2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
27 #include "xfs_mount.h"
28 #include "xfs_error.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_alloc.h"
36 #include "xfs_ialloc.h"
37 #include "xfs_log_priv.h"
38 #include "xfs_buf_item.h"
39 #include "xfs_log_recover.h"
40 #include "xfs_extfree_item.h"
41 #include "xfs_trans_priv.h"
42 #include "xfs_quota.h"
44 #include "xfs_utils.h"
45 #include "xfs_trace.h"
47 STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *);
48 STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t);
50 STATIC void xlog_recover_check_summary(xlog_t *);
52 #define xlog_recover_check_summary(log)
56 * This structure is used during recovery to record the buf log items which
57 * have been canceled and should not be replayed.
59 struct xfs_buf_cancel {
63 struct list_head bc_list;
67 * Sector aligned buffer routines for buffer create/read/write/access
71 * Verify the given count of basic blocks is valid number of blocks
72 * to specify for an operation involving the given XFS log buffer.
73 * Returns nonzero if the count is valid, 0 otherwise.
77 xlog_buf_bbcount_valid(
81 return bbcount > 0 && bbcount <= log->l_logBBsize;
85 * Allocate a buffer to hold log data. The buffer needs to be able
86 * to map to a range of nbblks basic blocks at any valid (basic
87 * block) offset within the log.
94 if (!xlog_buf_bbcount_valid(log, nbblks)) {
95 xlog_warn("XFS: Invalid block length (0x%x) given for buffer",
97 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
102 * We do log I/O in units of log sectors (a power-of-2
103 * multiple of the basic block size), so we round up the
104 * requested size to acommodate the basic blocks required
105 * for complete log sectors.
107 * In addition, the buffer may be used for a non-sector-
108 * aligned block offset, in which case an I/O of the
109 * requested size could extend beyond the end of the
110 * buffer. If the requested size is only 1 basic block it
111 * will never straddle a sector boundary, so this won't be
112 * an issue. Nor will this be a problem if the log I/O is
113 * done in basic blocks (sector size 1). But otherwise we
114 * extend the buffer by one extra log sector to ensure
115 * there's space to accomodate this possiblility.
117 if (nbblks > 1 && log->l_sectBBsize > 1)
118 nbblks += log->l_sectBBsize;
119 nbblks = round_up(nbblks, log->l_sectBBsize);
121 return xfs_buf_get_uncached(log->l_mp->m_logdev_targp,
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.
143 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
145 ASSERT(BBTOB(offset + nbblks) <= XFS_BUF_SIZE(bp));
146 return XFS_BUF_PTR(bp) + BBTOB(offset);
151 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
162 if (!xlog_buf_bbcount_valid(log, nbblks)) {
163 xlog_warn("XFS: Invalid block length (0x%x) given for buffer",
165 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
169 blk_no = round_down(blk_no, log->l_sectBBsize);
170 nbblks = round_up(nbblks, log->l_sectBBsize);
173 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
175 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
178 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
179 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
181 xfsbdstrat(log->l_mp, bp);
182 error = xfs_buf_iowait(bp);
184 xfs_ioerror_alert("xlog_bread", log->l_mp,
185 bp, XFS_BUF_ADDR(bp));
199 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
203 *offset = xlog_align(log, blk_no, nbblks, bp);
208 * Write out the buffer at the given block for the given number of blocks.
209 * The buffer is kept locked across the write and is returned locked.
210 * This can only be used for synchronous log writes.
221 if (!xlog_buf_bbcount_valid(log, nbblks)) {
222 xlog_warn("XFS: Invalid block length (0x%x) given for buffer",
224 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
228 blk_no = round_down(blk_no, log->l_sectBBsize);
229 nbblks = round_up(nbblks, log->l_sectBBsize);
232 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
234 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
235 XFS_BUF_ZEROFLAGS(bp);
238 XFS_BUF_PSEMA(bp, PRIBIO);
239 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
240 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
242 if ((error = xfs_bwrite(log->l_mp, bp)))
243 xfs_ioerror_alert("xlog_bwrite", log->l_mp,
244 bp, XFS_BUF_ADDR(bp));
250 * dump debug superblock and log record information
253 xlog_header_check_dump(
255 xlog_rec_header_t *head)
257 cmn_err(CE_DEBUG, "%s: SB : uuid = %pU, fmt = %d\n",
258 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
259 cmn_err(CE_DEBUG, " log : uuid = %pU, fmt = %d\n",
260 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
263 #define xlog_header_check_dump(mp, head)
267 * check log record header for recovery
270 xlog_header_check_recover(
272 xlog_rec_header_t *head)
274 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
277 * IRIX doesn't write the h_fmt field and leaves it zeroed
278 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
279 * a dirty log created in IRIX.
281 if (unlikely(be32_to_cpu(head->h_fmt) != XLOG_FMT)) {
283 "XFS: dirty log written in incompatible format - can't recover");
284 xlog_header_check_dump(mp, head);
285 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
286 XFS_ERRLEVEL_HIGH, mp);
287 return XFS_ERROR(EFSCORRUPTED);
288 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
290 "XFS: dirty log entry has mismatched uuid - can't recover");
291 xlog_header_check_dump(mp, head);
292 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
293 XFS_ERRLEVEL_HIGH, mp);
294 return XFS_ERROR(EFSCORRUPTED);
300 * read the head block of the log and check the header
303 xlog_header_check_mount(
305 xlog_rec_header_t *head)
307 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
309 if (uuid_is_nil(&head->h_fs_uuid)) {
311 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
312 * h_fs_uuid is nil, we assume this log was last mounted
313 * by IRIX and continue.
315 xlog_warn("XFS: nil uuid in log - IRIX style log");
316 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
317 xlog_warn("XFS: log has mismatched uuid - can't recover");
318 xlog_header_check_dump(mp, head);
319 XFS_ERROR_REPORT("xlog_header_check_mount",
320 XFS_ERRLEVEL_HIGH, mp);
321 return XFS_ERROR(EFSCORRUPTED);
330 if (XFS_BUF_GETERROR(bp)) {
332 * We're not going to bother about retrying
333 * this during recovery. One strike!
335 xfs_ioerror_alert("xlog_recover_iodone",
336 bp->b_target->bt_mount, bp,
338 xfs_force_shutdown(bp->b_target->bt_mount,
339 SHUTDOWN_META_IO_ERROR);
341 XFS_BUF_CLR_IODONE_FUNC(bp);
342 xfs_buf_ioend(bp, 0);
346 * This routine finds (to an approximation) the first block in the physical
347 * log which contains the given cycle. It uses a binary search algorithm.
348 * Note that the algorithm can not be perfect because the disk will not
349 * necessarily be perfect.
352 xlog_find_cycle_start(
355 xfs_daddr_t first_blk,
356 xfs_daddr_t *last_blk,
366 mid_blk = BLK_AVG(first_blk, end_blk);
367 while (mid_blk != first_blk && mid_blk != end_blk) {
368 error = xlog_bread(log, mid_blk, 1, bp, &offset);
371 mid_cycle = xlog_get_cycle(offset);
372 if (mid_cycle == cycle)
373 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
375 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
376 mid_blk = BLK_AVG(first_blk, end_blk);
378 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
379 (mid_blk == end_blk && mid_blk-1 == first_blk));
387 * Check that a range of blocks does not contain stop_on_cycle_no.
388 * Fill in *new_blk with the block offset where such a block is
389 * found, or with -1 (an invalid block number) if there is no such
390 * block in the range. The scan needs to occur from front to back
391 * and the pointer into the region must be updated since a later
392 * routine will need to perform another test.
395 xlog_find_verify_cycle(
397 xfs_daddr_t start_blk,
399 uint stop_on_cycle_no,
400 xfs_daddr_t *new_blk)
406 xfs_caddr_t buf = NULL;
410 * Greedily allocate a buffer big enough to handle the full
411 * range of basic blocks we'll be examining. If that fails,
412 * try a smaller size. We need to be able to read at least
413 * a log sector, or we're out of luck.
415 bufblks = 1 << ffs(nbblks);
416 while (!(bp = xlog_get_bp(log, bufblks))) {
418 if (bufblks < log->l_sectBBsize)
422 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
425 bcount = min(bufblks, (start_blk + nbblks - i));
427 error = xlog_bread(log, i, bcount, bp, &buf);
431 for (j = 0; j < bcount; j++) {
432 cycle = xlog_get_cycle(buf);
433 if (cycle == stop_on_cycle_no) {
450 * Potentially backup over partial log record write.
452 * In the typical case, last_blk is the number of the block directly after
453 * a good log record. Therefore, we subtract one to get the block number
454 * of the last block in the given buffer. extra_bblks contains the number
455 * of blocks we would have read on a previous read. This happens when the
456 * last log record is split over the end of the physical log.
458 * extra_bblks is the number of blocks potentially verified on a previous
459 * call to this routine.
462 xlog_find_verify_log_record(
464 xfs_daddr_t start_blk,
465 xfs_daddr_t *last_blk,
470 xfs_caddr_t offset = NULL;
471 xlog_rec_header_t *head = NULL;
474 int num_blks = *last_blk - start_blk;
477 ASSERT(start_blk != 0 || *last_blk != start_blk);
479 if (!(bp = xlog_get_bp(log, num_blks))) {
480 if (!(bp = xlog_get_bp(log, 1)))
484 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
487 offset += ((num_blks - 1) << BBSHIFT);
490 for (i = (*last_blk) - 1; i >= 0; i--) {
492 /* valid log record not found */
494 "XFS: Log inconsistent (didn't find previous header)");
496 error = XFS_ERROR(EIO);
501 error = xlog_bread(log, i, 1, bp, &offset);
506 head = (xlog_rec_header_t *)offset;
508 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(head->h_magicno))
516 * We hit the beginning of the physical log & still no header. Return
517 * to caller. If caller can handle a return of -1, then this routine
518 * will be called again for the end of the physical log.
526 * We have the final block of the good log (the first block
527 * of the log record _before_ the head. So we check the uuid.
529 if ((error = xlog_header_check_mount(log->l_mp, head)))
533 * We may have found a log record header before we expected one.
534 * last_blk will be the 1st block # with a given cycle #. We may end
535 * up reading an entire log record. In this case, we don't want to
536 * reset last_blk. Only when last_blk points in the middle of a log
537 * record do we update last_blk.
539 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
540 uint h_size = be32_to_cpu(head->h_size);
542 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
543 if (h_size % XLOG_HEADER_CYCLE_SIZE)
549 if (*last_blk - i + extra_bblks !=
550 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
559 * Head is defined to be the point of the log where the next log write
560 * write could go. This means that incomplete LR writes at the end are
561 * eliminated when calculating the head. We aren't guaranteed that previous
562 * LR have complete transactions. We only know that a cycle number of
563 * current cycle number -1 won't be present in the log if we start writing
564 * from our current block number.
566 * last_blk contains the block number of the first block with a given
569 * Return: zero if normal, non-zero if error.
