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"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_trans.h"
32 #include "xfs_log_priv.h"
33 #include "xfs_log_recover.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_extfree_item.h"
36 #include "xfs_trans_priv.h"
37 #include "xfs_alloc.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_quota.h"
40 #include "xfs_cksum.h"
41 #include "xfs_trace.h"
42 #include "xfs_icache.h"
43 #include "xfs_bmap_btree.h"
44 #include "xfs_error.h"
47 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
54 xlog_clear_stale_blocks(
59 xlog_recover_check_summary(
62 #define xlog_recover_check_summary(log)
65 xlog_do_recovery_pass(
66 struct xlog *, xfs_daddr_t, xfs_daddr_t, int, xfs_daddr_t *);
69 * This structure is used during recovery to record the buf log items which
70 * have been canceled and should not be replayed.
72 struct xfs_buf_cancel {
76 struct list_head bc_list;
80 * Sector aligned buffer routines for buffer create/read/write/access
84 * Verify the given count of basic blocks is valid number of blocks
85 * to specify for an operation involving the given XFS log buffer.
86 * Returns nonzero if the count is valid, 0 otherwise.
90 xlog_buf_bbcount_valid(
94 return bbcount > 0 && bbcount <= log->l_logBBsize;
98 * Allocate a buffer to hold log data. The buffer needs to be able
99 * to map to a range of nbblks basic blocks at any valid (basic
100 * block) offset within the log.
109 if (!xlog_buf_bbcount_valid(log, nbblks)) {
110 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
112 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
117 * We do log I/O in units of log sectors (a power-of-2
118 * multiple of the basic block size), so we round up the
119 * requested size to accommodate the basic blocks required
120 * for complete log sectors.
122 * In addition, the buffer may be used for a non-sector-
123 * aligned block offset, in which case an I/O of the
124 * requested size could extend beyond the end of the
125 * buffer. If the requested size is only 1 basic block it
126 * will never straddle a sector boundary, so this won't be
127 * an issue. Nor will this be a problem if the log I/O is
128 * done in basic blocks (sector size 1). But otherwise we
129 * extend the buffer by one extra log sector to ensure
130 * there's space to accommodate this possibility.
132 if (nbblks > 1 && log->l_sectBBsize > 1)
133 nbblks += log->l_sectBBsize;
134 nbblks = round_up(nbblks, log->l_sectBBsize);
136 bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0);
150 * Return the address of the start of the given block number's data
151 * in a log buffer. The buffer covers a log sector-aligned region.
160 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
162 ASSERT(offset + nbblks <= bp->b_length);
163 return bp->b_addr + BBTOB(offset);
168 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
179 if (!xlog_buf_bbcount_valid(log, nbblks)) {
180 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
182 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
183 return -EFSCORRUPTED;
186 blk_no = round_down(blk_no, log->l_sectBBsize);
187 nbblks = round_up(nbblks, log->l_sectBBsize);
190 ASSERT(nbblks <= bp->b_length);
192 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
194 bp->b_io_length = nbblks;
197 error = xfs_buf_submit_wait(bp);
198 if (error && !XFS_FORCED_SHUTDOWN(log->l_mp))
199 xfs_buf_ioerror_alert(bp, __func__);
213 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
217 *offset = xlog_align(log, blk_no, nbblks, bp);
222 * Read at an offset into the buffer. Returns with the buffer in it's original
223 * state regardless of the result of the read.
228 xfs_daddr_t blk_no, /* block to read from */
229 int nbblks, /* blocks to read */
233 char *orig_offset = bp->b_addr;
234 int orig_len = BBTOB(bp->b_length);
237 error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks));
241 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
243 /* must reset buffer pointer even on error */
244 error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len);
251 * Write out the buffer at the given block for the given number of blocks.
252 * The buffer is kept locked across the write and is returned locked.
253 * This can only be used for synchronous log writes.
264 if (!xlog_buf_bbcount_valid(log, nbblks)) {
265 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
267 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
268 return -EFSCORRUPTED;
271 blk_no = round_down(blk_no, log->l_sectBBsize);
272 nbblks = round_up(nbblks, log->l_sectBBsize);
275 ASSERT(nbblks <= bp->b_length);
277 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
278 XFS_BUF_ZEROFLAGS(bp);
281 bp->b_io_length = nbblks;
284 error = xfs_bwrite(bp);
286 xfs_buf_ioerror_alert(bp, __func__);
293 * dump debug superblock and log record information
296 xlog_header_check_dump(
298 xlog_rec_header_t *head)
300 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d",
301 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
302 xfs_debug(mp, " log : uuid = %pU, fmt = %d",
303 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
306 #define xlog_header_check_dump(mp, head)
310 * check log record header for recovery
313 xlog_header_check_recover(
315 xlog_rec_header_t *head)
317 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
320 * IRIX doesn't write the h_fmt field and leaves it zeroed
321 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
322 * a dirty log created in IRIX.
