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_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_format.h"
30 #include "xfs_inode.h"
31 #include "xfs_trans.h"
33 #include "xfs_log_priv.h"
34 #include "xfs_log_recover.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_extfree_item.h"
37 #include "xfs_trans_priv.h"
38 #include "xfs_alloc.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_quota.h"
41 #include "xfs_cksum.h"
42 #include "xfs_trace.h"
43 #include "xfs_icache.h"
44 #include "xfs_bmap_btree.h"
45 #include "xfs_dinode.h"
46 #include "xfs_error.h"
49 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
56 xlog_clear_stale_blocks(
61 xlog_recover_check_summary(
64 #define xlog_recover_check_summary(log)
68 * This structure is used during recovery to record the buf log items which
69 * have been canceled and should not be replayed.
71 struct xfs_buf_cancel {
75 struct list_head bc_list;
79 * Sector aligned buffer routines for buffer create/read/write/access
83 * Verify the given count of basic blocks is valid number of blocks
84 * to specify for an operation involving the given XFS log buffer.
85 * Returns nonzero if the count is valid, 0 otherwise.
89 xlog_buf_bbcount_valid(
93 return bbcount > 0 && bbcount <= log->l_logBBsize;
97 * Allocate a buffer to hold log data. The buffer needs to be able
98 * to map to a range of nbblks basic blocks at any valid (basic
99 * block) offset within the log.
108 if (!xlog_buf_bbcount_valid(log, nbblks)) {
109 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
111 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
116 * We do log I/O in units of log sectors (a power-of-2
117 * multiple of the basic block size), so we round up the
118 * requested size to accommodate the basic blocks required
119 * for complete log sectors.
121 * In addition, the buffer may be used for a non-sector-
122 * aligned block offset, in which case an I/O of the
123 * requested size could extend beyond the end of the
124 * buffer. If the requested size is only 1 basic block it
125 * will never straddle a sector boundary, so this won't be
126 * an issue. Nor will this be a problem if the log I/O is
127 * done in basic blocks (sector size 1). But otherwise we
128 * extend the buffer by one extra log sector to ensure
129 * there's space to accommodate this possibility.
131 if (nbblks > 1 && log->l_sectBBsize > 1)
132 nbblks += log->l_sectBBsize;
133 nbblks = round_up(nbblks, log->l_sectBBsize);
135 bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0);
149 * Return the address of the start of the given block number's data
150 * in a log buffer. The buffer covers a log sector-aligned region.
159 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
161 ASSERT(offset + nbblks <= bp->b_length);
162 return bp->b_addr + BBTOB(offset);
167 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
178 if (!xlog_buf_bbcount_valid(log, nbblks)) {
179 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
181 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
182 return -EFSCORRUPTED;
185 blk_no = round_down(blk_no, log->l_sectBBsize);
186 nbblks = round_up(nbblks, log->l_sectBBsize);
189 ASSERT(nbblks <= bp->b_length);
191 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
193 bp->b_io_length = nbblks;
196 error = xfs_buf_submit_wait(bp);
197 if (error && !XFS_FORCED_SHUTDOWN(log->l_mp))
198 xfs_buf_ioerror_alert(bp, __func__);
212 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
216 *offset = xlog_align(log, blk_no, nbblks, bp);
221 * Read at an offset into the buffer. Returns with the buffer in it's original
222 * state regardless of the result of the read.
227 xfs_daddr_t blk_no, /* block to read from */
228 int nbblks, /* blocks to read */
232 xfs_caddr_t orig_offset = bp->b_addr;
233 int orig_len = BBTOB(bp->b_length);
236 error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks));
240 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
242 /* must reset buffer pointer even on error */
243 error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len);
250 * Write out the buffer at the given block for the given number of blocks.
251 * The buffer is kept locked across the write and is returned locked.
252 * This can only be used for synchronous log writes.
263 if (!xlog_buf_bbcount_valid(log, nbblks)) {
264 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
266 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
267 return -EFSCORRUPTED;
270 blk_no = round_down(blk_no, log->l_sectBBsize);
271 nbblks = round_up(nbblks, log->l_sectBBsize);
274 ASSERT(nbblks <= bp->b_length);
276 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
277 XFS_BUF_ZEROFLAGS(bp);
280 bp->b_io_length = nbblks;
283 error = xfs_bwrite(bp);
285 xfs_buf_ioerror_alert(bp, __func__);
292 * dump debug superblock and log record information
295 xlog_header_check_dump(
297 xlog_rec_header_t *head)
299 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d",
300 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
301 xfs_debug(mp, " log : uuid = %pU, fmt = %d",
302 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
305 #define xlog_header_check_dump(mp, head)
309 * check log record header for recovery
312 xlog_header_check_recover(
314 xlog_rec_header_t *head)
316 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
319 * IRIX doesn't write the h_fmt field and leaves it zeroed
320 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
321 * a dirty log created in IRIX.
323 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
325 "dirty log written in incompatible format - can't recover");
326 xlog_header_check_dump(mp, head);
327 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
328 XFS_ERRLEVEL_HIGH, mp);
329 return -EFSCORRUPTED;
330 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
332 "dirty log entry has mismatched uuid - can't recover");
333 xlog_header_check_dump(mp, head);
334 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
335 XFS_ERRLEVEL_HIGH, mp);
336 return -EFSCORRUPTED;
342 * read the head block of the log and check the header
345 xlog_header_check_mount(
347 xlog_rec_header_t *head)
349 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
351 if (uuid_is_nil(&head->h_fs_uuid)) {
353 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
354 * h_fs_uuid is nil, we assume this log was last mounted
355 * by IRIX and continue.
357 xfs_warn(mp, "nil uuid in log - IRIX style log");
358 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
359 xfs_warn(mp, "log has mismatched uuid - can't recover");
360 xlog_header_check_dump(mp, head);
361 XFS_ERROR_REPORT("xlog_header_check_mount",
362 XFS_ERRLEVEL_HIGH, mp);
363 return -EFSCORRUPTED;
374 * We're not going to bother about retrying
375 * this during recovery. One strike!
377 if (!XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
378 xfs_buf_ioerror_alert(bp, __func__);
379 xfs_force_shutdown(bp->b_target->bt_mount,
380 SHUTDOWN_META_IO_ERROR);
388 * This routine finds (to an approximation) the first block in the physical
389 * log which contains the given cycle. It uses a binary search algorithm.
390 * Note that the algorithm can not be perfect because the disk will not
391 * necessarily be perfect.
394 xlog_find_cycle_start(
397 xfs_daddr_t first_blk,
398 xfs_daddr_t *last_blk,
408 mid_blk = BLK_AVG(first_blk, end_blk);
409 while (mid_blk != first_blk && mid_blk != end_blk) {
410 error = xlog_bread(log, mid_blk, 1, bp, &offset);
413 mid_cycle = xlog_get_cycle(offset);
414 if (mid_cycle == cycle)
415 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
417 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
418 mid_blk = BLK_AVG(first_blk, end_blk);
420 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
421 (mid_blk == end_blk && mid_blk-1 == first_blk));
429 * Check that a range of blocks does not contain stop_on_cycle_no.
430 * Fill in *new_blk with the block offset where such a block is
431 * found, or with -1 (an invalid block number) if there is no such
432 * block in the range. The scan needs to occur from front to back
433 * and the pointer into the region must be updated since a later
434 * routine will need to perform another test.
437 xlog_find_verify_cycle(
439 xfs_daddr_t start_blk,
441 uint stop_on_cycle_no,
442 xfs_daddr_t *new_blk)
448 xfs_caddr_t buf = NULL;
452 * Greedily allocate a buffer big enough to handle the full
453 * range of basic blocks we'll be examining. If that fails,
454 * try a smaller size. We need to be able to read at least
455 * a log sector, or we're out of luck.
457 bufblks = 1 << ffs(nbblks);
458 while (bufblks > log->l_logBBsize)
460 while (!(bp = xlog_get_bp(log, bufblks))) {
462 if (bufblks < log->l_sectBBsize)
466 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
469 bcount = min(bufblks, (start_blk + nbblks - i));
471 error = xlog_bread(log, i, bcount, bp, &buf);
475 for (j = 0; j < bcount; j++) {
476 cycle = xlog_get_cycle(buf);
477 if (cycle == stop_on_cycle_no) {
494 * Potentially backup over partial log record write.
496 * In the typical case, last_blk is the number of the block directly after
497 * a good log record. Therefore, we subtract one to get the block number
498 * of the last block in the given buffer. extra_bblks contains the number
499 * of blocks we would have read on a previous read. This happens when the
500 * last log record is split over the end of the physical log.
502 * extra_bblks is the number of blocks potentially verified on a previous
503 * call to this routine.
506 xlog_find_verify_log_record(
508 xfs_daddr_t start_blk,
509 xfs_daddr_t *last_blk,
514 xfs_caddr_t offset = NULL;
515 xlog_rec_header_t *head = NULL;
518 int num_blks = *last_blk - start_blk;
521 ASSERT(start_blk != 0 || *last_blk != start_blk);
523 if (!(bp = xlog_get_bp(log, num_blks))) {
524 if (!(bp = xlog_get_bp(log, 1)))
528 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
531 offset += ((num_blks - 1) << BBSHIFT);
534 for (i = (*last_blk) - 1; i >= 0; i--) {
536 /* valid log record not found */
538 "Log inconsistent (didn't find previous header)");
545 error = xlog_bread(log, i, 1, bp, &offset);
550 head = (xlog_rec_header_t *)offset;
552 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
560 * We hit the beginning of the physical log & still no header. Return
561 * to caller. If caller can handle a return of -1, then this routine
562 * will be called again for the end of the physical log.
570 * We have the final block of the good log (the first block
571 * of the log record _before_ the head. So we check the uuid.
573 if ((error = xlog_header_check_mount(log->l_mp, head)))
577 * We may have found a log record header before we expected one.
578 * last_blk will be the 1st block # with a given cycle #. We may end
579 * up reading an entire log record. In this case, we don't want to
580 * reset last_blk. Only when last_blk points in the middle of a log
581 * record do we update last_blk.
583 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
584 uint h_size = be32_to_cpu(head->h_size);
586 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
587 if (h_size % XLOG_HEADER_CYCLE_SIZE)
593 if (*last_blk - i + extra_bblks !=
594 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
603 * Head is defined to be the point of the log where the next log write
604 * could go. This means that incomplete LR writes at the end are
605 * eliminated when calculating the head. We aren't guaranteed that previous
606 * LR have complete transactions. We only know that a cycle number of
607 * current cycle number -1 won't be present in the log if we start writing
608 * from our current block number.
610 * last_blk contains the block number of the first block with a given
613 * Return: zero if normal, non-zero if error.
618 xfs_daddr_t *return_head_blk)
622 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
624 uint first_half_cycle, last_half_cycle;
626 int error, log_bbnum = log->l_logBBsize;
628 /* Is the end of the log device zeroed? */
629 error = xlog_find_zeroed(log, &first_blk);
631 xfs_warn(log->l_mp, "empty log check failed");
635 *return_head_blk = first_blk;
637 /* Is the whole lot zeroed? */
639 /* Linux XFS shouldn't generate totally zeroed logs -
640 * mkfs etc write a dummy unmount record to a fresh
641 * log so we can store the uuid in there
643 xfs_warn(log->l_mp, "totally zeroed log");
649 first_blk = 0; /* get cycle # of 1st block */
650 bp = xlog_get_bp(log, 1);
654 error = xlog_bread(log, 0, 1, bp, &offset);
658 first_half_cycle = xlog_get_cycle(offset);
660 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
661 error = xlog_bread(log, last_blk, 1, bp, &offset);
665 last_half_cycle = xlog_get_cycle(offset);
666 ASSERT(last_half_cycle != 0);
669 * If the 1st half cycle number is equal to the last half cycle number,
670 * then the entire log is stamped with the same cycle number. In this
671 * case, head_blk can't be set to zero (which makes sense). The below
672 * math doesn't work out properly with head_blk equal to zero. Instead,
673 * we set it to log_bbnum which is an invalid block number, but this
674 * value makes the math correct. If head_blk doesn't changed through
675 * all the tests below, *head_blk is set to zero at the very end rather
676 * than log_bbnum. In a sense, log_bbnum and zero are the same block
677 * in a circular file.
679 if (first_half_cycle == last_half_cycle) {
681 * In this case we believe that the entire log should have
682 * cycle number last_half_cycle. We need to scan backwards
683 * from the end verifying that there are no holes still
684 * containing last_half_cycle - 1. If we find such a hole,
685 * then the start of that hole will be the new head. The
686 * simple case looks like
687 * x | x ... | x - 1 | x
688 * Another case that fits this picture would be
689 * x | x + 1 | x ... | x
690 * In this case the head really is somewhere at the end of the
691 * log, as one of the latest writes at the beginning was
694 * x | x + 1 | x ... | x - 1 | x
695 * This is really the combination of the above two cases, and
696 * the head has to end up at the start of the x-1 hole at the
699 * In the 256k log case, we will read from the beginning to the
700 * end of the log and search for cycle numbers equal to x-1.
