2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_error.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_alloc.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_log_priv.h"
42 #include "xfs_buf_item.h"
43 #include "xfs_log_recover.h"
44 #include "xfs_extfree_item.h"
45 #include "xfs_trans_priv.h"
46 #include "xfs_quota.h"
48 #include "xfs_utils.h"
49 #include "xfs_trace.h"
51 STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *);
52 STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t);
53 STATIC void xlog_recover_insert_item_backq(xlog_recover_item_t **q,
54 xlog_recover_item_t *item);
56 STATIC void xlog_recover_check_summary(xlog_t *);
58 #define xlog_recover_check_summary(log)
63 * Sector aligned buffer routines for buffer create/read/write/access
66 #define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \
67 ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \
68 ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) )
69 #define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask)
76 if (nbblks <= 0 || nbblks > log->l_logBBsize) {
77 xlog_warn("XFS: Invalid block length (0x%x) given for buffer", nbblks);
78 XFS_ERROR_REPORT("xlog_get_bp(1)",
79 XFS_ERRLEVEL_HIGH, log->l_mp);
83 if (log->l_sectbb_log) {
85 nbblks += XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
86 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
88 return xfs_buf_get_noaddr(BBTOB(nbblks), log->l_mp->m_logdev_targp);
107 if (!log->l_sectbb_log)
108 return XFS_BUF_PTR(bp);
110 ptr = XFS_BUF_PTR(bp) + BBTOB((int)blk_no & log->l_sectbb_mask);
111 ASSERT(XFS_BUF_SIZE(bp) >=
112 BBTOB(nbblks + (blk_no & log->l_sectbb_mask)));
118 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
129 if (nbblks <= 0 || nbblks > log->l_logBBsize) {
130 xlog_warn("XFS: Invalid block length (0x%x) given for buffer", nbblks);
131 XFS_ERROR_REPORT("xlog_bread(1)",
132 XFS_ERRLEVEL_HIGH, log->l_mp);
136 if (log->l_sectbb_log) {
137 blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
138 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
142 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
145 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
148 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
149 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
151 xfsbdstrat(log->l_mp, bp);
152 error = xfs_iowait(bp);
154 xfs_ioerror_alert("xlog_bread", log->l_mp,
155 bp, XFS_BUF_ADDR(bp));
169 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
173 *offset = xlog_align(log, blk_no, nbblks, bp);
178 * Write out the buffer at the given block for the given number of blocks.
179 * The buffer is kept locked across the write and is returned locked.
180 * This can only be used for synchronous log writes.
191 if (nbblks <= 0 || nbblks > log->l_logBBsize) {
192 xlog_warn("XFS: Invalid block length (0x%x) given for buffer", nbblks);
193 XFS_ERROR_REPORT("xlog_bwrite(1)",
194 XFS_ERRLEVEL_HIGH, log->l_mp);
198 if (log->l_sectbb_log) {
199 blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
200 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
204 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
206 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
207 XFS_BUF_ZEROFLAGS(bp);
210 XFS_BUF_PSEMA(bp, PRIBIO);
211 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
212 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
214 if ((error = xfs_bwrite(log->l_mp, bp)))
215 xfs_ioerror_alert("xlog_bwrite", log->l_mp,
216 bp, XFS_BUF_ADDR(bp));
222 * dump debug superblock and log record information
225 xlog_header_check_dump(
227 xlog_rec_header_t *head)
229 cmn_err(CE_DEBUG, "%s: SB : uuid = %pU, fmt = %d\n",
230 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
231 cmn_err(CE_DEBUG, " log : uuid = %pU, fmt = %d\n",
232 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
235 #define xlog_header_check_dump(mp, head)
239 * check log record header for recovery
242 xlog_header_check_recover(
244 xlog_rec_header_t *head)
246 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
249 * IRIX doesn't write the h_fmt field and leaves it zeroed
250 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
251 * a dirty log created in IRIX.
253 if (unlikely(be32_to_cpu(head->h_fmt) != XLOG_FMT)) {
255 "XFS: dirty log written in incompatible format - can't recover");
256 xlog_header_check_dump(mp, head);
257 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
258 XFS_ERRLEVEL_HIGH, mp);
259 return XFS_ERROR(EFSCORRUPTED);
260 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
262 "XFS: dirty log entry has mismatched uuid - can't recover");
263 xlog_header_check_dump(mp, head);
264 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
265 XFS_ERRLEVEL_HIGH, mp);
266 return XFS_ERROR(EFSCORRUPTED);
272 * read the head block of the log and check the header
275 xlog_header_check_mount(
277 xlog_rec_header_t *head)
279 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
281 if (uuid_is_nil(&head->h_fs_uuid)) {
283 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
284 * h_fs_uuid is nil, we assume this log was last mounted
285 * by IRIX and continue.
287 xlog_warn("XFS: nil uuid in log - IRIX style log");
288 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
289 xlog_warn("XFS: log has mismatched uuid - can't recover");
290 xlog_header_check_dump(mp, head);
291 XFS_ERROR_REPORT("xlog_header_check_mount",
292 XFS_ERRLEVEL_HIGH, mp);
293 return XFS_ERROR(EFSCORRUPTED);
302 if (XFS_BUF_GETERROR(bp)) {
304 * We're not going to bother about retrying
305 * this during recovery. One strike!
307 xfs_ioerror_alert("xlog_recover_iodone",
308 bp->b_mount, bp, XFS_BUF_ADDR(bp));
309 xfs_force_shutdown(bp->b_mount, SHUTDOWN_META_IO_ERROR);
312 XFS_BUF_CLR_IODONE_FUNC(bp);
317 * This routine finds (to an approximation) the first block in the physical
318 * log which contains the given cycle. It uses a binary search algorithm.
319 * Note that the algorithm can not be perfect because the disk will not
320 * necessarily be perfect.
323 xlog_find_cycle_start(
326 xfs_daddr_t first_blk,
327 xfs_daddr_t *last_blk,
335 mid_blk = BLK_AVG(first_blk, *last_blk);
336 while (mid_blk != first_blk && mid_blk != *last_blk) {
337 error = xlog_bread(log, mid_blk, 1, bp, &offset);
340 mid_cycle = xlog_get_cycle(offset);
341 if (mid_cycle == cycle) {
343 /* last_half_cycle == mid_cycle */
346 /* first_half_cycle == mid_cycle */
348 mid_blk = BLK_AVG(first_blk, *last_blk);
350 ASSERT((mid_blk == first_blk && mid_blk+1 == *last_blk) ||
351 (mid_blk == *last_blk && mid_blk-1 == first_blk));
357 * Check that the range of blocks does not contain the cycle number
358 * given. The scan needs to occur from front to back and the ptr into the
359 * region must be updated since a later routine will need to perform another
360 * test. If the region is completely good, we end up returning the same
363 * Set blkno to -1 if we encounter no errors. This is an invalid block number
364 * since we don't ever expect logs to get this large.
367 xlog_find_verify_cycle(
369 xfs_daddr_t start_blk,
371 uint stop_on_cycle_no,
372 xfs_daddr_t *new_blk)
378 xfs_caddr_t buf = NULL;
381 bufblks = 1 << ffs(nbblks);
383 while (!(bp = xlog_get_bp(log, bufblks))) {
384 /* can't get enough memory to do everything in one big buffer */
386 if (bufblks <= log->l_sectbb_log)
390 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
393 bcount = min(bufblks, (start_blk + nbblks - i));
395 error = xlog_bread(log, i, bcount, bp, &buf);
399 for (j = 0; j < bcount; j++) {
400 cycle = xlog_get_cycle(buf);
401 if (cycle == stop_on_cycle_no) {
418 * Potentially backup over partial log record write.
420 * In the typical case, last_blk is the number of the block directly after
421 * a good log record. Therefore, we subtract one to get the block number
422 * of the last block in the given buffer. extra_bblks contains the number
423 * of blocks we would have read on a previous read. This happens when the
424 * last log record is split over the end of the physical log.
426 * extra_bblks is the number of blocks potentially verified on a previous
427 * call to this routine.
430 xlog_find_verify_log_record(
432 xfs_daddr_t start_blk,
433 xfs_daddr_t *last_blk,
438 xfs_caddr_t offset = NULL;
439 xlog_rec_header_t *head = NULL;
442 int num_blks = *last_blk - start_blk;
445 ASSERT(start_blk != 0 || *last_blk != start_blk);
447 if (!(bp = xlog_get_bp(log, num_blks))) {
448 if (!(bp = xlog_get_bp(log, 1)))
452 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
455 offset += ((num_blks - 1) << BBSHIFT);
458 for (i = (*last_blk) - 1; i >= 0; i--) {
460 /* valid log record not found */
462 "XFS: Log inconsistent (didn't find previous header)");
464 error = XFS_ERROR(EIO);
469 error = xlog_bread(log, i, 1, bp, &offset);
474 head = (xlog_rec_header_t *)offset;
476 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(head->h_magicno))
484 * We hit the beginning of the physical log & still no header. Return
485 * to caller. If caller can handle a return of -1, then this routine
486 * will be called again for the end of the physical log.
494 * We have the final block of the good log (the first block
495 * of the log record _before_ the head. So we check the uuid.
497 if ((error = xlog_header_check_mount(log->l_mp, head)))
501 * We may have found a log record header before we expected one.
502 * last_blk will be the 1st block # with a given cycle #. We may end
503 * up reading an entire log record. In this case, we don't want to
504 * reset last_blk. Only when last_blk points in the middle of a log
505 * record do we update last_blk.
507 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
508 uint h_size = be32_to_cpu(head->h_size);
510 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
511 if (h_size % XLOG_HEADER_CYCLE_SIZE)
517 if (*last_blk - i + extra_bblks !=
518 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
527 * Head is defined to be the point of the log where the next log write
528 * write could go. This means that incomplete LR writes at the end are
529 * eliminated when calculating the head. We aren't guaranteed that previous
530 * LR have complete transactions. We only know that a cycle number of
531 * current cycle number -1 won't be present in the log if we start writing
532 * from our current block number.
534 * last_blk contains the block number of the first block with a given
537 * Return: zero if normal, non-zero if error.
542 xfs_daddr_t *return_head_blk)
546 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
548 uint first_half_cycle, last_half_cycle;
550 int error, log_bbnum = log->l_logBBsize;
552 /* Is the end of the log device zeroed? */
553 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
554 *return_head_blk = first_blk;
556 /* Is the whole lot zeroed? */
558 /* Linux XFS shouldn't generate totally zeroed logs -
559 * mkfs etc write a dummy unmount record to a fresh
560 * log so we can store the uuid in there
562 xlog_warn("XFS: totally zeroed log");
567 xlog_warn("XFS: empty log check failed");
571 first_blk = 0; /* get cycle # of 1st block */
572 bp = xlog_get_bp(log, 1);
576 error = xlog_bread(log, 0, 1, bp, &offset);
580 first_half_cycle = xlog_get_cycle(offset);
582 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
583 error = xlog_bread(log, last_blk, 1, bp, &offset);
587 last_half_cycle = xlog_get_cycle(offset);
588 ASSERT(last_half_cycle != 0);
591 * If the 1st half cycle number is equal to the last half cycle number,
592 * then the entire log is stamped with the same cycle number. In this
593 * case, head_blk can't be set to zero (which makes sense). The below
594 * math doesn't work out properly with head_blk equal to zero. Instead,
595 * we set it to log_bbnum which is an invalid block number, but this
596 * value makes the math correct. If head_blk doesn't changed through
597 * all the tests below, *head_blk is set to zero at the very end rather
598 * than log_bbnum. In a sense, log_bbnum and zero are the same block
599 * in a circular file.
