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 * 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>
34 #include <linux/slab.h>
38 * is_empty - determine whether a buffer is empty (contains all 0xff).
39 * @buf: buffer to clean
40 * @len: length of buffer
42 * This function returns %1 if the buffer is empty (contains all 0xff) otherwise
45 static int is_empty(void *buf, int len)
50 for (i = 0; i < len; i++)
57 * first_non_ff - find offset of the first non-0xff byte.
58 * @buf: buffer to search in
59 * @len: length of buffer
61 * This function returns offset of the first non-0xff byte in @buf or %-1 if
62 * the buffer contains only 0xff bytes.
64 static int first_non_ff(void *buf, int len)
69 for (i = 0; i < len; i++)
76 * get_master_node - get the last valid master node allowing for corruption.
77 * @c: UBIFS file-system description object
79 * @pbuf: buffer containing the LEB read, is returned here
80 * @mst: master node, if found, is returned here
81 * @cor: corruption, if found, is returned here
83 * This function allocates a buffer, reads the LEB into it, and finds and
84 * returns the last valid master node allowing for one area of corruption.
85 * The corrupt area, if there is one, must be consistent with the assumption
86 * that it is the result of an unclean unmount while the master node was being
87 * written. Under those circumstances, it is valid to use the previously written
90 * This function returns %0 on success and a negative error code on failure.
92 static int get_master_node(const struct ubifs_info *c, int lnum, void **pbuf,
93 struct ubifs_mst_node **mst, void **cor)
95 const int sz = c->mst_node_alsz;
99 sbuf = vmalloc(c->leb_size);
103 err = ubi_read(c->ubi, lnum, sbuf, 0, c->leb_size);
104 if (err && err != -EBADMSG)
107 /* Find the first position that is definitely not a node */
111 while (offs + UBIFS_MST_NODE_SZ <= c->leb_size) {
112 struct ubifs_ch *ch = buf;
114 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
120 /* See if there was a valid master node before that */
127 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
128 if (ret != SCANNED_A_NODE && offs) {
129 /* Could have been corruption so check one place back */
133 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
134 if (ret != SCANNED_A_NODE)
136 * We accept only one area of corruption because
137 * we are assuming that it was caused while
138 * trying to write a master node.
142 if (ret == SCANNED_A_NODE) {
143 struct ubifs_ch *ch = buf;
145 if (ch->node_type != UBIFS_MST_NODE)
147 dbg_rcvry("found a master node at %d:%d", lnum, offs);
154 /* Check for corruption */
155 if (offs < c->leb_size) {
156 if (!is_empty(buf, min_t(int, len, sz))) {
158 dbg_rcvry("found corruption at %d:%d", lnum, offs);
164 /* Check remaining empty space */
165 if (offs < c->leb_size)
166 if (!is_empty(buf, len))
181 * write_rcvrd_mst_node - write recovered master node.
182 * @c: UBIFS file-system description object
185 * This function returns %0 on success and a negative error code on failure.
187 static int write_rcvrd_mst_node(struct ubifs_info *c,
188 struct ubifs_mst_node *mst)
190 int err = 0, lnum = UBIFS_MST_LNUM, sz = c->mst_node_alsz;
193 dbg_rcvry("recovery");
195 save_flags = mst->flags;
196 mst->flags |= cpu_to_le32(UBIFS_MST_RCVRY);
198 ubifs_prepare_node(c, mst, UBIFS_MST_NODE_SZ, 1);
199 err = ubi_leb_change(c->ubi, lnum, mst, sz, UBI_SHORTTERM);
202 err = ubi_leb_change(c->ubi, lnum + 1, mst, sz, UBI_SHORTTERM);
206 mst->flags = save_flags;
211 * ubifs_recover_master_node - recover the master node.
212 * @c: UBIFS file-system description object
214 * This function recovers the master node from corruption that may occur due to
215 * an unclean unmount.
217 * This function returns %0 on success and a negative error code on failure.
