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.
101 * Pass log block 0 since we don't have an addr yet, buffer will be
104 if (!xlog_verify_bno(log, 0, nbblks)) {
105 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
107 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
112 * We do log I/O in units of log sectors (a power-of-2 multiple of the
113 * basic block size), so we round up the requested size to accommodate
114 * the basic blocks required for complete log sectors.
116 * In addition, the buffer may be used for a non-sector-aligned block
117 * offset, in which case an I/O of the requested size could extend
118 * beyond the end of the buffer. If the requested size is only 1 basic
119 * block it will never straddle a sector boundary, so this won't be an
120 * issue. Nor will this be a problem if the log I/O is done in basic
121 * blocks (sector size 1). But otherwise we extend the buffer by one
122 * extra log sector to ensure there's space to accommodate this
125 if (nbblks > 1 && log->l_sectBBsize > 1)
126 nbblks += log->l_sectBBsize;
127 nbblks = round_up(nbblks, log->l_sectBBsize);
128 return kmem_alloc_large(BBTOB(nbblks), KM_MAYFAIL);
132 * Return the address of the start of the given block number's data
133 * in a log buffer. The buffer covers a log sector-aligned region.
135 static inline unsigned int
140 return BBTOB(blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1));
153 if (!xlog_verify_bno(log, blk_no, nbblks)) {
155 "Invalid log block/length (0x%llx, 0x%x) for buffer",
157 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
158 return -EFSCORRUPTED;
161 blk_no = round_down(blk_no, log->l_sectBBsize);
162 nbblks = round_up(nbblks, log->l_sectBBsize);
165 error = xfs_rw_bdev(log->l_targ->bt_bdev, log->l_logBBstart + blk_no,
166 BBTOB(nbblks), data, op);
167 if (error && !XFS_FORCED_SHUTDOWN(log->l_mp)) {
169 "log recovery %s I/O error at daddr 0x%llx len %d error %d",
170 op == REQ_OP_WRITE ? "write" : "read",
171 blk_no, nbblks, error);
183 return xlog_do_io(log, blk_no, nbblks, data, REQ_OP_READ);
196 error = xlog_do_io(log, blk_no, nbblks, data, REQ_OP_READ);
198 *offset = data + xlog_align(log, blk_no);
209 return xlog_do_io(log, blk_no, nbblks, data, REQ_OP_WRITE);
214 * dump debug superblock and log record information
217 xlog_header_check_dump(
219 xlog_rec_header_t *head)
221 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d",
222 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
223 xfs_debug(mp, " log : uuid = %pU, fmt = %d",
224 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
227 #define xlog_header_check_dump(mp, head)
231 * check log record header for recovery
234 xlog_header_check_recover(
236 xlog_rec_header_t *head)
238 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
241 * IRIX doesn't write the h_fmt field and leaves it zeroed
242 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
243 * a dirty log created in IRIX.
245 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
247 "dirty log written in incompatible format - can't recover");
248 xlog_header_check_dump(mp, head);
249 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
250 XFS_ERRLEVEL_HIGH, mp);
251 return -EFSCORRUPTED;
252 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
254 "dirty log entry has mismatched uuid - can't recover");
255 xlog_header_check_dump(mp, head);
256 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
257 XFS_ERRLEVEL_HIGH, mp);
258 return -EFSCORRUPTED;
264 * read the head block of the log and check the header
267 xlog_header_check_mount(
269 xlog_rec_header_t *head)
271 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
273 if (uuid_is_null(&head->h_fs_uuid)) {
275 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
276 * h_fs_uuid is null, we assume this log was last mounted
277 * by IRIX and continue.
279 xfs_warn(mp, "null uuid in log - IRIX style log");
280 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
281 xfs_warn(mp, "log has mismatched uuid - can't recover");
282 xlog_header_check_dump(mp, head);
283 XFS_ERROR_REPORT("xlog_header_check_mount",
284 XFS_ERRLEVEL_HIGH, mp);
285 return -EFSCORRUPTED;
296 * We're not going to bother about retrying
297 * this during recovery. One strike!
299 if (!XFS_FORCED_SHUTDOWN(bp->b_mount)) {
300 xfs_buf_ioerror_alert(bp, __func__);
301 xfs_force_shutdown(bp->b_mount, SHUTDOWN_META_IO_ERROR);
306 * On v5 supers, a bli could be attached to update the metadata LSN.
310 xfs_buf_item_relse(bp);
311 ASSERT(bp->b_log_item == NULL);
318 * This routine finds (to an approximation) the first block in the physical
319 * log which contains the given cycle. It uses a binary search algorithm.
320 * Note that the algorithm can not be perfect because the disk will not
321 * necessarily be perfect.
324 xlog_find_cycle_start(
327 xfs_daddr_t first_blk,
328 xfs_daddr_t *last_blk,
338 mid_blk = BLK_AVG(first_blk, end_blk);
339 while (mid_blk != first_blk && mid_blk != end_blk) {
340 error = xlog_bread(log, mid_blk, 1, buffer, &offset);
343 mid_cycle = xlog_get_cycle(offset);
344 if (mid_cycle == cycle)
345 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
347 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
348 mid_blk = BLK_AVG(first_blk, end_blk);
350 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
351 (mid_blk == end_blk && mid_blk-1 == first_blk));
359 * Check that a range of blocks does not contain stop_on_cycle_no.
360 * Fill in *new_blk with the block offset where such a block is
361 * found, or with -1 (an invalid block number) if there is no such
362 * block in the range. The scan needs to occur from front to back
363 * and the pointer into the region must be updated since a later
364 * routine will need to perform another test.
367 xlog_find_verify_cycle(
369 xfs_daddr_t start_blk,
371 uint stop_on_cycle_no,
372 xfs_daddr_t *new_blk)
382 * Greedily allocate a buffer big enough to handle the full
383 * range of basic blocks we'll be examining. If that fails,
384 * try a smaller size. We need to be able to read at least
385 * a log sector, or we're out of luck.
387 bufblks = 1 << ffs(nbblks);
388 while (bufblks > log->l_logBBsize)
390 while (!(buffer = xlog_alloc_buffer(log, bufblks))) {
392 if (bufblks < log->l_sectBBsize)
396 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
399 bcount = min(bufblks, (start_blk + nbblks - i));
401 error = xlog_bread(log, i, bcount, buffer, &buf);
405 for (j = 0; j < bcount; j++) {
406 cycle = xlog_get_cycle(buf);
407 if (cycle == stop_on_cycle_no) {
424 * Potentially backup over partial log record write.
426 * In the typical case, last_blk is the number of the block directly after
427 * a good log record. Therefore, we subtract one to get the block number
428 * of the last block in the given buffer. extra_bblks contains the number
429 * of blocks we would have read on a previous read. This happens when the
430 * last log record is split over the end of the physical log.
432 * extra_bblks is the number of blocks potentially verified on a previous
433 * call to this routine.
436 xlog_find_verify_log_record(
438 xfs_daddr_t start_blk,
439 xfs_daddr_t *last_blk,
445 xlog_rec_header_t *head = NULL;
448 int num_blks = *last_blk - start_blk;
451 ASSERT(start_blk != 0 || *last_blk != start_blk);
453 buffer = xlog_alloc_buffer(log, num_blks);
455 buffer = xlog_alloc_buffer(log, 1);
460 error = xlog_bread(log, start_blk, num_blks, buffer, &offset);
463 offset += ((num_blks - 1) << BBSHIFT);
466 for (i = (*last_blk) - 1; i >= 0; i--) {
468 /* valid log record not found */
470 "Log inconsistent (didn't find previous header)");
477 error = xlog_bread(log, i, 1, buffer, &offset);
482 head = (xlog_rec_header_t *)offset;
484 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
492 * We hit the beginning of the physical log & still no header. Return
493 * to caller. If caller can handle a return of -1, then this routine
494 * will be called again for the end of the physical log.
502 * We have the final block of the good log (the first block
503 * of the log record _before_ the head. So we check the uuid.
505 if ((error = xlog_header_check_mount(log->l_mp, head)))
509 * We may have found a log record header before we expected one.
510 * last_blk will be the 1st block # with a given cycle #. We may end
511 * up reading an entire log record. In this case, we don't want to
512 * reset last_blk. Only when last_blk points in the middle of a log
513 * record do we update last_blk.
515 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
516 uint h_size = be32_to_cpu(head->h_size);
518 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
519 if (h_size % XLOG_HEADER_CYCLE_SIZE)
525 if (*last_blk - i + extra_bblks !=
526 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
535 * Head is defined to be the point of the log where the next log write
536 * could go. This means that incomplete LR writes at the end are
537 * eliminated when calculating the head. We aren't guaranteed that previous
538 * LR have complete transactions. We only know that a cycle number of
539 * current cycle number -1 won't be present in the log if we start writing
540 * from our current block number.
542 * last_blk contains the block number of the first block with a given
545 * Return: zero if normal, non-zero if error.
550 xfs_daddr_t *return_head_blk)
554 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
556 uint first_half_cycle, last_half_cycle;
558 int error, log_bbnum = log->l_logBBsize;
560 /* Is the end of the log device zeroed? */
561 error = xlog_find_zeroed(log, &first_blk);
563 xfs_warn(log->l_mp, "empty log check failed");
567 *return_head_blk = first_blk;
569 /* Is the whole lot zeroed? */
571 /* Linux XFS shouldn't generate totally zeroed logs -
572 * mkfs etc write a dummy unmount record to a fresh
573 * log so we can store the uuid in there
575 xfs_warn(log->l_mp, "totally zeroed log");
581 first_blk = 0; /* get cycle # of 1st block */
582 buffer = xlog_alloc_buffer(log, 1);
586 error = xlog_bread(log, 0, 1, buffer, &offset);
588 goto out_free_buffer;
590 first_half_cycle = xlog_get_cycle(offset);
592 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
593 error = xlog_bread(log, last_blk, 1, buffer, &offset);
595 goto out_free_buffer;
597 last_half_cycle = xlog_get_cycle(offset);
598 ASSERT(last_half_cycle != 0);
601 * If the 1st half cycle number is equal to the last half cycle number,
602 * then the entire log is stamped with the same cycle number. In this
603 * case, head_blk can't be set to zero (which makes sense). The below
604 * math doesn't work out properly with head_blk equal to zero. Instead,
605 * we set it to log_bbnum which is an invalid block number, but this
606 * value makes the math correct. If head_blk doesn't changed through
607 * all the tests below, *head_blk is set to zero at the very end rather
608 * than log_bbnum. In a sense, log_bbnum and zero are the same block
609 * in a circular file.
611 if (first_half_cycle == last_half_cycle) {
613 * In this case we believe that the entire log should have
614 * cycle number last_half_cycle. We need to scan backwards
615 * from the end verifying that there are no holes still
616 * containing last_half_cycle - 1. If we find such a hole,
617 * then the start of that hole will be the new head. The
618 * simple case looks like
619 * x | x ... | x - 1 | x
620 * Another case that fits this picture would be
621 * x | x + 1 | x ... | x
622 * In this case the head really is somewhere at the end of the
623 * log, as one of the latest writes at the beginning was
626 * x | x + 1 | x ... | x - 1 | x
627 * This is really the combination of the above two cases, and
628 * the head has to end up at the start of the x-1 hole at the
631 * In the 256k log case, we will read from the beginning to the
632 * end of the log and search for cycle numbers equal to x-1.
633 * We don't worry about the x+1 blocks that we encounter,
634 * because we know that they cannot be the head since the log
637 head_blk = log_bbnum;
638 stop_on_cycle = last_half_cycle - 1;
641 * In this case we want to find the first block with cycle
642 * number matching last_half_cycle. We expect the log to be
644 * x + 1 ... | x ... | x
645 * The first block with cycle number x (last_half_cycle) will
646 * be where the new head belongs. First we do a binary search
647 * for the first occurrence of last_half_cycle. The binary
648 * search may not be totally accurate, so then we scan back
649 * from there looking for occurrences of last_half_cycle before
650 * us. If that backwards scan wraps around the beginning of
651 * the log, then we look for occurrences of last_half_cycle - 1
652 * at the end of the log. The cases we're looking for look
654 * v binary search stopped here
655 * x + 1 ... | x | x + 1 | x ... | x
656 * ^ but we want to locate this spot
658 * <---------> less than scan distance
659 * x + 1 ... | x ... | x - 1 | x
660 * ^ we want to locate this spot
662 stop_on_cycle = last_half_cycle;
663 error = xlog_find_cycle_start(log, buffer, first_blk, &head_blk,
666 goto out_free_buffer;
670 * Now validate the answer. Scan back some number of maximum possible
671 * blocks and make sure each one has the expected cycle number. The
672 * maximum is determined by the total possible amount of buffering
673 * in the in-core log. The following number can be made tighter if
674 * we actually look at the block size of the filesystem.
676 num_scan_bblks = min_t(int, log_bbnum, XLOG_TOTAL_REC_SHIFT(log));
677 if (head_blk >= num_scan_bblks) {
679 * We are guaranteed that the entire check can be performed
682 start_blk = head_blk - num_scan_bblks;
683 if ((error = xlog_find_verify_cycle(log,
684 start_blk, num_scan_bblks,
685 stop_on_cycle, &new_blk)))
686 goto out_free_buffer;
689 } else { /* need to read 2 parts of log */
691 * We are going to scan backwards in the log in two parts.
692 * First we scan the physical end of the log. In this part
693 * of the log, we are looking for blocks with cycle number
694 * last_half_cycle - 1.
695 * If we find one, then we know that the log starts there, as
696 * we've found a hole that didn't get written in going around
697 * the end of the physical log. The simple case for this is
698 * x + 1 ... | x ... | x - 1 | x
699 * <---------> less than scan distance
700 * If all of the blocks at the end of the log have cycle number
701 * last_half_cycle, then we check the blocks at the start of
702 * the log looking for occurrences of last_half_cycle. If we
703 * find one, then our current estimate for the location of the
704 * first occurrence of last_half_cycle is wrong and we move
705 * back to the hole we've found. This case looks like
706 * x + 1 ... | x | x + 1 | x ...
707 * ^ binary search stopped here
708 * Another case we need to handle that only occurs in 256k
710 * x + 1 ... | x ... | x+1 | x ...
711 * ^ binary search stops here
712 * In a 256k log, the scan at the end of the log will see the
713 * x + 1 blocks. We need to skip past those since that is
714 * certainly not the head of the log. By searching for
715 * last_half_cycle-1 we accomplish that.
717 ASSERT(head_blk <= INT_MAX &&
718 (xfs_daddr_t) num_scan_bblks >= head_blk);
719 start_blk = log_bbnum - (num_scan_bblks - head_blk);
720 if ((error = xlog_find_verify_cycle(log, start_blk,
721 num_scan_bblks - (int)head_blk,
722 (stop_on_cycle - 1), &new_blk)))
723 goto out_free_buffer;
730 * Scan beginning of log now. The last part of the physical
731 * log is good. This scan needs to verify that it doesn't find
732 * the last_half_cycle.
735 ASSERT(head_blk <= INT_MAX);
736 if ((error = xlog_find_verify_cycle(log,
737 start_blk, (int)head_blk,
738 stop_on_cycle, &new_blk)))
739 goto out_free_buffer;
746 * Now we need to make sure head_blk is not pointing to a block in
747 * the middle of a log record.
749 num_scan_bblks = XLOG_REC_SHIFT(log);
750 if (head_blk >= num_scan_bblks) {
751 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
753 /* start ptr at last block ptr before head_blk */
754 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
758 goto out_free_buffer;
761 ASSERT(head_blk <= INT_MAX);
762 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
764 goto out_free_buffer;
766 /* We hit the beginning of the log during our search */
767 start_blk = log_bbnum - (num_scan_bblks - head_blk);
769 ASSERT(start_blk <= INT_MAX &&
770 (xfs_daddr_t) log_bbnum-start_blk >= 0);
771 ASSERT(head_blk <= INT_MAX);
772 error = xlog_find_verify_log_record(log, start_blk,
773 &new_blk, (int)head_blk);
777 goto out_free_buffer;
778 if (new_blk != log_bbnum)
781 goto out_free_buffer;
785 if (head_blk == log_bbnum)
786 *return_head_blk = 0;
788 *return_head_blk = head_blk;
790 * When returning here, we have a good block number. Bad block
791 * means that during a previous crash, we didn't have a clean break
792 * from cycle number N to cycle number N-1. In this case, we need
793 * to find the first block with cycle number N-1.
800 xfs_warn(log->l_mp, "failed to find log head");
805 * Seek backwards in the log for log record headers.
807 * Given a starting log block, walk backwards until we find the provided number
808 * of records or hit the provided tail block. The return value is the number of
809 * records encountered or a negative error code. The log block and buffer
810 * pointer of the last record seen are returned in rblk and rhead respectively.
813 xlog_rseek_logrec_hdr(
815 xfs_daddr_t head_blk,
816 xfs_daddr_t tail_blk,
820 struct xlog_rec_header **rhead,
832 * Walk backwards from the head block until we hit the tail or the first
835 end_blk = head_blk > tail_blk ? tail_blk : 0;
836 for (i = (int) head_blk - 1; i >= end_blk; i--) {
837 error = xlog_bread(log, i, 1, buffer, &offset);
841 if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
843 *rhead = (struct xlog_rec_header *) offset;
844 if (++found == count)
850 * If we haven't hit the tail block or the log record header count,
851 * start looking again from the end of the physical log. Note that
852 * callers can pass head == tail if the tail is not yet known.
854 if (tail_blk >= head_blk && found != count) {
855 for (i = log->l_logBBsize - 1; i >= (int) tail_blk; i--) {
856 error = xlog_bread(log, i, 1, buffer, &offset);
860 if (*(__be32 *)offset ==
861 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
864 *rhead = (struct xlog_rec_header *) offset;
865 if (++found == count)
878 * Seek forward in the log for log record headers.
880 * Given head and tail blocks, walk forward from the tail block until we find
881 * the provided number of records or hit the head block. The return value is the
882 * number of records encountered or a negative error code. The log block and
883 * buffer pointer of the last record seen are returned in rblk and rhead
887 xlog_seek_logrec_hdr(
889 xfs_daddr_t head_blk,
890 xfs_daddr_t tail_blk,
894 struct xlog_rec_header **rhead,
906 * Walk forward from the tail block until we hit the head or the last
909 end_blk = head_blk > tail_blk ? head_blk : log->l_logBBsize - 1;
910 for (i = (int) tail_blk; i <= end_blk; i++) {
911 error = xlog_bread(log, i, 1, buffer, &offset);
915 if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
917 *rhead = (struct xlog_rec_header *) offset;
918 if (++found == count)
924 * If we haven't hit the head block or the log record header count,
925 * start looking again from the start of the physical log.
