2 * This file is part of UBIFS.
4 * Copyright (C) 2006-2008 Nokia Corporation.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published by
8 * the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc., 51
17 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 * Authors: Adrian Hunter
20 * Artem Bityutskiy (Битюцкий Артём)
24 * This file implements functions that manage the running of the commit process.
25 * Each affected module has its own functions to accomplish their part in the
26 * commit and those functions are called here.
28 * The commit is the process whereby all updates to the index and LEB properties
29 * are written out together and the journal becomes empty. This keeps the
30 * file system consistent - at all times the state can be recreated by reading
31 * the index and LEB properties and then replaying the journal.
33 * The commit is split into two parts named "commit start" and "commit end".
34 * During commit start, the commit process has exclusive access to the journal
35 * by holding the commit semaphore down for writing. As few I/O operations as
36 * possible are performed during commit start, instead the nodes that are to be
37 * written are merely identified. During commit end, the commit semaphore is no
38 * longer held and the journal is again in operation, allowing users to continue
39 * to use the file system while the bulk of the commit I/O is performed. The
40 * purpose of this two-step approach is to prevent the commit from causing any
41 * latency blips. Note that in any case, the commit does not prevent lookups
42 * (as permitted by the TNC mutex), or access to VFS data structures e.g. page
46 #include <linux/freezer.h>
47 #include <linux/kthread.h>
51 * do_commit - commit the journal.
52 * @c: UBIFS file-system description object
54 * This function implements UBIFS commit. It has to be called with commit lock
55 * locked. Returns zero in case of success and a negative error code in case of
58 static int do_commit(struct ubifs_info
*c
)
60 int err
, new_ltail_lnum
, old_ltail_lnum
, i
;
61 struct ubifs_zbranch zroot
;
62 struct ubifs_lp_stats lst
;
70 /* Sync all write buffers (necessary for recovery) */
71 for (i
= 0; i
< c
->jhead_cnt
; i
++) {
72 err
= ubifs_wbuf_sync(&c
->jheads
[i
].wbuf
);
78 err
= ubifs_gc_start_commit(c
);
81 err
= dbg_check_lprops(c
);
84 err
= ubifs_log_start_commit(c
, &new_ltail_lnum
);
87 err
= ubifs_tnc_start_commit(c
, &zroot
);
90 err
= ubifs_lpt_start_commit(c
);
93 err
= ubifs_orphan_start_commit(c
);
97 ubifs_get_lp_stats(c
, &lst
);
99 up_write(&c
->commit_sem
);
101 err
= ubifs_tnc_end_commit(c
);
104 err
= ubifs_lpt_end_commit(c
);
107 err
= ubifs_orphan_end_commit(c
);
110 old_ltail_lnum
= c
->ltail_lnum
;
111 err
= ubifs_log_end_commit(c
, new_ltail_lnum
);
114 err
= dbg_check_old_index(c
, &zroot
);
118 mutex_lock(&c
->mst_mutex
);
119 c
->mst_node
->cmt_no
= cpu_to_le64(c
->cmt_no
);
120 c
->mst_node
->log_lnum
= cpu_to_le32(new_ltail_lnum
);
121 c
->mst_node
->root_lnum
= cpu_to_le32(zroot
.lnum
);
122 c
->mst_node
->root_offs
= cpu_to_le32(zroot
.offs
);
123 c
->mst_node
->root_len
= cpu_to_le32(zroot
.len
);
124 c
->mst_node
->ihead_lnum
= cpu_to_le32(c
->ihead_lnum
);
125 c
->mst_node
->ihead_offs
= cpu_to_le32(c
->ihead_offs
);
126 c
->mst_node
->index_size
= cpu_to_le64(c
->old_idx_sz
);
127 c
->mst_node
->lpt_lnum
= cpu_to_le32(c
->lpt_lnum
);
128 c
->mst_node
->lpt_offs
= cpu_to_le32(c
->lpt_offs
);
129 c
->mst_node
->nhead_lnum