219 int ubifs_recover_master_node(struct ubifs_info *c)
221 void *buf1 = NULL, *buf2 = NULL, *cor1 = NULL, *cor2 = NULL;
222 struct ubifs_mst_node *mst1 = NULL, *mst2 = NULL, *mst;
223 const int sz = c->mst_node_alsz;
224 int err, offs1, offs2;
226 dbg_rcvry("recovery");
228 err = get_master_node(c, UBIFS_MST_LNUM, &buf1, &mst1, &cor1);
232 err = get_master_node(c, UBIFS_MST_LNUM + 1, &buf2, &mst2, &cor2);
237 offs1 = (void *)mst1 - buf1;
238 if ((le32_to_cpu(mst1->flags) & UBIFS_MST_RCVRY) &&
239 (offs1 == 0 && !cor1)) {
241 * mst1 was written by recovery at offset 0 with no
244 dbg_rcvry("recovery recovery");
247 offs2 = (void *)mst2 - buf2;
248 if (offs1 == offs2) {
249 /* Same offset, so must be the same */
250 if (memcmp((void *)mst1 + UBIFS_CH_SZ,
251 (void *)mst2 + UBIFS_CH_SZ,
252 UBIFS_MST_NODE_SZ - UBIFS_CH_SZ))
255 } else if (offs2 + sz == offs1) {
256 /* 1st LEB was written, 2nd was not */
260 } else if (offs1 == 0 && offs2 + sz >= c->leb_size) {
261 /* 1st LEB was unmapped and written, 2nd not */
269 * 2nd LEB was unmapped and about to be written, so
270 * there must be only one master node in the first LEB
273 if (offs1 != 0 || cor1)
281 * 1st LEB was unmapped and about to be written, so there must
282 * be no room left in 2nd LEB.
284 offs2 = (void *)mst2 - buf2;
285 if (offs2 + sz + sz <= c->leb_size)
290 ubifs_msg("recovered master node from LEB %d",
291 (mst == mst1 ? UBIFS_MST_LNUM : UBIFS_MST_LNUM + 1));
293 memcpy(c->mst_node, mst, UBIFS_MST_NODE_SZ);
295 if ((c->vfs_sb->s_flags & MS_RDONLY)) {
296 /* Read-only mode. Keep a copy for switching to rw mode */
297 c->rcvrd_mst_node = kmalloc(sz, GFP_KERNEL);
298 if (!c->rcvrd_mst_node) {
302 memcpy(c->rcvrd_mst_node, c->mst_node, UBIFS_MST_NODE_SZ);
304 /* Write the recovered master node */
305 c->max_sqnum = le64_to_cpu(mst->ch.sqnum) - 1;
306 err = write_rcvrd_mst_node(c, c->mst_node);
319 ubifs_err("failed to recover master node");
321 dbg_err("dumping first master node");
322 dbg_dump_node(c, mst1);
325 dbg_err("dumping second master node");
326 dbg_dump_node(c, mst2);
334 * ubifs_write_rcvrd_mst_node - write the recovered master node.
335 * @c: UBIFS file-system description object
337 * This function writes the master node that was recovered during mounting in
338 * read-only mode and must now be written because we are remounting rw.
340 * This function returns %0 on success and a negative error code on failure.
342 int ubifs_write_rcvrd_mst_node(struct ubifs_info *c)
346 if (!c->rcvrd_mst_node)
348 c->rcvrd_mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
349 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
350 err = write_rcvrd_mst_node(c, c->rcvrd_mst_node);
353 kfree(c->rcvrd_mst_node);
354 c->rcvrd_mst_node = NULL;
359 * is_last_write - determine if an offset was in the last write to a LEB.
360 * @c: UBIFS file-system description object
361 * @buf: buffer to check
362 * @offs: offset to check
364 * This function returns %1 if @offs was in the last write to the LEB whose data
365 * is in @buf, otherwise %0 is returned. The determination is made by checking
366 * for subsequent empty space starting from the next @c->min_io_size boundary.
368 static int is_last_write(const struct ubifs_info *c, void *buf, int offs)
370 int empty_offs, check_len;
374 * Round up to the next @c->min_io_size boundary i.e. @offs is in the
375 * last wbuf written. After that should be empty space.
377 empty_offs = ALIGN(offs + 1, c->min_io_size);
378 check_len = c->leb_size - empty_offs;
379 p = buf + empty_offs - offs;
380 return is_empty(p, check_len);
384 * clean_buf - clean the data from an LEB sitting in a buffer.
385 * @c: UBIFS file-system description object
386 * @buf: buffer to clean
387 * @lnum: LEB number to clean
388 * @offs: offset from which to clean
389 * @len: length of buffer
391 * This function pads up to the next min_io_size boundary (if there is one) and
392 * sets empty space to all 0xff. @buf, @offs and @len are updated to the next
393 * @c->min_io_size boundary.
395 static void clean_buf(const struct ubifs_info *c, void **buf, int lnum,
398 int empty_offs, pad_len;
401 dbg_rcvry("cleaning corruption at %d:%d", lnum, *offs);
403 ubifs_assert(!(*offs & 7));
404 empty_offs = ALIGN(*offs, c->min_io_size);
405 pad_len = empty_offs - *offs;
406 ubifs_pad(c, *buf, pad_len);
410 memset(*buf, 0xff, c->leb_size - empty_offs);
414 * no_more_nodes - determine if there are no more nodes in a buffer.