601 if (first_half_cycle == last_half_cycle) {
603 * In this case we believe that the entire log should have
604 * cycle number last_half_cycle. We need to scan backwards
605 * from the end verifying that there are no holes still
606 * containing last_half_cycle - 1. If we find such a hole,
607 * then the start of that hole will be the new head. The
608 * simple case looks like
609 * x | x ... | x - 1 | x
610 * Another case that fits this picture would be
611 * x | x + 1 | x ... | x
612 * In this case the head really is somewhere at the end of the
613 * log, as one of the latest writes at the beginning was
616 * x | x + 1 | x ... | x - 1 | x
617 * This is really the combination of the above two cases, and
618 * the head has to end up at the start of the x-1 hole at the
621 * In the 256k log case, we will read from the beginning to the
622 * end of the log and search for cycle numbers equal to x-1.
623 * We don't worry about the x+1 blocks that we encounter,
624 * because we know that they cannot be the head since the log
627 head_blk = log_bbnum;
628 stop_on_cycle = last_half_cycle - 1;
631 * In this case we want to find the first block with cycle
632 * number matching last_half_cycle. We expect the log to be
635 * The first block with cycle number x (last_half_cycle) will
636 * be where the new head belongs. First we do a binary search
637 * for the first occurrence of last_half_cycle. The binary
638 * search may not be totally accurate, so then we scan back
639 * from there looking for occurrences of last_half_cycle before
640 * us. If that backwards scan wraps around the beginning of
641 * the log, then we look for occurrences of last_half_cycle - 1
642 * at the end of the log. The cases we're looking for look
644 * x + 1 ... | x | x + 1 | x ...
645 * ^ binary search stopped here
647 * x + 1 ... | x ... | x - 1 | x
648 * <---------> less than scan distance
650 stop_on_cycle = last_half_cycle;
651 if ((error = xlog_find_cycle_start(log, bp, first_blk,
652 &head_blk, last_half_cycle)))
657 * Now validate the answer. Scan back some number of maximum possible
658 * blocks and make sure each one has the expected cycle number. The
659 * maximum is determined by the total possible amount of buffering
660 * in the in-core log. The following number can be made tighter if
661 * we actually look at the block size of the filesystem.
663 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
664 if (head_blk >= num_scan_bblks) {
666 * We are guaranteed that the entire check can be performed
669 start_blk = head_blk - num_scan_bblks;
670 if ((error = xlog_find_verify_cycle(log,
671 start_blk, num_scan_bblks,
672 stop_on_cycle, &new_blk)))
676 } else { /* need to read 2 parts of log */
678 * We are going to scan backwards in the log in two parts.
679 * First we scan the physical end of the log. In this part
680 * of the log, we are looking for blocks with cycle number
681 * last_half_cycle - 1.
682 * If we find one, then we know that the log starts there, as
683 * we've found a hole that didn't get written in going around
684 * the end of the physical log. The simple case for this is
685 * x + 1 ... | x ... | x - 1 | x
686 * <---------> less than scan distance
687 * If all of the blocks at the end of the log have cycle number
688 * last_half_cycle, then we check the blocks at the start of
689 * the log looking for occurrences of last_half_cycle. If we
690 * find one, then our current estimate for the location of the
691 * first occurrence of last_half_cycle is wrong and we move
692 * back to the hole we've found. This case looks like
693 * x + 1 ... | x | x + 1 | x ...
694 * ^ binary search stopped here
695 * Another case we need to handle that only occurs in 256k
697 * x + 1 ... | x ... | x+1 | x ...
698 * ^ binary search stops here
699 * In a 256k log, the scan at the end of the log will see the
700 * x + 1 blocks. We need to skip past those since that is
701 * certainly not the head of the log. By searching for
702 * last_half_cycle-1 we accomplish that.
704 start_blk = log_bbnum - num_scan_bblks + head_blk;
705 ASSERT(head_blk <= INT_MAX &&
706 (xfs_daddr_t) num_scan_bblks - head_blk >= 0);
707 if ((error = xlog_find_verify_cycle(log, start_blk,
708 num_scan_bblks - (int)head_blk,
709 (stop_on_cycle - 1), &new_blk)))
717 * Scan beginning of log now. The last part of the physical
718 * log is good. This scan needs to verify that it doesn't find
719 * the last_half_cycle.
722 ASSERT(head_blk <= INT_MAX);
723 if ((error = xlog_find_verify_cycle(log,
724 start_blk, (int)head_blk,
725 stop_on_cycle, &new_blk)))
733 * Now we need to make sure head_blk is not pointing to a block in
734 * the middle of a log record.
736 num_scan_bblks = XLOG_REC_SHIFT(log);
737 if (head_blk >= num_scan_bblks) {
738 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
740 /* start ptr at last block ptr before head_blk */
741 if ((error = xlog_find_verify_log_record(log, start_blk,
742 &head_blk, 0)) == -1) {
743 error = XFS_ERROR(EIO);
749 ASSERT(head_blk <= INT_MAX);
750 if ((error = xlog_find_verify_log_record(log, start_blk,
751 &head_blk, 0)) == -1) {
752 /* We hit the beginning of the log during our search */
753 start_blk = log_bbnum - num_scan_bblks + head_blk;
755 ASSERT(start_blk <= INT_MAX &&
756 (xfs_daddr_t) log_bbnum-start_blk >= 0);
757 ASSERT(head_blk <= INT_MAX);
758 if ((error = xlog_find_verify_log_record(log,
760 (int)head_blk)) == -1) {
761 error = XFS_ERROR(EIO);
765 if (new_blk != log_bbnum)
772 if (head_blk == log_bbnum)
773 *return_head_blk = 0;
775 *return_head_blk = head_blk;
777 * When returning here, we have a good block number. Bad block
778 * means that during a previous crash, we didn't have a clean break
779 * from cycle number N to cycle number N-1. In this case, we need
780 * to find the first block with cycle number N-1.
788 xlog_warn("XFS: failed to find log head");
793 * Find the sync block number or the tail of the log.
795 * This will be the block number of the last record to have its
796 * associated buffers synced to disk. Every log record header has
797 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
798 * to get a sync block number. The only concern is to figure out which
799 * log record header to believe.
801 * The following algorithm uses the log record header with the largest
802 * lsn. The entire log record does not need to be valid. We only care
803 * that the header is valid.
805 * We could speed up search by using current head_blk buffer, but it is not
811 xfs_daddr_t *head_blk,
812 xfs_daddr_t *tail_blk)
814 xlog_rec_header_t *rhead;
815 xlog_op_header_t *op_head;
816 xfs_caddr_t offset = NULL;
819 xfs_daddr_t umount_data_blk;
820 xfs_daddr_t after_umount_blk;
827 * Find previous log record
829 if ((error = xlog_find_head(log, head_blk)))
832 bp = xlog_get_bp(log, 1);
835 if (*head_blk == 0) { /* special case */
836 error = xlog_bread(log, 0, 1, bp, &offset);
840 if (xlog_get_cycle(offset) == 0) {
842 /* leave all other log inited values alone */
848 * Search backwards looking for log record header block
850 ASSERT(*head_blk < INT_MAX);
851 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
852 error = xlog_bread(log, i, 1, bp, &offset);
856 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(*(__be32 *)offset)) {
862 * If we haven't found the log record header block, start looking
863 * again from the end of the physical log. XXXmiken: There should be
864 * a check here to make sure we didn't search more than N blocks in
868 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
869 error = xlog_bread(log, i, 1, bp, &offset);
873 if (XLOG_HEADER_MAGIC_NUM ==
874 be32_to_cpu(*(__be32 *)offset)) {
881 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
883 return XFS_ERROR(EIO);
886 /* find blk_no of tail of log */
887 rhead = (xlog_rec_header_t *)offset;
888 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
891 * Reset log values according to the state of the log when we
892 * crashed. In the case where head_blk == 0, we bump curr_cycle
893 * one because the next write starts a new cycle rather than
894 * continuing the cycle of the last good log record. At this
895 * point we have guaranteed that all partial log records have been
896 * accounted for. Therefore, we know that the last good log record
897 * written was complete and ended exactly on the end boundary
898 * of the physical log.
900 log->l_prev_block = i;
901 log->l_curr_block = (int)*head_blk;
902 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
905 log->l_tail_lsn = be64_to_cpu(rhead->h_tail_lsn);
906 log->l_last_sync_lsn = be64_to_cpu(rhead->h_lsn);
907 log->l_grant_reserve_cycle = log->l_curr_cycle;
908 log->l_grant_reserve_bytes = BBTOB(log->l_curr_block);
909 log->l_grant_write_cycle = log->l_curr_cycle;
910 log->l_grant_write_bytes = BBTOB(log->l_curr_block);
913 * Look for unmount record. If we find it, then we know there
914 * was a clean unmount. Since 'i' could be the last block in
915 * the physical log, we convert to a log block before comparing
918 * Save the current tail lsn to use to pass to
919 * xlog_clear_stale_blocks() below. We won't want to clear the
920 * unmount record if there is one, so we pass the lsn of the
921 * unmount record rather than the block after it.
923 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
924 int h_size = be32_to_cpu(rhead->h_size);
925 int h_version = be32_to_cpu(rhead->h_version);
927 if ((h_version & XLOG_VERSION_2) &&
928 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
929 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
930 if (h_size % XLOG_HEADER_CYCLE_SIZE)
938 after_umount_blk = (i + hblks + (int)
939 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
940 tail_lsn = log->l_tail_lsn;
941 if (*head_blk == after_umount_blk &&
942 be32_to_cpu(rhead->h_num_logops) == 1) {
943 umount_data_blk = (i + hblks) % log->l_logBBsize;
944 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
948 op_head = (xlog_op_header_t *)offset;
949 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
951 * Set tail and last sync so that newly written
952 * log records will point recovery to after the
953 * current unmount record.
956 xlog_assign_lsn(log->l_curr_cycle,
958 log->l_last_sync_lsn =
959 xlog_assign_lsn(log->l_curr_cycle,
961 *tail_blk = after_umount_blk;
964 * Note that the unmount was clean. If the unmount
965 * was not clean, we need to know this to rebuild the
966 * superblock counters from the perag headers if we
967 * have a filesystem using non-persistent counters.
969 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
974 * Make sure that there are no blocks in front of the head
975 * with the same cycle number as the head. This can happen
976 * because we allow multiple outstanding log writes concurrently,
977 * and the later writes might make it out before earlier ones.
979 * We use the lsn from before modifying it so that we'll never
980 * overwrite the unmount record after a clean unmount.
982 * Do this only if we are going to recover the filesystem
984 * NOTE: This used to say "if (!readonly)"
985 * However on Linux, we can & do recover a read-only filesystem.
986 * We only skip recovery if NORECOVERY is specified on mount,
987 * in which case we would not be here.
989 * But... if the -device- itself is readonly, just skip this.
990 * We can't recover this device anyway, so it won't matter.
992 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) {
993 error = xlog_clear_stale_blocks(log, tail_lsn);
1001 xlog_warn("XFS: failed to locate log tail");
1006 * Is the log zeroed at all?
1008 * The last binary search should be changed to perform an X block read
1009 * once X becomes small enough. You can then search linearly through
1010 * the X blocks. This will cut down on the number of reads we need to do.