701 * We don't worry about the x+1 blocks that we encounter,
702 * because we know that they cannot be the head since the log
705 head_blk = log_bbnum;
706 stop_on_cycle = last_half_cycle - 1;
709 * In this case we want to find the first block with cycle
710 * number matching last_half_cycle. We expect the log to be
712 * x + 1 ... | x ... | x
713 * The first block with cycle number x (last_half_cycle) will
714 * be where the new head belongs. First we do a binary search
715 * for the first occurrence of last_half_cycle. The binary
716 * search may not be totally accurate, so then we scan back
717 * from there looking for occurrences of last_half_cycle before
718 * us. If that backwards scan wraps around the beginning of
719 * the log, then we look for occurrences of last_half_cycle - 1
720 * at the end of the log. The cases we're looking for look
722 * v binary search stopped here
723 * x + 1 ... | x | x + 1 | x ... | x
724 * ^ but we want to locate this spot
726 * <---------> less than scan distance
727 * x + 1 ... | x ... | x - 1 | x
728 * ^ we want to locate this spot
730 stop_on_cycle = last_half_cycle;
731 if ((error = xlog_find_cycle_start(log, bp, first_blk,
732 &head_blk, last_half_cycle)))
737 * Now validate the answer. Scan back some number of maximum possible
738 * blocks and make sure each one has the expected cycle number. The
739 * maximum is determined by the total possible amount of buffering
740 * in the in-core log. The following number can be made tighter if
741 * we actually look at the block size of the filesystem.
743 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
744 if (head_blk >= num_scan_bblks) {
746 * We are guaranteed that the entire check can be performed
749 start_blk = head_blk - num_scan_bblks;
750 if ((error = xlog_find_verify_cycle(log,
751 start_blk, num_scan_bblks,
752 stop_on_cycle, &new_blk)))
756 } else { /* need to read 2 parts of log */
758 * We are going to scan backwards in the log in two parts.
759 * First we scan the physical end of the log. In this part
760 * of the log, we are looking for blocks with cycle number
761 * last_half_cycle - 1.
762 * If we find one, then we know that the log starts there, as
763 * we've found a hole that didn't get written in going around
764 * the end of the physical log. The simple case for this is
765 * x + 1 ... | x ... | x - 1 | x
766 * <---------> less than scan distance
767 * If all of the blocks at the end of the log have cycle number
768 * last_half_cycle, then we check the blocks at the start of
769 * the log looking for occurrences of last_half_cycle. If we
770 * find one, then our current estimate for the location of the
771 * first occurrence of last_half_cycle is wrong and we move
772 * back to the hole we've found. This case looks like
773 * x + 1 ... | x | x + 1 | x ...
774 * ^ binary search stopped here
775 * Another case we need to handle that only occurs in 256k
777 * x + 1 ... | x ... | x+1 | x ...
778 * ^ binary search stops here
779 * In a 256k log, the scan at the end of the log will see the
780 * x + 1 blocks. We need to skip past those since that is
781 * certainly not the head of the log. By searching for
782 * last_half_cycle-1 we accomplish that.
784 ASSERT(head_blk <= INT_MAX &&
785 (xfs_daddr_t) num_scan_bblks >= head_blk);
786 start_blk = log_bbnum - (num_scan_bblks - head_blk);
787 if ((error = xlog_find_verify_cycle(log, start_blk,
788 num_scan_bblks - (int)head_blk,
789 (stop_on_cycle - 1), &new_blk)))
797 * Scan beginning of log now. The last part of the physical
798 * log is good. This scan needs to verify that it doesn't find
799 * the last_half_cycle.
802 ASSERT(head_blk <= INT_MAX);
803 if ((error = xlog_find_verify_cycle(log,
804 start_blk, (int)head_blk,
805 stop_on_cycle, &new_blk)))
813 * Now we need to make sure head_blk is not pointing to a block in
814 * the middle of a log record.
816 num_scan_bblks = XLOG_REC_SHIFT(log);
817 if (head_blk >= num_scan_bblks) {
818 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
820 /* start ptr at last block ptr before head_blk */
821 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
828 ASSERT(head_blk <= INT_MAX);
829 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
833 /* We hit the beginning of the log during our search */
834 start_blk = log_bbnum - (num_scan_bblks - head_blk);
836 ASSERT(start_blk <= INT_MAX &&
837 (xfs_daddr_t) log_bbnum-start_blk >= 0);
838 ASSERT(head_blk <= INT_MAX);
839 error = xlog_find_verify_log_record(log, start_blk,
840 &new_blk, (int)head_blk);
845 if (new_blk != log_bbnum)
852 if (head_blk == log_bbnum)
853 *return_head_blk = 0;
855 *return_head_blk = head_blk;
857 * When returning here, we have a good block number. Bad block
858 * means that during a previous crash, we didn't have a clean break
859 * from cycle number N to cycle number N-1. In this case, we need
860 * to find the first block with cycle number N-1.
868 xfs_warn(log->l_mp, "failed to find log head");
873 * Find the sync block number or the tail of the log.
875 * This will be the block number of the last record to have its
876 * associated buffers synced to disk. Every log record header has
877 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
878 * to get a sync block number. The only concern is to figure out which
879 * log record header to believe.
881 * The following algorithm uses the log record header with the largest
882 * lsn. The entire log record does not need to be valid. We only care
883 * that the header is valid.
885 * We could speed up search by using current head_blk buffer, but it is not
891 xfs_daddr_t *head_blk,
892 xfs_daddr_t *tail_blk)
894 xlog_rec_header_t *rhead;
895 xlog_op_header_t *op_head;
896 xfs_caddr_t offset = NULL;
899 xfs_daddr_t umount_data_blk;
900 xfs_daddr_t after_umount_blk;
907 * Find previous log record
909 if ((error = xlog_find_head(log, head_blk)))
912 bp = xlog_get_bp(log, 1);
915 if (*head_blk == 0) { /* special case */
916 error = xlog_bread(log, 0, 1, bp, &offset);
920 if (xlog_get_cycle(offset) == 0) {
922 /* leave all other log inited values alone */
928 * Search backwards looking for log record header block
930 ASSERT(*head_blk < INT_MAX);
931 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
932 error = xlog_bread(log, i, 1, bp, &offset);
936 if (*(__be32 *)offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
942 * If we haven't found the log record header block, start looking
943 * again from the end of the physical log. XXXmiken: There should be
944 * a check here to make sure we didn't search more than N blocks in
948 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
949 error = xlog_bread(log, i, 1, bp, &offset);
953 if (*(__be32 *)offset ==
954 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
961 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
967 /* find blk_no of tail of log */
968 rhead = (xlog_rec_header_t *)offset;
969 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
972 * Reset log values according to the state of the log when we
973 * crashed. In the case where head_blk == 0, we bump curr_cycle
974 * one because the next write starts a new cycle rather than
975 * continuing the cycle of the last good log record. At this
976 * point we have guaranteed that all partial log records have been
977 * accounted for. Therefore, we know that the last good log record
978 * written was complete and ended exactly on the end boundary
979 * of the physical log.
981 log->l_prev_block = i;
982 log->l_curr_block = (int)*head_blk;
983 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
986 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
987 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
988 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
989 BBTOB(log->l_curr_block));
990 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
991 BBTOB(log->l_curr_block));
994 * Look for unmount record. If we find it, then we know there
995 * was a clean unmount. Since 'i' could be the last block in
996 * the physical log, we convert to a log block before comparing
999 * Save the current tail lsn to use to pass to
1000 * xlog_clear_stale_blocks() below. We won't want to clear the
1001 * unmount record if there is one, so we pass the lsn of the
1002 * unmount record rather than the block after it.
1004 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1005 int h_size = be32_to_cpu(rhead->h_size);
1006 int h_version = be32_to_cpu(rhead->h_version);
1008 if ((h_version & XLOG_VERSION_2) &&
1009 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
1010 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
1011 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1019 after_umount_blk = (i + hblks + (int)
1020 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
1021 tail_lsn = atomic64_read(&log->l_tail_lsn);
1022 if (*head_blk == after_umount_blk &&
1023 be32_to_cpu(rhead->h_num_logops) == 1) {
1024 umount_data_blk = (i + hblks) % log->l_logBBsize;
1025 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1029 op_head = (xlog_op_header_t *)offset;
1030 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1032 * Set tail and last sync so that newly written
1033 * log records will point recovery to after the
1034 * current unmount record.
1036 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1037 log->l_curr_cycle, after_umount_blk);
1038 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1039 log->l_curr_cycle, after_umount_blk);
1040 *tail_blk = after_umount_blk;
1043 * Note that the unmount was clean. If the unmount
1044 * was not clean, we need to know this to rebuild the
1045 * superblock counters from the perag headers if we
1046 * have a filesystem using non-persistent counters.
1048 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1053 * Make sure that there are no blocks in front of the head
1054 * with the same cycle number as the head. This can happen
1055 * because we allow multiple outstanding log writes concurrently,
1056 * and the later writes might make it out before earlier ones.
1058 * We use the lsn from before modifying it so that we'll never
1059 * overwrite the unmount record after a clean unmount.
1061 * Do this only if we are going to recover the filesystem
1063 * NOTE: This used to say "if (!readonly)"
1064 * However on Linux, we can & do recover a read-only filesystem.
1065 * We only skip recovery if NORECOVERY is specified on mount,
1066 * in which case we would not be here.
1068 * But... if the -device- itself is readonly, just skip this.
1069 * We can't recover this device anyway, so it won't matter.
1071 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
1072 error = xlog_clear_stale_blocks(log, tail_lsn);
1078 xfs_warn(log->l_mp, "failed to locate log tail");
1083 * Is the log zeroed at all?
1085 * The last binary search should be changed to perform an X block read
1086 * once X becomes small enough. You can then search linearly through
1087 * the X blocks. This will cut down on the number of reads we need to do.
1089 * If the log is partially zeroed, this routine will pass back the blkno
1090 * of the first block with cycle number 0. It won't have a complete LR
1094 * 0 => the log is completely written to
1095 * 1 => use *blk_no as the first block of the log
1096 * <0 => error has occurred
1101 xfs_daddr_t *blk_no)
1105 uint first_cycle, last_cycle;
1106 xfs_daddr_t new_blk, last_blk, start_blk;
1107 xfs_daddr_t num_scan_bblks;
1108 int error, log_bbnum = log->l_logBBsize;
1112 /* check totally zeroed log */
1113 bp = xlog_get_bp(log, 1);
1116 error = xlog_bread(log, 0, 1, bp, &offset);
1120 first_cycle = xlog_get_cycle(offset);
1121 if (first_cycle == 0) { /* completely zeroed log */
1127 /* check partially zeroed log */
1128 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1132 last_cycle = xlog_get_cycle(offset);
1133 if (last_cycle != 0) { /* log completely written to */
1136 } else if (first_cycle != 1) {
1138 * If the cycle of the last block is zero, the cycle of
1139 * the first block must be 1. If it's not, maybe we're
1140 * not looking at a log... Bail out.
1143 "Log inconsistent or not a log (last==0, first!=1)");
1148 /* we have a partially zeroed log */
1149 last_blk = log_bbnum-1;
1150 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1154 * Validate the answer. Because there is no way to guarantee that
1155 * the entire log is made up of log records which are the same size,
1156 * we scan over the defined maximum blocks. At this point, the maximum
1157 * is not chosen to mean anything special. XXXmiken
1159 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1160 ASSERT(num_scan_bblks <= INT_MAX);
1162 if (last_blk < num_scan_bblks)
1163 num_scan_bblks = last_blk;
1164 start_blk = last_blk - num_scan_bblks;
1167 * We search for any instances of cycle number 0 that occur before
1168 * our current estimate of the head. What we're trying to detect is
1169 * 1 ... | 0 | 1 | 0...
1170 * ^ binary search ends here
1172 if ((error = xlog_find_verify_cycle(log, start_blk,
1173 (int)num_scan_bblks, 0, &new_blk)))
1179 * Potentially backup over partial log record write. We don't need
1180 * to search the end of the log because we know it is zero.
1182 error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0);
1197 * These are simple subroutines used by xlog_clear_stale_blocks() below
1198 * to initialize a buffer full of empty log record headers and write
1199 * them into the log.
1210 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1212 memset(buf, 0, BBSIZE);
1213 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1214 recp->h_cycle = cpu_to_be32(cycle);
1215 recp->h_version = cpu_to_be32(
1216 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1217 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1218 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1219 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1220 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1224 xlog_write_log_records(
1235 int sectbb = log->l_sectBBsize;
1236 int end_block = start_block + blocks;
1242 * Greedily allocate a buffer big enough to handle the full
1243 * range of basic blocks to be written. If that fails, try
1244 * a smaller size. We need to be able to write at least a
1245 * log sector, or we're out of luck.
1247 bufblks = 1 << ffs(blocks);
1248 while (bufblks > log->l_logBBsize)
1250 while (!(bp = xlog_get_bp(log, bufblks))) {
1252 if (bufblks < sectbb)
1256 /* We may need to do a read at the start to fill in part of
1257 * the buffer in the starting sector not covered by the first
1260 balign = round_down(start_block, sectbb);
1261 if (balign != start_block) {
1262 error = xlog_bread_noalign(log, start_block, 1, bp);
1266 j = start_block - balign;
1269 for (i = start_block; i < end_block; i += bufblks) {
1270 int bcount, endcount;
1272 bcount = min(bufblks, end_block - start_block);
1273 endcount = bcount - j;
1275 /* We may need to do a read at the end to fill in part of
1276 * the buffer in the final sector not covered by the write.
