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_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_format.h"
30 #include "xfs_inode.h"
31 #include "xfs_trans.h"
33 #include "xfs_log_priv.h"
34 #include "xfs_log_recover.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_extfree_item.h"
37 #include "xfs_trans_priv.h"
38 #include "xfs_alloc.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_quota.h"
41 #include "xfs_cksum.h"
42 #include "xfs_trace.h"
43 #include "xfs_icache.h"
44 #include "xfs_bmap_btree.h"
45 #include "xfs_dinode.h"
46 #include "xfs_error.h"
49 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
56 xlog_clear_stale_blocks(
61 xlog_recover_check_summary(
64 #define xlog_recover_check_summary(log)
68 * This structure is used during recovery to record the buf log items which
69 * have been canceled and should not be replayed.
71 struct xfs_buf_cancel {
75 struct list_head bc_list;
79 * Sector aligned buffer routines for buffer create/read/write/access
83 * Verify the given count of basic blocks is valid number of blocks
84 * to specify for an operation involving the given XFS log buffer.
85 * Returns nonzero if the count is valid, 0 otherwise.
89 xlog_buf_bbcount_valid(
93 return bbcount > 0 && bbcount <= log->l_logBBsize;
97 * Allocate a buffer to hold log data. The buffer needs to be able
98 * to map to a range of nbblks basic blocks at any valid (basic
99 * block) offset within the log.
108 if (!xlog_buf_bbcount_valid(log, nbblks)) {
109 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
111 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
116 * We do log I/O in units of log sectors (a power-of-2
117 * multiple of the basic block size), so we round up the
118 * requested size to accommodate the basic blocks required
119 * for complete log sectors.
121 * In addition, the buffer may be used for a non-sector-
122 * aligned block offset, in which case an I/O of the
123 * requested size could extend beyond the end of the
124 * buffer. If the requested size is only 1 basic block it
125 * will never straddle a sector boundary, so this won't be
126 * an issue. Nor will this be a problem if the log I/O is
127 * done in basic blocks (sector size 1). But otherwise we
128 * extend the buffer by one extra log sector to ensure
129 * there's space to accommodate this possibility.
131 if (nbblks > 1 && log->l_sectBBsize > 1)
132 nbblks += log->l_sectBBsize;
133 nbblks = round_up(nbblks, log->l_sectBBsize);
135 bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0);
149 * Return the address of the start of the given block number's data
150 * in a log buffer. The buffer covers a log sector-aligned region.
159 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
161 ASSERT(offset + nbblks <= bp->b_length);
162 return bp->b_addr + BBTOB(offset);
167 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
178 if (!xlog_buf_bbcount_valid(log, nbblks)) {
179 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
181 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
185 blk_no = round_down(blk_no, log->l_sectBBsize);
186 nbblks = round_up(nbblks, log->l_sectBBsize);
189 ASSERT(nbblks <= bp->b_length);
191 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
193 bp->b_io_length = nbblks;
196 if (XFS_FORCED_SHUTDOWN(log->l_mp))
197 return XFS_ERROR(EIO);
199 xfs_buf_iorequest(bp);
200 error = xfs_buf_iowait(bp);
202 xfs_buf_ioerror_alert(bp, __func__);
216 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
220 *offset = xlog_align(log, blk_no, nbblks, bp);
225 * Read at an offset into the buffer. Returns with the buffer in it's original
226 * state regardless of the result of the read.
231 xfs_daddr_t blk_no, /* block to read from */
232 int nbblks, /* blocks to read */
236 xfs_caddr_t orig_offset = bp->b_addr;
237 int orig_len = BBTOB(bp->b_length);
240 error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks));
244 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
246 /* must reset buffer pointer even on error */
247 error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len);
254 * Write out the buffer at the given block for the given number of blocks.
255 * The buffer is kept locked across the write and is returned locked.
256 * This can only be used for synchronous log writes.
267 if (!xlog_buf_bbcount_valid(log, nbblks)) {
268 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
270 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
274 blk_no = round_down(blk_no, log->l_sectBBsize);
275 nbblks = round_up(nbblks, log->l_sectBBsize);
278 ASSERT(nbblks <= bp->b_length);
280 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
281 XFS_BUF_ZEROFLAGS(bp);
284 bp->b_io_length = nbblks;
287 error = xfs_bwrite(bp);
289 xfs_buf_ioerror_alert(bp, __func__);
296 * dump debug superblock and log record information
299 xlog_header_check_dump(
301 xlog_rec_header_t *head)
303 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d",
304 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
305 xfs_debug(mp, " log : uuid = %pU, fmt = %d",
306 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
309 #define xlog_header_check_dump(mp, head)
313 * check log record header for recovery
316 xlog_header_check_recover(
318 xlog_rec_header_t *head)
320 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
323 * IRIX doesn't write the h_fmt field and leaves it zeroed
324 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
325 * a dirty log created in IRIX.
327 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
329 "dirty log written in incompatible format - can't recover");
330 xlog_header_check_dump(mp, head);
331 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
332 XFS_ERRLEVEL_HIGH, mp);
333 return XFS_ERROR(EFSCORRUPTED);
334 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
336 "dirty log entry has mismatched uuid - can't recover");
337 xlog_header_check_dump(mp, head);
338 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
339 XFS_ERRLEVEL_HIGH, mp);
340 return XFS_ERROR(EFSCORRUPTED);
346 * read the head block of the log and check the header
349 xlog_header_check_mount(
351 xlog_rec_header_t *head)
353 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
355 if (uuid_is_nil(&head->h_fs_uuid)) {
357 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
358 * h_fs_uuid is nil, we assume this log was last mounted
359 * by IRIX and continue.
361 xfs_warn(mp, "nil uuid in log - IRIX style log");
362 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
363 xfs_warn(mp, "log has mismatched uuid - can't recover");
364 xlog_header_check_dump(mp, head);
365 XFS_ERROR_REPORT("xlog_header_check_mount",
366 XFS_ERRLEVEL_HIGH, mp);
367 return XFS_ERROR(EFSCORRUPTED);
378 * We're not going to bother about retrying
379 * this during recovery. One strike!
381 xfs_buf_ioerror_alert(bp, __func__);
382 xfs_force_shutdown(bp->b_target->bt_mount,
383 SHUTDOWN_META_IO_ERROR);
386 xfs_buf_ioend(bp, 0);
390 * This routine finds (to an approximation) the first block in the physical
391 * log which contains the given cycle. It uses a binary search algorithm.
392 * Note that the algorithm can not be perfect because the disk will not
393 * necessarily be perfect.
396 xlog_find_cycle_start(
399 xfs_daddr_t first_blk,
400 xfs_daddr_t *last_blk,
410 mid_blk = BLK_AVG(first_blk, end_blk);
411 while (mid_blk != first_blk && mid_blk != end_blk) {
412 error = xlog_bread(log, mid_blk, 1, bp, &offset);
415 mid_cycle = xlog_get_cycle(offset);
416 if (mid_cycle == cycle)
417 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
419 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
420 mid_blk = BLK_AVG(first_blk, end_blk);
422 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
423 (mid_blk == end_blk && mid_blk-1 == first_blk));
431 * Check that a range of blocks does not contain stop_on_cycle_no.
432 * Fill in *new_blk with the block offset where such a block is
433 * found, or with -1 (an invalid block number) if there is no such
434 * block in the range. The scan needs to occur from front to back
435 * and the pointer into the region must be updated since a later
436 * routine will need to perform another test.
439 xlog_find_verify_cycle(
441 xfs_daddr_t start_blk,
443 uint stop_on_cycle_no,
444 xfs_daddr_t *new_blk)
450 xfs_caddr_t buf = NULL;
454 * Greedily allocate a buffer big enough to handle the full
455 * range of basic blocks we'll be examining. If that fails,
456 * try a smaller size. We need to be able to read at least
457 * a log sector, or we're out of luck.
459 bufblks = 1 << ffs(nbblks);
460 while (bufblks > log->l_logBBsize)
462 while (!(bp = xlog_get_bp(log, bufblks))) {
464 if (bufblks < log->l_sectBBsize)
468 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
471 bcount = min(bufblks, (start_blk + nbblks - i));
473 error = xlog_bread(log, i, bcount, bp, &buf);
477 for (j = 0; j < bcount; j++) {
478 cycle = xlog_get_cycle(buf);
479 if (cycle == stop_on_cycle_no) {
496 * Potentially backup over partial log record write.
498 * In the typical case, last_blk is the number of the block directly after
499 * a good log record. Therefore, we subtract one to get the block number
500 * of the last block in the given buffer. extra_bblks contains the number
501 * of blocks we would have read on a previous read. This happens when the
502 * last log record is split over the end of the physical log.
504 * extra_bblks is the number of blocks potentially verified on a previous
505 * call to this routine.
508 xlog_find_verify_log_record(
510 xfs_daddr_t start_blk,
511 xfs_daddr_t *last_blk,
516 xfs_caddr_t offset = NULL;
517 xlog_rec_header_t *head = NULL;
520 int num_blks = *last_blk - start_blk;
523 ASSERT(start_blk != 0 || *last_blk != start_blk);
525 if (!(bp = xlog_get_bp(log, num_blks))) {
526 if (!(bp = xlog_get_bp(log, 1)))
530 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
533 offset += ((num_blks - 1) << BBSHIFT);
536 for (i = (*last_blk) - 1; i >= 0; i--) {
538 /* valid log record not found */
540 "Log inconsistent (didn't find previous header)");
542 error = XFS_ERROR(EIO);
547 error = xlog_bread(log, i, 1, bp, &offset);
552 head = (xlog_rec_header_t *)offset;
554 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
562 * We hit the beginning of the physical log & still no header. Return
563 * to caller. If caller can handle a return of -1, then this routine
564 * will be called again for the end of the physical log.
572 * We have the final block of the good log (the first block
573 * of the log record _before_ the head. So we check the uuid.
575 if ((error = xlog_header_check_mount(log->l_mp, head)))
579 * We may have found a log record header before we expected one.
580 * last_blk will be the 1st block # with a given cycle #. We may end
581 * up reading an entire log record. In this case, we don't want to
582 * reset last_blk. Only when last_blk points in the middle of a log
583 * record do we update last_blk.
585 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
586 uint h_size = be32_to_cpu(head->h_size);
588 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
589 if (h_size % XLOG_HEADER_CYCLE_SIZE)
595 if (*last_blk - i + extra_bblks !=
596 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
605 * Head is defined to be the point of the log where the next log write
606 * could go. This means that incomplete LR writes at the end are
607 * eliminated when calculating the head. We aren't guaranteed that previous
608 * LR have complete transactions. We only know that a cycle number of
609 * current cycle number -1 won't be present in the log if we start writing
610 * from our current block number.
612 * last_blk contains the block number of the first block with a given
615 * Return: zero if normal, non-zero if error.
620 xfs_daddr_t *return_head_blk)
624 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
626 uint first_half_cycle, last_half_cycle;
628 int error, log_bbnum = log->l_logBBsize;
630 /* Is the end of the log device zeroed? */
631 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
632 *return_head_blk = first_blk;
634 /* Is the whole lot zeroed? */
636 /* Linux XFS shouldn't generate totally zeroed logs -
637 * mkfs etc write a dummy unmount record to a fresh
638 * log so we can store the uuid in there
640 xfs_warn(log->l_mp, "totally zeroed log");
645 xfs_warn(log->l_mp, "empty log check failed");
649 first_blk = 0; /* get cycle # of 1st block */
650 bp = xlog_get_bp(log, 1);
654 error = xlog_bread(log, 0, 1, bp, &offset);
658 first_half_cycle = xlog_get_cycle(offset);
660 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
661 error = xlog_bread(log, last_blk, 1, bp, &offset);
665 last_half_cycle = xlog_get_cycle(offset);
666 ASSERT(last_half_cycle != 0);
669 * If the 1st half cycle number is equal to the last half cycle number,
670 * then the entire log is stamped with the same cycle number. In this
671 * case, head_blk can't be set to zero (which makes sense). The below
672 * math doesn't work out properly with head_blk equal to zero. Instead,
673 * we set it to log_bbnum which is an invalid block number, but this
674 * value makes the math correct. If head_blk doesn't changed through
675 * all the tests below, *head_blk is set to zero at the very end rather
676 * than log_bbnum. In a sense, log_bbnum and zero are the same block
677 * in a circular file.
679 if (first_half_cycle == last_half_cycle) {
681 * In this case we believe that the entire log should have
682 * cycle number last_half_cycle. We need to scan backwards
683 * from the end verifying that there are no holes still
684 * containing last_half_cycle - 1. If we find such a hole,
685 * then the start of that hole will be the new head. The
686 * simple case looks like
687 * x | x ... | x - 1 | x
688 * Another case that fits this picture would be
689 * x | x + 1 | x ... | x
690 * In this case the head really is somewhere at the end of the
691 * log, as one of the latest writes at the beginning was
694 * x | x + 1 | x ... | x - 1 | x
695 * This is really the combination of the above two cases, and
696 * the head has to end up at the start of the x-1 hole at the
699 * In the 256k log case, we will read from the beginning to the
700 * end of the log and search for cycle numbers equal to x-1.
