1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_inode.h"
17 #include "xfs_trans.h"
19 #include "xfs_log_priv.h"
20 #include "xfs_log_recover.h"
21 #include "xfs_inode_item.h"
22 #include "xfs_extfree_item.h"
23 #include "xfs_trans_priv.h"
24 #include "xfs_alloc.h"
25 #include "xfs_ialloc.h"
26 #include "xfs_quota.h"
27 #include "xfs_trace.h"
28 #include "xfs_icache.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_error.h"
32 #include "xfs_rmap_item.h"
33 #include "xfs_buf_item.h"
34 #include "xfs_refcount_item.h"
35 #include "xfs_bmap_item.h"
37 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
44 xlog_clear_stale_blocks(
49 xlog_recover_check_summary(
52 #define xlog_recover_check_summary(log)
55 xlog_do_recovery_pass(
56 struct xlog *, xfs_daddr_t, xfs_daddr_t, int, xfs_daddr_t *);
59 * This structure is used during recovery to record the buf log items which
60 * have been canceled and should not be replayed.
62 struct xfs_buf_cancel {
66 struct list_head bc_list;
70 * Sector aligned buffer routines for buffer create/read/write/access
74 * Verify the log-relative block number and length in basic blocks are valid for
75 * an operation involving the given XFS log buffer. Returns true if the fields
76 * are valid, false otherwise.
84 if (blk_no < 0 || blk_no >= log->l_logBBsize)
86 if (bbcount <= 0 || (blk_no + bbcount) > log->l_logBBsize)
92 * Allocate a buffer to hold log data. The buffer needs to be able to map to
93 * a range of nbblks basic blocks at any valid offset within the log.
100 int align_mask = xfs_buftarg_dma_alignment(log->l_targ);
103 * Pass log block 0 since we don't have an addr yet, buffer will be
106 if (!xlog_verify_bno(log, 0, nbblks)) {
107 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
109 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
114 * We do log I/O in units of log sectors (a power-of-2 multiple of the
115 * basic block size), so we round up the requested size to accommodate
116 * the basic blocks required for complete log sectors.
118 * In addition, the buffer may be used for a non-sector-aligned block
119 * offset, in which case an I/O of the requested size could extend
120 * beyond the end of the buffer. If the requested size is only 1 basic
121 * block it will never straddle a sector boundary, so this won't be an
122 * issue. Nor will this be a problem if the log I/O is done in basic
123 * blocks (sector size 1). But otherwise we extend the buffer by one
124 * extra log sector to ensure there's space to accommodate this
127 if (nbblks > 1 && log->l_sectBBsize > 1)
128 nbblks += log->l_sectBBsize;
129 nbblks = round_up(nbblks, log->l_sectBBsize);
130 return kmem_alloc_io(BBTOB(nbblks), align_mask, KM_MAYFAIL | KM_ZERO);
134 * Return the address of the start of the given block number's data
135 * in a log buffer. The buffer covers a log sector-aligned region.
137 static inline unsigned int
142 return BBTOB(blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1));
155 if (!xlog_verify_bno(log, blk_no, nbblks)) {
157 "Invalid log block/length (0x%llx, 0x%x) for buffer",
159 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
160 return -EFSCORRUPTED;
163 blk_no = round_down(blk_no, log->l_sectBBsize);
164 nbblks = round_up(nbblks, log->l_sectBBsize);
167 error = xfs_rw_bdev(log->l_targ->bt_bdev, log->l_logBBstart + blk_no,
168 BBTOB(nbblks), data, op);
169 if (error && !XFS_FORCED_SHUTDOWN(log->l_mp)) {
171 "log recovery %s I/O error at daddr 0x%llx len %d error %d",
172 op == REQ_OP_WRITE ? "write" : "read",
173 blk_no, nbblks, error);
185 return xlog_do_io(log, blk_no, nbblks, data, REQ_OP_READ);
198 error = xlog_do_io(log, blk_no, nbblks, data, REQ_OP_READ);
200 *offset = data + xlog_align(log, blk_no);
211 return xlog_do_io(log, blk_no, nbblks, data, REQ_OP_WRITE);
216 * dump debug superblock and log record information
219 xlog_header_check_dump(
221 xlog_rec_header_t *head)
223 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d",
224 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
225 xfs_debug(mp, " log : uuid = %pU, fmt = %d",
226 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
229 #define xlog_header_check_dump(mp, head)
233 * check log record header for recovery
236 xlog_header_check_recover(
238 xlog_rec_header_t *head)
240 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
243 * IRIX doesn't write the h_fmt field and leaves it zeroed
244 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
245 * a dirty log created in IRIX.
247 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
249 "dirty log written in incompatible format - can't recover");
250 xlog_header_check_dump(mp, head);
251 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
252 XFS_ERRLEVEL_HIGH, mp);
253 return -EFSCORRUPTED;
254 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
256 "dirty log entry has mismatched uuid - can't recover");
257 xlog_header_check_dump(mp, head);
258 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
259 XFS_ERRLEVEL_HIGH, mp);
260 return -EFSCORRUPTED;
266 * read the head block of the log and check the header
269 xlog_header_check_mount(
271 xlog_rec_header_t *head)
273 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
275 if (uuid_is_null(&head->h_fs_uuid)) {
277 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
278 * h_fs_uuid is null, we assume this log was last mounted
279 * by IRIX and continue.
281 xfs_warn(mp, "null uuid in log - IRIX style log");
282 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
283 xfs_warn(mp, "log has mismatched uuid - can't recover");
284 xlog_header_check_dump(mp, head);
285 XFS_ERROR_REPORT("xlog_header_check_mount",
286 XFS_ERRLEVEL_HIGH, mp);
287 return -EFSCORRUPTED;
298 * We're not going to bother about retrying
299 * this during recovery. One strike!
301 if (!XFS_FORCED_SHUTDOWN(bp->b_mount)) {
302 xfs_buf_ioerror_alert(bp, __func__);
303 xfs_force_shutdown(bp->b_mount, SHUTDOWN_META_IO_ERROR);
308 * On v5 supers, a bli could be attached to update the metadata LSN.
312 xfs_buf_item_relse(bp);
313 ASSERT(bp->b_log_item == NULL);
320 * This routine finds (to an approximation) the first block in the physical
321 * log which contains the given cycle. It uses a binary search algorithm.
322 * Note that the algorithm can not be perfect because the disk will not
323 * necessarily be perfect.
326 xlog_find_cycle_start(
329 xfs_daddr_t first_blk,
330 xfs_daddr_t *last_blk,
340 mid_blk = BLK_AVG(first_blk, end_blk);
341 while (mid_blk != first_blk && mid_blk != end_blk) {
342 error = xlog_bread(log, mid_blk, 1, buffer, &offset);
345 mid_cycle = xlog_get_cycle(offset);
346 if (mid_cycle == cycle)
347 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
349 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
350 mid_blk = BLK_AVG(first_blk, end_blk);
352 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
353 (mid_blk == end_blk && mid_blk-1 == first_blk));
361 * Check that a range of blocks does not contain stop_on_cycle_no.
362 * Fill in *new_blk with the block offset where such a block is
363 * found, or with -1 (an invalid block number) if there is no such
364 * block in the range. The scan needs to occur from front to back
365 * and the pointer into the region must be updated since a later
366 * routine will need to perform another test.
369 xlog_find_verify_cycle(
371 xfs_daddr_t start_blk,
373 uint stop_on_cycle_no,
374 xfs_daddr_t *new_blk)
384 * Greedily allocate a buffer big enough to handle the full
385 * range of basic blocks we'll be examining. If that fails,
386 * try a smaller size. We need to be able to read at least
387 * a log sector, or we're out of luck.
389 bufblks = 1 << ffs(nbblks);
390 while (bufblks > log->l_logBBsize)
392 while (!(buffer = xlog_alloc_buffer(log, bufblks))) {
394 if (bufblks < log->l_sectBBsize)
398 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
401 bcount = min(bufblks, (start_blk + nbblks - i));
403 error = xlog_bread(log, i, bcount, buffer, &buf);
407 for (j = 0; j < bcount; j++) {
408 cycle = xlog_get_cycle(buf);
409 if (cycle == stop_on_cycle_no) {
426 * Potentially backup over partial log record write.
428 * In the typical case, last_blk is the number of the block directly after
429 * a good log record. Therefore, we subtract one to get the block number
430 * of the last block in the given buffer. extra_bblks contains the number
431 * of blocks we would have read on a previous read. This happens when the
432 * last log record is split over the end of the physical log.
434 * extra_bblks is the number of blocks potentially verified on a previous
435 * call to this routine.
438 xlog_find_verify_log_record(
440 xfs_daddr_t start_blk,
441 xfs_daddr_t *last_blk,
447 xlog_rec_header_t *head = NULL;
450 int num_blks = *last_blk - start_blk;
453 ASSERT(start_blk != 0 || *last_blk != start_blk);
455 buffer = xlog_alloc_buffer(log, num_blks);
457 buffer = xlog_alloc_buffer(log, 1);
462 error = xlog_bread(log, start_blk, num_blks, buffer, &offset);
465 offset += ((num_blks - 1) << BBSHIFT);
468 for (i = (*last_blk) - 1; i >= 0; i--) {
470 /* valid log record not found */
472 "Log inconsistent (didn't find previous header)");
474 error = -EFSCORRUPTED;
479 error = xlog_bread(log, i, 1, buffer, &offset);
484 head = (xlog_rec_header_t *)offset;
486 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
494 * We hit the beginning of the physical log & still no header. Return
495 * to caller. If caller can handle a return of -1, then this routine
496 * will be called again for the end of the physical log.
504 * We have the final block of the good log (the first block
505 * of the log record _before_ the head. So we check the uuid.
507 if ((error = xlog_header_check_mount(log->l_mp, head)))
511 * We may have found a log record header before we expected one.
512 * last_blk will be the 1st block # with a given cycle #. We may end
513 * up reading an entire log record. In this case, we don't want to
514 * reset last_blk. Only when last_blk points in the middle of a log
515 * record do we update last_blk.
517 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
518 uint h_size = be32_to_cpu(head->h_size);
520 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
521 if (h_size % XLOG_HEADER_CYCLE_SIZE)
527 if (*last_blk - i + extra_bblks !=
528 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
537 * Head is defined to be the point of the log where the next log write
538 * could go. This means that incomplete LR writes at the end are
539 * eliminated when calculating the head. We aren't guaranteed that previous
540 * LR have complete transactions. We only know that a cycle number of
541 * current cycle number -1 won't be present in the log if we start writing
542 * from our current block number.
544 * last_blk contains the block number of the first block with a given
547 * Return: zero if normal, non-zero if error.
552 xfs_daddr_t *return_head_blk)
556 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
558 uint first_half_cycle, last_half_cycle;
560 int error, log_bbnum = log->l_logBBsize;
562 /* Is the end of the log device zeroed? */
563 error = xlog_find_zeroed(log, &first_blk);
565 xfs_warn(log->l_mp, "empty log check failed");
569 *return_head_blk = first_blk;
571 /* Is the whole lot zeroed? */
573 /* Linux XFS shouldn't generate totally zeroed logs -
574 * mkfs etc write a dummy unmount record to a fresh
575 * log so we can store the uuid in there
577 xfs_warn(log->l_mp, "totally zeroed log");
583 first_blk = 0; /* get cycle # of 1st block */
584 buffer = xlog_alloc_buffer(log, 1);
588 error = xlog_bread(log, 0, 1, buffer, &offset);
590 goto out_free_buffer;
592 first_half_cycle = xlog_get_cycle(offset);
594 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
595 error = xlog_bread(log, last_blk, 1, buffer, &offset);
597 goto out_free_buffer;
599 last_half_cycle = xlog_get_cycle(offset);
600 ASSERT(last_half_cycle != 0);
603 * If the 1st half cycle number is equal to the last half cycle number,
604 * then the entire log is stamped with the same cycle number. In this
605 * case, head_blk can't be set to zero (which makes sense). The below
606 * math doesn't work out properly with head_blk equal to zero. Instead,
607 * we set it to log_bbnum which is an invalid block number, but this
608 * value makes the math correct. If head_blk doesn't changed through
609 * all the tests below, *head_blk is set to zero at the very end rather
610 * than log_bbnum. In a sense, log_bbnum and zero are the same block
611 * in a circular file.
613 if (first_half_cycle == last_half_cycle) {
615 * In this case we believe that the entire log should have
616 * cycle number last_half_cycle. We need to scan backwards
617 * from the end verifying that there are no holes still
618 * containing last_half_cycle - 1. If we find such a hole,
619 * then the start of that hole will be the new head. The
620 * simple case looks like
621 * x | x ... | x - 1 | x
622 * Another case that fits this picture would be
623 * x | x + 1 | x ... | x
624 * In this case the head really is somewhere at the end of the
625 * log, as one of the latest writes at the beginning was
628 * x | x + 1 | x ... | x - 1 | x
629 * This is really the combination of the above two cases, and
630 * the head has to end up at the start of the x-1 hole at the
633 * In the 256k log case, we will read from the beginning to the
634 * end of the log and search for cycle numbers equal to x-1.
635 * We don't worry about the x+1 blocks that we encounter,
636 * because we know that they cannot be the head since the log
639 head_blk = log_bbnum;
640 stop_on_cycle = last_half_cycle - 1;
643 * In this case we want to find the first block with cycle
644 * number matching last_half_cycle. We expect the log to be
646 * x + 1 ... | x ... | x
647 * The first block with cycle number x (last_half_cycle) will
648 * be where the new head belongs. First we do a binary search
649 * for the first occurrence of last_half_cycle. The binary
650 * search may not be totally accurate, so then we scan back
651 * from there looking for occurrences of last_half_cycle before
652 * us. If that backwards scan wraps around the beginning of
653 * the log, then we look for occurrences of last_half_cycle - 1
654 * at the end of the log. The cases we're looking for look
656 * v binary search stopped here
657 * x + 1 ... | x | x + 1 | x ... | x
658 * ^ but we want to locate this spot
660 * <---------> less than scan distance
661 * x + 1 ... | x ... | x - 1 | x
662 * ^ we want to locate this spot
664 stop_on_cycle = last_half_cycle;
665 error = xlog_find_cycle_start(log, buffer, first_blk, &head_blk,
668 goto out_free_buffer;
672 * Now validate the answer. Scan back some number of maximum possible
673 * blocks and make sure each one has the expected cycle number. The
674 * maximum is determined by the total possible amount of buffering
675 * in the in-core log. The following number can be made tighter if
676 * we actually look at the block size of the filesystem.
678 num_scan_bblks = min_t(int, log_bbnum, XLOG_TOTAL_REC_SHIFT(log));
679 if (head_blk >= num_scan_bblks) {
681 * We are guaranteed that the entire check can be performed
684 start_blk = head_blk - num_scan_bblks;
685 if ((error = xlog_find_verify_cycle(log,
686 start_blk, num_scan_bblks,
687 stop_on_cycle, &new_blk)))
688 goto out_free_buffer;
691 } else { /* need to read 2 parts of log */
693 * We are going to scan backwards in the log in two parts.
694 * First we scan the physical end of the log. In this part
695 * of the log, we are looking for blocks with cycle number
696 * last_half_cycle - 1.
697 * If we find one, then we know that the log starts there, as
698 * we've found a hole that didn't get written in going around
699 * the end of the physical log. The simple case for this is
700 * x + 1 ... | x ... | x - 1 | x
701 * <---------> less than scan distance
702 * If all of the blocks at the end of the log have cycle number
703 * last_half_cycle, then we check the blocks at the start of
704 * the log looking for occurrences of last_half_cycle. If we
705 * find one, then our current estimate for the location of the
706 * first occurrence of last_half_cycle is wrong and we move
707 * back to the hole we've found. This case looks like
708 * x + 1 ... | x | x + 1 | x ...
709 * ^ binary search stopped here
710 * Another case we need to handle that only occurs in 256k
712 * x + 1 ... | x ... | x+1 | x ...
713 * ^ binary search stops here
714 * In a 256k log, the scan at the end of the log will see the
715 * x + 1 blocks. We need to skip past those since that is
716 * certainly not the head of the log. By searching for
717 * last_half_cycle-1 we accomplish that.
719 ASSERT(head_blk <= INT_MAX &&
720 (xfs_daddr_t) num_scan_bblks >= head_blk);
721 start_blk = log_bbnum - (num_scan_bblks - head_blk);
722 if ((error = xlog_find_verify_cycle(log, start_blk,
723 num_scan_bblks - (int)head_blk,
724 (stop_on_cycle - 1), &new_blk)))
725 goto out_free_buffer;
732 * Scan beginning of log now. The last part of the physical
733 * log is good. This scan needs to verify that it doesn't find
734 * the last_half_cycle.
737 ASSERT(head_blk <= INT_MAX);
738 if ((error = xlog_find_verify_cycle(log,
739 start_blk, (int)head_blk,
740 stop_on_cycle, &new_blk)))
741 goto out_free_buffer;
748 * Now we need to make sure head_blk is not pointing to a block in
749 * the middle of a log record.
751 num_scan_bblks = XLOG_REC_SHIFT(log);
752 if (head_blk >= num_scan_bblks) {
753 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
755 /* start ptr at last block ptr before head_blk */
756 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
760 goto out_free_buffer;
763 ASSERT(head_blk <= INT_MAX);
764 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
766 goto out_free_buffer;
768 /* We hit the beginning of the log during our search */
769 start_blk = log_bbnum - (num_scan_bblks - head_blk);
771 ASSERT(start_blk <= INT_MAX &&
772 (xfs_daddr_t) log_bbnum-start_blk >= 0);
773 ASSERT(head_blk <= INT_MAX);
774 error = xlog_find_verify_log_record(log, start_blk,
775 &new_blk, (int)head_blk);
779 goto out_free_buffer;
780 if (new_blk != log_bbnum)
783 goto out_free_buffer;
787 if (head_blk == log_bbnum)
788 *return_head_blk = 0;
790 *return_head_blk = head_blk;
792 * When returning here, we have a good block number. Bad block
793 * means that during a previous crash, we didn't have a clean break
794 * from cycle number N to cycle number N-1. In this case, we need
795 * to find the first block with cycle number N-1.
802 xfs_warn(log->l_mp, "failed to find log head");
807 * Seek backwards in the log for log record headers.
809 * Given a starting log block, walk backwards until we find the provided number
810 * of records or hit the provided tail block. The return value is the number of
811 * records encountered or a negative error code. The log block and buffer
812 * pointer of the last record seen are returned in rblk and rhead respectively.
