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;
117 else if (c->lhead_lnum != c->ltail_lnum)
124 * ubifs_add_bud - add bud LEB to the tree of buds and its journal head list.
125 * @c: UBIFS file-system description object
126 * @bud: the bud to add
128 void ubifs_add_bud(struct ubifs_info *c, struct ubifs_bud *bud)
130 struct rb_node **p, *parent = NULL;
132 struct ubifs_jhead *jhead;
134 spin_lock(&c->buds_lock);
135 p = &c->buds.rb_node;
138 b = rb_entry(parent, struct ubifs_bud, rb);
139 ubifs_assert(bud->lnum != b->lnum);
140 if (bud->lnum < b->lnum)
146 rb_link_node(&bud->rb, parent, p);
147 rb_insert_color(&bud->rb, &c->buds);
149 jhead = &c->jheads[bud->jhead];
150 list_add_tail(&bud->list, &jhead->buds_list);
152 ubifs_assert(c->replaying && c->ro_mount);
155 * Note, although this is a new bud, we anyway account this space now,
156 * before any data has been written to it, because this is about to
157 * guarantee fixed mount time, and this bud will anyway be read and
160 c->bud_bytes += c->leb_size - bud->start;
162 dbg_log("LEB %d:%d, jhead %s, bud_bytes %lld", bud->lnum,
163 bud->start, dbg_jhead(bud->jhead), c->bud_bytes);
164 spin_unlock(&c->buds_lock);
168 * ubifs_add_bud_to_log - add a new bud to the log.
169 * @c: UBIFS file-system description object
170 * @jhead: journal head the bud belongs to
171 * @lnum: LEB number of the bud
172 * @offs: starting offset of the bud
174 * This function writes reference node for the new bud LEB @lnum it to the log,
175 * and adds it to the buds tress. It also makes sure that log size does not
176 * exceed the 'c->max_bud_bytes' limit. Returns zero in case of success,
177 * %-EAGAIN if commit is required, and a negative error codes in case of
180 int ubifs_add_bud_to_log(struct ubifs_info *c, int jhead, int lnum, int offs)
183 struct ubifs_bud *bud;
184 struct ubifs_ref_node *ref;
186 bud = kmalloc(sizeof(struct ubifs_bud), GFP_NOFS);
189 ref = kzalloc(c->ref_node_alsz, GFP_NOFS);
195 mutex_lock(&c->log_mutex);
196 ubifs_assert(!c->ro_media && !c->ro_mount);
202 /* Make sure we have enough space in the log */
203 if (empty_log_bytes(c) - c->ref_node_alsz < c->min_log_bytes) {
204 dbg_log("not enough log space - %lld, required %d",
205 empty_log_bytes(c), c->min_log_bytes);
206 ubifs_commit_required(c);
212 * Make sure the amount of space in buds will not exceed the
213 * 'c->max_bud_bytes' limit, because we want to guarantee mount time
216 * It is not necessary to hold @c->buds_lock when reading @c->bud_bytes
217 * because we are holding @c->log_mutex. All @c->bud_bytes take place
218 * when both @c->log_mutex and @c->bud_bytes are locked.
220 if (c->bud_bytes + c->leb_size - offs > c->max_bud_bytes) {
221 dbg_log("bud bytes %lld (%lld max), require commit",
222 c->bud_bytes, c->max_bud_bytes);
223 ubifs_commit_required(c);
229 * If the journal is full enough - start background commit. Note, it is
230 * OK to read 'c->cmt_state' without spinlock because integer reads
231 * are atomic in the kernel.
233 if (c->bud_bytes >= c->bg_bud_bytes &&
234 c->cmt_state == COMMIT_RESTING) {
235 dbg_log("bud bytes %lld (%lld max), initiate BG commit",
236 c->bud_bytes, c->max_bud_bytes);
237 ubifs_request_bg_commit(c);
244 ref->ch.node_type = UBIFS_REF_NODE;
245 ref->lnum = cpu_to_le32(bud->lnum);
246 ref->offs = cpu_to_le32(bud->start);
247 ref->jhead = cpu_to_le32(jhead);
249 if (c->lhead_offs > c->leb_size - c->ref_node_alsz) {
250 c->lhead_lnum = ubifs_next_log_lnum(c, c->lhead_lnum);
254 if (c->lhead_offs == 0) {
255 /* Must ensure next log LEB has been unmapped */
256 err = ubifs_leb_unmap(c, c->lhead_lnum);
261 if (bud->start == 0) {
263 * Before writing the LEB reference which refers an empty LEB
264 * to the log, we have to make sure it is mapped, because
265 * otherwise we'd risk to refer an LEB with garbage in case of
266 * an unclean reboot, because the target LEB might have been
267 * unmapped, but not yet physically erased.
