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: Artem Bityutskiy (Битюцкий Артём)
24 * This file is a part of UBIFS journal implementation and contains various
25 * functions which manipulate the log. The log is a fixed area on the flash
26 * which does not contain any data but refers to buds. The log is a part of the
32 static int dbg_check_bud_bytes(struct ubifs_info
*c
);
35 * ubifs_search_bud - search bud LEB.
36 * @c: UBIFS file-system description object
37 * @lnum: logical eraseblock number to search
39 * This function searches bud LEB @lnum. Returns bud description object in case
40 * of success and %NULL if there is no bud with this LEB number.
42 struct ubifs_bud
*ubifs_search_bud(struct ubifs_info
*c
, int lnum
)
45 struct ubifs_bud
*bud
;
47 spin_lock(&c
->buds_lock
);
50 bud
= rb_entry(p
, struct ubifs_bud
, rb
);
53 else if (lnum
> bud
->lnum
)
56 spin_unlock(&c
->buds_lock
);
60 spin_unlock(&c
->buds_lock
);
65 * ubifs_get_wbuf - get the wbuf associated with a LEB, if there is one.
66 * @c: UBIFS file-system description object
67 * @lnum: logical eraseblock number to search
69 * This functions returns the wbuf for @lnum or %NULL if there is not one.
71 struct ubifs_wbuf
*ubifs_get_wbuf(struct ubifs_info
*c
, int lnum
)
74 struct ubifs_bud
*bud
;
80 spin_lock(&c
->buds_lock
);
83 bud
= rb_entry(p
, struct ubifs_bud
, rb
);
86 else if (lnum
> bud
->lnum
)
90 spin_unlock(&c
->buds_lock
);
91 return &c
->jheads
[jhead
].wbuf
;
94 spin_unlock(&c
->buds_lock
);
99 * empty_log_bytes - calculate amount of empty space in the log.
100 * @c: UBIFS file-system description object
102 static inline long long empty_log_bytes(const struct ubifs_info
*c
)
106 h
= (long long)c
->lhead_lnum
* c
->leb_size
+ c
->lhead_offs
;
107 t
= (long long)c
->ltail_lnum
* c
->leb_size
;
110 return c
->log_bytes
- h
+ t
;
113 else if (c
->lhead_lnum
!= c
->ltail_lnum
)
120 * ubifs_add_bud - add bud LEB to the tree of buds and its journal head list.
121 * @c: UBIFS file-system description object
122 * @bud: the bud to add
124 void ubifs_add_bud(struct ubifs_info
*c
, struct ubifs_bud
*bud
)
126 struct rb_node
**p
, *parent
= NULL
;
128 struct ubifs_jhead
*jhead
;
130 spin_lock(&c
->buds_lock
);
131 p
= &c
->buds
.rb_node
;
134 b
= rb_entry(parent
, struct ubifs_bud
, rb
);
135 ubifs_assert(c
, bud
->lnum
!= b
->lnum
);
136 if (bud
->lnum
< b
->lnum
)
142 rb_link_node(&bud
->rb
, parent
, p
);
143 rb_insert_color(&bud
->rb
, &c
->buds
);
145 jhead
= &c
->jheads
[bud
->jhead
];
146 list_add_tail(&bud
->list
, &jhead
->buds_list
);
148 ubifs_assert(c
, c
->replaying
&& c
->ro_mount
);
151 * Note, although this is a new bud, we anyway account this space now,
152 * before any data has been written to it, because this is about to
153 * guarantee fixed mount time, and this bud will anyway be read and
156 c
->bud_bytes
+= c
->leb_size
- bud
->start
;
158 dbg_log("LEB %d:%d, jhead %s, bud_bytes %lld", bud
->lnum
,
159 bud
->start
, dbg_jhead(bud
->jhead
), c
->bud_bytes
);
160 spin_unlock(&c
->buds_lock
);
164 * ubifs_add_bud_to_log - add a new bud to the log.