815 xlog_rseek_logrec_hdr(
817 xfs_daddr_t head_blk,
818 xfs_daddr_t tail_blk,
822 struct xlog_rec_header **rhead,
834 * Walk backwards from the head block until we hit the tail or the first
837 end_blk = head_blk > tail_blk ? tail_blk : 0;
838 for (i = (int) head_blk - 1; i >= end_blk; i--) {
839 error = xlog_bread(log, i, 1, buffer, &offset);
843 if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
845 *rhead = (struct xlog_rec_header *) offset;
846 if (++found == count)
852 * If we haven't hit the tail block or the log record header count,
853 * start looking again from the end of the physical log. Note that
854 * callers can pass head == tail if the tail is not yet known.
856 if (tail_blk >= head_blk && found != count) {
857 for (i = log->l_logBBsize - 1; i >= (int) tail_blk; i--) {
858 error = xlog_bread(log, i, 1, buffer, &offset);
862 if (*(__be32 *)offset ==
863 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
866 *rhead = (struct xlog_rec_header *) offset;
867 if (++found == count)
880 * Seek forward in the log for log record headers.
882 * Given head and tail blocks, walk forward from the tail block until we find
883 * the provided number of records or hit the head block. The return value is the
884 * number of records encountered or a negative error code. The log block and
885 * buffer pointer of the last record seen are returned in rblk and rhead
889 xlog_seek_logrec_hdr(
891 xfs_daddr_t head_blk,
892 xfs_daddr_t tail_blk,
896 struct xlog_rec_header **rhead,
908 * Walk forward from the tail block until we hit the head or the last
911 end_blk = head_blk > tail_blk ? head_blk : log->l_logBBsize - 1;
912 for (i = (int) tail_blk; i <= end_blk; i++) {
913 error = xlog_bread(log, i, 1, buffer, &offset);
917 if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
919 *rhead = (struct xlog_rec_header *) offset;
920 if (++found == count)
926 * If we haven't hit the head block or the log record header count,
927 * start looking again from the start of the physical log.
929 if (tail_blk > head_blk && found != count) {
930 for (i = 0; i < (int) head_blk; i++) {
931 error = xlog_bread(log, i, 1, buffer, &offset);
935 if (*(__be32 *)offset ==
936 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
939 *rhead = (struct xlog_rec_header *) offset;
940 if (++found == count)
953 * Calculate distance from head to tail (i.e., unused space in the log).
958 xfs_daddr_t head_blk,
959 xfs_daddr_t tail_blk)
961 if (head_blk < tail_blk)
962 return tail_blk - head_blk;
964 return tail_blk + (log->l_logBBsize - head_blk);
968 * Verify the log tail. This is particularly important when torn or incomplete
969 * writes have been detected near the front of the log and the head has been
970 * walked back accordingly.
972 * We also have to handle the case where the tail was pinned and the head
973 * blocked behind the tail right before a crash. If the tail had been pushed
974 * immediately prior to the crash and the subsequent checkpoint was only
975 * partially written, it's possible it overwrote the last referenced tail in the
976 * log with garbage. This is not a coherency problem because the tail must have
977 * been pushed before it can be overwritten, but appears as log corruption to
978 * recovery because we have no way to know the tail was updated if the
979 * subsequent checkpoint didn't write successfully.
981 * Therefore, CRC check the log from tail to head. If a failure occurs and the
982 * offending record is within max iclog bufs from the head, walk the tail
983 * forward and retry until a valid tail is found or corruption is detected out
984 * of the range of a possible overwrite.
989 xfs_daddr_t head_blk,
990 xfs_daddr_t *tail_blk,
993 struct xlog_rec_header *thead;
995 xfs_daddr_t first_bad;
998 xfs_daddr_t tmp_tail;
999 xfs_daddr_t orig_tail = *tail_blk;
1001 buffer = xlog_alloc_buffer(log, 1);
1006 * Make sure the tail points to a record (returns positive count on
1009 error = xlog_seek_logrec_hdr(log, head_blk, *tail_blk, 1, buffer,
1010 &tmp_tail, &thead, &wrapped);
1013 if (*tail_blk != tmp_tail)
1014 *tail_blk = tmp_tail;
1017 * Run a CRC check from the tail to the head. We can't just check
1018 * MAX_ICLOGS records past the tail because the tail may point to stale
1019 * blocks cleared during the search for the head/tail. These blocks are
1020 * overwritten with zero-length records and thus record count is not a
1021 * reliable indicator of the iclog state before a crash.
1024 error = xlog_do_recovery_pass(log, head_blk, *tail_blk,
1025 XLOG_RECOVER_CRCPASS, &first_bad);
1026 while ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) {
1030 * Is corruption within range of the head? If so, retry from
1031 * the next record. Otherwise return an error.
1033 tail_distance = xlog_tail_distance(log, head_blk, first_bad);
1034 if (tail_distance > BTOBB(XLOG_MAX_ICLOGS * hsize))
1037 /* skip to the next record; returns positive count on success */
1038 error = xlog_seek_logrec_hdr(log, head_blk, first_bad, 2,
1039 buffer, &tmp_tail, &thead, &wrapped);
1043 *tail_blk = tmp_tail;
1045 error = xlog_do_recovery_pass(log, head_blk, *tail_blk,
1046 XLOG_RECOVER_CRCPASS, &first_bad);
1049 if (!error && *tail_blk != orig_tail)
1051 "Tail block (0x%llx) overwrite detected. Updated to 0x%llx",
1052 orig_tail, *tail_blk);
1059 * Detect and trim torn writes from the head of the log.
1061 * Storage without sector atomicity guarantees can result in torn writes in the
1062 * log in the event of a crash. Our only means to detect this scenario is via
1063 * CRC verification. While we can't always be certain that CRC verification
1064 * failure is due to a torn write vs. an unrelated corruption, we do know that
1065 * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at
1066 * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of
1067 * the log and treat failures in this range as torn writes as a matter of
1068 * policy. In the event of CRC failure, the head is walked back to the last good
1069 * record in the log and the tail is updated from that record and verified.
1074 xfs_daddr_t *head_blk, /* in/out: unverified head */
1075 xfs_daddr_t *tail_blk, /* out: tail block */
1077 xfs_daddr_t *rhead_blk, /* start blk of last record */
1078 struct xlog_rec_header **rhead, /* ptr to last record */
1079 bool *wrapped) /* last rec. wraps phys. log */
1081 struct xlog_rec_header *tmp_rhead;
1083 xfs_daddr_t first_bad;
1084 xfs_daddr_t tmp_rhead_blk;
1090 * Check the head of the log for torn writes. Search backwards from the
1091 * head until we hit the tail or the maximum number of log record I/Os
1092 * that could have been in flight at one time. Use a temporary buffer so
1093 * we don't trash the rhead/buffer pointers from the caller.
1095 tmp_buffer = xlog_alloc_buffer(log, 1);
1098 error = xlog_rseek_logrec_hdr(log, *head_blk, *tail_blk,
1099 XLOG_MAX_ICLOGS, tmp_buffer,
1100 &tmp_rhead_blk, &tmp_rhead, &tmp_wrapped);
1101 kmem_free(tmp_buffer);
1106 * Now run a CRC verification pass over the records starting at the
1107 * block found above to the current head. If a CRC failure occurs, the
1108 * log block of the first bad record is saved in first_bad.
1110 error = xlog_do_recovery_pass(log, *head_blk, tmp_rhead_blk,
1111 XLOG_RECOVER_CRCPASS, &first_bad);
1112 if ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) {
1114 * We've hit a potential torn write. Reset the error and warn
1119 "Torn write (CRC failure) detected at log block 0x%llx. Truncating head block from 0x%llx.",
1120 first_bad, *head_blk);
1123 * Get the header block and buffer pointer for the last good
1124 * record before the bad record.
1126 * Note that xlog_find_tail() clears the blocks at the new head
1127 * (i.e., the records with invalid CRC) if the cycle number
1128 * matches the the current cycle.
1130 found = xlog_rseek_logrec_hdr(log, first_bad, *tail_blk, 1,
1131 buffer, rhead_blk, rhead, wrapped);
1134 if (found == 0) /* XXX: right thing to do here? */
1138 * Reset the head block to the starting block of the first bad
1139 * log record and set the tail block based on the last good
1142 * Bail out if the updated head/tail match as this indicates
1143 * possible corruption outside of the acceptable
1144 * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair...
1146 *head_blk = first_bad;
1147 *tail_blk = BLOCK_LSN(be64_to_cpu((*rhead)->h_tail_lsn));
1148 if (*head_blk == *tail_blk) {
1156 return xlog_verify_tail(log, *head_blk, tail_blk,
1157 be32_to_cpu((*rhead)->h_size));
1161 * We need to make sure we handle log wrapping properly, so we can't use the
1162 * calculated logbno directly. Make sure it wraps to the correct bno inside the
1165 * The log is limited to 32 bit sizes, so we use the appropriate modulus
1166 * operation here and cast it back to a 64 bit daddr on return.
1168 static inline xfs_daddr_t
1175 div_s64_rem(bno, log->l_logBBsize, &mod);
1180 * Check whether the head of the log points to an unmount record. In other
1181 * words, determine whether the log is clean. If so, update the in-core state
1185 xlog_check_unmount_rec(
1187 xfs_daddr_t *head_blk,
1188 xfs_daddr_t *tail_blk,
1189 struct xlog_rec_header *rhead,
1190 xfs_daddr_t rhead_blk,
1194 struct xlog_op_header *op_head;
1195 xfs_daddr_t umount_data_blk;
1196 xfs_daddr_t after_umount_blk;
1204 * Look for unmount record. If we find it, then we know there was a
1205 * clean unmount. Since 'i' could be the last block in the physical
1206 * log, we convert to a log block before comparing to the head_blk.
1208 * Save the current tail lsn to use to pass to xlog_clear_stale_blocks()
1209 * below. We won't want to clear the unmount record if there is one, so
1210 * we pass the lsn of the unmount record rather than the block after it.
1212 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1213 int h_size = be32_to_cpu(rhead->h_size);
1214 int h_version = be32_to_cpu(rhead->h_version);
1216 if ((h_version & XLOG_VERSION_2) &&
1217 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
1218 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
1219 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1228 after_umount_blk = xlog_wrap_logbno(log,
1229 rhead_blk + hblks + BTOBB(be32_to_cpu(rhead->h_len)));
1231 if (*head_blk == after_umount_blk &&
1232 be32_to_cpu(rhead->h_num_logops) == 1) {
1233 umount_data_blk = xlog_wrap_logbno(log, rhead_blk + hblks);
1234 error = xlog_bread(log, umount_data_blk, 1, buffer, &offset);
1238 op_head = (struct xlog_op_header *)offset;
1239 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1241 * Set tail and last sync so that newly written log
1242 * records will point recovery to after the current
1245 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1246 log->l_curr_cycle, after_umount_blk);
1247 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1248 log->l_curr_cycle, after_umount_blk);
1249 *tail_blk = after_umount_blk;
1261 xfs_daddr_t head_blk,
1262 struct xlog_rec_header *rhead,
1263 xfs_daddr_t rhead_blk,
1267 * Reset log values according to the state of the log when we
1268 * crashed. In the case where head_blk == 0, we bump curr_cycle
1269 * one because the next write starts a new cycle rather than
1270 * continuing the cycle of the last good log record. At this
1271 * point we have guaranteed that all partial log records have been
1272 * accounted for. Therefore, we know that the last good log record
1273 * written was complete and ended exactly on the end boundary
1274 * of the physical log.
1276 log->l_prev_block = rhead_blk;
1277 log->l_curr_block = (int)head_blk;
1278 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
1280 log->l_curr_cycle++;
1281 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
1282 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
1283 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
1284 BBTOB(log->l_curr_block));
1285 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
1286 BBTOB(log->l_curr_block));
1290 * Find the sync block number or the tail of the log.
1292 * This will be the block number of the last record to have its
1293 * associated buffers synced to disk. Every log record header has
1294 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
1295 * to get a sync block number. The only concern is to figure out which
1296 * log record header to believe.
1298 * The following algorithm uses the log record header with the largest
1299 * lsn. The entire log record does not need to be valid. We only care
1300 * that the header is valid.
1302 * We could speed up search by using current head_blk buffer, but it is not
1308 xfs_daddr_t *head_blk,
1309 xfs_daddr_t *tail_blk)
1311 xlog_rec_header_t *rhead;
1312 char *offset = NULL;
1315 xfs_daddr_t rhead_blk;
1317 bool wrapped = false;
1321 * Find previous log record
1323 if ((error = xlog_find_head(log, head_blk)))
1325 ASSERT(*head_blk < INT_MAX);
1327 buffer = xlog_alloc_buffer(log, 1);
1330 if (*head_blk == 0) { /* special case */
1331 error = xlog_bread(log, 0, 1, buffer, &offset);
1335 if (xlog_get_cycle(offset) == 0) {
1337 /* leave all other log inited values alone */
1343 * Search backwards through the log looking for the log record header
1344 * block. This wraps all the way back around to the head so something is
1345 * seriously wrong if we can't find it.
1347 error = xlog_rseek_logrec_hdr(log, *head_blk, *head_blk, 1, buffer,
1348 &rhead_blk, &rhead, &wrapped);
1352 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
1353 return -EFSCORRUPTED;
1355 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
1358 * Set the log state based on the current head record.
1360 xlog_set_state(log, *head_blk, rhead, rhead_blk, wrapped);
1361 tail_lsn = atomic64_read(&log->l_tail_lsn);
1364 * Look for an unmount record at the head of the log. This sets the log
1365 * state to determine whether recovery is necessary.
1367 error = xlog_check_unmount_rec(log, head_blk, tail_blk, rhead,
1368 rhead_blk, buffer, &clean);
1373 * Verify the log head if the log is not clean (e.g., we have anything
1374 * but an unmount record at the head). This uses CRC verification to
1375 * detect and trim torn writes. If discovered, CRC failures are
1376 * considered torn writes and the log head is trimmed accordingly.
1378 * Note that we can only run CRC verification when the log is dirty
1379 * because there's no guarantee that the log data behind an unmount
1380 * record is compatible with the current architecture.
1383 xfs_daddr_t orig_head = *head_blk;
1385 error = xlog_verify_head(log, head_blk, tail_blk, buffer,
1386 &rhead_blk, &rhead, &wrapped);
1390 /* update in-core state again if the head changed */
1391 if (*head_blk != orig_head) {
1392 xlog_set_state(log, *head_blk, rhead, rhead_blk,
1394 tail_lsn = atomic64_read(&log->l_tail_lsn);
1395 error = xlog_check_unmount_rec(log, head_blk, tail_blk,
1396 rhead, rhead_blk, buffer,
1404 * Note that the unmount was clean. If the unmount was not clean, we
1405 * need to know this to rebuild the superblock counters from the perag
1406 * headers if we have a filesystem using non-persistent counters.
1409 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1412 * Make sure that there are no blocks in front of the head
1413 * with the same cycle number as the head. This can happen
1414 * because we allow multiple outstanding log writes concurrently,
1415 * and the later writes might make it out before earlier ones.
1417 * We use the lsn from before modifying it so that we'll never
1418 * overwrite the unmount record after a clean unmount.
1420 * Do this only if we are going to recover the filesystem
1422 * NOTE: This used to say "if (!readonly)"
1423 * However on Linux, we can & do recover a read-only filesystem.
1424 * We only skip recovery if NORECOVERY is specified on mount,
1425 * in which case we would not be here.
1427 * But... if the -device- itself is readonly, just skip this.
1428 * We can't recover this device anyway, so it won't matter.
1430 if (!xfs_readonly_buftarg(log->l_targ))
1431 error = xlog_clear_stale_blocks(log, tail_lsn);
1437 xfs_warn(log->l_mp, "failed to locate log tail");
1442 * Is the log zeroed at all?
1444 * The last binary search should be changed to perform an X block read
1445 * once X becomes small enough. You can then search linearly through
1446 * the X blocks. This will cut down on the number of reads we need to do.
1448 * If the log is partially zeroed, this routine will pass back the blkno
1449 * of the first block with cycle number 0. It won't have a complete LR
1453 * 0 => the log is completely written to
1454 * 1 => use *blk_no as the first block of the log
1455 * <0 => error has occurred
1460 xfs_daddr_t *blk_no)
1464 uint first_cycle, last_cycle;
1465 xfs_daddr_t new_blk, last_blk, start_blk;
1466 xfs_daddr_t num_scan_bblks;
1467 int error, log_bbnum = log->l_logBBsize;
1471 /* check totally zeroed log */
1472 buffer = xlog_alloc_buffer(log, 1);
1475 error = xlog_bread(log, 0, 1, buffer, &offset);
1477 goto out_free_buffer;
1479 first_cycle = xlog_get_cycle(offset);
1480 if (first_cycle == 0) { /* completely zeroed log */
1486 /* check partially zeroed log */
1487 error = xlog_bread(log, log_bbnum-1, 1, buffer, &offset);
1489 goto out_free_buffer;
1491 last_cycle = xlog_get_cycle(offset);
1492 if (last_cycle != 0) { /* log completely written to */
1497 /* we have a partially zeroed log */
1498 last_blk = log_bbnum-1;
1499 error = xlog_find_cycle_start(log, buffer, 0, &last_blk, 0);
1501 goto out_free_buffer;
1504 * Validate the answer. Because there is no way to guarantee that
1505 * the entire log is made up of log records which are the same size,
1506 * we scan over the defined maximum blocks. At this point, the maximum
1507 * is not chosen to mean anything special. XXXmiken
1509 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1510 ASSERT(num_scan_bblks <= INT_MAX);
1512 if (last_blk < num_scan_bblks)
1513 num_scan_bblks = last_blk;
1514 start_blk = last_blk - num_scan_bblks;
1517 * We search for any instances of cycle number 0 that occur before
1518 * our current estimate of the head. What we're trying to detect is
1519 * 1 ... | 0 | 1 | 0...
1520 * ^ binary search ends here
1522 if ((error = xlog_find_verify_cycle(log, start_blk,
1523 (int)num_scan_bblks, 0, &new_blk)))
1524 goto out_free_buffer;
1529 * Potentially backup over partial log record write. We don't need
1530 * to search the end of the log because we know it is zero.
1532 error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0);
1536 goto out_free_buffer;
1547 * These are simple subroutines used by xlog_clear_stale_blocks() below
1548 * to initialize a buffer full of empty log record headers and write
1549 * them into the log.
1560 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1562 memset(buf, 0, BBSIZE);
1563 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1564 recp->h_cycle = cpu_to_be32(cycle);
1565 recp->h_version = cpu_to_be32(
1566 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1567 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1568 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1569 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1570 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1574 xlog_write_log_records(
1585 int sectbb = log->l_sectBBsize;
1586 int end_block = start_block + blocks;
1592 * Greedily allocate a buffer big enough to handle the full
1593 * range of basic blocks to be written. If that fails, try
1594 * a smaller size. We need to be able to write at least a
1595 * log sector, or we're out of luck.