1012 * If the log is partially zeroed, this routine will pass back the blkno
1013 * of the first block with cycle number 0. It won't have a complete LR
1017 * 0 => the log is completely written to
1018 * -1 => use *blk_no as the first block of the log
1019 * >0 => error has occurred
1024 xfs_daddr_t *blk_no)
1028 uint first_cycle, last_cycle;
1029 xfs_daddr_t new_blk, last_blk, start_blk;
1030 xfs_daddr_t num_scan_bblks;
1031 int error, log_bbnum = log->l_logBBsize;
1035 /* check totally zeroed log */
1036 bp = xlog_get_bp(log, 1);
1039 error = xlog_bread(log, 0, 1, bp, &offset);
1043 first_cycle = xlog_get_cycle(offset);
1044 if (first_cycle == 0) { /* completely zeroed log */
1050 /* check partially zeroed log */
1051 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1055 last_cycle = xlog_get_cycle(offset);
1056 if (last_cycle != 0) { /* log completely written to */
1059 } else if (first_cycle != 1) {
1061 * If the cycle of the last block is zero, the cycle of
1062 * the first block must be 1. If it's not, maybe we're
1063 * not looking at a log... Bail out.
1065 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1066 return XFS_ERROR(EINVAL);
1069 /* we have a partially zeroed log */
1070 last_blk = log_bbnum-1;
1071 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1075 * Validate the answer. Because there is no way to guarantee that
1076 * the entire log is made up of log records which are the same size,
1077 * we scan over the defined maximum blocks. At this point, the maximum
1078 * is not chosen to mean anything special. XXXmiken
1080 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1081 ASSERT(num_scan_bblks <= INT_MAX);
1083 if (last_blk < num_scan_bblks)
1084 num_scan_bblks = last_blk;
1085 start_blk = last_blk - num_scan_bblks;
1088 * We search for any instances of cycle number 0 that occur before
1089 * our current estimate of the head. What we're trying to detect is
1090 * 1 ... | 0 | 1 | 0...
1091 * ^ binary search ends here
1093 if ((error = xlog_find_verify_cycle(log, start_blk,
1094 (int)num_scan_bblks, 0, &new_blk)))
1100 * Potentially backup over partial log record write. We don't need
1101 * to search the end of the log because we know it is zero.
1103 if ((error = xlog_find_verify_log_record(log, start_blk,
1104 &last_blk, 0)) == -1) {
1105 error = XFS_ERROR(EIO);
1119 * These are simple subroutines used by xlog_clear_stale_blocks() below
1120 * to initialize a buffer full of empty log record headers and write
1121 * them into the log.
1132 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1134 memset(buf, 0, BBSIZE);
1135 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1136 recp->h_cycle = cpu_to_be32(cycle);
1137 recp->h_version = cpu_to_be32(
1138 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1139 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1140 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1141 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1142 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1146 xlog_write_log_records(
1157 int sectbb = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
1158 int end_block = start_block + blocks;
1163 bufblks = 1 << ffs(blocks);
1164 while (!(bp = xlog_get_bp(log, bufblks))) {
1166 if (bufblks <= log->l_sectbb_log)
1170 /* We may need to do a read at the start to fill in part of
1171 * the buffer in the starting sector not covered by the first
1174 balign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, start_block);
1175 if (balign != start_block) {
1176 error = xlog_bread_noalign(log, start_block, 1, bp);
1180 j = start_block - balign;
1183 for (i = start_block; i < end_block; i += bufblks) {
1184 int bcount, endcount;
1186 bcount = min(bufblks, end_block - start_block);
1187 endcount = bcount - j;
1189 /* We may need to do a read at the end to fill in part of
1190 * the buffer in the final sector not covered by the write.
1191 * If this is the same sector as the above read, skip it.
1193 ealign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, end_block);
1194 if (j == 0 && (start_block + endcount > ealign)) {
1195 offset = XFS_BUF_PTR(bp);
1196 balign = BBTOB(ealign - start_block);
1197 error = XFS_BUF_SET_PTR(bp, offset + balign,
1202 error = xlog_bread_noalign(log, ealign, sectbb, bp);
1206 error = XFS_BUF_SET_PTR(bp, offset, bufblks);
1211 offset = xlog_align(log, start_block, endcount, bp);
1212 for (; j < endcount; j++) {
1213 xlog_add_record(log, offset, cycle, i+j,
1214 tail_cycle, tail_block);
1217 error = xlog_bwrite(log, start_block, endcount, bp);
1220 start_block += endcount;
1230 * This routine is called to blow away any incomplete log writes out
1231 * in front of the log head. We do this so that we won't become confused
1232 * if we come up, write only a little bit more, and then crash again.
1233 * If we leave the partial log records out there, this situation could
1234 * cause us to think those partial writes are valid blocks since they
1235 * have the current cycle number. We get rid of them by overwriting them
1236 * with empty log records with the old cycle number rather than the
1239 * The tail lsn is passed in rather than taken from
1240 * the log so that we will not write over the unmount record after a
1241 * clean unmount in a 512 block log. Doing so would leave the log without
1242 * any valid log records in it until a new one was written. If we crashed
1243 * during that time we would not be able to recover.
1246 xlog_clear_stale_blocks(
1250 int tail_cycle, head_cycle;
1251 int tail_block, head_block;
1252 int tail_distance, max_distance;
1256 tail_cycle = CYCLE_LSN(tail_lsn);
1257 tail_block = BLOCK_LSN(tail_lsn);
1258 head_cycle = log->l_curr_cycle;
1259 head_block = log->l_curr_block;
1262 * Figure out the distance between the new head of the log
1263 * and the tail. We want to write over any blocks beyond the
1264 * head that we may have written just before the crash, but
1265 * we don't want to overwrite the tail of the log.
1267 if (head_cycle == tail_cycle) {
1269 * The tail is behind the head in the physical log,
1270 * so the distance from the head to the tail is the
1271 * distance from the head to the end of the log plus
1272 * the distance from the beginning of the log to the
1275 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1276 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1277 XFS_ERRLEVEL_LOW, log->l_mp);
1278 return XFS_ERROR(EFSCORRUPTED);
1280 tail_distance = tail_block + (log->l_logBBsize - head_block);
1283 * The head is behind the tail in the physical log,
1284 * so the distance from the head to the tail is just
1285 * the tail block minus the head block.
1287 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1288 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1289 XFS_ERRLEVEL_LOW, log->l_mp);
1290 return XFS_ERROR(EFSCORRUPTED);
1292 tail_distance = tail_block - head_block;
1296 * If the head is right up against the tail, we can't clear
1299 if (tail_distance <= 0) {
1300 ASSERT(tail_distance == 0);
1304 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1306 * Take the smaller of the maximum amount of outstanding I/O
1307 * we could have and the distance to the tail to clear out.
1308 * We take the smaller so that we don't overwrite the tail and
1309 * we don't waste all day writing from the head to the tail
1312 max_distance = MIN(max_distance, tail_distance);
1314 if ((head_block + max_distance) <= log->l_logBBsize) {
1316 * We can stomp all the blocks we need to without
1317 * wrapping around the end of the log. Just do it
1318 * in a single write. Use the cycle number of the
1319 * current cycle minus one so that the log will look like:
1322 error = xlog_write_log_records(log, (head_cycle - 1),
1323 head_block, max_distance, tail_cycle,
1329 * We need to wrap around the end of the physical log in
1330 * order to clear all the blocks. Do it in two separate
1331 * I/Os. The first write should be from the head to the
1332 * end of the physical log, and it should use the current
1333 * cycle number minus one just like above.
1335 distance = log->l_logBBsize - head_block;
1336 error = xlog_write_log_records(log, (head_cycle - 1),
1337 head_block, distance, tail_cycle,
1344 * Now write the blocks at the start of the physical log.
1345 * This writes the remainder of the blocks we want to clear.
1346 * It uses the current cycle number since we're now on the
1347 * same cycle as the head so that we get:
1348 * n ... n ... | n - 1 ...
1349 * ^^^^^ blocks we're writing
1351 distance = max_distance - (log->l_logBBsize - head_block);
1352 error = xlog_write_log_records(log, head_cycle, 0, distance,
1353 tail_cycle, tail_block);
1361 /******************************************************************************
1363 * Log recover routines
1365 ******************************************************************************
1368 STATIC xlog_recover_t *
1369 xlog_recover_find_tid(
1373 xlog_recover_t *p = q;
1376 if (p->r_log_tid == tid)
1384 xlog_recover_put_hashq(
1386 xlog_recover_t *trans)
1393 xlog_recover_add_item(
1394 xlog_recover_item_t **itemq)
1396 xlog_recover_item_t *item;
1398 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1399 xlog_recover_insert_item_backq(itemq, item);
1403 xlog_recover_add_to_cont_trans(
1404 xlog_recover_t *trans,
1408 xlog_recover_item_t *item;
1409 xfs_caddr_t ptr, old_ptr;
1412 item = trans->r_itemq;
1414 /* finish copying rest of trans header */
1415 xlog_recover_add_item(&trans->r_itemq);
1416 ptr = (xfs_caddr_t) &trans->r_theader +
1417 sizeof(xfs_trans_header_t) - len;
1418 memcpy(ptr, dp, len); /* d, s, l */
1421 item = item->ri_prev;
1423 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1424 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1426 ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
1427 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1428 item->ri_buf[item->ri_cnt-1].i_len += len;
1429 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1434 * The next region to add is the start of a new region. It could be
1435 * a whole region or it could be the first part of a new region. Because
1436 * of this, the assumption here is that the type and size fields of all
1437 * format structures fit into the first 32 bits of the structure.
1439 * This works because all regions must be 32 bit aligned. Therefore, we
1440 * either have both fields or we have neither field. In the case we have
1441 * neither field, the data part of the region is zero length. We only have
1442 * a log_op_header and can throw away the header since a new one will appear
1443 * later. If we have at least 4 bytes, then we can determine how many regions
1444 * will appear in the current log item.
1447 xlog_recover_add_to_trans(
1448 xlog_recover_t *trans,
1452 xfs_inode_log_format_t *in_f; /* any will do */
1453 xlog_recover_item_t *item;
1458 item = trans->r_itemq;
1460 /* we need to catch log corruptions here */
1461 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
1462 xlog_warn("XFS: xlog_recover_add_to_trans: "
1463 "bad header magic number");
1465 return XFS_ERROR(EIO);
1467 if (len == sizeof(xfs_trans_header_t))
1468 xlog_recover_add_item(&trans->r_itemq);
1469 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1473 ptr = kmem_alloc(len, KM_SLEEP);
1474 memcpy(ptr, dp, len);
1475 in_f = (xfs_inode_log_format_t *)ptr;
1477 if (item->ri_prev->ri_total != 0 &&
1478 item->ri_prev->ri_total == item->ri_prev->ri_cnt) {
1479 xlog_recover_add_item(&trans->r_itemq);
1481 item = trans->r_itemq;
1482 item = item->ri_prev;
1484 if (item->ri_total == 0) { /* first region to be added */
1485 if (in_f->ilf_size == 0 ||
1486 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
1488 "XFS: bad number of regions (%d) in inode log format",
1491 return XFS_ERROR(EIO);
1494 item->ri_total = in_f->ilf_size;
1496 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1499 ASSERT(item->ri_total > item->ri_cnt);
1500 /* Description region is ri_buf[0] */
1501 item->ri_buf[item->ri_cnt].i_addr = ptr;
1502 item->ri_buf[item->ri_cnt].i_len = len;
1508 xlog_recover_new_tid(
1513 xlog_recover_t *trans;
1515 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1516 trans->r_log_tid = tid;
1518 xlog_recover_put_hashq(q, trans);
1522 xlog_recover_unlink_tid(
1524 xlog_recover_t *trans)
1529 ASSERT(trans != NULL);
1535 if (tp->r_next == trans) {
1543 "XFS: xlog_recover_unlink_tid: trans not found");
1545 return XFS_ERROR(EIO);
1547 tp->r_next = tp->r_next->r_next;
1553 xlog_recover_insert_item_backq(
1554 xlog_recover_item_t **q,
1555 xlog_recover_item_t *item)
1558 item->ri_prev = item->ri_next = item;
1562 item->ri_prev = (*q)->ri_prev;
1563 (*q)->ri_prev = item;
1564 item->ri_prev->ri_next = item;
1569 xlog_recover_insert_item_frontq(
1570 xlog_recover_item_t **q,
1571 xlog_recover_item_t *item)
1573 xlog_recover_insert_item_backq(q, item);
1578 xlog_recover_reorder_trans(
1579 xlog_recover_t *trans)
1581 xlog_recover_item_t *first_item, *itemq, *itemq_next;
1582 xfs_buf_log_format_t *buf_f;
1585 first_item = itemq = trans->r_itemq;
1586 trans->r_itemq = NULL;
1588 itemq_next = itemq->ri_next;
1589 buf_f = (xfs_buf_log_format_t *)itemq->ri_buf[0].i_addr;
1591 switch (ITEM_TYPE(itemq)) {
1593 flags = buf_f->blf_flags;
1594 if (!(flags & XFS_BLI_CANCEL)) {
1595 xlog_recover_insert_item_frontq(&trans->r_itemq,
1601 case XFS_LI_QUOTAOFF:
1604 xlog_recover_insert_item_backq(&trans->r_itemq, itemq);
1608 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1610 return XFS_ERROR(EIO);
1613 } while (first_item != itemq);
1618 * Build up the table of buf cancel records so that we don't replay
1619 * cancelled data in the second pass. For buffer records that are
1620 * not cancel records, there is nothing to do here so we just return.
