1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_inode.h"
17 #include "xfs_trans.h"
19 #include "xfs_log_priv.h"
20 #include "xfs_log_recover.h"
21 #include "xfs_inode_item.h"
22 #include "xfs_extfree_item.h"
23 #include "xfs_trans_priv.h"
24 #include "xfs_alloc.h"
25 #include "xfs_ialloc.h"
26 #include "xfs_quota.h"
27 #include "xfs_trace.h"
28 #include "xfs_icache.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_error.h"
32 #include "xfs_rmap_item.h"
33 #include "xfs_buf_item.h"
34 #include "xfs_refcount_item.h"
35 #include "xfs_bmap_item.h"
37 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
44 xlog_clear_stale_blocks(
49 xlog_recover_check_summary(
52 #define xlog_recover_check_summary(log)
55 xlog_do_recovery_pass(
56 struct xlog *, xfs_daddr_t, xfs_daddr_t, int, xfs_daddr_t *);
59 * This structure is used during recovery to record the buf log items which
60 * have been canceled and should not be replayed.
62 struct xfs_buf_cancel {
66 struct list_head bc_list;
70 * Sector aligned buffer routines for buffer create/read/write/access
74 * Verify the log-relative block number and length in basic blocks are valid for
75 * an operation involving the given XFS log buffer. Returns true if the fields
76 * are valid, false otherwise.
84 if (blk_no < 0 || blk_no >= log->l_logBBsize)
86 if (bbcount <= 0 || (blk_no + bbcount) > log->l_logBBsize)
92 * Allocate a buffer to hold log data. The buffer needs to be able to map to
93 * a range of nbblks basic blocks at any valid offset within the log.
100 int align_mask = xfs_buftarg_dma_alignment(log->l_targ);
103 * Pass log block 0 since we don't have an addr yet, buffer will be
106 if (XFS_IS_CORRUPT(log->l_mp, !xlog_verify_bno(log, 0, nbblks))) {
107 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
113 * We do log I/O in units of log sectors (a power-of-2 multiple of the
114 * basic block size), so we round up the requested size to accommodate
115 * the basic blocks required for complete log sectors.
117 * In addition, the buffer may be used for a non-sector-aligned block
118 * offset, in which case an I/O of the requested size could extend
119 * beyond the end of the buffer. If the requested size is only 1 basic
120 * block it will never straddle a sector boundary, so this won't be an
121 * issue. Nor will this be a problem if the log I/O is done in basic
122 * blocks (sector size 1). But otherwise we extend the buffer by one
123 * extra log sector to ensure there's space to accommodate this
126 if (nbblks > 1 && log->l_sectBBsize > 1)
127 nbblks += log->l_sectBBsize;
128 nbblks = round_up(nbblks, log->l_sectBBsize);
129 return kmem_alloc_io(BBTOB(nbblks), align_mask, KM_MAYFAIL | KM_ZERO);
133 * Return the address of the start of the given block number's data
134 * in a log buffer. The buffer covers a log sector-aligned region.
136 static inline unsigned int
141 return BBTOB(blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1));
154 if (XFS_IS_CORRUPT(log->l_mp, !xlog_verify_bno(log, blk_no, nbblks))) {
156 "Invalid log block/length (0x%llx, 0x%x) for buffer",
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 (XFS_IS_CORRUPT(mp, 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 return -EFSCORRUPTED;
251 if (XFS_IS_CORRUPT(mp, !uuid_equal(&mp->m_sb.sb_uuid,
252 &head->h_fs_uuid))) {
254 "dirty log entry has mismatched uuid - can't recover");
255 xlog_header_check_dump(mp, head);
256 return -EFSCORRUPTED;
262 * read the head block of the log and check the header
265 xlog_header_check_mount(
267 xlog_rec_header_t *head)
269 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
271 if (uuid_is_null(&head->h_fs_uuid)) {
273 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
274 * h_fs_uuid is null, we assume this log was last mounted
275 * by IRIX and continue.
277 xfs_warn(mp, "null uuid in log - IRIX style log");
278 } else if (XFS_IS_CORRUPT(mp, !uuid_equal(&mp->m_sb.sb_uuid,
279 &head->h_fs_uuid))) {
280 xfs_warn(mp, "log has mismatched uuid - can't recover");
281 xlog_header_check_dump(mp, head);
282 return -EFSCORRUPTED;
293 * We're not going to bother about retrying
294 * this during recovery. One strike!
296 if (!XFS_FORCED_SHUTDOWN(bp->b_mount)) {
297 xfs_buf_ioerror_alert(bp, __func__);
298 xfs_force_shutdown(bp->b_mount, SHUTDOWN_META_IO_ERROR);
303 * On v5 supers, a bli could be attached to update the metadata LSN.
307 xfs_buf_item_relse(bp);
308 ASSERT(bp->b_log_item == NULL);
315 * This routine finds (to an approximation) the first block in the physical
316 * log which contains the given cycle. It uses a binary search algorithm.
317 * Note that the algorithm can not be perfect because the disk will not
318 * necessarily be perfect.
321 xlog_find_cycle_start(
324 xfs_daddr_t first_blk,
325 xfs_daddr_t *last_blk,
335 mid_blk = BLK_AVG(first_blk, end_blk);
336 while (mid_blk != first_blk && mid_blk != end_blk) {
337 error = xlog_bread(log, mid_blk, 1, buffer, &offset);
340 mid_cycle = xlog_get_cycle(offset);
341 if (mid_cycle == cycle)
342 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
344 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
345 mid_blk = BLK_AVG(first_blk, end_blk);
347 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
348 (mid_blk == end_blk && mid_blk-1 == first_blk));
356 * Check that a range of blocks does not contain stop_on_cycle_no.
357 * Fill in *new_blk with the block offset where such a block is
358 * found, or with -1 (an invalid block number) if there is no such
359 * block in the range. The scan needs to occur from front to back
360 * and the pointer into the region must be updated since a later
361 * routine will need to perform another test.
364 xlog_find_verify_cycle(
366 xfs_daddr_t start_blk,
368 uint stop_on_cycle_no,
369 xfs_daddr_t *new_blk)
379 * Greedily allocate a buffer big enough to handle the full
380 * range of basic blocks we'll be examining. If that fails,
381 * try a smaller size. We need to be able to read at least
382 * a log sector, or we're out of luck.
384 bufblks = 1 << ffs(nbblks);
385 while (bufblks > log->l_logBBsize)
387 while (!(buffer = xlog_alloc_buffer(log, bufblks))) {
389 if (bufblks < log->l_sectBBsize)
393 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
396 bcount = min(bufblks, (start_blk + nbblks - i));
398 error = xlog_bread(log, i, bcount, buffer, &buf);
402 for (j = 0; j < bcount; j++) {
403 cycle = xlog_get_cycle(buf);
404 if (cycle == stop_on_cycle_no) {
421 * Potentially backup over partial log record write.
423 * In the typical case, last_blk is the number of the block directly after
424 * a good log record. Therefore, we subtract one to get the block number
425 * of the last block in the given buffer. extra_bblks contains the number
426 * of blocks we would have read on a previous read. This happens when the
427 * last log record is split over the end of the physical log.
429 * extra_bblks is the number of blocks potentially verified on a previous
430 * call to this routine.
433 xlog_find_verify_log_record(
435 xfs_daddr_t start_blk,
436 xfs_daddr_t *last_blk,
442 xlog_rec_header_t *head = NULL;
445 int num_blks = *last_blk - start_blk;
448 ASSERT(start_blk != 0 || *last_blk != start_blk);
450 buffer = xlog_alloc_buffer(log, num_blks);
452 buffer = xlog_alloc_buffer(log, 1);
457 error = xlog_bread(log, start_blk, num_blks, buffer, &offset);
460 offset += ((num_blks - 1) << BBSHIFT);
463 for (i = (*last_blk) - 1; i >= 0; i--) {
465 /* valid log record not found */
467 "Log inconsistent (didn't find previous header)");
469 error = -EFSCORRUPTED;
474 error = xlog_bread(log, i, 1, buffer, &offset);
479 head = (xlog_rec_header_t *)offset;
481 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
489 * We hit the beginning of the physical log & still no header. Return
490 * to caller. If caller can handle a return of -1, then this routine
491 * will be called again for the end of the physical log.
499 * We have the final block of the good log (the first block
500 * of the log record _before_ the head. So we check the uuid.
502 if ((error = xlog_header_check_mount(log->l_mp, head)))
506 * We may have found a log record header before we expected one.
507 * last_blk will be the 1st block # with a given cycle #. We may end
508 * up reading an entire log record. In this case, we don't want to
509 * reset last_blk. Only when last_blk points in the middle of a log
510 * record do we update last_blk.
512 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
513 uint h_size = be32_to_cpu(head->h_size);
515 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
516 if (h_size % XLOG_HEADER_CYCLE_SIZE)
522 if (*last_blk - i + extra_bblks !=
523 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
532 * Head is defined to be the point of the log where the next log write
533 * could go. This means that incomplete LR writes at the end are
534 * eliminated when calculating the head. We aren't guaranteed that previous
535 * LR have complete transactions. We only know that a cycle number of
536 * current cycle number -1 won't be present in the log if we start writing
537 * from our current block number.
539 * last_blk contains the block number of the first block with a given
542 * Return: zero if normal, non-zero if error.
547 xfs_daddr_t *return_head_blk)
551 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
553 uint first_half_cycle, last_half_cycle;
555 int error, log_bbnum = log->l_logBBsize;
557 /* Is the end of the log device zeroed? */
558 error = xlog_find_zeroed(log, &first_blk);
560 xfs_warn(log->l_mp, "empty log check failed");
564 *return_head_blk = first_blk;
566 /* Is the whole lot zeroed? */
568 /* Linux XFS shouldn't generate totally zeroed logs -
569 * mkfs etc write a dummy unmount record to a fresh
570 * log so we can store the uuid in there
572 xfs_warn(log->l_mp, "totally zeroed log");
578 first_blk = 0; /* get cycle # of 1st block */
579 buffer = xlog_alloc_buffer(log, 1);
583 error = xlog_bread(log, 0, 1, buffer, &offset);
585 goto out_free_buffer;
587 first_half_cycle = xlog_get_cycle(offset);
589 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
590 error = xlog_bread(log, last_blk, 1, buffer, &offset);
592 goto out_free_buffer;
594 last_half_cycle = xlog_get_cycle(offset);
595 ASSERT(last_half_cycle != 0);
598 * If the 1st half cycle number is equal to the last half cycle number,
599 * then the entire log is stamped with the same cycle number. In this
600 * case, head_blk can't be set to zero (which makes sense). The below
601 * math doesn't work out properly with head_blk equal to zero. Instead,
602 * we set it to log_bbnum which is an invalid block number, but this
603 * value makes the math correct. If head_blk doesn't changed through
604 * all the tests below, *head_blk is set to zero at the very end rather
605 * than log_bbnum. In a sense, log_bbnum and zero are the same block
606 * in a circular file.
608 if (first_half_cycle == last_half_cycle) {
610 * In this case we believe that the entire log should have
611 * cycle number last_half_cycle. We need to scan backwards
612 * from the end verifying that there are no holes still
613 * containing last_half_cycle - 1. If we find such a hole,
614 * then the start of that hole will be the new head. The
615 * simple case looks like
616 * x | x ... | x - 1 | x
617 * Another case that fits this picture would be
618 * x | x + 1 | x ... | x
619 * In this case the head really is somewhere at the end of the
620 * log, as one of the latest writes at the beginning was
623 * x | x + 1 | x ... | x - 1 | x
624 * This is really the combination of the above two cases, and
625 * the head has to end up at the start of the x-1 hole at the
628 * In the 256k log case, we will read from the beginning to the
629 * end of the log and search for cycle numbers equal to x-1.
630 * We don't worry about the x+1 blocks that we encounter,
631 * because we know that they cannot be the head since the log
634 head_blk = log_bbnum;
635 stop_on_cycle = last_half_cycle - 1;
638 * In this case we want to find the first block with cycle
639 * number matching last_half_cycle. We expect the log to be
641 * x + 1 ... | x ... | x
642 * The first block with cycle number x (last_half_cycle) will
643 * be where the new head belongs. First we do a binary search
644 * for the first occurrence of last_half_cycle. The binary
645 * search may not be totally accurate, so then we scan back
646 * from there looking for occurrences of last_half_cycle before
647 * us. If that backwards scan wraps around the beginning of
648 * the log, then we look for occurrences of last_half_cycle - 1
649 * at the end of the log. The cases we're looking for look
651 * v binary search stopped here
652 * x + 1 ... | x | x + 1 | x ... | x
653 * ^ but we want to locate this spot
655 * <---------> less than scan distance
656 * x + 1 ... | x ... | x - 1 | x
657 * ^ we want to locate this spot
659 stop_on_cycle = last_half_cycle;
660 error = xlog_find_cycle_start(log, buffer, first_blk, &head_blk,
663 goto out_free_buffer;
667 * Now validate the answer. Scan back some number of maximum possible
668 * blocks and make sure each one has the expected cycle number. The
669 * maximum is determined by the total possible amount of buffering
670 * in the in-core log. The following number can be made tighter if
671 * we actually look at the block size of the filesystem.
673 num_scan_bblks = min_t(int, log_bbnum, XLOG_TOTAL_REC_SHIFT(log));
674 if (head_blk >= num_scan_bblks) {
676 * We are guaranteed that the entire check can be performed
679 start_blk = head_blk - num_scan_bblks;
680 if ((error = xlog_find_verify_cycle(log,
681 start_blk, num_scan_bblks,
682 stop_on_cycle, &new_blk)))
683 goto out_free_buffer;
686 } else { /* need to read 2 parts of log */
688 * We are going to scan backwards in the log in two parts.
689 * First we scan the physical end of the log. In this part
690 * of the log, we are looking for blocks with cycle number
691 * last_half_cycle - 1.
692 * If we find one, then we know that the log starts there, as
693 * we've found a hole that didn't get written in going around
694 * the end of the physical log. The simple case for this is
695 * x + 1 ... | x ... | x - 1 | x
696 * <---------> less than scan distance
697 * If all of the blocks at the end of the log have cycle number
698 * last_half_cycle, then we check the blocks at the start of
699 * the log looking for occurrences of last_half_cycle. If we
700 * find one, then our current estimate for the location of the
701 * first occurrence of last_half_cycle is wrong and we move
702 * back to the hole we've found. This case looks like
703 * x + 1 ... | x | x + 1 | x ...
704 * ^ binary search stopped here
705 * Another case we need to handle that only occurs in 256k
707 * x + 1 ... | x ... | x+1 | x ...
708 * ^ binary search stops here
709 * In a 256k log, the scan at the end of the log will see the
710 * x + 1 blocks. We need to skip past those since that is
711 * certainly not the head of the log. By searching for
712 * last_half_cycle-1 we accomplish that.
714 ASSERT(head_blk <= INT_MAX &&
715 (xfs_daddr_t) num_scan_bblks >= head_blk);
716 start_blk = log_bbnum - (num_scan_bblks - head_blk);
717 if ((error = xlog_find_verify_cycle(log, start_blk,
718 num_scan_bblks - (int)head_blk,
719 (stop_on_cycle - 1), &new_blk)))
720 goto out_free_buffer;
727 * Scan beginning of log now. The last part of the physical
728 * log is good. This scan needs to verify that it doesn't find
729 * the last_half_cycle.
732 ASSERT(head_blk <= INT_MAX);
733 if ((error = xlog_find_verify_cycle(log,
734 start_blk, (int)head_blk,
735 stop_on_cycle, &new_blk)))
736 goto out_free_buffer;
743 * Now we need to make sure head_blk is not pointing to a block in
744 * the middle of a log record.
746 num_scan_bblks = XLOG_REC_SHIFT(log);
747 if (head_blk >= num_scan_bblks) {
748 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
750 /* start ptr at last block ptr before head_blk */
751 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
755 goto out_free_buffer;
758 ASSERT(head_blk <= INT_MAX);
759 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
761 goto out_free_buffer;
763 /* We hit the beginning of the log during our search */
764 start_blk = log_bbnum - (num_scan_bblks - head_blk);
766 ASSERT(start_blk <= INT_MAX &&
767 (xfs_daddr_t) log_bbnum-start_blk >= 0);
768 ASSERT(head_blk <= INT_MAX);
769 error = xlog_find_verify_log_record(log, start_blk,
770 &new_blk, (int)head_blk);
774 goto out_free_buffer;
775 if (new_blk != log_bbnum)
778 goto out_free_buffer;
782 if (head_blk == log_bbnum)
783 *return_head_blk = 0;
785 *return_head_blk = head_blk;
787 * When returning here, we have a good block number. Bad block
788 * means that during a previous crash, we didn't have a clean break
789 * from cycle number N to cycle number N-1. In this case, we need
790 * to find the first block with cycle number N-1.
797 xfs_warn(log->l_mp, "failed to find log head");
802 * Seek backwards in the log for log record headers.
804 * Given a starting log block, walk backwards until we find the provided number
805 * of records or hit the provided tail block. The return value is the number of
806 * records encountered or a negative error code. The log block and buffer
807 * pointer of the last record seen are returned in rblk and rhead respectively.
810 xlog_rseek_logrec_hdr(
812 xfs_daddr_t head_blk,
813 xfs_daddr_t tail_blk,
817 struct xlog_rec_header **rhead,
829 * Walk backwards from the head block until we hit the tail or the first
832 end_blk = head_blk > tail_blk ? tail_blk : 0;
833 for (i = (int) head_blk - 1; i >= end_blk; i--) {
834 error = xlog_bread(log, i, 1, buffer, &offset);
838 if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
840 *rhead = (struct xlog_rec_header *) offset;
841 if (++found == count)
847 * If we haven't hit the tail block or the log record header count,
848 * start looking again from the end of the physical log. Note that
849 * callers can pass head == tail if the tail is not yet known.
851 if (tail_blk >= head_blk && found != count) {
852 for (i = log->l_logBBsize - 1; i >= (int) tail_blk; i--) {
853 error = xlog_bread(log, i, 1, buffer, &offset);
857 if (*(__be32 *)offset ==
858 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
861 *rhead = (struct xlog_rec_header *) offset;
862 if (++found == count)
875 * Seek forward in the log for log record headers.