269 err = ubifs_leb_map(c, bud->lnum, UBI_SHORTTERM);
274 dbg_log("write ref LEB %d:%d",
275 c->lhead_lnum, c->lhead_offs);
276 err = ubifs_write_node(c, ref, UBIFS_REF_NODE_SZ, c->lhead_lnum,
277 c->lhead_offs, UBI_SHORTTERM);
281 c->lhead_offs += c->ref_node_alsz;
283 ubifs_add_bud(c, bud);
285 mutex_unlock(&c->log_mutex);
290 mutex_unlock(&c->log_mutex);
297 * remove_buds - remove used buds.
298 * @c: UBIFS file-system description object
300 * This function removes use buds from the buds tree. It does not remove the
301 * buds which are pointed to by journal heads.
303 static void remove_buds(struct ubifs_info *c)
307 ubifs_assert(list_empty(&c->old_buds));
308 c->cmt_bud_bytes = 0;
309 spin_lock(&c->buds_lock);
310 p = rb_first(&c->buds);
312 struct rb_node *p1 = p;
313 struct ubifs_bud *bud;
314 struct ubifs_wbuf *wbuf;
317 bud = rb_entry(p1, struct ubifs_bud, rb);
318 wbuf = &c->jheads[bud->jhead].wbuf;
320 if (wbuf->lnum == bud->lnum) {
322 * Do not remove buds which are pointed to by journal
323 * heads (non-closed buds).
325 c->cmt_bud_bytes += wbuf->offs - bud->start;
326 dbg_log("preserve %d:%d, jhead %s, bud bytes %d, "
327 "cmt_bud_bytes %lld", bud->lnum, bud->start,
328 dbg_jhead(bud->jhead), wbuf->offs - bud->start,
330 bud->start = wbuf->offs;
332 c->cmt_bud_bytes += c->leb_size - bud->start;
333 dbg_log("remove %d:%d, jhead %s, bud bytes %d, "
334 "cmt_bud_bytes %lld", bud->lnum, bud->start,
335 dbg_jhead(bud->jhead), c->leb_size - bud->start,
337 rb_erase(p1, &c->buds);
339 * If the commit does not finish, the recovery will need
340 * to replay the journal, in which case the old buds
341 * must be unchanged. Do not release them until post
342 * commit i.e. do not allow them to be garbage
345 list_move(&bud->list, &c->old_buds);
348 spin_unlock(&c->buds_lock);
352 * ubifs_log_start_commit - start commit.
353 * @c: UBIFS file-system description object
354 * @ltail_lnum: return new log tail LEB number
356 * The commit operation starts with writing "commit start" node to the log and
357 * reference nodes for all journal heads which will define new journal after
358 * the commit has been finished. The commit start and reference nodes are
359 * written in one go to the nearest empty log LEB (hence, when commit is
360 * finished UBIFS may safely unmap all the previous log LEBs). This function
361 * returns zero in case of success and a negative error code in case of
364 int ubifs_log_start_commit(struct ubifs_info *c, int *ltail_lnum)
367 struct ubifs_cs_node *cs;
368 struct ubifs_ref_node *ref;
369 int err, i, max_len, len;
371 err = dbg_check_bud_bytes(c);
375 max_len = UBIFS_CS_NODE_SZ + c->jhead_cnt * UBIFS_REF_NODE_SZ;
376 max_len = ALIGN(max_len, c->min_io_size);
377 buf = cs = kmalloc(max_len, GFP_NOFS);
381 cs->ch.node_type = UBIFS_CS_NODE;
382 cs->cmt_no = cpu_to_le64(c->cmt_no);
383 ubifs_prepare_node(c, cs, UBIFS_CS_NODE_SZ, 0);
386 * Note, we do not lock 'c->log_mutex' because this is the commit start
387 * phase and we are exclusively using the log. And we do not lock
388 * write-buffer because nobody can write to the file-system at this
392 len = UBIFS_CS_NODE_SZ;
393 for (i = 0; i < c->jhead_cnt; i++) {
394 int lnum = c->jheads[i].wbuf.lnum;
395 int offs = c->jheads[i].wbuf.offs;
397 if (lnum == -1 || offs == c->leb_size)
400 dbg_log("add ref to LEB %d:%d for jhead %s",
401 lnum, offs, dbg_jhead(i));
403 ref->ch.node_type = UBIFS_REF_NODE;
404 ref->lnum = cpu_to_le32(lnum);
405 ref->offs = cpu_to_le32(offs);
406 ref->jhead = cpu_to_le32(i);
408 ubifs_prepare_node(c, ref, UBIFS_REF_NODE_SZ, 0);
409 len += UBIFS_REF_NODE_SZ;
412 ubifs_pad(c, buf + len, ALIGN(len, c->min_io_size) - len);
414 /* Switch to the next log LEB */
416 c->lhead_lnum = ubifs_next_log_lnum(c, c->lhead_lnum);
420 if (c->lhead_offs == 0) {
421 /* Must ensure next LEB has been unmapped */
422 err = ubifs_leb_unmap(c, c->lhead_lnum);
427 len = ALIGN(len, c->min_io_size);
428 dbg_log("writing commit start at LEB %d:0, len %d", c->lhead_lnum, len);
429 err = ubifs_leb_write(c, c->lhead_lnum, cs, 0, len, UBI_SHORTTERM);
433 *ltail_lnum = c->lhead_lnum;
435 c->lhead_offs += len;
436 if (c->lhead_offs == c->leb_size) {
437 c->lhead_lnum = ubifs_next_log_lnum(c, c->lhead_lnum);
444 * We have started the commit and now users may use the rest of the log
447 c->min_log_bytes = 0;
455 * ubifs_log_end_commit - end commit.
