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 * get_master_node - get the last valid master node allowing for corruption.
57 * @c: UBIFS file-system description object
59 * @pbuf: buffer containing the LEB read, is returned here
60 * @mst: master node, if found, is returned here
61 * @cor: corruption, if found, is returned here
63 * This function allocates a buffer, reads the LEB into it, and finds and
64 * returns the last valid master node allowing for one area of corruption.
65 * The corrupt area, if there is one, must be consistent with the assumption
66 * that it is the result of an unclean unmount while the master node was being
67 * written. Under those circumstances, it is valid to use the previously written
70 * This function returns %0 on success and a negative error code on failure.
72 static int get_master_node(const struct ubifs_info *c, int lnum, void **pbuf,
73 struct ubifs_mst_node **mst, void **cor)
75 const int sz = c->mst_node_alsz;
79 sbuf = vmalloc(c->leb_size);
83 err = ubi_read(c->ubi, lnum, sbuf, 0, c->leb_size);
84 if (err && err != -EBADMSG)
87 /* Find the first position that is definitely not a node */
91 while (offs + UBIFS_MST_NODE_SZ <= c->leb_size) {
92 struct ubifs_ch *ch = buf;
94 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
100 /* See if there was a valid master node before that */
107 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
108 if (ret != SCANNED_A_NODE && offs) {
109 /* Could have been corruption so check one place back */
113 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
114 if (ret != SCANNED_A_NODE)
116 * We accept only one area of corruption because
117 * we are assuming that it was caused while
118 * trying to write a master node.
122 if (ret == SCANNED_A_NODE) {
123 struct ubifs_ch *ch = buf;
125 if (ch->node_type != UBIFS_MST_NODE)
127 dbg_rcvry("found a master node at %d:%d", lnum, offs);
134 /* Check for corruption */
135 if (offs < c->leb_size) {
136 if (!is_empty(buf, min_t(int, len, sz))) {
138 dbg_rcvry("found corruption at %d:%d", lnum, offs);
144 /* Check remaining empty space */
145 if (offs < c->leb_size)
146 if (!is_empty(buf, len))
161 * write_rcvrd_mst_node - write recovered master node.
162 * @c: UBIFS file-system description object
165 * This function returns %0 on success and a negative error code on failure.
167 static int write_rcvrd_mst_node(struct ubifs_info *c,
168 struct ubifs_mst_node *mst)
170 int err = 0, lnum = UBIFS_MST_LNUM, sz = c->mst_node_alsz;
173 dbg_rcvry("recovery");
175 save_flags = mst->flags;
176 mst->flags = cpu_to_le32(le32_to_cpu(mst->flags) | UBIFS_MST_RCVRY);
178 ubifs_prepare_node(c, mst, UBIFS_MST_NODE_SZ, 1);
179 err = ubi_leb_change(c->ubi, lnum, mst, sz, UBI_SHORTTERM);
182 err = ubi_leb_change(c->ubi, lnum + 1, mst, sz, UBI_SHORTTERM);
186 mst->flags = save_flags;
191 * ubifs_recover_master_node - recover the master node.
192 * @c: UBIFS file-system description object
194 * This function recovers the master node from corruption that may occur due to
195 * an unclean unmount.
197 * This function returns %0 on success and a negative error code on failure.
199 int ubifs_recover_master_node(struct ubifs_info *c)
201 void *buf1 = NULL, *buf2 = NULL, *cor1 = NULL, *cor2 = NULL;
202 struct ubifs_mst_node *mst1 = NULL, *mst2 = NULL, *mst;
203 const int sz = c->mst_node_alsz;
204 int err, offs1, offs2;
206 dbg_rcvry("recovery");
208 err = get_master_node(c, UBIFS_MST_LNUM, &buf1, &mst1, &cor1);
212 err = get_master_node(c, UBIFS_MST_LNUM + 1, &buf2, &mst2, &cor2);
217 offs1 = (void *)mst1 - buf1;
218 if ((le32_to_cpu(mst1->flags) & UBIFS_MST_RCVRY) &&
219 (offs1 == 0 && !cor1)) {
221 * mst1 was written by recovery at offset 0 with no
224 dbg_rcvry("recovery recovery");
227 offs2 = (void *)mst2 - buf2;
228 if (offs1 == offs2) {
229 /* Same offset, so must be the same */
230 if (memcmp((void *)mst1 + UBIFS_CH_SZ,
231 (void *)mst2 + UBIFS_CH_SZ,
232 UBIFS_MST_NODE_SZ - UBIFS_CH_SZ))
235 } else if (offs2 + sz == offs1) {
236 /* 1st LEB was written, 2nd was not */
240 } else if (offs1 == 0 && offs2 + sz >= c->leb_size) {
241 /* 1st LEB was unmapped and written, 2nd not */
249 * 2nd LEB was unmapped and about to be written, so
250 * there must be only one master node in the first LEB
253 if (offs1 != 0 || cor1)
261 * 1st LEB was unmapped and about to be written, so there must
262 * be no room left in 2nd LEB.