701 * We don't worry about the x+1 blocks that we encounter,
702 * because we know that they cannot be the head since the log
705 head_blk = log_bbnum;
706 stop_on_cycle = last_half_cycle - 1;
709 * In this case we want to find the first block with cycle
710 * number matching last_half_cycle. We expect the log to be
712 * x + 1 ... | x ... | x
713 * The first block with cycle number x (last_half_cycle) will
714 * be where the new head belongs. First we do a binary search
715 * for the first occurrence of last_half_cycle. The binary
716 * search may not be totally accurate, so then we scan back
717 * from there looking for occurrences of last_half_cycle before
718 * us. If that backwards scan wraps around the beginning of
719 * the log, then we look for occurrences of last_half_cycle - 1
720 * at the end of the log. The cases we're looking for look
722 * v binary search stopped here
723 * x + 1 ... | x | x + 1 | x ... | x
724 * ^ but we want to locate this spot
726 * <---------> less than scan distance
727 * x + 1 ... | x ... | x - 1 | x
728 * ^ we want to locate this spot
730 stop_on_cycle = last_half_cycle;
731 if ((error = xlog_find_cycle_start(log, bp, first_blk,
732 &head_blk, last_half_cycle)))
737 * Now validate the answer. Scan back some number of maximum possible
738 * blocks and make sure each one has the expected cycle number. The
739 * maximum is determined by the total possible amount of buffering
740 * in the in-core log. The following number can be made tighter if
741 * we actually look at the block size of the filesystem.
743 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
744 if (head_blk >= num_scan_bblks) {
746 * We are guaranteed that the entire check can be performed
749 start_blk = head_blk - num_scan_bblks;
750 if ((error = xlog_find_verify_cycle(log,
751 start_blk, num_scan_bblks,
752 stop_on_cycle, &new_blk)))
756 } else { /* need to read 2 parts of log */
758 * We are going to scan backwards in the log in two parts.
759 * First we scan the physical end of the log. In this part
760 * of the log, we are looking for blocks with cycle number
761 * last_half_cycle - 1.
762 * If we find one, then we know that the log starts there, as
763 * we've found a hole that didn't get written in going around
764 * the end of the physical log. The simple case for this is
765 * x + 1 ... | x ... | x - 1 | x
766 * <---------> less than scan distance
767 * If all of the blocks at the end of the log have cycle number
768 * last_half_cycle, then we check the blocks at the start of
769 * the log looking for occurrences of last_half_cycle. If we
770 * find one, then our current estimate for the location of the
771 * first occurrence of last_half_cycle is wrong and we move
772 * back to the hole we've found. This case looks like
773 * x + 1 ... | x | x + 1 | x ...
774 * ^ binary search stopped here
775 * Another case we need to handle that only occurs in 256k
777 * x + 1 ... | x ... | x+1 | x ...
778 * ^ binary search stops here
779 * In a 256k log, the scan at the end of the log will see the
780 * x + 1 blocks. We need to skip past those since that is
781 * certainly not the head of the log. By searching for
782 * last_half_cycle-1 we accomplish that.
784 ASSERT(head_blk <= INT_MAX &&
785 (xfs_daddr_t) num_scan_bblks >= head_blk);
786 start_blk = log_bbnum - (num_scan_bblks - head_blk);
787 if ((error = xlog_find_verify_cycle(log, start_blk,
788 num_scan_bblks - (int)head_blk,
789 (stop_on_cycle - 1), &new_blk)))
797 * Scan beginning of log now. The last part of the physical
798 * log is good. This scan needs to verify that it doesn't find
799 * the last_half_cycle.
802 ASSERT(head_blk <= INT_MAX);
803 if ((error = xlog_find_verify_cycle(log,
804 start_blk, (int)head_blk,
805 stop_on_cycle, &new_blk)))
813 * Now we need to make sure head_blk is not pointing to a block in
814 * the middle of a log record.
816 num_scan_bblks = XLOG_REC_SHIFT(log);
817 if (head_blk >= num_scan_bblks) {
818 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
820 /* start ptr at last block ptr before head_blk */
821 if ((error = xlog_find_verify_log_record(log, start_blk,
822 &head_blk, 0)) == -1) {
823 error = XFS_ERROR(EIO);
829 ASSERT(head_blk <= INT_MAX);
830 if ((error = xlog_find_verify_log_record(log, start_blk,
831 &head_blk, 0)) == -1) {
832 /* We hit the beginning of the log during our search */
833 start_blk = log_bbnum - (num_scan_bblks - head_blk);
835 ASSERT(start_blk <= INT_MAX &&
836 (xfs_daddr_t) log_bbnum-start_blk >= 0);
837 ASSERT(head_blk <= INT_MAX);
838 if ((error = xlog_find_verify_log_record(log,
840 (int)head_blk)) == -1) {
841 error = XFS_ERROR(EIO);
845 if (new_blk != log_bbnum)
852 if (head_blk == log_bbnum)
853 *return_head_blk = 0;
855 *return_head_blk = head_blk;
857 * When returning here, we have a good block number. Bad block
858 * means that during a previous crash, we didn't have a clean break
859 * from cycle number N to cycle number N-1. In this case, we need
860 * to find the first block with cycle number N-1.
868 xfs_warn(log->l_mp, "failed to find log head");
873 * Find the sync block number or the tail of the log.
875 * This will be the block number of the last record to have its
876 * associated buffers synced to disk. Every log record header has
877 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
878 * to get a sync block number. The only concern is to figure out which
879 * log record header to believe.
881 * The following algorithm uses the log record header with the largest
882 * lsn. The entire log record does not need to be valid. We only care
883 * that the header is valid.
885 * We could speed up search by using current head_blk buffer, but it is not
891 xfs_daddr_t *head_blk,
892 xfs_daddr_t *tail_blk)
894 xlog_rec_header_t *rhead;
895 xlog_op_header_t *op_head;
896 xfs_caddr_t offset = NULL;
899 xfs_daddr_t umount_data_blk;
900 xfs_daddr_t after_umount_blk;
907 * Find previous log record
909 if ((error = xlog_find_head(log, head_blk)))
912 bp = xlog_get_bp(log, 1);
915 if (*head_blk == 0) { /* special case */
916 error = xlog_bread(log, 0, 1, bp, &offset);
920 if (xlog_get_cycle(offset) == 0) {
922 /* leave all other log inited values alone */
928 * Search backwards looking for log record header block
930 ASSERT(*head_blk < INT_MAX);
931 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
932 error = xlog_bread(log, i, 1, bp, &offset);
936 if (*(__be32 *)offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
942 * If we haven't found the log record header block, start looking
943 * again from the end of the physical log. XXXmiken: There should be
944 * a check here to make sure we didn't search more than N blocks in
948 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
949 error = xlog_bread(log, i, 1, bp, &offset);
953 if (*(__be32 *)offset ==
954 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
961 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
964 return XFS_ERROR(EIO);
967 /* find blk_no of tail of log */
968 rhead = (xlog_rec_header_t *)offset;
969 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
972 * Reset log values according to the state of the log when we
973 * crashed. In the case where head_blk == 0, we bump curr_cycle
974 * one because the next write starts a new cycle rather than
975 * continuing the cycle of the last good log record. At this
976 * point we have guaranteed that all partial log records have been
977 * accounted for. Therefore, we know that the last good log record
978 * written was complete and ended exactly on the end boundary
979 * of the physical log.
981 log->l_prev_block = i;
982 log->l_curr_block = (int)*head_blk;
983 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
986 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
987 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
988 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
989 BBTOB(log->l_curr_block));
990 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
991 BBTOB(log->l_curr_block));
994 * Look for unmount record. If we find it, then we know there
995 * was a clean unmount. Since 'i' could be the last block in
996 * the physical log, we convert to a log block before comparing
999 * Save the current tail lsn to use to pass to
1000 * xlog_clear_stale_blocks() below. We won't want to clear the
1001 * unmount record if there is one, so we pass the lsn of the
1002 * unmount record rather than the block after it.
1004 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1005 int h_size = be32_to_cpu(rhead->h_size);
1006 int h_version = be32_to_cpu(rhead->h_version);
1008 if ((h_version & XLOG_VERSION_2) &&
1009 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
1010 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
1011 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1019 after_umount_blk = (i + hblks + (int)
1020 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
1021 tail_lsn = atomic64_read(&log->l_tail_lsn);
1022 if (*head_blk == after_umount_blk &&
1023 be32_to_cpu(rhead->h_num_logops) == 1) {
1024 umount_data_blk = (i + hblks) % log->l_logBBsize;
1025 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1029 op_head = (xlog_op_header_t *)offset;
1030 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1032 * Set tail and last sync so that newly written
1033 * log records will point recovery to after the
1034 * current unmount record.
1036 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1037 log->l_curr_cycle, after_umount_blk);
1038 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1039 log->l_curr_cycle, after_umount_blk);
1040 *tail_blk = after_umount_blk;
1043 * Note that the unmount was clean. If the unmount
1044 * was not clean, we need to know this to rebuild the
1045 * superblock counters from the perag headers if we
1046 * have a filesystem using non-persistent counters.
1048 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1053 * Make sure that there are no blocks in front of the head
1054 * with the same cycle number as the head. This can happen
1055 * because we allow multiple outstanding log writes concurrently,
1056 * and the later writes might make it out before earlier ones.
1058 * We use the lsn from before modifying it so that we'll never
1059 * overwrite the unmount record after a clean unmount.
1061 * Do this only if we are going to recover the filesystem
1063 * NOTE: This used to say "if (!readonly)"
1064 * However on Linux, we can & do recover a read-only filesystem.
1065 * We only skip recovery if NORECOVERY is specified on mount,
1066 * in which case we would not be here.
1068 * But... if the -device- itself is readonly, just skip this.
1069 * We can't recover this device anyway, so it won't matter.
1071 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
1072 error = xlog_clear_stale_blocks(log, tail_lsn);
1078 xfs_warn(log->l_mp, "failed to locate log tail");
1083 * Is the log zeroed at all?
1085 * The last binary search should be changed to perform an X block read
1086 * once X becomes small enough. You can then search linearly through
1087 * the X blocks. This will cut down on the number of reads we need to do.
1089 * If the log is partially zeroed, this routine will pass back the blkno
1090 * of the first block with cycle number 0. It won't have a complete LR
1094 * 0 => the log is completely written to
1095 * -1 => use *blk_no as the first block of the log
1096 * >0 => error has occurred
1101 xfs_daddr_t *blk_no)
1105 uint first_cycle, last_cycle;
1106 xfs_daddr_t new_blk, last_blk, start_blk;
1107 xfs_daddr_t num_scan_bblks;
1108 int error, log_bbnum = log->l_logBBsize;
1112 /* check totally zeroed log */
1113 bp = xlog_get_bp(log, 1);
1116 error = xlog_bread(log, 0, 1, bp, &offset);
1120 first_cycle = xlog_get_cycle(offset);
1121 if (first_cycle == 0) { /* completely zeroed log */
1127 /* check partially zeroed log */
1128 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1132 last_cycle = xlog_get_cycle(offset);
1133 if (last_cycle != 0) { /* log completely written to */
1136 } else if (first_cycle != 1) {
1138 * If the cycle of the last block is zero, the cycle of
1139 * the first block must be 1. If it's not, maybe we're
1140 * not looking at a log... Bail out.
1143 "Log inconsistent or not a log (last==0, first!=1)");
1144 error = XFS_ERROR(EINVAL);
1148 /* we have a partially zeroed log */
1149 last_blk = log_bbnum-1;
1150 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1154 * Validate the answer. Because there is no way to guarantee that
1155 * the entire log is made up of log records which are the same size,
1156 * we scan over the defined maximum blocks. At this point, the maximum
1157 * is not chosen to mean anything special. XXXmiken
1159 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1160 ASSERT(num_scan_bblks <= INT_MAX);
1162 if (last_blk < num_scan_bblks)
1163 num_scan_bblks = last_blk;
1164 start_blk = last_blk - num_scan_bblks;
1167 * We search for any instances of cycle number 0 that occur before
1168 * our current estimate of the head. What we're trying to detect is
1169 * 1 ... | 0 | 1 | 0...
1170 * ^ binary search ends here
1172 if ((error = xlog_find_verify_cycle(log, start_blk,
1173 (int)num_scan_bblks, 0, &new_blk)))
1179 * Potentially backup over partial log record write. We don't need
1180 * to search the end of the log because we know it is zero.
1182 if ((error = xlog_find_verify_log_record(log, start_blk,
1183 &last_blk, 0)) == -1) {
1184 error = XFS_ERROR(EIO);
1198 * These are simple subroutines used by xlog_clear_stale_blocks() below
1199 * to initialize a buffer full of empty log record headers and write
1200 * them into the log.
1211 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1213 memset(buf, 0, BBSIZE);
1214 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1215 recp->h_cycle = cpu_to_be32(cycle);
1216 recp->h_version = cpu_to_be32(
1217 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1218 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1219 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1220 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1221 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1225 xlog_write_log_records(
1236 int sectbb = log->l_sectBBsize;
1237 int end_block = start_block + blocks;
1243 * Greedily allocate a buffer big enough to handle the full
1244 * range of basic blocks to be written. If that fails, try
1245 * a smaller size. We need to be able to write at least a
1246 * log sector, or we're out of luck.
1248 bufblks = 1 << ffs(blocks);
1249 while (bufblks > log->l_logBBsize)
1251 while (!(bp = xlog_get_bp(log, bufblks))) {
1253 if (bufblks < sectbb)
1257 /* We may need to do a read at the start to fill in part of
1258 * the buffer in the starting sector not covered by the first
1261 balign = round_down(start_block, sectbb);
1262 if (balign != start_block) {
1263 error = xlog_bread_noalign(log, start_block, 1, bp);
1267 j = start_block - balign;
1270 for (i = start_block; i < end_block; i += bufblks) {
1271 int bcount, endcount;
1273 bcount = min(bufblks, end_block - start_block);
1274 endcount = bcount - j;
1276 /* We may need to do a read at the end to fill in part of
1277 * the buffer in the final sector not covered by the write.