877 * Given head and tail blocks, walk forward from the tail block until we find
878 * the provided number of records or hit the head block. The return value is the
879 * number of records encountered or a negative error code. The log block and
880 * buffer pointer of the last record seen are returned in rblk and rhead
884 xlog_seek_logrec_hdr(
886 xfs_daddr_t head_blk,
887 xfs_daddr_t tail_blk,
891 struct xlog_rec_header **rhead,
903 * Walk forward from the tail block until we hit the head or the last
906 end_blk = head_blk > tail_blk ? head_blk : log->l_logBBsize - 1;
907 for (i = (int) tail_blk; i <= end_blk; i++) {
908 error = xlog_bread(log, i, 1, buffer, &offset);
912 if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
914 *rhead = (struct xlog_rec_header *) offset;
915 if (++found == count)
921 * If we haven't hit the head block or the log record header count,
922 * start looking again from the start of the physical log.
924 if (tail_blk > head_blk && found != count) {
925 for (i = 0; i < (int) head_blk; i++) {
926 error = xlog_bread(log, i, 1, buffer, &offset);
930 if (*(__be32 *)offset ==
931 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
934 *rhead = (struct xlog_rec_header *) offset;
935 if (++found == count)
948 * Calculate distance from head to tail (i.e., unused space in the log).
953 xfs_daddr_t head_blk,
954 xfs_daddr_t tail_blk)
956 if (head_blk < tail_blk)
957 return tail_blk - head_blk;
959 return tail_blk + (log->l_logBBsize - head_blk);
963 * Verify the log tail. This is particularly important when torn or incomplete
964 * writes have been detected near the front of the log and the head has been
965 * walked back accordingly.
967 * We also have to handle the case where the tail was pinned and the head
968 * blocked behind the tail right before a crash. If the tail had been pushed
969 * immediately prior to the crash and the subsequent checkpoint was only
970 * partially written, it's possible it overwrote the last referenced tail in the
971 * log with garbage. This is not a coherency problem because the tail must have
972 * been pushed before it can be overwritten, but appears as log corruption to
973 * recovery because we have no way to know the tail was updated if the
974 * subsequent checkpoint didn't write successfully.
976 * Therefore, CRC check the log from tail to head. If a failure occurs and the
977 * offending record is within max iclog bufs from the head, walk the tail
978 * forward and retry until a valid tail is found or corruption is detected out
979 * of the range of a possible overwrite.
984 xfs_daddr_t head_blk,
985 xfs_daddr_t *tail_blk,
988 struct xlog_rec_header *thead;
990 xfs_daddr_t first_bad;
993 xfs_daddr_t tmp_tail;
994 xfs_daddr_t orig_tail = *tail_blk;
996 buffer = xlog_alloc_buffer(log, 1);
1001 * Make sure the tail points to a record (returns positive count on
1004 error = xlog_seek_logrec_hdr(log, head_blk, *tail_blk, 1, buffer,
1005 &tmp_tail, &thead, &wrapped);
1008 if (*tail_blk != tmp_tail)
1009 *tail_blk = tmp_tail;
1012 * Run a CRC check from the tail to the head. We can't just check
1013 * MAX_ICLOGS records past the tail because the tail may point to stale
1014 * blocks cleared during the search for the head/tail. These blocks are
1015 * overwritten with zero-length records and thus record count is not a
1016 * reliable indicator of the iclog state before a crash.
1019 error = xlog_do_recovery_pass(log, head_blk, *tail_blk,
1020 XLOG_RECOVER_CRCPASS, &first_bad);
1021 while ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) {
1025 * Is corruption within range of the head? If so, retry from
1026 * the next record. Otherwise return an error.
1028 tail_distance = xlog_tail_distance(log, head_blk, first_bad);
1029 if (tail_distance > BTOBB(XLOG_MAX_ICLOGS * hsize))
1032 /* skip to the next record; returns positive count on success */
1033 error = xlog_seek_logrec_hdr(log, head_blk, first_bad, 2,
1034 buffer, &tmp_tail, &thead, &wrapped);
1038 *tail_blk = tmp_tail;
1040 error = xlog_do_recovery_pass(log, head_blk, *tail_blk,
1041 XLOG_RECOVER_CRCPASS, &first_bad);
1044 if (!error && *tail_blk != orig_tail)
1046 "Tail block (0x%llx) overwrite detected. Updated to 0x%llx",
1047 orig_tail, *tail_blk);
1054 * Detect and trim torn writes from the head of the log.
1056 * Storage without sector atomicity guarantees can result in torn writes in the
1057 * log in the event of a crash. Our only means to detect this scenario is via
1058 * CRC verification. While we can't always be certain that CRC verification
1059 * failure is due to a torn write vs. an unrelated corruption, we do know that
1060 * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at
1061 * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of
1062 * the log and treat failures in this range as torn writes as a matter of
1063 * policy. In the event of CRC failure, the head is walked back to the last good
1064 * record in the log and the tail is updated from that record and verified.
1069 xfs_daddr_t *head_blk, /* in/out: unverified head */
1070 xfs_daddr_t *tail_blk, /* out: tail block */
1072 xfs_daddr_t *rhead_blk, /* start blk of last record */
1073 struct xlog_rec_header **rhead, /* ptr to last record */
1074 bool *wrapped) /* last rec. wraps phys. log */
1076 struct xlog_rec_header *tmp_rhead;
1078 xfs_daddr_t first_bad;
1079 xfs_daddr_t tmp_rhead_blk;
1085 * Check the head of the log for torn writes. Search backwards from the
1086 * head until we hit the tail or the maximum number of log record I/Os
1087 * that could have been in flight at one time. Use a temporary buffer so
1088 * we don't trash the rhead/buffer pointers from the caller.
1090 tmp_buffer = xlog_alloc_buffer(log, 1);
1093 error = xlog_rseek_logrec_hdr(log, *head_blk, *tail_blk,
1094 XLOG_MAX_ICLOGS, tmp_buffer,
1095 &tmp_rhead_blk, &tmp_rhead, &tmp_wrapped);
1096 kmem_free(tmp_buffer);
1101 * Now run a CRC verification pass over the records starting at the
1102 * block found above to the current head. If a CRC failure occurs, the
1103 * log block of the first bad record is saved in first_bad.
1105 error = xlog_do_recovery_pass(log, *head_blk, tmp_rhead_blk,
1106 XLOG_RECOVER_CRCPASS, &first_bad);
1107 if ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) {
1109 * We've hit a potential torn write. Reset the error and warn
1114 "Torn write (CRC failure) detected at log block 0x%llx. Truncating head block from 0x%llx.",
1115 first_bad, *head_blk);
1118 * Get the header block and buffer pointer for the last good
1119 * record before the bad record.
1121 * Note that xlog_find_tail() clears the blocks at the new head
1122 * (i.e., the records with invalid CRC) if the cycle number
1123 * matches the the current cycle.
1125 found = xlog_rseek_logrec_hdr(log, first_bad, *tail_blk, 1,
1126 buffer, rhead_blk, rhead, wrapped);
1129 if (found == 0) /* XXX: right thing to do here? */
1133 * Reset the head block to the starting block of the first bad
1134 * log record and set the tail block based on the last good
1137 * Bail out if the updated head/tail match as this indicates
1138 * possible corruption outside of the acceptable
1139 * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair...
1141 *head_blk = first_bad;
1142 *tail_blk = BLOCK_LSN(be64_to_cpu((*rhead)->h_tail_lsn));
1143 if (*head_blk == *tail_blk) {
1151 return xlog_verify_tail(log, *head_blk, tail_blk,
1152 be32_to_cpu((*rhead)->h_size));
1156 * We need to make sure we handle log wrapping properly, so we can't use the
1157 * calculated logbno directly. Make sure it wraps to the correct bno inside the
1160 * The log is limited to 32 bit sizes, so we use the appropriate modulus
1161 * operation here and cast it back to a 64 bit daddr on return.
1163 static inline xfs_daddr_t
1170 div_s64_rem(bno, log->l_logBBsize, &mod);
1175 * Check whether the head of the log points to an unmount record. In other
1176 * words, determine whether the log is clean. If so, update the in-core state
1180 xlog_check_unmount_rec(
1182 xfs_daddr_t *head_blk,
1183 xfs_daddr_t *tail_blk,
1184 struct xlog_rec_header *rhead,
1185 xfs_daddr_t rhead_blk,
1189 struct xlog_op_header *op_head;
1190 xfs_daddr_t umount_data_blk;
1191 xfs_daddr_t after_umount_blk;
1199 * Look for unmount record. If we find it, then we know there was a
1200 * clean unmount. Since 'i' could be the last block in the physical
1201 * log, we convert to a log block before comparing to the head_blk.
1203 * Save the current tail lsn to use to pass to xlog_clear_stale_blocks()
1204 * below. We won't want to clear the unmount record if there is one, so
1205 * we pass the lsn of the unmount record rather than the block after it.
1207 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1208 int h_size = be32_to_cpu(rhead->h_size);
1209 int h_version = be32_to_cpu(rhead->h_version);
1211 if ((h_version & XLOG_VERSION_2) &&
1212 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
1213 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
1214 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1223 after_umount_blk = xlog_wrap_logbno(log,
1224 rhead_blk + hblks + BTOBB(be32_to_cpu(rhead->h_len)));
1226 if (*head_blk == after_umount_blk &&
1227 be32_to_cpu(rhead->h_num_logops) == 1) {
1228 umount_data_blk = xlog_wrap_logbno(log, rhead_blk + hblks);
1229 error = xlog_bread(log, umount_data_blk, 1, buffer, &offset);
1233 op_head = (struct xlog_op_header *)offset;
1234 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1236 * Set tail and last sync so that newly written log
1237 * records will point recovery to after the current
1240 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1241 log->l_curr_cycle, after_umount_blk);
1242 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1243 log->l_curr_cycle, after_umount_blk);
1244 *tail_blk = after_umount_blk;
1256 xfs_daddr_t head_blk,
1257 struct xlog_rec_header *rhead,
1258 xfs_daddr_t rhead_blk,
1262 * Reset log values according to the state of the log when we
1263 * crashed. In the case where head_blk == 0, we bump curr_cycle
1264 * one because the next write starts a new cycle rather than
1265 * continuing the cycle of the last good log record. At this
1266 * point we have guaranteed that all partial log records have been
1267 * accounted for. Therefore, we know that the last good log record
1268 * written was complete and ended exactly on the end boundary
1269 * of the physical log.
1271 log->l_prev_block = rhead_blk;
1272 log->l_curr_block = (int)head_blk;
1273 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
1275 log->l_curr_cycle++;
1276 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
1277 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
1278 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
1279 BBTOB(log->l_curr_block));
1280 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
1281 BBTOB(log->l_curr_block));
1285 * Find the sync block number or the tail of the log.
1287 * This will be the block number of the last record to have its
1288 * associated buffers synced to disk. Every log record header has
1289 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
1290 * to get a sync block number. The only concern is to figure out which
1291 * log record header to believe.
1293 * The following algorithm uses the log record header with the largest
1294 * lsn. The entire log record does not need to be valid. We only care
1295 * that the header is valid.
1297 * We could speed up search by using current head_blk buffer, but it is not
1303 xfs_daddr_t *head_blk,
1304 xfs_daddr_t *tail_blk)
1306 xlog_rec_header_t *rhead;
1307 char *offset = NULL;
1310 xfs_daddr_t rhead_blk;
1312 bool wrapped = false;
1316 * Find previous log record
1318 if ((error = xlog_find_head(log, head_blk)))
1320 ASSERT(*head_blk < INT_MAX);
1322 buffer = xlog_alloc_buffer(log, 1);
1325 if (*head_blk == 0) { /* special case */
1326 error = xlog_bread(log, 0, 1, buffer, &offset);
1330 if (xlog_get_cycle(offset) == 0) {
1332 /* leave all other log inited values alone */
1338 * Search backwards through the log looking for the log record header
1339 * block. This wraps all the way back around to the head so something is
1340 * seriously wrong if we can't find it.
1342 error = xlog_rseek_logrec_hdr(log, *head_blk, *head_blk, 1, buffer,
1343 &rhead_blk, &rhead, &wrapped);
1347 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
1348 error = -EFSCORRUPTED;
1351 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
1354 * Set the log state based on the current head record.
1356 xlog_set_state(log, *head_blk, rhead, rhead_blk, wrapped);
1357 tail_lsn = atomic64_read(&log->l_tail_lsn);
1360 * Look for an unmount record at the head of the log. This sets the log
1361 * state to determine whether recovery is necessary.
1363 error = xlog_check_unmount_rec(log, head_blk, tail_blk, rhead,
1364 rhead_blk, buffer, &clean);
1369 * Verify the log head if the log is not clean (e.g., we have anything
1370 * but an unmount record at the head). This uses CRC verification to
1371 * detect and trim torn writes. If discovered, CRC failures are
1372 * considered torn writes and the log head is trimmed accordingly.
1374 * Note that we can only run CRC verification when the log is dirty
1375 * because there's no guarantee that the log data behind an unmount
1376 * record is compatible with the current architecture.
1379 xfs_daddr_t orig_head = *head_blk;
1381 error = xlog_verify_head(log, head_blk, tail_blk, buffer,
1382 &rhead_blk, &rhead, &wrapped);
1386 /* update in-core state again if the head changed */
1387 if (*head_blk != orig_head) {
1388 xlog_set_state(log, *head_blk, rhead, rhead_blk,
1390 tail_lsn = atomic64_read(&log->l_tail_lsn);
1391 error = xlog_check_unmount_rec(log, head_blk, tail_blk,
1392 rhead, rhead_blk, buffer,
1400 * Note that the unmount was clean. If the unmount was not clean, we
1401 * need to know this to rebuild the superblock counters from the perag
1402 * headers if we have a filesystem using non-persistent counters.
1405 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1408 * Make sure that there are no blocks in front of the head
1409 * with the same cycle number as the head. This can happen
1410 * because we allow multiple outstanding log writes concurrently,
1411 * and the later writes might make it out before earlier ones.
1413 * We use the lsn from before modifying it so that we'll never
1414 * overwrite the unmount record after a clean unmount.
1416 * Do this only if we are going to recover the filesystem
1418 * NOTE: This used to say "if (!readonly)"
1419 * However on Linux, we can & do recover a read-only filesystem.
1420 * We only skip recovery if NORECOVERY is specified on mount,
1421 * in which case we would not be here.
1423 * But... if the -device- itself is readonly, just skip this.
1424 * We can't recover this device anyway, so it won't matter.
1426 if (!xfs_readonly_buftarg(log->l_targ))
1427 error = xlog_clear_stale_blocks(log, tail_lsn);
1433 xfs_warn(log->l_mp, "failed to locate log tail");
1438 * Is the log zeroed at all?
1440 * The last binary search should be changed to perform an X block read
1441 * once X becomes small enough. You can then search linearly through
1442 * the X blocks. This will cut down on the number of reads we need to do.
1444 * If the log is partially zeroed, this routine will pass back the blkno
1445 * of the first block with cycle number 0. It won't have a complete LR
1449 * 0 => the log is completely written to
1450 * 1 => use *blk_no as the first block of the log
1451 * <0 => error has occurred
1456 xfs_daddr_t *blk_no)
1460 uint first_cycle, last_cycle;
1461 xfs_daddr_t new_blk, last_blk, start_blk;
1462 xfs_daddr_t num_scan_bblks;
1463 int error, log_bbnum = log->l_logBBsize;
1467 /* check totally zeroed log */
1468 buffer = xlog_alloc_buffer(log, 1);
1471 error = xlog_bread(log, 0, 1, buffer, &offset);
1473 goto out_free_buffer;
1475 first_cycle = xlog_get_cycle(offset);
1476 if (first_cycle == 0) { /* completely zeroed log */
1482 /* check partially zeroed log */
1483 error = xlog_bread(log, log_bbnum-1, 1, buffer, &offset);
1485 goto out_free_buffer;
1487 last_cycle = xlog_get_cycle(offset);
1488 if (last_cycle != 0) { /* log completely written to */
1493 /* we have a partially zeroed log */
1494 last_blk = log_bbnum-1;
1495 error = xlog_find_cycle_start(log, buffer, 0, &last_blk, 0);
1497 goto out_free_buffer;
1500 * Validate the answer. Because there is no way to guarantee that
1501 * the entire log is made up of log records which are the same size,
1502 * we scan over the defined maximum blocks. At this point, the maximum
1503 * is not chosen to mean anything special. XXXmiken
1505 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1506 ASSERT(num_scan_bblks <= INT_MAX);
1508 if (last_blk < num_scan_bblks)
1509 num_scan_bblks = last_blk;
1510 start_blk = last_blk - num_scan_bblks;
1513 * We search for any instances of cycle number 0 that occur before
1514 * our current estimate of the head. What we're trying to detect is
1515 * 1 ... | 0 | 1 | 0...
1516 * ^ binary search ends here
1518 if ((error = xlog_find_verify_cycle(log, start_blk,
1519 (int)num_scan_bblks, 0, &new_blk)))
1520 goto out_free_buffer;
1525 * Potentially backup over partial log record write. We don't need
1526 * to search the end of the log because we know it is zero.
1528 error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0);
1532 goto out_free_buffer;
1543 * These are simple subroutines used by xlog_clear_stale_blocks() below
1544 * to initialize a buffer full of empty log record headers and write
1545 * them into the log.
1556 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1558 memset(buf, 0, BBSIZE);
1559 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1560 recp->h_cycle = cpu_to_be32(cycle);
1561 recp->h_version = cpu_to_be32(
1562 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1563 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1564 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1565 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1566 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1570 xlog_write_log_records(
1581 int sectbb = log->l_sectBBsize;
1582 int end_block = start_block + blocks;
1588 * Greedily allocate a buffer big enough to handle the full
1589 * range of basic blocks to be written. If that fails, try
1590 * a smaller size. We need to be able to write at least a
1591 * log sector, or we're out of luck.
1593 bufblks = 1 << ffs(blocks);
1594 while (bufblks > log->l_logBBsize)
1596 while (!(buffer = xlog_alloc_buffer(log, bufblks))) {
1598 if (bufblks < sectbb)
1602 /* We may need to do a read at the start to fill in part of
1603 * the buffer in the starting sector not covered by the first
1606 balign = round_down(start_block, sectbb);
1607 if (balign != start_block) {
1608 error = xlog_bread_noalign(log, start_block, 1, buffer);
1610 goto out_free_buffer;
1612 j = start_block - balign;
1615 for (i = start_block; i < end_block; i += bufblks) {
1616 int bcount, endcount;
1618 bcount = min(bufblks, end_block - start_block);
1619 endcount = bcount - j;
1621 /* We may need to do a read at the end to fill in part of
1622 * the buffer in the final sector not covered by the write.
1623 * If this is the same sector as the above read, skip it.
