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: Adrian Hunter
20 * Artem Bityutskiy (Битюцкий Артём)
24 * This file implements functions needed to recover from unclean un-mounts.
25 * When UBIFS is mounted, it checks a flag on the master node to determine if
26 * an un-mount was completed successfully. If not, the process of mounting
27 * incorporates additional checking and fixing of on-flash data structures.
28 * UBIFS always cleans away all remnants of an unclean un-mount, so that
29 * errors do not accumulate. However UBIFS defers recovery if it is mounted
30 * read-only, and the flash is not modified in that case.
32 * The general UBIFS approach to the recovery is that it recovers from
33 * corruptions which could be caused by power cuts, but it refuses to recover
34 * from corruption caused by other reasons. And UBIFS tries to distinguish
35 * between these 2 reasons of corruptions and silently recover in the former
36 * case and loudly complain in the latter case.
38 * UBIFS writes only to erased LEBs, so it writes only to the flash space
39 * containing only 0xFFs. UBIFS also always writes strictly from the beginning
40 * of the LEB to the end. And UBIFS assumes that the underlying flash media
41 * writes in @c->max_write_size bytes at a time.
43 * Hence, if UBIFS finds a corrupted node at offset X, it expects only the min.
44 * I/O unit corresponding to offset X to contain corrupted data, all the
45 * following min. I/O units have to contain empty space (all 0xFFs). If this is
46 * not true, the corruption cannot be the result of a power cut, and UBIFS
50 #include <linux/crc32.h>
51 #include <linux/slab.h>
55 * is_empty - determine whether a buffer is empty (contains all 0xff).
56 * @buf: buffer to clean
57 * @len: length of buffer
59 * This function returns %1 if the buffer is empty (contains all 0xff) otherwise
62 static int is_empty(void *buf, int len)
67 for (i = 0; i < len; i++)
74 * first_non_ff - find offset of the first non-0xff byte.
75 * @buf: buffer to search in
76 * @len: length of buffer
78 * This function returns offset of the first non-0xff byte in @buf or %-1 if
79 * the buffer contains only 0xff bytes.
81 static int first_non_ff(void *buf, int len)
86 for (i = 0; i < len; i++)
93 * get_master_node - get the last valid master node allowing for corruption.
94 * @c: UBIFS file-system description object
96 * @pbuf: buffer containing the LEB read, is returned here
97 * @mst: master node, if found, is returned here
98 * @cor: corruption, if found, is returned here
100 * This function allocates a buffer, reads the LEB into it, and finds and
101 * returns the last valid master node allowing for one area of corruption.
102 * The corrupt area, if there is one, must be consistent with the assumption
103 * that it is the result of an unclean unmount while the master node was being
104 * written. Under those circumstances, it is valid to use the previously written
107 * This function returns %0 on success and a negative error code on failure.
109 static int get_master_node(const struct ubifs_info *c, int lnum, void **pbuf,
110 struct ubifs_mst_node **mst, void **cor)
112 const int sz = c->mst_node_alsz;
116 sbuf = vmalloc(c->leb_size);
120 err = ubi_read(c->ubi, lnum, sbuf, 0, c->leb_size);
121 if (err && err != -EBADMSG)
124 /* Find the first position that is definitely not a node */
128 while (offs + UBIFS_MST_NODE_SZ <= c->leb_size) {
129 struct ubifs_ch *ch = buf;
131 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
137 /* See if there was a valid master node before that */
144 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
145 if (ret != SCANNED_A_NODE && offs) {
146 /* Could have been corruption so check one place back */
150 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
151 if (ret != SCANNED_A_NODE)
153 * We accept only one area of corruption because
154 * we are assuming that it was caused while
155 * trying to write a master node.
159 if (ret == SCANNED_A_NODE) {
160 struct ubifs_ch *ch = buf;
162 if (ch->node_type != UBIFS_MST_NODE)
164 dbg_rcvry("found a master node at %d:%d", lnum, offs);
171 /* Check for corruption */
172 if (offs < c->leb_size) {
173 if (!is_empty(buf, min_t(int, len, sz))) {
175 dbg_rcvry("found corruption at %d:%d", lnum, offs);
181 /* Check remaining empty space */
182 if (offs < c->leb_size)
183 if (!is_empty(buf, len))
198 * write_rcvrd_mst_node - write recovered master node.
199 * @c: UBIFS file-system description object
202 * This function returns %0 on success and a negative error code on failure.
204 static int write_rcvrd_mst_node(struct ubifs_info *c,
205 struct ubifs_mst_node *mst)
207 int err = 0, lnum = UBIFS_MST_LNUM, sz = c->mst_node_alsz;
210 dbg_rcvry("recovery");
212 save_flags = mst->flags;
213 mst->flags |= cpu_to_le32(UBIFS_MST_RCVRY);
215 ubifs_prepare_node(c, mst, UBIFS_MST_NODE_SZ, 1);
216 err = ubi_leb_change(c->ubi, lnum, mst, sz, UBI_SHORTTERM);
219 err = ubi_leb_change(c->ubi, lnum + 1, mst, sz, UBI_SHORTTERM);
223 mst->flags = save_flags;
228 * ubifs_recover_master_node - recover the master node.
229 * @c: UBIFS file-system description object
231 * This function recovers the master node from corruption that may occur due to
232 * an unclean unmount.
234 * This function returns %0 on success and a negative error code on failure.