415 * @c: UBIFS file-system description object
416 * @buf: buffer to check
417 * @len: length of buffer
418 * @lnum: LEB number of the LEB from which @buf was read
419 * @offs: offset from which @buf was read
421 * This function ensures that the corrupted node at @offs is the last thing
422 * written to a LEB. This function returns %1 if more data is not found and
423 * %0 if more data is found.
425 static int no_more_nodes(const struct ubifs_info *c, void *buf, int len,
428 struct ubifs_ch *ch = buf;
429 int skip, dlen = le32_to_cpu(ch->len);
431 /* Check for empty space after the corrupt node's common header */
432 skip = ALIGN(offs + UBIFS_CH_SZ, c->min_io_size) - offs;
433 if (is_empty(buf + skip, len - skip))
436 * The area after the common header size is not empty, so the common
437 * header must be intact. Check it.
439 if (ubifs_check_node(c, buf, lnum, offs, 1, 0) != -EUCLEAN) {
440 dbg_rcvry("unexpected bad common header at %d:%d", lnum, offs);
443 /* Now we know the corrupt node's length we can skip over it */
444 skip = ALIGN(offs + dlen, c->min_io_size) - offs;
445 /* After which there should be empty space */
446 if (is_empty(buf + skip, len - skip))
448 dbg_rcvry("unexpected data at %d:%d", lnum, offs + skip);
453 * fix_unclean_leb - fix an unclean LEB.
454 * @c: UBIFS file-system description object
455 * @sleb: scanned LEB information
456 * @start: offset where scan started
458 static int fix_unclean_leb(struct ubifs_info *c, struct ubifs_scan_leb *sleb,
461 int lnum = sleb->lnum, endpt = start;
463 /* Get the end offset of the last node we are keeping */
464 if (!list_empty(&sleb->nodes)) {
465 struct ubifs_scan_node *snod;
467 snod = list_entry(sleb->nodes.prev,
468 struct ubifs_scan_node, list);
469 endpt = snod->offs + snod->len;
472 if ((c->vfs_sb->s_flags & MS_RDONLY) && !c->remounting_rw) {
473 /* Add to recovery list */
474 struct ubifs_unclean_leb *ucleb;
476 dbg_rcvry("need to fix LEB %d start %d endpt %d",
477 lnum, start, sleb->endpt);
478 ucleb = kzalloc(sizeof(struct ubifs_unclean_leb), GFP_NOFS);
482 ucleb->endpt = endpt;
483 list_add_tail(&ucleb->list, &c->unclean_leb_list);
485 /* Write the fixed LEB back to flash */
488 dbg_rcvry("fixing LEB %d start %d endpt %d",
489 lnum, start, sleb->endpt);
491 err = ubifs_leb_unmap(c, lnum);
495 int len = ALIGN(endpt, c->min_io_size);
498 err = ubi_read(c->ubi, lnum, sleb->buf, 0,
503 /* Pad to min_io_size */
505 int pad_len = len - ALIGN(endpt, 8);
508 void *buf = sleb->buf + len - pad_len;
510 ubifs_pad(c, buf, pad_len);
513 err = ubi_leb_change(c->ubi, lnum, sleb->buf, len,
523 * drop_incomplete_group - drop nodes from an incomplete group.
524 * @sleb: scanned LEB information
525 * @offs: offset of dropped nodes is returned here
527 * This function returns %1 if nodes are dropped and %0 otherwise.
529 static int drop_incomplete_group(struct ubifs_scan_leb *sleb, int *offs)
533 while (!list_empty(&sleb->nodes)) {
534 struct ubifs_scan_node *snod;
537 snod = list_entry(sleb->nodes.prev, struct ubifs_scan_node,
540 if (ch->group_type != UBIFS_IN_NODE_GROUP)
542 dbg_rcvry("dropping node at %d:%d", sleb->lnum, snod->offs);
544 list_del(&snod->list);
546 sleb->nodes_cnt -= 1;
553 * ubifs_recover_leb - scan and recover a LEB.
554 * @c: UBIFS file-system description object
557 * @sbuf: LEB-sized buffer to use
558 * @grouped: nodes may be grouped for recovery
560 * This function does a scan of a LEB, but caters for errors that might have
561 * been caused by the unclean unmount from which we are attempting to recover.
562 * Returns %0 in case of success, %-EUCLEAN if an unrecoverable corruption is
563 * found, and a negative error code in case of failure.