1622 * If we get a cancel record which is already in the table, this indicates
1623 * that the buffer was cancelled multiple times. In order to ensure
1624 * that during pass 2 we keep the record in the table until we reach its
1625 * last occurrence in the log, we keep a reference count in the cancel
1626 * record in the table to tell us how many times we expect to see this
1627 * record during the second pass.
1630 xlog_recover_do_buffer_pass1(
1632 xfs_buf_log_format_t *buf_f)
1634 xfs_buf_cancel_t *bcp;
1635 xfs_buf_cancel_t *nextp;
1636 xfs_buf_cancel_t *prevp;
1637 xfs_buf_cancel_t **bucket;
1638 xfs_daddr_t blkno = 0;
1642 switch (buf_f->blf_type) {
1644 blkno = buf_f->blf_blkno;
1645 len = buf_f->blf_len;
1646 flags = buf_f->blf_flags;
1651 * If this isn't a cancel buffer item, then just return.
1653 if (!(flags & XFS_BLI_CANCEL))
1657 * Insert an xfs_buf_cancel record into the hash table of
1658 * them. If there is already an identical record, bump
1659 * its reference count.
1661 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1662 XLOG_BC_TABLE_SIZE];
1664 * If the hash bucket is empty then just insert a new record into
1667 if (*bucket == NULL) {
1668 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1670 bcp->bc_blkno = blkno;
1672 bcp->bc_refcount = 1;
1673 bcp->bc_next = NULL;
1679 * The hash bucket is not empty, so search for duplicates of our
1680 * record. If we find one them just bump its refcount. If not
1681 * then add us at the end of the list.
1685 while (nextp != NULL) {
1686 if (nextp->bc_blkno == blkno && nextp->bc_len == len) {
1687 nextp->bc_refcount++;
1691 nextp = nextp->bc_next;
1693 ASSERT(prevp != NULL);
1694 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1696 bcp->bc_blkno = blkno;
1698 bcp->bc_refcount = 1;
1699 bcp->bc_next = NULL;
1700 prevp->bc_next = bcp;
1704 * Check to see whether the buffer being recovered has a corresponding
1705 * entry in the buffer cancel record table. If it does then return 1
1706 * so that it will be cancelled, otherwise return 0. If the buffer is
1707 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
1708 * the refcount on the entry in the table and remove it from the table
1709 * if this is the last reference.
1711 * We remove the cancel record from the table when we encounter its
1712 * last occurrence in the log so that if the same buffer is re-used
1713 * again after its last cancellation we actually replay the changes
1714 * made at that point.
1717 xlog_check_buffer_cancelled(
1723 xfs_buf_cancel_t *bcp;
1724 xfs_buf_cancel_t *prevp;
1725 xfs_buf_cancel_t **bucket;
1727 if (log->l_buf_cancel_table == NULL) {
1729 * There is nothing in the table built in pass one,
1730 * so this buffer must not be cancelled.
1732 ASSERT(!(flags & XFS_BLI_CANCEL));
1736 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1737 XLOG_BC_TABLE_SIZE];
1741 * There is no corresponding entry in the table built
1742 * in pass one, so this buffer has not been cancelled.
1744 ASSERT(!(flags & XFS_BLI_CANCEL));
1749 * Search for an entry in the buffer cancel table that
1750 * matches our buffer.
1753 while (bcp != NULL) {
1754 if (bcp->bc_blkno == blkno && bcp->bc_len == len) {
1756 * We've go a match, so return 1 so that the
1757 * recovery of this buffer is cancelled.
1758 * If this buffer is actually a buffer cancel
1759 * log item, then decrement the refcount on the
1760 * one in the table and remove it if this is the
1763 if (flags & XFS_BLI_CANCEL) {
1765 if (bcp->bc_refcount == 0) {
1766 if (prevp == NULL) {
1767 *bucket = bcp->bc_next;
1769 prevp->bc_next = bcp->bc_next;
1780 * We didn't find a corresponding entry in the table, so
1781 * return 0 so that the buffer is NOT cancelled.
1783 ASSERT(!(flags & XFS_BLI_CANCEL));
1788 xlog_recover_do_buffer_pass2(
1790 xfs_buf_log_format_t *buf_f)
1792 xfs_daddr_t blkno = 0;
1796 switch (buf_f->blf_type) {
1798 blkno = buf_f->blf_blkno;
1799 flags = buf_f->blf_flags;
1800 len = buf_f->blf_len;
1804 return xlog_check_buffer_cancelled(log, blkno, len, flags);
1808 * Perform recovery for a buffer full of inodes. In these buffers,
1809 * the only data which should be recovered is that which corresponds
1810 * to the di_next_unlinked pointers in the on disk inode structures.
1811 * The rest of the data for the inodes is always logged through the
1812 * inodes themselves rather than the inode buffer and is recovered
1813 * in xlog_recover_do_inode_trans().
1815 * The only time when buffers full of inodes are fully recovered is
1816 * when the buffer is full of newly allocated inodes. In this case
1817 * the buffer will not be marked as an inode buffer and so will be
1818 * sent to xlog_recover_do_reg_buffer() below during recovery.
1821 xlog_recover_do_inode_buffer(
1823 xlog_recover_item_t *item,
1825 xfs_buf_log_format_t *buf_f)
1833 int next_unlinked_offset;
1835 xfs_agino_t *logged_nextp;
1836 xfs_agino_t *buffer_nextp;
1837 unsigned int *data_map = NULL;
1838 unsigned int map_size = 0;
1840 switch (buf_f->blf_type) {
1842 data_map = buf_f->blf_data_map;
1843 map_size = buf_f->blf_map_size;
1847 * Set the variables corresponding to the current region to
1848 * 0 so that we'll initialize them on the first pass through
1856 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1857 for (i = 0; i < inodes_per_buf; i++) {
1858 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1859 offsetof(xfs_dinode_t, di_next_unlinked);
1861 while (next_unlinked_offset >=
1862 (reg_buf_offset + reg_buf_bytes)) {
1864 * The next di_next_unlinked field is beyond
1865 * the current logged region. Find the next
1866 * logged region that contains or is beyond
1867 * the current di_next_unlinked field.
1870 bit = xfs_next_bit(data_map, map_size, bit);
1873 * If there are no more logged regions in the
1874 * buffer, then we're done.
1880 nbits = xfs_contig_bits(data_map, map_size,
1883 reg_buf_offset = bit << XFS_BLI_SHIFT;
1884 reg_buf_bytes = nbits << XFS_BLI_SHIFT;
1889 * If the current logged region starts after the current
1890 * di_next_unlinked field, then move on to the next
1891 * di_next_unlinked field.
1893 if (next_unlinked_offset < reg_buf_offset) {
1897 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1898 ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0);
1899 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1902 * The current logged region contains a copy of the
1903 * current di_next_unlinked field. Extract its value
1904 * and copy it to the buffer copy.
1906 logged_nextp = (xfs_agino_t *)
1907 ((char *)(item->ri_buf[item_index].i_addr) +
1908 (next_unlinked_offset - reg_buf_offset));
1909 if (unlikely(*logged_nextp == 0)) {
1910 xfs_fs_cmn_err(CE_ALERT, mp,
1911 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1913 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1914 XFS_ERRLEVEL_LOW, mp);
1915 return XFS_ERROR(EFSCORRUPTED);
1918 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1919 next_unlinked_offset);
1920 *buffer_nextp = *logged_nextp;
1927 * Perform a 'normal' buffer recovery. Each logged region of the
1928 * buffer should be copied over the corresponding region in the
1929 * given buffer. The bitmap in the buf log format structure indicates
1930 * where to place the logged data.
1934 xlog_recover_do_reg_buffer(
1935 xlog_recover_item_t *item,
1937 xfs_buf_log_format_t *buf_f)
1942 unsigned int *data_map = NULL;
1943 unsigned int map_size = 0;
1946 switch (buf_f->blf_type) {
1948 data_map = buf_f->blf_data_map;
1949 map_size = buf_f->blf_map_size;
1953 i = 1; /* 0 is the buf format structure */
1955 bit = xfs_next_bit(data_map, map_size, bit);
1958 nbits = xfs_contig_bits(data_map, map_size, bit);
1960 ASSERT(item->ri_buf[i].i_addr != NULL);
1961 ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0);
1962 ASSERT(XFS_BUF_COUNT(bp) >=
1963 ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT));
1966 * Do a sanity check if this is a dquot buffer. Just checking
1967 * the first dquot in the buffer should do. XXXThis is
1968 * probably a good thing to do for other buf types also.
1971 if (buf_f->blf_flags &
1972 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
1973 if (item->ri_buf[i].i_addr == NULL) {
1975 "XFS: NULL dquot in %s.", __func__);
1978 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
1980 "XFS: dquot too small (%d) in %s.",
1981 item->ri_buf[i].i_len, __func__);
1984 error = xfs_qm_dqcheck((xfs_disk_dquot_t *)
1985 item->ri_buf[i].i_addr,
1986 -1, 0, XFS_QMOPT_DOWARN,
1987 "dquot_buf_recover");
1992 memcpy(xfs_buf_offset(bp,
1993 (uint)bit << XFS_BLI_SHIFT), /* dest */
1994 item->ri_buf[i].i_addr, /* source */
1995 nbits<<XFS_BLI_SHIFT); /* length */
2001 /* Shouldn't be any more regions */
2002 ASSERT(i == item->ri_total);
2006 * Do some primitive error checking on ondisk dquot data structures.
2010 xfs_disk_dquot_t *ddq,
2012 uint type, /* used only when IO_dorepair is true */
2016 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
2020 * We can encounter an uninitialized dquot buffer for 2 reasons:
2021 * 1. If we crash while deleting the quotainode(s), and those blks got
2022 * used for user data. This is because we take the path of regular
2023 * file deletion; however, the size field of quotainodes is never
2024 * updated, so all the tricks that we play in itruncate_finish
2025 * don't quite matter.
2027 * 2. We don't play the quota buffers when there's a quotaoff logitem.
2028 * But the allocation will be replayed so we'll end up with an
2029 * uninitialized quota block.
2031 * This is all fine; things are still consistent, and we haven't lost
2032 * any quota information. Just don't complain about bad dquot blks.