= cpu_to_le32(c
->nhead_lnum
);
130 c
->mst_node
->nhead_offs
= cpu_to_le32(c
->nhead_offs
);
131 c
->mst_node
->ltab_lnum
= cpu_to_le32(c
->ltab_lnum
);
132 c
->mst_node
->ltab_offs
= cpu_to_le32(c
->ltab_offs
);
133 c
->mst_node
->lsave_lnum
= cpu_to_le32(c
->lsave_lnum
);
134 c
->mst_node
->lsave_offs
= cpu_to_le32(c
->lsave_offs
);
135 c
->mst_node
->lscan_lnum
= cpu_to_le32(c
->lscan_lnum
);
136 c
->mst_node
->empty_lebs
= cpu_to_le32(lst
.empty_lebs
);
137 c
->mst_node
->idx_lebs
= cpu_to_le32(lst
.idx_lebs
);
138 c
->mst_node
->total_free
= cpu_to_le64(lst
.total_free
);
139 c
->mst_node
->total_dirty
= cpu_to_le64(lst
.total_dirty
);
140 c
->mst_node
->total_used
= cpu_to_le64(lst
.total_used
);
141 c
->mst_node
->total_dead
= cpu_to_le64(lst
.total_dead
);
142 c
->mst_node
->total_dark
= cpu_to_le64(lst
.total_dark
);
144 c
->mst_node
->flags
|= cpu_to_le32(UBIFS_MST_NO_ORPHS
);
146 c
->mst_node
->flags
&= ~cpu_to_le32(UBIFS_MST_NO_ORPHS
);
147 err
= ubifs_write_master(c
);
148 mutex_unlock(&c
->mst_mutex
);
152 err
= ubifs_log_post_commit(c
, old_ltail_lnum
);
155 err
= ubifs_gc_end_commit(c
);
158 err
= ubifs_lpt_post_commit(c
);
162 spin_lock(&c
->cs_lock
);
163 c
->cmt_state
= COMMIT_RESTING
;
165 dbg_cmt("commit end");
166 spin_unlock(&c
->cs_lock
);
171 up_write(&c
->commit_sem
);
173 ubifs_err("commit failed, error %d", err
);
174 spin_lock(&c
->cs_lock
);
175 c
->cmt_state
= COMMIT_BROKEN
;
177 spin_unlock(&c
->cs_lock
);
178 ubifs_ro_mode(c
, err
);
183 * run_bg_commit - run background commit if it is needed.
184 * @c: UBIFS file-system description object
186 * This function runs background commit if it is needed. Returns zero in case
187 * of success and a negative error code in case of failure.
189 static int run_bg_commit(struct ubifs_info
*c
)
191 spin_lock(&c
->cs_lock
);
193 * Run background commit only if background commit was requested or if
194 * commit is required.
196 if (c
->cmt_state
!= COMMIT_BACKGROUND
&&
197 c
->cmt_state
!= COMMIT_REQUIRED
)
199 spin_unlock(&c
->cs_lock
);
201 down_write(&c
->commit_sem
);
202 spin_lock(&c
->cs_lock
);
203 if (c
->cmt_state
== COMMIT_REQUIRED
)
204 c
->cmt_state
= COMMIT_RUNNING_REQUIRED
;
205 else if (c
->cmt_state
== COMMIT_BACKGROUND
)
206 c
->cmt_state
= COMMIT_RUNNING_BACKGROUND
;
209 spin_unlock(&c
->cs_lock
);
214 up_write(&c
->commit_sem
);
216 spin_unlock(&c
->cs_lock
);
221 * ubifs_bg_thread - UBIFS background thread function.
222 * @info: points to the file-system description object
224 * This function implements various file-system background activities:
225 * o when a write-buffer timer expires it synchronizes the appropriate
227 * o when the journal is about to be full, it starts in-advance commit.
229 * Note, other stuff like background garbage collection may be added here in
232 int ubifs_bg_thread(void *info
)
235 struct ubifs_info
*c
= info
;
237 dbg_msg("background thread \"%s\" started, PID %d",
238 c
->bgt_name
, current
->pid
);
242 if (kthread_should_stop())
248 set_current_state(TASK_INTERRUPTIBLE
);
249 /* Check if there is something to do */
252 * Nothing prevents us from going sleep now and
253 * be never woken up and block the task which
254 * could wait in 'kthread_stop()' forever.
256 if (kthread_should_stop())
261 __set_current_state(TASK_RUNNING
);
264 err
= ubifs_bg_wbufs_sync(c
);
266 ubifs_ro_mode(c
, err
);
272 dbg_msg("background thread \"%s\" stops", c
->bgt_name
);
277 * ubifs_commit_required - set commit state to "required".