1625 ealign = round_down(end_block, sectbb);
1626 if (j == 0 && (start_block + endcount > ealign)) {
1627 error = xlog_bread_noalign(log, ealign, sectbb,
1628 buffer + BBTOB(ealign - start_block));
1634 offset = buffer + xlog_align(log, start_block);
1635 for (; j < endcount; j++) {
1636 xlog_add_record(log, offset, cycle, i+j,
1637 tail_cycle, tail_block);
1640 error = xlog_bwrite(log, start_block, endcount, buffer);
1643 start_block += endcount;
1653 * This routine is called to blow away any incomplete log writes out
1654 * in front of the log head. We do this so that we won't become confused
1655 * if we come up, write only a little bit more, and then crash again.
1656 * If we leave the partial log records out there, this situation could
1657 * cause us to think those partial writes are valid blocks since they
1658 * have the current cycle number. We get rid of them by overwriting them
1659 * with empty log records with the old cycle number rather than the
1662 * The tail lsn is passed in rather than taken from
1663 * the log so that we will not write over the unmount record after a
1664 * clean unmount in a 512 block log. Doing so would leave the log without
1665 * any valid log records in it until a new one was written. If we crashed
1666 * during that time we would not be able to recover.
1669 xlog_clear_stale_blocks(
1673 int tail_cycle, head_cycle;
1674 int tail_block, head_block;
1675 int tail_distance, max_distance;
1679 tail_cycle = CYCLE_LSN(tail_lsn);
1680 tail_block = BLOCK_LSN(tail_lsn);
1681 head_cycle = log->l_curr_cycle;
1682 head_block = log->l_curr_block;
1685 * Figure out the distance between the new head of the log
1686 * and the tail. We want to write over any blocks beyond the
1687 * head that we may have written just before the crash, but
1688 * we don't want to overwrite the tail of the log.
1690 if (head_cycle == tail_cycle) {
1692 * The tail is behind the head in the physical log,
1693 * so the distance from the head to the tail is the
1694 * distance from the head to the end of the log plus
1695 * the distance from the beginning of the log to the
1698 if (XFS_IS_CORRUPT(log->l_mp,
1699 head_block < tail_block ||
1700 head_block >= log->l_logBBsize))
1701 return -EFSCORRUPTED;
1702 tail_distance = tail_block + (log->l_logBBsize - head_block);
1705 * The head is behind the tail in the physical log,
1706 * so the distance from the head to the tail is just
1707 * the tail block minus the head block.
1709 if (XFS_IS_CORRUPT(log->l_mp,
1710 head_block >= tail_block ||
1711 head_cycle != tail_cycle + 1))
1712 return -EFSCORRUPTED;
1713 tail_distance = tail_block - head_block;
1717 * If the head is right up against the tail, we can't clear
1720 if (tail_distance <= 0) {
1721 ASSERT(tail_distance == 0);
1725 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1727 * Take the smaller of the maximum amount of outstanding I/O
1728 * we could have and the distance to the tail to clear out.
1729 * We take the smaller so that we don't overwrite the tail and
1730 * we don't waste all day writing from the head to the tail
1733 max_distance = min(max_distance, tail_distance);
1735 if ((head_block + max_distance) <= log->l_logBBsize) {
1737 * We can stomp all the blocks we need to without
1738 * wrapping around the end of the log. Just do it
1739 * in a single write. Use the cycle number of the
1740 * current cycle minus one so that the log will look like:
1743 error = xlog_write_log_records(log, (head_cycle - 1),
1744 head_block, max_distance, tail_cycle,
1750 * We need to wrap around the end of the physical log in
1751 * order to clear all the blocks. Do it in two separate
1752 * I/Os. The first write should be from the head to the
1753 * end of the physical log, and it should use the current
1754 * cycle number minus one just like above.
1756 distance = log->l_logBBsize - head_block;
1757 error = xlog_write_log_records(log, (head_cycle - 1),
1758 head_block, distance, tail_cycle,
1765 * Now write the blocks at the start of the physical log.
1766 * This writes the remainder of the blocks we want to clear.
1767 * It uses the current cycle number since we're now on the
1768 * same cycle as the head so that we get:
1769 * n ... n ... | n - 1 ...
1770 * ^^^^^ blocks we're writing
1772 distance = max_distance - (log->l_logBBsize - head_block);
1773 error = xlog_write_log_records(log, head_cycle, 0, distance,
1774 tail_cycle, tail_block);
1782 /******************************************************************************
1784 * Log recover routines
1786 ******************************************************************************
1790 * Sort the log items in the transaction.
1792 * The ordering constraints are defined by the inode allocation and unlink
1793 * behaviour. The rules are:
1795 * 1. Every item is only logged once in a given transaction. Hence it
1796 * represents the last logged state of the item. Hence ordering is
1797 * dependent on the order in which operations need to be performed so
1798 * required initial conditions are always met.
1800 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1801 * there's nothing to replay from them so we can simply cull them
1802 * from the transaction. However, we can't do that until after we've
1803 * replayed all the other items because they may be dependent on the
1804 * cancelled buffer and replaying the cancelled buffer can remove it
1805 * form the cancelled buffer table. Hence they have tobe done last.
1807 * 3. Inode allocation buffers must be replayed before inode items that
1808 * read the buffer and replay changes into it. For filesystems using the
1809 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1810 * treated the same as inode allocation buffers as they create and
1811 * initialise the buffers directly.
1813 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1814 * This ensures that inodes are completely flushed to the inode buffer
1815 * in a "free" state before we remove the unlinked inode list pointer.
1817 * Hence the ordering needs to be inode allocation buffers first, inode items
1818 * second, inode unlink buffers third and cancelled buffers last.
1820 * But there's a problem with that - we can't tell an inode allocation buffer
1821 * apart from a regular buffer, so we can't separate them. We can, however,
1822 * tell an inode unlink buffer from the others, and so we can separate them out
1823 * from all the other buffers and move them to last.
1825 * Hence, 4 lists, in order from head to tail:
1826 * - buffer_list for all buffers except cancelled/inode unlink buffers
1827 * - item_list for all non-buffer items
1828 * - inode_buffer_list for inode unlink buffers
1829 * - cancel_list for the cancelled buffers
1831 * Note that we add objects to the tail of the lists so that first-to-last
1832 * ordering is preserved within the lists. Adding objects to the head of the
1833 * list means when we traverse from the head we walk them in last-to-first
1834 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1835 * but for all other items there may be specific ordering that we need to
1839 xlog_recover_reorder_trans(
1841 struct xlog_recover *trans,
1844 xlog_recover_item_t *item, *n;
1846 LIST_HEAD(sort_list);
1847 LIST_HEAD(cancel_list);
1848 LIST_HEAD(buffer_list);
1849 LIST_HEAD(inode_buffer_list);
1850 LIST_HEAD(inode_list);
1852 list_splice_init(&trans->r_itemq, &sort_list);
1853 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1854 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1856 switch (ITEM_TYPE(item)) {
1857 case XFS_LI_ICREATE:
1858 list_move_tail(&item->ri_list, &buffer_list);
1861 if (buf_f->blf_flags & XFS_BLF_CANCEL) {
1862 trace_xfs_log_recover_item_reorder_head(log,
1864 list_move(&item->ri_list, &cancel_list);
1867 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1868 list_move(&item->ri_list, &inode_buffer_list);
1871 list_move_tail(&item->ri_list, &buffer_list);
1875 case XFS_LI_QUOTAOFF:
1884 trace_xfs_log_recover_item_reorder_tail(log,
1886 list_move_tail(&item->ri_list, &inode_list);
1890 "%s: unrecognized type of log operation",
1894 * return the remaining items back to the transaction
1895 * item list so they can be freed in caller.
1897 if (!list_empty(&sort_list))
1898 list_splice_init(&sort_list, &trans->r_itemq);
1904 ASSERT(list_empty(&sort_list));
1905 if (!list_empty(&buffer_list))
1906 list_splice(&buffer_list, &trans->r_itemq);
1907 if (!list_empty(&inode_list))
1908 list_splice_tail(&inode_list, &trans->r_itemq);
1909 if (!list_empty(&inode_buffer_list))
1910 list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1911 if (!list_empty(&cancel_list))
1912 list_splice_tail(&cancel_list, &trans->r_itemq);
1917 * Build up the table of buf cancel records so that we don't replay
1918 * cancelled data in the second pass. For buffer records that are
1919 * not cancel records, there is nothing to do here so we just return.
1921 * If we get a cancel record which is already in the table, this indicates
1922 * that the buffer was cancelled multiple times. In order to ensure
1923 * that during pass 2 we keep the record in the table until we reach its
1924 * last occurrence in the log, we keep a reference count in the cancel
1925 * record in the table to tell us how many times we expect to see this
1926 * record during the second pass.
1929 xlog_recover_buffer_pass1(
1931 struct xlog_recover_item *item)
1933 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1934 struct list_head *bucket;
1935 struct xfs_buf_cancel *bcp;
1938 * If this isn't a cancel buffer item, then just return.
1940 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1941 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
1946 * Insert an xfs_buf_cancel record into the hash table of them.
1947 * If there is already an identical record, bump its reference count.
1949 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1950 list_for_each_entry(bcp, bucket, bc_list) {
1951 if (bcp->bc_blkno == buf_f->blf_blkno &&
1952 bcp->bc_len == buf_f->blf_len) {
1954 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
1959 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), 0);
1960 bcp->bc_blkno = buf_f->blf_blkno;
1961 bcp->bc_len = buf_f->blf_len;
1962 bcp->bc_refcount = 1;
1963 list_add_tail(&bcp->bc_list, bucket);
1965 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
1970 * Check to see whether the buffer being recovered has a corresponding
1971 * entry in the buffer cancel record table. If it is, return the cancel
1972 * buffer structure to the caller.
1974 STATIC struct xfs_buf_cancel *
1975 xlog_peek_buffer_cancelled(
1979 unsigned short flags)
1981 struct list_head *bucket;
1982 struct xfs_buf_cancel *bcp;
1984 if (!log->l_buf_cancel_table) {
1985 /* empty table means no cancelled buffers in the log */
1986 ASSERT(!(flags & XFS_BLF_CANCEL));
1990 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1991 list_for_each_entry(bcp, bucket, bc_list) {
1992 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
1997 * We didn't find a corresponding entry in the table, so return 0 so
1998 * that the buffer is NOT cancelled.
2000 ASSERT(!(flags & XFS_BLF_CANCEL));
2005 * If the buffer is being cancelled then return 1 so that it will be cancelled,
2006 * otherwise return 0. If the buffer is actually a buffer cancel item
2007 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
2008 * table and remove it from the table if this is the last reference.
2010 * We remove the cancel record from the table when we encounter its last
2011 * occurrence in the log so that if the same buffer is re-used again after its
2012 * last cancellation we actually replay the changes made at that point.
2015 xlog_check_buffer_cancelled(
2019 unsigned short flags)
2021 struct xfs_buf_cancel *bcp;
2023 bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags);
2028 * We've go a match, so return 1 so that the recovery of this buffer
2029 * is cancelled. If this buffer is actually a buffer cancel log
2030 * item, then decrement the refcount on the one in the table and
2031 * remove it if this is the last reference.
2033 if (flags & XFS_BLF_CANCEL) {
2034 if (--bcp->bc_refcount == 0) {
2035 list_del(&bcp->bc_list);
2043 * Perform recovery for a buffer full of inodes. In these buffers, the only
2044 * data which should be recovered is that which corresponds to the
2045 * di_next_unlinked pointers in the on disk inode structures. The rest of the
2046 * data for the inodes is always logged through the inodes themselves rather
2047 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
2049 * The only time when buffers full of inodes are fully recovered is when the
2050 * buffer is full of newly allocated inodes. In this case the buffer will
2051 * not be marked as an inode buffer and so will be sent to
2052 * xlog_recover_do_reg_buffer() below during recovery.
2055 xlog_recover_do_inode_buffer(
2056 struct xfs_mount *mp,
2057 xlog_recover_item_t *item,
2059 xfs_buf_log_format_t *buf_f)
2065 int reg_buf_offset = 0;
2066 int reg_buf_bytes = 0;
2067 int next_unlinked_offset;
2069 xfs_agino_t *logged_nextp;
2070 xfs_agino_t *buffer_nextp;
2072 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
2075 * Post recovery validation only works properly on CRC enabled
2078 if (xfs_sb_version_hascrc(&mp->m_sb))
2079 bp->b_ops = &xfs_inode_buf_ops;
2081 inodes_per_buf = BBTOB(bp->b_length) >> mp->m_sb.sb_inodelog;
2082 for (i = 0; i < inodes_per_buf; i++) {
2083 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
2084 offsetof(xfs_dinode_t, di_next_unlinked);
2086 while (next_unlinked_offset >=
2087 (reg_buf_offset + reg_buf_bytes)) {
2089 * The next di_next_unlinked field is beyond
2090 * the current logged region. Find the next
2091 * logged region that contains or is beyond
2092 * the current di_next_unlinked field.
2095 bit = xfs_next_bit(buf_f->blf_data_map,
2096 buf_f->blf_map_size, bit);
2099 * If there are no more logged regions in the
2100 * buffer, then we're done.
2105 nbits = xfs_contig_bits(buf_f->blf_data_map,
2106 buf_f->blf_map_size, bit);
2108 reg_buf_offset = bit << XFS_BLF_SHIFT;
2109 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
2114 * If the current logged region starts after the current
2115 * di_next_unlinked field, then move on to the next
2116 * di_next_unlinked field.
2118 if (next_unlinked_offset < reg_buf_offset)
2121 ASSERT(item->ri_buf[item_index].i_addr != NULL);
2122 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
2123 ASSERT((reg_buf_offset + reg_buf_bytes) <= BBTOB(bp->b_length));
2126 * The current logged region contains a copy of the
2127 * current di_next_unlinked field. Extract its value
2128 * and copy it to the buffer copy.
2130 logged_nextp = item->ri_buf[item_index].i_addr +
2131 next_unlinked_offset - reg_buf_offset;
2132 if (XFS_IS_CORRUPT(mp, *logged_nextp == 0)) {
2134 "Bad inode buffer log record (ptr = "PTR_FMT", bp = "PTR_FMT"). "
2135 "Trying to replay bad (0) inode di_next_unlinked field.",
2137 return -EFSCORRUPTED;
2140 buffer_nextp = xfs_buf_offset(bp, next_unlinked_offset);
2141 *buffer_nextp = *logged_nextp;
2144 * If necessary, recalculate the CRC in the on-disk inode. We
2145 * have to leave the inode in a consistent state for whoever
2148 xfs_dinode_calc_crc(mp,
2149 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
2157 * V5 filesystems know the age of the buffer on disk being recovered. We can
2158 * have newer objects on disk than we are replaying, and so for these cases we
2159 * don't want to replay the current change as that will make the buffer contents
2160 * temporarily invalid on disk.
2162 * The magic number might not match the buffer type we are going to recover
2163 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
2164 * extract the LSN of the existing object in the buffer based on it's current
2165 * magic number. If we don't recognise the magic number in the buffer, then
2166 * return a LSN of -1 so that the caller knows it was an unrecognised block and
2167 * so can recover the buffer.
2169 * Note: we cannot rely solely on magic number matches to determine that the
2170 * buffer has a valid LSN - we also need to verify that it belongs to this
2171 * filesystem, so we need to extract the object's LSN and compare it to that
2172 * which we read from the superblock. If the UUIDs don't match, then we've got a
2173 * stale metadata block from an old filesystem instance that we need to recover
2177 xlog_recover_get_buf_lsn(
2178 struct xfs_mount *mp,
2184 void *blk = bp->b_addr;
2188 /* v4 filesystems always recover immediately */
2189 if (!xfs_sb_version_hascrc(&mp->m_sb))
2190 goto recover_immediately;
2192 magic32 = be32_to_cpu(*(__be32 *)blk);
2194 case XFS_ABTB_CRC_MAGIC:
2195 case XFS_ABTC_CRC_MAGIC:
2196 case XFS_ABTB_MAGIC:
2197 case XFS_ABTC_MAGIC:
2198 case XFS_RMAP_CRC_MAGIC:
2199 case XFS_REFC_CRC_MAGIC:
2200 case XFS_IBT_CRC_MAGIC:
2201 case XFS_IBT_MAGIC: {
2202 struct xfs_btree_block *btb = blk;
2204 lsn = be64_to_cpu(btb->bb_u.s.bb_lsn);
2205 uuid = &btb->bb_u.s.bb_uuid;
2208 case XFS_BMAP_CRC_MAGIC:
2209 case XFS_BMAP_MAGIC: {
2210 struct xfs_btree_block *btb = blk;
2212 lsn = be64_to_cpu(btb->bb_u.l.bb_lsn);
2213 uuid = &btb->bb_u.l.bb_uuid;
2217 lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn);
2218 uuid = &((struct xfs_agf *)blk)->agf_uuid;
2220 case XFS_AGFL_MAGIC:
2221 lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn);
2222 uuid = &((struct xfs_agfl *)blk)->agfl_uuid;
2225 lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn);
2226 uuid = &((struct xfs_agi *)blk)->agi_uuid;
2228 case XFS_SYMLINK_MAGIC:
2229 lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn);
2230 uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid;
2232 case XFS_DIR3_BLOCK_MAGIC:
2233 case XFS_DIR3_DATA_MAGIC:
2234 case XFS_DIR3_FREE_MAGIC:
2235 lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn);
2236 uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid;
2238 case XFS_ATTR3_RMT_MAGIC:
2240 * Remote attr blocks are written synchronously, rather than
2241 * being logged. That means they do not contain a valid LSN
2242 * (i.e. transactionally ordered) in them, and hence any time we
2243 * see a buffer to replay over the top of a remote attribute
2244 * block we should simply do so.
2246 goto recover_immediately;
2249 * superblock uuids are magic. We may or may not have a
2250 * sb_meta_uuid on disk, but it will be set in the in-core
2251 * superblock. We set the uuid pointer for verification
2252 * according to the superblock feature mask to ensure we check
2253 * the relevant UUID in the superblock.
2255 lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn);
2256 if (xfs_sb_version_hasmetauuid(&mp->m_sb))
2257 uuid = &((struct xfs_dsb *)blk)->sb_meta_uuid;
2259 uuid = &((struct xfs_dsb *)blk)->sb_uuid;
2265 if (lsn != (xfs_lsn_t)-1) {
2266 if (!uuid_equal(&mp->m_sb.sb_meta_uuid, uuid))
2267 goto recover_immediately;
2271 magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic);
2273 case XFS_DIR3_LEAF1_MAGIC:
2274 case XFS_DIR3_LEAFN_MAGIC:
2275 case XFS_DA3_NODE_MAGIC:
2276 lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn);
2277 uuid = &((struct xfs_da3_blkinfo *)blk)->uuid;
2283 if (lsn != (xfs_lsn_t)-1) {
2284 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
2285 goto recover_immediately;
2290 * We do individual object checks on dquot and inode buffers as they
2291 * have their own individual LSN records. Also, we could have a stale
2292 * buffer here, so we have to at least recognise these buffer types.
