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(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);
247 ubifs_assert(c->lhead_lnum != c->ltail_lnum);
251 if (c->lhead_offs == 0) {
252 /* Must ensure next log LEB has been unmapped */
253 err = ubifs_leb_unmap(c, c->lhead_lnum);
258 if (bud->start == 0) {
260 * Before writing the LEB reference which refers an empty LEB
261 * to the log, we have to make sure it is mapped, because
262 * otherwise we'd risk to refer an LEB with garbage in case of
263 * an unclean reboot, because the target LEB might have been
264 * unmapped, but not yet physically erased.
266 err = ubifs_leb_map(c, bud->lnum);
271 dbg_log("write ref LEB %d:%d",
272 c->lhead_lnum, c->lhead_offs);
273 err = ubifs_write_node(c, ref, UBIFS_REF_NODE_SZ, c->lhead_lnum,
278 c->lhead_offs += c->ref_node_alsz;
280 ubifs_add_bud(c, bud);
282 mutex_unlock(&c->log_mutex);
287 mutex_unlock(&c->log_mutex);
294 * remove_buds - remove used buds.
295 * @c: UBIFS file-system description object
297 * This function removes use buds from the buds tree. It does not remove the
298 * buds which are pointed to by journal heads.
300 static void remove_buds(struct ubifs_info *c)
304 ubifs_assert(list_empty(&c->old_buds));
305 c->cmt_bud_bytes = 0;
306 spin_lock(&c->buds_lock);
307 p = rb_first(&c->buds);
309 struct rb_node *p1 = p;
310 struct ubifs_bud *bud;
311 struct ubifs_wbuf *wbuf;
314 bud = rb_entry(p1, struct ubifs_bud, rb);
315 wbuf = &c->jheads[bud->jhead].wbuf;
317 if (wbuf->lnum == bud->lnum) {
319 * Do not remove buds which are pointed to by journal
320 * heads (non-closed buds).
322 c->cmt_bud_bytes += wbuf->offs - bud->start;
323 dbg_log("preserve %d:%d, jhead %s, bud bytes %d, cmt_bud_bytes %lld",
324 bud->lnum, bud->start, dbg_jhead(bud->jhead),
325 wbuf->offs - bud->start, c->cmt_bud_bytes);
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, cmt_bud_bytes %lld",
330 bud->lnum, bud->start, dbg_jhead(bud->jhead),
331 c->leb_size - bud->start, c->cmt_bud_bytes);
332 rb_erase(p1, &c->buds);
334 * If the commit does not finish, the recovery will need
335 * to replay the journal, in which case the old buds
336 * must be unchanged. Do not release them until post
337 * commit i.e. do not allow them to be garbage
340 list_move(&bud->list, &c->old_buds);
343 spin_unlock(&c->buds_lock);
347 * ubifs_log_start_commit - start commit.
348 * @c: UBIFS file-system description object
349 * @ltail_lnum: return new log tail LEB number
351 * The commit operation starts with writing "commit start" node to the log and
352 * reference nodes for all journal heads which will define new journal after
353 * the commit has been finished. The commit start and reference nodes are
354 * written in one go to the nearest empty log LEB (hence, when commit is
355 * finished UBIFS may safely unmap all the previous log LEBs). This function
356 * returns zero in case of success and a negative error code in case of
359 int ubifs_log_start_commit(struct ubifs_info *c, int *ltail_lnum)
362 struct ubifs_cs_node *cs;
363 struct ubifs_ref_node *ref;
364 int err, i, max_len, len;
366 err = dbg_check_bud_bytes(c);
370 max_len = UBIFS_CS_NODE_SZ + c->jhead_cnt * UBIFS_REF_NODE_SZ;
371 max_len = ALIGN(max_len, c->min_io_size);
372 buf = cs = kmalloc(max_len, GFP_NOFS);
376 cs->ch.node_type = UBIFS_CS_NODE;
377 cs->cmt_no = cpu_to_le64(c->cmt_no);
378 ubifs_prepare_node(c, cs, UBIFS_CS_NODE_SZ, 0);
381 * Note, we do not lock 'c->log_mutex' because this is the commit start
382 * phase and we are exclusively using the log. And we do not lock
383 * write-buffer because nobody can write to the file-system at this
387 len = UBIFS_CS_NODE_SZ;
388 for (i = 0; i < c->jhead_cnt; i++) {
389 int lnum = c->jheads[i].wbuf.lnum;
390 int offs = c->jheads[i].wbuf.offs;
392 if (lnum == -1 || offs == c->leb_size)
395 dbg_log("add ref to LEB %d:%d for jhead %s",