324 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
326 "dirty log written in incompatible format - can't recover");
327 xlog_header_check_dump(mp, head);
328 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
329 XFS_ERRLEVEL_HIGH, mp);
330 return -EFSCORRUPTED;
331 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
333 "dirty log entry has mismatched uuid - can't recover");
334 xlog_header_check_dump(mp, head);
335 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
336 XFS_ERRLEVEL_HIGH, mp);
337 return -EFSCORRUPTED;
343 * read the head block of the log and check the header
346 xlog_header_check_mount(
348 xlog_rec_header_t *head)
350 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
352 if (uuid_is_nil(&head->h_fs_uuid)) {
354 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
355 * h_fs_uuid is nil, we assume this log was last mounted
356 * by IRIX and continue.
358 xfs_warn(mp, "nil uuid in log - IRIX style log");
359 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
360 xfs_warn(mp, "log has mismatched uuid - can't recover");
361 xlog_header_check_dump(mp, head);
362 XFS_ERROR_REPORT("xlog_header_check_mount",
363 XFS_ERRLEVEL_HIGH, mp);
364 return -EFSCORRUPTED;
375 * We're not going to bother about retrying
376 * this during recovery. One strike!
378 if (!XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
379 xfs_buf_ioerror_alert(bp, __func__);
380 xfs_force_shutdown(bp->b_target->bt_mount,
381 SHUTDOWN_META_IO_ERROR);
389 * This routine finds (to an approximation) the first block in the physical
390 * log which contains the given cycle. It uses a binary search algorithm.
391 * Note that the algorithm can not be perfect because the disk will not
392 * necessarily be perfect.
395 xlog_find_cycle_start(
398 xfs_daddr_t first_blk,
399 xfs_daddr_t *last_blk,
409 mid_blk = BLK_AVG(first_blk, end_blk);
410 while (mid_blk != first_blk && mid_blk != end_blk) {
411 error = xlog_bread(log, mid_blk, 1, bp, &offset);
414 mid_cycle = xlog_get_cycle(offset);
415 if (mid_cycle == cycle)
416 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
418 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
419 mid_blk = BLK_AVG(first_blk, end_blk);
421 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
422 (mid_blk == end_blk && mid_blk-1 == first_blk));
430 * Check that a range of blocks does not contain stop_on_cycle_no.
431 * Fill in *new_blk with the block offset where such a block is
432 * found, or with -1 (an invalid block number) if there is no such
433 * block in the range. The scan needs to occur from front to back
434 * and the pointer into the region must be updated since a later
435 * routine will need to perform another test.
438 xlog_find_verify_cycle(
440 xfs_daddr_t start_blk,
442 uint stop_on_cycle_no,
443 xfs_daddr_t *new_blk)
453 * Greedily allocate a buffer big enough to handle the full
454 * range of basic blocks we'll be examining. If that fails,
455 * try a smaller size. We need to be able to read at least
456 * a log sector, or we're out of luck.
458 bufblks = 1 << ffs(nbblks);
459 while (bufblks > log->l_logBBsize)
461 while (!(bp = xlog_get_bp(log, bufblks))) {
463 if (bufblks < log->l_sectBBsize)
467 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
470 bcount = min(bufblks, (start_blk + nbblks - i));
472 error = xlog_bread(log, i, bcount, bp, &buf);
476 for (j = 0; j < bcount; j++) {
477 cycle = xlog_get_cycle(buf);
478 if (cycle == stop_on_cycle_no) {
495 * Potentially backup over partial log record write.
497 * In the typical case, last_blk is the number of the block directly after
498 * a good log record. Therefore, we subtract one to get the block number
499 * of the last block in the given buffer. extra_bblks contains the number
500 * of blocks we would have read on a previous read. This happens when the
501 * last log record is split over the end of the physical log.
503 * extra_bblks is the number of blocks potentially verified on a previous
504 * call to this routine.