2294 * A notd complexity here is inode unlinked list processing - it logs
2295 * the inode directly in the buffer, but we don't know which inodes have
2296 * been modified, and there is no global buffer LSN. Hence we need to
2297 * recover all inode buffer types immediately. This problem will be
2298 * fixed by logical logging of the unlinked list modifications.
2300 magic16 = be16_to_cpu(*(__be16 *)blk);
2302 case XFS_DQUOT_MAGIC:
2303 case XFS_DINODE_MAGIC:
2304 goto recover_immediately;
2309 /* unknown buffer contents, recover immediately */
2311 recover_immediately:
2312 return (xfs_lsn_t)-1;
2317 * Validate the recovered buffer is of the correct type and attach the
2318 * appropriate buffer operations to them for writeback. Magic numbers are in a
2320 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2321 * the first 32 bits of the buffer (most blocks),
2322 * inside a struct xfs_da_blkinfo at the start of the buffer.
2325 xlog_recover_validate_buf_type(
2326 struct xfs_mount *mp,
2328 xfs_buf_log_format_t *buf_f,
2329 xfs_lsn_t current_lsn)
2331 struct xfs_da_blkinfo *info = bp->b_addr;
2335 char *warnmsg = NULL;
2338 * We can only do post recovery validation on items on CRC enabled
2339 * fielsystems as we need to know when the buffer was written to be able
2340 * to determine if we should have replayed the item. If we replay old
2341 * metadata over a newer buffer, then it will enter a temporarily
2342 * inconsistent state resulting in verification failures. Hence for now
2343 * just avoid the verification stage for non-crc filesystems
2345 if (!xfs_sb_version_hascrc(&mp->m_sb))
2348 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
2349 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
2350 magicda = be16_to_cpu(info->magic);
2351 switch (xfs_blft_from_flags(buf_f)) {
2352 case XFS_BLFT_BTREE_BUF:
2354 case XFS_ABTB_CRC_MAGIC:
2355 case XFS_ABTB_MAGIC:
2356 bp->b_ops = &xfs_bnobt_buf_ops;
2358 case XFS_ABTC_CRC_MAGIC:
2359 case XFS_ABTC_MAGIC:
2360 bp->b_ops = &xfs_cntbt_buf_ops;
2362 case XFS_IBT_CRC_MAGIC:
2364 bp->b_ops = &xfs_inobt_buf_ops;
2366 case XFS_FIBT_CRC_MAGIC:
2367 case XFS_FIBT_MAGIC:
2368 bp->b_ops = &xfs_finobt_buf_ops;
2370 case XFS_BMAP_CRC_MAGIC:
2371 case XFS_BMAP_MAGIC:
2372 bp->b_ops = &xfs_bmbt_buf_ops;
2374 case XFS_RMAP_CRC_MAGIC:
2375 bp->b_ops = &xfs_rmapbt_buf_ops;
2377 case XFS_REFC_CRC_MAGIC:
2378 bp->b_ops = &xfs_refcountbt_buf_ops;
2381 warnmsg = "Bad btree block magic!";
2385 case XFS_BLFT_AGF_BUF:
2386 if (magic32 != XFS_AGF_MAGIC) {
2387 warnmsg = "Bad AGF block magic!";
2390 bp->b_ops = &xfs_agf_buf_ops;
2392 case XFS_BLFT_AGFL_BUF:
2393 if (magic32 != XFS_AGFL_MAGIC) {
2394 warnmsg = "Bad AGFL block magic!";
2397 bp->b_ops = &xfs_agfl_buf_ops;
2399 case XFS_BLFT_AGI_BUF:
2400 if (magic32 != XFS_AGI_MAGIC) {
2401 warnmsg = "Bad AGI block magic!";
2404 bp->b_ops = &xfs_agi_buf_ops;
2406 case XFS_BLFT_UDQUOT_BUF:
2407 case XFS_BLFT_PDQUOT_BUF:
2408 case XFS_BLFT_GDQUOT_BUF:
2409 #ifdef CONFIG_XFS_QUOTA
2410 if (magic16 != XFS_DQUOT_MAGIC) {
2411 warnmsg = "Bad DQUOT block magic!";
2414 bp->b_ops = &xfs_dquot_buf_ops;
2417 "Trying to recover dquots without QUOTA support built in!");
2421 case XFS_BLFT_DINO_BUF:
2422 if (magic16 != XFS_DINODE_MAGIC) {
2423 warnmsg = "Bad INODE block magic!";
2426 bp->b_ops = &xfs_inode_buf_ops;
2428 case XFS_BLFT_SYMLINK_BUF:
2429 if (magic32 != XFS_SYMLINK_MAGIC) {
2430 warnmsg = "Bad symlink block magic!";
2433 bp->b_ops = &xfs_symlink_buf_ops;
2435 case XFS_BLFT_DIR_BLOCK_BUF:
2436 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2437 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2438 warnmsg = "Bad dir block magic!";
2441 bp->b_ops = &xfs_dir3_block_buf_ops;
2443 case XFS_BLFT_DIR_DATA_BUF:
2444 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2445 magic32 != XFS_DIR3_DATA_MAGIC) {
2446 warnmsg = "Bad dir data magic!";
2449 bp->b_ops = &xfs_dir3_data_buf_ops;
2451 case XFS_BLFT_DIR_FREE_BUF:
2452 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2453 magic32 != XFS_DIR3_FREE_MAGIC) {
2454 warnmsg = "Bad dir3 free magic!";
2457 bp->b_ops = &xfs_dir3_free_buf_ops;
2459 case XFS_BLFT_DIR_LEAF1_BUF:
2460 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2461 magicda != XFS_DIR3_LEAF1_MAGIC) {
2462 warnmsg = "Bad dir leaf1 magic!";
2465 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2467 case XFS_BLFT_DIR_LEAFN_BUF:
2468 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2469 magicda != XFS_DIR3_LEAFN_MAGIC) {
2470 warnmsg = "Bad dir leafn magic!";
2473 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2475 case XFS_BLFT_DA_NODE_BUF:
2476 if (magicda != XFS_DA_NODE_MAGIC &&
2477 magicda != XFS_DA3_NODE_MAGIC) {
2478 warnmsg = "Bad da node magic!";
2481 bp->b_ops = &xfs_da3_node_buf_ops;
2483 case XFS_BLFT_ATTR_LEAF_BUF:
2484 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2485 magicda != XFS_ATTR3_LEAF_MAGIC) {
2486 warnmsg = "Bad attr leaf magic!";
2489 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2491 case XFS_BLFT_ATTR_RMT_BUF:
2492 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
2493 warnmsg = "Bad attr remote magic!";
2496 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2498 case XFS_BLFT_SB_BUF:
2499 if (magic32 != XFS_SB_MAGIC) {
2500 warnmsg = "Bad SB block magic!";
2503 bp->b_ops = &xfs_sb_buf_ops;
2505 #ifdef CONFIG_XFS_RT
2506 case XFS_BLFT_RTBITMAP_BUF:
2507 case XFS_BLFT_RTSUMMARY_BUF:
2508 /* no magic numbers for verification of RT buffers */
2509 bp->b_ops = &xfs_rtbuf_ops;
2511 #endif /* CONFIG_XFS_RT */
2513 xfs_warn(mp, "Unknown buffer type %d!",
2514 xfs_blft_from_flags(buf_f));
2519 * Nothing else to do in the case of a NULL current LSN as this means
2520 * the buffer is more recent than the change in the log and will be
2523 if (current_lsn == NULLCOMMITLSN)
2527 xfs_warn(mp, warnmsg);
2532 * We must update the metadata LSN of the buffer as it is written out to
2533 * ensure that older transactions never replay over this one and corrupt
2534 * the buffer. This can occur if log recovery is interrupted at some
2535 * point after the current transaction completes, at which point a
2536 * subsequent mount starts recovery from the beginning.
2538 * Write verifiers update the metadata LSN from log items attached to
2539 * the buffer. Therefore, initialize a bli purely to carry the LSN to
2540 * the verifier. We'll clean it up in our ->iodone() callback.
2543 struct xfs_buf_log_item *bip;
2545 ASSERT(!bp->b_iodone || bp->b_iodone == xlog_recover_iodone);
2546 bp->b_iodone = xlog_recover_iodone;
2547 xfs_buf_item_init(bp, mp);
2548 bip = bp->b_log_item;
2549 bip->bli_item.li_lsn = current_lsn;
2554 * Perform a 'normal' buffer recovery. Each logged region of the
2555 * buffer should be copied over the corresponding region in the
2556 * given buffer. The bitmap in the buf log format structure indicates
2557 * where to place the logged data.
2560 xlog_recover_do_reg_buffer(
2561 struct xfs_mount *mp,
2562 xlog_recover_item_t *item,
2564 xfs_buf_log_format_t *buf_f,
2565 xfs_lsn_t current_lsn)
2571 const size_t size_disk_dquot = sizeof(struct xfs_disk_dquot);
2573 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2576 i = 1; /* 0 is the buf format structure */
2578 bit = xfs_next_bit(buf_f->blf_data_map,
2579 buf_f->blf_map_size, bit);
2582 nbits = xfs_contig_bits(buf_f->blf_data_map,
2583 buf_f->blf_map_size, bit);
2585 ASSERT(item->ri_buf[i].i_addr != NULL);
2586 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
2587 ASSERT(BBTOB(bp->b_length) >=
2588 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
2591 * The dirty regions logged in the buffer, even though
2592 * contiguous, may span multiple chunks. This is because the
2593 * dirty region may span a physical page boundary in a buffer
2594 * and hence be split into two separate vectors for writing into
2595 * the log. Hence we need to trim nbits back to the length of
2596 * the current region being copied out of the log.
2598 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2599 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2602 * Do a sanity check if this is a dquot buffer. Just checking
2603 * the first dquot in the buffer should do. XXXThis is
2604 * probably a good thing to do for other buf types also.
2607 if (buf_f->blf_flags &
2608 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2609 if (item->ri_buf[i].i_addr == NULL) {
2611 "XFS: NULL dquot in %s.", __func__);
2614 if (item->ri_buf[i].i_len < size_disk_dquot) {
2616 "XFS: dquot too small (%d) in %s.",
2617 item->ri_buf[i].i_len, __func__);
2620 fa = xfs_dquot_verify(mp, item->ri_buf[i].i_addr,
2624 "dquot corrupt at %pS trying to replay into block 0x%llx",
2630 memcpy(xfs_buf_offset(bp,
2631 (uint)bit << XFS_BLF_SHIFT), /* dest */
2632 item->ri_buf[i].i_addr, /* source */
2633 nbits<<XFS_BLF_SHIFT); /* length */
2639 /* Shouldn't be any more regions */
2640 ASSERT(i == item->ri_total);
2642 xlog_recover_validate_buf_type(mp, bp, buf_f, current_lsn);
2646 * Perform a dquot buffer recovery.
2647 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2648 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2649 * Else, treat it as a regular buffer and do recovery.
2651 * Return false if the buffer was tossed and true if we recovered the buffer to
2652 * indicate to the caller if the buffer needs writing.
2655 xlog_recover_do_dquot_buffer(
2656 struct xfs_mount *mp,
2658 struct xlog_recover_item *item,
2660 struct xfs_buf_log_format *buf_f)
2664 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2667 * Filesystems are required to send in quota flags at mount time.
2673 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2674 type |= XFS_DQ_USER;
2675 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2676 type |= XFS_DQ_PROJ;
2677 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2678 type |= XFS_DQ_GROUP;
2680 * This type of quotas was turned off, so ignore this buffer
2682 if (log->l_quotaoffs_flag & type)
2685 xlog_recover_do_reg_buffer(mp, item, bp, buf_f, NULLCOMMITLSN);
2690 * This routine replays a modification made to a buffer at runtime.
2691 * There are actually two types of buffer, regular and inode, which
2692 * are handled differently. Inode buffers are handled differently
2693 * in that we only recover a specific set of data from them, namely
2694 * the inode di_next_unlinked fields. This is because all other inode
2695 * data is actually logged via inode records and any data we replay
2696 * here which overlaps that may be stale.
2698 * When meta-data buffers are freed at run time we log a buffer item
2699 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2700 * of the buffer in the log should not be replayed at recovery time.
2701 * This is so that if the blocks covered by the buffer are reused for
2702 * file data before we crash we don't end up replaying old, freed
2703 * meta-data into a user's file.
2705 * To handle the cancellation of buffer log items, we make two passes
2706 * over the log during recovery. During the first we build a table of
2707 * those buffers which have been cancelled, and during the second we
2708 * only replay those buffers which do not have corresponding cancel
2709 * records in the table. See xlog_recover_buffer_pass[1,2] above
2710 * for more details on the implementation of the table of cancel records.
2713 xlog_recover_buffer_pass2(
2715 struct list_head *buffer_list,
2716 struct xlog_recover_item *item,
2717 xfs_lsn_t current_lsn)
2719 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2720 xfs_mount_t *mp = log->l_mp;
2727 * In this pass we only want to recover all the buffers which have
2728 * not been cancelled and are not cancellation buffers themselves.
2730 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2731 buf_f->blf_len, buf_f->blf_flags)) {
2732 trace_xfs_log_recover_buf_cancel(log, buf_f);
2736 trace_xfs_log_recover_buf_recover(log, buf_f);
2739 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2740 buf_flags |= XBF_UNMAPPED;
2742 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2746 error = bp->b_error;
2748 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
2753 * Recover the buffer only if we get an LSN from it and it's less than
2754 * the lsn of the transaction we are replaying.
2756 * Note that we have to be extremely careful of readahead here.
2757 * Readahead does not attach verfiers to the buffers so if we don't
2758 * actually do any replay after readahead because of the LSN we found
2759 * in the buffer if more recent than that current transaction then we
2760 * need to attach the verifier directly. Failure to do so can lead to
2761 * future recovery actions (e.g. EFI and unlinked list recovery) can
2762 * operate on the buffers and they won't get the verifier attached. This
2763 * can lead to blocks on disk having the correct content but a stale
2766 * It is safe to assume these clean buffers are currently up to date.
2767 * If the buffer is dirtied by a later transaction being replayed, then
2768 * the verifier will be reset to match whatever recover turns that
2771 lsn = xlog_recover_get_buf_lsn(mp, bp);
2772 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2773 trace_xfs_log_recover_buf_skip(log, buf_f);
2774 xlog_recover_validate_buf_type(mp, bp, buf_f, NULLCOMMITLSN);
2778 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
2779 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2782 } else if (buf_f->blf_flags &
2783 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2786 dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2790 xlog_recover_do_reg_buffer(mp, item, bp, buf_f, current_lsn);
2794 * Perform delayed write on the buffer. Asynchronous writes will be
2795 * slower when taking into account all the buffers to be flushed.
2797 * Also make sure that only inode buffers with good sizes stay in
2798 * the buffer cache. The kernel moves inodes in buffers of 1 block
2799 * or inode_cluster_size bytes, whichever is bigger. The inode
2800 * buffers in the log can be a different size if the log was generated
2801 * by an older kernel using unclustered inode buffers or a newer kernel
2802 * running with a different inode cluster size. Regardless, if the
2803 * the inode buffer size isn't max(blocksize, inode_cluster_size)
2804 * for *our* value of inode_cluster_size, then we need to keep
2805 * the buffer out of the buffer cache so that the buffer won't
2806 * overlap with future reads of those inodes.
2808 if (XFS_DINODE_MAGIC ==
2809 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2810 (BBTOB(bp->b_length) != M_IGEO(log->l_mp)->inode_cluster_size)) {
2812 error = xfs_bwrite(bp);
2814 ASSERT(bp->b_mount == mp);
2815 bp->b_iodone = xlog_recover_iodone;
2816 xfs_buf_delwri_queue(bp, buffer_list);
2825 * Inode fork owner changes
2827 * If we have been told that we have to reparent the inode fork, it's because an
2828 * extent swap operation on a CRC enabled filesystem has been done and we are
2829 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2832 * The complexity here is that we don't have an inode context to work with, so
2833 * after we've replayed the inode we need to instantiate one. This is where the
2836 * We are in the middle of log recovery, so we can't run transactions. That
2837 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2838 * that will result in the corresponding iput() running the inode through
2839 * xfs_inactive(). If we've just replayed an inode core that changes the link
2840 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2841 * transactions (bad!).
2843 * So, to avoid this, we instantiate an inode directly from the inode core we've
2844 * just recovered. We have the buffer still locked, and all we really need to
2845 * instantiate is the inode core and the forks being modified. We can do this
2846 * manually, then run the inode btree owner change, and then tear down the
2847 * xfs_inode without having to run any transactions at all.
2849 * Also, because we don't have a transaction context available here but need to
2850 * gather all the buffers we modify for writeback so we pass the buffer_list
2851 * instead for the operation to use.
2855 xfs_recover_inode_owner_change(
2856 struct xfs_mount *mp,
2857 struct xfs_dinode *dip,
2858 struct xfs_inode_log_format *in_f,
2859 struct list_head *buffer_list)
2861 struct xfs_inode *ip;
2864 ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER));
2866 ip = xfs_inode_alloc(mp, in_f->ilf_ino);
2870 /* instantiate the inode */
2871 xfs_inode_from_disk(ip, dip);
2872 ASSERT(ip->i_d.di_version >= 3);
2874 error = xfs_iformat_fork(ip, dip);
2878 if (!xfs_inode_verify_forks(ip)) {
2879 error = -EFSCORRUPTED;
2883 if (in_f->ilf_fields & XFS_ILOG_DOWNER) {
2884 ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT);
2885 error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK,
2886 ip->i_ino, buffer_list);
2891 if (in_f->ilf_fields & XFS_ILOG_AOWNER) {
2892 ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT);
2893 error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK,
2894 ip->i_ino, buffer_list);
2905 xlog_recover_inode_pass2(
2907 struct list_head *buffer_list,
2908 struct xlog_recover_item *item,
2909 xfs_lsn_t current_lsn)
2911 struct xfs_inode_log_format *in_f;
2912 xfs_mount_t *mp = log->l_mp;
2921 struct xfs_log_dinode *ldip;
2925 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
2926 in_f = item->ri_buf[0].i_addr;
2928 in_f = kmem_alloc(sizeof(struct xfs_inode_log_format), 0);
2930 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2936 * Inode buffers can be freed, look out for it,
2937 * and do not replay the inode.
