aboutsummaryrefslogtreecommitdiffstats
path: root/fs/btrfs
diff options
context:
space:
mode:
Diffstat (limited to 'fs/btrfs')
-rw-r--r--fs/btrfs/acl.c6
-rw-r--r--fs/btrfs/async-thread.c81
-rw-r--r--fs/btrfs/async-thread.h10
-rw-r--r--fs/btrfs/btrfs_inode.h18
-rw-r--r--fs/btrfs/ctree.h19
-rw-r--r--fs/btrfs/disk-io.c50
-rw-r--r--fs/btrfs/extent-tree.c365
-rw-r--r--fs/btrfs/extent_io.c42
-rw-r--r--fs/btrfs/extent_io.h18
-rw-r--r--fs/btrfs/extent_map.c4
-rw-r--r--fs/btrfs/file.c48
-rw-r--r--fs/btrfs/free-space-cache.c2
-rw-r--r--fs/btrfs/inode.c239
-rw-r--r--fs/btrfs/ioctl.c7
-rw-r--r--fs/btrfs/ordered-data.c6
-rw-r--r--fs/btrfs/relocation.c4
-rw-r--r--fs/btrfs/root-tree.c2
-rw-r--r--fs/btrfs/super.c9
-rw-r--r--fs/btrfs/transaction.c64
-rw-r--r--fs/btrfs/transaction.h5
-rw-r--r--fs/btrfs/tree-log.c56
-rw-r--r--fs/btrfs/tree-log.h3
-rw-r--r--fs/btrfs/xattr.c2
23 files changed, 810 insertions, 250 deletions
diff --git a/fs/btrfs/acl.c b/fs/btrfs/acl.c
index 69b355ae7f49..361604244271 100644
--- a/fs/btrfs/acl.c
+++ b/fs/btrfs/acl.c
@@ -27,7 +27,7 @@
27#include "btrfs_inode.h" 27#include "btrfs_inode.h"
28#include "xattr.h" 28#include "xattr.h"
29 29
30#ifdef CONFIG_BTRFS_POSIX_ACL 30#ifdef CONFIG_BTRFS_FS_POSIX_ACL
31 31
32static struct posix_acl *btrfs_get_acl(struct inode *inode, int type) 32static struct posix_acl *btrfs_get_acl(struct inode *inode, int type)
33{ 33{
@@ -313,7 +313,7 @@ struct xattr_handler btrfs_xattr_acl_access_handler = {
313 .set = btrfs_xattr_acl_access_set, 313 .set = btrfs_xattr_acl_access_set,
314}; 314};
315 315
316#else /* CONFIG_BTRFS_POSIX_ACL */ 316#else /* CONFIG_BTRFS_FS_POSIX_ACL */
317 317
318int btrfs_acl_chmod(struct inode *inode) 318int btrfs_acl_chmod(struct inode *inode)
319{ 319{
@@ -325,4 +325,4 @@ int btrfs_init_acl(struct inode *inode, struct inode *dir)
325 return 0; 325 return 0;
326} 326}
327 327
328#endif /* CONFIG_BTRFS_POSIX_ACL */ 328#endif /* CONFIG_BTRFS_FS_POSIX_ACL */
diff --git a/fs/btrfs/async-thread.c b/fs/btrfs/async-thread.c
index 282ca085c2fb..c0861e781cdb 100644
--- a/fs/btrfs/async-thread.c
+++ b/fs/btrfs/async-thread.c
@@ -64,6 +64,51 @@ struct btrfs_worker_thread {
64}; 64};
65 65
66/* 66/*
67 * btrfs_start_workers uses kthread_run, which can block waiting for memory
68 * for a very long time. It will actually throttle on page writeback,
69 * and so it may not make progress until after our btrfs worker threads
70 * process all of the pending work structs in their queue
71 *
72 * This means we can't use btrfs_start_workers from inside a btrfs worker
73 * thread that is used as part of cleaning dirty memory, which pretty much
74 * involves all of the worker threads.
75 *
76 * Instead we have a helper queue who never has more than one thread
77 * where we scheduler thread start operations. This worker_start struct
78 * is used to contain the work and hold a pointer to the queue that needs
79 * another worker.
80 */
81struct worker_start {
82 struct btrfs_work work;
83 struct btrfs_workers *queue;
84};
85
86static void start_new_worker_func(struct btrfs_work *work)
87{
88 struct worker_start *start;
89 start = container_of(work, struct worker_start, work);
90 btrfs_start_workers(start->queue, 1);
91 kfree(start);
92}
93
94static int start_new_worker(struct btrfs_workers *queue)
95{
96 struct worker_start *start;
97 int ret;
98
99 start = kzalloc(sizeof(*start), GFP_NOFS);
100 if (!start)
101 return -ENOMEM;
102
103 start->work.func = start_new_worker_func;
104 start->queue = queue;
105 ret = btrfs_queue_worker(queue->atomic_worker_start, &start->work);
106 if (ret)
107 kfree(start);
108 return ret;
109}
110
111/*
67 * helper function to move a thread onto the idle list after it 112 * helper function to move a thread onto the idle list after it
68 * has finished some requests. 113 * has finished some requests.
69 */ 114 */
@@ -118,11 +163,13 @@ static void check_pending_worker_creates(struct btrfs_worker_thread *worker)
118 goto out; 163 goto out;
119 164
120 workers->atomic_start_pending = 0; 165 workers->atomic_start_pending = 0;
121 if (workers->num_workers >= workers->max_workers) 166 if (workers->num_workers + workers->num_workers_starting >=
167 workers->max_workers)
122 goto out; 168 goto out;
123 169
170 workers->num_workers_starting += 1;
124 spin_unlock_irqrestore(&workers->lock, flags); 171 spin_unlock_irqrestore(&workers->lock, flags);
125 btrfs_start_workers(workers, 1); 172 start_new_worker(workers);
126 return; 173 return;
127 174
128out: 175out:
@@ -390,9 +437,11 @@ int btrfs_stop_workers(struct btrfs_workers *workers)
390/* 437/*
391 * simple init on struct btrfs_workers 438 * simple init on struct btrfs_workers
392 */ 439 */
393void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max) 440void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max,
441 struct btrfs_workers *async_helper)
394{ 442{
395 workers->num_workers = 0; 443 workers->num_workers = 0;
444 workers->num_workers_starting = 0;
396 INIT_LIST_HEAD(&workers->worker_list); 445 INIT_LIST_HEAD(&workers->worker_list);
397 INIT_LIST_HEAD(&workers->idle_list); 446 INIT_LIST_HEAD(&workers->idle_list);
398 INIT_LIST_HEAD(&workers->order_list); 447 INIT_LIST_HEAD(&workers->order_list);
@@ -404,14 +453,15 @@ void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max)
404 workers->name = name; 453 workers->name = name;
405 workers->ordered = 0; 454 workers->ordered = 0;
406 workers->atomic_start_pending = 0; 455 workers->atomic_start_pending = 0;
407 workers->atomic_worker_start = 0; 456 workers->atomic_worker_start = async_helper;
408} 457}
409 458
410/* 459/*
411 * starts new worker threads. This does not enforce the max worker 460 * starts new worker threads. This does not enforce the max worker
412 * count in case you need to temporarily go past it. 461 * count in case you need to temporarily go past it.
413 */ 462 */
414int btrfs_start_workers(struct btrfs_workers *workers, int num_workers) 463static int __btrfs_start_workers(struct btrfs_workers *workers,
464 int num_workers)
415{ 465{
416 struct btrfs_worker_thread *worker; 466 struct btrfs_worker_thread *worker;
417 int ret = 0; 467 int ret = 0;
@@ -444,6 +494,8 @@ int btrfs_start_workers(struct btrfs_workers *workers, int num_workers)
444 list_add_tail(&worker->worker_list, &workers->idle_list); 494 list_add_tail(&worker->worker_list, &workers->idle_list);
445 worker->idle = 1; 495 worker->idle = 1;
446 workers->num_workers++; 496 workers->num_workers++;
497 workers->num_workers_starting--;
498 WARN_ON(workers->num_workers_starting < 0);
447 spin_unlock_irq(&workers->lock); 499 spin_unlock_irq(&workers->lock);
448 } 500 }
449 return 0; 501 return 0;
@@ -452,6 +504,14 @@ fail:
452 return ret; 504 return ret;
453} 505}
454 506
507int btrfs_start_workers(struct btrfs_workers *workers, int num_workers)
508{
509 spin_lock_irq(&workers->lock);
510 workers->num_workers_starting += num_workers;
511 spin_unlock_irq(&workers->lock);
512 return __btrfs_start_workers(workers, num_workers);
513}
514
455/* 515/*
456 * run through the list and find a worker thread that doesn't have a lot 516 * run through the list and find a worker thread that doesn't have a lot
457 * to do right now. This can return null if we aren't yet at the thread 517 * to do right now. This can return null if we aren't yet at the thread
@@ -461,7 +521,10 @@ static struct btrfs_worker_thread *next_worker(struct btrfs_workers *workers)
461{ 521{
462 struct btrfs_worker_thread *worker; 522 struct btrfs_worker_thread *worker;
463 struct list_head *next; 523 struct list_head *next;
464 int enforce_min = workers->num_workers < workers->max_workers; 524 int enforce_min;
525
526 enforce_min = (workers->num_workers + workers->num_workers_starting) <
527 workers->max_workers;
465 528
466 /* 529 /*
467 * if we find an idle thread, don't move it to the end of the 530 * if we find an idle thread, don't move it to the end of the
@@ -509,15 +572,17 @@ again:
509 worker = next_worker(workers); 572 worker = next_worker(workers);
510 573
511 if (!worker) { 574 if (!worker) {
512 if (workers->num_workers >= workers->max_workers) { 575 if (workers->num_workers + workers->num_workers_starting >=
576 workers->max_workers) {
513 goto fallback; 577 goto fallback;
514 } else if (workers->atomic_worker_start) { 578 } else if (workers->atomic_worker_start) {
515 workers->atomic_start_pending = 1; 579 workers->atomic_start_pending = 1;
516 goto fallback; 580 goto fallback;
517 } else { 581 } else {
582 workers->num_workers_starting++;
518 spin_unlock_irqrestore(&workers->lock, flags); 583 spin_unlock_irqrestore(&workers->lock, flags);
519 /* we're below the limit, start another worker */ 584 /* we're below the limit, start another worker */
520 btrfs_start_workers(workers, 1); 585 __btrfs_start_workers(workers, 1);
521 goto again; 586 goto again;
522 } 587 }
523 } 588 }
diff --git a/fs/btrfs/async-thread.h b/fs/btrfs/async-thread.h
index fc089b95ec14..5077746cf85e 100644
--- a/fs/btrfs/async-thread.h
+++ b/fs/btrfs/async-thread.h
@@ -64,6 +64,8 @@ struct btrfs_workers {
64 /* current number of running workers */ 64 /* current number of running workers */
65 int num_workers; 65 int num_workers;
66 66
67 int num_workers_starting;
68
67 /* max number of workers allowed. changed by btrfs_start_workers */ 69 /* max number of workers allowed. changed by btrfs_start_workers */
68 int max_workers; 70 int max_workers;
69 71
@@ -78,9 +80,10 @@ struct btrfs_workers {
78 80
79 /* 81 /*
80 * are we allowed to sleep while starting workers or are we required 82 * are we allowed to sleep while starting workers or are we required
81 * to start them at a later time? 83 * to start them at a later time? If we can't sleep, this indicates
84 * which queue we need to use to schedule thread creation.
82 */ 85 */
83 int atomic_worker_start; 86 struct btrfs_workers *atomic_worker_start;
84 87
85 /* list with all the work threads. The workers on the idle thread 88 /* list with all the work threads. The workers on the idle thread
86 * may be actively servicing jobs, but they haven't yet hit the 89 * may be actively servicing jobs, but they haven't yet hit the
@@ -109,7 +112,8 @@ struct btrfs_workers {
109int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work); 112int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work);
110int btrfs_start_workers(struct btrfs_workers *workers, int num_workers); 113int btrfs_start_workers(struct btrfs_workers *workers, int num_workers);
111int btrfs_stop_workers(struct btrfs_workers *workers); 114int btrfs_stop_workers(struct btrfs_workers *workers);
112void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max); 115void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max,
116 struct btrfs_workers *async_starter);
113int btrfs_requeue_work(struct btrfs_work *work); 117int btrfs_requeue_work(struct btrfs_work *work);
114void btrfs_set_work_high_prio(struct btrfs_work *work); 118void btrfs_set_work_high_prio(struct btrfs_work *work);
115#endif 119#endif
diff --git a/fs/btrfs/btrfs_inode.h b/fs/btrfs/btrfs_inode.h
index a54d354cefcb..f6783a42f010 100644
--- a/fs/btrfs/btrfs_inode.h
+++ b/fs/btrfs/btrfs_inode.h
@@ -86,6 +86,12 @@ struct btrfs_inode {
86 * transid of the trans_handle that last modified this inode 86 * transid of the trans_handle that last modified this inode
87 */ 87 */
88 u64 last_trans; 88 u64 last_trans;
89
90 /*
91 * log transid when this inode was last modified
92 */
93 u64 last_sub_trans;
94
89 /* 95 /*
90 * transid that last logged this inode 96 * transid that last logged this inode
91 */ 97 */
@@ -128,12 +134,14 @@ struct btrfs_inode {
128 u64 last_unlink_trans; 134 u64 last_unlink_trans;
129 135
130 /* 136 /*
131 * These two counters are for delalloc metadata reservations. We keep 137 * Counters to keep track of the number of extent item's we may use due
132 * track of how many extents we've accounted for vs how many extents we 138 * to delalloc and such. outstanding_extents is the number of extent
133 * have. 139 * items we think we'll end up using, and reserved_extents is the number
140 * of extent items we've reserved metadata for.
