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.
96 if (!xlog_buf_bbcount_valid(log, nbblks)) {
97 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
99 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
104 * We do log I/O in units of log sectors (a power-of-2
105 * multiple of the basic block size), so we round up the
106 * requested size to accommodate the basic blocks required
107 * for complete log sectors.
109 * In addition, the buffer may be used for a non-sector-
110 * aligned block offset, in which case an I/O of the
111 * requested size could extend beyond the end of the
112 * buffer. If the requested size is only 1 basic block it
113 * will never straddle a sector boundary, so this won't be
114 * an issue. Nor will this be a problem if the log I/O is
115 * done in basic blocks (sector size 1). But otherwise we
116 * extend the buffer by one extra log sector to ensure
117 * there's space to accommodate this possibility.
119 if (nbblks > 1 && log->l_sectBBsize > 1)
120 nbblks += log->l_sectBBsize;
121 nbblks = round_up(nbblks, log->l_sectBBsize);
123 bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, BBTOB(nbblks), 0);
137 * Return the address of the start of the given block number's data
138 * in a log buffer. The buffer covers a log sector-aligned region.
147 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
149 ASSERT(BBTOB(offset + nbblks) <= XFS_BUF_SIZE(bp));
150 return bp->b_addr + BBTOB(offset);
155 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
166 if (!xlog_buf_bbcount_valid(log, nbblks)) {
167 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
169 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
173 blk_no = round_down(blk_no, log->l_sectBBsize);
174 nbblks = round_up(nbblks, log->l_sectBBsize);
177 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
179 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
181 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
183 xfsbdstrat(log->l_mp, bp);
184 error = xfs_buf_iowait(bp);
186 xfs_ioerror_alert("xlog_bread", log->l_mp,
187 bp, XFS_BUF_ADDR(bp));
201 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
205 *offset = xlog_align(log, blk_no, nbblks, bp);
210 * Read at an offset into the buffer. Returns with the buffer in it's original
211 * state regardless of the result of the read.
216 xfs_daddr_t blk_no, /* block to read from */
217 int nbblks, /* blocks to read */
221 xfs_caddr_t orig_offset = bp->b_addr;
222 int orig_len = bp->b_buffer_length;
225 error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks));
229 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
231 /* must reset buffer pointer even on error */
232 error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len);
239 * Write out the buffer at the given block for the given number of blocks.
240 * The buffer is kept locked across the write and is returned locked.
241 * This can only be used for synchronous log writes.
252 if (!xlog_buf_bbcount_valid(log, nbblks)) {
253 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
255 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
259 blk_no = round_down(blk_no, log->l_sectBBsize);
260 nbblks = round_up(nbblks, log->l_sectBBsize);
263 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
265 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
266 XFS_BUF_ZEROFLAGS(bp);
269 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
271 error = xfs_bwrite(bp);
273 xfs_ioerror_alert("xlog_bwrite", log->l_mp,
274 bp, XFS_BUF_ADDR(bp));
282 * dump debug superblock and log record information
285 xlog_header_check_dump(
287 xlog_rec_header_t *head)
289 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d\n",
290 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
291 xfs_debug(mp, " log : uuid = %pU, fmt = %d\n",
292 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
295 #define xlog_header_check_dump(mp, head)
299 * check log record header for recovery
302 xlog_header_check_recover(
304 xlog_rec_header_t *head)
306 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
309 * IRIX doesn't write the h_fmt field and leaves it zeroed
310 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
311 * a dirty log created in IRIX.
313 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
315 "dirty log written in incompatible format - can't recover");
316 xlog_header_check_dump(mp, head);
317 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
318 XFS_ERRLEVEL_HIGH, mp);
319 return XFS_ERROR(EFSCORRUPTED);
320 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
322 "dirty log entry has mismatched uuid - can't recover");
323 xlog_header_check_dump(mp, head);
324 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
325 XFS_ERRLEVEL_HIGH, mp);
326 return XFS_ERROR(EFSCORRUPTED);
332 * read the head block of the log and check the header
335 xlog_header_check_mount(
337 xlog_rec_header_t *head)
339 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
341 if (uuid_is_nil(&head->h_fs_uuid)) {
343 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
344 * h_fs_uuid is nil, we assume this log was last mounted
345 * by IRIX and continue.
347 xfs_warn(mp, "nil uuid in log - IRIX style log");
348 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
349 xfs_warn(mp, "log has mismatched uuid - can't recover");
350 xlog_header_check_dump(mp, head);
351 XFS_ERROR_REPORT("xlog_header_check_mount",
352 XFS_ERRLEVEL_HIGH, mp);
353 return XFS_ERROR(EFSCORRUPTED);
364 * We're not going to bother about retrying
365 * this during recovery. One strike!
367 xfs_ioerror_alert("xlog_recover_iodone",
368 bp->b_target->bt_mount, bp,
370 xfs_force_shutdown(bp->b_target->bt_mount,
371 SHUTDOWN_META_IO_ERROR);
374 xfs_buf_ioend(bp, 0);
378 * This routine finds (to an approximation) the first block in the physical
379 * log which contains the given cycle. It uses a binary search algorithm.
380 * Note that the algorithm can not be perfect because the disk will not
381 * necessarily be perfect.
384 xlog_find_cycle_start(
387 xfs_daddr_t first_blk,
388 xfs_daddr_t *last_blk,
398 mid_blk = BLK_AVG(first_blk, end_blk);
399 while (mid_blk != first_blk && mid_blk != end_blk) {
400 error = xlog_bread(log, mid_blk, 1, bp, &offset);
403 mid_cycle = xlog_get_cycle(offset);
404 if (mid_cycle == cycle)
405 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
407 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
408 mid_blk = BLK_AVG(first_blk, end_blk);
410 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
411 (mid_blk == end_blk && mid_blk-1 == first_blk));
419 * Check that a range of blocks does not contain stop_on_cycle_no.
420 * Fill in *new_blk with the block offset where such a block is
421 * found, or with -1 (an invalid block number) if there is no such
422 * block in the range. The scan needs to occur from front to back
423 * and the pointer into the region must be updated since a later
424 * routine will need to perform another test.
427 xlog_find_verify_cycle(
429 xfs_daddr_t start_blk,
431 uint stop_on_cycle_no,
432 xfs_daddr_t *new_blk)
438 xfs_caddr_t buf = NULL;
442 * Greedily allocate a buffer big enough to handle the full
443 * range of basic blocks we'll be examining. If that fails,
444 * try a smaller size. We need to be able to read at least
445 * a log sector, or we're out of luck.
447 bufblks = 1 << ffs(nbblks);
448 while (!(bp = xlog_get_bp(log, bufblks))) {
450 if (bufblks < log->l_sectBBsize)
454 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
457 bcount = min(bufblks, (start_blk + nbblks - i));
459 error = xlog_bread(log, i, bcount, bp, &buf);
463 for (j = 0; j < bcount; j++) {
464 cycle = xlog_get_cycle(buf);
465 if (cycle == stop_on_cycle_no) {
482 * Potentially backup over partial log record write.
484 * In the typical case, last_blk is the number of the block directly after
485 * a good log record. Therefore, we subtract one to get the block number
486 * of the last block in the given buffer. extra_bblks contains the number
487 * of blocks we would have read on a previous read. This happens when the
488 * last log record is split over the end of the physical log.
490 * extra_bblks is the number of blocks potentially verified on a previous
491 * call to this routine.
494 xlog_find_verify_log_record(
496 xfs_daddr_t start_blk,
497 xfs_daddr_t *last_blk,
502 xfs_caddr_t offset = NULL;
503 xlog_rec_header_t *head = NULL;
506 int num_blks = *last_blk - start_blk;
509 ASSERT(start_blk != 0 || *last_blk != start_blk);
511 if (!(bp = xlog_get_bp(log, num_blks))) {
512 if (!(bp = xlog_get_bp(log, 1)))
516 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
519 offset += ((num_blks - 1) << BBSHIFT);
522 for (i = (*last_blk) - 1; i >= 0; i--) {
524 /* valid log record not found */
526 "Log inconsistent (didn't find previous header)");
528 error = XFS_ERROR(EIO);
533 error = xlog_bread(log, i, 1, bp, &offset);
538 head = (xlog_rec_header_t *)offset;
540 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
548 * We hit the beginning of the physical log & still no header. Return
549 * to caller. If caller can handle a return of -1, then this routine
550 * will be called again for the end of the physical log.
558 * We have the final block of the good log (the first block
559 * of the log record _before_ the head. So we check the uuid.
561 if ((error = xlog_header_check_mount(log->l_mp, head)))
565 * We may have found a log record header before we expected one.
566 * last_blk will be the 1st block # with a given cycle #. We may end
567 * up reading an entire log record. In this case, we don't want to
568 * reset last_blk. Only when last_blk points in the middle of a log
569 * record do we update last_blk.
571 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
572 uint h_size = be32_to_cpu(head->h_size);
574 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
575 if (h_size % XLOG_HEADER_CYCLE_SIZE)
581 if (*last_blk - i + extra_bblks !=
582 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
591 * Head is defined to be the point of the log where the next log write
592 * write could go. This means that incomplete LR writes at the end are
593 * eliminated when calculating the head. We aren't guaranteed that previous
594 * LR have complete transactions. We only know that a cycle number of
595 * current cycle number -1 won't be present in the log if we start writing
596 * from our current block number.
598 * last_blk contains the block number of the first block with a given
601 * Return: zero if normal, non-zero if error.
606 xfs_daddr_t *return_head_blk)
610 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
612 uint first_half_cycle, last_half_cycle;
614 int error, log_bbnum = log->l_logBBsize;
616 /* Is the end of the log device zeroed? */
617 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
618 *return_head_blk = first_blk;
620 /* Is the whole lot zeroed? */
622 /* Linux XFS shouldn't generate totally zeroed logs -
623 * mkfs etc write a dummy unmount record to a fresh
624 * log so we can store the uuid in there
626 xfs_warn(log->l_mp, "totally zeroed log");
631 xfs_warn(log->l_mp, "empty log check failed");
635 first_blk = 0; /* get cycle # of 1st block */
636 bp = xlog_get_bp(log, 1);
640 error = xlog_bread(log, 0, 1, bp, &offset);
644 first_half_cycle = xlog_get_cycle(offset);
646 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
647 error = xlog_bread(log, last_blk, 1, bp, &offset);
651 last_half_cycle = xlog_get_cycle(offset);
652 ASSERT(last_half_cycle != 0);
655 * If the 1st half cycle number is equal to the last half cycle number,
656 * then the entire log is stamped with the same cycle number. In this
657 * case, head_blk can't be set to zero (which makes sense). The below
658 * math doesn't work out properly with head_blk equal to zero. Instead,
659 * we set it to log_bbnum which is an invalid block number, but this
660 * value makes the math correct. If head_blk doesn't changed through
661 * all the tests below, *head_blk is set to zero at the very end rather
662 * than log_bbnum. In a sense, log_bbnum and zero are the same block
663 * in a circular file.
