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 sucessfully. If not, the process of mounting
27 * incorparates 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.
33 #include <linux/crc32.h>
37 * is_empty - determine whether a buffer is empty (contains all 0xff).
38 * @buf: buffer to clean
39 * @len: length of buffer
41 * This function returns %1 if the buffer is empty (contains all 0xff) otherwise
44 static int is_empty(void *buf, int len)
49 for (i = 0; i < len; i++)
56 * first_non_ff - find offset of the first non-0xff byte.
57 * @buf: buffer to search in
58 * @len: length of buffer
60 * This function returns offset of the first non-0xff byte in @buf or %-1 if
61 * the buffer contains only 0xff bytes.
63 static int first_non_ff(void *buf, int len)
68 for (i = 0; i < len; i++)
75 * get_master_node - get the last valid master node allowing for corruption.
76 * @c: UBIFS file-system description object
78 * @pbuf: buffer containing the LEB read, is returned here
79 * @mst: master node, if found, is returned here
80 * @cor: corruption, if found, is returned here
82 * This function allocates a buffer, reads the LEB into it, and finds and
83 * returns the last valid master node allowing for one area of corruption.
84 * The corrupt area, if there is one, must be consistent with the assumption
85 * that it is the result of an unclean unmount while the master node was being
86 * written. Under those circumstances, it is valid to use the previously written
89 * This function returns %0 on success and a negative error code on failure.
91 static int get_master_node(const struct ubifs_info *c, int lnum, void **pbuf,
92 struct ubifs_mst_node **mst, void **cor)
94 const int sz = c->mst_node_alsz;
98 sbuf = vmalloc(c->leb_size);
102 err = ubi_read(c->ubi, lnum, sbuf, 0, c->leb_size);
103 if (err && err != -EBADMSG)
106 /* Find the first position that is definitely not a node */
110 while (offs + UBIFS_MST_NODE_SZ <= c->leb_size) {
111 struct ubifs_ch *ch = buf;
113 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
119 /* See if there was a valid master node before that */
126 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
127 if (ret != SCANNED_A_NODE && offs) {
128 /* Could have been corruption so check one place back */
132 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
133 if (ret != SCANNED_A_NODE)
135 * We accept only one area of corruption because
136 * we are assuming that it was caused while
137 * trying to write a master node.
141 if (ret == SCANNED_A_NODE) {
142 struct ubifs_ch *ch = buf;
144 if (ch->node_type != UBIFS_MST_NODE)
146 dbg_rcvry("found a master node at %d:%d", lnum, offs);
153 /* Check for corruption */
154 if (offs < c->leb_size) {
155 if (!is_empty(buf, min_t(int, len, sz))) {
157 dbg_rcvry("found corruption at %d:%d", lnum, offs);
163 /* Check remaining empty space */
164 if (offs < c->leb_size)
165 if (!is_empty(buf, len))
180 * write_rcvrd_mst_node - write recovered master node.
181 * @c: UBIFS file-system description object
184 * This function returns %0 on success and a negative error code on failure.
186 static int write_rcvrd_mst_node(struct ubifs_info *c,
187 struct ubifs_mst_node *mst)
189 int err = 0, lnum = UBIFS_MST_LNUM, sz = c->mst_node_alsz;
192 dbg_rcvry("recovery");
194 save_flags = mst->flags;
195 mst->flags |= cpu_to_le32(UBIFS_MST_RCVRY);
197 ubifs_prepare_node(c, mst, UBIFS_MST_NODE_SZ, 1);
198 err = ubi_leb_change(c->ubi, lnum, mst, sz, UBI_SHORTTERM);
201 err = ubi_leb_change(c->ubi, lnum + 1, mst, sz, UBI_SHORTTERM);
205 mst->flags = save_flags;
210 * ubifs_recover_master_node - recover the master node.
211 * @c: UBIFS file-system description object
213 * This function recovers the master node from corruption that may occur due to
214 * an unclean unmount.
216 * This function returns %0 on success and a negative error code on failure.