134 */ 141 */
135 int delalloc_reserved_extents; 142 spinlock_t accounting_lock;
136 int delalloc_extents; 143 int reserved_extents;
144 int outstanding_extents;
137 145
138 /* 146 /*
139 * ordered_data_close is set by truncate when a file that used 147 * ordered_data_close is set by truncate when a file that used
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index dd8ced9814c4..444b3e9b92a4 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -691,14 +691,17 @@ struct btrfs_space_info {
691 691
692 struct list_head list; 692 struct list_head list;
693 693
694 /* for controlling how we free up space for allocations */
695 wait_queue_head_t allocate_wait;
696 wait_queue_head_t flush_wait;
697 int allocating_chunk;
698 int flushing;
699
694 /* for block groups in our same type */ 700 /* for block groups in our same type */
695 struct list_head block_groups; 701 struct list_head block_groups;
696 spinlock_t lock; 702 spinlock_t lock;
697 struct rw_semaphore groups_sem; 703 struct rw_semaphore groups_sem;
698 atomic_t caching_threads; 704 atomic_t caching_threads;
699
700 int allocating_chunk;
701 wait_queue_head_t wait;
702}; 705};
703 706
704/* 707/*
@@ -907,6 +910,7 @@ struct btrfs_fs_info {
907 * A third pool does submit_bio to avoid deadlocking with the other 910 * A third pool does submit_bio to avoid deadlocking with the other
908 * two 911 * two
909 */ 912 */
913 struct btrfs_workers generic_worker;
910 struct btrfs_workers workers; 914 struct btrfs_workers workers;
911 struct btrfs_workers delalloc_workers; 915 struct btrfs_workers delalloc_workers;
912 struct btrfs_workers endio_workers; 916 struct btrfs_workers endio_workers;
@@ -914,6 +918,7 @@ struct btrfs_fs_info {
914 struct btrfs_workers endio_meta_write_workers; 918 struct btrfs_workers endio_meta_write_workers;
915 struct btrfs_workers endio_write_workers; 919 struct btrfs_workers endio_write_workers;
916 struct btrfs_workers submit_workers; 920 struct btrfs_workers submit_workers;
921 struct btrfs_workers enospc_workers;
917 /* 922 /*
918 * fixup workers take dirty pages that didn't properly go through 923 * fixup workers take dirty pages that didn't properly go through
919 * the cow mechanism and make them safe to write. It happens 924 * the cow mechanism and make them safe to write. It happens
@@ -1004,7 +1009,10 @@ struct btrfs_root {
1004 atomic_t log_writers; 1009 atomic_t log_writers;
1005 atomic_t log_commit[2]; 1010 atomic_t log_commit[2];
1006 unsigned long log_transid; 1011 unsigned long log_transid;
1012 unsigned long last_log_commit;
1007 unsigned long log_batch; 1013 unsigned long log_batch;
1014 pid_t log_start_pid;
1015 bool log_multiple_pids;
1008 1016
1009 u64 objectid; 1017 u64 objectid;
1010 u64 last_trans; 1018 u64 last_trans;
@@ -1145,6 +1153,7 @@ struct btrfs_root {
1145#define BTRFS_MOUNT_FLUSHONCOMMIT (1 << 7) 1153#define BTRFS_MOUNT_FLUSHONCOMMIT (1 << 7)
1146#define BTRFS_MOUNT_SSD_SPREAD (1 << 8) 1154#define BTRFS_MOUNT_SSD_SPREAD (1 << 8)
1147#define BTRFS_MOUNT_NOSSD (1 << 9) 1155#define BTRFS_MOUNT_NOSSD (1 << 9)
1156#define BTRFS_MOUNT_DISCARD (1 << 10)
1148 1157
1149#define btrfs_clear_opt(o, opt) ((o) &= ~BTRFS_MOUNT_##opt) 1158#define btrfs_clear_opt(o, opt) ((o) &= ~BTRFS_MOUNT_##opt)
1150#define btrfs_set_opt(o, opt) ((o) |= BTRFS_MOUNT_##opt) 1159#define btrfs_set_opt(o, opt) ((o) |= BTRFS_MOUNT_##opt)
@@ -2323,7 +2332,7 @@ int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode);
2323void btrfs_orphan_cleanup(struct btrfs_root *root); 2332void btrfs_orphan_cleanup(struct btrfs_root *root);
2324int btrfs_cont_expand(struct inode *inode, loff_t size); 2333int btrfs_cont_expand(struct inode *inode, loff_t size);
2325int btrfs_invalidate_inodes(struct btrfs_root *root); 2334int btrfs_invalidate_inodes(struct btrfs_root *root);
2326extern struct dentry_operations btrfs_dentry_operations; 2335extern const struct dentry_operations btrfs_dentry_operations;
2327 2336
2328/* ioctl.c */ 2337/* ioctl.c */
2329long btrfs_ioctl(struct file *file, unsigned int cmd, unsigned long arg); 2338long btrfs_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
@@ -2366,7 +2375,7 @@ int btrfs_parse_options(struct btrfs_root *root, char *options);
2366int btrfs_sync_fs(struct super_block *sb, int wait); 2375int btrfs_sync_fs(struct super_block *sb, int wait);
2367 2376
2368/* acl.c */ 2377/* acl.c */
2369#ifdef CONFIG_BTRFS_POSIX_ACL 2378#ifdef CONFIG_BTRFS_FS_POSIX_ACL
2370int btrfs_check_acl(struct inode *inode, int mask); 2379int btrfs_check_acl(struct inode *inode, int mask);
2371#else 2380#else
2372#define btrfs_check_acl NULL 2381#define btrfs_check_acl NULL
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index af0435f79fa6..02b6afbd7450 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -917,6 +917,7 @@ static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
917 atomic_set(&root->log_writers, 0); 917 atomic_set(&root->log_writers, 0);
918 root->log_batch = 0; 918 root->log_batch = 0;
919 root->log_transid = 0; 919 root->log_transid = 0;
920 root->last_log_commit = 0;
920 extent_io_tree_init(&root->dirty_log_pages, 921 extent_io_tree_init(&root->dirty_log_pages,
921 fs_info->btree_inode->i_mapping, GFP_NOFS); 922 fs_info->btree_inode->i_mapping, GFP_NOFS);
922 923
@@ -1087,6 +1088,7 @@ int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1087 WARN_ON(root->log_root); 1088 WARN_ON(root->log_root);
1088 root->log_root = log_root; 1089 root->log_root = log_root;
1089 root->log_transid = 0; 1090 root->log_transid = 0;
1091 root->last_log_commit = 0;
1090 return 0; 1092 return 0;
1091} 1093}
1092 1094
@@ -1746,21 +1748,25 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1746 err = -EINVAL; 1748 err = -EINVAL;
1747 goto fail_iput; 1749 goto fail_iput;
1748 } 1750 }
1749printk("thread pool is %d\n", fs_info->thread_pool_size); 1751
1750 /* 1752 btrfs_init_workers(&fs_info->generic_worker,
1751 * we need to start all the end_io workers up front because the 1753 "genwork", 1, NULL);
1752 * queue work function gets called at interrupt time, and so it 1754
1753 * cannot dynamically grow.
1754 */
1755 btrfs_init_workers(&fs_info->workers, "worker", 1755 btrfs_init_workers(&fs_info->workers, "worker",
1756 fs_info->thread_pool_size); 1756 fs_info->thread_pool_size,
1757 &fs_info->generic_worker);
1757 1758
1758 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc", 1759 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1759 fs_info->thread_pool_size); 1760 fs_info->thread_pool_size,
1761 &fs_info->generic_worker);
1760 1762
1761 btrfs_init_workers(&fs_info->submit_workers, "submit", 1763 btrfs_init_workers(&fs_info->submit_workers, "submit",
1762 min_t(u64, fs_devices->num_devices, 1764 min_t(u64, fs_devices->num_devices,
1763 fs_info->thread_pool_size)); 1765 fs_info->thread_pool_size),
1766 &fs_info->generic_worker);
1767 btrfs_init_workers(&fs_info->enospc_workers, "enospc",
1768 fs_info->thread_pool_size,
1769 &fs_info->generic_worker);
1764 1770
1765 /* a higher idle thresh on the submit workers makes it much more 1771 /* a higher idle thresh on the submit workers makes it much more
1766 * likely that bios will be send down in a sane order to the 1772 * likely that bios will be send down in a sane order to the
@@ -1774,15 +1780,20 @@ printk("thread pool is %d\n", fs_info->thread_pool_size);
1774 fs_info->delalloc_workers.idle_thresh = 2; 1780 fs_info->delalloc_workers.idle_thresh = 2;
1775 fs_info->delalloc_workers.ordered = 1; 1781 fs_info->delalloc_workers.ordered = 1;
1776 1782
1777 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1); 1783 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1784 &fs_info->generic_worker);
1778 btrfs_init_workers(&fs_info->endio_workers, "endio", 1785 btrfs_init_workers(&fs_info->endio_workers, "endio",
1779 fs_info->thread_pool_size); 1786 fs_info->thread_pool_size,
1787 &fs_info->generic_worker);
1780 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta", 1788 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1781 fs_info->thread_pool_size); 1789 fs_info->thread_pool_size,
1790 &fs_info->generic_worker);
1782 btrfs_init_workers(&fs_info->endio_meta_write_workers, 1791 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1783 "endio-meta-write", fs_info->thread_pool_size); 1792 "endio-meta-write", fs_info->thread_pool_size,
1793 &fs_info->generic_worker);
1784 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write", 1794 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1785 fs_info->thread_pool_size); 1795 fs_info->thread_pool_size,
1796 &fs_info->generic_worker);
1786 1797
1787 /* 1798 /*
1788 * endios are largely parallel and should have a very 1799 * endios are largely parallel and should have a very
@@ -1794,12 +1805,8 @@ printk("thread pool is %d\n", fs_info->thread_pool_size);
1794 fs_info->endio_write_workers.idle_thresh = 2; 1805 fs_info->endio_write_workers.idle_thresh = 2;
1795 fs_info->endio_meta_write_workers.idle_thresh = 2; 1806 fs_info->endio_meta_write_workers.idle_thresh = 2;
1796 1807
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;
1801
1802 btrfs_start_workers(&fs_info->workers, 1); 1808 btrfs_start_workers(&fs_info->workers, 1);
1809 btrfs_start_workers(&fs_info->generic_worker, 1);
1803 btrfs_start_workers(&fs_info->submit_workers, 1); 1810 btrfs_start_workers(&fs_info->submit_workers, 1);
1804 btrfs_start_workers(&fs_info->delalloc_workers, 1); 1811 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1805 btrfs_start_workers(&fs_info->fixup_workers, 1); 1812 btrfs_start_workers(&fs_info->fixup_workers, 1);
@@ -1807,6 +1814,7 @@ printk("thread pool is %d\n", fs_info->thread_pool_size);
1807 btrfs_start_workers(&fs_info->endio_meta_workers, 1); 1814 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1808 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1); 1815 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1809 btrfs_start_workers(&fs_info->endio_write_workers, 1); 1816 btrfs_start_workers(&fs_info->endio_write_workers, 1);
1817 btrfs_start_workers(&fs_info->enospc_workers, 1);
1810 1818
1811 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super); 1819 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1812 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages, 1820 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
@@ -2012,6 +2020,7 @@ fail_chunk_root:
2012 free_extent_buffer(chunk_root->node); 2020 free_extent_buffer(chunk_root->node);
2013 free_extent_buffer(chunk_root->commit_root); 2021 free_extent_buffer(chunk_root->commit_root);
2014fail_sb_buffer: 2022fail_sb_buffer:
2023 btrfs_stop_workers(&fs_info->generic_worker);
2015 btrfs_stop_workers(&fs_info->fixup_workers); 2024 btrfs_stop_workers(&fs_info->fixup_workers);
2016 btrfs_stop_workers(&fs_info->delalloc_workers); 2025 btrfs_stop_workers(&fs_info->delalloc_workers);
2017 btrfs_stop_workers(&fs_info->workers); 2026 btrfs_stop_workers(&fs_info->workers);
@@ -2020,6 +2029,7 @@ fail_sb_buffer:
2020 btrfs_stop_workers(&fs_info->endio_meta_write_workers); 2029 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2021 btrfs_stop_workers(&fs_info->endio_write_workers); 2030 btrfs_stop_workers(&fs_info->endio_write_workers);
2022 btrfs_stop_workers(&fs_info->submit_workers); 2031 btrfs_stop_workers(&fs_info->submit_workers);
2032 btrfs_stop_workers(&fs_info->enospc_workers);
2023fail_iput: 2033fail_iput:
2024 invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 2034 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2025 iput(fs_info->btree_inode); 2035 iput(fs_info->btree_inode);
@@ -2437,6 +2447,7 @@ int close_ctree(struct btrfs_root *root)
2437 2447
2438 iput(fs_info->btree_inode); 2448 iput(fs_info->btree_inode);
2439 2449
2450 btrfs_stop_workers(&fs_info->generic_worker);
2440 btrfs_stop_workers(&fs_info->fixup_workers); 2451 btrfs_stop_workers(&fs_info->fixup_workers);
2441 btrfs_stop_workers(&fs_info->delalloc_workers); 2452 btrfs_stop_workers(&fs_info->delalloc_workers);
2442 btrfs_stop_workers(&fs_info->workers); 2453 btrfs_stop_workers(&fs_info->workers);
@@ -2445,6 +2456,7 @@ int close_ctree(struct btrfs_root *root)
2445 btrfs_stop_workers(&fs_info->endio_meta_write_workers); 2456 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2446 btrfs_stop_workers(&fs_info->endio_write_workers); 2457 btrfs_stop_workers(&fs_info->endio_write_workers);
2447 btrfs_stop_workers(&fs_info->submit_workers); 2458 btrfs_stop_workers(&fs_info->submit_workers);
2459 btrfs_stop_workers(&fs_info->enospc_workers);
2448 2460
2449 btrfs_close_devices(fs_info->fs_devices); 2461 btrfs_close_devices(fs_info->fs_devices);
2450 btrfs_mapping_tree_free(&fs_info->mapping_tree); 2462 btrfs_mapping_tree_free(&fs_info->mapping_tree);
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 359a754c782c..94627c4cc193 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -1568,23 +1568,23 @@ static int remove_extent_backref(struct btrfs_trans_handle *trans,
1568 return ret; 1568 return ret;
1569} 1569}
1570 1570
1571#ifdef BIO_RW_DISCARD
1572static void btrfs_issue_discard(struct block_device *bdev, 1571static void btrfs_issue_discard(struct block_device *bdev,
1573 u64 start, u64 len) 1572 u64 start, u64 len)
1574{ 1573{
1575 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL, 1574 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
1576 DISCARD_FL_BARRIER); 1575 DISCARD_FL_BARRIER);
1577} 1576}
1578#endif
1579 1577
1580static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, 1578static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1581 u64 num_bytes) 1579 u64 num_bytes)
1582{ 1580{
1583#ifdef BIO_RW_DISCARD
1584 int ret; 1581 int ret;
1585 u64 map_length = num_bytes; 1582 u64 map_length = num_bytes;
1586 struct btrfs_multi_bio *multi = NULL; 1583 struct btrfs_multi_bio *multi = NULL;
1587 1584
1585 if (!btrfs_test_opt(root, DISCARD))
1586 return 0;
1587
1588 /* Tell the block device(s) that the sectors can be discarded */ 1588 /* Tell the block device(s) that the sectors can be discarded */
1589 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ, 1589 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1590 bytenr, &map_length, &multi, 0); 1590 bytenr, &map_length, &multi, 0);
@@ -1604,9 +1604,6 @@ static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1604 } 1604 }
1605 1605
1606 return ret; 1606 return ret;
1607#else
1608 return 0;
1609#endif
1610} 1607}
1611 1608
1612int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, 1609int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
@@ -2824,14 +2821,17 @@ int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
2824 num_items); 2821 num_items);
2825 2822
2826 spin_lock(&meta_sinfo->lock); 2823 spin_lock(&meta_sinfo->lock);
2827 if (BTRFS_I(inode)->delalloc_reserved_extents <= 2824 spin_lock(&BTRFS_I(inode)->accounting_lock);
2828 BTRFS_I(inode)->delalloc_extents) { 2825 if (BTRFS_I(inode)->reserved_extents <=
2826 BTRFS_I(inode)->outstanding_extents) {
2827 spin_unlock(&BTRFS_I(inode)->accounting_lock);
2829 spin_unlock(&meta_sinfo->lock); 2828 spin_unlock(&meta_sinfo->lock);
2830 return 0; 2829 return 0;
2831 } 2830 }
2831 spin_unlock(&BTRFS_I(inode)->accounting_lock);
2832 2832
2833 BTRFS_I(inode)->delalloc_reserved_extents--; 2833 BTRFS_I(inode)->reserved_extents--;
2834 BUG_ON(BTRFS_I(inode)->delalloc_reserved_extents < 0); 2834 BUG_ON(BTRFS_I(inode)->reserved_extents < 0);
2835 2835
2836 if (meta_sinfo->bytes_delalloc < num_bytes) { 2836 if (meta_sinfo->bytes_delalloc < num_bytes) {
2837 bug = true; 2837 bug = true;
@@ -2864,6 +2864,107 @@ static void check_force_delalloc(struct btrfs_space_info *meta_sinfo)
2864 meta_sinfo->force_delalloc = 0; 2864 meta_sinfo->force_delalloc = 0;
2865} 2865}
2866 2866
2867struct async_flush {
2868 struct btrfs_root *root;
2869 struct btrfs_space_info *info;
2870 struct btrfs_work work;
2871};
2872
2873static noinline void flush_delalloc_async(struct btrfs_work *work)
2874{
2875 struct async_flush *async;
2876 struct btrfs_root *root;
2877 struct btrfs_space_info *info;
2878
2879 async = container_of(work, struct async_flush, work);
2880 root = async->root;
2881 info = async->info;
2882
2883 btrfs_start_delalloc_inodes(root);
2884 wake_up(&info->flush_wait);
2885 btrfs_wait_ordered_extents(root, 0);
2886
2887 spin_lock(&info->lock);
2888 info->flushing = 0;
2889 spin_unlock(&info->lock);
2890 wake_up(&info->flush_wait);
2891
2892 kfree(async);
2893}
2894
2895static void wait_on_flush(struct btrfs_space_info *info)
2896{
2897 DEFINE_WAIT(wait);
2898 u64 used;
2899
2900 while (1) {
2901 prepare_to_wait(&info->flush_wait, &wait,
2902 TASK_UNINTERRUPTIBLE);
2903 spin_lock(&info->lock);
2904 if (!info->flushing) {
2905 spin_unlock(&info->lock);
2906 break;
2907 }
2908
2909 used = info->bytes_used + info->bytes_reserved +
2910 info->bytes_pinned + info->bytes_readonly +
2911 info->bytes_super + info->bytes_root +
2912 info->bytes_may_use + info->bytes_delalloc;
2913 if (used < info->total_bytes) {
2914 spin_unlock(&info->lock);
2915 break;
2916 }
2917 spin_unlock(&info->lock);
2918 schedule();
2919 }
2920 finish_wait(&info->flush_wait, &wait);
2921}
2922
2923static void flush_delalloc(struct btrfs_root *root,
2924 struct btrfs_space_info *info)
2925{
2926 struct async_flush *async;
2927 bool wait = false;
2928
2929 spin_lock(&info->lock);
2930
2931 if (!info->flushing) {
2932 info->flushing = 1;
2933 init_waitqueue_head(&info->flush_wait);
2934 } else {
2935 wait = true;
2936 }
2937
2938 spin_unlock(&info->lock);
2939
2940 if (wait) {
2941 wait_on_flush(info);
2942 return;
2943 }
2944
2945 async = kzalloc(sizeof(*async), GFP_NOFS);
2946 if (!async)
2947 goto flush;
2948
2949 async->root = root;
2950 async->info = info;
2951 async->work.func = flush_delalloc_async;
2952
2953 btrfs_queue_worker(&root->fs_info->enospc_workers,
2954 &async->work);
2955 wait_on_flush(info);
2956 return;
2957
2958flush:
2959 btrfs_start_delalloc_inodes(root);
2960 btrfs_wait_ordered_extents(root, 0);
2961
2962 spin_lock(&info->lock);
2963 info->flushing = 0;
2964 spin_unlock(&info->lock);
2965 wake_up(&info->flush_wait);
2966}
2967
2867static int maybe_allocate_chunk(struct btrfs_root *root, 2968static int maybe_allocate_chunk(struct btrfs_root *root,
2868 struct btrfs_space_info *info) 2969 struct btrfs_space_info *info)
2869{ 2970{
@@ -2876,10 +2977,10 @@ static int maybe_allocate_chunk(struct btrfs_root *root,
2876 2977
2877 free_space = btrfs_super_total_bytes(disk_super); 2978 free_space = btrfs_super_total_bytes(disk_super);
2878 /* 2979 /*
2879 * we allow the metadata to grow to a max of either 5gb or 5% of the 2980 * we allow the metadata to grow to a max of either 10gb or 5% of the
2880 * space in the volume. 2981 * space in the volume.