2034 if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) {
2035 if (flags & XFS_QMOPT_DOWARN)
2037 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
2038 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
2041 if (ddq->d_version != XFS_DQUOT_VERSION) {
2042 if (flags & XFS_QMOPT_DOWARN)
2044 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
2045 str, id, ddq->d_version, XFS_DQUOT_VERSION);
2049 if (ddq->d_flags != XFS_DQ_USER &&
2050 ddq->d_flags != XFS_DQ_PROJ &&
2051 ddq->d_flags != XFS_DQ_GROUP) {
2052 if (flags & XFS_QMOPT_DOWARN)
2054 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
2055 str, id, ddq->d_flags);
2059 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
2060 if (flags & XFS_QMOPT_DOWARN)
2062 "%s : ondisk-dquot 0x%p, ID mismatch: "
2063 "0x%x expected, found id 0x%x",
2064 str, ddq, id, be32_to_cpu(ddq->d_id));
2068 if (!errs && ddq->d_id) {
2069 if (ddq->d_blk_softlimit &&
2070 be64_to_cpu(ddq->d_bcount) >=
2071 be64_to_cpu(ddq->d_blk_softlimit)) {
2072 if (!ddq->d_btimer) {
2073 if (flags & XFS_QMOPT_DOWARN)
2075 "%s : Dquot ID 0x%x (0x%p) "
2076 "BLK TIMER NOT STARTED",
2077 str, (int)be32_to_cpu(ddq->d_id), ddq);
2081 if (ddq->d_ino_softlimit &&
2082 be64_to_cpu(ddq->d_icount) >=
2083 be64_to_cpu(ddq->d_ino_softlimit)) {
2084 if (!ddq->d_itimer) {
2085 if (flags & XFS_QMOPT_DOWARN)
2087 "%s : Dquot ID 0x%x (0x%p) "
2088 "INODE TIMER NOT STARTED",
2089 str, (int)be32_to_cpu(ddq->d_id), ddq);
2093 if (ddq->d_rtb_softlimit &&
2094 be64_to_cpu(ddq->d_rtbcount) >=
2095 be64_to_cpu(ddq->d_rtb_softlimit)) {
2096 if (!ddq->d_rtbtimer) {
2097 if (flags & XFS_QMOPT_DOWARN)
2099 "%s : Dquot ID 0x%x (0x%p) "
2100 "RTBLK TIMER NOT STARTED",
2101 str, (int)be32_to_cpu(ddq->d_id), ddq);
2107 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2110 if (flags & XFS_QMOPT_DOWARN)
2111 cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id);
2114 * Typically, a repair is only requested by quotacheck.
2117 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2118 memset(d, 0, sizeof(xfs_dqblk_t));
2120 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2121 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2122 d->dd_diskdq.d_flags = type;
2123 d->dd_diskdq.d_id = cpu_to_be32(id);
2129 * Perform a dquot buffer recovery.
2130 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2131 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2132 * Else, treat it as a regular buffer and do recovery.
2135 xlog_recover_do_dquot_buffer(
2138 xlog_recover_item_t *item,
2140 xfs_buf_log_format_t *buf_f)
2145 * Filesystems are required to send in quota flags at mount time.
2147 if (mp->m_qflags == 0) {
2152 if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF)
2153 type |= XFS_DQ_USER;
2154 if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF)
2155 type |= XFS_DQ_PROJ;
2156 if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF)
2157 type |= XFS_DQ_GROUP;
2159 * This type of quotas was turned off, so ignore this buffer
2161 if (log->l_quotaoffs_flag & type)
2164 xlog_recover_do_reg_buffer(item, bp, buf_f);
2168 * This routine replays a modification made to a buffer at runtime.
2169 * There are actually two types of buffer, regular and inode, which
2170 * are handled differently. Inode buffers are handled differently
2171 * in that we only recover a specific set of data from them, namely
2172 * the inode di_next_unlinked fields. This is because all other inode
2173 * data is actually logged via inode records and any data we replay
2174 * here which overlaps that may be stale.
2176 * When meta-data buffers are freed at run time we log a buffer item
2177 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
2178 * of the buffer in the log should not be replayed at recovery time.
2179 * This is so that if the blocks covered by the buffer are reused for
2180 * file data before we crash we don't end up replaying old, freed
2181 * meta-data into a user's file.
2183 * To handle the cancellation of buffer log items, we make two passes
2184 * over the log during recovery. During the first we build a table of
2185 * those buffers which have been cancelled, and during the second we
2186 * only replay those buffers which do not have corresponding cancel
2187 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2188 * for more details on the implementation of the table of cancel records.
2191 xlog_recover_do_buffer_trans(
2193 xlog_recover_item_t *item,
2196 xfs_buf_log_format_t *buf_f;
2206 buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr;
2208 if (pass == XLOG_RECOVER_PASS1) {
2210 * In this pass we're only looking for buf items
2211 * with the XFS_BLI_CANCEL bit set.
2213 xlog_recover_do_buffer_pass1(log, buf_f);
2217 * In this pass we want to recover all the buffers
2218 * which have not been cancelled and are not
2219 * cancellation buffers themselves. The routine
2220 * we call here will tell us whether or not to
2221 * continue with the replay of this buffer.
2223 cancel = xlog_recover_do_buffer_pass2(log, buf_f);
2228 switch (buf_f->blf_type) {
2230 blkno = buf_f->blf_blkno;
2231 len = buf_f->blf_len;
2232 flags = buf_f->blf_flags;
2235 xfs_fs_cmn_err(CE_ALERT, log->l_mp,
2236 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2237 buf_f->blf_type, log->l_mp->m_logname ?
2238 log->l_mp->m_logname : "internal");
2239 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2240 XFS_ERRLEVEL_LOW, log->l_mp);
2241 return XFS_ERROR(EFSCORRUPTED);
2245 buf_flags = XFS_BUF_LOCK;
2246 if (!(flags & XFS_BLI_INODE_BUF))
2247 buf_flags |= XFS_BUF_MAPPED;
2249 bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, buf_flags);
2250 if (XFS_BUF_ISERROR(bp)) {
2251 xfs_ioerror_alert("xlog_recover_do..(read#1)", log->l_mp,
2253 error = XFS_BUF_GETERROR(bp);
2259 if (flags & XFS_BLI_INODE_BUF) {
2260 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2262 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
2263 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2265 xlog_recover_do_reg_buffer(item, bp, buf_f);
2268 return XFS_ERROR(error);
2271 * Perform delayed write on the buffer. Asynchronous writes will be
2272 * slower when taking into account all the buffers to be flushed.
2274 * Also make sure that only inode buffers with good sizes stay in
2275 * the buffer cache. The kernel moves inodes in buffers of 1 block
2276 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2277 * buffers in the log can be a different size if the log was generated
2278 * by an older kernel using unclustered inode buffers or a newer kernel
2279 * running with a different inode cluster size. Regardless, if the
2280 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2281 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2282 * the buffer out of the buffer cache so that the buffer won't
2283 * overlap with future reads of those inodes.
2285 if (XFS_DINODE_MAGIC ==
2286 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2287 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2288 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2290 error = xfs_bwrite(mp, bp);
2292 ASSERT(bp->b_mount == NULL || bp->b_mount == mp);
2294 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2295 xfs_bdwrite(mp, bp);
2302 xlog_recover_do_inode_trans(
2304 xlog_recover_item_t *item,
2307 xfs_inode_log_format_t *in_f;
2318 xfs_icdinode_t *dicp;
2321 if (pass == XLOG_RECOVER_PASS1) {
2325 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2326 in_f = (xfs_inode_log_format_t *)item->ri_buf[0].i_addr;
2328 in_f = (xfs_inode_log_format_t *)kmem_alloc(
2329 sizeof(xfs_inode_log_format_t), KM_SLEEP);
2331 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2335 ino = in_f->ilf_ino;
2339 * Inode buffers can be freed, look out for it,
2340 * and do not replay the inode.
2342 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2343 in_f->ilf_len, 0)) {
2348 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len,
2350 if (XFS_BUF_ISERROR(bp)) {
2351 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
2352 bp, in_f->ilf_blkno);
2353 error = XFS_BUF_GETERROR(bp);
2358 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2359 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
2362 * Make sure the place we're flushing out to really looks
2365 if (unlikely(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC)) {
2367 xfs_fs_cmn_err(CE_ALERT, mp,
2368 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2370 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2371 XFS_ERRLEVEL_LOW, mp);
2372 error = EFSCORRUPTED;
2375 dicp = (xfs_icdinode_t *)(item->ri_buf[1].i_addr);
2376 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2378 xfs_fs_cmn_err(CE_ALERT, mp,
2379 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2381 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2382 XFS_ERRLEVEL_LOW, mp);
2383 error = EFSCORRUPTED;
2387 /* Skip replay when the on disk inode is newer than the log one */
2388 if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
2390 * Deal with the wrap case, DI_MAX_FLUSH is less
2391 * than smaller numbers
2393 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
2394 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2402 /* Take the opportunity to reset the flush iteration count */
2403 dicp->di_flushiter = 0;
2405 if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) {
2406 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2407 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2408 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2409 XFS_ERRLEVEL_LOW, mp, dicp);
2411 xfs_fs_cmn_err(CE_ALERT, mp,
2412 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2413 item, dip, bp, ino);
2414 error = EFSCORRUPTED;
2417 } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) {
2418 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2419 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2420 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2421 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2422 XFS_ERRLEVEL_LOW, mp, dicp);
2424 xfs_fs_cmn_err(CE_ALERT, mp,
2425 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2426 item, dip, bp, ino);
2427 error = EFSCORRUPTED;
2431 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2432 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2433 XFS_ERRLEVEL_LOW, mp, dicp);
2435 xfs_fs_cmn_err(CE_ALERT, mp,
2436 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2438 dicp->di_nextents + dicp->di_anextents,
2440 error = EFSCORRUPTED;
2443 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2444 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2445 XFS_ERRLEVEL_LOW, mp, dicp);
2447 xfs_fs_cmn_err(CE_ALERT, mp,
2448 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2449 item, dip, bp, ino, dicp->di_forkoff);
2450 error = EFSCORRUPTED;
2453 if (unlikely(item->ri_buf[1].i_len > sizeof(struct xfs_icdinode))) {
2454 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2455 XFS_ERRLEVEL_LOW, mp, dicp);
2457 xfs_fs_cmn_err(CE_ALERT, mp,
2458 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2459 item->ri_buf[1].i_len, item);
2460 error = EFSCORRUPTED;
2464 /* The core is in in-core format */
2465 xfs_dinode_to_disk(dip, (xfs_icdinode_t *)item->ri_buf[1].i_addr);
2467 /* the rest is in on-disk format */
2468 if (item->ri_buf[1].i_len > sizeof(struct xfs_icdinode)) {
2469 memcpy((xfs_caddr_t) dip + sizeof(struct xfs_icdinode),
2470 item->ri_buf[1].i_addr + sizeof(struct xfs_icdinode),
2471 item->ri_buf[1].i_len - sizeof(struct xfs_icdinode));
2474 fields = in_f->ilf_fields;
2475 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2477 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
2480 memcpy(XFS_DFORK_DPTR(dip),
2481 &in_f->ilf_u.ilfu_uuid,
2486 if (in_f->ilf_size == 2)
2487 goto write_inode_buffer;
2488 len = item->ri_buf[2].i_len;
2489 src = item->ri_buf[2].i_addr;
2490 ASSERT(in_f->ilf_size <= 4);
2491 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2492 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2493 (len == in_f->ilf_dsize));
2495 switch (fields & XFS_ILOG_DFORK) {
2496 case XFS_ILOG_DDATA:
2498 memcpy(XFS_DFORK_DPTR(dip), src, len);
2501 case XFS_ILOG_DBROOT:
2502 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
2503 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
2504 XFS_DFORK_DSIZE(dip, mp));
2509 * There are no data fork flags set.
