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 #ifdef CONFIG_UBIFS_FS_DEBUG
33 static int dbg_check_bud_bytes(struct ubifs_info *c);
35 #define dbg_check_bud_bytes(c) 0
39 * ubifs_search_bud - search bud LEB.
40 * @c: UBIFS file-system description object
41 * @lnum: logical eraseblock number to search
43 * This function searches bud LEB @lnum. Returns bud description object in case
44 * of success and %NULL if there is no bud with this LEB number.
46 struct ubifs_bud *ubifs_search_bud(struct ubifs_info *c, int lnum)
49 struct ubifs_bud *bud;
51 spin_lock(&c->buds_lock);
54 bud = rb_entry(p, struct ubifs_bud, rb);
57 else if (lnum > bud->lnum)
60 spin_unlock(&c->buds_lock);
64 spin_unlock(&c->buds_lock);
69 * ubifs_get_wbuf - get the wbuf associated with a LEB, if there is one.
70 * @c: UBIFS file-system description object
71 * @lnum: logical eraseblock number to search
73 * This functions returns the wbuf for @lnum or %NULL if there is not one.
75 struct ubifs_wbuf *ubifs_get_wbuf(struct ubifs_info *c, int lnum)
78 struct ubifs_bud *bud;
84 spin_lock(&c->buds_lock);
87 bud = rb_entry(p, struct ubifs_bud, rb);
90 else if (lnum > bud->lnum)
94 spin_unlock(&c->buds_lock);
95 return &c->jheads[jhead].wbuf;
98 spin_unlock(&c->buds_lock);
103 * empty_log_bytes - calculate amount of empty space in the log.
104 * @c: UBIFS file-system description object
106 static inline long long empty_log_bytes(const struct ubifs_info *c)
110 h = (long long)c->lhead_lnum * c->leb_size + c->lhead_offs;
111 t = (long long)c->ltail_lnum * c->leb_size;
114 return c->log_bytes - h + t;
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(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->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 reference node for the new bud LEB @lnum it to the log,
171 * and adds it to the buds tress. 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 codes 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->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);
240 ref->ch.node_type = UBIFS_REF_NODE;
241 ref->lnum = cpu_to_le32(bud->lnum);
242 ref->offs = cpu_to_le32(bud->start);
243 ref->jhead = cpu_to_le32(jhead);
245 if (c->lhead_offs > c->leb_size - c->ref_node_alsz) {
246 c->lhead_lnum = ubifs_next_log_lnum(c, c->lhead_lnum);
250 if (c->lhead_offs == 0) {
251 /* Must ensure next log LEB has been unmapped */
252 err = ubifs_leb_unmap(c, c->lhead_lnum);
257 if (bud->start == 0) {
259 * Before writing the LEB reference which refers an empty LEB
260 * to the log, we have to make sure it is mapped, because
261 * otherwise we'd risk to refer an LEB with garbage in case of
262 * an unclean reboot, because the target LEB might have been
263 * unmapped, but not yet physically erased.
265 err = ubifs_leb_map(c, bud->lnum, UBI_SHORTTERM);
270 dbg_log("write ref LEB %d:%d",
271 c->lhead_lnum, c->lhead_offs);
272 err = ubifs_write_node(c, ref, UBIFS_REF_NODE_SZ, c->lhead_lnum,
273 c->lhead_offs, UBI_SHORTTERM);
277 c->lhead_offs += c->ref_node_alsz;
279 ubifs_add_bud(c, bud);
281 mutex_unlock(&c->log_mutex);
286 mutex_unlock(&c->log_mutex);
293 * remove_buds - remove used buds.
294 * @c: UBIFS file-system description object
296 * This function removes use buds from the buds tree. It does not remove the
297 * buds which are pointed to by journal heads.
299 static void remove_buds(struct ubifs_info *c)
303 ubifs_assert(list_empty(&c->old_buds));
304 c->cmt_bud_bytes = 0;
305 spin_lock(&c->buds_lock);
306 p = rb_first(&c->buds);
308 struct rb_node *p1 = p;
309 struct ubifs_bud *bud;
310 struct ubifs_wbuf *wbuf;
313 bud = rb_entry(p1, struct ubifs_bud, rb);
314 wbuf = &c->jheads[bud->jhead].wbuf;
316 if (wbuf->lnum == bud->lnum) {
318 * Do not remove buds which are pointed to by journal
319 * heads (non-closed buds).