1277 * If this is the same sector as the above read, skip it.
1279 ealign = round_down(end_block, sectbb);
1280 if (j == 0 && (start_block + endcount > ealign)) {
1281 offset = bp->b_addr + BBTOB(ealign - start_block);
1282 error = xlog_bread_offset(log, ealign, sectbb,
1289 offset = xlog_align(log, start_block, endcount, bp);
1290 for (; j < endcount; j++) {
1291 xlog_add_record(log, offset, cycle, i+j,
1292 tail_cycle, tail_block);
1295 error = xlog_bwrite(log, start_block, endcount, bp);
1298 start_block += endcount;
1308 * This routine is called to blow away any incomplete log writes out
1309 * in front of the log head. We do this so that we won't become confused
1310 * if we come up, write only a little bit more, and then crash again.
1311 * If we leave the partial log records out there, this situation could
1312 * cause us to think those partial writes are valid blocks since they
1313 * have the current cycle number. We get rid of them by overwriting them
1314 * with empty log records with the old cycle number rather than the
1317 * The tail lsn is passed in rather than taken from
1318 * the log so that we will not write over the unmount record after a
1319 * clean unmount in a 512 block log. Doing so would leave the log without
1320 * any valid log records in it until a new one was written. If we crashed
1321 * during that time we would not be able to recover.
1324 xlog_clear_stale_blocks(
1328 int tail_cycle, head_cycle;
1329 int tail_block, head_block;
1330 int tail_distance, max_distance;
1334 tail_cycle = CYCLE_LSN(tail_lsn);
1335 tail_block = BLOCK_LSN(tail_lsn);
1336 head_cycle = log->l_curr_cycle;
1337 head_block = log->l_curr_block;
1340 * Figure out the distance between the new head of the log
1341 * and the tail. We want to write over any blocks beyond the
1342 * head that we may have written just before the crash, but
1343 * we don't want to overwrite the tail of the log.
1345 if (head_cycle == tail_cycle) {
1347 * The tail is behind the head in the physical log,
1348 * so the distance from the head to the tail is the
1349 * distance from the head to the end of the log plus
1350 * the distance from the beginning of the log to the
1353 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1354 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1355 XFS_ERRLEVEL_LOW, log->l_mp);
1356 return -EFSCORRUPTED;
1358 tail_distance = tail_block + (log->l_logBBsize - head_block);
1361 * The head is behind the tail in the physical log,
1362 * so the distance from the head to the tail is just
1363 * the tail block minus the head block.
1365 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1366 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1367 XFS_ERRLEVEL_LOW, log->l_mp);
1368 return -EFSCORRUPTED;
1370 tail_distance = tail_block - head_block;
1374 * If the head is right up against the tail, we can't clear
1377 if (tail_distance <= 0) {
1378 ASSERT(tail_distance == 0);
1382 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1384 * Take the smaller of the maximum amount of outstanding I/O
1385 * we could have and the distance to the tail to clear out.
1386 * We take the smaller so that we don't overwrite the tail and
1387 * we don't waste all day writing from the head to the tail
1390 max_distance = MIN(max_distance, tail_distance);
1392 if ((head_block + max_distance) <= log->l_logBBsize) {
1394 * We can stomp all the blocks we need to without
1395 * wrapping around the end of the log. Just do it
1396 * in a single write. Use the cycle number of the
1397 * current cycle minus one so that the log will look like:
1400 error = xlog_write_log_records(log, (head_cycle - 1),
1401 head_block, max_distance, tail_cycle,
1407 * We need to wrap around the end of the physical log in
1408 * order to clear all the blocks. Do it in two separate
1409 * I/Os. The first write should be from the head to the
1410 * end of the physical log, and it should use the current
1411 * cycle number minus one just like above.
1413 distance = log->l_logBBsize - head_block;
1414 error = xlog_write_log_records(log, (head_cycle - 1),
1415 head_block, distance, tail_cycle,
1422 * Now write the blocks at the start of the physical log.
1423 * This writes the remainder of the blocks we want to clear.
1424 * It uses the current cycle number since we're now on the
1425 * same cycle as the head so that we get:
1426 * n ... n ... | n - 1 ...
1427 * ^^^^^ blocks we're writing
1429 distance = max_distance - (log->l_logBBsize - head_block);
1430 error = xlog_write_log_records(log, head_cycle, 0, distance,
1431 tail_cycle, tail_block);
1439 /******************************************************************************
1441 * Log recover routines
1443 ******************************************************************************
1446 STATIC xlog_recover_t *
1447 xlog_recover_find_tid(
1448 struct hlist_head *head,
1451 xlog_recover_t *trans;
1453 hlist_for_each_entry(trans, head, r_list) {
1454 if (trans->r_log_tid == tid)
1461 xlog_recover_new_tid(
1462 struct hlist_head *head,
1466 xlog_recover_t *trans;
1468 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1469 trans->r_log_tid = tid;
1471 INIT_LIST_HEAD(&trans->r_itemq);
1473 INIT_HLIST_NODE(&trans->r_list);
1474 hlist_add_head(&trans->r_list, head);
1478 xlog_recover_add_item(
1479 struct list_head *head)
1481 xlog_recover_item_t *item;
1483 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1484 INIT_LIST_HEAD(&item->ri_list);
1485 list_add_tail(&item->ri_list, head);
1489 xlog_recover_add_to_cont_trans(
1491 struct xlog_recover *trans,
1495 xlog_recover_item_t *item;
1496 xfs_caddr_t ptr, old_ptr;
1499 if (list_empty(&trans->r_itemq)) {
1500 /* finish copying rest of trans header */
1501 xlog_recover_add_item(&trans->r_itemq);
1502 ptr = (xfs_caddr_t) &trans->r_theader +
1503 sizeof(xfs_trans_header_t) - len;
1504 memcpy(ptr, dp, len); /* d, s, l */
1507 /* take the tail entry */
1508 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1510 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1511 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1513 ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP);
1514 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1515 item->ri_buf[item->ri_cnt-1].i_len += len;
1516 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1517 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
1522 * The next region to add is the start of a new region. It could be
1523 * a whole region or it could be the first part of a new region. Because
1524 * of this, the assumption here is that the type and size fields of all
1525 * format structures fit into the first 32 bits of the structure.
1527 * This works because all regions must be 32 bit aligned. Therefore, we
1528 * either have both fields or we have neither field. In the case we have
1529 * neither field, the data part of the region is zero length. We only have
1530 * a log_op_header and can throw away the header since a new one will appear
1531 * later. If we have at least 4 bytes, then we can determine how many regions
1532 * will appear in the current log item.
1535 xlog_recover_add_to_trans(
1537 struct xlog_recover *trans,
1541 xfs_inode_log_format_t *in_f; /* any will do */
1542 xlog_recover_item_t *item;
1547 if (list_empty(&trans->r_itemq)) {
1548 /* we need to catch log corruptions here */
1549 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
1550 xfs_warn(log->l_mp, "%s: bad header magic number",
1555 if (len == sizeof(xfs_trans_header_t))
1556 xlog_recover_add_item(&trans->r_itemq);
1557 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1561 ptr = kmem_alloc(len, KM_SLEEP);
1562 memcpy(ptr, dp, len);
1563 in_f = (xfs_inode_log_format_t *)ptr;
1565 /* take the tail entry */
1566 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1567 if (item->ri_total != 0 &&
1568 item->ri_total == item->ri_cnt) {
1569 /* tail item is in use, get a new one */
1570 xlog_recover_add_item(&trans->r_itemq);
1571 item = list_entry(trans->r_itemq.prev,
1572 xlog_recover_item_t, ri_list);
1575 if (item->ri_total == 0) { /* first region to be added */
1576 if (in_f->ilf_size == 0 ||
1577 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
1579 "bad number of regions (%d) in inode log format",
1586 item->ri_total = in_f->ilf_size;
1588 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1591 ASSERT(item->ri_total > item->ri_cnt);
1592 /* Description region is ri_buf[0] */
1593 item->ri_buf[item->ri_cnt].i_addr = ptr;
1594 item->ri_buf[item->ri_cnt].i_len = len;
1596 trace_xfs_log_recover_item_add(log, trans, item, 0);
1601 * Sort the log items in the transaction.
1603 * The ordering constraints are defined by the inode allocation and unlink
1604 * behaviour. The rules are:
1606 * 1. Every item is only logged once in a given transaction. Hence it
1607 * represents the last logged state of the item. Hence ordering is
1608 * dependent on the order in which operations need to be performed so
1609 * required initial conditions are always met.
1611 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1612 * there's nothing to replay from them so we can simply cull them
1613 * from the transaction. However, we can't do that until after we've
1614 * replayed all the other items because they may be dependent on the
1615 * cancelled buffer and replaying the cancelled buffer can remove it
1616 * form the cancelled buffer table. Hence they have tobe done last.
1618 * 3. Inode allocation buffers must be replayed before inode items that
1619 * read the buffer and replay changes into it. For filesystems using the
1620 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1621 * treated the same as inode allocation buffers as they create and
1622 * initialise the buffers directly.
1624 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1625 * This ensures that inodes are completely flushed to the inode buffer
1626 * in a "free" state before we remove the unlinked inode list pointer.
1628 * Hence the ordering needs to be inode allocation buffers first, inode items
1629 * second, inode unlink buffers third and cancelled buffers last.
1631 * But there's a problem with that - we can't tell an inode allocation buffer
1632 * apart from a regular buffer, so we can't separate them. We can, however,
1633 * tell an inode unlink buffer from the others, and so we can separate them out
1634 * from all the other buffers and move them to last.
1636 * Hence, 4 lists, in order from head to tail:
1637 * - buffer_list for all buffers except cancelled/inode unlink buffers
1638 * - item_list for all non-buffer items
1639 * - inode_buffer_list for inode unlink buffers
1640 * - cancel_list for the cancelled buffers
1642 * Note that we add objects to the tail of the lists so that first-to-last
1643 * ordering is preserved within the lists. Adding objects to the head of the
1644 * list means when we traverse from the head we walk them in last-to-first
1645 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1646 * but for all other items there may be specific ordering that we need to
1650 xlog_recover_reorder_trans(
1652 struct xlog_recover *trans,
1655 xlog_recover_item_t *item, *n;
1657 LIST_HEAD(sort_list);
1658 LIST_HEAD(cancel_list);
1659 LIST_HEAD(buffer_list);
1660 LIST_HEAD(inode_buffer_list);
1661 LIST_HEAD(inode_list);
1663 list_splice_init(&trans->r_itemq, &sort_list);
1664 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1665 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1667 switch (ITEM_TYPE(item)) {
1668 case XFS_LI_ICREATE:
1669 list_move_tail(&item->ri_list, &buffer_list);
1672 if (buf_f->blf_flags & XFS_BLF_CANCEL) {
1673 trace_xfs_log_recover_item_reorder_head(log,
1675 list_move(&item->ri_list, &cancel_list);
1678 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1679 list_move(&item->ri_list, &inode_buffer_list);
1682 list_move_tail(&item->ri_list, &buffer_list);
1686 case XFS_LI_QUOTAOFF:
1689 trace_xfs_log_recover_item_reorder_tail(log,
1691 list_move_tail(&item->ri_list, &inode_list);
1695 "%s: unrecognized type of log operation",
1699 * return the remaining items back to the transaction
1700 * item list so they can be freed in caller.
1702 if (!list_empty(&sort_list))
1703 list_splice_init(&sort_list, &trans->r_itemq);
1709 ASSERT(list_empty(&sort_list));
1710 if (!list_empty(&buffer_list))
1711 list_splice(&buffer_list, &trans->r_itemq);
1712 if (!list_empty(&inode_list))
1713 list_splice_tail(&inode_list, &trans->r_itemq);
1714 if (!list_empty(&inode_buffer_list))
1715 list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1716 if (!list_empty(&cancel_list))
1717 list_splice_tail(&cancel_list, &trans->r_itemq);
1722 * Build up the table of buf cancel records so that we don't replay
1723 * cancelled data in the second pass. For buffer records that are
1724 * not cancel records, there is nothing to do here so we just return.
1726 * If we get a cancel record which is already in the table, this indicates
1727 * that the buffer was cancelled multiple times. In order to ensure
1728 * that during pass 2 we keep the record in the table until we reach its
1729 * last occurrence in the log, we keep a reference count in the cancel
1730 * record in the table to tell us how many times we expect to see this
1731 * record during the second pass.
1734 xlog_recover_buffer_pass1(
1736 struct xlog_recover_item *item)
1738 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1739 struct list_head *bucket;
1740 struct xfs_buf_cancel *bcp;
1743 * If this isn't a cancel buffer item, then just return.
1745 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1746 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
1751 * Insert an xfs_buf_cancel record into the hash table of them.
1752 * If there is already an identical record, bump its reference count.
1754 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1755 list_for_each_entry(bcp, bucket, bc_list) {
1756 if (bcp->bc_blkno == buf_f->blf_blkno &&
1757 bcp->bc_len == buf_f->blf_len) {
1759 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
1764 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1765 bcp->bc_blkno = buf_f->blf_blkno;
1766 bcp->bc_len = buf_f->blf_len;
1767 bcp->bc_refcount = 1;
1768 list_add_tail(&bcp->bc_list, bucket);
1770 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
1775 * Check to see whether the buffer being recovered has a corresponding
1776 * entry in the buffer cancel record table. If it is, return the cancel
1777 * buffer structure to the caller.