2939 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2940 in_f->ilf_len, 0)) {
2942 trace_xfs_log_recover_inode_cancel(log, in_f);
2945 trace_xfs_log_recover_inode_recover(log, in_f);
2947 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
2948 &xfs_inode_buf_ops);
2953 error = bp->b_error;
2955 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
2958 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2959 dip = xfs_buf_offset(bp, in_f->ilf_boffset);
2962 * Make sure the place we're flushing out to really looks
2965 if (XFS_IS_CORRUPT(mp, !xfs_verify_magic16(bp, dip->di_magic))) {
2967 "%s: Bad inode magic number, dip = "PTR_FMT", dino bp = "PTR_FMT", ino = %Ld",
2968 __func__, dip, bp, in_f->ilf_ino);
2969 error = -EFSCORRUPTED;
2972 ldip = item->ri_buf[1].i_addr;
2973 if (XFS_IS_CORRUPT(mp, ldip->di_magic != XFS_DINODE_MAGIC)) {
2975 "%s: Bad inode log record, rec ptr "PTR_FMT", ino %Ld",
2976 __func__, item, in_f->ilf_ino);
2977 error = -EFSCORRUPTED;
2982 * If the inode has an LSN in it, recover the inode only if it's less
2983 * than the lsn of the transaction we are replaying. Note: we still
2984 * need to replay an owner change even though the inode is more recent
2985 * than the transaction as there is no guarantee that all the btree
2986 * blocks are more recent than this transaction, too.
2988 if (dip->di_version >= 3) {
2989 xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn);
2991 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2992 trace_xfs_log_recover_inode_skip(log, in_f);
2994 goto out_owner_change;
2999 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
3000 * are transactional and if ordering is necessary we can determine that
3001 * more accurately by the LSN field in the V3 inode core. Don't trust
3002 * the inode versions we might be changing them here - use the
3003 * superblock flag to determine whether we need to look at di_flushiter
3004 * to skip replay when the on disk inode is newer than the log one
3006 if (!xfs_sb_version_hascrc(&mp->m_sb) &&
3007 ldip->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
3009 * Deal with the wrap case, DI_MAX_FLUSH is less
3010 * than smaller numbers
3012 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
3013 ldip->di_flushiter < (DI_MAX_FLUSH >> 1)) {
3016 trace_xfs_log_recover_inode_skip(log, in_f);
3022 /* Take the opportunity to reset the flush iteration count */
3023 ldip->di_flushiter = 0;
3025 if (unlikely(S_ISREG(ldip->di_mode))) {
3026 if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) &&
3027 (ldip->di_format != XFS_DINODE_FMT_BTREE)) {
3028 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
3029 XFS_ERRLEVEL_LOW, mp, ldip,
3032 "%s: Bad regular inode log record, rec ptr "PTR_FMT", "
3033 "ino ptr = "PTR_FMT", ino bp = "PTR_FMT", ino %Ld",
3034 __func__, item, dip, bp, in_f->ilf_ino);
3035 error = -EFSCORRUPTED;
3038 } else if (unlikely(S_ISDIR(ldip->di_mode))) {
3039 if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) &&
3040 (ldip->di_format != XFS_DINODE_FMT_BTREE) &&
3041 (ldip->di_format != XFS_DINODE_FMT_LOCAL)) {
3042 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
3043 XFS_ERRLEVEL_LOW, mp, ldip,
3046 "%s: Bad dir inode log record, rec ptr "PTR_FMT", "
3047 "ino ptr = "PTR_FMT", ino bp = "PTR_FMT", ino %Ld",
3048 __func__, item, dip, bp, in_f->ilf_ino);
3049 error = -EFSCORRUPTED;
3053 if (unlikely(ldip->di_nextents + ldip->di_anextents > ldip->di_nblocks)){
3054 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
3055 XFS_ERRLEVEL_LOW, mp, ldip,
3058 "%s: Bad inode log record, rec ptr "PTR_FMT", dino ptr "PTR_FMT", "
3059 "dino bp "PTR_FMT", ino %Ld, total extents = %d, nblocks = %Ld",
3060 __func__, item, dip, bp, in_f->ilf_ino,
3061 ldip->di_nextents + ldip->di_anextents,
3063 error = -EFSCORRUPTED;
3066 if (unlikely(ldip->di_forkoff > mp->m_sb.sb_inodesize)) {
3067 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
3068 XFS_ERRLEVEL_LOW, mp, ldip,
3071 "%s: Bad inode log record, rec ptr "PTR_FMT", dino ptr "PTR_FMT", "
3072 "dino bp "PTR_FMT", ino %Ld, forkoff 0x%x", __func__,
3073 item, dip, bp, in_f->ilf_ino, ldip->di_forkoff);
3074 error = -EFSCORRUPTED;
3077 isize = xfs_log_dinode_size(ldip->di_version);
3078 if (unlikely(item->ri_buf[1].i_len > isize)) {
3079 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
3080 XFS_ERRLEVEL_LOW, mp, ldip,
3083 "%s: Bad inode log record length %d, rec ptr "PTR_FMT,
3084 __func__, item->ri_buf[1].i_len, item);
3085 error = -EFSCORRUPTED;
3089 /* recover the log dinode inode into the on disk inode */
3090 xfs_log_dinode_to_disk(ldip, dip);
3092 fields = in_f->ilf_fields;
3093 if (fields & XFS_ILOG_DEV)
3094 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
3096 if (in_f->ilf_size == 2)
3097 goto out_owner_change;
3098 len = item->ri_buf[2].i_len;
3099 src = item->ri_buf[2].i_addr;
3100 ASSERT(in_f->ilf_size <= 4);
3101 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
3102 ASSERT(!(fields & XFS_ILOG_DFORK) ||
3103 (len == in_f->ilf_dsize));
3105 switch (fields & XFS_ILOG_DFORK) {
3106 case XFS_ILOG_DDATA:
3108 memcpy(XFS_DFORK_DPTR(dip), src, len);
3111 case XFS_ILOG_DBROOT:
3112 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
3113 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
3114 XFS_DFORK_DSIZE(dip, mp));
3119 * There are no data fork flags set.
3121 ASSERT((fields & XFS_ILOG_DFORK) == 0);
3126 * If we logged any attribute data, recover it. There may or
3127 * may not have been any other non-core data logged in this
3130 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
3131 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
3136 len = item->ri_buf[attr_index].i_len;
3137 src = item->ri_buf[attr_index].i_addr;
3138 ASSERT(len == in_f->ilf_asize);
3140 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
3141 case XFS_ILOG_ADATA:
3143 dest = XFS_DFORK_APTR(dip);
3144 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
3145 memcpy(dest, src, len);
3148 case XFS_ILOG_ABROOT:
3149 dest = XFS_DFORK_APTR(dip);
3150 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
3151 len, (xfs_bmdr_block_t*)dest,
3152 XFS_DFORK_ASIZE(dip, mp));
3156 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
3158 error = -EFSCORRUPTED;
3164 /* Recover the swapext owner change unless inode has been deleted */
3165 if ((in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER)) &&
3166 (dip->di_mode != 0))
3167 error = xfs_recover_inode_owner_change(mp, dip, in_f,
3169 /* re-generate the checksum. */
3170 xfs_dinode_calc_crc(log->l_mp, dip);
3172 ASSERT(bp->b_mount == mp);
3173 bp->b_iodone = xlog_recover_iodone;
3174 xfs_buf_delwri_queue(bp, buffer_list);
3185 * Recover QUOTAOFF records. We simply make a note of it in the xlog
3186 * structure, so that we know not to do any dquot item or dquot buffer recovery,
3190 xlog_recover_quotaoff_pass1(
3192 struct xlog_recover_item *item)
3194 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
3198 * The logitem format's flag tells us if this was user quotaoff,
3199 * group/project quotaoff or both.
3201 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
3202 log->l_quotaoffs_flag |= XFS_DQ_USER;
3203 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
3204 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
3205 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
3206 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
3212 * Recover a dquot record
3215 xlog_recover_dquot_pass2(
3217 struct list_head *buffer_list,
3218 struct xlog_recover_item *item,
3219 xfs_lsn_t current_lsn)
3221 xfs_mount_t *mp = log->l_mp;
3223 struct xfs_disk_dquot *ddq, *recddq;
3226 xfs_dq_logformat_t *dq_f;
3231 * Filesystems are required to send in quota flags at mount time.
3233 if (mp->m_qflags == 0)
3236 recddq = item->ri_buf[1].i_addr;
3237 if (recddq == NULL) {
3238 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
3239 return -EFSCORRUPTED;
3241 if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot)) {
3242 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
3243 item->ri_buf[1].i_len, __func__);
3244 return -EFSCORRUPTED;
3248 * This type of quotas was turned off, so ignore this record.
3250 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3252 if (log->l_quotaoffs_flag & type)
3256 * At this point we know that quota was _not_ turned off.
3257 * Since the mount flags are not indicating to us otherwise, this
3258 * must mean that quota is on, and the dquot needs to be replayed.
3259 * Remember that we may not have fully recovered the superblock yet,
3260 * so we can't do the usual trick of looking at the SB quota bits.
3262 * The other possibility, of course, is that the quota subsystem was
3263 * removed since the last mount - ENOSYS.
3265 dq_f = item->ri_buf[0].i_addr;
3267 fa = xfs_dquot_verify(mp, recddq, dq_f->qlf_id, 0);
3269 xfs_alert(mp, "corrupt dquot ID 0x%x in log at %pS",
3271 return -EFSCORRUPTED;
3273 ASSERT(dq_f->qlf_len == 1);
3276 * At this point we are assuming that the dquots have been allocated
3277 * and hence the buffer has valid dquots stamped in it. It should,
3278 * therefore, pass verifier validation. If the dquot is bad, then the
3279 * we'll return an error here, so we don't need to specifically check
3280 * the dquot in the buffer after the verifier has run.
3282 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
3283 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
3284 &xfs_dquot_buf_ops);
3289 ddq = xfs_buf_offset(bp, dq_f->qlf_boffset);
3292 * If the dquot has an LSN in it, recover the dquot only if it's less
3293 * than the lsn of the transaction we are replaying.
3295 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3296 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq;
3297 xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn);
3299 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3304 memcpy(ddq, recddq, item->ri_buf[1].i_len);
3305 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3306 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
3310 ASSERT(dq_f->qlf_size == 2);
3311 ASSERT(bp->b_mount == mp);
3312 bp->b_iodone = xlog_recover_iodone;
3313 xfs_buf_delwri_queue(bp, buffer_list);
3321 * This routine is called to create an in-core extent free intent
3322 * item from the efi format structure which was logged on disk.
3323 * It allocates an in-core efi, copies the extents from the format
3324 * structure into it, and adds the efi to the AIL with the given
3328 xlog_recover_efi_pass2(
3330 struct xlog_recover_item *item,
3334 struct xfs_mount *mp = log->l_mp;
3335 struct xfs_efi_log_item *efip;
3336 struct xfs_efi_log_format *efi_formatp;
3338 efi_formatp = item->ri_buf[0].i_addr;
3340 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
3341 error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format);
3343 xfs_efi_item_free(efip);
3346 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
3348 spin_lock(&log->l_ailp->ail_lock);
3350 * The EFI has two references. One for the EFD and one for EFI to ensure
3351 * it makes it into the AIL. Insert the EFI into the AIL directly and
3352 * drop the EFI reference. Note that xfs_trans_ail_update() drops the
3355 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
3356 xfs_efi_release(efip);
3362 * This routine is called when an EFD format structure is found in a committed
3363 * transaction in the log. Its purpose is to cancel the corresponding EFI if it
3364 * was still in the log. To do this it searches the AIL for the EFI with an id
3365 * equal to that in the EFD format structure. If we find it we drop the EFD
3366 * reference, which removes the EFI from the AIL and frees it.
3369 xlog_recover_efd_pass2(
3371 struct xlog_recover_item *item)
3373 xfs_efd_log_format_t *efd_formatp;
3374 xfs_efi_log_item_t *efip = NULL;
3375 struct xfs_log_item *lip;
3377 struct xfs_ail_cursor cur;
3378 struct xfs_ail *ailp = log->l_ailp;
3380 efd_formatp = item->ri_buf[0].i_addr;
3381 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
3382 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
3383 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
3384 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
3385 efi_id = efd_formatp->efd_efi_id;
3388 * Search for the EFI with the id in the EFD format structure in the
3391 spin_lock(&ailp->ail_lock);
3392 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3393 while (lip != NULL) {
3394 if (lip->li_type == XFS_LI_EFI) {
3395 efip = (xfs_efi_log_item_t *)lip;
3396 if (efip->efi_format.efi_id == efi_id) {
3398 * Drop the EFD reference to the EFI. This
3399 * removes the EFI from the AIL and frees it.
3401 spin_unlock(&ailp->ail_lock);
3402 xfs_efi_release(efip);
3403 spin_lock(&ailp->ail_lock);
3407 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3410 xfs_trans_ail_cursor_done(&cur);
3411 spin_unlock(&ailp->ail_lock);
3417 * This routine is called to create an in-core extent rmap update
3418 * item from the rui format structure which was logged on disk.
3419 * It allocates an in-core rui, copies the extents from the format
3420 * structure into it, and adds the rui to the AIL with the given
3424 xlog_recover_rui_pass2(
3426 struct xlog_recover_item *item,
3430 struct xfs_mount *mp = log->l_mp;
3431 struct xfs_rui_log_item *ruip;
3432 struct xfs_rui_log_format *rui_formatp;
3434 rui_formatp = item->ri_buf[0].i_addr;
3436 ruip = xfs_rui_init(mp, rui_formatp->rui_nextents);
3437 error = xfs_rui_copy_format(&item->ri_buf[0], &ruip->rui_format);
3439 xfs_rui_item_free(ruip);
3442 atomic_set(&ruip->rui_next_extent, rui_formatp->rui_nextents);
3444 spin_lock(&log->l_ailp->ail_lock);
3446 * The RUI has two references. One for the RUD and one for RUI to ensure
3447 * it makes it into the AIL. Insert the RUI into the AIL directly and
3448 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
3451 xfs_trans_ail_update(log->l_ailp, &ruip->rui_item, lsn);
3452 xfs_rui_release(ruip);
3458 * This routine is called when an RUD format structure is found in a committed
3459 * transaction in the log. Its purpose is to cancel the corresponding RUI if it
3460 * was still in the log. To do this it searches the AIL for the RUI with an id
3461 * equal to that in the RUD format structure. If we find it we drop the RUD
3462 * reference, which removes the RUI from the AIL and frees it.
3465 xlog_recover_rud_pass2(
3467 struct xlog_recover_item *item)
3469 struct xfs_rud_log_format *rud_formatp;
3470 struct xfs_rui_log_item *ruip = NULL;
3471 struct xfs_log_item *lip;
3473 struct xfs_ail_cursor cur;
3474 struct xfs_ail *ailp = log->l_ailp;
3476 rud_formatp = item->ri_buf[0].i_addr;
3477 ASSERT(item->ri_buf[0].i_len == sizeof(struct xfs_rud_log_format));
3478 rui_id = rud_formatp->rud_rui_id;
3481 * Search for the RUI with the id in the RUD format structure in the
3484 spin_lock(&ailp->ail_lock);
3485 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3486 while (lip != NULL) {
3487 if (lip->li_type == XFS_LI_RUI) {
3488 ruip = (struct xfs_rui_log_item *)lip;
3489 if (ruip->rui_format.rui_id == rui_id) {
3491 * Drop the RUD reference to the RUI. This
3492 * removes the RUI from the AIL and frees it.
3494 spin_unlock(&ailp->ail_lock);
3495 xfs_rui_release(ruip);
3496 spin_lock(&ailp->ail_lock);
3500 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3503 xfs_trans_ail_cursor_done(&cur);
3504 spin_unlock(&ailp->ail_lock);
3510 * Copy an CUI format buffer from the given buf, and into the destination
3511 * CUI format structure. The CUI/CUD items were designed not to need any
3512 * special alignment handling.
3515 xfs_cui_copy_format(
3516 struct xfs_log_iovec *buf,
3517 struct xfs_cui_log_format *dst_cui_fmt)
3519 struct xfs_cui_log_format *src_cui_fmt;
3522 src_cui_fmt = buf->i_addr;
3523 len = xfs_cui_log_format_sizeof(src_cui_fmt->cui_nextents);
3525 if (buf->i_len == len) {
3526 memcpy(dst_cui_fmt, src_cui_fmt, len);
3529 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, NULL);
3530 return -EFSCORRUPTED;
3534 * This routine is called to create an in-core extent refcount update
3535 * item from the cui format structure which was logged on disk.
3536 * It allocates an in-core cui, copies the extents from the format
3537 * structure into it, and adds the cui to the AIL with the given
3541 xlog_recover_cui_pass2(
3543 struct xlog_recover_item *item,
3547 struct xfs_mount *mp = log->l_mp;
3548 struct xfs_cui_log_item *cuip;
3549 struct xfs_cui_log_format *cui_formatp;
3551 cui_formatp = item->ri_buf[0].i_addr;
3553 cuip = xfs_cui_init(mp, cui_formatp->cui_nextents);
3554 error = xfs_cui_copy_format(&item->ri_buf[0], &cuip->cui_format);
3556 xfs_cui_item_free(cuip);
3559 atomic_set(&cuip->cui_next_extent, cui_formatp->cui_nextents);
3561 spin_lock(&log->l_ailp->ail_lock);
3563 * The CUI has two references. One for the CUD and one for CUI to ensure
3564 * it makes it into the AIL. Insert the CUI into the AIL directly and
3565 * drop the CUI reference. Note that xfs_trans_ail_update() drops the
3568 xfs_trans_ail_update(log->l_ailp, &cuip->cui_item, lsn);
3569 xfs_cui_release(cuip);
3575 * This routine is called when an CUD format structure is found in a committed
3576 * transaction in the log. Its purpose is to cancel the corresponding CUI if it
3577 * was still in the log. To do this it searches the AIL for the CUI with an id
3578 * equal to that in the CUD format structure. If we find it we drop the CUD
3579 * reference, which removes the CUI from the AIL and frees it.
3582 xlog_recover_cud_pass2(
3584 struct xlog_recover_item *item)
3586 struct xfs_cud_log_format *cud_formatp;
3587 struct xfs_cui_log_item *cuip = NULL;
3588 struct xfs_log_item *lip;
3590 struct xfs_ail_cursor cur;
3591 struct xfs_ail *ailp = log->l_ailp;
3593 cud_formatp = item->ri_buf[0].i_addr;
3594 if (item->ri_buf[0].i_len != sizeof(struct xfs_cud_log_format)) {
3595 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, log->l_mp);
3596 return -EFSCORRUPTED;
3598 cui_id = cud_formatp->cud_cui_id;
3601 * Search for the CUI with the id in the CUD format structure in the
3604 spin_lock(&ailp->ail_lock);
3605 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3606 while (lip != NULL) {
3607 if (lip->li_type == XFS_LI_CUI) {
3608 cuip = (struct xfs_cui_log_item *)lip;
3609 if (cuip->cui_format.cui_id == cui_id) {
3611 * Drop the CUD reference to the CUI. This
3612 * removes the CUI from the AIL and frees it.
