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;
116 * ubifs_add_bud - add bud LEB to the tree of buds and its journal head list.
117 * @c: UBIFS file-system description object
118 * @bud: the bud to add
120 void ubifs_add_bud(struct ubifs_info *c, struct ubifs_bud *bud)
122 struct rb_node **p, *parent = NULL;
124 struct ubifs_jhead *jhead;
126 spin_lock(&c->buds_lock);
127 p = &c->buds.rb_node;
130 b = rb_entry(parent, struct ubifs_bud, rb);
131 ubifs_assert(bud->lnum != b->lnum);
132 if (bud->lnum < b->lnum)
138 rb_link_node(&bud->rb, parent, p);
139 rb_insert_color(&bud->rb, &c->buds);
141 jhead = &c->jheads[bud->jhead];
142 list_add_tail(&bud->list, &jhead->buds_list);
144 ubifs_assert(c->replaying && c->ro_mount);
147 * Note, although this is a new bud, we anyway account this space now,
148 * before any data has been written to it, because this is about to
149 * guarantee fixed mount time, and this bud will anyway be read and
152 c->bud_bytes += c->leb_size - bud->start;
154 dbg_log("LEB %d:%d, jhead %s, bud_bytes %lld", bud->lnum,
155 bud->start, dbg_jhead(bud->jhead), c->bud_bytes);
156 spin_unlock(&c->buds_lock);
160 * ubifs_add_bud_to_log - add a new bud to the log.
161 * @c: UBIFS file-system description object
162 * @jhead: journal head the bud belongs to
163 * @lnum: LEB number of the bud
164 * @offs: starting offset of the bud
166 * This function writes reference node for the new bud LEB @lnum it to the log,
167 * and adds it to the buds tress. It also makes sure that log size does not
168 * exceed the 'c->max_bud_bytes' limit. Returns zero in case of success,
169 * %-EAGAIN if commit is required, and a negative error codes in case of
172 int ubifs_add_bud_to_log(struct ubifs_info *c, int jhead, int lnum, int offs)
175 struct ubifs_bud *bud;
176 struct ubifs_ref_node *ref;
178 bud = kmalloc(sizeof(struct ubifs_bud), GFP_NOFS);
181 ref = kzalloc(c->ref_node_alsz, GFP_NOFS);
187 mutex_lock(&c->log_mutex);
188 ubifs_assert(!c->ro_media && !c->ro_mount);
194 /* Make sure we have enough space in the log */
195 if (empty_log_bytes(c) - c->ref_node_alsz < c->min_log_bytes) {
196 dbg_log("not enough log space - %lld, required %d",
197 empty_log_bytes(c), c->min_log_bytes);
198 ubifs_commit_required(c);
204 * Make sure the amount of space in buds will not exceed the
205 * 'c->max_bud_bytes' limit, because we want to guarantee mount time
208 * It is not necessary to hold @c->buds_lock when reading @c->bud_bytes
209 * because we are holding @c->log_mutex. All @c->bud_bytes take place
210 * when both @c->log_mutex and @c->bud_bytes are locked.
212 if (c->bud_bytes + c->leb_size - offs > c->max_bud_bytes) {
213 dbg_log("bud bytes %lld (%lld max), require commit",
214 c->bud_bytes, c->max_bud_bytes);
215 ubifs_commit_required(c);
221 * If the journal is full enough - start background commit. Note, it is
222 * OK to read 'c->cmt_state' without spinlock because integer reads
223 * are atomic in the kernel.
225 if (c->bud_bytes >= c->bg_bud_bytes &&
226 c->cmt_state == COMMIT_RESTING) {
227 dbg_log("bud bytes %lld (%lld max), initiate BG commit",
228 c->bud_bytes, c->max_bud_bytes);
229 ubifs_request_bg_commit(c);
236 ref->ch.node_type = UBIFS_REF_NODE;
237 ref->lnum = cpu_to_le32(bud->lnum);
238 ref->offs = cpu_to_le32(bud->start);
239 ref->jhead = cpu_to_le32(jhead);
241 if (c->lhead_offs > c->leb_size - c->ref_node_alsz) {
242 c->lhead_lnum = ubifs_next_log_lnum(c, c->lhead_lnum);
246 if (c->lhead_offs == 0) {
247 /* Must ensure next log LEB has been unmapped */
248 err = ubifs_leb_unmap(c, c->lhead_lnum);
253 if (bud->start == 0) {
255 * Before writing the LEB reference which refers an empty LEB
256 * to the log, we have to make sure it is mapped, because
257 * otherwise we'd risk to refer an LEB with garbage in case of
258 * an unclean reboot, because the target LEB might have been
259 * unmapped, but not yet physically erased.