264 offs2 = (void *)mst2 - buf2;
265 if (offs2 + sz + sz <= c->leb_size)
270 dbg_rcvry("recovered master node from LEB %d",
271 (mst == mst1 ? UBIFS_MST_LNUM : UBIFS_MST_LNUM + 1));
273 memcpy(c->mst_node, mst, UBIFS_MST_NODE_SZ);
275 if ((c->vfs_sb->s_flags & MS_RDONLY)) {
276 /* Read-only mode. Keep a copy for switching to rw mode */
277 c->rcvrd_mst_node = kmalloc(sz, GFP_KERNEL);
278 if (!c->rcvrd_mst_node) {
282 memcpy(c->rcvrd_mst_node, c->mst_node, UBIFS_MST_NODE_SZ);
284 /* Write the recovered master node */
285 c->max_sqnum = le64_to_cpu(mst->ch.sqnum) - 1;
286 err = write_rcvrd_mst_node(c, c->mst_node);
299 ubifs_err("failed to recover master node");
301 dbg_err("dumping first master node");
302 dbg_dump_node(c, mst1);
305 dbg_err("dumping second master node");
306 dbg_dump_node(c, mst2);
314 * ubifs_write_rcvrd_mst_node - write the recovered master node.
315 * @c: UBIFS file-system description object
317 * This function writes the master node that was recovered during mounting in
318 * read-only mode and must now be written because we are remounting rw.
320 * This function returns %0 on success and a negative error code on failure.
322 int ubifs_write_rcvrd_mst_node(struct ubifs_info *c)
326 if (!c->rcvrd_mst_node)
328 c->rcvrd_mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
329 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
330 err = write_rcvrd_mst_node(c, c->rcvrd_mst_node);
333 kfree(c->rcvrd_mst_node);
334 c->rcvrd_mst_node = NULL;
339 * is_last_write - determine if an offset was in the last write to a LEB.
340 * @c: UBIFS file-system description object
341 * @buf: buffer to check
342 * @offs: offset to check
344 * This function returns %1 if @offs was in the last write to the LEB whose data
345 * is in @buf, otherwise %0 is returned. The determination is made by checking
346 * for subsequent empty space starting from the next min_io_size boundary (or a
347 * bit less than the common header size if min_io_size is one).
349 static int is_last_write(const struct ubifs_info *c, void *buf, int offs)
355 if (c->min_io_size == 1) {
356 check_len = c->leb_size - offs;
358 for (; check_len > 0; check_len--)
362 * 'check_len' is the size of the corruption which cannot be
363 * more than the size of 1 node if it was caused by an unclean
366 if (check_len > UBIFS_MAX_NODE_SZ)
372 * Round up to the next c->min_io_size boundary i.e. 'offs' is in the
373 * last wbuf written. After that should be empty space.
375 empty_offs = ALIGN(offs + 1, c->min_io_size);
376 check_len = c->leb_size - empty_offs;
377 p = buf + empty_offs - offs;
379 for (; check_len > 0; check_len--)
386 * clean_buf - clean the data from an LEB sitting in a buffer.
387 * @c: UBIFS file-system description object
388 * @buf: buffer to clean
389 * @lnum: LEB number to clean
390 * @offs: offset from which to clean
391 * @len: length of buffer
393 * This function pads up to the next min_io_size boundary (if there is one) and
394 * sets empty space to all 0xff. @buf, @offs and @len are updated to the next
395 * min_io_size boundary (if there is one).
397 static void clean_buf(const struct ubifs_info *c, void **buf, int lnum,
400 int empty_offs, pad_len;
403 dbg_rcvry("cleaning corruption at %d:%d", lnum, *offs);
405 if (c->min_io_size == 1) {
406 memset(*buf, 0xff, c->leb_size - *offs);
410 ubifs_assert(!(*offs & 7));
411 empty_offs = ALIGN(*offs, c->min_io_size);
412 pad_len = empty_offs - *offs;
413 ubifs_pad(c, *buf, pad_len);
417 memset(*buf, 0xff, c->leb_size - empty_offs);
421 * no_more_nodes - determine if there are no more nodes in a buffer.
422 * @c: UBIFS file-system description object
423 * @buf: buffer to check
424 * @len: length of buffer
425 * @lnum: LEB number of the LEB from which @buf was read
426 * @offs: offset from which @buf was read
428 * This function scans @buf for more nodes and returns %0 is a node is found and
429 * %1 if no more nodes are found.