2881 */ 2982 */
2882 min_metadata = min((u64)5 * 1024 * 1024 * 1024, 2983 min_metadata = min((u64)10 * 1024 * 1024 * 1024,
2883 div64_u64(free_space * 5, 100)); 2984 div64_u64(free_space * 5, 100));
2884 if (info->total_bytes >= min_metadata) { 2985 if (info->total_bytes >= min_metadata) {
2885 spin_unlock(&info->lock); 2986 spin_unlock(&info->lock);
@@ -2894,7 +2995,7 @@ static int maybe_allocate_chunk(struct btrfs_root *root,
2894 if (!info->allocating_chunk) { 2995 if (!info->allocating_chunk) {
2895 info->force_alloc = 1; 2996 info->force_alloc = 1;
2896 info->allocating_chunk = 1; 2997 info->allocating_chunk = 1;
2897 init_waitqueue_head(&info->wait); 2998 init_waitqueue_head(&info->allocate_wait);
2898 } else { 2999 } else {
2899 wait = true; 3000 wait = true;
2900 } 3001 }
@@ -2902,7 +3003,7 @@ static int maybe_allocate_chunk(struct btrfs_root *root,
2902 spin_unlock(&info->lock); 3003 spin_unlock(&info->lock);
2903 3004
2904 if (wait) { 3005 if (wait) {
2905 wait_event(info->wait, 3006 wait_event(info->allocate_wait,
2906 !info->allocating_chunk); 3007 !info->allocating_chunk);
2907 return 1; 3008 return 1;
2908 } 3009 }
@@ -2923,7 +3024,7 @@ out:
2923 spin_lock(&info->lock); 3024 spin_lock(&info->lock);
2924 info->allocating_chunk = 0; 3025 info->allocating_chunk = 0;
2925 spin_unlock(&info->lock); 3026 spin_unlock(&info->lock);
2926 wake_up(&info->wait); 3027 wake_up(&info->allocate_wait);
2927 3028
2928 if (ret) 3029 if (ret)
2929 return 0; 3030 return 0;
@@ -2981,21 +3082,20 @@ again:
2981 filemap_flush(inode->i_mapping); 3082 filemap_flush(inode->i_mapping);
2982 goto again; 3083 goto again;
2983 } else if (flushed == 3) { 3084 } else if (flushed == 3) {
2984 btrfs_start_delalloc_inodes(root); 3085 flush_delalloc(root, meta_sinfo);
2985 btrfs_wait_ordered_extents(root, 0);
2986 goto again; 3086 goto again;
2987 } 3087 }
2988 spin_lock(&meta_sinfo->lock); 3088 spin_lock(&meta_sinfo->lock);
2989 meta_sinfo->bytes_delalloc -= num_bytes; 3089 meta_sinfo->bytes_delalloc -= num_bytes;
2990 spin_unlock(&meta_sinfo->lock); 3090 spin_unlock(&meta_sinfo->lock);
2991 printk(KERN_ERR "enospc, has %d, reserved %d\n", 3091 printk(KERN_ERR "enospc, has %d, reserved %d\n",
2992 BTRFS_I(inode)->delalloc_extents, 3092 BTRFS_I(inode)->outstanding_extents,
2993 BTRFS_I(inode)->delalloc_reserved_extents); 3093 BTRFS_I(inode)->reserved_extents);
2994 dump_space_info(meta_sinfo, 0, 0); 3094 dump_space_info(meta_sinfo, 0, 0);
2995 return -ENOSPC; 3095 return -ENOSPC;
2996 } 3096 }
2997 3097
2998 BTRFS_I(inode)->delalloc_reserved_extents++; 3098 BTRFS_I(inode)->reserved_extents++;
2999 check_force_delalloc(meta_sinfo); 3099 check_force_delalloc(meta_sinfo);
3000 spin_unlock(&meta_sinfo->lock); 3100 spin_unlock(&meta_sinfo->lock);
3001 3101
@@ -3094,8 +3194,7 @@ again:
3094 } 3194 }
3095 3195
3096 if (retries == 2) { 3196 if (retries == 2) {
3097 btrfs_start_delalloc_inodes(root); 3197 flush_delalloc(root, meta_sinfo);
3098 btrfs_wait_ordered_extents(root, 0);
3099 goto again; 3198 goto again;
3100 } 3199 }
3101 spin_lock(&meta_sinfo->lock); 3200 spin_lock(&meta_sinfo->lock);
@@ -3588,6 +3687,14 @@ static int pin_down_bytes(struct btrfs_trans_handle *trans,
3588 if (is_data) 3687 if (is_data)
3589 goto pinit; 3688 goto pinit;
3590 3689
3690 /*
3691 * discard is sloooow, and so triggering discards on
3692 * individual btree blocks isn't a good plan. Just
3693 * pin everything in discard mode.
3694 */
3695 if (btrfs_test_opt(root, DISCARD))
3696 goto pinit;
3697
3591 buf = btrfs_find_tree_block(root, bytenr, num_bytes); 3698 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
3592 if (!buf) 3699 if (!buf)
3593 goto pinit; 3700 goto pinit;
@@ -3995,7 +4102,7 @@ wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
3995} 4102}
3996 4103
3997enum btrfs_loop_type { 4104enum btrfs_loop_type {
3998 LOOP_CACHED_ONLY = 0, 4105 LOOP_FIND_IDEAL = 0,
3999 LOOP_CACHING_NOWAIT = 1, 4106 LOOP_CACHING_NOWAIT = 1,
4000 LOOP_CACHING_WAIT = 2, 4107 LOOP_CACHING_WAIT = 2,
4001 LOOP_ALLOC_CHUNK = 3, 4108 LOOP_ALLOC_CHUNK = 3,
@@ -4024,11 +4131,15 @@ static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4024 struct btrfs_block_group_cache *block_group = NULL; 4131 struct btrfs_block_group_cache *block_group = NULL;
4025 int empty_cluster = 2 * 1024 * 1024; 4132 int empty_cluster = 2 * 1024 * 1024;
4026 int allowed_chunk_alloc = 0; 4133 int allowed_chunk_alloc = 0;
4134 int done_chunk_alloc = 0;
4027 struct btrfs_space_info *space_info; 4135 struct btrfs_space_info *space_info;
4028 int last_ptr_loop = 0; 4136 int last_ptr_loop = 0;
4029 int loop = 0; 4137 int loop = 0;
4030 bool found_uncached_bg = false; 4138 bool found_uncached_bg = false;
4031 bool failed_cluster_refill = false; 4139 bool failed_cluster_refill = false;
4140 bool failed_alloc = false;
4141 u64 ideal_cache_percent = 0;
4142 u64 ideal_cache_offset = 0;
4032 4143
4033 WARN_ON(num_bytes < root->sectorsize); 4144 WARN_ON(num_bytes < root->sectorsize);
4034 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); 4145 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
@@ -4064,14 +4175,19 @@ static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4064 empty_cluster = 0; 4175 empty_cluster = 0;
4065 4176
4066 if (search_start == hint_byte) { 4177 if (search_start == hint_byte) {
4178ideal_cache:
4067 block_group = btrfs_lookup_block_group(root->fs_info, 4179 block_group = btrfs_lookup_block_group(root->fs_info,
4068 search_start); 4180 search_start);
4069 /* 4181 /*
4070 * we don't want to use the block group if it doesn't match our 4182 * we don't want to use the block group if it doesn't match our
4071 * allocation bits, or if its not cached. 4183 * allocation bits, or if its not cached.
4184 *
4185 * However if we are re-searching with an ideal block group
4186 * picked out then we don't care that the block group is cached.
4072 */ 4187 */
4073 if (block_group && block_group_bits(block_group, data) && 4188 if (block_group && block_group_bits(block_group, data) &&
4074 block_group_cache_done(block_group)) { 4189 (block_group->cached != BTRFS_CACHE_NO ||
4190 search_start == ideal_cache_offset)) {
4075 down_read(&space_info->groups_sem); 4191 down_read(&space_info->groups_sem);
4076 if (list_empty(&block_group->list) || 4192 if (list_empty(&block_group->list) ||
4077 block_group->ro) { 4193 block_group->ro) {
@@ -4083,13 +4199,13 @@ static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4083 */ 4199 */
4084 btrfs_put_block_group(block_group); 4200 btrfs_put_block_group(block_group);
4085 up_read(&space_info->groups_sem); 4201 up_read(&space_info->groups_sem);
4086 } else 4202 } else {
4087 goto have_block_group; 4203 goto have_block_group;
4204 }
4088 } else if (block_group) { 4205 } else if (block_group) {
4089 btrfs_put_block_group(block_group); 4206 btrfs_put_block_group(block_group);
4090 } 4207 }
4091 } 4208 }
4092
4093search: 4209search:
4094 down_read(&space_info->groups_sem); 4210 down_read(&space_info->groups_sem);
4095 list_for_each_entry(block_group, &space_info->block_groups, list) { 4211 list_for_each_entry(block_group, &space_info->block_groups, list) {
@@ -4101,28 +4217,45 @@ search:
4101 4217
4102have_block_group: 4218have_block_group:
4103 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) { 4219 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4220 u64 free_percent;
4221
4222 free_percent = btrfs_block_group_used(&block_group->item);
4223 free_percent *= 100;
4224 free_percent = div64_u64(free_percent,
4225 block_group->key.offset);
4226 free_percent = 100 - free_percent;
4227 if (free_percent > ideal_cache_percent &&
4228 likely(!block_group->ro)) {
4229 ideal_cache_offset = block_group->key.objectid;
4230 ideal_cache_percent = free_percent;
4231 }
4232
4104 /* 4233 /*
4105 * we want to start caching kthreads, but not too many 4234 * We only want to start kthread caching if we are at
4106 * right off the bat so we don't overwhelm the system, 4235 * the point where we will wait for caching to make
4107 * so only start them if there are less than 2 and we're 4236 * progress, or if our ideal search is over and we've
4108 * in the initial allocation phase. 4237 * found somebody to start caching.
4109 */ 4238 */
4110 if (loop > LOOP_CACHING_NOWAIT || 4239 if (loop > LOOP_CACHING_NOWAIT ||
4111 atomic_read(&space_info->caching_threads) < 2) { 4240 (loop > LOOP_FIND_IDEAL &&
4241 atomic_read(&space_info->caching_threads) < 2)) {
4112 ret = cache_block_group(block_group); 4242 ret = cache_block_group(block_group);
4113 BUG_ON(ret); 4243 BUG_ON(ret);
4114 } 4244 }
4115 }
4116
4117 cached = block_group_cache_done(block_group);
4118 if (unlikely(!cached)) {
4119 found_uncached_bg = true; 4245 found_uncached_bg = true;
4120 4246
4121 /* if we only want cached bgs, loop */ 4247 /*
4122 if (loop == LOOP_CACHED_ONLY) 4248 * If loop is set for cached only, try the next block
4249 * group.
4250 */
4251 if (loop == LOOP_FIND_IDEAL)
4123 goto loop; 4252 goto loop;
4124 } 4253 }
4125 4254
4255 cached = block_group_cache_done(block_group);
4256 if (unlikely(!cached))
4257 found_uncached_bg = true;
4258
4126 if (unlikely(block_group->ro)) 4259 if (unlikely(block_group->ro))
4127 goto loop; 4260 goto loop;
4128 4261
@@ -4233,14 +4366,23 @@ refill_cluster:
4233 4366
4234 offset = btrfs_find_space_for_alloc(block_group, search_start, 4367 offset = btrfs_find_space_for_alloc(block_group, search_start,
4235 num_bytes, empty_size); 4368 num_bytes, empty_size);
4236 if (!offset && (cached || (!cached && 4369 /*
4237 loop == LOOP_CACHING_NOWAIT))) { 4370 * If we didn't find a chunk, and we haven't failed on this
4238 goto loop; 4371 * block group before, and this block group is in the middle of
4239 } else if (!offset && (!cached && 4372 * caching and we are ok with waiting, then go ahead and wait
4240 loop > LOOP_CACHING_NOWAIT)) { 4373 * for progress to be made, and set failed_alloc to true.
4374 *
4375 * If failed_alloc is true then we've already waited on this
4376 * block group once and should move on to the next block group.