565 struct ubifs_scan_leb *ubifs_recover_leb(struct ubifs_info *c, int lnum,
566 int offs, void *sbuf, int grouped)
568 int err, len = c->leb_size - offs, need_clean = 0, quiet = 1;
569 int empty_chkd = 0, start = offs;
570 struct ubifs_scan_leb *sleb;
571 void *buf = sbuf + offs;
573 dbg_rcvry("%d:%d", lnum, offs);
575 sleb = ubifs_start_scan(c, lnum, offs, sbuf);
585 dbg_scan("look at LEB %d:%d (%d bytes left)",
591 * Scan quietly until there is an error from which we cannot
594 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, quiet);
596 if (ret == SCANNED_A_NODE) {
597 /* A valid node, and not a padding node */
598 struct ubifs_ch *ch = buf;
601 err = ubifs_add_snod(c, sleb, buf, offs);
604 node_len = ALIGN(le32_to_cpu(ch->len), 8);
612 /* Padding bytes or a valid padding node */
619 if (ret == SCANNED_EMPTY_SPACE) {
620 if (!is_empty(buf, len)) {
621 if (!is_last_write(c, buf, offs))
623 clean_buf(c, &buf, lnum, &offs, &len);
630 if (ret == SCANNED_GARBAGE || ret == SCANNED_A_BAD_PAD_NODE)
631 if (is_last_write(c, buf, offs)) {
632 clean_buf(c, &buf, lnum, &offs, &len);
638 if (ret == SCANNED_A_CORRUPT_NODE)
639 if (no_more_nodes(c, buf, len, lnum, offs)) {
640 clean_buf(c, &buf, lnum, &offs, &len);
647 /* Redo the last scan but noisily */
653 case SCANNED_GARBAGE:
656 case SCANNED_A_CORRUPT_NODE:
657 case SCANNED_A_BAD_PAD_NODE:
667 if (!empty_chkd && !is_empty(buf, len)) {
668 if (is_last_write(c, buf, offs)) {
669 clean_buf(c, &buf, lnum, &offs, &len);
672 int corruption = first_non_ff(buf, len);
674 ubifs_err("corrupt empty space LEB %d:%d, corruption "
675 "starts at %d", lnum, offs, corruption);
676 /* Make sure we dump interesting non-0xFF data */
683 /* Drop nodes from incomplete group */
684 if (grouped && drop_incomplete_group(sleb, &offs)) {
686 len = c->leb_size - offs;
687 clean_buf(c, &buf, lnum, &offs, &len);
691 if (offs % c->min_io_size) {
692 clean_buf(c, &buf, lnum, &offs, &len);
696 ubifs_end_scan(c, sleb, lnum, offs);
699 err = fix_unclean_leb(c, sleb, start);
707 ubifs_scanned_corruption(c, lnum, offs, buf);
710 ubifs_err("LEB %d scanning failed", lnum);
711 ubifs_scan_destroy(sleb);
716 * get_cs_sqnum - get commit start sequence number.
717 * @c: UBIFS file-system description object
718 * @lnum: LEB number of commit start node
719 * @offs: offset of commit start node
720 * @cs_sqnum: commit start sequence number is returned here
722 * This function returns %0 on success and a negative error code on failure.
724 static int get_cs_sqnum(struct ubifs_info *c, int lnum, int offs,
725 unsigned long long *cs_sqnum)
727 struct ubifs_cs_node *cs_node = NULL;
730 dbg_rcvry("at %d:%d", lnum, offs);
731 cs_node = kmalloc(UBIFS_CS_NODE_SZ, GFP_KERNEL);
734 if (c->leb_size - offs < UBIFS_CS_NODE_SZ)
736 err = ubi_read(c->ubi, lnum, (void *)cs_node, offs, UBIFS_CS_NODE_SZ);
737 if (err && err != -EBADMSG)
739 ret = ubifs_scan_a_node(c, cs_node, UBIFS_CS_NODE_SZ, lnum, offs, 0);
740 if (ret != SCANNED_A_NODE) {
741 dbg_err("Not a valid node");
744 if (cs_node->ch.node_type != UBIFS_CS_NODE) {
745 dbg_err("Node a CS node, type is %d", cs_node->ch.node_type);
748 if (le64_to_cpu(cs_node->cmt_no) != c->cmt_no) {
749 dbg_err("CS node cmt_no %llu != current cmt_no %llu",
750 (unsigned long long)le64_to_cpu(cs_node->cmt_no),
754 *cs_sqnum = le64_to_cpu(cs_node->ch.sqnum);
755 dbg_rcvry("commit start sqnum %llu", *cs_sqnum);
762 ubifs_err("failed to get CS sqnum");
768 * ubifs_recover_log_leb - scan and recover a log LEB.
769 * @c: UBIFS file-system description object
772 * @sbuf: LEB-sized buffer to use
774 * This function does a scan of a LEB, but caters for errors that might have
775 * been caused by the unclean unmount from which we are attempting to recover.
777 * This function returns %0 on success and a negative error code on failure.