1597 bufblks = 1 << ffs(blocks);
1598 while (bufblks > log->l_logBBsize)
1600 while (!(buffer = xlog_alloc_buffer(log, bufblks))) {
1602 if (bufblks < sectbb)
1606 /* We may need to do a read at the start to fill in part of
1607 * the buffer in the starting sector not covered by the first
1610 balign = round_down(start_block, sectbb);
1611 if (balign != start_block) {
1612 error = xlog_bread_noalign(log, start_block, 1, buffer);
1614 goto out_free_buffer;
1616 j = start_block - balign;
1619 for (i = start_block; i < end_block; i += bufblks) {
1620 int bcount, endcount;
1622 bcount = min(bufblks, end_block - start_block);
1623 endcount = bcount - j;
1625 /* We may need to do a read at the end to fill in part of
1626 * the buffer in the final sector not covered by the write.
1627 * If this is the same sector as the above read, skip it.
1629 ealign = round_down(end_block, sectbb);
1630 if (j == 0 && (start_block + endcount > ealign)) {
1631 error = xlog_bread_noalign(log, ealign, sectbb,
1632 buffer + BBTOB(ealign - start_block));
1638 offset = buffer + xlog_align(log, start_block);
1639 for (; j < endcount; j++) {
1640 xlog_add_record(log, offset, cycle, i+j,
1641 tail_cycle, tail_block);
1644 error = xlog_bwrite(log, start_block, endcount, buffer);
1647 start_block += endcount;
1657 * This routine is called to blow away any incomplete log writes out
1658 * in front of the log head. We do this so that we won't become confused
1659 * if we come up, write only a little bit more, and then crash again.
1660 * If we leave the partial log records out there, this situation could
1661 * cause us to think those partial writes are valid blocks since they
1662 * have the current cycle number. We get rid of them by overwriting them
1663 * with empty log records with the old cycle number rather than the
1666 * The tail lsn is passed in rather than taken from
1667 * the log so that we will not write over the unmount record after a
1668 * clean unmount in a 512 block log. Doing so would leave the log without
1669 * any valid log records in it until a new one was written. If we crashed
1670 * during that time we would not be able to recover.
1673 xlog_clear_stale_blocks(
1677 int tail_cycle, head_cycle;
1678 int tail_block, head_block;
1679 int tail_distance, max_distance;
1683 tail_cycle = CYCLE_LSN(tail_lsn);
1684 tail_block = BLOCK_LSN(tail_lsn);
1685 head_cycle = log->l_curr_cycle;
1686 head_block = log->l_curr_block;
1689 * Figure out the distance between the new head of the log
1690 * and the tail. We want to write over any blocks beyond the
1691 * head that we may have written just before the crash, but
1692 * we don't want to overwrite the tail of the log.
1694 if (head_cycle == tail_cycle) {
1696 * The tail is behind the head in the physical log,
1697 * so the distance from the head to the tail is the
1698 * distance from the head to the end of the log plus
1699 * the distance from the beginning of the log to the
1702 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1703 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1704 XFS_ERRLEVEL_LOW, log->l_mp);
1705 return -EFSCORRUPTED;
1707 tail_distance = tail_block + (log->l_logBBsize - head_block);
1710 * The head is behind the tail in the physical log,
1711 * so the distance from the head to the tail is just
1712 * the tail block minus the head block.
1714 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1715 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1716 XFS_ERRLEVEL_LOW, log->l_mp);
1717 return -EFSCORRUPTED;
1719 tail_distance = tail_block - head_block;
1723 * If the head is right up against the tail, we can't clear
1726 if (tail_distance <= 0) {
1727 ASSERT(tail_distance == 0);
1731 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1733 * Take the smaller of the maximum amount of outstanding I/O
1734 * we could have and the distance to the tail to clear out.
1735 * We take the smaller so that we don't overwrite the tail and
1736 * we don't waste all day writing from the head to the tail
1739 max_distance = min(max_distance, tail_distance);
1741 if ((head_block + max_distance) <= log->l_logBBsize) {
1743 * We can stomp all the blocks we need to without
1744 * wrapping around the end of the log. Just do it
1745 * in a single write. Use the cycle number of the
1746 * current cycle minus one so that the log will look like:
1749 error = xlog_write_log_records(log, (head_cycle - 1),
1750 head_block, max_distance, tail_cycle,
1756 * We need to wrap around the end of the physical log in
1757 * order to clear all the blocks. Do it in two separate
1758 * I/Os. The first write should be from the head to the
1759 * end of the physical log, and it should use the current
1760 * cycle number minus one just like above.
1762 distance = log->l_logBBsize - head_block;
1763 error = xlog_write_log_records(log, (head_cycle - 1),
1764 head_block, distance, tail_cycle,
1771 * Now write the blocks at the start of the physical log.
1772 * This writes the remainder of the blocks we want to clear.
1773 * It uses the current cycle number since we're now on the
1774 * same cycle as the head so that we get:
1775 * n ... n ... | n - 1 ...
1776 * ^^^^^ blocks we're writing
1778 distance = max_distance - (log->l_logBBsize - head_block);
1779 error = xlog_write_log_records(log, head_cycle, 0, distance,
1780 tail_cycle, tail_block);
1788 /******************************************************************************
1790 * Log recover routines
1792 ******************************************************************************
1796 * Sort the log items in the transaction.
1798 * The ordering constraints are defined by the inode allocation and unlink
1799 * behaviour. The rules are:
1801 * 1. Every item is only logged once in a given transaction. Hence it
1802 * represents the last logged state of the item. Hence ordering is
1803 * dependent on the order in which operations need to be performed so
1804 * required initial conditions are always met.
1806 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1807 * there's nothing to replay from them so we can simply cull them
1808 * from the transaction. However, we can't do that until after we've
1809 * replayed all the other items because they may be dependent on the
1810 * cancelled buffer and replaying the cancelled buffer can remove it
1811 * form the cancelled buffer table. Hence they have tobe done last.
1813 * 3. Inode allocation buffers must be replayed before inode items that
1814 * read the buffer and replay changes into it. For filesystems using the
1815 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1816 * treated the same as inode allocation buffers as they create and
1817 * initialise the buffers directly.
1819 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1820 * This ensures that inodes are completely flushed to the inode buffer
1821 * in a "free" state before we remove the unlinked inode list pointer.
1823 * Hence the ordering needs to be inode allocation buffers first, inode items
1824 * second, inode unlink buffers third and cancelled buffers last.
1826 * But there's a problem with that - we can't tell an inode allocation buffer
1827 * apart from a regular buffer, so we can't separate them. We can, however,
1828 * tell an inode unlink buffer from the others, and so we can separate them out
1829 * from all the other buffers and move them to last.
1831 * Hence, 4 lists, in order from head to tail:
1832 * - buffer_list for all buffers except cancelled/inode unlink buffers
1833 * - item_list for all non-buffer items
1834 * - inode_buffer_list for inode unlink buffers
1835 * - cancel_list for the cancelled buffers
1837 * Note that we add objects to the tail of the lists so that first-to-last
1838 * ordering is preserved within the lists. Adding objects to the head of the
1839 * list means when we traverse from the head we walk them in last-to-first
1840 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1841 * but for all other items there may be specific ordering that we need to
1845 xlog_recover_reorder_trans(
1847 struct xlog_recover *trans,
1850 xlog_recover_item_t *item, *n;
1852 LIST_HEAD(sort_list);
1853 LIST_HEAD(cancel_list);
1854 LIST_HEAD(buffer_list);
1855 LIST_HEAD(inode_buffer_list);
1856 LIST_HEAD(inode_list);
1858 list_splice_init(&trans->r_itemq, &sort_list);
1859 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1860 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1862 switch (ITEM_TYPE(item)) {
1863 case XFS_LI_ICREATE:
1864 list_move_tail(&item->ri_list, &buffer_list);
1867 if (buf_f->blf_flags & XFS_BLF_CANCEL) {
1868 trace_xfs_log_recover_item_reorder_head(log,
1870 list_move(&item->ri_list, &cancel_list);
1873 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1874 list_move(&item->ri_list, &inode_buffer_list);
1877 list_move_tail(&item->ri_list, &buffer_list);
1881 case XFS_LI_QUOTAOFF:
1890 trace_xfs_log_recover_item_reorder_tail(log,
1892 list_move_tail(&item->ri_list, &inode_list);
1896 "%s: unrecognized type of log operation",
1900 * return the remaining items back to the transaction
1901 * item list so they can be freed in caller.
1903 if (!list_empty(&sort_list))
1904 list_splice_init(&sort_list, &trans->r_itemq);
1910 ASSERT(list_empty(&sort_list));
1911 if (!list_empty(&buffer_list))
1912 list_splice(&buffer_list, &trans->r_itemq);
1913 if (!list_empty(&inode_list))
1914 list_splice_tail(&inode_list, &trans->r_itemq);
1915 if (!list_empty(&inode_buffer_list))
1916 list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1917 if (!list_empty(&cancel_list))
1918 list_splice_tail(&cancel_list, &trans->r_itemq);
1923 * Build up the table of buf cancel records so that we don't replay
1924 * cancelled data in the second pass. For buffer records that are
1925 * not cancel records, there is nothing to do here so we just return.
1927 * If we get a cancel record which is already in the table, this indicates
1928 * that the buffer was cancelled multiple times. In order to ensure
1929 * that during pass 2 we keep the record in the table until we reach its
1930 * last occurrence in the log, we keep a reference count in the cancel
1931 * record in the table to tell us how many times we expect to see this
1932 * record during the second pass.
1935 xlog_recover_buffer_pass1(
1937 struct xlog_recover_item *item)
1939 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1940 struct list_head *bucket;
1941 struct xfs_buf_cancel *bcp;
1944 * If this isn't a cancel buffer item, then just return.
1946 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1947 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
1952 * Insert an xfs_buf_cancel record into the hash table of them.
1953 * If there is already an identical record, bump its reference count.
1955 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1956 list_for_each_entry(bcp, bucket, bc_list) {
1957 if (bcp->bc_blkno == buf_f->blf_blkno &&
1958 bcp->bc_len == buf_f->blf_len) {
1960 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
1965 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), 0);
1966 bcp->bc_blkno = buf_f->blf_blkno;
1967 bcp->bc_len = buf_f->blf_len;
1968 bcp->bc_refcount = 1;
1969 list_add_tail(&bcp->bc_list, bucket);
1971 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
1976 * Check to see whether the buffer being recovered has a corresponding
1977 * entry in the buffer cancel record table. If it is, return the cancel
1978 * buffer structure to the caller.
1980 STATIC struct xfs_buf_cancel *
1981 xlog_peek_buffer_cancelled(
1985 unsigned short flags)
1987 struct list_head *bucket;
1988 struct xfs_buf_cancel *bcp;
1990 if (!log->l_buf_cancel_table) {
1991 /* empty table means no cancelled buffers in the log */
1992 ASSERT(!(flags & XFS_BLF_CANCEL));
1996 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1997 list_for_each_entry(bcp, bucket, bc_list) {
1998 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
2003 * We didn't find a corresponding entry in the table, so return 0 so
2004 * that the buffer is NOT cancelled.
2006 ASSERT(!(flags & XFS_BLF_CANCEL));
2011 * If the buffer is being cancelled then return 1 so that it will be cancelled,
2012 * otherwise return 0. If the buffer is actually a buffer cancel item
2013 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
2014 * table and remove it from the table if this is the last reference.
2016 * We remove the cancel record from the table when we encounter its last
2017 * occurrence in the log so that if the same buffer is re-used again after its
2018 * last cancellation we actually replay the changes made at that point.
2021 xlog_check_buffer_cancelled(
2025 unsigned short flags)
2027 struct xfs_buf_cancel *bcp;
2029 bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags);
2034 * We've go a match, so return 1 so that the recovery of this buffer
2035 * is cancelled. If this buffer is actually a buffer cancel log
2036 * item, then decrement the refcount on the one in the table and
2037 * remove it if this is the last reference.
2039 if (flags & XFS_BLF_CANCEL) {
2040 if (--bcp->bc_refcount == 0) {
2041 list_del(&bcp->bc_list);
2049 * Perform recovery for a buffer full of inodes. In these buffers, the only
2050 * data which should be recovered is that which corresponds to the
2051 * di_next_unlinked pointers in the on disk inode structures. The rest of the
2052 * data for the inodes is always logged through the inodes themselves rather
2053 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
2055 * The only time when buffers full of inodes are fully recovered is when the
2056 * buffer is full of newly allocated inodes. In this case the buffer will
2057 * not be marked as an inode buffer and so will be sent to
2058 * xlog_recover_do_reg_buffer() below during recovery.
2061 xlog_recover_do_inode_buffer(
2062 struct xfs_mount *mp,
2063 xlog_recover_item_t *item,
2065 xfs_buf_log_format_t *buf_f)
2071 int reg_buf_offset = 0;
2072 int reg_buf_bytes = 0;
2073 int next_unlinked_offset;
2075 xfs_agino_t *logged_nextp;
2076 xfs_agino_t *buffer_nextp;
2078 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
2081 * Post recovery validation only works properly on CRC enabled
2084 if (xfs_sb_version_hascrc(&mp->m_sb))
2085 bp->b_ops = &xfs_inode_buf_ops;
2087 inodes_per_buf = BBTOB(bp->b_length) >> mp->m_sb.sb_inodelog;
2088 for (i = 0; i < inodes_per_buf; i++) {
2089 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
2090 offsetof(xfs_dinode_t, di_next_unlinked);
2092 while (next_unlinked_offset >=
2093 (reg_buf_offset + reg_buf_bytes)) {
2095 * The next di_next_unlinked field is beyond
2096 * the current logged region. Find the next
2097 * logged region that contains or is beyond
2098 * the current di_next_unlinked field.
2101 bit = xfs_next_bit(buf_f->blf_data_map,
2102 buf_f->blf_map_size, bit);
2105 * If there are no more logged regions in the
2106 * buffer, then we're done.
2111 nbits = xfs_contig_bits(buf_f->blf_data_map,
2112 buf_f->blf_map_size, bit);
2114 reg_buf_offset = bit << XFS_BLF_SHIFT;
2115 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
2120 * If the current logged region starts after the current
2121 * di_next_unlinked field, then move on to the next
2122 * di_next_unlinked field.
2124 if (next_unlinked_offset < reg_buf_offset)
2127 ASSERT(item->ri_buf[item_index].i_addr != NULL);
2128 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
2129 ASSERT((reg_buf_offset + reg_buf_bytes) <= BBTOB(bp->b_length));
2132 * The current logged region contains a copy of the
2133 * current di_next_unlinked field. Extract its value
2134 * and copy it to the buffer copy.
2136 logged_nextp = item->ri_buf[item_index].i_addr +
2137 next_unlinked_offset - reg_buf_offset;
2138 if (unlikely(*logged_nextp == 0)) {
2140 "Bad inode buffer log record (ptr = "PTR_FMT", bp = "PTR_FMT"). "
2141 "Trying to replay bad (0) inode di_next_unlinked field.",
2143 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
2144 XFS_ERRLEVEL_LOW, mp);
2145 return -EFSCORRUPTED;
2148 buffer_nextp = xfs_buf_offset(bp, next_unlinked_offset);
2149 *buffer_nextp = *logged_nextp;
2152 * If necessary, recalculate the CRC in the on-disk inode. We
2153 * have to leave the inode in a consistent state for whoever
2156 xfs_dinode_calc_crc(mp,
2157 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
2165 * V5 filesystems know the age of the buffer on disk being recovered. We can
2166 * have newer objects on disk than we are replaying, and so for these cases we
2167 * don't want to replay the current change as that will make the buffer contents
2168 * temporarily invalid on disk.
2170 * The magic number might not match the buffer type we are going to recover
2171 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
2172 * extract the LSN of the existing object in the buffer based on it's current
2173 * magic number. If we don't recognise the magic number in the buffer, then
2174 * return a LSN of -1 so that the caller knows it was an unrecognised block and
2175 * so can recover the buffer.
2177 * Note: we cannot rely solely on magic number matches to determine that the
2178 * buffer has a valid LSN - we also need to verify that it belongs to this
2179 * filesystem, so we need to extract the object's LSN and compare it to that
2180 * which we read from the superblock. If the UUIDs don't match, then we've got a
2181 * stale metadata block from an old filesystem instance that we need to recover
2185 xlog_recover_get_buf_lsn(
2186 struct xfs_mount *mp,
2192 void *blk = bp->b_addr;
2196 /* v4 filesystems always recover immediately */
2197 if (!xfs_sb_version_hascrc(&mp->m_sb))
2198 goto recover_immediately;
2200 magic32 = be32_to_cpu(*(__be32 *)blk);
2202 case XFS_ABTB_CRC_MAGIC:
2203 case XFS_ABTC_CRC_MAGIC:
2204 case XFS_ABTB_MAGIC:
2205 case XFS_ABTC_MAGIC:
2206 case XFS_RMAP_CRC_MAGIC:
2207 case XFS_REFC_CRC_MAGIC:
2208 case XFS_IBT_CRC_MAGIC:
2209 case XFS_IBT_MAGIC: {
2210 struct xfs_btree_block *btb = blk;
2212 lsn = be64_to_cpu(btb->bb_u.s.bb_lsn);
2213 uuid = &btb->bb_u.s.bb_uuid;
2216 case XFS_BMAP_CRC_MAGIC:
2217 case XFS_BMAP_MAGIC: {
2218 struct xfs_btree_block *btb = blk;
2220 lsn = be64_to_cpu(btb->bb_u.l.bb_lsn);
2221 uuid = &btb->bb_u.l.bb_uuid;
2225 lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn);
2226 uuid = &((struct xfs_agf *)blk)->agf_uuid;
2228 case XFS_AGFL_MAGIC:
2229 lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn);
2230 uuid = &((struct xfs_agfl *)blk)->agfl_uuid;
2233 lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn);
2234 uuid = &((struct xfs_agi *)blk)->agi_uuid;
2236 case XFS_SYMLINK_MAGIC:
2237 lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn);
2238 uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid;
2240 case XFS_DIR3_BLOCK_MAGIC:
2241 case XFS_DIR3_DATA_MAGIC:
2242 case XFS_DIR3_FREE_MAGIC:
2243 lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn);
2244 uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid;
2246 case XFS_ATTR3_RMT_MAGIC:
2248 * Remote attr blocks are written synchronously, rather than
2249 * being logged. That means they do not contain a valid LSN
2250 * (i.e. transactionally ordered) in them, and hence any time we
2251 * see a buffer to replay over the top of a remote attribute
2252 * block we should simply do so.
2254 goto recover_immediately;
2257 * superblock uuids are magic. We may or may not have a
2258 * sb_meta_uuid on disk, but it will be set in the in-core
2259 * superblock. We set the uuid pointer for verification
2260 * according to the superblock feature mask to ensure we check
2261 * the relevant UUID in the superblock.