4377 */
4378 if (!offset && !failed_alloc && !cached &&
4379 loop > LOOP_CACHING_NOWAIT) {
4241 wait_block_group_cache_progress(block_group, 4380 wait_block_group_cache_progress(block_group,
4242 num_bytes + empty_size); 4381 num_bytes + empty_size);
4382 failed_alloc = true;
4243 goto have_block_group; 4383 goto have_block_group;
4384 } else if (!offset) {
4385 goto loop;
4244 } 4386 }
4245checks: 4387checks:
4246 search_start = stripe_align(root, offset); 4388 search_start = stripe_align(root, offset);
@@ -4288,13 +4430,16 @@ checks:
4288 break; 4430 break;
4289loop: 4431loop:
4290 failed_cluster_refill = false; 4432 failed_cluster_refill = false;
4433 failed_alloc = false;
4291 btrfs_put_block_group(block_group); 4434 btrfs_put_block_group(block_group);
4292 } 4435 }
4293 up_read(&space_info->groups_sem); 4436 up_read(&space_info->groups_sem);
4294 4437
4295 /* LOOP_CACHED_ONLY, only search fully cached block groups 4438 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
4296 * LOOP_CACHING_NOWAIT, search partially cached block groups, but 4439 * for them to make caching progress. Also
4297 * dont wait foR them to finish caching 4440 * determine the best possible bg to cache
4441 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
4442 * caching kthreads as we move along
4298 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching 4443 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4299 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again 4444 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4300 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try 4445 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
@@ -4303,12 +4448,47 @@ loop:
4303 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE && 4448 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
4304 (found_uncached_bg || empty_size || empty_cluster || 4449 (found_uncached_bg || empty_size || empty_cluster ||
4305 allowed_chunk_alloc)) { 4450 allowed_chunk_alloc)) {
4306 if (found_uncached_bg) { 4451 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
4307 found_uncached_bg = false; 4452 found_uncached_bg = false;
4308 if (loop < LOOP_CACHING_WAIT) { 4453 loop++;
4309 loop++; 4454 if (!ideal_cache_percent &&
4455 atomic_read(&space_info->caching_threads))
4310 goto search; 4456 goto search;
4311 } 4457
4458 /*
4459 * 1 of the following 2 things have happened so far
4460 *
4461 * 1) We found an ideal block group for caching that
4462 * is mostly full and will cache quickly, so we might
4463 * as well wait for it.
4464 *
4465 * 2) We searched for cached only and we didn't find
4466 * anything, and we didn't start any caching kthreads
4467 * either, so chances are we will loop through and
4468 * start a couple caching kthreads, and then come back
4469 * around and just wait for them. This will be slower
4470 * because we will have 2 caching kthreads reading at
4471 * the same time when we could have just started one
4472 * and waited for it to get far enough to give us an
4473 * allocation, so go ahead and go to the wait caching
4474 * loop.
4475 */
4476 loop = LOOP_CACHING_WAIT;
4477 search_start = ideal_cache_offset;
4478 ideal_cache_percent = 0;
4479 goto ideal_cache;
4480 } else if (loop == LOOP_FIND_IDEAL) {
4481 /*
4482 * Didn't find a uncached bg, wait on anything we find
4483 * next.
4484 */
4485 loop = LOOP_CACHING_WAIT;
4486 goto search;
4487 }
4488
4489 if (loop < LOOP_CACHING_WAIT) {
4490 loop++;
4491 goto search;
4312 } 4492 }
4313 4493
4314 if (loop == LOOP_ALLOC_CHUNK) { 4494 if (loop == LOOP_ALLOC_CHUNK) {
@@ -4320,7 +4500,8 @@ loop:
4320 ret = do_chunk_alloc(trans, root, num_bytes + 4500 ret = do_chunk_alloc(trans, root, num_bytes +
4321 2 * 1024 * 1024, data, 1); 4501 2 * 1024 * 1024, data, 1);
4322 allowed_chunk_alloc = 0; 4502 allowed_chunk_alloc = 0;
4323 } else { 4503 done_chunk_alloc = 1;
4504 } else if (!done_chunk_alloc) {
4324 space_info->force_alloc = 1; 4505 space_info->force_alloc = 1;
4325 } 4506 }
4326 4507
@@ -4799,6 +4980,7 @@ static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4799 u64 bytenr; 4980 u64 bytenr;
4800 u64 generation; 4981 u64 generation;
4801 u64 refs; 4982 u64 refs;
4983 u64 flags;
4802 u64 last = 0; 4984 u64 last = 0;
4803 u32 nritems; 4985 u32 nritems;
4804 u32 blocksize; 4986 u32 blocksize;
@@ -4836,15 +5018,19 @@ static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4836 generation <= root->root_key.offset) 5018 generation <= root->root_key.offset)
4837 continue; 5019 continue;
4838 5020
5021 /* We don't lock the tree block, it's OK to be racy here */
5022 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5023 &refs, &flags);
5024 BUG_ON(ret);
5025 BUG_ON(refs == 0);
5026
4839 if (wc->stage == DROP_REFERENCE) { 5027 if (wc->stage == DROP_REFERENCE) {
4840 ret = btrfs_lookup_extent_info(trans, root,
4841 bytenr, blocksize,
4842 &refs, NULL);
4843 BUG_ON(ret);
4844 BUG_ON(refs == 0);
4845 if (refs == 1) 5028 if (refs == 1)
4846 goto reada; 5029 goto reada;
4847 5030
5031 if (wc->level == 1 &&
5032 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5033 continue;
4848 if (!wc->update_ref || 5034 if (!wc->update_ref ||
4849 generation <= root->root_key.offset) 5035 generation <= root->root_key.offset)
4850 continue; 5036 continue;
@@ -4853,6 +5039,10 @@ static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4853 &wc->update_progress); 5039 &wc->update_progress);
4854 if (ret < 0) 5040 if (ret < 0)
4855 continue; 5041 continue;
5042 } else {
5043 if (wc->level == 1 &&
5044 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5045 continue;
4856 } 5046 }
4857reada: 5047reada:
4858 ret = readahead_tree_block(root, bytenr, blocksize, 5048 ret = readahead_tree_block(root, bytenr, blocksize,
@@ -4876,7 +5066,7 @@ reada:
4876static noinline int walk_down_proc(struct btrfs_trans_handle *trans, 5066static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4877 struct btrfs_root *root, 5067 struct btrfs_root *root,
4878 struct btrfs_path *path, 5068 struct btrfs_path *path,
4879 struct walk_control *wc) 5069 struct walk_control *wc, int lookup_info)
4880{ 5070{
4881 int level = wc->level; 5071 int level = wc->level;
4882 struct extent_buffer *eb = path->nodes[level]; 5072 struct extent_buffer *eb = path->nodes[level];
@@ -4891,8 +5081,9 @@ static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4891 * when reference count of tree block is 1, it won't increase 5081 * when reference count of tree block is 1, it won't increase
4892 * again. once full backref flag is set, we never clear it. 5082 * again. once full backref flag is set, we never clear it.
4893 */ 5083 */
4894 if ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || 5084 if (lookup_info &&
4895 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag))) { 5085 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5086 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
4896 BUG_ON(!path->locks[level]); 5087 BUG_ON(!path->locks[level]);
4897 ret = btrfs_lookup_extent_info(trans, root, 5088 ret = btrfs_lookup_extent_info(trans, root,
4898 eb->start, eb->len, 5089 eb->start, eb->len,
@@ -4953,7 +5144,7 @@ static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4953static noinline int do_walk_down(struct btrfs_trans_handle *trans, 5144static noinline int do_walk_down(struct btrfs_trans_handle *trans,
4954 struct btrfs_root *root, 5145 struct btrfs_root *root,
4955 struct btrfs_path *path, 5146 struct btrfs_path *path,
4956 struct walk_control *wc) 5147 struct walk_control *wc, int *lookup_info)
4957{ 5148{
4958 u64 bytenr; 5149 u64 bytenr;
4959 u64 generation; 5150 u64 generation;
@@ -4973,8 +5164,10 @@ static noinline int do_walk_down(struct btrfs_trans_handle *trans,
4973 * for the subtree 5164 * for the subtree
4974 */ 5165 */
4975 if (wc->stage == UPDATE_BACKREF && 5166 if (wc->stage == UPDATE_BACKREF &&
4976 generation <= root->root_key.offset) 5167 generation <= root->root_key.offset) {
5168 *lookup_info = 1;
4977 return 1; 5169 return 1;
5170 }
4978 5171
4979 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); 5172 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
4980 blocksize = btrfs_level_size(root, level - 1); 5173 blocksize = btrfs_level_size(root, level - 1);
@@ -4987,14 +5180,19 @@ static noinline int do_walk_down(struct btrfs_trans_handle *trans,
4987 btrfs_tree_lock(next); 5180 btrfs_tree_lock(next);
4988 btrfs_set_lock_blocking(next); 5181 btrfs_set_lock_blocking(next);
4989 5182
4990 if (wc->stage == DROP_REFERENCE) { 5183 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
4991 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, 5184 &wc->refs[level - 1],
4992 &wc->refs[level - 1], 5185 &wc->flags[level - 1]);
4993 &wc->flags[level - 1]); 5186 BUG_ON(ret);
4994 BUG_ON(ret); 5187 BUG_ON(wc->refs[level - 1] == 0);
4995 BUG_ON(wc->refs[level - 1] == 0); 5188 *lookup_info = 0;
4996 5189
5190 if (wc->stage == DROP_REFERENCE) {
4997 if (wc->refs[level - 1] > 1) { 5191 if (wc->refs[level - 1] > 1) {
5192 if (level == 1 &&
5193 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5194 goto skip;
5195
4998 if (!wc->update_ref || 5196 if (!wc->update_ref ||
4999 generation <= root->root_key.offset) 5197 generation <= root->root_key.offset)
5000 goto skip; 5198 goto skip;
@@ -5008,12 +5206,17 @@ static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5008 wc->stage = UPDATE_BACKREF; 5206 wc->stage = UPDATE_BACKREF;
5009 wc->shared_level = level - 1; 5207 wc->shared_level = level - 1;
5010 } 5208 }
5209 } else {
5210 if (level == 1 &&
5211 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5212 goto skip;
5011 } 5213 }
5012 5214
5013 if (!btrfs_buffer_uptodate(next, generation)) { 5215 if (!btrfs_buffer_uptodate(next, generation)) {
5014 btrfs_tree_unlock(next); 5216 btrfs_tree_unlock(next);
5015 free_extent_buffer(next); 5217 free_extent_buffer(next);
5016 next = NULL; 5218 next = NULL;
5219 *lookup_info = 1;
5017 } 5220 }
5018 5221
5019 if (!next) { 5222 if (!next) {
@@ -5036,21 +5239,22 @@ static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5036skip: 5239skip:
5037 wc->refs[level - 1] = 0; 5240 wc->refs[level - 1] = 0;
5038 wc->flags[level - 1] = 0; 5241 wc->flags[level - 1] = 0;
5242 if (wc->stage == DROP_REFERENCE) {
5243 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5244 parent = path->nodes[level]->start;
5245 } else {
5246 BUG_ON(root->root_key.objectid !=
5247 btrfs_header_owner(path->nodes[level]));
5248 parent = 0;
5249 }
5039 5250
5040 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { 5251 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5041 parent = path->nodes[level]->start; 5252 root->root_key.objectid, level - 1, 0);
5042 } else { 5253 BUG_ON(ret);
5043 BUG_ON(root->root_key.objectid !=
5044 btrfs_header_owner(path->nodes[level]));
5045 parent = 0;
5046 } 5254 }
5047
5048 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5049 root->root_key.objectid, level - 1, 0);
5050 BUG_ON(ret);
5051
5052 btrfs_tree_unlock(next); 5255 btrfs_tree_unlock(next);
5053 free_extent_buffer(next); 5256 free_extent_buffer(next);
5257 *lookup_info = 1;
5054 return 1; 5258 return 1;
5055} 5259}
5056 5260
@@ -5164,6 +5368,7 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5164 struct walk_control *wc) 5368 struct walk_control *wc)
5165{ 5369{
5166 int level = wc->level; 5370 int level = wc->level;
5371 int lookup_info = 1;
5167 int ret; 5372 int ret;
5168 5373
5169 while (level >= 0) { 5374 while (level >= 0) {
@@ -5171,14 +5376,14 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5171 btrfs_header_nritems(path->nodes[level])) 5376 btrfs_header_nritems(path->nodes[level]))
5172 break; 5377 break;
5173 5378
5174 ret = walk_down_proc(trans, root, path, wc); 5379 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5175 if (ret > 0) 5380 if (ret > 0)
5176 break; 5381 break;
5177 5382
5178 if (level == 0) 5383 if (level == 0)
5179 break; 5384 break;
5180 5385
5181 ret = do_walk_down(trans, root, path, wc); 5386 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5182 if (ret > 0) { 5387 if (ret > 0) {
5183 path->slots[level]++; 5388 path->slots[level]++;
5184 continue; 5389 continue;
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index de1793ba004a..96577e8bf9fd 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -460,7 +460,8 @@ static int clear_state_bit(struct extent_io_tree *tree,
460 struct extent_state *state, int bits, int wake, 460 struct extent_state *state, int bits, int wake,
461 int delete) 461 int delete)
462{ 462{
463 int ret = state->state & bits; 463 int bits_to_clear = bits & ~EXTENT_DO_ACCOUNTING;
464 int ret = state->state & bits_to_clear;
464 465
465 if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { 466 if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
466 u64 range = state->end - state->start + 1; 467 u64 range = state->end - state->start + 1;
@@ -468,7 +469,7 @@ static int clear_state_bit(struct extent_io_tree *tree,
468 tree->dirty_bytes -= range; 469 tree->dirty_bytes -= range;
469 } 470 }
470 clear_state_cb(tree, state, bits); 471 clear_state_cb(tree, state, bits);
471 state->state &= ~bits; 472 state->state &= ~bits_to_clear;
472 if (wake) 473 if (wake)
473 wake_up(&state->wq); 474 wake_up(&state->wq);
474 if (delete || state->state == 0) { 475 if (delete || state->state == 0) {
@@ -956,7 +957,8 @@ int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
956 gfp_t mask) 957 gfp_t mask)
957{ 958{
958 return clear_extent_bit(tree, start, end, 959 return clear_extent_bit(tree, start, end,
959 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, 960 EXTENT_DIRTY | EXTENT_DELALLOC |
961 EXTENT_DO_ACCOUNTING, 0, 0,
960 NULL, mask); 962 NULL, mask);
961} 963}
962 964
@@ -1401,12 +1403,7 @@ out_failed:
1401int extent_clear_unlock_delalloc(struct inode *inode, 1403int extent_clear_unlock_delalloc(struct inode *inode,
1402 struct extent_io_tree *tree, 1404 struct extent_io_tree *tree,
1403 u64 start, u64 end, struct page *locked_page, 1405 u64 start, u64 end, struct page *locked_page,
1404 int unlock_pages, 1406 unsigned long op)
1405 int clear_unlock,
1406 int clear_delalloc, int clear_dirty,
1407 int set_writeback,
1408 int end_writeback,
1409 int set_private2)
1410{ 1407{
1411 int ret; 1408 int ret;
1412 struct page *pages[16]; 1409 struct page *pages[16];