665 if (first_half_cycle == last_half_cycle) {
667 * In this case we believe that the entire log should have
668 * cycle number last_half_cycle. We need to scan backwards
669 * from the end verifying that there are no holes still
670 * containing last_half_cycle - 1. If we find such a hole,
671 * then the start of that hole will be the new head. The
672 * simple case looks like
673 * x | x ... | x - 1 | x
674 * Another case that fits this picture would be
675 * x | x + 1 | x ... | x
676 * In this case the head really is somewhere at the end of the
677 * log, as one of the latest writes at the beginning was
680 * x | x + 1 | x ... | x - 1 | x
681 * This is really the combination of the above two cases, and
682 * the head has to end up at the start of the x-1 hole at the
685 * In the 256k log case, we will read from the beginning to the
686 * end of the log and search for cycle numbers equal to x-1.
687 * We don't worry about the x+1 blocks that we encounter,
688 * because we know that they cannot be the head since the log
691 head_blk = log_bbnum;
692 stop_on_cycle = last_half_cycle - 1;
695 * In this case we want to find the first block with cycle
696 * number matching last_half_cycle. We expect the log to be
698 * x + 1 ... | x ... | x
699 * The first block with cycle number x (last_half_cycle) will
700 * be where the new head belongs. First we do a binary search
701 * for the first occurrence of last_half_cycle. The binary
702 * search may not be totally accurate, so then we scan back
703 * from there looking for occurrences of last_half_cycle before
704 * us. If that backwards scan wraps around the beginning of
705 * the log, then we look for occurrences of last_half_cycle - 1
706 * at the end of the log. The cases we're looking for look
708 * v binary search stopped here
709 * x + 1 ... | x | x + 1 | x ... | x
710 * ^ but we want to locate this spot
712 * <---------> less than scan distance
713 * x + 1 ... | x ... | x - 1 | x
714 * ^ we want to locate this spot
716 stop_on_cycle = last_half_cycle;
717 if ((error = xlog_find_cycle_start(log, bp, first_blk,
718 &head_blk, last_half_cycle)))
723 * Now validate the answer. Scan back some number of maximum possible
724 * blocks and make sure each one has the expected cycle number. The
725 * maximum is determined by the total possible amount of buffering
726 * in the in-core log. The following number can be made tighter if
727 * we actually look at the block size of the filesystem.
729 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
730 if (head_blk >= num_scan_bblks) {
732 * We are guaranteed that the entire check can be performed
735 start_blk = head_blk - num_scan_bblks;
736 if ((error = xlog_find_verify_cycle(log,
737 start_blk, num_scan_bblks,
738 stop_on_cycle, &new_blk)))
742 } else { /* need to read 2 parts of log */
744 * We are going to scan backwards in the log in two parts.
745 * First we scan the physical end of the log. In this part
746 * of the log, we are looking for blocks with cycle number
747 * last_half_cycle - 1.
748 * If we find one, then we know that the log starts there, as
749 * we've found a hole that didn't get written in going around
750 * the end of the physical log. The simple case for this is
751 * x + 1 ... | x ... | x - 1 | x
752 * <---------> less than scan distance
753 * If all of the blocks at the end of the log have cycle number
754 * last_half_cycle, then we check the blocks at the start of
755 * the log looking for occurrences of last_half_cycle. If we
756 * find one, then our current estimate for the location of the
757 * first occurrence of last_half_cycle is wrong and we move
758 * back to the hole we've found. This case looks like
759 * x + 1 ... | x | x + 1 | x ...
760 * ^ binary search stopped here
761 * Another case we need to handle that only occurs in 256k
763 * x + 1 ... | x ... | x+1 | x ...
764 * ^ binary search stops here
765 * In a 256k log, the scan at the end of the log will see the
766 * x + 1 blocks. We need to skip past those since that is
767 * certainly not the head of the log. By searching for
768 * last_half_cycle-1 we accomplish that.
770 ASSERT(head_blk <= INT_MAX &&
771 (xfs_daddr_t) num_scan_bblks >= head_blk);
772 start_blk = log_bbnum - (num_scan_bblks - head_blk);
773 if ((error = xlog_find_verify_cycle(log, start_blk,
774 num_scan_bblks - (int)head_blk,
775 (stop_on_cycle - 1), &new_blk)))
783 * Scan beginning of log now. The last part of the physical
784 * log is good. This scan needs to verify that it doesn't find
785 * the last_half_cycle.
788 ASSERT(head_blk <= INT_MAX);
789 if ((error = xlog_find_verify_cycle(log,
790 start_blk, (int)head_blk,
791 stop_on_cycle, &new_blk)))
799 * Now we need to make sure head_blk is not pointing to a block in
800 * the middle of a log record.
802 num_scan_bblks = XLOG_REC_SHIFT(log);
803 if (head_blk >= num_scan_bblks) {
804 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
806 /* start ptr at last block ptr before head_blk */
807 if ((error = xlog_find_verify_log_record(log, start_blk,
808 &head_blk, 0)) == -1) {
809 error = XFS_ERROR(EIO);
815 ASSERT(head_blk <= INT_MAX);
816 if ((error = xlog_find_verify_log_record(log, start_blk,
817 &head_blk, 0)) == -1) {
818 /* We hit the beginning of the log during our search */
819 start_blk = log_bbnum - (num_scan_bblks - head_blk);
821 ASSERT(start_blk <= INT_MAX &&
822 (xfs_daddr_t) log_bbnum-start_blk >= 0);
823 ASSERT(head_blk <= INT_MAX);
824 if ((error = xlog_find_verify_log_record(log,
826 (int)head_blk)) == -1) {
827 error = XFS_ERROR(EIO);
831 if (new_blk != log_bbnum)
838 if (head_blk == log_bbnum)
839 *return_head_blk = 0;
841 *return_head_blk = head_blk;
843 * When returning here, we have a good block number. Bad block
844 * means that during a previous crash, we didn't have a clean break
845 * from cycle number N to cycle number N-1. In this case, we need
846 * to find the first block with cycle number N-1.
854 xfs_warn(log->l_mp, "failed to find log head");
859 * Find the sync block number or the tail of the log.
861 * This will be the block number of the last record to have its
862 * associated buffers synced to disk. Every log record header has
863 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
864 * to get a sync block number. The only concern is to figure out which
865 * log record header to believe.
867 * The following algorithm uses the log record header with the largest
868 * lsn. The entire log record does not need to be valid. We only care
869 * that the header is valid.
871 * We could speed up search by using current head_blk buffer, but it is not
877 xfs_daddr_t *head_blk,
878 xfs_daddr_t *tail_blk)
880 xlog_rec_header_t *rhead;
881 xlog_op_header_t *op_head;
882 xfs_caddr_t offset = NULL;
885 xfs_daddr_t umount_data_blk;
886 xfs_daddr_t after_umount_blk;
893 * Find previous log record
895 if ((error = xlog_find_head(log, head_blk)))
898 bp = xlog_get_bp(log, 1);
901 if (*head_blk == 0) { /* special case */
902 error = xlog_bread(log, 0, 1, bp, &offset);
906 if (xlog_get_cycle(offset) == 0) {
908 /* leave all other log inited values alone */
914 * Search backwards looking for log record header block
916 ASSERT(*head_blk < INT_MAX);
917 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
918 error = xlog_bread(log, i, 1, bp, &offset);
922 if (*(__be32 *)offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
928 * If we haven't found the log record header block, start looking
929 * again from the end of the physical log. XXXmiken: There should be
930 * a check here to make sure we didn't search more than N blocks in
934 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
935 error = xlog_bread(log, i, 1, bp, &offset);
939 if (*(__be32 *)offset ==
940 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
947 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
949 return XFS_ERROR(EIO);
952 /* find blk_no of tail of log */
953 rhead = (xlog_rec_header_t *)offset;
954 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
957 * Reset log values according to the state of the log when we
958 * crashed. In the case where head_blk == 0, we bump curr_cycle
959 * one because the next write starts a new cycle rather than
960 * continuing the cycle of the last good log record. At this
961 * point we have guaranteed that all partial log records have been
962 * accounted for. Therefore, we know that the last good log record
963 * written was complete and ended exactly on the end boundary
964 * of the physical log.
966 log->l_prev_block = i;
967 log->l_curr_block = (int)*head_blk;
968 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
971 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
972 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
973 xlog_assign_grant_head(&log->l_grant_reserve_head, log->l_curr_cycle,
974 BBTOB(log->l_curr_block));
975 xlog_assign_grant_head(&log->l_grant_write_head, log->l_curr_cycle,
976 BBTOB(log->l_curr_block));
979 * Look for unmount record. If we find it, then we know there
980 * was a clean unmount. Since 'i' could be the last block in
981 * the physical log, we convert to a log block before comparing
984 * Save the current tail lsn to use to pass to
985 * xlog_clear_stale_blocks() below. We won't want to clear the
986 * unmount record if there is one, so we pass the lsn of the
987 * unmount record rather than the block after it.
989 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
990 int h_size = be32_to_cpu(rhead->h_size);
991 int h_version = be32_to_cpu(rhead->h_version);
993 if ((h_version & XLOG_VERSION_2) &&
994 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
995 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
996 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1004 after_umount_blk = (i + hblks + (int)
1005 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
1006 tail_lsn = atomic64_read(&log->l_tail_lsn);
1007 if (*head_blk == after_umount_blk &&
1008 be32_to_cpu(rhead->h_num_logops) == 1) {
1009 umount_data_blk = (i + hblks) % log->l_logBBsize;
1010 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1014 op_head = (xlog_op_header_t *)offset;
1015 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1017 * Set tail and last sync so that newly written
1018 * log records will point recovery to after the
1019 * current unmount record.
1021 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1022 log->l_curr_cycle, after_umount_blk);
1023 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1024 log->l_curr_cycle, after_umount_blk);
1025 *tail_blk = after_umount_blk;
1028 * Note that the unmount was clean. If the unmount
1029 * was not clean, we need to know this to rebuild the
1030 * superblock counters from the perag headers if we
1031 * have a filesystem using non-persistent counters.
1033 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1038 * Make sure that there are no blocks in front of the head
1039 * with the same cycle number as the head. This can happen
1040 * because we allow multiple outstanding log writes concurrently,
1041 * and the later writes might make it out before earlier ones.
1043 * We use the lsn from before modifying it so that we'll never
1044 * overwrite the unmount record after a clean unmount.