2511 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2516 * If we logged any attribute data, recover it. There may or
2517 * may not have been any other non-core data logged in this
2520 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2521 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2526 len = item->ri_buf[attr_index].i_len;
2527 src = item->ri_buf[attr_index].i_addr;
2528 ASSERT(len == in_f->ilf_asize);
2530 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2531 case XFS_ILOG_ADATA:
2533 dest = XFS_DFORK_APTR(dip);
2534 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2535 memcpy(dest, src, len);
2538 case XFS_ILOG_ABROOT:
2539 dest = XFS_DFORK_APTR(dip);
2540 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2541 len, (xfs_bmdr_block_t*)dest,
2542 XFS_DFORK_ASIZE(dip, mp));
2546 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2555 ASSERT(bp->b_mount == NULL || bp->b_mount == mp);
2557 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2558 xfs_bdwrite(mp, bp);
2562 return XFS_ERROR(error);
2566 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2567 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2571 xlog_recover_do_quotaoff_trans(
2573 xlog_recover_item_t *item,
2576 xfs_qoff_logformat_t *qoff_f;
2578 if (pass == XLOG_RECOVER_PASS2) {
2582 qoff_f = (xfs_qoff_logformat_t *)item->ri_buf[0].i_addr;
2586 * The logitem format's flag tells us if this was user quotaoff,
2587 * group/project quotaoff or both.
2589 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2590 log->l_quotaoffs_flag |= XFS_DQ_USER;
2591 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2592 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2593 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2594 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2600 * Recover a dquot record
2603 xlog_recover_do_dquot_trans(
2605 xlog_recover_item_t *item,
2610 struct xfs_disk_dquot *ddq, *recddq;
2612 xfs_dq_logformat_t *dq_f;
2615 if (pass == XLOG_RECOVER_PASS1) {
2621 * Filesystems are required to send in quota flags at mount time.
2623 if (mp->m_qflags == 0)
2626 recddq = (xfs_disk_dquot_t *)item->ri_buf[1].i_addr;
2628 if (item->ri_buf[1].i_addr == NULL) {
2630 "XFS: NULL dquot in %s.", __func__);
2631 return XFS_ERROR(EIO);
2633 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
2635 "XFS: dquot too small (%d) in %s.",
2636 item->ri_buf[1].i_len, __func__);
2637 return XFS_ERROR(EIO);
2641 * This type of quotas was turned off, so ignore this record.
2643 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
2645 if (log->l_quotaoffs_flag & type)
2649 * At this point we know that quota was _not_ turned off.
2650 * Since the mount flags are not indicating to us otherwise, this
2651 * must mean that quota is on, and the dquot needs to be replayed.
2652 * Remember that we may not have fully recovered the superblock yet,
2653 * so we can't do the usual trick of looking at the SB quota bits.
2655 * The other possibility, of course, is that the quota subsystem was
2656 * removed since the last mount - ENOSYS.
2658 dq_f = (xfs_dq_logformat_t *)item->ri_buf[0].i_addr;
2660 if ((error = xfs_qm_dqcheck(recddq,
2662 0, XFS_QMOPT_DOWARN,
2663 "xlog_recover_do_dquot_trans (log copy)"))) {
2664 return XFS_ERROR(EIO);
2666 ASSERT(dq_f->qlf_len == 1);
2668 error = xfs_read_buf(mp, mp->m_ddev_targp,
2670 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2673 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
2674 bp, dq_f->qlf_blkno);
2678 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2681 * At least the magic num portion should be on disk because this
2682 * was among a chunk of dquots created earlier, and we did some
2683 * minimal initialization then.
2685 if (xfs_qm_dqcheck(ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2686 "xlog_recover_do_dquot_trans")) {
2688 return XFS_ERROR(EIO);
2691 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2693 ASSERT(dq_f->qlf_size == 2);
2694 ASSERT(bp->b_mount == NULL || bp->b_mount == mp);
2696 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2697 xfs_bdwrite(mp, bp);
2703 * This routine is called to create an in-core extent free intent
2704 * item from the efi format structure which was logged on disk.
2705 * It allocates an in-core efi, copies the extents from the format
2706 * structure into it, and adds the efi to the AIL with the given
2710 xlog_recover_do_efi_trans(
2712 xlog_recover_item_t *item,
2718 xfs_efi_log_item_t *efip;
2719 xfs_efi_log_format_t *efi_formatp;
2721 if (pass == XLOG_RECOVER_PASS1) {
2725 efi_formatp = (xfs_efi_log_format_t *)item->ri_buf[0].i_addr;
2728 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
2729 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2730 &(efip->efi_format)))) {
2731 xfs_efi_item_free(efip);
2734 efip->efi_next_extent = efi_formatp->efi_nextents;
2735 efip->efi_flags |= XFS_EFI_COMMITTED;
2737 spin_lock(&log->l_ailp->xa_lock);
2739 * xfs_trans_ail_update() drops the AIL lock.
2741 xfs_trans_ail_update(log->l_ailp, (xfs_log_item_t *)efip, lsn);
2747 * This routine is called when an efd format structure is found in
2748 * a committed transaction in the log. It's purpose is to cancel
2749 * the corresponding efi if it was still in the log. To do this
2750 * it searches the AIL for the efi with an id equal to that in the
2751 * efd format structure. If we find it, we remove the efi from the
2755 xlog_recover_do_efd_trans(
2757 xlog_recover_item_t *item,
2760 xfs_efd_log_format_t *efd_formatp;
2761 xfs_efi_log_item_t *efip = NULL;
2762 xfs_log_item_t *lip;
2764 struct xfs_ail_cursor cur;
2765 struct xfs_ail *ailp = log->l_ailp;
2767 if (pass == XLOG_RECOVER_PASS1) {
2771 efd_formatp = (xfs_efd_log_format_t *)item->ri_buf[0].i_addr;
2772 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2773 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2774 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2775 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
2776 efi_id = efd_formatp->efd_efi_id;
2779 * Search for the efi with the id in the efd format structure
2782 spin_lock(&ailp->xa_lock);
2783 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
2784 while (lip != NULL) {
2785 if (lip->li_type == XFS_LI_EFI) {
2786 efip = (xfs_efi_log_item_t *)lip;
2787 if (efip->efi_format.efi_id == efi_id) {
2789 * xfs_trans_ail_delete() drops the
2792 xfs_trans_ail_delete(ailp, lip);
2793 xfs_efi_item_free(efip);
2794 spin_lock(&ailp->xa_lock);
2798 lip = xfs_trans_ail_cursor_next(ailp, &cur);
2800 xfs_trans_ail_cursor_done(ailp, &cur);
2801 spin_unlock(&ailp->xa_lock);
2805 * Perform the transaction
2807 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2808 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2811 xlog_recover_do_trans(
2813 xlog_recover_t *trans,
2817 xlog_recover_item_t *item, *first_item;
2819 error = xlog_recover_reorder_trans(trans);
2823 first_item = item = trans->r_itemq;
2825 switch (ITEM_TYPE(item)) {
2827 error = xlog_recover_do_buffer_trans(log, item, pass);
2830 error = xlog_recover_do_inode_trans(log, item, pass);
2833 error = xlog_recover_do_efi_trans(log, item,
2834 trans->r_lsn, pass);
2837 xlog_recover_do_efd_trans(log, item, pass);
2841 error = xlog_recover_do_dquot_trans(log, item, pass);
2843 case XFS_LI_QUOTAOFF:
2844 error = xlog_recover_do_quotaoff_trans(log, item,
2849 "XFS: invalid item type (%d) xlog_recover_do_trans", ITEM_TYPE(item));
2851 error = XFS_ERROR(EIO);
2857 item = item->ri_next;
2858 } while (first_item != item);
2864 * Free up any resources allocated by the transaction
2866 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2869 xlog_recover_free_trans(
2870 xlog_recover_t *trans)
2872 xlog_recover_item_t *first_item, *item, *free_item;
2875 item = first_item = trans->r_itemq;
2878 item = item->ri_next;
2879 /* Free the regions in the item. */
2880 for (i = 0; i < free_item->ri_cnt; i++) {
2881 kmem_free(free_item->ri_buf[i].i_addr);
2883 /* Free the item itself */
2884 kmem_free(free_item->ri_buf);
2885 kmem_free(free_item);
2886 } while (first_item != item);
2887 /* Free the transaction recover structure */
2892 xlog_recover_commit_trans(
2895 xlog_recover_t *trans,
2900 if ((error = xlog_recover_unlink_tid(q, trans)))
2902 if ((error = xlog_recover_do_trans(log, trans, pass)))
2904 xlog_recover_free_trans(trans); /* no error */
2909 xlog_recover_unmount_trans(
2910 xlog_recover_t *trans)
2912 /* Do nothing now */
2913 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2918 * There are two valid states of the r_state field. 0 indicates that the
2919 * transaction structure is in a normal state. We have either seen the
2920 * start of the transaction or the last operation we added was not a partial
2921 * operation. If the last operation we added to the transaction was a
2922 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2924 * NOTE: skip LRs with 0 data length.
2927 xlog_recover_process_data(
2929 xlog_recover_t *rhash[],
2930 xlog_rec_header_t *rhead,
2936 xlog_op_header_t *ohead;
2937 xlog_recover_t *trans;
2943 lp = dp + be32_to_cpu(rhead->h_len);
2944 num_logops = be32_to_cpu(rhead->h_num_logops);
2946 /* check the log format matches our own - else we can't recover */
2947 if (xlog_header_check_recover(log->l_mp, rhead))
2948 return (XFS_ERROR(EIO));
2950 while ((dp < lp) && num_logops) {
2951 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2952 ohead = (xlog_op_header_t *)dp;
2953 dp += sizeof(xlog_op_header_t);
2954 if (ohead->oh_clientid != XFS_TRANSACTION &&
2955 ohead->oh_clientid != XFS_LOG) {
2957 "XFS: xlog_recover_process_data: bad clientid");
2959 return (XFS_ERROR(EIO));
2961 tid = be32_to_cpu(ohead->oh_tid);
2962 hash = XLOG_RHASH(tid);
2963 trans = xlog_recover_find_tid(rhash[hash], tid);
2964 if (trans == NULL) { /* not found; add new tid */
2965 if (ohead->oh_flags & XLOG_START_TRANS)
2966 xlog_recover_new_tid(&rhash[hash], tid,
2967 be64_to_cpu(rhead->h_lsn));
2969 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
2971 "XFS: xlog_recover_process_data: bad length");
2973 return (XFS_ERROR(EIO));
2975 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2976 if (flags & XLOG_WAS_CONT_TRANS)
2977 flags &= ~XLOG_CONTINUE_TRANS;
2979 case XLOG_COMMIT_TRANS:
2980 error = xlog_recover_commit_trans(log,
2981 &rhash[hash], trans, pass);
2983 case XLOG_UNMOUNT_TRANS:
2984 error = xlog_recover_unmount_trans(trans);
2986 case XLOG_WAS_CONT_TRANS:
2987 error = xlog_recover_add_to_cont_trans(trans,
2988 dp, be32_to_cpu(ohead->oh_len));
2990 case XLOG_START_TRANS:
2992 "XFS: xlog_recover_process_data: bad transaction");
2994 error = XFS_ERROR(EIO);
2997 case XLOG_CONTINUE_TRANS:
2998 error = xlog_recover_add_to_trans(trans,
2999 dp, be32_to_cpu(ohead->oh_len));
3003 "XFS: xlog_recover_process_data: bad flag");
3005 error = XFS_ERROR(EIO);
3011 dp += be32_to_cpu(ohead->oh_len);
3018 * Process an extent free intent item that was recovered from
3019 * the log. We need to free the extents that it describes.