2263 lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn);
2264 if (xfs_sb_version_hasmetauuid(&mp->m_sb))
2265 uuid = &((struct xfs_dsb *)blk)->sb_meta_uuid;
2267 uuid = &((struct xfs_dsb *)blk)->sb_uuid;
2273 if (lsn != (xfs_lsn_t)-1) {
2274 if (!uuid_equal(&mp->m_sb.sb_meta_uuid, uuid))
2275 goto recover_immediately;
2279 magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic);
2281 case XFS_DIR3_LEAF1_MAGIC:
2282 case XFS_DIR3_LEAFN_MAGIC:
2283 case XFS_DA3_NODE_MAGIC:
2284 lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn);
2285 uuid = &((struct xfs_da3_blkinfo *)blk)->uuid;
2291 if (lsn != (xfs_lsn_t)-1) {
2292 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
2293 goto recover_immediately;
2298 * We do individual object checks on dquot and inode buffers as they
2299 * have their own individual LSN records. Also, we could have a stale
2300 * buffer here, so we have to at least recognise these buffer types.
2302 * A notd complexity here is inode unlinked list processing - it logs
2303 * the inode directly in the buffer, but we don't know which inodes have
2304 * been modified, and there is no global buffer LSN. Hence we need to
2305 * recover all inode buffer types immediately. This problem will be
2306 * fixed by logical logging of the unlinked list modifications.
2308 magic16 = be16_to_cpu(*(__be16 *)blk);
2310 case XFS_DQUOT_MAGIC:
2311 case XFS_DINODE_MAGIC:
2312 goto recover_immediately;
2317 /* unknown buffer contents, recover immediately */
2319 recover_immediately:
2320 return (xfs_lsn_t)-1;
2325 * Validate the recovered buffer is of the correct type and attach the
2326 * appropriate buffer operations to them for writeback. Magic numbers are in a
2328 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2329 * the first 32 bits of the buffer (most blocks),
2330 * inside a struct xfs_da_blkinfo at the start of the buffer.
2333 xlog_recover_validate_buf_type(
2334 struct xfs_mount *mp,
2336 xfs_buf_log_format_t *buf_f,
2337 xfs_lsn_t current_lsn)
2339 struct xfs_da_blkinfo *info = bp->b_addr;
2343 char *warnmsg = NULL;
2346 * We can only do post recovery validation on items on CRC enabled
2347 * fielsystems as we need to know when the buffer was written to be able
2348 * to determine if we should have replayed the item. If we replay old
2349 * metadata over a newer buffer, then it will enter a temporarily
2350 * inconsistent state resulting in verification failures. Hence for now
2351 * just avoid the verification stage for non-crc filesystems
2353 if (!xfs_sb_version_hascrc(&mp->m_sb))
2356 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
2357 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
2358 magicda = be16_to_cpu(info->magic);
2359 switch (xfs_blft_from_flags(buf_f)) {
2360 case XFS_BLFT_BTREE_BUF:
2362 case XFS_ABTB_CRC_MAGIC:
2363 case XFS_ABTB_MAGIC:
2364 bp->b_ops = &xfs_bnobt_buf_ops;
2366 case XFS_ABTC_CRC_MAGIC:
2367 case XFS_ABTC_MAGIC:
2368 bp->b_ops = &xfs_cntbt_buf_ops;
2370 case XFS_IBT_CRC_MAGIC:
2372 bp->b_ops = &xfs_inobt_buf_ops;
2374 case XFS_FIBT_CRC_MAGIC:
2375 case XFS_FIBT_MAGIC:
2376 bp->b_ops = &xfs_finobt_buf_ops;
2378 case XFS_BMAP_CRC_MAGIC:
2379 case XFS_BMAP_MAGIC:
2380 bp->b_ops = &xfs_bmbt_buf_ops;
2382 case XFS_RMAP_CRC_MAGIC:
2383 bp->b_ops = &xfs_rmapbt_buf_ops;
2385 case XFS_REFC_CRC_MAGIC:
2386 bp->b_ops = &xfs_refcountbt_buf_ops;
2389 warnmsg = "Bad btree block magic!";
2393 case XFS_BLFT_AGF_BUF:
2394 if (magic32 != XFS_AGF_MAGIC) {
2395 warnmsg = "Bad AGF block magic!";
2398 bp->b_ops = &xfs_agf_buf_ops;
2400 case XFS_BLFT_AGFL_BUF:
2401 if (magic32 != XFS_AGFL_MAGIC) {
2402 warnmsg = "Bad AGFL block magic!";
2405 bp->b_ops = &xfs_agfl_buf_ops;
2407 case XFS_BLFT_AGI_BUF:
2408 if (magic32 != XFS_AGI_MAGIC) {
2409 warnmsg = "Bad AGI block magic!";
2412 bp->b_ops = &xfs_agi_buf_ops;
2414 case XFS_BLFT_UDQUOT_BUF:
2415 case XFS_BLFT_PDQUOT_BUF:
2416 case XFS_BLFT_GDQUOT_BUF:
2417 #ifdef CONFIG_XFS_QUOTA
2418 if (magic16 != XFS_DQUOT_MAGIC) {
2419 warnmsg = "Bad DQUOT block magic!";
2422 bp->b_ops = &xfs_dquot_buf_ops;
2425 "Trying to recover dquots without QUOTA support built in!");
2429 case XFS_BLFT_DINO_BUF:
2430 if (magic16 != XFS_DINODE_MAGIC) {
2431 warnmsg = "Bad INODE block magic!";
2434 bp->b_ops = &xfs_inode_buf_ops;
2436 case XFS_BLFT_SYMLINK_BUF:
2437 if (magic32 != XFS_SYMLINK_MAGIC) {
2438 warnmsg = "Bad symlink block magic!";
2441 bp->b_ops = &xfs_symlink_buf_ops;
2443 case XFS_BLFT_DIR_BLOCK_BUF:
2444 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2445 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2446 warnmsg = "Bad dir block magic!";
2449 bp->b_ops = &xfs_dir3_block_buf_ops;
2451 case XFS_BLFT_DIR_DATA_BUF:
2452 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2453 magic32 != XFS_DIR3_DATA_MAGIC) {
2454 warnmsg = "Bad dir data magic!";
2457 bp->b_ops = &xfs_dir3_data_buf_ops;
2459 case XFS_BLFT_DIR_FREE_BUF:
2460 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2461 magic32 != XFS_DIR3_FREE_MAGIC) {
2462 warnmsg = "Bad dir3 free magic!";
2465 bp->b_ops = &xfs_dir3_free_buf_ops;
2467 case XFS_BLFT_DIR_LEAF1_BUF:
2468 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2469 magicda != XFS_DIR3_LEAF1_MAGIC) {
2470 warnmsg = "Bad dir leaf1 magic!";
2473 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2475 case XFS_BLFT_DIR_LEAFN_BUF:
2476 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2477 magicda != XFS_DIR3_LEAFN_MAGIC) {
2478 warnmsg = "Bad dir leafn magic!";
2481 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2483 case XFS_BLFT_DA_NODE_BUF:
2484 if (magicda != XFS_DA_NODE_MAGIC &&
2485 magicda != XFS_DA3_NODE_MAGIC) {
2486 warnmsg = "Bad da node magic!";
2489 bp->b_ops = &xfs_da3_node_buf_ops;
2491 case XFS_BLFT_ATTR_LEAF_BUF:
2492 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2493 magicda != XFS_ATTR3_LEAF_MAGIC) {
2494 warnmsg = "Bad attr leaf magic!";
2497 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2499 case XFS_BLFT_ATTR_RMT_BUF:
2500 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
2501 warnmsg = "Bad attr remote magic!";
2504 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2506 case XFS_BLFT_SB_BUF:
2507 if (magic32 != XFS_SB_MAGIC) {
2508 warnmsg = "Bad SB block magic!";
2511 bp->b_ops = &xfs_sb_buf_ops;
2513 #ifdef CONFIG_XFS_RT
2514 case XFS_BLFT_RTBITMAP_BUF:
2515 case XFS_BLFT_RTSUMMARY_BUF:
2516 /* no magic numbers for verification of RT buffers */
2517 bp->b_ops = &xfs_rtbuf_ops;
2519 #endif /* CONFIG_XFS_RT */
2521 xfs_warn(mp, "Unknown buffer type %d!",
2522 xfs_blft_from_flags(buf_f));
2527 * Nothing else to do in the case of a NULL current LSN as this means
2528 * the buffer is more recent than the change in the log and will be
2531 if (current_lsn == NULLCOMMITLSN)
2535 xfs_warn(mp, warnmsg);
2540 * We must update the metadata LSN of the buffer as it is written out to
2541 * ensure that older transactions never replay over this one and corrupt
2542 * the buffer. This can occur if log recovery is interrupted at some
2543 * point after the current transaction completes, at which point a
2544 * subsequent mount starts recovery from the beginning.
2546 * Write verifiers update the metadata LSN from log items attached to
2547 * the buffer. Therefore, initialize a bli purely to carry the LSN to
2548 * the verifier. We'll clean it up in our ->iodone() callback.
2551 struct xfs_buf_log_item *bip;
2553 ASSERT(!bp->b_iodone || bp->b_iodone == xlog_recover_iodone);
2554 bp->b_iodone = xlog_recover_iodone;
2555 xfs_buf_item_init(bp, mp);
2556 bip = bp->b_log_item;
2557 bip->bli_item.li_lsn = current_lsn;
2562 * Perform a 'normal' buffer recovery. Each logged region of the
2563 * buffer should be copied over the corresponding region in the
2564 * given buffer. The bitmap in the buf log format structure indicates
2565 * where to place the logged data.
2568 xlog_recover_do_reg_buffer(
2569 struct xfs_mount *mp,
2570 xlog_recover_item_t *item,
2572 xfs_buf_log_format_t *buf_f,
2573 xfs_lsn_t current_lsn)
2580 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2583 i = 1; /* 0 is the buf format structure */
2585 bit = xfs_next_bit(buf_f->blf_data_map,
2586 buf_f->blf_map_size, bit);
2589 nbits = xfs_contig_bits(buf_f->blf_data_map,
2590 buf_f->blf_map_size, bit);
2592 ASSERT(item->ri_buf[i].i_addr != NULL);
2593 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
2594 ASSERT(BBTOB(bp->b_length) >=
2595 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
2598 * The dirty regions logged in the buffer, even though
2599 * contiguous, may span multiple chunks. This is because the
2600 * dirty region may span a physical page boundary in a buffer
2601 * and hence be split into two separate vectors for writing into
2602 * the log. Hence we need to trim nbits back to the length of
2603 * the current region being copied out of the log.
2605 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2606 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2609 * Do a sanity check if this is a dquot buffer. Just checking
2610 * the first dquot in the buffer should do. XXXThis is
2611 * probably a good thing to do for other buf types also.
2614 if (buf_f->blf_flags &
2615 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2616 if (item->ri_buf[i].i_addr == NULL) {
2618 "XFS: NULL dquot in %s.", __func__);
2621 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
2623 "XFS: dquot too small (%d) in %s.",
2624 item->ri_buf[i].i_len, __func__);
2627 fa = xfs_dquot_verify(mp, item->ri_buf[i].i_addr,
2631 "dquot corrupt at %pS trying to replay into block 0x%llx",
2637 memcpy(xfs_buf_offset(bp,
2638 (uint)bit << XFS_BLF_SHIFT), /* dest */
2639 item->ri_buf[i].i_addr, /* source */
2640 nbits<<XFS_BLF_SHIFT); /* length */
2646 /* Shouldn't be any more regions */
2647 ASSERT(i == item->ri_total);
2649 xlog_recover_validate_buf_type(mp, bp, buf_f, current_lsn);
2653 * Perform a dquot buffer recovery.
2654 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2655 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2656 * Else, treat it as a regular buffer and do recovery.
2658 * Return false if the buffer was tossed and true if we recovered the buffer to
2659 * indicate to the caller if the buffer needs writing.
2662 xlog_recover_do_dquot_buffer(
2663 struct xfs_mount *mp,
2665 struct xlog_recover_item *item,
2667 struct xfs_buf_log_format *buf_f)
2671 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2674 * Filesystems are required to send in quota flags at mount time.
2680 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2681 type |= XFS_DQ_USER;
2682 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2683 type |= XFS_DQ_PROJ;
2684 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2685 type |= XFS_DQ_GROUP;
2687 * This type of quotas was turned off, so ignore this buffer
2689 if (log->l_quotaoffs_flag & type)
2692 xlog_recover_do_reg_buffer(mp, item, bp, buf_f, NULLCOMMITLSN);
2697 * This routine replays a modification made to a buffer at runtime.
2698 * There are actually two types of buffer, regular and inode, which
2699 * are handled differently. Inode buffers are handled differently
2700 * in that we only recover a specific set of data from them, namely
2701 * the inode di_next_unlinked fields. This is because all other inode
2702 * data is actually logged via inode records and any data we replay
2703 * here which overlaps that may be stale.
2705 * When meta-data buffers are freed at run time we log a buffer item
2706 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2707 * of the buffer in the log should not be replayed at recovery time.
2708 * This is so that if the blocks covered by the buffer are reused for
2709 * file data before we crash we don't end up replaying old, freed
2710 * meta-data into a user's file.
2712 * To handle the cancellation of buffer log items, we make two passes
2713 * over the log during recovery. During the first we build a table of
2714 * those buffers which have been cancelled, and during the second we
2715 * only replay those buffers which do not have corresponding cancel
2716 * records in the table. See xlog_recover_buffer_pass[1,2] above
2717 * for more details on the implementation of the table of cancel records.
2720 xlog_recover_buffer_pass2(
2722 struct list_head *buffer_list,
2723 struct xlog_recover_item *item,
2724 xfs_lsn_t current_lsn)
2726 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2727 xfs_mount_t *mp = log->l_mp;
2734 * In this pass we only want to recover all the buffers which have
2735 * not been cancelled and are not cancellation buffers themselves.
2737 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2738 buf_f->blf_len, buf_f->blf_flags)) {
2739 trace_xfs_log_recover_buf_cancel(log, buf_f);
2743 trace_xfs_log_recover_buf_recover(log, buf_f);
2746 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2747 buf_flags |= XBF_UNMAPPED;
2749 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2753 error = bp->b_error;
2755 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
2760 * Recover the buffer only if we get an LSN from it and it's less than
2761 * the lsn of the transaction we are replaying.
2763 * Note that we have to be extremely careful of readahead here.
2764 * Readahead does not attach verfiers to the buffers so if we don't
2765 * actually do any replay after readahead because of the LSN we found
2766 * in the buffer if more recent than that current transaction then we
2767 * need to attach the verifier directly. Failure to do so can lead to
2768 * future recovery actions (e.g. EFI and unlinked list recovery) can
2769 * operate on the buffers and they won't get the verifier attached. This
2770 * can lead to blocks on disk having the correct content but a stale
2773 * It is safe to assume these clean buffers are currently up to date.
2774 * If the buffer is dirtied by a later transaction being replayed, then
2775 * the verifier will be reset to match whatever recover turns that
2778 lsn = xlog_recover_get_buf_lsn(mp, bp);
2779 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2780 trace_xfs_log_recover_buf_skip(log, buf_f);
2781 xlog_recover_validate_buf_type(mp, bp, buf_f, NULLCOMMITLSN);
2785 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
2786 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2789 } else if (buf_f->blf_flags &
2790 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2793 dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2797 xlog_recover_do_reg_buffer(mp, item, bp, buf_f, current_lsn);
2801 * Perform delayed write on the buffer. Asynchronous writes will be
2802 * slower when taking into account all the buffers to be flushed.
2804 * Also make sure that only inode buffers with good sizes stay in
2805 * the buffer cache. The kernel moves inodes in buffers of 1 block
2806 * or inode_cluster_size bytes, whichever is bigger. The inode
2807 * buffers in the log can be a different size if the log was generated
2808 * by an older kernel using unclustered inode buffers or a newer kernel
2809 * running with a different inode cluster size. Regardless, if the
2810 * the inode buffer size isn't max(blocksize, inode_cluster_size)
2811 * for *our* value of inode_cluster_size, then we need to keep
2812 * the buffer out of the buffer cache so that the buffer won't
2813 * overlap with future reads of those inodes.
2815 if (XFS_DINODE_MAGIC ==
2816 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2817 (BBTOB(bp->b_length) != M_IGEO(log->l_mp)->inode_cluster_size)) {
2819 error = xfs_bwrite(bp);
2821 ASSERT(bp->b_mount == mp);
2822 bp->b_iodone = xlog_recover_iodone;
2823 xfs_buf_delwri_queue(bp, buffer_list);
2832 * Inode fork owner changes
2834 * If we have been told that we have to reparent the inode fork, it's because an
2835 * extent swap operation on a CRC enabled filesystem has been done and we are
2836 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2839 * The complexity here is that we don't have an inode context to work with, so
2840 * after we've replayed the inode we need to instantiate one. This is where the
2843 * We are in the middle of log recovery, so we can't run transactions. That
2844 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2845 * that will result in the corresponding iput() running the inode through
2846 * xfs_inactive(). If we've just replayed an inode core that changes the link
2847 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2848 * transactions (bad!).
2850 * So, to avoid this, we instantiate an inode directly from the inode core we've
2851 * just recovered. We have the buffer still locked, and all we really need to
2852 * instantiate is the inode core and the forks being modified. We can do this
2853 * manually, then run the inode btree owner change, and then tear down the
2854 * xfs_inode without having to run any transactions at all.
2856 * Also, because we don't have a transaction context available here but need to
2857 * gather all the buffers we modify for writeback so we pass the buffer_list
2858 * instead for the operation to use.
2862 xfs_recover_inode_owner_change(
2863 struct xfs_mount *mp,
2864 struct xfs_dinode *dip,
2865 struct xfs_inode_log_format *in_f,
2866 struct list_head *buffer_list)
2868 struct xfs_inode *ip;
2871 ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER));
2873 ip = xfs_inode_alloc(mp, in_f->ilf_ino);
2877 /* instantiate the inode */
2878 xfs_inode_from_disk(ip, dip);
2879 ASSERT(ip->i_d.di_version >= 3);
2881 error = xfs_iformat_fork(ip, dip);
2885 if (!xfs_inode_verify_forks(ip)) {
2886 error = -EFSCORRUPTED;
2890 if (in_f->ilf_fields & XFS_ILOG_DOWNER) {
2891 ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT);
2892 error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK,
2893 ip->i_ino, buffer_list);
2898 if (in_f->ilf_fields & XFS_ILOG_AOWNER) {
2899 ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT);
2900 error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK,
2901 ip->i_ino, buffer_list);
2912 xlog_recover_inode_pass2(
2914 struct list_head *buffer_list,
2915 struct xlog_recover_item *item,
2916 xfs_lsn_t current_lsn)
2918 struct xfs_inode_log_format *in_f;
2919 xfs_mount_t *mp = log->l_mp;
2928 struct xfs_log_dinode *ldip;
2932 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
2933 in_f = item->ri_buf[0].i_addr;
2935 in_f = kmem_alloc(sizeof(struct xfs_inode_log_format), 0);
2937 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2943 * Inode buffers can be freed, look out for it,
2944 * and do not replay the inode.