321 c->cmt_bud_bytes += wbuf->offs - bud->start;
322 dbg_log("preserve %d:%d, jhead %s, bud bytes %d, "
323 "cmt_bud_bytes %lld", bud->lnum, bud->start,
324 dbg_jhead(bud->jhead), wbuf->offs - bud->start,
326 bud->start = wbuf->offs;
328 c->cmt_bud_bytes += c->leb_size - bud->start;
329 dbg_log("remove %d:%d, jhead %s, bud bytes %d, "
330 "cmt_bud_bytes %lld", bud->lnum, bud->start,
331 dbg_jhead(bud->jhead), c->leb_size - bud->start,
333 rb_erase(p1, &c->buds);
335 * If the commit does not finish, the recovery will need
336 * to replay the journal, in which case the old buds
337 * must be unchanged. Do not release them until post
338 * commit i.e. do not allow them to be garbage
341 list_move(&bud->list, &c->old_buds);
344 spin_unlock(&c->buds_lock);
348 * ubifs_log_start_commit - start commit.
349 * @c: UBIFS file-system description object
350 * @ltail_lnum: return new log tail LEB number
352 * The commit operation starts with writing "commit start" node to the log and
353 * reference nodes for all journal heads which will define new journal after
354 * the commit has been finished. The commit start and reference nodes are
355 * written in one go to the nearest empty log LEB (hence, when commit is
356 * finished UBIFS may safely unmap all the previous log LEBs). This function
357 * returns zero in case of success and a negative error code in case of
360 int ubifs_log_start_commit(struct ubifs_info *c, int *ltail_lnum)
363 struct ubifs_cs_node *cs;
364 struct ubifs_ref_node *ref;
365 int err, i, max_len, len;
367 err = dbg_check_bud_bytes(c);
371 max_len = UBIFS_CS_NODE_SZ + c->jhead_cnt * UBIFS_REF_NODE_SZ;
372 max_len = ALIGN(max_len, c->min_io_size);
373 buf = cs = kmalloc(max_len, GFP_NOFS);
377 cs->ch.node_type = UBIFS_CS_NODE;
378 cs->cmt_no = cpu_to_le64(c->cmt_no);
379 ubifs_prepare_node(c, cs, UBIFS_CS_NODE_SZ, 0);
382 * Note, we do not lock 'c->log_mutex' because this is the commit start
383 * phase and we are exclusively using the log. And we do not lock
384 * write-buffer because nobody can write to the file-system at this
388 len = UBIFS_CS_NODE_SZ;
389 for (i = 0; i < c->jhead_cnt; i++) {
390 int lnum = c->jheads[i].wbuf.lnum;
391 int offs = c->jheads[i].wbuf.offs;
393 if (lnum == -1 || offs == c->leb_size)
396 dbg_log("add ref to LEB %d:%d for jhead %s",
397 lnum, offs, dbg_jhead(i));
399 ref->ch.node_type = UBIFS_REF_NODE;
400 ref->lnum = cpu_to_le32(lnum);
401 ref->offs = cpu_to_le32(offs);
402 ref->jhead = cpu_to_le32(i);
404 ubifs_prepare_node(c, ref, UBIFS_REF_NODE_SZ, 0);
405 len += UBIFS_REF_NODE_SZ;
408 ubifs_pad(c, buf + len, ALIGN(len, c->min_io_size) - len);
410 /* Switch to the next log LEB */
412 c->lhead_lnum = ubifs_next_log_lnum(c, c->lhead_lnum);
416 if (c->lhead_offs == 0) {
417 /* Must ensure next LEB has been unmapped */
418 err = ubifs_leb_unmap(c, c->lhead_lnum);
423 len = ALIGN(len, c->min_io_size);
424 dbg_log("writing commit start at LEB %d:0, len %d", c->lhead_lnum, len);
425 err = ubifs_leb_write(c, c->lhead_lnum, cs, 0, len, UBI_SHORTTERM);
429 *ltail_lnum = c->lhead_lnum;
431 c->lhead_offs += len;
432 if (c->lhead_offs == c->leb_size) {
433 c->lhead_lnum = ubifs_next_log_lnum(c, c->lhead_lnum);
440 * We have started the commit and now users may use the rest of the log
443 c->min_log_bytes = 0;
451 * ubifs_log_end_commit - end commit.