507 xlog_find_verify_log_record(
509 xfs_daddr_t start_blk,
510 xfs_daddr_t *last_blk,
516 xlog_rec_header_t *head = NULL;
519 int num_blks = *last_blk - start_blk;
522 ASSERT(start_blk != 0 || *last_blk != start_blk);
524 if (!(bp = xlog_get_bp(log, num_blks))) {
525 if (!(bp = xlog_get_bp(log, 1)))
529 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
532 offset += ((num_blks - 1) << BBSHIFT);
535 for (i = (*last_blk) - 1; i >= 0; i--) {
537 /* valid log record not found */
539 "Log inconsistent (didn't find previous header)");
546 error = xlog_bread(log, i, 1, bp, &offset);
551 head = (xlog_rec_header_t *)offset;
553 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
561 * We hit the beginning of the physical log & still no header. Return
562 * to caller. If caller can handle a return of -1, then this routine
563 * will be called again for the end of the physical log.
571 * We have the final block of the good log (the first block
572 * of the log record _before_ the head. So we check the uuid.
574 if ((error = xlog_header_check_mount(log->l_mp, head)))
578 * We may have found a log record header before we expected one.
579 * last_blk will be the 1st block # with a given cycle #. We may end
580 * up reading an entire log record. In this case, we don't want to
581 * reset last_blk. Only when last_blk points in the middle of a log
582 * record do we update last_blk.
584 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
585 uint h_size = be32_to_cpu(head->h_size);
587 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
588 if (h_size % XLOG_HEADER_CYCLE_SIZE)
594 if (*last_blk - i + extra_bblks !=
595 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
604 * Head is defined to be the point of the log where the next log write
605 * could go. This means that incomplete LR writes at the end are
606 * eliminated when calculating the head. We aren't guaranteed that previous
607 * LR have complete transactions. We only know that a cycle number of
608 * current cycle number -1 won't be present in the log if we start writing
609 * from our current block number.
611 * last_blk contains the block number of the first block with a given
614 * Return: zero if normal, non-zero if error.
619 xfs_daddr_t *return_head_blk)
623 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
625 uint first_half_cycle, last_half_cycle;
627 int error, log_bbnum = log->l_logBBsize;
629 /* Is the end of the log device zeroed? */
630 error = xlog_find_zeroed(log, &first_blk);
632 xfs_warn(log->l_mp, "empty log check failed");
636 *return_head_blk = first_blk;
638 /* Is the whole lot zeroed? */
640 /* Linux XFS shouldn't generate totally zeroed logs -
641 * mkfs etc write a dummy unmount record to a fresh
642 * log so we can store the uuid in there
644 xfs_warn(log->l_mp, "totally zeroed log");
650 first_blk = 0; /* get cycle # of 1st block */
651 bp = xlog_get_bp(log, 1);
655 error = xlog_bread(log, 0, 1, bp, &offset);
659 first_half_cycle = xlog_get_cycle(offset);
661 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
662 error = xlog_bread(log, last_blk, 1, bp, &offset);
666 last_half_cycle = xlog_get_cycle(offset);
667 ASSERT(last_half_cycle != 0);
670 * If the 1st half cycle number is equal to the last half cycle number,
671 * then the entire log is stamped with the same cycle number. In this
672 * case, head_blk can't be set to zero (which makes sense). The below
673 * math doesn't work out properly with head_blk equal to zero. Instead,
674 * we set it to log_bbnum which is an invalid block number, but this
675 * value makes the math correct. If head_blk doesn't changed through
676 * all the tests below, *head_blk is set to zero at the very end rather
677 * than log_bbnum. In a sense, log_bbnum and zero are the same block
678 * in a circular file.
680 if (first_half_cycle == last_half_cycle) {
682 * In this case we believe that the entire log should have
683 * cycle number last_half_cycle. We need to scan backwards
684 * from the end verifying that there are no holes still
685 * containing last_half_cycle - 1. If we find such a hole,
686 * then the start of that hole will be the new head. The
687 * simple case looks like
688 * x | x ... | x - 1 | x
689 * Another case that fits this picture would be
690 * x | x + 1 | x ... | x
691 * In this case the head really is somewhere at the end of the
692 * log, as one of the latest writes at the beginning was
695 * x | x + 1 | x ... | x - 1 | x
696 * This is really the combination of the above two cases, and
697 * the head has to end up at the start of the x-1 hole at the
700 * In the 256k log case, we will read from the beginning to the
701 * end of the log and search for cycle numbers equal to x-1.
