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_format.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_btree.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_alloc.h"
37 #include "xfs_ialloc.h"
38 #include "xfs_log_priv.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_log_recover.h"
41 #include "xfs_extfree_item.h"
42 #include "xfs_trans_priv.h"
43 #include "xfs_quota.h"
44 #include "xfs_utils.h"
45 #include "xfs_cksum.h"
46 #include "xfs_trace.h"
47 #include "xfs_icache.h"
48 #include "xfs_icreate_item.h"
50 /* Need all the magic numbers and buffer ops structures from these headers */
51 #include "xfs_symlink.h"
52 #include "xfs_da_btree.h"
53 #include "xfs_dir2_format.h"
55 #include "xfs_attr_leaf.h"
56 #include "xfs_attr_remote.h"
58 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
65 xlog_clear_stale_blocks(
70 xlog_recover_check_summary(
73 #define xlog_recover_check_summary(log)
77 * This structure is used during recovery to record the buf log items which
78 * have been canceled and should not be replayed.
80 struct xfs_buf_cancel {
84 struct list_head bc_list;
88 * Sector aligned buffer routines for buffer create/read/write/access
92 * Verify the given count of basic blocks is valid number of blocks
93 * to specify for an operation involving the given XFS log buffer.
94 * Returns nonzero if the count is valid, 0 otherwise.
98 xlog_buf_bbcount_valid(
102 return bbcount > 0 && bbcount <= log->l_logBBsize;
106 * Allocate a buffer to hold log data. The buffer needs to be able
107 * to map to a range of nbblks basic blocks at any valid (basic
108 * block) offset within the log.
117 if (!xlog_buf_bbcount_valid(log, nbblks)) {
118 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
120 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
125 * We do log I/O in units of log sectors (a power-of-2
126 * multiple of the basic block size), so we round up the
127 * requested size to accommodate the basic blocks required
128 * for complete log sectors.
130 * In addition, the buffer may be used for a non-sector-
131 * aligned block offset, in which case an I/O of the
132 * requested size could extend beyond the end of the
133 * buffer. If the requested size is only 1 basic block it
134 * will never straddle a sector boundary, so this won't be
135 * an issue. Nor will this be a problem if the log I/O is
136 * done in basic blocks (sector size 1). But otherwise we
137 * extend the buffer by one extra log sector to ensure
138 * there's space to accommodate this possibility.
140 if (nbblks > 1 && log->l_sectBBsize > 1)
141 nbblks += log->l_sectBBsize;
142 nbblks = round_up(nbblks, log->l_sectBBsize);
144 bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0);
158 * Return the address of the start of the given block number's data
159 * in a log buffer. The buffer covers a log sector-aligned region.
168 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
170 ASSERT(offset + nbblks <= bp->b_length);
171 return bp->b_addr + BBTOB(offset);
176 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
187 if (!xlog_buf_bbcount_valid(log, nbblks)) {
188 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
190 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
194 blk_no = round_down(blk_no, log->l_sectBBsize);
195 nbblks = round_up(nbblks, log->l_sectBBsize);
198 ASSERT(nbblks <= bp->b_length);
200 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
202 bp->b_io_length = nbblks;
205 xfsbdstrat(log->l_mp, bp);
206 error = xfs_buf_iowait(bp);
208 xfs_buf_ioerror_alert(bp, __func__);
222 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
226 *offset = xlog_align(log, blk_no, nbblks, bp);
231 * Read at an offset into the buffer. Returns with the buffer in it's original
232 * state regardless of the result of the read.
237 xfs_daddr_t blk_no, /* block to read from */
238 int nbblks, /* blocks to read */
242 xfs_caddr_t orig_offset = bp->b_addr;
243 int orig_len = BBTOB(bp->b_length);
246 error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks));
250 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
252 /* must reset buffer pointer even on error */
253 error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len);
260 * Write out the buffer at the given block for the given number of blocks.
261 * The buffer is kept locked across the write and is returned locked.
262 * This can only be used for synchronous log writes.
273 if (!xlog_buf_bbcount_valid(log, nbblks)) {
274 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
276 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
280 blk_no = round_down(blk_no, log->l_sectBBsize);
281 nbblks = round_up(nbblks, log->l_sectBBsize);
284 ASSERT(nbblks <= bp->b_length);
286 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
287 XFS_BUF_ZEROFLAGS(bp);
290 bp->b_io_length = nbblks;
293 error = xfs_bwrite(bp);
295 xfs_buf_ioerror_alert(bp, __func__);
302 * dump debug superblock and log record information
305 xlog_header_check_dump(
307 xlog_rec_header_t *head)
309 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d\n",
310 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
311 xfs_debug(mp, " log : uuid = %pU, fmt = %d\n",
312 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
315 #define xlog_header_check_dump(mp, head)
319 * check log record header for recovery
322 xlog_header_check_recover(
324 xlog_rec_header_t *head)
326 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
329 * IRIX doesn't write the h_fmt field and leaves it zeroed
330 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
331 * a dirty log created in IRIX.
333 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
335 "dirty log written in incompatible format - can't recover");
336 xlog_header_check_dump(mp, head);
337 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
338 XFS_ERRLEVEL_HIGH, mp);
339 return XFS_ERROR(EFSCORRUPTED);
340 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
342 "dirty log entry has mismatched uuid - can't recover");
343 xlog_header_check_dump(mp, head);
344 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
345 XFS_ERRLEVEL_HIGH, mp);
346 return XFS_ERROR(EFSCORRUPTED);
352 * read the head block of the log and check the header
355 xlog_header_check_mount(
357 xlog_rec_header_t *head)
359 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
361 if (uuid_is_nil(&head->h_fs_uuid)) {
363 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
364 * h_fs_uuid is nil, we assume this log was last mounted
365 * by IRIX and continue.
367 xfs_warn(mp, "nil uuid in log - IRIX style log");
368 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
369 xfs_warn(mp, "log has mismatched uuid - can't recover");
370 xlog_header_check_dump(mp, head);
371 XFS_ERROR_REPORT("xlog_header_check_mount",
372 XFS_ERRLEVEL_HIGH, mp);
373 return XFS_ERROR(EFSCORRUPTED);
384 * We're not going to bother about retrying
385 * this during recovery. One strike!
387 xfs_buf_ioerror_alert(bp, __func__);
388 xfs_force_shutdown(bp->b_target->bt_mount,
389 SHUTDOWN_META_IO_ERROR);
392 xfs_buf_ioend(bp, 0);
396 * This routine finds (to an approximation) the first block in the physical
397 * log which contains the given cycle. It uses a binary search algorithm.
398 * Note that the algorithm can not be perfect because the disk will not
399 * necessarily be perfect.
402 xlog_find_cycle_start(
405 xfs_daddr_t first_blk,
406 xfs_daddr_t *last_blk,
416 mid_blk = BLK_AVG(first_blk, end_blk);
417 while (mid_blk != first_blk && mid_blk != end_blk) {
418 error = xlog_bread(log, mid_blk, 1, bp, &offset);
421 mid_cycle = xlog_get_cycle(offset);
422 if (mid_cycle == cycle)
423 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
425 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
426 mid_blk = BLK_AVG(first_blk, end_blk);
428 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
429 (mid_blk == end_blk && mid_blk-1 == first_blk));
437 * Check that a range of blocks does not contain stop_on_cycle_no.
438 * Fill in *new_blk with the block offset where such a block is
439 * found, or with -1 (an invalid block number) if there is no such
440 * block in the range. The scan needs to occur from front to back
441 * and the pointer into the region must be updated since a later
442 * routine will need to perform another test.
445 xlog_find_verify_cycle(
447 xfs_daddr_t start_blk,
449 uint stop_on_cycle_no,
450 xfs_daddr_t *new_blk)
456 xfs_caddr_t buf = NULL;
460 * Greedily allocate a buffer big enough to handle the full
461 * range of basic blocks we'll be examining. If that fails,
462 * try a smaller size. We need to be able to read at least
463 * a log sector, or we're out of luck.
465 bufblks = 1 << ffs(nbblks);
466 while (bufblks > log->l_logBBsize)
468 while (!(bp = xlog_get_bp(log, bufblks))) {
470 if (bufblks < log->l_sectBBsize)
474 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
477 bcount = min(bufblks, (start_blk + nbblks - i));
479 error = xlog_bread(log, i, bcount, bp, &buf);
483 for (j = 0; j < bcount; j++) {
484 cycle = xlog_get_cycle(buf);
485 if (cycle == stop_on_cycle_no) {
502 * Potentially backup over partial log record write.
504 * In the typical case, last_blk is the number of the block directly after
505 * a good log record. Therefore, we subtract one to get the block number
506 * of the last block in the given buffer. extra_bblks contains the number
507 * of blocks we would have read on a previous read. This happens when the
508 * last log record is split over the end of the physical log.
510 * extra_bblks is the number of blocks potentially verified on a previous
511 * call to this routine.
514 xlog_find_verify_log_record(
516 xfs_daddr_t start_blk,
517 xfs_daddr_t *last_blk,
522 xfs_caddr_t offset = NULL;
523 xlog_rec_header_t *head = NULL;
526 int num_blks = *last_blk - start_blk;
529 ASSERT(start_blk != 0 || *last_blk != start_blk);
531 if (!(bp = xlog_get_bp(log, num_blks))) {
532 if (!(bp = xlog_get_bp(log, 1)))
536 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
539 offset += ((num_blks - 1) << BBSHIFT);
542 for (i = (*last_blk) - 1; i >= 0; i--) {
544 /* valid log record not found */
546 "Log inconsistent (didn't find previous header)");
548 error = XFS_ERROR(EIO);
553 error = xlog_bread(log, i, 1, bp, &offset);
558 head = (xlog_rec_header_t *)offset;
560 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
568 * We hit the beginning of the physical log & still no header. Return
569 * to caller. If caller can handle a return of -1, then this routine
570 * will be called again for the end of the physical log.
578 * We have the final block of the good log (the first block
579 * of the log record _before_ the head. So we check the uuid.
581 if ((error = xlog_header_check_mount(log->l_mp, head)))
585 * We may have found a log record header before we expected one.
586 * last_blk will be the 1st block # with a given cycle #. We may end
587 * up reading an entire log record. In this case, we don't want to
588 * reset last_blk. Only when last_blk points in the middle of a log
589 * record do we update last_blk.
591 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
592 uint h_size = be32_to_cpu(head->h_size);
594 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
595 if (h_size % XLOG_HEADER_CYCLE_SIZE)
601 if (*last_blk - i + extra_bblks !=
602 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
611 * Head is defined to be the point of the log where the next log write
612 * write could go. This means that incomplete LR writes at the end are
613 * eliminated when calculating the head. We aren't guaranteed that previous
614 * LR have complete transactions. We only know that a cycle number of
615 * current cycle number -1 won't be present in the log if we start writing
616 * from our current block number.
618 * last_blk contains the block number of the first block with a given
621 * Return: zero if normal, non-zero if error.
626 xfs_daddr_t *return_head_blk)
630 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
632 uint first_half_cycle, last_half_cycle;
634 int error, log_bbnum = log->l_logBBsize;
636 /* Is the end of the log device zeroed? */
637 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
638 *return_head_blk = first_blk;
640 /* Is the whole lot zeroed? */
642 /* Linux XFS shouldn't generate totally zeroed logs -
643 * mkfs etc write a dummy unmount record to a fresh
644 * log so we can store the uuid in there
646 xfs_warn(log->l_mp, "totally zeroed log");
651 xfs_warn(log->l_mp, "empty log check failed");
655 first_blk = 0; /* get cycle # of 1st block */
656 bp = xlog_get_bp(log, 1);
660 error = xlog_bread(log, 0, 1, bp, &offset);
664 first_half_cycle = xlog_get_cycle(offset);
666 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
667 error = xlog_bread(log, last_blk, 1, bp, &offset);
671 last_half_cycle = xlog_get_cycle(offset);
672 ASSERT(last_half_cycle != 0);
675 * If the 1st half cycle number is equal to the last half cycle number,
676 * then the entire log is stamped with the same cycle number. In this
677 * case, head_blk can't be set to zero (which makes sense). The below
678 * math doesn't work out properly with head_blk equal to zero. Instead,
679 * we set it to log_bbnum which is an invalid block number, but this
680 * value makes the math correct. If head_blk doesn't changed through
681 * all the tests below, *head_blk is set to zero at the very end rather
682 * than log_bbnum. In a sense, log_bbnum and zero are the same block
683 * in a circular file.