452 * @c: UBIFS file-system description object
453 * @ltail_lnum: new log tail LEB number
455 * This function is called on when the commit operation was finished. It
456 * moves log tail to new position and unmaps LEBs which contain obsolete data.
457 * Returns zero in case of success and a negative error code in case of
460 int ubifs_log_end_commit(struct ubifs_info *c, int ltail_lnum)
465 * At this phase we have to lock 'c->log_mutex' because UBIFS allows FS
466 * writes during commit. Its only short "commit" start phase when
467 * writers are blocked.
469 mutex_lock(&c->log_mutex);
471 dbg_log("old tail was LEB %d:0, new tail is LEB %d:0",
472 c->ltail_lnum, ltail_lnum);
474 c->ltail_lnum = ltail_lnum;
476 * The commit is finished and from now on it must be guaranteed that
477 * there is always enough space for the next commit.
479 c->min_log_bytes = c->leb_size;
481 spin_lock(&c->buds_lock);
482 c->bud_bytes -= c->cmt_bud_bytes;
483 spin_unlock(&c->buds_lock);
485 err = dbg_check_bud_bytes(c);
487 mutex_unlock(&c->log_mutex);
492 * ubifs_log_post_commit - things to do after commit is completed.
493 * @c: UBIFS file-system description object
494 * @old_ltail_lnum: old log tail LEB number
496 * Release buds only after commit is completed, because they must be unchanged
497 * if recovery is needed.
499 * Unmap log LEBs only after commit is completed, because they may be needed for
502 * This function returns %0 on success and a negative error code on failure.
504 int ubifs_log_post_commit(struct ubifs_info *c, int old_ltail_lnum)
508 while (!list_empty(&c->old_buds)) {
509 struct ubifs_bud *bud;
511 bud = list_entry(c->old_buds.next, struct ubifs_bud, list);
512 err = ubifs_return_leb(c, bud->lnum);
515 list_del(&bud->list);
518 mutex_lock(&c->log_mutex);
519 for (lnum = old_ltail_lnum; lnum != c->ltail_lnum;
520 lnum = ubifs_next_log_lnum(c, lnum)) {
521 dbg_log("unmap log LEB %d", lnum);
522 err = ubifs_leb_unmap(c, lnum);
527 mutex_unlock(&c->log_mutex);
532 * struct done_ref - references that have been done.
542 * done_already - determine if a reference has been done already.
543 * @done_tree: rb-tree to store references that have been done
544 * @lnum: LEB number of reference
546 * This function returns %1 if the reference has been done, %0 if not, otherwise
547 * a negative error code is returned.
549 static int done_already(struct rb_root *done_tree, int lnum)
551 struct rb_node **p = &done_tree->rb_node, *parent = NULL;
556 dr = rb_entry(parent, struct done_ref, rb);
559 else if (lnum > dr->lnum)
565 dr = kzalloc(sizeof(struct done_ref), GFP_NOFS);
571 rb_link_node(&dr->rb, parent, p);
572 rb_insert_color(&dr->rb, done_tree);
578 * destroy_done_tree - destroy the done tree.
579 * @done_tree: done tree to destroy
581 static void destroy_done_tree(struct rb_root *done_tree)
583 struct rb_node *this = done_tree->rb_node;
588 this = this->rb_left;
590 } else if (this->rb_right) {
591 this = this->rb_right;
594 dr = rb_entry(this, struct done_ref, rb);
595 this = rb_parent(this);
597 if (this->rb_left == &dr->rb)
598 this->rb_left = NULL;
600 this->rb_right = NULL;
607 * add_node - add a node to the consolidated log.
608 * @c: UBIFS file-system description object
609 * @buf: buffer to which to add
610 * @lnum: LEB number to which to write is passed and returned here
611 * @offs: offset to where to write is passed and returned here
614 * This function returns %0 on success and a negative error code on failure.