1779 STATIC struct xfs_buf_cancel *
1780 xlog_peek_buffer_cancelled(
1786 struct list_head *bucket;
1787 struct xfs_buf_cancel *bcp;
1789 if (!log->l_buf_cancel_table) {
1790 /* empty table means no cancelled buffers in the log */
1791 ASSERT(!(flags & XFS_BLF_CANCEL));
1795 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1796 list_for_each_entry(bcp, bucket, bc_list) {
1797 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
1802 * We didn't find a corresponding entry in the table, so return 0 so
1803 * that the buffer is NOT cancelled.
1805 ASSERT(!(flags & XFS_BLF_CANCEL));
1810 * If the buffer is being cancelled then return 1 so that it will be cancelled,
1811 * otherwise return 0. If the buffer is actually a buffer cancel item
1812 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
1813 * table and remove it from the table if this is the last reference.
1815 * We remove the cancel record from the table when we encounter its last
1816 * occurrence in the log so that if the same buffer is re-used again after its
1817 * last cancellation we actually replay the changes made at that point.
1820 xlog_check_buffer_cancelled(
1826 struct xfs_buf_cancel *bcp;
1828 bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags);
1833 * We've go a match, so return 1 so that the recovery of this buffer
1834 * is cancelled. If this buffer is actually a buffer cancel log
1835 * item, then decrement the refcount on the one in the table and
1836 * remove it if this is the last reference.
1838 if (flags & XFS_BLF_CANCEL) {
1839 if (--bcp->bc_refcount == 0) {
1840 list_del(&bcp->bc_list);
1848 * Perform recovery for a buffer full of inodes. In these buffers, the only
1849 * data which should be recovered is that which corresponds to the
1850 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1851 * data for the inodes is always logged through the inodes themselves rather
1852 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1854 * The only time when buffers full of inodes are fully recovered is when the
1855 * buffer is full of newly allocated inodes. In this case the buffer will
1856 * not be marked as an inode buffer and so will be sent to
1857 * xlog_recover_do_reg_buffer() below during recovery.
1860 xlog_recover_do_inode_buffer(
1861 struct xfs_mount *mp,
1862 xlog_recover_item_t *item,
1864 xfs_buf_log_format_t *buf_f)
1870 int reg_buf_offset = 0;
1871 int reg_buf_bytes = 0;
1872 int next_unlinked_offset;
1874 xfs_agino_t *logged_nextp;
1875 xfs_agino_t *buffer_nextp;
1877 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
1880 * Post recovery validation only works properly on CRC enabled
1883 if (xfs_sb_version_hascrc(&mp->m_sb))
1884 bp->b_ops = &xfs_inode_buf_ops;
1886 inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog;
1887 for (i = 0; i < inodes_per_buf; i++) {
1888 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1889 offsetof(xfs_dinode_t, di_next_unlinked);
1891 while (next_unlinked_offset >=
1892 (reg_buf_offset + reg_buf_bytes)) {
1894 * The next di_next_unlinked field is beyond
1895 * the current logged region. Find the next
1896 * logged region that contains or is beyond
1897 * the current di_next_unlinked field.
1900 bit = xfs_next_bit(buf_f->blf_data_map,
1901 buf_f->blf_map_size, bit);
1904 * If there are no more logged regions in the
1905 * buffer, then we're done.
1910 nbits = xfs_contig_bits(buf_f->blf_data_map,
1911 buf_f->blf_map_size, bit);
1913 reg_buf_offset = bit << XFS_BLF_SHIFT;
1914 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
1919 * If the current logged region starts after the current
1920 * di_next_unlinked field, then move on to the next
1921 * di_next_unlinked field.
1923 if (next_unlinked_offset < reg_buf_offset)
1926 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1927 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
1928 ASSERT((reg_buf_offset + reg_buf_bytes) <=
1929 BBTOB(bp->b_io_length));
1932 * The current logged region contains a copy of the
1933 * current di_next_unlinked field. Extract its value
1934 * and copy it to the buffer copy.
1936 logged_nextp = item->ri_buf[item_index].i_addr +
1937 next_unlinked_offset - reg_buf_offset;
1938 if (unlikely(*logged_nextp == 0)) {
1940 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1941 "Trying to replay bad (0) inode di_next_unlinked field.",
1943 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1944 XFS_ERRLEVEL_LOW, mp);
1945 return -EFSCORRUPTED;
1948 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1949 next_unlinked_offset);
1950 *buffer_nextp = *logged_nextp;
1953 * If necessary, recalculate the CRC in the on-disk inode. We
1954 * have to leave the inode in a consistent state for whoever
1957 xfs_dinode_calc_crc(mp, (struct xfs_dinode *)
1958 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
1966 * V5 filesystems know the age of the buffer on disk being recovered. We can
1967 * have newer objects on disk than we are replaying, and so for these cases we
1968 * don't want to replay the current change as that will make the buffer contents
1969 * temporarily invalid on disk.
1971 * The magic number might not match the buffer type we are going to recover
1972 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
1973 * extract the LSN of the existing object in the buffer based on it's current
1974 * magic number. If we don't recognise the magic number in the buffer, then
1975 * return a LSN of -1 so that the caller knows it was an unrecognised block and
1976 * so can recover the buffer.
1978 * Note: we cannot rely solely on magic number matches to determine that the
1979 * buffer has a valid LSN - we also need to verify that it belongs to this
1980 * filesystem, so we need to extract the object's LSN and compare it to that
1981 * which we read from the superblock. If the UUIDs don't match, then we've got a
1982 * stale metadata block from an old filesystem instance that we need to recover
1986 xlog_recover_get_buf_lsn(
1987 struct xfs_mount *mp,
1993 void *blk = bp->b_addr;
1997 /* v4 filesystems always recover immediately */
1998 if (!xfs_sb_version_hascrc(&mp->m_sb))
1999 goto recover_immediately;
2001 magic32 = be32_to_cpu(*(__be32 *)blk);
2003 case XFS_ABTB_CRC_MAGIC:
2004 case XFS_ABTC_CRC_MAGIC:
2005 case XFS_ABTB_MAGIC:
2006 case XFS_ABTC_MAGIC:
2007 case XFS_IBT_CRC_MAGIC:
2008 case XFS_IBT_MAGIC: {
2009 struct xfs_btree_block *btb = blk;
2011 lsn = be64_to_cpu(btb->bb_u.s.bb_lsn);
2012 uuid = &btb->bb_u.s.bb_uuid;
2015 case XFS_BMAP_CRC_MAGIC:
2016 case XFS_BMAP_MAGIC: {
2017 struct xfs_btree_block *btb = blk;
2019 lsn = be64_to_cpu(btb->bb_u.l.bb_lsn);
2020 uuid = &btb->bb_u.l.bb_uuid;
2024 lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn);
2025 uuid = &((struct xfs_agf *)blk)->agf_uuid;
2027 case XFS_AGFL_MAGIC:
2028 lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn);
2029 uuid = &((struct xfs_agfl *)blk)->agfl_uuid;
2032 lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn);
2033 uuid = &((struct xfs_agi *)blk)->agi_uuid;
2035 case XFS_SYMLINK_MAGIC:
2036 lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn);
2037 uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid;
2039 case XFS_DIR3_BLOCK_MAGIC:
2040 case XFS_DIR3_DATA_MAGIC:
2041 case XFS_DIR3_FREE_MAGIC:
2042 lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn);
2043 uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid;
2045 case XFS_ATTR3_RMT_MAGIC:
2046 lsn = be64_to_cpu(((struct xfs_attr3_rmt_hdr *)blk)->rm_lsn);
2047 uuid = &((struct xfs_attr3_rmt_hdr *)blk)->rm_uuid;
2050 lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn);
2051 uuid = &((struct xfs_dsb *)blk)->sb_uuid;
2057 if (lsn != (xfs_lsn_t)-1) {
2058 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
2059 goto recover_immediately;
2063 magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic);
2065 case XFS_DIR3_LEAF1_MAGIC:
2066 case XFS_DIR3_LEAFN_MAGIC:
2067 case XFS_DA3_NODE_MAGIC:
2068 lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn);
2069 uuid = &((struct xfs_da3_blkinfo *)blk)->uuid;
2075 if (lsn != (xfs_lsn_t)-1) {
2076 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
2077 goto recover_immediately;
2082 * We do individual object checks on dquot and inode buffers as they
2083 * have their own individual LSN records. Also, we could have a stale
2084 * buffer here, so we have to at least recognise these buffer types.
2086 * A notd complexity here is inode unlinked list processing - it logs
2087 * the inode directly in the buffer, but we don't know which inodes have
2088 * been modified, and there is no global buffer LSN. Hence we need to
2089 * recover all inode buffer types immediately. This problem will be
2090 * fixed by logical logging of the unlinked list modifications.
2092 magic16 = be16_to_cpu(*(__be16 *)blk);
2094 case XFS_DQUOT_MAGIC:
2095 case XFS_DINODE_MAGIC:
2096 goto recover_immediately;
2101 /* unknown buffer contents, recover immediately */
2103 recover_immediately:
2104 return (xfs_lsn_t)-1;
2109 * Validate the recovered buffer is of the correct type and attach the
2110 * appropriate buffer operations to them for writeback. Magic numbers are in a
2112 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2113 * the first 32 bits of the buffer (most blocks),
2114 * inside a struct xfs_da_blkinfo at the start of the buffer.
2117 xlog_recover_validate_buf_type(
2118 struct xfs_mount *mp,
2120 xfs_buf_log_format_t *buf_f)
2122 struct xfs_da_blkinfo *info = bp->b_addr;
2128 * We can only do post recovery validation on items on CRC enabled
2129 * fielsystems as we need to know when the buffer was written to be able
2130 * to determine if we should have replayed the item. If we replay old
2131 * metadata over a newer buffer, then it will enter a temporarily
2132 * inconsistent state resulting in verification failures. Hence for now
2133 * just avoid the verification stage for non-crc filesystems
2135 if (!xfs_sb_version_hascrc(&mp->m_sb))
2138 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
2139 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
2140 magicda = be16_to_cpu(info->magic);
2141 switch (xfs_blft_from_flags(buf_f)) {
2142 case XFS_BLFT_BTREE_BUF:
2144 case XFS_ABTB_CRC_MAGIC:
2145 case XFS_ABTC_CRC_MAGIC:
2146 case XFS_ABTB_MAGIC:
2147 case XFS_ABTC_MAGIC:
2148 bp->b_ops = &xfs_allocbt_buf_ops;
2150 case XFS_IBT_CRC_MAGIC:
2151 case XFS_FIBT_CRC_MAGIC:
2153 case XFS_FIBT_MAGIC:
2154 bp->b_ops = &xfs_inobt_buf_ops;
2156 case XFS_BMAP_CRC_MAGIC:
2157 case XFS_BMAP_MAGIC:
2158 bp->b_ops = &xfs_bmbt_buf_ops;
2161 xfs_warn(mp, "Bad btree block magic!");
2166 case XFS_BLFT_AGF_BUF:
2167 if (magic32 != XFS_AGF_MAGIC) {
2168 xfs_warn(mp, "Bad AGF block magic!");
2172 bp->b_ops = &xfs_agf_buf_ops;
2174 case XFS_BLFT_AGFL_BUF:
2175 if (magic32 != XFS_AGFL_MAGIC) {
2176 xfs_warn(mp, "Bad AGFL block magic!");
2180 bp->b_ops = &xfs_agfl_buf_ops;
2182 case XFS_BLFT_AGI_BUF:
2183 if (magic32 != XFS_AGI_MAGIC) {
2184 xfs_warn(mp, "Bad AGI block magic!");
2188 bp->b_ops = &xfs_agi_buf_ops;
2190 case XFS_BLFT_UDQUOT_BUF:
2191 case XFS_BLFT_PDQUOT_BUF:
2192 case XFS_BLFT_GDQUOT_BUF:
2193 #ifdef CONFIG_XFS_QUOTA
2194 if (magic16 != XFS_DQUOT_MAGIC) {
2195 xfs_warn(mp, "Bad DQUOT block magic!");
2199 bp->b_ops = &xfs_dquot_buf_ops;
2202 "Trying to recover dquots without QUOTA support built in!");
2206 case XFS_BLFT_DINO_BUF:
2207 if (magic16 != XFS_DINODE_MAGIC) {
2208 xfs_warn(mp, "Bad INODE block magic!");
2212 bp->b_ops = &xfs_inode_buf_ops;
2214 case XFS_BLFT_SYMLINK_BUF:
2215 if (magic32 != XFS_SYMLINK_MAGIC) {
2216 xfs_warn(mp, "Bad symlink block magic!");
2220 bp->b_ops = &xfs_symlink_buf_ops;
2222 case XFS_BLFT_DIR_BLOCK_BUF:
2223 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2224 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2225 xfs_warn(mp, "Bad dir block magic!");
2229 bp->b_ops = &xfs_dir3_block_buf_ops;
2231 case XFS_BLFT_DIR_DATA_BUF:
2232 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2233 magic32 != XFS_DIR3_DATA_MAGIC) {
2234 xfs_warn(mp, "Bad dir data magic!");
2238 bp->b_ops = &xfs_dir3_data_buf_ops;
2240 case XFS_BLFT_DIR_FREE_BUF:
2241 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2242 magic32 != XFS_DIR3_FREE_MAGIC) {
2243 xfs_warn(mp, "Bad dir3 free magic!");
2247 bp->b_ops = &xfs_dir3_free_buf_ops;
2249 case XFS_BLFT_DIR_LEAF1_BUF:
2250 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2251 magicda != XFS_DIR3_LEAF1_MAGIC) {
2252 xfs_warn(mp, "Bad dir leaf1 magic!");
2256 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2258 case XFS_BLFT_DIR_LEAFN_BUF:
2259 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2260 magicda != XFS_DIR3_LEAFN_MAGIC) {
2261 xfs_warn(mp, "Bad dir leafn magic!");
2265 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2267 case XFS_BLFT_DA_NODE_BUF:
2268 if (magicda != XFS_DA_NODE_MAGIC &&
2269 magicda != XFS_DA3_NODE_MAGIC) {
2270 xfs_warn(mp, "Bad da node magic!");
2274 bp->b_ops = &xfs_da3_node_buf_ops;
2276 case XFS_BLFT_ATTR_LEAF_BUF:
2277 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2278 magicda != XFS_ATTR3_LEAF_MAGIC) {
2279 xfs_warn(mp, "Bad attr leaf magic!");
2283 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2285 case XFS_BLFT_ATTR_RMT_BUF:
2286 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
2287 xfs_warn(mp, "Bad attr remote magic!");
2291 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2293 case XFS_BLFT_SB_BUF:
2294 if (magic32 != XFS_SB_MAGIC) {
2295 xfs_warn(mp, "Bad SB block magic!");
2299 bp->b_ops = &xfs_sb_buf_ops;
2302 xfs_warn(mp, "Unknown buffer type %d!",
2303 xfs_blft_from_flags(buf_f));
2309 * Perform a 'normal' buffer recovery. Each logged region of the
2310 * buffer should be copied over the corresponding region in the
2311 * given buffer. The bitmap in the buf log format structure indicates
2312 * where to place the logged data.