218 int ubifs_recover_master_node(struct ubifs_info *c)
220 void *buf1 = NULL, *buf2 = NULL, *cor1 = NULL, *cor2 = NULL;
221 struct ubifs_mst_node *mst1 = NULL, *mst2 = NULL, *mst;
222 const int sz = c->mst_node_alsz;
223 int err, offs1, offs2;
225 dbg_rcvry("recovery");
227 err = get_master_node(c, UBIFS_MST_LNUM, &buf1, &mst1, &cor1);
231 err = get_master_node(c, UBIFS_MST_LNUM + 1, &buf2, &mst2, &cor2);
236 offs1 = (void *)mst1 - buf1;
237 if ((le32_to_cpu(mst1->flags) & UBIFS_MST_RCVRY) &&
238 (offs1 == 0 && !cor1)) {
240 * mst1 was written by recovery at offset 0 with no
243 dbg_rcvry("recovery recovery");
246 offs2 = (void *)mst2 - buf2;
247 if (offs1 == offs2) {
248 /* Same offset, so must be the same */
249 if (memcmp((void *)mst1 + UBIFS_CH_SZ,
250 (void *)mst2 + UBIFS_CH_SZ,
251 UBIFS_MST_NODE_SZ - UBIFS_CH_SZ))
254 } else if (offs2 + sz == offs1) {
255 /* 1st LEB was written, 2nd was not */
259 } else if (offs1 == 0 && offs2 + sz >= c->leb_size) {
260 /* 1st LEB was unmapped and written, 2nd not */
268 * 2nd LEB was unmapped and about to be written, so
269 * there must be only one master node in the first LEB
272 if (offs1 != 0 || cor1)
280 * 1st LEB was unmapped and about to be written, so there must
281 * be no room left in 2nd LEB.
283 offs2 = (void *)mst2 - buf2;
284 if (offs2 + sz + sz <= c->leb_size)
289 dbg_rcvry("recovered master node from LEB %d",
290 (mst == mst1 ? UBIFS_MST_LNUM : UBIFS_MST_LNUM + 1));
292 memcpy(c->mst_node, mst, UBIFS_MST_NODE_SZ);
294 if ((c->vfs_sb->s_flags & MS_RDONLY)) {
295 /* Read-only mode. Keep a copy for switching to rw mode */
296 c->rcvrd_mst_node = kmalloc(sz, GFP_KERNEL);
297 if (!c->rcvrd_mst_node) {
301 memcpy(c->rcvrd_mst_node, c->mst_node, UBIFS_MST_NODE_SZ);
303 /* Write the recovered master node */
304 c->max_sqnum = le64_to_cpu(mst->ch.sqnum) - 1;
305 err = write_rcvrd_mst_node(c, c->mst_node);
318 ubifs_err("failed to recover master node");
320 dbg_err("dumping first master node");
321 dbg_dump_node(c, mst1);
324 dbg_err("dumping second master node");
325 dbg_dump_node(c, mst2);
333 * ubifs_write_rcvrd_mst_node - write the recovered master node.
334 * @c: UBIFS file-system description object
336 * This function writes the master node that was recovered during mounting in
337 * read-only mode and must now be written because we are remounting rw.
339 * This function returns %0 on success and a negative error code on failure.
341 int ubifs_write_rcvrd_mst_node(struct ubifs_info *c)
345 if (!c->rcvrd_mst_node)
347 c->rcvrd_mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
348 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
349 err = write_rcvrd_mst_node(c, c->rcvrd_mst_node);
352 kfree(c->rcvrd_mst_node);
353 c->rcvrd_mst_node = NULL;
358 * is_last_write - determine if an offset was in the last write to a LEB.
359 * @c: UBIFS file-system description object
360 * @buf: buffer to check
361 * @offs: offset to check
363 * This function returns %1 if @offs was in the last write to the LEB whose data
364 * is in @buf, otherwise %0 is returned. The determination is made by checking
365 * for subsequent empty space starting from the next @c->min_io_size boundary.
367 static int is_last_write(const struct ubifs_info *c, void *buf, int offs)
369 int empty_offs, check_len;
373 * Round up to the next @c->min_io_size boundary i.e. @offs is in the
374 * last wbuf written. After that should be empty space.
376 empty_offs = ALIGN(offs + 1, c->min_io_size);
377 check_len = c->leb_size - empty_offs;
378 p = buf + empty_offs - offs;
379 return is_empty(p, check_len);
383 * clean_buf - clean the data from an LEB sitting in a buffer.
384 * @c: UBIFS file-system description object
385 * @buf: buffer to clean
386 * @lnum: LEB number to clean
387 * @offs: offset from which to clean
388 * @len: length of buffer
390 * This function pads up to the next min_io_size boundary (if there is one) and
391 * sets empty space to all 0xff. @buf, @offs and @len are updated to the next
392 * @c->min_io_size boundary.
394 static void clean_buf(const struct ubifs_info *c, void **buf, int lnum,
397 int empty_offs, pad_len;
400 dbg_rcvry("cleaning corruption at %d:%d", lnum, *offs);
402 ubifs_assert(!(*offs & 7));
403 empty_offs = ALIGN(*offs, c->min_io_size);
404 pad_len = empty_offs - *offs;
405 ubifs_pad(c, *buf, pad_len);
409 memset(*buf, 0xff, c->leb_size - empty_offs);
413 * no_more_nodes - determine if there are no more nodes in a buffer.