616 static int add_node(struct ubifs_info *c, void *buf, int *lnum, int *offs,
619 struct ubifs_ch *ch = node;
620 int len = le32_to_cpu(ch->len), remains = c->leb_size - *offs;
623 int sz = ALIGN(*offs, c->min_io_size), err;
625 ubifs_pad(c, buf + *offs, sz - *offs);
626 err = ubifs_leb_change(c, *lnum, buf, sz, UBI_SHORTTERM);
629 *lnum = ubifs_next_log_lnum(c, *lnum);
632 memcpy(buf + *offs, node, len);
633 *offs += ALIGN(len, 8);
638 * ubifs_consolidate_log - consolidate the log.
639 * @c: UBIFS file-system description object
641 * Repeated failed commits could cause the log to be full, but at least 1 LEB is
642 * needed for commit. This function rewrites the reference nodes in the log
643 * omitting duplicates, and failed CS nodes, and leaving no gaps.
645 * This function returns %0 on success and a negative error code on failure.
647 int ubifs_consolidate_log(struct ubifs_info *c)
649 struct ubifs_scan_leb *sleb;
650 struct ubifs_scan_node *snod;
651 struct rb_root done_tree = RB_ROOT;
652 int lnum, err, first = 1, write_lnum, offs = 0;
655 dbg_rcvry("log tail LEB %d, log head LEB %d", c->ltail_lnum,
657 buf = vmalloc(c->leb_size);
660 lnum = c->ltail_lnum;
663 sleb = ubifs_scan(c, lnum, 0, c->sbuf, 0);
668 list_for_each_entry(snod, &sleb->nodes, list) {
669 switch (snod->type) {
670 case UBIFS_REF_NODE: {
671 struct ubifs_ref_node *ref = snod->node;
672 int ref_lnum = le32_to_cpu(ref->lnum);
674 err = done_already(&done_tree, ref_lnum);
678 err = add_node(c, buf, &write_lnum,
688 err = add_node(c, buf, &write_lnum, &offs,
696 ubifs_scan_destroy(sleb);
697 if (lnum == c->lhead_lnum)
699 lnum = ubifs_next_log_lnum(c, lnum);
702 int sz = ALIGN(offs, c->min_io_size);
704 ubifs_pad(c, buf + offs, sz - offs);
705 err = ubifs_leb_change(c, write_lnum, buf, sz, UBI_SHORTTERM);
708 offs = ALIGN(offs, c->min_io_size);
710 destroy_done_tree(&done_tree);
712 if (write_lnum == c->lhead_lnum) {
713 ubifs_err("log is too full");
716 /* Unmap remaining LEBs */
719 lnum = ubifs_next_log_lnum(c, lnum);
720 err = ubifs_leb_unmap(c, lnum);
723 } while (lnum != c->lhead_lnum);
724 c->lhead_lnum = write_lnum;
725 c->lhead_offs = offs;
726 dbg_rcvry("new log head at %d:%d", c->lhead_lnum, c->lhead_offs);
730 ubifs_scan_destroy(sleb);
732 destroy_done_tree(&done_tree);
737 #ifdef CONFIG_UBIFS_FS_DEBUG
740 * dbg_check_bud_bytes - make sure bud bytes calculation are all right.
741 * @c: UBIFS file-system description object
743 * This function makes sure the amount of flash space used by closed buds
744 * ('c->bud_bytes' is correct). Returns zero in case of success and %-EINVAL in
747 static int dbg_check_bud_bytes(struct ubifs_info *c)
750 struct ubifs_bud *bud;
751 long long bud_bytes = 0;
753 if (!dbg_is_chk_gen(c))
756 spin_lock(&c->buds_lock);
757 for (i = 0; i < c->jhead_cnt; i++)
758 list_for_each_entry(bud, &c->jheads[i].buds_list, list)
759 bud_bytes += c->leb_size - bud->start;
761 if (c->bud_bytes != bud_bytes) {
762 ubifs_err("bad bud_bytes %lld, calculated %lld",
763 c->bud_bytes, bud_bytes);
766 spin_unlock(&c->buds_lock);
771 #endif /* CONFIG_UBIFS_FS_DEBUG */