcu);
2381 2454
2382 kfree(fs_info->extent_root); 2455 kfree(fs_info->extent_root);
2383 kfree(fs_info->tree_root); 2456 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 535f85ba104f..993f93ff7ba6 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 +57,17 @@ 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);
64 71
65static noinline int 72static noinline int
66block_group_cache_done(struct btrfs_block_group_cache *cache) 73block_group_cache_done(struct btrfs_block_group_cache *cache)
@@ -153,34 +160,34 @@ block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
153 return ret; 160 return ret;
154} 161}
155 162
156/* 163static int add_excluded_extent(struct btrfs_root *root,
157 * We always set EXTENT_LOCKED for the super mirror extents so we don't 164 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{ 165{
164 u64 start, end, last = 0; 166 u64 end = start + num_bytes - 1;
165 int ret; 167 set_extent_bits(&root->fs_info->freed_extents[0],
168 start, end, EXTENT_UPTODATE, GFP_NOFS);
169 set_extent_bits(&root->fs_info->freed_extents[1],
170 start, end, EXTENT_UPTODATE, GFP_NOFS);
171 return 0;
172}
166 173
167 while (1) { 174static void free_excluded_extents(struct btrfs_root *root,
168 ret = find_first_extent_bit(&info->pinned_extents, last, 175 struct btrfs_block_group_cache *cache)
169 &start, &end, 176{
170 EXTENT_LOCKED|EXTENT_DIRTY); 177 u64 start, end;
171 if (ret)
172 break;
173 178
174 clear_extent_bits(&info->pinned_extents, start, end, 179 start = cache->key.objectid;
175 EXTENT_LOCKED|EXTENT_DIRTY, GFP_NOFS); 180 end = start + cache->key.offset - 1;
176 last = end+1; 181
177 } 182 clear_extent_bits(&root->fs_info->freed_extents[0],
183 start, end, EXTENT_UPTODATE, GFP_NOFS);
184 clear_extent_bits(&root->fs_info->freed_extents[1],
185 start, end, EXTENT_UPTODATE, GFP_NOFS);
178} 186}
179 187
180static int remove_sb_from_cache(struct btrfs_root *root, 188static int exclude_super_stripes(struct btrfs_root *root,
181 struct btrfs_block_group_cache *cache) 189 struct btrfs_block_group_cache *cache)
182{ 190{
183 struct btrfs_fs_info *fs_info = root->fs_info;
184 u64 bytenr; 191 u64 bytenr;
185 u64 *logical; 192 u64 *logical;
186 int stripe_len; 193 int stripe_len;
@@ -192,17 +199,42 @@ static int remove_sb_from_cache(struct btrfs_root *root,
192 cache->key.objectid, bytenr, 199 cache->key.objectid, bytenr,
193 0, &logical, &nr, &stripe_len); 200 0, &logical, &nr, &stripe_len);
194 BUG_ON(ret); 201 BUG_ON(ret);
202
195 while (nr--) { 203 while (nr--) {
196 try_lock_extent(&fs_info->pinned_extents, 204 cache->bytes_super += stripe_len;
197 logical[nr], 205 ret = add_excluded_extent(root, logical[nr],
198 logical[nr] + stripe_len - 1, GFP_NOFS); 206 stripe_len);
207 BUG_ON(ret);
199 } 208 }
209
200 kfree(logical); 210 kfree(logical);
201 } 211 }
202
203 return 0; 212 return 0;
204} 213}
205 214
215static struct btrfs_caching_control *
216get_caching_control(struct btrfs_block_group_cache *cache)
217{
218 struct btrfs_caching_control *ctl;
219
220 spin_lock(&cache->lock);
221 if (cache->cached != BTRFS_CACHE_STARTED) {
222 spin_unlock(&cache->lock);
223 return NULL;
224 }
225
226 ctl = cache->caching_ctl;
227 atomic_inc(&ctl->count);
228 spin_unlock(&cache->lock);
229 return ctl;
230}
231
232static void put_caching_control(struct btrfs_caching_control *ctl)
233{
234 if (atomic_dec_and_test(&ctl->count))
235 kfree(ctl);
236}
237
206/* 238/*
207 * this is only called by cache_block_group, since we could have freed extents 239 * 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 240 * we need to check the pinned_extents for any extents that can't be used yet
@@ -215,9 +247,9 @@ static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
215 int ret; 247 int ret;
216 248
217 while (start < end) { 249 while (start < end) {
218 ret = find_first_extent_bit(&info->pinned_extents, start, 250 ret = find_first_extent_bit(info->pinned_extents, start,
219 &extent_start, &extent_end, 251 &extent_start, &extent_end,
220 EXTENT_DIRTY|EXTENT_LOCKED); 252 EXTENT_DIRTY | EXTENT_UPTODATE);
221 if (ret) 253 if (ret)
222 break; 254 break;
223 255
@@ -249,22 +281,27 @@ static int caching_kthread(void *data)
249{ 281{
250 struct btrfs_block_group_cache *block_group = data; 282 struct btrfs_block_group_cache *block_group = data;
251 struct btrfs_fs_info *fs_info = block_group->fs_info; 283 struct btrfs_fs_info *fs_info = block_group->fs_info;
252 u64 last = 0; 284 struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
285 struct btrfs_root *extent_root = fs_info->extent_root;
253 struct btrfs_path *path; 286 struct btrfs_path *path;
254 int ret = 0;
255 struct btrfs_key key;
256 struct extent_buffer *leaf; 287 struct extent_buffer *leaf;
257 int slot; 288 struct btrfs_key key;
258 u64 total_found = 0; 289 u64 total_found = 0;
259 290 u64 last = 0;
260 BUG_ON(!fs_info); 291 u32 nritems;
292 int ret = 0;
261 293
262 path = btrfs_alloc_path(); 294 path = btrfs_alloc_path();
263 if (!path) 295 if (!path)
264 return -ENOMEM; 296 return -ENOMEM;
265 297
266 atomic_inc(&block_group->space_info->caching_threads); 298 exclude_super_stripes(extent_root, block_group);
299 spin_lock(&block_group->space_info->lock);
300 block_group->space_info->bytes_super += block_group->bytes_super;
301 spin_unlock(&block_group->space_info->lock);
302
267 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); 303 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
304
268 /* 305 /*
269 * We don't want to deadlock with somebody trying to allocate a new 306 * 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 307 * extent for the extent root while also trying to search the extent
@@ -277,74 +314,64 @@ static int caching_kthread(void *data)
277 314
278 key.objectid = last; 315 key.objectid = last;
279 key.offset = 0; 316 key.offset = 0;
280 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); 317 key.type = BTRFS_EXTENT_ITEM_KEY;
281again: 318again:
319 mutex_lock(&caching_ctl->mutex);
282 /* need to make sure the commit_root doesn't disappear */ 320 /* need to make sure the commit_root doesn't disappear */
283 down_read(&fs_info->extent_commit_sem); 321 down_read(&fs_info->extent_commit_sem);
284 322
285 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); 323 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
286 if (ret < 0) 324 if (ret < 0)
287 goto err; 325 goto err;
288 326
327 leaf = path->nodes[0];
328 nritems = btrfs_header_nritems(leaf);
329
289 while (1) { 330 while (1) {
290 smp_mb(); 331 smp_mb();
291 if (block_group->fs_info->closing > 1) { 332 if (fs_info->closing > 1) {
292 last = (u64)-1; 333 last = (u64)-1;
293 break; 334 break;
294 } 335 }
295 336
296 leaf = path->nodes[0]; 337 if (path->slots[0] < nritems) {
297 slot = path->slots[0]; 338 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
298 if (slot >= btrfs_header_nritems(leaf)) { 339 } else {
299 ret = btrfs_next_leaf(fs_info->extent_root, path); 340 ret = find_next_key(path, 0, &key);
300 if (ret < 0) 341 if (ret)
301 goto err;
302 else if (ret)
303 break; 342 break;
304 343
305 if (need_resched() || 344 caching_ctl->progress = last;
306 btrfs_transaction_in_commit(fs_info)) { 345 btrfs_release_path(extent_root, path);
307 leaf = path->nodes[0]; 346 up_read(&fs_info->extent_commit_sem);
308 347 mutex_unlock(&caching_ctl->mutex);
309 /* this shouldn't happen, but if the 348 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); 349 schedule_timeout(1);
323 goto again; 350 else
324 } 351 cond_resched();
352 goto again;
353 }
325 354
355 if (key.objectid < block_group->key.objectid) {
356 path->slots[0]++;
326 continue; 357 continue;
327 } 358 }
328 btrfs_item_key_to_cpu(leaf, &key, slot);
329 if (key.objectid < block_group->key.objectid)
330 goto next;
331 359
332 if (key.objectid >= block_group->key.objectid + 360 if (key.objectid >= block_group->key.objectid +
333 block_group->key.offset) 361 block_group->key.offset)
334 break; 362 break;
335 363
336 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) { 364 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
337 total_found += add_new_free_space(block_group, 365 total_found += add_new_free_space(block_group,
338 fs_info, last, 366 fs_info, last,
339 key.objectid); 367 key.objectid);
340 last = key.objectid + key.offset; 368 last = key.objectid + key.offset;
341 }
342 369
343 if (total_found > (1024 * 1024 * 2)) { 370 if (total_found > (1024 * 1024 * 2)) {
344 total_found = 0; 371 total_found = 0;
345 wake_up(&block_group->caching_q); 372 wake_up(&caching_ctl->wait);
373 }
346 } 374 }
347next:
348 path->slots[0]++; 375 path->slots[0]++;
349 } 376 }
350 ret = 0; 377 ret = 0;
@@ -352,33 +379,65 @@ next:
352 total_found += add_new_free_space(block_group, fs_info, last, 379 total_found += add_new_free_space(block_group, fs_info, last,
353 block_group->key.objectid + 380 block_group->key.objectid +
354 block_group->key.offset); 381 block_group->key.offset);
382 caching_ctl->progress = (u64)-1;
355 383
356 spin_lock(&block_group->lock); 384 spin_lock(&block_group->lock);
385 block_group->caching_ctl = NULL;
357 block_group->cached = BTRFS_CACHE_FINISHED; 386 block_group->cached = BTRFS_CACHE_FINISHED;
358 spin_unlock(&block_group->lock); 387 spin_unlock(&block_group->lock);
359 388
360err: 389err:
361 btrfs_free_path(path); 390 btrfs_free_path(path);
362 up_read(&fs_info->extent_commit_sem); 391 up_read(&fs_info->extent_commit_sem);
363 atomic_dec(&block_group->space_info->caching_threads);
364 wake_up(&block_group->caching_q);
365 392
393 free_excluded_extents(extent_root, block_group);
394
395 mutex_unlock(&caching_ctl->mutex);
396 wake_up(&caching_ctl->wait);
397
398 put_caching_control(caching_ctl);
399 atomic_dec(&block_group->space_info->caching_threads);
366 return 0; 400 return 0;
367} 401}
368 402
369static int cache_block_group(struct btrfs_block_group_cache *cache) 403static int cache_block_group(struct btrfs_block_group_cache *cache)
370{ 404{
405 struct btrfs_fs_info *fs_info = cache->fs_info;
406 struct btrfs_caching_control *caching_ctl;
371 struct task_struct *tsk; 407 struct task_struct *tsk;
372 int ret = 0; 408 int ret = 0;
373 409
410 smp_mb();
411 if (cache->cached != BTRFS_CACHE_NO)
412 return 0;
413
414 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
415 BUG_ON(!caching_ctl);
416
417 INIT_LIST_HEAD(&caching_ctl->list);
418 mutex_init(&caching_ctl->mutex);
419 init_waitqueue_head(&caching_ctl->wait);
420 caching_ctl->block_group = cache;
421 caching_ctl->progress = cache->key.objectid;
422 /* one for caching kthread, one for caching block group list */
423 atomic_set(&caching_ctl->count, 2);
424
374 spin_lock(&cache->lock); 425 spin_lock(&cache->lock);
375 if (cache->cached != BTRFS_CACHE_NO) { 426 if (cache->cached != BTRFS_CACHE_NO) {
376 spin_unlock(&cache->lock); 427 spin_unlock(&cache->lock);
377 return ret; 428 kfree(caching_ctl);
429 return 0;
378 } 430 }
431 cache->caching_ctl = caching_ctl;
379 cache->cached = BTRFS_CACHE_STARTED; 432 cache->cached = BTRFS_CACHE_STARTED;
380 spin_unlock(&cache->lock); 433 spin_unlock(&cache->lock);
381 434
435 down_write(&fs_info->extent_commit_sem);
436 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
437 up_write(&fs_info->extent_commit_sem);
438
439 atomic_inc(&cache->space_info->caching_threads);
440
382 tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n", 441 tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
383 cache->key.objectid); 442 cache->key.objectid);
384 if (IS_ERR(tsk)) { 443 if (IS_ERR(tsk)) {
@@ -1657,7 +1716,6 @@ static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1657 parent, ref_root, flags, 1716 parent, ref_root, flags,
1658 ref->objectid, ref->offset, 1717 ref->objectid, ref->offset,
1659 &ins, node->ref_mod); 1718 &ins, node->ref_mod);
1660 update_reserved_extents(root, ins.objectid, ins.offset, 0);
1661 } else if (node->action == BTRFS_ADD_DELAYED_REF) { 1719 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1662 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, 1720 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1663 node->num_bytes, parent, 1721 node->num_bytes, parent,
@@ -1783,7 +1841,6 @@ static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1783 extent_op->flags_to_set, 1841 extent_op->flags_to_set,
1784 &extent_op->key, 1842 &extent_op->key,
1785 ref->level, &ins); 1843 ref->level, &ins);
1786 update_reserved_extents(root, ins.objectid, ins.offset, 0);
1787 } else if (node->action == BTRFS_ADD_DELAYED_REF) { 1844 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1788 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, 1845 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1789 node->num_bytes, parent, ref_root, 1846 node->num_bytes, parent, ref_root,
@@ -1818,16 +1875,32 @@ static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1818 BUG_ON(extent_op); 1875 BUG_ON(extent_op);
1819 head = btrfs_delayed_node_to_head(node); 1876 head = btrfs_delayed_node_to_head(node);
1820 if (insert_reserved) { 1877 if (insert_reserved) {
1878 int mark_free = 0;
1879 struct extent_buffer *must_clean = NULL;
1880
1881 ret = pin_down_bytes(trans, root, NULL,
1882 node->bytenr, node->num_bytes,
1883 head->is_data, 1, &must_clean);
1884 if (ret > 0)
1885 mark_free = 1;
1886
1887 if (must_clean) {
1888 clean_tree_block(NULL, root, must_clean);
1889 btrfs_tree_unlock(must_clean);
1890 free_extent_buffer(must_clean);
1891 }
1821 if (head->is_data) { 1892 if (head->is_data) {
1822 ret = btrfs_del_csums(trans, root, 1893 ret = btrfs_del_csums(trans, root,
1823 node->bytenr, 1894 node->bytenr,
1824 node->num_bytes); 1895 node->num_bytes);
1825 BUG_ON(ret); 1896 BUG_ON(ret);
1826 } 1897 }
1827 btrfs_update_pinned_extents(root, node->bytenr, 1898 if (mark_free) {
1828 node->num_bytes, 1); 1899 ret = btrfs_free_reserved_extent(root,
1829 update_reserved_extents(root, node->bytenr, 1900 node->bytenr,
1830 node->num_bytes, 0); 1901 node->num_bytes);
1902 BUG_ON(ret);
1903 }
1831 } 1904 }
1832 mutex_unlock(&head->mutex); 1905 mutex_unlock(&head->mutex);
1833 return 0; 1906 return 0;
@@ -2706,6 +2779,8 @@ int btrfs_check_metadata_free_space(struct btrfs_root *root)
2706 /* get the space info for where the metadata will live */ 2779 /* get the space info for where the metadata will live */
2707 alloc_target = btrfs_get_alloc_profile(root, 0); 2780 alloc_target = btrfs_get_alloc_profile(root, 0);
2708 meta_sinfo = __find_space_info(info, alloc_target); 2781 meta_sinfo = __find_space_info(info, alloc_target);
2782 if (!meta_sinfo)
2783 goto alloc;
2709 2784
2710again: 2785again:
2711 spin_lock(&meta_sinfo->lock); 2786 spin_lock(&meta_sinfo->lock);
@@ -2717,12 +2792,13 @@ again:
2717 do_div(thresh, 100); 2792 do_div(thresh, 100);
2718 2793
2719 if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved + 2794 if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2720 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) { 2795 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
2796 meta_sinfo->bytes_super > thresh) {
2721 struct btrfs_trans_handle *trans; 2797 struct btrfs_trans_handle *trans;
2722 if (!meta_sinfo->full) { 2798 if (!meta_sinfo->full) {
2723 meta_sinfo->force_alloc = 1; 2799 meta_sinfo->force_alloc = 1;
2724 spin_unlock(&meta_sinfo->lock); 2800 spin_unlock(&meta_sinfo->lock);
2725 2801alloc:
2726 trans = btrfs_start_transaction(root, 1); 2802 trans = btrfs_start_transaction(root, 1);
2727 if (!trans) 2803 if (!trans)
2728 return -ENOMEM; 2804 return -ENOMEM;
@@ -2730,6 +2806,10 @@ again:
2730 ret = do_chunk_alloc(trans, root->fs_info->extent_root, 2806 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2731 2 * 1024 * 1024, alloc_target, 0); 2807 2 * 1024 * 1024, alloc_target, 0);
2732 btrfs_end_transaction(trans, root); 2808 btrfs_end_transaction(trans, root);
2809 if (!meta_sinfo) {
2810 meta_sinfo = __find_space_info(info,
2811 alloc_target);
2812 }
2733 goto again; 2813 goto again;
2734 } 2814 }
2735 spin_unlock(&meta_sinfo->lock); 2815 spin_unlock(&meta_sinfo->lock);
@@ -2765,13 +2845,16 @@ int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
2765 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); 2845 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2766 2846
2767 data_sinfo = BTRFS_I(inode)->space_info; 2847 data_sinfo = BTRFS_I(inode)->space_info;
2848 if (!data_sinfo)
2849 goto alloc;
2850
2768again: 2851again:
2769 /* make sure we have enough space to handle the data first */ 2852 /* make sure we have enough space to handle the data first */
2770 spin_lock(&data_sinfo->lock); 2853 spin_lock(&data_sinfo->lock);
2771 if (data_sinfo->total_bytes - data_sinfo->bytes_used - 2854 if (data_sinfo->total_bytes - data_sinfo->bytes_used -
2772 data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved - 2855 data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
2773 data_sinfo->bytes_pinned - data_sinfo->bytes_readonly - 2856 data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
2774 data_sinfo->bytes_may_use < bytes) { 2857 data_sinfo->bytes_may_use - data_sinfo->bytes_super < bytes) {
2775 struct btrfs_trans_handle *trans; 2858 struct btrfs_trans_handle *trans;
2776 2859
2777 /* 2860 /*
@@ -2783,7 +2866,7 @@ again:
2783 2866
2784 data_sinfo->force_alloc = 1; 2867 data_sinfo->force_alloc = 1;
2785 spin_unlock(&data_sinfo->lock); 2868 spin_unlock(&data_sinfo->lock);
2786 2869alloc:
2787 alloc_target = btrfs_get_alloc_profile(root, 1); 2870 alloc_target = btrfs_get_alloc_profile(root, 1);
2788 trans = btrfs_start_transaction(root, 1); 2871 trans = btrfs_start_transaction(root, 1);
2789 if (!trans) 2872 if (!trans)
@@ -2795,6 +2878,11 @@ again:
2795 btrfs_end_transaction(trans, root); 2878 btrfs_end_transaction(trans, root);
2796 if (ret) 2879 if (ret)
2797 return ret; 2880 return ret;
2881
2882 if (!data_sinfo) {
2883 btrfs_set_inode_space_info(root, inode);
2884 data_sinfo = BTRFS_I(inode)->space_info;
2885 }
2798 goto again; 2886 goto again;
2799 } 2887 }
2800 spin_unlock(&data_sinfo->lock); 2888 spin_unlock(&data_sinfo->lock);
@@ -3009,10 +3097,12 @@ static int update_block_group(struct btrfs_trans_handle *trans,
3009 num_bytes = min(total, cache->key.offset - byte_in_group); 3097 num_bytes = min(total, cache->key.offset - byte_in_group);
3010 if (alloc) { 3098 if (alloc) {
3011 old_val += num_bytes; 3099 old_val += num_bytes;
3100 btrfs_set_block_group_used(&cache->item, old_val);
3101 cache->reserved -= num_bytes;
3012 cache->space_info->bytes_used += num_bytes; 3102 cache->space_info->bytes_used += num_bytes;
3103 cache->space_info->bytes_reserved -= num_bytes;
3013 if (cache->ro) 3104 if (cache->ro)
3014 cache->space_info->bytes_readonly -= num_bytes; 3105 cache->space_info->bytes_readonly -= num_bytes;
3015 btrfs_set_block_group_used(&cache->item, old_val);
3016 spin_unlock(&cache->lock); 3106 spin_unlock(&cache->lock);
3017 spin_unlock(&cache->space_info->lock); 3107 spin_unlock(&cache->space_info->lock);
3018 } else { 3108 } else {
@@ -3057,127 +3147,136 @@ static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3057 return bytenr; 3147 return bytenr;
3058} 3148}
3059 3149
3060int btrfs_update_pinned_extents(struct btrfs_root *root, 3150/*
3061 u64 bytenr, u64 num, int pin) 3151 * this function must be called within transaction
3152 */
3153int btrfs_pin_extent(struct btrfs_root *root,
3154 u64 bytenr, u64 num_bytes, int reserved)
3062{ 3155{
3063 u64 len;
3064 struct btrfs_block_group_cache *cache;
3065 struct btrfs_fs_info *fs_info = root->fs_info; 3156 struct btrfs_fs_info *fs_info = root->fs_info;
3157 struct btrfs_block_group_cache *cache;
3066 3158
3067 if (pin) 3159 cache = btrfs_lookup_block_group(fs_info, bytenr);
3068 set_extent_dirty(&fs_info->pinned_extents, 3160 BUG_ON(!cache);
3069 bytenr, bytenr + num - 1, GFP_NOFS);
3070
3071 while (num > 0) {
3072 cache = btrfs_lookup_block_group(fs_info, bytenr);
3073 BUG_ON(!cache);
3074 len = min(num, cache->key.offset -
3075 (bytenr - cache->key.objectid));
3076 if (pin) {
3077 spin_lock(&cache->space_info->lock);
3078 spin_lock(&cache->lock);
3079 cache->pinned += len;
3080 cache->space_info->bytes_pinned += len;
3081 spin_unlock(&cache->lock);
3082 spin_unlock(&cache->space_info->lock);
3083 fs_info->total_pinned += len;
3084 } else {
3085 int unpin = 0;
3086 3161
3087 /* 3162 spin_lock(&cache->space_info->lock);
3088 * in order to not race with the block group caching, we 3163 spin_lock(&cache->lock);
3089 * only want to unpin the extent if we are cached. If 3164 cache->pinned += num_bytes;
3090 * we aren't cached, we want to start async caching this 3165 cache->space_info->bytes_pinned += num_bytes;
3091 * block group so we can free the extent the next time 3166 if (reserved) {
3092 * around. 3167 cache->reserved -= num_bytes;
3093 */ 3168 cache->space_info->bytes_reserved -= num_bytes;
3094 spin_lock(&cache->space_info->lock); 3169 }
3095 spin_lock(&cache->lock); 3170 spin_unlock(&cache->lock);
3096 unpin = (cache->cached == BTRFS_CACHE_FINISHED); 3171 spin_unlock(&cache->space_info->lock);
3097 if (likely(unpin)) {
3098 cache->pinned -= len;
3099 cache->space_info->bytes_pinned -= len;
3100 fs_info->total_pinned -= len;
3101 }
3102 spin_unlock(&cache->lock);
3103 spin_unlock(&cache->space_info->lock);
3104 3172
3105 if (likely(unpin)) 3173 btrfs_put_block_group(cache);
3106 clear_extent_dirty(&fs_info->pinned_extents,
3107 bytenr, bytenr + len -1,
3108 GFP_NOFS);
3109 else
3110 cache_block_group(cache);
3111 3174
3112 if (unpin) 3175 set_extent_dirty(fs_info->pinned_extents,
3113 btrfs_add_free_space(cache, bytenr, len); 3176 bytenr, bytenr + num_bytes - 1, GFP_NOFS);
3114 } 3177 return 0;
3115 btrfs_put_block_group(cache); 3178}
3116 bytenr += len; 3179
3117 num -= len; 3180static int update_reserved_extents(struct btrfs_block_group_cache *cache,
3181 u64 num_bytes, int reserve)
3182{
3183 spin_lock(&cache->space_info->lock);
3184 spin_lock(&cache->lock);
3185 if (reserve) {
3186 cache->reserved += num_bytes;
3187 cache->space_info->bytes_reserved += num_bytes;
3188 } else {
3189 cache->reserved -= num_bytes;
3190 cache->space_info->bytes_reserved -= num_bytes;
3118 } 3191 }
3192 spin_unlock(&cache->lock);
3193 spin_unlock(&cache->space_info->lock);
3119 return 0; 3194 return 0;
3120} 3195}
3121 3196
3122static int update_reserved_extents(struct btrfs_root *root, 3197int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3123 u64 bytenr, u64 num, int reserve) 3198 struct btrfs_root *root)
3124{ 3199{
3125 u64 len;
3126 struct btrfs_block_group_cache *cache;
3127 struct btrfs_fs_info *fs_info = root->fs_info; 3200 struct btrfs_fs_info *fs_info = root->fs_info;
3201 struct btrfs_caching_control *next;
3202 struct btrfs_caching_control *caching_ctl;
3203 struct btrfs_block_group_cache *cache;
3128 3204
3129 while (num > 0) { 3205 down_write(&fs_info->extent_commit_sem);
3130 cache = btrfs_lookup_block_group(fs_info, bytenr);
3131 BUG_ON(!cache);
3132 len = min(num, cache->key.offset -
3133 (bytenr - cache->key.objectid));
3134 3206
3135 spin_lock(&cache->space_info->lock); 3207 list_for_each_entry_safe(caching_ctl, next,
3136 spin_lock(&cache->lock); 3208 &fs_info->caching_block_groups, list) {
3137 if (reserve) { 3209 cache = caching_ctl->block_group;
3138 cache->reserved += len; 3210 if (block_group_cache_done(cache)) {
3139 cache->space_info->bytes_reserved += len; 3211 cache->last_byte_to_unpin = (u64)-1;
3212 list_del_init(&caching_ctl->list);
3213 put_caching_control(caching_ctl);
3140 } else { 3214 } else {
3141 cache->reserved -= len; 3215 cache->last_byte_to_unpin = caching_ctl->progress;
3142 cache->space_info->bytes_reserved -= len;
3143 } 3216 }
3144 spin_unlock(&cache->lock);
3145 spin_unlock(&cache->space_info->lock);
3146 btrfs_put_block_group(cache);
3147 bytenr += len;
3148 num -= len;
3149 } 3217 }
3218
3219 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3220 fs_info->pinned_extents = &fs_info->freed_extents[1];
3221 else
3222 fs_info->pinned_extents = &fs_info->freed_extents[0];
3223
3224 up_write(&fs_info->extent_commit_sem);
3150 return 0; 3225 return 0;
3151} 3226}
3152 3227
3153int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy) 3228static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3154{ 3229{
3155 u64 last = 0; 3230 struct btrfs_fs_info *fs_info = root->fs_info;
3156 u64 start; 3231 struct btrfs_block_group_cache *cache = NULL;
3157 u64 end; 3232 u64 len;
3158 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
3159 int ret;
3160 3233
3161 while (1) { 3234 while (start <= end) {
3162 ret = find_first_extent_bit(pinned_extents, last, 3235 if (!cache ||
3163 &start, &end, EXTENT_DIRTY); 3236 start >= cache->key.objectid + cache->key.offset) {
3164 if (ret) 3237 if (cache)
3165 break; 3238 btrfs_put_block_group(cache);
3239 cache = btrfs_lookup_block_group(fs_info, start);
3240 BUG_ON(!cache);
3241 }
3242
3243 len = cache->key.objectid + cache->key.offset - start;
3244 len = min(len, end + 1 - start);
3166 3245
3167 set_extent_dirty(copy, start, end, GFP_NOFS); 3246 if (start < cache->last_byte_to_unpin) {
3168 last = end + 1; 3247 len = min(len, cache->last_byte_to_unpin - start);
3248 btrfs_add_free_space(cache, start, len);
3249 }
3250
3251 spin_lock(&cache->space_info->lock);
3252 spin_lock(&cache->lock);
3253 cache->pinned -= len;