431 static int no_more_nodes(const struct ubifs_info *c, void *buf, int len,
434 int skip, next_offs = 0;
436 if (len > UBIFS_DATA_NODE_SZ) {
437 struct ubifs_ch *ch = buf;
438 int dlen = le32_to_cpu(ch->len);
440 if (ch->node_type == UBIFS_DATA_NODE && dlen >= UBIFS_CH_SZ &&
441 dlen <= UBIFS_MAX_DATA_NODE_SZ)
442 /* The corrupt node looks like a data node */
443 next_offs = ALIGN(offs + dlen, 8);
446 if (c->min_io_size == 1)
449 skip = ALIGN(offs + 1, c->min_io_size) - offs;
455 struct ubifs_ch *ch = buf;
456 uint32_t magic = le32_to_cpu(ch->magic);
459 if (magic == UBIFS_NODE_MAGIC) {
460 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
461 if (ret == SCANNED_A_NODE || ret > 0) {
463 * There is a small chance this is just data in
464 * a data node, so check that possibility. e.g.
465 * this is part of a file that itself contains
468 if (next_offs && offs + le32_to_cpu(ch->len) <=
471 dbg_rcvry("unexpected node at %d:%d", lnum,
484 * fix_unclean_leb - fix an unclean LEB.
485 * @c: UBIFS file-system description object
486 * @sleb: scanned LEB information
487 * @start: offset where scan started
489 static int fix_unclean_leb(struct ubifs_info *c, struct ubifs_scan_leb *sleb,
492 int lnum = sleb->lnum, endpt = start;
494 /* Get the end offset of the last node we are keeping */
495 if (!list_empty(&sleb->nodes)) {
496 struct ubifs_scan_node *snod;
498 snod = list_entry(sleb->nodes.prev,
499 struct ubifs_scan_node, list);
500 endpt = snod->offs + snod->len;
503 if ((c->vfs_sb->s_flags & MS_RDONLY) && !c->remounting_rw) {
504 /* Add to recovery list */
505 struct ubifs_unclean_leb *ucleb;
507 dbg_rcvry("need to fix LEB %d start %d endpt %d",
508 lnum, start, sleb->endpt);
509 ucleb = kzalloc(sizeof(struct ubifs_unclean_leb), GFP_NOFS);
513 ucleb->endpt = endpt;
514 list_add_tail(&ucleb->list, &c->unclean_leb_list);
516 /* Write the fixed LEB back to flash */
519 dbg_rcvry("fixing LEB %d start %d endpt %d",
520 lnum, start, sleb->endpt);
522 err = ubifs_leb_unmap(c, lnum);
526 int len = ALIGN(endpt, c->min_io_size);
529 err = ubi_read(c->ubi, lnum, sleb->buf, 0,
534 /* Pad to min_io_size */
536 int pad_len = len - ALIGN(endpt, 8);
539 void *buf = sleb->buf + len - pad_len;
541 ubifs_pad(c, buf, pad_len);
544 err = ubi_leb_change(c->ubi, lnum, sleb->buf, len,
554 * drop_incomplete_group - drop nodes from an incomplete group.
555 * @sleb: scanned LEB information
556 * @offs: offset of dropped nodes is returned here
558 * This function returns %1 if nodes are dropped and %0 otherwise.
560 static int drop_incomplete_group(struct ubifs_scan_leb *sleb, int *offs)
564 while (!list_empty(&sleb->nodes)) {
565 struct ubifs_scan_node *snod;
568 snod = list_entry(sleb->nodes.prev, struct ubifs_scan_node,
571 if (ch->group_type != UBIFS_IN_NODE_GROUP)
573 dbg_rcvry("dropping node at %d:%d", sleb->lnum, snod->offs);
575 list_del(&snod->list);
577 sleb->nodes_cnt -= 1;
584 * ubifs_recover_leb - scan and recover a LEB.
585 * @c: UBIFS file-system description object
588 * @sbuf: LEB-sized buffer to use
589 * @grouped: nodes may be grouped for recovery
591 * This function does a scan of a LEB, but caters for errors that might have
592 * been caused by the unclean unmount from which we are attempting to recover.
594 * This function returns %0 on success and a negative error code on failure.
