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"
27 #include "xfs_mount.h"
28 #include "xfs_error.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_alloc.h"
36 #include "xfs_ialloc.h"
37 #include "xfs_log_priv.h"
38 #include "xfs_buf_item.h"
39 #include "xfs_log_recover.h"
40 #include "xfs_extfree_item.h"
41 #include "xfs_trans_priv.h"
42 #include "xfs_quota.h"
44 #include "xfs_utils.h"
45 #include "xfs_trace.h"
47 STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *);
48 STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t);
50 STATIC void xlog_recover_check_summary(xlog_t *);
52 #define xlog_recover_check_summary(log)
56 * This structure is used during recovery to record the buf log items which
57 * have been canceled and should not be replayed.
59 struct xfs_buf_cancel {
63 struct list_head bc_list;
67 * Sector aligned buffer routines for buffer create/read/write/access
71 * Verify the given count of basic blocks is valid number of blocks
72 * to specify for an operation involving the given XFS log buffer.
73 * Returns nonzero if the count is valid, 0 otherwise.
77 xlog_buf_bbcount_valid(
81 return bbcount > 0 && bbcount <= log->l_logBBsize;
85 * Allocate a buffer to hold log data. The buffer needs to be able
86 * to map to a range of nbblks basic blocks at any valid (basic
87 * block) offset within the log.
96 if (!xlog_buf_bbcount_valid(log, nbblks)) {
97 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
99 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
104 * We do log I/O in units of log sectors (a power-of-2
105 * multiple of the basic block size), so we round up the
106 * requested size to accommodate the basic blocks required
107 * for complete log sectors.
109 * In addition, the buffer may be used for a non-sector-
110 * aligned block offset, in which case an I/O of the
111 * requested size could extend beyond the end of the
112 * buffer. If the requested size is only 1 basic block it
113 * will never straddle a sector boundary, so this won't be
114 * an issue. Nor will this be a problem if the log I/O is
115 * done in basic blocks (sector size 1). But otherwise we
116 * extend the buffer by one extra log sector to ensure
117 * there's space to accommodate this possibility.
119 if (nbblks > 1 && log->l_sectBBsize > 1)
120 nbblks += log->l_sectBBsize;
121 nbblks = round_up(nbblks, log->l_sectBBsize);
123 bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, BBTOB(nbblks), 0);
137 * Return the address of the start of the given block number's data
138 * in a log buffer. The buffer covers a log sector-aligned region.
147 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
149 ASSERT(BBTOB(offset + nbblks) <= XFS_BUF_SIZE(bp));
150 return bp->b_addr + BBTOB(offset);
155 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
166 if (!xlog_buf_bbcount_valid(log, nbblks)) {
167 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
169 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
173 blk_no = round_down(blk_no, log->l_sectBBsize);
174 nbblks = round_up(nbblks, log->l_sectBBsize);
177 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
179 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
181 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
183 xfsbdstrat(log->l_mp, bp);
184 error = xfs_buf_iowait(bp);
186 xfs_buf_ioerror_alert(bp, __func__);
200 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
204 *offset = xlog_align(log, blk_no, nbblks, bp);
209 * Read at an offset into the buffer. Returns with the buffer in it's original
210 * state regardless of the result of the read.
215 xfs_daddr_t blk_no, /* block to read from */
216 int nbblks, /* blocks to read */
220 xfs_caddr_t orig_offset = bp->b_addr;
221 int orig_len = bp->b_buffer_length;
224 error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks));
228 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
230 /* must reset buffer pointer even on error */
231 error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len);
238 * Write out the buffer at the given block for the given number of blocks.
239 * The buffer is kept locked across the write and is returned locked.
240 * This can only be used for synchronous log writes.
251 if (!xlog_buf_bbcount_valid(log, nbblks)) {
252 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
254 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
258 blk_no = round_down(blk_no, log->l_sectBBsize);
259 nbblks = round_up(nbblks, log->l_sectBBsize);
262 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
264 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
265 XFS_BUF_ZEROFLAGS(bp);
268 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
270 error = xfs_bwrite(bp);
272 xfs_buf_ioerror_alert(bp, __func__);
279 * dump debug superblock and log record information
282 xlog_header_check_dump(
284 xlog_rec_header_t *head)
286 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d\n",
287 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
288 xfs_debug(mp, " log : uuid = %pU, fmt = %d\n",
289 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
292 #define xlog_header_check_dump(mp, head)
296 * check log record header for recovery
299 xlog_header_check_recover(
301 xlog_rec_header_t *head)
303 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
306 * IRIX doesn't write the h_fmt field and leaves it zeroed
307 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
308 * a dirty log created in IRIX.
310 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
312 "dirty log written in incompatible format - can't recover");
313 xlog_header_check_dump(mp, head);
314 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
315 XFS_ERRLEVEL_HIGH, mp);
316 return XFS_ERROR(EFSCORRUPTED);
317 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
319 "dirty log entry has mismatched uuid - can't recover");
320 xlog_header_check_dump(mp, head);
321 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
322 XFS_ERRLEVEL_HIGH, mp);
323 return XFS_ERROR(EFSCORRUPTED);
329 * read the head block of the log and check the header
332 xlog_header_check_mount(
334 xlog_rec_header_t *head)
336 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
338 if (uuid_is_nil(&head->h_fs_uuid)) {
340 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
341 * h_fs_uuid is nil, we assume this log was last mounted
342 * by IRIX and continue.
344 xfs_warn(mp, "nil uuid in log - IRIX style log");
345 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
346 xfs_warn(mp, "log has mismatched uuid - can't recover");
347 xlog_header_check_dump(mp, head);
348 XFS_ERROR_REPORT("xlog_header_check_mount",
349 XFS_ERRLEVEL_HIGH, mp);
350 return XFS_ERROR(EFSCORRUPTED);
361 * We're not going to bother about retrying
362 * this during recovery. One strike!
364 xfs_buf_ioerror_alert(bp, __func__);
365 xfs_force_shutdown(bp->b_target->bt_mount,
366 SHUTDOWN_META_IO_ERROR);
369 xfs_buf_ioend(bp, 0);
373 * This routine finds (to an approximation) the first block in the physical
374 * log which contains the given cycle. It uses a binary search algorithm.
375 * Note that the algorithm can not be perfect because the disk will not
376 * necessarily be perfect.
379 xlog_find_cycle_start(
382 xfs_daddr_t first_blk,
383 xfs_daddr_t *last_blk,
393 mid_blk = BLK_AVG(first_blk, end_blk);
394 while (mid_blk != first_blk && mid_blk != end_blk) {
395 error = xlog_bread(log, mid_blk, 1, bp, &offset);
398 mid_cycle = xlog_get_cycle(offset);
399 if (mid_cycle == cycle)
400 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
402 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
403 mid_blk = BLK_AVG(first_blk, end_blk);
405 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
406 (mid_blk == end_blk && mid_blk-1 == first_blk));
414 * Check that a range of blocks does not contain stop_on_cycle_no.
415 * Fill in *new_blk with the block offset where such a block is
416 * found, or with -1 (an invalid block number) if there is no such
417 * block in the range. The scan needs to occur from front to back
418 * and the pointer into the region must be updated since a later
419 * routine will need to perform another test.
422 xlog_find_verify_cycle(
424 xfs_daddr_t start_blk,
426 uint stop_on_cycle_no,
427 xfs_daddr_t *new_blk)
433 xfs_caddr_t buf = NULL;
437 * Greedily allocate a buffer big enough to handle the full
438 * range of basic blocks we'll be examining. If that fails,
439 * try a smaller size. We need to be able to read at least
440 * a log sector, or we're out of luck.
442 bufblks = 1 << ffs(nbblks);
443 while (bufblks > log->l_logBBsize)
445 while (!(bp = xlog_get_bp(log, bufblks))) {
447 if (bufblks < log->l_sectBBsize)
451 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
454 bcount = min(bufblks, (start_blk + nbblks - i));
456 error = xlog_bread(log, i, bcount, bp, &buf);
460 for (j = 0; j < bcount; j++) {
461 cycle = xlog_get_cycle(buf);
462 if (cycle == stop_on_cycle_no) {
479 * Potentially backup over partial log record write.
481 * In the typical case, last_blk is the number of the block directly after
482 * a good log record. Therefore, we subtract one to get the block number
483 * of the last block in the given buffer. extra_bblks contains the number
484 * of blocks we would have read on a previous read. This happens when the
485 * last log record is split over the end of the physical log.
487 * extra_bblks is the number of blocks potentially verified on a previous
488 * call to this routine.
491 xlog_find_verify_log_record(
493 xfs_daddr_t start_blk,
494 xfs_daddr_t *last_blk,
499 xfs_caddr_t offset = NULL;
500 xlog_rec_header_t *head = NULL;
503 int num_blks = *last_blk - start_blk;
506 ASSERT(start_blk != 0 || *last_blk != start_blk);
508 if (!(bp = xlog_get_bp(log, num_blks))) {
509 if (!(bp = xlog_get_bp(log, 1)))
513 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
516 offset += ((num_blks - 1) << BBSHIFT);
519 for (i = (*last_blk) - 1; i >= 0; i--) {
521 /* valid log record not found */
523 "Log inconsistent (didn't find previous header)");
525 error = XFS_ERROR(EIO);
530 error = xlog_bread(log, i, 1, bp, &offset);
535 head = (xlog_rec_header_t *)offset;
537 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
545 * We hit the beginning of the physical log & still no header. Return
546 * to caller. If caller can handle a return of -1, then this routine
547 * will be called again for the end of the physical log.
555 * We have the final block of the good log (the first block
556 * of the log record _before_ the head. So we check the uuid.
558 if ((error = xlog_header_check_mount(log->l_mp, head)))
562 * We may have found a log record header before we expected one.
563 * last_blk will be the 1st block # with a given cycle #. We may end
564 * up reading an entire log record. In this case, we don't want to
565 * reset last_blk. Only when last_blk points in the middle of a log
566 * record do we update last_blk.
568 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
569 uint h_size = be32_to_cpu(head->h_size);
571 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
572 if (h_size % XLOG_HEADER_CYCLE_SIZE)
578 if (*last_blk - i + extra_bblks !=
579 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
588 * Head is defined to be the point of the log where the next log write
589 * write could go. This means that incomplete LR writes at the end are
590 * eliminated when calculating the head. We aren't guaranteed that previous
591 * LR have complete transactions. We only know that a cycle number of
592 * current cycle number -1 won't be present in the log if we start writing
593 * from our current block number.
595 * last_blk contains the block number of the first block with a given
598 * Return: zero if normal, non-zero if error.