456 * @c: UBIFS file-system description object
457 * @ltail_lnum: new log tail LEB number
459 * This function is called on when the commit operation was finished. It
460 * moves log tail to new position and updates the master node so that it stores
461 * the new log tail LEB number. Returns zero in case of success and a negative
462 * error code in case of failure.
464 int ubifs_log_end_commit(struct ubifs_info *c, int ltail_lnum)
469 * At this phase we have to lock 'c->log_mutex' because UBIFS allows FS
470 * writes during commit. Its only short "commit" start phase when
471 * writers are blocked.
473 mutex_lock(&c->log_mutex);
475 dbg_log("old tail was LEB %d:0, new tail is LEB %d:0",
476 c->ltail_lnum, ltail_lnum);
478 c->ltail_lnum = ltail_lnum;
480 * The commit is finished and from now on it must be guaranteed that
481 * there is always enough space for the next commit.
483 c->min_log_bytes = c->leb_size;
485 spin_lock(&c->buds_lock);
486 c->bud_bytes -= c->cmt_bud_bytes;
487 spin_unlock(&c->buds_lock);
489 err = dbg_check_bud_bytes(c);
493 err = ubifs_write_master(c);
496 mutex_unlock(&c->log_mutex);
501 * ubifs_log_post_commit - things to do after commit is completed.
502 * @c: UBIFS file-system description object
503 * @old_ltail_lnum: old log tail LEB number
505 * Release buds only after commit is completed, because they must be unchanged
506 * if recovery is needed.
508 * Unmap log LEBs only after commit is completed, because they may be needed for
511 * This function returns %0 on success and a negative error code on failure.
513 int ubifs_log_post_commit(struct ubifs_info *c, int old_ltail_lnum)
517 while (!list_empty(&c->old_buds)) {
518 struct ubifs_bud *bud;
520 bud = list_entry(c->old_buds.next, struct ubifs_bud, list);
521 err = ubifs_return_leb(c, bud->lnum);
524 list_del(&bud->list);
527 mutex_lock(&c->log_mutex);
528 for (lnum = old_ltail_lnum; lnum != c->ltail_lnum;
529 lnum = ubifs_next_log_lnum(c, lnum)) {
530 dbg_log("unmap log LEB %d", lnum);
531 err = ubifs_leb_unmap(c, lnum);
536 mutex_unlock(&c->log_mutex);
541 * struct done_ref - references that have been done.
551 * done_already - determine if a reference has been done already.
552 * @done_tree: rb-tree to store references that have been done
553 * @lnum: LEB number of reference
555 * This function returns %1 if the reference has been done, %0 if not, otherwise
556 * a negative error code is returned.
558 static int done_already(struct rb_root *done_tree, int lnum)
560 struct rb_node **p = &done_tree->rb_node, *parent = NULL;
565 dr = rb_entry(parent, struct done_ref, rb);
568 else if (lnum > dr->lnum)
574 dr = kzalloc(sizeof(struct done_ref), GFP_NOFS);
580 rb_link_node(&dr->rb, parent, p);
581 rb_insert_color(&dr->rb, done_tree);
587 * destroy_done_tree - destroy the done tree.
588 * @done_tree: done tree to destroy
590 static void destroy_done_tree(struct rb_root *done_tree)
592 struct rb_node *this = done_tree->rb_node;
597 this = this->rb_left;
599 } else if (this->rb_right) {
600 this = this->rb_right;
603 dr = rb_entry(this, struct done_ref, rb);
604 this = rb_parent(this);
606 if (this->rb_left == &dr->rb)
607 this->rb_left = NULL;
609 this->rb_right = NULL;
616 * add_node - add a node to the consolidated log.