596 struct ubifs_scan_leb *ubifs_recover_leb(struct ubifs_info *c, int lnum,
597 int offs, void *sbuf, int grouped)
599 int err, len = c->leb_size - offs, need_clean = 0, quiet = 1;
600 int empty_chkd = 0, start = offs;
601 struct ubifs_scan_leb *sleb;
602 void *buf = sbuf + offs;
604 dbg_rcvry("%d:%d", lnum, offs);
606 sleb = ubifs_start_scan(c, lnum, offs, sbuf);
616 dbg_scan("look at LEB %d:%d (%d bytes left)",
622 * Scan quietly until there is an error from which we cannot
625 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, quiet);
627 if (ret == SCANNED_A_NODE) {
628 /* A valid node, and not a padding node */
629 struct ubifs_ch *ch = buf;
632 err = ubifs_add_snod(c, sleb, buf, offs);
635 node_len = ALIGN(le32_to_cpu(ch->len), 8);
643 /* Padding bytes or a valid padding node */
650 if (ret == SCANNED_EMPTY_SPACE) {
651 if (!is_empty(buf, len)) {
652 if (!is_last_write(c, buf, offs))
654 clean_buf(c, &buf, lnum, &offs, &len);
661 if (ret == SCANNED_GARBAGE || ret == SCANNED_A_BAD_PAD_NODE)
662 if (is_last_write(c, buf, offs)) {
663 clean_buf(c, &buf, lnum, &offs, &len);
669 if (ret == SCANNED_A_CORRUPT_NODE)
670 if (no_more_nodes(c, buf, len, lnum, offs)) {
671 clean_buf(c, &buf, lnum, &offs, &len);
678 /* Redo the last scan but noisily */
684 case SCANNED_GARBAGE:
687 case SCANNED_A_CORRUPT_NODE:
688 case SCANNED_A_BAD_PAD_NODE:
697 if (!empty_chkd && !is_empty(buf, len)) {
698 if (is_last_write(c, buf, offs)) {
699 clean_buf(c, &buf, lnum, &offs, &len);
702 ubifs_err("corrupt empty space at LEB %d:%d",
708 /* Drop nodes from incomplete group */
709 if (grouped && drop_incomplete_group(sleb, &offs)) {
711 len = c->leb_size - offs;
712 clean_buf(c, &buf, lnum, &offs, &len);
716 if (offs % c->min_io_size) {
717 clean_buf(c, &buf, lnum, &offs, &len);
721 ubifs_end_scan(c, sleb, lnum, offs);
724 err = fix_unclean_leb(c, sleb, start);
732 ubifs_scanned_corruption(c, lnum, offs, buf);
735 ubifs_err("LEB %d scanning failed", lnum);
736 ubifs_scan_destroy(sleb);
741 * get_cs_sqnum - get commit start sequence number.
742 * @c: UBIFS file-system description object
743 * @lnum: LEB number of commit start node
744 * @offs: offset of commit start node
745 * @cs_sqnum: commit start sequence number is returned here
747 * This function returns %0 on success and a negative error code on failure.
749 static int get_cs_sqnum(struct ubifs_info *c, int lnum, int offs,
750 unsigned long long *cs_sqnum)
752 struct ubifs_cs_node *cs_node = NULL;
755 dbg_rcvry("at %d:%d", lnum, offs);
756 cs_node = kmalloc(UBIFS_CS_NODE_SZ, GFP_KERNEL);
759 if (c->leb_size - offs < UBIFS_CS_NODE_SZ)
761 err = ubi_read(c->ubi, lnum, (void *)cs_node, offs, UBIFS_CS_NODE_SZ);
762 if (err && err != -EBADMSG)
764 ret = ubifs_scan_a_node(c, cs_node, UBIFS_CS_NODE_SZ, lnum, offs, 0);
765 if (ret != SCANNED_A_NODE) {
766 dbg_err("Not a valid node");
769 if (cs_node->ch.node_type != UBIFS_CS_NODE) {
770 dbg_err("Node a CS node, type is %d", cs_node->ch.node_type);
773 if (le64_to_cpu(cs_node->cmt_no) != c->cmt_no) {
774 dbg_err("CS node cmt_no %llu != current cmt_no %llu",
775 (unsigned long long)le64_to_cpu(cs_node->cmt_no),
779 *cs_sqnum = le64_to_cpu(cs_node->ch.sqnum);
780 dbg_rcvry("commit start sqnum %llu", *cs_sqnum);
787 ubifs_err("failed to get CS sqnum");
793 * ubifs_recover_log_leb - scan and recover a log LEB.
794 * @c: UBIFS file-system description object
797 * @sbuf: LEB-sized buffer to use
799 * This function does a scan of a LEB, but caters for errors that might have
800 * been caused by the unclean unmount from which we are attempting to recover.
802 * This function returns %0 on success and a negative error code on failure.