685 if (first_half_cycle == last_half_cycle) {
687 * In this case we believe that the entire log should have
688 * cycle number last_half_cycle. We need to scan backwards
689 * from the end verifying that there are no holes still
690 * containing last_half_cycle - 1. If we find such a hole,
691 * then the start of that hole will be the new head. The
692 * simple case looks like
693 * x | x ... | x - 1 | x
694 * Another case that fits this picture would be
695 * x | x + 1 | x ... | x
696 * In this case the head really is somewhere at the end of the
697 * log, as one of the latest writes at the beginning was
700 * x | x + 1 | x ... | x - 1 | x
701 * This is really the combination of the above two cases, and
702 * the head has to end up at the start of the x-1 hole at the
705 * In the 256k log case, we will read from the beginning to the
706 * end of the log and search for cycle numbers equal to x-1.
707 * We don't worry about the x+1 blocks that we encounter,
708 * because we know that they cannot be the head since the log
711 head_blk = log_bbnum;
712 stop_on_cycle = last_half_cycle - 1;
715 * In this case we want to find the first block with cycle
716 * number matching last_half_cycle. We expect the log to be
718 * x + 1 ... | x ... | x
719 * The first block with cycle number x (last_half_cycle) will
720 * be where the new head belongs. First we do a binary search
721 * for the first occurrence of last_half_cycle. The binary
722 * search may not be totally accurate, so then we scan back
723 * from there looking for occurrences of last_half_cycle before
724 * us. If that backwards scan wraps around the beginning of
725 * the log, then we look for occurrences of last_half_cycle - 1
726 * at the end of the log. The cases we're looking for look
728 * v binary search stopped here
729 * x + 1 ... | x | x + 1 | x ... | x
730 * ^ but we want to locate this spot
732 * <---------> less than scan distance
733 * x + 1 ... | x ... | x - 1 | x
734 * ^ we want to locate this spot
736 stop_on_cycle = last_half_cycle;
737 if ((error = xlog_find_cycle_start(log, bp, first_blk,
738 &head_blk, last_half_cycle)))
743 * Now validate the answer. Scan back some number of maximum possible
744 * blocks and make sure each one has the expected cycle number. The
745 * maximum is determined by the total possible amount of buffering
746 * in the in-core log. The following number can be made tighter if
747 * we actually look at the block size of the filesystem.
749 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
750 if (head_blk >= num_scan_bblks) {
752 * We are guaranteed that the entire check can be performed
755 start_blk = head_blk - num_scan_bblks;
756 if ((error = xlog_find_verify_cycle(log,
757 start_blk, num_scan_bblks,
758 stop_on_cycle, &new_blk)))
762 } else { /* need to read 2 parts of log */
764 * We are going to scan backwards in the log in two parts.
765 * First we scan the physical end of the log. In this part
766 * of the log, we are looking for blocks with cycle number
767 * last_half_cycle - 1.
768 * If we find one, then we know that the log starts there, as
769 * we've found a hole that didn't get written in going around
770 * the end of the physical log. The simple case for this is
771 * x + 1 ... | x ... | x - 1 | x
772 * <---------> less than scan distance
773 * If all of the blocks at the end of the log have cycle number
774 * last_half_cycle, then we check the blocks at the start of
775 * the log looking for occurrences of last_half_cycle. If we
776 * find one, then our current estimate for the location of the
777 * first occurrence of last_half_cycle is wrong and we move
778 * back to the hole we've found. This case looks like
779 * x + 1 ... | x | x + 1 | x ...
780 * ^ binary search stopped here
781 * Another case we need to handle that only occurs in 256k
783 * x + 1 ... | x ... | x+1 | x ...
784 * ^ binary search stops here
785 * In a 256k log, the scan at the end of the log will see the
786 * x + 1 blocks. We need to skip past those since that is
787 * certainly not the head of the log. By searching for
788 * last_half_cycle-1 we accomplish that.
790 ASSERT(head_blk <= INT_MAX &&
791 (xfs_daddr_t) num_scan_bblks >= head_blk);
792 start_blk = log_bbnum - (num_scan_bblks - head_blk);
793 if ((error = xlog_find_verify_cycle(log, start_blk,
794 num_scan_bblks - (int)head_blk,
795 (stop_on_cycle - 1), &new_blk)))
803 * Scan beginning of log now. The last part of the physical
804 * log is good. This scan needs to verify that it doesn't find
805 * the last_half_cycle.
808 ASSERT(head_blk <= INT_MAX);
809 if ((error = xlog_find_verify_cycle(log,
810 start_blk, (int)head_blk,
811 stop_on_cycle, &new_blk)))
819 * Now we need to make sure head_blk is not pointing to a block in
820 * the middle of a log record.
822 num_scan_bblks = XLOG_REC_SHIFT(log);
823 if (head_blk >= num_scan_bblks) {
824 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
826 /* start ptr at last block ptr before head_blk */
827 if ((error = xlog_find_verify_log_record(log, start_blk,
828 &head_blk, 0)) == -1) {
829 error = XFS_ERROR(EIO);
835 ASSERT(head_blk <= INT_MAX);
836 if ((error = xlog_find_verify_log_record(log, start_blk,
837 &head_blk, 0)) == -1) {
838 /* We hit the beginning of the log during our search */
839 start_blk = log_bbnum - (num_scan_bblks - head_blk);
841 ASSERT(start_blk <= INT_MAX &&
842 (xfs_daddr_t) log_bbnum-start_blk >= 0);
843 ASSERT(head_blk <= INT_MAX);
844 if ((error = xlog_find_verify_log_record(log,
846 (int)head_blk)) == -1) {
847 error = XFS_ERROR(EIO);
851 if (new_blk != log_bbnum)
858 if (head_blk == log_bbnum)
859 *return_head_blk = 0;
861 *return_head_blk = head_blk;
863 * When returning here, we have a good block number. Bad block
864 * means that during a previous crash, we didn't have a clean break
865 * from cycle number N to cycle number N-1. In this case, we need
866 * to find the first block with cycle number N-1.
874 xfs_warn(log->l_mp, "failed to find log head");
879 * Find the sync block number or the tail of the log.
881 * This will be the block number of the last record to have its
882 * associated buffers synced to disk. Every log record header has
883 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
884 * to get a sync block number. The only concern is to figure out which
885 * log record header to believe.
887 * The following algorithm uses the log record header with the largest
888 * lsn. The entire log record does not need to be valid. We only care
889 * that the header is valid.
891 * We could speed up search by using current head_blk buffer, but it is not
897 xfs_daddr_t *head_blk,
898 xfs_daddr_t *tail_blk)
900 xlog_rec_header_t *rhead;
901 xlog_op_header_t *op_head;
902 xfs_caddr_t offset = NULL;
905 xfs_daddr_t umount_data_blk;
906 xfs_daddr_t after_umount_blk;
913 * Find previous log record
915 if ((error = xlog_find_head(log, head_blk)))
918 bp = xlog_get_bp(log, 1);
921 if (*head_blk == 0) { /* special case */
922 error = xlog_bread(log, 0, 1, bp, &offset);
926 if (xlog_get_cycle(offset) == 0) {
928 /* leave all other log inited values alone */
934 * Search backwards looking for log record header block
936 ASSERT(*head_blk < INT_MAX);
937 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
938 error = xlog_bread(log, i, 1, bp, &offset);
942 if (*(__be32 *)offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
948 * If we haven't found the log record header block, start looking
949 * again from the end of the physical log. XXXmiken: There should be
950 * a check here to make sure we didn't search more than N blocks in
954 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
955 error = xlog_bread(log, i, 1, bp, &offset);
959 if (*(__be32 *)offset ==
960 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
967 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
969 return XFS_ERROR(EIO);
972 /* find blk_no of tail of log */
973 rhead = (xlog_rec_header_t *)offset;
974 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
977 * Reset log values according to the state of the log when we
978 * crashed. In the case where head_blk == 0, we bump curr_cycle
979 * one because the next write starts a new cycle rather than
980 * continuing the cycle of the last good log record. At this
981 * point we have guaranteed that all partial log records have been
982 * accounted for. Therefore, we know that the last good log record
983 * written was complete and ended exactly on the end boundary
984 * of the physical log.
986 log->l_prev_block = i;
987 log->l_curr_block = (int)*head_blk;
988 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
991 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
992 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
993 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
994 BBTOB(log->l_curr_block));
995 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
996 BBTOB(log->l_curr_block));
999 * Look for unmount record. If we find it, then we know there
1000 * was a clean unmount. Since 'i' could be the last block in
1001 * the physical log, we convert to a log block before comparing
1004 * Save the current tail lsn to use to pass to
1005 * xlog_clear_stale_blocks() below. We won't want to clear the
1006 * unmount record if there is one, so we pass the lsn of the
1007 * unmount record rather than the block after it.
1009 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1010 int h_size = be32_to_cpu(rhead->h_size);
1011 int h_version = be32_to_cpu(rhead->h_version);
1013 if ((h_version & XLOG_VERSION_2) &&
1014 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
1015 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
1016 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1024 after_umount_blk = (i + hblks + (int)
1025 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
1026 tail_lsn = atomic64_read(&log->l_tail_lsn);
1027 if (*head_blk == after_umount_blk &&
1028 be32_to_cpu(rhead->h_num_logops) == 1) {
1029 umount_data_blk = (i + hblks) % log->l_logBBsize;
1030 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1034 op_head = (xlog_op_header_t *)offset;
1035 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1037 * Set tail and last sync so that newly written
1038 * log records will point recovery to after the
1039 * current unmount record.
1041 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1042 log->l_curr_cycle, after_umount_blk);
1043 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1044 log->l_curr_cycle, after_umount_blk);
1045 *tail_blk = after_umount_blk;
1048 * Note that the unmount was clean. If the unmount
1049 * was not clean, we need to know this to rebuild the
1050 * superblock counters from the perag headers if we
1051 * have a filesystem using non-persistent counters.
1053 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1058 * Make sure that there are no blocks in front of the head
1059 * with the same cycle number as the head. This can happen
1060 * because we allow multiple outstanding log writes concurrently,
1061 * and the later writes might make it out before earlier ones.
1063 * We use the lsn from before modifying it so that we'll never
1064 * overwrite the unmount record after a clean unmount.
1066 * Do this only if we are going to recover the filesystem
1068 * NOTE: This used to say "if (!readonly)"
1069 * However on Linux, we can & do recover a read-only filesystem.
1070 * We only skip recovery if NORECOVERY is specified on mount,
1071 * in which case we would not be here.
1073 * But... if the -device- itself is readonly, just skip this.
1074 * We can't recover this device anyway, so it won't matter.
1076 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
1077 error = xlog_clear_stale_blocks(log, tail_lsn);
1083 xfs_warn(log->l_mp, "failed to locate log tail");
1088 * Is the log zeroed at all?
1090 * The last binary search should be changed to perform an X block read
1091 * once X becomes small enough. You can then search linearly through
1092 * the X blocks. This will cut down on the number of reads we need to do.
1094 * If the log is partially zeroed, this routine will pass back the blkno
1095 * of the first block with cycle number 0. It won't have a complete LR
1099 * 0 => the log is completely written to
1100 * -1 => use *blk_no as the first block of the log
1101 * >0 => error has occurred
1106 xfs_daddr_t *blk_no)
1110 uint first_cycle, last_cycle;
1111 xfs_daddr_t new_blk, last_blk, start_blk;
1112 xfs_daddr_t num_scan_bblks;
1113 int error, log_bbnum = log->l_logBBsize;
1117 /* check totally zeroed log */
1118 bp = xlog_get_bp(log, 1);
1121 error = xlog_bread(log, 0, 1, bp, &offset);
1125 first_cycle = xlog_get_cycle(offset);
1126 if (first_cycle == 0) { /* completely zeroed log */
1132 /* check partially zeroed log */
1133 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1137 last_cycle = xlog_get_cycle(offset);
1138 if (last_cycle != 0) { /* log completely written to */
1141 } else if (first_cycle != 1) {
1143 * If the cycle of the last block is zero, the cycle of
1144 * the first block must be 1. If it's not, maybe we're
1145 * not looking at a log... Bail out.
1148 "Log inconsistent or not a log (last==0, first!=1)");
1149 return XFS_ERROR(EINVAL);
1152 /* we have a partially zeroed log */
1153 last_blk = log_bbnum-1;
1154 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1158 * Validate the answer. Because there is no way to guarantee that
1159 * the entire log is made up of log records which are the same size,
1160 * we scan over the defined maximum blocks. At this point, the maximum
1161 * is not chosen to mean anything special. XXXmiken
1163 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1164 ASSERT(num_scan_bblks <= INT_MAX);
1166 if (last_blk < num_scan_bblks)
1167 num_scan_bblks = last_blk;
1168 start_blk = last_blk - num_scan_bblks;
1171 * We search for any instances of cycle number 0 that occur before
1172 * our current estimate of the head. What we're trying to detect is
1173 * 1 ... | 0 | 1 | 0...