165 * @c: UBIFS file-system description object
166 * @jhead: journal head the bud belongs to
167 * @lnum: LEB number of the bud
168 * @offs: starting offset of the bud
170 * This function writes a reference node for the new bud LEB @lnum to the log,
171 * and adds it to the buds trees. It also makes sure that log size does not
172 * exceed the 'c->max_bud_bytes' limit. Returns zero in case of success,
173 * %-EAGAIN if commit is required, and a negative error code in case of
176 int ubifs_add_bud_to_log(struct ubifs_info
*c
, int jhead
, int lnum
, int offs
)
179 struct ubifs_bud
*bud
;
180 struct ubifs_ref_node
*ref
;
182 bud
= kmalloc(sizeof(struct ubifs_bud
), GFP_NOFS
);
185 ref
= kzalloc(c
->ref_node_alsz
, GFP_NOFS
);
191 mutex_lock(&c
->log_mutex
);
192 ubifs_assert(c
, !c
->ro_media
&& !c
->ro_mount
);
198 /* Make sure we have enough space in the log */
199 if (empty_log_bytes(c
) - c
->ref_node_alsz
< c
->min_log_bytes
) {
200 dbg_log("not enough log space - %lld, required %d",
201 empty_log_bytes(c
), c
->min_log_bytes
);
202 ubifs_commit_required(c
);
208 * Make sure the amount of space in buds will not exceed the
209 * 'c->max_bud_bytes' limit, because we want to guarantee mount time
212 * It is not necessary to hold @c->buds_lock when reading @c->bud_bytes
213 * because we are holding @c->log_mutex. All @c->bud_bytes take place
214 * when both @c->log_mutex and @c->bud_bytes are locked.
216 if (c
->bud_bytes
+ c
->leb_size
- offs
> c
->max_bud_bytes
) {
217 dbg_log("bud bytes %lld (%lld max), require commit",
218 c
->bud_bytes
, c
->max_bud_bytes
);
219 ubifs_commit_required(c
);
225 * If the journal is full enough - start background commit. Note, it is
226 * OK to read 'c->cmt_state' without spinlock because integer reads
227 * are atomic in the kernel.
229 if (c
->bud_bytes
>= c
->bg_bud_bytes
&&
230 c
->cmt_state
== COMMIT_RESTING
) {
231 dbg_log("bud bytes %lld (%lld max), initiate BG commit",
232 c
->bud_bytes
, c
->max_bud_bytes
);
233 ubifs_request_bg_commit(c
);
239 bud
->log_hash
= NULL
;
241 ref
->ch
.node_type
= UBIFS_REF_NODE
;
242 ref
->lnum
= cpu_to_le32(bud
->lnum
);
243 ref
->offs
= cpu_to_le32(bud
->start
);
244 ref
->jhead
= cpu_to_le32(jhead
);
246 if (c
->lhead_offs
> c
->leb_size
- c
->ref_node_alsz
) {
247 c
->lhead_lnum
= ubifs_next_log_lnum(c
, c
->lhead_lnum
);
248 ubifs_assert(c
, c
->lhead_lnum
!= c
->ltail_lnum
);
252 if (c
->lhead_offs
== 0) {
253 /* Must ensure next log LEB has been unmapped */
254 err
= ubifs_leb_unmap(c
, c
->lhead_lnum
);
259 if (bud
->start
== 0) {
261 * Before writing the LEB reference which refers an empty LEB
262 * to the log, we have to make sure it is mapped, because
263 * otherwise we'd risk to refer an LEB with garbage in case of
264 * an unclean reboot, because the target LEB might have been
265 * unmapped, but not yet physically erased.
267 err
= ubifs_leb_map(c
, bud
->lnum
);
272 dbg_log("write ref LEB %d:%d",
273 c
->lhead_lnum
, c
->lhead_offs
);
274 err
= ubifs_write_node(c
, ref
, UBIFS_REF_NODE_SZ
, c
->lhead_lnum
,
279 err
= ubifs_shash_update(c
, c
->log_hash
, ref
, UBIFS_REF_NODE_SZ
);
283 err
= ubifs_shash_copy_state(c
, c
->log_hash
, c
->jheads
[jhead
].log_hash
);
287 c
->lhead_offs
+= c
->ref_node_alsz
;
289 ubifs_add_bud(c
, bud
);
291 mutex_unlock(&c
->log_mutex
);
296 mutex_unlock(&c
->log_mutex
);
303 * remove_buds - remove used buds.
304 * @c: UBIFS file-system description object
306 * This function removes use buds from the buds tree. It does not remove the
307 * buds which are pointed to by journal heads.
309 static void remove_buds(struct ubifs_info
*c
)
313 ubifs_assert(c
, list_empty(&c
->old_buds
));
314 c
->cmt_bud_bytes
= 0;
315 spin_lock(&c
->buds_lock
);
316 p
= rb_first(&c
->buds
);
318 struct rb_node
*p1
= p
;
319 struct ubifs_bud
*bud
;
320 struct ubifs_wbuf
*wbuf
;
323 bud
= rb_entry(p1
, struct ubifs_bud
, rb
);
324 wbuf
= &c
->jheads
[bud
->jhead
].wbuf
;
326 if (wbuf
->lnum
== bud
->lnum
) {
328 * Do not remove buds which are pointed to by journal
329 * heads (non-closed buds).