3254 cache->space_info->bytes_pinned -= len;
3255 spin_unlock(&cache->lock);
3256 spin_unlock(&cache->space_info->lock);
3257
3258 start += len;
3169 } 3259 }
3260
3261 if (cache)
3262 btrfs_put_block_group(cache);
3170 return 0; 3263 return 0;
3171} 3264}
3172 3265
3173int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, 3266int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3174 struct btrfs_root *root, 3267 struct btrfs_root *root)
3175 struct extent_io_tree *unpin)
3176{ 3268{
3269 struct btrfs_fs_info *fs_info = root->fs_info;
3270 struct extent_io_tree *unpin;
3177 u64 start; 3271 u64 start;
3178 u64 end; 3272 u64 end;
3179 int ret; 3273 int ret;
3180 3274
3275 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3276 unpin = &fs_info->freed_extents[1];
3277 else
3278 unpin = &fs_info->freed_extents[0];
3279
3181 while (1) { 3280 while (1) {
3182 ret = find_first_extent_bit(unpin, 0, &start, &end, 3281 ret = find_first_extent_bit(unpin, 0, &start, &end,
3183 EXTENT_DIRTY); 3282 EXTENT_DIRTY);
@@ -3186,10 +3285,8 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3186 3285
3187 ret = btrfs_discard_extent(root, start, end + 1 - start); 3286 ret = btrfs_discard_extent(root, start, end + 1 - start);
3188 3287
3189 /* unlocks the pinned mutex */
3190 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
3191 clear_extent_dirty(unpin, start, end, GFP_NOFS); 3288 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3192 3289 unpin_extent_range(root, start, end);
3193 cond_resched(); 3290 cond_resched();
3194 } 3291 }
3195 3292
@@ -3199,7 +3296,8 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3199static int pin_down_bytes(struct btrfs_trans_handle *trans, 3296static int pin_down_bytes(struct btrfs_trans_handle *trans,
3200 struct btrfs_root *root, 3297 struct btrfs_root *root,
3201 struct btrfs_path *path, 3298 struct btrfs_path *path,
3202 u64 bytenr, u64 num_bytes, int is_data, 3299 u64 bytenr, u64 num_bytes,
3300 int is_data, int reserved,
3203 struct extent_buffer **must_clean) 3301 struct extent_buffer **must_clean)
3204{ 3302{
3205 int err = 0; 3303 int err = 0;
@@ -3231,15 +3329,15 @@ static int pin_down_bytes(struct btrfs_trans_handle *trans,
3231 } 3329 }
3232 free_extent_buffer(buf); 3330 free_extent_buffer(buf);
3233pinit: 3331pinit:
3234 btrfs_set_path_blocking(path); 3332 if (path)
3333 btrfs_set_path_blocking(path);
3235 /* unlocks the pinned mutex */ 3334 /* unlocks the pinned mutex */
3236 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); 3335 btrfs_pin_extent(root, bytenr, num_bytes, reserved);
3237 3336
3238 BUG_ON(err < 0); 3337 BUG_ON(err < 0);
3239 return 0; 3338 return 0;
3240} 3339}
3241 3340
3242
3243static int __btrfs_free_extent(struct btrfs_trans_handle *trans, 3341static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3244 struct btrfs_root *root, 3342 struct btrfs_root *root,
3245 u64 bytenr, u64 num_bytes, u64 parent, 3343 u64 bytenr, u64 num_bytes, u64 parent,
@@ -3413,7 +3511,7 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3413 } 3511 }
3414 3512
3415 ret = pin_down_bytes(trans, root, path, bytenr, 3513 ret = pin_down_bytes(trans, root, path, bytenr,
3416 num_bytes, is_data, &must_clean); 3514 num_bytes, is_data, 0, &must_clean);
3417 if (ret > 0) 3515 if (ret > 0)
3418 mark_free = 1; 3516 mark_free = 1;
3419 BUG_ON(ret < 0); 3517 BUG_ON(ret < 0);
@@ -3544,8 +3642,7 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans,
3544 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { 3642 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
3545 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); 3643 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
3546 /* unlocks the pinned mutex */ 3644 /* unlocks the pinned mutex */
3547 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); 3645 btrfs_pin_extent(root, bytenr, num_bytes, 1);
3548 update_reserved_extents(root, bytenr, num_bytes, 0);
3549 ret = 0; 3646 ret = 0;
3550 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { 3647 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
3551 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, 3648 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
@@ -3585,19 +3682,33 @@ static noinline int
3585wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, 3682wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
3586 u64 num_bytes) 3683 u64 num_bytes)
3587{ 3684{
3685 struct btrfs_caching_control *caching_ctl;
3588 DEFINE_WAIT(wait); 3686 DEFINE_WAIT(wait);
3589 3687
3590 prepare_to_wait(&cache->caching_q, &wait, TASK_UNINTERRUPTIBLE); 3688 caching_ctl = get_caching_control(cache);
3591 3689 if (!caching_ctl)
3592 if (block_group_cache_done(cache)) {
3593 finish_wait(&cache->caching_q, &wait);
3594 return 0; 3690 return 0;
3595 }
3596 schedule();
3597 finish_wait(&cache->caching_q, &wait);
3598 3691
3599 wait_event(cache->caching_q, block_group_cache_done(cache) || 3692 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
3600 (cache->free_space >= num_bytes)); 3693 (cache->free_space >= num_bytes));
3694
3695 put_caching_control(caching_ctl);
3696 return 0;
3697}
3698
3699static noinline int
3700wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
3701{
3702 struct btrfs_caching_control *caching_ctl;
3703 DEFINE_WAIT(wait);
3704
3705 caching_ctl = get_caching_control(cache);
3706 if (!caching_ctl)
3707 return 0;
3708
3709 wait_event(caching_ctl->wait, block_group_cache_done(cache));
3710
3711 put_caching_control(caching_ctl);
3601 return 0; 3712 return 0;
3602} 3713}
3603 3714
@@ -3635,6 +3746,7 @@ static noinline int find_free_extent(struct btrfs_trans_handle *trans,
3635 int last_ptr_loop = 0; 3746 int last_ptr_loop = 0;
3636 int loop = 0; 3747 int loop = 0;
3637 bool found_uncached_bg = false; 3748 bool found_uncached_bg = false;
3749 bool failed_cluster_refill = false;
3638 3750
3639 WARN_ON(num_bytes < root->sectorsize); 3751 WARN_ON(num_bytes < root->sectorsize);
3640 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); 3752 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
@@ -3732,7 +3844,16 @@ have_block_group:
3732 if (unlikely(block_group->ro)) 3844 if (unlikely(block_group->ro))
3733 goto loop; 3845 goto loop;
3734 3846
3735 if (last_ptr) { 3847 /*
3848 * Ok we want to try and use the cluster allocator, so lets look
3849 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
3850 * have tried the cluster allocator plenty of times at this
3851 * point and not have found anything, so we are likely way too
3852 * fragmented for the clustering stuff to find anything, so lets
3853 * just skip it and let the allocator find whatever block it can
3854 * find
3855 */
3856 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
3736 /* 3857 /*
3737 * the refill lock keeps out other 3858 * the refill lock keeps out other
3738 * people trying to start a new cluster 3859 * people trying to start a new cluster
@@ -3807,9 +3928,11 @@ refill_cluster:
3807 spin_unlock(&last_ptr->refill_lock); 3928 spin_unlock(&last_ptr->refill_lock);
3808 goto checks; 3929 goto checks;
3809 } 3930 }
3810 } else if (!cached && loop > LOOP_CACHING_NOWAIT) { 3931 } else if (!cached && loop > LOOP_CACHING_NOWAIT
3932 && !failed_cluster_refill) {
3811 spin_unlock(&last_ptr->refill_lock); 3933 spin_unlock(&last_ptr->refill_lock);
3812 3934
3935 failed_cluster_refill = true;
3813 wait_block_group_cache_progress(block_group, 3936 wait_block_group_cache_progress(block_group,
3814 num_bytes + empty_cluster + empty_size); 3937 num_bytes + empty_cluster + empty_size);
3815 goto have_block_group; 3938 goto have_block_group;
@@ -3821,13 +3944,9 @@ refill_cluster:
3821 * cluster. Free the cluster we've been trying 3944 * cluster. Free the cluster we've been trying
3822 * to use, and go to the next block group 3945 * to use, and go to the next block group
3823 */ 3946 */
3824 if (loop < LOOP_NO_EMPTY_SIZE) { 3947 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3825 btrfs_return_cluster_to_free_space(NULL,
3826 last_ptr);
3827 spin_unlock(&last_ptr->refill_lock);
3828 goto loop;
3829 }
3830 spin_unlock(&last_ptr->refill_lock); 3948 spin_unlock(&last_ptr->refill_lock);
3949 goto loop;
3831 } 3950 }
3832 3951
3833 offset = btrfs_find_space_for_alloc(block_group, search_start, 3952 offset = btrfs_find_space_for_alloc(block_group, search_start,
@@ -3881,9 +4000,12 @@ checks:
3881 search_start - offset); 4000 search_start - offset);
3882 BUG_ON(offset > search_start); 4001 BUG_ON(offset > search_start);
3883 4002
4003 update_reserved_extents(block_group, num_bytes, 1);
4004
3884 /* we are all good, lets return */ 4005 /* we are all good, lets return */
3885 break; 4006 break;
3886loop: 4007loop:
4008 failed_cluster_refill = false;
3887 btrfs_put_block_group(block_group); 4009 btrfs_put_block_group(block_group);
3888 } 4010 }
3889 up_read(&space_info->groups_sem); 4011 up_read(&space_info->groups_sem);
@@ -3973,12 +4095,12 @@ static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3973 up_read(&info->groups_sem); 4095 up_read(&info->groups_sem);
3974} 4096}
3975 4097
3976static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans, 4098int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3977 struct btrfs_root *root, 4099 struct btrfs_root *root,
3978 u64 num_bytes, u64 min_alloc_size, 4100 u64 num_bytes, u64 min_alloc_size,
3979 u64 empty_size, u64 hint_byte, 4101 u64 empty_size, u64 hint_byte,
3980 u64 search_end, struct btrfs_key *ins, 4102 u64 search_end, struct btrfs_key *ins,
3981 u64 data) 4103 u64 data)
3982{ 4104{
3983 int ret; 4105 int ret;
3984 u64 search_start = 0; 4106 u64 search_start = 0;
@@ -4044,25 +4166,8 @@ int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4044 ret = btrfs_discard_extent(root, start, len); 4166 ret = btrfs_discard_extent(root, start, len);
4045 4167
4046 btrfs_add_free_space(cache, start, len); 4168 btrfs_add_free_space(cache, start, len);
4169 update_reserved_extents(cache, len, 0);
4047 btrfs_put_block_group(cache); 4170 btrfs_put_block_group(cache);
4048 update_reserved_extents(root, start, len, 0);
4049
4050 return ret;
4051}
4052
4053int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4054 struct btrfs_root *root,
4055 u64 num_bytes, u64 min_alloc_size,
4056 u64 empty_size, u64 hint_byte,
4057 u64 search_end, struct btrfs_key *ins,
4058 u64 data)
4059{
4060 int ret;
4061 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
4062 empty_size, hint_byte, search_end, ins,
4063 data);
4064 if (!ret)
4065 update_reserved_extents(root, ins->objectid, ins->offset, 1);
4066 4171
4067 return ret; 4172 return ret;
4068} 4173}
@@ -4223,15 +4328,46 @@ int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4223{ 4328{
4224 int ret; 4329 int ret;
4225 struct btrfs_block_group_cache *block_group; 4330 struct btrfs_block_group_cache *block_group;
4331 struct btrfs_caching_control *caching_ctl;
4332 u64 start = ins->objectid;
4333 u64 num_bytes = ins->offset;
4226 4334
4227 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); 4335 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4228 cache_block_group(block_group); 4336 cache_block_group(block_group);
4229 wait_event(block_group->caching_q, 4337 caching_ctl = get_caching_control(block_group);
4230 block_group_cache_done(block_group));
4231 4338
4232 ret = btrfs_remove_free_space(block_group, ins->objectid, 4339 if (!caching_ctl) {
4233 ins->offset); 4340 BUG_ON(!block_group_cache_done(block_group));
4234 BUG_ON(ret); 4341 ret = btrfs_remove_free_space(block_group, start, num_bytes);
4342 BUG_ON(ret);
4343 } else {
4344 mutex_lock(&caching_ctl->mutex);
4345
4346 if (start >= caching_ctl->progress) {
4347 ret = add_excluded_extent(root, start, num_bytes);
4348 BUG_ON(ret);
4349 } else if (start + num_bytes <= caching_ctl->progress) {
4350 ret = btrfs_remove_free_space(block_group,
4351 start, num_bytes);
4352 BUG_ON(ret);
4353 } else {
4354 num_bytes = caching_ctl->progress - start;
4355 ret = btrfs_remove_free_space(block_group,
4356 start, num_bytes);
4357 BUG_ON(ret);
4358
4359 start = caching_ctl->progress;
4360 num_bytes = ins->objectid + ins->offset -
4361 caching_ctl->progress;
4362 ret = add_excluded_extent(root, start, num_bytes);
4363 BUG_ON(ret);
4364 }
4365
4366 mutex_unlock(&caching_ctl->mutex);
4367 put_caching_control(caching_ctl);
4368 }
4369
4370 update_reserved_extents(block_group, ins->offset, 1);
4235 btrfs_put_block_group(block_group); 4371 btrfs_put_block_group(block_group);
4236 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, 4372 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4237 0, owner, offset, ins, 1); 4373 0, owner, offset, ins, 1);
@@ -4255,9 +4391,9 @@ static int alloc_tree_block(struct btrfs_trans_handle *trans,
4255 int ret; 4391 int ret;
4256 u64 flags = 0; 4392 u64 flags = 0;
4257 4393
4258 ret = __btrfs_reserve_extent(trans, root, num_bytes, num_bytes, 4394 ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4259 empty_size, hint_byte, search_end, 4395 empty_size, hint_byte, search_end,
4260 ins, 0); 4396 ins, 0);
4261 if (ret) 4397 if (ret)
4262 return ret; 4398 return ret;
4263 4399
@@ -4268,7 +4404,6 @@ static int alloc_tree_block(struct btrfs_trans_handle *trans,
4268 } else 4404 } else
4269 BUG_ON(parent > 0); 4405 BUG_ON(parent > 0);
4270 4406
4271 update_reserved_extents(root, ins->objectid, ins->offset, 1);
4272 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { 4407 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4273 struct btrfs_delayed_extent_op *extent_op; 4408 struct btrfs_delayed_extent_op *extent_op;
4274 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); 4409 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
@@ -4347,452 +4482,99 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4347 return buf; 4482 return buf;
4348} 4483}
4349 4484
4350#if 0 4485struct walk_control {
4351int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans, 4486 u64 refs[BTRFS_MAX_LEVEL];
4352 struct btrfs_root *root, struct extent_buffer *leaf) 4487 u64 flags[BTRFS_MAX_LEVEL];
4353{ 4488 struct btrfs_key update_progress;
4354 u64 disk_bytenr; 4489 int stage;
4355 u64 num_bytes; 4490 int level;
4356 struct btrfs_key key; 4491 int shared_level;
4357 struct btrfs_file_extent_item *fi; 4492 int update_ref;
4358 u32 nritems; 4493 int keep_locks;
4359 int i; 4494 int reada_slot;
4360 int ret; 4495 int reada_count;
4361 4496};
4362 BUG_ON(!btrfs_is_leaf(leaf));
4363 nritems = btrfs_header_nritems(leaf);
4364
4365 for (i = 0; i < nritems; i++) {
4366 cond_resched();
4367 btrfs_item_key_to_cpu(leaf, &key, i);
4368
4369 /* only extents have references, skip everything else */
4370 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4371 continue;
4372
4373 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4374
4375 /* inline extents live in the btree, they don't have refs */
4376 if (btrfs_file_extent_type(leaf, fi) ==
4377 BTRFS_FILE_EXTENT_INLINE)
4378 continue;
4379
4380 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4381
4382 /* holes don't have refs */
4383 if (disk_bytenr == 0)
4384 continue;
4385
4386 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4387 ret = btrfs_free_extent(trans, root, disk_bytenr, num_bytes,
4388 leaf->start, 0, key.objectid, 0);
4389 BUG_ON(ret);
4390 }
4391 return 0;
4392}
4393
4394static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
4395 struct btrfs_root *root,
4396 struct btrfs_leaf_ref *ref)
4397{
4398 int i;
4399 int ret;
4400 struct btrfs_extent_info *info;
4401 struct refsort *sorted;
4402
4403 if (ref->nritems == 0)
4404 return 0;
4405
4406 sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
4407 for (i = 0; i < ref->nritems; i++) {
4408 sorted[i].bytenr = ref->extents[i].bytenr;
4409 sorted[i].slot = i;
4410 }
4411 sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
4412
4413 /*
4414 * the items in the ref were sorted when the ref was inserted
4415 * into the ref cache, so this is already in order
4416 */
4417 for (i = 0; i < ref->nritems; i++) {
4418 info = ref->extents + sorted[i].slot;
4419 ret = btrfs_free_extent(trans, root, info->bytenr,
4420 info->num_bytes, ref->bytenr,
4421 ref->owner, ref->generation,
4422 info->objectid, 0);
4423
4424 atomic_inc(&root->fs_info->throttle_gen);
4425 wake_up(&root->fs_info->transaction_throttle);
4426 cond_resched();
4427
4428 BUG_ON(ret);
4429 info++;
4430 }
4431
4432 kfree(sorted);
4433 return 0;
4434}
4435
4436
4437static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
4438 struct btrfs_root *root, u64 start,
4439 u64 len, u32 *refs)
4440{
4441 int ret;
4442
4443 ret = btrfs_lookup_extent_refs(trans, root, start, len, refs);
4444 BUG_ON(ret);
4445
4446#if 0 /* some debugging code in case we see problems here */
4447 /* if the refs count is one, it won't get increased again. But
4448 * if the ref count is > 1, someone may be decreasing it at
4449 * the same time we are.
4450 */
4451 if (*refs != 1) {
4452 struct extent_buffer *eb = NULL;
4453 eb = btrfs_find_create_tree_block(root, start, len);
4454 if (eb)
4455 btrfs_tree_lock(eb);
4456
4457 mutex_lock(&root->fs_info->alloc_mutex);
4458 ret = lookup_extent_ref(NULL, root, start, len, refs);
4459 BUG_ON(ret);
4460 mutex_unlock(&root->fs_info->alloc_mutex);
4461
4462 if (eb) {
4463 btrfs_tree_unlock(eb);
4464 free_extent_buffer(eb);
4465 }
4466 if (*refs == 1) {
4467 printk(KERN_ERR "btrfs block %llu went down to one "
4468 "during drop_snap\n", (unsigned long long)start);
4469 }
4470
4471 }
4472#endif
4473
4474 cond_resched();
4475 return ret;
4476}
4477 4497
4498#define DROP_REFERENCE 1
4499#define UPDATE_BACKREF 2
4478 4500
4479/* 4501static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4480 * this is used while deleting old snapshots, and it drops the refs 4502 struct btrfs_root *root,
4481 * on a whole subtree starting from a level 1 node. 4503 struct walk_control *wc,
4482 * 4504 struct btrfs_path *path)
4483 * The idea is to sort all the leaf pointers, and then drop the
4484 * ref on all the leaves in order. Most of the time the leaves
4485 * will have ref cache entries, so no leaf IOs will be required to
4486 * find the extents they have references on.
4487 *
4488 * For each leaf, any references it has are also dropped in order
4489 *
4490 * This ends up dropping the references in something close to optimal
4491 * order for reading and modifying the extent allocation tree.
4492 */
4493static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
4494 struct btrfs_root *root,
4495 struct btrfs_path *path)
4496{ 4505{
4497 u64 bytenr; 4506 u64 bytenr;
4498 u64 root_owner; 4507 u64 generation;
4499 u64 root_gen; 4508 u64 refs;
4500 struct extent_buffer *eb = path->nodes[1]; 4509 u64 last = 0;
4501 struct extent_buffer *leaf; 4510 u32 nritems;
4502 struct btrfs_leaf_ref *ref; 4511 u32 blocksize;
4503 struct refsort *sorted = NULL; 4512 struct btrfs_key key;
4504 int nritems = btrfs_header_nritems(eb); 4513 struct extent_buffer *eb;
4505 int ret; 4514 int ret;
4506 int i; 4515 int slot;
4507 int refi = 0; 4516 int nread = 0;
4508 int slot = path->slots[1];
4509 u32 blocksize = btrfs_level_size(root, 0);
4510 u32 refs;
4511
4512 if (nritems == 0)
4513 goto out;
4514
4515 root_owner = btrfs_header_owner(eb);
4516 root_gen = btrfs_header_generation(eb);
4517 sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
4518 4517
4519 /* 4518 if (path->slots[wc->level] < wc->reada_slot) {
4520 * step one, sort all the leaf pointers so we don't scribble 4519 wc->reada_count = wc->reada_count * 2 / 3;
4521 * randomly into the extent allocation tree 4520 wc->reada_count = max(wc->reada_count, 2);
4522 */ 4521 } else {
4523 for (i = slot; i < nritems; i++) { 4522 wc->reada_count = wc->reada_count * 3 / 2;
4524 sorted[refi].bytenr = btrfs_node_blockptr(eb, i); 4523 wc->reada_count = min_t(int, wc->reada_count,
4525 sorted[refi].slot = i; 4524 BTRFS_NODEPTRS_PER_BLOCK(root));
4526 refi++;
4527 } 4525 }
4528 4526
4529 /* 4527 eb = path->nodes[wc->level];
4530 * nritems won't be zero, but if we're picking up drop_snapshot 4528 nritems = btrfs_header_nritems(eb);
4531 * after a crash, slot might be > 0, so double check things 4529 blocksize = btrfs_level_size(root, wc->level - 1);
4532 * just in case.
4533 */
4534 if (refi == 0)
4535 goto out;
4536 4530
4537 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL); 4531 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
4532 if (nread >= wc->reada_count)
4533 break;
4538 4534
4539 /* 4535 cond_resched();
4540 * the first loop frees everything the leaves point to 4536 bytenr = btrfs_node_blockptr(eb, slot);
4541 */ 4537 generation = btrfs_node_ptr_generation(eb, slot);
4542 for (i = 0; i < refi; i++) {
4543 u64 ptr_gen;
4544 4538
4545 bytenr = sorted[i].bytenr; 4539 if (slot == path->slots[wc->level])
4540 goto reada;
4546 4541
4547 /* 4542 if (wc->stage == UPDATE_BACKREF &&
4548 * check the reference count on this leaf. If it is > 1 4543 generation <= root->root_key.offset)
4549 * we just decrement it below and don't update any
4550 * of the refs the leaf points to.
4551 */
4552 ret = drop_snap_lookup_refcount(trans, root, bytenr,
4553 blocksize, &refs);
4554 BUG_ON(ret);
4555 if (refs != 1)
4556 continue; 4544 continue;
4557 4545
4558 ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot); 4546 if (wc->stage == DROP_REFERENCE) {
4559 4547 ret = btrfs_lookup_extent_info(trans, root,
4560 /* 4548 bytenr, blocksize,
4561 * the leaf only had one reference, which means the 4549 &refs, NULL);
4562 * only thing pointing to this leaf is the snapshot
4563 * we're deleting. It isn't possible for the reference
4564 * count to increase again later
4565 *
4566 * The reference cache is checked for the leaf,
4567 * and if found we'll be able to drop any refs held by
4568 * the leaf without needing to read it in.
4569 */
4570 ref = btrfs_lookup_leaf_ref(root, bytenr);
4571 if (ref && ref->generation != ptr_gen) {
4572 btrfs_free_leaf_ref(root, ref);
4573 ref = NULL;
4574 }
4575 if (ref) {
4576 ret = cache_drop_leaf_ref(trans, root, ref);
4577 BUG_ON(ret);
4578 btrfs_remove_leaf_ref(root, ref);
4579 btrfs_free_leaf_ref(root, ref);
4580 } else {
4581 /*
4582 * the leaf wasn't in the reference cache, so
4583 * we have to read it.
4584 */
4585 leaf = read_tree_block(root, bytenr, blocksize,
4586 ptr_gen);
4587 ret = btrfs_drop_leaf_ref(trans, root, leaf);
4588 BUG_ON(ret); 4550 BUG_ON(ret);
4589 free_extent_buffer(leaf); 4551 BUG_ON(refs == 0);
4590 } 4552 if (refs == 1)
4591 atomic_inc(&root->fs_info->throttle_gen); 4553 goto reada;
4592 wake_up(&root->fs_info->transaction_throttle);
4593 cond_resched();
4594 }
4595
4596 /*
4597 * run through the loop again to free the refs on the leaves.
4598 * This is faster than doing it in the loop above because
4599 * the leaves are likely to be clustered together. We end up
4600 * working in nice chunks on the extent allocation tree.
4601 */
4602 for (i = 0; i < refi; i++) {
4603 bytenr = sorted[i].bytenr;
4604 ret = btrfs_free_extent(trans, root, bytenr,
4605 blocksize, eb->start,
4606 root_owner, root_gen, 0, 1);
4607 BUG_ON(ret);
4608
4609 atomic_inc(&root->fs_info->throttle_gen);
4610 wake_up(&root->fs_info->transaction_throttle);
4611 cond_resched();
4612 }
4613out:
4614 kfree(sorted);
4615
4616 /*
4617 * update the path to show we've processed the entire level 1
4618 * node. This will get saved into the root's drop_snapshot_progress
4619 * field so these drops are not repeated again if this transaction
4620 * commits.
4621 */
4622 path->slots[1] = nritems;
4623 return 0;
4624}
4625
4626/*
4627 * helper function for drop_snapshot, this walks down the tree dropping ref
4628 * counts as it goes.
4629 */
4630static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4631 struct btrfs_root *root,
4632 struct btrfs_path *path, int *level)
4633{
4634 u64 root_owner;
4635 u64 root_gen;
4636 u64 bytenr;
4637 u64 ptr_gen;
4638 struct extent_buffer *next;
4639 struct extent_buffer *cur;
4640 struct extent_buffer *parent;
4641 u32 blocksize;
4642 int ret;
4643 u32 refs;
4644
4645 WARN_ON(*level < 0);
4646 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4647 ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start,
4648 path->nodes[*level]->len, &refs);
4649 BUG_ON(ret);
4650 if (refs > 1)
4651 goto out;
4652
4653 /*
4654 * walk down to the last node level and free all the leaves
4655 */
4656 while (*level >= 0) {
4657 WARN_ON(*level < 0);
4658 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4659 cur = path->nodes[*level];
4660
4661 if (btrfs_header_level(cur) != *level)
4662 WARN_ON(1);
4663 4554
4664 if (path->slots[*level] >= 4555 if (!wc->update_ref ||
4665 btrfs_header_nritems(cur)) 4556 generation <= root->root_key.offset)
4666 break; 4557 continue;
4667 4558 btrfs_node_key_to_cpu(eb, &key, slot);
4668 /* the new code goes down to level 1 and does all the 4559 ret = btrfs_comp_cpu_keys(&key,
4669 * leaves pointed to that node in bulk. So, this check 4560 &wc->update_progress);
4670 * for level 0 will always be false. 4561 if (ret < 0)
4671 * 4562 continue;
4672 * But, the disk format allows the drop_snapshot_progress
4673 * field in the root to leave things in a state where
4674 * a leaf will need cleaning up here. If someone crashes
4675 * with the old code and then boots with the new code,
4676 * we might find a leaf here.