804 struct ubifs_scan_leb *ubifs_recover_log_leb(struct ubifs_info *c, int lnum,
805 int offs, void *sbuf)
807 struct ubifs_scan_leb *sleb;
810 dbg_rcvry("LEB %d", lnum);
811 next_lnum = lnum + 1;
812 if (next_lnum >= UBIFS_LOG_LNUM + c->log_lebs)
813 next_lnum = UBIFS_LOG_LNUM;
814 if (next_lnum != c->ltail_lnum) {
816 * We can only recover at the end of the log, so check that the
817 * next log LEB is empty or out of date.
819 sleb = ubifs_scan(c, next_lnum, 0, sbuf);
822 if (sleb->nodes_cnt) {
823 struct ubifs_scan_node *snod;
824 unsigned long long cs_sqnum = c->cs_sqnum;
826 snod = list_entry(sleb->nodes.next,
827 struct ubifs_scan_node, list);
831 err = get_cs_sqnum(c, lnum, offs, &cs_sqnum);
833 ubifs_scan_destroy(sleb);
837 if (snod->sqnum > cs_sqnum) {
838 ubifs_err("unrecoverable log corruption "
840 ubifs_scan_destroy(sleb);
841 return ERR_PTR(-EUCLEAN);
844 ubifs_scan_destroy(sleb);
846 return ubifs_recover_leb(c, lnum, offs, sbuf, 0);
850 * recover_head - recover a head.
851 * @c: UBIFS file-system description object
852 * @lnum: LEB number of head to recover
853 * @offs: offset of head to recover
854 * @sbuf: LEB-sized buffer to use
856 * This function ensures that there is no data on the flash at a head location.
858 * This function returns %0 on success and a negative error code on failure.
860 static int recover_head(const struct ubifs_info *c, int lnum, int offs,
863 int len, err, need_clean = 0;
865 if (c->min_io_size > 1)
866 len = c->min_io_size;
869 if (offs + len > c->leb_size)
870 len = c->leb_size - offs;
875 /* Read at the head location and check it is empty flash */
876 err = ubi_read(c->ubi, lnum, sbuf, offs, len);
890 dbg_rcvry("cleaning head at %d:%d", lnum, offs);
892 return ubifs_leb_unmap(c, lnum);
893 err = ubi_read(c->ubi, lnum, sbuf, 0, offs);
896 return ubi_leb_change(c->ubi, lnum, sbuf, offs, UBI_UNKNOWN);
903 * ubifs_recover_inl_heads - recover index and LPT heads.
904 * @c: UBIFS file-system description object
905 * @sbuf: LEB-sized buffer to use
907 * This function ensures that there is no data on the flash at the index and
908 * LPT head locations.
910 * This deals with the recovery of a half-completed journal commit. UBIFS is
911 * careful never to overwrite the last version of the index or the LPT. Because
912 * the index and LPT are wandering trees, data from a half-completed commit will
913 * not be referenced anywhere in UBIFS. The data will be either in LEBs that are
914 * assumed to be empty and will be unmapped anyway before use, or in the index
917 * This function returns %0 on success and a negative error code on failure.
919 int ubifs_recover_inl_heads(const struct ubifs_info *c, void *sbuf)
923 ubifs_assert(!(c->vfs_sb->s_flags & MS_RDONLY) || c->remounting_rw);
925 dbg_rcvry("checking index head at %d:%d", c->ihead_lnum, c->ihead_offs);
926 err = recover_head(c, c->ihead_lnum, c->ihead_offs, sbuf);
930 dbg_rcvry("checking LPT head at %d:%d", c->nhead_lnum, c->nhead_offs);
931 err = recover_head(c, c->nhead_lnum, c->nhead_offs, sbuf);
939 * clean_an_unclean_leb - read and write a LEB to remove corruption.
940 * @c: UBIFS file-system description object
941 * @ucleb: unclean LEB information
942 * @sbuf: LEB-sized buffer to use
944 * This function reads a LEB up to a point pre-determined by the mount recovery,
945 * checks the nodes, and writes the result back to the flash, thereby cleaning
946 * off any following corruption, or non-fatal ECC errors.
948 * This function returns %0 on success and a negative error code on failure.