278 * @c: UBIFS file-system description object
280 * This function is called if a commit is required but cannot be done from the
281 * calling function, so it is just flagged instead.
283 void ubifs_commit_required(struct ubifs_info
*c
)
285 spin_lock(&c
->cs_lock
);
286 switch (c
->cmt_state
) {
288 case COMMIT_BACKGROUND
:
289 dbg_cmt("old: %s, new: %s", dbg_cstate(c
->cmt_state
),
290 dbg_cstate(COMMIT_REQUIRED
));
291 c
->cmt_state
= COMMIT_REQUIRED
;
293 case COMMIT_RUNNING_BACKGROUND
:
294 dbg_cmt("old: %s, new: %s", dbg_cstate(c
->cmt_state
),
295 dbg_cstate(COMMIT_RUNNING_REQUIRED
));
296 c
->cmt_state
= COMMIT_RUNNING_REQUIRED
;
298 case COMMIT_REQUIRED
:
299 case COMMIT_RUNNING_REQUIRED
:
303 spin_unlock(&c
->cs_lock
);
307 * ubifs_request_bg_commit - notify the background thread to do a commit.
308 * @c: UBIFS file-system description object
310 * This function is called if the journal is full enough to make a commit
311 * worthwhile, so background thread is kicked to start it.
313 void ubifs_request_bg_commit(struct ubifs_info
*c
)
315 spin_lock(&c
->cs_lock
);
316 if (c
->cmt_state
== COMMIT_RESTING
) {
317 dbg_cmt("old: %s, new: %s", dbg_cstate(c
->cmt_state
),
318 dbg_cstate(COMMIT_BACKGROUND
));
319 c
->cmt_state
= COMMIT_BACKGROUND
;
320 spin_unlock(&c
->cs_lock
);
321 ubifs_wake_up_bgt(c
);
323 spin_unlock(&c
->cs_lock
);
327 * wait_for_commit - wait for commit.
328 * @c: UBIFS file-system description object
330 * This function sleeps until the commit operation is no longer running.
332 static int wait_for_commit(struct ubifs_info
*c
)
334 dbg_cmt("pid %d goes sleep", current
->pid
);
337 * The following sleeps if the condition is false, and will be woken
338 * when the commit ends. It is possible, although very unlikely, that we
339 * will wake up and see the subsequent commit running, rather than the
340 * one we were waiting for, and go back to sleep. However, we will be
341 * woken again, so there is no danger of sleeping forever.
343 wait_event(c
->cmt_wq
, c
->cmt_state
!= COMMIT_RUNNING_BACKGROUND
&&
344 c
->cmt_state
!= COMMIT_RUNNING_REQUIRED
);
345 dbg_cmt("commit finished, pid %d woke up", current
->pid
);
350 * ubifs_run_commit - run or wait for commit.
351 * @c: UBIFS file-system description object
353 * This function runs commit and returns zero in case of success and a negative
354 * error code in case of failure.
356 int ubifs_run_commit(struct ubifs_info
*c
)
360 spin_lock(&c
->cs_lock
);
361 if (c
->cmt_state
== COMMIT_BROKEN
) {
366 if (c
->cmt_state
== COMMIT_RUNNING_BACKGROUND
)
368 * We set the commit state to 'running required' to indicate
369 * that we want it to complete as quickly as possible.
371 c
->cmt_state
= COMMIT_RUNNING_REQUIRED
;
373 if (c
->cmt_state
== COMMIT_RUNNING_REQUIRED
) {
374 spin_unlock(&c
->cs_lock
);
375 return wait_for_commit(c
);
377 spin_unlock(&c
->cs_lock
);
379 /* Ok, the commit is indeed needed */
381 down_write(&c
->commit_sem
);
382 spin_lock(&c
->cs_lock
);
384 * Since we unlocked 'c->cs_lock', the state may have changed, so
387 if (c
->cmt_state
== COMMIT_BROKEN
) {
392 if (c
->cmt_state
== COMMIT_RUNNING_BACKGROUND
)
393 c
->cmt_state
= COMMIT_RUNNING_REQUIRED
;
395 if (c
->cmt_state
== COMMIT_RUNNING_REQUIRED
) {
396 up_write(&c
->commit_sem
);
397 spin_unlock(&c
->cs_lock
);
398 return wait_for_commit(c
);
400 c
->cmt_state
= COMMIT_RUNNING_REQUIRED
;
401 spin_unlock(&c
->cs_lock
);
407 up_write(&c
->commit_sem
);
409 spin_unlock(&c
->cs_lock
);
414 * ubifs_gc_should_commit - determine if it is time for GC to run commit.