1174 * ^ binary search ends here
1176 if ((error = xlog_find_verify_cycle(log, start_blk,
1177 (int)num_scan_bblks, 0, &new_blk)))
1183 * Potentially backup over partial log record write. We don't need
1184 * to search the end of the log because we know it is zero.
1186 if ((error = xlog_find_verify_log_record(log, start_blk,
1187 &last_blk, 0)) == -1) {
1188 error = XFS_ERROR(EIO);
1202 * These are simple subroutines used by xlog_clear_stale_blocks() below
1203 * to initialize a buffer full of empty log record headers and write
1204 * them into the log.
1215 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1217 memset(buf, 0, BBSIZE);
1218 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1219 recp->h_cycle = cpu_to_be32(cycle);
1220 recp->h_version = cpu_to_be32(
1221 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1222 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1223 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1224 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1225 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1229 xlog_write_log_records(
1240 int sectbb = log->l_sectBBsize;
1241 int end_block = start_block + blocks;
1247 * Greedily allocate a buffer big enough to handle the full
1248 * range of basic blocks to be written. If that fails, try
1249 * a smaller size. We need to be able to write at least a
1250 * log sector, or we're out of luck.
1252 bufblks = 1 << ffs(blocks);
1253 while (bufblks > log->l_logBBsize)
1255 while (!(bp = xlog_get_bp(log, bufblks))) {
1257 if (bufblks < sectbb)
1261 /* We may need to do a read at the start to fill in part of
1262 * the buffer in the starting sector not covered by the first
1265 balign = round_down(start_block, sectbb);
1266 if (balign != start_block) {
1267 error = xlog_bread_noalign(log, start_block, 1, bp);
1271 j = start_block - balign;
1274 for (i = start_block; i < end_block; i += bufblks) {
1275 int bcount, endcount;
1277 bcount = min(bufblks, end_block - start_block);
1278 endcount = bcount - j;
1280 /* We may need to do a read at the end to fill in part of
1281 * the buffer in the final sector not covered by the write.
1282 * If this is the same sector as the above read, skip it.
1284 ealign = round_down(end_block, sectbb);
1285 if (j == 0 && (start_block + endcount > ealign)) {
1286 offset = bp->b_addr + BBTOB(ealign - start_block);
1287 error = xlog_bread_offset(log, ealign, sectbb,
1294 offset = xlog_align(log, start_block, endcount, bp);
1295 for (; j < endcount; j++) {
1296 xlog_add_record(log, offset, cycle, i+j,
1297 tail_cycle, tail_block);
1300 error = xlog_bwrite(log, start_block, endcount, bp);
1303 start_block += endcount;
1313 * This routine is called to blow away any incomplete log writes out
1314 * in front of the log head. We do this so that we won't become confused
1315 * if we come up, write only a little bit more, and then crash again.
1316 * If we leave the partial log records out there, this situation could
1317 * cause us to think those partial writes are valid blocks since they
1318 * have the current cycle number. We get rid of them by overwriting them
1319 * with empty log records with the old cycle number rather than the
1322 * The tail lsn is passed in rather than taken from
1323 * the log so that we will not write over the unmount record after a
1324 * clean unmount in a 512 block log. Doing so would leave the log without
1325 * any valid log records in it until a new one was written. If we crashed
1326 * during that time we would not be able to recover.
1329 xlog_clear_stale_blocks(
1333 int tail_cycle, head_cycle;
1334 int tail_block, head_block;
1335 int tail_distance, max_distance;
1339 tail_cycle = CYCLE_LSN(tail_lsn);
1340 tail_block = BLOCK_LSN(tail_lsn);
1341 head_cycle = log->l_curr_cycle;
1342 head_block = log->l_curr_block;
1345 * Figure out the distance between the new head of the log
1346 * and the tail. We want to write over any blocks beyond the
1347 * head that we may have written just before the crash, but
1348 * we don't want to overwrite the tail of the log.
1350 if (head_cycle == tail_cycle) {
1352 * The tail is behind the head in the physical log,
1353 * so the distance from the head to the tail is the
1354 * distance from the head to the end of the log plus
1355 * the distance from the beginning of the log to the
1358 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1359 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1360 XFS_ERRLEVEL_LOW, log->l_mp);
1361 return XFS_ERROR(EFSCORRUPTED);
1363 tail_distance = tail_block + (log->l_logBBsize - head_block);
1366 * The head is behind the tail in the physical log,
1367 * so the distance from the head to the tail is just
1368 * the tail block minus the head block.
1370 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1371 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1372 XFS_ERRLEVEL_LOW, log->l_mp);
1373 return XFS_ERROR(EFSCORRUPTED);
1375 tail_distance = tail_block - head_block;
1379 * If the head is right up against the tail, we can't clear
1382 if (tail_distance <= 0) {
1383 ASSERT(tail_distance == 0);
1387 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1389 * Take the smaller of the maximum amount of outstanding I/O
1390 * we could have and the distance to the tail to clear out.
1391 * We take the smaller so that we don't overwrite the tail and
1392 * we don't waste all day writing from the head to the tail
1395 max_distance = MIN(max_distance, tail_distance);
1397 if ((head_block + max_distance) <= log->l_logBBsize) {
1399 * We can stomp all the blocks we need to without
1400 * wrapping around the end of the log. Just do it
1401 * in a single write. Use the cycle number of the
1402 * current cycle minus one so that the log will look like:
1405 error = xlog_write_log_records(log, (head_cycle - 1),
1406 head_block, max_distance, tail_cycle,
1412 * We need to wrap around the end of the physical log in
1413 * order to clear all the blocks. Do it in two separate
1414 * I/Os. The first write should be from the head to the
1415 * end of the physical log, and it should use the current
1416 * cycle number minus one just like above.
1418 distance = log->l_logBBsize - head_block;
1419 error = xlog_write_log_records(log, (head_cycle - 1),
1420 head_block, distance, tail_cycle,
1427 * Now write the blocks at the start of the physical log.
1428 * This writes the remainder of the blocks we want to clear.
1429 * It uses the current cycle number since we're now on the
1430 * same cycle as the head so that we get:
1431 * n ... n ... | n - 1 ...
1432 * ^^^^^ blocks we're writing
1434 distance = max_distance - (log->l_logBBsize - head_block);
1435 error = xlog_write_log_records(log, head_cycle, 0, distance,
1436 tail_cycle, tail_block);
1444 /******************************************************************************
1446 * Log recover routines
1448 ******************************************************************************
1451 STATIC xlog_recover_t *
1452 xlog_recover_find_tid(
1453 struct hlist_head *head,
1456 xlog_recover_t *trans;
1458 hlist_for_each_entry(trans, head, r_list) {
1459 if (trans->r_log_tid == tid)
1466 xlog_recover_new_tid(
1467 struct hlist_head *head,
1471 xlog_recover_t *trans;
1473 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1474 trans->r_log_tid = tid;
1476 INIT_LIST_HEAD(&trans->r_itemq);
1478 INIT_HLIST_NODE(&trans->r_list);
1479 hlist_add_head(&trans->r_list, head);
1483 xlog_recover_add_item(
1484 struct list_head *head)
1486 xlog_recover_item_t *item;
1488 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1489 INIT_LIST_HEAD(&item->ri_list);
1490 list_add_tail(&item->ri_list, head);
1494 xlog_recover_add_to_cont_trans(
1496 struct xlog_recover *trans,
1500 xlog_recover_item_t *item;
1501 xfs_caddr_t ptr, old_ptr;
1504 if (list_empty(&trans->r_itemq)) {
1505 /* finish copying rest of trans header */
1506 xlog_recover_add_item(&trans->r_itemq);
1507 ptr = (xfs_caddr_t) &trans->r_theader +
1508 sizeof(xfs_trans_header_t) - len;
1509 memcpy(ptr, dp, len); /* d, s, l */
1512 /* take the tail entry */
1513 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1515 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1516 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1518 ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP);
1519 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1520 item->ri_buf[item->ri_cnt-1].i_len += len;
1521 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1522 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
1527 * The next region to add is the start of a new region. It could be
1528 * a whole region or it could be the first part of a new region. Because
1529 * of this, the assumption here is that the type and size fields of all
1530 * format structures fit into the first 32 bits of the structure.
1532 * This works because all regions must be 32 bit aligned. Therefore, we
1533 * either have both fields or we have neither field. In the case we have
1534 * neither field, the data part of the region is zero length. We only have
1535 * a log_op_header and can throw away the header since a new one will appear
1536 * later. If we have at least 4 bytes, then we can determine how many regions
1537 * will appear in the current log item.
1540 xlog_recover_add_to_trans(
1542 struct xlog_recover *trans,
1546 xfs_inode_log_format_t *in_f; /* any will do */
1547 xlog_recover_item_t *item;
1552 if (list_empty(&trans->r_itemq)) {
1553 /* we need to catch log corruptions here */
1554 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
1555 xfs_warn(log->l_mp, "%s: bad header magic number",
1558 return XFS_ERROR(EIO);
1560 if (len == sizeof(xfs_trans_header_t))
1561 xlog_recover_add_item(&trans->r_itemq);
1562 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1566 ptr = kmem_alloc(len, KM_SLEEP);
1567 memcpy(ptr, dp, len);
1568 in_f = (xfs_inode_log_format_t *)ptr;
1570 /* take the tail entry */
1571 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1572 if (item->ri_total != 0 &&
1573 item->ri_total == item->ri_cnt) {
1574 /* tail item is in use, get a new one */
1575 xlog_recover_add_item(&trans->r_itemq);
1576 item = list_entry(trans->r_itemq.prev,
1577 xlog_recover_item_t, ri_list);
1580 if (item->ri_total == 0) { /* first region to be added */
1581 if (in_f->ilf_size == 0 ||
1582 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
1584 "bad number of regions (%d) in inode log format",
1587 return XFS_ERROR(EIO);
1590 item->ri_total = in_f->ilf_size;
1592 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1595 ASSERT(item->ri_total > item->ri_cnt);
1596 /* Description region is ri_buf[0] */
1597 item->ri_buf[item->ri_cnt].i_addr = ptr;
1598 item->ri_buf[item->ri_cnt].i_len = len;
1600 trace_xfs_log_recover_item_add(log, trans, item, 0);
1605 * Sort the log items in the transaction.
1607 * The ordering constraints are defined by the inode allocation and unlink
1608 * behaviour. The rules are:
1610 * 1. Every item is only logged once in a given transaction. Hence it
1611 * represents the last logged state of the item. Hence ordering is
1612 * dependent on the order in which operations need to be performed so
1613 * required initial conditions are always met.
1615 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1616 * there's nothing to replay from them so we can simply cull them
1617 * from the transaction. However, we can't do that until after we've
1618 * replayed all the other items because they may be dependent on the
1619 * cancelled buffer and replaying the cancelled buffer can remove it
1620 * form the cancelled buffer table. Hence they have tobe done last.
1622 * 3. Inode allocation buffers must be replayed before inode items that
1623 * read the buffer and replay changes into it. For filesystems using the
1624 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1625 * treated the same as inode allocation buffers as they create and
1626 * initialise the buffers directly.
1628 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1629 * This ensures that inodes are completely flushed to the inode buffer
1630 * in a "free" state before we remove the unlinked inode list pointer.
1632 * Hence the ordering needs to be inode allocation buffers first, inode items
1633 * second, inode unlink buffers third and cancelled buffers last.
1635 * But there's a problem with that - we can't tell an inode allocation buffer
1636 * apart from a regular buffer, so we can't separate them. We can, however,
1637 * tell an inode unlink buffer from the others, and so we can separate them out
1638 * from all the other buffers and move them to last.