2946 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2947 in_f->ilf_len, 0)) {
2949 trace_xfs_log_recover_inode_cancel(log, in_f);
2952 trace_xfs_log_recover_inode_recover(log, in_f);
2954 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
2955 &xfs_inode_buf_ops);
2960 error = bp->b_error;
2962 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
2965 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2966 dip = xfs_buf_offset(bp, in_f->ilf_boffset);
2969 * Make sure the place we're flushing out to really looks
2972 if (unlikely(!xfs_verify_magic16(bp, dip->di_magic))) {
2974 "%s: Bad inode magic number, dip = "PTR_FMT", dino bp = "PTR_FMT", ino = %Ld",
2975 __func__, dip, bp, in_f->ilf_ino);
2976 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2977 XFS_ERRLEVEL_LOW, mp);
2978 error = -EFSCORRUPTED;
2981 ldip = item->ri_buf[1].i_addr;
2982 if (unlikely(ldip->di_magic != XFS_DINODE_MAGIC)) {
2984 "%s: Bad inode log record, rec ptr "PTR_FMT", ino %Ld",
2985 __func__, item, in_f->ilf_ino);
2986 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2987 XFS_ERRLEVEL_LOW, mp);
2988 error = -EFSCORRUPTED;
2993 * If the inode has an LSN in it, recover the inode only if it's less
2994 * than the lsn of the transaction we are replaying. Note: we still
2995 * need to replay an owner change even though the inode is more recent
2996 * than the transaction as there is no guarantee that all the btree
2997 * blocks are more recent than this transaction, too.
2999 if (dip->di_version >= 3) {
3000 xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn);
3002 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3003 trace_xfs_log_recover_inode_skip(log, in_f);
3005 goto out_owner_change;
3010 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
3011 * are transactional and if ordering is necessary we can determine that
3012 * more accurately by the LSN field in the V3 inode core. Don't trust
3013 * the inode versions we might be changing them here - use the
3014 * superblock flag to determine whether we need to look at di_flushiter
3015 * to skip replay when the on disk inode is newer than the log one
3017 if (!xfs_sb_version_hascrc(&mp->m_sb) &&
3018 ldip->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
3020 * Deal with the wrap case, DI_MAX_FLUSH is less
3021 * than smaller numbers
3023 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
3024 ldip->di_flushiter < (DI_MAX_FLUSH >> 1)) {
3027 trace_xfs_log_recover_inode_skip(log, in_f);
3033 /* Take the opportunity to reset the flush iteration count */
3034 ldip->di_flushiter = 0;
3036 if (unlikely(S_ISREG(ldip->di_mode))) {
3037 if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) &&
3038 (ldip->di_format != XFS_DINODE_FMT_BTREE)) {
3039 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
3040 XFS_ERRLEVEL_LOW, mp, ldip,
3043 "%s: Bad regular inode log record, rec ptr "PTR_FMT", "
3044 "ino ptr = "PTR_FMT", ino bp = "PTR_FMT", ino %Ld",
3045 __func__, item, dip, bp, in_f->ilf_ino);
3046 error = -EFSCORRUPTED;
3049 } else if (unlikely(S_ISDIR(ldip->di_mode))) {
3050 if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) &&
3051 (ldip->di_format != XFS_DINODE_FMT_BTREE) &&
3052 (ldip->di_format != XFS_DINODE_FMT_LOCAL)) {
3053 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
3054 XFS_ERRLEVEL_LOW, mp, ldip,
3057 "%s: Bad dir inode log record, rec ptr "PTR_FMT", "
3058 "ino ptr = "PTR_FMT", ino bp = "PTR_FMT", ino %Ld",
3059 __func__, item, dip, bp, in_f->ilf_ino);
3060 error = -EFSCORRUPTED;
3064 if (unlikely(ldip->di_nextents + ldip->di_anextents > ldip->di_nblocks)){
3065 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
3066 XFS_ERRLEVEL_LOW, mp, ldip,
3069 "%s: Bad inode log record, rec ptr "PTR_FMT", dino ptr "PTR_FMT", "
3070 "dino bp "PTR_FMT", ino %Ld, total extents = %d, nblocks = %Ld",
3071 __func__, item, dip, bp, in_f->ilf_ino,
3072 ldip->di_nextents + ldip->di_anextents,
3074 error = -EFSCORRUPTED;
3077 if (unlikely(ldip->di_forkoff > mp->m_sb.sb_inodesize)) {
3078 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
3079 XFS_ERRLEVEL_LOW, mp, ldip,
3082 "%s: Bad inode log record, rec ptr "PTR_FMT", dino ptr "PTR_FMT", "
3083 "dino bp "PTR_FMT", ino %Ld, forkoff 0x%x", __func__,
3084 item, dip, bp, in_f->ilf_ino, ldip->di_forkoff);
3085 error = -EFSCORRUPTED;
3088 isize = xfs_log_dinode_size(ldip->di_version);
3089 if (unlikely(item->ri_buf[1].i_len > isize)) {
3090 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
3091 XFS_ERRLEVEL_LOW, mp, ldip,
3094 "%s: Bad inode log record length %d, rec ptr "PTR_FMT,
3095 __func__, item->ri_buf[1].i_len, item);
3096 error = -EFSCORRUPTED;
3100 /* recover the log dinode inode into the on disk inode */
3101 xfs_log_dinode_to_disk(ldip, dip);
3103 fields = in_f->ilf_fields;
3104 if (fields & XFS_ILOG_DEV)
3105 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
3107 if (in_f->ilf_size == 2)
3108 goto out_owner_change;
3109 len = item->ri_buf[2].i_len;
3110 src = item->ri_buf[2].i_addr;
3111 ASSERT(in_f->ilf_size <= 4);
3112 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
3113 ASSERT(!(fields & XFS_ILOG_DFORK) ||
3114 (len == in_f->ilf_dsize));
3116 switch (fields & XFS_ILOG_DFORK) {
3117 case XFS_ILOG_DDATA:
3119 memcpy(XFS_DFORK_DPTR(dip), src, len);
3122 case XFS_ILOG_DBROOT:
3123 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
3124 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
3125 XFS_DFORK_DSIZE(dip, mp));
3130 * There are no data fork flags set.
3132 ASSERT((fields & XFS_ILOG_DFORK) == 0);
3137 * If we logged any attribute data, recover it. There may or
3138 * may not have been any other non-core data logged in this
3141 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
3142 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
3147 len = item->ri_buf[attr_index].i_len;
3148 src = item->ri_buf[attr_index].i_addr;
3149 ASSERT(len == in_f->ilf_asize);
3151 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
3152 case XFS_ILOG_ADATA:
3154 dest = XFS_DFORK_APTR(dip);
3155 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
3156 memcpy(dest, src, len);
3159 case XFS_ILOG_ABROOT:
3160 dest = XFS_DFORK_APTR(dip);
3161 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
3162 len, (xfs_bmdr_block_t*)dest,
3163 XFS_DFORK_ASIZE(dip, mp));
3167 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
3169 error = -EFSCORRUPTED;
3175 /* Recover the swapext owner change unless inode has been deleted */
3176 if ((in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER)) &&
3177 (dip->di_mode != 0))
3178 error = xfs_recover_inode_owner_change(mp, dip, in_f,
3180 /* re-generate the checksum. */
3181 xfs_dinode_calc_crc(log->l_mp, dip);
3183 ASSERT(bp->b_mount == mp);
3184 bp->b_iodone = xlog_recover_iodone;
3185 xfs_buf_delwri_queue(bp, buffer_list);
3196 * Recover QUOTAOFF records. We simply make a note of it in the xlog
3197 * structure, so that we know not to do any dquot item or dquot buffer recovery,
3201 xlog_recover_quotaoff_pass1(
3203 struct xlog_recover_item *item)
3205 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
3209 * The logitem format's flag tells us if this was user quotaoff,
3210 * group/project quotaoff or both.
3212 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
3213 log->l_quotaoffs_flag |= XFS_DQ_USER;
3214 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
3215 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
3216 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
3217 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
3223 * Recover a dquot record
3226 xlog_recover_dquot_pass2(
3228 struct list_head *buffer_list,
3229 struct xlog_recover_item *item,
3230 xfs_lsn_t current_lsn)
3232 xfs_mount_t *mp = log->l_mp;
3234 struct xfs_disk_dquot *ddq, *recddq;
3237 xfs_dq_logformat_t *dq_f;
3242 * Filesystems are required to send in quota flags at mount time.
3244 if (mp->m_qflags == 0)
3247 recddq = item->ri_buf[1].i_addr;
3248 if (recddq == NULL) {
3249 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
3250 return -EFSCORRUPTED;
3252 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
3253 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
3254 item->ri_buf[1].i_len, __func__);
3255 return -EFSCORRUPTED;
3259 * This type of quotas was turned off, so ignore this record.
3261 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3263 if (log->l_quotaoffs_flag & type)
3267 * At this point we know that quota was _not_ turned off.
3268 * Since the mount flags are not indicating to us otherwise, this
3269 * must mean that quota is on, and the dquot needs to be replayed.
3270 * Remember that we may not have fully recovered the superblock yet,
3271 * so we can't do the usual trick of looking at the SB quota bits.
3273 * The other possibility, of course, is that the quota subsystem was
3274 * removed since the last mount - ENOSYS.
3276 dq_f = item->ri_buf[0].i_addr;
3278 fa = xfs_dquot_verify(mp, recddq, dq_f->qlf_id, 0);
3280 xfs_alert(mp, "corrupt dquot ID 0x%x in log at %pS",
3282 return -EFSCORRUPTED;
3284 ASSERT(dq_f->qlf_len == 1);
3287 * At this point we are assuming that the dquots have been allocated
3288 * and hence the buffer has valid dquots stamped in it. It should,
3289 * therefore, pass verifier validation. If the dquot is bad, then the
3290 * we'll return an error here, so we don't need to specifically check
3291 * the dquot in the buffer after the verifier has run.
3293 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
3294 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
3295 &xfs_dquot_buf_ops);
3300 ddq = xfs_buf_offset(bp, dq_f->qlf_boffset);
3303 * If the dquot has an LSN in it, recover the dquot only if it's less
3304 * than the lsn of the transaction we are replaying.
3306 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3307 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq;
3308 xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn);
3310 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3315 memcpy(ddq, recddq, item->ri_buf[1].i_len);
3316 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3317 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
3321 ASSERT(dq_f->qlf_size == 2);
3322 ASSERT(bp->b_mount == mp);
3323 bp->b_iodone = xlog_recover_iodone;
3324 xfs_buf_delwri_queue(bp, buffer_list);
3332 * This routine is called to create an in-core extent free intent
3333 * item from the efi format structure which was logged on disk.
3334 * It allocates an in-core efi, copies the extents from the format
3335 * structure into it, and adds the efi to the AIL with the given
3339 xlog_recover_efi_pass2(
3341 struct xlog_recover_item *item,
3345 struct xfs_mount *mp = log->l_mp;
3346 struct xfs_efi_log_item *efip;
3347 struct xfs_efi_log_format *efi_formatp;
3349 efi_formatp = item->ri_buf[0].i_addr;
3351 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
3352 error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format);
3354 xfs_efi_item_free(efip);
3357 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
3359 spin_lock(&log->l_ailp->ail_lock);
3361 * The EFI has two references. One for the EFD and one for EFI to ensure
3362 * it makes it into the AIL. Insert the EFI into the AIL directly and
3363 * drop the EFI reference. Note that xfs_trans_ail_update() drops the
3366 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
3367 xfs_efi_release(efip);
3373 * This routine is called when an EFD format structure is found in a committed
3374 * transaction in the log. Its purpose is to cancel the corresponding EFI if it
3375 * was still in the log. To do this it searches the AIL for the EFI with an id
3376 * equal to that in the EFD format structure. If we find it we drop the EFD
3377 * reference, which removes the EFI from the AIL and frees it.
3380 xlog_recover_efd_pass2(
3382 struct xlog_recover_item *item)
3384 xfs_efd_log_format_t *efd_formatp;
3385 xfs_efi_log_item_t *efip = NULL;
3386 struct xfs_log_item *lip;
3388 struct xfs_ail_cursor cur;
3389 struct xfs_ail *ailp = log->l_ailp;
3391 efd_formatp = item->ri_buf[0].i_addr;
3392 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
3393 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
3394 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
3395 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
3396 efi_id = efd_formatp->efd_efi_id;
3399 * Search for the EFI with the id in the EFD format structure in the
3402 spin_lock(&ailp->ail_lock);
3403 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3404 while (lip != NULL) {
3405 if (lip->li_type == XFS_LI_EFI) {
3406 efip = (xfs_efi_log_item_t *)lip;
3407 if (efip->efi_format.efi_id == efi_id) {
3409 * Drop the EFD reference to the EFI. This
3410 * removes the EFI from the AIL and frees it.
3412 spin_unlock(&ailp->ail_lock);
3413 xfs_efi_release(efip);
3414 spin_lock(&ailp->ail_lock);
3418 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3421 xfs_trans_ail_cursor_done(&cur);
3422 spin_unlock(&ailp->ail_lock);
3428 * This routine is called to create an in-core extent rmap update
3429 * item from the rui format structure which was logged on disk.
3430 * It allocates an in-core rui, copies the extents from the format
3431 * structure into it, and adds the rui to the AIL with the given
3435 xlog_recover_rui_pass2(
3437 struct xlog_recover_item *item,
3441 struct xfs_mount *mp = log->l_mp;
3442 struct xfs_rui_log_item *ruip;
3443 struct xfs_rui_log_format *rui_formatp;
3445 rui_formatp = item->ri_buf[0].i_addr;
3447 ruip = xfs_rui_init(mp, rui_formatp->rui_nextents);
3448 error = xfs_rui_copy_format(&item->ri_buf[0], &ruip->rui_format);
3450 xfs_rui_item_free(ruip);
3453 atomic_set(&ruip->rui_next_extent, rui_formatp->rui_nextents);
3455 spin_lock(&log->l_ailp->ail_lock);
3457 * The RUI has two references. One for the RUD and one for RUI to ensure
3458 * it makes it into the AIL. Insert the RUI into the AIL directly and
3459 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
3462 xfs_trans_ail_update(log->l_ailp, &ruip->rui_item, lsn);
3463 xfs_rui_release(ruip);
3469 * This routine is called when an RUD format structure is found in a committed
3470 * transaction in the log. Its purpose is to cancel the corresponding RUI if it
3471 * was still in the log. To do this it searches the AIL for the RUI with an id
3472 * equal to that in the RUD format structure. If we find it we drop the RUD
3473 * reference, which removes the RUI from the AIL and frees it.
3476 xlog_recover_rud_pass2(
3478 struct xlog_recover_item *item)
3480 struct xfs_rud_log_format *rud_formatp;
3481 struct xfs_rui_log_item *ruip = NULL;
3482 struct xfs_log_item *lip;
3484 struct xfs_ail_cursor cur;
3485 struct xfs_ail *ailp = log->l_ailp;
3487 rud_formatp = item->ri_buf[0].i_addr;
3488 ASSERT(item->ri_buf[0].i_len == sizeof(struct xfs_rud_log_format));
3489 rui_id = rud_formatp->rud_rui_id;
3492 * Search for the RUI with the id in the RUD format structure in the
3495 spin_lock(&ailp->ail_lock);
3496 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3497 while (lip != NULL) {
3498 if (lip->li_type == XFS_LI_RUI) {
3499 ruip = (struct xfs_rui_log_item *)lip;
3500 if (ruip->rui_format.rui_id == rui_id) {
3502 * Drop the RUD reference to the RUI. This
3503 * removes the RUI from the AIL and frees it.
3505 spin_unlock(&ailp->ail_lock);
3506 xfs_rui_release(ruip);
3507 spin_lock(&ailp->ail_lock);
3511 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3514 xfs_trans_ail_cursor_done(&cur);
3515 spin_unlock(&ailp->ail_lock);
3521 * Copy an CUI format buffer from the given buf, and into the destination
3522 * CUI format structure. The CUI/CUD items were designed not to need any
3523 * special alignment handling.
3526 xfs_cui_copy_format(
3527 struct xfs_log_iovec *buf,
3528 struct xfs_cui_log_format *dst_cui_fmt)
3530 struct xfs_cui_log_format *src_cui_fmt;
3533 src_cui_fmt = buf->i_addr;
3534 len = xfs_cui_log_format_sizeof(src_cui_fmt->cui_nextents);
3536 if (buf->i_len == len) {
3537 memcpy(dst_cui_fmt, src_cui_fmt, len);
3540 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, NULL);
3541 return -EFSCORRUPTED;
3545 * This routine is called to create an in-core extent refcount update
3546 * item from the cui format structure which was logged on disk.
3547 * It allocates an in-core cui, copies the extents from the format
3548 * structure into it, and adds the cui to the AIL with the given
3552 xlog_recover_cui_pass2(
3554 struct xlog_recover_item *item,
3558 struct xfs_mount *mp = log->l_mp;
3559 struct xfs_cui_log_item *cuip;
3560 struct xfs_cui_log_format *cui_formatp;
3562 cui_formatp = item->ri_buf[0].i_addr;
3564 cuip = xfs_cui_init(mp, cui_formatp->cui_nextents);
3565 error = xfs_cui_copy_format(&item->ri_buf[0], &cuip->cui_format);
3567 xfs_cui_item_free(cuip);
3570 atomic_set(&cuip->cui_next_extent, cui_formatp->cui_nextents);
3572 spin_lock(&log->l_ailp->ail_lock);
3574 * The CUI has two references. One for the CUD and one for CUI to ensure
3575 * it makes it into the AIL. Insert the CUI into the AIL directly and
3576 * drop the CUI reference. Note that xfs_trans_ail_update() drops the
3579 xfs_trans_ail_update(log->l_ailp, &cuip->cui_item, lsn);
3580 xfs_cui_release(cuip);
3586 * This routine is called when an CUD format structure is found in a committed
3587 * transaction in the log. Its purpose is to cancel the corresponding CUI if it
3588 * was still in the log. To do this it searches the AIL for the CUI with an id
3589 * equal to that in the CUD format structure. If we find it we drop the CUD
3590 * reference, which removes the CUI from the AIL and frees it.