2315 xlog_recover_do_reg_buffer(
2316 struct xfs_mount *mp,
2317 xlog_recover_item_t *item,
2319 xfs_buf_log_format_t *buf_f)
2326 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2329 i = 1; /* 0 is the buf format structure */
2331 bit = xfs_next_bit(buf_f->blf_data_map,
2332 buf_f->blf_map_size, bit);
2335 nbits = xfs_contig_bits(buf_f->blf_data_map,
2336 buf_f->blf_map_size, bit);
2338 ASSERT(item->ri_buf[i].i_addr != NULL);
2339 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
2340 ASSERT(BBTOB(bp->b_io_length) >=
2341 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
2344 * The dirty regions logged in the buffer, even though
2345 * contiguous, may span multiple chunks. This is because the
2346 * dirty region may span a physical page boundary in a buffer
2347 * and hence be split into two separate vectors for writing into
2348 * the log. Hence we need to trim nbits back to the length of
2349 * the current region being copied out of the log.
2351 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2352 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2355 * Do a sanity check if this is a dquot buffer. Just checking
2356 * the first dquot in the buffer should do. XXXThis is
2357 * probably a good thing to do for other buf types also.
2360 if (buf_f->blf_flags &
2361 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2362 if (item->ri_buf[i].i_addr == NULL) {
2364 "XFS: NULL dquot in %s.", __func__);
2367 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
2369 "XFS: dquot too small (%d) in %s.",
2370 item->ri_buf[i].i_len, __func__);
2373 error = xfs_dqcheck(mp, item->ri_buf[i].i_addr,
2374 -1, 0, XFS_QMOPT_DOWARN,
2375 "dquot_buf_recover");
2380 memcpy(xfs_buf_offset(bp,
2381 (uint)bit << XFS_BLF_SHIFT), /* dest */
2382 item->ri_buf[i].i_addr, /* source */
2383 nbits<<XFS_BLF_SHIFT); /* length */
2389 /* Shouldn't be any more regions */
2390 ASSERT(i == item->ri_total);
2392 xlog_recover_validate_buf_type(mp, bp, buf_f);
2396 * Perform a dquot buffer recovery.
2397 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2398 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2399 * Else, treat it as a regular buffer and do recovery.
2401 * Return false if the buffer was tossed and true if we recovered the buffer to
2402 * indicate to the caller if the buffer needs writing.
2405 xlog_recover_do_dquot_buffer(
2406 struct xfs_mount *mp,
2408 struct xlog_recover_item *item,
2410 struct xfs_buf_log_format *buf_f)
2414 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2417 * Filesystems are required to send in quota flags at mount time.
2423 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2424 type |= XFS_DQ_USER;
2425 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2426 type |= XFS_DQ_PROJ;
2427 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2428 type |= XFS_DQ_GROUP;
2430 * This type of quotas was turned off, so ignore this buffer
2432 if (log->l_quotaoffs_flag & type)
2435 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2440 * This routine replays a modification made to a buffer at runtime.
2441 * There are actually two types of buffer, regular and inode, which
2442 * are handled differently. Inode buffers are handled differently
2443 * in that we only recover a specific set of data from them, namely
2444 * the inode di_next_unlinked fields. This is because all other inode
2445 * data is actually logged via inode records and any data we replay
2446 * here which overlaps that may be stale.
2448 * When meta-data buffers are freed at run time we log a buffer item
2449 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2450 * of the buffer in the log should not be replayed at recovery time.
2451 * This is so that if the blocks covered by the buffer are reused for
2452 * file data before we crash we don't end up replaying old, freed
2453 * meta-data into a user's file.
2455 * To handle the cancellation of buffer log items, we make two passes
2456 * over the log during recovery. During the first we build a table of
2457 * those buffers which have been cancelled, and during the second we
2458 * only replay those buffers which do not have corresponding cancel
2459 * records in the table. See xlog_recover_buffer_pass[1,2] above
2460 * for more details on the implementation of the table of cancel records.
2463 xlog_recover_buffer_pass2(
2465 struct list_head *buffer_list,
2466 struct xlog_recover_item *item,
2467 xfs_lsn_t current_lsn)
2469 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2470 xfs_mount_t *mp = log->l_mp;
2477 * In this pass we only want to recover all the buffers which have
2478 * not been cancelled and are not cancellation buffers themselves.
2480 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2481 buf_f->blf_len, buf_f->blf_flags)) {
2482 trace_xfs_log_recover_buf_cancel(log, buf_f);
2486 trace_xfs_log_recover_buf_recover(log, buf_f);
2489 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2490 buf_flags |= XBF_UNMAPPED;
2492 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2496 error = bp->b_error;
2498 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
2503 * Recover the buffer only if we get an LSN from it and it's less than
2504 * the lsn of the transaction we are replaying.
2506 * Note that we have to be extremely careful of readahead here.
2507 * Readahead does not attach verfiers to the buffers so if we don't
2508 * actually do any replay after readahead because of the LSN we found
2509 * in the buffer if more recent than that current transaction then we
2510 * need to attach the verifier directly. Failure to do so can lead to
2511 * future recovery actions (e.g. EFI and unlinked list recovery) can
2512 * operate on the buffers and they won't get the verifier attached. This
2513 * can lead to blocks on disk having the correct content but a stale
2516 * It is safe to assume these clean buffers are currently up to date.
2517 * If the buffer is dirtied by a later transaction being replayed, then
2518 * the verifier will be reset to match whatever recover turns that
2521 lsn = xlog_recover_get_buf_lsn(mp, bp);
2522 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2523 xlog_recover_validate_buf_type(mp, bp, buf_f);
2527 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
2528 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2531 } else if (buf_f->blf_flags &
2532 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2535 dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2539 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2543 * Perform delayed write on the buffer. Asynchronous writes will be
2544 * slower when taking into account all the buffers to be flushed.
2546 * Also make sure that only inode buffers with good sizes stay in
2547 * the buffer cache. The kernel moves inodes in buffers of 1 block
2548 * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode
2549 * buffers in the log can be a different size if the log was generated
2550 * by an older kernel using unclustered inode buffers or a newer kernel
2551 * running with a different inode cluster size. Regardless, if the
2552 * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size)
2553 * for *our* value of mp->m_inode_cluster_size, then we need to keep
2554 * the buffer out of the buffer cache so that the buffer won't
2555 * overlap with future reads of those inodes.
2557 if (XFS_DINODE_MAGIC ==
2558 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2559 (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize,
2560 (__uint32_t)log->l_mp->m_inode_cluster_size))) {
2562 error = xfs_bwrite(bp);
2564 ASSERT(bp->b_target->bt_mount == mp);
2565 bp->b_iodone = xlog_recover_iodone;
2566 xfs_buf_delwri_queue(bp, buffer_list);
2575 * Inode fork owner changes
2577 * If we have been told that we have to reparent the inode fork, it's because an
2578 * extent swap operation on a CRC enabled filesystem has been done and we are
2579 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2582 * The complexity here is that we don't have an inode context to work with, so
2583 * after we've replayed the inode we need to instantiate one. This is where the
2586 * We are in the middle of log recovery, so we can't run transactions. That
2587 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2588 * that will result in the corresponding iput() running the inode through
2589 * xfs_inactive(). If we've just replayed an inode core that changes the link
2590 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2591 * transactions (bad!).
2593 * So, to avoid this, we instantiate an inode directly from the inode core we've
2594 * just recovered. We have the buffer still locked, and all we really need to
2595 * instantiate is the inode core and the forks being modified. We can do this
2596 * manually, then run the inode btree owner change, and then tear down the
2597 * xfs_inode without having to run any transactions at all.
2599 * Also, because we don't have a transaction context available here but need to
2600 * gather all the buffers we modify for writeback so we pass the buffer_list
2601 * instead for the operation to use.
2605 xfs_recover_inode_owner_change(
2606 struct xfs_mount *mp,
2607 struct xfs_dinode *dip,
2608 struct xfs_inode_log_format *in_f,
2609 struct list_head *buffer_list)
2611 struct xfs_inode *ip;
2614 ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER));
2616 ip = xfs_inode_alloc(mp, in_f->ilf_ino);
2620 /* instantiate the inode */
2621 xfs_dinode_from_disk(&ip->i_d, dip);
2622 ASSERT(ip->i_d.di_version >= 3);
2624 error = xfs_iformat_fork(ip, dip);
2629 if (in_f->ilf_fields & XFS_ILOG_DOWNER) {
2630 ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT);
2631 error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK,
2632 ip->i_ino, buffer_list);
2637 if (in_f->ilf_fields & XFS_ILOG_AOWNER) {
2638 ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT);
2639 error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK,
2640 ip->i_ino, buffer_list);
2651 xlog_recover_inode_pass2(
2653 struct list_head *buffer_list,
2654 struct xlog_recover_item *item,
2655 xfs_lsn_t current_lsn)
2657 xfs_inode_log_format_t *in_f;
2658 xfs_mount_t *mp = log->l_mp;
2667 xfs_icdinode_t *dicp;
2671 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2672 in_f = item->ri_buf[0].i_addr;
2674 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
2676 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2682 * Inode buffers can be freed, look out for it,
2683 * and do not replay the inode.
2685 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2686 in_f->ilf_len, 0)) {
2688 trace_xfs_log_recover_inode_cancel(log, in_f);
2691 trace_xfs_log_recover_inode_recover(log, in_f);
2693 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
2694 &xfs_inode_buf_ops);
2699 error = bp->b_error;
2701 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
2704 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2705 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
2708 * Make sure the place we're flushing out to really looks
2711 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
2713 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2714 __func__, dip, bp, in_f->ilf_ino);
2715 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2716 XFS_ERRLEVEL_LOW, mp);
2717 error = -EFSCORRUPTED;
2720 dicp = item->ri_buf[1].i_addr;
2721 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2723 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2724 __func__, item, in_f->ilf_ino);
2725 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2726 XFS_ERRLEVEL_LOW, mp);
2727 error = -EFSCORRUPTED;
2732 * If the inode has an LSN in it, recover the inode only if it's less
2733 * than the lsn of the transaction we are replaying. Note: we still
2734 * need to replay an owner change even though the inode is more recent
2735 * than the transaction as there is no guarantee that all the btree
2736 * blocks are more recent than this transaction, too.