574 xfs_daddr_t *return_head_blk)
578 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
580 uint first_half_cycle, last_half_cycle;
582 int error, log_bbnum = log->l_logBBsize;
584 /* Is the end of the log device zeroed? */
585 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
586 *return_head_blk = first_blk;
588 /* Is the whole lot zeroed? */
590 /* Linux XFS shouldn't generate totally zeroed logs -
591 * mkfs etc write a dummy unmount record to a fresh
592 * log so we can store the uuid in there
594 xlog_warn("XFS: totally zeroed log");
599 xlog_warn("XFS: empty log check failed");
603 first_blk = 0; /* get cycle # of 1st block */
604 bp = xlog_get_bp(log, 1);
608 error = xlog_bread(log, 0, 1, bp, &offset);
612 first_half_cycle = xlog_get_cycle(offset);
614 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
615 error = xlog_bread(log, last_blk, 1, bp, &offset);
619 last_half_cycle = xlog_get_cycle(offset);
620 ASSERT(last_half_cycle != 0);
623 * If the 1st half cycle number is equal to the last half cycle number,
624 * then the entire log is stamped with the same cycle number. In this
625 * case, head_blk can't be set to zero (which makes sense). The below
626 * math doesn't work out properly with head_blk equal to zero. Instead,
627 * we set it to log_bbnum which is an invalid block number, but this
628 * value makes the math correct. If head_blk doesn't changed through
629 * all the tests below, *head_blk is set to zero at the very end rather
630 * than log_bbnum. In a sense, log_bbnum and zero are the same block
631 * in a circular file.
633 if (first_half_cycle == last_half_cycle) {
635 * In this case we believe that the entire log should have
636 * cycle number last_half_cycle. We need to scan backwards
637 * from the end verifying that there are no holes still
638 * containing last_half_cycle - 1. If we find such a hole,
639 * then the start of that hole will be the new head. The
640 * simple case looks like
641 * x | x ... | x - 1 | x
642 * Another case that fits this picture would be
643 * x | x + 1 | x ... | x
644 * In this case the head really is somewhere at the end of the
645 * log, as one of the latest writes at the beginning was
648 * x | x + 1 | x ... | x - 1 | x
649 * This is really the combination of the above two cases, and
650 * the head has to end up at the start of the x-1 hole at the
653 * In the 256k log case, we will read from the beginning to the
654 * end of the log and search for cycle numbers equal to x-1.
655 * We don't worry about the x+1 blocks that we encounter,
656 * because we know that they cannot be the head since the log
659 head_blk = log_bbnum;
660 stop_on_cycle = last_half_cycle - 1;
663 * In this case we want to find the first block with cycle
664 * number matching last_half_cycle. We expect the log to be
666 * x + 1 ... | x ... | x
667 * The first block with cycle number x (last_half_cycle) will
668 * be where the new head belongs. First we do a binary search
669 * for the first occurrence of last_half_cycle. The binary
670 * search may not be totally accurate, so then we scan back
671 * from there looking for occurrences of last_half_cycle before
672 * us. If that backwards scan wraps around the beginning of
673 * the log, then we look for occurrences of last_half_cycle - 1
674 * at the end of the log. The cases we're looking for look
676 * v binary search stopped here
677 * x + 1 ... | x | x + 1 | x ... | x
678 * ^ but we want to locate this spot
680 * <---------> less than scan distance
681 * x + 1 ... | x ... | x - 1 | x
682 * ^ we want to locate this spot
684 stop_on_cycle = last_half_cycle;
685 if ((error = xlog_find_cycle_start(log, bp, first_blk,
686 &head_blk, last_half_cycle)))
691 * Now validate the answer. Scan back some number of maximum possible
692 * blocks and make sure each one has the expected cycle number. The
693 * maximum is determined by the total possible amount of buffering
694 * in the in-core log. The following number can be made tighter if
695 * we actually look at the block size of the filesystem.
697 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
698 if (head_blk >= num_scan_bblks) {
700 * We are guaranteed that the entire check can be performed
703 start_blk = head_blk - num_scan_bblks;
704 if ((error = xlog_find_verify_cycle(log,
705 start_blk, num_scan_bblks,
706 stop_on_cycle, &new_blk)))
710 } else { /* need to read 2 parts of log */
712 * We are going to scan backwards in the log in two parts.
713 * First we scan the physical end of the log. In this part
714 * of the log, we are looking for blocks with cycle number
715 * last_half_cycle - 1.
716 * If we find one, then we know that the log starts there, as
717 * we've found a hole that didn't get written in going around
718 * the end of the physical log. The simple case for this is
719 * x + 1 ... | x ... | x - 1 | x
720 * <---------> less than scan distance
721 * If all of the blocks at the end of the log have cycle number
722 * last_half_cycle, then we check the blocks at the start of
723 * the log looking for occurrences of last_half_cycle. If we
724 * find one, then our current estimate for the location of the
725 * first occurrence of last_half_cycle is wrong and we move
726 * back to the hole we've found. This case looks like
727 * x + 1 ... | x | x + 1 | x ...
728 * ^ binary search stopped here
729 * Another case we need to handle that only occurs in 256k
731 * x + 1 ... | x ... | x+1 | x ...
732 * ^ binary search stops here
733 * In a 256k log, the scan at the end of the log will see the
734 * x + 1 blocks. We need to skip past those since that is
735 * certainly not the head of the log. By searching for
736 * last_half_cycle-1 we accomplish that.
738 ASSERT(head_blk <= INT_MAX &&
739 (xfs_daddr_t) num_scan_bblks >= head_blk);
740 start_blk = log_bbnum - (num_scan_bblks - head_blk);
741 if ((error = xlog_find_verify_cycle(log, start_blk,
742 num_scan_bblks - (int)head_blk,
743 (stop_on_cycle - 1), &new_blk)))
751 * Scan beginning of log now. The last part of the physical
752 * log is good. This scan needs to verify that it doesn't find
753 * the last_half_cycle.
756 ASSERT(head_blk <= INT_MAX);
757 if ((error = xlog_find_verify_cycle(log,
758 start_blk, (int)head_blk,
759 stop_on_cycle, &new_blk)))
767 * Now we need to make sure head_blk is not pointing to a block in
768 * the middle of a log record.
770 num_scan_bblks = XLOG_REC_SHIFT(log);
771 if (head_blk >= num_scan_bblks) {
772 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
774 /* start ptr at last block ptr before head_blk */
775 if ((error = xlog_find_verify_log_record(log, start_blk,
776 &head_blk, 0)) == -1) {
777 error = XFS_ERROR(EIO);
783 ASSERT(head_blk <= INT_MAX);
784 if ((error = xlog_find_verify_log_record(log, start_blk,
785 &head_blk, 0)) == -1) {
786 /* We hit the beginning of the log during our search */
787 start_blk = log_bbnum - (num_scan_bblks - head_blk);
789 ASSERT(start_blk <= INT_MAX &&
790 (xfs_daddr_t) log_bbnum-start_blk >= 0);
791 ASSERT(head_blk <= INT_MAX);
792 if ((error = xlog_find_verify_log_record(log,
794 (int)head_blk)) == -1) {
795 error = XFS_ERROR(EIO);
799 if (new_blk != log_bbnum)
806 if (head_blk == log_bbnum)
807 *return_head_blk = 0;
809 *return_head_blk = head_blk;
811 * When returning here, we have a good block number. Bad block
812 * means that during a previous crash, we didn't have a clean break
813 * from cycle number N to cycle number N-1. In this case, we need
814 * to find the first block with cycle number N-1.
822 xlog_warn("XFS: failed to find log head");
827 * Find the sync block number or the tail of the log.
829 * This will be the block number of the last record to have its
830 * associated buffers synced to disk. Every log record header has
831 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
832 * to get a sync block number. The only concern is to figure out which
833 * log record header to believe.
835 * The following algorithm uses the log record header with the largest
836 * lsn. The entire log record does not need to be valid. We only care
837 * that the header is valid.
839 * We could speed up search by using current head_blk buffer, but it is not
845 xfs_daddr_t *head_blk,
846 xfs_daddr_t *tail_blk)
848 xlog_rec_header_t *rhead;
849 xlog_op_header_t *op_head;
850 xfs_caddr_t offset = NULL;
853 xfs_daddr_t umount_data_blk;
854 xfs_daddr_t after_umount_blk;
861 * Find previous log record
863 if ((error = xlog_find_head(log, head_blk)))
866 bp = xlog_get_bp(log, 1);
869 if (*head_blk == 0) { /* special case */
870 error = xlog_bread(log, 0, 1, bp, &offset);
874 if (xlog_get_cycle(offset) == 0) {
876 /* leave all other log inited values alone */
882 * Search backwards looking for log record header block
884 ASSERT(*head_blk < INT_MAX);
885 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
886 error = xlog_bread(log, i, 1, bp, &offset);
890 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(*(__be32 *)offset)) {
896 * If we haven't found the log record header block, start looking
897 * again from the end of the physical log. XXXmiken: There should be
898 * a check here to make sure we didn't search more than N blocks in
902 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
903 error = xlog_bread(log, i, 1, bp, &offset);
907 if (XLOG_HEADER_MAGIC_NUM ==
908 be32_to_cpu(*(__be32 *)offset)) {
915 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
917 return XFS_ERROR(EIO);
920 /* find blk_no of tail of log */
921 rhead = (xlog_rec_header_t *)offset;
922 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
925 * Reset log values according to the state of the log when we
926 * crashed. In the case where head_blk == 0, we bump curr_cycle
927 * one because the next write starts a new cycle rather than
928 * continuing the cycle of the last good log record. At this
929 * point we have guaranteed that all partial log records have been
930 * accounted for. Therefore, we know that the last good log record
931 * written was complete and ended exactly on the end boundary
932 * of the physical log.
934 log->l_prev_block = i;
935 log->l_curr_block = (int)*head_blk;
936 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
939 log->l_tail_lsn = be64_to_cpu(rhead->h_tail_lsn);
940 log->l_last_sync_lsn = be64_to_cpu(rhead->h_lsn);
941 log->l_grant_reserve_cycle = log->l_curr_cycle;
942 log->l_grant_reserve_bytes = BBTOB(log->l_curr_block);
943 log->l_grant_write_cycle = log->l_curr_cycle;
944 log->l_grant_write_bytes = BBTOB(log->l_curr_block);
947 * Look for unmount record. If we find it, then we know there
948 * was a clean unmount. Since 'i' could be the last block in
949 * the physical log, we convert to a log block before comparing
952 * Save the current tail lsn to use to pass to
953 * xlog_clear_stale_blocks() below. We won't want to clear the
954 * unmount record if there is one, so we pass the lsn of the
955 * unmount record rather than the block after it.
957 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
958 int h_size = be32_to_cpu(rhead->h_size);
959 int h_version = be32_to_cpu(rhead->h_version);
961 if ((h_version & XLOG_VERSION_2) &&
962 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
963 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
964 if (h_size % XLOG_HEADER_CYCLE_SIZE)
972 after_umount_blk = (i + hblks + (int)
973 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
974 tail_lsn = log->l_tail_lsn;
975 if (*head_blk == after_umount_blk &&
976 be32_to_cpu(rhead->h_num_logops) == 1) {
977 umount_data_blk = (i + hblks) % log->l_logBBsize;
978 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
982 op_head = (xlog_op_header_t *)offset;
983 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
985 * Set tail and last sync so that newly written
986 * log records will point recovery to after the
987 * current unmount record.