3593 xlog_recover_cud_pass2(
3595 struct xlog_recover_item *item)
3597 struct xfs_cud_log_format *cud_formatp;
3598 struct xfs_cui_log_item *cuip = NULL;
3599 struct xfs_log_item *lip;
3601 struct xfs_ail_cursor cur;
3602 struct xfs_ail *ailp = log->l_ailp;
3604 cud_formatp = item->ri_buf[0].i_addr;
3605 if (item->ri_buf[0].i_len != sizeof(struct xfs_cud_log_format)) {
3606 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, log->l_mp);
3607 return -EFSCORRUPTED;
3609 cui_id = cud_formatp->cud_cui_id;
3612 * Search for the CUI with the id in the CUD format structure in the
3615 spin_lock(&ailp->ail_lock);
3616 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3617 while (lip != NULL) {
3618 if (lip->li_type == XFS_LI_CUI) {
3619 cuip = (struct xfs_cui_log_item *)lip;
3620 if (cuip->cui_format.cui_id == cui_id) {
3622 * Drop the CUD reference to the CUI. This
3623 * removes the CUI from the AIL and frees it.
3625 spin_unlock(&ailp->ail_lock);
3626 xfs_cui_release(cuip);
3627 spin_lock(&ailp->ail_lock);
3631 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3634 xfs_trans_ail_cursor_done(&cur);
3635 spin_unlock(&ailp->ail_lock);
3641 * Copy an BUI format buffer from the given buf, and into the destination
3642 * BUI format structure. The BUI/BUD items were designed not to need any
3643 * special alignment handling.
3646 xfs_bui_copy_format(
3647 struct xfs_log_iovec *buf,
3648 struct xfs_bui_log_format *dst_bui_fmt)
3650 struct xfs_bui_log_format *src_bui_fmt;
3653 src_bui_fmt = buf->i_addr;
3654 len = xfs_bui_log_format_sizeof(src_bui_fmt->bui_nextents);
3656 if (buf->i_len == len) {
3657 memcpy(dst_bui_fmt, src_bui_fmt, len);
3660 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, NULL);
3661 return -EFSCORRUPTED;
3665 * This routine is called to create an in-core extent bmap update
3666 * item from the bui format structure which was logged on disk.
3667 * It allocates an in-core bui, copies the extents from the format
3668 * structure into it, and adds the bui to the AIL with the given
3672 xlog_recover_bui_pass2(
3674 struct xlog_recover_item *item,
3678 struct xfs_mount *mp = log->l_mp;
3679 struct xfs_bui_log_item *buip;
3680 struct xfs_bui_log_format *bui_formatp;
3682 bui_formatp = item->ri_buf[0].i_addr;
3684 if (bui_formatp->bui_nextents != XFS_BUI_MAX_FAST_EXTENTS) {
3685 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, log->l_mp);
3686 return -EFSCORRUPTED;
3688 buip = xfs_bui_init(mp);
3689 error = xfs_bui_copy_format(&item->ri_buf[0], &buip->bui_format);
3691 xfs_bui_item_free(buip);
3694 atomic_set(&buip->bui_next_extent, bui_formatp->bui_nextents);
3696 spin_lock(&log->l_ailp->ail_lock);
3698 * The RUI has two references. One for the RUD and one for RUI to ensure
3699 * it makes it into the AIL. Insert the RUI into the AIL directly and
3700 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
3703 xfs_trans_ail_update(log->l_ailp, &buip->bui_item, lsn);
3704 xfs_bui_release(buip);
3710 * This routine is called when an BUD format structure is found in a committed
3711 * transaction in the log. Its purpose is to cancel the corresponding BUI if it
3712 * was still in the log. To do this it searches the AIL for the BUI with an id
3713 * equal to that in the BUD format structure. If we find it we drop the BUD
3714 * reference, which removes the BUI from the AIL and frees it.
3717 xlog_recover_bud_pass2(
3719 struct xlog_recover_item *item)
3721 struct xfs_bud_log_format *bud_formatp;
3722 struct xfs_bui_log_item *buip = NULL;
3723 struct xfs_log_item *lip;
3725 struct xfs_ail_cursor cur;
3726 struct xfs_ail *ailp = log->l_ailp;
3728 bud_formatp = item->ri_buf[0].i_addr;
3729 if (item->ri_buf[0].i_len != sizeof(struct xfs_bud_log_format)) {
3730 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, log->l_mp);
3731 return -EFSCORRUPTED;
3733 bui_id = bud_formatp->bud_bui_id;
3736 * Search for the BUI with the id in the BUD format structure in the
3739 spin_lock(&ailp->ail_lock);
3740 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3741 while (lip != NULL) {
3742 if (lip->li_type == XFS_LI_BUI) {
3743 buip = (struct xfs_bui_log_item *)lip;
3744 if (buip->bui_format.bui_id == bui_id) {
3746 * Drop the BUD reference to the BUI. This
3747 * removes the BUI from the AIL and frees it.
3749 spin_unlock(&ailp->ail_lock);
3750 xfs_bui_release(buip);
3751 spin_lock(&ailp->ail_lock);
3755 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3758 xfs_trans_ail_cursor_done(&cur);
3759 spin_unlock(&ailp->ail_lock);
3765 * This routine is called when an inode create format structure is found in a
3766 * committed transaction in the log. It's purpose is to initialise the inodes
3767 * being allocated on disk. This requires us to get inode cluster buffers that
3768 * match the range to be initialised, stamped with inode templates and written
3769 * by delayed write so that subsequent modifications will hit the cached buffer
3770 * and only need writing out at the end of recovery.
3773 xlog_recover_do_icreate_pass2(
3775 struct list_head *buffer_list,
3776 xlog_recover_item_t *item)
3778 struct xfs_mount *mp = log->l_mp;
3779 struct xfs_icreate_log *icl;
3780 struct xfs_ino_geometry *igeo = M_IGEO(mp);
3781 xfs_agnumber_t agno;
3782 xfs_agblock_t agbno;
3785 xfs_agblock_t length;
3791 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3792 if (icl->icl_type != XFS_LI_ICREATE) {
3793 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
3797 if (icl->icl_size != 1) {
3798 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
3802 agno = be32_to_cpu(icl->icl_ag);
3803 if (agno >= mp->m_sb.sb_agcount) {
3804 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
3807 agbno = be32_to_cpu(icl->icl_agbno);
3808 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3809 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
3812 isize = be32_to_cpu(icl->icl_isize);
3813 if (isize != mp->m_sb.sb_inodesize) {
3814 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
3817 count = be32_to_cpu(icl->icl_count);
3819 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
3822 length = be32_to_cpu(icl->icl_length);
3823 if (!length || length >= mp->m_sb.sb_agblocks) {
3824 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
3829 * The inode chunk is either full or sparse and we only support
3830 * m_ino_geo.ialloc_min_blks sized sparse allocations at this time.
3832 if (length != igeo->ialloc_blks &&
3833 length != igeo->ialloc_min_blks) {
3835 "%s: unsupported chunk length", __FUNCTION__);
3839 /* verify inode count is consistent with extent length */
3840 if ((count >> mp->m_sb.sb_inopblog) != length) {
3842 "%s: inconsistent inode count and chunk length",
3848 * The icreate transaction can cover multiple cluster buffers and these
3849 * buffers could have been freed and reused. Check the individual
3850 * buffers for cancellation so we don't overwrite anything written after
3853 bb_per_cluster = XFS_FSB_TO_BB(mp, igeo->blocks_per_cluster);
3854 nbufs = length / igeo->blocks_per_cluster;
3855 for (i = 0, cancel_count = 0; i < nbufs; i++) {
3858 daddr = XFS_AGB_TO_DADDR(mp, agno,
3859 agbno + i * igeo->blocks_per_cluster);
3860 if (xlog_check_buffer_cancelled(log, daddr, bb_per_cluster, 0))
3865 * We currently only use icreate for a single allocation at a time. This
3866 * means we should expect either all or none of the buffers to be
3867 * cancelled. Be conservative and skip replay if at least one buffer is
3868 * cancelled, but warn the user that something is awry if the buffers
3869 * are not consistent.
3871 * XXX: This must be refined to only skip cancelled clusters once we use
3872 * icreate for multiple chunk allocations.
3874 ASSERT(!cancel_count || cancel_count == nbufs);
3876 if (cancel_count != nbufs)
3878 "WARNING: partial inode chunk cancellation, skipped icreate.");
3879 trace_xfs_log_recover_icreate_cancel(log, icl);
3883 trace_xfs_log_recover_icreate_recover(log, icl);
3884 return xfs_ialloc_inode_init(mp, NULL, buffer_list, count, agno, agbno,
3885 length, be32_to_cpu(icl->icl_gen));
3889 xlog_recover_buffer_ra_pass2(
3891 struct xlog_recover_item *item)
3893 struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr;
3894 struct xfs_mount *mp = log->l_mp;
3896 if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
3897 buf_f->blf_len, buf_f->blf_flags)) {
3901 xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
3902 buf_f->blf_len, NULL);
3906 xlog_recover_inode_ra_pass2(
3908 struct xlog_recover_item *item)
3910 struct xfs_inode_log_format ilf_buf;
3911 struct xfs_inode_log_format *ilfp;
3912 struct xfs_mount *mp = log->l_mp;
3915 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3916 ilfp = item->ri_buf[0].i_addr;
3919 memset(ilfp, 0, sizeof(*ilfp));
3920 error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
3925 if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
3928 xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
3929 ilfp->ilf_len, &xfs_inode_buf_ra_ops);
3933 xlog_recover_dquot_ra_pass2(
3935 struct xlog_recover_item *item)
3937 struct xfs_mount *mp = log->l_mp;
3938 struct xfs_disk_dquot *recddq;
3939 struct xfs_dq_logformat *dq_f;
3944 if (mp->m_qflags == 0)
3947 recddq = item->ri_buf[1].i_addr;
3950 if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
3953 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3955 if (log->l_quotaoffs_flag & type)
3958 dq_f = item->ri_buf[0].i_addr;
3960 ASSERT(dq_f->qlf_len == 1);
3962 len = XFS_FSB_TO_BB(mp, dq_f->qlf_len);
3963 if (xlog_peek_buffer_cancelled(log, dq_f->qlf_blkno, len, 0))
3966 xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno, len,
3967 &xfs_dquot_buf_ra_ops);
3971 xlog_recover_ra_pass2(
3973 struct xlog_recover_item *item)
3975 switch (ITEM_TYPE(item)) {
3977 xlog_recover_buffer_ra_pass2(log, item);
3980 xlog_recover_inode_ra_pass2(log, item);
3983 xlog_recover_dquot_ra_pass2(log, item);
3987 case XFS_LI_QUOTAOFF:
4000 xlog_recover_commit_pass1(
4002 struct xlog_recover *trans,
4003 struct xlog_recover_item *item)
4005 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
4007 switch (ITEM_TYPE(item)) {
4009 return xlog_recover_buffer_pass1(log, item);
4010 case XFS_LI_QUOTAOFF:
4011 return xlog_recover_quotaoff_pass1(log, item);
4016 case XFS_LI_ICREATE:
4023 /* nothing to do in pass 1 */
4026 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
4027 __func__, ITEM_TYPE(item));
4029 return -EFSCORRUPTED;
4034 xlog_recover_commit_pass2(
4036 struct xlog_recover *trans,
4037 struct list_head *buffer_list,
4038 struct xlog_recover_item *item)
4040 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
4042 switch (ITEM_TYPE(item)) {
4044 return xlog_recover_buffer_pass2(log, buffer_list, item,
4047 return xlog_recover_inode_pass2(log, buffer_list, item,
4050 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
4052 return xlog_recover_efd_pass2(log, item);
4054 return xlog_recover_rui_pass2(log, item, trans->r_lsn);
4056 return xlog_recover_rud_pass2(log, item);
4058 return xlog_recover_cui_pass2(log, item, trans->r_lsn);
4060 return xlog_recover_cud_pass2(log, item);
4062 return xlog_recover_bui_pass2(log, item, trans->r_lsn);
4064 return xlog_recover_bud_pass2(log, item);
4066 return xlog_recover_dquot_pass2(log, buffer_list, item,
4068 case XFS_LI_ICREATE:
4069 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
4070 case XFS_LI_QUOTAOFF:
4071 /* nothing to do in pass2 */
4074 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
4075 __func__, ITEM_TYPE(item));
4077 return -EFSCORRUPTED;
4082 xlog_recover_items_pass2(
4084 struct xlog_recover *trans,
4085 struct list_head *buffer_list,
4086 struct list_head *item_list)
4088 struct xlog_recover_item *item;
4091 list_for_each_entry(item, item_list, ri_list) {
4092 error = xlog_recover_commit_pass2(log, trans,
4102 * Perform the transaction.
4104 * If the transaction modifies a buffer or inode, do it now. Otherwise,
4105 * EFIs and EFDs get queued up by adding entries into the AIL for them.
4108 xlog_recover_commit_trans(
4110 struct xlog_recover *trans,
4112 struct list_head *buffer_list)
4115 int items_queued = 0;
4116 struct xlog_recover_item *item;
4117 struct xlog_recover_item *next;
4118 LIST_HEAD (ra_list);
4119 LIST_HEAD (done_list);
4121 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
4123 hlist_del_init(&trans->r_list);
4125 error = xlog_recover_reorder_trans(log, trans, pass);
4129 list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
4131 case XLOG_RECOVER_PASS1:
4132 error = xlog_recover_commit_pass1(log, trans, item);
4134 case XLOG_RECOVER_PASS2:
4135 xlog_recover_ra_pass2(log, item);
4136 list_move_tail(&item->ri_list, &ra_list);
4138 if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
4139 error = xlog_recover_items_pass2(log, trans,
4140 buffer_list, &ra_list);
4141 list_splice_tail_init(&ra_list, &done_list);
4155 if (!list_empty(&ra_list)) {
4157 error = xlog_recover_items_pass2(log, trans,
4158 buffer_list, &ra_list);
4159 list_splice_tail_init(&ra_list, &done_list);
4162 if (!list_empty(&done_list))
4163 list_splice_init(&done_list, &trans->r_itemq);
4169 xlog_recover_add_item(
4170 struct list_head *head)
4172 xlog_recover_item_t *item;
4174 item = kmem_zalloc(sizeof(xlog_recover_item_t), 0);
4175 INIT_LIST_HEAD(&item->ri_list);
4176 list_add_tail(&item->ri_list, head);
4180 xlog_recover_add_to_cont_trans(
4182 struct xlog_recover *trans,
4186 xlog_recover_item_t *item;
4187 char *ptr, *old_ptr;
4191 * If the transaction is empty, the header was split across this and the
4192 * previous record. Copy the rest of the header.
4194 if (list_empty(&trans->r_itemq)) {
4195 ASSERT(len <= sizeof(struct xfs_trans_header));
4196 if (len > sizeof(struct xfs_trans_header)) {
4197 xfs_warn(log->l_mp, "%s: bad header length", __func__);
4198 return -EFSCORRUPTED;
4201 xlog_recover_add_item(&trans->r_itemq);
4202 ptr = (char *)&trans->r_theader +
4203 sizeof(struct xfs_trans_header) - len;
4204 memcpy(ptr, dp, len);
4208 /* take the tail entry */
4209 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
4211 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
4212 old_len = item->ri_buf[item->ri_cnt-1].i_len;
4214 ptr = kmem_realloc(old_ptr, len + old_len, 0);
4215 memcpy(&ptr[old_len], dp, len);
4216 item->ri_buf[item->ri_cnt-1].i_len += len;
4217 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
4218 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
4223 * The next region to add is the start of a new region. It could be
4224 * a whole region or it could be the first part of a new region. Because
4225 * of this, the assumption here is that the type and size fields of all
4226 * format structures fit into the first 32 bits of the structure.
4228 * This works because all regions must be 32 bit aligned. Therefore, we
4229 * either have both fields or we have neither field. In the case we have
4230 * neither field, the data part of the region is zero length. We only have
4231 * a log_op_header and can throw away the header since a new one will appear
4232 * later. If we have at least 4 bytes, then we can determine how many regions
4233 * will appear in the current log item.
4236 xlog_recover_add_to_trans(
4238 struct xlog_recover *trans,
4242 struct xfs_inode_log_format *in_f; /* any will do */
4243 xlog_recover_item_t *item;
4248 if (list_empty(&trans->r_itemq)) {
4249 /* we need to catch log corruptions here */
4250 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
4251 xfs_warn(log->l_mp, "%s: bad header magic number",
4254 return -EFSCORRUPTED;
4257 if (len > sizeof(struct xfs_trans_header)) {
4258 xfs_warn(log->l_mp, "%s: bad header length", __func__);
4260 return -EFSCORRUPTED;
4264 * The transaction header can be arbitrarily split across op
4265 * records. If we don't have the whole thing here, copy what we
4266 * do have and handle the rest in the next record.
4268 if (len == sizeof(struct xfs_trans_header))
4269 xlog_recover_add_item(&trans->r_itemq);
4270 memcpy(&trans->r_theader, dp, len);
4274 ptr = kmem_alloc(len, 0);
4275 memcpy(ptr, dp, len);
4276 in_f = (struct xfs_inode_log_format *)ptr;
4278 /* take the tail entry */
4279 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
4280 if (item->ri_total != 0 &&
4281 item->ri_total == item->ri_cnt) {
4282 /* tail item is in use, get a new one */
4283 xlog_recover_add_item(&trans->r_itemq);
4284 item = list_entry(trans->r_itemq.prev,
4285 xlog_recover_item_t, ri_list);
4288 if (item->ri_total == 0) { /* first region to be added */
4289 if (in_f->ilf_size == 0 ||
4290 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
4292 "bad number of regions (%d) in inode log format",
4296 return -EFSCORRUPTED;
4299 item->ri_total = in_f->ilf_size;
4301 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
4305 if (item->ri_total <= item->ri_cnt) {
4307 "log item region count (%d) overflowed size (%d)",
4308 item->ri_cnt, item->ri_total);
4311 return -EFSCORRUPTED;
4314 /* Description region is ri_buf[0] */
4315 item->ri_buf[item->ri_cnt].i_addr = ptr;
4316 item->ri_buf[item->ri_cnt].i_len = len;
4318 trace_xfs_log_recover_item_add(log, trans, item, 0);
4323 * Free up any resources allocated by the transaction
4325 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
4328 xlog_recover_free_trans(
4329 struct xlog_recover *trans)
4331 xlog_recover_item_t *item, *n;
4334 hlist_del_init(&trans->r_list);
4336 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
4337 /* Free the regions in the item. */
4338 list_del(&item->ri_list);
4339 for (i = 0; i < item->ri_cnt; i++)
4340 kmem_free(item->ri_buf[i].i_addr);
4341 /* Free the item itself */
4342 kmem_free(item->ri_buf);
4345 /* Free the transaction recover structure */
4350 * On error or completion, trans is freed.