603 xfs_daddr_t *return_head_blk)
607 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
609 uint first_half_cycle, last_half_cycle;
611 int error, log_bbnum = log->l_logBBsize;
613 /* Is the end of the log device zeroed? */
614 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
615 *return_head_blk = first_blk;
617 /* Is the whole lot zeroed? */
619 /* Linux XFS shouldn't generate totally zeroed logs -
620 * mkfs etc write a dummy unmount record to a fresh
621 * log so we can store the uuid in there
623 xfs_warn(log->l_mp, "totally zeroed log");
628 xfs_warn(log->l_mp, "empty log check failed");
632 first_blk = 0; /* get cycle # of 1st block */
633 bp = xlog_get_bp(log, 1);
637 error = xlog_bread(log, 0, 1, bp, &offset);
641 first_half_cycle = xlog_get_cycle(offset);
643 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
644 error = xlog_bread(log, last_blk, 1, bp, &offset);
648 last_half_cycle = xlog_get_cycle(offset);
649 ASSERT(last_half_cycle != 0);
652 * If the 1st half cycle number is equal to the last half cycle number,
653 * then the entire log is stamped with the same cycle number. In this
654 * case, head_blk can't be set to zero (which makes sense). The below
655 * math doesn't work out properly with head_blk equal to zero. Instead,
656 * we set it to log_bbnum which is an invalid block number, but this
657 * value makes the math correct. If head_blk doesn't changed through
658 * all the tests below, *head_blk is set to zero at the very end rather
659 * than log_bbnum. In a sense, log_bbnum and zero are the same block
660 * in a circular file.
662 if (first_half_cycle == last_half_cycle) {
664 * In this case we believe that the entire log should have
665 * cycle number last_half_cycle. We need to scan backwards
666 * from the end verifying that there are no holes still
667 * containing last_half_cycle - 1. If we find such a hole,
668 * then the start of that hole will be the new head. The
669 * simple case looks like
670 * x | x ... | x - 1 | x
671 * Another case that fits this picture would be
672 * x | x + 1 | x ... | x
673 * In this case the head really is somewhere at the end of the
674 * log, as one of the latest writes at the beginning was
677 * x | x + 1 | x ... | x - 1 | x
678 * This is really the combination of the above two cases, and
679 * the head has to end up at the start of the x-1 hole at the
682 * In the 256k log case, we will read from the beginning to the
683 * end of the log and search for cycle numbers equal to x-1.
684 * We don't worry about the x+1 blocks that we encounter,
685 * because we know that they cannot be the head since the log
688 head_blk = log_bbnum;
689 stop_on_cycle = last_half_cycle - 1;
692 * In this case we want to find the first block with cycle
693 * number matching last_half_cycle. We expect the log to be
695 * x + 1 ... | x ... | x
696 * The first block with cycle number x (last_half_cycle) will
697 * be where the new head belongs. First we do a binary search
698 * for the first occurrence of last_half_cycle. The binary
699 * search may not be totally accurate, so then we scan back
700 * from there looking for occurrences of last_half_cycle before
701 * us. If that backwards scan wraps around the beginning of
702 * the log, then we look for occurrences of last_half_cycle - 1
703 * at the end of the log. The cases we're looking for look
705 * v binary search stopped here
706 * x + 1 ... | x | x + 1 | x ... | x
707 * ^ but we want to locate this spot
709 * <---------> less than scan distance
710 * x + 1 ... | x ... | x - 1 | x
711 * ^ we want to locate this spot
713 stop_on_cycle = last_half_cycle;
714 if ((error = xlog_find_cycle_start(log, bp, first_blk,
715 &head_blk, last_half_cycle)))
720 * Now validate the answer. Scan back some number of maximum possible
721 * blocks and make sure each one has the expected cycle number. The
722 * maximum is determined by the total possible amount of buffering
723 * in the in-core log. The following number can be made tighter if
724 * we actually look at the block size of the filesystem.
726 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
727 if (head_blk >= num_scan_bblks) {
729 * We are guaranteed that the entire check can be performed
732 start_blk = head_blk - num_scan_bblks;
733 if ((error = xlog_find_verify_cycle(log,
734 start_blk, num_scan_bblks,
735 stop_on_cycle, &new_blk)))
739 } else { /* need to read 2 parts of log */
741 * We are going to scan backwards in the log in two parts.
742 * First we scan the physical end of the log. In this part
743 * of the log, we are looking for blocks with cycle number
744 * last_half_cycle - 1.
745 * If we find one, then we know that the log starts there, as
746 * we've found a hole that didn't get written in going around
747 * the end of the physical log. The simple case for this is
748 * x + 1 ... | x ... | x - 1 | x
749 * <---------> less than scan distance
750 * If all of the blocks at the end of the log have cycle number
751 * last_half_cycle, then we check the blocks at the start of
752 * the log looking for occurrences of last_half_cycle. If we
753 * find one, then our current estimate for the location of the
754 * first occurrence of last_half_cycle is wrong and we move
755 * back to the hole we've found. This case looks like
756 * x + 1 ... | x | x + 1 | x ...
757 * ^ binary search stopped here
758 * Another case we need to handle that only occurs in 256k
760 * x + 1 ... | x ... | x+1 | x ...
761 * ^ binary search stops here
762 * In a 256k log, the scan at the end of the log will see the
763 * x + 1 blocks. We need to skip past those since that is
764 * certainly not the head of the log. By searching for
765 * last_half_cycle-1 we accomplish that.
767 ASSERT(head_blk <= INT_MAX &&
768 (xfs_daddr_t) num_scan_bblks >= head_blk);
769 start_blk = log_bbnum - (num_scan_bblks - head_blk);
770 if ((error = xlog_find_verify_cycle(log, start_blk,
771 num_scan_bblks - (int)head_blk,
772 (stop_on_cycle - 1), &new_blk)))
780 * Scan beginning of log now. The last part of the physical
781 * log is good. This scan needs to verify that it doesn't find
782 * the last_half_cycle.
785 ASSERT(head_blk <= INT_MAX);
786 if ((error = xlog_find_verify_cycle(log,
787 start_blk, (int)head_blk,
788 stop_on_cycle, &new_blk)))
796 * Now we need to make sure head_blk is not pointing to a block in
797 * the middle of a log record.
799 num_scan_bblks = XLOG_REC_SHIFT(log);
800 if (head_blk >= num_scan_bblks) {
801 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
803 /* start ptr at last block ptr before head_blk */
804 if ((error = xlog_find_verify_log_record(log, start_blk,
805 &head_blk, 0)) == -1) {
806 error = XFS_ERROR(EIO);
812 ASSERT(head_blk <= INT_MAX);
813 if ((error = xlog_find_verify_log_record(log, start_blk,
814 &head_blk, 0)) == -1) {
815 /* We hit the beginning of the log during our search */
816 start_blk = log_bbnum - (num_scan_bblks - head_blk);
818 ASSERT(start_blk <= INT_MAX &&
819 (xfs_daddr_t) log_bbnum-start_blk >= 0);
820 ASSERT(head_blk <= INT_MAX);
821 if ((error = xlog_find_verify_log_record(log,
823 (int)head_blk)) == -1) {
824 error = XFS_ERROR(EIO);
828 if (new_blk != log_bbnum)
835 if (head_blk == log_bbnum)
836 *return_head_blk = 0;
838 *return_head_blk = head_blk;
840 * When returning here, we have a good block number. Bad block
841 * means that during a previous crash, we didn't have a clean break
842 * from cycle number N to cycle number N-1. In this case, we need
843 * to find the first block with cycle number N-1.
851 xfs_warn(log->l_mp, "failed to find log head");
856 * Find the sync block number or the tail of the log.
858 * This will be the block number of the last record to have its
859 * associated buffers synced to disk. Every log record header has
860 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
861 * to get a sync block number. The only concern is to figure out which
862 * log record header to believe.
864 * The following algorithm uses the log record header with the largest
865 * lsn. The entire log record does not need to be valid. We only care
866 * that the header is valid.
868 * We could speed up search by using current head_blk buffer, but it is not
874 xfs_daddr_t *head_blk,
875 xfs_daddr_t *tail_blk)
877 xlog_rec_header_t *rhead;
878 xlog_op_header_t *op_head;
879 xfs_caddr_t offset = NULL;
882 xfs_daddr_t umount_data_blk;
883 xfs_daddr_t after_umount_blk;
890 * Find previous log record
892 if ((error = xlog_find_head(log, head_blk)))
895 bp = xlog_get_bp(log, 1);
898 if (*head_blk == 0) { /* special case */
899 error = xlog_bread(log, 0, 1, bp, &offset);
903 if (xlog_get_cycle(offset) == 0) {
905 /* leave all other log inited values alone */
911 * Search backwards looking for log record header block
913 ASSERT(*head_blk < INT_MAX);
914 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
915 error = xlog_bread(log, i, 1, bp, &offset);
919 if (*(__be32 *)offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
925 * If we haven't found the log record header block, start looking
926 * again from the end of the physical log. XXXmiken: There should be
927 * a check here to make sure we didn't search more than N blocks in
931 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
932 error = xlog_bread(log, i, 1, bp, &offset);
936 if (*(__be32 *)offset ==
937 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
944 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
946 return XFS_ERROR(EIO);
949 /* find blk_no of tail of log */
950 rhead = (xlog_rec_header_t *)offset;
951 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
954 * Reset log values according to the state of the log when we
955 * crashed. In the case where head_blk == 0, we bump curr_cycle
956 * one because the next write starts a new cycle rather than
957 * continuing the cycle of the last good log record. At this
958 * point we have guaranteed that all partial log records have been
959 * accounted for. Therefore, we know that the last good log record
960 * written was complete and ended exactly on the end boundary
961 * of the physical log.
963 log->l_prev_block = i;
964 log->l_curr_block = (int)*head_blk;
965 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
968 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
969 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
970 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
971 BBTOB(log->l_curr_block));
972 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
973 BBTOB(log->l_curr_block));
976 * Look for unmount record. If we find it, then we know there
977 * was a clean unmount. Since 'i' could be the last block in
978 * the physical log, we convert to a log block before comparing
981 * Save the current tail lsn to use to pass to
982 * xlog_clear_stale_blocks() below. We won't want to clear the
983 * unmount record if there is one, so we pass the lsn of the
984 * unmount record rather than the block after it.
986 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
987 int h_size = be32_to_cpu(rhead->h_size);
988 int h_version = be32_to_cpu(rhead->h_version);
990 if ((h_version & XLOG_VERSION_2) &&
991 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
992 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
993 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1001 after_umount_blk = (i + hblks + (int)
1002 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
1003 tail_lsn = atomic64_read(&log->l_tail_lsn);
1004 if (*head_blk == after_umount_blk &&
1005 be32_to_cpu(rhead->h_num_logops) == 1) {
1006 umount_data_blk = (i + hblks) % log->l_logBBsize;
1007 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1011 op_head = (xlog_op_header_t *)offset;
1012 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1014 * Set tail and last sync so that newly written
1015 * log records will point recovery to after the
1016 * current unmount record.