779 struct ubifs_scan_leb *ubifs_recover_log_leb(struct ubifs_info *c, int lnum,
780 int offs, void *sbuf)
782 struct ubifs_scan_leb *sleb;
785 dbg_rcvry("LEB %d", lnum);
786 next_lnum = lnum + 1;
787 if (next_lnum >= UBIFS_LOG_LNUM + c->log_lebs)
788 next_lnum = UBIFS_LOG_LNUM;
789 if (next_lnum != c->ltail_lnum) {
791 * We can only recover at the end of the log, so check that the
792 * next log LEB is empty or out of date.
794 sleb = ubifs_scan(c, next_lnum, 0, sbuf, 0);
797 if (sleb->nodes_cnt) {
798 struct ubifs_scan_node *snod;
799 unsigned long long cs_sqnum = c->cs_sqnum;
801 snod = list_entry(sleb->nodes.next,
802 struct ubifs_scan_node, list);
806 err = get_cs_sqnum(c, lnum, offs, &cs_sqnum);
808 ubifs_scan_destroy(sleb);
812 if (snod->sqnum > cs_sqnum) {
813 ubifs_err("unrecoverable log corruption "
815 ubifs_scan_destroy(sleb);
816 return ERR_PTR(-EUCLEAN);
819 ubifs_scan_destroy(sleb);
821 return ubifs_recover_leb(c, lnum, offs, sbuf, 0);
825 * recover_head - recover a head.
826 * @c: UBIFS file-system description object
827 * @lnum: LEB number of head to recover
828 * @offs: offset of head to recover
829 * @sbuf: LEB-sized buffer to use
831 * This function ensures that there is no data on the flash at a head location.
833 * This function returns %0 on success and a negative error code on failure.
835 static int recover_head(const struct ubifs_info *c, int lnum, int offs,
840 if (c->min_io_size > 1)
841 len = c->min_io_size;
844 if (offs + len > c->leb_size)
845 len = c->leb_size - offs;
850 /* Read at the head location and check it is empty flash */
851 err = ubi_read(c->ubi, lnum, sbuf, offs, len);
852 if (err || !is_empty(sbuf, len)) {
853 dbg_rcvry("cleaning head at %d:%d", lnum, offs);
855 return ubifs_leb_unmap(c, lnum);
856 err = ubi_read(c->ubi, lnum, sbuf, 0, offs);
859 return ubi_leb_change(c->ubi, lnum, sbuf, offs, UBI_UNKNOWN);
866 * ubifs_recover_inl_heads - recover index and LPT heads.
867 * @c: UBIFS file-system description object
868 * @sbuf: LEB-sized buffer to use
870 * This function ensures that there is no data on the flash at the index and
871 * LPT head locations.
873 * This deals with the recovery of a half-completed journal commit. UBIFS is
874 * careful never to overwrite the last version of the index or the LPT. Because
875 * the index and LPT are wandering trees, data from a half-completed commit will
876 * not be referenced anywhere in UBIFS. The data will be either in LEBs that are
877 * assumed to be empty and will be unmapped anyway before use, or in the index
880 * This function returns %0 on success and a negative error code on failure.
882 int ubifs_recover_inl_heads(const struct ubifs_info *c, void *sbuf)
886 ubifs_assert(!(c->vfs_sb->s_flags & MS_RDONLY) || c->remounting_rw);
888 dbg_rcvry("checking index head at %d:%d", c->ihead_lnum, c->ihead_offs);
889 err = recover_head(c, c->ihead_lnum, c->ihead_offs, sbuf);
893 dbg_rcvry("checking LPT head at %d:%d", c->nhead_lnum, c->nhead_offs);
894 err = recover_head(c, c->nhead_lnum, c->nhead_offs, sbuf);
902 * clean_an_unclean_leb - read and write a LEB to remove corruption.
903 * @c: UBIFS file-system description object
904 * @ucleb: unclean LEB information
905 * @sbuf: LEB-sized buffer to use
907 * This function reads a LEB up to a point pre-determined by the mount recovery,
908 * checks the nodes, and writes the result back to the flash, thereby cleaning
909 * off any following corruption, or non-fatal ECC errors.
911 * This function returns %0 on success and a negative error code on failure.