4677 */
4678 if (*level == 0) {
4679 ret = btrfs_drop_leaf_ref(trans, root, cur);
4680 BUG_ON(ret);
4681 break;
4682 } 4563 }
4683 4564reada:
4684 /* 4565 ret = readahead_tree_block(root, bytenr, blocksize,
4685 * once we get to level one, process the whole node 4566 generation);
4686 * at once, including everything below it. 4567 if (ret)
4687 */
4688 if (*level == 1) {
4689 ret = drop_level_one_refs(trans, root, path);
4690 BUG_ON(ret);
4691 break; 4568 break;
4692 } 4569 last = bytenr + blocksize;
4693 4570 nread++;
4694 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
4695 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
4696 blocksize = btrfs_level_size(root, *level - 1);
4697
4698 ret = drop_snap_lookup_refcount(trans, root, bytenr,
4699 blocksize, &refs);
4700 BUG_ON(ret);
4701
4702 /*
4703 * if there is more than one reference, we don't need
4704 * to read that node to drop any references it has. We
4705 * just drop the ref we hold on that node and move on to the
4706 * next slot in this level.
4707 */
4708 if (refs != 1) {
4709 parent = path->nodes[*level];
4710 root_owner = btrfs_header_owner(parent);
4711 root_gen = btrfs_header_generation(parent);
4712 path->slots[*level]++;
4713
4714 ret = btrfs_free_extent(trans, root, bytenr,
4715 blocksize, parent->start,
4716 root_owner, root_gen,
4717 *level - 1, 1);
4718 BUG_ON(ret);
4719
4720 atomic_inc(&root->fs_info->throttle_gen);
4721 wake_up(&root->fs_info->transaction_throttle);
4722 cond_resched();
4723
4724 continue;
4725 }
4726
4727 /*
4728 * we need to keep freeing things in the next level down.
4729 * read the block and loop around to process it
4730 */
4731 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
4732 WARN_ON(*level <= 0);
4733 if (path->nodes[*level-1])
4734 free_extent_buffer(path->nodes[*level-1]);
4735 path->nodes[*level-1] = next;
4736 *level = btrfs_header_level(next);
4737 path->slots[*level] = 0;
4738 cond_resched();
4739 } 4571 }
4740out: 4572 wc->reada_slot = slot;
4741 WARN_ON(*level < 0);
4742 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4743
4744 if (path->nodes[*level] == root->node) {
4745 parent = path->nodes[*level];
4746 bytenr = path->nodes[*level]->start;
4747 } else {
4748 parent = path->nodes[*level + 1];
4749 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
4750 }
4751
4752 blocksize = btrfs_level_size(root, *level);
4753 root_owner = btrfs_header_owner(parent);
4754 root_gen = btrfs_header_generation(parent);
4755
4756 /*
4757 * cleanup and free the reference on the last node
4758 * we processed
4759 */
4760 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
4761 parent->start, root_owner, root_gen,
4762 *level, 1);
4763 free_extent_buffer(path->nodes[*level]);
4764 path->nodes[*level] = NULL;
4765
4766 *level += 1;
4767 BUG_ON(ret);
4768
4769 cond_resched();
4770 return 0;
4771} 4573}
4772#endif
4773
4774struct walk_control {
4775 u64 refs[BTRFS_MAX_LEVEL];
4776 u64 flags[BTRFS_MAX_LEVEL];
4777 struct btrfs_key update_progress;
4778 int stage;
4779 int level;
4780 int shared_level;
4781 int update_ref;
4782 int keep_locks;
4783};
4784
4785#define DROP_REFERENCE 1
4786#define UPDATE_BACKREF 2
4787 4574
4788/* 4575/*
4789 * hepler to process tree block while walking down the tree. 4576 * hepler to process tree block while walking down the tree.
4790 * 4577 *
4791 * when wc->stage == DROP_REFERENCE, this function checks
4792 * reference count of the block. if the block is shared and
4793 * we need update back refs for the subtree rooted at the
4794 * block, this function changes wc->stage to UPDATE_BACKREF
4795 *
4796 * when wc->stage == UPDATE_BACKREF, this function updates 4578 * when wc->stage == UPDATE_BACKREF, this function updates
4797 * back refs for pointers in the block. 4579 * back refs for pointers in the block.
4798 * 4580 *
@@ -4805,7 +4587,6 @@ static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4805{ 4587{
4806 int level = wc->level; 4588 int level = wc->level;
4807 struct extent_buffer *eb = path->nodes[level]; 4589 struct extent_buffer *eb = path->nodes[level];
4808 struct btrfs_key key;
4809 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; 4590 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4810 int ret; 4591 int ret;
4811 4592
@@ -4828,21 +4609,6 @@ static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4828 BUG_ON(wc->refs[level] == 0); 4609 BUG_ON(wc->refs[level] == 0);
4829 } 4610 }
4830 4611
4831 if (wc->stage == DROP_REFERENCE &&
4832 wc->update_ref && wc->refs[level] > 1) {
4833 BUG_ON(eb == root->node);
4834 BUG_ON(path->slots[level] > 0);
4835 if (level == 0)
4836 btrfs_item_key_to_cpu(eb, &key, path->slots[level]);
4837 else
4838 btrfs_node_key_to_cpu(eb, &key, path->slots[level]);
4839 if (btrfs_header_owner(eb) == root->root_key.objectid &&
4840 btrfs_comp_cpu_keys(&key, &wc->update_progress) >= 0) {
4841 wc->stage = UPDATE_BACKREF;
4842 wc->shared_level = level;
4843 }
4844 }
4845
4846 if (wc->stage == DROP_REFERENCE) { 4612 if (wc->stage == DROP_REFERENCE) {
4847 if (wc->refs[level] > 1) 4613 if (wc->refs[level] > 1)
4848 return 1; 4614 return 1;
@@ -4879,6 +4645,123 @@ static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4879} 4645}
4880 4646
4881/* 4647/*
4648 * hepler to process tree block pointer.
4649 *
4650 * when wc->stage == DROP_REFERENCE, this function checks
4651 * reference count of the block pointed to. if the block
4652 * is shared and we need update back refs for the subtree
4653 * rooted at the block, this function changes wc->stage to
4654 * UPDATE_BACKREF. if the block is shared and there is no
4655 * need to update back, this function drops the reference
4656 * to the block.
4657 *
4658 * NOTE: return value 1 means we should stop walking down.
4659 */
4660static noinline int do_walk_down(struct btrfs_trans_handle *trans,
4661 struct btrfs_root *root,
4662 struct btrfs_path *path,
4663 struct walk_control *wc)
4664{
4665 u64 bytenr;
4666 u64 generation;
4667 u64 parent;
4668 u32 blocksize;
4669 struct btrfs_key key;
4670 struct extent_buffer *next;
4671 int level = wc->level;
4672 int reada = 0;
4673 int ret = 0;
4674
4675 generation = btrfs_node_ptr_generation(path->nodes[level],
4676 path->slots[level]);
4677 /*
4678 * if the lower level block was created before the snapshot
4679 * was created, we know there is no need to update back refs
4680 * for the subtree
4681 */
4682 if (wc->stage == UPDATE_BACKREF &&
4683 generation <= root->root_key.offset)
4684 return 1;
4685
4686 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
4687 blocksize = btrfs_level_size(root, level - 1);
4688
4689 next = btrfs_find_tree_block(root, bytenr, blocksize);
4690 if (!next) {
4691 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
4692 reada = 1;
4693 }
4694 btrfs_tree_lock(next);
4695 btrfs_set_lock_blocking(next);
4696
4697 if (wc->stage == DROP_REFERENCE) {
4698 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
4699 &wc->refs[level - 1],
4700 &wc->flags[level - 1]);
4701 BUG_ON(ret);
4702 BUG_ON(wc->refs[level - 1] == 0);
4703
4704 if (wc->refs[level - 1] > 1) {
4705 if (!wc->update_ref ||
4706 generation <= root->root_key.offset)
4707 goto skip;
4708
4709 btrfs_node_key_to_cpu(path->nodes[level], &key,
4710 path->slots[level]);
4711 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
4712 if (ret < 0)
4713 goto skip;
4714
4715 wc->stage = UPDATE_BACKREF;
4716 wc->shared_level = level - 1;
4717 }
4718 }
4719
4720 if (!btrfs_buffer_uptodate(next, generation)) {
4721 btrfs_tree_unlock(next);
4722 free_extent_buffer(next);
4723 next = NULL;
4724 }
4725
4726 if (!next) {
4727 if (reada && level == 1)
4728 reada_walk_down(trans, root, wc, path);
4729 next = read_tree_block(root, bytenr, blocksize, generation);
4730 btrfs_tree_lock(next);
4731 btrfs_set_lock_blocking(next);
4732 }
4733
4734 level--;
4735 BUG_ON(level != btrfs_header_level(next));
4736 path->nodes[level] = next;
4737 path->slots[level] = 0;
4738 path->locks[level] = 1;
4739 wc->level = level;
4740 if (wc->level == 1)
4741 wc->reada_slot = 0;
4742 return 0;
4743skip:
4744 wc->refs[level - 1] = 0;
4745 wc->flags[level - 1] = 0;
4746
4747 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
4748 parent = path->nodes[level]->start;
4749 } else {
4750 BUG_ON(root->root_key.objectid !=
4751 btrfs_header_owner(path->nodes[level]));
4752 parent = 0;
4753 }
4754
4755 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
4756 root->root_key.objectid, level - 1, 0);
4757 BUG_ON(ret);
4758
4759 btrfs_tree_unlock(next);
4760 free_extent_buffer(next);
4761 return 1;
4762}
4763
4764/*
4882 * hepler to process tree block while walking up the tree. 4765 * hepler to process tree block while walking up the tree.
4883 * 4766 *
4884 * when wc->stage == DROP_REFERENCE, this function drops 4767 * when wc->stage == DROP_REFERENCE, this function drops
@@ -4905,7 +4788,6 @@ static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
4905 if (level < wc->shared_level) 4788 if (level < wc->shared_level)
4906 goto out; 4789 goto out;
4907 4790
4908 BUG_ON(wc->refs[level] <= 1);
4909 ret = find_next_key(path, level + 1, &wc->update_progress); 4791 ret = find_next_key(path, level + 1, &wc->update_progress);
4910 if (ret > 0) 4792 if (ret > 0)
4911 wc->update_ref = 0; 4793 wc->update_ref = 0;
@@ -4936,8 +4818,6 @@ static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
4936 path->locks[level] = 0; 4818 path->locks[level] = 0;
4937 return 1; 4819 return 1;
4938 } 4820 }
4939 } else {
4940 BUG_ON(level != 0);
4941 } 4821 }
4942 } 4822 }
4943 4823
@@ -4990,17 +4870,13 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4990 struct btrfs_path *path, 4870 struct btrfs_path *path,
4991 struct walk_control *wc) 4871 struct walk_control *wc)
4992{ 4872{
4993 struct extent_buffer *next;
4994 struct extent_buffer *cur;
4995 u64 bytenr;
4996 u64 ptr_gen;
4997 u32 blocksize;
4998 int level = wc->level; 4873 int level = wc->level;
4999 int ret; 4874 int ret;
5000 4875
5001 while (level >= 0) { 4876 while (level >= 0) {
5002 cur = path->nodes[level]; 4877 if (path->slots[level] >=
5003 BUG_ON(path->slots[level] >= btrfs_header_nritems(cur)); 4878 btrfs_header_nritems(path->nodes[level]))
4879 break;
5004 4880
5005 ret = walk_down_proc(trans, root, path, wc); 4881 ret = walk_down_proc(trans, root, path, wc);
5006 if (ret > 0) 4882 if (ret > 0)
@@ -5009,20 +4885,12 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5009 if (level == 0) 4885 if (level == 0)
5010 break; 4886 break;
5011 4887
5012 bytenr = btrfs_node_blockptr(cur, path->slots[level]); 4888 ret = do_walk_down(trans, root, path, wc);
5013 blocksize = btrfs_level_size(root, level - 1); 4889 if (ret > 0) {
5014 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[level]); 4890 path->slots[level]++;
5015 4891 continue;
5016 next = read_tree_block(root, bytenr, blocksize, ptr_gen); 4892 }
5017 btrfs_tree_lock(next); 4893 level = wc->level;
5018 btrfs_set_lock_blocking(next);
5019
5020 level--;
5021 BUG_ON(level != btrfs_header_level(next));
5022 path->nodes[level] = next;
5023 path->slots[level] = 0;
5024 path->locks[level] = 1;
5025 wc->level = level;
5026 } 4894 }
5027 return 0; 4895 return 0;
5028} 4896}
@@ -5112,9 +4980,7 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5112 err = ret; 4980 err = ret;
5113 goto out; 4981 goto out;
5114 } 4982 }
5115 btrfs_node_key_to_cpu(path->nodes[level], &key, 4983 WARN_ON(ret > 0);
5116 path->slots[level]);
5117 WARN_ON(memcmp(&key, &wc->update_progress, sizeof(key)));
5118 4984
5119 /* 4985 /*
5120 * unlock our path, this is safe because only this 4986 * unlock our path, this is safe because only this
@@ -5149,6 +5015,7 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5149 wc->stage = DROP_REFERENCE; 5015 wc->stage = DROP_REFERENCE;
5150 wc->update_ref = update_ref; 5016 wc->update_ref = update_ref;
5151 wc->keep_locks = 0; 5017 wc->keep_locks = 0;
5018 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5152 5019
5153 while (1) { 5020 while (1) {
5154 ret = walk_down_tree(trans, root, path, wc); 5021 ret = walk_down_tree(trans, root, path, wc);
@@ -5201,9 +5068,24 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5201 ret = btrfs_del_root(trans, tree_root, &root->root_key); 5068 ret = btrfs_del_root(trans, tree_root, &root->root_key);
5202 BUG_ON(ret); 5069 BUG_ON(ret);
5203 5070
5204 free_extent_buffer(root->node); 5071 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5205 free_extent_buffer(root->commit_root); 5072 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
5206 kfree(root); 5073 NULL, NULL);
5074 BUG_ON(ret < 0);
5075 if (ret > 0) {
5076 ret = btrfs_del_orphan_item(trans, tree_root,
5077 root->root_key.objectid);
5078 BUG_ON(ret);
5079 }
5080 }
5081
5082 if (root->in_radix) {
5083 btrfs_free_fs_root(tree_root->fs_info, root);
5084 } else {
5085 free_extent_buffer(root->node);
5086 free_extent_buffer(root->commit_root);
5087 kfree(root);
5088 }
5207out: 5089out:
5208 btrfs_end_transaction(trans, tree_root); 5090 btrfs_end_transaction(trans, tree_root);
5209 kfree(wc); 5091 kfree(wc);
@@ -5255,6 +5137,7 @@ int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5255 wc->stage = DROP_REFERENCE; 5137 wc->stage = DROP_REFERENCE;
5256 wc->update_ref = 0; 5138 wc->update_ref = 0;
5257 wc->keep_locks = 1; 5139 wc->keep_locks = 1;
5140 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5258 5141
5259 while (1) { 5142 while (1) {
5260 wret = walk_down_tree(trans, root, path, wc); 5143 wret = walk_down_tree(trans, root, path, wc);
@@ -5397,9 +5280,9 @@ static noinline int relocate_data_extent(struct inode *reloc_inode,
5397 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); 5280 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5398 while (1) { 5281 while (1) {
5399 int ret; 5282 int ret;
5400 spin_lock(&em_tree->lock); 5283 write_lock(&em_tree->lock);
5401 ret = add_extent_mapping(em_tree, em); 5284 ret = add_extent_mapping(em_tree, em);
5402 spin_unlock(&em_tree->lock); 5285 write_unlock(&em_tree->lock);
5403 if (ret != -EEXIST) { 5286 if (ret != -EEXIST) {
5404 free_extent_map(em); 5287 free_extent_map(em);
5405 break; 5288 break;
@@ -6842,287 +6725,86 @@ int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
6842 return 0; 6725 return 0;
6843} 6726}
6844 6727
6845#if 0 6728/*
6846static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 6729 * checks to see if its even possible to relocate this block group.
6847 struct btrfs_root *root, 6730 *
6848 u64 objectid, u64 size) 6731 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6849{ 6732 * ok to go ahead and try.
6850 struct btrfs_path *path; 6733 */
6851 struct btrfs_inode_item *item; 6734int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
6852 struct extent_buffer *leaf;
6853 int ret;
6854
6855 path = btrfs_alloc_path();
6856 if (!path)
6857 return -ENOMEM;
6858
6859 path->leave_spinning = 1;
6860 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
6861 if (ret)
6862 goto out;
6863
6864 leaf = path->nodes[0];
6865 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
6866 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
6867 btrfs_set_inode_generation(leaf, item, 1);
6868 btrfs_set_inode_size(leaf, item, size);
6869 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
6870 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
6871 btrfs_mark_buffer_dirty(leaf);
6872 btrfs_release_path(root, path);
6873out:
6874 btrfs_free_path(path);
6875 return ret;
6876}
6877
6878static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
6879 struct btrfs_block_group_cache *group)
6880{ 6735{
6881 struct inode *inode = NULL; 6736 struct btrfs_block_group_cache *block_group;
6882 struct btrfs_trans_handle *trans; 6737 struct btrfs_space_info *space_info;
6883 struct btrfs_root *root; 6738 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6884 struct btrfs_key root_key; 6739 struct btrfs_device *device;
6885 u64 objectid = BTRFS_FIRST_FREE_OBJECTID; 6740 int full = 0;
6886 int err = 0; 6741 int ret = 0;
6887 6742
6888 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID; 6743 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
6889 root_key.type = BTRFS_ROOT_ITEM_KEY;
6890 root_key.offset = (u64)-1;
6891 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
6892 if (IS_ERR(root))
6893 return ERR_CAST(root);
6894 6744
6895 trans = btrfs_start_transaction(root, 1); 6745 /* odd, couldn't find the block group, leave it alone */
6896 BUG_ON(!trans); 6746 if (!block_group)
6747 return -1;
6897 6748
6898 err = btrfs_find_free_objectid(trans, root, objectid, &objectid); 6749 /* no bytes used, we're good */
6899 if (err) 6750 if (!btrfs_block_group_used(&block_group->item))
6900 goto out; 6751 goto out;
6901 6752
6902 err = __insert_orphan_inode(trans, root, objectid, group->key.offset); 6753 space_info = block_group->space_info;
6903 BUG_ON(err); 6754 spin_lock(&space_info->lock);
6904
6905 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
6906 group->key.offset, 0, group->key.offset,
6907 0, 0, 0);
6908 BUG_ON(err);
6909
6910 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
6911 if (inode->i_state & I_NEW) {
6912 BTRFS_I(inode)->root = root;
6913 BTRFS_I(inode)->location.objectid = objectid;
6914 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
6915 BTRFS_I(inode)->location.offset = 0;
6916 btrfs_read_locked_inode(inode);
6917 unlock_new_inode(inode);
6918 BUG_ON(is_bad_inode(inode));
6919 } else {
6920 BUG_ON(1);
6921 }
6922 BTRFS_I(inode)->index_cnt = group->key.objectid;
6923
6924 err = btrfs_orphan_add(trans, inode);
6925out:
6926 btrfs_end_transaction(trans, root);
6927 if (err) {
6928 if (inode)
6929 iput(inode);
6930 inode = ERR_PTR(err);
6931 }
6932 return inode;
6933}
6934
6935int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
6936{
6937
6938 struct btrfs_ordered_sum *sums;
6939 struct btrfs_sector_sum *sector_sum;
6940 struct btrfs_ordered_extent *ordered;
6941 struct btrfs_root *root = BTRFS_I(inode)->root;
6942 struct list_head list;
6943 size_t offset;
6944 int ret;
6945 u64 disk_bytenr;
6946
6947 INIT_LIST_HEAD(&list);
6948
6949 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
6950 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
6951
6952 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
6953 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
6954 disk_bytenr + len - 1, &list);
6955
6956 while (!list_empty(&list)) {
6957 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
6958 list_del_init(&sums->list);
6959
6960 sector_sum = sums->sums;
6961 sums->bytenr = ordered->start;
6962 6755
6963 offset = 0; 6756 full = space_info->full;
6964 while (offset < sums->len) {
6965 sector_sum->bytenr += ordered->start - disk_bytenr;
6966 sector_sum++;
6967 offset += root->sectorsize;
6968 }
6969 6757
6970 btrfs_add_ordered_sum(inode, ordered, sums); 6758 /*
6759 * if this is the last block group we have in this space, we can't
6760 * relocate it unless we're able to allocate a new chunk below.
6761 *
6762 * Otherwise, we need to make sure we have room in the space to handle
6763 * all of the extents from this block group. If we can, we're good
6764 */
6765 if ((space_info->total_bytes != block_group->key.offset) &&
6766 (space_info->bytes_used + space_info->bytes_reserved +
6767 space_info->bytes_pinned + space_info->bytes_readonly +
6768 btrfs_block_group_used(&block_group->item) <
6769 space_info->total_bytes)) {
6770 spin_unlock(&space_info->lock);
6771 goto out;
6971 } 6772 }
6972 btrfs_put_ordered_extent(ordered); 6773 spin_unlock(&space_info->lock);
6973 return 0;
6974}
6975
6976int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
6977{
6978 struct btrfs_trans_handle *trans;
6979 struct btrfs_path *path;
6980 struct btrfs_fs_info *info = root->fs_info;
6981 struct extent_buffer *leaf;
6982 struct inode *reloc_inode;
6983 struct btrfs_block_group_cache *block_group;
6984 struct btrfs_key key;
6985 u64 skipped;
6986 u64 cur_byte;
6987 u64 total_found;
6988 u32 nritems;
6989 int ret;
6990 int progress;
6991 int pass = 0;
6992
6993 root = root->fs_info->extent_root;
6994
6995 block_group = btrfs_lookup_block_group(info, group_start);
6996 BUG_ON(!block_group);
6997
6998 printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
6999 (unsigned long long)block_group->key.objectid,
7000 (unsigned long long)block_group->flags);
7001
7002 path = btrfs_alloc_path();
7003 BUG_ON(!path);
7004
7005 reloc_inode = create_reloc_inode(info, block_group);
7006 BUG_ON(IS_ERR(reloc_inode));
7007
7008 __alloc_chunk_for_shrink(root, block_group, 1);
7009 set_block_group_readonly(block_group);
7010
7011 btrfs_start_delalloc_inodes(info->tree_root);
7012 btrfs_wait_ordered_extents(info->tree_root, 0);
7013again:
7014 skipped = 0;
7015 total_found = 0;
7016 progress = 0;
7017 key.objectid = block_group->key.objectid;
7018 key.offset = 0;
7019 key.type = 0;
7020 cur_byte = key.objectid;
7021
7022 trans = btrfs_start_transaction(info->tree_root, 1);
7023 btrfs_commit_transaction(trans, info->tree_root);
7024 6774
7025 mutex_lock(&root->fs_info->cleaner_mutex); 6775 /*
7026 btrfs_clean_old_snapshots(info->tree_root); 6776 * ok we don't have enough space, but maybe we have free space on our
7027 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1); 6777 * devices to allocate new chunks for relocation, so loop through our
7028 mutex_unlock(&root->fs_info->cleaner_mutex); 6778 * alloc devices and guess if we have enough space. However, if we
6779 * were marked as full, then we know there aren't enough chunks, and we
6780 * can just return.
6781 */
6782 ret = -1;
6783 if (full)
6784 goto out;
7029 6785
7030 trans = btrfs_start_transaction(info->tree_root, 1); 6786 mutex_lock(&root->fs_info->chunk_mutex);
7031 btrfs_commit_transaction(trans, info->tree_root); 6787 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
6788 u64 min_free = btrfs_block_group_used(&block_group->item);
6789 u64 dev_offset, max_avail;
7032 6790
7033 while (1) { 6791 /*
7034 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 6792 * check to make sure we can actually find a chunk with enough
7035 if (ret < 0) 6793 * space to fit our block group in.
7036 goto out; 6794 */
7037next: 6795 if (device->total_bytes > device->bytes_used + min_free) {
7038 leaf = path->nodes[0]; 6796 ret = find_free_dev_extent(NULL, device, min_free,
7039 nritems = btrfs_header_nritems(leaf); 6797 &dev_offset, &max_avail);
7040 if (path->slots[0] >= nritems) { 6798 if (!ret)
7041 ret = btrfs_next_leaf(root, path);
7042 if (ret < 0)
7043 goto out;
7044 if (ret == 1) {
7045 ret = 0;
7046 break; 6799 break;
7047 } 6800 ret = -1;
7048 leaf = path->nodes[0];
7049 nritems = btrfs_header_nritems(leaf);
7050 } 6801 }
7051
7052 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7053
7054 if (key.objectid >= block_group->key.objectid +
7055 block_group->key.offset)
7056 break;
7057
7058 if (progress && need_resched()) {
7059 btrfs_release_path(root, path);
7060 cond_resched();
7061 progress = 0;
7062 continue;
7063 }
7064 progress = 1;
7065
7066 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
7067 key.objectid + key.offset <= cur_byte) {
7068 path->slots[0]++;
7069 goto next;
7070 }
7071
7072 total_found++;
7073 cur_byte = key.objectid + key.offset;
7074 btrfs_release_path(root, path);
7075
7076 __alloc_chunk_for_shrink(root, block_group, 0);
7077 ret = relocate_one_extent(root, path, &key, block_group,
7078 reloc_inode, pass);
7079 BUG_ON(ret < 0);
7080 if (ret > 0)
7081 skipped++;
7082
7083 key.objectid = cur_byte;
7084 key.type = 0;
7085 key.offset = 0;
7086 }
7087
7088 btrfs_release_path(root, path);
7089
7090 if (pass == 0) {
7091 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
7092 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
7093 }
7094
7095 if (total_found > 0) {
7096 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
7097 (unsigned long long)total_found, pass);
7098 pass++;
7099 if (total_found == skipped && pass > 2) {
7100 iput(reloc_inode);
7101 reloc_inode = create_reloc_inode(info, block_group);
7102 pass = 0;
7103 }
7104 goto again;
7105 } 6802 }
7106 6803 mutex_unlock(&root->fs_info->chunk_mutex);
7107 /* delete reloc_inode */
7108 iput(reloc_inode);
7109
7110 /* unpin extents in this range */
7111 trans = btrfs_start_transaction(info->tree_root, 1);
7112 btrfs_commit_transaction(trans, info->tree_root);
7113
7114 spin_lock(&block_group->lock);
7115 WARN_ON(block_group->pinned > 0);
7116 WARN_ON(block_group->reserved > 0);
7117 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
7118 spin_unlock(&block_group->lock);
7119 btrfs_put_block_group(block_group);
7120 ret = 0;
7121out: 6804out:
7122 btrfs_free_path(path); 6805 btrfs_put_block_group(block_group);
7123 return ret; 6806 return ret;
7124} 6807}
7125#endif
7126 6808
7127static int find_first_block_group(struct btrfs_root *root, 6809static int find_first_block_group(struct btrfs_root *root,
7128 struct btrfs_path *path, struct btrfs_key *key) 6810 struct btrfs_path *path, struct btrfs_key *key)
@@ -7165,8 +6847,18 @@ int btrfs_free_block_groups(struct btrfs_fs_info *info)
7165{ 6847{
7166 struct btrfs_block_group_cache *block_group; 6848 struct btrfs_block_group_cache *block_group;
7167 struct btrfs_space_info *space_info; 6849 struct btrfs_space_info *space_info;
6850 struct btrfs_caching_control *caching_ctl;
7168 struct rb_node *n; 6851 struct rb_node *n;
7169 6852
6853 down_write(&info->extent_commit_sem);
6854 while (!list_empty(&info->caching_block_groups)) {
6855 caching_ctl = list_entry(info->caching_block_groups.next,
6856 struct btrfs_caching_control, list);
6857 list_del(&caching_ctl->list);
6858 put_caching_control(caching_ctl);
6859 }
6860 up_write(&info->extent_commit_sem);
6861
7170 spin_lock(&info->block_group_cache_lock); 6862 spin_lock(&info->block_group_cache_lock);
7171 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { 6863 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7172 block_group = rb_entry(n, struct btrfs_block_group_cache, 6864 block_group = rb_entry(n, struct btrfs_block_group_cache,
@@ -7180,8 +6872,7 @@ int btrfs_free_block_groups(struct btrfs_fs_info *info)
7180 up_write(&block_group->space_info->groups_sem); 6872 up_write(&block_group->space_info->groups_sem);
7181 6873
7182 if (block_group->cached == BTRFS_CACHE_STARTED) 6874 if (block_group->cached == BTRFS_CACHE_STARTED)
7183 wait_event(block_group->caching_q, 6875 wait_block_group_cache_done(block_group);
7184 block_group_cache_done(block_group));
7185 6876
7186 btrfs_remove_free_space_cache(block_group); 6877 btrfs_remove_free_space_cache(block_group);
7187 6878
@@ -7251,7 +6942,6 @@ int btrfs_read_block_groups(struct btrfs_root *root)
7251 spin_lock_init(&cache->lock); 6942 spin_lock_init(&cache->lock);
7252 spin_lock_init(&cache->tree_lock); 6943 spin_lock_init(&cache->tree_lock);
7253 cache->fs_info = info; 6944 cache->fs_info = info;
7254 init_waitqueue_head(&cache->caching_q);
7255 INIT_LIST_HEAD(&cache->list); 6945 INIT_LIST_HEAD(&cache->list);
7256 INIT_LIST_HEAD(&cache->cluster_list); 6946 INIT_LIST_HEAD(&cache->cluster_list);
7257 6947
@@ -7273,8 +6963,6 @@ int btrfs_read_block_groups(struct btrfs_root *root)
7273 cache->flags = btrfs_block_group_flags(&cache->item); 6963 cache->flags = btrfs_block_group_flags(&cache->item);
7274 cache->sectorsize = root->sectorsize; 6964 cache->sectorsize = root->sectorsize;
7275 6965
7276 remove_sb_from_cache(root, cache);
7277
7278 /* 6966 /*
7279 * check for two cases, either we are full, and therefore 6967 * check for two cases, either we are full, and therefore
7280 * don't need to bother with the caching work since we won't 6968 * don't need to bother with the caching work since we won't
@@ -7283,13 +6971,19 @@ int btrfs_read_block_groups(struct btrfs_root *root)
7283 * time, particularly in the full case. 6971 * time, particularly in the full case.