702 * We don't worry about the x+1 blocks that we encounter,
703 * because we know that they cannot be the head since the log
706 head_blk = log_bbnum;
707 stop_on_cycle = last_half_cycle - 1;
710 * In this case we want to find the first block with cycle
711 * number matching last_half_cycle. We expect the log to be
713 * x + 1 ... | x ... | x
714 * The first block with cycle number x (last_half_cycle) will
715 * be where the new head belongs. First we do a binary search
716 * for the first occurrence of last_half_cycle. The binary
717 * search may not be totally accurate, so then we scan back
718 * from there looking for occurrences of last_half_cycle before
719 * us. If that backwards scan wraps around the beginning of
720 * the log, then we look for occurrences of last_half_cycle - 1
721 * at the end of the log. The cases we're looking for look
723 * v binary search stopped here
724 * x + 1 ... | x | x + 1 | x ... | x
725 * ^ but we want to locate this spot
727 * <---------> less than scan distance
728 * x + 1 ... | x ... | x - 1 | x
729 * ^ we want to locate this spot
731 stop_on_cycle = last_half_cycle;
732 if ((error = xlog_find_cycle_start(log, bp, first_blk,
733 &head_blk, last_half_cycle)))
738 * Now validate the answer. Scan back some number of maximum possible
739 * blocks and make sure each one has the expected cycle number. The
740 * maximum is determined by the total possible amount of buffering
741 * in the in-core log. The following number can be made tighter if
742 * we actually look at the block size of the filesystem.
744 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
745 if (head_blk >= num_scan_bblks) {
747 * We are guaranteed that the entire check can be performed
750 start_blk = head_blk - num_scan_bblks;
751 if ((error = xlog_find_verify_cycle(log,
752 start_blk, num_scan_bblks,
753 stop_on_cycle, &new_blk)))
757 } else { /* need to read 2 parts of log */
759 * We are going to scan backwards in the log in two parts.
760 * First we scan the physical end of the log. In this part
761 * of the log, we are looking for blocks with cycle number
762 * last_half_cycle - 1.
763 * If we find one, then we know that the log starts there, as
764 * we've found a hole that didn't get written in going around
765 * the end of the physical log. The simple case for this is
766 * x + 1 ... | x ... | x - 1 | x
767 * <---------> less than scan distance
768 * If all of the blocks at the end of the log have cycle number
769 * last_half_cycle, then we check the blocks at the start of
770 * the log looking for occurrences of last_half_cycle. If we
771 * find one, then our current estimate for the location of the
772 * first occurrence of last_half_cycle is wrong and we move
773 * back to the hole we've found. This case looks like
774 * x + 1 ... | x | x + 1 | x ...
775 * ^ binary search stopped here
776 * Another case we need to handle that only occurs in 256k
778 * x + 1 ... | x ... | x+1 | x ...
779 * ^ binary search stops here
780 * In a 256k log, the scan at the end of the log will see the
781 * x + 1 blocks. We need to skip past those since that is
782 * certainly not the head of the log. By searching for
783 * last_half_cycle-1 we accomplish that.
785 ASSERT(head_blk <= INT_MAX &&
786 (xfs_daddr_t) num_scan_bblks >= head_blk);
787 start_blk = log_bbnum - (num_scan_bblks - head_blk);
788 if ((error = xlog_find_verify_cycle(log, start_blk,
789 num_scan_bblks - (int)head_blk,
790 (stop_on_cycle - 1), &new_blk)))
798 * Scan beginning of log now. The last part of the physical
799 * log is good. This scan needs to verify that it doesn't find
800 * the last_half_cycle.
803 ASSERT(head_blk <= INT_MAX);
804 if ((error = xlog_find_verify_cycle(log,
805 start_blk, (int)head_blk,
806 stop_on_cycle, &new_blk)))
814 * Now we need to make sure head_blk is not pointing to a block in
815 * the middle of a log record.
817 num_scan_bblks = XLOG_REC_SHIFT(log);
818 if (head_blk >= num_scan_bblks) {
819 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
821 /* start ptr at last block ptr before head_blk */
822 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
829 ASSERT(head_blk <= INT_MAX);
830 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
834 /* We hit the beginning of the log during our search */
835 start_blk = log_bbnum - (num_scan_bblks - head_blk);
837 ASSERT(start_blk <= INT_MAX &&
838 (xfs_daddr_t) log_bbnum-start_blk >= 0);
839 ASSERT(head_blk <= INT_MAX);
840 error = xlog_find_verify_log_record(log, start_blk,
841 &new_blk, (int)head_blk);
846 if (new_blk != log_bbnum)
853 if (head_blk == log_bbnum)
854 *return_head_blk = 0;
856 *return_head_blk = head_blk;
858 * When returning here, we have a good block number. Bad block
859 * means that during a previous crash, we didn't have a clean break
860 * from cycle number N to cycle number N-1. In this case, we need
861 * to find the first block with cycle number N-1.