236 int ubifs_recover_master_node(struct ubifs_info *c)
238 void *buf1 = NULL, *buf2 = NULL, *cor1 = NULL, *cor2 = NULL;
239 struct ubifs_mst_node *mst1 = NULL, *mst2 = NULL, *mst;
240 const int sz = c->mst_node_alsz;
241 int err, offs1, offs2;
243 dbg_rcvry("recovery");
245 err = get_master_node(c, UBIFS_MST_LNUM, &buf1, &mst1, &cor1);
249 err = get_master_node(c, UBIFS_MST_LNUM + 1, &buf2, &mst2, &cor2);
254 offs1 = (void *)mst1 - buf1;
255 if ((le32_to_cpu(mst1->flags) & UBIFS_MST_RCVRY) &&
256 (offs1 == 0 && !cor1)) {
258 * mst1 was written by recovery at offset 0 with no
261 dbg_rcvry("recovery recovery");
264 offs2 = (void *)mst2 - buf2;
265 if (offs1 == offs2) {
266 /* Same offset, so must be the same */
267 if (memcmp((void *)mst1 + UBIFS_CH_SZ,
268 (void *)mst2 + UBIFS_CH_SZ,
269 UBIFS_MST_NODE_SZ - UBIFS_CH_SZ))
272 } else if (offs2 + sz == offs1) {
273 /* 1st LEB was written, 2nd was not */
277 } else if (offs1 == 0 && offs2 + sz >= c->leb_size) {
278 /* 1st LEB was unmapped and written, 2nd not */
286 * 2nd LEB was unmapped and about to be written, so
287 * there must be only one master node in the first LEB
290 if (offs1 != 0 || cor1)
298 * 1st LEB was unmapped and about to be written, so there must
299 * be no room left in 2nd LEB.
301 offs2 = (void *)mst2 - buf2;
302 if (offs2 + sz + sz <= c->leb_size)
307 ubifs_msg("recovered master node from LEB %d",
308 (mst == mst1 ? UBIFS_MST_LNUM : UBIFS_MST_LNUM + 1));
310 memcpy(c->mst_node, mst, UBIFS_MST_NODE_SZ);
313 /* Read-only mode. Keep a copy for switching to rw mode */
314 c->rcvrd_mst_node = kmalloc(sz, GFP_KERNEL);
315 if (!c->rcvrd_mst_node) {
319 memcpy(c->rcvrd_mst_node, c->mst_node, UBIFS_MST_NODE_SZ);
322 * We had to recover the master node, which means there was an
323 * unclean reboot. However, it is possible that the master node
324 * is clean at this point, i.e., %UBIFS_MST_DIRTY is not set.
325 * E.g., consider the following chain of events:
327 * 1. UBIFS was cleanly unmounted, so the master node is clean
328 * 2. UBIFS is being mounted R/W and starts changing the master
329 * node in the first (%UBIFS_MST_LNUM). A power cut happens,
330 * so this LEB ends up with some amount of garbage at the
332 * 3. UBIFS is being mounted R/O. We reach this place and
333 * recover the master node from the second LEB
334 * (%UBIFS_MST_LNUM + 1). But we cannot update the media
335 * because we are being mounted R/O. We have to defer the
337 * 4. However, this master node (@c->mst_node) is marked as
338 * clean (since the step 1). And if we just return, the
339 * mount code will be confused and won't recover the master
340 * node when it is re-mounter R/W later.
342 * Thus, to force the recovery by marking the master node as
345 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
347 /* Write the recovered master node */
348 c->max_sqnum = le64_to_cpu(mst->ch.sqnum) - 1;
349 err = write_rcvrd_mst_node(c, c->mst_node);
362 ubifs_err("failed to recover master node");
364 dbg_err("dumping first master node");
365 dbg_dump_node(c, mst1);
368 dbg_err("dumping second master node");
369 dbg_dump_node(c, mst2);
377 * ubifs_write_rcvrd_mst_node - write the recovered master node.
378 * @c: UBIFS file-system description object
380 * This function writes the master node that was recovered during mounting in
381 * read-only mode and must now be written because we are remounting rw.
383 * This function returns %0 on success and a negative error code on failure.
385 int ubifs_write_rcvrd_mst_node(struct ubifs_info *c)
389 if (!c->rcvrd_mst_node)
391 c->rcvrd_mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
392 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
393 err = write_rcvrd_mst_node(c, c->rcvrd_mst_node);
396 kfree(c->rcvrd_mst_node);
397 c->rcvrd_mst_node = NULL;
402 * is_last_write - determine if an offset was in the last write to a LEB.
403 * @c: UBIFS file-system description object
404 * @buf: buffer to check
405 * @offs: offset to check
407 * This function returns %1 if @offs was in the last write to the LEB whose data
408 * is in @buf, otherwise %0 is returned. The determination is made by checking
409 * for subsequent empty space starting from the next @c->max_write_size
412 static int is_last_write(const struct ubifs_info *c, void *buf, int offs)
414 int empty_offs, check_len;
418 * Round up to the next @c->max_write_size boundary i.e. @offs is in
419 * the last wbuf written. After that should be empty space.
421 empty_offs = ALIGN(offs + 1, c->max_write_size);
422 check_len = c->leb_size - empty_offs;
423 p = buf + empty_offs - offs;
424 return is_empty(p, check_len);
428 * clean_buf - clean the data from an LEB sitting in a buffer.