396 lnum, offs, dbg_jhead(i));
398 ref->ch.node_type = UBIFS_REF_NODE;
399 ref->lnum = cpu_to_le32(lnum);
400 ref->offs = cpu_to_le32(offs);
401 ref->jhead = cpu_to_le32(i);
403 ubifs_prepare_node(c, ref, UBIFS_REF_NODE_SZ, 0);
404 len += UBIFS_REF_NODE_SZ;
407 ubifs_pad(c, buf + len, ALIGN(len, c->min_io_size) - len);
409 /* Switch to the next log LEB */
411 c->lhead_lnum = ubifs_next_log_lnum(c, c->lhead_lnum);
412 ubifs_assert(c->lhead_lnum != c->ltail_lnum);
416 /* Must ensure next LEB has been unmapped */
417 err = ubifs_leb_unmap(c, c->lhead_lnum);
421 len = ALIGN(len, c->min_io_size);
422 dbg_log("writing commit start at LEB %d:0, len %d", c->lhead_lnum, len);
423 err = ubifs_leb_write(c, c->lhead_lnum, cs, 0, len);
427 *ltail_lnum = c->lhead_lnum;
429 c->lhead_offs += len;
430 if (c->lhead_offs == c->leb_size) {
431 c->lhead_lnum = ubifs_next_log_lnum(c, c->lhead_lnum);
438 * We have started the commit and now users may use the rest of the log
441 c->min_log_bytes = 0;
449 * ubifs_log_end_commit - end commit.
450 * @c: UBIFS file-system description object
451 * @ltail_lnum: new log tail LEB number
453 * This function is called on when the commit operation was finished. It
454 * moves log tail to new position and updates the master node so that it stores
455 * the new log tail LEB number. Returns zero in case of success and a negative
456 * error code in case of failure.
458 int ubifs_log_end_commit(struct ubifs_info *c, int ltail_lnum)
463 * At this phase we have to lock 'c->log_mutex' because UBIFS allows FS
464 * writes during commit. Its only short "commit" start phase when
465 * writers are blocked.
467 mutex_lock(&c->log_mutex);
469 dbg_log("old tail was LEB %d:0, new tail is LEB %d:0",
470 c->ltail_lnum, ltail_lnum);
472 c->ltail_lnum = ltail_lnum;
474 * The commit is finished and from now on it must be guaranteed that
475 * there is always enough space for the next commit.
477 c->min_log_bytes = c->leb_size;
479 spin_lock(&c->buds_lock);
480 c->bud_bytes -= c->cmt_bud_bytes;
481 spin_unlock(&c->buds_lock);
483 err = dbg_check_bud_bytes(c);
487 err = ubifs_write_master(c);
490 mutex_unlock(&c->log_mutex);
495 * ubifs_log_post_commit - things to do after commit is completed.
496 * @c: UBIFS file-system description object
497 * @old_ltail_lnum: old log tail LEB number
499 * Release buds only after commit is completed, because they must be unchanged
500 * if recovery is needed.
502 * Unmap log LEBs only after commit is completed, because they may be needed for
505 * This function returns %0 on success and a negative error code on failure.
507 int ubifs_log_post_commit(struct ubifs_info *c, int old_ltail_lnum)
511 while (!list_empty(&c->old_buds)) {
512 struct ubifs_bud *bud;
514 bud = list_entry(c->old_buds.next, struct ubifs_bud, list);
515 err = ubifs_return_leb(c, bud->lnum);
518 list_del(&bud->list);
521 mutex_lock(&c->log_mutex);
522 for (lnum = old_ltail_lnum; lnum != c->ltail_lnum;
523 lnum = ubifs_next_log_lnum(c, lnum)) {
524 dbg_log("unmap log LEB %d", lnum);
525 err = ubifs_leb_unmap(c, lnum);
530 mutex_unlock(&c->log_mutex);
535 * struct done_ref - references that have been done.
545 * done_already - determine if a reference has been done already.
546 * @done_tree: rb-tree to store references that have been done
547 * @lnum: LEB number of reference
549 * This function returns %1 if the reference has been done, %0 if not, otherwise
550 * a negative error code is returned.
552 static int done_already(struct rb_root *done_tree, int lnum)
554 struct rb_node **p = &done_tree->rb_node, *parent = NULL;
559 dr = rb_entry(parent, struct done_ref, rb);
562 else if (lnum > dr->lnum)
568 dr = kzalloc(sizeof(struct done_ref), GFP_NOFS);
574 rb_link_node(&dr->rb, parent, p);
575 rb_insert_color(&dr->rb, done_tree);
581 * destroy_done_tree - destroy the done tree.