2738 if (dip->di_version >= 3) {
2739 xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn);
2741 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2742 trace_xfs_log_recover_inode_skip(log, in_f);
2744 goto out_owner_change;
2749 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2750 * are transactional and if ordering is necessary we can determine that
2751 * more accurately by the LSN field in the V3 inode core. Don't trust
2752 * the inode versions we might be changing them here - use the
2753 * superblock flag to determine whether we need to look at di_flushiter
2754 * to skip replay when the on disk inode is newer than the log one
2756 if (!xfs_sb_version_hascrc(&mp->m_sb) &&
2757 dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
2759 * Deal with the wrap case, DI_MAX_FLUSH is less
2760 * than smaller numbers
2762 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
2763 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2766 trace_xfs_log_recover_inode_skip(log, in_f);
2772 /* Take the opportunity to reset the flush iteration count */
2773 dicp->di_flushiter = 0;
2775 if (unlikely(S_ISREG(dicp->di_mode))) {
2776 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2777 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2778 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2779 XFS_ERRLEVEL_LOW, mp, dicp);
2781 "%s: Bad regular inode log record, rec ptr 0x%p, "
2782 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2783 __func__, item, dip, bp, in_f->ilf_ino);
2784 error = -EFSCORRUPTED;
2787 } else if (unlikely(S_ISDIR(dicp->di_mode))) {
2788 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2789 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2790 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2791 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2792 XFS_ERRLEVEL_LOW, mp, dicp);
2794 "%s: Bad dir inode log record, rec ptr 0x%p, "
2795 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2796 __func__, item, dip, bp, in_f->ilf_ino);
2797 error = -EFSCORRUPTED;
2801 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2802 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2803 XFS_ERRLEVEL_LOW, mp, dicp);
2805 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2806 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2807 __func__, item, dip, bp, in_f->ilf_ino,
2808 dicp->di_nextents + dicp->di_anextents,
2810 error = -EFSCORRUPTED;
2813 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2814 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2815 XFS_ERRLEVEL_LOW, mp, dicp);
2817 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2818 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
2819 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
2820 error = -EFSCORRUPTED;
2823 isize = xfs_icdinode_size(dicp->di_version);
2824 if (unlikely(item->ri_buf[1].i_len > isize)) {
2825 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
2826 XFS_ERRLEVEL_LOW, mp, dicp);
2828 "%s: Bad inode log record length %d, rec ptr 0x%p",
2829 __func__, item->ri_buf[1].i_len, item);
2830 error = -EFSCORRUPTED;
2834 /* The core is in in-core format */
2835 xfs_dinode_to_disk(dip, dicp);
2837 /* the rest is in on-disk format */
2838 if (item->ri_buf[1].i_len > isize) {
2839 memcpy((char *)dip + isize,
2840 item->ri_buf[1].i_addr + isize,
2841 item->ri_buf[1].i_len - isize);
2844 fields = in_f->ilf_fields;
2845 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2847 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
2850 memcpy(XFS_DFORK_DPTR(dip),
2851 &in_f->ilf_u.ilfu_uuid,
2856 if (in_f->ilf_size == 2)
2857 goto out_owner_change;
2858 len = item->ri_buf[2].i_len;
2859 src = item->ri_buf[2].i_addr;
2860 ASSERT(in_f->ilf_size <= 4);
2861 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2862 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2863 (len == in_f->ilf_dsize));
2865 switch (fields & XFS_ILOG_DFORK) {
2866 case XFS_ILOG_DDATA:
2868 memcpy(XFS_DFORK_DPTR(dip), src, len);
2871 case XFS_ILOG_DBROOT:
2872 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
2873 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
2874 XFS_DFORK_DSIZE(dip, mp));
2879 * There are no data fork flags set.
2881 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2886 * If we logged any attribute data, recover it. There may or
2887 * may not have been any other non-core data logged in this
2890 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2891 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2896 len = item->ri_buf[attr_index].i_len;
2897 src = item->ri_buf[attr_index].i_addr;
2898 ASSERT(len == in_f->ilf_asize);
2900 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2901 case XFS_ILOG_ADATA:
2903 dest = XFS_DFORK_APTR(dip);
2904 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2905 memcpy(dest, src, len);
2908 case XFS_ILOG_ABROOT:
2909 dest = XFS_DFORK_APTR(dip);
2910 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2911 len, (xfs_bmdr_block_t*)dest,
2912 XFS_DFORK_ASIZE(dip, mp));
2916 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
2924 if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER))
2925 error = xfs_recover_inode_owner_change(mp, dip, in_f,
2927 /* re-generate the checksum. */
2928 xfs_dinode_calc_crc(log->l_mp, dip);
2930 ASSERT(bp->b_target->bt_mount == mp);
2931 bp->b_iodone = xlog_recover_iodone;
2932 xfs_buf_delwri_queue(bp, buffer_list);
2943 * Recover QUOTAOFF records. We simply make a note of it in the xlog
2944 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2948 xlog_recover_quotaoff_pass1(
2950 struct xlog_recover_item *item)
2952 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
2956 * The logitem format's flag tells us if this was user quotaoff,
2957 * group/project quotaoff or both.
2959 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2960 log->l_quotaoffs_flag |= XFS_DQ_USER;
2961 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2962 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2963 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2964 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2970 * Recover a dquot record
2973 xlog_recover_dquot_pass2(
2975 struct list_head *buffer_list,
2976 struct xlog_recover_item *item,
2977 xfs_lsn_t current_lsn)
2979 xfs_mount_t *mp = log->l_mp;
2981 struct xfs_disk_dquot *ddq, *recddq;
2983 xfs_dq_logformat_t *dq_f;
2988 * Filesystems are required to send in quota flags at mount time.
2990 if (mp->m_qflags == 0)
2993 recddq = item->ri_buf[1].i_addr;
2994 if (recddq == NULL) {
2995 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
2998 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
2999 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
3000 item->ri_buf[1].i_len, __func__);
3005 * This type of quotas was turned off, so ignore this record.
3007 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3009 if (log->l_quotaoffs_flag & type)
3013 * At this point we know that quota was _not_ turned off.
3014 * Since the mount flags are not indicating to us otherwise, this
3015 * must mean that quota is on, and the dquot needs to be replayed.
3016 * Remember that we may not have fully recovered the superblock yet,
3017 * so we can't do the usual trick of looking at the SB quota bits.
3019 * The other possibility, of course, is that the quota subsystem was
3020 * removed since the last mount - ENOSYS.
3022 dq_f = item->ri_buf[0].i_addr;
3024 error = xfs_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
3025 "xlog_recover_dquot_pass2 (log copy)");
3028 ASSERT(dq_f->qlf_len == 1);
3031 * At this point we are assuming that the dquots have been allocated
3032 * and hence the buffer has valid dquots stamped in it. It should,
3033 * therefore, pass verifier validation. If the dquot is bad, then the
3034 * we'll return an error here, so we don't need to specifically check
3035 * the dquot in the buffer after the verifier has run.
3037 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
3038 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
3039 &xfs_dquot_buf_ops);
3044 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
3047 * If the dquot has an LSN in it, recover the dquot only if it's less
3048 * than the lsn of the transaction we are replaying.
3050 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3051 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq;
3052 xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn);
3054 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3059 memcpy(ddq, recddq, item->ri_buf[1].i_len);
3060 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3061 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
3065 ASSERT(dq_f->qlf_size == 2);
3066 ASSERT(bp->b_target->bt_mount == mp);
3067 bp->b_iodone = xlog_recover_iodone;
3068 xfs_buf_delwri_queue(bp, buffer_list);
3076 * This routine is called to create an in-core extent free intent
3077 * item from the efi format structure which was logged on disk.
3078 * It allocates an in-core efi, copies the extents from the format
3079 * structure into it, and adds the efi to the AIL with the given
3083 xlog_recover_efi_pass2(
3085 struct xlog_recover_item *item,
3089 xfs_mount_t *mp = log->l_mp;
3090 xfs_efi_log_item_t *efip;
3091 xfs_efi_log_format_t *efi_formatp;
3093 efi_formatp = item->ri_buf[0].i_addr;
3095 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
3096 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
3097 &(efip->efi_format)))) {
3098 xfs_efi_item_free(efip);
3101 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
3103 spin_lock(&log->l_ailp->xa_lock);
3105 * xfs_trans_ail_update() drops the AIL lock.
3107 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
3113 * This routine is called when an efd format structure is found in
3114 * a committed transaction in the log. It's purpose is to cancel
3115 * the corresponding efi if it was still in the log. To do this
3116 * it searches the AIL for the efi with an id equal to that in the
3117 * efd format structure. If we find it, we remove the efi from the
3121 xlog_recover_efd_pass2(
3123 struct xlog_recover_item *item)
3125 xfs_efd_log_format_t *efd_formatp;
3126 xfs_efi_log_item_t *efip = NULL;
3127 xfs_log_item_t *lip;
3129 struct xfs_ail_cursor cur;
3130 struct xfs_ail *ailp = log->l_ailp;
3132 efd_formatp = item->ri_buf[0].i_addr;
3133 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
3134 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
3135 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
3136 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
3137 efi_id = efd_formatp->efd_efi_id;
3140 * Search for the efi with the id in the efd format structure
3143 spin_lock(&ailp->xa_lock);
3144 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3145 while (lip != NULL) {
3146 if (lip->li_type == XFS_LI_EFI) {
3147 efip = (xfs_efi_log_item_t *)lip;
3148 if (efip->efi_format.efi_id == efi_id) {
3150 * xfs_trans_ail_delete() drops the
3153 xfs_trans_ail_delete(ailp, lip,
3154 SHUTDOWN_CORRUPT_INCORE);
3155 xfs_efi_item_free(efip);
3156 spin_lock(&ailp->xa_lock);
3160 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3162 xfs_trans_ail_cursor_done(&cur);
3163 spin_unlock(&ailp->xa_lock);
3169 * This routine is called when an inode create format structure is found in a
3170 * committed transaction in the log. It's purpose is to initialise the inodes
3171 * being allocated on disk. This requires us to get inode cluster buffers that
3172 * match the range to be intialised, stamped with inode templates and written
3173 * by delayed write so that subsequent modifications will hit the cached buffer
3174 * and only need writing out at the end of recovery.
3177 xlog_recover_do_icreate_pass2(
3179 struct list_head *buffer_list,
3180 xlog_recover_item_t *item)
3182 struct xfs_mount *mp = log->l_mp;
3183 struct xfs_icreate_log *icl;
3184 xfs_agnumber_t agno;
3185 xfs_agblock_t agbno;
3188 xfs_agblock_t length;
3190 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3191 if (icl->icl_type != XFS_LI_ICREATE) {
3192 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
3196 if (icl->icl_size != 1) {
3197 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
3201 agno = be32_to_cpu(icl->icl_ag);
3202 if (agno >= mp->m_sb.sb_agcount) {
3203 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
3206 agbno = be32_to_cpu(icl->icl_agbno);
3207 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3208 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
3211 isize = be32_to_cpu(icl->icl_isize);
3212 if (isize != mp->m_sb.sb_inodesize) {
3213 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
3216 count = be32_to_cpu(icl->icl_count);
3218 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
3221 length = be32_to_cpu(icl->icl_length);
3222 if (!length || length >= mp->m_sb.sb_agblocks) {
3223 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
3227 /* existing allocation is fixed value */
3228 ASSERT(count == mp->m_ialloc_inos);
3229 ASSERT(length == mp->m_ialloc_blks);
3230 if (count != mp->m_ialloc_inos ||
3231 length != mp->m_ialloc_blks) {
3232 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count 2");
3237 * Inode buffers can be freed. Do not replay the inode initialisation as
3238 * we could be overwriting something written after this inode buffer was
3241 * XXX: we need to iterate all buffers and only init those that are not
3242 * cancelled. I think that a more fine grained factoring of
3243 * xfs_ialloc_inode_init may be appropriate here to enable this to be
3246 if (xlog_check_buffer_cancelled(log,
3247 XFS_AGB_TO_DADDR(mp, agno, agbno), length, 0))
3250 xfs_ialloc_inode_init(mp, NULL, buffer_list, agno, agbno, length,
3251 be32_to_cpu(icl->icl_gen));
3256 * Free up any resources allocated by the transaction
3258 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3261 xlog_recover_free_trans(
3262 struct xlog_recover *trans)
3264 xlog_recover_item_t *item, *n;
3267 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
3268 /* Free the regions in the item. */
3269 list_del(&item->ri_list);
3270 for (i = 0; i < item->ri_cnt; i++)
3271 kmem_free(item->ri_buf[i].i_addr);
3272 /* Free the item itself */
3273 kmem_free(item->ri_buf);
3276 /* Free the transaction recover structure */
3281 xlog_recover_buffer_ra_pass2(
3283 struct xlog_recover_item *item)
3285 struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr;
3286 struct xfs_mount *mp = log->l_mp;
3288 if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
3289 buf_f->blf_len, buf_f->blf_flags)) {
3293 xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
3294 buf_f->blf_len, NULL);
3298 xlog_recover_inode_ra_pass2(
3300 struct xlog_recover_item *item)
3302 struct xfs_inode_log_format ilf_buf;
3303 struct xfs_inode_log_format *ilfp;
3304 struct xfs_mount *mp = log->l_mp;
3307 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3308 ilfp = item->ri_buf[0].i_addr;
3311 memset(ilfp, 0, sizeof(*ilfp));
3312 error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
3317 if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
3320 xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
3321 ilfp->ilf_len, &xfs_inode_buf_ra_ops);
3325 xlog_recover_dquot_ra_pass2(
3327 struct xlog_recover_item *item)
3329 struct xfs_mount *mp = log->l_mp;
3330 struct xfs_disk_dquot *recddq;
3331 struct xfs_dq_logformat *dq_f;
3335 if (mp->m_qflags == 0)
3338 recddq = item->ri_buf[1].i_addr;
3341 if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
3344 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3346 if (log->l_quotaoffs_flag & type)
3349 dq_f = item->ri_buf[0].i_addr;
3351 ASSERT(dq_f->qlf_len == 1);
3353 xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno,
3354 XFS_FSB_TO_BB(mp, dq_f->qlf_len), NULL);
3358 xlog_recover_ra_pass2(
3360 struct xlog_recover_item *item)
3362 switch (ITEM_TYPE(item)) {
3364 xlog_recover_buffer_ra_pass2(log, item);
3367 xlog_recover_inode_ra_pass2(log, item);
3370 xlog_recover_dquot_ra_pass2(log, item);
3374 case XFS_LI_QUOTAOFF:
3381 xlog_recover_commit_pass1(
3383 struct xlog_recover *trans,
3384 struct xlog_recover_item *item)
3386 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
3388 switch (ITEM_TYPE(item)) {
3390 return xlog_recover_buffer_pass1(log, item);
3391 case XFS_LI_QUOTAOFF:
3392 return xlog_recover_quotaoff_pass1(log, item);
3397 case XFS_LI_ICREATE:
3398 /* nothing to do in pass 1 */
3401 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3402 __func__, ITEM_TYPE(item));
3409 xlog_recover_commit_pass2(
3411 struct xlog_recover *trans,
3412 struct list_head *buffer_list,
3413 struct xlog_recover_item *item)
3415 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
3417 switch (ITEM_TYPE(item)) {
3419 return xlog_recover_buffer_pass2(log, buffer_list, item,
3422 return xlog_recover_inode_pass2(log, buffer_list, item,
3425 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
3427 return xlog_recover_efd_pass2(log, item);
3429 return xlog_recover_dquot_pass2(log, buffer_list, item,
3431 case XFS_LI_ICREATE:
3432 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
3433 case XFS_LI_QUOTAOFF:
3434 /* nothing to do in pass2 */
3437 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3438 __func__, ITEM_TYPE(item));
3445 xlog_recover_items_pass2(
3447 struct xlog_recover *trans,
3448 struct list_head *buffer_list,
3449 struct list_head *item_list)
3451 struct xlog_recover_item *item;
3454 list_for_each_entry(item, item_list, ri_list) {
3455 error = xlog_recover_commit_pass2(log, trans,
3465 * Perform the transaction.