429 * @c: UBIFS file-system description object
430 * @buf: buffer to clean
431 * @lnum: LEB number to clean
432 * @offs: offset from which to clean
433 * @len: length of buffer
435 * This function pads up to the next min_io_size boundary (if there is one) and
436 * sets empty space to all 0xff. @buf, @offs and @len are updated to the next
437 * @c->min_io_size boundary.
439 static void clean_buf(const struct ubifs_info *c, void **buf, int lnum,
442 int empty_offs, pad_len;
445 dbg_rcvry("cleaning corruption at %d:%d", lnum, *offs);
447 ubifs_assert(!(*offs & 7));
448 empty_offs = ALIGN(*offs, c->min_io_size);
449 pad_len = empty_offs - *offs;
450 ubifs_pad(c, *buf, pad_len);
454 memset(*buf, 0xff, c->leb_size - empty_offs);
458 * no_more_nodes - determine if there are no more nodes in a buffer.
459 * @c: UBIFS file-system description object
460 * @buf: buffer to check
461 * @len: length of buffer
462 * @lnum: LEB number of the LEB from which @buf was read
463 * @offs: offset from which @buf was read
465 * This function ensures that the corrupted node at @offs is the last thing
466 * written to a LEB. This function returns %1 if more data is not found and
467 * %0 if more data is found.
469 static int no_more_nodes(const struct ubifs_info *c, void *buf, int len,
472 struct ubifs_ch *ch = buf;
473 int skip, dlen = le32_to_cpu(ch->len);
475 /* Check for empty space after the corrupt node's common header */
476 skip = ALIGN(offs + UBIFS_CH_SZ, c->max_write_size) - offs;
477 if (is_empty(buf + skip, len - skip))
480 * The area after the common header size is not empty, so the common
481 * header must be intact. Check it.
483 if (ubifs_check_node(c, buf, lnum, offs, 1, 0) != -EUCLEAN) {
484 dbg_rcvry("unexpected bad common header at %d:%d", lnum, offs);
487 /* Now we know the corrupt node's length we can skip over it */
488 skip = ALIGN(offs + dlen, c->max_write_size) - offs;
489 /* After which there should be empty space */
490 if (is_empty(buf + skip, len - skip))
492 dbg_rcvry("unexpected data at %d:%d", lnum, offs + skip);
497 * fix_unclean_leb - fix an unclean LEB.
498 * @c: UBIFS file-system description object
499 * @sleb: scanned LEB information
500 * @start: offset where scan started
502 static int fix_unclean_leb(struct ubifs_info *c, struct ubifs_scan_leb *sleb,
505 int lnum = sleb->lnum, endpt = start;
507 /* Get the end offset of the last node we are keeping */
508 if (!list_empty(&sleb->nodes)) {
509 struct ubifs_scan_node *snod;
511 snod = list_entry(sleb->nodes.prev,
512 struct ubifs_scan_node, list);
513 endpt = snod->offs + snod->len;
516 if (c->ro_mount && !c->remounting_rw) {
517 /* Add to recovery list */
518 struct ubifs_unclean_leb *ucleb;
520 dbg_rcvry("need to fix LEB %d start %d endpt %d",
521 lnum, start, sleb->endpt);
522 ucleb = kzalloc(sizeof(struct ubifs_unclean_leb), GFP_NOFS);
526 ucleb->endpt = endpt;
527 list_add_tail(&ucleb->list, &c->unclean_leb_list);
529 /* Write the fixed LEB back to flash */
532 dbg_rcvry("fixing LEB %d start %d endpt %d",
533 lnum, start, sleb->endpt);
535 err = ubifs_leb_unmap(c, lnum);
539 int len = ALIGN(endpt, c->min_io_size);
542 err = ubi_read(c->ubi, lnum, sleb->buf, 0,
547 /* Pad to min_io_size */
549 int pad_len = len - ALIGN(endpt, 8);
552 void *buf = sleb->buf + len - pad_len;
554 ubifs_pad(c, buf, pad_len);
557 err = ubi_leb_change(c->ubi, lnum, sleb->buf, len,
567 * drop_last_node - drop the last node or group of nodes.
568 * @sleb: scanned LEB information
569 * @offs: offset of dropped nodes is returned here
570 * @grouped: non-zero if whole group of nodes have to be dropped
572 * This is a helper function for 'ubifs_recover_leb()' which drops the last
573 * node of the scanned LEB or the last group of nodes if @grouped is not zero.
574 * This function returns %1 if a node was dropped and %0 otherwise.
576 static int drop_last_node(struct ubifs_scan_leb *sleb, int *offs, int grouped)
580 while (!list_empty(&sleb->nodes)) {
581 struct ubifs_scan_node *snod;
584 snod = list_entry(sleb->nodes.prev, struct ubifs_scan_node,
587 if (ch->group_type != UBIFS_IN_NODE_GROUP)
589 dbg_rcvry("dropping node at %d:%d", sleb->lnum, snod->offs);
591 list_del(&snod->list);
593 sleb->nodes_cnt -= 1;
602 * ubifs_recover_leb - scan and recover a LEB.
603 * @c: UBIFS file-system description object
606 * @sbuf: LEB-sized buffer to use
607 * @grouped: nodes may be grouped for recovery
609 * This function does a scan of a LEB, but caters for errors that might have
610 * been caused by the unclean unmount from which we are attempting to recover.