990 xlog_assign_lsn(log->l_curr_cycle,
992 log->l_last_sync_lsn =
993 xlog_assign_lsn(log->l_curr_cycle,
995 *tail_blk = after_umount_blk;
998 * Note that the unmount was clean. If the unmount
999 * was not clean, we need to know this to rebuild the
1000 * superblock counters from the perag headers if we
1001 * have a filesystem using non-persistent counters.
1003 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1008 * Make sure that there are no blocks in front of the head
1009 * with the same cycle number as the head. This can happen
1010 * because we allow multiple outstanding log writes concurrently,
1011 * and the later writes might make it out before earlier ones.
1013 * We use the lsn from before modifying it so that we'll never
1014 * overwrite the unmount record after a clean unmount.
1016 * Do this only if we are going to recover the filesystem
1018 * NOTE: This used to say "if (!readonly)"
1019 * However on Linux, we can & do recover a read-only filesystem.
1020 * We only skip recovery if NORECOVERY is specified on mount,
1021 * in which case we would not be here.
1023 * But... if the -device- itself is readonly, just skip this.
1024 * We can't recover this device anyway, so it won't matter.
1026 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
1027 error = xlog_clear_stale_blocks(log, tail_lsn);
1033 xlog_warn("XFS: failed to locate log tail");
1038 * Is the log zeroed at all?
1040 * The last binary search should be changed to perform an X block read
1041 * once X becomes small enough. You can then search linearly through
1042 * the X blocks. This will cut down on the number of reads we need to do.
1044 * If the log is partially zeroed, this routine will pass back the blkno
1045 * of the first block with cycle number 0. It won't have a complete LR
1049 * 0 => the log is completely written to
1050 * -1 => use *blk_no as the first block of the log
1051 * >0 => error has occurred
1056 xfs_daddr_t *blk_no)
1060 uint first_cycle, last_cycle;
1061 xfs_daddr_t new_blk, last_blk, start_blk;
1062 xfs_daddr_t num_scan_bblks;
1063 int error, log_bbnum = log->l_logBBsize;
1067 /* check totally zeroed log */
1068 bp = xlog_get_bp(log, 1);
1071 error = xlog_bread(log, 0, 1, bp, &offset);
1075 first_cycle = xlog_get_cycle(offset);
1076 if (first_cycle == 0) { /* completely zeroed log */
1082 /* check partially zeroed log */
1083 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1087 last_cycle = xlog_get_cycle(offset);
1088 if (last_cycle != 0) { /* log completely written to */
1091 } else if (first_cycle != 1) {
1093 * If the cycle of the last block is zero, the cycle of
1094 * the first block must be 1. If it's not, maybe we're
1095 * not looking at a log... Bail out.
1097 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1098 return XFS_ERROR(EINVAL);
1101 /* we have a partially zeroed log */
1102 last_blk = log_bbnum-1;
1103 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1107 * Validate the answer. Because there is no way to guarantee that
1108 * the entire log is made up of log records which are the same size,
1109 * we scan over the defined maximum blocks. At this point, the maximum
1110 * is not chosen to mean anything special. XXXmiken
1112 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1113 ASSERT(num_scan_bblks <= INT_MAX);
1115 if (last_blk < num_scan_bblks)
1116 num_scan_bblks = last_blk;
1117 start_blk = last_blk - num_scan_bblks;
1120 * We search for any instances of cycle number 0 that occur before
1121 * our current estimate of the head. What we're trying to detect is
1122 * 1 ... | 0 | 1 | 0...
1123 * ^ binary search ends here
1125 if ((error = xlog_find_verify_cycle(log, start_blk,
1126 (int)num_scan_bblks, 0, &new_blk)))
1132 * Potentially backup over partial log record write. We don't need
1133 * to search the end of the log because we know it is zero.
1135 if ((error = xlog_find_verify_log_record(log, start_blk,
1136 &last_blk, 0)) == -1) {
1137 error = XFS_ERROR(EIO);
1151 * These are simple subroutines used by xlog_clear_stale_blocks() below
1152 * to initialize a buffer full of empty log record headers and write
1153 * them into the log.
1164 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1166 memset(buf, 0, BBSIZE);
1167 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1168 recp->h_cycle = cpu_to_be32(cycle);
1169 recp->h_version = cpu_to_be32(
1170 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1171 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1172 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1173 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1174 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1178 xlog_write_log_records(
1189 int sectbb = log->l_sectBBsize;
1190 int end_block = start_block + blocks;
1196 * Greedily allocate a buffer big enough to handle the full
1197 * range of basic blocks to be written. If that fails, try
1198 * a smaller size. We need to be able to write at least a
1199 * log sector, or we're out of luck.
1201 bufblks = 1 << ffs(blocks);
1202 while (!(bp = xlog_get_bp(log, bufblks))) {
1204 if (bufblks < sectbb)
1208 /* We may need to do a read at the start to fill in part of
1209 * the buffer in the starting sector not covered by the first
1212 balign = round_down(start_block, sectbb);
1213 if (balign != start_block) {
1214 error = xlog_bread_noalign(log, start_block, 1, bp);
1218 j = start_block - balign;
1221 for (i = start_block; i < end_block; i += bufblks) {
1222 int bcount, endcount;
1224 bcount = min(bufblks, end_block - start_block);
1225 endcount = bcount - j;
1227 /* We may need to do a read at the end to fill in part of
1228 * the buffer in the final sector not covered by the write.
1229 * If this is the same sector as the above read, skip it.
1231 ealign = round_down(end_block, sectbb);
1232 if (j == 0 && (start_block + endcount > ealign)) {
1233 offset = XFS_BUF_PTR(bp);
1234 balign = BBTOB(ealign - start_block);
1235 error = XFS_BUF_SET_PTR(bp, offset + balign,
1240 error = xlog_bread_noalign(log, ealign, sectbb, bp);
1244 error = XFS_BUF_SET_PTR(bp, offset, bufblks);
1249 offset = xlog_align(log, start_block, endcount, bp);
1250 for (; j < endcount; j++) {
1251 xlog_add_record(log, offset, cycle, i+j,
1252 tail_cycle, tail_block);
1255 error = xlog_bwrite(log, start_block, endcount, bp);
1258 start_block += endcount;
1268 * This routine is called to blow away any incomplete log writes out
1269 * in front of the log head. We do this so that we won't become confused
1270 * if we come up, write only a little bit more, and then crash again.
1271 * If we leave the partial log records out there, this situation could
1272 * cause us to think those partial writes are valid blocks since they
1273 * have the current cycle number. We get rid of them by overwriting them
1274 * with empty log records with the old cycle number rather than the
1277 * The tail lsn is passed in rather than taken from
1278 * the log so that we will not write over the unmount record after a
1279 * clean unmount in a 512 block log. Doing so would leave the log without
1280 * any valid log records in it until a new one was written. If we crashed
1281 * during that time we would not be able to recover.
1284 xlog_clear_stale_blocks(
1288 int tail_cycle, head_cycle;
1289 int tail_block, head_block;
1290 int tail_distance, max_distance;
1294 tail_cycle = CYCLE_LSN(tail_lsn);
1295 tail_block = BLOCK_LSN(tail_lsn);
1296 head_cycle = log->l_curr_cycle;
1297 head_block = log->l_curr_block;
1300 * Figure out the distance between the new head of the log
1301 * and the tail. We want to write over any blocks beyond the
1302 * head that we may have written just before the crash, but
1303 * we don't want to overwrite the tail of the log.
1305 if (head_cycle == tail_cycle) {
1307 * The tail is behind the head in the physical log,
1308 * so the distance from the head to the tail is the
1309 * distance from the head to the end of the log plus
1310 * the distance from the beginning of the log to the
1313 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1314 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1315 XFS_ERRLEVEL_LOW, log->l_mp);
1316 return XFS_ERROR(EFSCORRUPTED);
1318 tail_distance = tail_block + (log->l_logBBsize - head_block);
1321 * The head is behind the tail in the physical log,
1322 * so the distance from the head to the tail is just
1323 * the tail block minus the head block.
1325 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1326 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1327 XFS_ERRLEVEL_LOW, log->l_mp);
1328 return XFS_ERROR(EFSCORRUPTED);
1330 tail_distance = tail_block - head_block;
1334 * If the head is right up against the tail, we can't clear
1337 if (tail_distance <= 0) {
1338 ASSERT(tail_distance == 0);
1342 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1344 * Take the smaller of the maximum amount of outstanding I/O
1345 * we could have and the distance to the tail to clear out.
1346 * We take the smaller so that we don't overwrite the tail and
1347 * we don't waste all day writing from the head to the tail
1350 max_distance = MIN(max_distance, tail_distance);
1352 if ((head_block + max_distance) <= log->l_logBBsize) {
1354 * We can stomp all the blocks we need to without
1355 * wrapping around the end of the log. Just do it
1356 * in a single write. Use the cycle number of the
1357 * current cycle minus one so that the log will look like:
1360 error = xlog_write_log_records(log, (head_cycle - 1),
1361 head_block, max_distance, tail_cycle,
1367 * We need to wrap around the end of the physical log in
1368 * order to clear all the blocks. Do it in two separate
1369 * I/Os. The first write should be from the head to the
1370 * end of the physical log, and it should use the current
1371 * cycle number minus one just like above.
1373 distance = log->l_logBBsize - head_block;
1374 error = xlog_write_log_records(log, (head_cycle - 1),
1375 head_block, distance, tail_cycle,
1382 * Now write the blocks at the start of the physical log.
1383 * This writes the remainder of the blocks we want to clear.
1384 * It uses the current cycle number since we're now on the
1385 * same cycle as the head so that we get:
1386 * n ... n ... | n - 1 ...
1387 * ^^^^^ blocks we're writing
1389 distance = max_distance - (log->l_logBBsize - head_block);
1390 error = xlog_write_log_records(log, head_cycle, 0, distance,
1391 tail_cycle, tail_block);
1399 /******************************************************************************
1401 * Log recover routines
1403 ******************************************************************************
1406 STATIC xlog_recover_t *
1407 xlog_recover_find_tid(
1408 struct hlist_head *head,
1411 xlog_recover_t *trans;
1412 struct hlist_node *n;
1414 hlist_for_each_entry(trans, n, head, r_list) {
1415 if (trans->r_log_tid == tid)
1422 xlog_recover_new_tid(
1423 struct hlist_head *head,
1427 xlog_recover_t *trans;
1429 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1430 trans->r_log_tid = tid;
1432 INIT_LIST_HEAD(&trans->r_itemq);
1434 INIT_HLIST_NODE(&trans->r_list);
1435 hlist_add_head(&trans->r_list, head);
1439 xlog_recover_add_item(
1440 struct list_head *head)
1442 xlog_recover_item_t *item;
1444 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1445 INIT_LIST_HEAD(&item->ri_list);
1446 list_add_tail(&item->ri_list, head);
1450 xlog_recover_add_to_cont_trans(
1452 xlog_recover_t *trans,
1456 xlog_recover_item_t *item;
1457 xfs_caddr_t ptr, old_ptr;
1460 if (list_empty(&trans->r_itemq)) {
1461 /* finish copying rest of trans header */
1462 xlog_recover_add_item(&trans->r_itemq);
1463 ptr = (xfs_caddr_t) &trans->r_theader +
1464 sizeof(xfs_trans_header_t) - len;
1465 memcpy(ptr, dp, len); /* d, s, l */
1468 /* take the tail entry */
1469 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1471 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1472 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1474 ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
1475 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1476 item->ri_buf[item->ri_cnt-1].i_len += len;
1477 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1478 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
1483 * The next region to add is the start of a new region. It could be
1484 * a whole region or it could be the first part of a new region. Because
1485 * of this, the assumption here is that the type and size fields of all
1486 * format structures fit into the first 32 bits of the structure.