927 if (tail_blk > head_blk && found != count) {
928 for (i = 0; i < (int) head_blk; i++) {
929 error = xlog_bread(log, i, 1, buffer, &offset);
933 if (*(__be32 *)offset ==
934 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
937 *rhead = (struct xlog_rec_header *) offset;
938 if (++found == count)
951 * Calculate distance from head to tail (i.e., unused space in the log).
956 xfs_daddr_t head_blk,
957 xfs_daddr_t tail_blk)
959 if (head_blk < tail_blk)
960 return tail_blk - head_blk;
962 return tail_blk + (log->l_logBBsize - head_blk);
966 * Verify the log tail. This is particularly important when torn or incomplete
967 * writes have been detected near the front of the log and the head has been
968 * walked back accordingly.
970 * We also have to handle the case where the tail was pinned and the head
971 * blocked behind the tail right before a crash. If the tail had been pushed
972 * immediately prior to the crash and the subsequent checkpoint was only
973 * partially written, it's possible it overwrote the last referenced tail in the
974 * log with garbage. This is not a coherency problem because the tail must have
975 * been pushed before it can be overwritten, but appears as log corruption to
976 * recovery because we have no way to know the tail was updated if the
977 * subsequent checkpoint didn't write successfully.
979 * Therefore, CRC check the log from tail to head. If a failure occurs and the
980 * offending record is within max iclog bufs from the head, walk the tail
981 * forward and retry until a valid tail is found or corruption is detected out
982 * of the range of a possible overwrite.
987 xfs_daddr_t head_blk,
988 xfs_daddr_t *tail_blk,
991 struct xlog_rec_header *thead;
993 xfs_daddr_t first_bad;
996 xfs_daddr_t tmp_tail;
997 xfs_daddr_t orig_tail = *tail_blk;
999 buffer = xlog_alloc_buffer(log, 1);
1004 * Make sure the tail points to a record (returns positive count on
1007 error = xlog_seek_logrec_hdr(log, head_blk, *tail_blk, 1, buffer,
1008 &tmp_tail, &thead, &wrapped);
1011 if (*tail_blk != tmp_tail)
1012 *tail_blk = tmp_tail;
1015 * Run a CRC check from the tail to the head. We can't just check
1016 * MAX_ICLOGS records past the tail because the tail may point to stale
1017 * blocks cleared during the search for the head/tail. These blocks are
1018 * overwritten with zero-length records and thus record count is not a
1019 * reliable indicator of the iclog state before a crash.
1022 error = xlog_do_recovery_pass(log, head_blk, *tail_blk,
1023 XLOG_RECOVER_CRCPASS, &first_bad);
1024 while ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) {
1028 * Is corruption within range of the head? If so, retry from
1029 * the next record. Otherwise return an error.
1031 tail_distance = xlog_tail_distance(log, head_blk, first_bad);
1032 if (tail_distance > BTOBB(XLOG_MAX_ICLOGS * hsize))
1035 /* skip to the next record; returns positive count on success */
1036 error = xlog_seek_logrec_hdr(log, head_blk, first_bad, 2,
1037 buffer, &tmp_tail, &thead, &wrapped);
1041 *tail_blk = tmp_tail;
1043 error = xlog_do_recovery_pass(log, head_blk, *tail_blk,
1044 XLOG_RECOVER_CRCPASS, &first_bad);
1047 if (!error && *tail_blk != orig_tail)
1049 "Tail block (0x%llx) overwrite detected. Updated to 0x%llx",
1050 orig_tail, *tail_blk);
1057 * Detect and trim torn writes from the head of the log.
1059 * Storage without sector atomicity guarantees can result in torn writes in the
1060 * log in the event of a crash. Our only means to detect this scenario is via
1061 * CRC verification. While we can't always be certain that CRC verification
1062 * failure is due to a torn write vs. an unrelated corruption, we do know that
1063 * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at
1064 * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of
1065 * the log and treat failures in this range as torn writes as a matter of
1066 * policy. In the event of CRC failure, the head is walked back to the last good
1067 * record in the log and the tail is updated from that record and verified.
1072 xfs_daddr_t *head_blk, /* in/out: unverified head */
1073 xfs_daddr_t *tail_blk, /* out: tail block */
1075 xfs_daddr_t *rhead_blk, /* start blk of last record */
1076 struct xlog_rec_header **rhead, /* ptr to last record */
1077 bool *wrapped) /* last rec. wraps phys. log */
1079 struct xlog_rec_header *tmp_rhead;
1081 xfs_daddr_t first_bad;
1082 xfs_daddr_t tmp_rhead_blk;
1088 * Check the head of the log for torn writes. Search backwards from the
1089 * head until we hit the tail or the maximum number of log record I/Os
1090 * that could have been in flight at one time. Use a temporary buffer so
1091 * we don't trash the rhead/buffer pointers from the caller.
1093 tmp_buffer = xlog_alloc_buffer(log, 1);
1096 error = xlog_rseek_logrec_hdr(log, *head_blk, *tail_blk,
1097 XLOG_MAX_ICLOGS, tmp_buffer,
1098 &tmp_rhead_blk, &tmp_rhead, &tmp_wrapped);
1099 kmem_free(tmp_buffer);
1104 * Now run a CRC verification pass over the records starting at the
1105 * block found above to the current head. If a CRC failure occurs, the
1106 * log block of the first bad record is saved in first_bad.
1108 error = xlog_do_recovery_pass(log, *head_blk, tmp_rhead_blk,
1109 XLOG_RECOVER_CRCPASS, &first_bad);
1110 if ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) {
1112 * We've hit a potential torn write. Reset the error and warn
1117 "Torn write (CRC failure) detected at log block 0x%llx. Truncating head block from 0x%llx.",
1118 first_bad, *head_blk);
1121 * Get the header block and buffer pointer for the last good
1122 * record before the bad record.
1124 * Note that xlog_find_tail() clears the blocks at the new head
1125 * (i.e., the records with invalid CRC) if the cycle number
1126 * matches the the current cycle.
1128 found = xlog_rseek_logrec_hdr(log, first_bad, *tail_blk, 1,
1129 buffer, rhead_blk, rhead, wrapped);
1132 if (found == 0) /* XXX: right thing to do here? */
1136 * Reset the head block to the starting block of the first bad
1137 * log record and set the tail block based on the last good
1140 * Bail out if the updated head/tail match as this indicates
1141 * possible corruption outside of the acceptable
1142 * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair...
1144 *head_blk = first_bad;
1145 *tail_blk = BLOCK_LSN(be64_to_cpu((*rhead)->h_tail_lsn));
1146 if (*head_blk == *tail_blk) {
1154 return xlog_verify_tail(log, *head_blk, tail_blk,
1155 be32_to_cpu((*rhead)->h_size));
1159 * We need to make sure we handle log wrapping properly, so we can't use the
1160 * calculated logbno directly. Make sure it wraps to the correct bno inside the
1163 * The log is limited to 32 bit sizes, so we use the appropriate modulus
1164 * operation here and cast it back to a 64 bit daddr on return.
1166 static inline xfs_daddr_t
1173 div_s64_rem(bno, log->l_logBBsize, &mod);
1178 * Check whether the head of the log points to an unmount record. In other
1179 * words, determine whether the log is clean. If so, update the in-core state
1183 xlog_check_unmount_rec(
1185 xfs_daddr_t *head_blk,
1186 xfs_daddr_t *tail_blk,
1187 struct xlog_rec_header *rhead,
1188 xfs_daddr_t rhead_blk,
1192 struct xlog_op_header *op_head;
1193 xfs_daddr_t umount_data_blk;
1194 xfs_daddr_t after_umount_blk;
1202 * Look for unmount record. If we find it, then we know there was a
1203 * clean unmount. Since 'i' could be the last block in the physical
1204 * log, we convert to a log block before comparing to the head_blk.
1206 * Save the current tail lsn to use to pass to xlog_clear_stale_blocks()
1207 * below. We won't want to clear the unmount record if there is one, so
1208 * we pass the lsn of the unmount record rather than the block after it.
1210 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1211 int h_size = be32_to_cpu(rhead->h_size);
1212 int h_version = be32_to_cpu(rhead->h_version);
1214 if ((h_version & XLOG_VERSION_2) &&
1215 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
1216 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
1217 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1226 after_umount_blk = xlog_wrap_logbno(log,
1227 rhead_blk + hblks + BTOBB(be32_to_cpu(rhead->h_len)));
1229 if (*head_blk == after_umount_blk &&
1230 be32_to_cpu(rhead->h_num_logops) == 1) {
1231 umount_data_blk = xlog_wrap_logbno(log, rhead_blk + hblks);
1232 error = xlog_bread(log, umount_data_blk, 1, buffer, &offset);
1236 op_head = (struct xlog_op_header *)offset;
1237 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1239 * Set tail and last sync so that newly written log
1240 * records will point recovery to after the current
1243 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1244 log->l_curr_cycle, after_umount_blk);
1245 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1246 log->l_curr_cycle, after_umount_blk);
1247 *tail_blk = after_umount_blk;
1259 xfs_daddr_t head_blk,
1260 struct xlog_rec_header *rhead,
1261 xfs_daddr_t rhead_blk,
1265 * Reset log values according to the state of the log when we
1266 * crashed. In the case where head_blk == 0, we bump curr_cycle
1267 * one because the next write starts a new cycle rather than
1268 * continuing the cycle of the last good log record. At this
1269 * point we have guaranteed that all partial log records have been
1270 * accounted for. Therefore, we know that the last good log record
1271 * written was complete and ended exactly on the end boundary
1272 * of the physical log.
1274 log->l_prev_block = rhead_blk;
1275 log->l_curr_block = (int)head_blk;
1276 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
1278 log->l_curr_cycle++;
1279 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
1280 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
1281 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
1282 BBTOB(log->l_curr_block));
1283 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
1284 BBTOB(log->l_curr_block));
1288 * Find the sync block number or the tail of the log.
1290 * This will be the block number of the last record to have its
1291 * associated buffers synced to disk. Every log record header has
1292 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
1293 * to get a sync block number. The only concern is to figure out which
1294 * log record header to believe.
1296 * The following algorithm uses the log record header with the largest
1297 * lsn. The entire log record does not need to be valid. We only care
1298 * that the header is valid.
1300 * We could speed up search by using current head_blk buffer, but it is not
1306 xfs_daddr_t *head_blk,
1307 xfs_daddr_t *tail_blk)
1309 xlog_rec_header_t *rhead;
1310 char *offset = NULL;
1313 xfs_daddr_t rhead_blk;
1315 bool wrapped = false;
1319 * Find previous log record
1321 if ((error = xlog_find_head(log, head_blk)))
1323 ASSERT(*head_blk < INT_MAX);
1325 buffer = xlog_alloc_buffer(log, 1);
1328 if (*head_blk == 0) { /* special case */
1329 error = xlog_bread(log, 0, 1, buffer, &offset);
1333 if (xlog_get_cycle(offset) == 0) {
1335 /* leave all other log inited values alone */
1341 * Search backwards through the log looking for the log record header
1342 * block. This wraps all the way back around to the head so something is
1343 * seriously wrong if we can't find it.
1345 error = xlog_rseek_logrec_hdr(log, *head_blk, *head_blk, 1, buffer,
1346 &rhead_blk, &rhead, &wrapped);
1350 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
1353 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
1356 * Set the log state based on the current head record.
1358 xlog_set_state(log, *head_blk, rhead, rhead_blk, wrapped);
1359 tail_lsn = atomic64_read(&log->l_tail_lsn);
1362 * Look for an unmount record at the head of the log. This sets the log
1363 * state to determine whether recovery is necessary.
1365 error = xlog_check_unmount_rec(log, head_blk, tail_blk, rhead,
1366 rhead_blk, buffer, &clean);
1371 * Verify the log head if the log is not clean (e.g., we have anything
1372 * but an unmount record at the head). This uses CRC verification to
1373 * detect and trim torn writes. If discovered, CRC failures are
1374 * considered torn writes and the log head is trimmed accordingly.
1376 * Note that we can only run CRC verification when the log is dirty
1377 * because there's no guarantee that the log data behind an unmount
1378 * record is compatible with the current architecture.
1381 xfs_daddr_t orig_head = *head_blk;
1383 error = xlog_verify_head(log, head_blk, tail_blk, buffer,
1384 &rhead_blk, &rhead, &wrapped);
1388 /* update in-core state again if the head changed */
1389 if (*head_blk != orig_head) {
1390 xlog_set_state(log, *head_blk, rhead, rhead_blk,
1392 tail_lsn = atomic64_read(&log->l_tail_lsn);
1393 error = xlog_check_unmount_rec(log, head_blk, tail_blk,
1394 rhead, rhead_blk, buffer,
1402 * Note that the unmount was clean. If the unmount was not clean, we
1403 * need to know this to rebuild the superblock counters from the perag
1404 * headers if we have a filesystem using non-persistent counters.
1407 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1410 * Make sure that there are no blocks in front of the head
1411 * with the same cycle number as the head. This can happen
1412 * because we allow multiple outstanding log writes concurrently,
1413 * and the later writes might make it out before earlier ones.
1415 * We use the lsn from before modifying it so that we'll never
1416 * overwrite the unmount record after a clean unmount.
1418 * Do this only if we are going to recover the filesystem
1420 * NOTE: This used to say "if (!readonly)"
1421 * However on Linux, we can & do recover a read-only filesystem.
1422 * We only skip recovery if NORECOVERY is specified on mount,
1423 * in which case we would not be here.
1425 * But... if the -device- itself is readonly, just skip this.
1426 * We can't recover this device anyway, so it won't matter.
1428 if (!xfs_readonly_buftarg(log->l_targ))
1429 error = xlog_clear_stale_blocks(log, tail_lsn);
1435 xfs_warn(log->l_mp, "failed to locate log tail");
1440 * Is the log zeroed at all?
1442 * The last binary search should be changed to perform an X block read
1443 * once X becomes small enough. You can then search linearly through
1444 * the X blocks. This will cut down on the number of reads we need to do.
1446 * If the log is partially zeroed, this routine will pass back the blkno
1447 * of the first block with cycle number 0. It won't have a complete LR
1451 * 0 => the log is completely written to
1452 * 1 => use *blk_no as the first block of the log
1453 * <0 => error has occurred
1458 xfs_daddr_t *blk_no)
1462 uint first_cycle, last_cycle;
1463 xfs_daddr_t new_blk, last_blk, start_blk;
1464 xfs_daddr_t num_scan_bblks;
1465 int error, log_bbnum = log->l_logBBsize;
1469 /* check totally zeroed log */
1470 buffer = xlog_alloc_buffer(log, 1);
1473 error = xlog_bread(log, 0, 1, buffer, &offset);
1475 goto out_free_buffer;
1477 first_cycle = xlog_get_cycle(offset);
1478 if (first_cycle == 0) { /* completely zeroed log */
1484 /* check partially zeroed log */
1485 error = xlog_bread(log, log_bbnum-1, 1, buffer, &offset);
1487 goto out_free_buffer;
1489 last_cycle = xlog_get_cycle(offset);
1490 if (last_cycle != 0) { /* log completely written to */
1495 /* we have a partially zeroed log */
1496 last_blk = log_bbnum-1;
1497 error = xlog_find_cycle_start(log, buffer, 0, &last_blk, 0);
1499 goto out_free_buffer;
1502 * Validate the answer. Because there is no way to guarantee that
1503 * the entire log is made up of log records which are the same size,
1504 * we scan over the defined maximum blocks. At this point, the maximum
1505 * is not chosen to mean anything special. XXXmiken
1507 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1508 ASSERT(num_scan_bblks <= INT_MAX);
1510 if (last_blk < num_scan_bblks)
1511 num_scan_bblks = last_blk;
1512 start_blk = last_blk - num_scan_bblks;
1515 * We search for any instances of cycle number 0 that occur before
1516 * our current estimate of the head. What we're trying to detect is
1517 * 1 ... | 0 | 1 | 0...
1518 * ^ binary search ends here
1520 if ((error = xlog_find_verify_cycle(log, start_blk,
1521 (int)num_scan_bblks, 0, &new_blk)))
1522 goto out_free_buffer;
1527 * Potentially backup over partial log record write. We don't need
1528 * to search the end of the log because we know it is zero.
1530 error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0);
1534 goto out_free_buffer;
1545 * These are simple subroutines used by xlog_clear_stale_blocks() below
1546 * to initialize a buffer full of empty log record headers and write
1547 * them into the log.
1558 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1560 memset(buf, 0, BBSIZE);
1561 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1562 recp->h_cycle = cpu_to_be32(cycle);
1563 recp->h_version = cpu_to_be32(
1564 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1565 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1566 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1567 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1568 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1572 xlog_write_log_records(
1583 int sectbb = log->l_sectBBsize;
1584 int end_block = start_block + blocks;
1590 * Greedily allocate a buffer big enough to handle the full
1591 * range of basic blocks to be written. If that fails, try
1592 * a smaller size. We need to be able to write at least a
1593 * log sector, or we're out of luck.
1595 bufblks = 1 << ffs(blocks);
1596 while (bufblks > log->l_logBBsize)
1598 while (!(buffer = xlog_alloc_buffer(log, bufblks))) {
1600 if (bufblks < sectbb)
1604 /* We may need to do a read at the start to fill in part of
1605 * the buffer in the starting sector not covered by the first
1608 balign = round_down(start_block, sectbb);
1609 if (balign != start_block) {
1610 error = xlog_bread_noalign(log, start_block, 1, buffer);
1612 goto out_free_buffer;
1614 j = start_block - balign;
1617 for (i = start_block; i < end_block; i += bufblks) {
1618 int bcount, endcount;
1620 bcount = min(bufblks, end_block - start_block);
1621 endcount = bcount - j;
1623 /* We may need to do a read at the end to fill in part of
1624 * the buffer in the final sector not covered by the write.
1625 * If this is the same sector as the above read, skip it.