611 * Returns %0 in case of success, %-EUCLEAN if an unrecoverable corruption is
612 * found, and a negative error code in case of failure.
614 struct ubifs_scan_leb *ubifs_recover_leb(struct ubifs_info *c, int lnum,
615 int offs, void *sbuf, int grouped)
617 int ret = 0, err, len = c->leb_size - offs, start = offs, min_io_unit;
618 struct ubifs_scan_leb *sleb;
619 void *buf = sbuf + offs;
621 dbg_rcvry("%d:%d", lnum, offs);
623 sleb = ubifs_start_scan(c, lnum, offs, sbuf);
627 ubifs_assert(len >= 8);
629 dbg_scan("look at LEB %d:%d (%d bytes left)",
635 * Scan quietly until there is an error from which we cannot
638 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 0);
639 if (ret == SCANNED_A_NODE) {
640 /* A valid node, and not a padding node */
641 struct ubifs_ch *ch = buf;
644 err = ubifs_add_snod(c, sleb, buf, offs);
647 node_len = ALIGN(le32_to_cpu(ch->len), 8);
651 } else if (ret > 0) {
652 /* Padding bytes or a valid padding node */
656 } else if (ret == SCANNED_EMPTY_SPACE ||
657 ret == SCANNED_GARBAGE ||
658 ret == SCANNED_A_BAD_PAD_NODE ||
659 ret == SCANNED_A_CORRUPT_NODE) {
660 dbg_rcvry("found corruption - %d", ret);
663 dbg_err("unexpected return value %d", ret);
669 if (ret == SCANNED_GARBAGE || ret == SCANNED_A_BAD_PAD_NODE) {
670 if (!is_last_write(c, buf, offs))
671 goto corrupted_rescan;
672 } else if (ret == SCANNED_A_CORRUPT_NODE) {
673 if (!no_more_nodes(c, buf, len, lnum, offs))
674 goto corrupted_rescan;
675 } else if (!is_empty(buf, len)) {
676 if (!is_last_write(c, buf, offs)) {
677 int corruption = first_non_ff(buf, len);
680 * See header comment for this file for more
681 * explanations about the reasons we have this check.
683 ubifs_err("corrupt empty space LEB %d:%d, corruption "
684 "starts at %d", lnum, offs, corruption);
685 /* Make sure we dump interesting non-0xFF data */
692 min_io_unit = round_down(offs, c->min_io_size);
695 * If nodes are grouped, always drop the incomplete group at
698 drop_last_node(sleb, &offs, 1);
701 * While we are in the middle of the same min. I/O unit keep dropping
702 * nodes. So basically, what we want is to make sure that the last min.
703 * I/O unit where we saw the corruption is dropped completely with all
704 * the uncorrupted node which may possibly sit there.
706 * In other words, let's name the min. I/O unit where the corruption
707 * starts B, and the previous min. I/O unit A. The below code tries to
708 * deal with a situation when half of B contains valid nodes or the end
709 * of a valid node, and the second half of B contains corrupted data or
710 * garbage. This means that UBIFS had been writing to B just before the
711 * power cut happened. I do not know how realistic is this scenario
712 * that half of the min. I/O unit had been written successfully and the
713 * other half not, but this is possible in our 'failure mode emulation'
714 * infrastructure at least.
716 * So what is the problem, why we need to drop those nodes? Whey can't
717 * we just clean-up the second half of B by putting a padding node
718 * there? We can, and this works fine with one exception which was
719 * reproduced with power cut emulation testing and happens extremely
720 * rarely. The description follows, but it is worth noting that that is
721 * only about the GC head, so we could do this trick only if the bud
722 * belongs to the GC head, but it does not seem to be worth an
723 * additional "if" statement.
725 * So, imagine the file-system is full, we run GC which is moving valid
726 * nodes from LEB X to LEB Y (obviously, LEB Y is the current GC head
727 * LEB). The @c->gc_lnum is -1, which means that GC will retain LEB X
728 * and will try to continue. Imagine that LEB X is currently the
729 * dirtiest LEB, and the amount of used space in LEB Y is exactly the
730 * same as amount of free space in LEB X.
732 * And a power cut happens when nodes are moved from LEB X to LEB Y. We
733 * are here trying to recover LEB Y which is the GC head LEB. We find
734 * the min. I/O unit B as described above. Then we clean-up LEB Y by
735 * padding min. I/O unit. And later 'ubifs_rcvry_gc_commit()' function
736 * fails, because it cannot find a dirty LEB which could be GC'd into
737 * LEB Y! Even LEB X does not match because the amount of valid nodes
738 * there does not fit the free space in LEB Y any more! And this is
739 * because of the padding node which we added to LEB Y. The
740 * user-visible effect of this which I once observed and analysed is
741 * that we cannot mount the file-system with -ENOSPC error.
743 * So obviously, to make sure that situation does not happen we should
744 * free min. I/O unit B in LEB Y completely and the last used min. I/O
745 * unit in LEB Y should be A. This is basically what the below code
748 while (min_io_unit == round_down(offs, c->min_io_size) &&
749 min_io_unit != offs &&
750 drop_last_node(sleb, &offs, grouped));
753 len = c->leb_size - offs;
755 clean_buf(c, &buf, lnum, &offs, &len);
756 ubifs_end_scan(c, sleb, lnum, offs);
758 err = fix_unclean_leb(c, sleb, start);
765 /* Re-scan the corrupted data with verbose messages */
766 dbg_err("corruptio %d", ret);
767 ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
769 ubifs_scanned_corruption(c, lnum, offs, buf);
772 ubifs_err("LEB %d scanning failed", lnum);
773 ubifs_scan_destroy(sleb);
778 * get_cs_sqnum - get commit start sequence number.