1278 * If this is the same sector as the above read, skip it.
1280 ealign = round_down(end_block, sectbb);
1281 if (j == 0 && (start_block + endcount > ealign)) {
1282 offset = bp->b_addr + BBTOB(ealign - start_block);
1283 error = xlog_bread_offset(log, ealign, sectbb,
1290 offset = xlog_align(log, start_block, endcount, bp);
1291 for (; j < endcount; j++) {
1292 xlog_add_record(log, offset, cycle, i+j,
1293 tail_cycle, tail_block);
1296 error = xlog_bwrite(log, start_block, endcount, bp);
1299 start_block += endcount;
1309 * This routine is called to blow away any incomplete log writes out
1310 * in front of the log head. We do this so that we won't become confused
1311 * if we come up, write only a little bit more, and then crash again.
1312 * If we leave the partial log records out there, this situation could
1313 * cause us to think those partial writes are valid blocks since they
1314 * have the current cycle number. We get rid of them by overwriting them
1315 * with empty log records with the old cycle number rather than the
1318 * The tail lsn is passed in rather than taken from
1319 * the log so that we will not write over the unmount record after a
1320 * clean unmount in a 512 block log. Doing so would leave the log without
1321 * any valid log records in it until a new one was written. If we crashed
1322 * during that time we would not be able to recover.
1325 xlog_clear_stale_blocks(
1329 int tail_cycle, head_cycle;
1330 int tail_block, head_block;
1331 int tail_distance, max_distance;
1335 tail_cycle = CYCLE_LSN(tail_lsn);
1336 tail_block = BLOCK_LSN(tail_lsn);
1337 head_cycle = log->l_curr_cycle;
1338 head_block = log->l_curr_block;
1341 * Figure out the distance between the new head of the log
1342 * and the tail. We want to write over any blocks beyond the
1343 * head that we may have written just before the crash, but
1344 * we don't want to overwrite the tail of the log.
1346 if (head_cycle == tail_cycle) {
1348 * The tail is behind the head in the physical log,
1349 * so the distance from the head to the tail is the
1350 * distance from the head to the end of the log plus
1351 * the distance from the beginning of the log to the
1354 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1355 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1356 XFS_ERRLEVEL_LOW, log->l_mp);
1357 return XFS_ERROR(EFSCORRUPTED);
1359 tail_distance = tail_block + (log->l_logBBsize - head_block);
1362 * The head is behind the tail in the physical log,
1363 * so the distance from the head to the tail is just
1364 * the tail block minus the head block.
1366 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1367 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1368 XFS_ERRLEVEL_LOW, log->l_mp);
1369 return XFS_ERROR(EFSCORRUPTED);
1371 tail_distance = tail_block - head_block;
1375 * If the head is right up against the tail, we can't clear
1378 if (tail_distance <= 0) {
1379 ASSERT(tail_distance == 0);
1383 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1385 * Take the smaller of the maximum amount of outstanding I/O
1386 * we could have and the distance to the tail to clear out.
1387 * We take the smaller so that we don't overwrite the tail and
1388 * we don't waste all day writing from the head to the tail
1391 max_distance = MIN(max_distance, tail_distance);
1393 if ((head_block + max_distance) <= log->l_logBBsize) {
1395 * We can stomp all the blocks we need to without
1396 * wrapping around the end of the log. Just do it
1397 * in a single write. Use the cycle number of the
1398 * current cycle minus one so that the log will look like:
1401 error = xlog_write_log_records(log, (head_cycle - 1),
1402 head_block, max_distance, tail_cycle,
1408 * We need to wrap around the end of the physical log in
1409 * order to clear all the blocks. Do it in two separate
1410 * I/Os. The first write should be from the head to the
1411 * end of the physical log, and it should use the current
1412 * cycle number minus one just like above.
1414 distance = log->l_logBBsize - head_block;
1415 error = xlog_write_log_records(log, (head_cycle - 1),
1416 head_block, distance, tail_cycle,
1423 * Now write the blocks at the start of the physical log.
1424 * This writes the remainder of the blocks we want to clear.
1425 * It uses the current cycle number since we're now on the
1426 * same cycle as the head so that we get:
1427 * n ... n ... | n - 1 ...
1428 * ^^^^^ blocks we're writing
1430 distance = max_distance - (log->l_logBBsize - head_block);
1431 error = xlog_write_log_records(log, head_cycle, 0, distance,
1432 tail_cycle, tail_block);
1440 /******************************************************************************
1442 * Log recover routines
1444 ******************************************************************************
1447 STATIC xlog_recover_t *
1448 xlog_recover_find_tid(
1449 struct hlist_head *head,
1452 xlog_recover_t *trans;
1454 hlist_for_each_entry(trans, head, r_list) {
1455 if (trans->r_log_tid == tid)
1462 xlog_recover_new_tid(
1463 struct hlist_head *head,
1467 xlog_recover_t *trans;
1469 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1470 trans->r_log_tid = tid;
1472 INIT_LIST_HEAD(&trans->r_itemq);
1474 INIT_HLIST_NODE(&trans->r_list);
1475 hlist_add_head(&trans->r_list, head);
1479 xlog_recover_add_item(
1480 struct list_head *head)
1482 xlog_recover_item_t *item;
1484 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1485 INIT_LIST_HEAD(&item->ri_list);
1486 list_add_tail(&item->ri_list, head);
1490 xlog_recover_add_to_cont_trans(
1492 struct xlog_recover *trans,
1496 xlog_recover_item_t *item;
1497 xfs_caddr_t ptr, old_ptr;
1500 if (list_empty(&trans->r_itemq)) {
1501 /* finish copying rest of trans header */
1502 xlog_recover_add_item(&trans->r_itemq);
1503 ptr = (xfs_caddr_t) &trans->r_theader +
1504 sizeof(xfs_trans_header_t) - len;
1505 memcpy(ptr, dp, len); /* d, s, l */
1508 /* take the tail entry */
1509 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1511 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1512 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1514 ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP);
1515 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1516 item->ri_buf[item->ri_cnt-1].i_len += len;
1517 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1518 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
1523 * The next region to add is the start of a new region. It could be
1524 * a whole region or it could be the first part of a new region. Because
1525 * of this, the assumption here is that the type and size fields of all
1526 * format structures fit into the first 32 bits of the structure.
1528 * This works because all regions must be 32 bit aligned. Therefore, we
1529 * either have both fields or we have neither field. In the case we have
1530 * neither field, the data part of the region is zero length. We only have
1531 * a log_op_header and can throw away the header since a new one will appear
1532 * later. If we have at least 4 bytes, then we can determine how many regions
1533 * will appear in the current log item.
1536 xlog_recover_add_to_trans(
1538 struct xlog_recover *trans,
1542 xfs_inode_log_format_t *in_f; /* any will do */
1543 xlog_recover_item_t *item;
1548 if (list_empty(&trans->r_itemq)) {
1549 /* we need to catch log corruptions here */
1550 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
1551 xfs_warn(log->l_mp, "%s: bad header magic number",
1554 return XFS_ERROR(EIO);
1556 if (len == sizeof(xfs_trans_header_t))
1557 xlog_recover_add_item(&trans->r_itemq);
1558 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1562 ptr = kmem_alloc(len, KM_SLEEP);
1563 memcpy(ptr, dp, len);
1564 in_f = (xfs_inode_log_format_t *)ptr;
1566 /* take the tail entry */
1567 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1568 if (item->ri_total != 0 &&
1569 item->ri_total == item->ri_cnt) {
1570 /* tail item is in use, get a new one */
1571 xlog_recover_add_item(&trans->r_itemq);
1572 item = list_entry(trans->r_itemq.prev,
1573 xlog_recover_item_t, ri_list);
1576 if (item->ri_total == 0) { /* first region to be added */
1577 if (in_f->ilf_size == 0 ||
1578 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
1580 "bad number of regions (%d) in inode log format",
1584 return XFS_ERROR(EIO);
1587 item->ri_total = in_f->ilf_size;
1589 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1592 ASSERT(item->ri_total > item->ri_cnt);
1593 /* Description region is ri_buf[0] */
1594 item->ri_buf[item->ri_cnt].i_addr = ptr;
1595 item->ri_buf[item->ri_cnt].i_len = len;
1597 trace_xfs_log_recover_item_add(log, trans, item, 0);
1602 * Sort the log items in the transaction.
1604 * The ordering constraints are defined by the inode allocation and unlink
1605 * behaviour. The rules are:
1607 * 1. Every item is only logged once in a given transaction. Hence it
1608 * represents the last logged state of the item. Hence ordering is
1609 * dependent on the order in which operations need to be performed so
1610 * required initial conditions are always met.
1612 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1613 * there's nothing to replay from them so we can simply cull them
1614 * from the transaction. However, we can't do that until after we've
1615 * replayed all the other items because they may be dependent on the
1616 * cancelled buffer and replaying the cancelled buffer can remove it
1617 * form the cancelled buffer table. Hence they have tobe done last.
1619 * 3. Inode allocation buffers must be replayed before inode items that
1620 * read the buffer and replay changes into it. For filesystems using the
1621 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1622 * treated the same as inode allocation buffers as they create and
1623 * initialise the buffers directly.
1625 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1626 * This ensures that inodes are completely flushed to the inode buffer
1627 * in a "free" state before we remove the unlinked inode list pointer.
1629 * Hence the ordering needs to be inode allocation buffers first, inode items
1630 * second, inode unlink buffers third and cancelled buffers last.
1632 * But there's a problem with that - we can't tell an inode allocation buffer
1633 * apart from a regular buffer, so we can't separate them. We can, however,
1634 * tell an inode unlink buffer from the others, and so we can separate them out
1635 * from all the other buffers and move them to last.
1637 * Hence, 4 lists, in order from head to tail:
1638 * - buffer_list for all buffers except cancelled/inode unlink buffers
1639 * - item_list for all non-buffer items
1640 * - inode_buffer_list for inode unlink buffers
1641 * - cancel_list for the cancelled buffers
1643 * Note that we add objects to the tail of the lists so that first-to-last
1644 * ordering is preserved within the lists. Adding objects to the head of the
1645 * list means when we traverse from the head we walk them in last-to-first
1646 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1647 * but for all other items there may be specific ordering that we need to
1651 xlog_recover_reorder_trans(
1653 struct xlog_recover *trans,
1656 xlog_recover_item_t *item, *n;
1658 LIST_HEAD(sort_list);
1659 LIST_HEAD(cancel_list);
1660 LIST_HEAD(buffer_list);
1661 LIST_HEAD(inode_buffer_list);
1662 LIST_HEAD(inode_list);
1664 list_splice_init(&trans->r_itemq, &sort_list);
1665 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1666 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1668 switch (ITEM_TYPE(item)) {
1669 case XFS_LI_ICREATE:
1670 list_move_tail(&item->ri_list, &buffer_list);
1673 if (buf_f->blf_flags & XFS_BLF_CANCEL) {
1674 trace_xfs_log_recover_item_reorder_head(log,
1676 list_move(&item->ri_list, &cancel_list);
1679 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1680 list_move(&item->ri_list, &inode_buffer_list);
1683 list_move_tail(&item->ri_list, &buffer_list);
1687 case XFS_LI_QUOTAOFF:
1690 trace_xfs_log_recover_item_reorder_tail(log,
1692 list_move_tail(&item->ri_list, &inode_list);
1696 "%s: unrecognized type of log operation",
1700 * return the remaining items back to the transaction
1701 * item list so they can be freed in caller.
1703 if (!list_empty(&sort_list))
1704 list_splice_init(&sort_list, &trans->r_itemq);
1705 error = XFS_ERROR(EIO);
1710 ASSERT(list_empty(&sort_list));
1711 if (!list_empty(&buffer_list))
1712 list_splice(&buffer_list, &trans->r_itemq);
1713 if (!list_empty(&inode_list))
1714 list_splice_tail(&inode_list, &trans->r_itemq);
1715 if (!list_empty(&inode_buffer_list))
1716 list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1717 if (!list_empty(&cancel_list))
1718 list_splice_tail(&cancel_list, &trans->r_itemq);
1723 * Build up the table of buf cancel records so that we don't replay
1724 * cancelled data in the second pass. For buffer records that are
1725 * not cancel records, there is nothing to do here so we just return.
1727 * If we get a cancel record which is already in the table, this indicates
1728 * that the buffer was cancelled multiple times. In order to ensure
1729 * that during pass 2 we keep the record in the table until we reach its
1730 * last occurrence in the log, we keep a reference count in the cancel
1731 * record in the table to tell us how many times we expect to see this
1732 * record during the second pass.
1735 xlog_recover_buffer_pass1(
1737 struct xlog_recover_item *item)
1739 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1740 struct list_head *bucket;
1741 struct xfs_buf_cancel *bcp;
1744 * If this isn't a cancel buffer item, then just return.
1746 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1747 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
1752 * Insert an xfs_buf_cancel record into the hash table of them.
1753 * If there is already an identical record, bump its reference count.