7284 */ 6972 */
7285 if (found_key.offset == btrfs_block_group_used(&cache->item)) { 6973 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
6974 exclude_super_stripes(root, cache);
6975 cache->last_byte_to_unpin = (u64)-1;
7286 cache->cached = BTRFS_CACHE_FINISHED; 6976 cache->cached = BTRFS_CACHE_FINISHED;
6977 free_excluded_extents(root, cache);
7287 } else if (btrfs_block_group_used(&cache->item) == 0) { 6978 } else if (btrfs_block_group_used(&cache->item) == 0) {
6979 exclude_super_stripes(root, cache);
6980 cache->last_byte_to_unpin = (u64)-1;
7288 cache->cached = BTRFS_CACHE_FINISHED; 6981 cache->cached = BTRFS_CACHE_FINISHED;
7289 add_new_free_space(cache, root->fs_info, 6982 add_new_free_space(cache, root->fs_info,
7290 found_key.objectid, 6983 found_key.objectid,
7291 found_key.objectid + 6984 found_key.objectid +
7292 found_key.offset); 6985 found_key.offset);
6986 free_excluded_extents(root, cache);
7293 } 6987 }
7294 6988
7295 ret = update_space_info(info, cache->flags, found_key.offset, 6989 ret = update_space_info(info, cache->flags, found_key.offset,
@@ -7297,6 +6991,10 @@ int btrfs_read_block_groups(struct btrfs_root *root)
7297 &space_info); 6991 &space_info);
7298 BUG_ON(ret); 6992 BUG_ON(ret);
7299 cache->space_info = space_info; 6993 cache->space_info = space_info;
6994 spin_lock(&cache->space_info->lock);
6995 cache->space_info->bytes_super += cache->bytes_super;
6996 spin_unlock(&cache->space_info->lock);
6997
7300 down_write(&space_info->groups_sem); 6998 down_write(&space_info->groups_sem);
7301 list_add_tail(&cache->list, &space_info->block_groups); 6999 list_add_tail(&cache->list, &space_info->block_groups);
7302 up_write(&space_info->groups_sem); 7000 up_write(&space_info->groups_sem);
@@ -7346,7 +7044,6 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7346 atomic_set(&cache->count, 1); 7044 atomic_set(&cache->count, 1);
7347 spin_lock_init(&cache->lock); 7045 spin_lock_init(&cache->lock);
7348 spin_lock_init(&cache->tree_lock); 7046 spin_lock_init(&cache->tree_lock);
7349 init_waitqueue_head(&cache->caching_q);
7350 INIT_LIST_HEAD(&cache->list); 7047 INIT_LIST_HEAD(&cache->list);
7351 INIT_LIST_HEAD(&cache->cluster_list); 7048 INIT_LIST_HEAD(&cache->cluster_list);
7352 7049
@@ -7355,15 +7052,23 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7355 cache->flags = type; 7052 cache->flags = type;
7356 btrfs_set_block_group_flags(&cache->item, type); 7053 btrfs_set_block_group_flags(&cache->item, type);
7357 7054
7055 cache->last_byte_to_unpin = (u64)-1;
7358 cache->cached = BTRFS_CACHE_FINISHED; 7056 cache->cached = BTRFS_CACHE_FINISHED;
7359 remove_sb_from_cache(root, cache); 7057 exclude_super_stripes(root, cache);
7360 7058
7361 add_new_free_space(cache, root->fs_info, chunk_offset, 7059 add_new_free_space(cache, root->fs_info, chunk_offset,
7362 chunk_offset + size); 7060 chunk_offset + size);
7363 7061
7062 free_excluded_extents(root, cache);
7063
7364 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, 7064 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7365 &cache->space_info); 7065 &cache->space_info);
7366 BUG_ON(ret); 7066 BUG_ON(ret);
7067
7068 spin_lock(&cache->space_info->lock);
7069 cache->space_info->bytes_super += cache->bytes_super;
7070 spin_unlock(&cache->space_info->lock);
7071
7367 down_write(&cache->space_info->groups_sem); 7072 down_write(&cache->space_info->groups_sem);
7368 list_add_tail(&cache->list, &cache->space_info->block_groups); 7073 list_add_tail(&cache->list, &cache->space_info->block_groups);
7369 up_write(&cache->space_info->groups_sem); 7074 up_write(&cache->space_info->groups_sem);
@@ -7429,8 +7134,7 @@ int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7429 up_write(&block_group->space_info->groups_sem); 7134 up_write(&block_group->space_info->groups_sem);
7430 7135
7431 if (block_group->cached == BTRFS_CACHE_STARTED) 7136 if (block_group->cached == BTRFS_CACHE_STARTED)
7432 wait_event(block_group->caching_q, 7137 wait_block_group_cache_done(block_group);
7433 block_group_cache_done(block_group));
7434 7138
7435 btrfs_remove_free_space_cache(block_group); 7139 btrfs_remove_free_space_cache(block_group);
7436 7140
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 68260180f587..0cb88f8146ea 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -367,10 +367,10 @@ static int insert_state(struct extent_io_tree *tree,
367 } 367 }
368 if (bits & EXTENT_DIRTY) 368 if (bits & EXTENT_DIRTY)
369 tree->dirty_bytes += end - start + 1; 369 tree->dirty_bytes += end - start + 1;
370 set_state_cb(tree, state, bits);
371 state->state |= bits;
372 state->start = start; 370 state->start = start;
373 state->end = end; 371 state->end = end;
372 set_state_cb(tree, state, bits);
373 state->state |= bits;
374 node = tree_insert(&tree->state, end, &state->rb_node); 374 node = tree_insert(&tree->state, end, &state->rb_node);
375 if (node) { 375 if (node) {
376 struct extent_state *found; 376 struct extent_state *found;
@@ -471,10 +471,14 @@ 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. 471 * bits were already set, or zero if none of the bits were already set.
472 */ 472 */
473int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, 473int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
474 int bits, int wake, int delete, gfp_t mask) 474 int bits, int wake, int delete,
475 struct extent_state **cached_state,
476 gfp_t mask)
475{ 477{
476 struct extent_state *state; 478 struct extent_state *state;
479 struct extent_state *cached;
477 struct extent_state *prealloc = NULL; 480 struct extent_state *prealloc = NULL;
481 struct rb_node *next_node;
478 struct rb_node *node; 482 struct rb_node *node;
479 u64 last_end; 483 u64 last_end;
480 int err; 484 int err;
@@ -488,6 +492,17 @@ again:
488 } 492 }
489 493
490 spin_lock(&tree->lock); 494 spin_lock(&tree->lock);
495 if (cached_state) {
496 cached = *cached_state;
497 *cached_state = NULL;
498 cached_state = NULL;
499 if (cached && cached->tree && cached->start == start) {
500 atomic_dec(&cached->refs);
501 state = cached;
502 goto hit_next;
503 }
504 free_extent_state(cached);
505 }
491 /* 506 /*
492 * this search will find the extents that end after 507 * this search will find the extents that end after
493 * our range starts 508 * our range starts
@@ -496,6 +511,7 @@ again:
496 if (!node) 511 if (!node)
497 goto out; 512 goto out;
498 state = rb_entry(node, struct extent_state, rb_node); 513 state = rb_entry(node, struct extent_state, rb_node);
514hit_next:
499 if (state->start > end) 515 if (state->start > end)
500 goto out; 516 goto out;
501 WARN_ON(state->end < start); 517 WARN_ON(state->end < start);
@@ -531,8 +547,6 @@ again:
531 if (last_end == (u64)-1) 547 if (last_end == (u64)-1)
532 goto out; 548 goto out;
533 start = last_end + 1; 549 start = last_end + 1;
534 } else {
535 start = state->start;
536 } 550 }
537 goto search_again; 551 goto search_again;
538 } 552 }
@@ -550,16 +564,28 @@ again:
550 564
551 if (wake) 565 if (wake)
552 wake_up(&state->wq); 566 wake_up(&state->wq);
567
553 set |= clear_state_bit(tree, prealloc, bits, 568 set |= clear_state_bit(tree, prealloc, bits,
554 wake, delete); 569 wake, delete);
555 prealloc = NULL; 570 prealloc = NULL;
556 goto out; 571 goto out;
557 } 572 }
558 573
574 if (state->end < end && prealloc && !need_resched())
575 next_node = rb_next(&state->rb_node);
576 else
577 next_node = NULL;
578
559 set |= clear_state_bit(tree, state, bits, wake, delete); 579 set |= clear_state_bit(tree, state, bits, wake, delete);
560 if (last_end == (u64)-1) 580 if (last_end == (u64)-1)
561 goto out; 581 goto out;
562 start = last_end + 1; 582 start = last_end + 1;
583 if (start <= end && next_node) {
584 state = rb_entry(next_node, struct extent_state,
585 rb_node);
586 if (state->start == start)
587 goto hit_next;
588 }
563 goto search_again; 589 goto search_again;
564 590
565out: 591out:
@@ -653,28 +679,40 @@ static void set_state_bits(struct extent_io_tree *tree,
653 state->state |= bits; 679 state->state |= bits;
654} 680}
655 681
682static void cache_state(struct extent_state *state,
683 struct extent_state **cached_ptr)
684{
685 if (cached_ptr && !(*cached_ptr)) {
686 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
687 *cached_ptr = state;
688 atomic_inc(&state->refs);
689 }
690 }
691}
692
656/* 693/*
657 * set some bits on a range in the tree. This may require allocations 694 * 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. 695 * sleeping, so the gfp mask is used to indicate what is allowed.
659 * 696 *
660 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the 697 * 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 698 * part of the range already has the desired bits set. The start of the
662 * range is returned in failed_start in this case. 699 * existing range is returned in failed_start in this case.
663 * 700 *
664 * [start, end] is inclusive 701 * [start, end] is inclusive This takes the tree lock.
665 * This takes the tree lock.
666 */ 702 */
703
667static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, 704static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
668 int bits, int exclusive, u64 *failed_start, 705 int bits, int exclusive_bits, u64 *failed_start,
706 struct extent_state **cached_state,
669 gfp_t mask) 707 gfp_t mask)
670{ 708{
671 struct extent_state *state; 709 struct extent_state *state;
672 struct extent_state *prealloc = NULL; 710 struct extent_state *prealloc = NULL;
673 struct rb_node *node; 711 struct rb_node *node;
674 int err = 0; 712 int err = 0;
675 int set;
676 u64 last_start; 713 u64 last_start;
677 u64 last_end; 714 u64 last_end;
715
678again: 716again:
679 if (!prealloc && (mask & __GFP_WAIT)) { 717 if (!prealloc && (mask & __GFP_WAIT)) {
680 prealloc = alloc_extent_state(mask); 718 prealloc = alloc_extent_state(mask);
@@ -683,6 +721,13 @@ again:
683 } 721 }
684 722
685 spin_lock(&tree->lock); 723 spin_lock(&tree->lock);
724 if (cached_state && *cached_state) {
725 state = *cached_state;
726 if (state->start == start && state->tree) {
727 node = &state->rb_node;
728 goto hit_next;
729 }
730 }
686 /* 731 /*
687 * this search will find all the extents that end after 732 * this search will find all the extents that end after
688 * our range starts. 733 * our range starts.
@@ -694,8 +739,8 @@ again:
694 BUG_ON(err == -EEXIST); 739 BUG_ON(err == -EEXIST);
695 goto out; 740 goto out;
696 } 741 }
697
698 state = rb_entry(node, struct extent_state, rb_node); 742 state = rb_entry(node, struct extent_state, rb_node);
743hit_next:
699 last_start = state->start; 744 last_start = state->start;
700 last_end = state->end; 745 last_end = state->end;
701 746
@@ -706,17 +751,29 @@ again:
706 * Just lock what we found and keep going 751 * Just lock what we found and keep going
707 */ 752 */
708 if (state->start == start && state->end <= end) { 753 if (state->start == start && state->end <= end) {
709 set = state->state & bits; 754 struct rb_node *next_node;
710 if (set && exclusive) { 755 if (state->state & exclusive_bits) {
711 *failed_start = state->start; 756 *failed_start = state->start;
712 err = -EEXIST; 757 err = -EEXIST;
713 goto out; 758 goto out;
714 } 759 }
760
715 set_state_bits(tree, state, bits); 761 set_state_bits(tree, state, bits);
762 cache_state(state, cached_state);
716 merge_state(tree, state); 763 merge_state(tree, state);
717 if (last_end == (u64)-1) 764 if (last_end == (u64)-1)
718 goto out; 765 goto out;
766
719 start = last_end + 1; 767 start = last_end + 1;
768 if (start < end && prealloc && !need_resched()) {
769 next_node = rb_next(node);
770 if (next_node) {
771 state = rb_entry(next_node, struct extent_state,
772 rb_node);
773 if (state->start == start)
774 goto hit_next;
775 }
776 }
720 goto search_again; 777 goto search_again;
721 } 778 }
722 779
@@ -737,8 +794,7 @@ again:
737 * desired bit on it. 794 * desired bit on it.
738 */ 795 */
739 if (state->start < start) { 796 if (state->start < start) {
740 set = state->state & bits; 797 if (state->state & exclusive_bits) {
741 if (exclusive && set) {
742 *failed_start = start; 798 *failed_start = start;
743 err = -EEXIST; 799 err = -EEXIST;
744 goto out; 800 goto out;
@@ -750,12 +806,11 @@ again:
750 goto out; 806 goto out;
751 if (state->end <= end) { 807 if (state->end <= end) {
752 set_state_bits(tree, state, bits); 808 set_state_bits(tree, state, bits);
809 cache_state(state, cached_state);
753 merge_state(tree, state); 810 merge_state(tree, state);
754 if (last_end == (u64)-1) 811 if (last_end == (u64)-1)
755 goto out; 812 goto out;
756 start = last_end + 1; 813 start = last_end + 1;
757 } else {
758 start = state->start;
759 } 814 }
760 goto search_again; 815 goto search_again;
761 } 816 }
@@ -774,6 +829,7 @@ again:
774 this_end = last_start - 1; 829 this_end = last_start - 1;
775 err = insert_state(tree, prealloc, start, this_end, 830 err = insert_state(tree, prealloc, start, this_end,
776 bits); 831 bits);
832 cache_state(prealloc, cached_state);
777 prealloc = NULL; 833 prealloc = NULL;
778 BUG_ON(err == -EEXIST); 834 BUG_ON(err == -EEXIST);
779 if (err) 835 if (err)
@@ -788,8 +844,7 @@ again:
788 * on the first half 844 * on the first half
789 */ 845 */
790 if (state->start <= end && state->end > end) { 846 if (state->start <= end && state->end > end) {
791 set = state->state & bits; 847 if (state->state & exclusive_bits) {
792 if (exclusive && set) {
793 *failed_start = start; 848 *failed_start = start;
794 err = -EEXIST; 849 err = -EEXIST;
795 goto out; 850 goto out;
@@ -798,6 +853,7 @@ again:
798 BUG_ON(err == -EEXIST); 853 BUG_ON(err == -EEXIST);
799 854
800 set_state_bits(tree, prealloc, bits); 855 set_state_bits(tree, prealloc, bits);
856 cache_state(prealloc, cached_state);
801 merge_state(tree, prealloc); 857 merge_state(tree, prealloc);
802 prealloc = NULL; 858 prealloc = NULL;
803 goto out; 859 goto out;
@@ -826,86 +882,64 @@ int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
826 gfp_t mask) 882 gfp_t mask)
827{ 883{
828 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL, 884 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
829 mask); 885 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} 886}
837 887
838int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, 888int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
839 int bits, gfp_t mask) 889 int bits, gfp_t mask)
840{ 890{
841 return set_extent_bit(tree, start, end, bits, 0, NULL, 891 return set_extent_bit(tree, start, end, bits, 0, NULL,
842 mask); 892 NULL, mask);
843} 893}
844 894
845int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, 895int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
846 int bits, gfp_t mask) 896 int bits, gfp_t mask)
847{ 897{
848 return clear_extent_bit(tree, start, end, bits, 0, 0, mask); 898 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
849} 899}
850 900
851int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, 901int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
852 gfp_t mask) 902 gfp_t mask)
853{ 903{
854 return set_extent_bit(tree, start, end, 904 return set_extent_bit(tree, start, end,
855 EXTENT_DELALLOC | EXTENT_DIRTY, 905 EXTENT_DELALLOC | EXTENT_DIRTY | EXTENT_UPTODATE,
856 0, NULL, mask); 906 0, NULL, NULL, mask);
857} 907}
858 908
859int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, 909int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
860 gfp_t mask) 910 gfp_t mask)
861{ 911{
862 return clear_extent_bit(tree, start, end, 912 return clear_extent_bit(tree, start, end,
863 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask); 913 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
864} 914 NULL, mask);
865
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} 915}
871 916
872int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, 917int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
873 gfp_t mask) 918 gfp_t mask)
874{ 919{
875 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL, 920 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
876 mask); 921 NULL, mask);
877} 922}
878 923
879static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end, 924static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
880 gfp_t mask) 925 gfp_t mask)
881{ 926{
882 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask); 927 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0,
928 NULL, mask);
883} 929}
884 930
885int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, 931int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
886 gfp_t mask) 932 gfp_t mask)
887{ 933{
888 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL, 934 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
889 mask); 935 NULL, mask);
890} 936}
891 937
892static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, 938static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
893 u64 end, gfp_t mask) 939 u64 end, gfp_t mask)
894{ 940{
895 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask); 941 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
896} 942 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} 943}
910 944
911int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end) 945int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
@@ -917,13 +951,15 @@ 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 951 * either insert or lock state struct between start and end use mask to tell
918 * us if waiting is desired. 952 * us if waiting is desired.
919 */ 953 */
920int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) 954int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
955 int bits, struct extent_state **cached_state, gfp_t mask)
921{ 956{
922 int err; 957 int err;
923 u64 failed_start; 958 u64 failed_start;
924 while (1) { 959 while (1) {
925 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 960 err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
926 &failed_start, mask); 961 EXTENT_LOCKED, &failed_start,
962 cached_state, mask);
927 if (err == -EEXIST && (mask & __GFP_WAIT)) { 963 if (err == -EEXIST && (mask & __GFP_WAIT)) {
928 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); 964 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
929 start = failed_start; 965 start = failed_start;
@@ -935,27 +971,40 @@ int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
935 return err; 971 return err;
936} 972}
937 973
974int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
975{
976 return lock_extent_bits(tree, start, end, 0, NULL, mask);
977}
978
938int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end, 979int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
939 gfp_t mask) 980 gfp_t mask)
940{ 981{
941 int err; 982 int err;
942 u64 failed_start; 983 u64 failed_start;
943 984
944 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 985 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
945 &failed_start, mask); 986 &failed_start, NULL, mask);
946 if (err == -EEXIST) { 987 if (err == -EEXIST) {
947 if (failed_start > start) 988 if (failed_start > start)
948 clear_extent_bit(tree, start, failed_start - 1, 989 clear_extent_bit(tree, start, failed_start - 1,
949 EXTENT_LOCKED, 1, 0, mask); 990 EXTENT_LOCKED, 1, 0, NULL, mask);
950 return 0; 991 return 0;
951 } 992 }
952 return 1; 993 return 1;
953} 994}
954 995
996int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
997 struct extent_state **cached, gfp_t mask)
998{
999 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1000 mask);
1001}
1002
955int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, 1003int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
956 gfp_t mask) 1004 gfp_t mask)
957{ 1005{
958 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask); 1006 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1007 mask);
959} 1008}
960 1009
961/* 1010/*
@@ -974,7 +1023,6 @@ int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
974 page_cache_release(page); 1023 page_cache_release(page);
975 index++; 1024 index++;
976 } 1025 }
977 set_extent_dirty(tree, start, end, GFP_NOFS);
978 return 0; 1026 return 0;
979} 1027}
980 1028
@@ -994,7 +1042,6 @@ static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
994 page_cache_release(page); 1042 page_cache_release(page);
995 index++; 1043 index++;
996 } 1044 }
997 set_extent_writeback(tree, start, end, GFP_NOFS);
998 return 0; 1045 return 0;
999} 1046}
1000 1047
@@ -1232,6 +1279,7 @@ static noinline u64 find_lock_delalloc_range(struct inode *inode,
1232 u64 delalloc_start; 1279 u64 delalloc_start;
1233 u64 delalloc_end; 1280 u64 delalloc_end;
1234 u64 found; 1281 u64 found;
1282 struct extent_state *cached_state = NULL;
1235 int ret; 1283 int ret;
1236 int loops = 0; 1284 int loops = 0;
1237 1285
@@ -1269,6 +1317,7 @@ again:
1269 /* some of the pages are gone, lets avoid looping by 1317 /* some of the pages are gone, lets avoid looping by
1270 * shortening the size of the delalloc range we're searching 1318 * shortening the size of the delalloc range we're searching
1271 */ 1319 */
1320 free_extent_state(cached_state);
1272 if (!loops) { 1321 if (!loops) {
1273 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1); 1322 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1274 max_bytes = PAGE_CACHE_SIZE - offset; 1323 max_bytes = PAGE_CACHE_SIZE - offset;
@@ -1282,18 +1331,21 @@ again:
1282 BUG_ON(ret); 1331 BUG_ON(ret);
1283 1332
1284 /* step three, lock the state bits for the whole range */ 1333 /* step three, lock the state bits for the whole range */
1285 lock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS); 1334 lock_extent_bits(tree, delalloc_start, delalloc_end,
1335 0, &cached_state, GFP_NOFS);
1286 1336
1287 /* then test to make sure it is all still delalloc */ 1337 /* then test to make sure it is all still delalloc */
1288 ret = test_range_bit(tree, delalloc_start, delalloc_end, 1338 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1289 EXTENT_DELALLOC, 1); 1339 EXTENT_DELALLOC, 1, cached_state);
1290 if (!ret) { 1340 if (!ret) {
1291 unlock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS); 1341 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1342 &cached_state, GFP_NOFS);
1292 __unlock_for_delalloc(inode, locked_page, 1343 __unlock_for_delalloc(inode, locked_page,
1293 delalloc_start, delalloc_end); 1344 delalloc_start, delalloc_end);
1294 cond_resched(); 1345 cond_resched();
1295 goto again; 1346 goto again;
1296 } 1347 }
1348 free_extent_state(cached_state);
1297 *start = delalloc_start; 1349 *start = delalloc_start;
1298 *end = delalloc_end; 1350 *end = delalloc_end;
1299out_failed: 1351out_failed:
@@ -1307,7 +1359,8 @@ int extent_clear_unlock_delalloc(struct inode *inode,
1307 int clear_unlock, 1359 int clear_unlock,
1308 int clear_delalloc, int clear_dirty, 1360 int clear_delalloc, int clear_dirty,
1309 int set_writeback, 1361 int set_writeback,
1310 int end_writeback) 1362 int end_writeback,
1363 int set_private2)
1311{ 1364{
1312 int ret; 1365 int ret;
1313 struct page *pages[16]; 1366 struct page *pages[16];
@@ -1325,8 +1378,9 @@ int extent_clear_unlock_delalloc(struct inode *inode,
1325 if (clear_delalloc) 1378 if (clear_delalloc)
1326 clear_bits |= EXTENT_DELALLOC; 1379 clear_bits |= EXTENT_DELALLOC;
1327 1380
1328 clear_extent_bit(tree, start, end, clear_bits, 1, 0, GFP_NOFS); 1381 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1329 if (!(unlock_pages || clear_dirty || set_writeback || end_writeback)) 1382 if (!(unlock_pages || clear_dirty || set_writeback || end_writeback ||
1383 set_private2))
1330 return 0; 1384 return 0;
1331 1385
1332 while (nr_pages > 0) { 1386 while (nr_pages > 0) {
@@ -1334,6 +1388,10 @@ int extent_clear_unlock_delalloc(struct inode *inode,
1334 min_t(unsigned long, 1388 min_t(unsigned long,
1335 nr_pages, ARRAY_SIZE(pages)), pages); 1389 nr_pages, ARRAY_SIZE(pages)), pages);
1336 for (i = 0; i < ret; i++) { 1390 for (i = 0; i < ret; i++) {
1391
1392 if (set_private2)
1393 SetPagePrivate2(pages[i]);
1394
1337 if (pages[i] == locked_page) { 1395 if (pages[i] == locked_page) {
1338 page_cache_release(pages[i]); 1396 page_cache_release(pages[i]);
1339 continue; 1397 continue;
@@ -1476,14 +1534,17 @@ out:
1476 * range is found set. 1534 * range is found set.