1046 * Do this only if we are going to recover the filesystem
1048 * NOTE: This used to say "if (!readonly)"
1049 * However on Linux, we can & do recover a read-only filesystem.
1050 * We only skip recovery if NORECOVERY is specified on mount,
1051 * in which case we would not be here.
1053 * But... if the -device- itself is readonly, just skip this.
1054 * We can't recover this device anyway, so it won't matter.
1056 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
1057 error = xlog_clear_stale_blocks(log, tail_lsn);
1063 xfs_warn(log->l_mp, "failed to locate log tail");
1068 * Is the log zeroed at all?
1070 * The last binary search should be changed to perform an X block read
1071 * once X becomes small enough. You can then search linearly through
1072 * the X blocks. This will cut down on the number of reads we need to do.
1074 * If the log is partially zeroed, this routine will pass back the blkno
1075 * of the first block with cycle number 0. It won't have a complete LR
1079 * 0 => the log is completely written to
1080 * -1 => use *blk_no as the first block of the log
1081 * >0 => error has occurred
1086 xfs_daddr_t *blk_no)
1090 uint first_cycle, last_cycle;
1091 xfs_daddr_t new_blk, last_blk, start_blk;
1092 xfs_daddr_t num_scan_bblks;
1093 int error, log_bbnum = log->l_logBBsize;
1097 /* check totally zeroed log */
1098 bp = xlog_get_bp(log, 1);
1101 error = xlog_bread(log, 0, 1, bp, &offset);
1105 first_cycle = xlog_get_cycle(offset);
1106 if (first_cycle == 0) { /* completely zeroed log */
1112 /* check partially zeroed log */
1113 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1117 last_cycle = xlog_get_cycle(offset);
1118 if (last_cycle != 0) { /* log completely written to */
1121 } else if (first_cycle != 1) {
1123 * If the cycle of the last block is zero, the cycle of
1124 * the first block must be 1. If it's not, maybe we're
1125 * not looking at a log... Bail out.
1128 "Log inconsistent or not a log (last==0, first!=1)");
1129 return XFS_ERROR(EINVAL);
1132 /* we have a partially zeroed log */
1133 last_blk = log_bbnum-1;
1134 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1138 * Validate the answer. Because there is no way to guarantee that
1139 * the entire log is made up of log records which are the same size,
1140 * we scan over the defined maximum blocks. At this point, the maximum
1141 * is not chosen to mean anything special. XXXmiken
1143 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1144 ASSERT(num_scan_bblks <= INT_MAX);
1146 if (last_blk < num_scan_bblks)
1147 num_scan_bblks = last_blk;
1148 start_blk = last_blk - num_scan_bblks;
1151 * We search for any instances of cycle number 0 that occur before
1152 * our current estimate of the head. What we're trying to detect is
1153 * 1 ... | 0 | 1 | 0...
1154 * ^ binary search ends here
1156 if ((error = xlog_find_verify_cycle(log, start_blk,
1157 (int)num_scan_bblks, 0, &new_blk)))
1163 * Potentially backup over partial log record write. We don't need
1164 * to search the end of the log because we know it is zero.
1166 if ((error = xlog_find_verify_log_record(log, start_blk,
1167 &last_blk, 0)) == -1) {
1168 error = XFS_ERROR(EIO);
1182 * These are simple subroutines used by xlog_clear_stale_blocks() below
1183 * to initialize a buffer full of empty log record headers and write
1184 * them into the log.
1195 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1197 memset(buf, 0, BBSIZE);
1198 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1199 recp->h_cycle = cpu_to_be32(cycle);
1200 recp->h_version = cpu_to_be32(
1201 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1202 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1203 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1204 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1205 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1209 xlog_write_log_records(
1220 int sectbb = log->l_sectBBsize;
1221 int end_block = start_block + blocks;
1227 * Greedily allocate a buffer big enough to handle the full
1228 * range of basic blocks to be written. If that fails, try
1229 * a smaller size. We need to be able to write at least a
1230 * log sector, or we're out of luck.
1232 bufblks = 1 << ffs(blocks);
1233 while (!(bp = xlog_get_bp(log, bufblks))) {
1235 if (bufblks < sectbb)
1239 /* We may need to do a read at the start to fill in part of
1240 * the buffer in the starting sector not covered by the first
1243 balign = round_down(start_block, sectbb);
1244 if (balign != start_block) {
1245 error = xlog_bread_noalign(log, start_block, 1, bp);
1249 j = start_block - balign;
1252 for (i = start_block; i < end_block; i += bufblks) {
1253 int bcount, endcount;
1255 bcount = min(bufblks, end_block - start_block);
1256 endcount = bcount - j;
1258 /* We may need to do a read at the end to fill in part of
1259 * the buffer in the final sector not covered by the write.
1260 * If this is the same sector as the above read, skip it.
1262 ealign = round_down(end_block, sectbb);
1263 if (j == 0 && (start_block + endcount > ealign)) {
1264 offset = bp->b_addr + BBTOB(ealign - start_block);
1265 error = xlog_bread_offset(log, ealign, sectbb,
1272 offset = xlog_align(log, start_block, endcount, bp);
1273 for (; j < endcount; j++) {
1274 xlog_add_record(log, offset, cycle, i+j,
1275 tail_cycle, tail_block);
1278 error = xlog_bwrite(log, start_block, endcount, bp);
1281 start_block += endcount;
1291 * This routine is called to blow away any incomplete log writes out
1292 * in front of the log head. We do this so that we won't become confused
1293 * if we come up, write only a little bit more, and then crash again.
1294 * If we leave the partial log records out there, this situation could
1295 * cause us to think those partial writes are valid blocks since they
1296 * have the current cycle number. We get rid of them by overwriting them
1297 * with empty log records with the old cycle number rather than the
1300 * The tail lsn is passed in rather than taken from
1301 * the log so that we will not write over the unmount record after a
1302 * clean unmount in a 512 block log. Doing so would leave the log without
1303 * any valid log records in it until a new one was written. If we crashed
1304 * during that time we would not be able to recover.
1307 xlog_clear_stale_blocks(
1311 int tail_cycle, head_cycle;
1312 int tail_block, head_block;
1313 int tail_distance, max_distance;
1317 tail_cycle = CYCLE_LSN(tail_lsn);
1318 tail_block = BLOCK_LSN(tail_lsn);
1319 head_cycle = log->l_curr_cycle;
1320 head_block = log->l_curr_block;
1323 * Figure out the distance between the new head of the log
1324 * and the tail. We want to write over any blocks beyond the
1325 * head that we may have written just before the crash, but
1326 * we don't want to overwrite the tail of the log.
1328 if (head_cycle == tail_cycle) {
1330 * The tail is behind the head in the physical log,
1331 * so the distance from the head to the tail is the
1332 * distance from the head to the end of the log plus
1333 * the distance from the beginning of the log to the
1336 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1337 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1338 XFS_ERRLEVEL_LOW, log->l_mp);
1339 return XFS_ERROR(EFSCORRUPTED);
1341 tail_distance = tail_block + (log->l_logBBsize - head_block);
1344 * The head is behind the tail in the physical log,
1345 * so the distance from the head to the tail is just
1346 * the tail block minus the head block.
1348 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1349 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1350 XFS_ERRLEVEL_LOW, log->l_mp);
1351 return XFS_ERROR(EFSCORRUPTED);
1353 tail_distance = tail_block - head_block;
1357 * If the head is right up against the tail, we can't clear
1360 if (tail_distance <= 0) {
1361 ASSERT(tail_distance == 0);
1365 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1367 * Take the smaller of the maximum amount of outstanding I/O
1368 * we could have and the distance to the tail to clear out.
1369 * We take the smaller so that we don't overwrite the tail and
1370 * we don't waste all day writing from the head to the tail
1373 max_distance = MIN(max_distance, tail_distance);
1375 if ((head_block + max_distance) <= log->l_logBBsize) {
1377 * We can stomp all the blocks we need to without
1378 * wrapping around the end of the log. Just do it
1379 * in a single write. Use the cycle number of the
1380 * current cycle minus one so that the log will look like:
1383 error = xlog_write_log_records(log, (head_cycle - 1),
1384 head_block, max_distance, tail_cycle,
1390 * We need to wrap around the end of the physical log in
1391 * order to clear all the blocks. Do it in two separate
1392 * I/Os. The first write should be from the head to the
1393 * end of the physical log, and it should use the current
1394 * cycle number minus one just like above.
1396 distance = log->l_logBBsize - head_block;
1397 error = xlog_write_log_records(log, (head_cycle - 1),
1398 head_block, distance, tail_cycle,
1405 * Now write the blocks at the start of the physical log.
1406 * This writes the remainder of the blocks we want to clear.
1407 * It uses the current cycle number since we're now on the
1408 * same cycle as the head so that we get:
1409 * n ... n ... | n - 1 ...
1410 * ^^^^^ blocks we're writing
1412 distance = max_distance - (log->l_logBBsize - head_block);
1413 error = xlog_write_log_records(log, head_cycle, 0, distance,
1414 tail_cycle, tail_block);
1422 /******************************************************************************
1424 * Log recover routines
1426 ******************************************************************************
1429 STATIC xlog_recover_t *
1430 xlog_recover_find_tid(
1431 struct hlist_head *head,
1434 xlog_recover_t *trans;
1435 struct hlist_node *n;
1437 hlist_for_each_entry(trans, n, head, r_list) {
1438 if (trans->r_log_tid == tid)
1445 xlog_recover_new_tid(
1446 struct hlist_head *head,
1450 xlog_recover_t *trans;
1452 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1453 trans->r_log_tid = tid;
1455 INIT_LIST_HEAD(&trans->r_itemq);
1457 INIT_HLIST_NODE(&trans->r_list);
1458 hlist_add_head(&trans->r_list, head);
1462 xlog_recover_add_item(
1463 struct list_head *head)
1465 xlog_recover_item_t *item;
1467 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1468 INIT_LIST_HEAD(&item->ri_list);
1469 list_add_tail(&item->ri_list, head);
1473 xlog_recover_add_to_cont_trans(
1475 xlog_recover_t *trans,
1479 xlog_recover_item_t *item;
1480 xfs_caddr_t ptr, old_ptr;
1483 if (list_empty(&trans->r_itemq)) {
1484 /* finish copying rest of trans header */
1485 xlog_recover_add_item(&trans->r_itemq);
1486 ptr = (xfs_caddr_t) &trans->r_theader +
1487 sizeof(xfs_trans_header_t) - len;
1488 memcpy(ptr, dp, len); /* d, s, l */
1491 /* take the tail entry */
1492 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1494 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1495 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1497 ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
1498 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1499 item->ri_buf[item->ri_cnt-1].i_len += len;
1500 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1501 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
1506 * The next region to add is the start of a new region. It could be
1507 * a whole region or it could be the first part of a new region. Because
1508 * of this, the assumption here is that the type and size fields of all
1509 * format structures fit into the first 32 bits of the structure.