1755 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1756 list_for_each_entry(bcp, bucket, bc_list) {
1757 if (bcp->bc_blkno == buf_f->blf_blkno &&
1758 bcp->bc_len == buf_f->blf_len) {
1760 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
1765 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1766 bcp->bc_blkno = buf_f->blf_blkno;
1767 bcp->bc_len = buf_f->blf_len;
1768 bcp->bc_refcount = 1;
1769 list_add_tail(&bcp->bc_list, bucket);
1771 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
1776 * Check to see whether the buffer being recovered has a corresponding
1777 * entry in the buffer cancel record table. If it is, return the cancel
1778 * buffer structure to the caller.
1780 STATIC struct xfs_buf_cancel *
1781 xlog_peek_buffer_cancelled(
1787 struct list_head *bucket;
1788 struct xfs_buf_cancel *bcp;
1790 if (!log->l_buf_cancel_table) {
1791 /* empty table means no cancelled buffers in the log */
1792 ASSERT(!(flags & XFS_BLF_CANCEL));
1796 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1797 list_for_each_entry(bcp, bucket, bc_list) {
1798 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
1803 * We didn't find a corresponding entry in the table, so return 0 so
1804 * that the buffer is NOT cancelled.
1806 ASSERT(!(flags & XFS_BLF_CANCEL));
1811 * If the buffer is being cancelled then return 1 so that it will be cancelled,
1812 * otherwise return 0. If the buffer is actually a buffer cancel item
1813 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
1814 * table and remove it from the table if this is the last reference.
1816 * We remove the cancel record from the table when we encounter its last
1817 * occurrence in the log so that if the same buffer is re-used again after its
1818 * last cancellation we actually replay the changes made at that point.
1821 xlog_check_buffer_cancelled(
1827 struct xfs_buf_cancel *bcp;
1829 bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags);
1834 * We've go a match, so return 1 so that the recovery of this buffer
1835 * is cancelled. If this buffer is actually a buffer cancel log
1836 * item, then decrement the refcount on the one in the table and
1837 * remove it if this is the last reference.
1839 if (flags & XFS_BLF_CANCEL) {
1840 if (--bcp->bc_refcount == 0) {
1841 list_del(&bcp->bc_list);
1849 * Perform recovery for a buffer full of inodes. In these buffers, the only
1850 * data which should be recovered is that which corresponds to the
1851 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1852 * data for the inodes is always logged through the inodes themselves rather
1853 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1855 * The only time when buffers full of inodes are fully recovered is when the
1856 * buffer is full of newly allocated inodes. In this case the buffer will
1857 * not be marked as an inode buffer and so will be sent to
1858 * xlog_recover_do_reg_buffer() below during recovery.
1861 xlog_recover_do_inode_buffer(
1862 struct xfs_mount *mp,
1863 xlog_recover_item_t *item,
1865 xfs_buf_log_format_t *buf_f)
1871 int reg_buf_offset = 0;
1872 int reg_buf_bytes = 0;
1873 int next_unlinked_offset;
1875 xfs_agino_t *logged_nextp;
1876 xfs_agino_t *buffer_nextp;
1878 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
1881 * Post recovery validation only works properly on CRC enabled
1884 if (xfs_sb_version_hascrc(&mp->m_sb))
1885 bp->b_ops = &xfs_inode_buf_ops;
1887 inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog;
1888 for (i = 0; i < inodes_per_buf; i++) {
1889 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1890 offsetof(xfs_dinode_t, di_next_unlinked);
1892 while (next_unlinked_offset >=
1893 (reg_buf_offset + reg_buf_bytes)) {
1895 * The next di_next_unlinked field is beyond
1896 * the current logged region. Find the next
1897 * logged region that contains or is beyond
1898 * the current di_next_unlinked field.
1901 bit = xfs_next_bit(buf_f->blf_data_map,
1902 buf_f->blf_map_size, bit);
1905 * If there are no more logged regions in the
1906 * buffer, then we're done.
1911 nbits = xfs_contig_bits(buf_f->blf_data_map,
1912 buf_f->blf_map_size, bit);
1914 reg_buf_offset = bit << XFS_BLF_SHIFT;
1915 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
1920 * If the current logged region starts after the current
1921 * di_next_unlinked field, then move on to the next
1922 * di_next_unlinked field.
1924 if (next_unlinked_offset < reg_buf_offset)
1927 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1928 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
1929 ASSERT((reg_buf_offset + reg_buf_bytes) <=
1930 BBTOB(bp->b_io_length));
1933 * The current logged region contains a copy of the
1934 * current di_next_unlinked field. Extract its value
1935 * and copy it to the buffer copy.
1937 logged_nextp = item->ri_buf[item_index].i_addr +
1938 next_unlinked_offset - reg_buf_offset;
1939 if (unlikely(*logged_nextp == 0)) {
1941 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1942 "Trying to replay bad (0) inode di_next_unlinked field.",
1944 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1945 XFS_ERRLEVEL_LOW, mp);
1946 return XFS_ERROR(EFSCORRUPTED);
1949 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1950 next_unlinked_offset);
1951 *buffer_nextp = *logged_nextp;
1954 * If necessary, recalculate the CRC in the on-disk inode. We
1955 * have to leave the inode in a consistent state for whoever
1958 xfs_dinode_calc_crc(mp, (struct xfs_dinode *)
1959 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
1967 * V5 filesystems know the age of the buffer on disk being recovered. We can
1968 * have newer objects on disk than we are replaying, and so for these cases we
1969 * don't want to replay the current change as that will make the buffer contents
1970 * temporarily invalid on disk.
1972 * The magic number might not match the buffer type we are going to recover
1973 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
1974 * extract the LSN of the existing object in the buffer based on it's current
1975 * magic number. If we don't recognise the magic number in the buffer, then
1976 * return a LSN of -1 so that the caller knows it was an unrecognised block and
1977 * so can recover the buffer.
1979 * Note: we cannot rely solely on magic number matches to determine that the
1980 * buffer has a valid LSN - we also need to verify that it belongs to this
1981 * filesystem, so we need to extract the object's LSN and compare it to that
1982 * which we read from the superblock. If the UUIDs don't match, then we've got a
1983 * stale metadata block from an old filesystem instance that we need to recover
1987 xlog_recover_get_buf_lsn(
1988 struct xfs_mount *mp,
1994 void *blk = bp->b_addr;
1998 /* v4 filesystems always recover immediately */
1999 if (!xfs_sb_version_hascrc(&mp->m_sb))
2000 goto recover_immediately;
2002 magic32 = be32_to_cpu(*(__be32 *)blk);
2004 case XFS_ABTB_CRC_MAGIC:
2005 case XFS_ABTC_CRC_MAGIC:
2006 case XFS_ABTB_MAGIC:
2007 case XFS_ABTC_MAGIC:
2008 case XFS_IBT_CRC_MAGIC:
2009 case XFS_IBT_MAGIC: {
2010 struct xfs_btree_block *btb = blk;
2012 lsn = be64_to_cpu(btb->bb_u.s.bb_lsn);
2013 uuid = &btb->bb_u.s.bb_uuid;
2016 case XFS_BMAP_CRC_MAGIC:
2017 case XFS_BMAP_MAGIC: {
2018 struct xfs_btree_block *btb = blk;
2020 lsn = be64_to_cpu(btb->bb_u.l.bb_lsn);
2021 uuid = &btb->bb_u.l.bb_uuid;
2025 lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn);
2026 uuid = &((struct xfs_agf *)blk)->agf_uuid;
2028 case XFS_AGFL_MAGIC:
2029 lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn);
2030 uuid = &((struct xfs_agfl *)blk)->agfl_uuid;
2033 lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn);
2034 uuid = &((struct xfs_agi *)blk)->agi_uuid;
2036 case XFS_SYMLINK_MAGIC:
2037 lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn);
2038 uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid;
2040 case XFS_DIR3_BLOCK_MAGIC:
2041 case XFS_DIR3_DATA_MAGIC:
2042 case XFS_DIR3_FREE_MAGIC:
2043 lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn);
2044 uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid;
2046 case XFS_ATTR3_RMT_MAGIC:
2047 lsn = be64_to_cpu(((struct xfs_attr3_rmt_hdr *)blk)->rm_lsn);
2048 uuid = &((struct xfs_attr3_rmt_hdr *)blk)->rm_uuid;
2051 lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn);
2052 uuid = &((struct xfs_dsb *)blk)->sb_uuid;
2058 if (lsn != (xfs_lsn_t)-1) {
2059 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
2060 goto recover_immediately;
2064 magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic);
2066 case XFS_DIR3_LEAF1_MAGIC:
2067 case XFS_DIR3_LEAFN_MAGIC:
2068 case XFS_DA3_NODE_MAGIC:
2069 lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn);
2070 uuid = &((struct xfs_da3_blkinfo *)blk)->uuid;
2076 if (lsn != (xfs_lsn_t)-1) {
2077 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
2078 goto recover_immediately;
2083 * We do individual object checks on dquot and inode buffers as they
2084 * have their own individual LSN records. Also, we could have a stale
2085 * buffer here, so we have to at least recognise these buffer types.
2087 * A notd complexity here is inode unlinked list processing - it logs
2088 * the inode directly in the buffer, but we don't know which inodes have
2089 * been modified, and there is no global buffer LSN. Hence we need to
2090 * recover all inode buffer types immediately. This problem will be
2091 * fixed by logical logging of the unlinked list modifications.
2093 magic16 = be16_to_cpu(*(__be16 *)blk);
2095 case XFS_DQUOT_MAGIC:
2096 case XFS_DINODE_MAGIC:
2097 goto recover_immediately;
2102 /* unknown buffer contents, recover immediately */
2104 recover_immediately:
2105 return (xfs_lsn_t)-1;
2110 * Validate the recovered buffer is of the correct type and attach the
2111 * appropriate buffer operations to them for writeback. Magic numbers are in a
2113 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2114 * the first 32 bits of the buffer (most blocks),
2115 * inside a struct xfs_da_blkinfo at the start of the buffer.
2118 xlog_recover_validate_buf_type(
2119 struct xfs_mount *mp,
2121 xfs_buf_log_format_t *buf_f)
2123 struct xfs_da_blkinfo *info = bp->b_addr;
2129 * We can only do post recovery validation on items on CRC enabled
2130 * fielsystems as we need to know when the buffer was written to be able
2131 * to determine if we should have replayed the item. If we replay old
2132 * metadata over a newer buffer, then it will enter a temporarily
2133 * inconsistent state resulting in verification failures. Hence for now
2134 * just avoid the verification stage for non-crc filesystems
2136 if (!xfs_sb_version_hascrc(&mp->m_sb))
2139 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
2140 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
2141 magicda = be16_to_cpu(info->magic);
2142 switch (xfs_blft_from_flags(buf_f)) {
2143 case XFS_BLFT_BTREE_BUF:
2145 case XFS_ABTB_CRC_MAGIC:
2146 case XFS_ABTC_CRC_MAGIC:
2147 case XFS_ABTB_MAGIC:
2148 case XFS_ABTC_MAGIC:
2149 bp->b_ops = &xfs_allocbt_buf_ops;
2151 case XFS_IBT_CRC_MAGIC:
2153 bp->b_ops = &xfs_inobt_buf_ops;
2155 case XFS_BMAP_CRC_MAGIC:
2156 case XFS_BMAP_MAGIC:
2157 bp->b_ops = &xfs_bmbt_buf_ops;
2160 xfs_warn(mp, "Bad btree block magic!");
2165 case XFS_BLFT_AGF_BUF:
2166 if (magic32 != XFS_AGF_MAGIC) {
2167 xfs_warn(mp, "Bad AGF block magic!");
2171 bp->b_ops = &xfs_agf_buf_ops;
2173 case XFS_BLFT_AGFL_BUF:
2174 if (magic32 != XFS_AGFL_MAGIC) {
2175 xfs_warn(mp, "Bad AGFL block magic!");
2179 bp->b_ops = &xfs_agfl_buf_ops;
2181 case XFS_BLFT_AGI_BUF:
2182 if (magic32 != XFS_AGI_MAGIC) {
2183 xfs_warn(mp, "Bad AGI block magic!");
2187 bp->b_ops = &xfs_agi_buf_ops;
2189 case XFS_BLFT_UDQUOT_BUF:
2190 case XFS_BLFT_PDQUOT_BUF:
2191 case XFS_BLFT_GDQUOT_BUF:
2192 #ifdef CONFIG_XFS_QUOTA
2193 if (magic16 != XFS_DQUOT_MAGIC) {
2194 xfs_warn(mp, "Bad DQUOT block magic!");
2198 bp->b_ops = &xfs_dquot_buf_ops;
2201 "Trying to recover dquots without QUOTA support built in!");
2205 case XFS_BLFT_DINO_BUF:
2206 if (magic16 != XFS_DINODE_MAGIC) {
2207 xfs_warn(mp, "Bad INODE block magic!");
2211 bp->b_ops = &xfs_inode_buf_ops;
2213 case XFS_BLFT_SYMLINK_BUF:
2214 if (magic32 != XFS_SYMLINK_MAGIC) {
2215 xfs_warn(mp, "Bad symlink block magic!");
2219 bp->b_ops = &xfs_symlink_buf_ops;
2221 case XFS_BLFT_DIR_BLOCK_BUF:
2222 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2223 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2224 xfs_warn(mp, "Bad dir block magic!");
2228 bp->b_ops = &xfs_dir3_block_buf_ops;
2230 case XFS_BLFT_DIR_DATA_BUF:
2231 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2232 magic32 != XFS_DIR3_DATA_MAGIC) {
2233 xfs_warn(mp, "Bad dir data magic!");
2237 bp->b_ops = &xfs_dir3_data_buf_ops;
2239 case XFS_BLFT_DIR_FREE_BUF:
2240 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2241 magic32 != XFS_DIR3_FREE_MAGIC) {
2242 xfs_warn(mp, "Bad dir3 free magic!");
2246 bp->b_ops = &xfs_dir3_free_buf_ops;
2248 case XFS_BLFT_DIR_LEAF1_BUF:
2249 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2250 magicda != XFS_DIR3_LEAF1_MAGIC) {
2251 xfs_warn(mp, "Bad dir leaf1 magic!");
2255 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2257 case XFS_BLFT_DIR_LEAFN_BUF:
2258 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2259 magicda != XFS_DIR3_LEAFN_MAGIC) {
2260 xfs_warn(mp, "Bad dir leafn magic!");
2264 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2266 case XFS_BLFT_DA_NODE_BUF:
2267 if (magicda != XFS_DA_NODE_MAGIC &&
2268 magicda != XFS_DA3_NODE_MAGIC) {
2269 xfs_warn(mp, "Bad da node magic!");
2273 bp->b_ops = &xfs_da3_node_buf_ops;
2275 case XFS_BLFT_ATTR_LEAF_BUF:
2276 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2277 magicda != XFS_ATTR3_LEAF_MAGIC) {
2278 xfs_warn(mp, "Bad attr leaf magic!");
2282 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2284 case XFS_BLFT_ATTR_RMT_BUF:
2285 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
2286 xfs_warn(mp, "Bad attr remote magic!");
2290 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2292 case XFS_BLFT_SB_BUF:
2293 if (magic32 != XFS_SB_MAGIC) {
2294 xfs_warn(mp, "Bad SB block magic!");
2298 bp->b_ops = &xfs_sb_buf_ops;
2301 xfs_warn(mp, "Unknown buffer type %d!",
2302 xfs_blft_from_flags(buf_f));
2308 * Perform a 'normal' buffer recovery. Each logged region of the
2309 * buffer should be copied over the corresponding region in the
2310 * given buffer. The bitmap in the buf log format structure indicates
2311 * where to place the logged data.