414 * @c: UBIFS file-system description object
415 * @buf: buffer to check
416 * @len: length of buffer
417 * @lnum: LEB number of the LEB from which @buf was read
418 * @offs: offset from which @buf was read
420 * This function ensures that the corrupted node at @offs is the last thing
421 * written to a LEB. This function returns %1 if more data is not found and
422 * %0 if more data is found.
424 static int no_more_nodes(const struct ubifs_info *c, void *buf, int len,
427 struct ubifs_ch *ch = buf;
428 int skip, dlen = le32_to_cpu(ch->len);
430 /* Check for empty space after the corrupt node's common header */
431 skip = ALIGN(offs + UBIFS_CH_SZ, c->min_io_size) - offs;
432 if (is_empty(buf + skip, len - skip))
435 * The area after the common header size is not empty, so the common
436 * header must be intact. Check it.
438 if (ubifs_check_node(c, buf, lnum, offs, 1, 0) != -EUCLEAN) {
439 dbg_rcvry("unexpected bad common header at %d:%d", lnum, offs);
442 /* Now we know the corrupt node's length we can skip over it */
443 skip = ALIGN(offs + dlen, c->min_io_size) - offs;
444 /* After which there should be empty space */
445 if (is_empty(buf + skip, len - skip))
447 dbg_rcvry("unexpected data at %d:%d", lnum, offs + skip);
452 * fix_unclean_leb - fix an unclean LEB.
453 * @c: UBIFS file-system description object
454 * @sleb: scanned LEB information
455 * @start: offset where scan started
457 static int fix_unclean_leb(struct ubifs_info *c, struct ubifs_scan_leb *sleb,
460 int lnum = sleb->lnum, endpt = start;
462 /* Get the end offset of the last node we are keeping */
463 if (!list_empty(&sleb->nodes)) {
464 struct ubifs_scan_node *snod;
466 snod = list_entry(sleb->nodes.prev,
467 struct ubifs_scan_node, list);
468 endpt = snod->offs + snod->len;
471 if ((c->vfs_sb->s_flags & MS_RDONLY) && !c->remounting_rw) {
472 /* Add to recovery list */
473 struct ubifs_unclean_leb *ucleb;
475 dbg_rcvry("need to fix LEB %d start %d endpt %d",
476 lnum, start, sleb->endpt);
477 ucleb = kzalloc(sizeof(struct ubifs_unclean_leb), GFP_NOFS);
481 ucleb->endpt = endpt;
482 list_add_tail(&ucleb->list, &c->unclean_leb_list);
484 /* Write the fixed LEB back to flash */
487 dbg_rcvry("fixing LEB %d start %d endpt %d",
488 lnum, start, sleb->endpt);
490 err = ubifs_leb_unmap(c, lnum);
494 int len = ALIGN(endpt, c->min_io_size);
497 err = ubi_read(c->ubi, lnum, sleb->buf, 0,
502 /* Pad to min_io_size */
504 int pad_len = len - ALIGN(endpt, 8);
507 void *buf = sleb->buf + len - pad_len;
509 ubifs_pad(c, buf, pad_len);
512 err = ubi_leb_change(c->ubi, lnum, sleb->buf, len,
522 * drop_incomplete_group - drop nodes from an incomplete group.
523 * @sleb: scanned LEB information
524 * @offs: offset of dropped nodes is returned here
526 * This function returns %1 if nodes are dropped and %0 otherwise.
528 static int drop_incomplete_group(struct ubifs_scan_leb *sleb, int *offs)
532 while (!list_empty(&sleb->nodes)) {
533 struct ubifs_scan_node *snod;
536 snod = list_entry(sleb->nodes.prev, struct ubifs_scan_node,
539 if (ch->group_type != UBIFS_IN_NODE_GROUP)
541 dbg_rcvry("dropping node at %d:%d", sleb->lnum, snod->offs);
543 list_del(&snod->list);
545 sleb->nodes_cnt -= 1;
552 * ubifs_recover_leb - scan and recover a LEB.
553 * @c: UBIFS file-system description object
556 * @sbuf: LEB-sized buffer to use
557 * @grouped: nodes may be grouped for recovery
559 * This function does a scan of a LEB, but caters for errors that might have
560 * been caused by the unclean unmount from which we are attempting to recover.
561 * Returns %0 in case of success, %-EUCLEAN if an unrecoverable corruption is
562 * found, and a negative error code in case of failure.