913 static int clean_an_unclean_leb(const struct ubifs_info *c,
914 struct ubifs_unclean_leb *ucleb, void *sbuf)
916 int err, lnum = ucleb->lnum, offs = 0, len = ucleb->endpt, quiet = 1;
919 dbg_rcvry("LEB %d len %d", lnum, len);
922 /* Nothing to read, just unmap it */
923 err = ubifs_leb_unmap(c, lnum);
929 err = ubi_read(c->ubi, lnum, buf, offs, len);
930 if (err && err != -EBADMSG)
938 /* Scan quietly until there is an error */
939 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, quiet);
941 if (ret == SCANNED_A_NODE) {
942 /* A valid node, and not a padding node */
943 struct ubifs_ch *ch = buf;
946 node_len = ALIGN(le32_to_cpu(ch->len), 8);
954 /* Padding bytes or a valid padding node */
961 if (ret == SCANNED_EMPTY_SPACE) {
962 ubifs_err("unexpected empty space at %d:%d",
968 /* Redo the last scan but noisily */
973 ubifs_scanned_corruption(c, lnum, offs, buf);
977 /* Pad to min_io_size */
978 len = ALIGN(ucleb->endpt, c->min_io_size);
979 if (len > ucleb->endpt) {
980 int pad_len = len - ALIGN(ucleb->endpt, 8);
983 buf = c->sbuf + len - pad_len;
984 ubifs_pad(c, buf, pad_len);
988 /* Write back the LEB atomically */
989 err = ubi_leb_change(c->ubi, lnum, sbuf, len, UBI_UNKNOWN);
993 dbg_rcvry("cleaned LEB %d", lnum);
999 * ubifs_clean_lebs - clean LEBs recovered during read-only mount.
1000 * @c: UBIFS file-system description object
1001 * @sbuf: LEB-sized buffer to use
1003 * This function cleans a LEB identified during recovery that needs to be
1004 * written but was not because UBIFS was mounted read-only. This happens when
1005 * remounting to read-write mode.
1007 * This function returns %0 on success and a negative error code on failure.
1009 int ubifs_clean_lebs(const struct ubifs_info *c, void *sbuf)
1011 dbg_rcvry("recovery");
1012 while (!list_empty(&c->unclean_leb_list)) {
1013 struct ubifs_unclean_leb *ucleb;
1016 ucleb = list_entry(c->unclean_leb_list.next,
1017 struct ubifs_unclean_leb, list);
1018 err = clean_an_unclean_leb(c, ucleb, sbuf);
1021 list_del(&ucleb->list);
1028 * ubifs_rcvry_gc_commit - recover the GC LEB number and run the commit.
1029 * @c: UBIFS file-system description object
1031 * Out-of-place garbage collection requires always one empty LEB with which to
1032 * start garbage collection. The LEB number is recorded in c->gc_lnum and is
1033 * written to the master node on unmounting. In the case of an unclean unmount
1034 * the value of gc_lnum recorded in the master node is out of date and cannot
1035 * be used. Instead, recovery must allocate an empty LEB for this purpose.
1036 * However, there may not be enough empty space, in which case it must be
1037 * possible to GC the dirtiest LEB into the GC head LEB.
1039 * This function also runs the commit which causes the TNC updates from
1040 * size-recovery and orphans to be written to the flash. That is important to
1041 * ensure correct replay order for subsequent mounts.
1043 * This function returns %0 on success and a negative error code on failure.
1045 int ubifs_rcvry_gc_commit(struct ubifs_info *c)
1047 struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf;
1048 struct ubifs_lprops lp;
1052 if (wbuf->lnum == -1) {
1053 dbg_rcvry("no GC head LEB");
1057 * See whether the used space in the dirtiest LEB fits in the GC head
1060 if (wbuf->offs == c->leb_size) {
1061 dbg_rcvry("no room in GC head LEB");
1064 err = ubifs_find_dirty_leb(c, &lp, wbuf->offs, 2);
1067 dbg_err("could not find a dirty LEB");
1070 ubifs_assert(!(lp.flags & LPROPS_INDEX));
1072 if (lp.free + lp.dirty == c->leb_size) {
1073 /* An empty LEB was returned */
1074 if (lp.free != c->leb_size) {
1075 err = ubifs_change_one_lp(c, lnum, c->leb_size,
1080 err = ubifs_leb_unmap(c, lnum);
1084 dbg_rcvry("allocated LEB %d for GC", lnum);
1085 /* Run the commit */
1086 dbg_rcvry("committing");
1087 return ubifs_run_commit(c);
1090 * There was no empty LEB so the used space in the dirtiest LEB must fit
1091 * in the GC head LEB.
1093 if (lp.free + lp.dirty < wbuf->offs) {
1094 dbg_rcvry("LEB %d doesn't fit in GC head LEB %d:%d",
1095 lnum, wbuf->lnum, wbuf->offs);
1096 err = ubifs_return_leb(c, lnum);
1102 * We run the commit before garbage collection otherwise subsequent
1103 * mounts will see the GC and orphan deletion in a different order.
1105 dbg_rcvry("committing");
1106 err = ubifs_run_commit(c);
1110 * The data in the dirtiest LEB fits in the GC head LEB, so do the GC
1111 * - use locking to keep 'ubifs_assert()' happy.