1477 */ 1535 */
1478int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, 1536int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1479 int bits, int filled) 1537 int bits, int filled, struct extent_state *cached)
1480{ 1538{
1481 struct extent_state *state = NULL; 1539 struct extent_state *state = NULL;
1482 struct rb_node *node; 1540 struct rb_node *node;
1483 int bitset = 0; 1541 int bitset = 0;
1484 1542
1485 spin_lock(&tree->lock); 1543 spin_lock(&tree->lock);
1486 node = tree_search(tree, start); 1544 if (cached && cached->tree && cached->start == start)
1545 node = &cached->rb_node;
1546 else
1547 node = tree_search(tree, start);
1487 while (node && start <= end) { 1548 while (node && start <= end) {
1488 state = rb_entry(node, struct extent_state, rb_node); 1549 state = rb_entry(node, struct extent_state, rb_node);
1489 1550
@@ -1503,6 +1564,10 @@ int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1503 bitset = 0; 1564 bitset = 0;
1504 break; 1565 break;
1505 } 1566 }
1567
1568 if (state->end == (u64)-1)
1569 break;
1570
1506 start = state->end + 1; 1571 start = state->end + 1;
1507 if (start > end) 1572 if (start > end)
1508 break; 1573 break;
@@ -1526,7 +1591,7 @@ static int check_page_uptodate(struct extent_io_tree *tree,
1526{ 1591{
1527 u64 start = (u64)page->index << PAGE_CACHE_SHIFT; 1592 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1528 u64 end = start + PAGE_CACHE_SIZE - 1; 1593 u64 end = start + PAGE_CACHE_SIZE - 1;
1529 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1)) 1594 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1530 SetPageUptodate(page); 1595 SetPageUptodate(page);
1531 return 0; 1596 return 0;
1532} 1597}
@@ -1540,7 +1605,7 @@ static int check_page_locked(struct extent_io_tree *tree,
1540{ 1605{
1541 u64 start = (u64)page->index << PAGE_CACHE_SHIFT; 1606 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1542 u64 end = start + PAGE_CACHE_SIZE - 1; 1607 u64 end = start + PAGE_CACHE_SIZE - 1;
1543 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0)) 1608 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1544 unlock_page(page); 1609 unlock_page(page);
1545 return 0; 1610 return 0;
1546} 1611}
@@ -1552,10 +1617,7 @@ static int check_page_locked(struct extent_io_tree *tree,
1552static int check_page_writeback(struct extent_io_tree *tree, 1617static int check_page_writeback(struct extent_io_tree *tree,
1553 struct page *page) 1618 struct page *page)
1554{ 1619{
1555 u64 start = (u64)page->index << PAGE_CACHE_SHIFT; 1620 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; 1621 return 0;
1560} 1622}
1561 1623
@@ -1613,13 +1675,11 @@ static void end_bio_extent_writepage(struct bio *bio, int err)
1613 } 1675 }
1614 1676
1615 if (!uptodate) { 1677 if (!uptodate) {
1616 clear_extent_uptodate(tree, start, end, GFP_ATOMIC); 1678 clear_extent_uptodate(tree, start, end, GFP_NOFS);
1617 ClearPageUptodate(page); 1679 ClearPageUptodate(page);
1618 SetPageError(page); 1680 SetPageError(page);
1619 } 1681 }
1620 1682
1621 clear_extent_writeback(tree, start, end, GFP_ATOMIC);
1622
1623 if (whole_page) 1683 if (whole_page)
1624 end_page_writeback(page); 1684 end_page_writeback(page);
1625 else 1685 else
@@ -1983,7 +2043,8 @@ static int __extent_read_full_page(struct extent_io_tree *tree,
1983 continue; 2043 continue;
1984 } 2044 }
1985 /* the get_extent function already copied into the page */ 2045 /* the get_extent function already copied into the page */
1986 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) { 2046 if (test_range_bit(tree, cur, cur_end,
2047 EXTENT_UPTODATE, 1, NULL)) {
1987 check_page_uptodate(tree, page); 2048 check_page_uptodate(tree, page);
1988 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); 2049 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1989 cur = cur + iosize; 2050 cur = cur + iosize;
@@ -2078,6 +2139,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2078 u64 iosize; 2139 u64 iosize;
2079 u64 unlock_start; 2140 u64 unlock_start;
2080 sector_t sector; 2141 sector_t sector;
2142 struct extent_state *cached_state = NULL;
2081 struct extent_map *em; 2143 struct extent_map *em;
2082 struct block_device *bdev; 2144 struct block_device *bdev;
2083 int ret; 2145 int ret;
@@ -2124,6 +2186,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2124 delalloc_end = 0; 2186 delalloc_end = 0;
2125 page_started = 0; 2187 page_started = 0;
2126 if (!epd->extent_locked) { 2188 if (!epd->extent_locked) {
2189 u64 delalloc_to_write = 0;
2127 /* 2190 /*
2128 * make sure the wbc mapping index is at least updated 2191 * make sure the wbc mapping index is at least updated
2129 * to this page. 2192 * to this page.
@@ -2143,8 +2206,24 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2143 tree->ops->fill_delalloc(inode, page, delalloc_start, 2206 tree->ops->fill_delalloc(inode, page, delalloc_start,
2144 delalloc_end, &page_started, 2207 delalloc_end, &page_started,
2145 &nr_written); 2208 &nr_written);
2209 /*
2210 * delalloc_end is already one less than the total
2211 * length, so we don't subtract one from
2212 * PAGE_CACHE_SIZE
2213 */
2214 delalloc_to_write += (delalloc_end - delalloc_start +
2215 PAGE_CACHE_SIZE) >>
2216 PAGE_CACHE_SHIFT;
2146 delalloc_start = delalloc_end + 1; 2217 delalloc_start = delalloc_end + 1;
2147 } 2218 }
2219 if (wbc->nr_to_write < delalloc_to_write) {
2220 int thresh = 8192;
2221
2222 if (delalloc_to_write < thresh * 2)
2223 thresh = delalloc_to_write;
2224 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2225 thresh);
2226 }
2148 2227
2149 /* did the fill delalloc function already unlock and start 2228 /* did the fill delalloc function already unlock and start
2150 * the IO? 2229 * the IO?
@@ -2160,15 +2239,10 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2160 goto done_unlocked; 2239 goto done_unlocked;
2161 } 2240 }
2162 } 2241 }
2163 lock_extent(tree, start, page_end, GFP_NOFS);
2164
2165 unlock_start = start;
2166
2167 if (tree->ops && tree->ops->writepage_start_hook) { 2242 if (tree->ops && tree->ops->writepage_start_hook) {
2168 ret = tree->ops->writepage_start_hook(page, start, 2243 ret = tree->ops->writepage_start_hook(page, start,
2169 page_end); 2244 page_end);
2170 if (ret == -EAGAIN) { 2245 if (ret == -EAGAIN) {
2171 unlock_extent(tree, start, page_end, GFP_NOFS);
2172 redirty_page_for_writepage(wbc, page); 2246 redirty_page_for_writepage(wbc, page);
2173 update_nr_written(page, wbc, nr_written); 2247 update_nr_written(page, wbc, nr_written);
2174 unlock_page(page); 2248 unlock_page(page);
@@ -2184,12 +2258,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2184 update_nr_written(page, wbc, nr_written + 1); 2258 update_nr_written(page, wbc, nr_written + 1);
2185 2259
2186 end = page_end; 2260 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) { 2261 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) 2262 if (tree->ops && tree->ops->writepage_end_io_hook)
2194 tree->ops->writepage_end_io_hook(page, start, 2263 tree->ops->writepage_end_io_hook(page, start,
2195 page_end, NULL, 1); 2264 page_end, NULL, 1);
@@ -2197,13 +2266,10 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2197 goto done; 2266 goto done;
2198 } 2267 }
2199 2268
2200 set_extent_uptodate(tree, start, page_end, GFP_NOFS);
2201 blocksize = inode->i_sb->s_blocksize; 2269 blocksize = inode->i_sb->s_blocksize;
2202 2270
2203 while (cur <= end) { 2271 while (cur <= end) {
2204 if (cur >= last_byte) { 2272 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) 2273 if (tree->ops && tree->ops->writepage_end_io_hook)
2208 tree->ops->writepage_end_io_hook(page, cur, 2274 tree->ops->writepage_end_io_hook(page, cur,
2209 page_end, NULL, 1); 2275 page_end, NULL, 1);
@@ -2235,12 +2301,6 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2235 */ 2301 */
2236 if (compressed || block_start == EXTENT_MAP_HOLE || 2302 if (compressed || block_start == EXTENT_MAP_HOLE ||
2237 block_start == EXTENT_MAP_INLINE) { 2303 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 /* 2304 /*
2245 * end_io notification does not happen here for 2305 * end_io notification does not happen here for
2246 * compressed extents 2306 * compressed extents
@@ -2265,13 +2325,12 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2265 } 2325 }
2266 /* leave this out until we have a page_mkwrite call */ 2326 /* leave this out until we have a page_mkwrite call */
2267 if (0 && !test_range_bit(tree, cur, cur + iosize - 1, 2327 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2268 EXTENT_DIRTY, 0)) { 2328 EXTENT_DIRTY, 0, NULL)) {
2269 cur = cur + iosize; 2329 cur = cur + iosize;
2270 pg_offset += iosize; 2330 pg_offset += iosize;
2271 continue; 2331 continue;
2272 } 2332 }
2273 2333
2274 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
2275 if (tree->ops && tree->ops->writepage_io_hook) { 2334 if (tree->ops && tree->ops->writepage_io_hook) {
2276 ret = tree->ops->writepage_io_hook(page, cur, 2335 ret = tree->ops->writepage_io_hook(page, cur,
2277 cur + iosize - 1); 2336 cur + iosize - 1);
@@ -2309,12 +2368,12 @@ done:
2309 set_page_writeback(page); 2368 set_page_writeback(page);
2310 end_page_writeback(page); 2369 end_page_writeback(page);
2311 } 2370 }
2312 if (unlock_start <= page_end)
2313 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2314 unlock_page(page); 2371 unlock_page(page);
2315 2372
2316done_unlocked: 2373done_unlocked:
2317 2374
2375 /* drop our reference on any cached states */
2376 free_extent_state(cached_state);
2318 return 0; 2377 return 0;
2319} 2378}
2320 2379
@@ -2339,9 +2398,9 @@ static int extent_write_cache_pages(struct extent_io_tree *tree,
2339 writepage_t writepage, void *data, 2398 writepage_t writepage, void *data,
2340 void (*flush_fn)(void *)) 2399 void (*flush_fn)(void *))
2341{ 2400{
2342 struct backing_dev_info *bdi = mapping->backing_dev_info;
2343 int ret = 0; 2401 int ret = 0;
2344 int done = 0; 2402 int done = 0;
2403 int nr_to_write_done = 0;
2345 struct pagevec pvec; 2404 struct pagevec pvec;
2346 int nr_pages; 2405 int nr_pages;
2347 pgoff_t index; 2406 pgoff_t index;
@@ -2361,7 +2420,7 @@ static int extent_write_cache_pages(struct extent_io_tree *tree,
2361 scanned = 1; 2420 scanned = 1;
2362 } 2421 }
2363retry: 2422retry:
2364 while (!done && (index <= end) && 2423 while (!done && !nr_to_write_done && (index <= end) &&
2365 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, 2424 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2366 PAGECACHE_TAG_DIRTY, min(end - index, 2425 PAGECACHE_TAG_DIRTY, min(end - index,
2367 (pgoff_t)PAGEVEC_SIZE-1) + 1))) { 2426 (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
@@ -2412,12 +2471,15 @@ retry:
2412 unlock_page(page); 2471 unlock_page(page);
2413 ret = 0; 2472 ret = 0;
2414 } 2473 }
2415 if (ret || wbc->nr_to_write <= 0) 2474 if (ret)
2416 done = 1;
2417 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2418 wbc->encountered_congestion = 1;
2419 done = 1; 2475 done = 1;
2420 } 2476
2477 /*
2478 * the filesystem may choose to bump up nr_to_write.
2479 * We have to make sure to honor the new nr_to_write
2480 * at any time
2481 */
2482 nr_to_write_done = wbc->nr_to_write <= 0;
2421 } 2483 }
2422 pagevec_release(&pvec); 2484 pagevec_release(&pvec);
2423 cond_resched(); 2485 cond_resched();
@@ -2604,10 +2666,10 @@ int extent_invalidatepage(struct extent_io_tree *tree,
2604 return 0; 2666 return 0;
2605 2667
2606 lock_extent(tree, start, end, GFP_NOFS); 2668 lock_extent(tree, start, end, GFP_NOFS);
2607 wait_on_extent_writeback(tree, start, end); 2669 wait_on_page_writeback(page);
2608 clear_extent_bit(tree, start, end, 2670 clear_extent_bit(tree, start, end,
2609 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC, 2671 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2610 1, 1, GFP_NOFS); 2672 1, 1, NULL, GFP_NOFS);
2611 return 0; 2673 return 0;
2612} 2674}
2613 2675
@@ -2687,7 +2749,7 @@ int extent_prepare_write(struct extent_io_tree *tree,
2687 !isnew && !PageUptodate(page) && 2749 !isnew && !PageUptodate(page) &&
2688 (block_off_end > to || block_off_start < from) && 2750 (block_off_end > to || block_off_start < from) &&
2689 !test_range_bit(tree, block_start, cur_end, 2751 !test_range_bit(tree, block_start, cur_end,
2690 EXTENT_UPTODATE, 1)) { 2752 EXTENT_UPTODATE, 1, NULL)) {
2691 u64 sector; 2753 u64 sector;
2692 u64 extent_offset = block_start - em->start; 2754 u64 extent_offset = block_start - em->start;
2693 size_t iosize; 2755 size_t iosize;
@@ -2701,7 +2763,7 @@ int extent_prepare_write(struct extent_io_tree *tree,
2701 */ 2763 */
2702 set_extent_bit(tree, block_start, 2764 set_extent_bit(tree, block_start,
2703 block_start + iosize - 1, 2765 block_start + iosize - 1,
2704 EXTENT_LOCKED, 0, NULL, GFP_NOFS); 2766 EXTENT_LOCKED, 0, NULL, NULL, GFP_NOFS);
2705 ret = submit_extent_page(READ, tree, page, 2767 ret = submit_extent_page(READ, tree, page,
2706 sector, iosize, page_offset, em->bdev, 2768 sector, iosize, page_offset, em->bdev,
2707 NULL, 1, 2769 NULL, 1,
@@ -2742,13 +2804,18 @@ int try_release_extent_state(struct extent_map_tree *map,
2742 int ret = 1; 2804 int ret = 1;
2743 2805
2744 if (test_range_bit(tree, start, end, 2806 if (test_range_bit(tree, start, end,
2745 EXTENT_IOBITS | EXTENT_ORDERED, 0)) 2807 EXTENT_IOBITS, 0, NULL))
2746 ret = 0; 2808 ret = 0;
2747 else { 2809 else {
2748 if ((mask & GFP_NOFS) == GFP_NOFS) 2810 if ((mask & GFP_NOFS) == GFP_NOFS)
2749 mask = GFP_NOFS; 2811 mask = GFP_NOFS;
2750 clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 2812 /*
2751 1, 1, mask); 2813 * at this point we can safely clear everything except the
2814 * locked bit and the nodatasum bit
2815 */
2816 clear_extent_bit(tree, start, end,
2817 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
2818 0, 0, NULL, mask);
2752 } 2819 }
2753 return ret; 2820 return ret;
2754} 2821}
@@ -2771,29 +2838,28 @@ int try_release_extent_mapping(struct extent_map_tree *map,
2771 u64 len; 2838 u64 len;
2772 while (start <= end) { 2839 while (start <= end) {
2773 len = end - start + 1; 2840 len = end - start + 1;
2774 spin_lock(&map->lock); 2841 write_lock(&map->lock);
2775 em = lookup_extent_mapping(map, start, len); 2842 em = lookup_extent_mapping(map, start, len);
2776 if (!em || IS_ERR(em)) { 2843 if (!em || IS_ERR(em)) {
2777 spin_unlock(&map->lock); 2844 write_unlock(&map->lock);
2778 break; 2845 break;
2779 } 2846 }
2780 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || 2847 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2781 em->start != start) { 2848 em->start != start) {
2782 spin_unlock(&map->lock); 2849 write_unlock(&map->lock);
2783 free_extent_map(em); 2850 free_extent_map(em);
2784 break; 2851 break;
2785 } 2852 }
2786 if (!test_range_bit(tree, em->start, 2853 if (!test_range_bit(tree, em->start,
2787 extent_map_end(em) - 1, 2854 extent_map_end(em) - 1,
2788 EXTENT_LOCKED | EXTENT_WRITEBACK | 2855 EXTENT_LOCKED | EXTENT_WRITEBACK,
2789 EXTENT_ORDERED, 2856 0, NULL)) {
2790 0)) {
2791 remove_extent_mapping(map, em); 2857 remove_extent_mapping(map, em);
2792 /* once for the rb tree */ 2858 /* once for the rb tree */
2793 free_extent_map(em); 2859 free_extent_map(em);
2794 } 2860 }
2795 start = extent_map_end(em); 2861 start = extent_map_end(em);
2796 spin_unlock(&map->lock); 2862 write_unlock(&map->lock);
2797 2863
2798 /* once for us */ 2864 /* once for us */
2799 free_extent_map(em); 2865 free_extent_map(em);
@@ -3203,7 +3269,7 @@ int extent_range_uptodate(struct extent_io_tree *tree,
3203 int uptodate; 3269 int uptodate;
3204 unsigned long index; 3270 unsigned long index;
3205 3271
3206 ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1); 3272 ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL);
3207 if (ret) 3273 if (ret)
3208 return 1; 3274 return 1;
3209 while (start <= end) { 3275 while (start <= end) {
@@ -3233,7 +3299,7 @@ int extent_buffer_uptodate(struct extent_io_tree *tree,
3233 return 1; 3299 return 1;
3234 3300
3235 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1, 3301 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3236 EXTENT_UPTODATE, 1); 3302 EXTENT_UPTODATE, 1, NULL);
3237 if (ret) 3303 if (ret)
3238 return ret; 3304 return ret;
3239 3305
@@ -3269,7 +3335,7 @@ int read_extent_buffer_pages(struct extent_io_tree *tree,
3269 return 0; 3335 return 0;
3270 3336
3271 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1, 3337 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3272 EXTENT_UPTODATE, 1)) { 3338 EXTENT_UPTODATE, 1, NULL)) {
3273 return 0; 3339 return 0;
3274 } 3340 }
3275 3341
diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h
index 5bc20abf3f3d..14ed16fd862d 100644
--- a/fs/btrfs/extent_io.h
+++ b/fs/btrfs/extent_io.h
@@ -13,10 +13,8 @@
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)
19#define EXTENT_NODATASUM (1 << 12)
20#define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK) 18#define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK)
21 19
22/* flags for bio submission */ 20/* flags for bio submission */
@@ -142,6 +140,8 @@ int try_release_extent_state(struct extent_map_tree *map,
142 struct extent_io_tree *tree, struct page *page, 140 struct extent_io_tree *tree, struct page *page,
143 gfp_t mask); 141 gfp_t mask);
144int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask); 142int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask);
143int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
144 int bits, struct extent_state **cached, gfp_t mask);
145int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask); 145int 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, 146int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
147 gfp_t mask); 147 gfp_t mask);
@@ -155,11 +155,12 @@ u64 count_range_bits(struct extent_io_tree *tree,
155 u64 max_bytes, unsigned long bits); 155 u64 max_bytes, unsigned long bits);
156 156
157int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, 157int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
158 int bits, int filled); 158 int bits, int filled, struct extent_state *cached_state);
159int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, 159int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
160 int bits, gfp_t mask); 160 int bits, gfp_t mask);
161int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, 161int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
162 int bits, int wake, int delete, gfp_t mask); 162 int bits, int wake, int delete, struct extent_state **cached,
163 gfp_t mask);
163int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, 164int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
164 int bits, gfp_t mask); 165 int bits, gfp_t mask);
165int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, 166int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
@@ -282,5 +283,6 @@ int extent_clear_unlock_delalloc(struct inode *inode,
282 int clear_unlock, 283 int clear_unlock,
283 int clear_delalloc, int clear_dirty, 284 int clear_delalloc, int clear_dirty,
284 int set_writeback, 285 int set_writeback,
285 int end_writeback); 286 int end_writeback,
287 int set_private2);
286#endif 288#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..571ad3c13b47 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;
@@ -125,22 +123,6 @@ 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;
128
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); 126 btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
145 for (i = 0; i < num_pages; i++) { 127 for (i = 0; i < num_pages; i++) {
146 struct page *p = pages[i]; 128 struct page *p = pages[i];
@@ -155,9 +137,6 @@ static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
155 * at this time. 137 * at this time.
156 */ 138 */
157 } 139 }
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; 140 return err;
162} 141}
163 142
@@ -189,18 +168,18 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
189 if (!split2) 168 if (!split2)
190 split2 = alloc_extent_map(GFP_NOFS); 169 split2 = alloc_extent_map(GFP_NOFS);
191 170
192 spin_lock(&em_tree->lock); 171 write_lock(&em_tree->lock);
193 em = lookup_extent_mapping(em_tree, start, len); 172 em = lookup_extent_mapping(em_tree, start, len);
194 if (!em) { 173 if (!em) {
195 spin_unlock(&em_tree->lock); 174 write_unlock(&em_tree->lock);
196 break; 175 break;
197 } 176 }
198 flags = em->flags; 177 flags = em->flags;
199 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { 178 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
200 spin_unlock(&em_tree->lock);
201 if (em->start <= start && 179 if (em->start <= start &&
202 (!testend || em->start + em->len >= start + len)) { 180 (!testend || em->start + em->len >= start + len)) {
203 free_extent_map(em); 181 free_extent_map(em);
182 write_unlock(&em_tree->lock);
204 break; 183 break;
205 } 184 }
206 if (start < em->start) { 185 if (start < em->start) {
@@ -210,6 +189,7 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
210 start = em->start + em->len; 189 start = em->start + em->len;
211 } 190 }
212 free_extent_map(em); 191 free_extent_map(em);
192 write_unlock(&em_tree->lock);
213 continue; 193 continue;
214 } 194 }
215 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); 195 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
@@ -260,7 +240,7 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
260 free_extent_map(split); 240 free_extent_map(split);
261 split = NULL; 241 split = NULL;
262 } 242 }
263 spin_unlock(&em_tree->lock); 243 write_unlock(&em_tree->lock);
264 244
265 /* once for us */ 245 /* once for us */
266 free_extent_map(em); 246 free_extent_map(em);
@@ -289,7 +269,7 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
289noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans, 269noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
290 struct btrfs_root *root, struct inode *inode, 270 struct btrfs_root *root, struct inode *inode,
291 u64 start, u64 end, u64 locked_end, 271 u64 start, u64 end, u64 locked_end,
292 u64 inline_limit, u64 *hint_byte) 272 u64 inline_limit, u64 *hint_byte, int drop_cache)
293{ 273{
294 u64 extent_end = 0; 274 u64 extent_end = 0;
295 u64 search_start = start; 275 u64 search_start = start;
@@ -314,7 +294,8 @@ noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
314 int ret; 294 int ret;
315 295
316 inline_limit = 0; 296 inline_limit = 0;
317 btrfs_drop_extent_cache(inode, start, end - 1, 0); 297 if (drop_cache)
298 btrfs_drop_extent_cache(inode, start, end - 1, 0);
318 299
319 path = btrfs_alloc_path(); 300 path = btrfs_alloc_path();
320 if (!path) 301 if (!path)
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 d154a3f365d5..e9b76bcd1c12 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -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
@@ -425,7 +426,7 @@ again:
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, 1, 0,
428 0, 1, 1, 1); 429 0, 1, 1, 1, 0);
429 ret = 0; 430 ret = 0;
430 goto free_pages_out; 431 goto free_pages_out;
431 } 432 }
@@ -611,9 +612,9 @@ static noinline int submit_compressed_extents(struct inode *inode,
611 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); 612 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
612 613
613 while (1) { 614 while (1) {
614 spin_lock(&em_tree->lock); 615 write_lock(&em_tree->lock);
615 ret = add_extent_mapping(em_tree, em); 616 ret = add_extent_mapping(em_tree, em);
616 spin_unlock(&em_tree->lock); 617 write_unlock(&em_tree->lock);
617 if (ret != -EEXIST) { 618 if (ret != -EEXIST) {
618 free_extent_map(em); 619 free_extent_map(em);
619 break; 620 break;
@@ -640,7 +641,7 @@ static noinline int submit_compressed_extents(struct inode *inode,
640 async_extent->start, 641 async_extent->start,
641 async_extent->start + 642 async_extent->start +
642 async_extent->ram_size - 1, 643 async_extent->ram_size - 1,
643 NULL, 1, 1, 0, 1, 1, 0); 644 NULL, 1, 1, 0, 1, 1, 0, 0);
644 645
645 ret = btrfs_submit_compressed_write(inode, 646 ret = btrfs_submit_compressed_write(inode,
646 async_extent->start, 647 async_extent->start,
@@ -713,7 +714,7 @@ static noinline int cow_file_range(struct inode *inode,
713 extent_clear_unlock_delalloc(inode, 714 extent_clear_unlock_delalloc(inode,
714 &BTRFS_I(inode)->io_tree, 715 &BTRFS_I(inode)->io_tree,
715 start, end, NULL, 1, 1, 716 start, end, NULL, 1, 1,
716 1, 1, 1, 1); 717 1, 1, 1, 1, 0);
717 *nr_written = *nr_written + 718 *nr_written = *nr_written +
718 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; 719 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
719 *page_started = 1; 720 *page_started = 1;
@@ -725,6 +726,15 @@ static noinline int cow_file_range(struct inode *inode,
725 BUG_ON(disk_num_bytes > 726 BUG_ON(disk_num_bytes >
726 btrfs_super_total_bytes(&root->fs_info->super_copy)); 727 btrfs_super_total_bytes(&root->fs_info->super_copy));
727 728
729
730 read_lock(&BTRFS_I(inode)->extent_tree.lock);
731 em = search_extent_mapping(&BTRFS_I(inode)->extent_tree,
732 start, num_bytes);
733 if (em) {
734 alloc_hint = em->block_start;
735 free_extent_map(em);
736 }
737 read_unlock(&BTRFS_I(inode)->extent_tree.lock);
728 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); 738 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
729 739
730 while (disk_num_bytes > 0) { 740 while (disk_num_bytes > 0) {
@@ -737,7 +747,6 @@ static noinline int cow_file_range(struct inode *inode,
737 em = alloc_extent_map(GFP_NOFS); 747 em = alloc_extent_map(GFP_NOFS);
738 em->start = start; 748 em->start = start;
739 em->orig_start = em->start; 749 em->orig_start = em->start;
740
741 ram_size = ins.offset; 750 ram_size = ins.offset;
742 em->len = ins.offset; 751 em->len = ins.offset;
743 752
@@ -747,9 +756,9 @@ static noinline int cow_file_range(struct inode *inode,
747 set_bit(EXTENT_FLAG_PINNED, &em->flags); 756 set_bit(EXTENT_FLAG_PINNED, &em->flags);
748 757
749 while (1) { 758 while (1) {
750 spin_lock(&em_tree->lock); 759 write_lock(&em_tree->lock);
751 ret = add_extent_mapping(em_tree, em); 760 ret = add_extent_mapping(em_tree, em);
752 spin_unlock(&em_tree->lock); 761 write_unlock(&em_tree->lock);
753 if (ret != -EEXIST) { 762 if (ret != -EEXIST) {
754 free_extent_map(em); 763 free_extent_map(em);
755 break; 764 break;
@@ -776,11 +785,14 @@ static noinline int cow_file_range(struct inode *inode,
776 /* we're not doing compressed IO, don't unlock the first 785 /* we're not doing compressed IO, don't unlock the first
777 * page (which the caller expects to stay locked), don't 786 * page (which the caller expects to stay locked), don't
778 * clear any dirty bits and don't set any writeback bits 787 * clear any dirty bits and don't set any writeback bits
788 *
789 * Do set the Private2 bit so we know this page was properly
790 * setup for writepage
779 */ 791 */
780 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, 792 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
781 start, start + ram_size - 1, 793 start, start + ram_size - 1,
782 locked_page, unlock, 1, 794 locked_page, unlock, 1,
783 1, 0, 0, 0); 795 1, 0, 0, 0, 1);
784 disk_num_bytes -= cur_alloc_size; 796 disk_num_bytes -= cur_alloc_size;
785 num_bytes -= cur_alloc_size; 797 num_bytes -= cur_alloc_size;
786 alloc_hint = ins.objectid + ins.offset; 798 alloc_hint = ins.objectid + ins.offset;
@@ -853,7 +865,7 @@ static int cow_file_range_async(struct inode *inode, struct page *locked_page,
853 int limit = 10 * 1024 * 1042; 865 int limit = 10 * 1024 * 1042;
854 866
855 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED | 867 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
856 EXTENT_DELALLOC, 1, 0, GFP_NOFS); 868 EXTENT_DELALLOC, 1, 0, NULL, GFP_NOFS);
857 while (start < end) { 869 while (start < end) {
858 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); 870 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
859 async_cow->inode = inode; 871 async_cow->inode = inode;
@@ -1080,9 +1092,9 @@ out_check:
1080 em->bdev = root->fs_info->fs_devices->latest_bdev; 1092 em->bdev = root->fs_info->fs_devices->latest_bdev;
1081 set_bit(EXTENT_FLAG_PINNED, &em->flags); 1093 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1082 while (1) { 1094 while (1) {
1083 spin_lock(&em_tree->lock); 1095 write_lock(&em_tree->lock);
1084 ret = add_extent_mapping(em_tree, em); 1096 ret = add_extent_mapping(em_tree, em);
1085 spin_unlock(&em_tree->lock); 1097 write_unlock(&em_tree->lock);
1086 if (ret != -EEXIST) { 1098 if (ret != -EEXIST) {
1087 free_extent_map(em); 1099 free_extent_map(em);
1088 break; 1100 break;
@@ -1101,7 +1113,7 @@ out_check:
1101 1113
1102 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, 1114 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1103 cur_offset, cur_offset + num_bytes - 1, 1115 cur_offset, cur_offset + num_bytes - 1,
1104 locked_page, 1, 1, 1, 0, 0, 0); 1116 locked_page, 1, 1, 1, 0, 0, 0, 1);
1105 cur_offset = extent_end; 1117 cur_offset = extent_end;
1106 if (cur_offset > end) 1118 if (cur_offset > end)
1107 break; 1119 break;
@@ -1374,10 +1386,8 @@ again:
1374 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); 1386 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1375 1387
1376 /* already ordered? We're done */ 1388 /* already ordered? We're done */
1377 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, 1389 if (PagePrivate2(page))
1378 EXTENT_ORDERED, 0)) {
1379 goto out; 1390 goto out;
1380 }
1381 1391
1382 ordered = btrfs_lookup_ordered_extent(inode, page_start); 1392 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1383 if (ordered) { 1393 if (ordered) {
@@ -1413,11 +1423,9 @@ static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1413 struct inode *inode = page->mapping->host; 1423 struct inode *inode = page->mapping->host;
1414 struct btrfs_writepage_fixup *fixup; 1424 struct btrfs_writepage_fixup *fixup;
1415 struct btrfs_root *root = BTRFS_I(inode)->root; 1425 struct btrfs_root *root = BTRFS_I(inode)->root;
1416 int ret;
1417 1426
1418 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end, 1427 /* this page is properly in the ordered list */
1419 EXTENT_ORDERED, 0); 1428 if (TestClearPagePrivate2(page))
1420 if (ret)
1421 return 0; 1429 return 0;
1422 1430
1423 if (PageChecked(page)) 1431 if (PageChecked(page))
@@ -1455,9 +1463,19 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1455 BUG_ON(!path); 1463 BUG_ON(!path);
1456 1464
1457 path->leave_spinning = 1; 1465 path->leave_spinning = 1;
1466
1467 /*
1468 * we may be replacing one extent in the tree with another.