779 * @c: UBIFS file-system description object
780 * @lnum: LEB number of commit start node
781 * @offs: offset of commit start node
782 * @cs_sqnum: commit start sequence number is returned here
784 * This function returns %0 on success and a negative error code on failure.
786 static int get_cs_sqnum(struct ubifs_info *c, int lnum, int offs,
787 unsigned long long *cs_sqnum)
789 struct ubifs_cs_node *cs_node = NULL;
792 dbg_rcvry("at %d:%d", lnum, offs);
793 cs_node = kmalloc(UBIFS_CS_NODE_SZ, GFP_KERNEL);
796 if (c->leb_size - offs < UBIFS_CS_NODE_SZ)
798 err = ubi_read(c->ubi, lnum, (void *)cs_node, offs, UBIFS_CS_NODE_SZ);
799 if (err && err != -EBADMSG)
801 ret = ubifs_scan_a_node(c, cs_node, UBIFS_CS_NODE_SZ, lnum, offs, 0);
802 if (ret != SCANNED_A_NODE) {
803 dbg_err("Not a valid node");
806 if (cs_node->ch.node_type != UBIFS_CS_NODE) {
807 dbg_err("Node a CS node, type is %d", cs_node->ch.node_type);
810 if (le64_to_cpu(cs_node->cmt_no) != c->cmt_no) {
811 dbg_err("CS node cmt_no %llu != current cmt_no %llu",
812 (unsigned long long)le64_to_cpu(cs_node->cmt_no),
816 *cs_sqnum = le64_to_cpu(cs_node->ch.sqnum);
817 dbg_rcvry("commit start sqnum %llu", *cs_sqnum);
824 ubifs_err("failed to get CS sqnum");
830 * ubifs_recover_log_leb - scan and recover a log LEB.
831 * @c: UBIFS file-system description object
834 * @sbuf: LEB-sized buffer to use
836 * This function does a scan of a LEB, but caters for errors that might have
837 * been caused by unclean reboots from which we are attempting to recover
838 * (assume that only the last log LEB can be corrupted by an unclean reboot).
840 * This function returns %0 on success and a negative error code on failure.
842 struct ubifs_scan_leb *ubifs_recover_log_leb(struct ubifs_info *c, int lnum,
843 int offs, void *sbuf)
845 struct ubifs_scan_leb *sleb;
848 dbg_rcvry("LEB %d", lnum);
849 next_lnum = lnum + 1;
850 if (next_lnum >= UBIFS_LOG_LNUM + c->log_lebs)
851 next_lnum = UBIFS_LOG_LNUM;
852 if (next_lnum != c->ltail_lnum) {
854 * We can only recover at the end of the log, so check that the
855 * next log LEB is empty or out of date.
857 sleb = ubifs_scan(c, next_lnum, 0, sbuf, 0);
860 if (sleb->nodes_cnt) {
861 struct ubifs_scan_node *snod;
862 unsigned long long cs_sqnum = c->cs_sqnum;
864 snod = list_entry(sleb->nodes.next,
865 struct ubifs_scan_node, list);
869 err = get_cs_sqnum(c, lnum, offs, &cs_sqnum);
871 ubifs_scan_destroy(sleb);
875 if (snod->sqnum > cs_sqnum) {
876 ubifs_err("unrecoverable log corruption "
878 ubifs_scan_destroy(sleb);
879 return ERR_PTR(-EUCLEAN);
882 ubifs_scan_destroy(sleb);
884 return ubifs_recover_leb(c, lnum, offs, sbuf, 0);
888 * recover_head - recover a head.
889 * @c: UBIFS file-system description object
890 * @lnum: LEB number of head to recover
891 * @offs: offset of head to recover
892 * @sbuf: LEB-sized buffer to use
894 * This function ensures that there is no data on the flash at a head location.
896 * This function returns %0 on success and a negative error code on failure.
898 static int recover_head(const struct ubifs_info *c, int lnum, int offs,
901 int len = c->max_write_size, err;
903 if (offs + len > c->leb_size)
904 len = c->leb_size - offs;
909 /* Read at the head location and check it is empty flash */
910 err = ubi_read(c->ubi, lnum, sbuf, offs, len);
911 if (err || !is_empty(sbuf, len)) {
912 dbg_rcvry("cleaning head at %d:%d", lnum, offs);
914 return ubifs_leb_unmap(c, lnum);
915 err = ubi_read(c->ubi, lnum, sbuf, 0, offs);
918 return ubi_leb_change(c->ubi, lnum, sbuf, offs, UBI_UNKNOWN);
925 * ubifs_recover_inl_heads - recover index and LPT heads.
926 * @c: UBIFS file-system description object
927 * @sbuf: LEB-sized buffer to use
929 * This function ensures that there is no data on the flash at the index and
930 * LPT head locations.
932 * This deals with the recovery of a half-completed journal commit. UBIFS is
933 * careful never to overwrite the last version of the index or the LPT. Because
934 * the index and LPT are wandering trees, data from a half-completed commit will
935 * not be referenced anywhere in UBIFS. The data will be either in LEBs that are
936 * assumed to be empty and will be unmapped anyway before use, or in the index
939 * This function returns %0 on success and a negative error code on failure.