1018 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1019 log->l_curr_cycle, after_umount_blk);
1020 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1021 log->l_curr_cycle, after_umount_blk);
1022 *tail_blk = after_umount_blk;
1025 * Note that the unmount was clean. If the unmount
1026 * was not clean, we need to know this to rebuild the
1027 * superblock counters from the perag headers if we
1028 * have a filesystem using non-persistent counters.
1030 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1035 * Make sure that there are no blocks in front of the head
1036 * with the same cycle number as the head. This can happen
1037 * because we allow multiple outstanding log writes concurrently,
1038 * and the later writes might make it out before earlier ones.
1040 * We use the lsn from before modifying it so that we'll never
1041 * overwrite the unmount record after a clean unmount.
1043 * Do this only if we are going to recover the filesystem
1045 * NOTE: This used to say "if (!readonly)"
1046 * However on Linux, we can & do recover a read-only filesystem.
1047 * We only skip recovery if NORECOVERY is specified on mount,
1048 * in which case we would not be here.
1050 * But... if the -device- itself is readonly, just skip this.
1051 * We can't recover this device anyway, so it won't matter.
1053 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
1054 error = xlog_clear_stale_blocks(log, tail_lsn);
1060 xfs_warn(log->l_mp, "failed to locate log tail");
1065 * Is the log zeroed at all?
1067 * The last binary search should be changed to perform an X block read
1068 * once X becomes small enough. You can then search linearly through
1069 * the X blocks. This will cut down on the number of reads we need to do.
1071 * If the log is partially zeroed, this routine will pass back the blkno
1072 * of the first block with cycle number 0. It won't have a complete LR
1076 * 0 => the log is completely written to
1077 * -1 => use *blk_no as the first block of the log
1078 * >0 => error has occurred
1083 xfs_daddr_t *blk_no)
1087 uint first_cycle, last_cycle;
1088 xfs_daddr_t new_blk, last_blk, start_blk;
1089 xfs_daddr_t num_scan_bblks;
1090 int error, log_bbnum = log->l_logBBsize;
1094 /* check totally zeroed log */
1095 bp = xlog_get_bp(log, 1);
1098 error = xlog_bread(log, 0, 1, bp, &offset);
1102 first_cycle = xlog_get_cycle(offset);
1103 if (first_cycle == 0) { /* completely zeroed log */
1109 /* check partially zeroed log */
1110 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1114 last_cycle = xlog_get_cycle(offset);
1115 if (last_cycle != 0) { /* log completely written to */
1118 } else if (first_cycle != 1) {
1120 * If the cycle of the last block is zero, the cycle of
1121 * the first block must be 1. If it's not, maybe we're
1122 * not looking at a log... Bail out.
1125 "Log inconsistent or not a log (last==0, first!=1)");
1126 return XFS_ERROR(EINVAL);
1129 /* we have a partially zeroed log */
1130 last_blk = log_bbnum-1;
1131 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1135 * Validate the answer. Because there is no way to guarantee that
1136 * the entire log is made up of log records which are the same size,
1137 * we scan over the defined maximum blocks. At this point, the maximum
1138 * is not chosen to mean anything special. XXXmiken
1140 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1141 ASSERT(num_scan_bblks <= INT_MAX);
1143 if (last_blk < num_scan_bblks)
1144 num_scan_bblks = last_blk;
1145 start_blk = last_blk - num_scan_bblks;
1148 * We search for any instances of cycle number 0 that occur before
1149 * our current estimate of the head. What we're trying to detect is
1150 * 1 ... | 0 | 1 | 0...
1151 * ^ binary search ends here
1153 if ((error = xlog_find_verify_cycle(log, start_blk,
1154 (int)num_scan_bblks, 0, &new_blk)))
1160 * Potentially backup over partial log record write. We don't need
1161 * to search the end of the log because we know it is zero.
1163 if ((error = xlog_find_verify_log_record(log, start_blk,
1164 &last_blk, 0)) == -1) {
1165 error = XFS_ERROR(EIO);
1179 * These are simple subroutines used by xlog_clear_stale_blocks() below
1180 * to initialize a buffer full of empty log record headers and write
1181 * them into the log.
1192 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1194 memset(buf, 0, BBSIZE);
1195 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1196 recp->h_cycle = cpu_to_be32(cycle);
1197 recp->h_version = cpu_to_be32(
1198 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1199 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1200 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1201 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1202 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1206 xlog_write_log_records(
1217 int sectbb = log->l_sectBBsize;
1218 int end_block = start_block + blocks;
1224 * Greedily allocate a buffer big enough to handle the full
1225 * range of basic blocks to be written. If that fails, try
1226 * a smaller size. We need to be able to write at least a
1227 * log sector, or we're out of luck.
1229 bufblks = 1 << ffs(blocks);
1230 while (bufblks > log->l_logBBsize)
1232 while (!(bp = xlog_get_bp(log, bufblks))) {
1234 if (bufblks < sectbb)
1238 /* We may need to do a read at the start to fill in part of
1239 * the buffer in the starting sector not covered by the first
1242 balign = round_down(start_block, sectbb);
1243 if (balign != start_block) {
1244 error = xlog_bread_noalign(log, start_block, 1, bp);
1248 j = start_block - balign;
1251 for (i = start_block; i < end_block; i += bufblks) {
1252 int bcount, endcount;
1254 bcount = min(bufblks, end_block - start_block);
1255 endcount = bcount - j;
1257 /* We may need to do a read at the end to fill in part of
1258 * the buffer in the final sector not covered by the write.
1259 * If this is the same sector as the above read, skip it.
1261 ealign = round_down(end_block, sectbb);
1262 if (j == 0 && (start_block + endcount > ealign)) {
1263 offset = bp->b_addr + BBTOB(ealign - start_block);
1264 error = xlog_bread_offset(log, ealign, sectbb,
1271 offset = xlog_align(log, start_block, endcount, bp);
1272 for (; j < endcount; j++) {
1273 xlog_add_record(log, offset, cycle, i+j,
1274 tail_cycle, tail_block);
1277 error = xlog_bwrite(log, start_block, endcount, bp);
1280 start_block += endcount;
1290 * This routine is called to blow away any incomplete log writes out
1291 * in front of the log head. We do this so that we won't become confused
1292 * if we come up, write only a little bit more, and then crash again.
1293 * If we leave the partial log records out there, this situation could
1294 * cause us to think those partial writes are valid blocks since they
1295 * have the current cycle number. We get rid of them by overwriting them
1296 * with empty log records with the old cycle number rather than the
1299 * The tail lsn is passed in rather than taken from
1300 * the log so that we will not write over the unmount record after a
1301 * clean unmount in a 512 block log. Doing so would leave the log without
1302 * any valid log records in it until a new one was written. If we crashed
1303 * during that time we would not be able to recover.
1306 xlog_clear_stale_blocks(
1310 int tail_cycle, head_cycle;
1311 int tail_block, head_block;
1312 int tail_distance, max_distance;
1316 tail_cycle = CYCLE_LSN(tail_lsn);
1317 tail_block = BLOCK_LSN(tail_lsn);
1318 head_cycle = log->l_curr_cycle;
1319 head_block = log->l_curr_block;
1322 * Figure out the distance between the new head of the log
1323 * and the tail. We want to write over any blocks beyond the
1324 * head that we may have written just before the crash, but
1325 * we don't want to overwrite the tail of the log.
1327 if (head_cycle == tail_cycle) {
1329 * The tail is behind the head in the physical log,
1330 * so the distance from the head to the tail is the
1331 * distance from the head to the end of the log plus
1332 * the distance from the beginning of the log to the
1335 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1336 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1337 XFS_ERRLEVEL_LOW, log->l_mp);
1338 return XFS_ERROR(EFSCORRUPTED);
1340 tail_distance = tail_block + (log->l_logBBsize - head_block);
1343 * The head is behind the tail in the physical log,
1344 * so the distance from the head to the tail is just
1345 * the tail block minus the head block.
1347 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1348 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1349 XFS_ERRLEVEL_LOW, log->l_mp);
1350 return XFS_ERROR(EFSCORRUPTED);
1352 tail_distance = tail_block - head_block;
1356 * If the head is right up against the tail, we can't clear
1359 if (tail_distance <= 0) {
1360 ASSERT(tail_distance == 0);
1364 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1366 * Take the smaller of the maximum amount of outstanding I/O
1367 * we could have and the distance to the tail to clear out.
1368 * We take the smaller so that we don't overwrite the tail and
1369 * we don't waste all day writing from the head to the tail
1372 max_distance = MIN(max_distance, tail_distance);
1374 if ((head_block + max_distance) <= log->l_logBBsize) {
1376 * We can stomp all the blocks we need to without
1377 * wrapping around the end of the log. Just do it
1378 * in a single write. Use the cycle number of the
1379 * current cycle minus one so that the log will look like:
1382 error = xlog_write_log_records(log, (head_cycle - 1),
1383 head_block, max_distance, tail_cycle,
1389 * We need to wrap around the end of the physical log in
1390 * order to clear all the blocks. Do it in two separate
1391 * I/Os. The first write should be from the head to the
1392 * end of the physical log, and it should use the current
1393 * cycle number minus one just like above.
1395 distance = log->l_logBBsize - head_block;
1396 error = xlog_write_log_records(log, (head_cycle - 1),
1397 head_block, distance, tail_cycle,
1404 * Now write the blocks at the start of the physical log.
1405 * This writes the remainder of the blocks we want to clear.
1406 * It uses the current cycle number since we're now on the
1407 * same cycle as the head so that we get:
1408 * n ... n ... | n - 1 ...
1409 * ^^^^^ blocks we're writing
1411 distance = max_distance - (log->l_logBBsize - head_block);
1412 error = xlog_write_log_records(log, head_cycle, 0, distance,
1413 tail_cycle, tail_block);
1421 /******************************************************************************
1423 * Log recover routines
1425 ******************************************************************************
1428 STATIC xlog_recover_t *
1429 xlog_recover_find_tid(
1430 struct hlist_head *head,
1433 xlog_recover_t *trans;
1434 struct hlist_node *n;
1436 hlist_for_each_entry(trans, n, head, r_list) {
1437 if (trans->r_log_tid == tid)
1444 xlog_recover_new_tid(
1445 struct hlist_head *head,
1449 xlog_recover_t *trans;
1451 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1452 trans->r_log_tid = tid;
1454 INIT_LIST_HEAD(&trans->r_itemq);
1456 INIT_HLIST_NODE(&trans->r_list);
1457 hlist_add_head(&trans->r_list, head);
1461 xlog_recover_add_item(
1462 struct list_head *head)
1464 xlog_recover_item_t *item;
1466 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1467 INIT_LIST_HEAD(&item->ri_list);
1468 list_add_tail(&item->ri_list, head);
1472 xlog_recover_add_to_cont_trans(
1474 xlog_recover_t *trans,
1478 xlog_recover_item_t *item;
1479 xfs_caddr_t ptr, old_ptr;
1482 if (list_empty(&trans->r_itemq)) {
1483 /* finish copying rest of trans header */
1484 xlog_recover_add_item(&trans->r_itemq);
1485 ptr = (xfs_caddr_t) &trans->r_theader +
1486 sizeof(xfs_trans_header_t) - len;
1487 memcpy(ptr, dp, len); /* d, s, l */
1490 /* take the tail entry */
1491 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1493 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1494 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1496 ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP);
1497 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1498 item->ri_buf[item->ri_cnt-1].i_len += len;
1499 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1500 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
1505 * The next region to add is the start of a new region. It could be
1506 * a whole region or it could be the first part of a new region. Because
1507 * of this, the assumption here is that the type and size fields of all
1508 * format structures fit into the first 32 bits of the structure.