1640 * Hence, 4 lists, in order from head to tail:
1641 * - buffer_list for all buffers except cancelled/inode unlink buffers
1642 * - item_list for all non-buffer items
1643 * - inode_buffer_list for inode unlink buffers
1644 * - cancel_list for the cancelled buffers
1646 * Note that we add objects to the tail of the lists so that first-to-last
1647 * ordering is preserved within the lists. Adding objects to the head of the
1648 * list means when we traverse from the head we walk them in last-to-first
1649 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1650 * but for all other items there may be specific ordering that we need to
1654 xlog_recover_reorder_trans(
1656 struct xlog_recover *trans,
1659 xlog_recover_item_t *item, *n;
1660 LIST_HEAD(sort_list);
1661 LIST_HEAD(cancel_list);
1662 LIST_HEAD(buffer_list);
1663 LIST_HEAD(inode_buffer_list);
1664 LIST_HEAD(inode_list);
1666 list_splice_init(&trans->r_itemq, &sort_list);
1667 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1668 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1670 switch (ITEM_TYPE(item)) {
1671 case XFS_LI_ICREATE:
1672 list_move_tail(&item->ri_list, &buffer_list);
1675 if (buf_f->blf_flags & XFS_BLF_CANCEL) {
1676 trace_xfs_log_recover_item_reorder_head(log,
1678 list_move(&item->ri_list, &cancel_list);
1681 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1682 list_move(&item->ri_list, &inode_buffer_list);
1685 list_move_tail(&item->ri_list, &buffer_list);
1689 case XFS_LI_QUOTAOFF:
1692 trace_xfs_log_recover_item_reorder_tail(log,
1694 list_move_tail(&item->ri_list, &inode_list);
1698 "%s: unrecognized type of log operation",
1701 return XFS_ERROR(EIO);
1704 ASSERT(list_empty(&sort_list));
1705 if (!list_empty(&buffer_list))
1706 list_splice(&buffer_list, &trans->r_itemq);
1707 if (!list_empty(&inode_list))
1708 list_splice_tail(&inode_list, &trans->r_itemq);
1709 if (!list_empty(&inode_buffer_list))
1710 list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1711 if (!list_empty(&cancel_list))
1712 list_splice_tail(&cancel_list, &trans->r_itemq);
1717 * Build up the table of buf cancel records so that we don't replay
1718 * cancelled data in the second pass. For buffer records that are
1719 * not cancel records, there is nothing to do here so we just return.
1721 * If we get a cancel record which is already in the table, this indicates
1722 * that the buffer was cancelled multiple times. In order to ensure
1723 * that during pass 2 we keep the record in the table until we reach its
1724 * last occurrence in the log, we keep a reference count in the cancel
1725 * record in the table to tell us how many times we expect to see this
1726 * record during the second pass.
1729 xlog_recover_buffer_pass1(
1731 struct xlog_recover_item *item)
1733 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1734 struct list_head *bucket;
1735 struct xfs_buf_cancel *bcp;
1738 * If this isn't a cancel buffer item, then just return.
1740 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1741 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
1746 * Insert an xfs_buf_cancel record into the hash table of them.
1747 * If there is already an identical record, bump its reference count.
1749 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1750 list_for_each_entry(bcp, bucket, bc_list) {
1751 if (bcp->bc_blkno == buf_f->blf_blkno &&
1752 bcp->bc_len == buf_f->blf_len) {
1754 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
1759 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1760 bcp->bc_blkno = buf_f->blf_blkno;
1761 bcp->bc_len = buf_f->blf_len;
1762 bcp->bc_refcount = 1;
1763 list_add_tail(&bcp->bc_list, bucket);
1765 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
1770 * Check to see whether the buffer being recovered has a corresponding
1771 * entry in the buffer cancel record table. If it does then return 1
1772 * so that it will be cancelled, otherwise return 0. If the buffer is
1773 * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement
1774 * the refcount on the entry in the table and remove it from the table
1775 * if this is the last reference.
1777 * We remove the cancel record from the table when we encounter its
1778 * last occurrence in the log so that if the same buffer is re-used
1779 * again after its last cancellation we actually replay the changes
1780 * made at that point.
1783 xlog_check_buffer_cancelled(
1789 struct list_head *bucket;
1790 struct xfs_buf_cancel *bcp;
1792 if (log->l_buf_cancel_table == NULL) {
1794 * There is nothing in the table built in pass one,
1795 * so this buffer must not be cancelled.
1797 ASSERT(!(flags & XFS_BLF_CANCEL));
1802 * Search for an entry in the cancel table that matches our buffer.
1804 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1805 list_for_each_entry(bcp, bucket, bc_list) {
1806 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
1811 * We didn't find a corresponding entry in the table, so return 0 so
1812 * that the buffer is NOT cancelled.
1814 ASSERT(!(flags & XFS_BLF_CANCEL));
1819 * We've go a match, so return 1 so that the recovery of this buffer
1820 * is cancelled. If this buffer is actually a buffer cancel log
1821 * item, then decrement the refcount on the one in the table and
1822 * remove it if this is the last reference.
1824 if (flags & XFS_BLF_CANCEL) {
1825 if (--bcp->bc_refcount == 0) {
1826 list_del(&bcp->bc_list);
1834 * Perform recovery for a buffer full of inodes. In these buffers, the only
1835 * data which should be recovered is that which corresponds to the
1836 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1837 * data for the inodes is always logged through the inodes themselves rather
1838 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1840 * The only time when buffers full of inodes are fully recovered is when the
1841 * buffer is full of newly allocated inodes. In this case the buffer will
1842 * not be marked as an inode buffer and so will be sent to
1843 * xlog_recover_do_reg_buffer() below during recovery.
1846 xlog_recover_do_inode_buffer(
1847 struct xfs_mount *mp,
1848 xlog_recover_item_t *item,
1850 xfs_buf_log_format_t *buf_f)
1856 int reg_buf_offset = 0;
1857 int reg_buf_bytes = 0;
1858 int next_unlinked_offset;
1860 xfs_agino_t *logged_nextp;
1861 xfs_agino_t *buffer_nextp;
1863 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
1866 * Post recovery validation only works properly on CRC enabled
1869 if (xfs_sb_version_hascrc(&mp->m_sb))
1870 bp->b_ops = &xfs_inode_buf_ops;
1872 inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog;
1873 for (i = 0; i < inodes_per_buf; i++) {
1874 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1875 offsetof(xfs_dinode_t, di_next_unlinked);
1877 while (next_unlinked_offset >=
1878 (reg_buf_offset + reg_buf_bytes)) {
1880 * The next di_next_unlinked field is beyond
1881 * the current logged region. Find the next
1882 * logged region that contains or is beyond
1883 * the current di_next_unlinked field.
1886 bit = xfs_next_bit(buf_f->blf_data_map,
1887 buf_f->blf_map_size, bit);
1890 * If there are no more logged regions in the
1891 * buffer, then we're done.
1896 nbits = xfs_contig_bits(buf_f->blf_data_map,
1897 buf_f->blf_map_size, bit);
1899 reg_buf_offset = bit << XFS_BLF_SHIFT;
1900 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
1905 * If the current logged region starts after the current
1906 * di_next_unlinked field, then move on to the next
1907 * di_next_unlinked field.
1909 if (next_unlinked_offset < reg_buf_offset)
1912 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1913 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
1914 ASSERT((reg_buf_offset + reg_buf_bytes) <=
1915 BBTOB(bp->b_io_length));
1918 * The current logged region contains a copy of the
1919 * current di_next_unlinked field. Extract its value
1920 * and copy it to the buffer copy.
1922 logged_nextp = item->ri_buf[item_index].i_addr +
1923 next_unlinked_offset - reg_buf_offset;
1924 if (unlikely(*logged_nextp == 0)) {
1926 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1927 "Trying to replay bad (0) inode di_next_unlinked field.",
1929 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1930 XFS_ERRLEVEL_LOW, mp);
1931 return XFS_ERROR(EFSCORRUPTED);
1934 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1935 next_unlinked_offset);
1936 *buffer_nextp = *logged_nextp;
1939 * If necessary, recalculate the CRC in the on-disk inode. We
1940 * have to leave the inode in a consistent state for whoever
1943 xfs_dinode_calc_crc(mp, (struct xfs_dinode *)
1944 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
1952 * Validate the recovered buffer is of the correct type and attach the
1953 * appropriate buffer operations to them for writeback. Magic numbers are in a
1955 * the first 16 bits of the buffer (inode buffer, dquot buffer),
1956 * the first 32 bits of the buffer (most blocks),
1957 * inside a struct xfs_da_blkinfo at the start of the buffer.
1960 xlog_recovery_validate_buf_type(
1961 struct xfs_mount *mp,
1963 xfs_buf_log_format_t *buf_f)
1965 struct xfs_da_blkinfo *info = bp->b_addr;
1970 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
1971 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
1972 magicda = be16_to_cpu(info->magic);
1973 switch (xfs_blft_from_flags(buf_f)) {
1974 case XFS_BLFT_BTREE_BUF:
1976 case XFS_ABTB_CRC_MAGIC:
1977 case XFS_ABTC_CRC_MAGIC:
1978 case XFS_ABTB_MAGIC:
1979 case XFS_ABTC_MAGIC:
1980 bp->b_ops = &xfs_allocbt_buf_ops;
1982 case XFS_IBT_CRC_MAGIC:
1984 bp->b_ops = &xfs_inobt_buf_ops;
1986 case XFS_BMAP_CRC_MAGIC:
1987 case XFS_BMAP_MAGIC:
1988 bp->b_ops = &xfs_bmbt_buf_ops;
1991 xfs_warn(mp, "Bad btree block magic!");
1996 case XFS_BLFT_AGF_BUF:
1997 if (magic32 != XFS_AGF_MAGIC) {
1998 xfs_warn(mp, "Bad AGF block magic!");
2002 bp->b_ops = &xfs_agf_buf_ops;
2004 case XFS_BLFT_AGFL_BUF:
2005 if (!xfs_sb_version_hascrc(&mp->m_sb))
2007 if (magic32 != XFS_AGFL_MAGIC) {
2008 xfs_warn(mp, "Bad AGFL block magic!");
2012 bp->b_ops = &xfs_agfl_buf_ops;
2014 case XFS_BLFT_AGI_BUF:
2015 if (magic32 != XFS_AGI_MAGIC) {
2016 xfs_warn(mp, "Bad AGI block magic!");
2020 bp->b_ops = &xfs_agi_buf_ops;
2022 case XFS_BLFT_UDQUOT_BUF:
2023 case XFS_BLFT_PDQUOT_BUF:
2024 case XFS_BLFT_GDQUOT_BUF:
2025 #ifdef CONFIG_XFS_QUOTA
2026 if (magic16 != XFS_DQUOT_MAGIC) {
2027 xfs_warn(mp, "Bad DQUOT block magic!");
2031 bp->b_ops = &xfs_dquot_buf_ops;
2034 "Trying to recover dquots without QUOTA support built in!");
2038 case XFS_BLFT_DINO_BUF:
2040 * we get here with inode allocation buffers, not buffers that
2041 * track unlinked list changes.
2043 if (magic16 != XFS_DINODE_MAGIC) {
2044 xfs_warn(mp, "Bad INODE block magic!");
2048 bp->b_ops = &xfs_inode_buf_ops;
2050 case XFS_BLFT_SYMLINK_BUF:
2051 if (magic32 != XFS_SYMLINK_MAGIC) {
2052 xfs_warn(mp, "Bad symlink block magic!");
2056 bp->b_ops = &xfs_symlink_buf_ops;
2058 case XFS_BLFT_DIR_BLOCK_BUF:
2059 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2060 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2061 xfs_warn(mp, "Bad dir block magic!");
2065 bp->b_ops = &xfs_dir3_block_buf_ops;
2067 case XFS_BLFT_DIR_DATA_BUF:
2068 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2069 magic32 != XFS_DIR3_DATA_MAGIC) {
2070 xfs_warn(mp, "Bad dir data magic!");
2074 bp->b_ops = &xfs_dir3_data_buf_ops;
2076 case XFS_BLFT_DIR_FREE_BUF:
2077 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2078 magic32 != XFS_DIR3_FREE_MAGIC) {
2079 xfs_warn(mp, "Bad dir3 free magic!");
2083 bp->b_ops = &xfs_dir3_free_buf_ops;
2085 case XFS_BLFT_DIR_LEAF1_BUF:
2086 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2087 magicda != XFS_DIR3_LEAF1_MAGIC) {
2088 xfs_warn(mp, "Bad dir leaf1 magic!");
2092 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2094 case XFS_BLFT_DIR_LEAFN_BUF:
2095 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2096 magicda != XFS_DIR3_LEAFN_MAGIC) {
2097 xfs_warn(mp, "Bad dir leafn magic!");
2101 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2103 case XFS_BLFT_DA_NODE_BUF:
2104 if (magicda != XFS_DA_NODE_MAGIC &&
2105 magicda != XFS_DA3_NODE_MAGIC) {
2106 xfs_warn(mp, "Bad da node magic!");
2110 bp->b_ops = &xfs_da3_node_buf_ops;
2112 case XFS_BLFT_ATTR_LEAF_BUF:
2113 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2114 magicda != XFS_ATTR3_LEAF_MAGIC) {
2115 xfs_warn(mp, "Bad attr leaf magic!");
2119 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2121 case XFS_BLFT_ATTR_RMT_BUF:
2122 if (!xfs_sb_version_hascrc(&mp->m_sb))
2124 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
2125 xfs_warn(mp, "Bad attr remote magic!");
2129 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2131 case XFS_BLFT_SB_BUF:
2132 if (magic32 != XFS_SB_MAGIC) {
2133 xfs_warn(mp, "Bad SB block magic!");
2137 bp->b_ops = &xfs_sb_buf_ops;
2140 xfs_warn(mp, "Unknown buffer type %d!",
2141 xfs_blft_from_flags(buf_f));
2147 * Perform a 'normal' buffer recovery. Each logged region of the
2148 * buffer should be copied over the corresponding region in the
2149 * given buffer. The bitmap in the buf log format structure indicates
2150 * where to place the logged data.