1511 * This works because all regions must be 32 bit aligned. Therefore, we
1512 * either have both fields or we have neither field. In the case we have
1513 * neither field, the data part of the region is zero length. We only have
1514 * a log_op_header and can throw away the header since a new one will appear
1515 * later. If we have at least 4 bytes, then we can determine how many regions
1516 * will appear in the current log item.
1519 xlog_recover_add_to_trans(
1521 xlog_recover_t *trans,
1525 xfs_inode_log_format_t *in_f; /* any will do */
1526 xlog_recover_item_t *item;
1531 if (list_empty(&trans->r_itemq)) {
1532 /* we need to catch log corruptions here */
1533 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
1534 xfs_warn(log->l_mp, "%s: bad header magic number",
1537 return XFS_ERROR(EIO);
1539 if (len == sizeof(xfs_trans_header_t))
1540 xlog_recover_add_item(&trans->r_itemq);
1541 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1545 ptr = kmem_alloc(len, KM_SLEEP);
1546 memcpy(ptr, dp, len);
1547 in_f = (xfs_inode_log_format_t *)ptr;
1549 /* take the tail entry */
1550 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1551 if (item->ri_total != 0 &&
1552 item->ri_total == item->ri_cnt) {
1553 /* tail item is in use, get a new one */
1554 xlog_recover_add_item(&trans->r_itemq);
1555 item = list_entry(trans->r_itemq.prev,
1556 xlog_recover_item_t, ri_list);
1559 if (item->ri_total == 0) { /* first region to be added */
1560 if (in_f->ilf_size == 0 ||
1561 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
1563 "bad number of regions (%d) in inode log format",
1566 return XFS_ERROR(EIO);
1569 item->ri_total = in_f->ilf_size;
1571 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1574 ASSERT(item->ri_total > item->ri_cnt);
1575 /* Description region is ri_buf[0] */
1576 item->ri_buf[item->ri_cnt].i_addr = ptr;
1577 item->ri_buf[item->ri_cnt].i_len = len;
1579 trace_xfs_log_recover_item_add(log, trans, item, 0);
1584 * Sort the log items in the transaction. Cancelled buffers need
1585 * to be put first so they are processed before any items that might
1586 * modify the buffers. If they are cancelled, then the modifications
1587 * don't need to be replayed.
1590 xlog_recover_reorder_trans(
1592 xlog_recover_t *trans,
1595 xlog_recover_item_t *item, *n;
1596 LIST_HEAD(sort_list);
1598 list_splice_init(&trans->r_itemq, &sort_list);
1599 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1600 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1602 switch (ITEM_TYPE(item)) {
1604 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1605 trace_xfs_log_recover_item_reorder_head(log,
1607 list_move(&item->ri_list, &trans->r_itemq);
1612 case XFS_LI_QUOTAOFF:
1615 trace_xfs_log_recover_item_reorder_tail(log,
1617 list_move_tail(&item->ri_list, &trans->r_itemq);
1621 "%s: unrecognized type of log operation",
1624 return XFS_ERROR(EIO);
1627 ASSERT(list_empty(&sort_list));
1632 * Build up the table of buf cancel records so that we don't replay
1633 * cancelled data in the second pass. For buffer records that are
1634 * not cancel records, there is nothing to do here so we just return.
1636 * If we get a cancel record which is already in the table, this indicates
1637 * that the buffer was cancelled multiple times. In order to ensure
1638 * that during pass 2 we keep the record in the table until we reach its
1639 * last occurrence in the log, we keep a reference count in the cancel
1640 * record in the table to tell us how many times we expect to see this
1641 * record during the second pass.
1644 xlog_recover_buffer_pass1(
1646 xlog_recover_item_t *item)
1648 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1649 struct list_head *bucket;
1650 struct xfs_buf_cancel *bcp;
1653 * If this isn't a cancel buffer item, then just return.
1655 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1656 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
1661 * Insert an xfs_buf_cancel record into the hash table of them.
1662 * If there is already an identical record, bump its reference count.
1664 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1665 list_for_each_entry(bcp, bucket, bc_list) {
1666 if (bcp->bc_blkno == buf_f->blf_blkno &&
1667 bcp->bc_len == buf_f->blf_len) {
1669 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
1674 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1675 bcp->bc_blkno = buf_f->blf_blkno;
1676 bcp->bc_len = buf_f->blf_len;
1677 bcp->bc_refcount = 1;
1678 list_add_tail(&bcp->bc_list, bucket);
1680 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
1685 * Check to see whether the buffer being recovered has a corresponding
1686 * entry in the buffer cancel record table. If it does then return 1
1687 * so that it will be cancelled, otherwise return 0. If the buffer is
1688 * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement
1689 * the refcount on the entry in the table and remove it from the table
1690 * if this is the last reference.
1692 * We remove the cancel record from the table when we encounter its
1693 * last occurrence in the log so that if the same buffer is re-used
1694 * again after its last cancellation we actually replay the changes
1695 * made at that point.
1698 xlog_check_buffer_cancelled(
1704 struct list_head *bucket;
1705 struct xfs_buf_cancel *bcp;
1707 if (log->l_buf_cancel_table == NULL) {
1709 * There is nothing in the table built in pass one,
1710 * so this buffer must not be cancelled.
1712 ASSERT(!(flags & XFS_BLF_CANCEL));
1717 * Search for an entry in the cancel table that matches our buffer.
1719 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1720 list_for_each_entry(bcp, bucket, bc_list) {
1721 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
1726 * We didn't find a corresponding entry in the table, so return 0 so
1727 * that the buffer is NOT cancelled.
1729 ASSERT(!(flags & XFS_BLF_CANCEL));
1734 * We've go a match, so return 1 so that the recovery of this buffer
1735 * is cancelled. If this buffer is actually a buffer cancel log
1736 * item, then decrement the refcount on the one in the table and
1737 * remove it if this is the last reference.
1739 if (flags & XFS_BLF_CANCEL) {
1740 if (--bcp->bc_refcount == 0) {
1741 list_del(&bcp->bc_list);
1749 * Perform recovery for a buffer full of inodes. In these buffers, the only
1750 * data which should be recovered is that which corresponds to the
1751 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1752 * data for the inodes is always logged through the inodes themselves rather
1753 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1755 * The only time when buffers full of inodes are fully recovered is when the
1756 * buffer is full of newly allocated inodes. In this case the buffer will
1757 * not be marked as an inode buffer and so will be sent to
1758 * xlog_recover_do_reg_buffer() below during recovery.
1761 xlog_recover_do_inode_buffer(
1762 struct xfs_mount *mp,
1763 xlog_recover_item_t *item,
1765 xfs_buf_log_format_t *buf_f)
1771 int reg_buf_offset = 0;
1772 int reg_buf_bytes = 0;
1773 int next_unlinked_offset;
1775 xfs_agino_t *logged_nextp;
1776 xfs_agino_t *buffer_nextp;
1778 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
1780 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1781 for (i = 0; i < inodes_per_buf; i++) {
1782 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1783 offsetof(xfs_dinode_t, di_next_unlinked);
1785 while (next_unlinked_offset >=
1786 (reg_buf_offset + reg_buf_bytes)) {
1788 * The next di_next_unlinked field is beyond
1789 * the current logged region. Find the next
1790 * logged region that contains or is beyond
1791 * the current di_next_unlinked field.
1794 bit = xfs_next_bit(buf_f->blf_data_map,
1795 buf_f->blf_map_size, bit);
1798 * If there are no more logged regions in the
1799 * buffer, then we're done.
1804 nbits = xfs_contig_bits(buf_f->blf_data_map,
1805 buf_f->blf_map_size, bit);
1807 reg_buf_offset = bit << XFS_BLF_SHIFT;
1808 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
1813 * If the current logged region starts after the current
1814 * di_next_unlinked field, then move on to the next
1815 * di_next_unlinked field.
1817 if (next_unlinked_offset < reg_buf_offset)
1820 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1821 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
1822 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1825 * The current logged region contains a copy of the
1826 * current di_next_unlinked field. Extract its value
1827 * and copy it to the buffer copy.
1829 logged_nextp = item->ri_buf[item_index].i_addr +
1830 next_unlinked_offset - reg_buf_offset;
1831 if (unlikely(*logged_nextp == 0)) {
1833 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1834 "Trying to replay bad (0) inode di_next_unlinked field.",
1836 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1837 XFS_ERRLEVEL_LOW, mp);
1838 return XFS_ERROR(EFSCORRUPTED);
1841 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1842 next_unlinked_offset);
1843 *buffer_nextp = *logged_nextp;
1850 * Perform a 'normal' buffer recovery. Each logged region of the
1851 * buffer should be copied over the corresponding region in the
1852 * given buffer. The bitmap in the buf log format structure indicates
1853 * where to place the logged data.
1856 xlog_recover_do_reg_buffer(
1857 struct xfs_mount *mp,
1858 xlog_recover_item_t *item,
1860 xfs_buf_log_format_t *buf_f)
1867 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
1870 i = 1; /* 0 is the buf format structure */
1872 bit = xfs_next_bit(buf_f->blf_data_map,
1873 buf_f->blf_map_size, bit);
1876 nbits = xfs_contig_bits(buf_f->blf_data_map,
1877 buf_f->blf_map_size, bit);
1879 ASSERT(item->ri_buf[i].i_addr != NULL);
1880 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
1881 ASSERT(XFS_BUF_COUNT(bp) >=
1882 ((uint)bit << XFS_BLF_SHIFT)+(nbits<<XFS_BLF_SHIFT));
1885 * Do a sanity check if this is a dquot buffer. Just checking
1886 * the first dquot in the buffer should do. XXXThis is
1887 * probably a good thing to do for other buf types also.
1890 if (buf_f->blf_flags &
1891 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
1892 if (item->ri_buf[i].i_addr == NULL) {
1894 "XFS: NULL dquot in %s.", __func__);
1897 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
1899 "XFS: dquot too small (%d) in %s.",
1900 item->ri_buf[i].i_len, __func__);
1903 error = xfs_qm_dqcheck(mp, item->ri_buf[i].i_addr,
1904 -1, 0, XFS_QMOPT_DOWARN,
1905 "dquot_buf_recover");
1910 memcpy(xfs_buf_offset(bp,
1911 (uint)bit << XFS_BLF_SHIFT), /* dest */
1912 item->ri_buf[i].i_addr, /* source */
1913 nbits<<XFS_BLF_SHIFT); /* length */
1919 /* Shouldn't be any more regions */
1920 ASSERT(i == item->ri_total);
1924 * Do some primitive error checking on ondisk dquot data structures.