1627 ealign = round_down(end_block, sectbb);
1628 if (j == 0 && (start_block + endcount > ealign)) {
1629 error = xlog_bread_noalign(log, ealign, sectbb,
1630 buffer + BBTOB(ealign - start_block));
1636 offset = buffer + xlog_align(log, start_block);
1637 for (; j < endcount; j++) {
1638 xlog_add_record(log, offset, cycle, i+j,
1639 tail_cycle, tail_block);
1642 error = xlog_bwrite(log, start_block, endcount, buffer);
1645 start_block += endcount;
1655 * This routine is called to blow away any incomplete log writes out
1656 * in front of the log head. We do this so that we won't become confused
1657 * if we come up, write only a little bit more, and then crash again.
1658 * If we leave the partial log records out there, this situation could
1659 * cause us to think those partial writes are valid blocks since they
1660 * have the current cycle number. We get rid of them by overwriting them
1661 * with empty log records with the old cycle number rather than the
1664 * The tail lsn is passed in rather than taken from
1665 * the log so that we will not write over the unmount record after a
1666 * clean unmount in a 512 block log. Doing so would leave the log without
1667 * any valid log records in it until a new one was written. If we crashed
1668 * during that time we would not be able to recover.
1671 xlog_clear_stale_blocks(
1675 int tail_cycle, head_cycle;
1676 int tail_block, head_block;
1677 int tail_distance, max_distance;
1681 tail_cycle = CYCLE_LSN(tail_lsn);
1682 tail_block = BLOCK_LSN(tail_lsn);
1683 head_cycle = log->l_curr_cycle;
1684 head_block = log->l_curr_block;
1687 * Figure out the distance between the new head of the log
1688 * and the tail. We want to write over any blocks beyond the
1689 * head that we may have written just before the crash, but
1690 * we don't want to overwrite the tail of the log.
1692 if (head_cycle == tail_cycle) {
1694 * The tail is behind the head in the physical log,
1695 * so the distance from the head to the tail is the
1696 * distance from the head to the end of the log plus
1697 * the distance from the beginning of the log to the
1700 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1701 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1702 XFS_ERRLEVEL_LOW, log->l_mp);
1703 return -EFSCORRUPTED;
1705 tail_distance = tail_block + (log->l_logBBsize - head_block);
1708 * The head is behind the tail in the physical log,
1709 * so the distance from the head to the tail is just
1710 * the tail block minus the head block.
1712 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1713 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1714 XFS_ERRLEVEL_LOW, log->l_mp);
1715 return -EFSCORRUPTED;
1717 tail_distance = tail_block - head_block;
1721 * If the head is right up against the tail, we can't clear
1724 if (tail_distance <= 0) {
1725 ASSERT(tail_distance == 0);
1729 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1731 * Take the smaller of the maximum amount of outstanding I/O
1732 * we could have and the distance to the tail to clear out.
1733 * We take the smaller so that we don't overwrite the tail and
1734 * we don't waste all day writing from the head to the tail
1737 max_distance = min(max_distance, tail_distance);
1739 if ((head_block + max_distance) <= log->l_logBBsize) {
1741 * We can stomp all the blocks we need to without
1742 * wrapping around the end of the log. Just do it
1743 * in a single write. Use the cycle number of the
1744 * current cycle minus one so that the log will look like:
1747 error = xlog_write_log_records(log, (head_cycle - 1),
1748 head_block, max_distance, tail_cycle,
1754 * We need to wrap around the end of the physical log in
1755 * order to clear all the blocks. Do it in two separate
1756 * I/Os. The first write should be from the head to the
1757 * end of the physical log, and it should use the current
1758 * cycle number minus one just like above.
1760 distance = log->l_logBBsize - head_block;
1761 error = xlog_write_log_records(log, (head_cycle - 1),
1762 head_block, distance, tail_cycle,
1769 * Now write the blocks at the start of the physical log.
1770 * This writes the remainder of the blocks we want to clear.
1771 * It uses the current cycle number since we're now on the
1772 * same cycle as the head so that we get:
1773 * n ... n ... | n - 1 ...
1774 * ^^^^^ blocks we're writing
1776 distance = max_distance - (log->l_logBBsize - head_block);
1777 error = xlog_write_log_records(log, head_cycle, 0, distance,
1778 tail_cycle, tail_block);
1786 /******************************************************************************
1788 * Log recover routines
1790 ******************************************************************************
1794 * Sort the log items in the transaction.
1796 * The ordering constraints are defined by the inode allocation and unlink
1797 * behaviour. The rules are:
1799 * 1. Every item is only logged once in a given transaction. Hence it
1800 * represents the last logged state of the item. Hence ordering is
1801 * dependent on the order in which operations need to be performed so
1802 * required initial conditions are always met.
1804 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1805 * there's nothing to replay from them so we can simply cull them
1806 * from the transaction. However, we can't do that until after we've
1807 * replayed all the other items because they may be dependent on the
1808 * cancelled buffer and replaying the cancelled buffer can remove it
1809 * form the cancelled buffer table. Hence they have tobe done last.
1811 * 3. Inode allocation buffers must be replayed before inode items that
1812 * read the buffer and replay changes into it. For filesystems using the
1813 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1814 * treated the same as inode allocation buffers as they create and
1815 * initialise the buffers directly.
1817 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1818 * This ensures that inodes are completely flushed to the inode buffer
1819 * in a "free" state before we remove the unlinked inode list pointer.
1821 * Hence the ordering needs to be inode allocation buffers first, inode items
1822 * second, inode unlink buffers third and cancelled buffers last.
1824 * But there's a problem with that - we can't tell an inode allocation buffer
1825 * apart from a regular buffer, so we can't separate them. We can, however,
1826 * tell an inode unlink buffer from the others, and so we can separate them out
1827 * from all the other buffers and move them to last.
1829 * Hence, 4 lists, in order from head to tail:
1830 * - buffer_list for all buffers except cancelled/inode unlink buffers
1831 * - item_list for all non-buffer items
1832 * - inode_buffer_list for inode unlink buffers
1833 * - cancel_list for the cancelled buffers
1835 * Note that we add objects to the tail of the lists so that first-to-last
1836 * ordering is preserved within the lists. Adding objects to the head of the
1837 * list means when we traverse from the head we walk them in last-to-first
1838 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1839 * but for all other items there may be specific ordering that we need to
1843 xlog_recover_reorder_trans(
1845 struct xlog_recover *trans,
1848 xlog_recover_item_t *item, *n;
1850 LIST_HEAD(sort_list);
1851 LIST_HEAD(cancel_list);
1852 LIST_HEAD(buffer_list);
1853 LIST_HEAD(inode_buffer_list);
1854 LIST_HEAD(inode_list);
1856 list_splice_init(&trans->r_itemq, &sort_list);
1857 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1858 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1860 switch (ITEM_TYPE(item)) {
1861 case XFS_LI_ICREATE:
1862 list_move_tail(&item->ri_list, &buffer_list);
1865 if (buf_f->blf_flags & XFS_BLF_CANCEL) {
1866 trace_xfs_log_recover_item_reorder_head(log,
1868 list_move(&item->ri_list, &cancel_list);
1871 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1872 list_move(&item->ri_list, &inode_buffer_list);
1875 list_move_tail(&item->ri_list, &buffer_list);
1879 case XFS_LI_QUOTAOFF:
1888 trace_xfs_log_recover_item_reorder_tail(log,
1890 list_move_tail(&item->ri_list, &inode_list);
1894 "%s: unrecognized type of log operation",
1898 * return the remaining items back to the transaction
1899 * item list so they can be freed in caller.
1901 if (!list_empty(&sort_list))
1902 list_splice_init(&sort_list, &trans->r_itemq);
1908 ASSERT(list_empty(&sort_list));
1909 if (!list_empty(&buffer_list))
1910 list_splice(&buffer_list, &trans->r_itemq);
1911 if (!list_empty(&inode_list))
1912 list_splice_tail(&inode_list, &trans->r_itemq);
1913 if (!list_empty(&inode_buffer_list))
1914 list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1915 if (!list_empty(&cancel_list))
1916 list_splice_tail(&cancel_list, &trans->r_itemq);
1921 * Build up the table of buf cancel records so that we don't replay
1922 * cancelled data in the second pass. For buffer records that are
1923 * not cancel records, there is nothing to do here so we just return.
1925 * If we get a cancel record which is already in the table, this indicates
1926 * that the buffer was cancelled multiple times. In order to ensure
1927 * that during pass 2 we keep the record in the table until we reach its
1928 * last occurrence in the log, we keep a reference count in the cancel
1929 * record in the table to tell us how many times we expect to see this
1930 * record during the second pass.
1933 xlog_recover_buffer_pass1(
1935 struct xlog_recover_item *item)
1937 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1938 struct list_head *bucket;
1939 struct xfs_buf_cancel *bcp;
1942 * If this isn't a cancel buffer item, then just return.
1944 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1945 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
1950 * Insert an xfs_buf_cancel record into the hash table of them.
1951 * If there is already an identical record, bump its reference count.
1953 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1954 list_for_each_entry(bcp, bucket, bc_list) {
1955 if (bcp->bc_blkno == buf_f->blf_blkno &&
1956 bcp->bc_len == buf_f->blf_len) {
1958 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
1963 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1964 bcp->bc_blkno = buf_f->blf_blkno;
1965 bcp->bc_len = buf_f->blf_len;
1966 bcp->bc_refcount = 1;
1967 list_add_tail(&bcp->bc_list, bucket);
1969 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
1974 * Check to see whether the buffer being recovered has a corresponding
1975 * entry in the buffer cancel record table. If it is, return the cancel
1976 * buffer structure to the caller.
1978 STATIC struct xfs_buf_cancel *
1979 xlog_peek_buffer_cancelled(
1983 unsigned short flags)
1985 struct list_head *bucket;
1986 struct xfs_buf_cancel *bcp;
1988 if (!log->l_buf_cancel_table) {
1989 /* empty table means no cancelled buffers in the log */
1990 ASSERT(!(flags & XFS_BLF_CANCEL));
1994 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1995 list_for_each_entry(bcp, bucket, bc_list) {
1996 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
2001 * We didn't find a corresponding entry in the table, so return 0 so
2002 * that the buffer is NOT cancelled.
2004 ASSERT(!(flags & XFS_BLF_CANCEL));
2009 * If the buffer is being cancelled then return 1 so that it will be cancelled,
2010 * otherwise return 0. If the buffer is actually a buffer cancel item
2011 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
2012 * table and remove it from the table if this is the last reference.
2014 * We remove the cancel record from the table when we encounter its last
2015 * occurrence in the log so that if the same buffer is re-used again after its
2016 * last cancellation we actually replay the changes made at that point.
2019 xlog_check_buffer_cancelled(
2023 unsigned short flags)
2025 struct xfs_buf_cancel *bcp;
2027 bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags);
2032 * We've go a match, so return 1 so that the recovery of this buffer
2033 * is cancelled. If this buffer is actually a buffer cancel log
2034 * item, then decrement the refcount on the one in the table and
2035 * remove it if this is the last reference.
2037 if (flags & XFS_BLF_CANCEL) {
2038 if (--bcp->bc_refcount == 0) {
2039 list_del(&bcp->bc_list);
2047 * Perform recovery for a buffer full of inodes. In these buffers, the only
2048 * data which should be recovered is that which corresponds to the
2049 * di_next_unlinked pointers in the on disk inode structures. The rest of the
2050 * data for the inodes is always logged through the inodes themselves rather
2051 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
2053 * The only time when buffers full of inodes are fully recovered is when the
2054 * buffer is full of newly allocated inodes. In this case the buffer will
2055 * not be marked as an inode buffer and so will be sent to
2056 * xlog_recover_do_reg_buffer() below during recovery.
2059 xlog_recover_do_inode_buffer(
2060 struct xfs_mount *mp,
2061 xlog_recover_item_t *item,
2063 xfs_buf_log_format_t *buf_f)
2069 int reg_buf_offset = 0;
2070 int reg_buf_bytes = 0;
2071 int next_unlinked_offset;
2073 xfs_agino_t *logged_nextp;
2074 xfs_agino_t *buffer_nextp;
2076 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
2079 * Post recovery validation only works properly on CRC enabled
2082 if (xfs_sb_version_hascrc(&mp->m_sb))
2083 bp->b_ops = &xfs_inode_buf_ops;
2085 inodes_per_buf = BBTOB(bp->b_length) >> mp->m_sb.sb_inodelog;
2086 for (i = 0; i < inodes_per_buf; i++) {
2087 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
2088 offsetof(xfs_dinode_t, di_next_unlinked);
2090 while (next_unlinked_offset >=
2091 (reg_buf_offset + reg_buf_bytes)) {
2093 * The next di_next_unlinked field is beyond
2094 * the current logged region. Find the next
2095 * logged region that contains or is beyond
2096 * the current di_next_unlinked field.
2099 bit = xfs_next_bit(buf_f->blf_data_map,
2100 buf_f->blf_map_size, bit);
2103 * If there are no more logged regions in the
2104 * buffer, then we're done.
2109 nbits = xfs_contig_bits(buf_f->blf_data_map,
2110 buf_f->blf_map_size, bit);
2112 reg_buf_offset = bit << XFS_BLF_SHIFT;
2113 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
2118 * If the current logged region starts after the current
2119 * di_next_unlinked field, then move on to the next
2120 * di_next_unlinked field.
2122 if (next_unlinked_offset < reg_buf_offset)
2125 ASSERT(item->ri_buf[item_index].i_addr != NULL);
2126 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
2127 ASSERT((reg_buf_offset + reg_buf_bytes) <= BBTOB(bp->b_length));
2130 * The current logged region contains a copy of the
2131 * current di_next_unlinked field. Extract its value
2132 * and copy it to the buffer copy.
2134 logged_nextp = item->ri_buf[item_index].i_addr +
2135 next_unlinked_offset - reg_buf_offset;
2136 if (unlikely(*logged_nextp == 0)) {
2138 "Bad inode buffer log record (ptr = "PTR_FMT", bp = "PTR_FMT"). "
2139 "Trying to replay bad (0) inode di_next_unlinked field.",
2141 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
2142 XFS_ERRLEVEL_LOW, mp);
2143 return -EFSCORRUPTED;
2146 buffer_nextp = xfs_buf_offset(bp, next_unlinked_offset);
2147 *buffer_nextp = *logged_nextp;
2150 * If necessary, recalculate the CRC in the on-disk inode. We
2151 * have to leave the inode in a consistent state for whoever
2154 xfs_dinode_calc_crc(mp,
2155 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
2163 * V5 filesystems know the age of the buffer on disk being recovered. We can
2164 * have newer objects on disk than we are replaying, and so for these cases we
2165 * don't want to replay the current change as that will make the buffer contents
2166 * temporarily invalid on disk.
2168 * The magic number might not match the buffer type we are going to recover
2169 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
2170 * extract the LSN of the existing object in the buffer based on it's current
2171 * magic number. If we don't recognise the magic number in the buffer, then
2172 * return a LSN of -1 so that the caller knows it was an unrecognised block and
2173 * so can recover the buffer.
2175 * Note: we cannot rely solely on magic number matches to determine that the
2176 * buffer has a valid LSN - we also need to verify that it belongs to this
2177 * filesystem, so we need to extract the object's LSN and compare it to that
2178 * which we read from the superblock. If the UUIDs don't match, then we've got a
2179 * stale metadata block from an old filesystem instance that we need to recover
2183 xlog_recover_get_buf_lsn(
2184 struct xfs_mount *mp,
2190 void *blk = bp->b_addr;
2194 /* v4 filesystems always recover immediately */
2195 if (!xfs_sb_version_hascrc(&mp->m_sb))
2196 goto recover_immediately;
2198 magic32 = be32_to_cpu(*(__be32 *)blk);
2200 case XFS_ABTB_CRC_MAGIC:
2201 case XFS_ABTC_CRC_MAGIC:
2202 case XFS_ABTB_MAGIC:
2203 case XFS_ABTC_MAGIC:
2204 case XFS_RMAP_CRC_MAGIC:
2205 case XFS_REFC_CRC_MAGIC:
2206 case XFS_IBT_CRC_MAGIC:
2207 case XFS_IBT_MAGIC: {
2208 struct xfs_btree_block *btb = blk;
2210 lsn = be64_to_cpu(btb->bb_u.s.bb_lsn);
2211 uuid = &btb->bb_u.s.bb_uuid;
2214 case XFS_BMAP_CRC_MAGIC:
2215 case XFS_BMAP_MAGIC: {
2216 struct xfs_btree_block *btb = blk;
2218 lsn = be64_to_cpu(btb->bb_u.l.bb_lsn);
2219 uuid = &btb->bb_u.l.bb_uuid;
2223 lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn);
2224 uuid = &((struct xfs_agf *)blk)->agf_uuid;
2226 case XFS_AGFL_MAGIC:
2227 lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn);
2228 uuid = &((struct xfs_agfl *)blk)->agfl_uuid;
2231 lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn);
2232 uuid = &((struct xfs_agi *)blk)->agi_uuid;
2234 case XFS_SYMLINK_MAGIC:
2235 lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn);
2236 uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid;
2238 case XFS_DIR3_BLOCK_MAGIC:
2239 case XFS_DIR3_DATA_MAGIC:
2240 case XFS_DIR3_FREE_MAGIC:
2241 lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn);
2242 uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid;
2244 case XFS_ATTR3_RMT_MAGIC:
2246 * Remote attr blocks are written synchronously, rather than
2247 * being logged. That means they do not contain a valid LSN
2248 * (i.e. transactionally ordered) in them, and hence any time we
2249 * see a buffer to replay over the top of a remote attribute
2250 * block we should simply do so.
2252 goto recover_immediately;
2255 * superblock uuids are magic. We may or may not have a
2256 * sb_meta_uuid on disk, but it will be set in the in-core
2257 * superblock. We set the uuid pointer for verification
2258 * according to the superblock feature mask to ensure we check
2259 * the relevant UUID in the superblock.