2153 xlog_recover_do_reg_buffer(
2154 struct xfs_mount *mp,
2155 xlog_recover_item_t *item,
2157 xfs_buf_log_format_t *buf_f)
2164 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2167 i = 1; /* 0 is the buf format structure */
2169 bit = xfs_next_bit(buf_f->blf_data_map,
2170 buf_f->blf_map_size, bit);
2173 nbits = xfs_contig_bits(buf_f->blf_data_map,
2174 buf_f->blf_map_size, bit);
2176 ASSERT(item->ri_buf[i].i_addr != NULL);
2177 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
2178 ASSERT(BBTOB(bp->b_io_length) >=
2179 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
2182 * The dirty regions logged in the buffer, even though
2183 * contiguous, may span multiple chunks. This is because the
2184 * dirty region may span a physical page boundary in a buffer
2185 * and hence be split into two separate vectors for writing into
2186 * the log. Hence we need to trim nbits back to the length of
2187 * the current region being copied out of the log.
2189 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2190 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2193 * Do a sanity check if this is a dquot buffer. Just checking
2194 * the first dquot in the buffer should do. XXXThis is
2195 * probably a good thing to do for other buf types also.
2198 if (buf_f->blf_flags &
2199 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2200 if (item->ri_buf[i].i_addr == NULL) {
2202 "XFS: NULL dquot in %s.", __func__);
2205 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
2207 "XFS: dquot too small (%d) in %s.",
2208 item->ri_buf[i].i_len, __func__);
2211 error = xfs_qm_dqcheck(mp, item->ri_buf[i].i_addr,
2212 -1, 0, XFS_QMOPT_DOWARN,
2213 "dquot_buf_recover");
2218 memcpy(xfs_buf_offset(bp,
2219 (uint)bit << XFS_BLF_SHIFT), /* dest */
2220 item->ri_buf[i].i_addr, /* source */
2221 nbits<<XFS_BLF_SHIFT); /* length */
2227 /* Shouldn't be any more regions */
2228 ASSERT(i == item->ri_total);
2231 * We can only do post recovery validation on items on CRC enabled
2232 * fielsystems as we need to know when the buffer was written to be able
2233 * to determine if we should have replayed the item. If we replay old
2234 * metadata over a newer buffer, then it will enter a temporarily
2235 * inconsistent state resulting in verification failures. Hence for now
2236 * just avoid the verification stage for non-crc filesystems
2238 if (xfs_sb_version_hascrc(&mp->m_sb))
2239 xlog_recovery_validate_buf_type(mp, bp, buf_f);
2243 * Do some primitive error checking on ondisk dquot data structures.
2247 struct xfs_mount *mp,
2248 xfs_disk_dquot_t *ddq,
2250 uint type, /* used only when IO_dorepair is true */
2254 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
2258 * We can encounter an uninitialized dquot buffer for 2 reasons:
2259 * 1. If we crash while deleting the quotainode(s), and those blks got
2260 * used for user data. This is because we take the path of regular
2261 * file deletion; however, the size field of quotainodes is never
2262 * updated, so all the tricks that we play in itruncate_finish
2263 * don't quite matter.
2265 * 2. We don't play the quota buffers when there's a quotaoff logitem.
2266 * But the allocation will be replayed so we'll end up with an
2267 * uninitialized quota block.
2269 * This is all fine; things are still consistent, and we haven't lost
2270 * any quota information. Just don't complain about bad dquot blks.
2272 if (ddq->d_magic != cpu_to_be16(XFS_DQUOT_MAGIC)) {
2273 if (flags & XFS_QMOPT_DOWARN)
2275 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
2276 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
2279 if (ddq->d_version != XFS_DQUOT_VERSION) {
2280 if (flags & XFS_QMOPT_DOWARN)
2282 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
2283 str, id, ddq->d_version, XFS_DQUOT_VERSION);
2287 if (ddq->d_flags != XFS_DQ_USER &&
2288 ddq->d_flags != XFS_DQ_PROJ &&
2289 ddq->d_flags != XFS_DQ_GROUP) {
2290 if (flags & XFS_QMOPT_DOWARN)
2292 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
2293 str, id, ddq->d_flags);
2297 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
2298 if (flags & XFS_QMOPT_DOWARN)
2300 "%s : ondisk-dquot 0x%p, ID mismatch: "
2301 "0x%x expected, found id 0x%x",
2302 str, ddq, id, be32_to_cpu(ddq->d_id));
2306 if (!errs && ddq->d_id) {
2307 if (ddq->d_blk_softlimit &&
2308 be64_to_cpu(ddq->d_bcount) >
2309 be64_to_cpu(ddq->d_blk_softlimit)) {
2310 if (!ddq->d_btimer) {
2311 if (flags & XFS_QMOPT_DOWARN)
2313 "%s : Dquot ID 0x%x (0x%p) BLK TIMER NOT STARTED",
2314 str, (int)be32_to_cpu(ddq->d_id), ddq);
2318 if (ddq->d_ino_softlimit &&
2319 be64_to_cpu(ddq->d_icount) >
2320 be64_to_cpu(ddq->d_ino_softlimit)) {
2321 if (!ddq->d_itimer) {
2322 if (flags & XFS_QMOPT_DOWARN)
2324 "%s : Dquot ID 0x%x (0x%p) INODE TIMER NOT STARTED",
2325 str, (int)be32_to_cpu(ddq->d_id), ddq);
2329 if (ddq->d_rtb_softlimit &&
2330 be64_to_cpu(ddq->d_rtbcount) >
2331 be64_to_cpu(ddq->d_rtb_softlimit)) {
2332 if (!ddq->d_rtbtimer) {
2333 if (flags & XFS_QMOPT_DOWARN)
2335 "%s : Dquot ID 0x%x (0x%p) RTBLK TIMER NOT STARTED",
2336 str, (int)be32_to_cpu(ddq->d_id), ddq);
2342 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2345 if (flags & XFS_QMOPT_DOWARN)
2346 xfs_notice(mp, "Re-initializing dquot ID 0x%x", id);
2349 * Typically, a repair is only requested by quotacheck.
2352 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2353 memset(d, 0, sizeof(xfs_dqblk_t));
2355 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2356 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2357 d->dd_diskdq.d_flags = type;
2358 d->dd_diskdq.d_id = cpu_to_be32(id);
2360 if (xfs_sb_version_hascrc(&mp->m_sb)) {
2361 uuid_copy(&d->dd_uuid, &mp->m_sb.sb_uuid);
2362 xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
2370 * Perform a dquot buffer recovery.
2371 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2372 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2373 * Else, treat it as a regular buffer and do recovery.
2376 xlog_recover_do_dquot_buffer(
2377 struct xfs_mount *mp,
2379 struct xlog_recover_item *item,
2381 struct xfs_buf_log_format *buf_f)
2385 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2388 * Filesystems are required to send in quota flags at mount time.
2390 if (mp->m_qflags == 0) {
2395 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2396 type |= XFS_DQ_USER;
2397 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2398 type |= XFS_DQ_PROJ;
2399 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2400 type |= XFS_DQ_GROUP;
2402 * This type of quotas was turned off, so ignore this buffer
2404 if (log->l_quotaoffs_flag & type)
2407 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2411 * This routine replays a modification made to a buffer at runtime.
2412 * There are actually two types of buffer, regular and inode, which
2413 * are handled differently. Inode buffers are handled differently
2414 * in that we only recover a specific set of data from them, namely
2415 * the inode di_next_unlinked fields. This is because all other inode
2416 * data is actually logged via inode records and any data we replay
2417 * here which overlaps that may be stale.
2419 * When meta-data buffers are freed at run time we log a buffer item
2420 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2421 * of the buffer in the log should not be replayed at recovery time.
2422 * This is so that if the blocks covered by the buffer are reused for
2423 * file data before we crash we don't end up replaying old, freed
2424 * meta-data into a user's file.
2426 * To handle the cancellation of buffer log items, we make two passes
2427 * over the log during recovery. During the first we build a table of
2428 * those buffers which have been cancelled, and during the second we
2429 * only replay those buffers which do not have corresponding cancel
2430 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2431 * for more details on the implementation of the table of cancel records.
2434 xlog_recover_buffer_pass2(
2436 struct list_head *buffer_list,
2437 struct xlog_recover_item *item)
2439 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2440 xfs_mount_t *mp = log->l_mp;
2446 * In this pass we only want to recover all the buffers which have
2447 * not been cancelled and are not cancellation buffers themselves.
2449 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2450 buf_f->blf_len, buf_f->blf_flags)) {
2451 trace_xfs_log_recover_buf_cancel(log, buf_f);
2455 trace_xfs_log_recover_buf_recover(log, buf_f);
2458 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2459 buf_flags |= XBF_UNMAPPED;
2461 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2464 return XFS_ERROR(ENOMEM);
2465 error = bp->b_error;
2467 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
2472 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
2473 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2474 } else if (buf_f->blf_flags &
2475 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2476 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2478 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2481 return XFS_ERROR(error);
2484 * Perform delayed write on the buffer. Asynchronous writes will be
2485 * slower when taking into account all the buffers to be flushed.
2487 * Also make sure that only inode buffers with good sizes stay in
2488 * the buffer cache. The kernel moves inodes in buffers of 1 block
2489 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2490 * buffers in the log can be a different size if the log was generated
2491 * by an older kernel using unclustered inode buffers or a newer kernel
2492 * running with a different inode cluster size. Regardless, if the
2493 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2494 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2495 * the buffer out of the buffer cache so that the buffer won't
2496 * overlap with future reads of those inodes.
2498 if (XFS_DINODE_MAGIC ==
2499 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2500 (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize,
2501 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2503 error = xfs_bwrite(bp);
2505 ASSERT(bp->b_target->bt_mount == mp);
2506 bp->b_iodone = xlog_recover_iodone;
2507 xfs_buf_delwri_queue(bp, buffer_list);
2515 xlog_recover_inode_pass2(
2517 struct list_head *buffer_list,
2518 struct xlog_recover_item *item)
2520 xfs_inode_log_format_t *in_f;
2521 xfs_mount_t *mp = log->l_mp;
2530 xfs_icdinode_t *dicp;
2534 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2535 in_f = item->ri_buf[0].i_addr;
2537 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
2539 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2545 * Inode buffers can be freed, look out for it,
2546 * and do not replay the inode.