941 int ubifs_recover_inl_heads(const struct ubifs_info *c, void *sbuf)
945 ubifs_assert(!c->ro_mount || c->remounting_rw);
947 dbg_rcvry("checking index head at %d:%d", c->ihead_lnum, c->ihead_offs);
948 err = recover_head(c, c->ihead_lnum, c->ihead_offs, sbuf);
952 dbg_rcvry("checking LPT head at %d:%d", c->nhead_lnum, c->nhead_offs);
953 err = recover_head(c, c->nhead_lnum, c->nhead_offs, sbuf);
961 * clean_an_unclean_leb - read and write a LEB to remove corruption.
962 * @c: UBIFS file-system description object
963 * @ucleb: unclean LEB information
964 * @sbuf: LEB-sized buffer to use
966 * This function reads a LEB up to a point pre-determined by the mount recovery,
967 * checks the nodes, and writes the result back to the flash, thereby cleaning
968 * off any following corruption, or non-fatal ECC errors.
970 * This function returns %0 on success and a negative error code on failure.
972 static int clean_an_unclean_leb(const struct ubifs_info *c,
973 struct ubifs_unclean_leb *ucleb, void *sbuf)
975 int err, lnum = ucleb->lnum, offs = 0, len = ucleb->endpt, quiet = 1;
978 dbg_rcvry("LEB %d len %d", lnum, len);
981 /* Nothing to read, just unmap it */
982 err = ubifs_leb_unmap(c, lnum);
988 err = ubi_read(c->ubi, lnum, buf, offs, len);
989 if (err && err != -EBADMSG)
997 /* Scan quietly until there is an error */
998 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, quiet);
1000 if (ret == SCANNED_A_NODE) {
1001 /* A valid node, and not a padding node */
1002 struct ubifs_ch *ch = buf;
1005 node_len = ALIGN(le32_to_cpu(ch->len), 8);
1013 /* Padding bytes or a valid padding node */
1020 if (ret == SCANNED_EMPTY_SPACE) {
1021 ubifs_err("unexpected empty space at %d:%d",
1027 /* Redo the last scan but noisily */
1032 ubifs_scanned_corruption(c, lnum, offs, buf);
1036 /* Pad to min_io_size */
1037 len = ALIGN(ucleb->endpt, c->min_io_size);
1038 if (len > ucleb->endpt) {
1039 int pad_len = len - ALIGN(ucleb->endpt, 8);
1042 buf = c->sbuf + len - pad_len;
1043 ubifs_pad(c, buf, pad_len);
1047 /* Write back the LEB atomically */
1048 err = ubi_leb_change(c->ubi, lnum, sbuf, len, UBI_UNKNOWN);
1052 dbg_rcvry("cleaned LEB %d", lnum);
1058 * ubifs_clean_lebs - clean LEBs recovered during read-only mount.
1059 * @c: UBIFS file-system description object
1060 * @sbuf: LEB-sized buffer to use
1062 * This function cleans a LEB identified during recovery that needs to be
1063 * written but was not because UBIFS was mounted read-only. This happens when
1064 * remounting to read-write mode.
1066 * This function returns %0 on success and a negative error code on failure.
1068 int ubifs_clean_lebs(const struct ubifs_info *c, void *sbuf)
1070 dbg_rcvry("recovery");
1071 while (!list_empty(&c->unclean_leb_list)) {
1072 struct ubifs_unclean_leb *ucleb;
1075 ucleb = list_entry(c->unclean_leb_list.next,
1076 struct ubifs_unclean_leb, list);
1077 err = clean_an_unclean_leb(c, ucleb, sbuf);
1080 list_del(&ucleb->list);
1087 * grab_empty_leb - grab an empty LEB to use as GC LEB and run commit.
1088 * @c: UBIFS file-system description object
1090 * This is a helper function for 'ubifs_rcvry_gc_commit()' which grabs an empty
1091 * LEB to be used as GC LEB (@c->gc_lnum), and then runs the commit. Returns
1092 * zero in case of success and a negative error code in case of failure.
1094 static int grab_empty_leb(struct ubifs_info *c)
1099 * Note, it is very important to first search for an empty LEB and then
1100 * run the commit, not vice-versa. The reason is that there might be
1101 * only one empty LEB at the moment, the one which has been the
1102 * @c->gc_lnum just before the power cut happened. During the regular
1103 * UBIFS operation (not now) @c->gc_lnum is marked as "taken", so no
1104 * one but GC can grab it. But at this moment this single empty LEB is
1105 * not marked as taken, so if we run commit - what happens? Right, the
1106 * commit will grab it and write the index there. Remember that the
1107 * index always expands as long as there is free space, and it only
1108 * starts consolidating when we run out of space.
1110 * IOW, if we run commit now, we might not be able to find a free LEB
1113 lnum = ubifs_find_free_leb_for_idx(c);
1115 dbg_err("could not find an empty LEB");
1117 dbg_dump_budg(c, &c->bi);
1121 /* Reset the index flag */
1122 err = ubifs_change_one_lp(c, lnum, LPROPS_NC, LPROPS_NC, 0,
1128 dbg_rcvry("found empty LEB %d, run commit", lnum);
1130 return ubifs_run_commit(c);
1134 * ubifs_rcvry_gc_commit - recover the GC LEB number and run the commit.