617 * @c: UBIFS file-system description object
618 * @buf: buffer to which to add
619 * @lnum: LEB number to which to write is passed and returned here
620 * @offs: offset to where to write is passed and returned here
623 * This function returns %0 on success and a negative error code on failure.
625 static int add_node(struct ubifs_info *c, void *buf, int *lnum, int *offs,
628 struct ubifs_ch *ch = node;
629 int len = le32_to_cpu(ch->len), remains = c->leb_size - *offs;
632 int sz = ALIGN(*offs, c->min_io_size), err;
634 ubifs_pad(c, buf + *offs, sz - *offs);
635 err = ubifs_leb_change(c, *lnum, buf, sz, UBI_SHORTTERM);
638 *lnum = ubifs_next_log_lnum(c, *lnum);
641 memcpy(buf + *offs, node, len);
642 *offs += ALIGN(len, 8);
647 * ubifs_consolidate_log - consolidate the log.
648 * @c: UBIFS file-system description object
650 * Repeated failed commits could cause the log to be full, but at least 1 LEB is
651 * needed for commit. This function rewrites the reference nodes in the log
652 * omitting duplicates, and failed CS nodes, and leaving no gaps.
654 * This function returns %0 on success and a negative error code on failure.
656 int ubifs_consolidate_log(struct ubifs_info *c)
658 struct ubifs_scan_leb *sleb;
659 struct ubifs_scan_node *snod;
660 struct rb_root done_tree = RB_ROOT;
661 int lnum, err, first = 1, write_lnum, offs = 0;
664 dbg_rcvry("log tail LEB %d, log head LEB %d", c->ltail_lnum,
666 buf = vmalloc(c->leb_size);
669 lnum = c->ltail_lnum;
672 sleb = ubifs_scan(c, lnum, 0, c->sbuf, 0);
677 list_for_each_entry(snod, &sleb->nodes, list) {
678 switch (snod->type) {
679 case UBIFS_REF_NODE: {
680 struct ubifs_ref_node *ref = snod->node;
681 int ref_lnum = le32_to_cpu(ref->lnum);
683 err = done_already(&done_tree, ref_lnum);
687 err = add_node(c, buf, &write_lnum,
697 err = add_node(c, buf, &write_lnum, &offs,
705 ubifs_scan_destroy(sleb);
706 if (lnum == c->lhead_lnum)
708 lnum = ubifs_next_log_lnum(c, lnum);
711 int sz = ALIGN(offs, c->min_io_size);
713 ubifs_pad(c, buf + offs, sz - offs);
714 err = ubifs_leb_change(c, write_lnum, buf, sz, UBI_SHORTTERM);
717 offs = ALIGN(offs, c->min_io_size);
719 destroy_done_tree(&done_tree);
721 if (write_lnum == c->lhead_lnum) {
722 ubifs_err("log is too full");
725 /* Unmap remaining LEBs */
728 lnum = ubifs_next_log_lnum(c, lnum);
729 err = ubifs_leb_unmap(c, lnum);
732 } while (lnum != c->lhead_lnum);
733 c->lhead_lnum = write_lnum;
734 c->lhead_offs = offs;
735 dbg_rcvry("new log head at %d:%d", c->lhead_lnum, c->lhead_offs);
739 ubifs_scan_destroy(sleb);
741 destroy_done_tree(&done_tree);
746 #ifdef CONFIG_UBIFS_FS_DEBUG
749 * dbg_check_bud_bytes - make sure bud bytes calculation are all right.
750 * @c: UBIFS file-system description object
752 * This function makes sure the amount of flash space used by closed buds
753 * ('c->bud_bytes' is correct). Returns zero in case of success and %-EINVAL in
756 static int dbg_check_bud_bytes(struct ubifs_info *c)
759 struct ubifs_bud *bud;
760 long long bud_bytes = 0;
762 if (!dbg_is_chk_gen(c))
765 spin_lock(&c->buds_lock);
766 for (i = 0; i < c->jhead_cnt; i++)
767 list_for_each_entry(bud, &c->jheads[i].buds_list, list)
768 bud_bytes += c->leb_size - bud->start;
770 if (c->bud_bytes != bud_bytes) {
771 ubifs_err("bad bud_bytes %lld, calculated %lld",
772 c->bud_bytes, bud_bytes);
775 spin_unlock(&c->buds_lock);
780 #endif /* CONFIG_UBIFS_FS_DEBUG */