2261 lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn);
2262 if (xfs_sb_version_hasmetauuid(&mp->m_sb))
2263 uuid = &((struct xfs_dsb *)blk)->sb_meta_uuid;
2265 uuid = &((struct xfs_dsb *)blk)->sb_uuid;
2271 if (lsn != (xfs_lsn_t)-1) {
2272 if (!uuid_equal(&mp->m_sb.sb_meta_uuid, uuid))
2273 goto recover_immediately;
2277 magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic);
2279 case XFS_DIR3_LEAF1_MAGIC:
2280 case XFS_DIR3_LEAFN_MAGIC:
2281 case XFS_DA3_NODE_MAGIC:
2282 lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn);
2283 uuid = &((struct xfs_da3_blkinfo *)blk)->uuid;
2289 if (lsn != (xfs_lsn_t)-1) {
2290 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
2291 goto recover_immediately;
2296 * We do individual object checks on dquot and inode buffers as they
2297 * have their own individual LSN records. Also, we could have a stale
2298 * buffer here, so we have to at least recognise these buffer types.
2300 * A notd complexity here is inode unlinked list processing - it logs
2301 * the inode directly in the buffer, but we don't know which inodes have
2302 * been modified, and there is no global buffer LSN. Hence we need to
2303 * recover all inode buffer types immediately. This problem will be
2304 * fixed by logical logging of the unlinked list modifications.
2306 magic16 = be16_to_cpu(*(__be16 *)blk);
2308 case XFS_DQUOT_MAGIC:
2309 case XFS_DINODE_MAGIC:
2310 goto recover_immediately;
2315 /* unknown buffer contents, recover immediately */
2317 recover_immediately:
2318 return (xfs_lsn_t)-1;
2323 * Validate the recovered buffer is of the correct type and attach the
2324 * appropriate buffer operations to them for writeback. Magic numbers are in a
2326 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2327 * the first 32 bits of the buffer (most blocks),
2328 * inside a struct xfs_da_blkinfo at the start of the buffer.
2331 xlog_recover_validate_buf_type(
2332 struct xfs_mount *mp,
2334 xfs_buf_log_format_t *buf_f,
2335 xfs_lsn_t current_lsn)
2337 struct xfs_da_blkinfo *info = bp->b_addr;
2341 char *warnmsg = NULL;
2344 * We can only do post recovery validation on items on CRC enabled
2345 * fielsystems as we need to know when the buffer was written to be able
2346 * to determine if we should have replayed the item. If we replay old
2347 * metadata over a newer buffer, then it will enter a temporarily
2348 * inconsistent state resulting in verification failures. Hence for now
2349 * just avoid the verification stage for non-crc filesystems
2351 if (!xfs_sb_version_hascrc(&mp->m_sb))
2354 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
2355 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
2356 magicda = be16_to_cpu(info->magic);
2357 switch (xfs_blft_from_flags(buf_f)) {
2358 case XFS_BLFT_BTREE_BUF:
2360 case XFS_ABTB_CRC_MAGIC:
2361 case XFS_ABTB_MAGIC:
2362 bp->b_ops = &xfs_bnobt_buf_ops;
2364 case XFS_ABTC_CRC_MAGIC:
2365 case XFS_ABTC_MAGIC:
2366 bp->b_ops = &xfs_cntbt_buf_ops;
2368 case XFS_IBT_CRC_MAGIC:
2370 bp->b_ops = &xfs_inobt_buf_ops;
2372 case XFS_FIBT_CRC_MAGIC:
2373 case XFS_FIBT_MAGIC:
2374 bp->b_ops = &xfs_finobt_buf_ops;
2376 case XFS_BMAP_CRC_MAGIC:
2377 case XFS_BMAP_MAGIC:
2378 bp->b_ops = &xfs_bmbt_buf_ops;
2380 case XFS_RMAP_CRC_MAGIC:
2381 bp->b_ops = &xfs_rmapbt_buf_ops;
2383 case XFS_REFC_CRC_MAGIC:
2384 bp->b_ops = &xfs_refcountbt_buf_ops;
2387 warnmsg = "Bad btree block magic!";
2391 case XFS_BLFT_AGF_BUF:
2392 if (magic32 != XFS_AGF_MAGIC) {
2393 warnmsg = "Bad AGF block magic!";
2396 bp->b_ops = &xfs_agf_buf_ops;
2398 case XFS_BLFT_AGFL_BUF:
2399 if (magic32 != XFS_AGFL_MAGIC) {
2400 warnmsg = "Bad AGFL block magic!";
2403 bp->b_ops = &xfs_agfl_buf_ops;
2405 case XFS_BLFT_AGI_BUF:
2406 if (magic32 != XFS_AGI_MAGIC) {
2407 warnmsg = "Bad AGI block magic!";
2410 bp->b_ops = &xfs_agi_buf_ops;
2412 case XFS_BLFT_UDQUOT_BUF:
2413 case XFS_BLFT_PDQUOT_BUF:
2414 case XFS_BLFT_GDQUOT_BUF:
2415 #ifdef CONFIG_XFS_QUOTA
2416 if (magic16 != XFS_DQUOT_MAGIC) {
2417 warnmsg = "Bad DQUOT block magic!";
2420 bp->b_ops = &xfs_dquot_buf_ops;
2423 "Trying to recover dquots without QUOTA support built in!");
2427 case XFS_BLFT_DINO_BUF:
2428 if (magic16 != XFS_DINODE_MAGIC) {
2429 warnmsg = "Bad INODE block magic!";
2432 bp->b_ops = &xfs_inode_buf_ops;
2434 case XFS_BLFT_SYMLINK_BUF:
2435 if (magic32 != XFS_SYMLINK_MAGIC) {
2436 warnmsg = "Bad symlink block magic!";
2439 bp->b_ops = &xfs_symlink_buf_ops;
2441 case XFS_BLFT_DIR_BLOCK_BUF:
2442 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2443 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2444 warnmsg = "Bad dir block magic!";
2447 bp->b_ops = &xfs_dir3_block_buf_ops;
2449 case XFS_BLFT_DIR_DATA_BUF:
2450 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2451 magic32 != XFS_DIR3_DATA_MAGIC) {
2452 warnmsg = "Bad dir data magic!";
2455 bp->b_ops = &xfs_dir3_data_buf_ops;
2457 case XFS_BLFT_DIR_FREE_BUF:
2458 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2459 magic32 != XFS_DIR3_FREE_MAGIC) {
2460 warnmsg = "Bad dir3 free magic!";
2463 bp->b_ops = &xfs_dir3_free_buf_ops;
2465 case XFS_BLFT_DIR_LEAF1_BUF:
2466 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2467 magicda != XFS_DIR3_LEAF1_MAGIC) {
2468 warnmsg = "Bad dir leaf1 magic!";
2471 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2473 case XFS_BLFT_DIR_LEAFN_BUF:
2474 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2475 magicda != XFS_DIR3_LEAFN_MAGIC) {
2476 warnmsg = "Bad dir leafn magic!";
2479 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2481 case XFS_BLFT_DA_NODE_BUF:
2482 if (magicda != XFS_DA_NODE_MAGIC &&
2483 magicda != XFS_DA3_NODE_MAGIC) {
2484 warnmsg = "Bad da node magic!";
2487 bp->b_ops = &xfs_da3_node_buf_ops;
2489 case XFS_BLFT_ATTR_LEAF_BUF:
2490 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2491 magicda != XFS_ATTR3_LEAF_MAGIC) {
2492 warnmsg = "Bad attr leaf magic!";
2495 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2497 case XFS_BLFT_ATTR_RMT_BUF:
2498 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
2499 warnmsg = "Bad attr remote magic!";
2502 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2504 case XFS_BLFT_SB_BUF:
2505 if (magic32 != XFS_SB_MAGIC) {
2506 warnmsg = "Bad SB block magic!";
2509 bp->b_ops = &xfs_sb_buf_ops;
2511 #ifdef CONFIG_XFS_RT
2512 case XFS_BLFT_RTBITMAP_BUF:
2513 case XFS_BLFT_RTSUMMARY_BUF:
2514 /* no magic numbers for verification of RT buffers */
2515 bp->b_ops = &xfs_rtbuf_ops;
2517 #endif /* CONFIG_XFS_RT */
2519 xfs_warn(mp, "Unknown buffer type %d!",
2520 xfs_blft_from_flags(buf_f));
2525 * Nothing else to do in the case of a NULL current LSN as this means
2526 * the buffer is more recent than the change in the log and will be
2529 if (current_lsn == NULLCOMMITLSN)
2533 xfs_warn(mp, warnmsg);
2538 * We must update the metadata LSN of the buffer as it is written out to
2539 * ensure that older transactions never replay over this one and corrupt
2540 * the buffer. This can occur if log recovery is interrupted at some
2541 * point after the current transaction completes, at which point a
2542 * subsequent mount starts recovery from the beginning.
2544 * Write verifiers update the metadata LSN from log items attached to
2545 * the buffer. Therefore, initialize a bli purely to carry the LSN to
2546 * the verifier. We'll clean it up in our ->iodone() callback.
2549 struct xfs_buf_log_item *bip;
2551 ASSERT(!bp->b_iodone || bp->b_iodone == xlog_recover_iodone);
2552 bp->b_iodone = xlog_recover_iodone;
2553 xfs_buf_item_init(bp, mp);
2554 bip = bp->b_log_item;
2555 bip->bli_item.li_lsn = current_lsn;
2560 * Perform a 'normal' buffer recovery. Each logged region of the
2561 * buffer should be copied over the corresponding region in the
2562 * given buffer. The bitmap in the buf log format structure indicates
2563 * where to place the logged data.
2566 xlog_recover_do_reg_buffer(
2567 struct xfs_mount *mp,
2568 xlog_recover_item_t *item,
2570 xfs_buf_log_format_t *buf_f,
2571 xfs_lsn_t current_lsn)
2578 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2581 i = 1; /* 0 is the buf format structure */
2583 bit = xfs_next_bit(buf_f->blf_data_map,
2584 buf_f->blf_map_size, bit);
2587 nbits = xfs_contig_bits(buf_f->blf_data_map,
2588 buf_f->blf_map_size, bit);
2590 ASSERT(item->ri_buf[i].i_addr != NULL);
2591 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
2592 ASSERT(BBTOB(bp->b_length) >=
2593 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
2596 * The dirty regions logged in the buffer, even though
2597 * contiguous, may span multiple chunks. This is because the
2598 * dirty region may span a physical page boundary in a buffer
2599 * and hence be split into two separate vectors for writing into
2600 * the log. Hence we need to trim nbits back to the length of
2601 * the current region being copied out of the log.
2603 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2604 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2607 * Do a sanity check if this is a dquot buffer. Just checking
2608 * the first dquot in the buffer should do. XXXThis is
2609 * probably a good thing to do for other buf types also.
2612 if (buf_f->blf_flags &
2613 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2614 if (item->ri_buf[i].i_addr == NULL) {
2616 "XFS: NULL dquot in %s.", __func__);
2619 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
2621 "XFS: dquot too small (%d) in %s.",
2622 item->ri_buf[i].i_len, __func__);
2625 fa = xfs_dquot_verify(mp, item->ri_buf[i].i_addr,
2629 "dquot corrupt at %pS trying to replay into block 0x%llx",
2635 memcpy(xfs_buf_offset(bp,
2636 (uint)bit << XFS_BLF_SHIFT), /* dest */
2637 item->ri_buf[i].i_addr, /* source */
2638 nbits<<XFS_BLF_SHIFT); /* length */
2644 /* Shouldn't be any more regions */
2645 ASSERT(i == item->ri_total);
2647 xlog_recover_validate_buf_type(mp, bp, buf_f, current_lsn);
2651 * Perform a dquot buffer recovery.
2652 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2653 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2654 * Else, treat it as a regular buffer and do recovery.
2656 * Return false if the buffer was tossed and true if we recovered the buffer to
2657 * indicate to the caller if the buffer needs writing.
2660 xlog_recover_do_dquot_buffer(
2661 struct xfs_mount *mp,
2663 struct xlog_recover_item *item,
2665 struct xfs_buf_log_format *buf_f)
2669 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2672 * Filesystems are required to send in quota flags at mount time.
2678 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2679 type |= XFS_DQ_USER;
2680 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2681 type |= XFS_DQ_PROJ;
2682 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2683 type |= XFS_DQ_GROUP;
2685 * This type of quotas was turned off, so ignore this buffer
2687 if (log->l_quotaoffs_flag & type)
2690 xlog_recover_do_reg_buffer(mp, item, bp, buf_f, NULLCOMMITLSN);
2695 * This routine replays a modification made to a buffer at runtime.
2696 * There are actually two types of buffer, regular and inode, which
2697 * are handled differently. Inode buffers are handled differently
2698 * in that we only recover a specific set of data from them, namely
2699 * the inode di_next_unlinked fields. This is because all other inode
2700 * data is actually logged via inode records and any data we replay
2701 * here which overlaps that may be stale.
2703 * When meta-data buffers are freed at run time we log a buffer item
2704 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2705 * of the buffer in the log should not be replayed at recovery time.
2706 * This is so that if the blocks covered by the buffer are reused for
2707 * file data before we crash we don't end up replaying old, freed
2708 * meta-data into a user's file.
2710 * To handle the cancellation of buffer log items, we make two passes
2711 * over the log during recovery. During the first we build a table of
2712 * those buffers which have been cancelled, and during the second we
2713 * only replay those buffers which do not have corresponding cancel
2714 * records in the table. See xlog_recover_buffer_pass[1,2] above
2715 * for more details on the implementation of the table of cancel records.
2718 xlog_recover_buffer_pass2(
2720 struct list_head *buffer_list,
2721 struct xlog_recover_item *item,
2722 xfs_lsn_t current_lsn)
2724 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2725 xfs_mount_t *mp = log->l_mp;
2732 * In this pass we only want to recover all the buffers which have
2733 * not been cancelled and are not cancellation buffers themselves.
2735 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2736 buf_f->blf_len, buf_f->blf_flags)) {
2737 trace_xfs_log_recover_buf_cancel(log, buf_f);
2741 trace_xfs_log_recover_buf_recover(log, buf_f);
2744 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2745 buf_flags |= XBF_UNMAPPED;
2747 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2751 error = bp->b_error;
2753 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
2758 * Recover the buffer only if we get an LSN from it and it's less than
2759 * the lsn of the transaction we are replaying.
2761 * Note that we have to be extremely careful of readahead here.
2762 * Readahead does not attach verfiers to the buffers so if we don't
2763 * actually do any replay after readahead because of the LSN we found
2764 * in the buffer if more recent than that current transaction then we
2765 * need to attach the verifier directly. Failure to do so can lead to
2766 * future recovery actions (e.g. EFI and unlinked list recovery) can
2767 * operate on the buffers and they won't get the verifier attached. This
2768 * can lead to blocks on disk having the correct content but a stale
2771 * It is safe to assume these clean buffers are currently up to date.
2772 * If the buffer is dirtied by a later transaction being replayed, then
2773 * the verifier will be reset to match whatever recover turns that
2776 lsn = xlog_recover_get_buf_lsn(mp, bp);
2777 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2778 trace_xfs_log_recover_buf_skip(log, buf_f);
2779 xlog_recover_validate_buf_type(mp, bp, buf_f, NULLCOMMITLSN);
2783 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
2784 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2787 } else if (buf_f->blf_flags &
2788 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2791 dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2795 xlog_recover_do_reg_buffer(mp, item, bp, buf_f, current_lsn);
2799 * Perform delayed write on the buffer. Asynchronous writes will be
2800 * slower when taking into account all the buffers to be flushed.
2802 * Also make sure that only inode buffers with good sizes stay in
2803 * the buffer cache. The kernel moves inodes in buffers of 1 block
2804 * or inode_cluster_size bytes, whichever is bigger. The inode
2805 * buffers in the log can be a different size if the log was generated
2806 * by an older kernel using unclustered inode buffers or a newer kernel
2807 * running with a different inode cluster size. Regardless, if the
2808 * the inode buffer size isn't max(blocksize, inode_cluster_size)
2809 * for *our* value of inode_cluster_size, then we need to keep
2810 * the buffer out of the buffer cache so that the buffer won't
2811 * overlap with future reads of those inodes.
2813 if (XFS_DINODE_MAGIC ==
2814 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2815 (BBTOB(bp->b_length) != M_IGEO(log->l_mp)->inode_cluster_size)) {
2817 error = xfs_bwrite(bp);
2819 ASSERT(bp->b_mount == mp);
2820 bp->b_iodone = xlog_recover_iodone;
2821 xfs_buf_delwri_queue(bp, buffer_list);
2830 * Inode fork owner changes
2832 * If we have been told that we have to reparent the inode fork, it's because an
2833 * extent swap operation on a CRC enabled filesystem has been done and we are
2834 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2837 * The complexity here is that we don't have an inode context to work with, so
2838 * after we've replayed the inode we need to instantiate one. This is where the
2841 * We are in the middle of log recovery, so we can't run transactions. That
2842 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2843 * that will result in the corresponding iput() running the inode through
2844 * xfs_inactive(). If we've just replayed an inode core that changes the link
2845 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2846 * transactions (bad!).
2848 * So, to avoid this, we instantiate an inode directly from the inode core we've
2849 * just recovered. We have the buffer still locked, and all we really need to
2850 * instantiate is the inode core and the forks being modified. We can do this
2851 * manually, then run the inode btree owner change, and then tear down the
2852 * xfs_inode without having to run any transactions at all.
2854 * Also, because we don't have a transaction context available here but need to
2855 * gather all the buffers we modify for writeback so we pass the buffer_list
2856 * instead for the operation to use.
2860 xfs_recover_inode_owner_change(
2861 struct xfs_mount *mp,
2862 struct xfs_dinode *dip,
2863 struct xfs_inode_log_format *in_f,
2864 struct list_head *buffer_list)
2866 struct xfs_inode *ip;
2869 ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER));
2871 ip = xfs_inode_alloc(mp, in_f->ilf_ino);
2875 /* instantiate the inode */
2876 xfs_inode_from_disk(ip, dip);
2877 ASSERT(ip->i_d.di_version >= 3);
2879 error = xfs_iformat_fork(ip, dip);
2883 if (!xfs_inode_verify_forks(ip)) {
2884 error = -EFSCORRUPTED;
2888 if (in_f->ilf_fields & XFS_ILOG_DOWNER) {
2889 ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT);
2890 error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK,
2891 ip->i_ino, buffer_list);
2896 if (in_f->ilf_fields & XFS_ILOG_AOWNER) {
2897 ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT);
2898 error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK,
2899 ip->i_ino, buffer_list);
2910 xlog_recover_inode_pass2(
2912 struct list_head *buffer_list,
2913 struct xlog_recover_item *item,
2914 xfs_lsn_t current_lsn)
2916 struct xfs_inode_log_format *in_f;
2917 xfs_mount_t *mp = log->l_mp;
2926 struct xfs_log_dinode *ldip;
2930 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
2931 in_f = item->ri_buf[0].i_addr;
2933 in_f = kmem_alloc(sizeof(struct xfs_inode_log_format), KM_SLEEP);
2935 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2941 * Inode buffers can be freed, look out for it,
2942 * and do not replay the inode.