331 c
->cmt_bud_bytes
+= wbuf
->offs
- bud
->start
;
332 dbg_log("preserve %d:%d, jhead %s, bud bytes %d, cmt_bud_bytes %lld",
333 bud
->lnum
, bud
->start
, dbg_jhead(bud
->jhead
),
334 wbuf
->offs
- bud
->start
, c
->cmt_bud_bytes
);
335 bud
->start
= wbuf
->offs
;
337 c
->cmt_bud_bytes
+= c
->leb_size
- bud
->start
;
338 dbg_log("remove %d:%d, jhead %s, bud bytes %d, cmt_bud_bytes %lld",
339 bud
->lnum
, bud
->start
, dbg_jhead(bud
->jhead
),
340 c
->leb_size
- bud
->start
, c
->cmt_bud_bytes
);
341 rb_erase(p1
, &c
->buds
);
343 * If the commit does not finish, the recovery will need
344 * to replay the journal, in which case the old buds
345 * must be unchanged. Do not release them until post
346 * commit i.e. do not allow them to be garbage
349 list_move(&bud
->list
, &c
->old_buds
);
352 spin_unlock(&c
->buds_lock
);
356 * ubifs_log_start_commit - start commit.
357 * @c: UBIFS file-system description object
358 * @ltail_lnum: return new log tail LEB number
360 * The commit operation starts with writing "commit start" node to the log and
361 * reference nodes for all journal heads which will define new journal after
362 * the commit has been finished. The commit start and reference nodes are
363 * written in one go to the nearest empty log LEB (hence, when commit is
364 * finished UBIFS may safely unmap all the previous log LEBs). This function
365 * returns zero in case of success and a negative error code in case of
368 int ubifs_log_start_commit(struct ubifs_info
*c
, int *ltail_lnum
)
371 struct ubifs_cs_node
*cs
;
372 struct ubifs_ref_node
*ref
;
373 int err
, i
, max_len
, len
;
375 err
= dbg_check_bud_bytes(c
);
379 max_len
= UBIFS_CS_NODE_SZ
+ c
->jhead_cnt
* UBIFS_REF_NODE_SZ
;
380 max_len
= ALIGN(max_len
, c
->min_io_size
);
381 buf
= cs
= kmalloc(max_len
, GFP_NOFS
);
385 cs
->ch
.node_type
= UBIFS_CS_NODE
;
386 cs
->cmt_no
= cpu_to_le64(c
->cmt_no
);
387 ubifs_prepare_node(c
, cs
, UBIFS_CS_NODE_SZ
, 0);
389 err
= ubifs_shash_init(c
, c
->log_hash
);
393 err
= ubifs_shash_update(c
, c
->log_hash
, cs
, UBIFS_CS_NODE_SZ
);
398 * Note, we do not lock 'c->log_mutex' because this is the commit start
399 * phase and we are exclusively using the log. And we do not lock
400 * write-buffer because nobody can write to the file-system at this
404 len
= UBIFS_CS_NODE_SZ
;
405 for (i
= 0; i
< c
->jhead_cnt
; i
++) {
406 int lnum
= c
->jheads
[i
].wbuf
.lnum
;
407 int offs
= c
->jheads
[i
].wbuf
.offs
;
409 if (lnum
== -1 || offs
== c
->leb_size
)
412 dbg_log("add ref to LEB %d:%d for jhead %s",
413 lnum
, offs
, dbg_jhead(i
));
415 ref
->ch
.node_type
= UBIFS_REF_NODE
;
416 ref
->lnum
= cpu_to_le32(lnum
);
417 ref
->offs
= cpu_to_le32(offs
);
418 ref
->jhead
= cpu_to_le32(i
);
420 ubifs_prepare_node(c
, ref
, UBIFS_REF_NODE_SZ
, 0);
421 len
+= UBIFS_REF_NODE_SZ
;
423 err
= ubifs_shash_update(c
, c
->log_hash
, ref
,
427 ubifs_shash_copy_state(c
, c
->log_hash
, c
->jheads
[i
].log_hash
);
430 ubifs_pad(c
, buf
+ len
, ALIGN(len
, c
->min_io_size
) - len
);
432 /* Switch to the next log LEB */
434 c
->lhead_lnum
= ubifs_next_log_lnum(c
, c
->lhead_lnum
);
435 ubifs_assert(c
, c
->lhead_lnum
!= c
->ltail_lnum
);
439 /* Must ensure next LEB has been unmapped */
440 err
= ubifs_leb_unmap(c
, c
->lhead_lnum
);
444 len
= ALIGN(len
, c
->min_io_size
);
445 dbg_log("writing commit start at LEB %d:0, len %d", c
->lhead_lnum
, len
);
446 err
= ubifs_leb_write(c
, c
->lhead_lnum
, cs
, 0, len
);
450 *ltail_lnum
= c
->lhead_lnum
;
452 c
->lhead_offs
+= len
;
453 if (c
->lhead_offs
== c
->leb_size
) {
454 c
->lhead_lnum
= ubifs_next_log_lnum(c
, c
->lhead_lnum
);
461 * We have started the commit and now users may use the rest of the log
464 c
->min_log_bytes
= 0;
472 * ubifs_log_end_commit - end commit.