1488 * This works because all regions must be 32 bit aligned. Therefore, we
1489 * either have both fields or we have neither field. In the case we have
1490 * neither field, the data part of the region is zero length. We only have
1491 * a log_op_header and can throw away the header since a new one will appear
1492 * later. If we have at least 4 bytes, then we can determine how many regions
1493 * will appear in the current log item.
1496 xlog_recover_add_to_trans(
1498 xlog_recover_t *trans,
1502 xfs_inode_log_format_t *in_f; /* any will do */
1503 xlog_recover_item_t *item;
1508 if (list_empty(&trans->r_itemq)) {
1509 /* we need to catch log corruptions here */
1510 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
1511 xlog_warn("XFS: xlog_recover_add_to_trans: "
1512 "bad header magic number");
1514 return XFS_ERROR(EIO);
1516 if (len == sizeof(xfs_trans_header_t))
1517 xlog_recover_add_item(&trans->r_itemq);
1518 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1522 ptr = kmem_alloc(len, KM_SLEEP);
1523 memcpy(ptr, dp, len);
1524 in_f = (xfs_inode_log_format_t *)ptr;
1526 /* take the tail entry */
1527 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1528 if (item->ri_total != 0 &&
1529 item->ri_total == item->ri_cnt) {
1530 /* tail item is in use, get a new one */
1531 xlog_recover_add_item(&trans->r_itemq);
1532 item = list_entry(trans->r_itemq.prev,
1533 xlog_recover_item_t, ri_list);
1536 if (item->ri_total == 0) { /* first region to be added */
1537 if (in_f->ilf_size == 0 ||
1538 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
1540 "XFS: bad number of regions (%d) in inode log format",
1543 return XFS_ERROR(EIO);
1546 item->ri_total = in_f->ilf_size;
1548 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1551 ASSERT(item->ri_total > item->ri_cnt);
1552 /* Description region is ri_buf[0] */
1553 item->ri_buf[item->ri_cnt].i_addr = ptr;
1554 item->ri_buf[item->ri_cnt].i_len = len;
1556 trace_xfs_log_recover_item_add(log, trans, item, 0);
1561 * Sort the log items in the transaction. Cancelled buffers need
1562 * to be put first so they are processed before any items that might
1563 * modify the buffers. If they are cancelled, then the modifications
1564 * don't need to be replayed.
1567 xlog_recover_reorder_trans(
1569 xlog_recover_t *trans,
1572 xlog_recover_item_t *item, *n;
1573 LIST_HEAD(sort_list);
1575 list_splice_init(&trans->r_itemq, &sort_list);
1576 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1577 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1579 switch (ITEM_TYPE(item)) {
1581 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1582 trace_xfs_log_recover_item_reorder_head(log,
1584 list_move(&item->ri_list, &trans->r_itemq);
1589 case XFS_LI_QUOTAOFF:
1592 trace_xfs_log_recover_item_reorder_tail(log,
1594 list_move_tail(&item->ri_list, &trans->r_itemq);
1598 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1600 return XFS_ERROR(EIO);
1603 ASSERT(list_empty(&sort_list));
1608 * Build up the table of buf cancel records so that we don't replay
1609 * cancelled data in the second pass. For buffer records that are
1610 * not cancel records, there is nothing to do here so we just return.
1612 * If we get a cancel record which is already in the table, this indicates
1613 * that the buffer was cancelled multiple times. In order to ensure
1614 * that during pass 2 we keep the record in the table until we reach its
1615 * last occurrence in the log, we keep a reference count in the cancel
1616 * record in the table to tell us how many times we expect to see this
1617 * record during the second pass.
1620 xlog_recover_buffer_pass1(
1622 xlog_recover_item_t *item)
1624 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1625 struct list_head *bucket;
1626 struct xfs_buf_cancel *bcp;
1629 * If this isn't a cancel buffer item, then just return.
1631 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1632 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
1637 * Insert an xfs_buf_cancel record into the hash table of them.
1638 * If there is already an identical record, bump its reference count.
1640 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1641 list_for_each_entry(bcp, bucket, bc_list) {
1642 if (bcp->bc_blkno == buf_f->blf_blkno &&
1643 bcp->bc_len == buf_f->blf_len) {
1645 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
1650 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1651 bcp->bc_blkno = buf_f->blf_blkno;
1652 bcp->bc_len = buf_f->blf_len;
1653 bcp->bc_refcount = 1;
1654 list_add_tail(&bcp->bc_list, bucket);
1656 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
1661 * Check to see whether the buffer being recovered has a corresponding
1662 * entry in the buffer cancel record table. If it does then return 1
1663 * so that it will be cancelled, otherwise return 0. If the buffer is
1664 * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement
1665 * the refcount on the entry in the table and remove it from the table
1666 * if this is the last reference.
1668 * We remove the cancel record from the table when we encounter its
1669 * last occurrence in the log so that if the same buffer is re-used
1670 * again after its last cancellation we actually replay the changes
1671 * made at that point.
1674 xlog_check_buffer_cancelled(
1680 struct list_head *bucket;
1681 struct xfs_buf_cancel *bcp;
1683 if (log->l_buf_cancel_table == NULL) {
1685 * There is nothing in the table built in pass one,
1686 * so this buffer must not be cancelled.
1688 ASSERT(!(flags & XFS_BLF_CANCEL));
1693 * Search for an entry in the cancel table that matches our buffer.
1695 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1696 list_for_each_entry(bcp, bucket, bc_list) {
1697 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
1702 * We didn't find a corresponding entry in the table, so return 0 so
1703 * that the buffer is NOT cancelled.
1705 ASSERT(!(flags & XFS_BLF_CANCEL));
1710 * We've go a match, so return 1 so that the recovery of this buffer
1711 * is cancelled. If this buffer is actually a buffer cancel log
1712 * item, then decrement the refcount on the one in the table and
1713 * remove it if this is the last reference.
1715 if (flags & XFS_BLF_CANCEL) {
1716 if (--bcp->bc_refcount == 0) {
1717 list_del(&bcp->bc_list);
1725 * Perform recovery for a buffer full of inodes. In these buffers, the only
1726 * data which should be recovered is that which corresponds to the
1727 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1728 * data for the inodes is always logged through the inodes themselves rather
1729 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1731 * The only time when buffers full of inodes are fully recovered is when the
1732 * buffer is full of newly allocated inodes. In this case the buffer will
1733 * not be marked as an inode buffer and so will be sent to
1734 * xlog_recover_do_reg_buffer() below during recovery.
1737 xlog_recover_do_inode_buffer(
1738 struct xfs_mount *mp,
1739 xlog_recover_item_t *item,
1741 xfs_buf_log_format_t *buf_f)
1747 int reg_buf_offset = 0;
1748 int reg_buf_bytes = 0;
1749 int next_unlinked_offset;
1751 xfs_agino_t *logged_nextp;
1752 xfs_agino_t *buffer_nextp;
1754 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
1756 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1757 for (i = 0; i < inodes_per_buf; i++) {
1758 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1759 offsetof(xfs_dinode_t, di_next_unlinked);
1761 while (next_unlinked_offset >=
1762 (reg_buf_offset + reg_buf_bytes)) {
1764 * The next di_next_unlinked field is beyond
1765 * the current logged region. Find the next
1766 * logged region that contains or is beyond
1767 * the current di_next_unlinked field.
1770 bit = xfs_next_bit(buf_f->blf_data_map,
1771 buf_f->blf_map_size, bit);
1774 * If there are no more logged regions in the
1775 * buffer, then we're done.
1780 nbits = xfs_contig_bits(buf_f->blf_data_map,
1781 buf_f->blf_map_size, bit);
1783 reg_buf_offset = bit << XFS_BLF_SHIFT;
1784 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
1789 * If the current logged region starts after the current
1790 * di_next_unlinked field, then move on to the next
1791 * di_next_unlinked field.
1793 if (next_unlinked_offset < reg_buf_offset)
1796 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1797 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
1798 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1801 * The current logged region contains a copy of the
1802 * current di_next_unlinked field. Extract its value
1803 * and copy it to the buffer copy.
1805 logged_nextp = item->ri_buf[item_index].i_addr +
1806 next_unlinked_offset - reg_buf_offset;
1807 if (unlikely(*logged_nextp == 0)) {
1808 xfs_fs_cmn_err(CE_ALERT, mp,
1809 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1811 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1812 XFS_ERRLEVEL_LOW, mp);
1813 return XFS_ERROR(EFSCORRUPTED);
1816 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1817 next_unlinked_offset);
1818 *buffer_nextp = *logged_nextp;
1825 * Perform a 'normal' buffer recovery. Each logged region of the
1826 * buffer should be copied over the corresponding region in the
1827 * given buffer. The bitmap in the buf log format structure indicates
1828 * where to place the logged data.
1831 xlog_recover_do_reg_buffer(
1832 struct xfs_mount *mp,
1833 xlog_recover_item_t *item,
1835 xfs_buf_log_format_t *buf_f)
1842 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
1845 i = 1; /* 0 is the buf format structure */
1847 bit = xfs_next_bit(buf_f->blf_data_map,
1848 buf_f->blf_map_size, bit);
1851 nbits = xfs_contig_bits(buf_f->blf_data_map,
1852 buf_f->blf_map_size, bit);
1854 ASSERT(item->ri_buf[i].i_addr != NULL);
1855 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
1856 ASSERT(XFS_BUF_COUNT(bp) >=
1857 ((uint)bit << XFS_BLF_SHIFT)+(nbits<<XFS_BLF_SHIFT));
1860 * Do a sanity check if this is a dquot buffer. Just checking
1861 * the first dquot in the buffer should do. XXXThis is
1862 * probably a good thing to do for other buf types also.
1865 if (buf_f->blf_flags &
1866 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
1867 if (item->ri_buf[i].i_addr == NULL) {
1869 "XFS: NULL dquot in %s.", __func__);
1872 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
1874 "XFS: dquot too small (%d) in %s.",
1875 item->ri_buf[i].i_len, __func__);
1878 error = xfs_qm_dqcheck(item->ri_buf[i].i_addr,
1879 -1, 0, XFS_QMOPT_DOWARN,
1880 "dquot_buf_recover");
1885 memcpy(xfs_buf_offset(bp,
1886 (uint)bit << XFS_BLF_SHIFT), /* dest */
1887 item->ri_buf[i].i_addr, /* source */
1888 nbits<<XFS_BLF_SHIFT); /* length */
1894 /* Shouldn't be any more regions */
1895 ASSERT(i == item->ri_total);
1899 * Do some primitive error checking on ondisk dquot data structures.