564 struct ubifs_scan_leb *ubifs_recover_leb(struct ubifs_info *c, int lnum,
565 int offs, void *sbuf, int grouped)
567 int err, len = c->leb_size - offs, need_clean = 0, quiet = 1;
568 int empty_chkd = 0, start = offs;
569 struct ubifs_scan_leb *sleb;
570 void *buf = sbuf + offs;
572 dbg_rcvry("%d:%d", lnum, offs);
574 sleb = ubifs_start_scan(c, lnum, offs, sbuf);
584 dbg_scan("look at LEB %d:%d (%d bytes left)",
590 * Scan quietly until there is an error from which we cannot
593 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, quiet);
595 if (ret == SCANNED_A_NODE) {
596 /* A valid node, and not a padding node */
597 struct ubifs_ch *ch = buf;
600 err = ubifs_add_snod(c, sleb, buf, offs);
603 node_len = ALIGN(le32_to_cpu(ch->len), 8);
611 /* Padding bytes or a valid padding node */
618 if (ret == SCANNED_EMPTY_SPACE) {
619 if (!is_empty(buf, len)) {
620 if (!is_last_write(c, buf, offs))
622 clean_buf(c, &buf, lnum, &offs, &len);
629 if (ret == SCANNED_GARBAGE || ret == SCANNED_A_BAD_PAD_NODE)
630 if (is_last_write(c, buf, offs)) {
631 clean_buf(c, &buf, lnum, &offs, &len);
637 if (ret == SCANNED_A_CORRUPT_NODE)
638 if (no_more_nodes(c, buf, len, lnum, offs)) {
639 clean_buf(c, &buf, lnum, &offs, &len);
646 /* Redo the last scan but noisily */
652 case SCANNED_GARBAGE:
655 case SCANNED_A_CORRUPT_NODE:
656 case SCANNED_A_BAD_PAD_NODE:
666 if (!empty_chkd && !is_empty(buf, len)) {
667 if (is_last_write(c, buf, offs)) {
668 clean_buf(c, &buf, lnum, &offs, &len);
671 int corruption = first_non_ff(buf, len);
673 ubifs_err("corrupt empty space LEB %d:%d, corruption "
674 "starts at %d", lnum, offs, corruption);
675 /* Make sure we dump interesting non-0xFF data */
682 /* Drop nodes from incomplete group */
683 if (grouped && drop_incomplete_group(sleb, &offs)) {
685 len = c->leb_size - offs;
686 clean_buf(c, &buf, lnum, &offs, &len);
690 if (offs % c->min_io_size) {
691 clean_buf(c, &buf, lnum, &offs, &len);
695 ubifs_end_scan(c, sleb, lnum, offs);
698 err = fix_unclean_leb(c, sleb, start);
706 ubifs_scanned_corruption(c, lnum, offs, buf);
709 ubifs_err("LEB %d scanning failed", lnum);
710 ubifs_scan_destroy(sleb);
715 * get_cs_sqnum - get commit start sequence number.
716 * @c: UBIFS file-system description object
717 * @lnum: LEB number of commit start node
718 * @offs: offset of commit start node
719 * @cs_sqnum: commit start sequence number is returned here
721 * This function returns %0 on success and a negative error code on failure.
723 static int get_cs_sqnum(struct ubifs_info *c, int lnum, int offs,
724 unsigned long long *cs_sqnum)
726 struct ubifs_cs_node *cs_node = NULL;
729 dbg_rcvry("at %d:%d", lnum, offs);
730 cs_node = kmalloc(UBIFS_CS_NODE_SZ, GFP_KERNEL);
733 if (c->leb_size - offs < UBIFS_CS_NODE_SZ)
735 err = ubi_read(c->ubi, lnum, (void *)cs_node, offs, UBIFS_CS_NODE_SZ);
736 if (err && err != -EBADMSG)
738 ret = ubifs_scan_a_node(c, cs_node, UBIFS_CS_NODE_SZ, lnum, offs, 0);
739 if (ret != SCANNED_A_NODE) {
740 dbg_err("Not a valid node");
743 if (cs_node->ch.node_type != UBIFS_CS_NODE) {
744 dbg_err("Node a CS node, type is %d", cs_node->ch.node_type);
747 if (le64_to_cpu(cs_node->cmt_no) != c->cmt_no) {
748 dbg_err("CS node cmt_no %llu != current cmt_no %llu",
749 (unsigned long long)le64_to_cpu(cs_node->cmt_no),
753 *cs_sqnum = le64_to_cpu(cs_node->ch.sqnum);
754 dbg_rcvry("commit start sqnum %llu", *cs_sqnum);
761 ubifs_err("failed to get CS sqnum");
767 * ubifs_recover_log_leb - scan and recover a log LEB.
768 * @c: UBIFS file-system description object
771 * @sbuf: LEB-sized buffer to use
773 * This function does a scan of a LEB, but caters for errors that might have
774 * been caused by the unclean unmount from which we are attempting to recover.
776 * This function returns %0 on success and a negative error code on failure.