1510 * This works because all regions must be 32 bit aligned. Therefore, we
1511 * either have both fields or we have neither field. In the case we have
1512 * neither field, the data part of the region is zero length. We only have
1513 * a log_op_header and can throw away the header since a new one will appear
1514 * later. If we have at least 4 bytes, then we can determine how many regions
1515 * will appear in the current log item.
1518 xlog_recover_add_to_trans(
1520 xlog_recover_t *trans,
1524 xfs_inode_log_format_t *in_f; /* any will do */
1525 xlog_recover_item_t *item;
1530 if (list_empty(&trans->r_itemq)) {
1531 /* we need to catch log corruptions here */
1532 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
1533 xfs_warn(log->l_mp, "%s: bad header magic number",
1536 return XFS_ERROR(EIO);
1538 if (len == sizeof(xfs_trans_header_t))
1539 xlog_recover_add_item(&trans->r_itemq);
1540 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1544 ptr = kmem_alloc(len, KM_SLEEP);
1545 memcpy(ptr, dp, len);
1546 in_f = (xfs_inode_log_format_t *)ptr;
1548 /* take the tail entry */
1549 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1550 if (item->ri_total != 0 &&
1551 item->ri_total == item->ri_cnt) {
1552 /* tail item is in use, get a new one */
1553 xlog_recover_add_item(&trans->r_itemq);
1554 item = list_entry(trans->r_itemq.prev,
1555 xlog_recover_item_t, ri_list);
1558 if (item->ri_total == 0) { /* first region to be added */
1559 if (in_f->ilf_size == 0 ||
1560 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
1562 "bad number of regions (%d) in inode log format",
1565 return XFS_ERROR(EIO);
1568 item->ri_total = in_f->ilf_size;
1570 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1573 ASSERT(item->ri_total > item->ri_cnt);
1574 /* Description region is ri_buf[0] */
1575 item->ri_buf[item->ri_cnt].i_addr = ptr;
1576 item->ri_buf[item->ri_cnt].i_len = len;
1578 trace_xfs_log_recover_item_add(log, trans, item, 0);
1583 * Sort the log items in the transaction. Cancelled buffers need
1584 * to be put first so they are processed before any items that might
1585 * modify the buffers. If they are cancelled, then the modifications
1586 * don't need to be replayed.
1589 xlog_recover_reorder_trans(
1591 xlog_recover_t *trans,
1594 xlog_recover_item_t *item, *n;
1595 LIST_HEAD(sort_list);
1597 list_splice_init(&trans->r_itemq, &sort_list);
1598 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1599 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1601 switch (ITEM_TYPE(item)) {
1603 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1604 trace_xfs_log_recover_item_reorder_head(log,
1606 list_move(&item->ri_list, &trans->r_itemq);
1611 case XFS_LI_QUOTAOFF:
1614 trace_xfs_log_recover_item_reorder_tail(log,
1616 list_move_tail(&item->ri_list, &trans->r_itemq);
1620 "%s: unrecognized type of log operation",
1623 return XFS_ERROR(EIO);
1626 ASSERT(list_empty(&sort_list));
1631 * Build up the table of buf cancel records so that we don't replay
1632 * cancelled data in the second pass. For buffer records that are
1633 * not cancel records, there is nothing to do here so we just return.
1635 * If we get a cancel record which is already in the table, this indicates
1636 * that the buffer was cancelled multiple times. In order to ensure
1637 * that during pass 2 we keep the record in the table until we reach its
1638 * last occurrence in the log, we keep a reference count in the cancel
1639 * record in the table to tell us how many times we expect to see this
1640 * record during the second pass.
1643 xlog_recover_buffer_pass1(
1645 xlog_recover_item_t *item)
1647 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1648 struct list_head *bucket;
1649 struct xfs_buf_cancel *bcp;
1652 * If this isn't a cancel buffer item, then just return.
1654 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1655 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
1660 * Insert an xfs_buf_cancel record into the hash table of them.
1661 * If there is already an identical record, bump its reference count.
1663 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1664 list_for_each_entry(bcp, bucket, bc_list) {
1665 if (bcp->bc_blkno == buf_f->blf_blkno &&
1666 bcp->bc_len == buf_f->blf_len) {
1668 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
1673 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1674 bcp->bc_blkno = buf_f->blf_blkno;
1675 bcp->bc_len = buf_f->blf_len;
1676 bcp->bc_refcount = 1;
1677 list_add_tail(&bcp->bc_list, bucket);
1679 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
1684 * Check to see whether the buffer being recovered has a corresponding
1685 * entry in the buffer cancel record table. If it does then return 1
1686 * so that it will be cancelled, otherwise return 0. If the buffer is
1687 * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement
1688 * the refcount on the entry in the table and remove it from the table
1689 * if this is the last reference.
1691 * We remove the cancel record from the table when we encounter its
1692 * last occurrence in the log so that if the same buffer is re-used
1693 * again after its last cancellation we actually replay the changes
1694 * made at that point.
1697 xlog_check_buffer_cancelled(
1703 struct list_head *bucket;
1704 struct xfs_buf_cancel *bcp;
1706 if (log->l_buf_cancel_table == NULL) {
1708 * There is nothing in the table built in pass one,
1709 * so this buffer must not be cancelled.
1711 ASSERT(!(flags & XFS_BLF_CANCEL));
1716 * Search for an entry in the cancel table that matches our buffer.
1718 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1719 list_for_each_entry(bcp, bucket, bc_list) {
1720 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
1725 * We didn't find a corresponding entry in the table, so return 0 so
1726 * that the buffer is NOT cancelled.
1728 ASSERT(!(flags & XFS_BLF_CANCEL));
1733 * We've go a match, so return 1 so that the recovery of this buffer
1734 * is cancelled. If this buffer is actually a buffer cancel log
1735 * item, then decrement the refcount on the one in the table and
1736 * remove it if this is the last reference.
1738 if (flags & XFS_BLF_CANCEL) {
1739 if (--bcp->bc_refcount == 0) {
1740 list_del(&bcp->bc_list);
1748 * Perform recovery for a buffer full of inodes. In these buffers, the only
1749 * data which should be recovered is that which corresponds to the
1750 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1751 * data for the inodes is always logged through the inodes themselves rather
1752 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1754 * The only time when buffers full of inodes are fully recovered is when the
1755 * buffer is full of newly allocated inodes. In this case the buffer will
1756 * not be marked as an inode buffer and so will be sent to
1757 * xlog_recover_do_reg_buffer() below during recovery.
1760 xlog_recover_do_inode_buffer(
1761 struct xfs_mount *mp,
1762 xlog_recover_item_t *item,
1764 xfs_buf_log_format_t *buf_f)
1770 int reg_buf_offset = 0;
1771 int reg_buf_bytes = 0;
1772 int next_unlinked_offset;
1774 xfs_agino_t *logged_nextp;
1775 xfs_agino_t *buffer_nextp;
1777 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
1779 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1780 for (i = 0; i < inodes_per_buf; i++) {
1781 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1782 offsetof(xfs_dinode_t, di_next_unlinked);
1784 while (next_unlinked_offset >=
1785 (reg_buf_offset + reg_buf_bytes)) {
1787 * The next di_next_unlinked field is beyond
1788 * the current logged region. Find the next
1789 * logged region that contains or is beyond
1790 * the current di_next_unlinked field.
1793 bit = xfs_next_bit(buf_f->blf_data_map,
1794 buf_f->blf_map_size, bit);
1797 * If there are no more logged regions in the
1798 * buffer, then we're done.
1803 nbits = xfs_contig_bits(buf_f->blf_data_map,
1804 buf_f->blf_map_size, bit);
1806 reg_buf_offset = bit << XFS_BLF_SHIFT;
1807 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
1812 * If the current logged region starts after the current
1813 * di_next_unlinked field, then move on to the next
1814 * di_next_unlinked field.
1816 if (next_unlinked_offset < reg_buf_offset)
1819 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1820 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
1821 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1824 * The current logged region contains a copy of the
1825 * current di_next_unlinked field. Extract its value
1826 * and copy it to the buffer copy.
1828 logged_nextp = item->ri_buf[item_index].i_addr +
1829 next_unlinked_offset - reg_buf_offset;
1830 if (unlikely(*logged_nextp == 0)) {
1832 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1833 "Trying to replay bad (0) inode di_next_unlinked field.",
1835 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1836 XFS_ERRLEVEL_LOW, mp);
1837 return XFS_ERROR(EFSCORRUPTED);
1840 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1841 next_unlinked_offset);
1842 *buffer_nextp = *logged_nextp;
1849 * Perform a 'normal' buffer recovery. Each logged region of the
1850 * buffer should be copied over the corresponding region in the
1851 * given buffer. The bitmap in the buf log format structure indicates
1852 * where to place the logged data.
1855 xlog_recover_do_reg_buffer(
1856 struct xfs_mount *mp,
1857 xlog_recover_item_t *item,
1859 xfs_buf_log_format_t *buf_f)
1866 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
1869 i = 1; /* 0 is the buf format structure */
1871 bit = xfs_next_bit(buf_f->blf_data_map,
1872 buf_f->blf_map_size, bit);
1875 nbits = xfs_contig_bits(buf_f->blf_data_map,
1876 buf_f->blf_map_size, bit);
1878 ASSERT(item->ri_buf[i].i_addr != NULL);
1879 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
1880 ASSERT(XFS_BUF_COUNT(bp) >=
1881 ((uint)bit << XFS_BLF_SHIFT)+(nbits<<XFS_BLF_SHIFT));
1884 * Do a sanity check if this is a dquot buffer. Just checking
1885 * the first dquot in the buffer should do. XXXThis is
1886 * probably a good thing to do for other buf types also.
1889 if (buf_f->blf_flags &
1890 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
1891 if (item->ri_buf[i].i_addr == NULL) {
1893 "XFS: NULL dquot in %s.", __func__);
1896 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
1898 "XFS: dquot too small (%d) in %s.",
1899 item->ri_buf[i].i_len, __func__);
1902 error = xfs_qm_dqcheck(mp, item->ri_buf[i].i_addr,
1903 -1, 0, XFS_QMOPT_DOWARN,
1904 "dquot_buf_recover");
1909 memcpy(xfs_buf_offset(bp,
1910 (uint)bit << XFS_BLF_SHIFT), /* dest */
1911 item->ri_buf[i].i_addr, /* source */
1912 nbits<<XFS_BLF_SHIFT); /* length */
1918 /* Shouldn't be any more regions */
1919 ASSERT(i == item->ri_total);
1923 * Do some primitive error checking on ondisk dquot data structures.