1469 * The new extent is pinned in the extent map, and we don't want
1470 * to drop it from the cache until it is completely in the btree.
1471 *
1472 * So, tell btrfs_drop_extents to leave this extent in the cache.
1473 * the caller is expected to unpin it and allow it to be merged
1474 * with the others.
1475 */
1458 ret = btrfs_drop_extents(trans, root, inode, file_pos, 1476 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1459 file_pos + num_bytes, locked_end, 1477 file_pos + num_bytes, locked_end,
1460 file_pos, &hint); 1478 file_pos, &hint, 0);
1461 BUG_ON(ret); 1479 BUG_ON(ret);
1462 1480
1463 ins.objectid = inode->i_ino; 1481 ins.objectid = inode->i_ino;
@@ -1485,7 +1503,6 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1485 btrfs_mark_buffer_dirty(leaf); 1503 btrfs_mark_buffer_dirty(leaf);
1486 1504
1487 inode_add_bytes(inode, num_bytes); 1505 inode_add_bytes(inode, num_bytes);
1488 btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1489 1506
1490 ins.objectid = disk_bytenr; 1507 ins.objectid = disk_bytenr;
1491 ins.offset = disk_num_bytes; 1508 ins.offset = disk_num_bytes;
@@ -1596,6 +1613,9 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1596 ordered_extent->len, 1613 ordered_extent->len,
1597 compressed, 0, 0, 1614 compressed, 0, 0,
1598 BTRFS_FILE_EXTENT_REG); 1615 BTRFS_FILE_EXTENT_REG);
1616 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1617 ordered_extent->file_offset,
1618 ordered_extent->len);
1599 BUG_ON(ret); 1619 BUG_ON(ret);
1600 } 1620 }
1601 unlock_extent(io_tree, ordered_extent->file_offset, 1621 unlock_extent(io_tree, ordered_extent->file_offset,
@@ -1623,6 +1643,7 @@ nocow:
1623static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, 1643static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1624 struct extent_state *state, int uptodate) 1644 struct extent_state *state, int uptodate)
1625{ 1645{
1646 ClearPagePrivate2(page);
1626 return btrfs_finish_ordered_io(page->mapping->host, start, end); 1647 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1627} 1648}
1628 1649
@@ -1669,13 +1690,13 @@ static int btrfs_io_failed_hook(struct bio *failed_bio,
1669 failrec->last_mirror = 0; 1690 failrec->last_mirror = 0;
1670 failrec->bio_flags = 0; 1691 failrec->bio_flags = 0;
1671 1692
1672 spin_lock(&em_tree->lock); 1693 read_lock(&em_tree->lock);
1673 em = lookup_extent_mapping(em_tree, start, failrec->len); 1694 em = lookup_extent_mapping(em_tree, start, failrec->len);
1674 if (em->start > start || em->start + em->len < start) { 1695 if (em->start > start || em->start + em->len < start) {
1675 free_extent_map(em); 1696 free_extent_map(em);
1676 em = NULL; 1697 em = NULL;
1677 } 1698 }
1678 spin_unlock(&em_tree->lock); 1699 read_unlock(&em_tree->lock);
1679 1700
1680 if (!em || IS_ERR(em)) { 1701 if (!em || IS_ERR(em)) {
1681 kfree(failrec); 1702 kfree(failrec);
@@ -1794,7 +1815,7 @@ static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1794 return 0; 1815 return 0;
1795 1816
1796 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && 1817 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1797 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) { 1818 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1798 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, 1819 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1799 GFP_NOFS); 1820 GFP_NOFS);
1800 return 0; 1821 return 0;
@@ -2352,6 +2373,69 @@ static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2352 return ret; 2373 return ret;
2353} 2374}
2354 2375
2376int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2377 struct btrfs_root *root,
2378 struct inode *dir, u64 objectid,
2379 const char *name, int name_len)
2380{
2381 struct btrfs_path *path;
2382 struct extent_buffer *leaf;
2383 struct btrfs_dir_item *di;
2384 struct btrfs_key key;
2385 u64 index;
2386 int ret;
2387
2388 path = btrfs_alloc_path();
2389 if (!path)
2390 return -ENOMEM;
2391
2392 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2393 name, name_len, -1);
2394 BUG_ON(!di || IS_ERR(di));
2395
2396 leaf = path->nodes[0];
2397 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2398 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2399 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2400 BUG_ON(ret);
2401 btrfs_release_path(root, path);
2402
2403 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2404 objectid, root->root_key.objectid,
2405 dir->i_ino, &index, name, name_len);
2406 if (ret < 0) {
2407 BUG_ON(ret != -ENOENT);
2408 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2409 name, name_len);
2410 BUG_ON(!di || IS_ERR(di));
2411
2412 leaf = path->nodes[0];
2413 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2414 btrfs_release_path(root, path);
2415 index = key.offset;
2416 }
2417
2418 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2419 index, name, name_len, -1);
2420 BUG_ON(!di || IS_ERR(di));
2421
2422 leaf = path->nodes[0];
2423 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2424 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2425 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2426 BUG_ON(ret);
2427 btrfs_release_path(root, path);
2428
2429 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2430 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2431 ret = btrfs_update_inode(trans, root, dir);
2432 BUG_ON(ret);
2433 dir->i_sb->s_dirt = 1;
2434
2435 btrfs_free_path(path);
2436 return 0;
2437}
2438
2355static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) 2439static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2356{ 2440{
2357 struct inode *inode = dentry->d_inode; 2441 struct inode *inode = dentry->d_inode;
@@ -2361,29 +2445,31 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2361 struct btrfs_trans_handle *trans; 2445 struct btrfs_trans_handle *trans;
2362 unsigned long nr = 0; 2446 unsigned long nr = 0;
2363 2447
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 || 2448 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2369 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) { 2449 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
2370 return -ENOTEMPTY; 2450 return -ENOTEMPTY;
2371 }
2372 2451
2373 trans = btrfs_start_transaction(root, 1); 2452 trans = btrfs_start_transaction(root, 1);
2374 btrfs_set_trans_block_group(trans, dir); 2453 btrfs_set_trans_block_group(trans, dir);
2375 2454
2455 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2456 err = btrfs_unlink_subvol(trans, root, dir,
2457 BTRFS_I(inode)->location.objectid,
2458 dentry->d_name.name,
2459 dentry->d_name.len);
2460 goto out;
2461 }
2462
2376 err = btrfs_orphan_add(trans, inode); 2463 err = btrfs_orphan_add(trans, inode);
2377 if (err) 2464 if (err)
2378 goto fail_trans; 2465 goto out;
2379 2466
2380 /* now the directory is empty */ 2467 /* now the directory is empty */
2381 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, 2468 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2382 dentry->d_name.name, dentry->d_name.len); 2469 dentry->d_name.name, dentry->d_name.len);
2383 if (!err) 2470 if (!err)
2384 btrfs_i_size_write(inode, 0); 2471 btrfs_i_size_write(inode, 0);
2385 2472out:
2386fail_trans:
2387 nr = trans->blocks_used; 2473 nr = trans->blocks_used;
2388 ret = btrfs_end_transaction_throttle(trans, root); 2474 ret = btrfs_end_transaction_throttle(trans, root);
2389 btrfs_btree_balance_dirty(root, nr); 2475 btrfs_btree_balance_dirty(root, nr);
@@ -2935,7 +3021,7 @@ int btrfs_cont_expand(struct inode *inode, loff_t size)
2935 cur_offset, 3021 cur_offset,
2936 cur_offset + hole_size, 3022 cur_offset + hole_size,
2937 block_end, 3023 block_end,
2938 cur_offset, &hint_byte); 3024 cur_offset, &hint_byte, 1);
2939 if (err) 3025 if (err)
2940 break; 3026 break;
2941 err = btrfs_insert_file_extent(trans, root, 3027 err = btrfs_insert_file_extent(trans, root,
@@ -3003,6 +3089,11 @@ void btrfs_delete_inode(struct inode *inode)
3003 } 3089 }
3004 btrfs_wait_ordered_range(inode, 0, (u64)-1); 3090 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3005 3091
3092 if (inode->i_nlink > 0) {
3093 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3094 goto no_delete;
3095 }
3096
3006 btrfs_i_size_write(inode, 0); 3097 btrfs_i_size_write(inode, 0);
3007 trans = btrfs_join_transaction(root, 1); 3098 trans = btrfs_join_transaction(root, 1);
3008 3099
@@ -3070,29 +3161,67 @@ out_err:
3070 * is kind of like crossing a mount point. 3161 * is kind of like crossing a mount point.
3071 */ 3162 */
3072static int fixup_tree_root_location(struct btrfs_root *root, 3163static int fixup_tree_root_location(struct btrfs_root *root,
3073 struct btrfs_key *location, 3164 struct inode *dir,
3074 struct btrfs_root **sub_root, 3165 struct dentry *dentry,
3075 struct dentry *dentry) 3166 struct btrfs_key *location,
3167 struct btrfs_root **sub_root)
3076{ 3168{
3077 struct btrfs_root_item *ri; 3169 struct btrfs_path *path;
3170 struct btrfs_root *new_root;
3171 struct btrfs_root_ref *ref;
3172 struct extent_buffer *leaf;
3173 int ret;
3174 int err = 0;
3078 3175
3079 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY) 3176 path = btrfs_alloc_path();
3080 return 0; 3177 if (!path) {
3081 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 3178 err = -ENOMEM;
3082 return 0; 3179 goto out;
3180 }
3083 3181
3084 *sub_root = btrfs_read_fs_root(root->fs_info, location, 3182 err = -ENOENT;
3085 dentry->d_name.name, 3183 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3086 dentry->d_name.len); 3184 BTRFS_I(dir)->root->root_key.objectid,
3087 if (IS_ERR(*sub_root)) 3185 location->objectid);
3088 return PTR_ERR(*sub_root); 3186 if (ret) {
3187 if (ret < 0)
3188 err = ret;
3189 goto out;
3190 }
3089 3191
3090 ri = &(*sub_root)->root_item; 3192 leaf = path->nodes[0];
3091 location->objectid = btrfs_root_dirid(ri); 3193 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3092 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); 3194 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3093 location->offset = 0; 3195 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3196 goto out;
3094 3197
3095 return 0; 3198 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3199 (unsigned long)(ref + 1),
3200 dentry->d_name.len);
3201 if (ret)
3202 goto out;
3203
3204 btrfs_release_path(root->fs_info->tree_root, path);
3205
3206 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3207 if (IS_ERR(new_root)) {
3208 err = PTR_ERR(new_root);
3209 goto out;
3210 }
3211
3212 if (btrfs_root_refs(&new_root->root_item) == 0) {
3213 err = -ENOENT;
3214 goto out;
3215 }
3216
3217 *sub_root = new_root;
3218 location->objectid = btrfs_root_dirid(&new_root->root_item);
3219 location->type = BTRFS_INODE_ITEM_KEY;
3220 location->offset = 0;
3221 err = 0;
3222out:
3223 btrfs_free_path(path);
3224 return err;
3096} 3225}
3097 3226
3098static void inode_tree_add(struct inode *inode) 3227static void inode_tree_add(struct inode *inode)
@@ -3101,11 +3230,13 @@ static void inode_tree_add(struct inode *inode)
3101 struct btrfs_inode *entry; 3230 struct btrfs_inode *entry;
3102 struct rb_node **p; 3231 struct rb_node **p;
3103 struct rb_node *parent; 3232 struct rb_node *parent;
3104
3105again: 3233again:
3106 p = &root->inode_tree.rb_node; 3234 p = &root->inode_tree.rb_node;
3107 parent = NULL; 3235 parent = NULL;
3108 3236
3237 if (hlist_unhashed(&inode->i_hash))
3238 return;
3239
3109 spin_lock(&root->inode_lock); 3240 spin_lock(&root->inode_lock);
3110 while (*p) { 3241 while (*p) {
3111 parent = *p; 3242 parent = *p;
@@ -3132,13 +3263,87 @@ again:
3132static void inode_tree_del(struct inode *inode) 3263static void inode_tree_del(struct inode *inode)
3133{ 3264{
3134 struct btrfs_root *root = BTRFS_I(inode)->root; 3265 struct btrfs_root *root = BTRFS_I(inode)->root;
3266 int empty = 0;
3135 3267
3136 spin_lock(&root->inode_lock); 3268 spin_lock(&root->inode_lock);
3137 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { 3269 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3138 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); 3270 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3139 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 3271 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3272 empty = RB_EMPTY_ROOT(&root->inode_tree);
3140 } 3273 }
3141 spin_unlock(&root->inode_lock); 3274 spin_unlock(&root->inode_lock);
3275
3276 if (empty && btrfs_root_refs(&root->root_item) == 0) {
3277 synchronize_srcu(&root->fs_info->subvol_srcu);
3278 spin_lock(&root->inode_lock);
3279 empty = RB_EMPTY_ROOT(&root->inode_tree);
3280 spin_unlock(&root->inode_lock);
3281 if (empty)
3282 btrfs_add_dead_root(root);
3283 }
3284}
3285
3286int btrfs_invalidate_inodes(struct btrfs_root *root)
3287{
3288 struct rb_node *node;
3289 struct rb_node *prev;
3290 struct btrfs_inode *entry;
3291 struct inode *inode;
3292 u64 objectid = 0;
3293
3294 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3295
3296 spin_lock(&root->inode_lock);
3297again:
3298 node = root->inode_tree.rb_node;
3299 prev = NULL;
3300 while (node) {
3301 prev = node;
3302 entry = rb_entry(node, struct btrfs_inode, rb_node);
3303
3304 if (objectid < entry->vfs_inode.i_ino)
3305 node = node->rb_left;
3306 else if (objectid > entry->vfs_inode.i_ino)
3307 node = node->rb_right;
3308 else
3309 break;
3310 }
3311 if (!node) {
3312 while (prev) {
3313 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3314 if (objectid <= entry->vfs_inode.i_ino) {
3315 node = prev;
3316 break;
3317 }
3318 prev = rb_next(prev);
3319 }
3320 }
3321 while (node) {
3322 entry = rb_entry(node, struct btrfs_inode, rb_node);
3323 objectid = entry->vfs_inode.i_ino + 1;
3324 inode = igrab(&entry->vfs_inode);
3325 if (inode) {
3326 spin_unlock(&root->inode_lock);
3327 if (atomic_read(&inode->i_count) > 1)
3328 d_prune_aliases(inode);
3329 /*
3330 * btrfs_drop_inode will remove it from
3331 * the inode cache when its usage count
3332 * hits zero.
3333 */
3334 iput(inode);
3335 cond_resched();
3336 spin_lock(&root->inode_lock);
3337 goto again;
3338 }
3339
3340 if (cond_resched_lock(&root->inode_lock))
3341 goto again;
3342
3343 node = rb_next(node);
3344 }
3345 spin_unlock(&root->inode_lock);
3346 return 0;
3142} 3347}
3143 3348
3144static noinline void init_btrfs_i(struct inode *inode) 3349static noinline void init_btrfs_i(struct inode *inode)
@@ -3225,15 +3430,41 @@ struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3225 return inode; 3430 return inode;
3226} 3431}
3227 3432
3433static struct inode *new_simple_dir(struct super_block *s,
3434 struct btrfs_key *key,
3435 struct btrfs_root *root)
3436{
3437 struct inode *inode = new_inode(s);
3438
3439 if (!inode)
3440 return ERR_PTR(-ENOMEM);
3441
3442 init_btrfs_i(inode);
3443
3444 BTRFS_I(inode)->root = root;
3445 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3446 BTRFS_I(inode)->dummy_inode = 1;
3447
3448 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3449 inode->i_op = &simple_dir_inode_operations;
3450 inode->i_fop = &simple_dir_operations;
3451 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3452 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3453
3454 return inode;
3455}
3456
3228struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) 3457struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3229{ 3458{
3230 struct inode *inode; 3459 struct inode *inode;
3231 struct btrfs_inode *bi = BTRFS_I(dir); 3460 struct btrfs_root *root = BTRFS_I(dir)->root;
3232 struct btrfs_root *root = bi->root;
3233 struct btrfs_root *sub_root = root; 3461 struct btrfs_root *sub_root = root;
3234 struct btrfs_key location; 3462 struct btrfs_key location;
3463 int index;
3235 int ret; 3464 int ret;
3236 3465
3466 dentry->d_op = &btrfs_dentry_operations;
3467
3237 if (dentry->d_name.len > BTRFS_NAME_LEN) 3468 if (dentry->d_name.len > BTRFS_NAME_LEN)
3238 return ERR_PTR(-ENAMETOOLONG); 3469 return ERR_PTR(-ENAMETOOLONG);
3239 3470
@@ -3242,29 +3473,50 @@ struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3242 if (ret < 0) 3473 if (ret < 0)
3243 return ERR_PTR(ret); 3474 return ERR_PTR(ret);
3244 3475
3245 inode = NULL; 3476 if (location.objectid == 0)
3246 if (location.objectid) { 3477 return NULL;
3247 ret = fixup_tree_root_location(root, &location, &sub_root, 3478
3248 dentry); 3479 if (location.type == BTRFS_INODE_ITEM_KEY) {
3249 if (ret < 0) 3480 inode = btrfs_iget(dir->i_sb, &location, root);
3250 return ERR_PTR(ret); 3481 return inode;
3251 if (ret > 0) 3482 }
3252 return ERR_PTR(-ENOENT); 3483
3484 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3485
3486 index = srcu_read_lock(&root->fs_info->subvol_srcu);
3487 ret = fixup_tree_root_location(root, dir, dentry,
3488 &location, &sub_root);
3489 if (ret < 0) {
3490 if (ret != -ENOENT)
3491 inode = ERR_PTR(ret);
3492 else
3493 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3494 } else {
3253 inode = btrfs_iget(dir->i_sb, &location, sub_root); 3495 inode = btrfs_iget(dir->i_sb, &location, sub_root);
3254 if (IS_ERR(inode))
3255 return ERR_CAST(inode);
3256 } 3496 }
3497 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3498
3257 return inode; 3499 return inode;
3258} 3500}
3259 3501
3502static int btrfs_dentry_delete(struct dentry *dentry)
3503{
3504 struct btrfs_root *root;
3505
3506 if (!dentry->d_inode)
3507 return 0;
3508
3509 root = BTRFS_I(dentry->d_inode)->root;
3510 if (btrfs_root_refs(&root->root_item) == 0)
3511 return 1;
3512 return 0;
3513}
3514
3260static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, 3515static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3261 struct nameidata *nd) 3516 struct nameidata *nd)
3262{ 3517{
3263 struct inode *inode; 3518 struct inode *inode;
3264 3519
3265 if (dentry->d_name.len > BTRFS_NAME_LEN)
3266 return ERR_PTR(-ENAMETOOLONG);
3267
3268 inode = btrfs_lookup_dentry(dir, dentry); 3520 inode = btrfs_lookup_dentry(dir, dentry);
3269 if (IS_ERR(inode)) 3521 if (IS_ERR(inode))
3270 return ERR_CAST(inode); 3522 return ERR_CAST(inode);
@@ -3603,9 +3855,6 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3603 if (ret != 0) 3855 if (ret != 0)
3604 goto fail; 3856 goto fail;
3605 3857
3606 if (objectid > root->highest_inode)
3607 root->highest_inode = objectid;
3608
3609 inode->i_uid = current_fsuid(); 3858 inode->i_uid = current_fsuid();
3610 3859
3611 if (dir && (dir->i_mode & S_ISGID)) { 3860 if (dir && (dir->i_mode & S_ISGID)) {
@@ -3673,26 +3922,35 @@ int btrfs_add_link(struct btrfs_trans_handle *trans,
3673 struct inode *parent_inode, struct inode *inode, 3922 struct inode *parent_inode, struct inode *inode,
3674 const char *name, int name_len, int add_backref, u64 index) 3923 const char *name, int name_len, int add_backref, u64 index)
3675{ 3924{
3676 int ret; 3925 int ret = 0;
3677 struct btrfs_key key; 3926 struct btrfs_key key;
3678 struct btrfs_root *root = BTRFS_I(parent_inode)->root; 3927 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3679 3928
3680 key.objectid = inode->i_ino; 3929 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
3681 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); 3930 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
3682 key.offset = 0; 3931 } else {
3932 key.objectid = inode->i_ino;
3933 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3934 key.offset = 0;
3935 }
3936
3937 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
3938 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
3939 key.objectid, root->root_key.objectid,
3940 parent_inode->i_ino,
3941 index, name, name_len);
3942 } else if (add_backref) {
3943 ret = btrfs_insert_inode_ref(trans, root,
3944 name, name_len, inode->i_ino,
3945 parent_inode->i_ino, index);
3946 }
3683 3947
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) { 3948 if (ret == 0) {
3689 if (add_backref) { 3949 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3690 ret = btrfs_insert_inode_ref(trans, root, 3950 parent_inode->i_ino, &key,
3691 name, name_len, 3951 btrfs_inode_type(inode), index);
3692 inode->i_ino, 3952 BUG_ON(ret);
3693 parent_inode->i_ino, 3953
3694 index);
3695 }
3696 btrfs_i_size_write(parent_inode, parent_inode->i_size + 3954 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3697 name_len * 2); 3955 name_len * 2);
3698 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; 3956 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
@@ -3875,18 +4133,16 @@ static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3875 4133
3876 err = btrfs_add_nondir(trans, dentry, inode, 1, index); 4134 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3877 4135
3878 if (err) 4136 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; 4137 drop_inode = 1;
4138 } else {
4139 btrfs_update_inode_block_group(trans, dir);
4140 err = btrfs_update_inode(trans, root, inode);
4141 BUG_ON(err);
4142 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
4143 }
3886 4144
3887 nr = trans->blocks_used; 4145 nr = trans->blocks_used;
3888
3889 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
3890 btrfs_end_transaction_throttle(trans, root); 4146 btrfs_end_transaction_throttle(trans, root);
3891fail: 4147fail:
3892 if (drop_inode) { 4148 if (drop_inode) {
@@ -4064,11 +4320,11 @@ struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4064 int compressed; 4320 int compressed;
4065 4321
4066again: 4322again:
4067 spin_lock(&em_tree->lock); 4323 read_lock(&em_tree->lock);
4068 em = lookup_extent_mapping(em_tree, start, len); 4324 em = lookup_extent_mapping(em_tree, start, len);
4069 if (em) 4325 if (em)
4070 em->bdev = root->fs_info->fs_devices->latest_bdev; 4326 em->bdev = root->fs_info->fs_devices->latest_bdev;
4071 spin_unlock(&em_tree->lock); 4327 read_unlock(&em_tree->lock);
4072 4328
4073 if (em) { 4329 if (em) {
4074 if (em->start > start || em->start + em->len <= start) 4330 if (em->start > start || em->start + em->len <= start)
@@ -4215,6 +4471,11 @@ again:
4215 map = kmap(page); 4471 map = kmap(page);
4216 read_extent_buffer(leaf, map + pg_offset, ptr, 4472 read_extent_buffer(leaf, map + pg_offset, ptr,
4217 copy_size); 4473 copy_size);
4474 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
4475 memset(map + pg_offset + copy_size, 0,
4476 PAGE_CACHE_SIZE - pg_offset -
4477 copy_size);
4478 }
4218 kunmap(page); 4479 kunmap(page);
4219 } 4480 }
4220 flush_dcache_page(page); 4481 flush_dcache_page(page);
@@ -4259,7 +4520,7 @@ insert:
4259 } 4520 }
4260 4521
4261 err = 0; 4522 err = 0;
4262 spin_lock(&em_tree->lock); 4523 write_lock(&em_tree->lock);
4263 ret = add_extent_mapping(em_tree, em); 4524 ret = add_extent_mapping(em_tree, em);
4264 /* it is possible that someone inserted the extent into the tree 4525 /* it is possible that someone inserted the extent into the tree
4265 * while we had the lock dropped. It is also possible that 4526 * while we had the lock dropped. It is also possible that
@@ -4299,7 +4560,7 @@ insert:
4299 err = 0; 4560 err = 0;
4300 } 4561 }
4301 } 4562 }
4302 spin_unlock(&em_tree->lock); 4563 write_unlock(&em_tree->lock);
4303out: 4564out:
4304 if (path) 4565 if (path)
4305 btrfs_free_path(path); 4566 btrfs_free_path(path);
@@ -4398,13 +4659,21 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4398 u64 page_start = page_offset(page); 4659 u64 page_start = page_offset(page);
4399 u64 page_end = page_start + PAGE_CACHE_SIZE - 1; 4660 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4400 4661
4662
4663 /*
4664 * we have the page locked, so new writeback can't start,
4665 * and the dirty bit won't be cleared while we are here.
4666 *
4667 * Wait for IO on this page so that we can safely clear
4668 * the PagePrivate2 bit and do ordered accounting
4669 */
4401 wait_on_page_writeback(page); 4670 wait_on_page_writeback(page);
4671
4402 tree = &BTRFS_I(page->mapping->host)->io_tree; 4672 tree = &BTRFS_I(page->mapping->host)->io_tree;
4403 if (offset) { 4673 if (offset) {
4404 btrfs_releasepage(page, GFP_NOFS); 4674 btrfs_releasepage(page, GFP_NOFS);
4405 return; 4675 return;
4406 } 4676 }
4407
4408 lock_extent(tree, page_start, page_end, GFP_NOFS); 4677 lock_extent(tree, page_start, page_end, GFP_NOFS);
4409 ordered = btrfs_lookup_ordered_extent(page->mapping->host, 4678 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4410 page_offset(page)); 4679 page_offset(page));
@@ -4415,16 +4684,21 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4415 */ 4684 */
4416 clear_extent_bit(tree, page_start, page_end, 4685 clear_extent_bit(tree, page_start, page_end,
4417 EXTENT_DIRTY | EXTENT_DELALLOC | 4686 EXTENT_DIRTY | EXTENT_DELALLOC |
4418 EXTENT_LOCKED, 1, 0, GFP_NOFS); 4687 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
4419 btrfs_finish_ordered_io(page->mapping->host, 4688 /*
4420 page_start, page_end); 4689 * whoever cleared the private bit is responsible
4690 * for the finish_ordered_io
4691 */
4692 if (TestClearPagePrivate2(page)) {
4693 btrfs_finish_ordered_io(page->mapping->host,
4694 page_start, page_end);
4695 }
4421 btrfs_put_ordered_extent(ordered); 4696 btrfs_put_ordered_extent(ordered);
4422 lock_extent(tree, page_start, page_end, GFP_NOFS); 4697 lock_extent(tree, page_start, page_end, GFP_NOFS);
4423 } 4698 }
4424 clear_extent_bit(tree, page_start, page_end, 4699 clear_extent_bit(tree, page_start, page_end,
4425 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | 4700 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
4426 EXTENT_ORDERED, 4701 1, 1, NULL, GFP_NOFS);
4427 1, 1, GFP_NOFS);
4428 __btrfs_releasepage(page, GFP_NOFS); 4702 __btrfs_releasepage(page, GFP_NOFS);
4429 4703
4430 ClearPageChecked(page); 4704 ClearPageChecked(page);
@@ -4521,11 +4795,14 @@ again:
4521 } 4795 }
4522 ClearPageChecked(page); 4796 ClearPageChecked(page);
4523 set_page_dirty(page); 4797 set_page_dirty(page);
4798 SetPageUptodate(page);
4524 4799
4525 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1; 4800 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
4526 unlock_extent(io_tree, page_start, page_end, GFP_NOFS); 4801 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4527 4802
4528out_unlock: 4803out_unlock:
4804 if (!ret)
4805 return VM_FAULT_LOCKED;
4529 unlock_page(page); 4806 unlock_page(page);
4530out: 4807out:
4531 return ret; 4808 return ret;
@@ -4594,11 +4871,11 @@ out:
4594 * create a new subvolume directory/inode (helper for the ioctl). 4871 * create a new subvolume directory/inode (helper for the ioctl).