2944 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2945 in_f->ilf_len, 0)) {
2947 trace_xfs_log_recover_inode_cancel(log, in_f);
2950 trace_xfs_log_recover_inode_recover(log, in_f);
2952 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
2953 &xfs_inode_buf_ops);
2958 error = bp->b_error;
2960 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
2963 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2964 dip = xfs_buf_offset(bp, in_f->ilf_boffset);
2967 * Make sure the place we're flushing out to really looks
2970 if (unlikely(!xfs_verify_magic16(bp, dip->di_magic))) {
2972 "%s: Bad inode magic number, dip = "PTR_FMT", dino bp = "PTR_FMT", ino = %Ld",
2973 __func__, dip, bp, in_f->ilf_ino);
2974 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2975 XFS_ERRLEVEL_LOW, mp);
2976 error = -EFSCORRUPTED;
2979 ldip = item->ri_buf[1].i_addr;
2980 if (unlikely(ldip->di_magic != XFS_DINODE_MAGIC)) {
2982 "%s: Bad inode log record, rec ptr "PTR_FMT", ino %Ld",
2983 __func__, item, in_f->ilf_ino);
2984 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2985 XFS_ERRLEVEL_LOW, mp);
2986 error = -EFSCORRUPTED;
2991 * If the inode has an LSN in it, recover the inode only if it's less
2992 * than the lsn of the transaction we are replaying. Note: we still
2993 * need to replay an owner change even though the inode is more recent
2994 * than the transaction as there is no guarantee that all the btree
2995 * blocks are more recent than this transaction, too.
2997 if (dip->di_version >= 3) {
2998 xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn);
3000 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3001 trace_xfs_log_recover_inode_skip(log, in_f);
3003 goto out_owner_change;
3008 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
3009 * are transactional and if ordering is necessary we can determine that
3010 * more accurately by the LSN field in the V3 inode core. Don't trust
3011 * the inode versions we might be changing them here - use the
3012 * superblock flag to determine whether we need to look at di_flushiter
3013 * to skip replay when the on disk inode is newer than the log one
3015 if (!xfs_sb_version_hascrc(&mp->m_sb) &&
3016 ldip->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
3018 * Deal with the wrap case, DI_MAX_FLUSH is less
3019 * than smaller numbers
3021 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
3022 ldip->di_flushiter < (DI_MAX_FLUSH >> 1)) {
3025 trace_xfs_log_recover_inode_skip(log, in_f);
3031 /* Take the opportunity to reset the flush iteration count */
3032 ldip->di_flushiter = 0;
3034 if (unlikely(S_ISREG(ldip->di_mode))) {
3035 if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) &&
3036 (ldip->di_format != XFS_DINODE_FMT_BTREE)) {
3037 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
3038 XFS_ERRLEVEL_LOW, mp, ldip,
3041 "%s: Bad regular inode log record, rec ptr "PTR_FMT", "
3042 "ino ptr = "PTR_FMT", ino bp = "PTR_FMT", ino %Ld",
3043 __func__, item, dip, bp, in_f->ilf_ino);
3044 error = -EFSCORRUPTED;
3047 } else if (unlikely(S_ISDIR(ldip->di_mode))) {
3048 if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) &&
3049 (ldip->di_format != XFS_DINODE_FMT_BTREE) &&
3050 (ldip->di_format != XFS_DINODE_FMT_LOCAL)) {
3051 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
3052 XFS_ERRLEVEL_LOW, mp, ldip,
3055 "%s: Bad dir inode log record, rec ptr "PTR_FMT", "
3056 "ino ptr = "PTR_FMT", ino bp = "PTR_FMT", ino %Ld",
3057 __func__, item, dip, bp, in_f->ilf_ino);
3058 error = -EFSCORRUPTED;
3062 if (unlikely(ldip->di_nextents + ldip->di_anextents > ldip->di_nblocks)){
3063 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
3064 XFS_ERRLEVEL_LOW, mp, ldip,
3067 "%s: Bad inode log record, rec ptr "PTR_FMT", dino ptr "PTR_FMT", "
3068 "dino bp "PTR_FMT", ino %Ld, total extents = %d, nblocks = %Ld",
3069 __func__, item, dip, bp, in_f->ilf_ino,
3070 ldip->di_nextents + ldip->di_anextents,
3072 error = -EFSCORRUPTED;
3075 if (unlikely(ldip->di_forkoff > mp->m_sb.sb_inodesize)) {
3076 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
3077 XFS_ERRLEVEL_LOW, mp, ldip,
3080 "%s: Bad inode log record, rec ptr "PTR_FMT", dino ptr "PTR_FMT", "
3081 "dino bp "PTR_FMT", ino %Ld, forkoff 0x%x", __func__,
3082 item, dip, bp, in_f->ilf_ino, ldip->di_forkoff);
3083 error = -EFSCORRUPTED;
3086 isize = xfs_log_dinode_size(ldip->di_version);
3087 if (unlikely(item->ri_buf[1].i_len > isize)) {
3088 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
3089 XFS_ERRLEVEL_LOW, mp, ldip,
3092 "%s: Bad inode log record length %d, rec ptr "PTR_FMT,
3093 __func__, item->ri_buf[1].i_len, item);
3094 error = -EFSCORRUPTED;
3098 /* recover the log dinode inode into the on disk inode */
3099 xfs_log_dinode_to_disk(ldip, dip);
3101 fields = in_f->ilf_fields;
3102 if (fields & XFS_ILOG_DEV)
3103 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
3105 if (in_f->ilf_size == 2)
3106 goto out_owner_change;
3107 len = item->ri_buf[2].i_len;
3108 src = item->ri_buf[2].i_addr;
3109 ASSERT(in_f->ilf_size <= 4);
3110 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
3111 ASSERT(!(fields & XFS_ILOG_DFORK) ||
3112 (len == in_f->ilf_dsize));
3114 switch (fields & XFS_ILOG_DFORK) {
3115 case XFS_ILOG_DDATA:
3117 memcpy(XFS_DFORK_DPTR(dip), src, len);
3120 case XFS_ILOG_DBROOT:
3121 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
3122 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
3123 XFS_DFORK_DSIZE(dip, mp));
3128 * There are no data fork flags set.
3130 ASSERT((fields & XFS_ILOG_DFORK) == 0);
3135 * If we logged any attribute data, recover it. There may or
3136 * may not have been any other non-core data logged in this
3139 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
3140 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
3145 len = item->ri_buf[attr_index].i_len;
3146 src = item->ri_buf[attr_index].i_addr;
3147 ASSERT(len == in_f->ilf_asize);
3149 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
3150 case XFS_ILOG_ADATA:
3152 dest = XFS_DFORK_APTR(dip);
3153 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
3154 memcpy(dest, src, len);
3157 case XFS_ILOG_ABROOT:
3158 dest = XFS_DFORK_APTR(dip);
3159 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
3160 len, (xfs_bmdr_block_t*)dest,
3161 XFS_DFORK_ASIZE(dip, mp));
3165 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
3173 /* Recover the swapext owner change unless inode has been deleted */
3174 if ((in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER)) &&
3175 (dip->di_mode != 0))
3176 error = xfs_recover_inode_owner_change(mp, dip, in_f,
3178 /* re-generate the checksum. */
3179 xfs_dinode_calc_crc(log->l_mp, dip);
3181 ASSERT(bp->b_mount == mp);
3182 bp->b_iodone = xlog_recover_iodone;
3183 xfs_buf_delwri_queue(bp, buffer_list);
3194 * Recover QUOTAOFF records. We simply make a note of it in the xlog
3195 * structure, so that we know not to do any dquot item or dquot buffer recovery,
3199 xlog_recover_quotaoff_pass1(
3201 struct xlog_recover_item *item)
3203 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
3207 * The logitem format's flag tells us if this was user quotaoff,
3208 * group/project quotaoff or both.
3210 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
3211 log->l_quotaoffs_flag |= XFS_DQ_USER;
3212 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
3213 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
3214 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
3215 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
3221 * Recover a dquot record
3224 xlog_recover_dquot_pass2(
3226 struct list_head *buffer_list,
3227 struct xlog_recover_item *item,
3228 xfs_lsn_t current_lsn)
3230 xfs_mount_t *mp = log->l_mp;
3232 struct xfs_disk_dquot *ddq, *recddq;
3235 xfs_dq_logformat_t *dq_f;
3240 * Filesystems are required to send in quota flags at mount time.
3242 if (mp->m_qflags == 0)
3245 recddq = item->ri_buf[1].i_addr;
3246 if (recddq == NULL) {
3247 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
3250 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
3251 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
3252 item->ri_buf[1].i_len, __func__);
3257 * This type of quotas was turned off, so ignore this record.
3259 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3261 if (log->l_quotaoffs_flag & type)
3265 * At this point we know that quota was _not_ turned off.
3266 * Since the mount flags are not indicating to us otherwise, this
3267 * must mean that quota is on, and the dquot needs to be replayed.
3268 * Remember that we may not have fully recovered the superblock yet,
3269 * so we can't do the usual trick of looking at the SB quota bits.
3271 * The other possibility, of course, is that the quota subsystem was
3272 * removed since the last mount - ENOSYS.
3274 dq_f = item->ri_buf[0].i_addr;
3276 fa = xfs_dquot_verify(mp, recddq, dq_f->qlf_id, 0);
3278 xfs_alert(mp, "corrupt dquot ID 0x%x in log at %pS",
3282 ASSERT(dq_f->qlf_len == 1);
3285 * At this point we are assuming that the dquots have been allocated
3286 * and hence the buffer has valid dquots stamped in it. It should,
3287 * therefore, pass verifier validation. If the dquot is bad, then the
3288 * we'll return an error here, so we don't need to specifically check
3289 * the dquot in the buffer after the verifier has run.
3291 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
3292 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
3293 &xfs_dquot_buf_ops);
3298 ddq = xfs_buf_offset(bp, dq_f->qlf_boffset);
3301 * If the dquot has an LSN in it, recover the dquot only if it's less
3302 * than the lsn of the transaction we are replaying.
3304 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3305 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq;
3306 xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn);
3308 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3313 memcpy(ddq, recddq, item->ri_buf[1].i_len);
3314 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3315 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
3319 ASSERT(dq_f->qlf_size == 2);
3320 ASSERT(bp->b_mount == mp);
3321 bp->b_iodone = xlog_recover_iodone;
3322 xfs_buf_delwri_queue(bp, buffer_list);
3330 * This routine is called to create an in-core extent free intent
3331 * item from the efi format structure which was logged on disk.
3332 * It allocates an in-core efi, copies the extents from the format
3333 * structure into it, and adds the efi to the AIL with the given
3337 xlog_recover_efi_pass2(
3339 struct xlog_recover_item *item,
3343 struct xfs_mount *mp = log->l_mp;
3344 struct xfs_efi_log_item *efip;
3345 struct xfs_efi_log_format *efi_formatp;
3347 efi_formatp = item->ri_buf[0].i_addr;
3349 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
3350 error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format);
3352 xfs_efi_item_free(efip);
3355 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
3357 spin_lock(&log->l_ailp->ail_lock);
3359 * The EFI has two references. One for the EFD and one for EFI to ensure
3360 * it makes it into the AIL. Insert the EFI into the AIL directly and
3361 * drop the EFI reference. Note that xfs_trans_ail_update() drops the
3364 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
3365 xfs_efi_release(efip);
3371 * This routine is called when an EFD format structure is found in a committed
3372 * transaction in the log. Its purpose is to cancel the corresponding EFI if it
3373 * was still in the log. To do this it searches the AIL for the EFI with an id
3374 * equal to that in the EFD format structure. If we find it we drop the EFD
3375 * reference, which removes the EFI from the AIL and frees it.
3378 xlog_recover_efd_pass2(
3380 struct xlog_recover_item *item)
3382 xfs_efd_log_format_t *efd_formatp;
3383 xfs_efi_log_item_t *efip = NULL;
3384 struct xfs_log_item *lip;
3386 struct xfs_ail_cursor cur;
3387 struct xfs_ail *ailp = log->l_ailp;
3389 efd_formatp = item->ri_buf[0].i_addr;
3390 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
3391 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
3392 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
3393 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
3394 efi_id = efd_formatp->efd_efi_id;
3397 * Search for the EFI with the id in the EFD format structure in the
3400 spin_lock(&ailp->ail_lock);
3401 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3402 while (lip != NULL) {
3403 if (lip->li_type == XFS_LI_EFI) {
3404 efip = (xfs_efi_log_item_t *)lip;
3405 if (efip->efi_format.efi_id == efi_id) {
3407 * Drop the EFD reference to the EFI. This
3408 * removes the EFI from the AIL and frees it.
3410 spin_unlock(&ailp->ail_lock);
3411 xfs_efi_release(efip);
3412 spin_lock(&ailp->ail_lock);
3416 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3419 xfs_trans_ail_cursor_done(&cur);
3420 spin_unlock(&ailp->ail_lock);
3426 * This routine is called to create an in-core extent rmap update
3427 * item from the rui format structure which was logged on disk.
3428 * It allocates an in-core rui, copies the extents from the format
3429 * structure into it, and adds the rui to the AIL with the given
3433 xlog_recover_rui_pass2(
3435 struct xlog_recover_item *item,
3439 struct xfs_mount *mp = log->l_mp;
3440 struct xfs_rui_log_item *ruip;
3441 struct xfs_rui_log_format *rui_formatp;
3443 rui_formatp = item->ri_buf[0].i_addr;
3445 ruip = xfs_rui_init(mp, rui_formatp->rui_nextents);
3446 error = xfs_rui_copy_format(&item->ri_buf[0], &ruip->rui_format);
3448 xfs_rui_item_free(ruip);
3451 atomic_set(&ruip->rui_next_extent, rui_formatp->rui_nextents);
3453 spin_lock(&log->l_ailp->ail_lock);
3455 * The RUI has two references. One for the RUD and one for RUI to ensure
3456 * it makes it into the AIL. Insert the RUI into the AIL directly and
3457 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
3460 xfs_trans_ail_update(log->l_ailp, &ruip->rui_item, lsn);
3461 xfs_rui_release(ruip);
3467 * This routine is called when an RUD format structure is found in a committed
3468 * transaction in the log. Its purpose is to cancel the corresponding RUI if it
3469 * was still in the log. To do this it searches the AIL for the RUI with an id
3470 * equal to that in the RUD format structure. If we find it we drop the RUD
3471 * reference, which removes the RUI from the AIL and frees it.
3474 xlog_recover_rud_pass2(
3476 struct xlog_recover_item *item)
3478 struct xfs_rud_log_format *rud_formatp;
3479 struct xfs_rui_log_item *ruip = NULL;
3480 struct xfs_log_item *lip;
3482 struct xfs_ail_cursor cur;
3483 struct xfs_ail *ailp = log->l_ailp;
3485 rud_formatp = item->ri_buf[0].i_addr;
3486 ASSERT(item->ri_buf[0].i_len == sizeof(struct xfs_rud_log_format));
3487 rui_id = rud_formatp->rud_rui_id;
3490 * Search for the RUI with the id in the RUD format structure in the
3493 spin_lock(&ailp->ail_lock);
3494 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3495 while (lip != NULL) {
3496 if (lip->li_type == XFS_LI_RUI) {
3497 ruip = (struct xfs_rui_log_item *)lip;
3498 if (ruip->rui_format.rui_id == rui_id) {
3500 * Drop the RUD reference to the RUI. This
3501 * removes the RUI from the AIL and frees it.
3503 spin_unlock(&ailp->ail_lock);
3504 xfs_rui_release(ruip);
3505 spin_lock(&ailp->ail_lock);
3509 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3512 xfs_trans_ail_cursor_done(&cur);
3513 spin_unlock(&ailp->ail_lock);
3519 * Copy an CUI format buffer from the given buf, and into the destination
3520 * CUI format structure. The CUI/CUD items were designed not to need any
3521 * special alignment handling.
3524 xfs_cui_copy_format(
3525 struct xfs_log_iovec *buf,
3526 struct xfs_cui_log_format *dst_cui_fmt)
3528 struct xfs_cui_log_format *src_cui_fmt;
3531 src_cui_fmt = buf->i_addr;
3532 len = xfs_cui_log_format_sizeof(src_cui_fmt->cui_nextents);
3534 if (buf->i_len == len) {
3535 memcpy(dst_cui_fmt, src_cui_fmt, len);
3538 return -EFSCORRUPTED;
3542 * This routine is called to create an in-core extent refcount update
3543 * item from the cui format structure which was logged on disk.
3544 * It allocates an in-core cui, copies the extents from the format
3545 * structure into it, and adds the cui to the AIL with the given
3549 xlog_recover_cui_pass2(
3551 struct xlog_recover_item *item,
3555 struct xfs_mount *mp = log->l_mp;
3556 struct xfs_cui_log_item *cuip;
3557 struct xfs_cui_log_format *cui_formatp;
3559 cui_formatp = item->ri_buf[0].i_addr;
3561 cuip = xfs_cui_init(mp, cui_formatp->cui_nextents);
3562 error = xfs_cui_copy_format(&item->ri_buf[0], &cuip->cui_format);
3564 xfs_cui_item_free(cuip);
3567 atomic_set(&cuip->cui_next_extent, cui_formatp->cui_nextents);
3569 spin_lock(&log->l_ailp->ail_lock);
3571 * The CUI has two references. One for the CUD and one for CUI to ensure
3572 * it makes it into the AIL. Insert the CUI into the AIL directly and
3573 * drop the CUI reference. Note that xfs_trans_ail_update() drops the
3576 xfs_trans_ail_update(log->l_ailp, &cuip->cui_item, lsn);
3577 xfs_cui_release(cuip);
3583 * This routine is called when an CUD format structure is found in a committed
3584 * transaction in the log. Its purpose is to cancel the corresponding CUI if it
3585 * was still in the log. To do this it searches the AIL for the CUI with an id
3586 * equal to that in the CUD format structure. If we find it we drop the CUD
3587 * reference, which removes the CUI from the AIL and frees it.