@@ -1416,17 +1413,21 @@ int extent_clear_unlock_delalloc(struct inode *inode,
1416 int i; 1413 int i;
1417 int clear_bits = 0; 1414 int clear_bits = 0;
1418 1415
1419 if (clear_unlock) 1416 if (op & EXTENT_CLEAR_UNLOCK)
1420 clear_bits |= EXTENT_LOCKED; 1417 clear_bits |= EXTENT_LOCKED;
1421 if (clear_dirty) 1418 if (op & EXTENT_CLEAR_DIRTY)
1422 clear_bits |= EXTENT_DIRTY; 1419 clear_bits |= EXTENT_DIRTY;
1423 1420
1424 if (clear_delalloc) 1421 if (op & EXTENT_CLEAR_DELALLOC)
1425 clear_bits |= EXTENT_DELALLOC; 1422 clear_bits |= EXTENT_DELALLOC;
1426 1423
1424 if (op & EXTENT_CLEAR_ACCOUNTING)
1425 clear_bits |= EXTENT_DO_ACCOUNTING;
1426
1427 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); 1427 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1428 if (!(unlock_pages || clear_dirty || set_writeback || end_writeback || 1428 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1429 set_private2)) 1429 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1430 EXTENT_SET_PRIVATE2)))
1430 return 0; 1431 return 0;
1431 1432
1432 while (nr_pages > 0) { 1433 while (nr_pages > 0) {
@@ -1435,20 +1436,20 @@ int extent_clear_unlock_delalloc(struct inode *inode,
1435 nr_pages, ARRAY_SIZE(pages)), pages); 1436 nr_pages, ARRAY_SIZE(pages)), pages);
1436 for (i = 0; i < ret; i++) { 1437 for (i = 0; i < ret; i++) {
1437 1438
1438 if (set_private2) 1439 if (op & EXTENT_SET_PRIVATE2)
1439 SetPagePrivate2(pages[i]); 1440 SetPagePrivate2(pages[i]);
1440 1441
1441 if (pages[i] == locked_page) { 1442 if (pages[i] == locked_page) {
1442 page_cache_release(pages[i]); 1443 page_cache_release(pages[i]);
1443 continue; 1444 continue;
1444 } 1445 }
1445 if (clear_dirty) 1446 if (op & EXTENT_CLEAR_DIRTY)
1446 clear_page_dirty_for_io(pages[i]); 1447 clear_page_dirty_for_io(pages[i]);
1447 if (set_writeback) 1448 if (op & EXTENT_SET_WRITEBACK)
1448 set_page_writeback(pages[i]); 1449 set_page_writeback(pages[i]);
1449 if (end_writeback) 1450 if (op & EXTENT_END_WRITEBACK)
1450 end_page_writeback(pages[i]); 1451 end_page_writeback(pages[i]);
1451 if (unlock_pages) 1452 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1452 unlock_page(pages[i]); 1453 unlock_page(pages[i]);
1453 page_cache_release(pages[i]); 1454 page_cache_release(pages[i]);
1454 } 1455 }
@@ -2714,7 +2715,8 @@ int extent_invalidatepage(struct extent_io_tree *tree,
2714 lock_extent(tree, start, end, GFP_NOFS); 2715 lock_extent(tree, start, end, GFP_NOFS);
2715 wait_on_page_writeback(page); 2716 wait_on_page_writeback(page);
2716 clear_extent_bit(tree, start, end, 2717 clear_extent_bit(tree, start, end,
2717 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC, 2718 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
2719 EXTENT_DO_ACCOUNTING,
2718 1, 1, NULL, GFP_NOFS); 2720 1, 1, NULL, GFP_NOFS);
2719 return 0; 2721 return 0;
2720} 2722}
diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h
index 4794ec891fed..36de250a7b2b 100644
--- a/fs/btrfs/extent_io.h
+++ b/fs/btrfs/extent_io.h
@@ -15,6 +15,7 @@
15#define EXTENT_BUFFER_FILLED (1 << 8) 15#define EXTENT_BUFFER_FILLED (1 << 8)
16#define EXTENT_BOUNDARY (1 << 9) 16#define EXTENT_BOUNDARY (1 << 9)
17#define EXTENT_NODATASUM (1 << 10) 17#define EXTENT_NODATASUM (1 << 10)
18#define EXTENT_DO_ACCOUNTING (1 << 11)
18#define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK) 19#define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK)
19 20
20/* flags for bio submission */ 21/* flags for bio submission */
@@ -25,6 +26,16 @@
25#define EXTENT_BUFFER_BLOCKING 1 26#define EXTENT_BUFFER_BLOCKING 1
26#define EXTENT_BUFFER_DIRTY 2 27#define EXTENT_BUFFER_DIRTY 2
27 28
29/* these are flags for extent_clear_unlock_delalloc */
30#define EXTENT_CLEAR_UNLOCK_PAGE 0x1
31#define EXTENT_CLEAR_UNLOCK 0x2
32#define EXTENT_CLEAR_DELALLOC 0x4
33#define EXTENT_CLEAR_DIRTY 0x8
34#define EXTENT_SET_WRITEBACK 0x10
35#define EXTENT_END_WRITEBACK 0x20
36#define EXTENT_SET_PRIVATE2 0x40
37#define EXTENT_CLEAR_ACCOUNTING 0x80
38
28/* 39/*
29 * page->private values. Every page that is controlled by the extent 40 * page->private values. Every page that is controlled by the extent
30 * map has page->private set to one. 41 * map has page->private set to one.
@@ -288,10 +299,5 @@ int extent_range_uptodate(struct extent_io_tree *tree,
288int extent_clear_unlock_delalloc(struct inode *inode, 299int extent_clear_unlock_delalloc(struct inode *inode,
289 struct extent_io_tree *tree, 300 struct extent_io_tree *tree,
290 u64 start, u64 end, struct page *locked_page, 301 u64 start, u64 end, struct page *locked_page,
291 int unlock_page, 302 unsigned long op);
292 int clear_unlock,
293 int clear_delalloc, int clear_dirty,
294 int set_writeback,
295 int end_writeback,
296 int set_private2);
297#endif 303#endif
diff --git a/fs/btrfs/extent_map.c b/fs/btrfs/extent_map.c
index 2c726b7b9faa..46bea0f4dc7b 100644
--- a/fs/btrfs/extent_map.c
+++ b/fs/btrfs/extent_map.c
@@ -208,7 +208,7 @@ int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len)
208 write_lock(&tree->lock); 208 write_lock(&tree->lock);
209 em = lookup_extent_mapping(tree, start, len); 209 em = lookup_extent_mapping(tree, start, len);
210 210
211 WARN_ON(em->start != start || !em); 211 WARN_ON(!em || em->start != start);
212 212
213 if (!em) 213 if (!em)
214 goto out; 214 goto out;
@@ -256,7 +256,7 @@ out:
256 * Insert @em into @tree or perform a simple forward/backward merge with 256 * Insert @em into @tree or perform a simple forward/backward merge with
257 * existing mappings. The extent_map struct passed in will be inserted 257 * existing mappings. The extent_map struct passed in will be inserted
258 * into the tree directly, with an additional reference taken, or a 258 * into the tree directly, with an additional reference taken, or a
259 * reference dropped if the merge attempt was sucessfull. 259 * reference dropped if the merge attempt was successfull.
260 */ 260 */
261int add_extent_mapping(struct extent_map_tree *tree, 261int add_extent_mapping(struct extent_map_tree *tree,
262 struct extent_map *em) 262 struct extent_map *em)
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index f19e1259a971..77f759302e12 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -878,7 +878,8 @@ again:
878 btrfs_put_ordered_extent(ordered); 878 btrfs_put_ordered_extent(ordered);
879 879
880 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos, 880 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
881 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC, 881 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
882 EXTENT_DO_ACCOUNTING,
882 GFP_NOFS); 883 GFP_NOFS);
883 unlock_extent(&BTRFS_I(inode)->io_tree, 884 unlock_extent(&BTRFS_I(inode)->io_tree,
884 start_pos, last_pos - 1, GFP_NOFS); 885 start_pos, last_pos - 1, GFP_NOFS);
@@ -908,7 +909,7 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
908 unsigned long last_index; 909 unsigned long last_index;
909 int will_write; 910 int will_write;
910 911
911 will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) || 912 will_write = ((file->f_flags & O_DSYNC) || IS_SYNC(inode) ||
912 (file->f_flags & O_DIRECT)); 913 (file->f_flags & O_DIRECT));
913 914
914 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE, 915 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
@@ -1075,7 +1076,7 @@ out_nolock:
1075 if (err) 1076 if (err)
1076 num_written = err; 1077 num_written = err;
1077 1078
1078 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) { 1079 if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) {
1079 trans = btrfs_start_transaction(root, 1); 1080 trans = btrfs_start_transaction(root, 1);
1080 ret = btrfs_log_dentry_safe(trans, root, 1081 ret = btrfs_log_dentry_safe(trans, root,
1081 file->f_dentry); 1082 file->f_dentry);
@@ -1085,8 +1086,10 @@ out_nolock:
1085 btrfs_end_transaction(trans, root); 1086 btrfs_end_transaction(trans, root);
1086 else 1087 else
1087 btrfs_commit_transaction(trans, root); 1088 btrfs_commit_transaction(trans, root);
1088 } else { 1089 } else if (ret != BTRFS_NO_LOG_SYNC) {
1089 btrfs_commit_transaction(trans, root); 1090 btrfs_commit_transaction(trans, root);
1091 } else {
1092 btrfs_end_transaction(trans, root);
1090 } 1093 }
1091 } 1094 }
1092 if (file->f_flags & O_DIRECT) { 1095 if (file->f_flags & O_DIRECT) {
@@ -1136,6 +1139,13 @@ int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1136 int ret = 0; 1139 int ret = 0;
1137 struct btrfs_trans_handle *trans; 1140 struct btrfs_trans_handle *trans;
1138 1141
1142
1143 /* we wait first, since the writeback may change the inode */
1144 root->log_batch++;
1145 /* the VFS called filemap_fdatawrite for us */
1146 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1147 root->log_batch++;
1148
1139 /* 1149 /*
1140 * check the transaction that last modified this inode 1150 * check the transaction that last modified this inode
1141 * and see if its already been committed 1151 * and see if its already been committed
@@ -1143,6 +1153,11 @@ int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1143 if (!BTRFS_I(inode)->last_trans) 1153 if (!BTRFS_I(inode)->last_trans)
1144 goto out; 1154 goto out;
1145 1155
1156 /*
1157 * if the last transaction that changed this file was before
1158 * the current transaction, we can bail out now without any
1159 * syncing
1160 */
1146 mutex_lock(&root->fs_info->trans_mutex); 1161 mutex_lock(&root->fs_info->trans_mutex);
1147 if (BTRFS_I(inode)->last_trans <= 1162 if (BTRFS_I(inode)->last_trans <=
1148 root->fs_info->last_trans_committed) { 1163 root->fs_info->last_trans_committed) {
@@ -1152,13 +1167,6 @@ int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1152 } 1167 }
1153 mutex_unlock(&root->fs_info->trans_mutex); 1168 mutex_unlock(&root->fs_info->trans_mutex);
1154 1169
1155 root->log_batch++;
1156 filemap_fdatawrite(inode->i_mapping);
1157 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1158 root->log_batch++;
1159
1160 if (datasync && !(inode->i_state & I_DIRTY_PAGES))
1161 goto out;
1162 /* 1170 /*
1163 * ok we haven't committed the transaction yet, lets do a commit 1171 * ok we haven't committed the transaction yet, lets do a commit
1164 */ 1172 */
@@ -1187,14 +1195,18 @@ int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1187 */ 1195 */
1188 mutex_unlock(&dentry->d_inode->i_mutex); 1196 mutex_unlock(&dentry->d_inode->i_mutex);
1189 1197
1190 if (ret > 0) { 1198 if (ret != BTRFS_NO_LOG_SYNC) {
1191 ret = btrfs_commit_transaction(trans, root); 1199 if (ret > 0) {
1192 } else {
1193 ret = btrfs_sync_log(trans, root);
1194 if (ret == 0)
1195 ret = btrfs_end_transaction(trans, root);
1196 else
1197 ret = btrfs_commit_transaction(trans, root); 1200 ret = btrfs_commit_transaction(trans, root);
1201 } else {
1202 ret = btrfs_sync_log(trans, root);
1203 if (ret == 0)
1204 ret = btrfs_end_transaction(trans, root);
1205 else
1206 ret = btrfs_commit_transaction(trans, root);
1207 }
1208 } else {
1209 ret = btrfs_end_transaction(trans, root);
1198 } 1210 }
1199 mutex_lock(&dentry->d_inode->i_mutex); 1211 mutex_lock(&dentry->d_inode->i_mutex);
1200out: 1212out:
diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c
index 5c2caad76212..cb2849f03251 100644
--- a/fs/btrfs/free-space-cache.c
+++ b/fs/btrfs/free-space-cache.c
@@ -1296,7 +1296,7 @@ again:
1296 window_start = entry->offset; 1296 window_start = entry->offset;
1297 window_free = entry->bytes; 1297 window_free = entry->bytes;
1298 last = entry; 1298 last = entry;
1299 max_extent = 0; 1299 max_extent = entry->bytes;
1300 } else { 1300 } else {
1301 last = next; 1301 last = next;
1302 window_free += next->bytes; 1302 window_free += next->bytes;
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 112e5aa85892..b3ad168a0bfc 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -424,9 +424,12 @@ again:
424 * and free up our temp pages. 424 * and free up our temp pages.
425 */ 425 */
426 extent_clear_unlock_delalloc(inode, 426 extent_clear_unlock_delalloc(inode,
427 &BTRFS_I(inode)->io_tree, 427 &BTRFS_I(inode)->io_tree,
428 start, end, NULL, 1, 0, 428 start, end, NULL,
429 0, 1, 1, 1, 0); 429 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
430 EXTENT_CLEAR_DELALLOC |
431 EXTENT_CLEAR_ACCOUNTING |
432 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
430 ret = 0; 433 ret = 0;
431 goto free_pages_out; 434 goto free_pages_out;
432 } 435 }
@@ -535,7 +538,7 @@ static noinline int submit_compressed_extents(struct inode *inode,
535 struct btrfs_root *root = BTRFS_I(inode)->root; 538 struct btrfs_root *root = BTRFS_I(inode)->root;
536 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 539 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
537 struct extent_io_tree *io_tree; 540 struct extent_io_tree *io_tree;
538 int ret; 541 int ret = 0;
539 542
540 if (list_empty(&async_cow->extents)) 543 if (list_empty(&async_cow->extents))
541 return 0; 544 return 0;
@@ -549,6 +552,7 @@ static noinline int submit_compressed_extents(struct inode *inode,
549 552
550 io_tree = &BTRFS_I(inode)->io_tree; 553 io_tree = &BTRFS_I(inode)->io_tree;
551 554
555retry:
552 /* did the compression code fall back to uncompressed IO? */ 556 /* did the compression code fall back to uncompressed IO? */
553 if (!async_extent->pages) { 557 if (!async_extent->pages) {
554 int page_started = 0; 558 int page_started = 0;
@@ -559,11 +563,11 @@ static noinline int submit_compressed_extents(struct inode *inode,
559 async_extent->ram_size - 1, GFP_NOFS); 563 async_extent->ram_size - 1, GFP_NOFS);
560 564
561 /* allocate blocks */ 565 /* allocate blocks */
562 cow_file_range(inode, async_cow->locked_page, 566 ret = cow_file_range(inode, async_cow->locked_page,
563 async_extent->start, 567 async_extent->start,
564 async_extent->start + 568 async_extent->start +
565 async_extent->ram_size - 1, 569 async_extent->ram_size - 1,
566 &page_started, &nr_written, 0); 570 &page_started, &nr_written, 0);
567 571
568 /* 572 /*
569 * if page_started, cow_file_range inserted an 573 * if page_started, cow_file_range inserted an
@@ -571,7 +575,7 @@ static noinline int submit_compressed_extents(struct inode *inode,
571 * and IO for us. Otherwise, we need to submit 575 * and IO for us. Otherwise, we need to submit
572 * all those pages down to the drive. 576 * all those pages down to the drive.