582 * @done_tree: done tree to destroy
584 static void destroy_done_tree(struct rb_root *done_tree)
586 struct done_ref *dr, *n;
588 rbtree_postorder_for_each_entry_safe(dr, n, done_tree, rb)
593 * add_node - add a node to the consolidated log.
594 * @c: UBIFS file-system description object
595 * @buf: buffer to which to add
596 * @lnum: LEB number to which to write is passed and returned here
597 * @offs: offset to where to write is passed and returned here
600 * This function returns %0 on success and a negative error code on failure.
602 static int add_node(struct ubifs_info *c, void *buf, int *lnum, int *offs,
605 struct ubifs_ch *ch = node;
606 int len = le32_to_cpu(ch->len), remains = c->leb_size - *offs;
609 int sz = ALIGN(*offs, c->min_io_size), err;
611 ubifs_pad(c, buf + *offs, sz - *offs);
612 err = ubifs_leb_change(c, *lnum, buf, sz);
615 *lnum = ubifs_next_log_lnum(c, *lnum);
618 memcpy(buf + *offs, node, len);
619 *offs += ALIGN(len, 8);
624 * ubifs_consolidate_log - consolidate the log.
625 * @c: UBIFS file-system description object
627 * Repeated failed commits could cause the log to be full, but at least 1 LEB is
628 * needed for commit. This function rewrites the reference nodes in the log
629 * omitting duplicates, and failed CS nodes, and leaving no gaps.
631 * This function returns %0 on success and a negative error code on failure.
633 int ubifs_consolidate_log(struct ubifs_info *c)
635 struct ubifs_scan_leb *sleb;
636 struct ubifs_scan_node *snod;
637 struct rb_root done_tree = RB_ROOT;
638 int lnum, err, first = 1, write_lnum, offs = 0;
641 dbg_rcvry("log tail LEB %d, log head LEB %d", c->ltail_lnum,
643 buf = vmalloc(c->leb_size);
646 lnum = c->ltail_lnum;
649 sleb = ubifs_scan(c, lnum, 0, c->sbuf, 0);
654 list_for_each_entry(snod, &sleb->nodes, list) {
655 switch (snod->type) {
656 case UBIFS_REF_NODE: {
657 struct ubifs_ref_node *ref = snod->node;
658 int ref_lnum = le32_to_cpu(ref->lnum);
660 err = done_already(&done_tree, ref_lnum);
664 err = add_node(c, buf, &write_lnum,
674 err = add_node(c, buf, &write_lnum, &offs,
682 ubifs_scan_destroy(sleb);
683 if (lnum == c->lhead_lnum)
685 lnum = ubifs_next_log_lnum(c, lnum);
688 int sz = ALIGN(offs, c->min_io_size);
690 ubifs_pad(c, buf + offs, sz - offs);
691 err = ubifs_leb_change(c, write_lnum, buf, sz);
694 offs = ALIGN(offs, c->min_io_size);
696 destroy_done_tree(&done_tree);
698 if (write_lnum == c->lhead_lnum) {
699 ubifs_err(c, "log is too full");
702 /* Unmap remaining LEBs */
705 lnum = ubifs_next_log_lnum(c, lnum);
706 err = ubifs_leb_unmap(c, lnum);
709 } while (lnum != c->lhead_lnum);
710 c->lhead_lnum = write_lnum;
711 c->lhead_offs = offs;
712 dbg_rcvry("new log head at %d:%d", c->lhead_lnum, c->lhead_offs);
716 ubifs_scan_destroy(sleb);
718 destroy_done_tree(&done_tree);
724 * dbg_check_bud_bytes - make sure bud bytes calculation are all right.
725 * @c: UBIFS file-system description object
727 * This function makes sure the amount of flash space used by closed buds
728 * ('c->bud_bytes' is correct). Returns zero in case of success and %-EINVAL in
731 static int dbg_check_bud_bytes(struct ubifs_info *c)
734 struct ubifs_bud *bud;
735 long long bud_bytes = 0;
737 if (!dbg_is_chk_gen(c))
740 spin_lock(&c->buds_lock);
741 for (i = 0; i < c->jhead_cnt; i++)
742 list_for_each_entry(bud, &c->jheads[i].buds_list, list)
743 bud_bytes += c->leb_size - bud->start;
745 if (c->bud_bytes != bud_bytes) {
746 ubifs_err(c, "bad bud_bytes %lld, calculated %lld",
747 c->bud_bytes, bud_bytes);
750 spin_unlock(&c->buds_lock);