4353 xlog_recovery_process_trans(
4355 struct xlog_recover *trans,
4360 struct list_head *buffer_list)
4363 bool freeit = false;
4365 /* mask off ophdr transaction container flags */
4366 flags &= ~XLOG_END_TRANS;
4367 if (flags & XLOG_WAS_CONT_TRANS)
4368 flags &= ~XLOG_CONTINUE_TRANS;
4371 * Callees must not free the trans structure. We'll decide if we need to
4372 * free it or not based on the operation being done and it's result.
4375 /* expected flag values */
4377 case XLOG_CONTINUE_TRANS:
4378 error = xlog_recover_add_to_trans(log, trans, dp, len);
4380 case XLOG_WAS_CONT_TRANS:
4381 error = xlog_recover_add_to_cont_trans(log, trans, dp, len);
4383 case XLOG_COMMIT_TRANS:
4384 error = xlog_recover_commit_trans(log, trans, pass,
4386 /* success or fail, we are now done with this transaction. */
4390 /* unexpected flag values */
4391 case XLOG_UNMOUNT_TRANS:
4392 /* just skip trans */
4393 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
4396 case XLOG_START_TRANS:
4398 xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags);
4400 error = -EFSCORRUPTED;
4403 if (error || freeit)
4404 xlog_recover_free_trans(trans);
4409 * Lookup the transaction recovery structure associated with the ID in the
4410 * current ophdr. If the transaction doesn't exist and the start flag is set in
4411 * the ophdr, then allocate a new transaction for future ID matches to find.
4412 * Either way, return what we found during the lookup - an existing transaction
4415 STATIC struct xlog_recover *
4416 xlog_recover_ophdr_to_trans(
4417 struct hlist_head rhash[],
4418 struct xlog_rec_header *rhead,
4419 struct xlog_op_header *ohead)
4421 struct xlog_recover *trans;
4423 struct hlist_head *rhp;
4425 tid = be32_to_cpu(ohead->oh_tid);
4426 rhp = &rhash[XLOG_RHASH(tid)];
4427 hlist_for_each_entry(trans, rhp, r_list) {
4428 if (trans->r_log_tid == tid)
4433 * skip over non-start transaction headers - we could be
4434 * processing slack space before the next transaction starts
4436 if (!(ohead->oh_flags & XLOG_START_TRANS))
4439 ASSERT(be32_to_cpu(ohead->oh_len) == 0);
4442 * This is a new transaction so allocate a new recovery container to
4443 * hold the recovery ops that will follow.
4445 trans = kmem_zalloc(sizeof(struct xlog_recover), 0);
4446 trans->r_log_tid = tid;
4447 trans->r_lsn = be64_to_cpu(rhead->h_lsn);
4448 INIT_LIST_HEAD(&trans->r_itemq);
4449 INIT_HLIST_NODE(&trans->r_list);
4450 hlist_add_head(&trans->r_list, rhp);
4453 * Nothing more to do for this ophdr. Items to be added to this new
4454 * transaction will be in subsequent ophdr containers.
4460 xlog_recover_process_ophdr(
4462 struct hlist_head rhash[],
4463 struct xlog_rec_header *rhead,
4464 struct xlog_op_header *ohead,
4468 struct list_head *buffer_list)
4470 struct xlog_recover *trans;
4474 /* Do we understand who wrote this op? */
4475 if (ohead->oh_clientid != XFS_TRANSACTION &&
4476 ohead->oh_clientid != XFS_LOG) {
4477 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
4478 __func__, ohead->oh_clientid);
4480 return -EFSCORRUPTED;
4484 * Check the ophdr contains all the data it is supposed to contain.
4486 len = be32_to_cpu(ohead->oh_len);
4487 if (dp + len > end) {
4488 xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len);
4490 return -EFSCORRUPTED;
4493 trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead);
4495 /* nothing to do, so skip over this ophdr */
4500 * The recovered buffer queue is drained only once we know that all
4501 * recovery items for the current LSN have been processed. This is
4504 * - Buffer write submission updates the metadata LSN of the buffer.
4505 * - Log recovery skips items with a metadata LSN >= the current LSN of
4506 * the recovery item.
4507 * - Separate recovery items against the same metadata buffer can share
4508 * a current LSN. I.e., consider that the LSN of a recovery item is
4509 * defined as the starting LSN of the first record in which its
4510 * transaction appears, that a record can hold multiple transactions,
4511 * and/or that a transaction can span multiple records.
4513 * In other words, we are allowed to submit a buffer from log recovery
4514 * once per current LSN. Otherwise, we may incorrectly skip recovery
4515 * items and cause corruption.
4517 * We don't know up front whether buffers are updated multiple times per
4518 * LSN. Therefore, track the current LSN of each commit log record as it
4519 * is processed and drain the queue when it changes. Use commit records
4520 * because they are ordered correctly by the logging code.
4522 if (log->l_recovery_lsn != trans->r_lsn &&
4523 ohead->oh_flags & XLOG_COMMIT_TRANS) {
4524 error = xfs_buf_delwri_submit(buffer_list);
4527 log->l_recovery_lsn = trans->r_lsn;
4530 return xlog_recovery_process_trans(log, trans, dp, len,
4531 ohead->oh_flags, pass, buffer_list);
4535 * There are two valid states of the r_state field. 0 indicates that the
4536 * transaction structure is in a normal state. We have either seen the
4537 * start of the transaction or the last operation we added was not a partial
4538 * operation. If the last operation we added to the transaction was a
4539 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
4541 * NOTE: skip LRs with 0 data length.
4544 xlog_recover_process_data(
4546 struct hlist_head rhash[],
4547 struct xlog_rec_header *rhead,
4550 struct list_head *buffer_list)
4552 struct xlog_op_header *ohead;
4557 end = dp + be32_to_cpu(rhead->h_len);
4558 num_logops = be32_to_cpu(rhead->h_num_logops);
4560 /* check the log format matches our own - else we can't recover */
4561 if (xlog_header_check_recover(log->l_mp, rhead))
4564 trace_xfs_log_recover_record(log, rhead, pass);
4565 while ((dp < end) && num_logops) {
4567 ohead = (struct xlog_op_header *)dp;
4568 dp += sizeof(*ohead);
4571 /* errors will abort recovery */
4572 error = xlog_recover_process_ophdr(log, rhash, rhead, ohead,
4573 dp, end, pass, buffer_list);
4577 dp += be32_to_cpu(ohead->oh_len);
4583 /* Recover the EFI if necessary. */
4585 xlog_recover_process_efi(
4586 struct xfs_mount *mp,
4587 struct xfs_ail *ailp,
4588 struct xfs_log_item *lip)
4590 struct xfs_efi_log_item *efip;
4594 * Skip EFIs that we've already processed.
4596 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
4597 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags))
4600 spin_unlock(&ailp->ail_lock);
4601 error = xfs_efi_recover(mp, efip);
4602 spin_lock(&ailp->ail_lock);
4607 /* Release the EFI since we're cancelling everything. */
4609 xlog_recover_cancel_efi(
4610 struct xfs_mount *mp,
4611 struct xfs_ail *ailp,
4612 struct xfs_log_item *lip)
4614 struct xfs_efi_log_item *efip;
4616 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
4618 spin_unlock(&ailp->ail_lock);
4619 xfs_efi_release(efip);
4620 spin_lock(&ailp->ail_lock);
4623 /* Recover the RUI if necessary. */
4625 xlog_recover_process_rui(
4626 struct xfs_mount *mp,
4627 struct xfs_ail *ailp,
4628 struct xfs_log_item *lip)
4630 struct xfs_rui_log_item *ruip;
4634 * Skip RUIs that we've already processed.
4636 ruip = container_of(lip, struct xfs_rui_log_item, rui_item);
4637 if (test_bit(XFS_RUI_RECOVERED, &ruip->rui_flags))
4640 spin_unlock(&ailp->ail_lock);
4641 error = xfs_rui_recover(mp, ruip);
4642 spin_lock(&ailp->ail_lock);
4647 /* Release the RUI since we're cancelling everything. */
4649 xlog_recover_cancel_rui(
4650 struct xfs_mount *mp,
4651 struct xfs_ail *ailp,
4652 struct xfs_log_item *lip)
4654 struct xfs_rui_log_item *ruip;
4656 ruip = container_of(lip, struct xfs_rui_log_item, rui_item);
4658 spin_unlock(&ailp->ail_lock);
4659 xfs_rui_release(ruip);
4660 spin_lock(&ailp->ail_lock);
4663 /* Recover the CUI if necessary. */
4665 xlog_recover_process_cui(
4666 struct xfs_trans *parent_tp,
4667 struct xfs_ail *ailp,
4668 struct xfs_log_item *lip)
4670 struct xfs_cui_log_item *cuip;
4674 * Skip CUIs that we've already processed.
4676 cuip = container_of(lip, struct xfs_cui_log_item, cui_item);
4677 if (test_bit(XFS_CUI_RECOVERED, &cuip->cui_flags))
4680 spin_unlock(&ailp->ail_lock);
4681 error = xfs_cui_recover(parent_tp, cuip);
4682 spin_lock(&ailp->ail_lock);
4687 /* Release the CUI since we're cancelling everything. */
4689 xlog_recover_cancel_cui(
4690 struct xfs_mount *mp,
4691 struct xfs_ail *ailp,
4692 struct xfs_log_item *lip)
4694 struct xfs_cui_log_item *cuip;
4696 cuip = container_of(lip, struct xfs_cui_log_item, cui_item);
4698 spin_unlock(&ailp->ail_lock);
4699 xfs_cui_release(cuip);
4700 spin_lock(&ailp->ail_lock);
4703 /* Recover the BUI if necessary. */
4705 xlog_recover_process_bui(
4706 struct xfs_trans *parent_tp,
4707 struct xfs_ail *ailp,
4708 struct xfs_log_item *lip)
4710 struct xfs_bui_log_item *buip;
4714 * Skip BUIs that we've already processed.
4716 buip = container_of(lip, struct xfs_bui_log_item, bui_item);
4717 if (test_bit(XFS_BUI_RECOVERED, &buip->bui_flags))
4720 spin_unlock(&ailp->ail_lock);
4721 error = xfs_bui_recover(parent_tp, buip);
4722 spin_lock(&ailp->ail_lock);
4727 /* Release the BUI since we're cancelling everything. */
4729 xlog_recover_cancel_bui(
4730 struct xfs_mount *mp,
4731 struct xfs_ail *ailp,
4732 struct xfs_log_item *lip)
4734 struct xfs_bui_log_item *buip;
4736 buip = container_of(lip, struct xfs_bui_log_item, bui_item);
4738 spin_unlock(&ailp->ail_lock);
4739 xfs_bui_release(buip);
4740 spin_lock(&ailp->ail_lock);
4743 /* Is this log item a deferred action intent? */
4744 static inline bool xlog_item_is_intent(struct xfs_log_item *lip)
4746 switch (lip->li_type) {
4757 /* Take all the collected deferred ops and finish them in order. */
4759 xlog_finish_defer_ops(
4760 struct xfs_trans *parent_tp)
4762 struct xfs_mount *mp = parent_tp->t_mountp;
4763 struct xfs_trans *tp;
4769 * We're finishing the defer_ops that accumulated as a result of
4770 * recovering unfinished intent items during log recovery. We
4771 * reserve an itruncate transaction because it is the largest
4772 * permanent transaction type. Since we're the only user of the fs
4773 * right now, take 93% (15/16) of the available free blocks. Use
4774 * weird math to avoid a 64-bit division.
4776 freeblks = percpu_counter_sum(&mp->m_fdblocks);
4779 resblks = min_t(int64_t, UINT_MAX, freeblks);
4780 resblks = (resblks * 15) >> 4;
4781 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, resblks,
4782 0, XFS_TRANS_RESERVE, &tp);
4785 /* transfer all collected dfops to this transaction */
4786 xfs_defer_move(tp, parent_tp);
4788 return xfs_trans_commit(tp);
4792 * When this is called, all of the log intent items which did not have
4793 * corresponding log done items should be in the AIL. What we do now
4794 * is update the data structures associated with each one.
4796 * Since we process the log intent items in normal transactions, they
4797 * will be removed at some point after the commit. This prevents us
4798 * from just walking down the list processing each one. We'll use a
4799 * flag in the intent item to skip those that we've already processed
4800 * and use the AIL iteration mechanism's generation count to try to
4801 * speed this up at least a bit.
4803 * When we start, we know that the intents are the only things in the
4804 * AIL. As we process them, however, other items are added to the
4808 xlog_recover_process_intents(
4811 struct xfs_trans *parent_tp;
4812 struct xfs_ail_cursor cur;
4813 struct xfs_log_item *lip;
4814 struct xfs_ail *ailp;
4816 #if defined(DEBUG) || defined(XFS_WARN)
4821 * The intent recovery handlers commit transactions to complete recovery
4822 * for individual intents, but any new deferred operations that are
4823 * queued during that process are held off until the very end. The
4824 * purpose of this transaction is to serve as a container for deferred
4825 * operations. Each intent recovery handler must transfer dfops here
4826 * before its local transaction commits, and we'll finish the entire
4829 error = xfs_trans_alloc_empty(log->l_mp, &parent_tp);
4834 spin_lock(&ailp->ail_lock);
4835 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
4836 #if defined(DEBUG) || defined(XFS_WARN)
4837 last_lsn = xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block);
4839 while (lip != NULL) {
4841 * We're done when we see something other than an intent.
4842 * There should be no intents left in the AIL now.
4844 if (!xlog_item_is_intent(lip)) {
4846 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
4847 ASSERT(!xlog_item_is_intent(lip));
4853 * We should never see a redo item with a LSN higher than
4854 * the last transaction we found in the log at the start
4857 ASSERT(XFS_LSN_CMP(last_lsn, lip->li_lsn) >= 0);
4860 * NOTE: If your intent processing routine can create more
4861 * deferred ops, you /must/ attach them to the dfops in this
4862 * routine or else those subsequent intents will get
4863 * replayed in the wrong order!
4865 switch (lip->li_type) {
4867 error = xlog_recover_process_efi(log->l_mp, ailp, lip);
4870 error = xlog_recover_process_rui(log->l_mp, ailp, lip);
4873 error = xlog_recover_process_cui(parent_tp, ailp, lip);
4876 error = xlog_recover_process_bui(parent_tp, ailp, lip);
4881 lip = xfs_trans_ail_cursor_next(ailp, &cur);
4884 xfs_trans_ail_cursor_done(&cur);
4885 spin_unlock(&ailp->ail_lock);
4887 error = xlog_finish_defer_ops(parent_tp);
4888 xfs_trans_cancel(parent_tp);
4894 * A cancel occurs when the mount has failed and we're bailing out.
4895 * Release all pending log intent items so they don't pin the AIL.
4898 xlog_recover_cancel_intents(
4901 struct xfs_log_item *lip;
4902 struct xfs_ail_cursor cur;
4903 struct xfs_ail *ailp;
4906 spin_lock(&ailp->ail_lock);
4907 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
4908 while (lip != NULL) {
4910 * We're done when we see something other than an intent.
4911 * There should be no intents left in the AIL now.
4913 if (!xlog_item_is_intent(lip)) {
4915 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
4916 ASSERT(!xlog_item_is_intent(lip));
4921 switch (lip->li_type) {
4923 xlog_recover_cancel_efi(log->l_mp, ailp, lip);
4926 xlog_recover_cancel_rui(log->l_mp, ailp, lip);
4929 xlog_recover_cancel_cui(log->l_mp, ailp, lip);
4932 xlog_recover_cancel_bui(log->l_mp, ailp, lip);
4936 lip = xfs_trans_ail_cursor_next(ailp, &cur);
4939 xfs_trans_ail_cursor_done(&cur);
4940 spin_unlock(&ailp->ail_lock);
4944 * This routine performs a transaction to null out a bad inode pointer
4945 * in an agi unlinked inode hash bucket.
4948 xlog_recover_clear_agi_bucket(
4950 xfs_agnumber_t agno,
4959 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_clearagi, 0, 0, 0, &tp);
4963 error = xfs_read_agi(mp, tp, agno, &agibp);
4967 agi = XFS_BUF_TO_AGI(agibp);
4968 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
4969 offset = offsetof(xfs_agi_t, agi_unlinked) +
4970 (sizeof(xfs_agino_t) * bucket);
4971 xfs_trans_log_buf(tp, agibp, offset,
4972 (offset + sizeof(xfs_agino_t) - 1));
4974 error = xfs_trans_commit(tp);
4980 xfs_trans_cancel(tp);
4982 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
4987 xlog_recover_process_one_iunlink(
4988 struct xfs_mount *mp,
4989 xfs_agnumber_t agno,
4993 struct xfs_buf *ibp;
4994 struct xfs_dinode *dip;
4995 struct xfs_inode *ip;
4999 ino = XFS_AGINO_TO_INO(mp, agno, agino);
5000 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
5005 * Get the on disk inode to find the next inode in the bucket.
5007 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
5011 xfs_iflags_clear(ip, XFS_IRECOVERY);
5012 ASSERT(VFS_I(ip)->i_nlink == 0);
5013 ASSERT(VFS_I(ip)->i_mode != 0);
5015 /* setup for the next pass */
5016 agino = be32_to_cpu(dip->di_next_unlinked);
5020 * Prevent any DMAPI event from being sent when the reference on
5021 * the inode is dropped.
5023 ip->i_d.di_dmevmask = 0;
5032 * We can't read in the inode this bucket points to, or this inode
5033 * is messed up. Just ditch this bucket of inodes. We will lose
5034 * some inodes and space, but at least we won't hang.
5036 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
5037 * clear the inode pointer in the bucket.
5039 xlog_recover_clear_agi_bucket(mp, agno, bucket);
5044 * Recover AGI unlinked lists
5046 * This is called during recovery to process any inodes which we unlinked but
5047 * not freed when the system crashed. These inodes will be on the lists in the
5048 * AGI blocks. What we do here is scan all the AGIs and fully truncate and free
5049 * any inodes found on the lists. Each inode is removed from the lists when it
5050 * has been fully truncated and is freed. The freeing of the inode and its
5051 * removal from the list must be atomic.
5053 * If everything we touch in the agi processing loop is already in memory, this
5054 * loop can hold the cpu for a long time. It runs without lock contention,
5055 * memory allocation contention, the need wait for IO, etc, and so will run
5056 * until we either run out of inodes to process, run low on memory or we run out
5059 * This behaviour is bad for latency on single CPU and non-preemptible kernels,
5060 * and can prevent other filesytem work (such as CIL pushes) from running. This
5061 * can lead to deadlocks if the recovery process runs out of log reservation
5062 * space. Hence we need to yield the CPU when there is other kernel work
5063 * scheduled on this CPU to ensure other scheduled work can run without undue
5067 xlog_recover_process_iunlinks(
5071 xfs_agnumber_t agno;
5080 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
5082 * Find the agi for this ag.