3467 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3468 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3471 xlog_recover_commit_trans(
3473 struct xlog_recover *trans,
3478 int items_queued = 0;
3479 struct xlog_recover_item *item;
3480 struct xlog_recover_item *next;
3481 LIST_HEAD (buffer_list);
3482 LIST_HEAD (ra_list);
3483 LIST_HEAD (done_list);
3485 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
3487 hlist_del(&trans->r_list);
3489 error = xlog_recover_reorder_trans(log, trans, pass);
3493 list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
3495 case XLOG_RECOVER_PASS1:
3496 error = xlog_recover_commit_pass1(log, trans, item);
3498 case XLOG_RECOVER_PASS2:
3499 xlog_recover_ra_pass2(log, item);
3500 list_move_tail(&item->ri_list, &ra_list);
3502 if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
3503 error = xlog_recover_items_pass2(log, trans,
3504 &buffer_list, &ra_list);
3505 list_splice_tail_init(&ra_list, &done_list);
3519 if (!list_empty(&ra_list)) {
3521 error = xlog_recover_items_pass2(log, trans,
3522 &buffer_list, &ra_list);
3523 list_splice_tail_init(&ra_list, &done_list);
3526 if (!list_empty(&done_list))
3527 list_splice_init(&done_list, &trans->r_itemq);
3529 xlog_recover_free_trans(trans);
3531 error2 = xfs_buf_delwri_submit(&buffer_list);
3532 return error ? error : error2;
3536 xlog_recover_unmount_trans(
3539 /* Do nothing now */
3540 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
3545 * There are two valid states of the r_state field. 0 indicates that the
3546 * transaction structure is in a normal state. We have either seen the
3547 * start of the transaction or the last operation we added was not a partial
3548 * operation. If the last operation we added to the transaction was a
3549 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
3551 * NOTE: skip LRs with 0 data length.
3554 xlog_recover_process_data(
3556 struct hlist_head rhash[],
3557 struct xlog_rec_header *rhead,
3563 xlog_op_header_t *ohead;
3564 xlog_recover_t *trans;
3570 lp = dp + be32_to_cpu(rhead->h_len);
3571 num_logops = be32_to_cpu(rhead->h_num_logops);
3573 /* check the log format matches our own - else we can't recover */
3574 if (xlog_header_check_recover(log->l_mp, rhead))
3577 while ((dp < lp) && num_logops) {
3578 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
3579 ohead = (xlog_op_header_t *)dp;
3580 dp += sizeof(xlog_op_header_t);
3581 if (ohead->oh_clientid != XFS_TRANSACTION &&
3582 ohead->oh_clientid != XFS_LOG) {
3583 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
3584 __func__, ohead->oh_clientid);
3588 tid = be32_to_cpu(ohead->oh_tid);
3589 hash = XLOG_RHASH(tid);
3590 trans = xlog_recover_find_tid(&rhash[hash], tid);
3591 if (trans == NULL) { /* not found; add new tid */
3592 if (ohead->oh_flags & XLOG_START_TRANS)
3593 xlog_recover_new_tid(&rhash[hash], tid,
3594 be64_to_cpu(rhead->h_lsn));
3596 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
3597 xfs_warn(log->l_mp, "%s: bad length 0x%x",
3598 __func__, be32_to_cpu(ohead->oh_len));
3602 flags = ohead->oh_flags & ~XLOG_END_TRANS;
3603 if (flags & XLOG_WAS_CONT_TRANS)
3604 flags &= ~XLOG_CONTINUE_TRANS;
3606 case XLOG_COMMIT_TRANS:
3607 error = xlog_recover_commit_trans(log,
3610 case XLOG_UNMOUNT_TRANS:
3611 error = xlog_recover_unmount_trans(log);
3613 case XLOG_WAS_CONT_TRANS:
3614 error = xlog_recover_add_to_cont_trans(log,
3616 be32_to_cpu(ohead->oh_len));
3618 case XLOG_START_TRANS:
3619 xfs_warn(log->l_mp, "%s: bad transaction",
3625 case XLOG_CONTINUE_TRANS:
3626 error = xlog_recover_add_to_trans(log, trans,
3627 dp, be32_to_cpu(ohead->oh_len));
3630 xfs_warn(log->l_mp, "%s: bad flag 0x%x",
3637 xlog_recover_free_trans(trans);
3641 dp += be32_to_cpu(ohead->oh_len);
3648 * Process an extent free intent item that was recovered from
3649 * the log. We need to free the extents that it describes.
3652 xlog_recover_process_efi(
3654 xfs_efi_log_item_t *efip)
3656 xfs_efd_log_item_t *efdp;
3661 xfs_fsblock_t startblock_fsb;
3663 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
3666 * First check the validity of the extents described by the
3667 * EFI. If any are bad, then assume that all are bad and
3668 * just toss the EFI.
3670 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3671 extp = &(efip->efi_format.efi_extents[i]);
3672 startblock_fsb = XFS_BB_TO_FSB(mp,
3673 XFS_FSB_TO_DADDR(mp, extp->ext_start));
3674 if ((startblock_fsb == 0) ||
3675 (extp->ext_len == 0) ||
3676 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3677 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3679 * This will pull the EFI from the AIL and
3680 * free the memory associated with it.
3682 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
3683 xfs_efi_release(efip, efip->efi_format.efi_nextents);
3688 tp = xfs_trans_alloc(mp, 0);
3689 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
3692 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3694 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3695 extp = &(efip->efi_format.efi_extents[i]);
3696 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3699 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3703 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
3704 error = xfs_trans_commit(tp, 0);
3708 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3713 * When this is called, all of the EFIs which did not have
3714 * corresponding EFDs should be in the AIL. What we do now
3715 * is free the extents associated with each one.
3717 * Since we process the EFIs in normal transactions, they
3718 * will be removed at some point after the commit. This prevents
3719 * us from just walking down the list processing each one.
3720 * We'll use a flag in the EFI to skip those that we've already
3721 * processed and use the AIL iteration mechanism's generation
3722 * count to try to speed this up at least a bit.
3724 * When we start, we know that the EFIs are the only things in
3725 * the AIL. As we process them, however, other items are added
3726 * to the AIL. Since everything added to the AIL must come after
3727 * everything already in the AIL, we stop processing as soon as
3728 * we see something other than an EFI in the AIL.
3731 xlog_recover_process_efis(
3734 xfs_log_item_t *lip;
3735 xfs_efi_log_item_t *efip;
3737 struct xfs_ail_cursor cur;
3738 struct xfs_ail *ailp;
3741 spin_lock(&ailp->xa_lock);
3742 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3743 while (lip != NULL) {
3745 * We're done when we see something other than an EFI.
3746 * There should be no EFIs left in the AIL now.
3748 if (lip->li_type != XFS_LI_EFI) {
3750 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
3751 ASSERT(lip->li_type != XFS_LI_EFI);
3757 * Skip EFIs that we've already processed.
3759 efip = (xfs_efi_log_item_t *)lip;
3760 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
3761 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3765 spin_unlock(&ailp->xa_lock);
3766 error = xlog_recover_process_efi(log->l_mp, efip);
3767 spin_lock(&ailp->xa_lock);
3770 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3773 xfs_trans_ail_cursor_done(&cur);
3774 spin_unlock(&ailp->xa_lock);
3779 * This routine performs a transaction to null out a bad inode pointer
3780 * in an agi unlinked inode hash bucket.
3783 xlog_recover_clear_agi_bucket(
3785 xfs_agnumber_t agno,
3794 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3795 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_clearagi, 0, 0);
3799 error = xfs_read_agi(mp, tp, agno, &agibp);
3803 agi = XFS_BUF_TO_AGI(agibp);
3804 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
3805 offset = offsetof(xfs_agi_t, agi_unlinked) +
3806 (sizeof(xfs_agino_t) * bucket);
3807 xfs_trans_log_buf(tp, agibp, offset,
3808 (offset + sizeof(xfs_agino_t) - 1));
3810 error = xfs_trans_commit(tp, 0);
3816 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3818 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
3823 xlog_recover_process_one_iunlink(
3824 struct xfs_mount *mp,
3825 xfs_agnumber_t agno,
3829 struct xfs_buf *ibp;
3830 struct xfs_dinode *dip;
3831 struct xfs_inode *ip;
3835 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3836 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
3841 * Get the on disk inode to find the next inode in the bucket.
3843 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
3847 ASSERT(ip->i_d.di_nlink == 0);
3848 ASSERT(ip->i_d.di_mode != 0);
3850 /* setup for the next pass */
3851 agino = be32_to_cpu(dip->di_next_unlinked);
3855 * Prevent any DMAPI event from being sent when the reference on
3856 * the inode is dropped.
3858 ip->i_d.di_dmevmask = 0;
3867 * We can't read in the inode this bucket points to, or this inode
3868 * is messed up. Just ditch this bucket of inodes. We will lose
3869 * some inodes and space, but at least we won't hang.
3871 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3872 * clear the inode pointer in the bucket.
3874 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3879 * xlog_iunlink_recover
3881 * This is called during recovery to process any inodes which
3882 * we unlinked but not freed when the system crashed. These
3883 * inodes will be on the lists in the AGI blocks. What we do
3884 * here is scan all the AGIs and fully truncate and free any
3885 * inodes found on the lists. Each inode is removed from the
3886 * lists when it has been fully truncated and is freed. The
3887 * freeing of the inode and its removal from the list must be
3891 xlog_recover_process_iunlinks(
3895 xfs_agnumber_t agno;
3906 * Prevent any DMAPI event from being sent while in this function.
3908 mp_dmevmask = mp->m_dmevmask;
3911 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3913 * Find the agi for this ag.
3915 error = xfs_read_agi(mp, NULL, agno, &agibp);
3918 * AGI is b0rked. Don't process it.
3920 * We should probably mark the filesystem as corrupt
3921 * after we've recovered all the ag's we can....
3926 * Unlock the buffer so that it can be acquired in the normal
3927 * course of the transaction to truncate and free each inode.
3928 * Because we are not racing with anyone else here for the AGI
3929 * buffer, we don't even need to hold it locked to read the
3930 * initial unlinked bucket entries out of the buffer. We keep
3931 * buffer reference though, so that it stays pinned in memory
3932 * while we need the buffer.
3934 agi = XFS_BUF_TO_AGI(agibp);
3935 xfs_buf_unlock(agibp);
3937 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3938 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3939 while (agino != NULLAGINO) {
3940 agino = xlog_recover_process_one_iunlink(mp,
3941 agno, agino, bucket);
3944 xfs_buf_rele(agibp);
3947 mp->m_dmevmask = mp_dmevmask;
3951 * Upack the log buffer data and crc check it. If the check fails, issue a
3952 * warning if and only if the CRC in the header is non-zero. This makes the
3953 * check an advisory warning, and the zero CRC check will prevent failure
3954 * warnings from being emitted when upgrading the kernel from one that does not
3955 * add CRCs by default.
3957 * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log
3958 * corruption failure
3961 xlog_unpack_data_crc(
3962 struct xlog_rec_header *rhead,
3968 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
3969 if (crc != rhead->h_crc) {
3970 if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
3971 xfs_alert(log->l_mp,
3972 "log record CRC mismatch: found 0x%x, expected 0x%x.",
3973 le32_to_cpu(rhead->h_crc),
3975 xfs_hex_dump(dp, 32);
3979 * If we've detected a log record corruption, then we can't
3980 * recover past this point. Abort recovery if we are enforcing
3981 * CRC protection by punting an error back up the stack.