2314 xlog_recover_do_reg_buffer(
2315 struct xfs_mount *mp,
2316 xlog_recover_item_t *item,
2318 xfs_buf_log_format_t *buf_f)
2325 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2328 i = 1; /* 0 is the buf format structure */
2330 bit = xfs_next_bit(buf_f->blf_data_map,
2331 buf_f->blf_map_size, bit);
2334 nbits = xfs_contig_bits(buf_f->blf_data_map,
2335 buf_f->blf_map_size, bit);
2337 ASSERT(item->ri_buf[i].i_addr != NULL);
2338 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
2339 ASSERT(BBTOB(bp->b_io_length) >=
2340 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
2343 * The dirty regions logged in the buffer, even though
2344 * contiguous, may span multiple chunks. This is because the
2345 * dirty region may span a physical page boundary in a buffer
2346 * and hence be split into two separate vectors for writing into
2347 * the log. Hence we need to trim nbits back to the length of
2348 * the current region being copied out of the log.
2350 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2351 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2354 * Do a sanity check if this is a dquot buffer. Just checking
2355 * the first dquot in the buffer should do. XXXThis is
2356 * probably a good thing to do for other buf types also.
2359 if (buf_f->blf_flags &
2360 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2361 if (item->ri_buf[i].i_addr == NULL) {
2363 "XFS: NULL dquot in %s.", __func__);
2366 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
2368 "XFS: dquot too small (%d) in %s.",
2369 item->ri_buf[i].i_len, __func__);
2372 error = xfs_dqcheck(mp, item->ri_buf[i].i_addr,
2373 -1, 0, XFS_QMOPT_DOWARN,
2374 "dquot_buf_recover");
2379 memcpy(xfs_buf_offset(bp,
2380 (uint)bit << XFS_BLF_SHIFT), /* dest */
2381 item->ri_buf[i].i_addr, /* source */
2382 nbits<<XFS_BLF_SHIFT); /* length */
2388 /* Shouldn't be any more regions */
2389 ASSERT(i == item->ri_total);
2391 xlog_recover_validate_buf_type(mp, bp, buf_f);
2395 * Perform a dquot buffer recovery.
2396 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2397 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2398 * Else, treat it as a regular buffer and do recovery.
2401 xlog_recover_do_dquot_buffer(
2402 struct xfs_mount *mp,
2404 struct xlog_recover_item *item,
2406 struct xfs_buf_log_format *buf_f)
2410 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2413 * Filesystems are required to send in quota flags at mount time.
2415 if (mp->m_qflags == 0) {
2420 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2421 type |= XFS_DQ_USER;
2422 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2423 type |= XFS_DQ_PROJ;
2424 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2425 type |= XFS_DQ_GROUP;
2427 * This type of quotas was turned off, so ignore this buffer
2429 if (log->l_quotaoffs_flag & type)
2432 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2436 * This routine replays a modification made to a buffer at runtime.
2437 * There are actually two types of buffer, regular and inode, which
2438 * are handled differently. Inode buffers are handled differently
2439 * in that we only recover a specific set of data from them, namely
2440 * the inode di_next_unlinked fields. This is because all other inode
2441 * data is actually logged via inode records and any data we replay
2442 * here which overlaps that may be stale.
2444 * When meta-data buffers are freed at run time we log a buffer item
2445 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2446 * of the buffer in the log should not be replayed at recovery time.
2447 * This is so that if the blocks covered by the buffer are reused for
2448 * file data before we crash we don't end up replaying old, freed
2449 * meta-data into a user's file.
2451 * To handle the cancellation of buffer log items, we make two passes
2452 * over the log during recovery. During the first we build a table of
2453 * those buffers which have been cancelled, and during the second we
2454 * only replay those buffers which do not have corresponding cancel
2455 * records in the table. See xlog_recover_buffer_pass[1,2] above
2456 * for more details on the implementation of the table of cancel records.
2459 xlog_recover_buffer_pass2(
2461 struct list_head *buffer_list,
2462 struct xlog_recover_item *item,
2463 xfs_lsn_t current_lsn)
2465 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2466 xfs_mount_t *mp = log->l_mp;
2473 * In this pass we only want to recover all the buffers which have
2474 * not been cancelled and are not cancellation buffers themselves.
2476 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2477 buf_f->blf_len, buf_f->blf_flags)) {
2478 trace_xfs_log_recover_buf_cancel(log, buf_f);
2482 trace_xfs_log_recover_buf_recover(log, buf_f);
2485 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2486 buf_flags |= XBF_UNMAPPED;
2488 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2491 return XFS_ERROR(ENOMEM);
2492 error = bp->b_error;
2494 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
2499 * Recover the buffer only if we get an LSN from it and it's less than
2500 * the lsn of the transaction we are replaying.
2502 * Note that we have to be extremely careful of readahead here.
2503 * Readahead does not attach verfiers to the buffers so if we don't
2504 * actually do any replay after readahead because of the LSN we found
2505 * in the buffer if more recent than that current transaction then we
2506 * need to attach the verifier directly. Failure to do so can lead to
2507 * future recovery actions (e.g. EFI and unlinked list recovery) can
2508 * operate on the buffers and they won't get the verifier attached. This
2509 * can lead to blocks on disk having the correct content but a stale
2512 * It is safe to assume these clean buffers are currently up to date.
2513 * If the buffer is dirtied by a later transaction being replayed, then
2514 * the verifier will be reset to match whatever recover turns that
2517 lsn = xlog_recover_get_buf_lsn(mp, bp);
2518 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2519 xlog_recover_validate_buf_type(mp, bp, buf_f);
2523 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
2524 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2525 } else if (buf_f->blf_flags &
2526 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2527 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2529 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2535 * Perform delayed write on the buffer. Asynchronous writes will be
2536 * slower when taking into account all the buffers to be flushed.
2538 * Also make sure that only inode buffers with good sizes stay in
2539 * the buffer cache. The kernel moves inodes in buffers of 1 block
2540 * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode
2541 * buffers in the log can be a different size if the log was generated
2542 * by an older kernel using unclustered inode buffers or a newer kernel
2543 * running with a different inode cluster size. Regardless, if the
2544 * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size)
2545 * for *our* value of mp->m_inode_cluster_size, then we need to keep
2546 * the buffer out of the buffer cache so that the buffer won't
2547 * overlap with future reads of those inodes.
2549 if (XFS_DINODE_MAGIC ==
2550 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2551 (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize,
2552 (__uint32_t)log->l_mp->m_inode_cluster_size))) {
2554 error = xfs_bwrite(bp);
2556 ASSERT(bp->b_target->bt_mount == mp);
2557 bp->b_iodone = xlog_recover_iodone;
2558 xfs_buf_delwri_queue(bp, buffer_list);
2567 * Inode fork owner changes
2569 * If we have been told that we have to reparent the inode fork, it's because an
2570 * extent swap operation on a CRC enabled filesystem has been done and we are
2571 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2574 * The complexity here is that we don't have an inode context to work with, so
2575 * after we've replayed the inode we need to instantiate one. This is where the
2578 * We are in the middle of log recovery, so we can't run transactions. That
2579 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2580 * that will result in the corresponding iput() running the inode through
2581 * xfs_inactive(). If we've just replayed an inode core that changes the link
2582 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2583 * transactions (bad!).
2585 * So, to avoid this, we instantiate an inode directly from the inode core we've
2586 * just recovered. We have the buffer still locked, and all we really need to
2587 * instantiate is the inode core and the forks being modified. We can do this
2588 * manually, then run the inode btree owner change, and then tear down the
2589 * xfs_inode without having to run any transactions at all.
2591 * Also, because we don't have a transaction context available here but need to
2592 * gather all the buffers we modify for writeback so we pass the buffer_list
2593 * instead for the operation to use.
2597 xfs_recover_inode_owner_change(
2598 struct xfs_mount *mp,
2599 struct xfs_dinode *dip,
2600 struct xfs_inode_log_format *in_f,
2601 struct list_head *buffer_list)
2603 struct xfs_inode *ip;
2606 ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER));
2608 ip = xfs_inode_alloc(mp, in_f->ilf_ino);
2612 /* instantiate the inode */
2613 xfs_dinode_from_disk(&ip->i_d, dip);
2614 ASSERT(ip->i_d.di_version >= 3);
2616 error = xfs_iformat_fork(ip, dip);
2621 if (in_f->ilf_fields & XFS_ILOG_DOWNER) {
2622 ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT);
2623 error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK,
2624 ip->i_ino, buffer_list);
2629 if (in_f->ilf_fields & XFS_ILOG_AOWNER) {
2630 ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT);
2631 error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK,
2632 ip->i_ino, buffer_list);
2643 xlog_recover_inode_pass2(
2645 struct list_head *buffer_list,
2646 struct xlog_recover_item *item,
2647 xfs_lsn_t current_lsn)
2649 xfs_inode_log_format_t *in_f;
2650 xfs_mount_t *mp = log->l_mp;
2659 xfs_icdinode_t *dicp;
2663 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2664 in_f = item->ri_buf[0].i_addr;
2666 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
2668 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2674 * Inode buffers can be freed, look out for it,
2675 * and do not replay the inode.
2677 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2678 in_f->ilf_len, 0)) {
2680 trace_xfs_log_recover_inode_cancel(log, in_f);
2683 trace_xfs_log_recover_inode_recover(log, in_f);
2685 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
2686 &xfs_inode_buf_ops);
2691 error = bp->b_error;
2693 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
2696 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2697 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
2700 * Make sure the place we're flushing out to really looks
2703 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
2705 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2706 __func__, dip, bp, in_f->ilf_ino);
2707 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2708 XFS_ERRLEVEL_LOW, mp);
2709 error = EFSCORRUPTED;
2712 dicp = item->ri_buf[1].i_addr;
2713 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2715 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2716 __func__, item, in_f->ilf_ino);
2717 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2718 XFS_ERRLEVEL_LOW, mp);
2719 error = EFSCORRUPTED;
2724 * If the inode has an LSN in it, recover the inode only if it's less
2725 * than the lsn of the transaction we are replaying. Note: we still
2726 * need to replay an owner change even though the inode is more recent
2727 * than the transaction as there is no guarantee that all the btree
2728 * blocks are more recent than this transaction, too.