261 err = ubifs_leb_map(c, bud->lnum);
266 dbg_log("write ref LEB %d:%d",
267 c->lhead_lnum, c->lhead_offs);
268 err = ubifs_write_node(c, ref, UBIFS_REF_NODE_SZ, c->lhead_lnum,
273 c->lhead_offs += c->ref_node_alsz;
275 ubifs_add_bud(c, bud);
277 mutex_unlock(&c->log_mutex);
282 mutex_unlock(&c->log_mutex);
289 * remove_buds - remove used buds.
290 * @c: UBIFS file-system description object
292 * This function removes use buds from the buds tree. It does not remove the
293 * buds which are pointed to by journal heads.
295 static void remove_buds(struct ubifs_info *c)
299 ubifs_assert(list_empty(&c->old_buds));
300 c->cmt_bud_bytes = 0;
301 spin_lock(&c->buds_lock);
302 p = rb_first(&c->buds);
304 struct rb_node *p1 = p;
305 struct ubifs_bud *bud;
306 struct ubifs_wbuf *wbuf;
309 bud = rb_entry(p1, struct ubifs_bud, rb);
310 wbuf = &c->jheads[bud->jhead].wbuf;
312 if (wbuf->lnum == bud->lnum) {
314 * Do not remove buds which are pointed to by journal
315 * heads (non-closed buds).
317 c->cmt_bud_bytes += wbuf->offs - bud->start;
318 dbg_log("preserve %d:%d, jhead %s, bud bytes %d, cmt_bud_bytes %lld",
319 bud->lnum, bud->start, dbg_jhead(bud->jhead),
320 wbuf->offs - bud->start, c->cmt_bud_bytes);
321 bud->start = wbuf->offs;
323 c->cmt_bud_bytes += c->leb_size - bud->start;
324 dbg_log("remove %d:%d, jhead %s, bud bytes %d, cmt_bud_bytes %lld",
325 bud->lnum, bud->start, dbg_jhead(bud->jhead),
326 c->leb_size - bud->start, c->cmt_bud_bytes);
327 rb_erase(p1, &c->buds);
329 * If the commit does not finish, the recovery will need
330 * to replay the journal, in which case the old buds
331 * must be unchanged. Do not release them until post
332 * commit i.e. do not allow them to be garbage
335 list_move(&bud->list, &c->old_buds);
338 spin_unlock(&c->buds_lock);
342 * ubifs_log_start_commit - start commit.
343 * @c: UBIFS file-system description object
344 * @ltail_lnum: return new log tail LEB number
346 * The commit operation starts with writing "commit start" node to the log and
347 * reference nodes for all journal heads which will define new journal after
348 * the commit has been finished. The commit start and reference nodes are
349 * written in one go to the nearest empty log LEB (hence, when commit is
350 * finished UBIFS may safely unmap all the previous log LEBs). This function
351 * returns zero in case of success and a negative error code in case of
354 int ubifs_log_start_commit(struct ubifs_info *c, int *ltail_lnum)
357 struct ubifs_cs_node *cs;
358 struct ubifs_ref_node *ref;
359 int err, i, max_len, len;
361 err = dbg_check_bud_bytes(c);
365 max_len = UBIFS_CS_NODE_SZ + c->jhead_cnt * UBIFS_REF_NODE_SZ;
366 max_len = ALIGN(max_len, c->min_io_size);
367 buf = cs = kmalloc(max_len, GFP_NOFS);
371 cs->ch.node_type = UBIFS_CS_NODE;
372 cs->cmt_no = cpu_to_le64(c->cmt_no);
373 ubifs_prepare_node(c, cs, UBIFS_CS_NODE_SZ, 0);
376 * Note, we do not lock 'c->log_mutex' because this is the commit start
377 * phase and we are exclusively using the log. And we do not lock
378 * write-buffer because nobody can write to the file-system at this
382 len = UBIFS_CS_NODE_SZ;
383 for (i = 0; i < c->jhead_cnt; i++) {
384 int lnum = c->jheads[i].wbuf.lnum;
385 int offs = c->jheads[i].wbuf.offs;
387 if (lnum == -1 || offs == c->leb_size)
390 dbg_log("add ref to LEB %d:%d for jhead %s",
391 lnum, offs, dbg_jhead(i));
393 ref->ch.node_type = UBIFS_REF_NODE;
394 ref->lnum = cpu_to_le32(lnum);
395 ref->offs = cpu_to_le32(offs);
396 ref->jhead = cpu_to_le32(i);
398 ubifs_prepare_node(c, ref, UBIFS_REF_NODE_SZ, 0);
399 len += UBIFS_REF_NODE_SZ;
402 ubifs_pad(c, buf + len, ALIGN(len, c->min_io_size) - len);
404 /* Switch to the next log LEB */
406 c->lhead_lnum = ubifs_next_log_lnum(c, c->lhead_lnum);
410 if (c->lhead_offs == 0) {
411 /* Must ensure next LEB has been unmapped */
412 err = ubifs_leb_unmap(c, c->lhead_lnum);
417 len = ALIGN(len, c->min_io_size);
418 dbg_log("writing commit start at LEB %d:0, len %d", c->lhead_lnum, len);
419 err = ubifs_leb_write(c, c->lhead_lnum, cs, 0, len);
423 *ltail_lnum = c->lhead_lnum;
425 c->lhead_offs += len;
426 if (c->lhead_offs == c->leb_size) {
427 c->lhead_lnum = ubifs_next_log_lnum(c, c->lhead_lnum);
434 * We have started the commit and now users may use the rest of the log
437 c->min_log_bytes = 0;
445 * ubifs_log_end_commit - end commit.