3590 xlog_recover_cud_pass2(
3592 struct xlog_recover_item *item)
3594 struct xfs_cud_log_format *cud_formatp;
3595 struct xfs_cui_log_item *cuip = NULL;
3596 struct xfs_log_item *lip;
3598 struct xfs_ail_cursor cur;
3599 struct xfs_ail *ailp = log->l_ailp;
3601 cud_formatp = item->ri_buf[0].i_addr;
3602 if (item->ri_buf[0].i_len != sizeof(struct xfs_cud_log_format))
3603 return -EFSCORRUPTED;
3604 cui_id = cud_formatp->cud_cui_id;
3607 * Search for the CUI with the id in the CUD format structure in the
3610 spin_lock(&ailp->ail_lock);
3611 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3612 while (lip != NULL) {
3613 if (lip->li_type == XFS_LI_CUI) {
3614 cuip = (struct xfs_cui_log_item *)lip;
3615 if (cuip->cui_format.cui_id == cui_id) {
3617 * Drop the CUD reference to the CUI. This
3618 * removes the CUI from the AIL and frees it.
3620 spin_unlock(&ailp->ail_lock);
3621 xfs_cui_release(cuip);
3622 spin_lock(&ailp->ail_lock);
3626 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3629 xfs_trans_ail_cursor_done(&cur);
3630 spin_unlock(&ailp->ail_lock);
3636 * Copy an BUI format buffer from the given buf, and into the destination
3637 * BUI format structure. The BUI/BUD items were designed not to need any
3638 * special alignment handling.
3641 xfs_bui_copy_format(
3642 struct xfs_log_iovec *buf,
3643 struct xfs_bui_log_format *dst_bui_fmt)
3645 struct xfs_bui_log_format *src_bui_fmt;
3648 src_bui_fmt = buf->i_addr;
3649 len = xfs_bui_log_format_sizeof(src_bui_fmt->bui_nextents);
3651 if (buf->i_len == len) {
3652 memcpy(dst_bui_fmt, src_bui_fmt, len);
3655 return -EFSCORRUPTED;
3659 * This routine is called to create an in-core extent bmap update
3660 * item from the bui format structure which was logged on disk.
3661 * It allocates an in-core bui, copies the extents from the format
3662 * structure into it, and adds the bui to the AIL with the given
3666 xlog_recover_bui_pass2(
3668 struct xlog_recover_item *item,
3672 struct xfs_mount *mp = log->l_mp;
3673 struct xfs_bui_log_item *buip;
3674 struct xfs_bui_log_format *bui_formatp;
3676 bui_formatp = item->ri_buf[0].i_addr;
3678 if (bui_formatp->bui_nextents != XFS_BUI_MAX_FAST_EXTENTS)
3679 return -EFSCORRUPTED;
3680 buip = xfs_bui_init(mp);
3681 error = xfs_bui_copy_format(&item->ri_buf[0], &buip->bui_format);
3683 xfs_bui_item_free(buip);
3686 atomic_set(&buip->bui_next_extent, bui_formatp->bui_nextents);
3688 spin_lock(&log->l_ailp->ail_lock);
3690 * The RUI has two references. One for the RUD and one for RUI to ensure
3691 * it makes it into the AIL. Insert the RUI into the AIL directly and
3692 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
3695 xfs_trans_ail_update(log->l_ailp, &buip->bui_item, lsn);
3696 xfs_bui_release(buip);
3702 * This routine is called when an BUD format structure is found in a committed
3703 * transaction in the log. Its purpose is to cancel the corresponding BUI if it
3704 * was still in the log. To do this it searches the AIL for the BUI with an id
3705 * equal to that in the BUD format structure. If we find it we drop the BUD
3706 * reference, which removes the BUI from the AIL and frees it.
3709 xlog_recover_bud_pass2(
3711 struct xlog_recover_item *item)
3713 struct xfs_bud_log_format *bud_formatp;
3714 struct xfs_bui_log_item *buip = NULL;
3715 struct xfs_log_item *lip;
3717 struct xfs_ail_cursor cur;
3718 struct xfs_ail *ailp = log->l_ailp;
3720 bud_formatp = item->ri_buf[0].i_addr;
3721 if (item->ri_buf[0].i_len != sizeof(struct xfs_bud_log_format))
3722 return -EFSCORRUPTED;
3723 bui_id = bud_formatp->bud_bui_id;
3726 * Search for the BUI with the id in the BUD format structure in the
3729 spin_lock(&ailp->ail_lock);
3730 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3731 while (lip != NULL) {
3732 if (lip->li_type == XFS_LI_BUI) {
3733 buip = (struct xfs_bui_log_item *)lip;
3734 if (buip->bui_format.bui_id == bui_id) {
3736 * Drop the BUD reference to the BUI. This
3737 * removes the BUI from the AIL and frees it.
3739 spin_unlock(&ailp->ail_lock);
3740 xfs_bui_release(buip);
3741 spin_lock(&ailp->ail_lock);
3745 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3748 xfs_trans_ail_cursor_done(&cur);
3749 spin_unlock(&ailp->ail_lock);
3755 * This routine is called when an inode create format structure is found in a
3756 * committed transaction in the log. It's purpose is to initialise the inodes
3757 * being allocated on disk. This requires us to get inode cluster buffers that
3758 * match the range to be initialised, stamped with inode templates and written
3759 * by delayed write so that subsequent modifications will hit the cached buffer
3760 * and only need writing out at the end of recovery.
3763 xlog_recover_do_icreate_pass2(
3765 struct list_head *buffer_list,
3766 xlog_recover_item_t *item)
3768 struct xfs_mount *mp = log->l_mp;
3769 struct xfs_icreate_log *icl;
3770 struct xfs_ino_geometry *igeo = M_IGEO(mp);
3771 xfs_agnumber_t agno;
3772 xfs_agblock_t agbno;
3775 xfs_agblock_t length;
3781 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3782 if (icl->icl_type != XFS_LI_ICREATE) {
3783 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
3787 if (icl->icl_size != 1) {
3788 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
3792 agno = be32_to_cpu(icl->icl_ag);
3793 if (agno >= mp->m_sb.sb_agcount) {
3794 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
3797 agbno = be32_to_cpu(icl->icl_agbno);
3798 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3799 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
3802 isize = be32_to_cpu(icl->icl_isize);
3803 if (isize != mp->m_sb.sb_inodesize) {
3804 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
3807 count = be32_to_cpu(icl->icl_count);
3809 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
3812 length = be32_to_cpu(icl->icl_length);
3813 if (!length || length >= mp->m_sb.sb_agblocks) {
3814 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
3819 * The inode chunk is either full or sparse and we only support
3820 * m_ino_geo.ialloc_min_blks sized sparse allocations at this time.
3822 if (length != igeo->ialloc_blks &&
3823 length != igeo->ialloc_min_blks) {
3825 "%s: unsupported chunk length", __FUNCTION__);
3829 /* verify inode count is consistent with extent length */
3830 if ((count >> mp->m_sb.sb_inopblog) != length) {
3832 "%s: inconsistent inode count and chunk length",
3838 * The icreate transaction can cover multiple cluster buffers and these
3839 * buffers could have been freed and reused. Check the individual
3840 * buffers for cancellation so we don't overwrite anything written after
3843 bb_per_cluster = XFS_FSB_TO_BB(mp, igeo->blocks_per_cluster);
3844 nbufs = length / igeo->blocks_per_cluster;
3845 for (i = 0, cancel_count = 0; i < nbufs; i++) {
3848 daddr = XFS_AGB_TO_DADDR(mp, agno,
3849 agbno + i * igeo->blocks_per_cluster);
3850 if (xlog_check_buffer_cancelled(log, daddr, bb_per_cluster, 0))
3855 * We currently only use icreate for a single allocation at a time. This
3856 * means we should expect either all or none of the buffers to be
3857 * cancelled. Be conservative and skip replay if at least one buffer is
3858 * cancelled, but warn the user that something is awry if the buffers
3859 * are not consistent.
3861 * XXX: This must be refined to only skip cancelled clusters once we use
3862 * icreate for multiple chunk allocations.
3864 ASSERT(!cancel_count || cancel_count == nbufs);
3866 if (cancel_count != nbufs)
3868 "WARNING: partial inode chunk cancellation, skipped icreate.");
3869 trace_xfs_log_recover_icreate_cancel(log, icl);
3873 trace_xfs_log_recover_icreate_recover(log, icl);
3874 return xfs_ialloc_inode_init(mp, NULL, buffer_list, count, agno, agbno,
3875 length, be32_to_cpu(icl->icl_gen));
3879 xlog_recover_buffer_ra_pass2(
3881 struct xlog_recover_item *item)
3883 struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr;
3884 struct xfs_mount *mp = log->l_mp;
3886 if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
3887 buf_f->blf_len, buf_f->blf_flags)) {
3891 xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
3892 buf_f->blf_len, NULL);
3896 xlog_recover_inode_ra_pass2(
3898 struct xlog_recover_item *item)
3900 struct xfs_inode_log_format ilf_buf;
3901 struct xfs_inode_log_format *ilfp;
3902 struct xfs_mount *mp = log->l_mp;
3905 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3906 ilfp = item->ri_buf[0].i_addr;
3909 memset(ilfp, 0, sizeof(*ilfp));
3910 error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
3915 if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
3918 xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
3919 ilfp->ilf_len, &xfs_inode_buf_ra_ops);
3923 xlog_recover_dquot_ra_pass2(
3925 struct xlog_recover_item *item)
3927 struct xfs_mount *mp = log->l_mp;
3928 struct xfs_disk_dquot *recddq;
3929 struct xfs_dq_logformat *dq_f;
3934 if (mp->m_qflags == 0)
3937 recddq = item->ri_buf[1].i_addr;
3940 if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
3943 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3945 if (log->l_quotaoffs_flag & type)
3948 dq_f = item->ri_buf[0].i_addr;
3950 ASSERT(dq_f->qlf_len == 1);
3952 len = XFS_FSB_TO_BB(mp, dq_f->qlf_len);
3953 if (xlog_peek_buffer_cancelled(log, dq_f->qlf_blkno, len, 0))
3956 xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno, len,
3957 &xfs_dquot_buf_ra_ops);
3961 xlog_recover_ra_pass2(
3963 struct xlog_recover_item *item)
3965 switch (ITEM_TYPE(item)) {
3967 xlog_recover_buffer_ra_pass2(log, item);
3970 xlog_recover_inode_ra_pass2(log, item);
3973 xlog_recover_dquot_ra_pass2(log, item);
3977 case XFS_LI_QUOTAOFF:
3990 xlog_recover_commit_pass1(
3992 struct xlog_recover *trans,
3993 struct xlog_recover_item *item)
3995 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
3997 switch (ITEM_TYPE(item)) {
3999 return xlog_recover_buffer_pass1(log, item);
4000 case XFS_LI_QUOTAOFF:
4001 return xlog_recover_quotaoff_pass1(log, item);
4006 case XFS_LI_ICREATE:
4013 /* nothing to do in pass 1 */
4016 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
4017 __func__, ITEM_TYPE(item));
4024 xlog_recover_commit_pass2(
4026 struct xlog_recover *trans,
4027 struct list_head *buffer_list,
4028 struct xlog_recover_item *item)
4030 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
4032 switch (ITEM_TYPE(item)) {
4034 return xlog_recover_buffer_pass2(log, buffer_list, item,
4037 return xlog_recover_inode_pass2(log, buffer_list, item,
4040 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
4042 return xlog_recover_efd_pass2(log, item);
4044 return xlog_recover_rui_pass2(log, item, trans->r_lsn);
4046 return xlog_recover_rud_pass2(log, item);
4048 return xlog_recover_cui_pass2(log, item, trans->r_lsn);
4050 return xlog_recover_cud_pass2(log, item);
4052 return xlog_recover_bui_pass2(log, item, trans->r_lsn);
4054 return xlog_recover_bud_pass2(log, item);
4056 return xlog_recover_dquot_pass2(log, buffer_list, item,
4058 case XFS_LI_ICREATE:
4059 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
4060 case XFS_LI_QUOTAOFF:
4061 /* nothing to do in pass2 */
4064 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
4065 __func__, ITEM_TYPE(item));
4072 xlog_recover_items_pass2(
4074 struct xlog_recover *trans,
4075 struct list_head *buffer_list,
4076 struct list_head *item_list)
4078 struct xlog_recover_item *item;
4081 list_for_each_entry(item, item_list, ri_list) {
4082 error = xlog_recover_commit_pass2(log, trans,
4092 * Perform the transaction.
4094 * If the transaction modifies a buffer or inode, do it now. Otherwise,
4095 * EFIs and EFDs get queued up by adding entries into the AIL for them.
4098 xlog_recover_commit_trans(
4100 struct xlog_recover *trans,
4102 struct list_head *buffer_list)
4105 int items_queued = 0;
4106 struct xlog_recover_item *item;
4107 struct xlog_recover_item *next;
4108 LIST_HEAD (ra_list);
4109 LIST_HEAD (done_list);
4111 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
4113 hlist_del_init(&trans->r_list);
4115 error = xlog_recover_reorder_trans(log, trans, pass);
4119 list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
4121 case XLOG_RECOVER_PASS1:
4122 error = xlog_recover_commit_pass1(log, trans, item);
4124 case XLOG_RECOVER_PASS2:
4125 xlog_recover_ra_pass2(log, item);
4126 list_move_tail(&item->ri_list, &ra_list);
4128 if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
4129 error = xlog_recover_items_pass2(log, trans,
4130 buffer_list, &ra_list);
4131 list_splice_tail_init(&ra_list, &done_list);
4145 if (!list_empty(&ra_list)) {
4147 error = xlog_recover_items_pass2(log, trans,
4148 buffer_list, &ra_list);
4149 list_splice_tail_init(&ra_list, &done_list);
4152 if (!list_empty(&done_list))
4153 list_splice_init(&done_list, &trans->r_itemq);
4159 xlog_recover_add_item(
4160 struct list_head *head)
4162 xlog_recover_item_t *item;
4164 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
4165 INIT_LIST_HEAD(&item->ri_list);
4166 list_add_tail(&item->ri_list, head);
4170 xlog_recover_add_to_cont_trans(
4172 struct xlog_recover *trans,
4176 xlog_recover_item_t *item;
4177 char *ptr, *old_ptr;
4181 * If the transaction is empty, the header was split across this and the
4182 * previous record. Copy the rest of the header.
4184 if (list_empty(&trans->r_itemq)) {
4185 ASSERT(len <= sizeof(struct xfs_trans_header));
4186 if (len > sizeof(struct xfs_trans_header)) {
4187 xfs_warn(log->l_mp, "%s: bad header length", __func__);
4191 xlog_recover_add_item(&trans->r_itemq);
4192 ptr = (char *)&trans->r_theader +
4193 sizeof(struct xfs_trans_header) - len;
4194 memcpy(ptr, dp, len);
4198 /* take the tail entry */
4199 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
4201 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
4202 old_len = item->ri_buf[item->ri_cnt-1].i_len;
4204 ptr = kmem_realloc(old_ptr, len + old_len, KM_SLEEP);
4205 memcpy(&ptr[old_len], dp, len);
4206 item->ri_buf[item->ri_cnt-1].i_len += len;
4207 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
4208 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
4213 * The next region to add is the start of a new region. It could be
4214 * a whole region or it could be the first part of a new region. Because
4215 * of this, the assumption here is that the type and size fields of all
4216 * format structures fit into the first 32 bits of the structure.
4218 * This works because all regions must be 32 bit aligned. Therefore, we
4219 * either have both fields or we have neither field. In the case we have
4220 * neither field, the data part of the region is zero length. We only have
4221 * a log_op_header and can throw away the header since a new one will appear
4222 * later. If we have at least 4 bytes, then we can determine how many regions
4223 * will appear in the current log item.
4226 xlog_recover_add_to_trans(
4228 struct xlog_recover *trans,
4232 struct xfs_inode_log_format *in_f; /* any will do */
4233 xlog_recover_item_t *item;
4238 if (list_empty(&trans->r_itemq)) {
4239 /* we need to catch log corruptions here */
4240 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
4241 xfs_warn(log->l_mp, "%s: bad header magic number",
4247 if (len > sizeof(struct xfs_trans_header)) {
4248 xfs_warn(log->l_mp, "%s: bad header length", __func__);
4254 * The transaction header can be arbitrarily split across op
4255 * records. If we don't have the whole thing here, copy what we
4256 * do have and handle the rest in the next record.
4258 if (len == sizeof(struct xfs_trans_header))
4259 xlog_recover_add_item(&trans->r_itemq);
4260 memcpy(&trans->r_theader, dp, len);
4264 ptr = kmem_alloc(len, KM_SLEEP);
4265 memcpy(ptr, dp, len);
4266 in_f = (struct xfs_inode_log_format *)ptr;
4268 /* take the tail entry */
4269 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
4270 if (item->ri_total != 0 &&
4271 item->ri_total == item->ri_cnt) {
4272 /* tail item is in use, get a new one */
4273 xlog_recover_add_item(&trans->r_itemq);
4274 item = list_entry(trans->r_itemq.prev,
4275 xlog_recover_item_t, ri_list);
4278 if (item->ri_total == 0) { /* first region to be added */
4279 if (in_f->ilf_size == 0 ||
4280 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
4282 "bad number of regions (%d) in inode log format",
4289 item->ri_total = in_f->ilf_size;
4291 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
4294 ASSERT(item->ri_total > item->ri_cnt);
4295 /* Description region is ri_buf[0] */
4296 item->ri_buf[item->ri_cnt].i_addr = ptr;
4297 item->ri_buf[item->ri_cnt].i_len = len;
4299 trace_xfs_log_recover_item_add(log, trans, item, 0);
4304 * Free up any resources allocated by the transaction
4306 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
4309 xlog_recover_free_trans(
4310 struct xlog_recover *trans)
4312 xlog_recover_item_t *item, *n;
4315 hlist_del_init(&trans->r_list);
4317 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
4318 /* Free the regions in the item. */
4319 list_del(&item->ri_list);
4320 for (i = 0; i < item->ri_cnt; i++)
4321 kmem_free(item->ri_buf[i].i_addr);
4322 /* Free the item itself */
4323 kmem_free(item->ri_buf);
4326 /* Free the transaction recover structure */
4331 * On error or completion, trans is freed.