950 static int clean_an_unclean_leb(const struct ubifs_info *c,
951 struct ubifs_unclean_leb *ucleb, void *sbuf)
953 int err, lnum = ucleb->lnum, offs = 0, len = ucleb->endpt, quiet = 1;
956 dbg_rcvry("LEB %d len %d", lnum, len);
959 /* Nothing to read, just unmap it */
960 err = ubifs_leb_unmap(c, lnum);
966 err = ubi_read(c->ubi, lnum, buf, offs, len);
967 if (err && err != -EBADMSG)
975 /* Scan quietly until there is an error */
976 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, quiet);
978 if (ret == SCANNED_A_NODE) {
979 /* A valid node, and not a padding node */
980 struct ubifs_ch *ch = buf;
983 node_len = ALIGN(le32_to_cpu(ch->len), 8);
991 /* Padding bytes or a valid padding node */
998 if (ret == SCANNED_EMPTY_SPACE) {
999 ubifs_err("unexpected empty space at %d:%d",
1005 /* Redo the last scan but noisily */
1010 ubifs_scanned_corruption(c, lnum, offs, buf);
1014 /* Pad to min_io_size */
1015 len = ALIGN(ucleb->endpt, c->min_io_size);
1016 if (len > ucleb->endpt) {
1017 int pad_len = len - ALIGN(ucleb->endpt, 8);
1020 buf = c->sbuf + len - pad_len;
1021 ubifs_pad(c, buf, pad_len);
1025 /* Write back the LEB atomically */
1026 err = ubi_leb_change(c->ubi, lnum, sbuf, len, UBI_UNKNOWN);
1030 dbg_rcvry("cleaned LEB %d", lnum);
1036 * ubifs_clean_lebs - clean LEBs recovered during read-only mount.
1037 * @c: UBIFS file-system description object
1038 * @sbuf: LEB-sized buffer to use
1040 * This function cleans a LEB identified during recovery that needs to be
1041 * written but was not because UBIFS was mounted read-only. This happens when
1042 * remounting to read-write mode.
1044 * This function returns %0 on success and a negative error code on failure.
1046 int ubifs_clean_lebs(const struct ubifs_info *c, void *sbuf)
1048 dbg_rcvry("recovery");
1049 while (!list_empty(&c->unclean_leb_list)) {
1050 struct ubifs_unclean_leb *ucleb;
1053 ucleb = list_entry(c->unclean_leb_list.next,
1054 struct ubifs_unclean_leb, list);
1055 err = clean_an_unclean_leb(c, ucleb, sbuf);
1058 list_del(&ucleb->list);
1065 * ubifs_rcvry_gc_commit - recover the GC LEB number and run the commit.
1066 * @c: UBIFS file-system description object
1068 * Out-of-place garbage collection requires always one empty LEB with which to
1069 * start garbage collection. The LEB number is recorded in c->gc_lnum and is
1070 * written to the master node on unmounting. In the case of an unclean unmount
1071 * the value of gc_lnum recorded in the master node is out of date and cannot
1072 * be used. Instead, recovery must allocate an empty LEB for this purpose.
1073 * However, there may not be enough empty space, in which case it must be
1074 * possible to GC the dirtiest LEB into the GC head LEB.
1076 * This function also runs the commit which causes the TNC updates from
1077 * size-recovery and orphans to be written to the flash. That is important to
1078 * ensure correct replay order for subsequent mounts.
1080 * This function returns %0 on success and a negative error code on failure.
1082 int ubifs_rcvry_gc_commit(struct ubifs_info *c)
1084 struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf;
1085 struct ubifs_lprops lp;
1089 if (wbuf->lnum == -1) {
1090 dbg_rcvry("no GC head LEB");
1094 * See whether the used space in the dirtiest LEB fits in the GC head
1097 if (wbuf->offs == c->leb_size) {
1098 dbg_rcvry("no room in GC head LEB");
1101 err = ubifs_find_dirty_leb(c, &lp, wbuf->offs, 2);
1104 dbg_err("could not find a dirty LEB");
1107 ubifs_assert(!(lp.flags & LPROPS_INDEX));
1109 if (lp.free + lp.dirty == c->leb_size) {
1110 /* An empty LEB was returned */
1111 if (lp.free != c->leb_size) {
1112 err = ubifs_change_one_lp(c, lnum, c->leb_size,
1117 err = ubifs_leb_unmap(c, lnum);
1121 dbg_rcvry("allocated LEB %d for GC", lnum);
1122 /* Run the commit */
1123 dbg_rcvry("committing");
1124 return ubifs_run_commit(c);
1127 * There was no empty LEB so the used space in the dirtiest LEB must fit
1128 * in the GC head LEB.
1130 if (lp.free + lp.dirty < wbuf->offs) {
1131 dbg_rcvry("LEB %d doesn't fit in GC head LEB %d:%d",
1132 lnum, wbuf->lnum, wbuf->offs);
1133 err = ubifs_return_leb(c, lnum);
1139 * We run the commit before garbage collection otherwise subsequent
1140 * mounts will see the GC and orphan deletion in a different order.
1142 dbg_rcvry("committing");
1143 err = ubifs_run_commit(c);
1147 * The data in the dirtiest LEB fits in the GC head LEB, so do the GC
1148 * - use locking to keep 'ubifs_assert()' happy.