1135 * @c: UBIFS file-system description object
1137 * Out-of-place garbage collection requires always one empty LEB with which to
1138 * start garbage collection. The LEB number is recorded in c->gc_lnum and is
1139 * written to the master node on unmounting. In the case of an unclean unmount
1140 * the value of gc_lnum recorded in the master node is out of date and cannot
1141 * be used. Instead, recovery must allocate an empty LEB for this purpose.
1142 * However, there may not be enough empty space, in which case it must be
1143 * possible to GC the dirtiest LEB into the GC head LEB.
1145 * This function also runs the commit which causes the TNC updates from
1146 * size-recovery and orphans to be written to the flash. That is important to
1147 * ensure correct replay order for subsequent mounts.
1149 * This function returns %0 on success and a negative error code on failure.
1151 int ubifs_rcvry_gc_commit(struct ubifs_info *c)
1153 struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf;
1154 struct ubifs_lprops lp;
1157 dbg_rcvry("GC head LEB %d, offs %d", wbuf->lnum, wbuf->offs);
1160 if (wbuf->lnum == -1 || wbuf->offs == c->leb_size)
1161 return grab_empty_leb(c);
1163 err = ubifs_find_dirty_leb(c, &lp, wbuf->offs, 2);
1168 dbg_rcvry("could not find a dirty LEB");
1169 return grab_empty_leb(c);
1172 ubifs_assert(!(lp.flags & LPROPS_INDEX));
1173 ubifs_assert(lp.free + lp.dirty >= wbuf->offs);
1176 * We run the commit before garbage collection otherwise subsequent
1177 * mounts will see the GC and orphan deletion in a different order.
1179 dbg_rcvry("committing");
1180 err = ubifs_run_commit(c);
1184 dbg_rcvry("GC'ing LEB %d", lp.lnum);
1185 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1186 err = ubifs_garbage_collect_leb(c, &lp);
1188 int err2 = ubifs_wbuf_sync_nolock(wbuf);
1193 mutex_unlock(&wbuf->io_mutex);
1195 dbg_err("GC failed, error %d", err);
1201 ubifs_assert(err == LEB_RETAINED);
1202 if (err != LEB_RETAINED)
1205 err = ubifs_leb_unmap(c, c->gc_lnum);
1209 dbg_rcvry("allocated LEB %d for GC", lp.lnum);
1214 * struct size_entry - inode size information for recovery.
1215 * @rb: link in the RB-tree of sizes
1216 * @inum: inode number
1217 * @i_size: size on inode
1218 * @d_size: maximum size based on data nodes
1219 * @exists: indicates whether the inode exists
1220 * @inode: inode if pinned in memory awaiting rw mode to fix it
1228 struct inode *inode;
1232 * add_ino - add an entry to the size tree.
1233 * @c: UBIFS file-system description object
1234 * @inum: inode number
1235 * @i_size: size on inode
1236 * @d_size: maximum size based on data nodes
1237 * @exists: indicates whether the inode exists
1239 static int add_ino(struct ubifs_info *c, ino_t inum, loff_t i_size,
1240 loff_t d_size, int exists)
1242 struct rb_node **p = &c->size_tree.rb_node, *parent = NULL;
1243 struct size_entry *e;
1247 e = rb_entry(parent, struct size_entry, rb);
1251 p = &(*p)->rb_right;
1254 e = kzalloc(sizeof(struct size_entry), GFP_KERNEL);
1263 rb_link_node(&e->rb, parent, p);
1264 rb_insert_color(&e->rb, &c->size_tree);
1270 * find_ino - find an entry on the size tree.
1271 * @c: UBIFS file-system description object
1272 * @inum: inode number
1274 static struct size_entry *find_ino(struct ubifs_info *c, ino_t inum)
1276 struct rb_node *p = c->size_tree.rb_node;
1277 struct size_entry *e;
1280 e = rb_entry(p, struct size_entry, rb);
1283 else if (inum > e->inum)
1292 * remove_ino - remove an entry from the size tree.
1293 * @c: UBIFS file-system description object
1294 * @inum: inode number
1296 static void remove_ino(struct ubifs_info *c, ino_t inum)
1298 struct size_entry *e = find_ino(c, inum);
1302 rb_erase(&e->rb, &c->size_tree);
1307 * ubifs_destroy_size_tree - free resources related to the size tree.
1308 * @c: UBIFS file-system description object
1310 void ubifs_destroy_size_tree(struct ubifs_info *c)
1312 struct rb_node *this = c->size_tree.rb_node;
1313 struct size_entry *e;
1316 if (this->rb_left) {
1317 this = this->rb_left;
1319 } else if (this->rb_right) {
1320 this = this->rb_right;
1323 e = rb_entry(this, struct size_entry, rb);
1326 this = rb_parent(this);
1328 if (this->rb_left == &e->rb)
1329 this->rb_left = NULL;
1331 this->rb_right = NULL;
1335 c->size_tree = RB_ROOT;
1339 * ubifs_recover_size_accum - accumulate inode sizes for recovery.
1340 * @c: UBIFS file-system description object
1342 * @deletion: node is for a deletion
1343 * @new_size: inode size
1345 * This function has two purposes:
1346 * 1) to ensure there are no data nodes that fall outside the inode size
1347 * 2) to ensure there are no data nodes for inodes that do not exist
1348 * To accomplish those purposes, a rb-tree is constructed containing an entry
1349 * for each inode number in the journal that has not been deleted, and recording
1350 * the size from the inode node, the maximum size of any data node (also altered
1351 * by truncations) and a flag indicating a inode number for which no inode node
1352 * was present in the journal.