3614 spin_unlock(&ailp->ail_lock);
3615 xfs_cui_release(cuip);
3616 spin_lock(&ailp->ail_lock);
3620 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3623 xfs_trans_ail_cursor_done(&cur);
3624 spin_unlock(&ailp->ail_lock);
3630 * Copy an BUI format buffer from the given buf, and into the destination
3631 * BUI format structure. The BUI/BUD items were designed not to need any
3632 * special alignment handling.
3635 xfs_bui_copy_format(
3636 struct xfs_log_iovec *buf,
3637 struct xfs_bui_log_format *dst_bui_fmt)
3639 struct xfs_bui_log_format *src_bui_fmt;
3642 src_bui_fmt = buf->i_addr;
3643 len = xfs_bui_log_format_sizeof(src_bui_fmt->bui_nextents);
3645 if (buf->i_len == len) {
3646 memcpy(dst_bui_fmt, src_bui_fmt, len);
3649 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, NULL);
3650 return -EFSCORRUPTED;
3654 * This routine is called to create an in-core extent bmap update
3655 * item from the bui format structure which was logged on disk.
3656 * It allocates an in-core bui, copies the extents from the format
3657 * structure into it, and adds the bui to the AIL with the given
3661 xlog_recover_bui_pass2(
3663 struct xlog_recover_item *item,
3667 struct xfs_mount *mp = log->l_mp;
3668 struct xfs_bui_log_item *buip;
3669 struct xfs_bui_log_format *bui_formatp;
3671 bui_formatp = item->ri_buf[0].i_addr;
3673 if (bui_formatp->bui_nextents != XFS_BUI_MAX_FAST_EXTENTS) {
3674 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, log->l_mp);
3675 return -EFSCORRUPTED;
3677 buip = xfs_bui_init(mp);
3678 error = xfs_bui_copy_format(&item->ri_buf[0], &buip->bui_format);
3680 xfs_bui_item_free(buip);
3683 atomic_set(&buip->bui_next_extent, bui_formatp->bui_nextents);
3685 spin_lock(&log->l_ailp->ail_lock);
3687 * The RUI has two references. One for the RUD and one for RUI to ensure
3688 * it makes it into the AIL. Insert the RUI into the AIL directly and
3689 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
3692 xfs_trans_ail_update(log->l_ailp, &buip->bui_item, lsn);
3693 xfs_bui_release(buip);
3699 * This routine is called when an BUD format structure is found in a committed
3700 * transaction in the log. Its purpose is to cancel the corresponding BUI if it
3701 * was still in the log. To do this it searches the AIL for the BUI with an id
3702 * equal to that in the BUD format structure. If we find it we drop the BUD
3703 * reference, which removes the BUI from the AIL and frees it.
3706 xlog_recover_bud_pass2(
3708 struct xlog_recover_item *item)
3710 struct xfs_bud_log_format *bud_formatp;
3711 struct xfs_bui_log_item *buip = NULL;
3712 struct xfs_log_item *lip;
3714 struct xfs_ail_cursor cur;
3715 struct xfs_ail *ailp = log->l_ailp;
3717 bud_formatp = item->ri_buf[0].i_addr;
3718 if (item->ri_buf[0].i_len != sizeof(struct xfs_bud_log_format)) {
3719 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, log->l_mp);
3720 return -EFSCORRUPTED;
3722 bui_id = bud_formatp->bud_bui_id;
3725 * Search for the BUI with the id in the BUD format structure in the
3728 spin_lock(&ailp->ail_lock);
3729 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3730 while (lip != NULL) {
3731 if (lip->li_type == XFS_LI_BUI) {
3732 buip = (struct xfs_bui_log_item *)lip;
3733 if (buip->bui_format.bui_id == bui_id) {
3735 * Drop the BUD reference to the BUI. This
3736 * removes the BUI from the AIL and frees it.
3738 spin_unlock(&ailp->ail_lock);
3739 xfs_bui_release(buip);
3740 spin_lock(&ailp->ail_lock);
3744 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3747 xfs_trans_ail_cursor_done(&cur);
3748 spin_unlock(&ailp->ail_lock);
3754 * This routine is called when an inode create format structure is found in a
3755 * committed transaction in the log. It's purpose is to initialise the inodes
3756 * being allocated on disk. This requires us to get inode cluster buffers that
3757 * match the range to be initialised, stamped with inode templates and written
3758 * by delayed write so that subsequent modifications will hit the cached buffer
3759 * and only need writing out at the end of recovery.
3762 xlog_recover_do_icreate_pass2(
3764 struct list_head *buffer_list,
3765 xlog_recover_item_t *item)
3767 struct xfs_mount *mp = log->l_mp;
3768 struct xfs_icreate_log *icl;
3769 struct xfs_ino_geometry *igeo = M_IGEO(mp);
3770 xfs_agnumber_t agno;
3771 xfs_agblock_t agbno;
3774 xfs_agblock_t length;
3780 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3781 if (icl->icl_type != XFS_LI_ICREATE) {
3782 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
3786 if (icl->icl_size != 1) {
3787 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
3791 agno = be32_to_cpu(icl->icl_ag);
3792 if (agno >= mp->m_sb.sb_agcount) {
3793 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
3796 agbno = be32_to_cpu(icl->icl_agbno);
3797 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3798 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
3801 isize = be32_to_cpu(icl->icl_isize);
3802 if (isize != mp->m_sb.sb_inodesize) {
3803 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
3806 count = be32_to_cpu(icl->icl_count);
3808 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
3811 length = be32_to_cpu(icl->icl_length);
3812 if (!length || length >= mp->m_sb.sb_agblocks) {
3813 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
3818 * The inode chunk is either full or sparse and we only support
3819 * m_ino_geo.ialloc_min_blks sized sparse allocations at this time.
3821 if (length != igeo->ialloc_blks &&
3822 length != igeo->ialloc_min_blks) {
3824 "%s: unsupported chunk length", __FUNCTION__);
3828 /* verify inode count is consistent with extent length */
3829 if ((count >> mp->m_sb.sb_inopblog) != length) {
3831 "%s: inconsistent inode count and chunk length",
3837 * The icreate transaction can cover multiple cluster buffers and these
3838 * buffers could have been freed and reused. Check the individual
3839 * buffers for cancellation so we don't overwrite anything written after
3842 bb_per_cluster = XFS_FSB_TO_BB(mp, igeo->blocks_per_cluster);
3843 nbufs = length / igeo->blocks_per_cluster;
3844 for (i = 0, cancel_count = 0; i < nbufs; i++) {
3847 daddr = XFS_AGB_TO_DADDR(mp, agno,
3848 agbno + i * igeo->blocks_per_cluster);
3849 if (xlog_check_buffer_cancelled(log, daddr, bb_per_cluster, 0))
3854 * We currently only use icreate for a single allocation at a time. This
3855 * means we should expect either all or none of the buffers to be
3856 * cancelled. Be conservative and skip replay if at least one buffer is
3857 * cancelled, but warn the user that something is awry if the buffers
3858 * are not consistent.
3860 * XXX: This must be refined to only skip cancelled clusters once we use
3861 * icreate for multiple chunk allocations.
3863 ASSERT(!cancel_count || cancel_count == nbufs);
3865 if (cancel_count != nbufs)
3867 "WARNING: partial inode chunk cancellation, skipped icreate.");
3868 trace_xfs_log_recover_icreate_cancel(log, icl);
3872 trace_xfs_log_recover_icreate_recover(log, icl);
3873 return xfs_ialloc_inode_init(mp, NULL, buffer_list, count, agno, agbno,
3874 length, be32_to_cpu(icl->icl_gen));
3878 xlog_recover_buffer_ra_pass2(
3880 struct xlog_recover_item *item)
3882 struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr;
3883 struct xfs_mount *mp = log->l_mp;
3885 if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
3886 buf_f->blf_len, buf_f->blf_flags)) {
3890 xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
3891 buf_f->blf_len, NULL);
3895 xlog_recover_inode_ra_pass2(
3897 struct xlog_recover_item *item)
3899 struct xfs_inode_log_format ilf_buf;
3900 struct xfs_inode_log_format *ilfp;
3901 struct xfs_mount *mp = log->l_mp;
3904 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3905 ilfp = item->ri_buf[0].i_addr;
3908 memset(ilfp, 0, sizeof(*ilfp));
3909 error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
3914 if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
3917 xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
3918 ilfp->ilf_len, &xfs_inode_buf_ra_ops);
3922 xlog_recover_dquot_ra_pass2(
3924 struct xlog_recover_item *item)
3926 struct xfs_mount *mp = log->l_mp;
3927 struct xfs_disk_dquot *recddq;
3928 struct xfs_dq_logformat *dq_f;
3933 if (mp->m_qflags == 0)
3936 recddq = item->ri_buf[1].i_addr;
3939 if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
3942 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3944 if (log->l_quotaoffs_flag & type)
3947 dq_f = item->ri_buf[0].i_addr;
3949 ASSERT(dq_f->qlf_len == 1);
3951 len = XFS_FSB_TO_BB(mp, dq_f->qlf_len);
3952 if (xlog_peek_buffer_cancelled(log, dq_f->qlf_blkno, len, 0))
3955 xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno, len,
3956 &xfs_dquot_buf_ra_ops);
3960 xlog_recover_ra_pass2(
3962 struct xlog_recover_item *item)
3964 switch (ITEM_TYPE(item)) {
3966 xlog_recover_buffer_ra_pass2(log, item);
3969 xlog_recover_inode_ra_pass2(log, item);
3972 xlog_recover_dquot_ra_pass2(log, item);
3976 case XFS_LI_QUOTAOFF:
3989 xlog_recover_commit_pass1(
3991 struct xlog_recover *trans,
3992 struct xlog_recover_item *item)
3994 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
3996 switch (ITEM_TYPE(item)) {
3998 return xlog_recover_buffer_pass1(log, item);
3999 case XFS_LI_QUOTAOFF:
4000 return xlog_recover_quotaoff_pass1(log, item);
4005 case XFS_LI_ICREATE:
4012 /* nothing to do in pass 1 */
4015 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
4016 __func__, ITEM_TYPE(item));
4018 return -EFSCORRUPTED;
4023 xlog_recover_commit_pass2(
4025 struct xlog_recover *trans,
4026 struct list_head *buffer_list,
4027 struct xlog_recover_item *item)
4029 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
4031 switch (ITEM_TYPE(item)) {
4033 return xlog_recover_buffer_pass2(log, buffer_list, item,
4036 return xlog_recover_inode_pass2(log, buffer_list, item,
4039 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
4041 return xlog_recover_efd_pass2(log, item);
4043 return xlog_recover_rui_pass2(log, item, trans->r_lsn);
4045 return xlog_recover_rud_pass2(log, item);
4047 return xlog_recover_cui_pass2(log, item, trans->r_lsn);
4049 return xlog_recover_cud_pass2(log, item);
4051 return xlog_recover_bui_pass2(log, item, trans->r_lsn);
4053 return xlog_recover_bud_pass2(log, item);
4055 return xlog_recover_dquot_pass2(log, buffer_list, item,
4057 case XFS_LI_ICREATE:
4058 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
4059 case XFS_LI_QUOTAOFF:
4060 /* nothing to do in pass2 */
4063 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
4064 __func__, ITEM_TYPE(item));
4066 return -EFSCORRUPTED;
4071 xlog_recover_items_pass2(
4073 struct xlog_recover *trans,
4074 struct list_head *buffer_list,
4075 struct list_head *item_list)
4077 struct xlog_recover_item *item;
4080 list_for_each_entry(item, item_list, ri_list) {
4081 error = xlog_recover_commit_pass2(log, trans,
4091 * Perform the transaction.
4093 * If the transaction modifies a buffer or inode, do it now. Otherwise,
4094 * EFIs and EFDs get queued up by adding entries into the AIL for them.
4097 xlog_recover_commit_trans(
4099 struct xlog_recover *trans,
4101 struct list_head *buffer_list)
4104 int items_queued = 0;
4105 struct xlog_recover_item *item;
4106 struct xlog_recover_item *next;
4107 LIST_HEAD (ra_list);
4108 LIST_HEAD (done_list);
4110 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
4112 hlist_del_init(&trans->r_list);
4114 error = xlog_recover_reorder_trans(log, trans, pass);
4118 list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
4120 case XLOG_RECOVER_PASS1:
4121 error = xlog_recover_commit_pass1(log, trans, item);
4123 case XLOG_RECOVER_PASS2:
4124 xlog_recover_ra_pass2(log, item);
4125 list_move_tail(&item->ri_list, &ra_list);
4127 if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
4128 error = xlog_recover_items_pass2(log, trans,
4129 buffer_list, &ra_list);
4130 list_splice_tail_init(&ra_list, &done_list);
4144 if (!list_empty(&ra_list)) {
4146 error = xlog_recover_items_pass2(log, trans,
4147 buffer_list, &ra_list);
4148 list_splice_tail_init(&ra_list, &done_list);
4151 if (!list_empty(&done_list))
4152 list_splice_init(&done_list, &trans->r_itemq);
4158 xlog_recover_add_item(
4159 struct list_head *head)
4161 xlog_recover_item_t *item;
4163 item = kmem_zalloc(sizeof(xlog_recover_item_t), 0);
4164 INIT_LIST_HEAD(&item->ri_list);
4165 list_add_tail(&item->ri_list, head);
4169 xlog_recover_add_to_cont_trans(
4171 struct xlog_recover *trans,
4175 xlog_recover_item_t *item;
4176 char *ptr, *old_ptr;
4180 * If the transaction is empty, the header was split across this and the
4181 * previous record. Copy the rest of the header.
4183 if (list_empty(&trans->r_itemq)) {
4184 ASSERT(len <= sizeof(struct xfs_trans_header));
4185 if (len > sizeof(struct xfs_trans_header)) {
4186 xfs_warn(log->l_mp, "%s: bad header length", __func__);
4187 return -EFSCORRUPTED;
4190 xlog_recover_add_item(&trans->r_itemq);
4191 ptr = (char *)&trans->r_theader +
4192 sizeof(struct xfs_trans_header) - len;
4193 memcpy(ptr, dp, len);
4197 /* take the tail entry */
4198 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
4200 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
4201 old_len = item->ri_buf[item->ri_cnt-1].i_len;
4203 ptr = kmem_realloc(old_ptr, len + old_len, 0);
4204 memcpy(&ptr[old_len], dp, len);
4205 item->ri_buf[item->ri_cnt-1].i_len += len;
4206 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
4207 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
4212 * The next region to add is the start of a new region. It could be
4213 * a whole region or it could be the first part of a new region. Because
4214 * of this, the assumption here is that the type and size fields of all
4215 * format structures fit into the first 32 bits of the structure.
4217 * This works because all regions must be 32 bit aligned. Therefore, we
4218 * either have both fields or we have neither field. In the case we have
4219 * neither field, the data part of the region is zero length. We only have
4220 * a log_op_header and can throw away the header since a new one will appear
4221 * later. If we have at least 4 bytes, then we can determine how many regions
4222 * will appear in the current log item.
4225 xlog_recover_add_to_trans(
4227 struct xlog_recover *trans,
4231 struct xfs_inode_log_format *in_f; /* any will do */
4232 xlog_recover_item_t *item;
4237 if (list_empty(&trans->r_itemq)) {
4238 /* we need to catch log corruptions here */
4239 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
4240 xfs_warn(log->l_mp, "%s: bad header magic number",
4243 return -EFSCORRUPTED;
4246 if (len > sizeof(struct xfs_trans_header)) {
4247 xfs_warn(log->l_mp, "%s: bad header length", __func__);
4249 return -EFSCORRUPTED;
4253 * The transaction header can be arbitrarily split across op
4254 * records. If we don't have the whole thing here, copy what we
4255 * do have and handle the rest in the next record.
4257 if (len == sizeof(struct xfs_trans_header))
4258 xlog_recover_add_item(&trans->r_itemq);
4259 memcpy(&trans->r_theader, dp, len);
4263 ptr = kmem_alloc(len, 0);
4264 memcpy(ptr, dp, len);
4265 in_f = (struct xfs_inode_log_format *)ptr;
4267 /* take the tail entry */
4268 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
4269 if (item->ri_total != 0 &&
4270 item->ri_total == item->ri_cnt) {
4271 /* tail item is in use, get a new one */
4272 xlog_recover_add_item(&trans->r_itemq);
4273 item = list_entry(trans->r_itemq.prev,
4274 xlog_recover_item_t, ri_list);
4277 if (item->ri_total == 0) { /* first region to be added */
4278 if (in_f->ilf_size == 0 ||
4279 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
4281 "bad number of regions (%d) in inode log format",
4285 return -EFSCORRUPTED;
4288 item->ri_total = in_f->ilf_size;
4290 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
4294 if (item->ri_total <= item->ri_cnt) {
4296 "log item region count (%d) overflowed size (%d)",
4297 item->ri_cnt, item->ri_total);
4300 return -EFSCORRUPTED;
4303 /* Description region is ri_buf[0] */
4304 item->ri_buf[item->ri_cnt].i_addr = ptr;
4305 item->ri_buf[item->ri_cnt].i_len = len;
4307 trace_xfs_log_recover_item_add(log, trans, item, 0);
4312 * Free up any resources allocated by the transaction
4314 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
4317 xlog_recover_free_trans(
4318 struct xlog_recover *trans)
4320 xlog_recover_item_t *item, *n;
4323 hlist_del_init(&trans->r_list);
4325 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
4326 /* Free the regions in the item. */
4327 list_del(&item->ri_list);
4328 for (i = 0; i < item->ri_cnt; i++)
4329 kmem_free(item->ri_buf[i].i_addr);
4330 /* Free the item itself */
4331 kmem_free(item->ri_buf);
4334 /* Free the transaction recover structure */
4339 * On error or completion, trans is freed.
4342 xlog_recovery_process_trans(
4344 struct xlog_recover *trans,
4349 struct list_head *buffer_list)
4352 bool freeit = false;
4354 /* mask off ophdr transaction container flags */
4355 flags &= ~XLOG_END_TRANS;
4356 if (flags & XLOG_WAS_CONT_TRANS)
4357 flags &= ~XLOG_CONTINUE_TRANS;
4360 * Callees must not free the trans structure. We'll decide if we need to
4361 * free it or not based on the operation being done and it's result.