1928 struct xfs_mount *mp,
1929 xfs_disk_dquot_t *ddq,
1931 uint type, /* used only when IO_dorepair is true */
1935 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
1939 * We can encounter an uninitialized dquot buffer for 2 reasons:
1940 * 1. If we crash while deleting the quotainode(s), and those blks got
1941 * used for user data. This is because we take the path of regular
1942 * file deletion; however, the size field of quotainodes is never
1943 * updated, so all the tricks that we play in itruncate_finish
1944 * don't quite matter.
1946 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1947 * But the allocation will be replayed so we'll end up with an
1948 * uninitialized quota block.
1950 * This is all fine; things are still consistent, and we haven't lost
1951 * any quota information. Just don't complain about bad dquot blks.
1953 if (ddq->d_magic != cpu_to_be16(XFS_DQUOT_MAGIC)) {
1954 if (flags & XFS_QMOPT_DOWARN)
1956 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
1957 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
1960 if (ddq->d_version != XFS_DQUOT_VERSION) {
1961 if (flags & XFS_QMOPT_DOWARN)
1963 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
1964 str, id, ddq->d_version, XFS_DQUOT_VERSION);
1968 if (ddq->d_flags != XFS_DQ_USER &&
1969 ddq->d_flags != XFS_DQ_PROJ &&
1970 ddq->d_flags != XFS_DQ_GROUP) {
1971 if (flags & XFS_QMOPT_DOWARN)
1973 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
1974 str, id, ddq->d_flags);
1978 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
1979 if (flags & XFS_QMOPT_DOWARN)
1981 "%s : ondisk-dquot 0x%p, ID mismatch: "
1982 "0x%x expected, found id 0x%x",
1983 str, ddq, id, be32_to_cpu(ddq->d_id));
1987 if (!errs && ddq->d_id) {
1988 if (ddq->d_blk_softlimit &&
1989 be64_to_cpu(ddq->d_bcount) >=
1990 be64_to_cpu(ddq->d_blk_softlimit)) {
1991 if (!ddq->d_btimer) {
1992 if (flags & XFS_QMOPT_DOWARN)
1994 "%s : Dquot ID 0x%x (0x%p) BLK TIMER NOT STARTED",
1995 str, (int)be32_to_cpu(ddq->d_id), ddq);
1999 if (ddq->d_ino_softlimit &&
2000 be64_to_cpu(ddq->d_icount) >=
2001 be64_to_cpu(ddq->d_ino_softlimit)) {
2002 if (!ddq->d_itimer) {
2003 if (flags & XFS_QMOPT_DOWARN)
2005 "%s : Dquot ID 0x%x (0x%p) INODE TIMER NOT STARTED",
2006 str, (int)be32_to_cpu(ddq->d_id), ddq);
2010 if (ddq->d_rtb_softlimit &&
2011 be64_to_cpu(ddq->d_rtbcount) >=
2012 be64_to_cpu(ddq->d_rtb_softlimit)) {
2013 if (!ddq->d_rtbtimer) {
2014 if (flags & XFS_QMOPT_DOWARN)
2016 "%s : Dquot ID 0x%x (0x%p) RTBLK TIMER NOT STARTED",
2017 str, (int)be32_to_cpu(ddq->d_id), ddq);
2023 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2026 if (flags & XFS_QMOPT_DOWARN)
2027 xfs_notice(mp, "Re-initializing dquot ID 0x%x", id);
2030 * Typically, a repair is only requested by quotacheck.
2033 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2034 memset(d, 0, sizeof(xfs_dqblk_t));
2036 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2037 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2038 d->dd_diskdq.d_flags = type;
2039 d->dd_diskdq.d_id = cpu_to_be32(id);
2045 * Perform a dquot buffer recovery.
2046 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2047 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2048 * Else, treat it as a regular buffer and do recovery.
2051 xlog_recover_do_dquot_buffer(
2054 xlog_recover_item_t *item,
2056 xfs_buf_log_format_t *buf_f)
2060 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2063 * Filesystems are required to send in quota flags at mount time.
2065 if (mp->m_qflags == 0) {
2070 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2071 type |= XFS_DQ_USER;
2072 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2073 type |= XFS_DQ_PROJ;
2074 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2075 type |= XFS_DQ_GROUP;
2077 * This type of quotas was turned off, so ignore this buffer
2079 if (log->l_quotaoffs_flag & type)
2082 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2086 * This routine replays a modification made to a buffer at runtime.
2087 * There are actually two types of buffer, regular and inode, which
2088 * are handled differently. Inode buffers are handled differently
2089 * in that we only recover a specific set of data from them, namely
2090 * the inode di_next_unlinked fields. This is because all other inode
2091 * data is actually logged via inode records and any data we replay
2092 * here which overlaps that may be stale.
2094 * When meta-data buffers are freed at run time we log a buffer item
2095 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2096 * of the buffer in the log should not be replayed at recovery time.
2097 * This is so that if the blocks covered by the buffer are reused for
2098 * file data before we crash we don't end up replaying old, freed
2099 * meta-data into a user's file.
2101 * To handle the cancellation of buffer log items, we make two passes
2102 * over the log during recovery. During the first we build a table of
2103 * those buffers which have been cancelled, and during the second we
2104 * only replay those buffers which do not have corresponding cancel
2105 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2106 * for more details on the implementation of the table of cancel records.
2109 xlog_recover_buffer_pass2(
2111 xlog_recover_item_t *item)
2113 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2114 xfs_mount_t *mp = log->l_mp;
2120 * In this pass we only want to recover all the buffers which have
2121 * not been cancelled and are not cancellation buffers themselves.
2123 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2124 buf_f->blf_len, buf_f->blf_flags)) {
2125 trace_xfs_log_recover_buf_cancel(log, buf_f);
2129 trace_xfs_log_recover_buf_recover(log, buf_f);
2131 buf_flags = XBF_LOCK;
2132 if (!(buf_f->blf_flags & XFS_BLF_INODE_BUF))
2133 buf_flags |= XBF_MAPPED;
2135 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2138 return XFS_ERROR(ENOMEM);
2139 error = bp->b_error;
2141 xfs_ioerror_alert("xlog_recover_do..(read#1)", mp,
2142 bp, buf_f->blf_blkno);
2147 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
2148 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2149 } else if (buf_f->blf_flags &
2150 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2151 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2153 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2156 return XFS_ERROR(error);
2159 * Perform delayed write on the buffer. Asynchronous writes will be
2160 * slower when taking into account all the buffers to be flushed.
2162 * Also make sure that only inode buffers with good sizes stay in
2163 * the buffer cache. The kernel moves inodes in buffers of 1 block
2164 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2165 * buffers in the log can be a different size if the log was generated
2166 * by an older kernel using unclustered inode buffers or a newer kernel
2167 * running with a different inode cluster size. Regardless, if the
2168 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2169 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2170 * the buffer out of the buffer cache so that the buffer won't
2171 * overlap with future reads of those inodes.
2173 if (XFS_DINODE_MAGIC ==
2174 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2175 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2176 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2178 error = xfs_bwrite(bp);
2180 ASSERT(bp->b_target->bt_mount == mp);
2181 bp->b_iodone = xlog_recover_iodone;
2182 xfs_buf_delwri_queue(bp);
2190 xlog_recover_inode_pass2(
2192 xlog_recover_item_t *item)
2194 xfs_inode_log_format_t *in_f;
2195 xfs_mount_t *mp = log->l_mp;
2204 xfs_icdinode_t *dicp;
2207 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2208 in_f = item->ri_buf[0].i_addr;
2210 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
2212 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2218 * Inode buffers can be freed, look out for it,
2219 * and do not replay the inode.
2221 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2222 in_f->ilf_len, 0)) {
2224 trace_xfs_log_recover_inode_cancel(log, in_f);
2227 trace_xfs_log_recover_inode_recover(log, in_f);
2229 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len,
2235 error = bp->b_error;
2237 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
2238 bp, in_f->ilf_blkno);
2242 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2243 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
2246 * Make sure the place we're flushing out to really looks
2249 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
2252 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2253 __func__, dip, bp, in_f->ilf_ino);
2254 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2255 XFS_ERRLEVEL_LOW, mp);
2256 error = EFSCORRUPTED;
2259 dicp = item->ri_buf[1].i_addr;
2260 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2263 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2264 __func__, item, in_f->ilf_ino);
2265 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2266 XFS_ERRLEVEL_LOW, mp);
2267 error = EFSCORRUPTED;
2271 /* Skip replay when the on disk inode is newer than the log one */
2272 if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
2274 * Deal with the wrap case, DI_MAX_FLUSH is less
2275 * than smaller numbers
2277 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
2278 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2282 trace_xfs_log_recover_inode_skip(log, in_f);
2287 /* Take the opportunity to reset the flush iteration count */
2288 dicp->di_flushiter = 0;
2290 if (unlikely(S_ISREG(dicp->di_mode))) {
2291 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2292 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2293 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2294 XFS_ERRLEVEL_LOW, mp, dicp);
2297 "%s: Bad regular inode log record, rec ptr 0x%p, "
2298 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2299 __func__, item, dip, bp, in_f->ilf_ino);
2300 error = EFSCORRUPTED;
2303 } else if (unlikely(S_ISDIR(dicp->di_mode))) {
2304 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2305 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2306 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2307 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2308 XFS_ERRLEVEL_LOW, mp, dicp);
2311 "%s: Bad dir inode log record, rec ptr 0x%p, "
2312 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2313 __func__, item, dip, bp, in_f->ilf_ino);
2314 error = EFSCORRUPTED;
2318 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2319 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2320 XFS_ERRLEVEL_LOW, mp, dicp);
2323 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2324 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2325 __func__, item, dip, bp, in_f->ilf_ino,
2326 dicp->di_nextents + dicp->di_anextents,
2328 error = EFSCORRUPTED;
2331 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2332 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2333 XFS_ERRLEVEL_LOW, mp, dicp);
2336 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2337 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
2338 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
2339 error = EFSCORRUPTED;
2342 if (unlikely(item->ri_buf[1].i_len > sizeof(struct xfs_icdinode))) {
2343 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
2344 XFS_ERRLEVEL_LOW, mp, dicp);
2347 "%s: Bad inode log record length %d, rec ptr 0x%p",
2348 __func__, item->ri_buf[1].i_len, item);
2349 error = EFSCORRUPTED;
2353 /* The core is in in-core format */
2354 xfs_dinode_to_disk(dip, item->ri_buf[1].i_addr);
2356 /* the rest is in on-disk format */
2357 if (item->ri_buf[1].i_len > sizeof(struct xfs_icdinode)) {
2358 memcpy((xfs_caddr_t) dip + sizeof(struct xfs_icdinode),
2359 item->ri_buf[1].i_addr + sizeof(struct xfs_icdinode),
2360 item->ri_buf[1].i_len - sizeof(struct xfs_icdinode));
2363 fields = in_f->ilf_fields;
2364 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2366 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
2369 memcpy(XFS_DFORK_DPTR(dip),
2370 &in_f->ilf_u.ilfu_uuid,
2375 if (in_f->ilf_size == 2)
2376 goto write_inode_buffer;
2377 len = item->ri_buf[2].i_len;
2378 src = item->ri_buf[2].i_addr;
2379 ASSERT(in_f->ilf_size <= 4);
2380 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2381 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2382 (len == in_f->ilf_dsize));
2384 switch (fields & XFS_ILOG_DFORK) {
2385 case XFS_ILOG_DDATA:
2387 memcpy(XFS_DFORK_DPTR(dip), src, len);
2390 case XFS_ILOG_DBROOT:
2391 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
2392 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
2393 XFS_DFORK_DSIZE(dip, mp));
2398 * There are no data fork flags set.