4595 */ 4872 */
4596int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, 4873int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4597 struct btrfs_root *new_root, struct dentry *dentry, 4874 struct btrfs_root *new_root,
4598 u64 new_dirid, u64 alloc_hint) 4875 u64 new_dirid, u64 alloc_hint)
4599{ 4876{
4600 struct inode *inode; 4877 struct inode *inode;
4601 int error; 4878 int err;
4602 u64 index = 0; 4879 u64 index = 0;
4603 4880
4604 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, 4881 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
@@ -4611,11 +4888,10 @@ int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4611 inode->i_nlink = 1; 4888 inode->i_nlink = 1;
4612 btrfs_i_size_write(inode, 0); 4889 btrfs_i_size_write(inode, 0);
4613 4890
4614 error = btrfs_update_inode(trans, new_root, inode); 4891 err = btrfs_update_inode(trans, new_root, inode);
4615 if (error) 4892 BUG_ON(err);
4616 return error;
4617 4893
4618 d_instantiate(dentry, inode); 4894 iput(inode);
4619 return 0; 4895 return 0;
4620} 4896}
4621 4897
@@ -4693,6 +4969,16 @@ void btrfs_destroy_inode(struct inode *inode)
4693 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); 4969 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4694} 4970}
4695 4971
4972void btrfs_drop_inode(struct inode *inode)
4973{
4974 struct btrfs_root *root = BTRFS_I(inode)->root;
4975
4976 if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0)
4977 generic_delete_inode(inode);
4978 else
4979 generic_drop_inode(inode);
4980}
4981
4696static void init_once(void *foo) 4982static void init_once(void *foo)
4697{ 4983{
4698 struct btrfs_inode *ei = (struct btrfs_inode *) foo; 4984 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
@@ -4761,31 +5047,32 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4761{ 5047{
4762 struct btrfs_trans_handle *trans; 5048 struct btrfs_trans_handle *trans;
4763 struct btrfs_root *root = BTRFS_I(old_dir)->root; 5049 struct btrfs_root *root = BTRFS_I(old_dir)->root;
5050 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
4764 struct inode *new_inode = new_dentry->d_inode; 5051 struct inode *new_inode = new_dentry->d_inode;
4765 struct inode *old_inode = old_dentry->d_inode; 5052 struct inode *old_inode = old_dentry->d_inode;
4766 struct timespec ctime = CURRENT_TIME; 5053 struct timespec ctime = CURRENT_TIME;
4767 u64 index = 0; 5054 u64 index = 0;
5055 u64 root_objectid;
4768 int ret; 5056 int ret;
4769 5057
4770 /* we're not allowed to rename between subvolumes */ 5058 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4771 if (BTRFS_I(old_inode)->root->root_key.objectid != 5059 return -EPERM;
4772 BTRFS_I(new_dir)->root->root_key.objectid) 5060
5061 /* we only allow rename subvolume link between subvolumes */
5062 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
4773 return -EXDEV; 5063 return -EXDEV;
4774 5064
4775 if (S_ISDIR(old_inode->i_mode) && new_inode && 5065 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
4776 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) { 5066 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
4777 return -ENOTEMPTY; 5067 return -ENOTEMPTY;
4778 }
4779 5068
4780 /* to rename a snapshot or subvolume, we need to juggle the 5069 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4781 * backrefs. This isn't coded yet 5070 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4782 */ 5071 return -ENOTEMPTY;
4783 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4784 return -EXDEV;
4785 5072
4786 ret = btrfs_check_metadata_free_space(root); 5073 ret = btrfs_check_metadata_free_space(root);
4787 if (ret) 5074 if (ret)
4788 goto out_unlock; 5075 return ret;
4789 5076
4790 /* 5077 /*
4791 * we're using rename to replace one file with another. 5078 * we're using rename to replace one file with another.
@@ -4796,8 +5083,40 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4796 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) 5083 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
4797 filemap_flush(old_inode->i_mapping); 5084 filemap_flush(old_inode->i_mapping);
4798 5085
5086 /* close the racy window with snapshot create/destroy ioctl */
5087 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5088 down_read(&root->fs_info->subvol_sem);
5089
4799 trans = btrfs_start_transaction(root, 1); 5090 trans = btrfs_start_transaction(root, 1);
5091 btrfs_set_trans_block_group(trans, new_dir);
5092
5093 if (dest != root)
5094 btrfs_record_root_in_trans(trans, dest);
4800 5095
5096 ret = btrfs_set_inode_index(new_dir, &index);
5097 if (ret)
5098 goto out_fail;
5099
5100 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5101 /* force full log commit if subvolume involved. */
5102 root->fs_info->last_trans_log_full_commit = trans->transid;
5103 } else {
5104 ret = btrfs_insert_inode_ref(trans, dest,
5105 new_dentry->d_name.name,
5106 new_dentry->d_name.len,
5107 old_inode->i_ino,
5108 new_dir->i_ino, index);
5109 if (ret)
5110 goto out_fail;
5111 /*
5112 * this is an ugly little race, but the rename is required
5113 * to make sure that if we crash, the inode is either at the
5114 * old name or the new one. pinning the log transaction lets
5115 * us make sure we don't allow a log commit to come in after
5116 * we unlink the name but before we add the new name back in.
5117 */
5118 btrfs_pin_log_trans(root);
5119 }
4801 /* 5120 /*
4802 * make sure the inode gets flushed if it is replacing 5121 * make sure the inode gets flushed if it is replacing
4803 * something. 5122 * something.
@@ -4807,18 +5126,6 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4807 btrfs_add_ordered_operation(trans, root, old_inode); 5126 btrfs_add_ordered_operation(trans, root, old_inode);
4808 } 5127 }
4809 5128
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; 5129 old_dir->i_ctime = old_dir->i_mtime = ctime;
4823 new_dir->i_ctime = new_dir->i_mtime = ctime; 5130 new_dir->i_ctime = new_dir->i_mtime = ctime;
4824 old_inode->i_ctime = ctime; 5131 old_inode->i_ctime = ctime;
@@ -4826,47 +5133,58 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4826 if (old_dentry->d_parent != new_dentry->d_parent) 5133 if (old_dentry->d_parent != new_dentry->d_parent)
4827 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); 5134 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
4828 5135
4829 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode, 5136 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4830 old_dentry->d_name.name, 5137 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
4831 old_dentry->d_name.len); 5138 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
4832 if (ret) 5139 old_dentry->d_name.name,
4833 goto out_fail; 5140 old_dentry->d_name.len);
5141 } else {
5142 btrfs_inc_nlink(old_dentry->d_inode);
5143 ret = btrfs_unlink_inode(trans, root, old_dir,
5144 old_dentry->d_inode,
5145 old_dentry->d_name.name,
5146 old_dentry->d_name.len);
5147 }
5148 BUG_ON(ret);
4834 5149
4835 if (new_inode) { 5150 if (new_inode) {
4836 new_inode->i_ctime = CURRENT_TIME; 5151 new_inode->i_ctime = CURRENT_TIME;
4837 ret = btrfs_unlink_inode(trans, root, new_dir, 5152 if (unlikely(new_inode->i_ino ==
4838 new_dentry->d_inode, 5153 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4839 new_dentry->d_name.name, 5154 root_objectid = BTRFS_I(new_inode)->location.objectid;
4840 new_dentry->d_name.len); 5155 ret = btrfs_unlink_subvol(trans, dest, new_dir,
4841 if (ret) 5156 root_objectid,
4842 goto out_fail; 5157 new_dentry->d_name.name,
5158 new_dentry->d_name.len);
5159 BUG_ON(new_inode->i_nlink == 0);
5160 } else {
5161 ret = btrfs_unlink_inode(trans, dest, new_dir,
5162 new_dentry->d_inode,
5163 new_dentry->d_name.name,
5164 new_dentry->d_name.len);
5165 }
5166 BUG_ON(ret);
4843 if (new_inode->i_nlink == 0) { 5167 if (new_inode->i_nlink == 0) {
4844 ret = btrfs_orphan_add(trans, new_dentry->d_inode); 5168 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4845 if (ret) 5169 BUG_ON(ret);
4846 goto out_fail;
4847 } 5170 }
4848
4849 } 5171 }
4850 ret = btrfs_set_inode_index(new_dir, &index);
4851 if (ret)
4852 goto out_fail;
4853 5172
4854 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode, 5173 ret = btrfs_add_link(trans, new_dir, old_inode,
4855 old_inode, new_dentry->d_name.name, 5174 new_dentry->d_name.name,
4856 new_dentry->d_name.len, 1, index); 5175 new_dentry->d_name.len, 0, index);
4857 if (ret) 5176 BUG_ON(ret);
4858 goto out_fail;
4859 5177
4860 btrfs_log_new_name(trans, old_inode, old_dir, 5178 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
4861 new_dentry->d_parent); 5179 btrfs_log_new_name(trans, old_inode, old_dir,
5180 new_dentry->d_parent);
5181 btrfs_end_log_trans(root);
5182 }
4862out_fail: 5183out_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); 5184 btrfs_end_transaction_throttle(trans, root);
4869out_unlock: 5185
5186 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5187 up_read(&root->fs_info->subvol_sem);
4870 return ret; 5188 return ret;
4871} 5189}
4872 5190
@@ -5058,6 +5376,8 @@ static int prealloc_file_range(struct btrfs_trans_handle *trans,
5058 0, 0, 0, 5376 0, 0, 0,
5059 BTRFS_FILE_EXTENT_PREALLOC); 5377 BTRFS_FILE_EXTENT_PREALLOC);
5060 BUG_ON(ret); 5378 BUG_ON(ret);
5379 btrfs_drop_extent_cache(inode, cur_offset,
5380 cur_offset + ins.offset -1, 0);
5061 num_bytes -= ins.offset; 5381 num_bytes -= ins.offset;
5062 cur_offset += ins.offset; 5382 cur_offset += ins.offset;
5063 alloc_hint = ins.objectid + ins.offset; 5383 alloc_hint = ins.objectid + ins.offset;
@@ -5223,6 +5543,7 @@ static const struct inode_operations btrfs_dir_ro_inode_operations = {
5223 .lookup = btrfs_lookup, 5543 .lookup = btrfs_lookup,
5224 .permission = btrfs_permission, 5544 .permission = btrfs_permission,
5225}; 5545};
5546
5226static struct file_operations btrfs_dir_file_operations = { 5547static struct file_operations btrfs_dir_file_operations = {
5227 .llseek = generic_file_llseek, 5548 .llseek = generic_file_llseek,
5228 .read = generic_read_dir, 5549 .read = generic_read_dir,
@@ -5310,3 +5631,7 @@ static const struct inode_operations btrfs_symlink_inode_operations = {
5310 .listxattr = btrfs_listxattr, 5631 .listxattr = btrfs_listxattr,
5311 .removexattr = btrfs_removexattr, 5632 .removexattr = btrfs_removexattr,
5312}; 5633};
5634
5635struct dentry_operations btrfs_dentry_operations = {
5636 .d_delete = btrfs_dentry_delete,
5637};
diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index bd88f25889f7..a8577a7f26ab 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;
@@ -241,7 +241,7 @@ static noinline int create_subvol(struct btrfs_root *root,
241 241
242 ret = btrfs_check_metadata_free_space(root); 242 ret = btrfs_check_metadata_free_space(root);
243 if (ret) 243 if (ret)
244 goto fail_commit; 244 return ret;
245 245
246 trans = btrfs_start_transaction(root, 1); 246 trans = btrfs_start_transaction(root, 1);
247 BUG_ON(!trans); 247 BUG_ON(!trans);
@@ -304,11 +304,17 @@ static noinline int create_subvol(struct btrfs_root *root,
304 if (ret) 304 if (ret)
305 goto fail; 305 goto fail;
306 306
307 key.offset = (u64)-1;
308 new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
309 BUG_ON(IS_ERR(new_root));
310
311 btrfs_record_root_in_trans(trans, new_root);
312
313 ret = btrfs_create_subvol_root(trans, new_root, new_dirid,
314 BTRFS_I(dir)->block_group);
307 /* 315 /*
308 * insert the directory item 316 * insert the directory item
309 */ 317 */
310 key.offset = (u64)-1;
311 dir = dentry->d_parent->d_inode;
312 ret = btrfs_set_inode_index(dir, &index); 318 ret = btrfs_set_inode_index(dir, &index);
313 BUG_ON(ret); 319 BUG_ON(ret);
314 320
@@ -322,44 +328,18 @@ static noinline int create_subvol(struct btrfs_root *root,
322 ret = btrfs_update_inode(trans, root, dir); 328 ret = btrfs_update_inode(trans, root, dir);
323 BUG_ON(ret); 329 BUG_ON(ret);
324 330
325 /* add the backref first */
326 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, 331 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
327 objectid, BTRFS_ROOT_BACKREF_KEY, 332 objectid, root->root_key.objectid,
328 root->root_key.objectid,
329 dir->i_ino, index, name, namelen); 333 dir->i_ino, index, name, namelen);
330 334
331 BUG_ON(ret); 335 BUG_ON(ret);
332 336
333 /* now add the forward ref */ 337 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: 338fail:
357 nr = trans->blocks_used; 339 nr = trans->blocks_used;
358 err = btrfs_commit_transaction(trans, new_root); 340 err = btrfs_commit_transaction(trans, root);
359 if (err && !ret) 341 if (err && !ret)
360 ret = err; 342 ret = err;
361fail_commit:
362 btrfs_btree_balance_dirty(root, nr);
363 return ret; 343 return ret;
364} 344}
365 345
@@ -420,14 +400,15 @@ 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 400 * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
421 * inside this filesystem so it's quite a bit simpler. 401 * inside this filesystem so it's quite a bit simpler.
422 */ 402 */
423static noinline int btrfs_mksubvol(struct path *parent, char *name, 403static noinline int btrfs_mksubvol(struct path *parent,
424 int mode, int namelen, 404 char *name, int namelen,
425 struct btrfs_root *snap_src) 405 struct btrfs_root *snap_src)
426{ 406{
407 struct inode *dir = parent->dentry->d_inode;
427 struct dentry *dentry; 408 struct dentry *dentry;
428 int error; 409 int error;
429 410
430 mutex_lock_nested(&parent->dentry->d_inode->i_mutex, I_MUTEX_PARENT); 411 mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
431 412
432 dentry = lookup_one_len(name, parent->dentry, namelen); 413 dentry = lookup_one_len(name, parent->dentry, namelen);
433 error = PTR_ERR(dentry); 414 error = PTR_ERR(dentry);
@@ -438,99 +419,39 @@ static noinline int btrfs_mksubvol(struct path *parent, char *name,
438 if (dentry->d_inode) 419 if (dentry->d_inode)
439 goto out_dput; 420 goto out_dput;
440 421
441 if (!IS_POSIXACL(parent->dentry->d_inode))
442 mode &= ~current_umask();
443
444 error = mnt_want_write(parent->mnt); 422 error = mnt_want_write(parent->mnt);
445 if (error) 423 if (error)
446 goto out_dput; 424 goto out_dput;
447 425
448 error = btrfs_may_create(parent->dentry->d_inode, dentry); 426 error = btrfs_may_create(dir, dentry);
449 if (error) 427 if (error)
450 goto out_drop_write; 428 goto out_drop_write;
451 429
452 /* 430 down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
453 * Actually perform the low-level subvolume creation after all 431
454 * this VFS fuzz. 432 if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
455 * 433 goto out_up_read;
456 * Eventually we want to pass in an inode under which we create this 434
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) { 435 if (snap_src) {
463 struct dentry *dir = dentry->d_parent; 436 error = create_snapshot(snap_src, dentry,
464 struct dentry *test = dir->d_parent; 437 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 { 438 } else {
517 error = create_subvol(BTRFS_I(parent->dentry->d_inode)->root, 439 error = create_subvol(BTRFS_I(dir)->root, dentry,
518 dentry, name, namelen); 440 name, namelen);
519 } 441 }
520 if (error) 442 if (!error)
521 goto out_drop_write; 443 fsnotify_mkdir(dir, dentry);
522 444out_up_read:
523 fsnotify_mkdir(parent->dentry->d_inode, dentry); 445 up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
524out_drop_write: 446out_drop_write:
525 mnt_drop_write(parent->mnt); 447 mnt_drop_write(parent->mnt);
526out_dput: 448out_dput:
527 dput(dentry); 449 dput(dentry);
528out_unlock: 450out_unlock:
529 mutex_unlock(&parent->dentry->d_inode->i_mutex); 451 mutex_unlock(&dir->i_mutex);
530 return error; 452 return error;
531} 453}
532 454
533
534static int btrfs_defrag_file(struct file *file) 455static int btrfs_defrag_file(struct file *file)
535{ 456{
536 struct inode *inode = fdentry(file)->d_inode; 457 struct inode *inode = fdentry(file)->d_inode;
@@ -596,9 +517,8 @@ again:
596 clear_page_dirty_for_io(page); 517 clear_page_dirty_for_io(page);
597 518
598 btrfs_set_extent_delalloc(inode, page_start, page_end); 519 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); 520 set_page_dirty(page);
521 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
602 unlock_page(page); 522 unlock_page(page);
603 page_cache_release(page); 523 page_cache_release(page);
604 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1); 524 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
@@ -609,7 +529,8 @@ out_unlock:
609 return 0; 529 return 0;
610} 530}
611 531
612static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg) 532static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
533 void __user *arg)
613{ 534{
614 u64 new_size; 535 u64 new_size;
615 u64 old_size; 536 u64 old_size;
@@ -718,10 +639,7 @@ static noinline int btrfs_ioctl_snap_create(struct file *file,
718{ 639{
719 struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root; 640 struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
720 struct btrfs_ioctl_vol_args *vol_args; 641 struct btrfs_ioctl_vol_args *vol_args;
721 struct btrfs_dir_item *di;
722 struct btrfs_path *path;
723 struct file *src_file; 642 struct file *src_file;
724 u64 root_dirid;
725 int namelen; 643 int namelen;
726 int ret = 0; 644 int ret = 0;
727 645
@@ -739,32 +657,9 @@ static noinline int btrfs_ioctl_snap_create(struct file *file,
739 goto out; 657 goto out;
740 } 658 }
741 659
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) { 660 if (subvol) {
765 ret = btrfs_mksubvol(&file->f_path, vol_args->name, 661 ret = btrfs_mksubvol(&file->f_path, vol_args->name, namelen,
766 file->f_path.dentry->d_inode->i_mode, 662 NULL);
767 namelen, NULL);
768 } else { 663 } else {
769 struct inode *src_inode; 664 struct inode *src_inode;
770 src_file = fget(vol_args->fd); 665 src_file = fget(vol_args->fd);
@@ -781,17 +676,156 @@ static noinline int btrfs_ioctl_snap_create(struct file *file,
781 fput(src_file); 676 fput(src_file);
782 goto out; 677 goto out;
783 } 678 }
784 ret = btrfs_mksubvol(&file->f_path, vol_args->name, 679 ret = btrfs_mksubvol(&file->f_path, vol_args->name, namelen,
785 file->f_path.dentry->d_inode->i_mode, 680 BTRFS_I(src_inode)->root);
786 namelen, BTRFS_I(src_inode)->root);
787 fput(src_file); 681 fput(src_file);
788 } 682 }
789
790out: 683out:
791 kfree(vol_args); 684 kfree(vol_args);
792 return ret; 685 return ret;
793} 686}
794 687
688/*
689 * helper to check if the subvolume references other subvolumes
690 */
691static noinline int may_destroy_subvol(struct btrfs_root *root)
692{
693 struct btrfs_path *path;
694 struct btrfs_key key;
695 int ret;
696
697 path = btrfs_alloc_path();
698 if (!path)
699 return -ENOMEM;
700
701 key.objectid = root->root_key.objectid;
702 key.type = BTRFS_ROOT_REF_KEY;
703 key.offset = (u64)-1;
704
705 ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
706 &key, path, 0, 0);
707 if (ret < 0)
708 goto out;
709 BUG_ON(ret == 0);
710
711 ret = 0;
712 if (path->slots[0] > 0) {
713 path->slots[0]--;
714 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
715 if (key.objectid == root->root_key.objectid &&
716 key.type == BTRFS_ROOT_REF_KEY)
717 ret = -ENOTEMPTY;
718 }
719out:
720 btrfs_free_path(path);
721 return ret;
722}
723
724static noinline int btrfs_ioctl_snap_destroy(struct file *file,
725 void __user *arg)
726{
727 struct dentry *parent = fdentry(file);
728 struct dentry *dentry;
729 struct inode *dir = parent->d_inode;
730 struct inode *inode;
731 struct btrfs_root *root = BTRFS_I(dir)->root;
732 struct btrfs_root *dest = NULL;
733 struct btrfs_ioctl_vol_args *vol_args;
734 struct btrfs_trans_handle *trans;
735 int namelen;
736 int ret;
737 int err = 0;
738
739 if (!capable(CAP_SYS_ADMIN))
740 return -EPERM;
741
742 vol_args = memdup_user(arg, sizeof(*vol_args));
743 if (IS_ERR(vol_args))
744 return PTR_ERR(vol_args);
745
746 vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
747 namelen = strlen(vol_args->name);
748 if (strchr(vol_args->name, '/') ||
749 strncmp(vol_args->name, "..", namelen) == 0) {
750 err = -EINVAL;
751 goto out;
752 }
753
754 err = mnt_want_write(file->f_path.mnt);
755 if (err)
756 goto out;
757
758 mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
759 dentry = lookup_one_len(vol_args->name, parent, namelen);
760 if (IS_ERR(dentry)) {
761 err = PTR_ERR(dentry);
762 goto out_unlock_dir;
763 }
764
765 if (!dentry->d_inode) {
766 err = -ENOENT;
767 goto out_dput;
768 }
769
770 inode = dentry->d_inode;
771 if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
772 err = -EINVAL;
773 goto out_dput;
774 }
775
776 dest = BTRFS_I(inode)->root;
777
778 mutex_lock(&inode->i_mutex);
779 err = d_invalidate(dentry);
780 if (err)
781 goto out_unlock;
782
783 down_write(&root->fs_info->subvol_sem);
784
785 err = may_destroy_subvol(dest);
786 if (err)
787 goto out_up_write;
788
789 trans = btrfs_start_transaction(root, 1);
790 ret = btrfs_unlink_subvol(trans, root, dir,
791 dest->root_key.objectid,
792 dentry->d_name.name,
793 dentry->d_name.len);
794 BUG_ON(ret);
795
796 btrfs_record_root_in_trans(trans, dest);
797
798 memset(&dest->root_item.drop_progress, 0,
799 sizeof(dest->root_item.drop_progress));
800 dest->root_item.drop_level = 0;
801 btrfs_set_root_refs(&dest->root_item, 0);
802
803 ret = btrfs_insert_orphan_item(trans,
804 root->fs_info->tree_root,
805 dest->root_key.objectid);
806 BUG_ON(ret);
807
808 ret = btrfs_commit_transaction(trans, root);
809 BUG_ON(ret);
810 inode->i_flags |= S_DEAD;
811out_up_write:
812 up_write(&root->fs_info->subvol_sem);
813out_unlock:
814 mutex_unlock(&inode->i_mutex);
815 if (!err) {
816 btrfs_invalidate_inodes(dest);
817 d_delete(dentry);
818 }
819out_dput:
820 dput(dentry);
821out_unlock_dir:
822 mutex_unlock(&dir->i_mutex);
823 mnt_drop_write(file->f_path.mnt);
824out:
825 kfree(vol_args);
826 return err;
827}
828
795static int btrfs_ioctl_defrag(struct file *file) 829static int btrfs_ioctl_defrag(struct file *file)
796{ 830{
797 struct inode *inode = fdentry(file)->d_inode; 831 struct inode *inode = fdentry(file)->d_inode;
@@ -865,8 +899,8 @@ static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
865 return ret; 899 return ret;
866} 900}
867 901
868static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd, 902static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
869 u64 off, u64 olen, u64 destoff) 903 u64 off, u64 olen, u64 destoff)
870{ 904{
871 struct inode *inode = fdentry(file)->d_inode; 905 struct inode *inode = fdentry(file)->d_inode;
872 struct btrfs_root *root = BTRFS_I(inode)->root; 906 struct btrfs_root *root = BTRFS_I(inode)->root;
@@ -976,7 +1010,7 @@ static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
976 1010
977 /* punch hole in destination first */ 1011 /* punch hole in destination first */
978 btrfs_drop_extents(trans, root, inode, off, off + len, 1012 btrfs_drop_extents(trans, root, inode, off, off + len,
979 off + len, 0, &hint_byte); 1013 off + len, 0, &hint_byte, 1);
980 1014
981 /* clone data */ 1015 /* clone data */
982 key.objectid = src->i_ino; 1016 key.objectid = src->i_ino;
@@ -1071,8 +1105,7 @@ static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
1071 datao += off - key.offset; 1105 datao += off - key.offset;
1072 datal -= off - key.offset; 1106 datal -= off - key.offset;
1073 } 1107 }
1074 if (key.offset + datao + datal + key.offset > 1108 if (key.offset + datao + datal > off + len)
1075 off + len)
1076 datal = off + len - key.offset - datao; 1109 datal = off + len - key.offset - datao;
1077 /* disko == 0 means it's a hole */ 1110 /* disko == 0 means it's a hole */
1078 if (!disko) 1111 if (!disko)
@@ -1258,6 +1291,8 @@ long btrfs_ioctl(struct file *file, unsigned int
1258 return btrfs_ioctl_snap_create(file, argp, 0); 1291 return btrfs_ioctl_snap_create(file, argp, 0);
1259 case BTRFS_IOC_SUBVOL_CREATE: 1292 case BTRFS_IOC_SUBVOL_CREATE:
1260 return btrfs_ioctl_snap_create(file, argp, 1); 1293 return btrfs_ioctl_snap_create(file, argp, 1);
1294 case BTRFS_IOC_SNAP_DESTROY:
1295 return btrfs_ioctl_snap_destroy(file, argp);
1261 case BTRFS_IOC_DEFRAG: 1296 case BTRFS_IOC_DEFRAG:
1262 return btrfs_ioctl_defrag(file); 1297 return btrfs_ioctl_defrag(file);
1263 case BTRFS_IOC_RESIZE: 1298 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 7b2f401e604e..b5d6d24726b0 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;
@@ -476,6 +474,7 @@ int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
476 u64 orig_end; 474 u64 orig_end;
477 u64 wait_end; 475 u64 wait_end;
478 struct btrfs_ordered_extent *ordered; 476 struct btrfs_ordered_extent *ordered;
477 int found;
479 478
480 if (start + len < start) { 479 if (start + len < start) {
481 orig_end = INT_LIMIT(loff_t); 480 orig_end = INT_LIMIT(loff_t);
@@ -502,6 +501,7 @@ again:
502 orig_end >> PAGE_CACHE_SHIFT); 501 orig_end >> PAGE_CACHE_SHIFT);
503 502
504 end = orig_end; 503 end = orig_end;
504 found = 0;
505 while (1) { 505 while (1) {
506 ordered = btrfs_lookup_first_ordered_extent(inode, end); 506 ordered = btrfs_lookup_first_ordered_extent(inode, end);
507 if (!ordered) 507 if (!ordered)
@@ -514,6 +514,7 @@ again:
514 btrfs_put_ordered_extent(ordered); 514 btrfs_put_ordered_extent(ordered);
515 break; 515 break;
516 } 516 }
517 found++;
517 btrfs_start_ordered_extent(inode, ordered, 1); 518 btrfs_start_ordered_extent(inode, ordered, 1);
518 end = ordered->file_offset; 519 end = ordered->file_offset;