3022 xlog_recover_process_efi(
3024 xfs_efi_log_item_t *efip)
3026 xfs_efd_log_item_t *efdp;
3031 xfs_fsblock_t startblock_fsb;
3033 ASSERT(!(efip->efi_flags & XFS_EFI_RECOVERED));
3036 * First check the validity of the extents described by the
3037 * EFI. If any are bad, then assume that all are bad and
3038 * just toss the EFI.
3040 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3041 extp = &(efip->efi_format.efi_extents[i]);
3042 startblock_fsb = XFS_BB_TO_FSB(mp,
3043 XFS_FSB_TO_DADDR(mp, extp->ext_start));
3044 if ((startblock_fsb == 0) ||
3045 (extp->ext_len == 0) ||
3046 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3047 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3049 * This will pull the EFI from the AIL and
3050 * free the memory associated with it.
3052 xfs_efi_release(efip, efip->efi_format.efi_nextents);
3053 return XFS_ERROR(EIO);
3057 tp = xfs_trans_alloc(mp, 0);
3058 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
3061 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3063 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3064 extp = &(efip->efi_format.efi_extents[i]);
3065 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3068 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3072 efip->efi_flags |= XFS_EFI_RECOVERED;
3073 error = xfs_trans_commit(tp, 0);
3077 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3082 * When this is called, all of the EFIs which did not have
3083 * corresponding EFDs should be in the AIL. What we do now
3084 * is free the extents associated with each one.
3086 * Since we process the EFIs in normal transactions, they
3087 * will be removed at some point after the commit. This prevents
3088 * us from just walking down the list processing each one.
3089 * We'll use a flag in the EFI to skip those that we've already
3090 * processed and use the AIL iteration mechanism's generation
3091 * count to try to speed this up at least a bit.
3093 * When we start, we know that the EFIs are the only things in
3094 * the AIL. As we process them, however, other items are added
3095 * to the AIL. Since everything added to the AIL must come after
3096 * everything already in the AIL, we stop processing as soon as
3097 * we see something other than an EFI in the AIL.
3100 xlog_recover_process_efis(
3103 xfs_log_item_t *lip;
3104 xfs_efi_log_item_t *efip;
3106 struct xfs_ail_cursor cur;
3107 struct xfs_ail *ailp;
3110 spin_lock(&ailp->xa_lock);
3111 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3112 while (lip != NULL) {
3114 * We're done when we see something other than an EFI.
3115 * There should be no EFIs left in the AIL now.
3117 if (lip->li_type != XFS_LI_EFI) {
3119 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
3120 ASSERT(lip->li_type != XFS_LI_EFI);
3126 * Skip EFIs that we've already processed.
3128 efip = (xfs_efi_log_item_t *)lip;
3129 if (efip->efi_flags & XFS_EFI_RECOVERED) {
3130 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3134 spin_unlock(&ailp->xa_lock);
3135 error = xlog_recover_process_efi(log->l_mp, efip);
3136 spin_lock(&ailp->xa_lock);
3139 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3142 xfs_trans_ail_cursor_done(ailp, &cur);
3143 spin_unlock(&ailp->xa_lock);
3148 * This routine performs a transaction to null out a bad inode pointer
3149 * in an agi unlinked inode hash bucket.
3152 xlog_recover_clear_agi_bucket(
3154 xfs_agnumber_t agno,
3163 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3164 error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp),
3169 error = xfs_read_agi(mp, tp, agno, &agibp);
3173 agi = XFS_BUF_TO_AGI(agibp);
3174 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
3175 offset = offsetof(xfs_agi_t, agi_unlinked) +
3176 (sizeof(xfs_agino_t) * bucket);
3177 xfs_trans_log_buf(tp, agibp, offset,
3178 (offset + sizeof(xfs_agino_t) - 1));
3180 error = xfs_trans_commit(tp, 0);
3186 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3188 xfs_fs_cmn_err(CE_WARN, mp, "xlog_recover_clear_agi_bucket: "
3189 "failed to clear agi %d. Continuing.", agno);
3194 xlog_recover_process_one_iunlink(
3195 struct xfs_mount *mp,
3196 xfs_agnumber_t agno,
3200 struct xfs_buf *ibp;
3201 struct xfs_dinode *dip;
3202 struct xfs_inode *ip;
3206 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3207 error = xfs_iget(mp, NULL, ino, 0, 0, &ip, 0);
3212 * Get the on disk inode to find the next inode in the bucket.
3214 error = xfs_itobp(mp, NULL, ip, &dip, &ibp, XFS_BUF_LOCK);
3218 ASSERT(ip->i_d.di_nlink == 0);
3219 ASSERT(ip->i_d.di_mode != 0);
3221 /* setup for the next pass */
3222 agino = be32_to_cpu(dip->di_next_unlinked);
3226 * Prevent any DMAPI event from being sent when the reference on
3227 * the inode is dropped.
3229 ip->i_d.di_dmevmask = 0;
3238 * We can't read in the inode this bucket points to, or this inode
3239 * is messed up. Just ditch this bucket of inodes. We will lose
3240 * some inodes and space, but at least we won't hang.
3242 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3243 * clear the inode pointer in the bucket.
3245 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3250 * xlog_iunlink_recover
3252 * This is called during recovery to process any inodes which
3253 * we unlinked but not freed when the system crashed. These
3254 * inodes will be on the lists in the AGI blocks. What we do
3255 * here is scan all the AGIs and fully truncate and free any
3256 * inodes found on the lists. Each inode is removed from the
3257 * lists when it has been fully truncated and is freed. The
3258 * freeing of the inode and its removal from the list must be
3262 xlog_recover_process_iunlinks(
3266 xfs_agnumber_t agno;
3277 * Prevent any DMAPI event from being sent while in this function.
3279 mp_dmevmask = mp->m_dmevmask;
3282 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3284 * Find the agi for this ag.
3286 error = xfs_read_agi(mp, NULL, agno, &agibp);
3289 * AGI is b0rked. Don't process it.
3291 * We should probably mark the filesystem as corrupt
3292 * after we've recovered all the ag's we can....
3296 agi = XFS_BUF_TO_AGI(agibp);
3298 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3299 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3300 while (agino != NULLAGINO) {
3302 * Release the agi buffer so that it can
3303 * be acquired in the normal course of the
3304 * transaction to truncate and free the inode.
3306 xfs_buf_relse(agibp);
3308 agino = xlog_recover_process_one_iunlink(mp,
3309 agno, agino, bucket);
3312 * Reacquire the agibuffer and continue around
3313 * the loop. This should never fail as we know
3314 * the buffer was good earlier on.
3316 error = xfs_read_agi(mp, NULL, agno, &agibp);
3318 agi = XFS_BUF_TO_AGI(agibp);
3323 * Release the buffer for the current agi so we can
3324 * go on to the next one.
3326 xfs_buf_relse(agibp);
3329 mp->m_dmevmask = mp_dmevmask;
3335 xlog_pack_data_checksum(
3337 xlog_in_core_t *iclog,
3344 up = (__be32 *)iclog->ic_datap;
3345 /* divide length by 4 to get # words */
3346 for (i = 0; i < (size >> 2); i++) {
3347 chksum ^= be32_to_cpu(*up);
3350 iclog->ic_header.h_chksum = cpu_to_be32(chksum);
3353 #define xlog_pack_data_checksum(log, iclog, size)
3357 * Stamp cycle number in every block
3362 xlog_in_core_t *iclog,
3366 int size = iclog->ic_offset + roundoff;
3370 xlog_pack_data_checksum(log, iclog, size);
3372 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3374 dp = iclog->ic_datap;
3375 for (i = 0; i < BTOBB(size) &&
3376 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3377 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
3378 *(__be32 *)dp = cycle_lsn;
3382 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3383 xlog_in_core_2_t *xhdr = iclog->ic_data;
3385 for ( ; i < BTOBB(size); i++) {
3386 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3387 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3388 xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
3389 *(__be32 *)dp = cycle_lsn;
3393 for (i = 1; i < log->l_iclog_heads; i++) {
3394 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3399 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
3401 xlog_unpack_data_checksum(
3402 xlog_rec_header_t *rhead,
3406 __be32 *up = (__be32 *)dp;
3410 /* divide length by 4 to get # words */
3411 for (i=0; i < be32_to_cpu(rhead->h_len) >> 2; i++) {
3412 chksum ^= be32_to_cpu(*up);
3415 if (chksum != be32_to_cpu(rhead->h_chksum)) {
3416 if (rhead->h_chksum ||
3417 ((log->l_flags & XLOG_CHKSUM_MISMATCH) == 0)) {
3419 "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n",
3420 be32_to_cpu(rhead->h_chksum), chksum);
3422 "XFS: Disregard message if filesystem was created with non-DEBUG kernel");
3423 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3425 "XFS: LogR this is a LogV2 filesystem\n");
3427 log->l_flags |= XLOG_CHKSUM_MISMATCH;
3432 #define xlog_unpack_data_checksum(rhead, dp, log)
3437 xlog_rec_header_t *rhead,
3443 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
3444 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3445 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
3449 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3450 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
3451 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
3452 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3453 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3454 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
3459 xlog_unpack_data_checksum(rhead, dp, log);
3463 xlog_valid_rec_header(
3465 xlog_rec_header_t *rhead,
3470 if (unlikely(be32_to_cpu(rhead->h_magicno) != XLOG_HEADER_MAGIC_NUM)) {
3471 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3472 XFS_ERRLEVEL_LOW, log->l_mp);
3473 return XFS_ERROR(EFSCORRUPTED);
3476 (!rhead->h_version ||
3477 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
3478 xlog_warn("XFS: %s: unrecognised log version (%d).",
3479 __func__, be32_to_cpu(rhead->h_version));
3480 return XFS_ERROR(EIO);
3483 /* LR body must have data or it wouldn't have been written */
3484 hlen = be32_to_cpu(rhead->h_len);
3485 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3486 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3487 XFS_ERRLEVEL_LOW, log->l_mp);
3488 return XFS_ERROR(EFSCORRUPTED);
3490 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3491 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3492 XFS_ERRLEVEL_LOW, log->l_mp);
3493 return XFS_ERROR(EFSCORRUPTED);
3499 * Read the log from tail to head and process the log records found.
3500 * Handle the two cases where the tail and head are in the same cycle
3501 * and where the active portion of the log wraps around the end of
3502 * the physical log separately. The pass parameter is passed through
3503 * to the routines called to process the data and is not looked at
3507 xlog_do_recovery_pass(
3509 xfs_daddr_t head_blk,
3510 xfs_daddr_t tail_blk,
3513 xlog_rec_header_t *rhead;
3516 xfs_buf_t *hbp, *dbp;
3517 int error = 0, h_size;
3518 int bblks, split_bblks;
3519 int hblks, split_hblks, wrapped_hblks;
3520 xlog_recover_t *rhash[XLOG_RHASH_SIZE];
3522 ASSERT(head_blk != tail_blk);
3525 * Read the header of the tail block and get the iclog buffer size from
3526 * h_size. Use this to tell how many sectors make up the log header.
3528 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3530 * When using variable length iclogs, read first sector of
3531 * iclog header and extract the header size from it. Get a
3532 * new hbp that is the correct size.