1150 dbg_rcvry("GC'ing LEB %d", lnum);
1151 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1152 err = ubifs_garbage_collect_leb(c, &lp);
1154 int err2 = ubifs_wbuf_sync_nolock(wbuf);
1159 mutex_unlock(&wbuf->io_mutex);
1161 dbg_err("GC failed, error %d", err);
1166 if (err != LEB_RETAINED) {
1167 dbg_err("GC returned %d", err);
1170 err = ubifs_leb_unmap(c, c->gc_lnum);
1173 dbg_rcvry("allocated LEB %d for GC", lnum);
1178 * There is no GC head LEB or the free space in the GC head LEB is too
1179 * small. Allocate gc_lnum by calling 'ubifs_find_free_leb_for_idx()' so
1182 lnum = ubifs_find_free_leb_for_idx(c);
1184 dbg_err("could not find an empty LEB");
1187 /* And reset the index flag */
1188 err = ubifs_change_one_lp(c, lnum, LPROPS_NC, LPROPS_NC, 0,
1193 dbg_rcvry("allocated LEB %d for GC", lnum);
1194 /* Run the commit */
1195 dbg_rcvry("committing");
1196 return ubifs_run_commit(c);
1200 * struct size_entry - inode size information for recovery.
1201 * @rb: link in the RB-tree of sizes
1202 * @inum: inode number
1203 * @i_size: size on inode
1204 * @d_size: maximum size based on data nodes
1205 * @exists: indicates whether the inode exists
1206 * @inode: inode if pinned in memory awaiting rw mode to fix it
1214 struct inode *inode;
1218 * add_ino - add an entry to the size tree.
1219 * @c: UBIFS file-system description object
1220 * @inum: inode number
1221 * @i_size: size on inode
1222 * @d_size: maximum size based on data nodes
1223 * @exists: indicates whether the inode exists
1225 static int add_ino(struct ubifs_info *c, ino_t inum, loff_t i_size,
1226 loff_t d_size, int exists)
1228 struct rb_node **p = &c->size_tree.rb_node, *parent = NULL;
1229 struct size_entry *e;
1233 e = rb_entry(parent, struct size_entry, rb);
1237 p = &(*p)->rb_right;
1240 e = kzalloc(sizeof(struct size_entry), GFP_KERNEL);
1249 rb_link_node(&e->rb, parent, p);
1250 rb_insert_color(&e->rb, &c->size_tree);
1256 * find_ino - find an entry on the size tree.
1257 * @c: UBIFS file-system description object
1258 * @inum: inode number
1260 static struct size_entry *find_ino(struct ubifs_info *c, ino_t inum)
1262 struct rb_node *p = c->size_tree.rb_node;
1263 struct size_entry *e;
1266 e = rb_entry(p, struct size_entry, rb);
1269 else if (inum > e->inum)
1278 * remove_ino - remove an entry from the size tree.
1279 * @c: UBIFS file-system description object
1280 * @inum: inode number
1282 static void remove_ino(struct ubifs_info *c, ino_t inum)
1284 struct size_entry *e = find_ino(c, inum);
1288 rb_erase(&e->rb, &c->size_tree);
1293 * ubifs_destroy_size_tree - free resources related to the size tree.
1294 * @c: UBIFS file-system description object
1296 void ubifs_destroy_size_tree(struct ubifs_info *c)
1298 struct rb_node *this = c->size_tree.rb_node;
1299 struct size_entry *e;
1302 if (this->rb_left) {
1303 this = this->rb_left;
1305 } else if (this->rb_right) {
1306 this = this->rb_right;
1309 e = rb_entry(this, struct size_entry, rb);
1312 this = rb_parent(this);
1314 if (this->rb_left == &e->rb)
1315 this->rb_left = NULL;
1317 this->rb_right = NULL;
1321 c->size_tree = RB_ROOT;
1325 * ubifs_recover_size_accum - accumulate inode sizes for recovery.
1326 * @c: UBIFS file-system description object
1328 * @deletion: node is for a deletion
1329 * @new_size: inode size
1331 * This function has two purposes:
1332 * 1) to ensure there are no data nodes that fall outside the inode size
1333 * 2) to ensure there are no data nodes for inodes that do not exist
1334 * To accomplish those purposes, a rb-tree is constructed containing an entry
1335 * for each inode number in the journal that has not been deleted, and recording
1336 * the size from the inode node, the maximum size of any data node (also altered
1337 * by truncations) and a flag indicating a inode number for which no inode node
1338 * was present in the journal.
1340 * Note that there is still the possibility that there are data nodes that have
1341 * been committed that are beyond the inode size, however the only way to find
1342 * them would be to scan the entire index. Alternatively, some provision could
1343 * be made to record the size of inodes at the start of commit, which would seem
1344 * very cumbersome for a scenario that is quite unlikely and the only negative
1345 * consequence of which is wasted space.