519 btrfs_put_ordered_extent(ordered); 520 btrfs_put_ordered_extent(ordered);
@@ -521,8 +522,8 @@ again:
521 break; 522 break;
522 end--; 523 end--;
523 } 524 }
524 if (test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end, 525 if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
525 EXTENT_ORDERED | EXTENT_DELALLOC, 0)) { 526 EXTENT_DELALLOC, 0, NULL)) {
526 schedule_timeout(1); 527 schedule_timeout(1);
527 goto again; 528 goto again;
528 } 529 }
@@ -613,7 +614,7 @@ int btrfs_ordered_update_i_size(struct inode *inode,
613 */ 614 */
614 if (test_range_bit(io_tree, disk_i_size, 615 if (test_range_bit(io_tree, disk_i_size,
615 ordered->file_offset + ordered->len - 1, 616 ordered->file_offset + ordered->len - 1,
616 EXTENT_DELALLOC, 0)) { 617 EXTENT_DELALLOC, 0, NULL)) {
617 goto out; 618 goto out;
618 } 619 }
619 /* 620 /*
@@ -664,7 +665,7 @@ int btrfs_ordered_update_i_size(struct inode *inode,
664 */ 665 */
665 if (i_size_test > entry_end(ordered) && 666 if (i_size_test > entry_end(ordered) &&
666 !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1, 667 !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1,
667 EXTENT_DELALLOC, 0)) { 668 EXTENT_DELALLOC, 0, NULL)) {
668 new_i_size = min_t(u64, i_size_test, i_size_read(inode)); 669 new_i_size = min_t(u64, i_size_test, i_size_read(inode));
669 } 670 }
670 BTRFS_I(inode)->disk_i_size = new_i_size; 671 BTRFS_I(inode)->disk_i_size = new_i_size;
diff --git a/fs/btrfs/ordered-data.h b/fs/btrfs/ordered-data.h
index 3d31c8827b01..993a7ea45c70 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
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..361ad323faac 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;
@@ -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
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 2db17cd66fc5..67035385444c 100644
--- a/fs/btrfs/super.c
+++ b/fs/btrfs/super.c
@@ -676,6 +676,7 @@ static int btrfs_unfreeze(struct super_block *sb)
676} 676}
677 677
678static const struct super_operations btrfs_super_ops = { 678static const struct super_operations btrfs_super_ops = {
679 .drop_inode = btrfs_drop_inode,
679 .delete_inode = btrfs_delete_inode, 680 .delete_inode = btrfs_delete_inode,
680 .put_super = btrfs_put_super, 681 .put_super = btrfs_put_super,
681 .sync_fs = btrfs_sync_fs, 682 .sync_fs = btrfs_sync_fs,
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index cdbb5022da52..88f866f85e7a 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,
@@ -720,7 +719,8 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
720 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 719 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
721 720
722 key.objectid = objectid; 721 key.objectid = objectid;
723 key.offset = 0; 722 /* record when the snapshot was created in key.offset */
723 key.offset = trans->transid;
724 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); 724 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
725 725
726 old = btrfs_lock_root_node(root); 726 old = btrfs_lock_root_node(root);
@@ -778,24 +778,14 @@ static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info,
778 ret = btrfs_update_inode(trans, parent_root, parent_inode); 778 ret = btrfs_update_inode(trans, parent_root, parent_inode);
779 BUG_ON(ret); 779 BUG_ON(ret);
780 780
781 /* add the backref first */
782 ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root, 781 ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
783 pending->root_key.objectid, 782 pending->root_key.objectid,
784 BTRFS_ROOT_BACKREF_KEY,
785 parent_root->root_key.objectid, 783 parent_root->root_key.objectid,
786 parent_inode->i_ino, index, pending->name, 784 parent_inode->i_ino, index, pending->name,
787 namelen); 785 namelen);
788 786
789 BUG_ON(ret); 787 BUG_ON(ret);
790 788
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); 789 inode = btrfs_lookup_dentry(parent_inode, pending->dentry);
800 d_instantiate(pending->dentry, inode); 790 d_instantiate(pending->dentry, inode);
801fail: 791fail:
@@ -874,7 +864,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
874 unsigned long timeout = 1; 864 unsigned long timeout = 1;
875 struct btrfs_transaction *cur_trans; 865 struct btrfs_transaction *cur_trans;
876 struct btrfs_transaction *prev_trans = NULL; 866 struct btrfs_transaction *prev_trans = NULL;
877 struct extent_io_tree *pinned_copy;
878 DEFINE_WAIT(wait); 867 DEFINE_WAIT(wait);
879 int ret; 868 int ret;
880 int should_grow = 0; 869 int should_grow = 0;
@@ -915,13 +904,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
915 return 0; 904 return 0;
916 } 905 }
917 906
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; 907 trans->transaction->in_commit = 1;
926 trans->transaction->blocked = 1; 908 trans->transaction->blocked = 1;
927 if (cur_trans->list.prev != &root->fs_info->trans_list) { 909 if (cur_trans->list.prev != &root->fs_info->trans_list) {
@@ -1019,6 +1001,8 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1019 ret = commit_cowonly_roots(trans, root); 1001 ret = commit_cowonly_roots(trans, root);
1020 BUG_ON(ret); 1002 BUG_ON(ret);
1021 1003
1004 btrfs_prepare_extent_commit(trans, root);
1005
1022 cur_trans = root->fs_info->running_transaction; 1006 cur_trans = root->fs_info->running_transaction;
1023 spin_lock(&root->fs_info->new_trans_lock); 1007 spin_lock(&root->fs_info->new_trans_lock);
1024 root->fs_info->running_transaction = NULL; 1008 root->fs_info->running_transaction = NULL;
@@ -1042,8 +1026,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1042 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, 1026 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1043 sizeof(root->fs_info->super_copy)); 1027 sizeof(root->fs_info->super_copy));
1044 1028
1045 btrfs_copy_pinned(root, pinned_copy);
1046
1047 trans->transaction->blocked = 0; 1029 trans->transaction->blocked = 0;
1048 1030
1049 wake_up(&root->fs_info->transaction_wait); 1031 wake_up(&root->fs_info->transaction_wait);
@@ -1059,8 +1041,7 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1059 */ 1041 */
1060 mutex_unlock(&root->fs_info->tree_log_mutex); 1042 mutex_unlock(&root->fs_info->tree_log_mutex);
1061 1043
1062 btrfs_finish_extent_commit(trans, root, pinned_copy); 1044 btrfs_finish_extent_commit(trans, root);
1063 kfree(pinned_copy);
1064 1045
1065 /* do the directory inserts of any pending snapshot creations */ 1046 /* do the directory inserts of any pending snapshot creations */
1066 finish_pending_snapshots(trans, root->fs_info); 1047 finish_pending_snapshots(trans, root->fs_info);
@@ -1096,8 +1077,13 @@ int btrfs_clean_old_snapshots(struct btrfs_root *root)
1096 1077
1097 while (!list_empty(&list)) { 1078 while (!list_empty(&list)) {
1098 root = list_entry(list.next, struct btrfs_root, root_list); 1079 root = list_entry(list.next, struct btrfs_root, root_list);
1099 list_del_init(&root->root_list); 1080 list_del(&root->root_list);
1100 btrfs_drop_snapshot(root, 0); 1081
1082 if (btrfs_header_backref_rev(root->node) <
1083 BTRFS_MIXED_BACKREF_REV)
1084 btrfs_drop_snapshot(root, 0);
1085 else
1086 btrfs_drop_snapshot(root, 1);
1101 } 1087 }
1102 return 0; 1088 return 0;
1103} 1089}
diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c
index 30c0d45c1b5e..7827841b55cb 100644
--- a/fs/btrfs/tree-log.c
+++ b/fs/btrfs/tree-log.c
@@ -263,8 +263,8 @@ static int process_one_buffer(struct btrfs_root *log,
263 struct walk_control *wc, u64 gen) 263 struct walk_control *wc, u64 gen)
264{ 264{
265 if (wc->pin) 265 if (wc->pin)
266 btrfs_update_pinned_extents(log->fs_info->extent_root, 266 btrfs_pin_extent(log->fs_info->extent_root,
267 eb->start, eb->len, 1); 267 eb->start, eb->len, 0);
268 268
269 if (btrfs_buffer_uptodate(eb, gen)) { 269 if (btrfs_buffer_uptodate(eb, gen)) {
270 if (wc->write) 270 if (wc->write)
@@ -534,7 +534,7 @@ static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
534 saved_nbytes = inode_get_bytes(inode); 534 saved_nbytes = inode_get_bytes(inode);
535 /* drop any overlapping extents */ 535 /* drop any overlapping extents */
536 ret = btrfs_drop_extents(trans, root, inode, 536 ret = btrfs_drop_extents(trans, root, inode,
537 start, extent_end, extent_end, start, &alloc_hint); 537 start, extent_end, extent_end, start, &alloc_hint, 1);
538 BUG_ON(ret); 538 BUG_ON(ret);
539 539
540 if (found_type == BTRFS_FILE_EXTENT_REG || 540 if (found_type == BTRFS_FILE_EXTENT_REG ||
@@ -2841,7 +2841,7 @@ 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) 2841 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2842 break; 2842 break;
2843 2843
2844 if (parent == sb->s_root) 2844 if (IS_ROOT(parent))
2845 break; 2845 break;
2846 2846
2847 parent = parent->d_parent; 2847 parent = parent->d_parent;
@@ -2880,6 +2880,12 @@ int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
2880 goto end_no_trans; 2880 goto end_no_trans;
2881 } 2881 }
2882 2882
2883 if (root != BTRFS_I(inode)->root ||
2884 btrfs_root_refs(&root->root_item) == 0) {
2885 ret = 1;
2886 goto end_no_trans;
2887 }
2888
2883 ret = check_parent_dirs_for_sync(trans, inode, parent, 2889 ret = check_parent_dirs_for_sync(trans, inode, parent,
2884 sb, last_committed); 2890 sb, last_committed);
2885 if (ret) 2891 if (ret)
@@ -2907,12 +2913,15 @@ int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
2907 break; 2913 break;
2908 2914
2909 inode = parent->d_inode; 2915 inode = parent->d_inode;
2916 if (root != BTRFS_I(inode)->root)
2917 break;
2918
2910 if (BTRFS_I(inode)->generation > 2919 if (BTRFS_I(inode)->generation >
2911 root->fs_info->last_trans_committed) { 2920 root->fs_info->last_trans_committed) {
2912 ret = btrfs_log_inode(trans, root, inode, inode_only); 2921 ret = btrfs_log_inode(trans, root, inode, inode_only);
2913 BUG_ON(ret); 2922 BUG_ON(ret);
2914 } 2923 }
2915 if (parent == sb->s_root) 2924 if (IS_ROOT(parent))
2916 break; 2925 break;
2917 2926
2918 parent = parent->d_parent; 2927 parent = parent->d_parent;
@@ -2951,7 +2960,6 @@ int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
2951 struct btrfs_key tmp_key; 2960 struct btrfs_key tmp_key;
2952 struct btrfs_root *log; 2961 struct btrfs_root *log;
2953 struct btrfs_fs_info *fs_info = log_root_tree->fs_info; 2962 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
2954 u64 highest_inode;
2955 struct walk_control wc = { 2963 struct walk_control wc = {
2956 .process_func = process_one_buffer, 2964 .process_func = process_one_buffer,
2957 .stage = 0, 2965 .stage = 0,
@@ -3010,11 +3018,6 @@ again:
3010 path); 3018 path);
3011 BUG_ON(ret); 3019 BUG_ON(ret);
3012 } 3020 }
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 3021
3019 key.offset = found_key.offset - 1; 3022 key.offset = found_key.offset - 1;
3020 wc.replay_dest->log_root = NULL; 3023 wc.replay_dest->log_root = NULL;
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 5cf405b0828d..23e7d36ff325 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -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
@@ -719,10 +719,9 @@ error:
719 * called very infrequently and that a given device has a small number 719 * called very infrequently and that a given device has a small number
720 * of extents 720 * of extents
721 */ 721 */
722static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans, 722int find_free_dev_extent(struct btrfs_trans_handle *trans,
723 struct btrfs_device *device, 723 struct btrfs_device *device, u64 num_bytes,
724 u64 num_bytes, u64 *start, 724 u64 *start, u64 *max_avail)
725 u64 *max_avail)
726{ 725{
727 struct btrfs_key key; 726 struct btrfs_key key;
728 struct btrfs_root *root = device->dev_root; 727 struct btrfs_root *root = device->dev_root;
@@ -1736,6 +1735,10 @@ static int btrfs_relocate_chunk(struct btrfs_root *root,
1736 extent_root = root->fs_info->extent_root; 1735 extent_root = root->fs_info->extent_root;
1737 em_tree = &root->fs_info->mapping_tree.map_tree; 1736 em_tree = &root->fs_info->mapping_tree.map_tree;
1738 1737
1738 ret = btrfs_can_relocate(extent_root, chunk_offset);
1739 if (ret)
1740 return -ENOSPC;
1741
1739 /* step one, relocate all the extents inside this chunk */ 1742 /* step one, relocate all the extents inside this chunk */
1740 ret = btrfs_relocate_block_group(extent_root, chunk_offset); 1743 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1741 BUG_ON(ret); 1744 BUG_ON(ret);
@@ -1749,9 +1752,9 @@ static int btrfs_relocate_chunk(struct btrfs_root *root,
1749 * step two, delete the device extents and the 1752 * step two, delete the device extents and the
1750 * chunk tree entries 1753 * chunk tree entries
1751 */ 1754 */
1752 spin_lock(&em_tree->lock); 1755 read_lock(&em_tree->lock);
1753 em = lookup_extent_mapping(em_tree, chunk_offset, 1); 1756 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1754 spin_unlock(&em_tree->lock); 1757 read_unlock(&em_tree->lock);
1755 1758
1756 BUG_ON(em->start > chunk_offset || 1759 BUG_ON(em->start > chunk_offset ||
1757 em->start + em->len < chunk_offset); 1760 em->start + em->len < chunk_offset);
@@ -1780,9 +1783,9 @@ static int btrfs_relocate_chunk(struct btrfs_root *root,
1780 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset); 1783 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1781 BUG_ON(ret); 1784 BUG_ON(ret);
1782 1785
1783 spin_lock(&em_tree->lock); 1786 write_lock(&em_tree->lock);
1784 remove_extent_mapping(em_tree, em); 1787 remove_extent_mapping(em_tree, em);
1785 spin_unlock(&em_tree->lock); 1788 write_unlock(&em_tree->lock);
1786 1789
1787 kfree(map); 1790 kfree(map);
1788 em->bdev = NULL; 1791 em->bdev = NULL;
@@ -1807,12 +1810,15 @@ static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1807 struct btrfs_key found_key; 1810 struct btrfs_key found_key;
1808 u64 chunk_tree = chunk_root->root_key.objectid; 1811 u64 chunk_tree = chunk_root->root_key.objectid;
1809 u64 chunk_type; 1812 u64 chunk_type;
1813 bool retried = false;
1814 int failed = 0;
1810 int ret; 1815 int ret;
1811 1816
1812 path = btrfs_alloc_path(); 1817 path = btrfs_alloc_path();
1813 if (!path) 1818 if (!path)
1814 return -ENOMEM; 1819 return -ENOMEM;
1815 1820
1821again:
1816 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 1822 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1817 key.offset = (u64)-1; 1823 key.offset = (u64)-1;
1818 key.type = BTRFS_CHUNK_ITEM_KEY; 1824 key.type = BTRFS_CHUNK_ITEM_KEY;
@@ -1842,7 +1848,10 @@ static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1842 ret = btrfs_relocate_chunk(chunk_root, chunk_tree, 1848 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1843 found_key.objectid, 1849 found_key.objectid,
1844 found_key.offset); 1850 found_key.offset);
1845 BUG_ON(ret); 1851 if (ret == -ENOSPC)
1852 failed++;
1853 else if (ret)
1854 BUG();
1846 } 1855 }
1847 1856
1848 if (found_key.offset == 0) 1857 if (found_key.offset == 0)
@@ -1850,6 +1859,14 @@ static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1850 key.offset = found_key.offset - 1; 1859 key.offset = found_key.offset - 1;
1851 } 1860 }
1852 ret = 0; 1861 ret = 0;
1862 if (failed && !retried) {
1863 failed = 0;
1864 retried = true;
1865 goto again;
1866 } else if (failed && retried) {
1867 WARN_ON(1);
1868 ret = -ENOSPC;
1869 }
1853error: 1870error:
1854 btrfs_free_path(path); 1871 btrfs_free_path(path);
1855 return ret; 1872 return ret;
@@ -1894,6 +1911,8 @@ int btrfs_balance(struct btrfs_root *dev_root)
1894 continue; 1911 continue;
1895 1912
1896 ret = btrfs_shrink_device(device, old_size - size_to_free); 1913 ret = btrfs_shrink_device(device, old_size - size_to_free);
1914 if (ret == -ENOSPC)
1915 break;
1897 BUG_ON(ret); 1916 BUG_ON(ret);
1898 1917
1899 trans = btrfs_start_transaction(dev_root, 1); 1918 trans = btrfs_start_transaction(dev_root, 1);
@@ -1938,9 +1957,8 @@ int btrfs_balance(struct btrfs_root *dev_root)
1938 chunk = btrfs_item_ptr(path->nodes[0], 1957 chunk = btrfs_item_ptr(path->nodes[0],
1939 path->slots[0], 1958 path->slots[0],
1940 struct btrfs_chunk); 1959 struct btrfs_chunk);
1941 key.offset = found_key.offset;
1942 /* chunk zero is special */ 1960 /* chunk zero is special */
1943 if (key.offset == 0) 1961 if (found_key.offset == 0)
1944 break; 1962 break;
1945 1963
1946 btrfs_release_path(chunk_root, path); 1964 btrfs_release_path(chunk_root, path);
@@ -1948,7 +1966,8 @@ int btrfs_balance(struct btrfs_root *dev_root)
1948 chunk_root->root_key.objectid, 1966 chunk_root->root_key.objectid,
1949 found_key.objectid, 1967 found_key.objectid,
1950 found_key.offset); 1968 found_key.offset);
1951 BUG_ON(ret); 1969 BUG_ON(ret && ret != -ENOSPC);
1970 key.offset = found_key.offset - 1;
1952 } 1971 }
1953 ret = 0; 1972 ret = 0;
1954error: 1973error:
@@ -1974,10 +1993,13 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1974 u64 chunk_offset; 1993 u64 chunk_offset;
1975 int ret; 1994 int ret;
1976 int slot; 1995 int slot;
1996 int failed = 0;
1997 bool retried = false;
1977 struct extent_buffer *l; 1998 struct extent_buffer *l;
1978 struct btrfs_key key; 1999 struct btrfs_key key;
1979 struct btrfs_super_block *super_copy = &root->fs_info->super_copy; 2000 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1980 u64 old_total = btrfs_super_total_bytes(super_copy); 2001 u64 old_total = btrfs_super_total_bytes(super_copy);
2002 u64 old_size = device->total_bytes;
1981 u64 diff = device->total_bytes - new_size; 2003 u64 diff = device->total_bytes - new_size;
1982 2004
1983 if (new_size >= device->total_bytes) 2005 if (new_size >= device->total_bytes)
@@ -1987,12 +2009,6 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1987 if (!path) 2009 if (!path)
1988 return -ENOMEM; 2010 return -ENOMEM;
1989 2011
1990 trans = btrfs_start_transaction(root, 1);
1991 if (!trans) {
1992 ret = -ENOMEM;
1993 goto done;
1994 }
1995
1996 path->reada = 2; 2012 path->reada = 2;
1997 2013
1998 lock_chunks(root); 2014 lock_chunks(root);
@@ -2001,8 +2017,8 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2001 if (device->writeable) 2017 if (device->writeable)
2002 device->fs_devices->total_rw_bytes -= diff; 2018 device->fs_devices->total_rw_bytes -= diff;
2003 unlock_chunks(root); 2019 unlock_chunks(root);
2004 btrfs_end_transaction(trans, root);
2005 2020
2021again:
2006 key.objectid = device->devid; 2022 key.objectid = device->devid;
2007 key.offset = (u64)-1; 2023 key.offset = (u64)-1;
2008 key.type = BTRFS_DEV_EXTENT_KEY; 2024 key.type = BTRFS_DEV_EXTENT_KEY;
@@ -2017,6 +2033,7 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2017 goto done; 2033 goto done;
2018 if (ret) { 2034 if (ret) {
2019 ret = 0; 2035 ret = 0;
2036 btrfs_release_path(root, path);
2020 break; 2037 break;
2021 } 2038 }
2022 2039
@@ -2024,14 +2041,18 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2024 slot = path->slots[0]; 2041 slot = path->slots[0];
2025 btrfs_item_key_to_cpu(l, &key, path->slots[0]); 2042 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2026 2043
2027 if (key.objectid != device->devid) 2044 if (key.objectid != device->devid) {
2045 btrfs_release_path(root, path);
2028 break; 2046 break;
2047 }
2029 2048
2030 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 2049 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2031 length = btrfs_dev_extent_length(l, dev_extent); 2050 length = btrfs_dev_extent_length(l, dev_extent);
2032 2051
2033 if (key.offset + length <= new_size) 2052 if (key.offset + length <= new_size) {
2053 btrfs_release_path(root, path);
2034 break; 2054 break;
2055 }
2035 2056
2036 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); 2057 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2037 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); 2058 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
@@ -2040,8 +2061,26 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2040 2061
2041 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid, 2062 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2042 chunk_offset); 2063 chunk_offset);
2043 if (ret) 2064 if (ret && ret != -ENOSPC)
2044 goto done; 2065 goto done;
2066 if (ret == -ENOSPC)
2067 failed++;
2068 key.offset -= 1;
2069 }
2070
2071 if (failed && !retried) {
2072 failed = 0;
2073 retried = true;
2074 goto again;
2075 } else if (failed && retried) {
2076 ret = -ENOSPC;
2077 lock_chunks(root);
2078
2079 device->total_bytes = old_size;
2080 if (device->writeable)
2081 device->fs_devices->total_rw_bytes += diff;
2082 unlock_chunks(root);
2083 goto done;
2045 } 2084 }
2046 2085
2047 /* Shrinking succeeded, else we would be at "done". */ 2086 /* Shrinking succeeded, else we would be at "done". */
@@ -2294,9 +2333,9 @@ again:
2294 em->block_len = em->len; 2333 em->block_len = em->len;
2295 2334
2296 em_tree = &extent_root->fs_info->mapping_tree.map_tree; 2335 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2297 spin_lock(&em_tree->lock); 2336 write_lock(&em_tree->lock);
2298 ret = add_extent_mapping(em_tree, em); 2337 ret = add_extent_mapping(em_tree, em);
2299 spin_unlock(&em_tree->lock); 2338 write_unlock(&em_tree->lock);
2300 BUG_ON(ret); 2339 BUG_ON(ret);
2301 free_extent_map(em); 2340 free_extent_map(em);
2302 2341
@@ -2491,9 +2530,9 @@ int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2491 int readonly = 0; 2530 int readonly = 0;
2492 int i; 2531 int i;
2493 2532
2494 spin_lock(&map_tree->map_tree.lock); 2533 read_lock(&map_tree->map_tree.lock);
2495 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); 2534 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2496 spin_unlock(&map_tree->map_tree.lock); 2535 read_unlock(&map_tree->map_tree.lock);
2497 if (!em) 2536 if (!em)
2498 return 1; 2537 return 1;
2499 2538
@@ -2518,11 +2557,11 @@ void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2518 struct extent_map *em; 2557 struct extent_map *em;
2519 2558
2520 while (1) { 2559 while (1) {
2521 spin_lock(&tree->map_tree.lock); 2560 write_lock(&tree->map_tree.lock);
2522 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); 2561 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2523 if (em) 2562 if (em)
2524 remove_extent_mapping(&tree->map_tree, em); 2563 remove_extent_mapping(&tree->map_tree, em);
2525 spin_unlock(&tree->map_tree.lock); 2564 write_unlock(&tree->map_tree.lock);
2526 if (!em) 2565 if (!em)
2527 break; 2566 break;
2528 kfree(em->bdev); 2567 kfree(em->bdev);
@@ -2540,9 +2579,9 @@ 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; 2579 struct extent_map_tree *em_tree = &map_tree->map_tree;
2541 int ret; 2580 int ret;
2542 2581
2543 spin_lock(&em_tree->lock); 2582 read_lock(&em_tree->lock);
2544 em = lookup_extent_mapping(em_tree, logical, len); 2583 em = lookup_extent_mapping(em_tree, logical, len);
2545 spin_unlock(&em_tree->lock); 2584 read_unlock(&em_tree->lock);
2546 BUG_ON(!em); 2585 BUG_ON(!em);
2547 2586
2548 BUG_ON(em->start > logical || em->start + em->len < logical); 2587 BUG_ON(em->start > logical || em->start + em->len < logical);
@@ -2604,9 +2643,9 @@ again:
2604 atomic_set(&multi->error, 0); 2643 atomic_set(&multi->error, 0);
2605 } 2644 }
2606 2645
2607 spin_lock(&em_tree->lock); 2646 read_lock(&em_tree->lock);
2608 em = lookup_extent_mapping(em_tree, logical, *length); 2647 em = lookup_extent_mapping(em_tree, logical, *length);
2609 spin_unlock(&em_tree->lock); 2648 read_unlock(&em_tree->lock);
2610 2649
2611 if (!em && unplug_page) 2650 if (!em && unplug_page)
2612 return 0; 2651 return 0;
@@ -2763,9 +2802,9 @@ int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2763 u64 stripe_nr; 2802 u64 stripe_nr;
2764 int i, j, nr = 0; 2803 int i, j, nr = 0;
2765 2804
2766 spin_lock(&em_tree->lock); 2805 read_lock(&em_tree->lock);
2767 em = lookup_extent_mapping(em_tree, chunk_start, 1); 2806 em = lookup_extent_mapping(em_tree, chunk_start, 1);
2768 spin_unlock(&em_tree->lock); 2807 read_unlock(&em_tree->lock);
2769 2808
2770 BUG_ON(!em || em->start != chunk_start); 2809 BUG_ON(!em || em->start != chunk_start);
2771 map = (struct map_lookup *)em->bdev; 2810 map = (struct map_lookup *)em->bdev;
@@ -3053,9 +3092,9 @@ static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3053 logical = key->offset; 3092 logical = key->offset;
3054 length = btrfs_chunk_length(leaf, chunk); 3093 length = btrfs_chunk_length(leaf, chunk);
3055 3094
3056 spin_lock(&map_tree->map_tree.lock); 3095 read_lock(&map_tree->map_tree.lock);
3057 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); 3096 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3058 spin_unlock(&map_tree->map_tree.lock); 3097 read_unlock(&map_tree->map_tree.lock);
3059 3098
3060 /* already mapped? */ 3099 /* already mapped? */
3061 if (em && em->start <= logical && em->start + em->len > logical) { 3100 if (em && em->start <= logical && em->start + em->len > logical) {
@@ -3114,9 +3153,9 @@ static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3114 map->stripes[i].dev->in_fs_metadata = 1; 3153 map->stripes[i].dev->in_fs_metadata = 1;
3115 } 3154 }
3116 3155
3117 spin_lock(&map_tree->map_tree.lock); 3156 write_lock(&map_tree->map_tree.lock);
3118 ret = add_extent_mapping(&map_tree->map_tree, em); 3157 ret = add_extent_mapping(&map_tree->map_tree, em);
3119 spin_unlock(&map_tree->map_tree.lock); 3158 write_unlock(&map_tree->map_tree.lock);
3120 BUG_ON(ret); 3159 BUG_ON(ret);
3121 free_extent_map(em); 3160 free_extent_map(em);
3122 3161
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
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-01 10:45:05 -0400