1927 struct xfs_mount *mp,
1928 xfs_disk_dquot_t *ddq,
1930 uint type, /* used only when IO_dorepair is true */
1934 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
1938 * We can encounter an uninitialized dquot buffer for 2 reasons:
1939 * 1. If we crash while deleting the quotainode(s), and those blks got
1940 * used for user data. This is because we take the path of regular
1941 * file deletion; however, the size field of quotainodes is never
1942 * updated, so all the tricks that we play in itruncate_finish
1943 * don't quite matter.
1945 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1946 * But the allocation will be replayed so we'll end up with an
1947 * uninitialized quota block.
1949 * This is all fine; things are still consistent, and we haven't lost
1950 * any quota information. Just don't complain about bad dquot blks.
1952 if (ddq->d_magic != cpu_to_be16(XFS_DQUOT_MAGIC)) {
1953 if (flags & XFS_QMOPT_DOWARN)
1955 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
1956 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
1959 if (ddq->d_version != XFS_DQUOT_VERSION) {
1960 if (flags & XFS_QMOPT_DOWARN)
1962 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
1963 str, id, ddq->d_version, XFS_DQUOT_VERSION);
1967 if (ddq->d_flags != XFS_DQ_USER &&
1968 ddq->d_flags != XFS_DQ_PROJ &&
1969 ddq->d_flags != XFS_DQ_GROUP) {
1970 if (flags & XFS_QMOPT_DOWARN)
1972 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
1973 str, id, ddq->d_flags);
1977 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
1978 if (flags & XFS_QMOPT_DOWARN)
1980 "%s : ondisk-dquot 0x%p, ID mismatch: "
1981 "0x%x expected, found id 0x%x",
1982 str, ddq, id, be32_to_cpu(ddq->d_id));
1986 if (!errs && ddq->d_id) {
1987 if (ddq->d_blk_softlimit &&
1988 be64_to_cpu(ddq->d_bcount) >
1989 be64_to_cpu(ddq->d_blk_softlimit)) {
1990 if (!ddq->d_btimer) {
1991 if (flags & XFS_QMOPT_DOWARN)
1993 "%s : Dquot ID 0x%x (0x%p) BLK TIMER NOT STARTED",
1994 str, (int)be32_to_cpu(ddq->d_id), ddq);
1998 if (ddq->d_ino_softlimit &&
1999 be64_to_cpu(ddq->d_icount) >
2000 be64_to_cpu(ddq->d_ino_softlimit)) {
2001 if (!ddq->d_itimer) {
2002 if (flags & XFS_QMOPT_DOWARN)
2004 "%s : Dquot ID 0x%x (0x%p) INODE TIMER NOT STARTED",
2005 str, (int)be32_to_cpu(ddq->d_id), ddq);
2009 if (ddq->d_rtb_softlimit &&
2010 be64_to_cpu(ddq->d_rtbcount) >
2011 be64_to_cpu(ddq->d_rtb_softlimit)) {
2012 if (!ddq->d_rtbtimer) {
2013 if (flags & XFS_QMOPT_DOWARN)
2015 "%s : Dquot ID 0x%x (0x%p) RTBLK TIMER NOT STARTED",
2016 str, (int)be32_to_cpu(ddq->d_id), ddq);
2022 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2025 if (flags & XFS_QMOPT_DOWARN)
2026 xfs_notice(mp, "Re-initializing dquot ID 0x%x", id);
2029 * Typically, a repair is only requested by quotacheck.
2032 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2033 memset(d, 0, sizeof(xfs_dqblk_t));
2035 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2036 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2037 d->dd_diskdq.d_flags = type;
2038 d->dd_diskdq.d_id = cpu_to_be32(id);
2044 * Perform a dquot buffer recovery.
2045 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2046 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2047 * Else, treat it as a regular buffer and do recovery.
2050 xlog_recover_do_dquot_buffer(
2053 xlog_recover_item_t *item,
2055 xfs_buf_log_format_t *buf_f)
2059 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2062 * Filesystems are required to send in quota flags at mount time.
2064 if (mp->m_qflags == 0) {
2069 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2070 type |= XFS_DQ_USER;
2071 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2072 type |= XFS_DQ_PROJ;
2073 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2074 type |= XFS_DQ_GROUP;
2076 * This type of quotas was turned off, so ignore this buffer
2078 if (log->l_quotaoffs_flag & type)
2081 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2085 * This routine replays a modification made to a buffer at runtime.
2086 * There are actually two types of buffer, regular and inode, which
2087 * are handled differently. Inode buffers are handled differently
2088 * in that we only recover a specific set of data from them, namely
2089 * the inode di_next_unlinked fields. This is because all other inode
2090 * data is actually logged via inode records and any data we replay
2091 * here which overlaps that may be stale.
2093 * When meta-data buffers are freed at run time we log a buffer item
2094 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2095 * of the buffer in the log should not be replayed at recovery time.
2096 * This is so that if the blocks covered by the buffer are reused for
2097 * file data before we crash we don't end up replaying old, freed
2098 * meta-data into a user's file.
2100 * To handle the cancellation of buffer log items, we make two passes
2101 * over the log during recovery. During the first we build a table of
2102 * those buffers which have been cancelled, and during the second we
2103 * only replay those buffers which do not have corresponding cancel
2104 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2105 * for more details on the implementation of the table of cancel records.
2108 xlog_recover_buffer_pass2(
2110 struct list_head *buffer_list,
2111 xlog_recover_item_t *item)
2113 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2114 xfs_mount_t *mp = log->l_mp;
2120 * In this pass we only want to recover all the buffers which have
2121 * not been cancelled and are not cancellation buffers themselves.
2123 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2124 buf_f->blf_len, buf_f->blf_flags)) {
2125 trace_xfs_log_recover_buf_cancel(log, buf_f);
2129 trace_xfs_log_recover_buf_recover(log, buf_f);
2131 buf_flags = XBF_LOCK;
2132 if (!(buf_f->blf_flags & XFS_BLF_INODE_BUF))
2133 buf_flags |= XBF_MAPPED;
2135 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2138 return XFS_ERROR(ENOMEM);
2139 error = bp->b_error;
2141 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
2146 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
2147 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2148 } else if (buf_f->blf_flags &
2149 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2150 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2152 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2155 return XFS_ERROR(error);
2158 * Perform delayed write on the buffer. Asynchronous writes will be
2159 * slower when taking into account all the buffers to be flushed.
2161 * Also make sure that only inode buffers with good sizes stay in
2162 * the buffer cache. The kernel moves inodes in buffers of 1 block
2163 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2164 * buffers in the log can be a different size if the log was generated
2165 * by an older kernel using unclustered inode buffers or a newer kernel
2166 * running with a different inode cluster size. Regardless, if the
2167 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2168 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2169 * the buffer out of the buffer cache so that the buffer won't
2170 * overlap with future reads of those inodes.
2172 if (XFS_DINODE_MAGIC ==
2173 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2174 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2175 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2177 error = xfs_bwrite(bp);
2179 ASSERT(bp->b_target->bt_mount == mp);
2180 bp->b_iodone = xlog_recover_iodone;
2181 xfs_buf_delwri_queue(bp, buffer_list);
2189 xlog_recover_inode_pass2(
2191 struct list_head *buffer_list,
2192 xlog_recover_item_t *item)
2194 xfs_inode_log_format_t *in_f;
2195 xfs_mount_t *mp = log->l_mp;
2204 xfs_icdinode_t *dicp;
2207 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2208 in_f = item->ri_buf[0].i_addr;
2210 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
2212 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2218 * Inode buffers can be freed, look out for it,
2219 * and do not replay the inode.
2221 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2222 in_f->ilf_len, 0)) {
2224 trace_xfs_log_recover_inode_cancel(log, in_f);
2227 trace_xfs_log_recover_inode_recover(log, in_f);
2229 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len,
2235 error = bp->b_error;
2237 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
2241 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2242 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
2245 * Make sure the place we're flushing out to really looks
2248 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
2251 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2252 __func__, dip, bp, in_f->ilf_ino);
2253 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2254 XFS_ERRLEVEL_LOW, mp);
2255 error = EFSCORRUPTED;
2258 dicp = item->ri_buf[1].i_addr;
2259 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2262 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2263 __func__, item, in_f->ilf_ino);
2264 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2265 XFS_ERRLEVEL_LOW, mp);
2266 error = EFSCORRUPTED;
2270 /* Skip replay when the on disk inode is newer than the log one */
2271 if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
2273 * Deal with the wrap case, DI_MAX_FLUSH is less
2274 * than smaller numbers
2276 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
2277 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2281 trace_xfs_log_recover_inode_skip(log, in_f);
2286 /* Take the opportunity to reset the flush iteration count */
2287 dicp->di_flushiter = 0;
2289 if (unlikely(S_ISREG(dicp->di_mode))) {
2290 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2291 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2292 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2293 XFS_ERRLEVEL_LOW, mp, dicp);
2296 "%s: Bad regular inode log record, rec ptr 0x%p, "
2297 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2298 __func__, item, dip, bp, in_f->ilf_ino);
2299 error = EFSCORRUPTED;
2302 } else if (unlikely(S_ISDIR(dicp->di_mode))) {
2303 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2304 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2305 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2306 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2307 XFS_ERRLEVEL_LOW, mp, dicp);
2310 "%s: Bad dir inode log record, rec ptr 0x%p, "
2311 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2312 __func__, item, dip, bp, in_f->ilf_ino);
2313 error = EFSCORRUPTED;
2317 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2318 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2319 XFS_ERRLEVEL_LOW, mp, dicp);
2322 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2323 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2324 __func__, item, dip, bp, in_f->ilf_ino,
2325 dicp->di_nextents + dicp->di_anextents,
2327 error = EFSCORRUPTED;
2330 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2331 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2332 XFS_ERRLEVEL_LOW, mp, dicp);
2335 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2336 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
2337 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
2338 error = EFSCORRUPTED;
2341 if (unlikely(item->ri_buf[1].i_len > sizeof(struct xfs_icdinode))) {
2342 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
2343 XFS_ERRLEVEL_LOW, mp, dicp);
2346 "%s: Bad inode log record length %d, rec ptr 0x%p",
2347 __func__, item->ri_buf[1].i_len, item);
2348 error = EFSCORRUPTED;
2352 /* The core is in in-core format */
2353 xfs_dinode_to_disk(dip, item->ri_buf[1].i_addr);
2355 /* the rest is in on-disk format */
2356 if (item->ri_buf[1].i_len > sizeof(struct xfs_icdinode)) {
2357 memcpy((xfs_caddr_t) dip + sizeof(struct xfs_icdinode),
2358 item->ri_buf[1].i_addr + sizeof(struct xfs_icdinode),
2359 item->ri_buf[1].i_len - sizeof(struct xfs_icdinode));
2362 fields = in_f->ilf_fields;
2363 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2365 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
2368 memcpy(XFS_DFORK_DPTR(dip),
2369 &in_f->ilf_u.ilfu_uuid,
2374 if (in_f->ilf_size == 2)
2375 goto write_inode_buffer;
2376 len = item->ri_buf[2].i_len;
2377 src = item->ri_buf[2].i_addr;
2378 ASSERT(in_f->ilf_size <= 4);
2379 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2380 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2381 (len == in_f->ilf_dsize));
2383 switch (fields & XFS_ILOG_DFORK) {
2384 case XFS_ILOG_DDATA:
2386 memcpy(XFS_DFORK_DPTR(dip), src, len);
2389 case XFS_ILOG_DBROOT:
2390 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
2391 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
2392 XFS_DFORK_DSIZE(dip, mp));
2397 * There are no data fork flags set.