778 struct ubifs_scan_leb *ubifs_recover_log_leb(struct ubifs_info *c, int lnum,
779 int offs, void *sbuf)
781 struct ubifs_scan_leb *sleb;
784 dbg_rcvry("LEB %d", lnum);
785 next_lnum = lnum + 1;
786 if (next_lnum >= UBIFS_LOG_LNUM + c->log_lebs)
787 next_lnum = UBIFS_LOG_LNUM;
788 if (next_lnum != c->ltail_lnum) {
790 * We can only recover at the end of the log, so check that the
791 * next log LEB is empty or out of date.
793 sleb = ubifs_scan(c, next_lnum, 0, sbuf);
796 if (sleb->nodes_cnt) {
797 struct ubifs_scan_node *snod;
798 unsigned long long cs_sqnum = c->cs_sqnum;
800 snod = list_entry(sleb->nodes.next,
801 struct ubifs_scan_node, list);
805 err = get_cs_sqnum(c, lnum, offs, &cs_sqnum);
807 ubifs_scan_destroy(sleb);
811 if (snod->sqnum > cs_sqnum) {
812 ubifs_err("unrecoverable log corruption "
814 ubifs_scan_destroy(sleb);
815 return ERR_PTR(-EUCLEAN);
818 ubifs_scan_destroy(sleb);
820 return ubifs_recover_leb(c, lnum, offs, sbuf, 0);
824 * recover_head - recover a head.
825 * @c: UBIFS file-system description object
826 * @lnum: LEB number of head to recover
827 * @offs: offset of head to recover
828 * @sbuf: LEB-sized buffer to use
830 * This function ensures that there is no data on the flash at a head location.
832 * This function returns %0 on success and a negative error code on failure.
834 static int recover_head(const struct ubifs_info *c, int lnum, int offs,
839 if (c->min_io_size > 1)
840 len = c->min_io_size;
843 if (offs + len > c->leb_size)
844 len = c->leb_size - offs;
849 /* Read at the head location and check it is empty flash */
850 err = ubi_read(c->ubi, lnum, sbuf, offs, len);
851 if (err || !is_empty(sbuf, len)) {
852 dbg_rcvry("cleaning head at %d:%d", lnum, offs);
854 return ubifs_leb_unmap(c, lnum);
855 err = ubi_read(c->ubi, lnum, sbuf, 0, offs);
858 return ubi_leb_change(c->ubi, lnum, sbuf, offs, UBI_UNKNOWN);
865 * ubifs_recover_inl_heads - recover index and LPT heads.
866 * @c: UBIFS file-system description object
867 * @sbuf: LEB-sized buffer to use
869 * This function ensures that there is no data on the flash at the index and
870 * LPT head locations.
872 * This deals with the recovery of a half-completed journal commit. UBIFS is
873 * careful never to overwrite the last version of the index or the LPT. Because
874 * the index and LPT are wandering trees, data from a half-completed commit will
875 * not be referenced anywhere in UBIFS. The data will be either in LEBs that are
876 * assumed to be empty and will be unmapped anyway before use, or in the index
879 * This function returns %0 on success and a negative error code on failure.
881 int ubifs_recover_inl_heads(const struct ubifs_info *c, void *sbuf)
885 ubifs_assert(!(c->vfs_sb->s_flags & MS_RDONLY) || c->remounting_rw);
887 dbg_rcvry("checking index head at %d:%d", c->ihead_lnum, c->ihead_offs);
888 err = recover_head(c, c->ihead_lnum, c->ihead_offs, sbuf);
892 dbg_rcvry("checking LPT head at %d:%d", c->nhead_lnum, c->nhead_offs);
893 err = recover_head(c, c->nhead_lnum, c->nhead_offs, sbuf);
901 * clean_an_unclean_leb - read and write a LEB to remove corruption.
902 * @c: UBIFS file-system description object
903 * @ucleb: unclean LEB information
904 * @sbuf: LEB-sized buffer to use
906 * This function reads a LEB up to a point pre-determined by the mount recovery,
907 * checks the nodes, and writes the result back to the flash, thereby cleaning
908 * off any following corruption, or non-fatal ECC errors.
910 * This function returns %0 on success and a negative error code on failure.