1113 dbg_rcvry("GC'ing LEB %d", lnum);
1114 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1115 err = ubifs_garbage_collect_leb(c, &lp);
1117 int err2 = ubifs_wbuf_sync_nolock(wbuf);
1122 mutex_unlock(&wbuf->io_mutex);
1124 dbg_err("GC failed, error %d", err);
1129 if (err != LEB_RETAINED) {
1130 dbg_err("GC returned %d", err);
1133 err = ubifs_leb_unmap(c, c->gc_lnum);
1136 dbg_rcvry("allocated LEB %d for GC", lnum);
1141 * There is no GC head LEB or the free space in the GC head LEB is too
1142 * small. Allocate gc_lnum by calling 'ubifs_find_free_leb_for_idx()' so
1145 lnum = ubifs_find_free_leb_for_idx(c);
1147 dbg_err("could not find an empty LEB");
1150 /* And reset the index flag */
1151 err = ubifs_change_one_lp(c, lnum, LPROPS_NC, LPROPS_NC, 0,
1156 dbg_rcvry("allocated LEB %d for GC", lnum);
1157 /* Run the commit */
1158 dbg_rcvry("committing");
1159 return ubifs_run_commit(c);
1163 * struct size_entry - inode size information for recovery.
1164 * @rb: link in the RB-tree of sizes
1165 * @inum: inode number
1166 * @i_size: size on inode
1167 * @d_size: maximum size based on data nodes
1168 * @exists: indicates whether the inode exists
1169 * @inode: inode if pinned in memory awaiting rw mode to fix it
1177 struct inode *inode;
1181 * add_ino - add an entry to the size tree.
1182 * @c: UBIFS file-system description object
1183 * @inum: inode number
1184 * @i_size: size on inode
1185 * @d_size: maximum size based on data nodes
1186 * @exists: indicates whether the inode exists
1188 static int add_ino(struct ubifs_info *c, ino_t inum, loff_t i_size,
1189 loff_t d_size, int exists)
1191 struct rb_node **p = &c->size_tree.rb_node, *parent = NULL;
1192 struct size_entry *e;
1196 e = rb_entry(parent, struct size_entry, rb);
1200 p = &(*p)->rb_right;
1203 e = kzalloc(sizeof(struct size_entry), GFP_KERNEL);
1212 rb_link_node(&e->rb, parent, p);
1213 rb_insert_color(&e->rb, &c->size_tree);
1219 * find_ino - find an entry on the size tree.
1220 * @c: UBIFS file-system description object
1221 * @inum: inode number
1223 static struct size_entry *find_ino(struct ubifs_info *c, ino_t inum)
1225 struct rb_node *p = c->size_tree.rb_node;
1226 struct size_entry *e;
1229 e = rb_entry(p, struct size_entry, rb);
1232 else if (inum > e->inum)
1241 * remove_ino - remove an entry from the size tree.
1242 * @c: UBIFS file-system description object
1243 * @inum: inode number
1245 static void remove_ino(struct ubifs_info *c, ino_t inum)
1247 struct size_entry *e = find_ino(c, inum);
1251 rb_erase(&e->rb, &c->size_tree);
1256 * ubifs_destroy_size_tree - free resources related to the size tree.
1257 * @c: UBIFS file-system description object
1259 void ubifs_destroy_size_tree(struct ubifs_info *c)
1261 struct rb_node *this = c->size_tree.rb_node;
1262 struct size_entry *e;
1265 if (this->rb_left) {
1266 this = this->rb_left;
1268 } else if (this->rb_right) {
1269 this = this->rb_right;
1272 e = rb_entry(this, struct size_entry, rb);
1275 this = rb_parent(this);
1277 if (this->rb_left == &e->rb)
1278 this->rb_left = NULL;
1280 this->rb_right = NULL;
1284 c->size_tree = RB_ROOT;
1288 * ubifs_recover_size_accum - accumulate inode sizes for recovery.
1289 * @c: UBIFS file-system description object
1291 * @deletion: node is for a deletion
1292 * @new_size: inode size
1294 * This function has two purposes:
1295 * 1) to ensure there are no data nodes that fall outside the inode size
1296 * 2) to ensure there are no data nodes for inodes that do not exist
1297 * To accomplish those purposes, a rb-tree is constructed containing an entry
1298 * for each inode number in the journal that has not been deleted, and recording
1299 * the size from the inode node, the maximum size of any data node (also altered
1300 * by truncations) and a flag indicating a inode number for which no inode node
1301 * was present in the journal.
1303 * Note that there is still the possibility that there are data nodes that have
1304 * been committed that are beyond the inode size, however the only way to find
1305 * them would be to scan the entire index. Alternatively, some provision could
1306 * be made to record the size of inodes at the start of commit, which would seem
1307 * very cumbersome for a scenario that is quite unlikely and the only negative
1308 * consequence of which is wasted space.
1310 * This functions returns %0 on success and a negative error code on failure.