2400 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2405 * If we logged any attribute data, recover it. There may or
2406 * may not have been any other non-core data logged in this
2409 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2410 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2415 len = item->ri_buf[attr_index].i_len;
2416 src = item->ri_buf[attr_index].i_addr;
2417 ASSERT(len == in_f->ilf_asize);
2419 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2420 case XFS_ILOG_ADATA:
2422 dest = XFS_DFORK_APTR(dip);
2423 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2424 memcpy(dest, src, len);
2427 case XFS_ILOG_ABROOT:
2428 dest = XFS_DFORK_APTR(dip);
2429 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2430 len, (xfs_bmdr_block_t*)dest,
2431 XFS_DFORK_ASIZE(dip, mp));
2435 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
2444 ASSERT(bp->b_target->bt_mount == mp);
2445 bp->b_iodone = xlog_recover_iodone;
2446 xfs_buf_delwri_queue(bp);
2451 return XFS_ERROR(error);
2455 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2456 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2460 xlog_recover_quotaoff_pass1(
2462 xlog_recover_item_t *item)
2464 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
2468 * The logitem format's flag tells us if this was user quotaoff,
2469 * group/project quotaoff or both.
2471 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2472 log->l_quotaoffs_flag |= XFS_DQ_USER;
2473 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2474 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2475 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2476 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2482 * Recover a dquot record
2485 xlog_recover_dquot_pass2(
2487 xlog_recover_item_t *item)
2489 xfs_mount_t *mp = log->l_mp;
2491 struct xfs_disk_dquot *ddq, *recddq;
2493 xfs_dq_logformat_t *dq_f;
2498 * Filesystems are required to send in quota flags at mount time.
2500 if (mp->m_qflags == 0)
2503 recddq = item->ri_buf[1].i_addr;
2504 if (recddq == NULL) {
2505 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
2506 return XFS_ERROR(EIO);
2508 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
2509 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
2510 item->ri_buf[1].i_len, __func__);
2511 return XFS_ERROR(EIO);
2515 * This type of quotas was turned off, so ignore this record.
2517 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
2519 if (log->l_quotaoffs_flag & type)
2523 * At this point we know that quota was _not_ turned off.
2524 * Since the mount flags are not indicating to us otherwise, this
2525 * must mean that quota is on, and the dquot needs to be replayed.
2526 * Remember that we may not have fully recovered the superblock yet,
2527 * so we can't do the usual trick of looking at the SB quota bits.
2529 * The other possibility, of course, is that the quota subsystem was
2530 * removed since the last mount - ENOSYS.
2532 dq_f = item->ri_buf[0].i_addr;
2534 error = xfs_qm_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2535 "xlog_recover_dquot_pass2 (log copy)");
2537 return XFS_ERROR(EIO);
2538 ASSERT(dq_f->qlf_len == 1);
2540 error = xfs_read_buf(mp, mp->m_ddev_targp,
2542 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2545 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
2546 bp, dq_f->qlf_blkno);
2550 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2553 * At least the magic num portion should be on disk because this
2554 * was among a chunk of dquots created earlier, and we did some
2555 * minimal initialization then.
2557 error = xfs_qm_dqcheck(mp, ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2558 "xlog_recover_dquot_pass2");
2561 return XFS_ERROR(EIO);
2564 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2566 ASSERT(dq_f->qlf_size == 2);
2567 ASSERT(bp->b_target->bt_mount == mp);
2568 bp->b_iodone = xlog_recover_iodone;
2569 xfs_buf_delwri_queue(bp);
2576 * This routine is called to create an in-core extent free intent
2577 * item from the efi format structure which was logged on disk.
2578 * It allocates an in-core efi, copies the extents from the format
2579 * structure into it, and adds the efi to the AIL with the given
2583 xlog_recover_efi_pass2(
2585 xlog_recover_item_t *item,
2589 xfs_mount_t *mp = log->l_mp;
2590 xfs_efi_log_item_t *efip;
2591 xfs_efi_log_format_t *efi_formatp;
2593 efi_formatp = item->ri_buf[0].i_addr;
2595 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
2596 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2597 &(efip->efi_format)))) {
2598 xfs_efi_item_free(efip);
2601 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
2603 spin_lock(&log->l_ailp->xa_lock);
2605 * xfs_trans_ail_update() drops the AIL lock.
2607 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
2613 * This routine is called when an efd format structure is found in
2614 * a committed transaction in the log. It's purpose is to cancel
2615 * the corresponding efi if it was still in the log. To do this
2616 * it searches the AIL for the efi with an id equal to that in the
2617 * efd format structure. If we find it, we remove the efi from the
2621 xlog_recover_efd_pass2(
2623 xlog_recover_item_t *item)
2625 xfs_efd_log_format_t *efd_formatp;
2626 xfs_efi_log_item_t *efip = NULL;
2627 xfs_log_item_t *lip;
2629 struct xfs_ail_cursor cur;
2630 struct xfs_ail *ailp = log->l_ailp;
2632 efd_formatp = item->ri_buf[0].i_addr;
2633 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2634 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2635 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2636 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
2637 efi_id = efd_formatp->efd_efi_id;
2640 * Search for the efi with the id in the efd format structure
2643 spin_lock(&ailp->xa_lock);
2644 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
2645 while (lip != NULL) {
2646 if (lip->li_type == XFS_LI_EFI) {
2647 efip = (xfs_efi_log_item_t *)lip;
2648 if (efip->efi_format.efi_id == efi_id) {
2650 * xfs_trans_ail_delete() drops the
2653 xfs_trans_ail_delete(ailp, lip);
2654 xfs_efi_item_free(efip);
2655 spin_lock(&ailp->xa_lock);
2659 lip = xfs_trans_ail_cursor_next(ailp, &cur);
2661 xfs_trans_ail_cursor_done(ailp, &cur);
2662 spin_unlock(&ailp->xa_lock);
2668 * Free up any resources allocated by the transaction
2670 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2673 xlog_recover_free_trans(
2674 struct xlog_recover *trans)
2676 xlog_recover_item_t *item, *n;
2679 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
2680 /* Free the regions in the item. */
2681 list_del(&item->ri_list);
2682 for (i = 0; i < item->ri_cnt; i++)
2683 kmem_free(item->ri_buf[i].i_addr);
2684 /* Free the item itself */
2685 kmem_free(item->ri_buf);
2688 /* Free the transaction recover structure */
2693 xlog_recover_commit_pass1(
2695 struct xlog_recover *trans,
2696 xlog_recover_item_t *item)
2698 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
2700 switch (ITEM_TYPE(item)) {
2702 return xlog_recover_buffer_pass1(log, item);
2703 case XFS_LI_QUOTAOFF:
2704 return xlog_recover_quotaoff_pass1(log, item);
2709 /* nothing to do in pass 1 */
2712 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
2713 __func__, ITEM_TYPE(item));
2715 return XFS_ERROR(EIO);
2720 xlog_recover_commit_pass2(
2722 struct xlog_recover *trans,
2723 xlog_recover_item_t *item)
2725 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
2727 switch (ITEM_TYPE(item)) {
2729 return xlog_recover_buffer_pass2(log, item);
2731 return xlog_recover_inode_pass2(log, item);
2733 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
2735 return xlog_recover_efd_pass2(log, item);
2737 return xlog_recover_dquot_pass2(log, item);
2738 case XFS_LI_QUOTAOFF:
2739 /* nothing to do in pass2 */
2742 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
2743 __func__, ITEM_TYPE(item));
2745 return XFS_ERROR(EIO);
2750 * Perform the transaction.
2752 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2753 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2756 xlog_recover_commit_trans(
2758 struct xlog_recover *trans,
2762 xlog_recover_item_t *item;
2764 hlist_del(&trans->r_list);
2766 error = xlog_recover_reorder_trans(log, trans, pass);
2770 list_for_each_entry(item, &trans->r_itemq, ri_list) {
2771 if (pass == XLOG_RECOVER_PASS1)
2772 error = xlog_recover_commit_pass1(log, trans, item);
2774 error = xlog_recover_commit_pass2(log, trans, item);
2779 xlog_recover_free_trans(trans);
2784 xlog_recover_unmount_trans(
2786 xlog_recover_t *trans)
2788 /* Do nothing now */
2789 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
2794 * There are two valid states of the r_state field. 0 indicates that the
2795 * transaction structure is in a normal state. We have either seen the
2796 * start of the transaction or the last operation we added was not a partial
2797 * operation. If the last operation we added to the transaction was a
2798 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2800 * NOTE: skip LRs with 0 data length.