415 * @c: UBIFS file-system description object
417 * This function is called by garbage collection to determine if commit should
418 * be run. If commit state is @COMMIT_BACKGROUND, which means that the journal
419 * is full enough to start commit, this function returns true. It is not
420 * absolutely necessary to commit yet, but it feels like this should be better
421 * then to keep doing GC. This function returns %1 if GC has to initiate commit
424 int ubifs_gc_should_commit(struct ubifs_info
*c
)
428 spin_lock(&c
->cs_lock
);
429 if (c
->cmt_state
== COMMIT_BACKGROUND
) {
430 dbg_cmt("commit required now");
431 c
->cmt_state
= COMMIT_REQUIRED
;
433 dbg_cmt("commit not requested");
434 if (c
->cmt_state
== COMMIT_REQUIRED
)
436 spin_unlock(&c
->cs_lock
);
440 #ifdef CONFIG_UBIFS_FS_DEBUG
443 * struct idx_node - hold index nodes during index tree traversal.
445 * @iip: index in parent (slot number of this indexing node in the parent
447 * @upper_key: all keys in this indexing node have to be less or equivalent to
449 * @idx: index node (8-byte aligned because all node structures must be 8-byte
453 struct list_head list
;
455 union ubifs_key upper_key
;
456 struct ubifs_idx_node idx
__attribute__((aligned(8)));
460 * dbg_old_index_check_init - get information for the next old index check.
461 * @c: UBIFS file-system description object
462 * @zroot: root of the index
464 * This function records information about the index that will be needed for the
465 * next old index check i.e. 'dbg_check_old_index()'.
467 * This function returns %0 on success and a negative error code on failure.
469 int dbg_old_index_check_init(struct ubifs_info
*c
, struct ubifs_zbranch
*zroot
)
471 struct ubifs_idx_node
*idx
;
472 int lnum
, offs
, len
, err
= 0;
473 struct ubifs_debug_info
*d
= c
->dbg
;
475 d
->old_zroot
= *zroot
;
476 lnum
= d
->old_zroot
.lnum
;
477 offs
= d
->old_zroot
.offs
;
478 len
= d
->old_zroot
.len
;
480 idx
= kmalloc(c
->max_idx_node_sz
, GFP_NOFS
);
484 err
= ubifs_read_node(c
, idx
, UBIFS_IDX_NODE
, len
, lnum
, offs
);
488 d
->old_zroot_level
= le16_to_cpu(idx
->level
);
489 d
->old_zroot_sqnum
= le64_to_cpu(idx
->ch
.sqnum
);
496 * dbg_check_old_index - check the old copy of the index.
497 * @c: UBIFS file-system description object
498 * @zroot: root of the new index
500 * In order to be able to recover from an unclean unmount, a complete copy of
501 * the index must exist on flash. This is the "old" index. The commit process
502 * must write the "new" index to flash without overwriting or destroying any
503 * part of the old index. This function is run at commit end in order to check
504 * that the old index does indeed exist completely intact.
506 * This function returns %0 on success and a negative error code on failure.
508 int dbg_check_old_index(struct ubifs_info
*c
, struct ubifs_zbranch
*zroot
)
510 int lnum
, offs
, len
, err
= 0, uninitialized_var(last_level
), child_cnt
;
512 struct ubifs_debug_info
*d
= c
->dbg
;
513 union ubifs_key
uninitialized_var(lower_key
), upper_key
, l_key
, u_key
;
514 unsigned long long uninitialized_var(last_sqnum
);
515 struct ubifs_idx_node
*idx
;
516 struct list_head list
;
520 if (!(ubifs_chk_flags
& UBIFS_CHK_OLD_IDX
))
523 INIT_LIST_HEAD(&list
);
525 sz
= sizeof(struct idx_node
) + ubifs_idx_node_sz(c
, c
->fanout
) -
528 /* Start at the old zroot */
529 lnum
= d
->old_zroot
.lnum
;
530 offs
= d
->old_zroot
.offs
;
531 len
= d
->old_zroot
.len
;
535 * Traverse the index tree preorder depth-first i.e. do a node and then
536 * its subtrees from left to right.