3983 if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
3984 return -EFSCORRUPTED;
3992 struct xlog_rec_header *rhead,
3999 error = xlog_unpack_data_crc(rhead, dp, log);
4003 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
4004 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
4005 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
4009 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
4010 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
4011 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
4012 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
4013 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
4014 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
4023 xlog_valid_rec_header(
4025 struct xlog_rec_header *rhead,
4030 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
4031 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
4032 XFS_ERRLEVEL_LOW, log->l_mp);
4033 return -EFSCORRUPTED;
4036 (!rhead->h_version ||
4037 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
4038 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
4039 __func__, be32_to_cpu(rhead->h_version));
4043 /* LR body must have data or it wouldn't have been written */
4044 hlen = be32_to_cpu(rhead->h_len);
4045 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
4046 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
4047 XFS_ERRLEVEL_LOW, log->l_mp);
4048 return -EFSCORRUPTED;
4050 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
4051 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
4052 XFS_ERRLEVEL_LOW, log->l_mp);
4053 return -EFSCORRUPTED;
4059 * Read the log from tail to head and process the log records found.
4060 * Handle the two cases where the tail and head are in the same cycle
4061 * and where the active portion of the log wraps around the end of
4062 * the physical log separately. The pass parameter is passed through
4063 * to the routines called to process the data and is not looked at
4067 xlog_do_recovery_pass(
4069 xfs_daddr_t head_blk,
4070 xfs_daddr_t tail_blk,
4073 xlog_rec_header_t *rhead;
4076 xfs_buf_t *hbp, *dbp;
4077 int error = 0, h_size;
4078 int bblks, split_bblks;
4079 int hblks, split_hblks, wrapped_hblks;
4080 struct hlist_head rhash[XLOG_RHASH_SIZE];
4082 ASSERT(head_blk != tail_blk);
4085 * Read the header of the tail block and get the iclog buffer size from
4086 * h_size. Use this to tell how many sectors make up the log header.
4088 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
4090 * When using variable length iclogs, read first sector of
4091 * iclog header and extract the header size from it. Get a
4092 * new hbp that is the correct size.
4094 hbp = xlog_get_bp(log, 1);
4098 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
4102 rhead = (xlog_rec_header_t *)offset;
4103 error = xlog_valid_rec_header(log, rhead, tail_blk);
4106 h_size = be32_to_cpu(rhead->h_size);
4107 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
4108 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
4109 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
4110 if (h_size % XLOG_HEADER_CYCLE_SIZE)
4113 hbp = xlog_get_bp(log, hblks);
4118 ASSERT(log->l_sectBBsize == 1);
4120 hbp = xlog_get_bp(log, 1);
4121 h_size = XLOG_BIG_RECORD_BSIZE;
4126 dbp = xlog_get_bp(log, BTOBB(h_size));
4132 memset(rhash, 0, sizeof(rhash));
4133 if (tail_blk <= head_blk) {
4134 for (blk_no = tail_blk; blk_no < head_blk; ) {
4135 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
4139 rhead = (xlog_rec_header_t *)offset;
4140 error = xlog_valid_rec_header(log, rhead, blk_no);
4144 /* blocks in data section */
4145 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4146 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
4151 error = xlog_unpack_data(rhead, offset, log);
4155 error = xlog_recover_process_data(log,
4156 rhash, rhead, offset, pass);
4159 blk_no += bblks + hblks;
4163 * Perform recovery around the end of the physical log.
4164 * When the head is not on the same cycle number as the tail,
4165 * we can't do a sequential recovery as above.
4168 while (blk_no < log->l_logBBsize) {
4170 * Check for header wrapping around physical end-of-log
4172 offset = hbp->b_addr;
4175 if (blk_no + hblks <= log->l_logBBsize) {
4176 /* Read header in one read */
4177 error = xlog_bread(log, blk_no, hblks, hbp,
4182 /* This LR is split across physical log end */
4183 if (blk_no != log->l_logBBsize) {
4184 /* some data before physical log end */
4185 ASSERT(blk_no <= INT_MAX);
4186 split_hblks = log->l_logBBsize - (int)blk_no;
4187 ASSERT(split_hblks > 0);
4188 error = xlog_bread(log, blk_no,
4196 * Note: this black magic still works with
4197 * large sector sizes (non-512) only because:
4198 * - we increased the buffer size originally
4199 * by 1 sector giving us enough extra space
4200 * for the second read;
4201 * - the log start is guaranteed to be sector
4203 * - we read the log end (LR header start)
4204 * _first_, then the log start (LR header end)
4205 * - order is important.
4207 wrapped_hblks = hblks - split_hblks;
4208 error = xlog_bread_offset(log, 0,
4210 offset + BBTOB(split_hblks));
4214 rhead = (xlog_rec_header_t *)offset;
4215 error = xlog_valid_rec_header(log, rhead,
4216 split_hblks ? blk_no : 0);
4220 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4223 /* Read in data for log record */
4224 if (blk_no + bblks <= log->l_logBBsize) {
4225 error = xlog_bread(log, blk_no, bblks, dbp,
4230 /* This log record is split across the
4231 * physical end of log */
4232 offset = dbp->b_addr;
4234 if (blk_no != log->l_logBBsize) {
4235 /* some data is before the physical
4237 ASSERT(!wrapped_hblks);
4238 ASSERT(blk_no <= INT_MAX);
4240 log->l_logBBsize - (int)blk_no;
4241 ASSERT(split_bblks > 0);
4242 error = xlog_bread(log, blk_no,
4250 * Note: this black magic still works with
4251 * large sector sizes (non-512) only because:
4252 * - we increased the buffer size originally
4253 * by 1 sector giving us enough extra space
4254 * for the second read;
4255 * - the log start is guaranteed to be sector
4257 * - we read the log end (LR header start)
4258 * _first_, then the log start (LR header end)
4259 * - order is important.
4261 error = xlog_bread_offset(log, 0,
4262 bblks - split_bblks, dbp,
4263 offset + BBTOB(split_bblks));
4268 error = xlog_unpack_data(rhead, offset, log);
4272 error = xlog_recover_process_data(log, rhash,
4273 rhead, offset, pass);
4279 ASSERT(blk_no >= log->l_logBBsize);
4280 blk_no -= log->l_logBBsize;
4282 /* read first part of physical log */
4283 while (blk_no < head_blk) {
4284 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
4288 rhead = (xlog_rec_header_t *)offset;
4289 error = xlog_valid_rec_header(log, rhead, blk_no);
4293 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4294 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
4299 error = xlog_unpack_data(rhead, offset, log);
4303 error = xlog_recover_process_data(log, rhash,
4304 rhead, offset, pass);
4307 blk_no += bblks + hblks;
4319 * Do the recovery of the log. We actually do this in two phases.
4320 * The two passes are necessary in order to implement the function
4321 * of cancelling a record written into the log. The first pass
4322 * determines those things which have been cancelled, and the
4323 * second pass replays log items normally except for those which
4324 * have been cancelled. The handling of the replay and cancellations
4325 * takes place in the log item type specific routines.
4327 * The table of items which have cancel records in the log is allocated
4328 * and freed at this level, since only here do we know when all of
4329 * the log recovery has been completed.
4332 xlog_do_log_recovery(
4334 xfs_daddr_t head_blk,
4335 xfs_daddr_t tail_blk)
4339 ASSERT(head_blk != tail_blk);
4342 * First do a pass to find all of the cancelled buf log items.
4343 * Store them in the buf_cancel_table for use in the second pass.
4345 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
4346 sizeof(struct list_head),
4348 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4349 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
4351 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4352 XLOG_RECOVER_PASS1);
4354 kmem_free(log->l_buf_cancel_table);
4355 log->l_buf_cancel_table = NULL;
4359 * Then do a second pass to actually recover the items in the log.
4360 * When it is complete free the table of buf cancel items.
4362 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4363 XLOG_RECOVER_PASS2);
4368 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4369 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
4373 kmem_free(log->l_buf_cancel_table);
4374 log->l_buf_cancel_table = NULL;
4380 * Do the actual recovery
4385 xfs_daddr_t head_blk,
4386 xfs_daddr_t tail_blk)
4393 * First replay the images in the log.
4395 error = xlog_do_log_recovery(log, head_blk, tail_blk);
4400 * If IO errors happened during recovery, bail out.
4402 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
4407 * We now update the tail_lsn since much of the recovery has completed
4408 * and there may be space available to use. If there were no extent
4409 * or iunlinks, we can free up the entire log and set the tail_lsn to
4410 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4411 * lsn of the last known good LR on disk. If there are extent frees
4412 * or iunlinks they will have some entries in the AIL; so we look at
4413 * the AIL to determine how to set the tail_lsn.
4415 xlog_assign_tail_lsn(log->l_mp);
4418 * Now that we've finished replaying all buffer and inode
4419 * updates, re-read in the superblock and reverify it.
4421 bp = xfs_getsb(log->l_mp, 0);
4423 ASSERT(!(XFS_BUF_ISWRITE(bp)));
4425 XFS_BUF_UNASYNC(bp);
4426 bp->b_ops = &xfs_sb_buf_ops;
4428 error = xfs_buf_submit_wait(bp);
4430 if (!XFS_FORCED_SHUTDOWN(log->l_mp)) {
4431 xfs_buf_ioerror_alert(bp, __func__);
4438 /* Convert superblock from on-disk format */
4439 sbp = &log->l_mp->m_sb;
4440 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
4441 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
4442 ASSERT(xfs_sb_good_version(sbp));
4445 /* We've re-read the superblock so re-initialize per-cpu counters */
4446 xfs_icsb_reinit_counters(log->l_mp);
4448 xlog_recover_check_summary(log);
4450 /* Normal transactions can now occur */
4451 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
4456 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4458 * Return error or zero.
4464 xfs_daddr_t head_blk, tail_blk;
4467 /* find the tail of the log */
4468 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
4471 if (tail_blk != head_blk) {
4472 /* There used to be a comment here:
4474 * disallow recovery on read-only mounts. note -- mount
4475 * checks for ENOSPC and turns it into an intelligent
4477 * ...but this is no longer true. Now, unless you specify
4478 * NORECOVERY (in which case this function would never be
4479 * called), we just go ahead and recover. We do this all
4480 * under the vfs layer, so we can get away with it unless
4481 * the device itself is read-only, in which case we fail.
4483 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
4488 * Version 5 superblock log feature mask validation. We know the
4489 * log is dirty so check if there are any unknown log features
4490 * in what we need to recover. If there are unknown features
4491 * (e.g. unsupported transactions, then simply reject the
4492 * attempt at recovery before touching anything.
4494 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
4495 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
4496 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
4498 "Superblock has unknown incompatible log features (0x%x) enabled.\n"
4499 "The log can not be fully and/or safely recovered by this kernel.\n"
4500 "Please recover the log on a kernel that supports the unknown features.",
4501 (log->l_mp->m_sb.sb_features_log_incompat &
4502 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
4506 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
4507 log->l_mp->m_logname ? log->l_mp->m_logname
4510 error = xlog_do_recover(log, head_blk, tail_blk);
4511 log->l_flags |= XLOG_RECOVERY_NEEDED;
4517 * In the first part of recovery we replay inodes and buffers and build
4518 * up the list of extent free items which need to be processed. Here
4519 * we process the extent free items and clean up the on disk unlinked
4520 * inode lists. This is separated from the first part of recovery so
4521 * that the root and real-time bitmap inodes can be read in from disk in
4522 * between the two stages. This is necessary so that we can free space
4523 * in the real-time portion of the file system.
4526 xlog_recover_finish(
4530 * Now we're ready to do the transactions needed for the
4531 * rest of recovery. Start with completing all the extent
4532 * free intent records and then process the unlinked inode
4533 * lists. At this point, we essentially run in normal mode
4534 * except that we're still performing recovery actions
4535 * rather than accepting new requests.
4537 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
4539 error = xlog_recover_process_efis(log);
4541 xfs_alert(log->l_mp, "Failed to recover EFIs");
4545 * Sync the log to get all the EFIs out of the AIL.
4546 * This isn't absolutely necessary, but it helps in
4547 * case the unlink transactions would have problems
4548 * pushing the EFIs out of the way.
4550 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
4552 xlog_recover_process_iunlinks(log);
4554 xlog_recover_check_summary(log);
4556 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
4557 log->l_mp->m_logname ? log->l_mp->m_logname
4559 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
4561 xfs_info(log->l_mp, "Ending clean mount");
4569 * Read all of the agf and agi counters and check that they
4570 * are consistent with the superblock counters.
4573 xlog_recover_check_summary(
4580 xfs_agnumber_t agno;
4581 __uint64_t freeblks;
4591 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
4592 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
4594 xfs_alert(mp, "%s agf read failed agno %d error %d",
4595 __func__, agno, error);
4597 agfp = XFS_BUF_TO_AGF(agfbp);
4598 freeblks += be32_to_cpu(agfp->agf_freeblks) +
4599 be32_to_cpu(agfp->agf_flcount);
4600 xfs_buf_relse(agfbp);
4603 error = xfs_read_agi(mp, NULL, agno, &agibp);
4605 xfs_alert(mp, "%s agi read failed agno %d error %d",
4606 __func__, agno, error);
4608 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
4610 itotal += be32_to_cpu(agi->agi_count);
4611 ifree += be32_to_cpu(agi->agi_freecount);
4612 xfs_buf_relse(agibp);