2803 xlog_recover_process_data(
2805 struct hlist_head rhash[],
2806 xlog_rec_header_t *rhead,
2812 xlog_op_header_t *ohead;
2813 xlog_recover_t *trans;
2819 lp = dp + be32_to_cpu(rhead->h_len);
2820 num_logops = be32_to_cpu(rhead->h_num_logops);
2822 /* check the log format matches our own - else we can't recover */
2823 if (xlog_header_check_recover(log->l_mp, rhead))
2824 return (XFS_ERROR(EIO));
2826 while ((dp < lp) && num_logops) {
2827 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2828 ohead = (xlog_op_header_t *)dp;
2829 dp += sizeof(xlog_op_header_t);
2830 if (ohead->oh_clientid != XFS_TRANSACTION &&
2831 ohead->oh_clientid != XFS_LOG) {
2832 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
2833 __func__, ohead->oh_clientid);
2835 return (XFS_ERROR(EIO));
2837 tid = be32_to_cpu(ohead->oh_tid);
2838 hash = XLOG_RHASH(tid);
2839 trans = xlog_recover_find_tid(&rhash[hash], tid);
2840 if (trans == NULL) { /* not found; add new tid */
2841 if (ohead->oh_flags & XLOG_START_TRANS)
2842 xlog_recover_new_tid(&rhash[hash], tid,
2843 be64_to_cpu(rhead->h_lsn));
2845 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
2846 xfs_warn(log->l_mp, "%s: bad length 0x%x",
2847 __func__, be32_to_cpu(ohead->oh_len));
2849 return (XFS_ERROR(EIO));
2851 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2852 if (flags & XLOG_WAS_CONT_TRANS)
2853 flags &= ~XLOG_CONTINUE_TRANS;
2855 case XLOG_COMMIT_TRANS:
2856 error = xlog_recover_commit_trans(log,
2859 case XLOG_UNMOUNT_TRANS:
2860 error = xlog_recover_unmount_trans(log, trans);
2862 case XLOG_WAS_CONT_TRANS:
2863 error = xlog_recover_add_to_cont_trans(log,
2865 be32_to_cpu(ohead->oh_len));
2867 case XLOG_START_TRANS:
2868 xfs_warn(log->l_mp, "%s: bad transaction",
2871 error = XFS_ERROR(EIO);
2874 case XLOG_CONTINUE_TRANS:
2875 error = xlog_recover_add_to_trans(log, trans,
2876 dp, be32_to_cpu(ohead->oh_len));
2879 xfs_warn(log->l_mp, "%s: bad flag 0x%x",
2882 error = XFS_ERROR(EIO);
2888 dp += be32_to_cpu(ohead->oh_len);
2895 * Process an extent free intent item that was recovered from
2896 * the log. We need to free the extents that it describes.
2899 xlog_recover_process_efi(
2901 xfs_efi_log_item_t *efip)
2903 xfs_efd_log_item_t *efdp;
2908 xfs_fsblock_t startblock_fsb;
2910 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
2913 * First check the validity of the extents described by the
2914 * EFI. If any are bad, then assume that all are bad and
2915 * just toss the EFI.
2917 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2918 extp = &(efip->efi_format.efi_extents[i]);
2919 startblock_fsb = XFS_BB_TO_FSB(mp,
2920 XFS_FSB_TO_DADDR(mp, extp->ext_start));
2921 if ((startblock_fsb == 0) ||
2922 (extp->ext_len == 0) ||
2923 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
2924 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
2926 * This will pull the EFI from the AIL and
2927 * free the memory associated with it.
2929 xfs_efi_release(efip, efip->efi_format.efi_nextents);
2930 return XFS_ERROR(EIO);
2934 tp = xfs_trans_alloc(mp, 0);
2935 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
2938 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
2940 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2941 extp = &(efip->efi_format.efi_extents[i]);
2942 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
2945 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
2949 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
2950 error = xfs_trans_commit(tp, 0);
2954 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
2959 * When this is called, all of the EFIs which did not have
2960 * corresponding EFDs should be in the AIL. What we do now
2961 * is free the extents associated with each one.
2963 * Since we process the EFIs in normal transactions, they
2964 * will be removed at some point after the commit. This prevents
2965 * us from just walking down the list processing each one.
2966 * We'll use a flag in the EFI to skip those that we've already
2967 * processed and use the AIL iteration mechanism's generation
2968 * count to try to speed this up at least a bit.
2970 * When we start, we know that the EFIs are the only things in
2971 * the AIL. As we process them, however, other items are added
2972 * to the AIL. Since everything added to the AIL must come after
2973 * everything already in the AIL, we stop processing as soon as
2974 * we see something other than an EFI in the AIL.
2977 xlog_recover_process_efis(
2980 xfs_log_item_t *lip;
2981 xfs_efi_log_item_t *efip;
2983 struct xfs_ail_cursor cur;
2984 struct xfs_ail *ailp;
2987 spin_lock(&ailp->xa_lock);
2988 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
2989 while (lip != NULL) {
2991 * We're done when we see something other than an EFI.
2992 * There should be no EFIs left in the AIL now.
2994 if (lip->li_type != XFS_LI_EFI) {
2996 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
2997 ASSERT(lip->li_type != XFS_LI_EFI);
3003 * Skip EFIs that we've already processed.
3005 efip = (xfs_efi_log_item_t *)lip;
3006 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
3007 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3011 spin_unlock(&ailp->xa_lock);
3012 error = xlog_recover_process_efi(log->l_mp, efip);
3013 spin_lock(&ailp->xa_lock);
3016 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3019 xfs_trans_ail_cursor_done(ailp, &cur);
3020 spin_unlock(&ailp->xa_lock);
3025 * This routine performs a transaction to null out a bad inode pointer
3026 * in an agi unlinked inode hash bucket.
3029 xlog_recover_clear_agi_bucket(
3031 xfs_agnumber_t agno,
3040 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3041 error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp),
3046 error = xfs_read_agi(mp, tp, agno, &agibp);
3050 agi = XFS_BUF_TO_AGI(agibp);
3051 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
3052 offset = offsetof(xfs_agi_t, agi_unlinked) +
3053 (sizeof(xfs_agino_t) * bucket);
3054 xfs_trans_log_buf(tp, agibp, offset,
3055 (offset + sizeof(xfs_agino_t) - 1));
3057 error = xfs_trans_commit(tp, 0);
3063 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3065 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
3070 xlog_recover_process_one_iunlink(
3071 struct xfs_mount *mp,
3072 xfs_agnumber_t agno,
3076 struct xfs_buf *ibp;
3077 struct xfs_dinode *dip;
3078 struct xfs_inode *ip;
3082 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3083 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
3088 * Get the on disk inode to find the next inode in the bucket.
3090 error = xfs_itobp(mp, NULL, ip, &dip, &ibp, XBF_LOCK);
3094 ASSERT(ip->i_d.di_nlink == 0);
3095 ASSERT(ip->i_d.di_mode != 0);
3097 /* setup for the next pass */
3098 agino = be32_to_cpu(dip->di_next_unlinked);
3102 * Prevent any DMAPI event from being sent when the reference on
3103 * the inode is dropped.
3105 ip->i_d.di_dmevmask = 0;
3114 * We can't read in the inode this bucket points to, or this inode
3115 * is messed up. Just ditch this bucket of inodes. We will lose
3116 * some inodes and space, but at least we won't hang.
3118 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3119 * clear the inode pointer in the bucket.
3121 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3126 * xlog_iunlink_recover
3128 * This is called during recovery to process any inodes which
3129 * we unlinked but not freed when the system crashed. These
3130 * inodes will be on the lists in the AGI blocks. What we do
3131 * here is scan all the AGIs and fully truncate and free any
3132 * inodes found on the lists. Each inode is removed from the
3133 * lists when it has been fully truncated and is freed. The
3134 * freeing of the inode and its removal from the list must be
3138 xlog_recover_process_iunlinks(
3142 xfs_agnumber_t agno;
3153 * Prevent any DMAPI event from being sent while in this function.
3155 mp_dmevmask = mp->m_dmevmask;
3158 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3160 * Find the agi for this ag.
3162 error = xfs_read_agi(mp, NULL, agno, &agibp);
3165 * AGI is b0rked. Don't process it.
3167 * We should probably mark the filesystem as corrupt
3168 * after we've recovered all the ag's we can....
3172 agi = XFS_BUF_TO_AGI(agibp);
3174 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3175 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3176 while (agino != NULLAGINO) {
3178 * Release the agi buffer so that it can
3179 * be acquired in the normal course of the
3180 * transaction to truncate and free the inode.
3182 xfs_buf_relse(agibp);
3184 agino = xlog_recover_process_one_iunlink(mp,
3185 agno, agino, bucket);
3188 * Reacquire the agibuffer and continue around
3189 * the loop. This should never fail as we know
3190 * the buffer was good earlier on.
3192 error = xfs_read_agi(mp, NULL, agno, &agibp);
3194 agi = XFS_BUF_TO_AGI(agibp);
3199 * Release the buffer for the current agi so we can
3200 * go on to the next one.
3202 xfs_buf_relse(agibp);
3205 mp->m_dmevmask = mp_dmevmask;
3211 xlog_pack_data_checksum(
3213 xlog_in_core_t *iclog,
3220 up = (__be32 *)iclog->ic_datap;
3221 /* divide length by 4 to get # words */
3222 for (i = 0; i < (size >> 2); i++) {
3223 chksum ^= be32_to_cpu(*up);
3226 iclog->ic_header.h_chksum = cpu_to_be32(chksum);
3229 #define xlog_pack_data_checksum(log, iclog, size)
3233 * Stamp cycle number in every block
3238 xlog_in_core_t *iclog,
3242 int size = iclog->ic_offset + roundoff;
3246 xlog_pack_data_checksum(log, iclog, size);
3248 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3250 dp = iclog->ic_datap;
3251 for (i = 0; i < BTOBB(size) &&
3252 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3253 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
3254 *(__be32 *)dp = cycle_lsn;
3258 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3259 xlog_in_core_2_t *xhdr = iclog->ic_data;
3261 for ( ; i < BTOBB(size); i++) {
3262 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3263 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3264 xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
3265 *(__be32 *)dp = cycle_lsn;
3269 for (i = 1; i < log->l_iclog_heads; i++) {
3270 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3277 xlog_rec_header_t *rhead,
3283 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
3284 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3285 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
3289 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3290 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
3291 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
3292 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3293 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3294 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
3301 xlog_valid_rec_header(
3303 xlog_rec_header_t *rhead,
3308 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
3309 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3310 XFS_ERRLEVEL_LOW, log->l_mp);
3311 return XFS_ERROR(EFSCORRUPTED);
3314 (!rhead->h_version ||
3315 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
3316 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
3317 __func__, be32_to_cpu(rhead->h_version));
3318 return XFS_ERROR(EIO);
3321 /* LR body must have data or it wouldn't have been written */
3322 hlen = be32_to_cpu(rhead->h_len);
3323 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3324 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3325 XFS_ERRLEVEL_LOW, log->l_mp);
3326 return XFS_ERROR(EFSCORRUPTED);
3328 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3329 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3330 XFS_ERRLEVEL_LOW, log->l_mp);
3331 return XFS_ERROR(EFSCORRUPTED);
3337 * Read the log from tail to head and process the log records found.
3338 * Handle the two cases where the tail and head are in the same cycle
3339 * and where the active portion of the log wraps around the end of
3340 * the physical log separately. The pass parameter is passed through
3341 * to the routines called to process the data and is not looked at
3345 xlog_do_recovery_pass(
3347 xfs_daddr_t head_blk,
3348 xfs_daddr_t tail_blk,
3351 xlog_rec_header_t *rhead;
3354 xfs_buf_t *hbp, *dbp;
3355 int error = 0, h_size;
3356 int bblks, split_bblks;
3357 int hblks, split_hblks, wrapped_hblks;
3358 struct hlist_head rhash[XLOG_RHASH_SIZE];
3360 ASSERT(head_blk != tail_blk);
3363 * Read the header of the tail block and get the iclog buffer size from
3364 * h_size. Use this to tell how many sectors make up the log header.