2730 if (dip->di_version >= 3) {
2731 xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn);
2733 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2734 trace_xfs_log_recover_inode_skip(log, in_f);
2736 goto out_owner_change;
2741 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2742 * are transactional and if ordering is necessary we can determine that
2743 * more accurately by the LSN field in the V3 inode core. Don't trust
2744 * the inode versions we might be changing them here - use the
2745 * superblock flag to determine whether we need to look at di_flushiter
2746 * to skip replay when the on disk inode is newer than the log one
2748 if (!xfs_sb_version_hascrc(&mp->m_sb) &&
2749 dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
2751 * Deal with the wrap case, DI_MAX_FLUSH is less
2752 * than smaller numbers
2754 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
2755 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2758 trace_xfs_log_recover_inode_skip(log, in_f);
2764 /* Take the opportunity to reset the flush iteration count */
2765 dicp->di_flushiter = 0;
2767 if (unlikely(S_ISREG(dicp->di_mode))) {
2768 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2769 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2770 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2771 XFS_ERRLEVEL_LOW, mp, dicp);
2773 "%s: Bad regular inode log record, rec ptr 0x%p, "
2774 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2775 __func__, item, dip, bp, in_f->ilf_ino);
2776 error = EFSCORRUPTED;
2779 } else if (unlikely(S_ISDIR(dicp->di_mode))) {
2780 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2781 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2782 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2783 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2784 XFS_ERRLEVEL_LOW, mp, dicp);
2786 "%s: Bad dir inode log record, rec ptr 0x%p, "
2787 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2788 __func__, item, dip, bp, in_f->ilf_ino);
2789 error = EFSCORRUPTED;
2793 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2794 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2795 XFS_ERRLEVEL_LOW, mp, dicp);
2797 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2798 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2799 __func__, item, dip, bp, in_f->ilf_ino,
2800 dicp->di_nextents + dicp->di_anextents,
2802 error = EFSCORRUPTED;
2805 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2806 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2807 XFS_ERRLEVEL_LOW, mp, dicp);
2809 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2810 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
2811 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
2812 error = EFSCORRUPTED;
2815 isize = xfs_icdinode_size(dicp->di_version);
2816 if (unlikely(item->ri_buf[1].i_len > isize)) {
2817 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
2818 XFS_ERRLEVEL_LOW, mp, dicp);
2820 "%s: Bad inode log record length %d, rec ptr 0x%p",
2821 __func__, item->ri_buf[1].i_len, item);
2822 error = EFSCORRUPTED;
2826 /* The core is in in-core format */
2827 xfs_dinode_to_disk(dip, dicp);
2829 /* the rest is in on-disk format */
2830 if (item->ri_buf[1].i_len > isize) {
2831 memcpy((char *)dip + isize,
2832 item->ri_buf[1].i_addr + isize,
2833 item->ri_buf[1].i_len - isize);
2836 fields = in_f->ilf_fields;
2837 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2839 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
2842 memcpy(XFS_DFORK_DPTR(dip),
2843 &in_f->ilf_u.ilfu_uuid,
2848 if (in_f->ilf_size == 2)
2849 goto out_owner_change;
2850 len = item->ri_buf[2].i_len;
2851 src = item->ri_buf[2].i_addr;
2852 ASSERT(in_f->ilf_size <= 4);
2853 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2854 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2855 (len == in_f->ilf_dsize));
2857 switch (fields & XFS_ILOG_DFORK) {
2858 case XFS_ILOG_DDATA:
2860 memcpy(XFS_DFORK_DPTR(dip), src, len);
2863 case XFS_ILOG_DBROOT:
2864 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
2865 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
2866 XFS_DFORK_DSIZE(dip, mp));
2871 * There are no data fork flags set.
2873 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2878 * If we logged any attribute data, recover it. There may or
2879 * may not have been any other non-core data logged in this
2882 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2883 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2888 len = item->ri_buf[attr_index].i_len;
2889 src = item->ri_buf[attr_index].i_addr;
2890 ASSERT(len == in_f->ilf_asize);
2892 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2893 case XFS_ILOG_ADATA:
2895 dest = XFS_DFORK_APTR(dip);
2896 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2897 memcpy(dest, src, len);
2900 case XFS_ILOG_ABROOT:
2901 dest = XFS_DFORK_APTR(dip);
2902 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2903 len, (xfs_bmdr_block_t*)dest,
2904 XFS_DFORK_ASIZE(dip, mp));
2908 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
2916 if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER))
2917 error = xfs_recover_inode_owner_change(mp, dip, in_f,
2919 /* re-generate the checksum. */
2920 xfs_dinode_calc_crc(log->l_mp, dip);
2922 ASSERT(bp->b_target->bt_mount == mp);
2923 bp->b_iodone = xlog_recover_iodone;
2924 xfs_buf_delwri_queue(bp, buffer_list);
2931 return XFS_ERROR(error);
2935 * Recover QUOTAOFF records. We simply make a note of it in the xlog
2936 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2940 xlog_recover_quotaoff_pass1(
2942 struct xlog_recover_item *item)
2944 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
2948 * The logitem format's flag tells us if this was user quotaoff,
2949 * group/project quotaoff or both.
2951 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2952 log->l_quotaoffs_flag |= XFS_DQ_USER;
2953 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2954 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2955 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2956 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2962 * Recover a dquot record
2965 xlog_recover_dquot_pass2(
2967 struct list_head *buffer_list,
2968 struct xlog_recover_item *item,
2969 xfs_lsn_t current_lsn)
2971 xfs_mount_t *mp = log->l_mp;
2973 struct xfs_disk_dquot *ddq, *recddq;
2975 xfs_dq_logformat_t *dq_f;
2980 * Filesystems are required to send in quota flags at mount time.
2982 if (mp->m_qflags == 0)
2985 recddq = item->ri_buf[1].i_addr;
2986 if (recddq == NULL) {
2987 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
2988 return XFS_ERROR(EIO);
2990 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
2991 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
2992 item->ri_buf[1].i_len, __func__);
2993 return XFS_ERROR(EIO);
2997 * This type of quotas was turned off, so ignore this record.
2999 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3001 if (log->l_quotaoffs_flag & type)
3005 * At this point we know that quota was _not_ turned off.
3006 * Since the mount flags are not indicating to us otherwise, this
3007 * must mean that quota is on, and the dquot needs to be replayed.
3008 * Remember that we may not have fully recovered the superblock yet,
3009 * so we can't do the usual trick of looking at the SB quota bits.
3011 * The other possibility, of course, is that the quota subsystem was
3012 * removed since the last mount - ENOSYS.
3014 dq_f = item->ri_buf[0].i_addr;
3016 error = xfs_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
3017 "xlog_recover_dquot_pass2 (log copy)");
3019 return XFS_ERROR(EIO);
3020 ASSERT(dq_f->qlf_len == 1);
3022 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
3023 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
3029 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
3032 * At least the magic num portion should be on disk because this
3033 * was among a chunk of dquots created earlier, and we did some
3034 * minimal initialization then.
3036 error = xfs_dqcheck(mp, ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
3037 "xlog_recover_dquot_pass2");
3040 return XFS_ERROR(EIO);
3044 * If the dquot has an LSN in it, recover the dquot only if it's less
3045 * than the lsn of the transaction we are replaying.
3047 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3048 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq;
3049 xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn);
3051 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3056 memcpy(ddq, recddq, item->ri_buf[1].i_len);
3057 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3058 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
3062 ASSERT(dq_f->qlf_size == 2);
3063 ASSERT(bp->b_target->bt_mount == mp);
3064 bp->b_iodone = xlog_recover_iodone;
3065 xfs_buf_delwri_queue(bp, buffer_list);
3073 * This routine is called to create an in-core extent free intent
3074 * item from the efi format structure which was logged on disk.
3075 * It allocates an in-core efi, copies the extents from the format
3076 * structure into it, and adds the efi to the AIL with the given
3080 xlog_recover_efi_pass2(
3082 struct xlog_recover_item *item,
3086 xfs_mount_t *mp = log->l_mp;
3087 xfs_efi_log_item_t *efip;
3088 xfs_efi_log_format_t *efi_formatp;
3090 efi_formatp = item->ri_buf[0].i_addr;
3092 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
3093 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
3094 &(efip->efi_format)))) {
3095 xfs_efi_item_free(efip);
3098 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
3100 spin_lock(&log->l_ailp->xa_lock);
3102 * xfs_trans_ail_update() drops the AIL lock.
3104 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
3110 * This routine is called when an efd format structure is found in
3111 * a committed transaction in the log. It's purpose is to cancel
3112 * the corresponding efi if it was still in the log. To do this
3113 * it searches the AIL for the efi with an id equal to that in the
3114 * efd format structure. If we find it, we remove the efi from the
3118 xlog_recover_efd_pass2(
3120 struct xlog_recover_item *item)
3122 xfs_efd_log_format_t *efd_formatp;
3123 xfs_efi_log_item_t *efip = NULL;
3124 xfs_log_item_t *lip;
3126 struct xfs_ail_cursor cur;
3127 struct xfs_ail *ailp = log->l_ailp;
3129 efd_formatp = item->ri_buf[0].i_addr;
3130 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
3131 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
3132 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
3133 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
3134 efi_id = efd_formatp->efd_efi_id;
3137 * Search for the efi with the id in the efd format structure
3140 spin_lock(&ailp->xa_lock);
3141 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3142 while (lip != NULL) {
3143 if (lip->li_type == XFS_LI_EFI) {
3144 efip = (xfs_efi_log_item_t *)lip;
3145 if (efip->efi_format.efi_id == efi_id) {
3147 * xfs_trans_ail_delete() drops the
3150 xfs_trans_ail_delete(ailp, lip,
3151 SHUTDOWN_CORRUPT_INCORE);
3152 xfs_efi_item_free(efip);
3153 spin_lock(&ailp->xa_lock);
3157 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3159 xfs_trans_ail_cursor_done(ailp, &cur);
3160 spin_unlock(&ailp->xa_lock);
3166 * This routine is called when an inode create format structure is found in a
3167 * committed transaction in the log. It's purpose is to initialise the inodes
3168 * being allocated on disk. This requires us to get inode cluster buffers that
3169 * match the range to be intialised, stamped with inode templates and written
3170 * by delayed write so that subsequent modifications will hit the cached buffer
3171 * and only need writing out at the end of recovery.
3174 xlog_recover_do_icreate_pass2(
3176 struct list_head *buffer_list,
3177 xlog_recover_item_t *item)
3179 struct xfs_mount *mp = log->l_mp;
3180 struct xfs_icreate_log *icl;
3181 xfs_agnumber_t agno;
3182 xfs_agblock_t agbno;
3185 xfs_agblock_t length;
3187 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3188 if (icl->icl_type != XFS_LI_ICREATE) {
3189 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
3193 if (icl->icl_size != 1) {
3194 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
3198 agno = be32_to_cpu(icl->icl_ag);
3199 if (agno >= mp->m_sb.sb_agcount) {
3200 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
3203 agbno = be32_to_cpu(icl->icl_agbno);
3204 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3205 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
3208 isize = be32_to_cpu(icl->icl_isize);
3209 if (isize != mp->m_sb.sb_inodesize) {
3210 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
3213 count = be32_to_cpu(icl->icl_count);
3215 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
3218 length = be32_to_cpu(icl->icl_length);
3219 if (!length || length >= mp->m_sb.sb_agblocks) {
3220 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
3224 /* existing allocation is fixed value */
3225 ASSERT(count == mp->m_ialloc_inos);
3226 ASSERT(length == mp->m_ialloc_blks);
3227 if (count != mp->m_ialloc_inos ||
3228 length != mp->m_ialloc_blks) {
3229 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count 2");
3234 * Inode buffers can be freed. Do not replay the inode initialisation as
3235 * we could be overwriting something written after this inode buffer was
3238 * XXX: we need to iterate all buffers and only init those that are not
3239 * cancelled. I think that a more fine grained factoring of
3240 * xfs_ialloc_inode_init may be appropriate here to enable this to be
3243 if (xlog_check_buffer_cancelled(log,
3244 XFS_AGB_TO_DADDR(mp, agno, agbno), length, 0))
3247 xfs_ialloc_inode_init(mp, NULL, buffer_list, agno, agbno, length,
3248 be32_to_cpu(icl->icl_gen));
3253 * Free up any resources allocated by the transaction
3255 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3258 xlog_recover_free_trans(
3259 struct xlog_recover *trans)
3261 xlog_recover_item_t *item, *n;
3264 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
3265 /* Free the regions in the item. */
3266 list_del(&item->ri_list);
3267 for (i = 0; i < item->ri_cnt; i++)
3268 kmem_free(item->ri_buf[i].i_addr);
3269 /* Free the item itself */
3270 kmem_free(item->ri_buf);
3273 /* Free the transaction recover structure */
3278 xlog_recover_buffer_ra_pass2(
3280 struct xlog_recover_item *item)
3282 struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr;
3283 struct xfs_mount *mp = log->l_mp;
3285 if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
3286 buf_f->blf_len, buf_f->blf_flags)) {
3290 xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
3291 buf_f->blf_len, NULL);
3295 xlog_recover_inode_ra_pass2(
3297 struct xlog_recover_item *item)
3299 struct xfs_inode_log_format ilf_buf;
3300 struct xfs_inode_log_format *ilfp;
3301 struct xfs_mount *mp = log->l_mp;
3304 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3305 ilfp = item->ri_buf[0].i_addr;
3308 memset(ilfp, 0, sizeof(*ilfp));
3309 error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
3314 if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
3317 xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
3318 ilfp->ilf_len, &xfs_inode_buf_ra_ops);
3322 xlog_recover_dquot_ra_pass2(
3324 struct xlog_recover_item *item)
3326 struct xfs_mount *mp = log->l_mp;
3327 struct xfs_disk_dquot *recddq;
3328 struct xfs_dq_logformat *dq_f;
3332 if (mp->m_qflags == 0)
3335 recddq = item->ri_buf[1].i_addr;
3338 if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
3341 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3343 if (log->l_quotaoffs_flag & type)
3346 dq_f = item->ri_buf[0].i_addr;
3348 ASSERT(dq_f->qlf_len == 1);
3350 xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno,
3351 XFS_FSB_TO_BB(mp, dq_f->qlf_len), NULL);
3355 xlog_recover_ra_pass2(
3357 struct xlog_recover_item *item)
3359 switch (ITEM_TYPE(item)) {
3361 xlog_recover_buffer_ra_pass2(log, item);
3364 xlog_recover_inode_ra_pass2(log, item);
3367 xlog_recover_dquot_ra_pass2(log, item);
3371 case XFS_LI_QUOTAOFF:
3378 xlog_recover_commit_pass1(
3380 struct xlog_recover *trans,
3381 struct xlog_recover_item *item)
3383 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
3385 switch (ITEM_TYPE(item)) {
3387 return xlog_recover_buffer_pass1(log, item);
3388 case XFS_LI_QUOTAOFF:
3389 return xlog_recover_quotaoff_pass1(log, item);
3394 case XFS_LI_ICREATE:
3395 /* nothing to do in pass 1 */
3398 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3399 __func__, ITEM_TYPE(item));
3401 return XFS_ERROR(EIO);
3406 xlog_recover_commit_pass2(
3408 struct xlog_recover *trans,
3409 struct list_head *buffer_list,
3410 struct xlog_recover_item *item)
3412 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
3414 switch (ITEM_TYPE(item)) {
3416 return xlog_recover_buffer_pass2(log, buffer_list, item,
3419 return xlog_recover_inode_pass2(log, buffer_list, item,
3422 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
3424 return xlog_recover_efd_pass2(log, item);
3426 return xlog_recover_dquot_pass2(log, buffer_list, item,
3428 case XFS_LI_ICREATE:
3429 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
3430 case XFS_LI_QUOTAOFF:
3431 /* nothing to do in pass2 */
3434 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3435 __func__, ITEM_TYPE(item));
3437 return XFS_ERROR(EIO);
3442 xlog_recover_items_pass2(
3444 struct xlog_recover *trans,
3445 struct list_head *buffer_list,
3446 struct list_head *item_list)
3448 struct xlog_recover_item *item;
3451 list_for_each_entry(item, item_list, ri_list) {
3452 error = xlog_recover_commit_pass2(log, trans,
3462 * Perform the transaction.