446 * @c: UBIFS file-system description object
447 * @ltail_lnum: new log tail LEB number
449 * This function is called on when the commit operation was finished. It
450 * moves log tail to new position and updates the master node so that it stores
451 * the new log tail LEB number. Returns zero in case of success and a negative
452 * error code in case of failure.
454 int ubifs_log_end_commit(struct ubifs_info *c, int ltail_lnum)
459 * At this phase we have to lock 'c->log_mutex' because UBIFS allows FS
460 * writes during commit. Its only short "commit" start phase when
461 * writers are blocked.
463 mutex_lock(&c->log_mutex);
465 dbg_log("old tail was LEB %d:0, new tail is LEB %d:0",
466 c->ltail_lnum, ltail_lnum);
468 c->ltail_lnum = ltail_lnum;
470 * The commit is finished and from now on it must be guaranteed that
471 * there is always enough space for the next commit.
473 c->min_log_bytes = c->leb_size;
475 spin_lock(&c->buds_lock);
476 c->bud_bytes -= c->cmt_bud_bytes;
477 spin_unlock(&c->buds_lock);
479 err = dbg_check_bud_bytes(c);
483 err = ubifs_write_master(c);
486 mutex_unlock(&c->log_mutex);
491 * ubifs_log_post_commit - things to do after commit is completed.
492 * @c: UBIFS file-system description object
493 * @old_ltail_lnum: old log tail LEB number
495 * Release buds only after commit is completed, because they must be unchanged
496 * if recovery is needed.
498 * Unmap log LEBs only after commit is completed, because they may be needed for
501 * This function returns %0 on success and a negative error code on failure.
503 int ubifs_log_post_commit(struct ubifs_info *c, int old_ltail_lnum)
507 while (!list_empty(&c->old_buds)) {
508 struct ubifs_bud *bud;
510 bud = list_entry(c->old_buds.next, struct ubifs_bud, list);
511 err = ubifs_return_leb(c, bud->lnum);
514 list_del(&bud->list);
517 mutex_lock(&c->log_mutex);
518 for (lnum = old_ltail_lnum; lnum != c->ltail_lnum;
519 lnum = ubifs_next_log_lnum(c, lnum)) {
520 dbg_log("unmap log LEB %d", lnum);
521 err = ubifs_leb_unmap(c, lnum);
526 mutex_unlock(&c->log_mutex);
531 * struct done_ref - references that have been done.
541 * done_already - determine if a reference has been done already.
542 * @done_tree: rb-tree to store references that have been done
543 * @lnum: LEB number of reference
545 * This function returns %1 if the reference has been done, %0 if not, otherwise
546 * a negative error code is returned.
548 static int done_already(struct rb_root *done_tree, int lnum)
550 struct rb_node **p = &done_tree->rb_node, *parent = NULL;
555 dr = rb_entry(parent, struct done_ref, rb);
558 else if (lnum > dr->lnum)
564 dr = kzalloc(sizeof(struct done_ref), GFP_NOFS);
570 rb_link_node(&dr->rb, parent, p);
571 rb_insert_color(&dr->rb, done_tree);
577 * destroy_done_tree - destroy the done tree.