4334 xlog_recovery_process_trans(
4336 struct xlog_recover *trans,
4341 struct list_head *buffer_list)
4344 bool freeit = false;
4346 /* mask off ophdr transaction container flags */
4347 flags &= ~XLOG_END_TRANS;
4348 if (flags & XLOG_WAS_CONT_TRANS)
4349 flags &= ~XLOG_CONTINUE_TRANS;
4352 * Callees must not free the trans structure. We'll decide if we need to
4353 * free it or not based on the operation being done and it's result.
4356 /* expected flag values */
4358 case XLOG_CONTINUE_TRANS:
4359 error = xlog_recover_add_to_trans(log, trans, dp, len);
4361 case XLOG_WAS_CONT_TRANS:
4362 error = xlog_recover_add_to_cont_trans(log, trans, dp, len);
4364 case XLOG_COMMIT_TRANS:
4365 error = xlog_recover_commit_trans(log, trans, pass,
4367 /* success or fail, we are now done with this transaction. */
4371 /* unexpected flag values */
4372 case XLOG_UNMOUNT_TRANS:
4373 /* just skip trans */
4374 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
4377 case XLOG_START_TRANS:
4379 xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags);
4384 if (error || freeit)
4385 xlog_recover_free_trans(trans);
4390 * Lookup the transaction recovery structure associated with the ID in the
4391 * current ophdr. If the transaction doesn't exist and the start flag is set in
4392 * the ophdr, then allocate a new transaction for future ID matches to find.
4393 * Either way, return what we found during the lookup - an existing transaction
4396 STATIC struct xlog_recover *
4397 xlog_recover_ophdr_to_trans(
4398 struct hlist_head rhash[],
4399 struct xlog_rec_header *rhead,
4400 struct xlog_op_header *ohead)
4402 struct xlog_recover *trans;
4404 struct hlist_head *rhp;
4406 tid = be32_to_cpu(ohead->oh_tid);
4407 rhp = &rhash[XLOG_RHASH(tid)];
4408 hlist_for_each_entry(trans, rhp, r_list) {
4409 if (trans->r_log_tid == tid)
4414 * skip over non-start transaction headers - we could be
4415 * processing slack space before the next transaction starts
4417 if (!(ohead->oh_flags & XLOG_START_TRANS))
4420 ASSERT(be32_to_cpu(ohead->oh_len) == 0);
4423 * This is a new transaction so allocate a new recovery container to
4424 * hold the recovery ops that will follow.
4426 trans = kmem_zalloc(sizeof(struct xlog_recover), KM_SLEEP);
4427 trans->r_log_tid = tid;
4428 trans->r_lsn = be64_to_cpu(rhead->h_lsn);
4429 INIT_LIST_HEAD(&trans->r_itemq);
4430 INIT_HLIST_NODE(&trans->r_list);
4431 hlist_add_head(&trans->r_list, rhp);
4434 * Nothing more to do for this ophdr. Items to be added to this new
4435 * transaction will be in subsequent ophdr containers.
4441 xlog_recover_process_ophdr(
4443 struct hlist_head rhash[],
4444 struct xlog_rec_header *rhead,
4445 struct xlog_op_header *ohead,
4449 struct list_head *buffer_list)
4451 struct xlog_recover *trans;
4455 /* Do we understand who wrote this op? */
4456 if (ohead->oh_clientid != XFS_TRANSACTION &&
4457 ohead->oh_clientid != XFS_LOG) {
4458 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
4459 __func__, ohead->oh_clientid);
4465 * Check the ophdr contains all the data it is supposed to contain.
4467 len = be32_to_cpu(ohead->oh_len);
4468 if (dp + len > end) {
4469 xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len);
4474 trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead);
4476 /* nothing to do, so skip over this ophdr */
4481 * The recovered buffer queue is drained only once we know that all
4482 * recovery items for the current LSN have been processed. This is
4485 * - Buffer write submission updates the metadata LSN of the buffer.
4486 * - Log recovery skips items with a metadata LSN >= the current LSN of
4487 * the recovery item.
4488 * - Separate recovery items against the same metadata buffer can share
4489 * a current LSN. I.e., consider that the LSN of a recovery item is
4490 * defined as the starting LSN of the first record in which its
4491 * transaction appears, that a record can hold multiple transactions,
4492 * and/or that a transaction can span multiple records.
4494 * In other words, we are allowed to submit a buffer from log recovery
4495 * once per current LSN. Otherwise, we may incorrectly skip recovery
4496 * items and cause corruption.
4498 * We don't know up front whether buffers are updated multiple times per
4499 * LSN. Therefore, track the current LSN of each commit log record as it
4500 * is processed and drain the queue when it changes. Use commit records
4501 * because they are ordered correctly by the logging code.
4503 if (log->l_recovery_lsn != trans->r_lsn &&
4504 ohead->oh_flags & XLOG_COMMIT_TRANS) {
4505 error = xfs_buf_delwri_submit(buffer_list);
4508 log->l_recovery_lsn = trans->r_lsn;
4511 return xlog_recovery_process_trans(log, trans, dp, len,
4512 ohead->oh_flags, pass, buffer_list);
4516 * There are two valid states of the r_state field. 0 indicates that the
4517 * transaction structure is in a normal state. We have either seen the
4518 * start of the transaction or the last operation we added was not a partial
4519 * operation. If the last operation we added to the transaction was a
4520 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
4522 * NOTE: skip LRs with 0 data length.
4525 xlog_recover_process_data(
4527 struct hlist_head rhash[],
4528 struct xlog_rec_header *rhead,
4531 struct list_head *buffer_list)
4533 struct xlog_op_header *ohead;
4538 end = dp + be32_to_cpu(rhead->h_len);
4539 num_logops = be32_to_cpu(rhead->h_num_logops);
4541 /* check the log format matches our own - else we can't recover */
4542 if (xlog_header_check_recover(log->l_mp, rhead))
4545 trace_xfs_log_recover_record(log, rhead, pass);
4546 while ((dp < end) && num_logops) {
4548 ohead = (struct xlog_op_header *)dp;
4549 dp += sizeof(*ohead);
4552 /* errors will abort recovery */
4553 error = xlog_recover_process_ophdr(log, rhash, rhead, ohead,
4554 dp, end, pass, buffer_list);
4558 dp += be32_to_cpu(ohead->oh_len);
4564 /* Recover the EFI if necessary. */
4566 xlog_recover_process_efi(
4567 struct xfs_mount *mp,
4568 struct xfs_ail *ailp,
4569 struct xfs_log_item *lip)
4571 struct xfs_efi_log_item *efip;
4575 * Skip EFIs that we've already processed.
4577 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
4578 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags))
4581 spin_unlock(&ailp->ail_lock);
4582 error = xfs_efi_recover(mp, efip);
4583 spin_lock(&ailp->ail_lock);
4588 /* Release the EFI since we're cancelling everything. */
4590 xlog_recover_cancel_efi(
4591 struct xfs_mount *mp,
4592 struct xfs_ail *ailp,
4593 struct xfs_log_item *lip)
4595 struct xfs_efi_log_item *efip;
4597 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
4599 spin_unlock(&ailp->ail_lock);
4600 xfs_efi_release(efip);
4601 spin_lock(&ailp->ail_lock);
4604 /* Recover the RUI if necessary. */
4606 xlog_recover_process_rui(
4607 struct xfs_mount *mp,
4608 struct xfs_ail *ailp,
4609 struct xfs_log_item *lip)
4611 struct xfs_rui_log_item *ruip;
4615 * Skip RUIs that we've already processed.
4617 ruip = container_of(lip, struct xfs_rui_log_item, rui_item);
4618 if (test_bit(XFS_RUI_RECOVERED, &ruip->rui_flags))
4621 spin_unlock(&ailp->ail_lock);
4622 error = xfs_rui_recover(mp, ruip);
4623 spin_lock(&ailp->ail_lock);
4628 /* Release the RUI since we're cancelling everything. */
4630 xlog_recover_cancel_rui(
4631 struct xfs_mount *mp,
4632 struct xfs_ail *ailp,
4633 struct xfs_log_item *lip)
4635 struct xfs_rui_log_item *ruip;
4637 ruip = container_of(lip, struct xfs_rui_log_item, rui_item);
4639 spin_unlock(&ailp->ail_lock);
4640 xfs_rui_release(ruip);
4641 spin_lock(&ailp->ail_lock);
4644 /* Recover the CUI if necessary. */
4646 xlog_recover_process_cui(
4647 struct xfs_trans *parent_tp,
4648 struct xfs_ail *ailp,
4649 struct xfs_log_item *lip)
4651 struct xfs_cui_log_item *cuip;
4655 * Skip CUIs that we've already processed.
4657 cuip = container_of(lip, struct xfs_cui_log_item, cui_item);
4658 if (test_bit(XFS_CUI_RECOVERED, &cuip->cui_flags))
4661 spin_unlock(&ailp->ail_lock);
4662 error = xfs_cui_recover(parent_tp, cuip);
4663 spin_lock(&ailp->ail_lock);
4668 /* Release the CUI since we're cancelling everything. */
4670 xlog_recover_cancel_cui(
4671 struct xfs_mount *mp,
4672 struct xfs_ail *ailp,
4673 struct xfs_log_item *lip)
4675 struct xfs_cui_log_item *cuip;
4677 cuip = container_of(lip, struct xfs_cui_log_item, cui_item);
4679 spin_unlock(&ailp->ail_lock);
4680 xfs_cui_release(cuip);
4681 spin_lock(&ailp->ail_lock);
4684 /* Recover the BUI if necessary. */
4686 xlog_recover_process_bui(
4687 struct xfs_trans *parent_tp,
4688 struct xfs_ail *ailp,
4689 struct xfs_log_item *lip)
4691 struct xfs_bui_log_item *buip;
4695 * Skip BUIs that we've already processed.
4697 buip = container_of(lip, struct xfs_bui_log_item, bui_item);
4698 if (test_bit(XFS_BUI_RECOVERED, &buip->bui_flags))
4701 spin_unlock(&ailp->ail_lock);
4702 error = xfs_bui_recover(parent_tp, buip);
4703 spin_lock(&ailp->ail_lock);
4708 /* Release the BUI since we're cancelling everything. */
4710 xlog_recover_cancel_bui(
4711 struct xfs_mount *mp,
4712 struct xfs_ail *ailp,
4713 struct xfs_log_item *lip)
4715 struct xfs_bui_log_item *buip;
4717 buip = container_of(lip, struct xfs_bui_log_item, bui_item);
4719 spin_unlock(&ailp->ail_lock);
4720 xfs_bui_release(buip);
4721 spin_lock(&ailp->ail_lock);
4724 /* Is this log item a deferred action intent? */
4725 static inline bool xlog_item_is_intent(struct xfs_log_item *lip)
4727 switch (lip->li_type) {
4738 /* Take all the collected deferred ops and finish them in order. */
4740 xlog_finish_defer_ops(
4741 struct xfs_trans *parent_tp)
4743 struct xfs_mount *mp = parent_tp->t_mountp;
4744 struct xfs_trans *tp;
4750 * We're finishing the defer_ops that accumulated as a result of
4751 * recovering unfinished intent items during log recovery. We
4752 * reserve an itruncate transaction because it is the largest
4753 * permanent transaction type. Since we're the only user of the fs
4754 * right now, take 93% (15/16) of the available free blocks. Use
4755 * weird math to avoid a 64-bit division.
4757 freeblks = percpu_counter_sum(&mp->m_fdblocks);
4760 resblks = min_t(int64_t, UINT_MAX, freeblks);
4761 resblks = (resblks * 15) >> 4;
4762 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, resblks,
4763 0, XFS_TRANS_RESERVE, &tp);
4766 /* transfer all collected dfops to this transaction */
4767 xfs_defer_move(tp, parent_tp);
4769 return xfs_trans_commit(tp);
4773 * When this is called, all of the log intent items which did not have
4774 * corresponding log done items should be in the AIL. What we do now
4775 * is update the data structures associated with each one.
4777 * Since we process the log intent items in normal transactions, they
4778 * will be removed at some point after the commit. This prevents us
4779 * from just walking down the list processing each one. We'll use a
4780 * flag in the intent item to skip those that we've already processed
4781 * and use the AIL iteration mechanism's generation count to try to
4782 * speed this up at least a bit.
4784 * When we start, we know that the intents are the only things in the
4785 * AIL. As we process them, however, other items are added to the
4789 xlog_recover_process_intents(
4792 struct xfs_trans *parent_tp;
4793 struct xfs_ail_cursor cur;
4794 struct xfs_log_item *lip;
4795 struct xfs_ail *ailp;
4797 #if defined(DEBUG) || defined(XFS_WARN)
4802 * The intent recovery handlers commit transactions to complete recovery
4803 * for individual intents, but any new deferred operations that are
4804 * queued during that process are held off until the very end. The
4805 * purpose of this transaction is to serve as a container for deferred
4806 * operations. Each intent recovery handler must transfer dfops here
4807 * before its local transaction commits, and we'll finish the entire
4810 error = xfs_trans_alloc_empty(log->l_mp, &parent_tp);
4815 spin_lock(&ailp->ail_lock);
4816 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
4817 #if defined(DEBUG) || defined(XFS_WARN)
4818 last_lsn = xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block);
4820 while (lip != NULL) {
4822 * We're done when we see something other than an intent.
4823 * There should be no intents left in the AIL now.
4825 if (!xlog_item_is_intent(lip)) {
4827 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
4828 ASSERT(!xlog_item_is_intent(lip));
4834 * We should never see a redo item with a LSN higher than
4835 * the last transaction we found in the log at the start
4838 ASSERT(XFS_LSN_CMP(last_lsn, lip->li_lsn) >= 0);
4841 * NOTE: If your intent processing routine can create more
4842 * deferred ops, you /must/ attach them to the dfops in this
4843 * routine or else those subsequent intents will get
4844 * replayed in the wrong order!
4846 switch (lip->li_type) {
4848 error = xlog_recover_process_efi(log->l_mp, ailp, lip);
4851 error = xlog_recover_process_rui(log->l_mp, ailp, lip);
4854 error = xlog_recover_process_cui(parent_tp, ailp, lip);
4857 error = xlog_recover_process_bui(parent_tp, ailp, lip);
4862 lip = xfs_trans_ail_cursor_next(ailp, &cur);
4865 xfs_trans_ail_cursor_done(&cur);
4866 spin_unlock(&ailp->ail_lock);
4868 error = xlog_finish_defer_ops(parent_tp);
4869 xfs_trans_cancel(parent_tp);
4875 * A cancel occurs when the mount has failed and we're bailing out.
4876 * Release all pending log intent items so they don't pin the AIL.
4879 xlog_recover_cancel_intents(
4882 struct xfs_log_item *lip;
4883 struct xfs_ail_cursor cur;
4884 struct xfs_ail *ailp;
4887 spin_lock(&ailp->ail_lock);
4888 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
4889 while (lip != NULL) {
4891 * We're done when we see something other than an intent.
4892 * There should be no intents left in the AIL now.
4894 if (!xlog_item_is_intent(lip)) {
4896 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
4897 ASSERT(!xlog_item_is_intent(lip));
4902 switch (lip->li_type) {
4904 xlog_recover_cancel_efi(log->l_mp, ailp, lip);
4907 xlog_recover_cancel_rui(log->l_mp, ailp, lip);
4910 xlog_recover_cancel_cui(log->l_mp, ailp, lip);
4913 xlog_recover_cancel_bui(log->l_mp, ailp, lip);
4917 lip = xfs_trans_ail_cursor_next(ailp, &cur);
4920 xfs_trans_ail_cursor_done(&cur);
4921 spin_unlock(&ailp->ail_lock);
4925 * This routine performs a transaction to null out a bad inode pointer
4926 * in an agi unlinked inode hash bucket.
4929 xlog_recover_clear_agi_bucket(
4931 xfs_agnumber_t agno,
4940 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_clearagi, 0, 0, 0, &tp);
4944 error = xfs_read_agi(mp, tp, agno, &agibp);
4948 agi = XFS_BUF_TO_AGI(agibp);
4949 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
4950 offset = offsetof(xfs_agi_t, agi_unlinked) +
4951 (sizeof(xfs_agino_t) * bucket);
4952 xfs_trans_log_buf(tp, agibp, offset,
4953 (offset + sizeof(xfs_agino_t) - 1));
4955 error = xfs_trans_commit(tp);
4961 xfs_trans_cancel(tp);
4963 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
4968 xlog_recover_process_one_iunlink(
4969 struct xfs_mount *mp,
4970 xfs_agnumber_t agno,
4974 struct xfs_buf *ibp;
4975 struct xfs_dinode *dip;
4976 struct xfs_inode *ip;
4980 ino = XFS_AGINO_TO_INO(mp, agno, agino);
4981 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
4986 * Get the on disk inode to find the next inode in the bucket.
4988 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
4992 xfs_iflags_clear(ip, XFS_IRECOVERY);
4993 ASSERT(VFS_I(ip)->i_nlink == 0);
4994 ASSERT(VFS_I(ip)->i_mode != 0);
4996 /* setup for the next pass */
4997 agino = be32_to_cpu(dip->di_next_unlinked);
5001 * Prevent any DMAPI event from being sent when the reference on
5002 * the inode is dropped.
5004 ip->i_d.di_dmevmask = 0;
5013 * We can't read in the inode this bucket points to, or this inode
5014 * is messed up. Just ditch this bucket of inodes. We will lose
5015 * some inodes and space, but at least we won't hang.
5017 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
5018 * clear the inode pointer in the bucket.
5020 xlog_recover_clear_agi_bucket(mp, agno, bucket);
5025 * xlog_iunlink_recover
5027 * This is called during recovery to process any inodes which
5028 * we unlinked but not freed when the system crashed. These
5029 * inodes will be on the lists in the AGI blocks. What we do
5030 * here is scan all the AGIs and fully truncate and free any
5031 * inodes found on the lists. Each inode is removed from the
5032 * lists when it has been fully truncated and is freed. The
5033 * freeing of the inode and its removal from the list must be
5037 xlog_recover_process_iunlinks(
5041 xfs_agnumber_t agno;
5050 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
5052 * Find the agi for this ag.
5054 error = xfs_read_agi(mp, NULL, agno, &agibp);
5057 * AGI is b0rked. Don't process it.