573 */ 577 */
574 if (!page_started) 578 if (!page_started && !ret)
575 extent_write_locked_range(io_tree, 579 extent_write_locked_range(io_tree,
576 inode, async_extent->start, 580 inode, async_extent->start,
577 async_extent->start + 581 async_extent->start +
@@ -599,7 +603,21 @@ static noinline int submit_compressed_extents(struct inode *inode,
599 async_extent->compressed_size, 603 async_extent->compressed_size,
600 0, alloc_hint, 604 0, alloc_hint,
601 (u64)-1, &ins, 1); 605 (u64)-1, &ins, 1);
602 BUG_ON(ret); 606 if (ret) {
607 int i;
608 for (i = 0; i < async_extent->nr_pages; i++) {
609 WARN_ON(async_extent->pages[i]->mapping);
610 page_cache_release(async_extent->pages[i]);
611 }
612 kfree(async_extent->pages);
613 async_extent->nr_pages = 0;
614 async_extent->pages = NULL;
615 unlock_extent(io_tree, async_extent->start,
616 async_extent->start +
617 async_extent->ram_size - 1, GFP_NOFS);
618 goto retry;
619 }
620
603 em = alloc_extent_map(GFP_NOFS); 621 em = alloc_extent_map(GFP_NOFS);
604 em->start = async_extent->start; 622 em->start = async_extent->start;
605 em->len = async_extent->ram_size; 623 em->len = async_extent->ram_size;
@@ -637,11 +655,14 @@ static noinline int submit_compressed_extents(struct inode *inode,
637 * clear dirty, set writeback and unlock the pages. 655 * clear dirty, set writeback and unlock the pages.
638 */ 656 */
639 extent_clear_unlock_delalloc(inode, 657 extent_clear_unlock_delalloc(inode,
640 &BTRFS_I(inode)->io_tree, 658 &BTRFS_I(inode)->io_tree,
641 async_extent->start, 659 async_extent->start,
642 async_extent->start + 660 async_extent->start +
643 async_extent->ram_size - 1, 661 async_extent->ram_size - 1,
644 NULL, 1, 1, 0, 1, 1, 0, 0); 662 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
663 EXTENT_CLEAR_UNLOCK |
664 EXTENT_CLEAR_DELALLOC |
665 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
645 666
646 ret = btrfs_submit_compressed_write(inode, 667 ret = btrfs_submit_compressed_write(inode,
647 async_extent->start, 668 async_extent->start,
@@ -712,9 +733,15 @@ static noinline int cow_file_range(struct inode *inode,
712 start, end, 0, NULL); 733 start, end, 0, NULL);
713 if (ret == 0) { 734 if (ret == 0) {
714 extent_clear_unlock_delalloc(inode, 735 extent_clear_unlock_delalloc(inode,
715 &BTRFS_I(inode)->io_tree, 736 &BTRFS_I(inode)->io_tree,
716 start, end, NULL, 1, 1, 737 start, end, NULL,
717 1, 1, 1, 1, 0); 738 EXTENT_CLEAR_UNLOCK_PAGE |
739 EXTENT_CLEAR_UNLOCK |
740 EXTENT_CLEAR_DELALLOC |
741 EXTENT_CLEAR_ACCOUNTING |
742 EXTENT_CLEAR_DIRTY |
743 EXTENT_SET_WRITEBACK |
744 EXTENT_END_WRITEBACK);
718 *nr_written = *nr_written + 745 *nr_written = *nr_written +
719 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; 746 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
720 *page_started = 1; 747 *page_started = 1;
@@ -731,13 +758,29 @@ static noinline int cow_file_range(struct inode *inode,
731 em = search_extent_mapping(&BTRFS_I(inode)->extent_tree, 758 em = search_extent_mapping(&BTRFS_I(inode)->extent_tree,
732 start, num_bytes); 759 start, num_bytes);
733 if (em) { 760 if (em) {
734 alloc_hint = em->block_start; 761 /*
735 free_extent_map(em); 762 * if block start isn't an actual block number then find the
763 * first block in this inode and use that as a hint. If that
764 * block is also bogus then just don't worry about it.
765 */
766 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
767 free_extent_map(em);
768 em = search_extent_mapping(em_tree, 0, 0);
769 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
770 alloc_hint = em->block_start;
771 if (em)
772 free_extent_map(em);
773 } else {
774 alloc_hint = em->block_start;
775 free_extent_map(em);
776 }
736 } 777 }
737 read_unlock(&BTRFS_I(inode)->extent_tree.lock); 778 read_unlock(&BTRFS_I(inode)->extent_tree.lock);
738 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); 779 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
739 780
740 while (disk_num_bytes > 0) { 781 while (disk_num_bytes > 0) {
782 unsigned long op;
783
741 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent); 784 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
742 ret = btrfs_reserve_extent(trans, root, cur_alloc_size, 785 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
743 root->sectorsize, 0, alloc_hint, 786 root->sectorsize, 0, alloc_hint,
@@ -789,10 +832,13 @@ static noinline int cow_file_range(struct inode *inode,
789 * Do set the Private2 bit so we know this page was properly 832 * Do set the Private2 bit so we know this page was properly
790 * setup for writepage 833 * setup for writepage
791 */ 834 */
835 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
836 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
837 EXTENT_SET_PRIVATE2;
838
792 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, 839 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
793 start, start + ram_size - 1, 840 start, start + ram_size - 1,
794 locked_page, unlock, 1, 841 locked_page, op);
795 1, 0, 0, 0, 1);
796 disk_num_bytes -= cur_alloc_size; 842 disk_num_bytes -= cur_alloc_size;
797 num_bytes -= cur_alloc_size; 843 num_bytes -= cur_alloc_size;
798 alloc_hint = ins.objectid + ins.offset; 844 alloc_hint = ins.objectid + ins.offset;
@@ -864,8 +910,8 @@ static int cow_file_range_async(struct inode *inode, struct page *locked_page,
864 u64 cur_end; 910 u64 cur_end;
865 int limit = 10 * 1024 * 1042; 911 int limit = 10 * 1024 * 1042;
866 912
867 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED | 913 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
868 EXTENT_DELALLOC, 1, 0, NULL, GFP_NOFS); 914 1, 0, NULL, GFP_NOFS);
869 while (start < end) { 915 while (start < end) {
870 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); 916 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
871 async_cow->inode = inode; 917 async_cow->inode = inode;
@@ -1006,6 +1052,7 @@ next_slot:
1006 1052
1007 if (found_key.offset > cur_offset) { 1053 if (found_key.offset > cur_offset) {
1008 extent_end = found_key.offset; 1054 extent_end = found_key.offset;
1055 extent_type = 0;
1009 goto out_check; 1056 goto out_check;
1010 } 1057 }
1011 1058
@@ -1112,8 +1159,10 @@ out_check:
1112 BUG_ON(ret); 1159 BUG_ON(ret);
1113 1160
1114 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, 1161 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1115 cur_offset, cur_offset + num_bytes - 1, 1162 cur_offset, cur_offset + num_bytes - 1,
1116 locked_page, 1, 1, 1, 0, 0, 0, 1); 1163 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1164 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1165 EXTENT_SET_PRIVATE2);
1117 cur_offset = extent_end; 1166 cur_offset = extent_end;
1118 if (cur_offset > end) 1167 if (cur_offset > end)
1119 break; 1168 break;
@@ -1178,15 +1227,17 @@ static int btrfs_split_extent_hook(struct inode *inode,
1178 root->fs_info->max_extent); 1227 root->fs_info->max_extent);
1179 1228
1180 /* 1229 /*
1181 * if we break a large extent up then leave delalloc_extents be, 1230 * if we break a large extent up then leave oustanding_extents
1182 * since we've already accounted for the large extent. 1231 * be, since we've already accounted for the large extent.
1183 */ 1232 */
1184 if (div64_u64(new_size + root->fs_info->max_extent - 1, 1233 if (div64_u64(new_size + root->fs_info->max_extent - 1,
1185 root->fs_info->max_extent) < num_extents) 1234 root->fs_info->max_extent) < num_extents)
1186 return 0; 1235 return 0;
1187 } 1236 }
1188 1237
1189 BTRFS_I(inode)->delalloc_extents++; 1238 spin_lock(&BTRFS_I(inode)->accounting_lock);
1239 BTRFS_I(inode)->outstanding_extents++;
1240 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1190 1241
1191 return 0; 1242 return 0;
1192} 1243}
@@ -1217,7 +1268,9 @@ static int btrfs_merge_extent_hook(struct inode *inode,
1217 1268
1218 /* we're not bigger than the max, unreserve the space and go */ 1269 /* we're not bigger than the max, unreserve the space and go */
1219 if (new_size <= root->fs_info->max_extent) { 1270 if (new_size <= root->fs_info->max_extent) {
1220 BTRFS_I(inode)->delalloc_extents--; 1271 spin_lock(&BTRFS_I(inode)->accounting_lock);
1272 BTRFS_I(inode)->outstanding_extents--;
1273 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1221 return 0; 1274 return 0;
1222 } 1275 }
1223 1276
@@ -1231,7 +1284,9 @@ static int btrfs_merge_extent_hook(struct inode *inode,
1231 root->fs_info->max_extent) > num_extents) 1284 root->fs_info->max_extent) > num_extents)
1232 return 0; 1285 return 0;
1233 1286
1234 BTRFS_I(inode)->delalloc_extents--; 1287 spin_lock(&BTRFS_I(inode)->accounting_lock);
1288 BTRFS_I(inode)->outstanding_extents--;
1289 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1235 1290
1236 return 0; 1291 return 0;
1237} 1292}
@@ -1253,7 +1308,9 @@ static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1253 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { 1308 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1254 struct btrfs_root *root = BTRFS_I(inode)->root; 1309 struct btrfs_root *root = BTRFS_I(inode)->root;
1255 1310
1256 BTRFS_I(inode)->delalloc_extents++; 1311 spin_lock(&BTRFS_I(inode)->accounting_lock);
1312 BTRFS_I(inode)->outstanding_extents++;
1313 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1257 btrfs_delalloc_reserve_space(root, inode, end - start + 1); 1314 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1258 spin_lock(&root->fs_info->delalloc_lock); 1315 spin_lock(&root->fs_info->delalloc_lock);
1259 BTRFS_I(inode)->delalloc_bytes += end - start + 1; 1316 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
@@ -1281,8 +1338,12 @@ static int btrfs_clear_bit_hook(struct inode *inode,
1281 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { 1338 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1282 struct btrfs_root *root = BTRFS_I(inode)->root; 1339 struct btrfs_root *root = BTRFS_I(inode)->root;
1283 1340
1284 BTRFS_I(inode)->delalloc_extents--; 1341 if (bits & EXTENT_DO_ACCOUNTING) {
1285 btrfs_unreserve_metadata_for_delalloc(root, inode, 1); 1342 spin_lock(&BTRFS_I(inode)->accounting_lock);
1343 BTRFS_I(inode)->outstanding_extents--;
1344 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1345 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
1346 }
1286 1347
1287 spin_lock(&root->fs_info->delalloc_lock); 1348 spin_lock(&root->fs_info->delalloc_lock);
1288 if (state->end - state->start + 1 > 1349 if (state->end - state->start + 1 >
@@ -2442,7 +2503,19 @@ static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2442 2503
2443 root = BTRFS_I(dir)->root; 2504 root = BTRFS_I(dir)->root;
2444 2505
2506 /*
2507 * 5 items for unlink inode
2508 * 1 for orphan
2509 */
2510 ret = btrfs_reserve_metadata_space(root, 6);
2511 if (ret)
2512 return ret;
2513
2445 trans = btrfs_start_transaction(root, 1); 2514 trans = btrfs_start_transaction(root, 1);
2515 if (IS_ERR(trans)) {
2516 btrfs_unreserve_metadata_space(root, 6);
2517 return PTR_ERR(trans);
2518 }
2446 2519
2447 btrfs_set_trans_block_group(trans, dir); 2520 btrfs_set_trans_block_group(trans, dir);
2448 2521
@@ -2457,6 +2530,7 @@ static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2457 nr = trans->blocks_used; 2530 nr = trans->blocks_used;
2458 2531
2459 btrfs_end_transaction_throttle(trans, root); 2532 btrfs_end_transaction_throttle(trans, root);
2533 btrfs_unreserve_metadata_space(root, 6);
2460 btrfs_btree_balance_dirty(root, nr); 2534 btrfs_btree_balance_dirty(root, nr);
2461 return ret; 2535 return ret;
2462} 2536}
@@ -2537,7 +2611,16 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2537 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) 2611 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
2538 return -ENOTEMPTY; 2612 return -ENOTEMPTY;
2539 2613
2614 ret = btrfs_reserve_metadata_space(root, 5);
2615 if (ret)
2616 return ret;
2617
2540 trans = btrfs_start_transaction(root, 1); 2618 trans = btrfs_start_transaction(root, 1);
2619 if (IS_ERR(trans)) {
2620 btrfs_unreserve_metadata_space(root, 5);
2621 return PTR_ERR(trans);
2622 }
2623
2541 btrfs_set_trans_block_group(trans, dir); 2624 btrfs_set_trans_block_group(trans, dir);
2542 2625
2543 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { 2626 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
@@ -2560,6 +2643,7 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2560out: 2643out:
2561 nr = trans->blocks_used; 2644 nr = trans->blocks_used;
2562 ret = btrfs_end_transaction_throttle(trans, root); 2645 ret = btrfs_end_transaction_throttle(trans, root);
2646 btrfs_unreserve_metadata_space(root, 5);
2563 btrfs_btree_balance_dirty(root, nr); 2647 btrfs_btree_balance_dirty(root, nr);
2564 2648
2565 if (ret && !err) 2649 if (ret && !err)
@@ -3000,12 +3084,22 @@ static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3000 3084
3001 if ((offset & (blocksize - 1)) == 0) 3085 if ((offset & (blocksize - 1)) == 0)
3002 goto out; 3086 goto out;
3087 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
3088 if (ret)
3089 goto out;
3090
3091 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
3092 if (ret)
3093 goto out;
3003 3094
3004 ret = -ENOMEM; 3095 ret = -ENOMEM;
3005again: 3096again:
3006 page = grab_cache_page(mapping, index); 3097 page = grab_cache_page(mapping, index);
3007 if (!page) 3098 if (!page) {
3099 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
3100 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
3008 goto out; 3101 goto out;
3102 }
3009 3103
3010 page_start = page_offset(page); 3104 page_start = page_offset(page);
3011 page_end = page_start + PAGE_CACHE_SIZE - 1; 3105 page_end = page_start + PAGE_CACHE_SIZE - 1;
@@ -3038,6 +3132,10 @@ again:
3038 goto again; 3132 goto again;
3039 } 3133 }
3040 3134
3135 clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
3136 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3137 GFP_NOFS);
3138
3041 ret = btrfs_set_extent_delalloc(inode, page_start, page_end); 3139 ret = btrfs_set_extent_delalloc(inode, page_start, page_end);
3042 if (ret) { 3140 if (ret) {
3043 unlock_extent(io_tree, page_start, page_end, GFP_NOFS); 3141 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
@@ -3056,6 +3154,9 @@ again:
3056 unlock_extent(io_tree, page_start, page_end, GFP_NOFS); 3154 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3057 3155
3058out_unlock: 3156out_unlock:
3157 if (ret)
3158 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
3159 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
3059 unlock_page(page); 3160 unlock_page(page);
3060 page_cache_release(page); 3161 page_cache_release(page);
3061out: 3162out:
@@ -3079,7 +3180,9 @@ int btrfs_cont_expand(struct inode *inode, loff_t size)
3079 if (size <= hole_start) 3180 if (size <= hole_start)
3080 return 0; 3181 return 0;
3081 3182
3082 btrfs_truncate_page(inode->i_mapping, inode->i_size); 3183 err = btrfs_truncate_page(inode->i_mapping, inode->i_size);
3184 if (err)
3185 return err;
3083 3186
3084 while (1) { 3187 while (1) {
3085 struct btrfs_ordered_extent *ordered; 3188 struct btrfs_ordered_extent *ordered;
@@ -3448,6 +3551,7 @@ static noinline void init_btrfs_i(struct inode *inode)
3448 bi->generation = 0; 3551 bi->generation = 0;
3449 bi->sequence = 0; 3552 bi->sequence = 0;
3450 bi->last_trans = 0; 3553 bi->last_trans = 0;
3554 bi->last_sub_trans = 0;
3451 bi->logged_trans = 0; 3555 bi->logged_trans = 0;
3452 bi->delalloc_bytes = 0; 3556 bi->delalloc_bytes = 0;
3453 bi->reserved_bytes = 0; 3557 bi->reserved_bytes = 0;
@@ -3598,12 +3702,14 @@ static int btrfs_dentry_delete(struct dentry *dentry)
3598{ 3702{
3599 struct btrfs_root *root; 3703 struct btrfs_root *root;
3600 3704
3601 if (!dentry->d_inode) 3705 if (!dentry->d_inode && !IS_ROOT(dentry))
3602 return 0; 3706 dentry = dentry->d_parent;
3603 3707
3604 root = BTRFS_I(dentry->d_inode)->root; 3708 if (dentry->d_inode) {
3605 if (btrfs_root_refs(&root->root_item) == 0) 3709 root = BTRFS_I(dentry->d_inode)->root;
3606 return 1; 3710 if (btrfs_root_refs(&root->root_item) == 0)
3711 return 1;
3712 }
3607 return 0; 3713 return 0;
3608} 3714}
3609 3715
@@ -4808,7 +4914,8 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4808 */ 4914 */
4809 clear_extent_bit(tree, page_start, page_end, 4915 clear_extent_bit(tree, page_start, page_end,
4810 EXTENT_DIRTY | EXTENT_DELALLOC | 4916 EXTENT_DIRTY | EXTENT_DELALLOC |
4811 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS); 4917 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
4918 NULL, GFP_NOFS);
4812 /* 4919 /*
4813 * whoever cleared the private bit is responsible 4920 * whoever cleared the private bit is responsible
4814 * for the finish_ordered_io 4921 * for the finish_ordered_io
@@ -4821,8 +4928,8 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4821 lock_extent(tree, page_start, page_end, GFP_NOFS); 4928 lock_extent(tree, page_start, page_end, GFP_NOFS);
4822 } 4929 }
4823 clear_extent_bit(tree, page_start, page_end, 4930 clear_extent_bit(tree, page_start, page_end,
4824 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC, 4931 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4825 1, 1, NULL, GFP_NOFS); 4932 EXTENT_DO_ACCOUNTING, 1, 1, NULL, GFP_NOFS);
4826 __btrfs_releasepage(page, GFP_NOFS); 4933 __btrfs_releasepage(page, GFP_NOFS);
4827 4934
4828 ClearPageChecked(page); 4935 ClearPageChecked(page);
@@ -4917,7 +5024,8 @@ again:
4917 * prepare_pages in the normal write path. 5024 * prepare_pages in the normal write path.