2548 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2549 in_f->ilf_len, 0)) {
2551 trace_xfs_log_recover_inode_cancel(log, in_f);
2554 trace_xfs_log_recover_inode_recover(log, in_f);
2556 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
2557 &xfs_inode_buf_ops);
2562 error = bp->b_error;
2564 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
2568 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2569 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
2572 * Make sure the place we're flushing out to really looks
2575 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
2578 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2579 __func__, dip, bp, in_f->ilf_ino);
2580 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2581 XFS_ERRLEVEL_LOW, mp);
2582 error = EFSCORRUPTED;
2585 dicp = item->ri_buf[1].i_addr;
2586 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2589 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2590 __func__, item, in_f->ilf_ino);
2591 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2592 XFS_ERRLEVEL_LOW, mp);
2593 error = EFSCORRUPTED;
2598 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2599 * are transactional and if ordering is necessary we can determine that
2600 * more accurately by the LSN field in the V3 inode core. Don't trust
2601 * the inode versions we might be changing them here - use the
2602 * superblock flag to determine whether we need to look at di_flushiter
2603 * to skip replay when the on disk inode is newer than the log one
2605 if (!xfs_sb_version_hascrc(&mp->m_sb) &&
2606 dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
2608 * Deal with the wrap case, DI_MAX_FLUSH is less
2609 * than smaller numbers
2611 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
2612 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2616 trace_xfs_log_recover_inode_skip(log, in_f);
2622 /* Take the opportunity to reset the flush iteration count */
2623 dicp->di_flushiter = 0;
2625 if (unlikely(S_ISREG(dicp->di_mode))) {
2626 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2627 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2628 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2629 XFS_ERRLEVEL_LOW, mp, dicp);
2632 "%s: Bad regular inode log record, rec ptr 0x%p, "
2633 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2634 __func__, item, dip, bp, in_f->ilf_ino);
2635 error = EFSCORRUPTED;
2638 } else if (unlikely(S_ISDIR(dicp->di_mode))) {
2639 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2640 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2641 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2642 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2643 XFS_ERRLEVEL_LOW, mp, dicp);
2646 "%s: Bad dir inode log record, rec ptr 0x%p, "
2647 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2648 __func__, item, dip, bp, in_f->ilf_ino);
2649 error = EFSCORRUPTED;
2653 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2654 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2655 XFS_ERRLEVEL_LOW, mp, dicp);
2658 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2659 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2660 __func__, item, dip, bp, in_f->ilf_ino,
2661 dicp->di_nextents + dicp->di_anextents,
2663 error = EFSCORRUPTED;
2666 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2667 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2668 XFS_ERRLEVEL_LOW, mp, dicp);
2671 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2672 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
2673 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
2674 error = EFSCORRUPTED;
2677 isize = xfs_icdinode_size(dicp->di_version);
2678 if (unlikely(item->ri_buf[1].i_len > isize)) {
2679 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
2680 XFS_ERRLEVEL_LOW, mp, dicp);
2683 "%s: Bad inode log record length %d, rec ptr 0x%p",
2684 __func__, item->ri_buf[1].i_len, item);
2685 error = EFSCORRUPTED;
2689 /* The core is in in-core format */
2690 xfs_dinode_to_disk(dip, dicp);
2692 /* the rest is in on-disk format */
2693 if (item->ri_buf[1].i_len > isize) {
2694 memcpy((char *)dip + isize,
2695 item->ri_buf[1].i_addr + isize,
2696 item->ri_buf[1].i_len - isize);
2699 fields = in_f->ilf_fields;
2700 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2702 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
2705 memcpy(XFS_DFORK_DPTR(dip),
2706 &in_f->ilf_u.ilfu_uuid,
2711 if (in_f->ilf_size == 2)
2712 goto write_inode_buffer;
2713 len = item->ri_buf[2].i_len;
2714 src = item->ri_buf[2].i_addr;
2715 ASSERT(in_f->ilf_size <= 4);
2716 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2717 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2718 (len == in_f->ilf_dsize));
2720 switch (fields & XFS_ILOG_DFORK) {
2721 case XFS_ILOG_DDATA:
2723 memcpy(XFS_DFORK_DPTR(dip), src, len);
2726 case XFS_ILOG_DBROOT:
2727 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
2728 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
2729 XFS_DFORK_DSIZE(dip, mp));
2734 * There are no data fork flags set.
2736 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2741 * If we logged any attribute data, recover it. There may or
2742 * may not have been any other non-core data logged in this
2745 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2746 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2751 len = item->ri_buf[attr_index].i_len;
2752 src = item->ri_buf[attr_index].i_addr;
2753 ASSERT(len == in_f->ilf_asize);
2755 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2756 case XFS_ILOG_ADATA:
2758 dest = XFS_DFORK_APTR(dip);
2759 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2760 memcpy(dest, src, len);
2763 case XFS_ILOG_ABROOT:
2764 dest = XFS_DFORK_APTR(dip);
2765 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2766 len, (xfs_bmdr_block_t*)dest,
2767 XFS_DFORK_ASIZE(dip, mp));
2771 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
2780 /* re-generate the checksum. */
2781 xfs_dinode_calc_crc(log->l_mp, dip);
2783 ASSERT(bp->b_target->bt_mount == mp);
2784 bp->b_iodone = xlog_recover_iodone;
2785 xfs_buf_delwri_queue(bp, buffer_list);
2790 return XFS_ERROR(error);
2794 * Recover QUOTAOFF records. We simply make a note of it in the xlog
2795 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2799 xlog_recover_quotaoff_pass1(
2801 struct xlog_recover_item *item)
2803 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
2807 * The logitem format's flag tells us if this was user quotaoff,
2808 * group/project quotaoff or both.
2810 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2811 log->l_quotaoffs_flag |= XFS_DQ_USER;
2812 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2813 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2814 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2815 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2821 * Recover a dquot record
2824 xlog_recover_dquot_pass2(
2826 struct list_head *buffer_list,
2827 struct xlog_recover_item *item)
2829 xfs_mount_t *mp = log->l_mp;
2831 struct xfs_disk_dquot *ddq, *recddq;
2833 xfs_dq_logformat_t *dq_f;
2838 * Filesystems are required to send in quota flags at mount time.
2840 if (mp->m_qflags == 0)
2843 recddq = item->ri_buf[1].i_addr;
2844 if (recddq == NULL) {
2845 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
2846 return XFS_ERROR(EIO);
2848 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
2849 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
2850 item->ri_buf[1].i_len, __func__);
2851 return XFS_ERROR(EIO);
2855 * This type of quotas was turned off, so ignore this record.
2857 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
2859 if (log->l_quotaoffs_flag & type)
2863 * At this point we know that quota was _not_ turned off.
2864 * Since the mount flags are not indicating to us otherwise, this
2865 * must mean that quota is on, and the dquot needs to be replayed.
2866 * Remember that we may not have fully recovered the superblock yet,
2867 * so we can't do the usual trick of looking at the SB quota bits.
2869 * The other possibility, of course, is that the quota subsystem was
2870 * removed since the last mount - ENOSYS.
2872 dq_f = item->ri_buf[0].i_addr;
2874 error = xfs_qm_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2875 "xlog_recover_dquot_pass2 (log copy)");
2877 return XFS_ERROR(EIO);
2878 ASSERT(dq_f->qlf_len == 1);
2880 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
2881 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
2887 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2890 * At least the magic num portion should be on disk because this
2891 * was among a chunk of dquots created earlier, and we did some
2892 * minimal initialization then.
2894 error = xfs_qm_dqcheck(mp, ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2895 "xlog_recover_dquot_pass2");
2898 return XFS_ERROR(EIO);
2901 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2902 if (xfs_sb_version_hascrc(&mp->m_sb)) {
2903 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
2907 ASSERT(dq_f->qlf_size == 2);
2908 ASSERT(bp->b_target->bt_mount == mp);
2909 bp->b_iodone = xlog_recover_iodone;
2910 xfs_buf_delwri_queue(bp, buffer_list);
2917 * This routine is called to create an in-core extent free intent
2918 * item from the efi format structure which was logged on disk.
2919 * It allocates an in-core efi, copies the extents from the format
2920 * structure into it, and adds the efi to the AIL with the given
2924 xlog_recover_efi_pass2(
2926 struct xlog_recover_item *item,
2930 xfs_mount_t *mp = log->l_mp;
2931 xfs_efi_log_item_t *efip;
2932 xfs_efi_log_format_t *efi_formatp;
2934 efi_formatp = item->ri_buf[0].i_addr;
2936 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
2937 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2938 &(efip->efi_format)))) {
2939 xfs_efi_item_free(efip);
2942 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
2944 spin_lock(&log->l_ailp->xa_lock);
2946 * xfs_trans_ail_update() drops the AIL lock.
2948 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
2954 * This routine is called when an efd format structure is found in
2955 * a committed transaction in the log. It's purpose is to cancel
2956 * the corresponding efi if it was still in the log. To do this
2957 * it searches the AIL for the efi with an id equal to that in the
2958 * efd format structure. If we find it, we remove the efi from the
2962 xlog_recover_efd_pass2(
2964 struct xlog_recover_item *item)
2966 xfs_efd_log_format_t *efd_formatp;
2967 xfs_efi_log_item_t *efip = NULL;
2968 xfs_log_item_t *lip;
2970 struct xfs_ail_cursor cur;
2971 struct xfs_ail *ailp = log->l_ailp;
2973 efd_formatp = item->ri_buf[0].i_addr;
2974 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2975 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2976 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2977 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
2978 efi_id = efd_formatp->efd_efi_id;
2981 * Search for the efi with the id in the efd format structure
2984 spin_lock(&ailp->xa_lock);
2985 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
2986 while (lip != NULL) {
2987 if (lip->li_type == XFS_LI_EFI) {
2988 efip = (xfs_efi_log_item_t *)lip;
2989 if (efip->efi_format.efi_id == efi_id) {
2991 * xfs_trans_ail_delete() drops the
2994 xfs_trans_ail_delete(ailp, lip,
2995 SHUTDOWN_CORRUPT_INCORE);
2996 xfs_efi_item_free(efip);
2997 spin_lock(&ailp->xa_lock);
3001 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3003 xfs_trans_ail_cursor_done(ailp, &cur);
3004 spin_unlock(&ailp->xa_lock);
3010 * This routine is called when an inode create format structure is found in a
3011 * committed transaction in the log. It's purpose is to initialise the inodes
3012 * being allocated on disk. This requires us to get inode cluster buffers that
3013 * match the range to be intialised, stamped with inode templates and written
3014 * by delayed write so that subsequent modifications will hit the cached buffer
3015 * and only need writing out at the end of recovery.
3018 xlog_recover_do_icreate_pass2(
3020 struct list_head *buffer_list,
3021 xlog_recover_item_t *item)
3023 struct xfs_mount *mp = log->l_mp;
3024 struct xfs_icreate_log *icl;
3025 xfs_agnumber_t agno;
3026 xfs_agblock_t agbno;
3029 xfs_agblock_t length;
3031 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3032 if (icl->icl_type != XFS_LI_ICREATE) {
3033 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
3037 if (icl->icl_size != 1) {
3038 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
3042 agno = be32_to_cpu(icl->icl_ag);
3043 if (agno >= mp->m_sb.sb_agcount) {
3044 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
3047 agbno = be32_to_cpu(icl->icl_agbno);
3048 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3049 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
3052 isize = be32_to_cpu(icl->icl_isize);
3053 if (isize != mp->m_sb.sb_inodesize) {
3054 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
3057 count = be32_to_cpu(icl->icl_count);
3059 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
3062 length = be32_to_cpu(icl->icl_length);
3063 if (!length || length >= mp->m_sb.sb_agblocks) {
3064 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
3068 /* existing allocation is fixed value */
3069 ASSERT(count == XFS_IALLOC_INODES(mp));
3070 ASSERT(length == XFS_IALLOC_BLOCKS(mp));
3071 if (count != XFS_IALLOC_INODES(mp) ||
3072 length != XFS_IALLOC_BLOCKS(mp)) {
3073 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count 2");
3078 * Inode buffers can be freed. Do not replay the inode initialisation as
3079 * we could be overwriting something written after this inode buffer was
3082 * XXX: we need to iterate all buffers and only init those that are not
3083 * cancelled. I think that a more fine grained factoring of
3084 * xfs_ialloc_inode_init may be appropriate here to enable this to be
3087 if (xlog_check_buffer_cancelled(log,
3088 XFS_AGB_TO_DADDR(mp, agno, agbno), length, 0))
3091 xfs_ialloc_inode_init(mp, NULL, buffer_list, agno, agbno, length,
3092 be32_to_cpu(icl->icl_gen));
3097 * Free up any resources allocated by the transaction
3099 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3102 xlog_recover_free_trans(
3103 struct xlog_recover *trans)
3105 xlog_recover_item_t *item, *n;
3108 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
3109 /* Free the regions in the item. */
3110 list_del(&item->ri_list);
3111 for (i = 0; i < item->ri_cnt; i++)
3112 kmem_free(item->ri_buf[i].i_addr);
3113 /* Free the item itself */
3114 kmem_free(item->ri_buf);
3117 /* Free the transaction recover structure */
3122 xlog_recover_commit_pass1(
3124 struct xlog_recover *trans,
3125 struct xlog_recover_item *item)
3127 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
3129 switch (ITEM_TYPE(item)) {
3131 return xlog_recover_buffer_pass1(log, item);
3132 case XFS_LI_QUOTAOFF:
3133 return xlog_recover_quotaoff_pass1(log, item);
3138 case XFS_LI_ICREATE:
3139 /* nothing to do in pass 1 */
3142 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3143 __func__, ITEM_TYPE(item));
3145 return XFS_ERROR(EIO);
3150 xlog_recover_commit_pass2(
3152 struct xlog_recover *trans,
3153 struct list_head *buffer_list,
3154 struct xlog_recover_item *item)
3156 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
3158 switch (ITEM_TYPE(item)) {
3160 return xlog_recover_buffer_pass2(log, buffer_list, item);
3162 return xlog_recover_inode_pass2(log, buffer_list, item);
3164 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
3166 return xlog_recover_efd_pass2(log, item);
3168 return xlog_recover_dquot_pass2(log, buffer_list, item);
3169 case XFS_LI_ICREATE:
3170 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
3171 case XFS_LI_QUOTAOFF:
3172 /* nothing to do in pass2 */
3175 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3176 __func__, ITEM_TYPE(item));
3178 return XFS_ERROR(EIO);
3183 * Perform the transaction.