539 struct ubifs_branch
*br
;
541 /* Get the next index node */
542 i
= kmalloc(sz
, GFP_NOFS
);
548 /* Keep the index nodes on our path in a linked list */
549 list_add_tail(&i
->list
, &list
);
550 /* Read the index node */
552 err
= ubifs_read_node(c
, idx
, UBIFS_IDX_NODE
, len
, lnum
, offs
);
555 /* Validate index node */
556 child_cnt
= le16_to_cpu(idx
->child_cnt
);
557 if (child_cnt
< 1 || child_cnt
> c
->fanout
) {
563 /* Check root level and sqnum */
564 if (le16_to_cpu(idx
->level
) != d
->old_zroot_level
) {
568 if (le64_to_cpu(idx
->ch
.sqnum
) != d
->old_zroot_sqnum
) {
572 /* Set last values as though root had a parent */
573 last_level
= le16_to_cpu(idx
->level
) + 1;
574 last_sqnum
= le64_to_cpu(idx
->ch
.sqnum
) + 1;
575 key_read(c
, ubifs_idx_key(c
, idx
), &lower_key
);
576 highest_ino_key(c
, &upper_key
, INUM_WATERMARK
);
578 key_copy(c
, &upper_key
, &i
->upper_key
);
579 if (le16_to_cpu(idx
->level
) != last_level
- 1) {
584 * The index is always written bottom up hence a child's sqnum
585 * is always less than the parents.
587 if (le64_to_cpu(idx
->ch
.sqnum
) >= last_sqnum
) {
591 /* Check key range */
592 key_read(c
, ubifs_idx_key(c
, idx
), &l_key
);
593 br
= ubifs_idx_branch(c
, idx
, child_cnt
- 1);
594 key_read(c
, &br
->key
, &u_key
);
595 if (keys_cmp(c
, &lower_key
, &l_key
) > 0) {
599 if (keys_cmp(c
, &upper_key
, &u_key
) < 0) {
603 if (keys_cmp(c
, &upper_key
, &u_key
) == 0)
604 if (!is_hash_key(c
, &u_key
)) {
608 /* Go to next index node */
609 if (le16_to_cpu(idx
->level
) == 0) {
610 /* At the bottom, so go up until can go right */
612 /* Drop the bottom of the list */
615 /* No more list means we are done */
616 if (list_empty(&list
))
618 /* Look at the new bottom */
619 i
= list_entry(list
.prev
, struct idx_node
,
622 /* Can we go right */
623 if (iip
+ 1 < le16_to_cpu(idx
->child_cnt
)) {
627 /* Nope, so go up again */
634 * We have the parent in 'idx' and now we set up for reading the
635 * child pointed to by slot 'iip'.
637 last_level
= le16_to_cpu(idx
->level
);
638 last_sqnum
= le64_to_cpu(idx
->ch
.sqnum
);
639 br
= ubifs_idx_branch(c
, idx
, iip
);
640 lnum
= le32_to_cpu(br
->lnum
);
641 offs
= le32_to_cpu(br
->offs
);
642 len
= le32_to_cpu(br
->len
);
643 key_read(c
, &br
->key
, &lower_key
);
644 if (iip
+ 1 < le16_to_cpu(idx
->child_cnt
)) {
645 br
= ubifs_idx_branch(c
, idx
, iip
+ 1);
646 key_read(c
, &br
->key
, &upper_key
);
648 key_copy(c
, &i
->upper_key
, &upper_key
);
651 err
= dbg_old_index_check_init(c
, zroot
);
658 dbg_err("dumping index node (iip=%d)", i
->iip
);
659 dbg_dump_node(c
, idx
);
662 if (!list_empty(&list
)) {
663 i
= list_entry(list
.prev
, struct idx_node
, list
);
664 dbg_err("dumping parent index node");
665 dbg_dump_node(c
, &i
->idx
);
668 while (!list_empty(&list
)) {
669 i
= list_entry(list
.next
, struct idx_node
, list
);
673 ubifs_err("failed, error %d", err
);
679 #endif /* CONFIG_UBIFS_FS_DEBUG */