1903 xfs_disk_dquot_t *ddq,
1905 uint type, /* used only when IO_dorepair is true */
1909 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
1913 * We can encounter an uninitialized dquot buffer for 2 reasons:
1914 * 1. If we crash while deleting the quotainode(s), and those blks got
1915 * used for user data. This is because we take the path of regular
1916 * file deletion; however, the size field of quotainodes is never
1917 * updated, so all the tricks that we play in itruncate_finish
1918 * don't quite matter.
1920 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1921 * But the allocation will be replayed so we'll end up with an
1922 * uninitialized quota block.
1924 * This is all fine; things are still consistent, and we haven't lost
1925 * any quota information. Just don't complain about bad dquot blks.
1927 if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) {
1928 if (flags & XFS_QMOPT_DOWARN)
1930 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
1931 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
1934 if (ddq->d_version != XFS_DQUOT_VERSION) {
1935 if (flags & XFS_QMOPT_DOWARN)
1937 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
1938 str, id, ddq->d_version, XFS_DQUOT_VERSION);
1942 if (ddq->d_flags != XFS_DQ_USER &&
1943 ddq->d_flags != XFS_DQ_PROJ &&
1944 ddq->d_flags != XFS_DQ_GROUP) {
1945 if (flags & XFS_QMOPT_DOWARN)
1947 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
1948 str, id, ddq->d_flags);
1952 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
1953 if (flags & XFS_QMOPT_DOWARN)
1955 "%s : ondisk-dquot 0x%p, ID mismatch: "
1956 "0x%x expected, found id 0x%x",
1957 str, ddq, id, be32_to_cpu(ddq->d_id));
1961 if (!errs && ddq->d_id) {
1962 if (ddq->d_blk_softlimit &&
1963 be64_to_cpu(ddq->d_bcount) >=
1964 be64_to_cpu(ddq->d_blk_softlimit)) {
1965 if (!ddq->d_btimer) {
1966 if (flags & XFS_QMOPT_DOWARN)
1968 "%s : Dquot ID 0x%x (0x%p) "
1969 "BLK TIMER NOT STARTED",
1970 str, (int)be32_to_cpu(ddq->d_id), ddq);
1974 if (ddq->d_ino_softlimit &&
1975 be64_to_cpu(ddq->d_icount) >=
1976 be64_to_cpu(ddq->d_ino_softlimit)) {
1977 if (!ddq->d_itimer) {
1978 if (flags & XFS_QMOPT_DOWARN)
1980 "%s : Dquot ID 0x%x (0x%p) "
1981 "INODE TIMER NOT STARTED",
1982 str, (int)be32_to_cpu(ddq->d_id), ddq);
1986 if (ddq->d_rtb_softlimit &&
1987 be64_to_cpu(ddq->d_rtbcount) >=
1988 be64_to_cpu(ddq->d_rtb_softlimit)) {
1989 if (!ddq->d_rtbtimer) {
1990 if (flags & XFS_QMOPT_DOWARN)
1992 "%s : Dquot ID 0x%x (0x%p) "
1993 "RTBLK TIMER NOT STARTED",
1994 str, (int)be32_to_cpu(ddq->d_id), ddq);
2000 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2003 if (flags & XFS_QMOPT_DOWARN)
2004 cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id);
2007 * Typically, a repair is only requested by quotacheck.
2010 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2011 memset(d, 0, sizeof(xfs_dqblk_t));
2013 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2014 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2015 d->dd_diskdq.d_flags = type;
2016 d->dd_diskdq.d_id = cpu_to_be32(id);
2022 * Perform a dquot buffer recovery.
2023 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2024 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2025 * Else, treat it as a regular buffer and do recovery.
2028 xlog_recover_do_dquot_buffer(
2031 xlog_recover_item_t *item,
2033 xfs_buf_log_format_t *buf_f)
2037 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2040 * Filesystems are required to send in quota flags at mount time.
2042 if (mp->m_qflags == 0) {
2047 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2048 type |= XFS_DQ_USER;
2049 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2050 type |= XFS_DQ_PROJ;
2051 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2052 type |= XFS_DQ_GROUP;
2054 * This type of quotas was turned off, so ignore this buffer
2056 if (log->l_quotaoffs_flag & type)
2059 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2063 * This routine replays a modification made to a buffer at runtime.
2064 * There are actually two types of buffer, regular and inode, which
2065 * are handled differently. Inode buffers are handled differently
2066 * in that we only recover a specific set of data from them, namely
2067 * the inode di_next_unlinked fields. This is because all other inode
2068 * data is actually logged via inode records and any data we replay
2069 * here which overlaps that may be stale.
2071 * When meta-data buffers are freed at run time we log a buffer item
2072 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2073 * of the buffer in the log should not be replayed at recovery time.
2074 * This is so that if the blocks covered by the buffer are reused for
2075 * file data before we crash we don't end up replaying old, freed
2076 * meta-data into a user's file.
2078 * To handle the cancellation of buffer log items, we make two passes
2079 * over the log during recovery. During the first we build a table of
2080 * those buffers which have been cancelled, and during the second we
2081 * only replay those buffers which do not have corresponding cancel
2082 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2083 * for more details on the implementation of the table of cancel records.
2086 xlog_recover_buffer_pass2(
2088 xlog_recover_item_t *item)
2090 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2091 xfs_mount_t *mp = log->l_mp;
2097 * In this pass we only want to recover all the buffers which have
2098 * not been cancelled and are not cancellation buffers themselves.
2100 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2101 buf_f->blf_len, buf_f->blf_flags)) {
2102 trace_xfs_log_recover_buf_cancel(log, buf_f);
2106 trace_xfs_log_recover_buf_recover(log, buf_f);
2108 buf_flags = XBF_LOCK;
2109 if (!(buf_f->blf_flags & XFS_BLF_INODE_BUF))
2110 buf_flags |= XBF_MAPPED;
2112 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2114 if (XFS_BUF_ISERROR(bp)) {
2115 xfs_ioerror_alert("xlog_recover_do..(read#1)", mp,
2116 bp, buf_f->blf_blkno);
2117 error = XFS_BUF_GETERROR(bp);
2123 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
2124 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2125 } else if (buf_f->blf_flags &
2126 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2127 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2129 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2132 return XFS_ERROR(error);
2135 * Perform delayed write on the buffer. Asynchronous writes will be
2136 * slower when taking into account all the buffers to be flushed.
2138 * Also make sure that only inode buffers with good sizes stay in
2139 * the buffer cache. The kernel moves inodes in buffers of 1 block
2140 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2141 * buffers in the log can be a different size if the log was generated
2142 * by an older kernel using unclustered inode buffers or a newer kernel
2143 * running with a different inode cluster size. Regardless, if the
2144 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2145 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2146 * the buffer out of the buffer cache so that the buffer won't
2147 * overlap with future reads of those inodes.
2149 if (XFS_DINODE_MAGIC ==
2150 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2151 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2152 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2154 error = xfs_bwrite(mp, bp);
2156 ASSERT(bp->b_target->bt_mount == mp);
2157 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2158 xfs_bdwrite(mp, bp);
2165 xlog_recover_inode_pass2(
2167 xlog_recover_item_t *item)
2169 xfs_inode_log_format_t *in_f;
2170 xfs_mount_t *mp = log->l_mp;
2179 xfs_icdinode_t *dicp;
2182 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2183 in_f = item->ri_buf[0].i_addr;
2185 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
2187 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2193 * Inode buffers can be freed, look out for it,
2194 * and do not replay the inode.
2196 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2197 in_f->ilf_len, 0)) {
2199 trace_xfs_log_recover_inode_cancel(log, in_f);
2202 trace_xfs_log_recover_inode_recover(log, in_f);
2204 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len,
2206 if (XFS_BUF_ISERROR(bp)) {
2207 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
2208 bp, in_f->ilf_blkno);
2209 error = XFS_BUF_GETERROR(bp);
2214 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2215 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
2218 * Make sure the place we're flushing out to really looks
2221 if (unlikely(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC)) {
2223 xfs_fs_cmn_err(CE_ALERT, mp,
2224 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2225 dip, bp, in_f->ilf_ino);
2226 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2227 XFS_ERRLEVEL_LOW, mp);
2228 error = EFSCORRUPTED;
2231 dicp = item->ri_buf[1].i_addr;
2232 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2234 xfs_fs_cmn_err(CE_ALERT, mp,
2235 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2236 item, in_f->ilf_ino);
2237 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2238 XFS_ERRLEVEL_LOW, mp);
2239 error = EFSCORRUPTED;
2243 /* Skip replay when the on disk inode is newer than the log one */
2244 if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
2246 * Deal with the wrap case, DI_MAX_FLUSH is less
2247 * than smaller numbers
2249 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
2250 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2254 trace_xfs_log_recover_inode_skip(log, in_f);
2259 /* Take the opportunity to reset the flush iteration count */
2260 dicp->di_flushiter = 0;
2262 if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) {
2263 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2264 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2265 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2266 XFS_ERRLEVEL_LOW, mp, dicp);
2268 xfs_fs_cmn_err(CE_ALERT, mp,
2269 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2270 item, dip, bp, in_f->ilf_ino);
2271 error = EFSCORRUPTED;
2274 } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) {
2275 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2276 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2277 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2278 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2279 XFS_ERRLEVEL_LOW, mp, dicp);
2281 xfs_fs_cmn_err(CE_ALERT, mp,
2282 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2283 item, dip, bp, in_f->ilf_ino);
2284 error = EFSCORRUPTED;
2288 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2289 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2290 XFS_ERRLEVEL_LOW, mp, dicp);
2292 xfs_fs_cmn_err(CE_ALERT, mp,
2293 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2294 item, dip, bp, in_f->ilf_ino,
2295 dicp->di_nextents + dicp->di_anextents,
2297 error = EFSCORRUPTED;
2300 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2301 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2302 XFS_ERRLEVEL_LOW, mp, dicp);
2304 xfs_fs_cmn_err(CE_ALERT, mp,
2305 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2306 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
2307 error = EFSCORRUPTED;
2310 if (unlikely(item->ri_buf[1].i_len > sizeof(struct xfs_icdinode))) {
2311 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
2312 XFS_ERRLEVEL_LOW, mp, dicp);
2314 xfs_fs_cmn_err(CE_ALERT, mp,
2315 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2316 item->ri_buf[1].i_len, item);
2317 error = EFSCORRUPTED;
2321 /* The core is in in-core format */
2322 xfs_dinode_to_disk(dip, item->ri_buf[1].i_addr);
2324 /* the rest is in on-disk format */
2325 if (item->ri_buf[1].i_len > sizeof(struct xfs_icdinode)) {
2326 memcpy((xfs_caddr_t) dip + sizeof(struct xfs_icdinode),
2327 item->ri_buf[1].i_addr + sizeof(struct xfs_icdinode),
2328 item->ri_buf[1].i_len - sizeof(struct xfs_icdinode));
2331 fields = in_f->ilf_fields;
2332 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2334 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
2337 memcpy(XFS_DFORK_DPTR(dip),
2338 &in_f->ilf_u.ilfu_uuid,
2343 if (in_f->ilf_size == 2)
2344 goto write_inode_buffer;
2345 len = item->ri_buf[2].i_len;
2346 src = item->ri_buf[2].i_addr;
2347 ASSERT(in_f->ilf_size <= 4);
2348 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2349 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2350 (len == in_f->ilf_dsize));
2352 switch (fields & XFS_ILOG_DFORK) {
2353 case XFS_ILOG_DDATA:
2355 memcpy(XFS_DFORK_DPTR(dip), src, len);
2358 case XFS_ILOG_DBROOT:
2359 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
2360 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
2361 XFS_DFORK_DSIZE(dip, mp));
2366 * There are no data fork flags set.