4364 /* expected flag values */
4366 case XLOG_CONTINUE_TRANS:
4367 error = xlog_recover_add_to_trans(log, trans, dp, len);
4369 case XLOG_WAS_CONT_TRANS:
4370 error = xlog_recover_add_to_cont_trans(log, trans, dp, len);
4372 case XLOG_COMMIT_TRANS:
4373 error = xlog_recover_commit_trans(log, trans, pass,
4375 /* success or fail, we are now done with this transaction. */
4379 /* unexpected flag values */
4380 case XLOG_UNMOUNT_TRANS:
4381 /* just skip trans */
4382 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
4385 case XLOG_START_TRANS:
4387 xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags);
4389 error = -EFSCORRUPTED;
4392 if (error || freeit)
4393 xlog_recover_free_trans(trans);
4398 * Lookup the transaction recovery structure associated with the ID in the
4399 * current ophdr. If the transaction doesn't exist and the start flag is set in
4400 * the ophdr, then allocate a new transaction for future ID matches to find.
4401 * Either way, return what we found during the lookup - an existing transaction
4404 STATIC struct xlog_recover *
4405 xlog_recover_ophdr_to_trans(
4406 struct hlist_head rhash[],
4407 struct xlog_rec_header *rhead,
4408 struct xlog_op_header *ohead)
4410 struct xlog_recover *trans;
4412 struct hlist_head *rhp;
4414 tid = be32_to_cpu(ohead->oh_tid);
4415 rhp = &rhash[XLOG_RHASH(tid)];
4416 hlist_for_each_entry(trans, rhp, r_list) {
4417 if (trans->r_log_tid == tid)
4422 * skip over non-start transaction headers - we could be
4423 * processing slack space before the next transaction starts
4425 if (!(ohead->oh_flags & XLOG_START_TRANS))
4428 ASSERT(be32_to_cpu(ohead->oh_len) == 0);
4431 * This is a new transaction so allocate a new recovery container to
4432 * hold the recovery ops that will follow.
4434 trans = kmem_zalloc(sizeof(struct xlog_recover), 0);
4435 trans->r_log_tid = tid;
4436 trans->r_lsn = be64_to_cpu(rhead->h_lsn);
4437 INIT_LIST_HEAD(&trans->r_itemq);
4438 INIT_HLIST_NODE(&trans->r_list);
4439 hlist_add_head(&trans->r_list, rhp);
4442 * Nothing more to do for this ophdr. Items to be added to this new
4443 * transaction will be in subsequent ophdr containers.
4449 xlog_recover_process_ophdr(
4451 struct hlist_head rhash[],
4452 struct xlog_rec_header *rhead,
4453 struct xlog_op_header *ohead,
4457 struct list_head *buffer_list)
4459 struct xlog_recover *trans;
4463 /* Do we understand who wrote this op? */
4464 if (ohead->oh_clientid != XFS_TRANSACTION &&
4465 ohead->oh_clientid != XFS_LOG) {
4466 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
4467 __func__, ohead->oh_clientid);
4469 return -EFSCORRUPTED;
4473 * Check the ophdr contains all the data it is supposed to contain.
4475 len = be32_to_cpu(ohead->oh_len);
4476 if (dp + len > end) {
4477 xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len);
4479 return -EFSCORRUPTED;
4482 trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead);
4484 /* nothing to do, so skip over this ophdr */
4489 * The recovered buffer queue is drained only once we know that all
4490 * recovery items for the current LSN have been processed. This is
4493 * - Buffer write submission updates the metadata LSN of the buffer.
4494 * - Log recovery skips items with a metadata LSN >= the current LSN of
4495 * the recovery item.
4496 * - Separate recovery items against the same metadata buffer can share
4497 * a current LSN. I.e., consider that the LSN of a recovery item is
4498 * defined as the starting LSN of the first record in which its
4499 * transaction appears, that a record can hold multiple transactions,
4500 * and/or that a transaction can span multiple records.
4502 * In other words, we are allowed to submit a buffer from log recovery
4503 * once per current LSN. Otherwise, we may incorrectly skip recovery
4504 * items and cause corruption.
4506 * We don't know up front whether buffers are updated multiple times per
4507 * LSN. Therefore, track the current LSN of each commit log record as it
4508 * is processed and drain the queue when it changes. Use commit records
4509 * because they are ordered correctly by the logging code.
4511 if (log->l_recovery_lsn != trans->r_lsn &&
4512 ohead->oh_flags & XLOG_COMMIT_TRANS) {
4513 error = xfs_buf_delwri_submit(buffer_list);
4516 log->l_recovery_lsn = trans->r_lsn;
4519 return xlog_recovery_process_trans(log, trans, dp, len,
4520 ohead->oh_flags, pass, buffer_list);
4524 * There are two valid states of the r_state field. 0 indicates that the
4525 * transaction structure is in a normal state. We have either seen the
4526 * start of the transaction or the last operation we added was not a partial
4527 * operation. If the last operation we added to the transaction was a
4528 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
4530 * NOTE: skip LRs with 0 data length.
4533 xlog_recover_process_data(
4535 struct hlist_head rhash[],
4536 struct xlog_rec_header *rhead,
4539 struct list_head *buffer_list)
4541 struct xlog_op_header *ohead;
4546 end = dp + be32_to_cpu(rhead->h_len);
4547 num_logops = be32_to_cpu(rhead->h_num_logops);
4549 /* check the log format matches our own - else we can't recover */
4550 if (xlog_header_check_recover(log->l_mp, rhead))
4553 trace_xfs_log_recover_record(log, rhead, pass);
4554 while ((dp < end) && num_logops) {
4556 ohead = (struct xlog_op_header *)dp;
4557 dp += sizeof(*ohead);
4560 /* errors will abort recovery */
4561 error = xlog_recover_process_ophdr(log, rhash, rhead, ohead,
4562 dp, end, pass, buffer_list);
4566 dp += be32_to_cpu(ohead->oh_len);
4572 /* Recover the EFI if necessary. */
4574 xlog_recover_process_efi(
4575 struct xfs_mount *mp,
4576 struct xfs_ail *ailp,
4577 struct xfs_log_item *lip)
4579 struct xfs_efi_log_item *efip;
4583 * Skip EFIs that we've already processed.
4585 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
4586 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags))
4589 spin_unlock(&ailp->ail_lock);
4590 error = xfs_efi_recover(mp, efip);
4591 spin_lock(&ailp->ail_lock);
4596 /* Release the EFI since we're cancelling everything. */
4598 xlog_recover_cancel_efi(
4599 struct xfs_mount *mp,
4600 struct xfs_ail *ailp,
4601 struct xfs_log_item *lip)
4603 struct xfs_efi_log_item *efip;
4605 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
4607 spin_unlock(&ailp->ail_lock);
4608 xfs_efi_release(efip);
4609 spin_lock(&ailp->ail_lock);
4612 /* Recover the RUI if necessary. */
4614 xlog_recover_process_rui(
4615 struct xfs_mount *mp,
4616 struct xfs_ail *ailp,
4617 struct xfs_log_item *lip)
4619 struct xfs_rui_log_item *ruip;
4623 * Skip RUIs that we've already processed.
4625 ruip = container_of(lip, struct xfs_rui_log_item, rui_item);
4626 if (test_bit(XFS_RUI_RECOVERED, &ruip->rui_flags))
4629 spin_unlock(&ailp->ail_lock);
4630 error = xfs_rui_recover(mp, ruip);
4631 spin_lock(&ailp->ail_lock);
4636 /* Release the RUI since we're cancelling everything. */
4638 xlog_recover_cancel_rui(
4639 struct xfs_mount *mp,
4640 struct xfs_ail *ailp,
4641 struct xfs_log_item *lip)
4643 struct xfs_rui_log_item *ruip;
4645 ruip = container_of(lip, struct xfs_rui_log_item, rui_item);
4647 spin_unlock(&ailp->ail_lock);
4648 xfs_rui_release(ruip);
4649 spin_lock(&ailp->ail_lock);
4652 /* Recover the CUI if necessary. */
4654 xlog_recover_process_cui(
4655 struct xfs_trans *parent_tp,
4656 struct xfs_ail *ailp,
4657 struct xfs_log_item *lip)
4659 struct xfs_cui_log_item *cuip;
4663 * Skip CUIs that we've already processed.
4665 cuip = container_of(lip, struct xfs_cui_log_item, cui_item);
4666 if (test_bit(XFS_CUI_RECOVERED, &cuip->cui_flags))
4669 spin_unlock(&ailp->ail_lock);
4670 error = xfs_cui_recover(parent_tp, cuip);
4671 spin_lock(&ailp->ail_lock);
4676 /* Release the CUI since we're cancelling everything. */
4678 xlog_recover_cancel_cui(
4679 struct xfs_mount *mp,
4680 struct xfs_ail *ailp,
4681 struct xfs_log_item *lip)
4683 struct xfs_cui_log_item *cuip;
4685 cuip = container_of(lip, struct xfs_cui_log_item, cui_item);
4687 spin_unlock(&ailp->ail_lock);
4688 xfs_cui_release(cuip);
4689 spin_lock(&ailp->ail_lock);
4692 /* Recover the BUI if necessary. */
4694 xlog_recover_process_bui(
4695 struct xfs_trans *parent_tp,
4696 struct xfs_ail *ailp,
4697 struct xfs_log_item *lip)
4699 struct xfs_bui_log_item *buip;
4703 * Skip BUIs that we've already processed.
4705 buip = container_of(lip, struct xfs_bui_log_item, bui_item);
4706 if (test_bit(XFS_BUI_RECOVERED, &buip->bui_flags))
4709 spin_unlock(&ailp->ail_lock);
4710 error = xfs_bui_recover(parent_tp, buip);
4711 spin_lock(&ailp->ail_lock);
4716 /* Release the BUI since we're cancelling everything. */
4718 xlog_recover_cancel_bui(
4719 struct xfs_mount *mp,
4720 struct xfs_ail *ailp,
4721 struct xfs_log_item *lip)
4723 struct xfs_bui_log_item *buip;
4725 buip = container_of(lip, struct xfs_bui_log_item, bui_item);
4727 spin_unlock(&ailp->ail_lock);
4728 xfs_bui_release(buip);
4729 spin_lock(&ailp->ail_lock);
4732 /* Is this log item a deferred action intent? */
4733 static inline bool xlog_item_is_intent(struct xfs_log_item *lip)
4735 switch (lip->li_type) {
4746 /* Take all the collected deferred ops and finish them in order. */
4748 xlog_finish_defer_ops(
4749 struct xfs_trans *parent_tp)
4751 struct xfs_mount *mp = parent_tp->t_mountp;
4752 struct xfs_trans *tp;
4758 * We're finishing the defer_ops that accumulated as a result of
4759 * recovering unfinished intent items during log recovery. We
4760 * reserve an itruncate transaction because it is the largest
4761 * permanent transaction type. Since we're the only user of the fs
4762 * right now, take 93% (15/16) of the available free blocks. Use
4763 * weird math to avoid a 64-bit division.
4765 freeblks = percpu_counter_sum(&mp->m_fdblocks);
4768 resblks = min_t(int64_t, UINT_MAX, freeblks);
4769 resblks = (resblks * 15) >> 4;
4770 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, resblks,
4771 0, XFS_TRANS_RESERVE, &tp);
4774 /* transfer all collected dfops to this transaction */
4775 xfs_defer_move(tp, parent_tp);
4777 return xfs_trans_commit(tp);
4781 * When this is called, all of the log intent items which did not have
4782 * corresponding log done items should be in the AIL. What we do now
4783 * is update the data structures associated with each one.
4785 * Since we process the log intent items in normal transactions, they
4786 * will be removed at some point after the commit. This prevents us
4787 * from just walking down the list processing each one. We'll use a
4788 * flag in the intent item to skip those that we've already processed
4789 * and use the AIL iteration mechanism's generation count to try to
4790 * speed this up at least a bit.
4792 * When we start, we know that the intents are the only things in the
4793 * AIL. As we process them, however, other items are added to the
4797 xlog_recover_process_intents(
4800 struct xfs_trans *parent_tp;
4801 struct xfs_ail_cursor cur;
4802 struct xfs_log_item *lip;
4803 struct xfs_ail *ailp;
4805 #if defined(DEBUG) || defined(XFS_WARN)
4810 * The intent recovery handlers commit transactions to complete recovery
4811 * for individual intents, but any new deferred operations that are
4812 * queued during that process are held off until the very end. The
4813 * purpose of this transaction is to serve as a container for deferred
4814 * operations. Each intent recovery handler must transfer dfops here
4815 * before its local transaction commits, and we'll finish the entire
4818 error = xfs_trans_alloc_empty(log->l_mp, &parent_tp);
4823 spin_lock(&ailp->ail_lock);
4824 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
4825 #if defined(DEBUG) || defined(XFS_WARN)
4826 last_lsn = xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block);
4828 while (lip != NULL) {
4830 * We're done when we see something other than an intent.
4831 * There should be no intents left in the AIL now.
4833 if (!xlog_item_is_intent(lip)) {
4835 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
4836 ASSERT(!xlog_item_is_intent(lip));
4842 * We should never see a redo item with a LSN higher than
4843 * the last transaction we found in the log at the start
4846 ASSERT(XFS_LSN_CMP(last_lsn, lip->li_lsn) >= 0);
4849 * NOTE: If your intent processing routine can create more
4850 * deferred ops, you /must/ attach them to the dfops in this
4851 * routine or else those subsequent intents will get
4852 * replayed in the wrong order!
4854 switch (lip->li_type) {
4856 error = xlog_recover_process_efi(log->l_mp, ailp, lip);
4859 error = xlog_recover_process_rui(log->l_mp, ailp, lip);
4862 error = xlog_recover_process_cui(parent_tp, ailp, lip);
4865 error = xlog_recover_process_bui(parent_tp, ailp, lip);
4870 lip = xfs_trans_ail_cursor_next(ailp, &cur);
4873 xfs_trans_ail_cursor_done(&cur);
4874 spin_unlock(&ailp->ail_lock);
4876 error = xlog_finish_defer_ops(parent_tp);
4877 xfs_trans_cancel(parent_tp);
4883 * A cancel occurs when the mount has failed and we're bailing out.
4884 * Release all pending log intent items so they don't pin the AIL.
4887 xlog_recover_cancel_intents(
4890 struct xfs_log_item *lip;
4891 struct xfs_ail_cursor cur;
4892 struct xfs_ail *ailp;
4895 spin_lock(&ailp->ail_lock);
4896 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
4897 while (lip != NULL) {
4899 * We're done when we see something other than an intent.
4900 * There should be no intents left in the AIL now.
4902 if (!xlog_item_is_intent(lip)) {
4904 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
4905 ASSERT(!xlog_item_is_intent(lip));
4910 switch (lip->li_type) {
4912 xlog_recover_cancel_efi(log->l_mp, ailp, lip);
4915 xlog_recover_cancel_rui(log->l_mp, ailp, lip);
4918 xlog_recover_cancel_cui(log->l_mp, ailp, lip);
4921 xlog_recover_cancel_bui(log->l_mp, ailp, lip);
4925 lip = xfs_trans_ail_cursor_next(ailp, &cur);
4928 xfs_trans_ail_cursor_done(&cur);
4929 spin_unlock(&ailp->ail_lock);
4933 * This routine performs a transaction to null out a bad inode pointer
4934 * in an agi unlinked inode hash bucket.
4937 xlog_recover_clear_agi_bucket(
4939 xfs_agnumber_t agno,
4948 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_clearagi, 0, 0, 0, &tp);
4952 error = xfs_read_agi(mp, tp, agno, &agibp);
4956 agi = XFS_BUF_TO_AGI(agibp);
4957 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
4958 offset = offsetof(xfs_agi_t, agi_unlinked) +
4959 (sizeof(xfs_agino_t) * bucket);
4960 xfs_trans_log_buf(tp, agibp, offset,
4961 (offset + sizeof(xfs_agino_t) - 1));
4963 error = xfs_trans_commit(tp);
4969 xfs_trans_cancel(tp);
4971 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
4976 xlog_recover_process_one_iunlink(
4977 struct xfs_mount *mp,
4978 xfs_agnumber_t agno,
4982 struct xfs_buf *ibp;
4983 struct xfs_dinode *dip;
4984 struct xfs_inode *ip;
4988 ino = XFS_AGINO_TO_INO(mp, agno, agino);
4989 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
4994 * Get the on disk inode to find the next inode in the bucket.
4996 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
5000 xfs_iflags_clear(ip, XFS_IRECOVERY);
5001 ASSERT(VFS_I(ip)->i_nlink == 0);
5002 ASSERT(VFS_I(ip)->i_mode != 0);
5004 /* setup for the next pass */
5005 agino = be32_to_cpu(dip->di_next_unlinked);
5009 * Prevent any DMAPI event from being sent when the reference on
5010 * the inode is dropped.
5012 ip->i_d.di_dmevmask = 0;
5021 * We can't read in the inode this bucket points to, or this inode
5022 * is messed up. Just ditch this bucket of inodes. We will lose
5023 * some inodes and space, but at least we won't hang.
5025 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
5026 * clear the inode pointer in the bucket.
5028 xlog_recover_clear_agi_bucket(mp, agno, bucket);
5033 * Recover AGI unlinked lists
5035 * This is called during recovery to process any inodes which we unlinked but
5036 * not freed when the system crashed. These inodes will be on the lists in the
5037 * AGI blocks. What we do here is scan all the AGIs and fully truncate and free
5038 * any inodes found on the lists. Each inode is removed from the lists when it
5039 * has been fully truncated and is freed. The freeing of the inode and its
5040 * removal from the list must be atomic.
5042 * If everything we touch in the agi processing loop is already in memory, this
5043 * loop can hold the cpu for a long time. It runs without lock contention,
5044 * memory allocation contention, the need wait for IO, etc, and so will run
5045 * until we either run out of inodes to process, run low on memory or we run out
5048 * This behaviour is bad for latency on single CPU and non-preemptible kernels,
5049 * and can prevent other filesytem work (such as CIL pushes) from running. This
5050 * can lead to deadlocks if the recovery process runs out of log reservation
5051 * space. Hence we need to yield the CPU when there is other kernel work
5052 * scheduled on this CPU to ensure other scheduled work can run without undue
5056 xlog_recover_process_iunlinks(
5060 xfs_agnumber_t agno;
5069 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
5071 * Find the agi for this ag.
5073 error = xfs_read_agi(mp, NULL, agno, &agibp);
5076 * AGI is b0rked. Don't process it.