1354 * Note that there is still the possibility that there are data nodes that have
1355 * been committed that are beyond the inode size, however the only way to find
1356 * them would be to scan the entire index. Alternatively, some provision could
1357 * be made to record the size of inodes at the start of commit, which would seem
1358 * very cumbersome for a scenario that is quite unlikely and the only negative
1359 * consequence of which is wasted space.
1361 * This functions returns %0 on success and a negative error code on failure.
1363 int ubifs_recover_size_accum(struct ubifs_info *c, union ubifs_key *key,
1364 int deletion, loff_t new_size)
1366 ino_t inum = key_inum(c, key);
1367 struct size_entry *e;
1370 switch (key_type(c, key)) {
1373 remove_ino(c, inum);
1375 e = find_ino(c, inum);
1377 e->i_size = new_size;
1380 err = add_ino(c, inum, new_size, 0, 1);
1386 case UBIFS_DATA_KEY:
1387 e = find_ino(c, inum);
1389 if (new_size > e->d_size)
1390 e->d_size = new_size;
1392 err = add_ino(c, inum, 0, new_size, 0);
1397 case UBIFS_TRUN_KEY:
1398 e = find_ino(c, inum);
1400 e->d_size = new_size;
1407 * fix_size_in_place - fix inode size in place on flash.
1408 * @c: UBIFS file-system description object
1409 * @e: inode size information for recovery
1411 static int fix_size_in_place(struct ubifs_info *c, struct size_entry *e)
1413 struct ubifs_ino_node *ino = c->sbuf;
1415 union ubifs_key key;
1416 int err, lnum, offs, len;
1420 /* Locate the inode node LEB number and offset */
1421 ino_key_init(c, &key, e->inum);
1422 err = ubifs_tnc_locate(c, &key, ino, &lnum, &offs);
1426 * If the size recorded on the inode node is greater than the size that
1427 * was calculated from nodes in the journal then don't change the inode.
1429 i_size = le64_to_cpu(ino->size);
1430 if (i_size >= e->d_size)
1433 err = ubi_read(c->ubi, lnum, c->sbuf, 0, c->leb_size);
1436 /* Change the size field and recalculate the CRC */
1437 ino = c->sbuf + offs;
1438 ino->size = cpu_to_le64(e->d_size);
1439 len = le32_to_cpu(ino->ch.len);
1440 crc = crc32(UBIFS_CRC32_INIT, (void *)ino + 8, len - 8);
1441 ino->ch.crc = cpu_to_le32(crc);
1442 /* Work out where data in the LEB ends and free space begins */
1444 len = c->leb_size - 1;
1445 while (p[len] == 0xff)
1447 len = ALIGN(len + 1, c->min_io_size);
1448 /* Atomically write the fixed LEB back again */
1449 err = ubi_leb_change(c->ubi, lnum, c->sbuf, len, UBI_UNKNOWN);
1452 dbg_rcvry("inode %lu at %d:%d size %lld -> %lld",
1453 (unsigned long)e->inum, lnum, offs, i_size, e->d_size);
1457 ubifs_warn("inode %lu failed to fix size %lld -> %lld error %d",
1458 (unsigned long)e->inum, e->i_size, e->d_size, err);
1463 * ubifs_recover_size - recover inode size.
1464 * @c: UBIFS file-system description object
1466 * This function attempts to fix inode size discrepancies identified by the
1467 * 'ubifs_recover_size_accum()' function.
1469 * This functions returns %0 on success and a negative error code on failure.
1471 int ubifs_recover_size(struct ubifs_info *c)
1473 struct rb_node *this = rb_first(&c->size_tree);
1476 struct size_entry *e;
1479 e = rb_entry(this, struct size_entry, rb);
1481 union ubifs_key key;
1483 ino_key_init(c, &key, e->inum);
1484 err = ubifs_tnc_lookup(c, &key, c->sbuf);
1485 if (err && err != -ENOENT)
1487 if (err == -ENOENT) {
1488 /* Remove data nodes that have no inode */
1489 dbg_rcvry("removing ino %lu",
1490 (unsigned long)e->inum);
1491 err = ubifs_tnc_remove_ino(c, e->inum);
1495 struct ubifs_ino_node *ino = c->sbuf;
1498 e->i_size = le64_to_cpu(ino->size);
1502 if (e->exists && e->i_size < e->d_size) {
1504 /* Fix the inode size and pin it in memory */
1505 struct inode *inode;
1506 struct ubifs_inode *ui;
1508 ubifs_assert(!e->inode);
1510 inode = ubifs_iget(c->vfs_sb, e->inum);
1512 return PTR_ERR(inode);
1514 ui = ubifs_inode(inode);
1515 if (inode->i_size < e->d_size) {
1516 dbg_rcvry("ino %lu size %lld -> %lld",
1517 (unsigned long)e->inum,
1518 inode->i_size, e->d_size);
1519 inode->i_size = e->d_size;
1520 ui->ui_size = e->d_size;
1521 ui->synced_i_size = e->d_size;
1523 this = rb_next(this);
1528 /* Fix the size in place */
1529 err = fix_size_in_place(c, e);
1537 this = rb_next(this);
1538 rb_erase(&e->rb, &c->size_tree);