2368 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2373 * If we logged any attribute data, recover it. There may or
2374 * may not have been any other non-core data logged in this
2377 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2378 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2383 len = item->ri_buf[attr_index].i_len;
2384 src = item->ri_buf[attr_index].i_addr;
2385 ASSERT(len == in_f->ilf_asize);
2387 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2388 case XFS_ILOG_ADATA:
2390 dest = XFS_DFORK_APTR(dip);
2391 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2392 memcpy(dest, src, len);
2395 case XFS_ILOG_ABROOT:
2396 dest = XFS_DFORK_APTR(dip);
2397 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2398 len, (xfs_bmdr_block_t*)dest,
2399 XFS_DFORK_ASIZE(dip, mp));
2403 xlog_warn("XFS: xlog_recover_inode_pass2: Invalid flag");
2412 ASSERT(bp->b_target->bt_mount == mp);
2413 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2414 xfs_bdwrite(mp, bp);
2418 return XFS_ERROR(error);
2422 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2423 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2427 xlog_recover_quotaoff_pass1(
2429 xlog_recover_item_t *item)
2431 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
2435 * The logitem format's flag tells us if this was user quotaoff,
2436 * group/project quotaoff or both.
2438 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2439 log->l_quotaoffs_flag |= XFS_DQ_USER;
2440 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2441 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2442 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2443 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2449 * Recover a dquot record
2452 xlog_recover_dquot_pass2(
2454 xlog_recover_item_t *item)
2456 xfs_mount_t *mp = log->l_mp;
2458 struct xfs_disk_dquot *ddq, *recddq;
2460 xfs_dq_logformat_t *dq_f;
2465 * Filesystems are required to send in quota flags at mount time.
2467 if (mp->m_qflags == 0)
2470 recddq = item->ri_buf[1].i_addr;
2471 if (recddq == NULL) {
2473 "XFS: NULL dquot in %s.", __func__);
2474 return XFS_ERROR(EIO);
2476 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
2478 "XFS: dquot too small (%d) in %s.",
2479 item->ri_buf[1].i_len, __func__);
2480 return XFS_ERROR(EIO);
2484 * This type of quotas was turned off, so ignore this record.
2486 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
2488 if (log->l_quotaoffs_flag & type)
2492 * At this point we know that quota was _not_ turned off.
2493 * Since the mount flags are not indicating to us otherwise, this
2494 * must mean that quota is on, and the dquot needs to be replayed.
2495 * Remember that we may not have fully recovered the superblock yet,
2496 * so we can't do the usual trick of looking at the SB quota bits.
2498 * The other possibility, of course, is that the quota subsystem was
2499 * removed since the last mount - ENOSYS.
2501 dq_f = item->ri_buf[0].i_addr;
2503 if ((error = xfs_qm_dqcheck(recddq,
2505 0, XFS_QMOPT_DOWARN,
2506 "xlog_recover_dquot_pass2 (log copy)"))) {
2507 return XFS_ERROR(EIO);
2509 ASSERT(dq_f->qlf_len == 1);
2511 error = xfs_read_buf(mp, mp->m_ddev_targp,
2513 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2516 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
2517 bp, dq_f->qlf_blkno);
2521 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2524 * At least the magic num portion should be on disk because this
2525 * was among a chunk of dquots created earlier, and we did some
2526 * minimal initialization then.
2528 if (xfs_qm_dqcheck(ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2529 "xlog_recover_dquot_pass2")) {
2531 return XFS_ERROR(EIO);
2534 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2536 ASSERT(dq_f->qlf_size == 2);
2537 ASSERT(bp->b_target->bt_mount == mp);
2538 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2539 xfs_bdwrite(mp, bp);
2545 * This routine is called to create an in-core extent free intent
2546 * item from the efi format structure which was logged on disk.
2547 * It allocates an in-core efi, copies the extents from the format
2548 * structure into it, and adds the efi to the AIL with the given
2552 xlog_recover_efi_pass2(
2554 xlog_recover_item_t *item,
2558 xfs_mount_t *mp = log->l_mp;
2559 xfs_efi_log_item_t *efip;
2560 xfs_efi_log_format_t *efi_formatp;
2562 efi_formatp = item->ri_buf[0].i_addr;
2564 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
2565 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2566 &(efip->efi_format)))) {
2567 xfs_efi_item_free(efip);
2570 efip->efi_next_extent = efi_formatp->efi_nextents;
2571 efip->efi_flags |= XFS_EFI_COMMITTED;
2573 spin_lock(&log->l_ailp->xa_lock);
2575 * xfs_trans_ail_update() drops the AIL lock.
2577 xfs_trans_ail_update(log->l_ailp, (xfs_log_item_t *)efip, lsn);
2583 * This routine is called when an efd format structure is found in
2584 * a committed transaction in the log. It's purpose is to cancel
2585 * the corresponding efi if it was still in the log. To do this
2586 * it searches the AIL for the efi with an id equal to that in the
2587 * efd format structure. If we find it, we remove the efi from the
2591 xlog_recover_efd_pass2(
2593 xlog_recover_item_t *item)
2595 xfs_efd_log_format_t *efd_formatp;
2596 xfs_efi_log_item_t *efip = NULL;
2597 xfs_log_item_t *lip;
2599 struct xfs_ail_cursor cur;
2600 struct xfs_ail *ailp = log->l_ailp;
2602 efd_formatp = item->ri_buf[0].i_addr;
2603 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2604 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2605 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2606 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
2607 efi_id = efd_formatp->efd_efi_id;
2610 * Search for the efi with the id in the efd format structure
2613 spin_lock(&ailp->xa_lock);
2614 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
2615 while (lip != NULL) {
2616 if (lip->li_type == XFS_LI_EFI) {
2617 efip = (xfs_efi_log_item_t *)lip;
2618 if (efip->efi_format.efi_id == efi_id) {
2620 * xfs_trans_ail_delete() drops the
2623 xfs_trans_ail_delete(ailp, lip);
2624 xfs_efi_item_free(efip);
2625 spin_lock(&ailp->xa_lock);
2629 lip = xfs_trans_ail_cursor_next(ailp, &cur);
2631 xfs_trans_ail_cursor_done(ailp, &cur);
2632 spin_unlock(&ailp->xa_lock);
2638 * Free up any resources allocated by the transaction
2640 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2643 xlog_recover_free_trans(
2644 struct xlog_recover *trans)
2646 xlog_recover_item_t *item, *n;
2649 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
2650 /* Free the regions in the item. */
2651 list_del(&item->ri_list);
2652 for (i = 0; i < item->ri_cnt; i++)
2653 kmem_free(item->ri_buf[i].i_addr);
2654 /* Free the item itself */
2655 kmem_free(item->ri_buf);
2658 /* Free the transaction recover structure */
2663 xlog_recover_commit_pass1(
2665 struct xlog_recover *trans,
2666 xlog_recover_item_t *item)
2668 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
2670 switch (ITEM_TYPE(item)) {
2672 return xlog_recover_buffer_pass1(log, item);
2673 case XFS_LI_QUOTAOFF:
2674 return xlog_recover_quotaoff_pass1(log, item);
2679 /* nothing to do in pass 1 */
2683 "XFS: invalid item type (%d) xlog_recover_commit_pass1",
2686 return XFS_ERROR(EIO);
2691 xlog_recover_commit_pass2(
2693 struct xlog_recover *trans,
2694 xlog_recover_item_t *item)
2696 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
2698 switch (ITEM_TYPE(item)) {
2700 return xlog_recover_buffer_pass2(log, item);
2702 return xlog_recover_inode_pass2(log, item);
2704 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
2706 return xlog_recover_efd_pass2(log, item);
2708 return xlog_recover_dquot_pass2(log, item);
2709 case XFS_LI_QUOTAOFF:
2710 /* nothing to do in pass2 */
2714 "XFS: invalid item type (%d) xlog_recover_commit_pass2",
2717 return XFS_ERROR(EIO);
2722 * Perform the transaction.
2724 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2725 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2728 xlog_recover_commit_trans(
2730 struct xlog_recover *trans,
2734 xlog_recover_item_t *item;
2736 hlist_del(&trans->r_list);
2738 error = xlog_recover_reorder_trans(log, trans, pass);
2742 list_for_each_entry(item, &trans->r_itemq, ri_list) {
2743 if (pass == XLOG_RECOVER_PASS1)
2744 error = xlog_recover_commit_pass1(log, trans, item);
2746 error = xlog_recover_commit_pass2(log, trans, item);
2751 xlog_recover_free_trans(trans);
2756 xlog_recover_unmount_trans(
2757 xlog_recover_t *trans)
2759 /* Do nothing now */
2760 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2765 * There are two valid states of the r_state field. 0 indicates that the
2766 * transaction structure is in a normal state. We have either seen the
2767 * start of the transaction or the last operation we added was not a partial
2768 * operation. If the last operation we added to the transaction was a
2769 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2771 * NOTE: skip LRs with 0 data length.
2774 xlog_recover_process_data(
2776 struct hlist_head rhash[],
2777 xlog_rec_header_t *rhead,
2783 xlog_op_header_t *ohead;
2784 xlog_recover_t *trans;
2790 lp = dp + be32_to_cpu(rhead->h_len);
2791 num_logops = be32_to_cpu(rhead->h_num_logops);
2793 /* check the log format matches our own - else we can't recover */
2794 if (xlog_header_check_recover(log->l_mp, rhead))
2795 return (XFS_ERROR(EIO));
2797 while ((dp < lp) && num_logops) {
2798 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2799 ohead = (xlog_op_header_t *)dp;
2800 dp += sizeof(xlog_op_header_t);
2801 if (ohead->oh_clientid != XFS_TRANSACTION &&
2802 ohead->oh_clientid != XFS_LOG) {
2804 "XFS: xlog_recover_process_data: bad clientid");
2806 return (XFS_ERROR(EIO));
2808 tid = be32_to_cpu(ohead->oh_tid);
2809 hash = XLOG_RHASH(tid);
2810 trans = xlog_recover_find_tid(&rhash[hash], tid);
2811 if (trans == NULL) { /* not found; add new tid */
2812 if (ohead->oh_flags & XLOG_START_TRANS)
2813 xlog_recover_new_tid(&rhash[hash], tid,
2814 be64_to_cpu(rhead->h_lsn));
2816 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
2818 "XFS: xlog_recover_process_data: bad length");
2820 return (XFS_ERROR(EIO));
2822 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2823 if (flags & XLOG_WAS_CONT_TRANS)
2824 flags &= ~XLOG_CONTINUE_TRANS;
2826 case XLOG_COMMIT_TRANS:
2827 error = xlog_recover_commit_trans(log,
2830 case XLOG_UNMOUNT_TRANS:
2831 error = xlog_recover_unmount_trans(trans);
2833 case XLOG_WAS_CONT_TRANS:
2834 error = xlog_recover_add_to_cont_trans(log,
2836 be32_to_cpu(ohead->oh_len));
2838 case XLOG_START_TRANS:
2840 "XFS: xlog_recover_process_data: bad transaction");
2842 error = XFS_ERROR(EIO);
2845 case XLOG_CONTINUE_TRANS:
2846 error = xlog_recover_add_to_trans(log, trans,
2847 dp, be32_to_cpu(ohead->oh_len));
2851 "XFS: xlog_recover_process_data: bad flag");
2853 error = XFS_ERROR(EIO);
2859 dp += be32_to_cpu(ohead->oh_len);
2866 * Process an extent free intent item that was recovered from
2867 * the log. We need to free the extents that it describes.