5078 * We should probably mark the filesystem as corrupt
5079 * after we've recovered all the ag's we can....
5084 * Unlock the buffer so that it can be acquired in the normal
5085 * course of the transaction to truncate and free each inode.
5086 * Because we are not racing with anyone else here for the AGI
5087 * buffer, we don't even need to hold it locked to read the
5088 * initial unlinked bucket entries out of the buffer. We keep
5089 * buffer reference though, so that it stays pinned in memory
5090 * while we need the buffer.
5092 agi = XFS_BUF_TO_AGI(agibp);
5093 xfs_buf_unlock(agibp);
5095 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
5096 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
5097 while (agino != NULLAGINO) {
5098 agino = xlog_recover_process_one_iunlink(mp,
5099 agno, agino, bucket);
5103 xfs_buf_rele(agibp);
5109 struct xlog_rec_header *rhead,
5115 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
5116 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
5117 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
5121 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
5122 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
5123 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
5124 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
5125 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
5126 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
5133 * CRC check, unpack and process a log record.
5136 xlog_recover_process(
5138 struct hlist_head rhash[],
5139 struct xlog_rec_header *rhead,
5142 struct list_head *buffer_list)
5144 __le32 old_crc = rhead->h_crc;
5147 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
5150 * Nothing else to do if this is a CRC verification pass. Just return
5151 * if this a record with a non-zero crc. Unfortunately, mkfs always
5152 * sets old_crc to 0 so we must consider this valid even on v5 supers.
5153 * Otherwise, return EFSBADCRC on failure so the callers up the stack
5154 * know precisely what failed.
5156 if (pass == XLOG_RECOVER_CRCPASS) {
5157 if (old_crc && crc != old_crc)
5163 * We're in the normal recovery path. Issue a warning if and only if the
5164 * CRC in the header is non-zero. This is an advisory warning and the
5165 * zero CRC check prevents warnings from being emitted when upgrading
5166 * the kernel from one that does not add CRCs by default.
5168 if (crc != old_crc) {
5169 if (old_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
5170 xfs_alert(log->l_mp,
5171 "log record CRC mismatch: found 0x%x, expected 0x%x.",
5172 le32_to_cpu(old_crc),
5174 xfs_hex_dump(dp, 32);
5178 * If the filesystem is CRC enabled, this mismatch becomes a
5179 * fatal log corruption failure.
5181 if (xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
5182 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, log->l_mp);
5183 return -EFSCORRUPTED;
5187 xlog_unpack_data(rhead, dp, log);
5189 return xlog_recover_process_data(log, rhash, rhead, dp, pass,
5194 xlog_valid_rec_header(
5196 struct xlog_rec_header *rhead,
5201 if (XFS_IS_CORRUPT(log->l_mp,
5202 rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM)))
5203 return -EFSCORRUPTED;
5204 if (XFS_IS_CORRUPT(log->l_mp,
5205 (!rhead->h_version ||
5206 (be32_to_cpu(rhead->h_version) &
5207 (~XLOG_VERSION_OKBITS))))) {
5208 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
5209 __func__, be32_to_cpu(rhead->h_version));
5210 return -EFSCORRUPTED;
5213 /* LR body must have data or it wouldn't have been written */
5214 hlen = be32_to_cpu(rhead->h_len);
5215 if (XFS_IS_CORRUPT(log->l_mp, hlen <= 0 || hlen > INT_MAX))
5216 return -EFSCORRUPTED;
5217 if (XFS_IS_CORRUPT(log->l_mp,
5218 blkno > log->l_logBBsize || blkno > INT_MAX))
5219 return -EFSCORRUPTED;
5224 * Read the log from tail to head and process the log records found.
5225 * Handle the two cases where the tail and head are in the same cycle
5226 * and where the active portion of the log wraps around the end of
5227 * the physical log separately. The pass parameter is passed through
5228 * to the routines called to process the data and is not looked at
5232 xlog_do_recovery_pass(
5234 xfs_daddr_t head_blk,
5235 xfs_daddr_t tail_blk,
5237 xfs_daddr_t *first_bad) /* out: first bad log rec */
5239 xlog_rec_header_t *rhead;
5240 xfs_daddr_t blk_no, rblk_no;
5241 xfs_daddr_t rhead_blk;
5244 int error = 0, h_size, h_len;
5246 int bblks, split_bblks;
5247 int hblks, split_hblks, wrapped_hblks;
5249 struct hlist_head rhash[XLOG_RHASH_SIZE];
5250 LIST_HEAD (buffer_list);
5252 ASSERT(head_blk != tail_blk);
5253 blk_no = rhead_blk = tail_blk;
5255 for (i = 0; i < XLOG_RHASH_SIZE; i++)
5256 INIT_HLIST_HEAD(&rhash[i]);
5259 * Read the header of the tail block and get the iclog buffer size from
5260 * h_size. Use this to tell how many sectors make up the log header.
5262 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
5264 * When using variable length iclogs, read first sector of
5265 * iclog header and extract the header size from it. Get a
5266 * new hbp that is the correct size.
5268 hbp = xlog_alloc_buffer(log, 1);
5272 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
5276 rhead = (xlog_rec_header_t *)offset;
5277 error = xlog_valid_rec_header(log, rhead, tail_blk);
5282 * xfsprogs has a bug where record length is based on lsunit but
5283 * h_size (iclog size) is hardcoded to 32k. Now that we
5284 * unconditionally CRC verify the unmount record, this means the
5285 * log buffer can be too small for the record and cause an
5288 * Detect this condition here. Use lsunit for the buffer size as
5289 * long as this looks like the mkfs case. Otherwise, return an
5290 * error to avoid a buffer overrun.
5292 h_size = be32_to_cpu(rhead->h_size);
5293 h_len = be32_to_cpu(rhead->h_len);
5294 if (h_len > h_size) {
5295 if (h_len <= log->l_mp->m_logbsize &&
5296 be32_to_cpu(rhead->h_num_logops) == 1) {
5298 "invalid iclog size (%d bytes), using lsunit (%d bytes)",
5299 h_size, log->l_mp->m_logbsize);
5300 h_size = log->l_mp->m_logbsize;
5302 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW,
5304 error = -EFSCORRUPTED;
5309 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
5310 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
5311 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
5312 if (h_size % XLOG_HEADER_CYCLE_SIZE)
5315 hbp = xlog_alloc_buffer(log, hblks);
5320 ASSERT(log->l_sectBBsize == 1);
5322 hbp = xlog_alloc_buffer(log, 1);
5323 h_size = XLOG_BIG_RECORD_BSIZE;
5328 dbp = xlog_alloc_buffer(log, BTOBB(h_size));
5334 memset(rhash, 0, sizeof(rhash));
5335 if (tail_blk > head_blk) {
5337 * Perform recovery around the end of the physical log.
5338 * When the head is not on the same cycle number as the tail,
5339 * we can't do a sequential recovery.
5341 while (blk_no < log->l_logBBsize) {
5343 * Check for header wrapping around physical end-of-log
5348 if (blk_no + hblks <= log->l_logBBsize) {
5349 /* Read header in one read */
5350 error = xlog_bread(log, blk_no, hblks, hbp,
5355 /* This LR is split across physical log end */
5356 if (blk_no != log->l_logBBsize) {
5357 /* some data before physical log end */
5358 ASSERT(blk_no <= INT_MAX);
5359 split_hblks = log->l_logBBsize - (int)blk_no;
5360 ASSERT(split_hblks > 0);
5361 error = xlog_bread(log, blk_no,
5369 * Note: this black magic still works with
5370 * large sector sizes (non-512) only because:
5371 * - we increased the buffer size originally
5372 * by 1 sector giving us enough extra space
5373 * for the second read;
5374 * - the log start is guaranteed to be sector
5376 * - we read the log end (LR header start)
5377 * _first_, then the log start (LR header end)
5378 * - order is important.
5380 wrapped_hblks = hblks - split_hblks;
5381 error = xlog_bread_noalign(log, 0,
5383 offset + BBTOB(split_hblks));
5387 rhead = (xlog_rec_header_t *)offset;
5388 error = xlog_valid_rec_header(log, rhead,
5389 split_hblks ? blk_no : 0);
5393 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
5397 * Read the log record data in multiple reads if it
5398 * wraps around the end of the log. Note that if the
5399 * header already wrapped, blk_no could point past the
5400 * end of the log. The record data is contiguous in
5403 if (blk_no + bblks <= log->l_logBBsize ||
5404 blk_no >= log->l_logBBsize) {
5405 rblk_no = xlog_wrap_logbno(log, blk_no);
5406 error = xlog_bread(log, rblk_no, bblks, dbp,
5411 /* This log record is split across the
5412 * physical end of log */
5415 if (blk_no != log->l_logBBsize) {
5416 /* some data is before the physical
5418 ASSERT(!wrapped_hblks);
5419 ASSERT(blk_no <= INT_MAX);
5421 log->l_logBBsize - (int)blk_no;
5422 ASSERT(split_bblks > 0);
5423 error = xlog_bread(log, blk_no,
5431 * Note: this black magic still works with
5432 * large sector sizes (non-512) only because:
5433 * - we increased the buffer size originally
5434 * by 1 sector giving us enough extra space
5435 * for the second read;
5436 * - the log start is guaranteed to be sector
5438 * - we read the log end (LR header start)
5439 * _first_, then the log start (LR header end)
5440 * - order is important.
5442 error = xlog_bread_noalign(log, 0,
5443 bblks - split_bblks,
5444 offset + BBTOB(split_bblks));
5449 error = xlog_recover_process(log, rhash, rhead, offset,
5450 pass, &buffer_list);
5458 ASSERT(blk_no >= log->l_logBBsize);
5459 blk_no -= log->l_logBBsize;
5463 /* read first part of physical log */
5464 while (blk_no < head_blk) {
5465 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
5469 rhead = (xlog_rec_header_t *)offset;
5470 error = xlog_valid_rec_header(log, rhead, blk_no);
5474 /* blocks in data section */
5475 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
5476 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
5481 error = xlog_recover_process(log, rhash, rhead, offset, pass,
5486 blk_no += bblks + hblks;
5496 * Submit buffers that have been added from the last record processed,
5497 * regardless of error status.
5499 if (!list_empty(&buffer_list))
5500 error2 = xfs_buf_delwri_submit(&buffer_list);
5502 if (error && first_bad)
5503 *first_bad = rhead_blk;
5506 * Transactions are freed at commit time but transactions without commit
5507 * records on disk are never committed. Free any that may be left in the
5510 for (i = 0; i < XLOG_RHASH_SIZE; i++) {
5511 struct hlist_node *tmp;
5512 struct xlog_recover *trans;
5514 hlist_for_each_entry_safe(trans, tmp, &rhash[i], r_list)
5515 xlog_recover_free_trans(trans);
5518 return error ? error : error2;
5522 * Do the recovery of the log. We actually do this in two phases.
5523 * The two passes are necessary in order to implement the function
5524 * of cancelling a record written into the log. The first pass
5525 * determines those things which have been cancelled, and the
5526 * second pass replays log items normally except for those which
5527 * have been cancelled. The handling of the replay and cancellations
5528 * takes place in the log item type specific routines.
5530 * The table of items which have cancel records in the log is allocated
5531 * and freed at this level, since only here do we know when all of
5532 * the log recovery has been completed.
5535 xlog_do_log_recovery(
5537 xfs_daddr_t head_blk,
5538 xfs_daddr_t tail_blk)
5542 ASSERT(head_blk != tail_blk);
5545 * First do a pass to find all of the cancelled buf log items.
5546 * Store them in the buf_cancel_table for use in the second pass.
5548 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
5549 sizeof(struct list_head),
5551 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
5552 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
5554 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
5555 XLOG_RECOVER_PASS1, NULL);
5557 kmem_free(log->l_buf_cancel_table);
5558 log->l_buf_cancel_table = NULL;
5562 * Then do a second pass to actually recover the items in the log.
5563 * When it is complete free the table of buf cancel items.
5565 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
5566 XLOG_RECOVER_PASS2, NULL);
5571 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
5572 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
5576 kmem_free(log->l_buf_cancel_table);
5577 log->l_buf_cancel_table = NULL;
5583 * Do the actual recovery
5588 xfs_daddr_t head_blk,
5589 xfs_daddr_t tail_blk)
5591 struct xfs_mount *mp = log->l_mp;
5596 trace_xfs_log_recover(log, head_blk, tail_blk);
5599 * First replay the images in the log.
5601 error = xlog_do_log_recovery(log, head_blk, tail_blk);
5606 * If IO errors happened during recovery, bail out.
5608 if (XFS_FORCED_SHUTDOWN(mp)) {
5613 * We now update the tail_lsn since much of the recovery has completed
5614 * and there may be space available to use. If there were no extent
5615 * or iunlinks, we can free up the entire log and set the tail_lsn to
5616 * be the last_sync_lsn. This was set in xlog_find_tail to be the
5617 * lsn of the last known good LR on disk. If there are extent frees
5618 * or iunlinks they will have some entries in the AIL; so we look at
5619 * the AIL to determine how to set the tail_lsn.
5621 xlog_assign_tail_lsn(mp);
5624 * Now that we've finished replaying all buffer and inode
5625 * updates, re-read in the superblock and reverify it.
5628 bp->b_flags &= ~(XBF_DONE | XBF_ASYNC);
5629 ASSERT(!(bp->b_flags & XBF_WRITE));
5630 bp->b_flags |= XBF_READ;
5631 bp->b_ops = &xfs_sb_buf_ops;
5633 error = xfs_buf_submit(bp);
5635 if (!XFS_FORCED_SHUTDOWN(mp)) {
5636 xfs_buf_ioerror_alert(bp, __func__);
5643 /* Convert superblock from on-disk format */
5645 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
5648 /* re-initialise in-core superblock and geometry structures */
5649 xfs_reinit_percpu_counters(mp);
5650 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
5652 xfs_warn(mp, "Failed post-recovery per-ag init: %d", error);
5655 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
5657 xlog_recover_check_summary(log);
5659 /* Normal transactions can now occur */
5660 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
5665 * Perform recovery and re-initialize some log variables in xlog_find_tail.
5667 * Return error or zero.
5673 xfs_daddr_t head_blk, tail_blk;
5676 /* find the tail of the log */
5677 error = xlog_find_tail(log, &head_blk, &tail_blk);
5682 * The superblock was read before the log was available and thus the LSN
5683 * could not be verified. Check the superblock LSN against the current
5684 * LSN now that it's known.
5686 if (xfs_sb_version_hascrc(&log->l_mp->m_sb) &&
5687 !xfs_log_check_lsn(log->l_mp, log->l_mp->m_sb.sb_lsn))
5690 if (tail_blk != head_blk) {
5691 /* There used to be a comment here:
5693 * disallow recovery on read-only mounts. note -- mount
5694 * checks for ENOSPC and turns it into an intelligent
5696 * ...but this is no longer true. Now, unless you specify
5697 * NORECOVERY (in which case this function would never be
5698 * called), we just go ahead and recover. We do this all
5699 * under the vfs layer, so we can get away with it unless
5700 * the device itself is read-only, in which case we fail.
5702 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
5707 * Version 5 superblock log feature mask validation. We know the
5708 * log is dirty so check if there are any unknown log features
5709 * in what we need to recover. If there are unknown features
5710 * (e.g. unsupported transactions, then simply reject the
5711 * attempt at recovery before touching anything.
5713 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
5714 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
5715 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
5717 "Superblock has unknown incompatible log features (0x%x) enabled.",
5718 (log->l_mp->m_sb.sb_features_log_incompat &
5719 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
5721 "The log can not be fully and/or safely recovered by this kernel.");
5723 "Please recover the log on a kernel that supports the unknown features.");
5728 * Delay log recovery if the debug hook is set. This is debug
5729 * instrumention to coordinate simulation of I/O failures with
5732 if (xfs_globals.log_recovery_delay) {
5733 xfs_notice(log->l_mp,
5734 "Delaying log recovery for %d seconds.",
5735 xfs_globals.log_recovery_delay);
5736 msleep(xfs_globals.log_recovery_delay * 1000);
5739 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
5740 log->l_mp->m_logname ? log->l_mp->m_logname
5743 error = xlog_do_recover(log, head_blk, tail_blk);
5744 log->l_flags |= XLOG_RECOVERY_NEEDED;
5750 * In the first part of recovery we replay inodes and buffers and build
5751 * up the list of extent free items which need to be processed. Here
5752 * we process the extent free items and clean up the on disk unlinked
5753 * inode lists. This is separated from the first part of recovery so
5754 * that the root and real-time bitmap inodes can be read in from disk in
5755 * between the two stages. This is necessary so that we can free space
5756 * in the real-time portion of the file system.
5759 xlog_recover_finish(
5763 * Now we're ready to do the transactions needed for the
5764 * rest of recovery. Start with completing all the extent
5765 * free intent records and then process the unlinked inode
5766 * lists. At this point, we essentially run in normal mode
5767 * except that we're still performing recovery actions
5768 * rather than accepting new requests.
5770 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
5772 error = xlog_recover_process_intents(log);
5774 xfs_alert(log->l_mp, "Failed to recover intents");
5779 * Sync the log to get all the intents out of the AIL.
5780 * This isn't absolutely necessary, but it helps in
5781 * case the unlink transactions would have problems
5782 * pushing the intents out of the way.
5784 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
5786 xlog_recover_process_iunlinks(log);
5788 xlog_recover_check_summary(log);
5790 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
5791 log->l_mp->m_logname ? log->l_mp->m_logname
5793 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
5795 xfs_info(log->l_mp, "Ending clean mount");
5801 xlog_recover_cancel(
5804 if (log->l_flags & XLOG_RECOVERY_NEEDED)
5805 xlog_recover_cancel_intents(log);
5810 * Read all of the agf and agi counters and check that they
5811 * are consistent with the superblock counters.
5814 xlog_recover_check_summary(
5821 xfs_agnumber_t agno;
5832 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
5833 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
5835 xfs_alert(mp, "%s agf read failed agno %d error %d",
5836 __func__, agno, error);
5838 agfp = XFS_BUF_TO_AGF(agfbp);
5839 freeblks += be32_to_cpu(agfp->agf_freeblks) +
5840 be32_to_cpu(agfp->agf_flcount);
5841 xfs_buf_relse(agfbp);
5844 error = xfs_read_agi(mp, NULL, agno, &agibp);
5846 xfs_alert(mp, "%s agi read failed agno %d error %d",
5847 __func__, agno, error);
5849 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
5851 itotal += be32_to_cpu(agi->agi_count);
5852 ifree += be32_to_cpu(agi->agi_freecount);
5853 xfs_buf_relse(agibp);