5059 * We should probably mark the filesystem as corrupt
5060 * after we've recovered all the ag's we can....
5065 * Unlock the buffer so that it can be acquired in the normal
5066 * course of the transaction to truncate and free each inode.
5067 * Because we are not racing with anyone else here for the AGI
5068 * buffer, we don't even need to hold it locked to read the
5069 * initial unlinked bucket entries out of the buffer. We keep
5070 * buffer reference though, so that it stays pinned in memory
5071 * while we need the buffer.
5073 agi = XFS_BUF_TO_AGI(agibp);
5074 xfs_buf_unlock(agibp);
5076 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
5077 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
5078 while (agino != NULLAGINO) {
5079 agino = xlog_recover_process_one_iunlink(mp,
5080 agno, agino, bucket);
5083 xfs_buf_rele(agibp);
5089 struct xlog_rec_header *rhead,
5095 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
5096 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
5097 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
5101 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
5102 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
5103 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
5104 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
5105 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
5106 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
5113 * CRC check, unpack and process a log record.
5116 xlog_recover_process(
5118 struct hlist_head rhash[],
5119 struct xlog_rec_header *rhead,
5122 struct list_head *buffer_list)
5124 __le32 old_crc = rhead->h_crc;
5127 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
5130 * Nothing else to do if this is a CRC verification pass. Just return
5131 * if this a record with a non-zero crc. Unfortunately, mkfs always
5132 * sets old_crc to 0 so we must consider this valid even on v5 supers.
5133 * Otherwise, return EFSBADCRC on failure so the callers up the stack
5134 * know precisely what failed.
5136 if (pass == XLOG_RECOVER_CRCPASS) {
5137 if (old_crc && crc != old_crc)
5143 * We're in the normal recovery path. Issue a warning if and only if the
5144 * CRC in the header is non-zero. This is an advisory warning and the
5145 * zero CRC check prevents warnings from being emitted when upgrading
5146 * the kernel from one that does not add CRCs by default.
5148 if (crc != old_crc) {
5149 if (old_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
5150 xfs_alert(log->l_mp,
5151 "log record CRC mismatch: found 0x%x, expected 0x%x.",
5152 le32_to_cpu(old_crc),
5154 xfs_hex_dump(dp, 32);
5158 * If the filesystem is CRC enabled, this mismatch becomes a
5159 * fatal log corruption failure.
5161 if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
5162 return -EFSCORRUPTED;
5165 xlog_unpack_data(rhead, dp, log);
5167 return xlog_recover_process_data(log, rhash, rhead, dp, pass,
5172 xlog_valid_rec_header(
5174 struct xlog_rec_header *rhead,
5179 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
5180 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
5181 XFS_ERRLEVEL_LOW, log->l_mp);
5182 return -EFSCORRUPTED;
5185 (!rhead->h_version ||
5186 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
5187 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
5188 __func__, be32_to_cpu(rhead->h_version));
5192 /* LR body must have data or it wouldn't have been written */
5193 hlen = be32_to_cpu(rhead->h_len);
5194 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
5195 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
5196 XFS_ERRLEVEL_LOW, log->l_mp);
5197 return -EFSCORRUPTED;
5199 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
5200 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
5201 XFS_ERRLEVEL_LOW, log->l_mp);
5202 return -EFSCORRUPTED;
5208 * Read the log from tail to head and process the log records found.
5209 * Handle the two cases where the tail and head are in the same cycle
5210 * and where the active portion of the log wraps around the end of
5211 * the physical log separately. The pass parameter is passed through
5212 * to the routines called to process the data and is not looked at
5216 xlog_do_recovery_pass(
5218 xfs_daddr_t head_blk,
5219 xfs_daddr_t tail_blk,
5221 xfs_daddr_t *first_bad) /* out: first bad log rec */
5223 xlog_rec_header_t *rhead;
5224 xfs_daddr_t blk_no, rblk_no;
5225 xfs_daddr_t rhead_blk;
5228 int error = 0, h_size, h_len;
5230 int bblks, split_bblks;
5231 int hblks, split_hblks, wrapped_hblks;
5233 struct hlist_head rhash[XLOG_RHASH_SIZE];
5234 LIST_HEAD (buffer_list);
5236 ASSERT(head_blk != tail_blk);
5237 blk_no = rhead_blk = tail_blk;
5239 for (i = 0; i < XLOG_RHASH_SIZE; i++)
5240 INIT_HLIST_HEAD(&rhash[i]);
5243 * Read the header of the tail block and get the iclog buffer size from
5244 * h_size. Use this to tell how many sectors make up the log header.
5246 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
5248 * When using variable length iclogs, read first sector of
5249 * iclog header and extract the header size from it. Get a
5250 * new hbp that is the correct size.
5252 hbp = xlog_alloc_buffer(log, 1);
5256 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
5260 rhead = (xlog_rec_header_t *)offset;
5261 error = xlog_valid_rec_header(log, rhead, tail_blk);
5266 * xfsprogs has a bug where record length is based on lsunit but
5267 * h_size (iclog size) is hardcoded to 32k. Now that we
5268 * unconditionally CRC verify the unmount record, this means the
5269 * log buffer can be too small for the record and cause an
5272 * Detect this condition here. Use lsunit for the buffer size as
5273 * long as this looks like the mkfs case. Otherwise, return an
5274 * error to avoid a buffer overrun.
5276 h_size = be32_to_cpu(rhead->h_size);
5277 h_len = be32_to_cpu(rhead->h_len);
5278 if (h_len > h_size) {
5279 if (h_len <= log->l_mp->m_logbsize &&
5280 be32_to_cpu(rhead->h_num_logops) == 1) {
5282 "invalid iclog size (%d bytes), using lsunit (%d bytes)",
5283 h_size, log->l_mp->m_logbsize);
5284 h_size = log->l_mp->m_logbsize;
5286 return -EFSCORRUPTED;
5289 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
5290 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
5291 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
5292 if (h_size % XLOG_HEADER_CYCLE_SIZE)
5295 hbp = xlog_alloc_buffer(log, hblks);
5300 ASSERT(log->l_sectBBsize == 1);
5302 hbp = xlog_alloc_buffer(log, 1);
5303 h_size = XLOG_BIG_RECORD_BSIZE;
5308 dbp = xlog_alloc_buffer(log, BTOBB(h_size));
5314 memset(rhash, 0, sizeof(rhash));
5315 if (tail_blk > head_blk) {
5317 * Perform recovery around the end of the physical log.
5318 * When the head is not on the same cycle number as the tail,
5319 * we can't do a sequential recovery.
5321 while (blk_no < log->l_logBBsize) {
5323 * Check for header wrapping around physical end-of-log
5328 if (blk_no + hblks <= log->l_logBBsize) {
5329 /* Read header in one read */
5330 error = xlog_bread(log, blk_no, hblks, hbp,
5335 /* This LR is split across physical log end */
5336 if (blk_no != log->l_logBBsize) {
5337 /* some data before physical log end */
5338 ASSERT(blk_no <= INT_MAX);
5339 split_hblks = log->l_logBBsize - (int)blk_no;
5340 ASSERT(split_hblks > 0);
5341 error = xlog_bread(log, blk_no,
5349 * Note: this black magic still works with
5350 * large sector sizes (non-512) only because:
5351 * - we increased the buffer size originally
5352 * by 1 sector giving us enough extra space
5353 * for the second read;
5354 * - the log start is guaranteed to be sector
5356 * - we read the log end (LR header start)
5357 * _first_, then the log start (LR header end)
5358 * - order is important.
5360 wrapped_hblks = hblks - split_hblks;
5361 error = xlog_bread_noalign(log, 0,
5363 offset + BBTOB(split_hblks));
5367 rhead = (xlog_rec_header_t *)offset;
5368 error = xlog_valid_rec_header(log, rhead,
5369 split_hblks ? blk_no : 0);
5373 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
5377 * Read the log record data in multiple reads if it
5378 * wraps around the end of the log. Note that if the
5379 * header already wrapped, blk_no could point past the
5380 * end of the log. The record data is contiguous in
5383 if (blk_no + bblks <= log->l_logBBsize ||
5384 blk_no >= log->l_logBBsize) {
5385 rblk_no = xlog_wrap_logbno(log, blk_no);
5386 error = xlog_bread(log, rblk_no, bblks, dbp,
5391 /* This log record is split across the
5392 * physical end of log */
5395 if (blk_no != log->l_logBBsize) {
5396 /* some data is before the physical
5398 ASSERT(!wrapped_hblks);
5399 ASSERT(blk_no <= INT_MAX);
5401 log->l_logBBsize - (int)blk_no;
5402 ASSERT(split_bblks > 0);
5403 error = xlog_bread(log, blk_no,
5411 * Note: this black magic still works with
5412 * large sector sizes (non-512) only because:
5413 * - we increased the buffer size originally
5414 * by 1 sector giving us enough extra space
5415 * for the second read;
5416 * - the log start is guaranteed to be sector
5418 * - we read the log end (LR header start)
5419 * _first_, then the log start (LR header end)
5420 * - order is important.
5422 error = xlog_bread_noalign(log, 0,
5423 bblks - split_bblks,
5424 offset + BBTOB(split_bblks));
5429 error = xlog_recover_process(log, rhash, rhead, offset,
5430 pass, &buffer_list);
5438 ASSERT(blk_no >= log->l_logBBsize);
5439 blk_no -= log->l_logBBsize;
5443 /* read first part of physical log */
5444 while (blk_no < head_blk) {
5445 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
5449 rhead = (xlog_rec_header_t *)offset;
5450 error = xlog_valid_rec_header(log, rhead, blk_no);
5454 /* blocks in data section */
5455 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
5456 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
5461 error = xlog_recover_process(log, rhash, rhead, offset, pass,
5466 blk_no += bblks + hblks;
5476 * Submit buffers that have been added from the last record processed,
5477 * regardless of error status.
5479 if (!list_empty(&buffer_list))
5480 error2 = xfs_buf_delwri_submit(&buffer_list);
5482 if (error && first_bad)
5483 *first_bad = rhead_blk;
5486 * Transactions are freed at commit time but transactions without commit
5487 * records on disk are never committed. Free any that may be left in the
5490 for (i = 0; i < XLOG_RHASH_SIZE; i++) {
5491 struct hlist_node *tmp;
5492 struct xlog_recover *trans;
5494 hlist_for_each_entry_safe(trans, tmp, &rhash[i], r_list)
5495 xlog_recover_free_trans(trans);
5498 return error ? error : error2;
5502 * Do the recovery of the log. We actually do this in two phases.
5503 * The two passes are necessary in order to implement the function
5504 * of cancelling a record written into the log. The first pass
5505 * determines those things which have been cancelled, and the
5506 * second pass replays log items normally except for those which
5507 * have been cancelled. The handling of the replay and cancellations
5508 * takes place in the log item type specific routines.
5510 * The table of items which have cancel records in the log is allocated
5511 * and freed at this level, since only here do we know when all of
5512 * the log recovery has been completed.
5515 xlog_do_log_recovery(
5517 xfs_daddr_t head_blk,
5518 xfs_daddr_t tail_blk)
5522 ASSERT(head_blk != tail_blk);
5525 * First do a pass to find all of the cancelled buf log items.
5526 * Store them in the buf_cancel_table for use in the second pass.
5528 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
5529 sizeof(struct list_head),
5531 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
5532 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
5534 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
5535 XLOG_RECOVER_PASS1, NULL);
5537 kmem_free(log->l_buf_cancel_table);
5538 log->l_buf_cancel_table = NULL;
5542 * Then do a second pass to actually recover the items in the log.
5543 * When it is complete free the table of buf cancel items.
5545 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
5546 XLOG_RECOVER_PASS2, NULL);
5551 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
5552 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
5556 kmem_free(log->l_buf_cancel_table);
5557 log->l_buf_cancel_table = NULL;
5563 * Do the actual recovery
5568 xfs_daddr_t head_blk,
5569 xfs_daddr_t tail_blk)
5571 struct xfs_mount *mp = log->l_mp;
5576 trace_xfs_log_recover(log, head_blk, tail_blk);
5579 * First replay the images in the log.
5581 error = xlog_do_log_recovery(log, head_blk, tail_blk);
5586 * If IO errors happened during recovery, bail out.
5588 if (XFS_FORCED_SHUTDOWN(mp)) {
5593 * We now update the tail_lsn since much of the recovery has completed
5594 * and there may be space available to use. If there were no extent
5595 * or iunlinks, we can free up the entire log and set the tail_lsn to
5596 * be the last_sync_lsn. This was set in xlog_find_tail to be the
5597 * lsn of the last known good LR on disk. If there are extent frees
5598 * or iunlinks they will have some entries in the AIL; so we look at
5599 * the AIL to determine how to set the tail_lsn.
5601 xlog_assign_tail_lsn(mp);
5604 * Now that we've finished replaying all buffer and inode
5605 * updates, re-read in the superblock and reverify it.
5608 bp->b_flags &= ~(XBF_DONE | XBF_ASYNC);
5609 ASSERT(!(bp->b_flags & XBF_WRITE));
5610 bp->b_flags |= XBF_READ;
5611 bp->b_ops = &xfs_sb_buf_ops;
5613 error = xfs_buf_submit(bp);
5615 if (!XFS_FORCED_SHUTDOWN(mp)) {
5616 xfs_buf_ioerror_alert(bp, __func__);
5623 /* Convert superblock from on-disk format */
5625 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
5628 /* re-initialise in-core superblock and geometry structures */
5629 xfs_reinit_percpu_counters(mp);
5630 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
5632 xfs_warn(mp, "Failed post-recovery per-ag init: %d", error);
5635 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
5637 xlog_recover_check_summary(log);
5639 /* Normal transactions can now occur */
5640 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
5645 * Perform recovery and re-initialize some log variables in xlog_find_tail.
5647 * Return error or zero.
5653 xfs_daddr_t head_blk, tail_blk;
5656 /* find the tail of the log */
5657 error = xlog_find_tail(log, &head_blk, &tail_blk);
5662 * The superblock was read before the log was available and thus the LSN
5663 * could not be verified. Check the superblock LSN against the current
5664 * LSN now that it's known.
5666 if (xfs_sb_version_hascrc(&log->l_mp->m_sb) &&
5667 !xfs_log_check_lsn(log->l_mp, log->l_mp->m_sb.sb_lsn))
5670 if (tail_blk != head_blk) {
5671 /* There used to be a comment here:
5673 * disallow recovery on read-only mounts. note -- mount
5674 * checks for ENOSPC and turns it into an intelligent
5676 * ...but this is no longer true. Now, unless you specify
5677 * NORECOVERY (in which case this function would never be
5678 * called), we just go ahead and recover. We do this all
5679 * under the vfs layer, so we can get away with it unless
5680 * the device itself is read-only, in which case we fail.
5682 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
5687 * Version 5 superblock log feature mask validation. We know the
5688 * log is dirty so check if there are any unknown log features
5689 * in what we need to recover. If there are unknown features
5690 * (e.g. unsupported transactions, then simply reject the
5691 * attempt at recovery before touching anything.
5693 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
5694 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
5695 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
5697 "Superblock has unknown incompatible log features (0x%x) enabled.",
5698 (log->l_mp->m_sb.sb_features_log_incompat &
5699 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
5701 "The log can not be fully and/or safely recovered by this kernel.");
5703 "Please recover the log on a kernel that supports the unknown features.");
5708 * Delay log recovery if the debug hook is set. This is debug
5709 * instrumention to coordinate simulation of I/O failures with
5712 if (xfs_globals.log_recovery_delay) {
5713 xfs_notice(log->l_mp,
5714 "Delaying log recovery for %d seconds.",
5715 xfs_globals.log_recovery_delay);
5716 msleep(xfs_globals.log_recovery_delay * 1000);
5719 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
5720 log->l_mp->m_logname ? log->l_mp->m_logname
5723 error = xlog_do_recover(log, head_blk, tail_blk);
5724 log->l_flags |= XLOG_RECOVERY_NEEDED;
5730 * In the first part of recovery we replay inodes and buffers and build
5731 * up the list of extent free items which need to be processed. Here
5732 * we process the extent free items and clean up the on disk unlinked
5733 * inode lists. This is separated from the first part of recovery so
5734 * that the root and real-time bitmap inodes can be read in from disk in
5735 * between the two stages. This is necessary so that we can free space
5736 * in the real-time portion of the file system.
5739 xlog_recover_finish(
5743 * Now we're ready to do the transactions needed for the
5744 * rest of recovery. Start with completing all the extent
5745 * free intent records and then process the unlinked inode
5746 * lists. At this point, we essentially run in normal mode
5747 * except that we're still performing recovery actions
5748 * rather than accepting new requests.
5750 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
5752 error = xlog_recover_process_intents(log);
5754 xfs_alert(log->l_mp, "Failed to recover intents");
5759 * Sync the log to get all the intents out of the AIL.
5760 * This isn't absolutely necessary, but it helps in
5761 * case the unlink transactions would have problems
5762 * pushing the intents out of the way.
5764 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
5766 xlog_recover_process_iunlinks(log);
5768 xlog_recover_check_summary(log);
5770 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
5771 log->l_mp->m_logname ? log->l_mp->m_logname
5773 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
5775 xfs_info(log->l_mp, "Ending clean mount");
5781 xlog_recover_cancel(
5784 if (log->l_flags & XLOG_RECOVERY_NEEDED)
5785 xlog_recover_cancel_intents(log);
5790 * Read all of the agf and agi counters and check that they
5791 * are consistent with the superblock counters.
5794 xlog_recover_check_summary(
5801 xfs_agnumber_t agno;
5812 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
5813 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
5815 xfs_alert(mp, "%s agf read failed agno %d error %d",
5816 __func__, agno, error);
5818 agfp = XFS_BUF_TO_AGF(agfbp);
5819 freeblks += be32_to_cpu(agfp->agf_freeblks) +
5820 be32_to_cpu(agfp->agf_flcount);
5821 xfs_buf_relse(agfbp);
5824 error = xfs_read_agi(mp, NULL, agno, &agibp);
5826 xfs_alert(mp, "%s agi read failed agno %d error %d",
5827 __func__, agno, error);
5829 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
5831 itotal += be32_to_cpu(agi->agi_count);
5832 ifree += be32_to_cpu(agi->agi_freecount);
5833 xfs_buf_relse(agibp);