1312 int ubifs_recover_size_accum(struct ubifs_info *c, union ubifs_key *key,
1313 int deletion, loff_t new_size)
1315 ino_t inum = key_inum(c, key);
1316 struct size_entry *e;
1319 switch (key_type(c, key)) {
1322 remove_ino(c, inum);
1324 e = find_ino(c, inum);
1326 e->i_size = new_size;
1329 err = add_ino(c, inum, new_size, 0, 1);
1335 case UBIFS_DATA_KEY:
1336 e = find_ino(c, inum);
1338 if (new_size > e->d_size)
1339 e->d_size = new_size;
1341 err = add_ino(c, inum, 0, new_size, 0);
1346 case UBIFS_TRUN_KEY:
1347 e = find_ino(c, inum);
1349 e->d_size = new_size;
1356 * fix_size_in_place - fix inode size in place on flash.
1357 * @c: UBIFS file-system description object
1358 * @e: inode size information for recovery
1360 static int fix_size_in_place(struct ubifs_info *c, struct size_entry *e)
1362 struct ubifs_ino_node *ino = c->sbuf;
1364 union ubifs_key key;
1365 int err, lnum, offs, len;
1369 /* Locate the inode node LEB number and offset */
1370 ino_key_init(c, &key, e->inum);
1371 err = ubifs_tnc_locate(c, &key, ino, &lnum, &offs);
1375 * If the size recorded on the inode node is greater than the size that
1376 * was calculated from nodes in the journal then don't change the inode.
1378 i_size = le64_to_cpu(ino->size);
1379 if (i_size >= e->d_size)
1382 err = ubi_read(c->ubi, lnum, c->sbuf, 0, c->leb_size);
1385 /* Change the size field and recalculate the CRC */
1386 ino = c->sbuf + offs;
1387 ino->size = cpu_to_le64(e->d_size);
1388 len = le32_to_cpu(ino->ch.len);
1389 crc = crc32(UBIFS_CRC32_INIT, (void *)ino + 8, len - 8);
1390 ino->ch.crc = cpu_to_le32(crc);
1391 /* Work out where data in the LEB ends and free space begins */
1393 len = c->leb_size - 1;
1394 while (p[len] == 0xff)
1396 len = ALIGN(len + 1, c->min_io_size);
1397 /* Atomically write the fixed LEB back again */
1398 err = ubi_leb_change(c->ubi, lnum, c->sbuf, len, UBI_UNKNOWN);
1401 dbg_rcvry("inode %lu at %d:%d size %lld -> %lld ",
1402 (unsigned long)e->inum, lnum, offs, i_size, e->d_size);
1406 ubifs_warn("inode %lu failed to fix size %lld -> %lld error %d",
1407 (unsigned long)e->inum, e->i_size, e->d_size, err);
1412 * ubifs_recover_size - recover inode size.
1413 * @c: UBIFS file-system description object
1415 * This function attempts to fix inode size discrepancies identified by the
1416 * 'ubifs_recover_size_accum()' function.
1418 * This functions returns %0 on success and a negative error code on failure.
1420 int ubifs_recover_size(struct ubifs_info *c)
1422 struct rb_node *this = rb_first(&c->size_tree);
1425 struct size_entry *e;
1428 e = rb_entry(this, struct size_entry, rb);
1430 union ubifs_key key;
1432 ino_key_init(c, &key, e->inum);
1433 err = ubifs_tnc_lookup(c, &key, c->sbuf);
1434 if (err && err != -ENOENT)
1436 if (err == -ENOENT) {
1437 /* Remove data nodes that have no inode */
1438 dbg_rcvry("removing ino %lu",
1439 (unsigned long)e->inum);
1440 err = ubifs_tnc_remove_ino(c, e->inum);
1444 struct ubifs_ino_node *ino = c->sbuf;
1447 e->i_size = le64_to_cpu(ino->size);
1450 if (e->exists && e->i_size < e->d_size) {
1451 if (!e->inode && (c->vfs_sb->s_flags & MS_RDONLY)) {
1452 /* Fix the inode size and pin it in memory */
1453 struct inode *inode;
1455 inode = ubifs_iget(c->vfs_sb, e->inum);
1457 return PTR_ERR(inode);
1458 if (inode->i_size < e->d_size) {
1459 dbg_rcvry("ino %lu size %lld -> %lld",
1460 (unsigned long)e->inum,
1461 e->d_size, inode->i_size);
1462 inode->i_size = e->d_size;
1463 ubifs_inode(inode)->ui_size = e->d_size;
1465 this = rb_next(this);
1470 /* Fix the size in place */
1471 err = fix_size_in_place(c, e);
1478 this = rb_next(this);
1479 rb_erase(&e->rb, &c->size_tree);