3366 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3368 * When using variable length iclogs, read first sector of
3369 * iclog header and extract the header size from it. Get a
3370 * new hbp that is the correct size.
3372 hbp = xlog_get_bp(log, 1);
3376 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
3380 rhead = (xlog_rec_header_t *)offset;
3381 error = xlog_valid_rec_header(log, rhead, tail_blk);
3384 h_size = be32_to_cpu(rhead->h_size);
3385 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
3386 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3387 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3388 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3391 hbp = xlog_get_bp(log, hblks);
3396 ASSERT(log->l_sectBBsize == 1);
3398 hbp = xlog_get_bp(log, 1);
3399 h_size = XLOG_BIG_RECORD_BSIZE;
3404 dbp = xlog_get_bp(log, BTOBB(h_size));
3410 memset(rhash, 0, sizeof(rhash));
3411 if (tail_blk <= head_blk) {
3412 for (blk_no = tail_blk; blk_no < head_blk; ) {
3413 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3417 rhead = (xlog_rec_header_t *)offset;
3418 error = xlog_valid_rec_header(log, rhead, blk_no);
3422 /* blocks in data section */
3423 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3424 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
3429 xlog_unpack_data(rhead, offset, log);
3430 if ((error = xlog_recover_process_data(log,
3431 rhash, rhead, offset, pass)))
3433 blk_no += bblks + hblks;
3437 * Perform recovery around the end of the physical log.
3438 * When the head is not on the same cycle number as the tail,
3439 * we can't do a sequential recovery as above.
3442 while (blk_no < log->l_logBBsize) {
3444 * Check for header wrapping around physical end-of-log
3446 offset = hbp->b_addr;
3449 if (blk_no + hblks <= log->l_logBBsize) {
3450 /* Read header in one read */
3451 error = xlog_bread(log, blk_no, hblks, hbp,
3456 /* This LR is split across physical log end */
3457 if (blk_no != log->l_logBBsize) {
3458 /* some data before physical log end */
3459 ASSERT(blk_no <= INT_MAX);
3460 split_hblks = log->l_logBBsize - (int)blk_no;
3461 ASSERT(split_hblks > 0);
3462 error = xlog_bread(log, blk_no,
3470 * Note: this black magic still works with
3471 * large sector sizes (non-512) only because:
3472 * - we increased the buffer size originally
3473 * by 1 sector giving us enough extra space
3474 * for the second read;
3475 * - the log start is guaranteed to be sector
3477 * - we read the log end (LR header start)
3478 * _first_, then the log start (LR header end)
3479 * - order is important.
3481 wrapped_hblks = hblks - split_hblks;
3482 error = xlog_bread_offset(log, 0,
3484 offset + BBTOB(split_hblks));
3488 rhead = (xlog_rec_header_t *)offset;
3489 error = xlog_valid_rec_header(log, rhead,
3490 split_hblks ? blk_no : 0);
3494 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3497 /* Read in data for log record */
3498 if (blk_no + bblks <= log->l_logBBsize) {
3499 error = xlog_bread(log, blk_no, bblks, dbp,
3504 /* This log record is split across the
3505 * physical end of log */
3506 offset = dbp->b_addr;
3508 if (blk_no != log->l_logBBsize) {
3509 /* some data is before the physical
3511 ASSERT(!wrapped_hblks);
3512 ASSERT(blk_no <= INT_MAX);
3514 log->l_logBBsize - (int)blk_no;
3515 ASSERT(split_bblks > 0);
3516 error = xlog_bread(log, blk_no,
3524 * Note: this black magic still works with
3525 * large sector sizes (non-512) only because:
3526 * - we increased the buffer size originally
3527 * by 1 sector giving us enough extra space
3528 * for the second read;
3529 * - the log start is guaranteed to be sector
3531 * - we read the log end (LR header start)
3532 * _first_, then the log start (LR header end)
3533 * - order is important.
3535 error = xlog_bread_offset(log, 0,
3536 bblks - split_bblks, hbp,
3537 offset + BBTOB(split_bblks));
3541 xlog_unpack_data(rhead, offset, log);
3542 if ((error = xlog_recover_process_data(log, rhash,
3543 rhead, offset, pass)))
3548 ASSERT(blk_no >= log->l_logBBsize);
3549 blk_no -= log->l_logBBsize;
3551 /* read first part of physical log */
3552 while (blk_no < head_blk) {
3553 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3557 rhead = (xlog_rec_header_t *)offset;
3558 error = xlog_valid_rec_header(log, rhead, blk_no);
3562 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3563 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
3568 xlog_unpack_data(rhead, offset, log);
3569 if ((error = xlog_recover_process_data(log, rhash,
3570 rhead, offset, pass)))
3572 blk_no += bblks + hblks;
3584 * Do the recovery of the log. We actually do this in two phases.
3585 * The two passes are necessary in order to implement the function
3586 * of cancelling a record written into the log. The first pass
3587 * determines those things which have been cancelled, and the
3588 * second pass replays log items normally except for those which
3589 * have been cancelled. The handling of the replay and cancellations
3590 * takes place in the log item type specific routines.
3592 * The table of items which have cancel records in the log is allocated
3593 * and freed at this level, since only here do we know when all of
3594 * the log recovery has been completed.
3597 xlog_do_log_recovery(
3599 xfs_daddr_t head_blk,
3600 xfs_daddr_t tail_blk)
3604 ASSERT(head_blk != tail_blk);
3607 * First do a pass to find all of the cancelled buf log items.
3608 * Store them in the buf_cancel_table for use in the second pass.
3610 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
3611 sizeof(struct list_head),
3613 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3614 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
3616 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3617 XLOG_RECOVER_PASS1);
3619 kmem_free(log->l_buf_cancel_table);
3620 log->l_buf_cancel_table = NULL;
3624 * Then do a second pass to actually recover the items in the log.
3625 * When it is complete free the table of buf cancel items.
3627 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3628 XLOG_RECOVER_PASS2);
3633 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3634 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
3638 kmem_free(log->l_buf_cancel_table);
3639 log->l_buf_cancel_table = NULL;
3645 * Do the actual recovery
3650 xfs_daddr_t head_blk,
3651 xfs_daddr_t tail_blk)
3658 * First replay the images in the log.
3660 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3665 XFS_bflush(log->l_mp->m_ddev_targp);
3668 * If IO errors happened during recovery, bail out.
3670 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3675 * We now update the tail_lsn since much of the recovery has completed
3676 * and there may be space available to use. If there were no extent
3677 * or iunlinks, we can free up the entire log and set the tail_lsn to
3678 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3679 * lsn of the last known good LR on disk. If there are extent frees
3680 * or iunlinks they will have some entries in the AIL; so we look at
3681 * the AIL to determine how to set the tail_lsn.
3683 xlog_assign_tail_lsn(log->l_mp);
3686 * Now that we've finished replaying all buffer and inode
3687 * updates, re-read in the superblock.
3689 bp = xfs_getsb(log->l_mp, 0);
3691 ASSERT(!(XFS_BUF_ISWRITE(bp)));
3692 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
3694 XFS_BUF_UNASYNC(bp);
3695 xfsbdstrat(log->l_mp, bp);
3696 error = xfs_buf_iowait(bp);
3698 xfs_ioerror_alert("xlog_do_recover",
3699 log->l_mp, bp, XFS_BUF_ADDR(bp));
3705 /* Convert superblock from on-disk format */
3706 sbp = &log->l_mp->m_sb;
3707 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
3708 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
3709 ASSERT(xfs_sb_good_version(sbp));
3712 /* We've re-read the superblock so re-initialize per-cpu counters */
3713 xfs_icsb_reinit_counters(log->l_mp);
3715 xlog_recover_check_summary(log);
3717 /* Normal transactions can now occur */
3718 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3723 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3725 * Return error or zero.
3731 xfs_daddr_t head_blk, tail_blk;
3734 /* find the tail of the log */
3735 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
3738 if (tail_blk != head_blk) {
3739 /* There used to be a comment here:
3741 * disallow recovery on read-only mounts. note -- mount
3742 * checks for ENOSPC and turns it into an intelligent
3744 * ...but this is no longer true. Now, unless you specify
3745 * NORECOVERY (in which case this function would never be
3746 * called), we just go ahead and recover. We do this all
3747 * under the vfs layer, so we can get away with it unless
3748 * the device itself is read-only, in which case we fail.
3750 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
3754 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
3755 log->l_mp->m_logname ? log->l_mp->m_logname
3758 error = xlog_do_recover(log, head_blk, tail_blk);
3759 log->l_flags |= XLOG_RECOVERY_NEEDED;
3765 * In the first part of recovery we replay inodes and buffers and build
3766 * up the list of extent free items which need to be processed. Here
3767 * we process the extent free items and clean up the on disk unlinked
3768 * inode lists. This is separated from the first part of recovery so
3769 * that the root and real-time bitmap inodes can be read in from disk in
3770 * between the two stages. This is necessary so that we can free space
3771 * in the real-time portion of the file system.
3774 xlog_recover_finish(
3778 * Now we're ready to do the transactions needed for the
3779 * rest of recovery. Start with completing all the extent
3780 * free intent records and then process the unlinked inode
3781 * lists. At this point, we essentially run in normal mode
3782 * except that we're still performing recovery actions
3783 * rather than accepting new requests.
3785 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
3787 error = xlog_recover_process_efis(log);
3789 xfs_alert(log->l_mp, "Failed to recover EFIs");
3793 * Sync the log to get all the EFIs out of the AIL.
3794 * This isn't absolutely necessary, but it helps in
3795 * case the unlink transactions would have problems
3796 * pushing the EFIs out of the way.
3798 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
3800 xlog_recover_process_iunlinks(log);
3802 xlog_recover_check_summary(log);
3804 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
3805 log->l_mp->m_logname ? log->l_mp->m_logname
3807 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
3809 xfs_info(log->l_mp, "Ending clean mount");
3817 * Read all of the agf and agi counters and check that they
3818 * are consistent with the superblock counters.
3821 xlog_recover_check_summary(
3828 xfs_agnumber_t agno;
3829 __uint64_t freeblks;
3839 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3840 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
3842 xfs_alert(mp, "%s agf read failed agno %d error %d",
3843 __func__, agno, error);
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 xfs_alert(mp, "%s agi read failed agno %d error %d",
3854 __func__, agno, error);
3856 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
3858 itotal += be32_to_cpu(agi->agi_count);
3859 ifree += be32_to_cpu(agi->agi_freecount);
3860 xfs_buf_relse(agibp);
projects / karo-tx-linux.git / blob