2399 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2404 * If we logged any attribute data, recover it. There may or
2405 * may not have been any other non-core data logged in this
2408 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2409 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2414 len = item->ri_buf[attr_index].i_len;
2415 src = item->ri_buf[attr_index].i_addr;
2416 ASSERT(len == in_f->ilf_asize);
2418 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2419 case XFS_ILOG_ADATA:
2421 dest = XFS_DFORK_APTR(dip);
2422 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2423 memcpy(dest, src, len);
2426 case XFS_ILOG_ABROOT:
2427 dest = XFS_DFORK_APTR(dip);
2428 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2429 len, (xfs_bmdr_block_t*)dest,
2430 XFS_DFORK_ASIZE(dip, mp));
2434 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
2443 ASSERT(bp->b_target->bt_mount == mp);
2444 bp->b_iodone = xlog_recover_iodone;
2445 xfs_buf_delwri_queue(bp, buffer_list);
2450 return XFS_ERROR(error);
2454 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2455 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2459 xlog_recover_quotaoff_pass1(
2461 xlog_recover_item_t *item)
2463 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
2467 * The logitem format's flag tells us if this was user quotaoff,
2468 * group/project quotaoff or both.
2470 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2471 log->l_quotaoffs_flag |= XFS_DQ_USER;
2472 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2473 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2474 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2475 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2481 * Recover a dquot record
2484 xlog_recover_dquot_pass2(
2486 struct list_head *buffer_list,
2487 xlog_recover_item_t *item)
2489 xfs_mount_t *mp = log->l_mp;
2491 struct xfs_disk_dquot *ddq, *recddq;
2493 xfs_dq_logformat_t *dq_f;
2498 * Filesystems are required to send in quota flags at mount time.
2500 if (mp->m_qflags == 0)
2503 recddq = item->ri_buf[1].i_addr;
2504 if (recddq == NULL) {
2505 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
2506 return XFS_ERROR(EIO);
2508 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
2509 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
2510 item->ri_buf[1].i_len, __func__);
2511 return XFS_ERROR(EIO);
2515 * This type of quotas was turned off, so ignore this record.
2517 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
2519 if (log->l_quotaoffs_flag & type)
2523 * At this point we know that quota was _not_ turned off.
2524 * Since the mount flags are not indicating to us otherwise, this
2525 * must mean that quota is on, and the dquot needs to be replayed.
2526 * Remember that we may not have fully recovered the superblock yet,
2527 * so we can't do the usual trick of looking at the SB quota bits.
2529 * The other possibility, of course, is that the quota subsystem was
2530 * removed since the last mount - ENOSYS.
2532 dq_f = item->ri_buf[0].i_addr;
2534 error = xfs_qm_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2535 "xlog_recover_dquot_pass2 (log copy)");
2537 return XFS_ERROR(EIO);
2538 ASSERT(dq_f->qlf_len == 1);
2540 error = xfs_read_buf(mp, mp->m_ddev_targp,
2542 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2545 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#3)");
2549 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2552 * At least the magic num portion should be on disk because this
2553 * was among a chunk of dquots created earlier, and we did some
2554 * minimal initialization then.
2556 error = xfs_qm_dqcheck(mp, ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2557 "xlog_recover_dquot_pass2");
2560 return XFS_ERROR(EIO);
2563 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2565 ASSERT(dq_f->qlf_size == 2);
2566 ASSERT(bp->b_target->bt_mount == mp);
2567 bp->b_iodone = xlog_recover_iodone;
2568 xfs_buf_delwri_queue(bp, buffer_list);
2575 * This routine is called to create an in-core extent free intent
2576 * item from the efi format structure which was logged on disk.
2577 * It allocates an in-core efi, copies the extents from the format
2578 * structure into it, and adds the efi to the AIL with the given
2582 xlog_recover_efi_pass2(
2584 xlog_recover_item_t *item,
2588 xfs_mount_t *mp = log->l_mp;
2589 xfs_efi_log_item_t *efip;
2590 xfs_efi_log_format_t *efi_formatp;
2592 efi_formatp = item->ri_buf[0].i_addr;
2594 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
2595 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2596 &(efip->efi_format)))) {
2597 xfs_efi_item_free(efip);
2600 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
2602 spin_lock(&log->l_ailp->xa_lock);
2604 * xfs_trans_ail_update() drops the AIL lock.
2606 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
2612 * This routine is called when an efd format structure is found in
2613 * a committed transaction in the log. It's purpose is to cancel
2614 * the corresponding efi if it was still in the log. To do this
2615 * it searches the AIL for the efi with an id equal to that in the
2616 * efd format structure. If we find it, we remove the efi from the
2620 xlog_recover_efd_pass2(
2622 xlog_recover_item_t *item)
2624 xfs_efd_log_format_t *efd_formatp;
2625 xfs_efi_log_item_t *efip = NULL;
2626 xfs_log_item_t *lip;
2628 struct xfs_ail_cursor cur;
2629 struct xfs_ail *ailp = log->l_ailp;
2631 efd_formatp = item->ri_buf[0].i_addr;
2632 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2633 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2634 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2635 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
2636 efi_id = efd_formatp->efd_efi_id;
2639 * Search for the efi with the id in the efd format structure
2642 spin_lock(&ailp->xa_lock);
2643 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
2644 while (lip != NULL) {
2645 if (lip->li_type == XFS_LI_EFI) {
2646 efip = (xfs_efi_log_item_t *)lip;
2647 if (efip->efi_format.efi_id == efi_id) {
2649 * xfs_trans_ail_delete() drops the
2652 xfs_trans_ail_delete(ailp, lip,
2653 SHUTDOWN_CORRUPT_INCORE);
2654 xfs_efi_item_free(efip);
2655 spin_lock(&ailp->xa_lock);
2659 lip = xfs_trans_ail_cursor_next(ailp, &cur);
2661 xfs_trans_ail_cursor_done(ailp, &cur);
2662 spin_unlock(&ailp->xa_lock);
2668 * Free up any resources allocated by the transaction
2670 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2673 xlog_recover_free_trans(
2674 struct xlog_recover *trans)
2676 xlog_recover_item_t *item, *n;
2679 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
2680 /* Free the regions in the item. */
2681 list_del(&item->ri_list);
2682 for (i = 0; i < item->ri_cnt; i++)
2683 kmem_free(item->ri_buf[i].i_addr);
2684 /* Free the item itself */
2685 kmem_free(item->ri_buf);
2688 /* Free the transaction recover structure */
2693 xlog_recover_commit_pass1(
2695 struct xlog_recover *trans,
2696 xlog_recover_item_t *item)
2698 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
2700 switch (ITEM_TYPE(item)) {
2702 return xlog_recover_buffer_pass1(log, item);
2703 case XFS_LI_QUOTAOFF:
2704 return xlog_recover_quotaoff_pass1(log, item);
2709 /* nothing to do in pass 1 */
2712 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
2713 __func__, ITEM_TYPE(item));
2715 return XFS_ERROR(EIO);
2720 xlog_recover_commit_pass2(
2722 struct xlog_recover *trans,
2723 struct list_head *buffer_list,
2724 xlog_recover_item_t *item)
2726 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
2728 switch (ITEM_TYPE(item)) {
2730 return xlog_recover_buffer_pass2(log, buffer_list, item);
2732 return xlog_recover_inode_pass2(log, buffer_list, item);
2734 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
2736 return xlog_recover_efd_pass2(log, item);
2738 return xlog_recover_dquot_pass2(log, buffer_list, item);
2739 case XFS_LI_QUOTAOFF:
2740 /* nothing to do in pass2 */
2743 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
2744 __func__, ITEM_TYPE(item));
2746 return XFS_ERROR(EIO);
2751 * Perform the transaction.
2753 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2754 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2757 xlog_recover_commit_trans(
2759 struct xlog_recover *trans,
2762 int error = 0, error2;
2763 xlog_recover_item_t *item;
2764 LIST_HEAD (buffer_list);
2766 hlist_del(&trans->r_list);
2768 error = xlog_recover_reorder_trans(log, trans, pass);
2772 list_for_each_entry(item, &trans->r_itemq, ri_list) {
2774 case XLOG_RECOVER_PASS1:
2775 error = xlog_recover_commit_pass1(log, trans, item);
2777 case XLOG_RECOVER_PASS2:
2778 error = xlog_recover_commit_pass2(log, trans,
2779 &buffer_list, item);
2789 xlog_recover_free_trans(trans);
2792 error2 = xfs_buf_delwri_submit(&buffer_list);
2793 return error ? error : error2;
2797 xlog_recover_unmount_trans(
2799 xlog_recover_t *trans)
2801 /* Do nothing now */
2802 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
2807 * There are two valid states of the r_state field. 0 indicates that the
2808 * transaction structure is in a normal state. We have either seen the
2809 * start of the transaction or the last operation we added was not a partial
2810 * operation. If the last operation we added to the transaction was a
2811 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2813 * NOTE: skip LRs with 0 data length.