3464 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3465 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3468 xlog_recover_commit_trans(
3470 struct xlog_recover *trans,
3475 int items_queued = 0;
3476 struct xlog_recover_item *item;
3477 struct xlog_recover_item *next;
3478 LIST_HEAD (buffer_list);
3479 LIST_HEAD (ra_list);
3480 LIST_HEAD (done_list);
3482 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
3484 hlist_del(&trans->r_list);
3486 error = xlog_recover_reorder_trans(log, trans, pass);
3490 list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
3492 case XLOG_RECOVER_PASS1:
3493 error = xlog_recover_commit_pass1(log, trans, item);
3495 case XLOG_RECOVER_PASS2:
3496 xlog_recover_ra_pass2(log, item);
3497 list_move_tail(&item->ri_list, &ra_list);
3499 if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
3500 error = xlog_recover_items_pass2(log, trans,
3501 &buffer_list, &ra_list);
3502 list_splice_tail_init(&ra_list, &done_list);
3516 if (!list_empty(&ra_list)) {
3518 error = xlog_recover_items_pass2(log, trans,
3519 &buffer_list, &ra_list);
3520 list_splice_tail_init(&ra_list, &done_list);
3523 if (!list_empty(&done_list))
3524 list_splice_init(&done_list, &trans->r_itemq);
3526 xlog_recover_free_trans(trans);
3528 error2 = xfs_buf_delwri_submit(&buffer_list);
3529 return error ? error : error2;
3533 xlog_recover_unmount_trans(
3535 struct xlog_recover *trans)
3537 /* Do nothing now */
3538 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
3543 * There are two valid states of the r_state field. 0 indicates that the
3544 * transaction structure is in a normal state. We have either seen the
3545 * start of the transaction or the last operation we added was not a partial
3546 * operation. If the last operation we added to the transaction was a
3547 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
3549 * NOTE: skip LRs with 0 data length.
3552 xlog_recover_process_data(
3554 struct hlist_head rhash[],
3555 struct xlog_rec_header *rhead,
3561 xlog_op_header_t *ohead;
3562 xlog_recover_t *trans;
3568 lp = dp + be32_to_cpu(rhead->h_len);
3569 num_logops = be32_to_cpu(rhead->h_num_logops);
3571 /* check the log format matches our own - else we can't recover */
3572 if (xlog_header_check_recover(log->l_mp, rhead))
3573 return (XFS_ERROR(EIO));
3575 while ((dp < lp) && num_logops) {
3576 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
3577 ohead = (xlog_op_header_t *)dp;
3578 dp += sizeof(xlog_op_header_t);
3579 if (ohead->oh_clientid != XFS_TRANSACTION &&
3580 ohead->oh_clientid != XFS_LOG) {
3581 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
3582 __func__, ohead->oh_clientid);
3584 return (XFS_ERROR(EIO));
3586 tid = be32_to_cpu(ohead->oh_tid);
3587 hash = XLOG_RHASH(tid);
3588 trans = xlog_recover_find_tid(&rhash[hash], tid);
3589 if (trans == NULL) { /* not found; add new tid */
3590 if (ohead->oh_flags & XLOG_START_TRANS)
3591 xlog_recover_new_tid(&rhash[hash], tid,
3592 be64_to_cpu(rhead->h_lsn));
3594 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
3595 xfs_warn(log->l_mp, "%s: bad length 0x%x",
3596 __func__, be32_to_cpu(ohead->oh_len));
3598 return (XFS_ERROR(EIO));
3600 flags = ohead->oh_flags & ~XLOG_END_TRANS;
3601 if (flags & XLOG_WAS_CONT_TRANS)
3602 flags &= ~XLOG_CONTINUE_TRANS;
3604 case XLOG_COMMIT_TRANS:
3605 error = xlog_recover_commit_trans(log,
3608 case XLOG_UNMOUNT_TRANS:
3609 error = xlog_recover_unmount_trans(log, trans);
3611 case XLOG_WAS_CONT_TRANS:
3612 error = xlog_recover_add_to_cont_trans(log,
3614 be32_to_cpu(ohead->oh_len));
3616 case XLOG_START_TRANS:
3617 xfs_warn(log->l_mp, "%s: bad transaction",
3620 error = XFS_ERROR(EIO);
3623 case XLOG_CONTINUE_TRANS:
3624 error = xlog_recover_add_to_trans(log, trans,
3625 dp, be32_to_cpu(ohead->oh_len));
3628 xfs_warn(log->l_mp, "%s: bad flag 0x%x",
3631 error = XFS_ERROR(EIO);
3635 xlog_recover_free_trans(trans);
3639 dp += be32_to_cpu(ohead->oh_len);
3646 * Process an extent free intent item that was recovered from
3647 * the log. We need to free the extents that it describes.
3650 xlog_recover_process_efi(
3652 xfs_efi_log_item_t *efip)
3654 xfs_efd_log_item_t *efdp;
3659 xfs_fsblock_t startblock_fsb;
3661 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
3664 * First check the validity of the extents described by the
3665 * EFI. If any are bad, then assume that all are bad and
3666 * just toss the EFI.
3668 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3669 extp = &(efip->efi_format.efi_extents[i]);
3670 startblock_fsb = XFS_BB_TO_FSB(mp,
3671 XFS_FSB_TO_DADDR(mp, extp->ext_start));
3672 if ((startblock_fsb == 0) ||
3673 (extp->ext_len == 0) ||
3674 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3675 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3677 * This will pull the EFI from the AIL and
3678 * free the memory associated with it.
3680 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
3681 xfs_efi_release(efip, efip->efi_format.efi_nextents);
3682 return XFS_ERROR(EIO);
3686 tp = xfs_trans_alloc(mp, 0);
3687 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
3690 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3692 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3693 extp = &(efip->efi_format.efi_extents[i]);
3694 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3697 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3701 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
3702 error = xfs_trans_commit(tp, 0);
3706 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3711 * When this is called, all of the EFIs which did not have
3712 * corresponding EFDs should be in the AIL. What we do now
3713 * is free the extents associated with each one.
3715 * Since we process the EFIs in normal transactions, they
3716 * will be removed at some point after the commit. This prevents
3717 * us from just walking down the list processing each one.
3718 * We'll use a flag in the EFI to skip those that we've already
3719 * processed and use the AIL iteration mechanism's generation
3720 * count to try to speed this up at least a bit.
3722 * When we start, we know that the EFIs are the only things in
3723 * the AIL. As we process them, however, other items are added
3724 * to the AIL. Since everything added to the AIL must come after
3725 * everything already in the AIL, we stop processing as soon as
3726 * we see something other than an EFI in the AIL.
3729 xlog_recover_process_efis(
3732 xfs_log_item_t *lip;
3733 xfs_efi_log_item_t *efip;
3735 struct xfs_ail_cursor cur;
3736 struct xfs_ail *ailp;
3739 spin_lock(&ailp->xa_lock);
3740 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3741 while (lip != NULL) {
3743 * We're done when we see something other than an EFI.
3744 * There should be no EFIs left in the AIL now.
3746 if (lip->li_type != XFS_LI_EFI) {
3748 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
3749 ASSERT(lip->li_type != XFS_LI_EFI);
3755 * Skip EFIs that we've already processed.
3757 efip = (xfs_efi_log_item_t *)lip;
3758 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
3759 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3763 spin_unlock(&ailp->xa_lock);
3764 error = xlog_recover_process_efi(log->l_mp, efip);
3765 spin_lock(&ailp->xa_lock);
3768 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3771 xfs_trans_ail_cursor_done(ailp, &cur);
3772 spin_unlock(&ailp->xa_lock);
3777 * This routine performs a transaction to null out a bad inode pointer
3778 * in an agi unlinked inode hash bucket.
3781 xlog_recover_clear_agi_bucket(
3783 xfs_agnumber_t agno,
3792 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3793 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_clearagi, 0, 0);
3797 error = xfs_read_agi(mp, tp, agno, &agibp);
3801 agi = XFS_BUF_TO_AGI(agibp);
3802 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
3803 offset = offsetof(xfs_agi_t, agi_unlinked) +
3804 (sizeof(xfs_agino_t) * bucket);
3805 xfs_trans_log_buf(tp, agibp, offset,
3806 (offset + sizeof(xfs_agino_t) - 1));
3808 error = xfs_trans_commit(tp, 0);
3814 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3816 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
3821 xlog_recover_process_one_iunlink(
3822 struct xfs_mount *mp,
3823 xfs_agnumber_t agno,
3827 struct xfs_buf *ibp;
3828 struct xfs_dinode *dip;
3829 struct xfs_inode *ip;
3833 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3834 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
3839 * Get the on disk inode to find the next inode in the bucket.
3841 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
3845 ASSERT(ip->i_d.di_nlink == 0);
3846 ASSERT(ip->i_d.di_mode != 0);
3848 /* setup for the next pass */
3849 agino = be32_to_cpu(dip->di_next_unlinked);
3853 * Prevent any DMAPI event from being sent when the reference on
3854 * the inode is dropped.
3856 ip->i_d.di_dmevmask = 0;
3865 * We can't read in the inode this bucket points to, or this inode
3866 * is messed up. Just ditch this bucket of inodes. We will lose
3867 * some inodes and space, but at least we won't hang.
3869 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3870 * clear the inode pointer in the bucket.
3872 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3877 * xlog_iunlink_recover
3879 * This is called during recovery to process any inodes which
3880 * we unlinked but not freed when the system crashed. These
3881 * inodes will be on the lists in the AGI blocks. What we do
3882 * here is scan all the AGIs and fully truncate and free any
3883 * inodes found on the lists. Each inode is removed from the
3884 * lists when it has been fully truncated and is freed. The
3885 * freeing of the inode and its removal from the list must be
3889 xlog_recover_process_iunlinks(
3893 xfs_agnumber_t agno;
3904 * Prevent any DMAPI event from being sent while in this function.
3906 mp_dmevmask = mp->m_dmevmask;
3909 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3911 * Find the agi for this ag.
3913 error = xfs_read_agi(mp, NULL, agno, &agibp);
3916 * AGI is b0rked. Don't process it.
3918 * We should probably mark the filesystem as corrupt
3919 * after we've recovered all the ag's we can....
3924 * Unlock the buffer so that it can be acquired in the normal
3925 * course of the transaction to truncate and free each inode.
3926 * Because we are not racing with anyone else here for the AGI
3927 * buffer, we don't even need to hold it locked to read the
3928 * initial unlinked bucket entries out of the buffer. We keep
3929 * buffer reference though, so that it stays pinned in memory
3930 * while we need the buffer.
3932 agi = XFS_BUF_TO_AGI(agibp);
3933 xfs_buf_unlock(agibp);
3935 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3936 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3937 while (agino != NULLAGINO) {
3938 agino = xlog_recover_process_one_iunlink(mp,
3939 agno, agino, bucket);
3942 xfs_buf_rele(agibp);
3945 mp->m_dmevmask = mp_dmevmask;
3949 * Upack the log buffer data and crc check it. If the check fails, issue a
3950 * warning if and only if the CRC in the header is non-zero. This makes the
3951 * check an advisory warning, and the zero CRC check will prevent failure
3952 * warnings from being emitted when upgrading the kernel from one that does not
3953 * add CRCs by default.
3955 * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log
3956 * corruption failure
3959 xlog_unpack_data_crc(
3960 struct xlog_rec_header *rhead,
3966 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
3967 if (crc != rhead->h_crc) {
3968 if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
3969 xfs_alert(log->l_mp,
3970 "log record CRC mismatch: found 0x%x, expected 0x%x.",
3971 le32_to_cpu(rhead->h_crc),
3973 xfs_hex_dump(dp, 32);
3977 * If we've detected a log record corruption, then we can't
3978 * recover past this point. Abort recovery if we are enforcing
3979 * CRC protection by punting an error back up the stack.