3185 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3186 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3189 xlog_recover_commit_trans(
3191 struct xlog_recover *trans,
3194 int error = 0, error2;
3195 xlog_recover_item_t *item;
3196 LIST_HEAD (buffer_list);
3198 hlist_del(&trans->r_list);
3200 error = xlog_recover_reorder_trans(log, trans, pass);
3204 list_for_each_entry(item, &trans->r_itemq, ri_list) {
3206 case XLOG_RECOVER_PASS1:
3207 error = xlog_recover_commit_pass1(log, trans, item);
3209 case XLOG_RECOVER_PASS2:
3210 error = xlog_recover_commit_pass2(log, trans,
3211 &buffer_list, item);
3221 xlog_recover_free_trans(trans);
3224 error2 = xfs_buf_delwri_submit(&buffer_list);
3225 return error ? error : error2;
3229 xlog_recover_unmount_trans(
3231 struct xlog_recover *trans)
3233 /* Do nothing now */
3234 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
3239 * There are two valid states of the r_state field. 0 indicates that the
3240 * transaction structure is in a normal state. We have either seen the
3241 * start of the transaction or the last operation we added was not a partial
3242 * operation. If the last operation we added to the transaction was a
3243 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
3245 * NOTE: skip LRs with 0 data length.
3248 xlog_recover_process_data(
3250 struct hlist_head rhash[],
3251 struct xlog_rec_header *rhead,
3257 xlog_op_header_t *ohead;
3258 xlog_recover_t *trans;
3264 lp = dp + be32_to_cpu(rhead->h_len);
3265 num_logops = be32_to_cpu(rhead->h_num_logops);
3267 /* check the log format matches our own - else we can't recover */
3268 if (xlog_header_check_recover(log->l_mp, rhead))
3269 return (XFS_ERROR(EIO));
3271 while ((dp < lp) && num_logops) {
3272 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
3273 ohead = (xlog_op_header_t *)dp;
3274 dp += sizeof(xlog_op_header_t);
3275 if (ohead->oh_clientid != XFS_TRANSACTION &&
3276 ohead->oh_clientid != XFS_LOG) {
3277 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
3278 __func__, ohead->oh_clientid);
3280 return (XFS_ERROR(EIO));
3282 tid = be32_to_cpu(ohead->oh_tid);
3283 hash = XLOG_RHASH(tid);
3284 trans = xlog_recover_find_tid(&rhash[hash], tid);
3285 if (trans == NULL) { /* not found; add new tid */
3286 if (ohead->oh_flags & XLOG_START_TRANS)
3287 xlog_recover_new_tid(&rhash[hash], tid,
3288 be64_to_cpu(rhead->h_lsn));
3290 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
3291 xfs_warn(log->l_mp, "%s: bad length 0x%x",
3292 __func__, be32_to_cpu(ohead->oh_len));
3294 return (XFS_ERROR(EIO));
3296 flags = ohead->oh_flags & ~XLOG_END_TRANS;
3297 if (flags & XLOG_WAS_CONT_TRANS)
3298 flags &= ~XLOG_CONTINUE_TRANS;
3300 case XLOG_COMMIT_TRANS:
3301 error = xlog_recover_commit_trans(log,
3304 case XLOG_UNMOUNT_TRANS:
3305 error = xlog_recover_unmount_trans(log, trans);
3307 case XLOG_WAS_CONT_TRANS:
3308 error = xlog_recover_add_to_cont_trans(log,
3310 be32_to_cpu(ohead->oh_len));
3312 case XLOG_START_TRANS:
3313 xfs_warn(log->l_mp, "%s: bad transaction",
3316 error = XFS_ERROR(EIO);
3319 case XLOG_CONTINUE_TRANS:
3320 error = xlog_recover_add_to_trans(log, trans,
3321 dp, be32_to_cpu(ohead->oh_len));
3324 xfs_warn(log->l_mp, "%s: bad flag 0x%x",
3327 error = XFS_ERROR(EIO);
3333 dp += be32_to_cpu(ohead->oh_len);
3340 * Process an extent free intent item that was recovered from
3341 * the log. We need to free the extents that it describes.
3344 xlog_recover_process_efi(
3346 xfs_efi_log_item_t *efip)
3348 xfs_efd_log_item_t *efdp;
3353 xfs_fsblock_t startblock_fsb;
3355 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
3358 * First check the validity of the extents described by the
3359 * EFI. If any are bad, then assume that all are bad and
3360 * just toss the EFI.
3362 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3363 extp = &(efip->efi_format.efi_extents[i]);
3364 startblock_fsb = XFS_BB_TO_FSB(mp,
3365 XFS_FSB_TO_DADDR(mp, extp->ext_start));
3366 if ((startblock_fsb == 0) ||
3367 (extp->ext_len == 0) ||
3368 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3369 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3371 * This will pull the EFI from the AIL and
3372 * free the memory associated with it.
3374 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
3375 xfs_efi_release(efip, efip->efi_format.efi_nextents);
3376 return XFS_ERROR(EIO);
3380 tp = xfs_trans_alloc(mp, 0);
3381 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
3384 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3386 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3387 extp = &(efip->efi_format.efi_extents[i]);
3388 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3391 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3395 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
3396 error = xfs_trans_commit(tp, 0);
3400 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3405 * When this is called, all of the EFIs which did not have
3406 * corresponding EFDs should be in the AIL. What we do now
3407 * is free the extents associated with each one.
3409 * Since we process the EFIs in normal transactions, they
3410 * will be removed at some point after the commit. This prevents
3411 * us from just walking down the list processing each one.
3412 * We'll use a flag in the EFI to skip those that we've already
3413 * processed and use the AIL iteration mechanism's generation
3414 * count to try to speed this up at least a bit.
3416 * When we start, we know that the EFIs are the only things in
3417 * the AIL. As we process them, however, other items are added
3418 * to the AIL. Since everything added to the AIL must come after
3419 * everything already in the AIL, we stop processing as soon as
3420 * we see something other than an EFI in the AIL.
3423 xlog_recover_process_efis(
3426 xfs_log_item_t *lip;
3427 xfs_efi_log_item_t *efip;
3429 struct xfs_ail_cursor cur;
3430 struct xfs_ail *ailp;
3433 spin_lock(&ailp->xa_lock);
3434 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3435 while (lip != NULL) {
3437 * We're done when we see something other than an EFI.
3438 * There should be no EFIs left in the AIL now.
3440 if (lip->li_type != XFS_LI_EFI) {
3442 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
3443 ASSERT(lip->li_type != XFS_LI_EFI);
3449 * Skip EFIs that we've already processed.
3451 efip = (xfs_efi_log_item_t *)lip;
3452 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
3453 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3457 spin_unlock(&ailp->xa_lock);
3458 error = xlog_recover_process_efi(log->l_mp, efip);
3459 spin_lock(&ailp->xa_lock);
3462 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3465 xfs_trans_ail_cursor_done(ailp, &cur);
3466 spin_unlock(&ailp->xa_lock);
3471 * This routine performs a transaction to null out a bad inode pointer
3472 * in an agi unlinked inode hash bucket.
3475 xlog_recover_clear_agi_bucket(
3477 xfs_agnumber_t agno,
3486 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3487 error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp),
3492 error = xfs_read_agi(mp, tp, agno, &agibp);
3496 agi = XFS_BUF_TO_AGI(agibp);
3497 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
3498 offset = offsetof(xfs_agi_t, agi_unlinked) +
3499 (sizeof(xfs_agino_t) * bucket);
3500 xfs_trans_log_buf(tp, agibp, offset,
3501 (offset + sizeof(xfs_agino_t) - 1));
3503 error = xfs_trans_commit(tp, 0);
3509 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3511 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
3516 xlog_recover_process_one_iunlink(
3517 struct xfs_mount *mp,
3518 xfs_agnumber_t agno,
3522 struct xfs_buf *ibp;
3523 struct xfs_dinode *dip;
3524 struct xfs_inode *ip;
3528 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3529 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
3534 * Get the on disk inode to find the next inode in the bucket.
3536 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
3540 ASSERT(ip->i_d.di_nlink == 0);
3541 ASSERT(ip->i_d.di_mode != 0);
3543 /* setup for the next pass */
3544 agino = be32_to_cpu(dip->di_next_unlinked);
3548 * Prevent any DMAPI event from being sent when the reference on
3549 * the inode is dropped.
3551 ip->i_d.di_dmevmask = 0;
3560 * We can't read in the inode this bucket points to, or this inode
3561 * is messed up. Just ditch this bucket of inodes. We will lose
3562 * some inodes and space, but at least we won't hang.
3564 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3565 * clear the inode pointer in the bucket.
3567 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3572 * xlog_iunlink_recover
3574 * This is called during recovery to process any inodes which
3575 * we unlinked but not freed when the system crashed. These
3576 * inodes will be on the lists in the AGI blocks. What we do
3577 * here is scan all the AGIs and fully truncate and free any
3578 * inodes found on the lists. Each inode is removed from the
3579 * lists when it has been fully truncated and is freed. The
3580 * freeing of the inode and its removal from the list must be
3584 xlog_recover_process_iunlinks(
3588 xfs_agnumber_t agno;
3599 * Prevent any DMAPI event from being sent while in this function.
3601 mp_dmevmask = mp->m_dmevmask;
3604 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3606 * Find the agi for this ag.
3608 error = xfs_read_agi(mp, NULL, agno, &agibp);
3611 * AGI is b0rked. Don't process it.
3613 * We should probably mark the filesystem as corrupt
3614 * after we've recovered all the ag's we can....
3619 * Unlock the buffer so that it can be acquired in the normal
3620 * course of the transaction to truncate and free each inode.
3621 * Because we are not racing with anyone else here for the AGI
3622 * buffer, we don't even need to hold it locked to read the
3623 * initial unlinked bucket entries out of the buffer. We keep
3624 * buffer reference though, so that it stays pinned in memory
3625 * while we need the buffer.
3627 agi = XFS_BUF_TO_AGI(agibp);
3628 xfs_buf_unlock(agibp);
3630 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3631 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3632 while (agino != NULLAGINO) {
3633 agino = xlog_recover_process_one_iunlink(mp,
3634 agno, agino, bucket);
3637 xfs_buf_rele(agibp);
3640 mp->m_dmevmask = mp_dmevmask;
3644 * Upack the log buffer data and crc check it. If the check fails, issue a
3645 * warning if and only if the CRC in the header is non-zero. This makes the
3646 * check an advisory warning, and the zero CRC check will prevent failure
3647 * warnings from being emitted when upgrading the kernel from one that does not
3648 * add CRCs by default.
3650 * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log
3651 * corruption failure
3654 xlog_unpack_data_crc(
3655 struct xlog_rec_header *rhead,
3661 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
3662 if (crc != rhead->h_crc) {
3663 if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
3664 xfs_alert(log->l_mp,
3665 "log record CRC mismatch: found 0x%x, expected 0x%x.\n",
3666 le32_to_cpu(rhead->h_crc),
3668 xfs_hex_dump(dp, 32);
3672 * If we've detected a log record corruption, then we can't
3673 * recover past this point. Abort recovery if we are enforcing
3674 * CRC protection by punting an error back up the stack.
3676 if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
3677 return EFSCORRUPTED;
3685 struct xlog_rec_header *rhead,
3692 error = xlog_unpack_data_crc(rhead, dp, log);
3696 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
3697 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3698 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
3702 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3703 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
3704 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
3705 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3706 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3707 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
3716 xlog_valid_rec_header(
3718 struct xlog_rec_header *rhead,
3723 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
3724 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3725 XFS_ERRLEVEL_LOW, log->l_mp);
3726 return XFS_ERROR(EFSCORRUPTED);
3729 (!rhead->h_version ||
3730 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
3731 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
3732 __func__, be32_to_cpu(rhead->h_version));
3733 return XFS_ERROR(EIO);
3736 /* LR body must have data or it wouldn't have been written */
3737 hlen = be32_to_cpu(rhead->h_len);
3738 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3739 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3740 XFS_ERRLEVEL_LOW, log->l_mp);
3741 return XFS_ERROR(EFSCORRUPTED);
3743 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3744 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3745 XFS_ERRLEVEL_LOW, log->l_mp);
3746 return XFS_ERROR(EFSCORRUPTED);
3752 * Read the log from tail to head and process the log records found.