2816 xlog_recover_process_data(
2818 struct hlist_head rhash[],
2819 xlog_rec_header_t *rhead,
2825 xlog_op_header_t *ohead;
2826 xlog_recover_t *trans;
2832 lp = dp + be32_to_cpu(rhead->h_len);
2833 num_logops = be32_to_cpu(rhead->h_num_logops);
2835 /* check the log format matches our own - else we can't recover */
2836 if (xlog_header_check_recover(log->l_mp, rhead))
2837 return (XFS_ERROR(EIO));
2839 while ((dp < lp) && num_logops) {
2840 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2841 ohead = (xlog_op_header_t *)dp;
2842 dp += sizeof(xlog_op_header_t);
2843 if (ohead->oh_clientid != XFS_TRANSACTION &&
2844 ohead->oh_clientid != XFS_LOG) {
2845 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
2846 __func__, ohead->oh_clientid);
2848 return (XFS_ERROR(EIO));
2850 tid = be32_to_cpu(ohead->oh_tid);
2851 hash = XLOG_RHASH(tid);
2852 trans = xlog_recover_find_tid(&rhash[hash], tid);
2853 if (trans == NULL) { /* not found; add new tid */
2854 if (ohead->oh_flags & XLOG_START_TRANS)
2855 xlog_recover_new_tid(&rhash[hash], tid,
2856 be64_to_cpu(rhead->h_lsn));
2858 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
2859 xfs_warn(log->l_mp, "%s: bad length 0x%x",
2860 __func__, be32_to_cpu(ohead->oh_len));
2862 return (XFS_ERROR(EIO));
2864 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2865 if (flags & XLOG_WAS_CONT_TRANS)
2866 flags &= ~XLOG_CONTINUE_TRANS;
2868 case XLOG_COMMIT_TRANS:
2869 error = xlog_recover_commit_trans(log,
2872 case XLOG_UNMOUNT_TRANS:
2873 error = xlog_recover_unmount_trans(log, trans);
2875 case XLOG_WAS_CONT_TRANS:
2876 error = xlog_recover_add_to_cont_trans(log,
2878 be32_to_cpu(ohead->oh_len));
2880 case XLOG_START_TRANS:
2881 xfs_warn(log->l_mp, "%s: bad transaction",
2884 error = XFS_ERROR(EIO);
2887 case XLOG_CONTINUE_TRANS:
2888 error = xlog_recover_add_to_trans(log, trans,
2889 dp, be32_to_cpu(ohead->oh_len));
2892 xfs_warn(log->l_mp, "%s: bad flag 0x%x",
2895 error = XFS_ERROR(EIO);
2901 dp += be32_to_cpu(ohead->oh_len);
2908 * Process an extent free intent item that was recovered from
2909 * the log. We need to free the extents that it describes.
2912 xlog_recover_process_efi(
2914 xfs_efi_log_item_t *efip)
2916 xfs_efd_log_item_t *efdp;
2921 xfs_fsblock_t startblock_fsb;
2923 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
2926 * First check the validity of the extents described by the
2927 * EFI. If any are bad, then assume that all are bad and
2928 * just toss the EFI.
2930 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2931 extp = &(efip->efi_format.efi_extents[i]);
2932 startblock_fsb = XFS_BB_TO_FSB(mp,
2933 XFS_FSB_TO_DADDR(mp, extp->ext_start));
2934 if ((startblock_fsb == 0) ||
2935 (extp->ext_len == 0) ||
2936 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
2937 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
2939 * This will pull the EFI from the AIL and
2940 * free the memory associated with it.
2942 xfs_efi_release(efip, efip->efi_format.efi_nextents);
2943 return XFS_ERROR(EIO);
2947 tp = xfs_trans_alloc(mp, 0);
2948 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
2951 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
2953 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2954 extp = &(efip->efi_format.efi_extents[i]);
2955 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
2958 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
2962 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
2963 error = xfs_trans_commit(tp, 0);
2967 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
2972 * When this is called, all of the EFIs which did not have
2973 * corresponding EFDs should be in the AIL. What we do now
2974 * is free the extents associated with each one.
2976 * Since we process the EFIs in normal transactions, they
2977 * will be removed at some point after the commit. This prevents
2978 * us from just walking down the list processing each one.
2979 * We'll use a flag in the EFI to skip those that we've already
2980 * processed and use the AIL iteration mechanism's generation
2981 * count to try to speed this up at least a bit.
2983 * When we start, we know that the EFIs are the only things in
2984 * the AIL. As we process them, however, other items are added
2985 * to the AIL. Since everything added to the AIL must come after
2986 * everything already in the AIL, we stop processing as soon as
2987 * we see something other than an EFI in the AIL.
2990 xlog_recover_process_efis(
2993 xfs_log_item_t *lip;
2994 xfs_efi_log_item_t *efip;
2996 struct xfs_ail_cursor cur;
2997 struct xfs_ail *ailp;
3000 spin_lock(&ailp->xa_lock);
3001 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3002 while (lip != NULL) {
3004 * We're done when we see something other than an EFI.
3005 * There should be no EFIs left in the AIL now.
3007 if (lip->li_type != XFS_LI_EFI) {
3009 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
3010 ASSERT(lip->li_type != XFS_LI_EFI);
3016 * Skip EFIs that we've already processed.
3018 efip = (xfs_efi_log_item_t *)lip;
3019 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
3020 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3024 spin_unlock(&ailp->xa_lock);
3025 error = xlog_recover_process_efi(log->l_mp, efip);
3026 spin_lock(&ailp->xa_lock);
3029 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3032 xfs_trans_ail_cursor_done(ailp, &cur);
3033 spin_unlock(&ailp->xa_lock);
3038 * This routine performs a transaction to null out a bad inode pointer
3039 * in an agi unlinked inode hash bucket.
3042 xlog_recover_clear_agi_bucket(
3044 xfs_agnumber_t agno,
3053 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3054 error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp),
3059 error = xfs_read_agi(mp, tp, agno, &agibp);
3063 agi = XFS_BUF_TO_AGI(agibp);
3064 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
3065 offset = offsetof(xfs_agi_t, agi_unlinked) +
3066 (sizeof(xfs_agino_t) * bucket);
3067 xfs_trans_log_buf(tp, agibp, offset,
3068 (offset + sizeof(xfs_agino_t) - 1));
3070 error = xfs_trans_commit(tp, 0);
3076 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3078 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
3083 xlog_recover_process_one_iunlink(
3084 struct xfs_mount *mp,
3085 xfs_agnumber_t agno,
3089 struct xfs_buf *ibp;
3090 struct xfs_dinode *dip;
3091 struct xfs_inode *ip;
3095 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3096 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
3101 * Get the on disk inode to find the next inode in the bucket.
3103 error = xfs_itobp(mp, NULL, ip, &dip, &ibp, XBF_LOCK);
3107 ASSERT(ip->i_d.di_nlink == 0);
3108 ASSERT(ip->i_d.di_mode != 0);
3110 /* setup for the next pass */
3111 agino = be32_to_cpu(dip->di_next_unlinked);
3115 * Prevent any DMAPI event from being sent when the reference on
3116 * the inode is dropped.
3118 ip->i_d.di_dmevmask = 0;
3127 * We can't read in the inode this bucket points to, or this inode
3128 * is messed up. Just ditch this bucket of inodes. We will lose
3129 * some inodes and space, but at least we won't hang.
3131 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3132 * clear the inode pointer in the bucket.
3134 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3139 * xlog_iunlink_recover
3141 * This is called during recovery to process any inodes which
3142 * we unlinked but not freed when the system crashed. These
3143 * inodes will be on the lists in the AGI blocks. What we do
3144 * here is scan all the AGIs and fully truncate and free any
3145 * inodes found on the lists. Each inode is removed from the
3146 * lists when it has been fully truncated and is freed. The
3147 * freeing of the inode and its removal from the list must be
3151 xlog_recover_process_iunlinks(
3155 xfs_agnumber_t agno;
3166 * Prevent any DMAPI event from being sent while in this function.
3168 mp_dmevmask = mp->m_dmevmask;
3171 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3173 * Find the agi for this ag.
3175 error = xfs_read_agi(mp, NULL, agno, &agibp);
3178 * AGI is b0rked. Don't process it.
3180 * We should probably mark the filesystem as corrupt
3181 * after we've recovered all the ag's we can....
3186 * Unlock the buffer so that it can be acquired in the normal
3187 * course of the transaction to truncate and free each inode.
3188 * Because we are not racing with anyone else here for the AGI
3189 * buffer, we don't even need to hold it locked to read the
3190 * initial unlinked bucket entries out of the buffer. We keep
3191 * buffer reference though, so that it stays pinned in memory
3192 * while we need the buffer.
3194 agi = XFS_BUF_TO_AGI(agibp);
3195 xfs_buf_unlock(agibp);
3197 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3198 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3199 while (agino != NULLAGINO) {
3200 agino = xlog_recover_process_one_iunlink(mp,
3201 agno, agino, bucket);
3204 xfs_buf_rele(agibp);
3207 mp->m_dmevmask = mp_dmevmask;
3213 xlog_pack_data_checksum(
3215 xlog_in_core_t *iclog,
3222 up = (__be32 *)iclog->ic_datap;
3223 /* divide length by 4 to get # words */
3224 for (i = 0; i < (size >> 2); i++) {
3225 chksum ^= be32_to_cpu(*up);
3228 iclog->ic_header.h_chksum = cpu_to_be32(chksum);
3231 #define xlog_pack_data_checksum(log, iclog, size)
3235 * Stamp cycle number in every block
3240 xlog_in_core_t *iclog,
3244 int size = iclog->ic_offset + roundoff;
3248 xlog_pack_data_checksum(log, iclog, size);
3250 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3252 dp = iclog->ic_datap;
3253 for (i = 0; i < BTOBB(size) &&
3254 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3255 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
3256 *(__be32 *)dp = cycle_lsn;
3260 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3261 xlog_in_core_2_t *xhdr = iclog->ic_data;
3263 for ( ; i < BTOBB(size); i++) {
3264 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3265 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3266 xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
3267 *(__be32 *)dp = cycle_lsn;
3271 for (i = 1; i < log->l_iclog_heads; i++) {
3272 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3279 xlog_rec_header_t *rhead,
3285 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
3286 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3287 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
3291 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3292 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
3293 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
3294 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3295 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3296 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
3303 xlog_valid_rec_header(
3305 xlog_rec_header_t *rhead,
3310 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
3311 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3312 XFS_ERRLEVEL_LOW, log->l_mp);
3313 return XFS_ERROR(EFSCORRUPTED);
3316 (!rhead->h_version ||
3317 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
3318 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
3319 __func__, be32_to_cpu(rhead->h_version));
3320 return XFS_ERROR(EIO);
3323 /* LR body must have data or it wouldn't have been written */
3324 hlen = be32_to_cpu(rhead->h_len);
3325 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3326 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3327 XFS_ERRLEVEL_LOW, log->l_mp);
3328 return XFS_ERROR(EFSCORRUPTED);
3330 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3331 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3332 XFS_ERRLEVEL_LOW, log->l_mp);
3333 return XFS_ERROR(EFSCORRUPTED);
3339 * Read the log from tail to head and process the log records found.
3340 * Handle the two cases where the tail and head are in the same cycle
3341 * and where the active portion of the log wraps around the end of
3342 * the physical log separately. The pass parameter is passed through
3343 * to the routines called to process the data and is not looked at
3347 xlog_do_recovery_pass(
3349 xfs_daddr_t head_blk,
3350 xfs_daddr_t tail_blk,
3353 xlog_rec_header_t *rhead;
3356 xfs_buf_t *hbp, *dbp;
3357 int error = 0, h_size;
3358 int bblks, split_bblks;
3359 int hblks, split_hblks, wrapped_hblks;
3360 struct hlist_head rhash[XLOG_RHASH_SIZE];
3362 ASSERT(head_blk != tail_blk);
3365 * Read the header of the tail block and get the iclog buffer size from
3366 * h_size. Use this to tell how many sectors make up the log header.