2870 xlog_recover_process_efi(
2872 xfs_efi_log_item_t *efip)
2874 xfs_efd_log_item_t *efdp;
2879 xfs_fsblock_t startblock_fsb;
2881 ASSERT(!(efip->efi_flags & XFS_EFI_RECOVERED));
2884 * First check the validity of the extents described by the
2885 * EFI. If any are bad, then assume that all are bad and
2886 * just toss the EFI.
2888 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2889 extp = &(efip->efi_format.efi_extents[i]);
2890 startblock_fsb = XFS_BB_TO_FSB(mp,
2891 XFS_FSB_TO_DADDR(mp, extp->ext_start));
2892 if ((startblock_fsb == 0) ||
2893 (extp->ext_len == 0) ||
2894 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
2895 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
2897 * This will pull the EFI from the AIL and
2898 * free the memory associated with it.
2900 xfs_efi_release(efip, efip->efi_format.efi_nextents);
2901 return XFS_ERROR(EIO);
2905 tp = xfs_trans_alloc(mp, 0);
2906 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
2909 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
2911 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2912 extp = &(efip->efi_format.efi_extents[i]);
2913 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
2916 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
2920 efip->efi_flags |= XFS_EFI_RECOVERED;
2921 error = xfs_trans_commit(tp, 0);
2925 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
2930 * When this is called, all of the EFIs which did not have
2931 * corresponding EFDs should be in the AIL. What we do now
2932 * is free the extents associated with each one.
2934 * Since we process the EFIs in normal transactions, they
2935 * will be removed at some point after the commit. This prevents
2936 * us from just walking down the list processing each one.
2937 * We'll use a flag in the EFI to skip those that we've already
2938 * processed and use the AIL iteration mechanism's generation
2939 * count to try to speed this up at least a bit.
2941 * When we start, we know that the EFIs are the only things in
2942 * the AIL. As we process them, however, other items are added
2943 * to the AIL. Since everything added to the AIL must come after
2944 * everything already in the AIL, we stop processing as soon as
2945 * we see something other than an EFI in the AIL.
2948 xlog_recover_process_efis(
2951 xfs_log_item_t *lip;
2952 xfs_efi_log_item_t *efip;
2954 struct xfs_ail_cursor cur;
2955 struct xfs_ail *ailp;
2958 spin_lock(&ailp->xa_lock);
2959 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
2960 while (lip != NULL) {
2962 * We're done when we see something other than an EFI.
2963 * There should be no EFIs left in the AIL now.
2965 if (lip->li_type != XFS_LI_EFI) {
2967 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
2968 ASSERT(lip->li_type != XFS_LI_EFI);
2974 * Skip EFIs that we've already processed.
2976 efip = (xfs_efi_log_item_t *)lip;
2977 if (efip->efi_flags & XFS_EFI_RECOVERED) {
2978 lip = xfs_trans_ail_cursor_next(ailp, &cur);
2982 spin_unlock(&ailp->xa_lock);
2983 error = xlog_recover_process_efi(log->l_mp, efip);
2984 spin_lock(&ailp->xa_lock);
2987 lip = xfs_trans_ail_cursor_next(ailp, &cur);
2990 xfs_trans_ail_cursor_done(ailp, &cur);
2991 spin_unlock(&ailp->xa_lock);
2996 * This routine performs a transaction to null out a bad inode pointer
2997 * in an agi unlinked inode hash bucket.
3000 xlog_recover_clear_agi_bucket(
3002 xfs_agnumber_t agno,
3011 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3012 error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp),
3017 error = xfs_read_agi(mp, tp, agno, &agibp);
3021 agi = XFS_BUF_TO_AGI(agibp);
3022 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
3023 offset = offsetof(xfs_agi_t, agi_unlinked) +
3024 (sizeof(xfs_agino_t) * bucket);
3025 xfs_trans_log_buf(tp, agibp, offset,
3026 (offset + sizeof(xfs_agino_t) - 1));
3028 error = xfs_trans_commit(tp, 0);
3034 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3036 xfs_fs_cmn_err(CE_WARN, mp, "xlog_recover_clear_agi_bucket: "
3037 "failed to clear agi %d. Continuing.", agno);
3042 xlog_recover_process_one_iunlink(
3043 struct xfs_mount *mp,
3044 xfs_agnumber_t agno,
3048 struct xfs_buf *ibp;
3049 struct xfs_dinode *dip;
3050 struct xfs_inode *ip;
3054 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3055 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
3060 * Get the on disk inode to find the next inode in the bucket.
3062 error = xfs_itobp(mp, NULL, ip, &dip, &ibp, XBF_LOCK);
3066 ASSERT(ip->i_d.di_nlink == 0);
3067 ASSERT(ip->i_d.di_mode != 0);
3069 /* setup for the next pass */
3070 agino = be32_to_cpu(dip->di_next_unlinked);
3074 * Prevent any DMAPI event from being sent when the reference on
3075 * the inode is dropped.
3077 ip->i_d.di_dmevmask = 0;
3086 * We can't read in the inode this bucket points to, or this inode
3087 * is messed up. Just ditch this bucket of inodes. We will lose
3088 * some inodes and space, but at least we won't hang.
3090 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3091 * clear the inode pointer in the bucket.
3093 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3098 * xlog_iunlink_recover
3100 * This is called during recovery to process any inodes which
3101 * we unlinked but not freed when the system crashed. These
3102 * inodes will be on the lists in the AGI blocks. What we do
3103 * here is scan all the AGIs and fully truncate and free any
3104 * inodes found on the lists. Each inode is removed from the
3105 * lists when it has been fully truncated and is freed. The
3106 * freeing of the inode and its removal from the list must be
3110 xlog_recover_process_iunlinks(
3114 xfs_agnumber_t agno;
3125 * Prevent any DMAPI event from being sent while in this function.
3127 mp_dmevmask = mp->m_dmevmask;
3130 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3132 * Find the agi for this ag.
3134 error = xfs_read_agi(mp, NULL, agno, &agibp);
3137 * AGI is b0rked. Don't process it.
3139 * We should probably mark the filesystem as corrupt
3140 * after we've recovered all the ag's we can....
3144 agi = XFS_BUF_TO_AGI(agibp);
3146 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3147 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3148 while (agino != NULLAGINO) {
3150 * Release the agi buffer so that it can
3151 * be acquired in the normal course of the
3152 * transaction to truncate and free the inode.
3154 xfs_buf_relse(agibp);
3156 agino = xlog_recover_process_one_iunlink(mp,
3157 agno, agino, bucket);
3160 * Reacquire the agibuffer and continue around
3161 * the loop. This should never fail as we know
3162 * the buffer was good earlier on.
3164 error = xfs_read_agi(mp, NULL, agno, &agibp);
3166 agi = XFS_BUF_TO_AGI(agibp);
3171 * Release the buffer for the current agi so we can
3172 * go on to the next one.
3174 xfs_buf_relse(agibp);
3177 mp->m_dmevmask = mp_dmevmask;
3183 xlog_pack_data_checksum(
3185 xlog_in_core_t *iclog,
3192 up = (__be32 *)iclog->ic_datap;
3193 /* divide length by 4 to get # words */
3194 for (i = 0; i < (size >> 2); i++) {
3195 chksum ^= be32_to_cpu(*up);
3198 iclog->ic_header.h_chksum = cpu_to_be32(chksum);
3201 #define xlog_pack_data_checksum(log, iclog, size)
3205 * Stamp cycle number in every block
3210 xlog_in_core_t *iclog,
3214 int size = iclog->ic_offset + roundoff;
3218 xlog_pack_data_checksum(log, iclog, size);
3220 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3222 dp = iclog->ic_datap;
3223 for (i = 0; i < BTOBB(size) &&
3224 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3225 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
3226 *(__be32 *)dp = cycle_lsn;
3230 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3231 xlog_in_core_2_t *xhdr = iclog->ic_data;
3233 for ( ; i < BTOBB(size); i++) {
3234 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3235 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3236 xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
3237 *(__be32 *)dp = cycle_lsn;
3241 for (i = 1; i < log->l_iclog_heads; i++) {
3242 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3249 xlog_rec_header_t *rhead,
3255 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
3256 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3257 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
3261 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3262 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
3263 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
3264 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3265 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3266 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
3273 xlog_valid_rec_header(
3275 xlog_rec_header_t *rhead,
3280 if (unlikely(be32_to_cpu(rhead->h_magicno) != XLOG_HEADER_MAGIC_NUM)) {
3281 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3282 XFS_ERRLEVEL_LOW, log->l_mp);
3283 return XFS_ERROR(EFSCORRUPTED);
3286 (!rhead->h_version ||
3287 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
3288 xlog_warn("XFS: %s: unrecognised log version (%d).",
3289 __func__, be32_to_cpu(rhead->h_version));
3290 return XFS_ERROR(EIO);
3293 /* LR body must have data or it wouldn't have been written */
3294 hlen = be32_to_cpu(rhead->h_len);
3295 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3296 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3297 XFS_ERRLEVEL_LOW, log->l_mp);
3298 return XFS_ERROR(EFSCORRUPTED);
3300 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3301 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3302 XFS_ERRLEVEL_LOW, log->l_mp);
3303 return XFS_ERROR(EFSCORRUPTED);
3309 * Read the log from tail to head and process the log records found.
3310 * Handle the two cases where the tail and head are in the same cycle
3311 * and where the active portion of the log wraps around the end of
3312 * the physical log separately. The pass parameter is passed through
3313 * to the routines called to process the data and is not looked at
3317 xlog_do_recovery_pass(
3319 xfs_daddr_t head_blk,
3320 xfs_daddr_t tail_blk,
3323 xlog_rec_header_t *rhead;
3326 xfs_buf_t *hbp, *dbp;
3327 int error = 0, h_size;
3328 int bblks, split_bblks;
3329 int hblks, split_hblks, wrapped_hblks;
3330 struct hlist_head rhash[XLOG_RHASH_SIZE];
3332 ASSERT(head_blk != tail_blk);
3335 * Read the header of the tail block and get the iclog buffer size from
3336 * h_size. Use this to tell how many sectors make up the log header.
3338 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3340 * When using variable length iclogs, read first sector of
3341 * iclog header and extract the header size from it. Get a
3342 * new hbp that is the correct size.