578 * @done_tree: done tree to destroy
580 static void destroy_done_tree(struct rb_root *done_tree)
582 struct rb_node *this = done_tree->rb_node;
587 this = this->rb_left;
589 } else if (this->rb_right) {
590 this = this->rb_right;
593 dr = rb_entry(this, struct done_ref, rb);
594 this = rb_parent(this);
596 if (this->rb_left == &dr->rb)
597 this->rb_left = NULL;
599 this->rb_right = NULL;
606 * add_node - add a node to the consolidated log.
607 * @c: UBIFS file-system description object
608 * @buf: buffer to which to add
609 * @lnum: LEB number to which to write is passed and returned here
610 * @offs: offset to where to write is passed and returned here
613 * This function returns %0 on success and a negative error code on failure.
615 static int add_node(struct ubifs_info *c, void *buf, int *lnum, int *offs,
618 struct ubifs_ch *ch = node;
619 int len = le32_to_cpu(ch->len), remains = c->leb_size - *offs;
622 int sz = ALIGN(*offs, c->min_io_size), err;
624 ubifs_pad(c, buf + *offs, sz - *offs);
625 err = ubifs_leb_change(c, *lnum, buf, sz);
628 *lnum = ubifs_next_log_lnum(c, *lnum);
631 memcpy(buf + *offs, node, len);
632 *offs += ALIGN(len, 8);
637 * ubifs_consolidate_log - consolidate the log.
638 * @c: UBIFS file-system description object
640 * Repeated failed commits could cause the log to be full, but at least 1 LEB is
641 * needed for commit. This function rewrites the reference nodes in the log
642 * omitting duplicates, and failed CS nodes, and leaving no gaps.
644 * This function returns %0 on success and a negative error code on failure.
646 int ubifs_consolidate_log(struct ubifs_info *c)
648 struct ubifs_scan_leb *sleb;
649 struct ubifs_scan_node *snod;
650 struct rb_root done_tree = RB_ROOT;
651 int lnum, err, first = 1, write_lnum, offs = 0;
654 dbg_rcvry("log tail LEB %d, log head LEB %d", c->ltail_lnum,
656 buf = vmalloc(c->leb_size);
659 lnum = c->ltail_lnum;
662 sleb = ubifs_scan(c, lnum, 0, c->sbuf, 0);
667 list_for_each_entry(snod, &sleb->nodes, list) {
668 switch (snod->type) {
669 case UBIFS_REF_NODE: {
670 struct ubifs_ref_node *ref = snod->node;
671 int ref_lnum = le32_to_cpu(ref->lnum);
673 err = done_already(&done_tree, ref_lnum);
677 err = add_node(c, buf, &write_lnum,
687 err = add_node(c, buf, &write_lnum, &offs,
695 ubifs_scan_destroy(sleb);
696 if (lnum == c->lhead_lnum)
698 lnum = ubifs_next_log_lnum(c, lnum);
701 int sz = ALIGN(offs, c->min_io_size);
703 ubifs_pad(c, buf + offs, sz - offs);
704 err = ubifs_leb_change(c, write_lnum, buf, sz);
707 offs = ALIGN(offs, c->min_io_size);
709 destroy_done_tree(&done_tree);
711 if (write_lnum == c->lhead_lnum) {
712 ubifs_err("log is too full");
715 /* Unmap remaining LEBs */
718 lnum = ubifs_next_log_lnum(c, lnum);
719 err = ubifs_leb_unmap(c, lnum);
722 } while (lnum != c->lhead_lnum);
723 c->lhead_lnum = write_lnum;
724 c->lhead_offs = offs;
725 dbg_rcvry("new log head at %d:%d", c->lhead_lnum, c->lhead_offs);
729 ubifs_scan_destroy(sleb);
731 destroy_done_tree(&done_tree);
737 * dbg_check_bud_bytes - make sure bud bytes calculation are all right.
738 * @c: UBIFS file-system description object
740 * This function makes sure the amount of flash space used by closed buds
741 * ('c->bud_bytes' is correct). Returns zero in case of success and %-EINVAL in
744 static int dbg_check_bud_bytes(struct ubifs_info *c)
747 struct ubifs_bud *bud;
748 long long bud_bytes = 0;
750 if (!dbg_is_chk_gen(c))
753 spin_lock(&c->buds_lock);
754 for (i = 0; i < c->jhead_cnt; i++)
755 list_for_each_entry(bud, &c->jheads[i].buds_list, list)
756 bud_bytes += c->leb_size - bud->start;
758 if (c->bud_bytes != bud_bytes) {
759 ubifs_err("bad bud_bytes %lld, calculated %lld",
760 c->bud_bytes, bud_bytes);
763 spin_unlock(&c->buds_lock);