473 * @c: UBIFS file-system description object
474 * @ltail_lnum: new log tail LEB number
476 * This function is called on when the commit operation was finished. It
477 * moves log tail to new position and updates the master node so that it stores
478 * the new log tail LEB number. Returns zero in case of success and a negative
479 * error code in case of failure.
481 int ubifs_log_end_commit(struct ubifs_info
*c
, int ltail_lnum
)
486 * At this phase we have to lock 'c->log_mutex' because UBIFS allows FS
487 * writes during commit. Its only short "commit" start phase when
488 * writers are blocked.
490 mutex_lock(&c
->log_mutex
);
492 dbg_log("old tail was LEB %d:0, new tail is LEB %d:0",
493 c
->ltail_lnum
, ltail_lnum
);
495 c
->ltail_lnum
= ltail_lnum
;
497 * The commit is finished and from now on it must be guaranteed that
498 * there is always enough space for the next commit.
500 c
->min_log_bytes
= c
->leb_size
;
502 spin_lock(&c
->buds_lock
);
503 c
->bud_bytes
-= c
->cmt_bud_bytes
;
504 spin_unlock(&c
->buds_lock
);
506 err
= dbg_check_bud_bytes(c
);
510 err
= ubifs_write_master(c
);
513 mutex_unlock(&c
->log_mutex
);
518 * ubifs_log_post_commit - things to do after commit is completed.
519 * @c: UBIFS file-system description object
520 * @old_ltail_lnum: old log tail LEB number
522 * Release buds only after commit is completed, because they must be unchanged
523 * if recovery is needed.
525 * Unmap log LEBs only after commit is completed, because they may be needed for
528 * This function returns %0 on success and a negative error code on failure.
530 int ubifs_log_post_commit(struct ubifs_info
*c
, int old_ltail_lnum
)
534 while (!list_empty(&c
->old_buds
)) {
535 struct ubifs_bud
*bud
;
537 bud
= list_entry(c
->old_buds
.next
, struct ubifs_bud
, list
);
538 err
= ubifs_return_leb(c
, bud
->lnum
);
541 list_del(&bud
->list
);
542 kfree(bud
->log_hash
);
545 mutex_lock(&c
->log_mutex
);
546 for (lnum
= old_ltail_lnum
; lnum
!= c
->ltail_lnum
;
547 lnum
= ubifs_next_log_lnum(c
, lnum
)) {
548 dbg_log("unmap log LEB %d", lnum
);
549 err
= ubifs_leb_unmap(c
, lnum
);
554 mutex_unlock(&c
->log_mutex
);
559 * struct done_ref - references that have been done.
569 * done_already - determine if a reference has been done already.
570 * @done_tree: rb-tree to store references that have been done
571 * @lnum: LEB number of reference
573 * This function returns %1 if the reference has been done, %0 if not, otherwise
574 * a negative error code is returned.
576 static int done_already(struct rb_root
*done_tree
, int lnum
)
578 struct rb_node
**p
= &done_tree
->rb_node
, *parent
= NULL
;
583 dr
= rb_entry(parent
, struct done_ref
, rb
);
586 else if (lnum
> dr
->lnum
)
592 dr
= kzalloc(sizeof(struct done_ref
), GFP_NOFS
);
598 rb_link_node(&dr
->rb
, parent
, p
);
599 rb_insert_color(&dr
->rb
, done_tree
);
605 * destroy_done_tree - destroy the done tree.
606 * @done_tree: done tree to destroy
608 static void destroy_done_tree(struct rb_root
*done_tree
)
610 struct done_ref
*dr
, *n
;
612 rbtree_postorder_for_each_entry_safe(dr
, n
, done_tree
, rb
)
617 * add_node - add a node to the consolidated log.