3368 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3370 * When using variable length iclogs, read first sector of
3371 * iclog header and extract the header size from it. Get a
3372 * new hbp that is the correct size.
3374 hbp = xlog_get_bp(log, 1);
3378 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
3382 rhead = (xlog_rec_header_t *)offset;
3383 error = xlog_valid_rec_header(log, rhead, tail_blk);
3386 h_size = be32_to_cpu(rhead->h_size);
3387 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
3388 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3389 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3390 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3393 hbp = xlog_get_bp(log, hblks);
3398 ASSERT(log->l_sectBBsize == 1);
3400 hbp = xlog_get_bp(log, 1);
3401 h_size = XLOG_BIG_RECORD_BSIZE;
3406 dbp = xlog_get_bp(log, BTOBB(h_size));
3412 memset(rhash, 0, sizeof(rhash));
3413 if (tail_blk <= head_blk) {
3414 for (blk_no = tail_blk; blk_no < head_blk; ) {
3415 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3419 rhead = (xlog_rec_header_t *)offset;
3420 error = xlog_valid_rec_header(log, rhead, blk_no);
3424 /* blocks in data section */
3425 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3426 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
3431 xlog_unpack_data(rhead, offset, log);
3432 if ((error = xlog_recover_process_data(log,
3433 rhash, rhead, offset, pass)))
3435 blk_no += bblks + hblks;
3439 * Perform recovery around the end of the physical log.
3440 * When the head is not on the same cycle number as the tail,
3441 * we can't do a sequential recovery as above.
3444 while (blk_no < log->l_logBBsize) {
3446 * Check for header wrapping around physical end-of-log
3448 offset = hbp->b_addr;
3451 if (blk_no + hblks <= log->l_logBBsize) {
3452 /* Read header in one read */
3453 error = xlog_bread(log, blk_no, hblks, hbp,
3458 /* This LR is split across physical log end */
3459 if (blk_no != log->l_logBBsize) {
3460 /* some data before physical log end */
3461 ASSERT(blk_no <= INT_MAX);
3462 split_hblks = log->l_logBBsize - (int)blk_no;
3463 ASSERT(split_hblks > 0);
3464 error = xlog_bread(log, blk_no,
3472 * Note: this black magic still works with
3473 * large sector sizes (non-512) only because:
3474 * - we increased the buffer size originally
3475 * by 1 sector giving us enough extra space
3476 * for the second read;
3477 * - the log start is guaranteed to be sector
3479 * - we read the log end (LR header start)
3480 * _first_, then the log start (LR header end)
3481 * - order is important.
3483 wrapped_hblks = hblks - split_hblks;
3484 error = xlog_bread_offset(log, 0,
3486 offset + BBTOB(split_hblks));
3490 rhead = (xlog_rec_header_t *)offset;
3491 error = xlog_valid_rec_header(log, rhead,
3492 split_hblks ? blk_no : 0);
3496 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3499 /* Read in data for log record */
3500 if (blk_no + bblks <= log->l_logBBsize) {
3501 error = xlog_bread(log, blk_no, bblks, dbp,
3506 /* This log record is split across the
3507 * physical end of log */
3508 offset = dbp->b_addr;
3510 if (blk_no != log->l_logBBsize) {
3511 /* some data is before the physical
3513 ASSERT(!wrapped_hblks);
3514 ASSERT(blk_no <= INT_MAX);
3516 log->l_logBBsize - (int)blk_no;
3517 ASSERT(split_bblks > 0);
3518 error = xlog_bread(log, blk_no,
3526 * Note: this black magic still works with
3527 * large sector sizes (non-512) only because:
3528 * - we increased the buffer size originally
3529 * by 1 sector giving us enough extra space
3530 * for the second read;
3531 * - the log start is guaranteed to be sector
3533 * - we read the log end (LR header start)
3534 * _first_, then the log start (LR header end)
3535 * - order is important.
3537 error = xlog_bread_offset(log, 0,
3538 bblks - split_bblks, hbp,
3539 offset + BBTOB(split_bblks));
3543 xlog_unpack_data(rhead, offset, log);
3544 if ((error = xlog_recover_process_data(log, rhash,
3545 rhead, offset, pass)))
3550 ASSERT(blk_no >= log->l_logBBsize);
3551 blk_no -= log->l_logBBsize;
3553 /* read first part of physical log */
3554 while (blk_no < head_blk) {
3555 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3559 rhead = (xlog_rec_header_t *)offset;
3560 error = xlog_valid_rec_header(log, rhead, blk_no);
3564 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3565 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
3570 xlog_unpack_data(rhead, offset, log);
3571 if ((error = xlog_recover_process_data(log, rhash,
3572 rhead, offset, pass)))
3574 blk_no += bblks + hblks;
3586 * Do the recovery of the log. We actually do this in two phases.
3587 * The two passes are necessary in order to implement the function
3588 * of cancelling a record written into the log. The first pass
3589 * determines those things which have been cancelled, and the
3590 * second pass replays log items normally except for those which
3591 * have been cancelled. The handling of the replay and cancellations
3592 * takes place in the log item type specific routines.
3594 * The table of items which have cancel records in the log is allocated
3595 * and freed at this level, since only here do we know when all of
3596 * the log recovery has been completed.
3599 xlog_do_log_recovery(
3601 xfs_daddr_t head_blk,
3602 xfs_daddr_t tail_blk)
3606 ASSERT(head_blk != tail_blk);
3609 * First do a pass to find all of the cancelled buf log items.
3610 * Store them in the buf_cancel_table for use in the second pass.
3612 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
3613 sizeof(struct list_head),
3615 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3616 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
3618 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3619 XLOG_RECOVER_PASS1);
3621 kmem_free(log->l_buf_cancel_table);
3622 log->l_buf_cancel_table = NULL;
3626 * Then do a second pass to actually recover the items in the log.
3627 * When it is complete free the table of buf cancel items.
3629 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3630 XLOG_RECOVER_PASS2);
3635 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3636 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
3640 kmem_free(log->l_buf_cancel_table);
3641 log->l_buf_cancel_table = NULL;
3647 * Do the actual recovery
3652 xfs_daddr_t head_blk,
3653 xfs_daddr_t tail_blk)
3660 * First replay the images in the log.
3662 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3667 * If IO errors happened during recovery, bail out.
3669 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3674 * We now update the tail_lsn since much of the recovery has completed
3675 * and there may be space available to use. If there were no extent
3676 * or iunlinks, we can free up the entire log and set the tail_lsn to
3677 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3678 * lsn of the last known good LR on disk. If there are extent frees
3679 * or iunlinks they will have some entries in the AIL; so we look at
3680 * the AIL to determine how to set the tail_lsn.
3682 xlog_assign_tail_lsn(log->l_mp);
3685 * Now that we've finished replaying all buffer and inode
3686 * updates, re-read in the superblock.
3688 bp = xfs_getsb(log->l_mp, 0);
3690 ASSERT(!(XFS_BUF_ISWRITE(bp)));
3692 XFS_BUF_UNASYNC(bp);
3693 xfsbdstrat(log->l_mp, bp);
3694 error = xfs_buf_iowait(bp);
3696 xfs_buf_ioerror_alert(bp, __func__);
3702 /* Convert superblock from on-disk format */
3703 sbp = &log->l_mp->m_sb;
3704 xfs_sb_from_disk(log->l_mp, XFS_BUF_TO_SBP(bp));
3705 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
3706 ASSERT(xfs_sb_good_version(sbp));
3709 /* We've re-read the superblock so re-initialize per-cpu counters */
3710 xfs_icsb_reinit_counters(log->l_mp);
3712 xlog_recover_check_summary(log);
3714 /* Normal transactions can now occur */
3715 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3720 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3722 * Return error or zero.
3728 xfs_daddr_t head_blk, tail_blk;
3731 /* find the tail of the log */
3732 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
3735 if (tail_blk != head_blk) {
3736 /* There used to be a comment here:
3738 * disallow recovery on read-only mounts. note -- mount
3739 * checks for ENOSPC and turns it into an intelligent
3741 * ...but this is no longer true. Now, unless you specify
3742 * NORECOVERY (in which case this function would never be
3743 * called), we just go ahead and recover. We do this all
3744 * under the vfs layer, so we can get away with it unless
3745 * the device itself is read-only, in which case we fail.
3747 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
3751 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
3752 log->l_mp->m_logname ? log->l_mp->m_logname
3755 error = xlog_do_recover(log, head_blk, tail_blk);
3756 log->l_flags |= XLOG_RECOVERY_NEEDED;
3762 * In the first part of recovery we replay inodes and buffers and build
3763 * up the list of extent free items which need to be processed. Here
3764 * we process the extent free items and clean up the on disk unlinked
3765 * inode lists. This is separated from the first part of recovery so
3766 * that the root and real-time bitmap inodes can be read in from disk in
3767 * between the two stages. This is necessary so that we can free space
3768 * in the real-time portion of the file system.
3771 xlog_recover_finish(
3775 * Now we're ready to do the transactions needed for the
3776 * rest of recovery. Start with completing all the extent
3777 * free intent records and then process the unlinked inode
3778 * lists. At this point, we essentially run in normal mode
3779 * except that we're still performing recovery actions
3780 * rather than accepting new requests.
3782 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
3784 error = xlog_recover_process_efis(log);
3786 xfs_alert(log->l_mp, "Failed to recover EFIs");
3790 * Sync the log to get all the EFIs out of the AIL.
3791 * This isn't absolutely necessary, but it helps in
3792 * case the unlink transactions would have problems
3793 * pushing the EFIs out of the way.
3795 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
3797 xlog_recover_process_iunlinks(log);
3799 xlog_recover_check_summary(log);
3801 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
3802 log->l_mp->m_logname ? log->l_mp->m_logname
3804 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
3806 xfs_info(log->l_mp, "Ending clean mount");
3814 * Read all of the agf and agi counters and check that they
3815 * are consistent with the superblock counters.
3818 xlog_recover_check_summary(
3825 xfs_agnumber_t agno;
3826 __uint64_t freeblks;
3836 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3837 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
3839 xfs_alert(mp, "%s agf read failed agno %d error %d",
3840 __func__, agno, error);
3842 agfp = XFS_BUF_TO_AGF(agfbp);
3843 freeblks += be32_to_cpu(agfp->agf_freeblks) +
3844 be32_to_cpu(agfp->agf_flcount);
3845 xfs_buf_relse(agfbp);
3848 error = xfs_read_agi(mp, NULL, agno, &agibp);
3850 xfs_alert(mp, "%s agi read failed agno %d error %d",
3851 __func__, agno, error);
3853 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
3855 itotal += be32_to_cpu(agi->agi_count);
3856 ifree += be32_to_cpu(agi->agi_freecount);
3857 xfs_buf_relse(agibp);