1347 * This functions returns %0 on success and a negative error code on failure.
1349 int ubifs_recover_size_accum(struct ubifs_info *c, union ubifs_key *key,
1350 int deletion, loff_t new_size)
1352 ino_t inum = key_inum(c, key);
1353 struct size_entry *e;
1356 switch (key_type(c, key)) {
1359 remove_ino(c, inum);
1361 e = find_ino(c, inum);
1363 e->i_size = new_size;
1366 err = add_ino(c, inum, new_size, 0, 1);
1372 case UBIFS_DATA_KEY:
1373 e = find_ino(c, inum);
1375 if (new_size > e->d_size)
1376 e->d_size = new_size;
1378 err = add_ino(c, inum, 0, new_size, 0);
1383 case UBIFS_TRUN_KEY:
1384 e = find_ino(c, inum);
1386 e->d_size = new_size;
1393 * fix_size_in_place - fix inode size in place on flash.
1394 * @c: UBIFS file-system description object
1395 * @e: inode size information for recovery
1397 static int fix_size_in_place(struct ubifs_info *c, struct size_entry *e)
1399 struct ubifs_ino_node *ino = c->sbuf;
1401 union ubifs_key key;
1402 int err, lnum, offs, len;
1406 /* Locate the inode node LEB number and offset */
1407 ino_key_init(c, &key, e->inum);
1408 err = ubifs_tnc_locate(c, &key, ino, &lnum, &offs);
1412 * If the size recorded on the inode node is greater than the size that
1413 * was calculated from nodes in the journal then don't change the inode.
1415 i_size = le64_to_cpu(ino->size);
1416 if (i_size >= e->d_size)
1419 err = ubi_read(c->ubi, lnum, c->sbuf, 0, c->leb_size);
1422 /* Change the size field and recalculate the CRC */
1423 ino = c->sbuf + offs;
1424 ino->size = cpu_to_le64(e->d_size);
1425 len = le32_to_cpu(ino->ch.len);
1426 crc = crc32(UBIFS_CRC32_INIT, (void *)ino + 8, len - 8);
1427 ino->ch.crc = cpu_to_le32(crc);
1428 /* Work out where data in the LEB ends and free space begins */
1430 len = c->leb_size - 1;
1431 while (p[len] == 0xff)
1433 len = ALIGN(len + 1, c->min_io_size);
1434 /* Atomically write the fixed LEB back again */
1435 err = ubi_leb_change(c->ubi, lnum, c->sbuf, len, UBI_UNKNOWN);
1438 dbg_rcvry("inode %lu at %d:%d size %lld -> %lld ", e->inum, lnum, offs,
1443 ubifs_warn("inode %lu failed to fix size %lld -> %lld error %d",
1444 e->inum, e->i_size, e->d_size, err);
1449 * ubifs_recover_size - recover inode size.
1450 * @c: UBIFS file-system description object
1452 * This function attempts to fix inode size discrepancies identified by the
1453 * 'ubifs_recover_size_accum()' function.
1455 * This functions returns %0 on success and a negative error code on failure.
1457 int ubifs_recover_size(struct ubifs_info *c)
1459 struct rb_node *this = rb_first(&c->size_tree);
1462 struct size_entry *e;
1465 e = rb_entry(this, struct size_entry, rb);
1467 union ubifs_key key;
1469 ino_key_init(c, &key, e->inum);
1470 err = ubifs_tnc_lookup(c, &key, c->sbuf);
1471 if (err && err != -ENOENT)
1473 if (err == -ENOENT) {
1474 /* Remove data nodes that have no inode */
1475 dbg_rcvry("removing ino %lu", e->inum);
1476 err = ubifs_tnc_remove_ino(c, e->inum);
1480 struct ubifs_ino_node *ino = c->sbuf;
1483 e->i_size = le64_to_cpu(ino->size);
1486 if (e->exists && e->i_size < e->d_size) {
1487 if (!e->inode && (c->vfs_sb->s_flags & MS_RDONLY)) {
1488 /* Fix the inode size and pin it in memory */
1489 struct inode *inode;
1491 inode = ubifs_iget(c->vfs_sb, e->inum);
1493 return PTR_ERR(inode);
1494 if (inode->i_size < e->d_size) {
1495 dbg_rcvry("ino %lu size %lld -> %lld",
1498 inode->i_size = e->d_size;
1499 ubifs_inode(inode)->ui_size = e->d_size;
1501 this = rb_next(this);
1506 /* Fix the size in place */
1507 err = fix_size_in_place(c, e);
1514 this = rb_next(this);
1515 rb_erase(&e->rb, &c->size_tree);