4918 */ 5025 */
4919 clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 5026 clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
4920 EXTENT_DIRTY | EXTENT_DELALLOC, GFP_NOFS); 5027 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
5028 GFP_NOFS);
4921 5029
4922 ret = btrfs_set_extent_delalloc(inode, page_start, page_end); 5030 ret = btrfs_set_extent_delalloc(inode, page_start, page_end);
4923 if (ret) { 5031 if (ret) {
@@ -4944,7 +5052,9 @@ again:
4944 set_page_dirty(page); 5052 set_page_dirty(page);
4945 SetPageUptodate(page); 5053 SetPageUptodate(page);
4946 5054
4947 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1; 5055 BTRFS_I(inode)->last_trans = root->fs_info->generation;
5056 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
5057
4948 unlock_extent(io_tree, page_start, page_end, GFP_NOFS); 5058 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4949 5059
4950out_unlock: 5060out_unlock:
@@ -4969,7 +5079,9 @@ static void btrfs_truncate(struct inode *inode)
4969 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 5079 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4970 return; 5080 return;
4971 5081
4972 btrfs_truncate_page(inode->i_mapping, inode->i_size); 5082 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
5083 if (ret)
5084 return;
4973 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); 5085 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4974 5086
4975 trans = btrfs_start_transaction(root, 1); 5087 trans = btrfs_start_transaction(root, 1);
@@ -5064,9 +5176,12 @@ struct inode *btrfs_alloc_inode(struct super_block *sb)
5064 if (!ei) 5176 if (!ei)
5065 return NULL; 5177 return NULL;
5066 ei->last_trans = 0; 5178 ei->last_trans = 0;
5179 ei->last_sub_trans = 0;
5067 ei->logged_trans = 0; 5180 ei->logged_trans = 0;
5068 ei->delalloc_extents = 0; 5181 ei->outstanding_extents = 0;
5069 ei->delalloc_reserved_extents = 0; 5182 ei->reserved_extents = 0;
5183 ei->root = NULL;
5184 spin_lock_init(&ei->accounting_lock);
5070 btrfs_ordered_inode_tree_init(&ei->ordered_tree); 5185 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
5071 INIT_LIST_HEAD(&ei->i_orphan); 5186 INIT_LIST_HEAD(&ei->i_orphan);
5072 INIT_LIST_HEAD(&ei->ordered_operations); 5187 INIT_LIST_HEAD(&ei->ordered_operations);
@@ -5082,6 +5197,14 @@ void btrfs_destroy_inode(struct inode *inode)
5082 WARN_ON(inode->i_data.nrpages); 5197 WARN_ON(inode->i_data.nrpages);
5083 5198
5084 /* 5199 /*
5200 * This can happen where we create an inode, but somebody else also
5201 * created the same inode and we need to destroy the one we already
5202 * created.
5203 */
5204 if (!root)
5205 goto free;
5206
5207 /*
5085 * Make sure we're properly removed from the ordered operation 5208 * Make sure we're properly removed from the ordered operation
5086 * lists. 5209 * lists.
5087 */ 5210 */
@@ -5116,6 +5239,7 @@ void btrfs_destroy_inode(struct inode *inode)
5116 } 5239 }
5117 inode_tree_del(inode); 5240 inode_tree_del(inode);
5118 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); 5241 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
5242free:
5119 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); 5243 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5120} 5244}
5121 5245
@@ -5221,11 +5345,14 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
5221 return -ENOTEMPTY; 5345 return -ENOTEMPTY;
5222 5346
5223 /* 5347 /*
5224 * 2 items for dir items 5348 * We want to reserve the absolute worst case amount of items. So if
5225 * 1 item for orphan entry 5349 * both inodes are subvols and we need to unlink them then that would
5226 * 1 item for ref 5350 * require 4 item modifications, but if they are both normal inodes it
5351 * would require 5 item modifications, so we'll assume their normal
5352 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
5353 * should cover the worst case number of items we'll modify.
5227 */ 5354 */
5228 ret = btrfs_reserve_metadata_space(root, 4); 5355 ret = btrfs_reserve_metadata_space(root, 11);
5229 if (ret) 5356 if (ret)
5230 return ret; 5357 return ret;
5231 5358
@@ -5341,7 +5468,7 @@ out_fail:
5341 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) 5468 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5342 up_read(&root->fs_info->subvol_sem); 5469 up_read(&root->fs_info->subvol_sem);
5343 5470
5344 btrfs_unreserve_metadata_space(root, 4); 5471 btrfs_unreserve_metadata_space(root, 11);
5345 return ret; 5472 return ret;
5346} 5473}
5347 5474
@@ -5805,6 +5932,6 @@ static const struct inode_operations btrfs_symlink_inode_operations = {
5805 .removexattr = btrfs_removexattr, 5932 .removexattr = btrfs_removexattr,
5806}; 5933};
5807 5934
5808struct dentry_operations btrfs_dentry_operations = { 5935const struct dentry_operations btrfs_dentry_operations = {
5809 .d_delete = btrfs_dentry_delete, 5936 .d_delete = btrfs_dentry_delete,
5810}; 5937};
diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 9a780c8d0ac8..cdbb054102b9 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -830,6 +830,7 @@ out_up_write:
830out_unlock: 830out_unlock:
831 mutex_unlock(&inode->i_mutex); 831 mutex_unlock(&inode->i_mutex);
832 if (!err) { 832 if (!err) {
833 shrink_dcache_sb(root->fs_info->sb);
833 btrfs_invalidate_inodes(dest); 834 btrfs_invalidate_inodes(dest);
834 d_delete(dentry); 835 d_delete(dentry);
835 } 836 }
@@ -1122,8 +1123,10 @@ static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
1122 datao += off - key.offset; 1123 datao += off - key.offset;
1123 datal -= off - key.offset; 1124 datal -= off - key.offset;
1124 } 1125 }
1125 if (key.offset + datao + datal > off + len) 1126
1126 datal = off + len - key.offset - datao; 1127 if (key.offset + datal > off + len)
1128 datal = off + len - key.offset;
1129
1127 /* disko == 0 means it's a hole */ 1130 /* disko == 0 means it's a hole */
1128 if (!disko) 1131 if (!disko)
1129 datao = 0; 1132 datao = 0;
diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c
index 897fba835f89..5799bc46a309 100644
--- a/fs/btrfs/ordered-data.c
+++ b/fs/btrfs/ordered-data.c
@@ -306,6 +306,12 @@ int btrfs_remove_ordered_extent(struct inode *inode,
306 tree->last = NULL; 306 tree->last = NULL;
307 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags); 307 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
308 308
309 spin_lock(&BTRFS_I(inode)->accounting_lock);
310 BTRFS_I(inode)->outstanding_extents--;
311 spin_unlock(&BTRFS_I(inode)->accounting_lock);
312 btrfs_unreserve_metadata_for_delalloc(BTRFS_I(inode)->root,
313 inode, 1);
314
309 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); 315 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
310 list_del_init(&entry->root_extent_list); 316 list_del_init(&entry->root_extent_list);
311 317
diff --git a/fs/btrfs/relocation.c b/fs/btrfs/relocation.c
index 361ad323faac..cfcc93c93a7b 100644
--- a/fs/btrfs/relocation.c
+++ b/fs/btrfs/relocation.c
@@ -3518,7 +3518,7 @@ int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
3518 BUG_ON(!rc->block_group); 3518 BUG_ON(!rc->block_group);
3519 3519
3520 btrfs_init_workers(&rc->workers, "relocate", 3520 btrfs_init_workers(&rc->workers, "relocate",
3521 fs_info->thread_pool_size); 3521 fs_info->thread_pool_size, NULL);
3522 3522
3523 rc->extent_root = extent_root; 3523 rc->extent_root = extent_root;
3524 btrfs_prepare_block_group_relocation(extent_root, rc->block_group); 3524 btrfs_prepare_block_group_relocation(extent_root, rc->block_group);
@@ -3701,7 +3701,7 @@ int btrfs_recover_relocation(struct btrfs_root *root)
3701 mapping_tree_init(&rc->reloc_root_tree); 3701 mapping_tree_init(&rc->reloc_root_tree);
3702 INIT_LIST_HEAD(&rc->reloc_roots); 3702 INIT_LIST_HEAD(&rc->reloc_roots);
3703 btrfs_init_workers(&rc->workers, "relocate", 3703 btrfs_init_workers(&rc->workers, "relocate",
3704 root->fs_info->thread_pool_size); 3704 root->fs_info->thread_pool_size, NULL);
3705 rc->extent_root = root->fs_info->extent_root; 3705 rc->extent_root = root->fs_info->extent_root;
3706 3706
3707 set_reloc_control(rc); 3707 set_reloc_control(rc);
diff --git a/fs/btrfs/root-tree.c b/fs/btrfs/root-tree.c
index 9351428f30e2..67fa2d29d663 100644
--- a/fs/btrfs/root-tree.c
+++ b/fs/btrfs/root-tree.c
@@ -159,7 +159,6 @@ int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
159 write_extent_buffer(l, item, ptr, sizeof(*item)); 159 write_extent_buffer(l, item, ptr, sizeof(*item));
160 btrfs_mark_buffer_dirty(path->nodes[0]); 160 btrfs_mark_buffer_dirty(path->nodes[0]);
161out: 161out:
162 btrfs_release_path(root, path);
163 btrfs_free_path(path); 162 btrfs_free_path(path);
164 return ret; 163 return ret;
165} 164}
@@ -332,7 +331,6 @@ int btrfs_del_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
332 BUG_ON(refs != 0); 331 BUG_ON(refs != 0);
333 ret = btrfs_del_item(trans, root, path); 332 ret = btrfs_del_item(trans, root, path);
334out: 333out:
335 btrfs_release_path(root, path);
336 btrfs_free_path(path); 334 btrfs_free_path(path);
337 return ret; 335 return ret;
338} 336}
diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
index 9de9b2236419..752a5463bf53 100644
--- a/fs/btrfs/super.c
+++ b/fs/btrfs/super.c
@@ -66,7 +66,8 @@ enum {
66 Opt_degraded, Opt_subvol, Opt_device, Opt_nodatasum, Opt_nodatacow, 66 Opt_degraded, Opt_subvol, Opt_device, Opt_nodatasum, Opt_nodatacow,
67 Opt_max_extent, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, 67 Opt_max_extent, Opt_max_inline, Opt_alloc_start, Opt_nobarrier,
68 Opt_ssd, Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, 68 Opt_ssd, Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl,
69 Opt_compress, Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_err, 69 Opt_compress, Opt_notreelog, Opt_ratio, Opt_flushoncommit,
70 Opt_discard, Opt_err,
70}; 71};
71 72
72static match_table_t tokens = { 73static match_table_t tokens = {
@@ -88,6 +89,7 @@ static match_table_t tokens = {
88 {Opt_notreelog, "notreelog"}, 89 {Opt_notreelog, "notreelog"},
89 {Opt_flushoncommit, "flushoncommit"}, 90 {Opt_flushoncommit, "flushoncommit"},
90 {Opt_ratio, "metadata_ratio=%d"}, 91 {Opt_ratio, "metadata_ratio=%d"},
92 {Opt_discard, "discard"},
91 {Opt_err, NULL}, 93 {Opt_err, NULL},
92}; 94};
93 95
@@ -257,6 +259,9 @@ int btrfs_parse_options(struct btrfs_root *root, char *options)
257 info->metadata_ratio); 259 info->metadata_ratio);
258 } 260 }
259 break; 261 break;
262 case Opt_discard:
263 btrfs_set_opt(info->mount_opt, DISCARD);
264 break;
260 default: 265 default:
261 break; 266 break;
262 } 267 }
@@ -344,7 +349,7 @@ static int btrfs_fill_super(struct super_block *sb,
344 sb->s_export_op = &btrfs_export_ops; 349 sb->s_export_op = &btrfs_export_ops;
345 sb->s_xattr = btrfs_xattr_handlers; 350 sb->s_xattr = btrfs_xattr_handlers;
346 sb->s_time_gran = 1; 351 sb->s_time_gran = 1;
347#ifdef CONFIG_BTRFS_POSIX_ACL 352#ifdef CONFIG_BTRFS_FS_POSIX_ACL
348 sb->s_flags |= MS_POSIXACL; 353 sb->s_flags |= MS_POSIXACL;
349#endif 354#endif
350 355
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 0b8f36d4400a..c207e8c32c9b 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -163,8 +163,14 @@ static void wait_current_trans(struct btrfs_root *root)
163 } 163 }
164} 164}
165 165
166enum btrfs_trans_type {
167 TRANS_START,
168 TRANS_JOIN,
169 TRANS_USERSPACE,
170};
171
166static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, 172static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
167 int num_blocks, int wait) 173 int num_blocks, int type)
168{ 174{
169 struct btrfs_trans_handle *h = 175 struct btrfs_trans_handle *h =
170 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); 176 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
@@ -172,7 +178,8 @@ static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
172 178
173 mutex_lock(&root->fs_info->trans_mutex); 179 mutex_lock(&root->fs_info->trans_mutex);
174 if (!root->fs_info->log_root_recovering && 180 if (!root->fs_info->log_root_recovering &&
175 ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2)) 181 ((type == TRANS_START && !root->fs_info->open_ioctl_trans) ||
182 type == TRANS_USERSPACE))
176 wait_current_trans(root); 183 wait_current_trans(root);
177 ret = join_transaction(root); 184 ret = join_transaction(root);
178 BUG_ON(ret); 185 BUG_ON(ret);
@@ -186,7 +193,7 @@ static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
186 h->alloc_exclude_start = 0; 193 h->alloc_exclude_start = 0;
187 h->delayed_ref_updates = 0; 194 h->delayed_ref_updates = 0;
188 195
189 if (!current->journal_info) 196 if (!current->journal_info && type != TRANS_USERSPACE)
190 current->journal_info = h; 197 current->journal_info = h;
191 198
192 root->fs_info->running_transaction->use_count++; 199 root->fs_info->running_transaction->use_count++;
@@ -198,18 +205,18 @@ static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
198struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 205struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
199 int num_blocks) 206 int num_blocks)
200{ 207{
201 return start_transaction(root, num_blocks, 1); 208 return start_transaction(root, num_blocks, TRANS_START);
202} 209}
203struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, 210struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
204 int num_blocks) 211 int num_blocks)
205{ 212{
206 return start_transaction(root, num_blocks, 0); 213 return start_transaction(root, num_blocks, TRANS_JOIN);
207} 214}
208 215
209struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, 216struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
210 int num_blocks) 217 int num_blocks)
211{ 218{
212 return start_transaction(r, num_blocks, 2); 219 return start_transaction(r, num_blocks, TRANS_USERSPACE);
213} 220}
214 221
215/* wait for a transaction commit to be fully complete */ 222/* wait for a transaction commit to be fully complete */
@@ -344,10 +351,10 @@ int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
344/* 351/*
345 * when btree blocks are allocated, they have some corresponding bits set for 352 * when btree blocks are allocated, they have some corresponding bits set for
346 * them in one of two extent_io trees. This is used to make sure all of 353 * them in one of two extent_io trees. This is used to make sure all of
347 * those extents are on disk for transaction or log commit 354 * those extents are sent to disk but does not wait on them
348 */ 355 */
349int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, 356int btrfs_write_marked_extents(struct btrfs_root *root,
350 struct extent_io_tree *dirty_pages) 357 struct extent_io_tree *dirty_pages)
351{ 358{
352 int ret; 359 int ret;
353 int err = 0; 360 int err = 0;
@@ -394,6 +401,29 @@ int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
394 page_cache_release(page); 401 page_cache_release(page);
395 } 402 }
396 } 403 }
404 if (err)
405 werr = err;
406 return werr;
407}
408
409/*
410 * when btree blocks are allocated, they have some corresponding bits set for
411 * them in one of two extent_io trees. This is used to make sure all of
412 * those extents are on disk for transaction or log commit. We wait
413 * on all the pages and clear them from the dirty pages state tree
414 */
415int btrfs_wait_marked_extents(struct btrfs_root *root,
416 struct extent_io_tree *dirty_pages)
417{
418 int ret;
419 int err = 0;
420 int werr = 0;
421 struct page *page;
422 struct inode *btree_inode = root->fs_info->btree_inode;
423 u64 start = 0;
424 u64 end;
425 unsigned long index;
426
397 while (1) { 427 while (1) {
398 ret = find_first_extent_bit(dirty_pages, 0, &start, &end, 428 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
399 EXTENT_DIRTY); 429 EXTENT_DIRTY);
@@ -424,6 +454,22 @@ int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
424 return werr; 454 return werr;
425} 455}
426 456
457/*
458 * when btree blocks are allocated, they have some corresponding bits set for
459 * them in one of two extent_io trees. This is used to make sure all of
460 * those extents are on disk for transaction or log commit
461 */
462int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
463 struct extent_io_tree *dirty_pages)
464{
465 int ret;
466 int ret2;
467
468 ret = btrfs_write_marked_extents(root, dirty_pages);
469 ret2 = btrfs_wait_marked_extents(root, dirty_pages);
470 return ret || ret2;
471}
472
427int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, 473int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
428 struct btrfs_root *root) 474 struct btrfs_root *root)
429{ 475{
diff --git a/fs/btrfs/transaction.h b/fs/btrfs/transaction.h
index 663c67404918..d4e3e7a6938c 100644
--- a/fs/btrfs/transaction.h
+++ b/fs/btrfs/transaction.h
@@ -79,6 +79,7 @@ static inline void btrfs_set_inode_last_trans(struct btrfs_trans_handle *trans,
79 struct inode *inode) 79 struct inode *inode)
80{ 80{
81 BTRFS_I(inode)->last_trans = trans->transaction->transid; 81 BTRFS_I(inode)->last_trans = trans->transaction->transid;
82 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
82} 83}
83 84
84int btrfs_end_transaction(struct btrfs_trans_handle *trans, 85int btrfs_end_transaction(struct btrfs_trans_handle *trans,
@@ -107,5 +108,9 @@ int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
107 struct btrfs_root *root); 108 struct btrfs_root *root);
108int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, 109int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
109 struct extent_io_tree *dirty_pages); 110 struct extent_io_tree *dirty_pages);
111int btrfs_write_marked_extents(struct btrfs_root *root,
112 struct extent_io_tree *dirty_pages);
113int btrfs_wait_marked_extents(struct btrfs_root *root,
114 struct extent_io_tree *dirty_pages);
110int btrfs_transaction_in_commit(struct btrfs_fs_info *info); 115int btrfs_transaction_in_commit(struct btrfs_fs_info *info);
111#endif 116#endif
diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c
index 7827841b55cb..741666a7676a 100644
--- a/fs/btrfs/tree-log.c
+++ b/fs/btrfs/tree-log.c
@@ -137,11 +137,20 @@ static int start_log_trans(struct btrfs_trans_handle *trans,
137 137
138 mutex_lock(&root->log_mutex); 138 mutex_lock(&root->log_mutex);
139 if (root->log_root) { 139 if (root->log_root) {
140 if (!root->log_start_pid) {
141 root->log_start_pid = current->pid;
142 root->log_multiple_pids = false;
143 } else if (root->log_start_pid != current->pid) {
144 root->log_multiple_pids = true;
145 }
146
140 root->log_batch++; 147 root->log_batch++;
141 atomic_inc(&root->log_writers); 148 atomic_inc(&root->log_writers);
142 mutex_unlock(&root->log_mutex); 149 mutex_unlock(&root->log_mutex);
143 return 0; 150 return 0;
144 } 151 }
152 root->log_multiple_pids = false;
153 root->log_start_pid = current->pid;
145 mutex_lock(&root->fs_info->tree_log_mutex); 154 mutex_lock(&root->fs_info->tree_log_mutex);
146 if (!root->fs_info->log_root_tree) { 155 if (!root->fs_info->log_root_tree) {
147 ret = btrfs_init_log_root_tree(trans, root->fs_info); 156 ret = btrfs_init_log_root_tree(trans, root->fs_info);
@@ -1971,6 +1980,7 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
1971 int ret; 1980 int ret;
1972 struct btrfs_root *log = root->log_root; 1981 struct btrfs_root *log = root->log_root;
1973 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree; 1982 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
1983 u64 log_transid = 0;
1974 1984
1975 mutex_lock(&root->log_mutex); 1985 mutex_lock(&root->log_mutex);
1976 index1 = root->log_transid % 2; 1986 index1 = root->log_transid % 2;
@@ -1987,10 +1997,11 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
1987 1997
1988 while (1) { 1998 while (1) {
1989 unsigned long batch = root->log_batch; 1999 unsigned long batch = root->log_batch;
1990 mutex_unlock(&root->log_mutex); 2000 if (root->log_multiple_pids) {
1991 schedule_timeout_uninterruptible(1); 2001 mutex_unlock(&root->log_mutex);
1992 mutex_lock(&root->log_mutex); 2002 schedule_timeout_uninterruptible(1);
1993 2003 mutex_lock(&root->log_mutex);
2004 }
1994 wait_for_writer(trans, root); 2005 wait_for_writer(trans, root);
1995 if (batch == root->log_batch) 2006 if (batch == root->log_batch)
1996 break; 2007 break;
@@ -2003,14 +2014,19 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
2003 goto out; 2014 goto out;
2004 } 2015 }
2005 2016
2006 ret = btrfs_write_and_wait_marked_extents(log, &log->dirty_log_pages); 2017 /* we start IO on all the marked extents here, but we don't actually
2018 * wait for them until later.
2019 */
2020 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages);
2007 BUG_ON(ret); 2021 BUG_ON(ret);
2008 2022
2009 btrfs_set_root_node(&log->root_item, log->node); 2023 btrfs_set_root_node(&log->root_item, log->node);
2010 2024
2011 root->log_batch = 0; 2025 root->log_batch = 0;
2026 log_transid = root->log_transid;
2012 root->log_transid++; 2027 root->log_transid++;
2013 log->log_transid = root->log_transid; 2028 log->log_transid = root->log_transid;
2029 root->log_start_pid = 0;
2014 smp_mb(); 2030 smp_mb();
2015 /* 2031 /*
2016 * log tree has been flushed to disk, new modifications of 2032 * log tree has been flushed to disk, new modifications of
@@ -2036,6 +2052,7 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
2036 2052
2037 index2 = log_root_tree->log_transid % 2; 2053 index2 = log_root_tree->log_transid % 2;
2038 if (atomic_read(&log_root_tree->log_commit[index2])) { 2054 if (atomic_read(&log_root_tree->log_commit[index2])) {
2055 btrfs_wait_marked_extents(log, &log->dirty_log_pages);
2039 wait_log_commit(trans, log_root_tree, 2056 wait_log_commit(trans, log_root_tree,
2040 log_root_tree->log_transid); 2057 log_root_tree->log_transid);
2041 mutex_unlock(&log_root_tree->log_mutex); 2058 mutex_unlock(&log_root_tree->log_mutex);
@@ -2055,6 +2072,7 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
2055 * check the full commit flag again 2072 * check the full commit flag again
2056 */ 2073 */
2057 if (root->fs_info->last_trans_log_full_commit == trans->transid) { 2074 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2075 btrfs_wait_marked_extents(log, &log->dirty_log_pages);
2058 mutex_unlock(&log_root_tree->log_mutex); 2076 mutex_unlock(&log_root_tree->log_mutex);
2059 ret = -EAGAIN; 2077 ret = -EAGAIN;
2060 goto out_wake_log_root; 2078 goto out_wake_log_root;
@@ -2063,6 +2081,7 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
2063 ret = btrfs_write_and_wait_marked_extents(log_root_tree, 2081 ret = btrfs_write_and_wait_marked_extents(log_root_tree,
2064 &log_root_tree->dirty_log_pages); 2082 &log_root_tree->dirty_log_pages);
2065 BUG_ON(ret); 2083 BUG_ON(ret);
2084 btrfs_wait_marked_extents(log, &log->dirty_log_pages);
2066 2085
2067 btrfs_set_super_log_root(&root->fs_info->super_for_commit, 2086 btrfs_set_super_log_root(&root->fs_info->super_for_commit,
2068 log_root_tree->node->start); 2087 log_root_tree->node->start);
@@ -2082,9 +2101,14 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
2082 * the running transaction open, so a full commit can't hop 2101 * the running transaction open, so a full commit can't hop
2083 * in and cause problems either. 2102 * in and cause problems either.
2084 */ 2103 */
2085 write_ctree_super(trans, root->fs_info->tree_root, 2); 2104 write_ctree_super(trans, root->fs_info->tree_root, 1);
2086 ret = 0; 2105 ret = 0;
2087 2106
2107 mutex_lock(&root->log_mutex);
2108 if (root->last_log_commit < log_transid)
2109 root->last_log_commit = log_transid;
2110 mutex_unlock(&root->log_mutex);
2111
2088out_wake_log_root: 2112out_wake_log_root:
2089 atomic_set(&log_root_tree->log_commit[index2], 0); 2113 atomic_set(&log_root_tree->log_commit[index2], 0);
2090 smp_mb(); 2114 smp_mb();
@@ -2852,6 +2876,21 @@ out:
2852 return ret; 2876 return ret;
2853} 2877}
2854 2878
2879static int inode_in_log(struct btrfs_trans_handle *trans,
2880 struct inode *inode)
2881{
2882 struct btrfs_root *root = BTRFS_I(inode)->root;
2883 int ret = 0;
2884
2885 mutex_lock(&root->log_mutex);
2886 if (BTRFS_I(inode)->logged_trans == trans->transid &&
2887 BTRFS_I(inode)->last_sub_trans <= root->last_log_commit)
2888 ret = 1;
2889 mutex_unlock(&root->log_mutex);
2890 return ret;
2891}
2892
2893
2855/* 2894/*
2856 * helper function around btrfs_log_inode to make sure newly created 2895 * helper function around btrfs_log_inode to make sure newly created
2857 * parent directories also end up in the log. A minimal inode and backref 2896 * parent directories also end up in the log. A minimal inode and backref
@@ -2891,6 +2930,11 @@ int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
2891 if (ret) 2930 if (ret)
2892 goto end_no_trans; 2931 goto end_no_trans;
2893 2932
2933 if (inode_in_log(trans, inode)) {
2934 ret = BTRFS_NO_LOG_SYNC;
2935 goto end_no_trans;
2936 }
2937
2894 start_log_trans(trans, root); 2938 start_log_trans(trans, root);
2895 2939
2896 ret = btrfs_log_inode(trans, root, inode, inode_only); 2940 ret = btrfs_log_inode(trans, root, inode, inode_only);
diff --git a/fs/btrfs/tree-log.h b/fs/btrfs/tree-log.h
index d09c7609e16b..0776eacb5083 100644
--- a/fs/btrfs/tree-log.h
+++ b/fs/btrfs/tree-log.h
@@ -19,6 +19,9 @@
19#ifndef __TREE_LOG_ 19#ifndef __TREE_LOG_
20#define __TREE_LOG_ 20#define __TREE_LOG_
21 21
22/* return value for btrfs_log_dentry_safe that means we don't need to log it at all */
23#define BTRFS_NO_LOG_SYNC 256
24
22int btrfs_sync_log(struct btrfs_trans_handle *trans, 25int btrfs_sync_log(struct btrfs_trans_handle *trans,
23 struct btrfs_root *root); 26 struct btrfs_root *root);
24int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root); 27int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root);
diff --git a/fs/btrfs/xattr.c b/fs/btrfs/xattr.c
index b0fc93f95fd0..b6dd5967c48a 100644
--- a/fs/btrfs/xattr.c
+++ b/fs/btrfs/xattr.c
@@ -260,7 +260,7 @@ err:
260 * attributes are handled directly. 260 * attributes are handled directly.
261 */ 261 */
262struct xattr_handler *btrfs_xattr_handlers[] = { 262struct xattr_handler *btrfs_xattr_handlers[] = {
263#ifdef CONFIG_BTRFS_POSIX_ACL 263#ifdef CONFIG_BTRFS_FS_POSIX_ACL
264 &btrfs_xattr_acl_access_handler, 264 &btrfs_xattr_acl_access_handler,
265 &btrfs_xattr_acl_default_handler, 265 &btrfs_xattr_acl_default_handler,
266#endif 266#endif
generated by cgit v1.2.2 (git 2.25.0) at 2025-12-07 20:22:40 -0500