912 static int clean_an_unclean_leb(const struct ubifs_info *c,
913 struct ubifs_unclean_leb *ucleb, void *sbuf)
915 int err, lnum = ucleb->lnum, offs = 0, len = ucleb->endpt, quiet = 1;
918 dbg_rcvry("LEB %d len %d", lnum, len);
921 /* Nothing to read, just unmap it */
922 err = ubifs_leb_unmap(c, lnum);
928 err = ubi_read(c->ubi, lnum, buf, offs, len);
929 if (err && err != -EBADMSG)
937 /* Scan quietly until there is an error */
938 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, quiet);
940 if (ret == SCANNED_A_NODE) {
941 /* A valid node, and not a padding node */
942 struct ubifs_ch *ch = buf;
945 node_len = ALIGN(le32_to_cpu(ch->len), 8);
953 /* Padding bytes or a valid padding node */
960 if (ret == SCANNED_EMPTY_SPACE) {
961 ubifs_err("unexpected empty space at %d:%d",
967 /* Redo the last scan but noisily */
972 ubifs_scanned_corruption(c, lnum, offs, buf);
976 /* Pad to min_io_size */
977 len = ALIGN(ucleb->endpt, c->min_io_size);
978 if (len > ucleb->endpt) {
979 int pad_len = len - ALIGN(ucleb->endpt, 8);
982 buf = c->sbuf + len - pad_len;
983 ubifs_pad(c, buf, pad_len);
987 /* Write back the LEB atomically */
988 err = ubi_leb_change(c->ubi, lnum, sbuf, len, UBI_UNKNOWN);
992 dbg_rcvry("cleaned LEB %d", lnum);
998 * ubifs_clean_lebs - clean LEBs recovered during read-only mount.
999 * @c: UBIFS file-system description object
1000 * @sbuf: LEB-sized buffer to use
1002 * This function cleans a LEB identified during recovery that needs to be
1003 * written but was not because UBIFS was mounted read-only. This happens when
1004 * remounting to read-write mode.
1006 * This function returns %0 on success and a negative error code on failure.
1008 int ubifs_clean_lebs(const struct ubifs_info *c, void *sbuf)
1010 dbg_rcvry("recovery");
1011 while (!list_empty(&c->unclean_leb_list)) {
1012 struct ubifs_unclean_leb *ucleb;
1015 ucleb = list_entry(c->unclean_leb_list.next,
1016 struct ubifs_unclean_leb, list);
1017 err = clean_an_unclean_leb(c, ucleb, sbuf);
1020 list_del(&ucleb->list);
1027 * ubifs_rcvry_gc_commit - recover the GC LEB number and run the commit.
1028 * @c: UBIFS file-system description object
1030 * Out-of-place garbage collection requires always one empty LEB with which to
1031 * start garbage collection. The LEB number is recorded in c->gc_lnum and is
1032 * written to the master node on unmounting. In the case of an unclean unmount
1033 * the value of gc_lnum recorded in the master node is out of date and cannot
1034 * be used. Instead, recovery must allocate an empty LEB for this purpose.
1035 * However, there may not be enough empty space, in which case it must be
1036 * possible to GC the dirtiest LEB into the GC head LEB.
1038 * This function also runs the commit which causes the TNC updates from
1039 * size-recovery and orphans to be written to the flash. That is important to
1040 * ensure correct replay order for subsequent mounts.
1042 * This function returns %0 on success and a negative error code on failure.
1044 int ubifs_rcvry_gc_commit(struct ubifs_info *c)
1046 struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf;
1047 struct ubifs_lprops lp;
1051 if (wbuf->lnum == -1) {
1052 dbg_rcvry("no GC head LEB");
1056 * See whether the used space in the dirtiest LEB fits in the GC head
1059 if (wbuf->offs == c->leb_size) {
1060 dbg_rcvry("no room in GC head LEB");
1063 err = ubifs_find_dirty_leb(c, &lp, wbuf->offs, 2);
1066 dbg_err("could not find a dirty LEB");
1069 ubifs_assert(!(lp.flags & LPROPS_INDEX));
1071 if (lp.free + lp.dirty == c->leb_size) {
1072 /* An empty LEB was returned */
1073 if (lp.free != c->leb_size) {
1074 err = ubifs_change_one_lp(c, lnum, c->leb_size,
1079 err = ubifs_leb_unmap(c, lnum);
1083 dbg_rcvry("allocated LEB %d for GC", lnum);
1084 /* Run the commit */
1085 dbg_rcvry("committing");
1086 return ubifs_run_commit(c);
1089 * There was no empty LEB so the used space in the dirtiest LEB must fit
1090 * in the GC head LEB.
1092 if (lp.free + lp.dirty < wbuf->offs) {
1093 dbg_rcvry("LEB %d doesn't fit in GC head LEB %d:%d",
1094 lnum, wbuf->lnum, wbuf->offs);
1095 err = ubifs_return_leb(c, lnum);
1101 * We run the commit before garbage collection otherwise subsequent
1102 * mounts will see the GC and orphan deletion in a different order.
1104 dbg_rcvry("committing");
1105 err = ubifs_run_commit(c);
1109 * The data in the dirtiest LEB fits in the GC head LEB, so do the GC
1110 * - use locking to keep 'ubifs_assert()' happy.