3981 if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
3982 return EFSCORRUPTED;
3990 struct xlog_rec_header *rhead,
3997 error = xlog_unpack_data_crc(rhead, dp, log);
4001 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
4002 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
4003 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
4007 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
4008 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
4009 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
4010 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
4011 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
4012 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
4021 xlog_valid_rec_header(
4023 struct xlog_rec_header *rhead,
4028 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
4029 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
4030 XFS_ERRLEVEL_LOW, log->l_mp);
4031 return XFS_ERROR(EFSCORRUPTED);
4034 (!rhead->h_version ||
4035 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
4036 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
4037 __func__, be32_to_cpu(rhead->h_version));
4038 return XFS_ERROR(EIO);
4041 /* LR body must have data or it wouldn't have been written */
4042 hlen = be32_to_cpu(rhead->h_len);
4043 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
4044 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
4045 XFS_ERRLEVEL_LOW, log->l_mp);
4046 return XFS_ERROR(EFSCORRUPTED);
4048 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
4049 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
4050 XFS_ERRLEVEL_LOW, log->l_mp);
4051 return XFS_ERROR(EFSCORRUPTED);
4057 * Read the log from tail to head and process the log records found.
4058 * Handle the two cases where the tail and head are in the same cycle
4059 * and where the active portion of the log wraps around the end of
4060 * the physical log separately. The pass parameter is passed through
4061 * to the routines called to process the data and is not looked at
4065 xlog_do_recovery_pass(
4067 xfs_daddr_t head_blk,
4068 xfs_daddr_t tail_blk,
4071 xlog_rec_header_t *rhead;
4074 xfs_buf_t *hbp, *dbp;
4075 int error = 0, h_size;
4076 int bblks, split_bblks;
4077 int hblks, split_hblks, wrapped_hblks;
4078 struct hlist_head rhash[XLOG_RHASH_SIZE];
4080 ASSERT(head_blk != tail_blk);
4083 * Read the header of the tail block and get the iclog buffer size from
4084 * h_size. Use this to tell how many sectors make up the log header.
4086 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
4088 * When using variable length iclogs, read first sector of
4089 * iclog header and extract the header size from it. Get a
4090 * new hbp that is the correct size.
4092 hbp = xlog_get_bp(log, 1);
4096 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
4100 rhead = (xlog_rec_header_t *)offset;
4101 error = xlog_valid_rec_header(log, rhead, tail_blk);
4104 h_size = be32_to_cpu(rhead->h_size);
4105 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
4106 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
4107 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
4108 if (h_size % XLOG_HEADER_CYCLE_SIZE)
4111 hbp = xlog_get_bp(log, hblks);
4116 ASSERT(log->l_sectBBsize == 1);
4118 hbp = xlog_get_bp(log, 1);
4119 h_size = XLOG_BIG_RECORD_BSIZE;
4124 dbp = xlog_get_bp(log, BTOBB(h_size));
4130 memset(rhash, 0, sizeof(rhash));
4131 if (tail_blk <= head_blk) {
4132 for (blk_no = tail_blk; blk_no < head_blk; ) {
4133 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
4137 rhead = (xlog_rec_header_t *)offset;
4138 error = xlog_valid_rec_header(log, rhead, blk_no);
4142 /* blocks in data section */
4143 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4144 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
4149 error = xlog_unpack_data(rhead, offset, log);
4153 error = xlog_recover_process_data(log,
4154 rhash, rhead, offset, pass);
4157 blk_no += bblks + hblks;
4161 * Perform recovery around the end of the physical log.
4162 * When the head is not on the same cycle number as the tail,
4163 * we can't do a sequential recovery as above.
4166 while (blk_no < log->l_logBBsize) {
4168 * Check for header wrapping around physical end-of-log
4170 offset = hbp->b_addr;
4173 if (blk_no + hblks <= log->l_logBBsize) {
4174 /* Read header in one read */
4175 error = xlog_bread(log, blk_no, hblks, hbp,
4180 /* This LR is split across physical log end */
4181 if (blk_no != log->l_logBBsize) {
4182 /* some data before physical log end */
4183 ASSERT(blk_no <= INT_MAX);
4184 split_hblks = log->l_logBBsize - (int)blk_no;
4185 ASSERT(split_hblks > 0);
4186 error = xlog_bread(log, blk_no,
4194 * Note: this black magic still works with
4195 * large sector sizes (non-512) only because:
4196 * - we increased the buffer size originally
4197 * by 1 sector giving us enough extra space
4198 * for the second read;
4199 * - the log start is guaranteed to be sector
4201 * - we read the log end (LR header start)
4202 * _first_, then the log start (LR header end)
4203 * - order is important.
4205 wrapped_hblks = hblks - split_hblks;
4206 error = xlog_bread_offset(log, 0,
4208 offset + BBTOB(split_hblks));
4212 rhead = (xlog_rec_header_t *)offset;
4213 error = xlog_valid_rec_header(log, rhead,
4214 split_hblks ? blk_no : 0);
4218 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4221 /* Read in data for log record */
4222 if (blk_no + bblks <= log->l_logBBsize) {
4223 error = xlog_bread(log, blk_no, bblks, dbp,
4228 /* This log record is split across the
4229 * physical end of log */
4230 offset = dbp->b_addr;
4232 if (blk_no != log->l_logBBsize) {
4233 /* some data is before the physical
4235 ASSERT(!wrapped_hblks);
4236 ASSERT(blk_no <= INT_MAX);
4238 log->l_logBBsize - (int)blk_no;
4239 ASSERT(split_bblks > 0);
4240 error = xlog_bread(log, blk_no,
4248 * Note: this black magic still works with
4249 * large sector sizes (non-512) only because:
4250 * - we increased the buffer size originally
4251 * by 1 sector giving us enough extra space
4252 * for the second read;
4253 * - the log start is guaranteed to be sector
4255 * - we read the log end (LR header start)
4256 * _first_, then the log start (LR header end)
4257 * - order is important.
4259 error = xlog_bread_offset(log, 0,
4260 bblks - split_bblks, dbp,
4261 offset + BBTOB(split_bblks));
4266 error = xlog_unpack_data(rhead, offset, log);
4270 error = xlog_recover_process_data(log, rhash,
4271 rhead, offset, pass);
4277 ASSERT(blk_no >= log->l_logBBsize);
4278 blk_no -= log->l_logBBsize;
4280 /* read first part of physical log */
4281 while (blk_no < head_blk) {
4282 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
4286 rhead = (xlog_rec_header_t *)offset;
4287 error = xlog_valid_rec_header(log, rhead, blk_no);
4291 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4292 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
4297 error = xlog_unpack_data(rhead, offset, log);
4301 error = xlog_recover_process_data(log, rhash,
4302 rhead, offset, pass);
4305 blk_no += bblks + hblks;
4317 * Do the recovery of the log. We actually do this in two phases.
4318 * The two passes are necessary in order to implement the function
4319 * of cancelling a record written into the log. The first pass
4320 * determines those things which have been cancelled, and the
4321 * second pass replays log items normally except for those which
4322 * have been cancelled. The handling of the replay and cancellations
4323 * takes place in the log item type specific routines.
4325 * The table of items which have cancel records in the log is allocated
4326 * and freed at this level, since only here do we know when all of
4327 * the log recovery has been completed.
4330 xlog_do_log_recovery(
4332 xfs_daddr_t head_blk,
4333 xfs_daddr_t tail_blk)
4337 ASSERT(head_blk != tail_blk);
4340 * First do a pass to find all of the cancelled buf log items.
4341 * Store them in the buf_cancel_table for use in the second pass.
4343 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
4344 sizeof(struct list_head),
4346 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4347 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
4349 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4350 XLOG_RECOVER_PASS1);
4352 kmem_free(log->l_buf_cancel_table);
4353 log->l_buf_cancel_table = NULL;
4357 * Then do a second pass to actually recover the items in the log.
4358 * When it is complete free the table of buf cancel items.
4360 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4361 XLOG_RECOVER_PASS2);
4366 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4367 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
4371 kmem_free(log->l_buf_cancel_table);
4372 log->l_buf_cancel_table = NULL;
4378 * Do the actual recovery
4383 xfs_daddr_t head_blk,
4384 xfs_daddr_t tail_blk)
4391 * First replay the images in the log.
4393 error = xlog_do_log_recovery(log, head_blk, tail_blk);
4398 * If IO errors happened during recovery, bail out.
4400 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
4405 * We now update the tail_lsn since much of the recovery has completed
4406 * and there may be space available to use. If there were no extent
4407 * or iunlinks, we can free up the entire log and set the tail_lsn to
4408 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4409 * lsn of the last known good LR on disk. If there are extent frees
4410 * or iunlinks they will have some entries in the AIL; so we look at
4411 * the AIL to determine how to set the tail_lsn.
4413 xlog_assign_tail_lsn(log->l_mp);
4416 * Now that we've finished replaying all buffer and inode
4417 * updates, re-read in the superblock and reverify it.
4419 bp = xfs_getsb(log->l_mp, 0);
4421 ASSERT(!(XFS_BUF_ISWRITE(bp)));
4423 XFS_BUF_UNASYNC(bp);
4424 bp->b_ops = &xfs_sb_buf_ops;
4426 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
4428 return XFS_ERROR(EIO);
4431 xfs_buf_iorequest(bp);
4432 error = xfs_buf_iowait(bp);
4434 xfs_buf_ioerror_alert(bp, __func__);
4440 /* Convert superblock from on-disk format */
4441 sbp = &log->l_mp->m_sb;
4442 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
4443 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
4444 ASSERT(xfs_sb_good_version(sbp));
4447 /* We've re-read the superblock so re-initialize per-cpu counters */
4448 xfs_icsb_reinit_counters(log->l_mp);
4450 xlog_recover_check_summary(log);
4452 /* Normal transactions can now occur */
4453 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
4458 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4460 * Return error or zero.
4466 xfs_daddr_t head_blk, tail_blk;
4469 /* find the tail of the log */
4470 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
4473 if (tail_blk != head_blk) {
4474 /* There used to be a comment here:
4476 * disallow recovery on read-only mounts. note -- mount
4477 * checks for ENOSPC and turns it into an intelligent
4479 * ...but this is no longer true. Now, unless you specify
4480 * NORECOVERY (in which case this function would never be
4481 * called), we just go ahead and recover. We do this all
4482 * under the vfs layer, so we can get away with it unless
4483 * the device itself is read-only, in which case we fail.
4485 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
4490 * Version 5 superblock log feature mask validation. We know the
4491 * log is dirty so check if there are any unknown log features
4492 * in what we need to recover. If there are unknown features
4493 * (e.g. unsupported transactions, then simply reject the
4494 * attempt at recovery before touching anything.
4496 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
4497 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
4498 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
4500 "Superblock has unknown incompatible log features (0x%x) enabled.\n"
4501 "The log can not be fully and/or safely recovered by this kernel.\n"
4502 "Please recover the log on a kernel that supports the unknown features.",
4503 (log->l_mp->m_sb.sb_features_log_incompat &
4504 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
4508 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
4509 log->l_mp->m_logname ? log->l_mp->m_logname
4512 error = xlog_do_recover(log, head_blk, tail_blk);
4513 log->l_flags |= XLOG_RECOVERY_NEEDED;
4519 * In the first part of recovery we replay inodes and buffers and build
4520 * up the list of extent free items which need to be processed. Here
4521 * we process the extent free items and clean up the on disk unlinked
4522 * inode lists. This is separated from the first part of recovery so
4523 * that the root and real-time bitmap inodes can be read in from disk in
4524 * between the two stages. This is necessary so that we can free space
4525 * in the real-time portion of the file system.
4528 xlog_recover_finish(
4532 * Now we're ready to do the transactions needed for the
4533 * rest of recovery. Start with completing all the extent
4534 * free intent records and then process the unlinked inode
4535 * lists. At this point, we essentially run in normal mode
4536 * except that we're still performing recovery actions
4537 * rather than accepting new requests.
4539 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
4541 error = xlog_recover_process_efis(log);
4543 xfs_alert(log->l_mp, "Failed to recover EFIs");
4547 * Sync the log to get all the EFIs out of the AIL.
4548 * This isn't absolutely necessary, but it helps in
4549 * case the unlink transactions would have problems
4550 * pushing the EFIs out of the way.
4552 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
4554 xlog_recover_process_iunlinks(log);
4556 xlog_recover_check_summary(log);
4558 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
4559 log->l_mp->m_logname ? log->l_mp->m_logname
4561 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
4563 xfs_info(log->l_mp, "Ending clean mount");
4571 * Read all of the agf and agi counters and check that they
4572 * are consistent with the superblock counters.
4575 xlog_recover_check_summary(
4582 xfs_agnumber_t agno;
4583 __uint64_t freeblks;
4593 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
4594 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
4596 xfs_alert(mp, "%s agf read failed agno %d error %d",
4597 __func__, agno, error);
4599 agfp = XFS_BUF_TO_AGF(agfbp);
4600 freeblks += be32_to_cpu(agfp->agf_freeblks) +
4601 be32_to_cpu(agfp->agf_flcount);
4602 xfs_buf_relse(agfbp);
4605 error = xfs_read_agi(mp, NULL, agno, &agibp);
4607 xfs_alert(mp, "%s agi read failed agno %d error %d",
4608 __func__, agno, error);
4610 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
4612 itotal += be32_to_cpu(agi->agi_count);
4613 ifree += be32_to_cpu(agi->agi_freecount);
4614 xfs_buf_relse(agibp);