5084 error = xfs_read_agi(mp, NULL, agno, &agibp);
5087 * AGI is b0rked. Don't process it.
5089 * We should probably mark the filesystem as corrupt
5090 * after we've recovered all the ag's we can....
5095 * Unlock the buffer so that it can be acquired in the normal
5096 * course of the transaction to truncate and free each inode.
5097 * Because we are not racing with anyone else here for the AGI
5098 * buffer, we don't even need to hold it locked to read the
5099 * initial unlinked bucket entries out of the buffer. We keep
5100 * buffer reference though, so that it stays pinned in memory
5101 * while we need the buffer.
5103 agi = XFS_BUF_TO_AGI(agibp);
5104 xfs_buf_unlock(agibp);
5106 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
5107 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
5108 while (agino != NULLAGINO) {
5109 agino = xlog_recover_process_one_iunlink(mp,
5110 agno, agino, bucket);
5114 xfs_buf_rele(agibp);
5120 struct xlog_rec_header *rhead,
5126 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
5127 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
5128 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
5132 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
5133 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
5134 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
5135 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
5136 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
5137 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
5144 * CRC check, unpack and process a log record.
5147 xlog_recover_process(
5149 struct hlist_head rhash[],
5150 struct xlog_rec_header *rhead,
5153 struct list_head *buffer_list)
5155 __le32 old_crc = rhead->h_crc;
5158 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
5161 * Nothing else to do if this is a CRC verification pass. Just return
5162 * if this a record with a non-zero crc. Unfortunately, mkfs always
5163 * sets old_crc to 0 so we must consider this valid even on v5 supers.
5164 * Otherwise, return EFSBADCRC on failure so the callers up the stack
5165 * know precisely what failed.
5167 if (pass == XLOG_RECOVER_CRCPASS) {
5168 if (old_crc && crc != old_crc)
5174 * We're in the normal recovery path. Issue a warning if and only if the
5175 * CRC in the header is non-zero. This is an advisory warning and the
5176 * zero CRC check prevents warnings from being emitted when upgrading
5177 * the kernel from one that does not add CRCs by default.
5179 if (crc != old_crc) {
5180 if (old_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
5181 xfs_alert(log->l_mp,
5182 "log record CRC mismatch: found 0x%x, expected 0x%x.",
5183 le32_to_cpu(old_crc),
5185 xfs_hex_dump(dp, 32);
5189 * If the filesystem is CRC enabled, this mismatch becomes a
5190 * fatal log corruption failure.
5192 if (xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
5193 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, log->l_mp);
5194 return -EFSCORRUPTED;
5198 xlog_unpack_data(rhead, dp, log);
5200 return xlog_recover_process_data(log, rhash, rhead, dp, pass,
5205 xlog_valid_rec_header(
5207 struct xlog_rec_header *rhead,
5212 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
5213 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
5214 XFS_ERRLEVEL_LOW, log->l_mp);
5215 return -EFSCORRUPTED;
5218 (!rhead->h_version ||
5219 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
5220 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
5221 __func__, be32_to_cpu(rhead->h_version));
5222 return -EFSCORRUPTED;
5225 /* LR body must have data or it wouldn't have been written */
5226 hlen = be32_to_cpu(rhead->h_len);
5227 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
5228 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
5229 XFS_ERRLEVEL_LOW, log->l_mp);
5230 return -EFSCORRUPTED;
5232 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
5233 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
5234 XFS_ERRLEVEL_LOW, log->l_mp);
5235 return -EFSCORRUPTED;
5241 * Read the log from tail to head and process the log records found.
5242 * Handle the two cases where the tail and head are in the same cycle
5243 * and where the active portion of the log wraps around the end of
5244 * the physical log separately. The pass parameter is passed through
5245 * to the routines called to process the data and is not looked at
5249 xlog_do_recovery_pass(
5251 xfs_daddr_t head_blk,
5252 xfs_daddr_t tail_blk,
5254 xfs_daddr_t *first_bad) /* out: first bad log rec */
5256 xlog_rec_header_t *rhead;
5257 xfs_daddr_t blk_no, rblk_no;
5258 xfs_daddr_t rhead_blk;
5261 int error = 0, h_size, h_len;
5263 int bblks, split_bblks;
5264 int hblks, split_hblks, wrapped_hblks;
5266 struct hlist_head rhash[XLOG_RHASH_SIZE];
5267 LIST_HEAD (buffer_list);
5269 ASSERT(head_blk != tail_blk);
5270 blk_no = rhead_blk = tail_blk;
5272 for (i = 0; i < XLOG_RHASH_SIZE; i++)
5273 INIT_HLIST_HEAD(&rhash[i]);
5276 * Read the header of the tail block and get the iclog buffer size from
5277 * h_size. Use this to tell how many sectors make up the log header.
5279 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
5281 * When using variable length iclogs, read first sector of
5282 * iclog header and extract the header size from it. Get a
5283 * new hbp that is the correct size.
5285 hbp = xlog_alloc_buffer(log, 1);
5289 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
5293 rhead = (xlog_rec_header_t *)offset;
5294 error = xlog_valid_rec_header(log, rhead, tail_blk);
5299 * xfsprogs has a bug where record length is based on lsunit but
5300 * h_size (iclog size) is hardcoded to 32k. Now that we
5301 * unconditionally CRC verify the unmount record, this means the
5302 * log buffer can be too small for the record and cause an
5305 * Detect this condition here. Use lsunit for the buffer size as
5306 * long as this looks like the mkfs case. Otherwise, return an
5307 * error to avoid a buffer overrun.
5309 h_size = be32_to_cpu(rhead->h_size);
5310 h_len = be32_to_cpu(rhead->h_len);
5311 if (h_len > h_size) {
5312 if (h_len <= log->l_mp->m_logbsize &&
5313 be32_to_cpu(rhead->h_num_logops) == 1) {
5315 "invalid iclog size (%d bytes), using lsunit (%d bytes)",
5316 h_size, log->l_mp->m_logbsize);
5317 h_size = log->l_mp->m_logbsize;
5319 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW,
5321 return -EFSCORRUPTED;
5325 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
5326 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
5327 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
5328 if (h_size % XLOG_HEADER_CYCLE_SIZE)
5331 hbp = xlog_alloc_buffer(log, hblks);
5336 ASSERT(log->l_sectBBsize == 1);
5338 hbp = xlog_alloc_buffer(log, 1);
5339 h_size = XLOG_BIG_RECORD_BSIZE;
5344 dbp = xlog_alloc_buffer(log, BTOBB(h_size));
5350 memset(rhash, 0, sizeof(rhash));
5351 if (tail_blk > head_blk) {
5353 * Perform recovery around the end of the physical log.
5354 * When the head is not on the same cycle number as the tail,
5355 * we can't do a sequential recovery.
5357 while (blk_no < log->l_logBBsize) {
5359 * Check for header wrapping around physical end-of-log
5364 if (blk_no + hblks <= log->l_logBBsize) {
5365 /* Read header in one read */
5366 error = xlog_bread(log, blk_no, hblks, hbp,
5371 /* This LR is split across physical log end */
5372 if (blk_no != log->l_logBBsize) {
5373 /* some data before physical log end */
5374 ASSERT(blk_no <= INT_MAX);
5375 split_hblks = log->l_logBBsize - (int)blk_no;
5376 ASSERT(split_hblks > 0);
5377 error = xlog_bread(log, blk_no,
5385 * Note: this black magic still works with
5386 * large sector sizes (non-512) only because:
5387 * - we increased the buffer size originally
5388 * by 1 sector giving us enough extra space
5389 * for the second read;
5390 * - the log start is guaranteed to be sector
5392 * - we read the log end (LR header start)
5393 * _first_, then the log start (LR header end)
5394 * - order is important.
5396 wrapped_hblks = hblks - split_hblks;
5397 error = xlog_bread_noalign(log, 0,
5399 offset + BBTOB(split_hblks));
5403 rhead = (xlog_rec_header_t *)offset;
5404 error = xlog_valid_rec_header(log, rhead,
5405 split_hblks ? blk_no : 0);
5409 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
5413 * Read the log record data in multiple reads if it
5414 * wraps around the end of the log. Note that if the
5415 * header already wrapped, blk_no could point past the
5416 * end of the log. The record data is contiguous in
5419 if (blk_no + bblks <= log->l_logBBsize ||
5420 blk_no >= log->l_logBBsize) {
5421 rblk_no = xlog_wrap_logbno(log, blk_no);
5422 error = xlog_bread(log, rblk_no, bblks, dbp,
5427 /* This log record is split across the
5428 * physical end of log */
5431 if (blk_no != log->l_logBBsize) {
5432 /* some data is before the physical
5434 ASSERT(!wrapped_hblks);
5435 ASSERT(blk_no <= INT_MAX);
5437 log->l_logBBsize - (int)blk_no;
5438 ASSERT(split_bblks > 0);
5439 error = xlog_bread(log, blk_no,
5447 * Note: this black magic still works with
5448 * large sector sizes (non-512) only because:
5449 * - we increased the buffer size originally
5450 * by 1 sector giving us enough extra space
5451 * for the second read;
5452 * - the log start is guaranteed to be sector
5454 * - we read the log end (LR header start)
5455 * _first_, then the log start (LR header end)
5456 * - order is important.
5458 error = xlog_bread_noalign(log, 0,
5459 bblks - split_bblks,
5460 offset + BBTOB(split_bblks));
5465 error = xlog_recover_process(log, rhash, rhead, offset,
5466 pass, &buffer_list);
5474 ASSERT(blk_no >= log->l_logBBsize);
5475 blk_no -= log->l_logBBsize;
5479 /* read first part of physical log */
5480 while (blk_no < head_blk) {
5481 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
5485 rhead = (xlog_rec_header_t *)offset;
5486 error = xlog_valid_rec_header(log, rhead, blk_no);
5490 /* blocks in data section */
5491 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
5492 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
5497 error = xlog_recover_process(log, rhash, rhead, offset, pass,
5502 blk_no += bblks + hblks;
5512 * Submit buffers that have been added from the last record processed,
5513 * regardless of error status.
5515 if (!list_empty(&buffer_list))
5516 error2 = xfs_buf_delwri_submit(&buffer_list);
5518 if (error && first_bad)
5519 *first_bad = rhead_blk;
5522 * Transactions are freed at commit time but transactions without commit
5523 * records on disk are never committed. Free any that may be left in the
5526 for (i = 0; i < XLOG_RHASH_SIZE; i++) {
5527 struct hlist_node *tmp;
5528 struct xlog_recover *trans;
5530 hlist_for_each_entry_safe(trans, tmp, &rhash[i], r_list)
5531 xlog_recover_free_trans(trans);
5534 return error ? error : error2;
5538 * Do the recovery of the log. We actually do this in two phases.
5539 * The two passes are necessary in order to implement the function
5540 * of cancelling a record written into the log. The first pass
5541 * determines those things which have been cancelled, and the
5542 * second pass replays log items normally except for those which
5543 * have been cancelled. The handling of the replay and cancellations
5544 * takes place in the log item type specific routines.
5546 * The table of items which have cancel records in the log is allocated
5547 * and freed at this level, since only here do we know when all of
5548 * the log recovery has been completed.
5551 xlog_do_log_recovery(
5553 xfs_daddr_t head_blk,
5554 xfs_daddr_t tail_blk)
5558 ASSERT(head_blk != tail_blk);
5561 * First do a pass to find all of the cancelled buf log items.
5562 * Store them in the buf_cancel_table for use in the second pass.
5564 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
5565 sizeof(struct list_head),
5567 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
5568 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
5570 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
5571 XLOG_RECOVER_PASS1, NULL);
5573 kmem_free(log->l_buf_cancel_table);
5574 log->l_buf_cancel_table = NULL;
5578 * Then do a second pass to actually recover the items in the log.
5579 * When it is complete free the table of buf cancel items.
5581 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
5582 XLOG_RECOVER_PASS2, NULL);
5587 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
5588 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
5592 kmem_free(log->l_buf_cancel_table);
5593 log->l_buf_cancel_table = NULL;
5599 * Do the actual recovery
5604 xfs_daddr_t head_blk,
5605 xfs_daddr_t tail_blk)
5607 struct xfs_mount *mp = log->l_mp;
5612 trace_xfs_log_recover(log, head_blk, tail_blk);
5615 * First replay the images in the log.
5617 error = xlog_do_log_recovery(log, head_blk, tail_blk);
5622 * If IO errors happened during recovery, bail out.
5624 if (XFS_FORCED_SHUTDOWN(mp)) {
5629 * We now update the tail_lsn since much of the recovery has completed
5630 * and there may be space available to use. If there were no extent
5631 * or iunlinks, we can free up the entire log and set the tail_lsn to
5632 * be the last_sync_lsn. This was set in xlog_find_tail to be the
5633 * lsn of the last known good LR on disk. If there are extent frees
5634 * or iunlinks they will have some entries in the AIL; so we look at
5635 * the AIL to determine how to set the tail_lsn.
5637 xlog_assign_tail_lsn(mp);
5640 * Now that we've finished replaying all buffer and inode
5641 * updates, re-read in the superblock and reverify it.
5644 bp->b_flags &= ~(XBF_DONE | XBF_ASYNC);
5645 ASSERT(!(bp->b_flags & XBF_WRITE));
5646 bp->b_flags |= XBF_READ;
5647 bp->b_ops = &xfs_sb_buf_ops;
5649 error = xfs_buf_submit(bp);
5651 if (!XFS_FORCED_SHUTDOWN(mp)) {
5652 xfs_buf_ioerror_alert(bp, __func__);
5659 /* Convert superblock from on-disk format */
5661 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
5664 /* re-initialise in-core superblock and geometry structures */
5665 xfs_reinit_percpu_counters(mp);
5666 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
5668 xfs_warn(mp, "Failed post-recovery per-ag init: %d", error);
5671 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
5673 xlog_recover_check_summary(log);
5675 /* Normal transactions can now occur */
5676 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
5681 * Perform recovery and re-initialize some log variables in xlog_find_tail.
5683 * Return error or zero.
5689 xfs_daddr_t head_blk, tail_blk;
5692 /* find the tail of the log */
5693 error = xlog_find_tail(log, &head_blk, &tail_blk);
5698 * The superblock was read before the log was available and thus the LSN
5699 * could not be verified. Check the superblock LSN against the current
5700 * LSN now that it's known.
5702 if (xfs_sb_version_hascrc(&log->l_mp->m_sb) &&
5703 !xfs_log_check_lsn(log->l_mp, log->l_mp->m_sb.sb_lsn))
5706 if (tail_blk != head_blk) {
5707 /* There used to be a comment here:
5709 * disallow recovery on read-only mounts. note -- mount
5710 * checks for ENOSPC and turns it into an intelligent
5712 * ...but this is no longer true. Now, unless you specify
5713 * NORECOVERY (in which case this function would never be
5714 * called), we just go ahead and recover. We do this all
5715 * under the vfs layer, so we can get away with it unless
5716 * the device itself is read-only, in which case we fail.
5718 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
5723 * Version 5 superblock log feature mask validation. We know the
5724 * log is dirty so check if there are any unknown log features
5725 * in what we need to recover. If there are unknown features
5726 * (e.g. unsupported transactions, then simply reject the
5727 * attempt at recovery before touching anything.
5729 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
5730 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
5731 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
5733 "Superblock has unknown incompatible log features (0x%x) enabled.",
5734 (log->l_mp->m_sb.sb_features_log_incompat &
5735 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
5737 "The log can not be fully and/or safely recovered by this kernel.");
5739 "Please recover the log on a kernel that supports the unknown features.");
5744 * Delay log recovery if the debug hook is set. This is debug
5745 * instrumention to coordinate simulation of I/O failures with
5748 if (xfs_globals.log_recovery_delay) {
5749 xfs_notice(log->l_mp,
5750 "Delaying log recovery for %d seconds.",
5751 xfs_globals.log_recovery_delay);
5752 msleep(xfs_globals.log_recovery_delay * 1000);
5755 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
5756 log->l_mp->m_logname ? log->l_mp->m_logname
5759 error = xlog_do_recover(log, head_blk, tail_blk);
5760 log->l_flags |= XLOG_RECOVERY_NEEDED;
5766 * In the first part of recovery we replay inodes and buffers and build
5767 * up the list of extent free items which need to be processed. Here
5768 * we process the extent free items and clean up the on disk unlinked
5769 * inode lists. This is separated from the first part of recovery so
5770 * that the root and real-time bitmap inodes can be read in from disk in
5771 * between the two stages. This is necessary so that we can free space
5772 * in the real-time portion of the file system.
5775 xlog_recover_finish(
5779 * Now we're ready to do the transactions needed for the
5780 * rest of recovery. Start with completing all the extent
5781 * free intent records and then process the unlinked inode
5782 * lists. At this point, we essentially run in normal mode
5783 * except that we're still performing recovery actions
5784 * rather than accepting new requests.
5786 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
5788 error = xlog_recover_process_intents(log);
5790 xfs_alert(log->l_mp, "Failed to recover intents");
5795 * Sync the log to get all the intents out of the AIL.
5796 * This isn't absolutely necessary, but it helps in
5797 * case the unlink transactions would have problems
5798 * pushing the intents out of the way.
5800 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
5802 xlog_recover_process_iunlinks(log);
5804 xlog_recover_check_summary(log);
5806 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
5807 log->l_mp->m_logname ? log->l_mp->m_logname
5809 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
5811 xfs_info(log->l_mp, "Ending clean mount");
5817 xlog_recover_cancel(
5820 if (log->l_flags & XLOG_RECOVERY_NEEDED)
5821 xlog_recover_cancel_intents(log);
5826 * Read all of the agf and agi counters and check that they
5827 * are consistent with the superblock counters.
5830 xlog_recover_check_summary(
5837 xfs_agnumber_t agno;
5848 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
5849 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
5851 xfs_alert(mp, "%s agf read failed agno %d error %d",
5852 __func__, agno, error);
5854 agfp = XFS_BUF_TO_AGF(agfbp);
5855 freeblks += be32_to_cpu(agfp->agf_freeblks) +
5856 be32_to_cpu(agfp->agf_flcount);
5857 xfs_buf_relse(agfbp);
5860 error = xfs_read_agi(mp, NULL, agno, &agibp);
5862 xfs_alert(mp, "%s agi read failed agno %d error %d",
5863 __func__, agno, error);
5865 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
5867 itotal += be32_to_cpu(agi->agi_count);
5868 ifree += be32_to_cpu(agi->agi_freecount);
5869 xfs_buf_relse(agibp);