1112 dbg_rcvry("GC'ing LEB %d", lnum);
1113 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1114 err = ubifs_garbage_collect_leb(c, &lp);
1116 int err2 = ubifs_wbuf_sync_nolock(wbuf);
1121 mutex_unlock(&wbuf->io_mutex);
1123 dbg_err("GC failed, error %d", err);
1128 if (err != LEB_RETAINED) {
1129 dbg_err("GC returned %d", err);
1132 err = ubifs_leb_unmap(c, c->gc_lnum);
1135 dbg_rcvry("allocated LEB %d for GC", lnum);
1140 * There is no GC head LEB or the free space in the GC head LEB is too
1141 * small. Allocate gc_lnum by calling 'ubifs_find_free_leb_for_idx()' so
1144 lnum = ubifs_find_free_leb_for_idx(c);
1146 dbg_err("could not find an empty LEB");
1149 /* And reset the index flag */
1150 err = ubifs_change_one_lp(c, lnum, LPROPS_NC, LPROPS_NC, 0,
1155 dbg_rcvry("allocated LEB %d for GC", lnum);
1156 /* Run the commit */
1157 dbg_rcvry("committing");
1158 return ubifs_run_commit(c);
1162 * struct size_entry - inode size information for recovery.
1163 * @rb: link in the RB-tree of sizes
1164 * @inum: inode number
1165 * @i_size: size on inode
1166 * @d_size: maximum size based on data nodes
1167 * @exists: indicates whether the inode exists
1168 * @inode: inode if pinned in memory awaiting rw mode to fix it
1176 struct inode *inode;
1180 * add_ino - add an entry to the size tree.
1181 * @c: UBIFS file-system description object
1182 * @inum: inode number
1183 * @i_size: size on inode
1184 * @d_size: maximum size based on data nodes
1185 * @exists: indicates whether the inode exists
1187 static int add_ino(struct ubifs_info *c, ino_t inum, loff_t i_size,
1188 loff_t d_size, int exists)
1190 struct rb_node **p = &c->size_tree.rb_node, *parent = NULL;
1191 struct size_entry *e;
1195 e = rb_entry(parent, struct size_entry, rb);
1199 p = &(*p)->rb_right;
1202 e = kzalloc(sizeof(struct size_entry), GFP_KERNEL);
1211 rb_link_node(&e->rb, parent, p);
1212 rb_insert_color(&e->rb, &c->size_tree);
1218 * find_ino - find an entry on the size tree.
1219 * @c: UBIFS file-system description object
1220 * @inum: inode number
1222 static struct size_entry *find_ino(struct ubifs_info *c, ino_t inum)
1224 struct rb_node *p = c->size_tree.rb_node;
1225 struct size_entry *e;
1228 e = rb_entry(p, struct size_entry, rb);
1231 else if (inum > e->inum)
1240 * remove_ino - remove an entry from the size tree.
1241 * @c: UBIFS file-system description object
1242 * @inum: inode number
1244 static void remove_ino(struct ubifs_info *c, ino_t inum)
1246 struct size_entry *e = find_ino(c, inum);
1250 rb_erase(&e->rb, &c->size_tree);
1255 * ubifs_destroy_size_tree - free resources related to the size tree.
1256 * @c: UBIFS file-system description object
1258 void ubifs_destroy_size_tree(struct ubifs_info *c)
1260 struct rb_node *this = c->size_tree.rb_node;
1261 struct size_entry *e;
1264 if (this->rb_left) {
1265 this = this->rb_left;
1267 } else if (this->rb_right) {
1268 this = this->rb_right;
1271 e = rb_entry(this, struct size_entry, rb);
1274 this = rb_parent(this);
1276 if (this->rb_left == &e->rb)
1277 this->rb_left = NULL;
1279 this->rb_right = NULL;
1283 c->size_tree = RB_ROOT;
1287 * ubifs_recover_size_accum - accumulate inode sizes for recovery.
1288 * @c: UBIFS file-system description object
1290 * @deletion: node is for a deletion
1291 * @new_size: inode size
1293 * This function has two purposes:
1294 * 1) to ensure there are no data nodes that fall outside the inode size
1295 * 2) to ensure there are no data nodes for inodes that do not exist
1296 * To accomplish those purposes, a rb-tree is constructed containing an entry
1297 * for each inode number in the journal that has not been deleted, and recording
1298 * the size from the inode node, the maximum size of any data node (also altered
1299 * by truncations) and a flag indicating a inode number for which no inode node
1300 * was present in the journal.
1302 * Note that there is still the possibility that there are data nodes that have
1303 * been committed that are beyond the inode size, however the only way to find
1304 * them would be to scan the entire index. Alternatively, some provision could
1305 * be made to record the size of inodes at the start of commit, which would seem
1306 * very cumbersome for a scenario that is quite unlikely and the only negative
1307 * consequence of which is wasted space.
1309 * This functions returns %0 on success and a negative error code on failure.