3344 hbp = xlog_get_bp(log, 1);
3348 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
3352 rhead = (xlog_rec_header_t *)offset;
3353 error = xlog_valid_rec_header(log, rhead, tail_blk);
3356 h_size = be32_to_cpu(rhead->h_size);
3357 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
3358 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3359 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3360 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3363 hbp = xlog_get_bp(log, hblks);
3368 ASSERT(log->l_sectBBsize == 1);
3370 hbp = xlog_get_bp(log, 1);
3371 h_size = XLOG_BIG_RECORD_BSIZE;
3376 dbp = xlog_get_bp(log, BTOBB(h_size));
3382 memset(rhash, 0, sizeof(rhash));
3383 if (tail_blk <= head_blk) {
3384 for (blk_no = tail_blk; blk_no < head_blk; ) {
3385 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3389 rhead = (xlog_rec_header_t *)offset;
3390 error = xlog_valid_rec_header(log, rhead, blk_no);
3394 /* blocks in data section */
3395 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3396 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
3401 xlog_unpack_data(rhead, offset, log);
3402 if ((error = xlog_recover_process_data(log,
3403 rhash, rhead, offset, pass)))
3405 blk_no += bblks + hblks;
3409 * Perform recovery around the end of the physical log.
3410 * When the head is not on the same cycle number as the tail,
3411 * we can't do a sequential recovery as above.
3414 while (blk_no < log->l_logBBsize) {
3416 * Check for header wrapping around physical end-of-log
3418 offset = XFS_BUF_PTR(hbp);
3421 if (blk_no + hblks <= log->l_logBBsize) {
3422 /* Read header in one read */
3423 error = xlog_bread(log, blk_no, hblks, hbp,
3428 /* This LR is split across physical log end */
3429 if (blk_no != log->l_logBBsize) {
3430 /* some data before physical log end */
3431 ASSERT(blk_no <= INT_MAX);
3432 split_hblks = log->l_logBBsize - (int)blk_no;
3433 ASSERT(split_hblks > 0);
3434 error = xlog_bread(log, blk_no,
3442 * Note: this black magic still works with
3443 * large sector sizes (non-512) only because:
3444 * - we increased the buffer size originally
3445 * by 1 sector giving us enough extra space
3446 * for the second read;
3447 * - the log start is guaranteed to be sector
3449 * - we read the log end (LR header start)
3450 * _first_, then the log start (LR header end)
3451 * - order is important.
3453 wrapped_hblks = hblks - split_hblks;
3454 error = XFS_BUF_SET_PTR(hbp,
3455 offset + BBTOB(split_hblks),
3456 BBTOB(hblks - split_hblks));
3460 error = xlog_bread_noalign(log, 0,
3461 wrapped_hblks, hbp);
3465 error = XFS_BUF_SET_PTR(hbp, offset,
3470 rhead = (xlog_rec_header_t *)offset;
3471 error = xlog_valid_rec_header(log, rhead,
3472 split_hblks ? blk_no : 0);
3476 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3479 /* Read in data for log record */
3480 if (blk_no + bblks <= log->l_logBBsize) {
3481 error = xlog_bread(log, blk_no, bblks, dbp,
3486 /* This log record is split across the
3487 * physical end of log */
3488 offset = XFS_BUF_PTR(dbp);
3490 if (blk_no != log->l_logBBsize) {
3491 /* some data is before the physical
3493 ASSERT(!wrapped_hblks);
3494 ASSERT(blk_no <= INT_MAX);
3496 log->l_logBBsize - (int)blk_no;
3497 ASSERT(split_bblks > 0);
3498 error = xlog_bread(log, blk_no,
3506 * Note: this black magic still works with
3507 * large sector sizes (non-512) only because:
3508 * - we increased the buffer size originally
3509 * by 1 sector giving us enough extra space
3510 * for the second read;
3511 * - the log start is guaranteed to be sector
3513 * - we read the log end (LR header start)
3514 * _first_, then the log start (LR header end)
3515 * - order is important.
3517 error = XFS_BUF_SET_PTR(dbp,
3518 offset + BBTOB(split_bblks),
3519 BBTOB(bblks - split_bblks));
3523 error = xlog_bread_noalign(log, wrapped_hblks,
3524 bblks - split_bblks,
3529 error = XFS_BUF_SET_PTR(dbp, offset, h_size);
3533 xlog_unpack_data(rhead, offset, log);
3534 if ((error = xlog_recover_process_data(log, rhash,
3535 rhead, offset, pass)))
3540 ASSERT(blk_no >= log->l_logBBsize);
3541 blk_no -= log->l_logBBsize;
3543 /* read first part of physical log */
3544 while (blk_no < head_blk) {
3545 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3549 rhead = (xlog_rec_header_t *)offset;
3550 error = xlog_valid_rec_header(log, rhead, blk_no);
3554 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3555 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
3560 xlog_unpack_data(rhead, offset, log);
3561 if ((error = xlog_recover_process_data(log, rhash,
3562 rhead, offset, pass)))
3564 blk_no += bblks + hblks;
3576 * Do the recovery of the log. We actually do this in two phases.
3577 * The two passes are necessary in order to implement the function
3578 * of cancelling a record written into the log. The first pass
3579 * determines those things which have been cancelled, and the
3580 * second pass replays log items normally except for those which
3581 * have been cancelled. The handling of the replay and cancellations
3582 * takes place in the log item type specific routines.
3584 * The table of items which have cancel records in the log is allocated
3585 * and freed at this level, since only here do we know when all of
3586 * the log recovery has been completed.
3589 xlog_do_log_recovery(
3591 xfs_daddr_t head_blk,
3592 xfs_daddr_t tail_blk)
3596 ASSERT(head_blk != tail_blk);
3599 * First do a pass to find all of the cancelled buf log items.
3600 * Store them in the buf_cancel_table for use in the second pass.
3602 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
3603 sizeof(struct list_head),
3605 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3606 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
3608 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3609 XLOG_RECOVER_PASS1);
3611 kmem_free(log->l_buf_cancel_table);
3612 log->l_buf_cancel_table = NULL;
3616 * Then do a second pass to actually recover the items in the log.
3617 * When it is complete free the table of buf cancel items.
3619 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3620 XLOG_RECOVER_PASS2);
3625 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3626 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
3630 kmem_free(log->l_buf_cancel_table);
3631 log->l_buf_cancel_table = NULL;
3637 * Do the actual recovery
3642 xfs_daddr_t head_blk,
3643 xfs_daddr_t tail_blk)
3650 * First replay the images in the log.
3652 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3657 XFS_bflush(log->l_mp->m_ddev_targp);
3660 * If IO errors happened during recovery, bail out.
3662 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3667 * We now update the tail_lsn since much of the recovery has completed
3668 * and there may be space available to use. If there were no extent
3669 * or iunlinks, we can free up the entire log and set the tail_lsn to
3670 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3671 * lsn of the last known good LR on disk. If there are extent frees
3672 * or iunlinks they will have some entries in the AIL; so we look at
3673 * the AIL to determine how to set the tail_lsn.
3675 xlog_assign_tail_lsn(log->l_mp);
3678 * Now that we've finished replaying all buffer and inode
3679 * updates, re-read in the superblock.
3681 bp = xfs_getsb(log->l_mp, 0);
3683 ASSERT(!(XFS_BUF_ISWRITE(bp)));
3684 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
3686 XFS_BUF_UNASYNC(bp);
3687 xfsbdstrat(log->l_mp, bp);
3688 error = xfs_buf_iowait(bp);
3690 xfs_ioerror_alert("xlog_do_recover",
3691 log->l_mp, bp, XFS_BUF_ADDR(bp));
3697 /* Convert superblock from on-disk format */
3698 sbp = &log->l_mp->m_sb;
3699 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
3700 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
3701 ASSERT(xfs_sb_good_version(sbp));
3704 /* We've re-read the superblock so re-initialize per-cpu counters */
3705 xfs_icsb_reinit_counters(log->l_mp);
3707 xlog_recover_check_summary(log);
3709 /* Normal transactions can now occur */
3710 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3715 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3717 * Return error or zero.
3723 xfs_daddr_t head_blk, tail_blk;
3726 /* find the tail of the log */
3727 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
3730 if (tail_blk != head_blk) {
3731 /* There used to be a comment here:
3733 * disallow recovery on read-only mounts. note -- mount
3734 * checks for ENOSPC and turns it into an intelligent
3736 * ...but this is no longer true. Now, unless you specify
3737 * NORECOVERY (in which case this function would never be
3738 * called), we just go ahead and recover. We do this all
3739 * under the vfs layer, so we can get away with it unless
3740 * the device itself is read-only, in which case we fail.
3742 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
3747 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3748 log->l_mp->m_fsname, log->l_mp->m_logname ?
3749 log->l_mp->m_logname : "internal");
3751 error = xlog_do_recover(log, head_blk, tail_blk);
3752 log->l_flags |= XLOG_RECOVERY_NEEDED;
3758 * In the first part of recovery we replay inodes and buffers and build
3759 * up the list of extent free items which need to be processed. Here
3760 * we process the extent free items and clean up the on disk unlinked
3761 * inode lists. This is separated from the first part of recovery so
3762 * that the root and real-time bitmap inodes can be read in from disk in
3763 * between the two stages. This is necessary so that we can free space
3764 * in the real-time portion of the file system.
3767 xlog_recover_finish(
3771 * Now we're ready to do the transactions needed for the
3772 * rest of recovery. Start with completing all the extent
3773 * free intent records and then process the unlinked inode
3774 * lists. At this point, we essentially run in normal mode
3775 * except that we're still performing recovery actions
3776 * rather than accepting new requests.
3778 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
3780 error = xlog_recover_process_efis(log);
3783 "Failed to recover EFIs on filesystem: %s",
3784 log->l_mp->m_fsname);
3788 * Sync the log to get all the EFIs out of the AIL.
3789 * This isn't absolutely necessary, but it helps in
3790 * case the unlink transactions would have problems
3791 * pushing the EFIs out of the way.
3793 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
3795 xlog_recover_process_iunlinks(log);
3797 xlog_recover_check_summary(log);
3800 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3801 log->l_mp->m_fsname, log->l_mp->m_logname ?
3802 log->l_mp->m_logname : "internal");
3803 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
3806 "!Ending clean XFS mount for filesystem: %s\n",
3807 log->l_mp->m_fsname);
3815 * Read all of the agf and agi counters and check that they
3816 * are consistent with the superblock counters.
3819 xlog_recover_check_summary(
3826 xfs_agnumber_t agno;
3827 __uint64_t freeblks;
3837 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3838 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
3840 xfs_fs_cmn_err(CE_ALERT, mp,
3841 "xlog_recover_check_summary(agf)"
3842 "agf read failed agno %d error %d",
3845 agfp = XFS_BUF_TO_AGF(agfbp);
3846 freeblks += be32_to_cpu(agfp->agf_freeblks) +
3847 be32_to_cpu(agfp->agf_flcount);
3848 xfs_buf_relse(agfbp);
3851 error = xfs_read_agi(mp, NULL, agno, &agibp);
3853 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
3855 itotal += be32_to_cpu(agi->agi_count);
3856 ifree += be32_to_cpu(agi->agi_freecount);
3857 xfs_buf_relse(agibp);