3534 hbp = xlog_get_bp(log, 1);
3538 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
3542 rhead = (xlog_rec_header_t *)offset;
3543 error = xlog_valid_rec_header(log, rhead, tail_blk);
3546 h_size = be32_to_cpu(rhead->h_size);
3547 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
3548 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3549 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3550 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3553 hbp = xlog_get_bp(log, hblks);
3558 ASSERT(log->l_sectbb_log == 0);
3560 hbp = xlog_get_bp(log, 1);
3561 h_size = XLOG_BIG_RECORD_BSIZE;
3566 dbp = xlog_get_bp(log, BTOBB(h_size));
3572 memset(rhash, 0, sizeof(rhash));
3573 if (tail_blk <= head_blk) {
3574 for (blk_no = tail_blk; blk_no < head_blk; ) {
3575 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3579 rhead = (xlog_rec_header_t *)offset;
3580 error = xlog_valid_rec_header(log, rhead, blk_no);
3584 /* blocks in data section */
3585 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3586 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
3591 xlog_unpack_data(rhead, offset, log);
3592 if ((error = xlog_recover_process_data(log,
3593 rhash, rhead, offset, pass)))
3595 blk_no += bblks + hblks;
3599 * Perform recovery around the end of the physical log.
3600 * When the head is not on the same cycle number as the tail,
3601 * we can't do a sequential recovery as above.
3604 while (blk_no < log->l_logBBsize) {
3606 * Check for header wrapping around physical end-of-log
3608 offset = XFS_BUF_PTR(hbp);
3611 if (blk_no + hblks <= log->l_logBBsize) {
3612 /* Read header in one read */
3613 error = xlog_bread(log, blk_no, hblks, hbp,
3618 /* This LR is split across physical log end */
3619 if (blk_no != log->l_logBBsize) {
3620 /* some data before physical log end */
3621 ASSERT(blk_no <= INT_MAX);
3622 split_hblks = log->l_logBBsize - (int)blk_no;
3623 ASSERT(split_hblks > 0);
3624 error = xlog_bread(log, blk_no,
3632 * Note: this black magic still works with
3633 * large sector sizes (non-512) only because:
3634 * - we increased the buffer size originally
3635 * by 1 sector giving us enough extra space
3636 * for the second read;
3637 * - the log start is guaranteed to be sector
3639 * - we read the log end (LR header start)
3640 * _first_, then the log start (LR header end)
3641 * - order is important.
3643 wrapped_hblks = hblks - split_hblks;
3644 error = XFS_BUF_SET_PTR(hbp,
3645 offset + BBTOB(split_hblks),
3646 BBTOB(hblks - split_hblks));
3650 error = xlog_bread_noalign(log, 0,
3651 wrapped_hblks, hbp);
3655 error = XFS_BUF_SET_PTR(hbp, offset,
3660 rhead = (xlog_rec_header_t *)offset;
3661 error = xlog_valid_rec_header(log, rhead,
3662 split_hblks ? blk_no : 0);
3666 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3669 /* Read in data for log record */
3670 if (blk_no + bblks <= log->l_logBBsize) {
3671 error = xlog_bread(log, blk_no, bblks, dbp,
3676 /* This log record is split across the
3677 * physical end of log */
3678 offset = XFS_BUF_PTR(dbp);
3680 if (blk_no != log->l_logBBsize) {
3681 /* some data is before the physical
3683 ASSERT(!wrapped_hblks);
3684 ASSERT(blk_no <= INT_MAX);
3686 log->l_logBBsize - (int)blk_no;
3687 ASSERT(split_bblks > 0);
3688 error = xlog_bread(log, blk_no,
3696 * Note: this black magic still works with
3697 * large sector sizes (non-512) only because:
3698 * - we increased the buffer size originally
3699 * by 1 sector giving us enough extra space
3700 * for the second read;
3701 * - the log start is guaranteed to be sector
3703 * - we read the log end (LR header start)
3704 * _first_, then the log start (LR header end)
3705 * - order is important.
3707 error = XFS_BUF_SET_PTR(dbp,
3708 offset + BBTOB(split_bblks),
3709 BBTOB(bblks - split_bblks));
3713 error = xlog_bread_noalign(log, wrapped_hblks,
3714 bblks - split_bblks,
3719 error = XFS_BUF_SET_PTR(dbp, offset, h_size);
3723 xlog_unpack_data(rhead, offset, log);
3724 if ((error = xlog_recover_process_data(log, rhash,
3725 rhead, offset, pass)))
3730 ASSERT(blk_no >= log->l_logBBsize);
3731 blk_no -= log->l_logBBsize;
3733 /* read first part of physical log */
3734 while (blk_no < head_blk) {
3735 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3739 rhead = (xlog_rec_header_t *)offset;
3740 error = xlog_valid_rec_header(log, rhead, blk_no);
3744 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3745 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
3750 xlog_unpack_data(rhead, offset, log);
3751 if ((error = xlog_recover_process_data(log, rhash,
3752 rhead, offset, pass)))
3754 blk_no += bblks + hblks;
3766 * Do the recovery of the log. We actually do this in two phases.
3767 * The two passes are necessary in order to implement the function
3768 * of cancelling a record written into the log. The first pass
3769 * determines those things which have been cancelled, and the
3770 * second pass replays log items normally except for those which
3771 * have been cancelled. The handling of the replay and cancellations
3772 * takes place in the log item type specific routines.
3774 * The table of items which have cancel records in the log is allocated
3775 * and freed at this level, since only here do we know when all of
3776 * the log recovery has been completed.
3779 xlog_do_log_recovery(
3781 xfs_daddr_t head_blk,
3782 xfs_daddr_t tail_blk)
3786 ASSERT(head_blk != tail_blk);
3789 * First do a pass to find all of the cancelled buf log items.
3790 * Store them in the buf_cancel_table for use in the second pass.
3792 log->l_buf_cancel_table =
3793 (xfs_buf_cancel_t **)kmem_zalloc(XLOG_BC_TABLE_SIZE *
3794 sizeof(xfs_buf_cancel_t*),
3796 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3797 XLOG_RECOVER_PASS1);
3799 kmem_free(log->l_buf_cancel_table);
3800 log->l_buf_cancel_table = NULL;
3804 * Then do a second pass to actually recover the items in the log.
3805 * When it is complete free the table of buf cancel items.
3807 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3808 XLOG_RECOVER_PASS2);
3813 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3814 ASSERT(log->l_buf_cancel_table[i] == NULL);
3818 kmem_free(log->l_buf_cancel_table);
3819 log->l_buf_cancel_table = NULL;
3825 * Do the actual recovery
3830 xfs_daddr_t head_blk,
3831 xfs_daddr_t tail_blk)
3838 * First replay the images in the log.
3840 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3845 XFS_bflush(log->l_mp->m_ddev_targp);
3848 * If IO errors happened during recovery, bail out.
3850 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3855 * We now update the tail_lsn since much of the recovery has completed
3856 * and there may be space available to use. If there were no extent
3857 * or iunlinks, we can free up the entire log and set the tail_lsn to
3858 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3859 * lsn of the last known good LR on disk. If there are extent frees
3860 * or iunlinks they will have some entries in the AIL; so we look at
3861 * the AIL to determine how to set the tail_lsn.
3863 xlog_assign_tail_lsn(log->l_mp);
3866 * Now that we've finished replaying all buffer and inode
3867 * updates, re-read in the superblock.
3869 bp = xfs_getsb(log->l_mp, 0);
3871 ASSERT(!(XFS_BUF_ISWRITE(bp)));
3872 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
3874 XFS_BUF_UNASYNC(bp);
3875 xfsbdstrat(log->l_mp, bp);
3876 error = xfs_iowait(bp);
3878 xfs_ioerror_alert("xlog_do_recover",
3879 log->l_mp, bp, XFS_BUF_ADDR(bp));
3885 /* Convert superblock from on-disk format */
3886 sbp = &log->l_mp->m_sb;
3887 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
3888 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
3889 ASSERT(xfs_sb_good_version(sbp));
3892 /* We've re-read the superblock so re-initialize per-cpu counters */
3893 xfs_icsb_reinit_counters(log->l_mp);
3895 xlog_recover_check_summary(log);
3897 /* Normal transactions can now occur */
3898 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3903 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3905 * Return error or zero.
3911 xfs_daddr_t head_blk, tail_blk;
3914 /* find the tail of the log */
3915 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
3918 if (tail_blk != head_blk) {
3919 /* There used to be a comment here:
3921 * disallow recovery on read-only mounts. note -- mount
3922 * checks for ENOSPC and turns it into an intelligent
3924 * ...but this is no longer true. Now, unless you specify
3925 * NORECOVERY (in which case this function would never be
3926 * called), we just go ahead and recover. We do this all
3927 * under the vfs layer, so we can get away with it unless
3928 * the device itself is read-only, in which case we fail.
3930 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
3935 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3936 log->l_mp->m_fsname, log->l_mp->m_logname ?
3937 log->l_mp->m_logname : "internal");
3939 error = xlog_do_recover(log, head_blk, tail_blk);
3940 log->l_flags |= XLOG_RECOVERY_NEEDED;
3946 * In the first part of recovery we replay inodes and buffers and build
3947 * up the list of extent free items which need to be processed. Here
3948 * we process the extent free items and clean up the on disk unlinked
3949 * inode lists. This is separated from the first part of recovery so
3950 * that the root and real-time bitmap inodes can be read in from disk in
3951 * between the two stages. This is necessary so that we can free space
3952 * in the real-time portion of the file system.
3955 xlog_recover_finish(
3959 * Now we're ready to do the transactions needed for the
3960 * rest of recovery. Start with completing all the extent
3961 * free intent records and then process the unlinked inode
3962 * lists. At this point, we essentially run in normal mode
3963 * except that we're still performing recovery actions
3964 * rather than accepting new requests.
3966 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
3968 error = xlog_recover_process_efis(log);
3971 "Failed to recover EFIs on filesystem: %s",
3972 log->l_mp->m_fsname);
3976 * Sync the log to get all the EFIs out of the AIL.
3977 * This isn't absolutely necessary, but it helps in
3978 * case the unlink transactions would have problems
3979 * pushing the EFIs out of the way.
3981 xfs_log_force(log->l_mp, (xfs_lsn_t)0,
3982 (XFS_LOG_FORCE | XFS_LOG_SYNC));
3984 xlog_recover_process_iunlinks(log);
3986 xlog_recover_check_summary(log);
3989 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3990 log->l_mp->m_fsname, log->l_mp->m_logname ?
3991 log->l_mp->m_logname : "internal");
3992 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
3995 "!Ending clean XFS mount for filesystem: %s\n",
3996 log->l_mp->m_fsname);
4004 * Read all of the agf and agi counters and check that they
4005 * are consistent with the superblock counters.
4008 xlog_recover_check_summary(
4016 #ifdef XFS_LOUD_RECOVERY
4019 xfs_agnumber_t agno;
4020 __uint64_t freeblks;
4030 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
4031 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
4033 xfs_fs_cmn_err(CE_ALERT, mp,
4034 "xlog_recover_check_summary(agf)"
4035 "agf read failed agno %d error %d",
4038 agfp = XFS_BUF_TO_AGF(agfbp);
4039 freeblks += be32_to_cpu(agfp->agf_freeblks) +
4040 be32_to_cpu(agfp->agf_flcount);
4041 xfs_buf_relse(agfbp);
4044 error = xfs_read_agi(mp, NULL, agno, &agibp);
4046 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
4048 itotal += be32_to_cpu(agi->agi_count);
4049 ifree += be32_to_cpu(agi->agi_freecount);
4050 xfs_buf_relse(agibp);
4054 sbbp = xfs_getsb(mp, 0);
4055 #ifdef XFS_LOUD_RECOVERY
4057 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(sbbp));
4059 "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
4060 sbp->sb_icount, itotal);
4062 "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
4063 sbp->sb_ifree, ifree);
4065 "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
4066 sbp->sb_fdblocks, freeblks);
4069 * This is turned off until I account for the allocation
4070 * btree blocks which live in free space.
4072 ASSERT(sbp->sb_icount == itotal);
4073 ASSERT(sbp->sb_ifree == ifree);
4074 ASSERT(sbp->sb_fdblocks == freeblks);
4077 xfs_buf_relse(sbbp);