1311 int ubifs_recover_size_accum(struct ubifs_info *c, union ubifs_key *key,
1312 int deletion, loff_t new_size)
1314 ino_t inum = key_inum(c, key);
1315 struct size_entry *e;
1318 switch (key_type(c, key)) {
1321 remove_ino(c, inum);
1323 e = find_ino(c, inum);
1325 e->i_size = new_size;
1328 err = add_ino(c, inum, new_size, 0, 1);
1334 case UBIFS_DATA_KEY:
1335 e = find_ino(c, inum);
1337 if (new_size > e->d_size)
1338 e->d_size = new_size;
1340 err = add_ino(c, inum, 0, new_size, 0);
1345 case UBIFS_TRUN_KEY:
1346 e = find_ino(c, inum);
1348 e->d_size = new_size;
1355 * fix_size_in_place - fix inode size in place on flash.
1356 * @c: UBIFS file-system description object
1357 * @e: inode size information for recovery
1359 static int fix_size_in_place(struct ubifs_info *c, struct size_entry *e)
1361 struct ubifs_ino_node *ino = c->sbuf;
1363 union ubifs_key key;
1364 int err, lnum, offs, len;
1368 /* Locate the inode node LEB number and offset */
1369 ino_key_init(c, &key, e->inum);
1370 err = ubifs_tnc_locate(c, &key, ino, &lnum, &offs);
1374 * If the size recorded on the inode node is greater than the size that
1375 * was calculated from nodes in the journal then don't change the inode.
1377 i_size = le64_to_cpu(ino->size);
1378 if (i_size >= e->d_size)
1381 err = ubi_read(c->ubi, lnum, c->sbuf, 0, c->leb_size);
1384 /* Change the size field and recalculate the CRC */
1385 ino = c->sbuf + offs;
1386 ino->size = cpu_to_le64(e->d_size);
1387 len = le32_to_cpu(ino->ch.len);
1388 crc = crc32(UBIFS_CRC32_INIT, (void *)ino + 8, len - 8);
1389 ino->ch.crc = cpu_to_le32(crc);
1390 /* Work out where data in the LEB ends and free space begins */
1392 len = c->leb_size - 1;
1393 while (p[len] == 0xff)
1395 len = ALIGN(len + 1, c->min_io_size);
1396 /* Atomically write the fixed LEB back again */
1397 err = ubi_leb_change(c->ubi, lnum, c->sbuf, len, UBI_UNKNOWN);
1400 dbg_rcvry("inode %lu at %d:%d size %lld -> %lld ",
1401 (unsigned long)e->inum, lnum, offs, i_size, e->d_size);
1405 ubifs_warn("inode %lu failed to fix size %lld -> %lld error %d",
1406 (unsigned long)e->inum, e->i_size, e->d_size, err);
1411 * ubifs_recover_size - recover inode size.
1412 * @c: UBIFS file-system description object
1414 * This function attempts to fix inode size discrepancies identified by the
1415 * 'ubifs_recover_size_accum()' function.
1417 * This functions returns %0 on success and a negative error code on failure.
1419 int ubifs_recover_size(struct ubifs_info *c)
1421 struct rb_node *this = rb_first(&c->size_tree);
1424 struct size_entry *e;
1427 e = rb_entry(this, struct size_entry, rb);
1429 union ubifs_key key;
1431 ino_key_init(c, &key, e->inum);
1432 err = ubifs_tnc_lookup(c, &key, c->sbuf);
1433 if (err && err != -ENOENT)
1435 if (err == -ENOENT) {
1436 /* Remove data nodes that have no inode */
1437 dbg_rcvry("removing ino %lu",
1438 (unsigned long)e->inum);
1439 err = ubifs_tnc_remove_ino(c, e->inum);
1443 struct ubifs_ino_node *ino = c->sbuf;
1446 e->i_size = le64_to_cpu(ino->size);
1449 if (e->exists && e->i_size < e->d_size) {
1450 if (!e->inode && (c->vfs_sb->s_flags & MS_RDONLY)) {
1451 /* Fix the inode size and pin it in memory */
1452 struct inode *inode;
1454 inode = ubifs_iget(c->vfs_sb, e->inum);
1456 return PTR_ERR(inode);
1457 if (inode->i_size < e->d_size) {
1458 dbg_rcvry("ino %lu size %lld -> %lld",
1459 (unsigned long)e->inum,
1460 e->d_size, inode->i_size);
1461 inode->i_size = e->d_size;
1462 ubifs_inode(inode)->ui_size = e->d_size;
1464 this = rb_next(this);
1469 /* Fix the size in place */
1470 err = fix_size_in_place(c, e);
1477 this = rb_next(this);
1478 rb_erase(&e->rb, &c->size_tree);