618 * @c: UBIFS file-system description object
619 * @buf: buffer to which to add
620 * @lnum: LEB number to which to write is passed and returned here
621 * @offs: offset to where to write is passed and returned here
624 * This function returns %0 on success and a negative error code on failure.
626 static int add_node(struct ubifs_info
*c
, void *buf
, int *lnum
, int *offs
,
629 struct ubifs_ch
*ch
= node
;
630 int len
= le32_to_cpu(ch
->len
), remains
= c
->leb_size
- *offs
;
633 int sz
= ALIGN(*offs
, c
->min_io_size
), err
;
635 ubifs_pad(c
, buf
+ *offs
, sz
- *offs
);
636 err
= ubifs_leb_change(c
, *lnum
, buf
, sz
);
639 *lnum
= ubifs_next_log_lnum(c
, *lnum
);
642 memcpy(buf
+ *offs
, node
, len
);
643 *offs
+= ALIGN(len
, 8);
648 * ubifs_consolidate_log - consolidate the log.
649 * @c: UBIFS file-system description object
651 * Repeated failed commits could cause the log to be full, but at least 1 LEB is
652 * needed for commit. This function rewrites the reference nodes in the log
653 * omitting duplicates, and failed CS nodes, and leaving no gaps.
655 * This function returns %0 on success and a negative error code on failure.
657 int ubifs_consolidate_log(struct ubifs_info
*c
)
659 struct ubifs_scan_leb
*sleb
;
660 struct ubifs_scan_node
*snod
;
661 struct rb_root done_tree
= RB_ROOT
;
662 int lnum
, err
, first
= 1, write_lnum
, offs
= 0;
665 dbg_rcvry("log tail LEB %d, log head LEB %d", c
->ltail_lnum
,
667 buf
= vmalloc(c
->leb_size
);
670 lnum
= c
->ltail_lnum
;
673 sleb
= ubifs_scan(c
, lnum
, 0, c
->sbuf
, 0);
678 list_for_each_entry(snod
, &sleb
->nodes
, list
) {
679 switch (snod
->type
) {
680 case UBIFS_REF_NODE
: {
681 struct ubifs_ref_node
*ref
= snod
->node
;
682 int ref_lnum
= le32_to_cpu(ref
->lnum
);
684 err
= done_already(&done_tree
, ref_lnum
);
688 err
= add_node(c
, buf
, &write_lnum
,
698 err
= add_node(c
, buf
, &write_lnum
, &offs
,
706 ubifs_scan_destroy(sleb
);
707 if (lnum
== c
->lhead_lnum
)
709 lnum
= ubifs_next_log_lnum(c
, lnum
);
712 int sz
= ALIGN(offs
, c
->min_io_size
);
714 ubifs_pad(c
, buf
+ offs
, sz
- offs
);
715 err
= ubifs_leb_change(c
, write_lnum
, buf
, sz
);
718 offs
= ALIGN(offs
, c
->min_io_size
);
720 destroy_done_tree(&done_tree
);
722 if (write_lnum
== c
->lhead_lnum
) {
723 ubifs_err(c
, "log is too full");
726 /* Unmap remaining LEBs */
729 lnum
= ubifs_next_log_lnum(c
, lnum
);
730 err
= ubifs_leb_unmap(c
, lnum
);
733 } while (lnum
!= c
->lhead_lnum
);
734 c
->lhead_lnum
= write_lnum
;
735 c
->lhead_offs
= offs
;
736 dbg_rcvry("new log head at %d:%d", c
->lhead_lnum
, c
->lhead_offs
);
740 ubifs_scan_destroy(sleb
);
742 destroy_done_tree(&done_tree
);
748 * dbg_check_bud_bytes - make sure bud bytes calculation are all right.
749 * @c: UBIFS file-system description object
751 * This function makes sure the amount of flash space used by closed buds
752 * ('c->bud_bytes' is correct). Returns zero in case of success and %-EINVAL in
755 static int dbg_check_bud_bytes(struct ubifs_info
*c
)
758 struct ubifs_bud
*bud
;
759 long long bud_bytes
= 0;
761 if (!dbg_is_chk_gen(c
))
764 spin_lock(&c
->buds_lock
);
765 for (i
= 0; i
< c
->jhead_cnt
; i
++)
766 list_for_each_entry(bud
, &c
->jheads
[i
].buds_list
, list
)
767 bud_bytes
+= c
->leb_size
- bud
->start
;
769 if (c
->bud_bytes
!= bud_bytes
) {
770 ubifs_err(c
, "bad bud_bytes %lld, calculated %lld",
771 c
->bud_bytes
, bud_bytes
);
774 spin_unlock(&c
->buds_lock
);