3753 * Handle the two cases where the tail and head are in the same cycle
3754 * and where the active portion of the log wraps around the end of
3755 * the physical log separately. The pass parameter is passed through
3756 * to the routines called to process the data and is not looked at
3760 xlog_do_recovery_pass(
3762 xfs_daddr_t head_blk,
3763 xfs_daddr_t tail_blk,
3766 xlog_rec_header_t *rhead;
3769 xfs_buf_t *hbp, *dbp;
3770 int error = 0, h_size;
3771 int bblks, split_bblks;
3772 int hblks, split_hblks, wrapped_hblks;
3773 struct hlist_head rhash[XLOG_RHASH_SIZE];
3775 ASSERT(head_blk != tail_blk);
3778 * Read the header of the tail block and get the iclog buffer size from
3779 * h_size. Use this to tell how many sectors make up the log header.
3781 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3783 * When using variable length iclogs, read first sector of
3784 * iclog header and extract the header size from it. Get a
3785 * new hbp that is the correct size.
3787 hbp = xlog_get_bp(log, 1);
3791 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
3795 rhead = (xlog_rec_header_t *)offset;
3796 error = xlog_valid_rec_header(log, rhead, tail_blk);
3799 h_size = be32_to_cpu(rhead->h_size);
3800 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
3801 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3802 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3803 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3806 hbp = xlog_get_bp(log, hblks);
3811 ASSERT(log->l_sectBBsize == 1);
3813 hbp = xlog_get_bp(log, 1);
3814 h_size = XLOG_BIG_RECORD_BSIZE;
3819 dbp = xlog_get_bp(log, BTOBB(h_size));
3825 memset(rhash, 0, sizeof(rhash));
3826 if (tail_blk <= head_blk) {
3827 for (blk_no = tail_blk; blk_no < head_blk; ) {
3828 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3832 rhead = (xlog_rec_header_t *)offset;
3833 error = xlog_valid_rec_header(log, rhead, blk_no);
3837 /* blocks in data section */
3838 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3839 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
3844 error = xlog_unpack_data(rhead, offset, log);
3848 error = xlog_recover_process_data(log,
3849 rhash, rhead, offset, pass);
3852 blk_no += bblks + hblks;
3856 * Perform recovery around the end of the physical log.
3857 * When the head is not on the same cycle number as the tail,
3858 * we can't do a sequential recovery as above.
3861 while (blk_no < log->l_logBBsize) {
3863 * Check for header wrapping around physical end-of-log
3865 offset = hbp->b_addr;
3868 if (blk_no + hblks <= log->l_logBBsize) {
3869 /* Read header in one read */
3870 error = xlog_bread(log, blk_no, hblks, hbp,
3875 /* This LR is split across physical log end */
3876 if (blk_no != log->l_logBBsize) {
3877 /* some data before physical log end */
3878 ASSERT(blk_no <= INT_MAX);
3879 split_hblks = log->l_logBBsize - (int)blk_no;
3880 ASSERT(split_hblks > 0);
3881 error = xlog_bread(log, blk_no,
3889 * Note: this black magic still works with
3890 * large sector sizes (non-512) only because:
3891 * - we increased the buffer size originally
3892 * by 1 sector giving us enough extra space
3893 * for the second read;
3894 * - the log start is guaranteed to be sector
3896 * - we read the log end (LR header start)
3897 * _first_, then the log start (LR header end)
3898 * - order is important.
3900 wrapped_hblks = hblks - split_hblks;
3901 error = xlog_bread_offset(log, 0,
3903 offset + BBTOB(split_hblks));
3907 rhead = (xlog_rec_header_t *)offset;
3908 error = xlog_valid_rec_header(log, rhead,
3909 split_hblks ? blk_no : 0);
3913 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3916 /* Read in data for log record */
3917 if (blk_no + bblks <= log->l_logBBsize) {
3918 error = xlog_bread(log, blk_no, bblks, dbp,
3923 /* This log record is split across the
3924 * physical end of log */
3925 offset = dbp->b_addr;
3927 if (blk_no != log->l_logBBsize) {
3928 /* some data is before the physical
3930 ASSERT(!wrapped_hblks);
3931 ASSERT(blk_no <= INT_MAX);
3933 log->l_logBBsize - (int)blk_no;
3934 ASSERT(split_bblks > 0);
3935 error = xlog_bread(log, blk_no,
3943 * Note: this black magic still works with
3944 * large sector sizes (non-512) only because:
3945 * - we increased the buffer size originally
3946 * by 1 sector giving us enough extra space
3947 * for the second read;
3948 * - the log start is guaranteed to be sector
3950 * - we read the log end (LR header start)
3951 * _first_, then the log start (LR header end)
3952 * - order is important.
3954 error = xlog_bread_offset(log, 0,
3955 bblks - split_bblks, dbp,
3956 offset + BBTOB(split_bblks));
3961 error = xlog_unpack_data(rhead, offset, log);
3965 error = xlog_recover_process_data(log, rhash,
3966 rhead, offset, pass);
3972 ASSERT(blk_no >= log->l_logBBsize);
3973 blk_no -= log->l_logBBsize;
3975 /* read first part of physical log */
3976 while (blk_no < head_blk) {
3977 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3981 rhead = (xlog_rec_header_t *)offset;
3982 error = xlog_valid_rec_header(log, rhead, blk_no);
3986 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3987 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
3992 error = xlog_unpack_data(rhead, offset, log);
3996 error = xlog_recover_process_data(log, rhash,
3997 rhead, offset, pass);
4000 blk_no += bblks + hblks;
4012 * Do the recovery of the log. We actually do this in two phases.
4013 * The two passes are necessary in order to implement the function
4014 * of cancelling a record written into the log. The first pass
4015 * determines those things which have been cancelled, and the
4016 * second pass replays log items normally except for those which
4017 * have been cancelled. The handling of the replay and cancellations
4018 * takes place in the log item type specific routines.
4020 * The table of items which have cancel records in the log is allocated
4021 * and freed at this level, since only here do we know when all of
4022 * the log recovery has been completed.
4025 xlog_do_log_recovery(
4027 xfs_daddr_t head_blk,
4028 xfs_daddr_t tail_blk)
4032 ASSERT(head_blk != tail_blk);
4035 * First do a pass to find all of the cancelled buf log items.
4036 * Store them in the buf_cancel_table for use in the second pass.
4038 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
4039 sizeof(struct list_head),
4041 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4042 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
4044 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4045 XLOG_RECOVER_PASS1);
4047 kmem_free(log->l_buf_cancel_table);
4048 log->l_buf_cancel_table = NULL;
4052 * Then do a second pass to actually recover the items in the log.
4053 * When it is complete free the table of buf cancel items.
4055 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4056 XLOG_RECOVER_PASS2);
4061 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4062 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
4066 kmem_free(log->l_buf_cancel_table);
4067 log->l_buf_cancel_table = NULL;
4073 * Do the actual recovery
4078 xfs_daddr_t head_blk,
4079 xfs_daddr_t tail_blk)
4086 * First replay the images in the log.
4088 error = xlog_do_log_recovery(log, head_blk, tail_blk);
4093 * If IO errors happened during recovery, bail out.
4095 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
4100 * We now update the tail_lsn since much of the recovery has completed
4101 * and there may be space available to use. If there were no extent
4102 * or iunlinks, we can free up the entire log and set the tail_lsn to
4103 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4104 * lsn of the last known good LR on disk. If there are extent frees
4105 * or iunlinks they will have some entries in the AIL; so we look at
4106 * the AIL to determine how to set the tail_lsn.
4108 xlog_assign_tail_lsn(log->l_mp);
4111 * Now that we've finished replaying all buffer and inode
4112 * updates, re-read in the superblock and reverify it.
4114 bp = xfs_getsb(log->l_mp, 0);
4116 ASSERT(!(XFS_BUF_ISWRITE(bp)));
4118 XFS_BUF_UNASYNC(bp);
4119 bp->b_ops = &xfs_sb_buf_ops;
4120 xfsbdstrat(log->l_mp, bp);
4121 error = xfs_buf_iowait(bp);
4123 xfs_buf_ioerror_alert(bp, __func__);
4129 /* Convert superblock from on-disk format */
4130 sbp = &log->l_mp->m_sb;
4131 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
4132 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
4133 ASSERT(xfs_sb_good_version(sbp));
4136 /* We've re-read the superblock so re-initialize per-cpu counters */
4137 xfs_icsb_reinit_counters(log->l_mp);
4139 xlog_recover_check_summary(log);
4141 /* Normal transactions can now occur */
4142 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
4147 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4149 * Return error or zero.
4155 xfs_daddr_t head_blk, tail_blk;
4158 /* find the tail of the log */
4159 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
4162 if (tail_blk != head_blk) {
4163 /* There used to be a comment here:
4165 * disallow recovery on read-only mounts. note -- mount
4166 * checks for ENOSPC and turns it into an intelligent
4168 * ...but this is no longer true. Now, unless you specify
4169 * NORECOVERY (in which case this function would never be
4170 * called), we just go ahead and recover. We do this all
4171 * under the vfs layer, so we can get away with it unless
4172 * the device itself is read-only, in which case we fail.
4174 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
4179 * Version 5 superblock log feature mask validation. We know the
4180 * log is dirty so check if there are any unknown log features
4181 * in what we need to recover. If there are unknown features
4182 * (e.g. unsupported transactions, then simply reject the
4183 * attempt at recovery before touching anything.
4185 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
4186 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
4187 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
4189 "Superblock has unknown incompatible log features (0x%x) enabled.\n"
4190 "The log can not be fully and/or safely recovered by this kernel.\n"
4191 "Please recover the log on a kernel that supports the unknown features.",
4192 (log->l_mp->m_sb.sb_features_log_incompat &
4193 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
4197 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
4198 log->l_mp->m_logname ? log->l_mp->m_logname
4201 error = xlog_do_recover(log, head_blk, tail_blk);
4202 log->l_flags |= XLOG_RECOVERY_NEEDED;
4208 * In the first part of recovery we replay inodes and buffers and build
4209 * up the list of extent free items which need to be processed. Here
4210 * we process the extent free items and clean up the on disk unlinked
4211 * inode lists. This is separated from the first part of recovery so
4212 * that the root and real-time bitmap inodes can be read in from disk in
4213 * between the two stages. This is necessary so that we can free space
4214 * in the real-time portion of the file system.
4217 xlog_recover_finish(
4221 * Now we're ready to do the transactions needed for the
4222 * rest of recovery. Start with completing all the extent
4223 * free intent records and then process the unlinked inode
4224 * lists. At this point, we essentially run in normal mode
4225 * except that we're still performing recovery actions
4226 * rather than accepting new requests.
4228 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
4230 error = xlog_recover_process_efis(log);
4232 xfs_alert(log->l_mp, "Failed to recover EFIs");
4236 * Sync the log to get all the EFIs out of the AIL.
4237 * This isn't absolutely necessary, but it helps in
4238 * case the unlink transactions would have problems
4239 * pushing the EFIs out of the way.
4241 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
4243 xlog_recover_process_iunlinks(log);
4245 xlog_recover_check_summary(log);
4247 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
4248 log->l_mp->m_logname ? log->l_mp->m_logname
4250 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
4252 xfs_info(log->l_mp, "Ending clean mount");
4260 * Read all of the agf and agi counters and check that they
4261 * are consistent with the superblock counters.
4264 xlog_recover_check_summary(
4271 xfs_agnumber_t agno;
4272 __uint64_t freeblks;
4282 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
4283 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
4285 xfs_alert(mp, "%s agf read failed agno %d error %d",
4286 __func__, agno, error);
4288 agfp = XFS_BUF_TO_AGF(agfbp);
4289 freeblks += be32_to_cpu(agfp->agf_freeblks) +
4290 be32_to_cpu(agfp->agf_flcount);
4291 xfs_buf_relse(agfbp);
4294 error = xfs_read_agi(mp, NULL, agno, &agibp);
4296 xfs_alert(mp, "%s agi read failed agno %d error %d",
4297 __func__, agno, error);
4299 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
4301 itotal += be32_to_cpu(agi->agi_count);
4302 ifree += be32_to_cpu(agi->agi_freecount);
4303 xfs_buf_relse(agibp);