869 xfs_warn(log->l_mp, "failed to find log head");
874 * Seek backwards in the log for log record headers.
876 * Given a starting log block, walk backwards until we find the provided number
877 * of records or hit the provided tail block. The return value is the number of
878 * records encountered or a negative error code. The log block and buffer
879 * pointer of the last record seen are returned in rblk and rhead respectively.
882 xlog_rseek_logrec_hdr(
884 xfs_daddr_t head_blk,
885 xfs_daddr_t tail_blk,
889 struct xlog_rec_header **rhead,
901 * Walk backwards from the head block until we hit the tail or the first
904 end_blk = head_blk > tail_blk ? tail_blk : 0;
905 for (i = (int) head_blk - 1; i >= end_blk; i--) {
906 error = xlog_bread(log, i, 1, bp, &offset);
910 if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
912 *rhead = (struct xlog_rec_header *) offset;
913 if (++found == count)
919 * If we haven't hit the tail block or the log record header count,
920 * start looking again from the end of the physical log. Note that
921 * callers can pass head == tail if the tail is not yet known.
923 if (tail_blk >= head_blk && found != count) {
924 for (i = log->l_logBBsize - 1; i >= (int) tail_blk; i--) {
925 error = xlog_bread(log, i, 1, bp, &offset);
929 if (*(__be32 *)offset ==
930 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
933 *rhead = (struct xlog_rec_header *) offset;
934 if (++found == count)
947 * Seek forward in the log for log record headers.
949 * Given head and tail blocks, walk forward from the tail block until we find
950 * the provided number of records or hit the head block. The return value is the
951 * number of records encountered or a negative error code. The log block and
952 * buffer pointer of the last record seen are returned in rblk and rhead
956 xlog_seek_logrec_hdr(
958 xfs_daddr_t head_blk,
959 xfs_daddr_t tail_blk,
963 struct xlog_rec_header **rhead,
975 * Walk forward from the tail block until we hit the head or the last
978 end_blk = head_blk > tail_blk ? head_blk : log->l_logBBsize - 1;
979 for (i = (int) tail_blk; i <= end_blk; i++) {
980 error = xlog_bread(log, i, 1, bp, &offset);
984 if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
986 *rhead = (struct xlog_rec_header *) offset;
987 if (++found == count)
993 * If we haven't hit the head block or the log record header count,
994 * start looking again from the start of the physical log.
996 if (tail_blk > head_blk && found != count) {
997 for (i = 0; i < (int) head_blk; i++) {
998 error = xlog_bread(log, i, 1, bp, &offset);
1002 if (*(__be32 *)offset ==
1003 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
1006 *rhead = (struct xlog_rec_header *) offset;
1007 if (++found == count)
1020 * Check the log tail for torn writes. This is required when torn writes are
1021 * detected at the head and the head had to be walked back to a previous record.
1022 * The tail of the previous record must now be verified to ensure the torn
1023 * writes didn't corrupt the previous tail.
1025 * Return an error if CRC verification fails as recovery cannot proceed.
1030 xfs_daddr_t head_blk,
1031 xfs_daddr_t tail_blk)
1033 struct xlog_rec_header *thead;
1035 xfs_daddr_t first_bad;
1039 xfs_daddr_t tmp_head;
1041 bp = xlog_get_bp(log, 1);
1046 * Seek XLOG_MAX_ICLOGS + 1 records past the current tail record to get
1047 * a temporary head block that points after the last possible
1048 * concurrently written record of the tail.
1050 count = xlog_seek_logrec_hdr(log, head_blk, tail_blk,
1051 XLOG_MAX_ICLOGS + 1, bp, &tmp_head, &thead,
1059 * If the call above didn't find XLOG_MAX_ICLOGS + 1 records, we ran
1060 * into the actual log head. tmp_head points to the start of the record
1061 * so update it to the actual head block.
1063 if (count < XLOG_MAX_ICLOGS + 1)
1064 tmp_head = head_blk;
1067 * We now have a tail and temporary head block that covers at least
1068 * XLOG_MAX_ICLOGS records from the tail. We need to verify that these
1069 * records were completely written. Run a CRC verification pass from
1070 * tail to head and return the result.
1072 error = xlog_do_recovery_pass(log, tmp_head, tail_blk,
1073 XLOG_RECOVER_CRCPASS, &first_bad);
1081 * Detect and trim torn writes from the head of the log.
1083 * Storage without sector atomicity guarantees can result in torn writes in the
1084 * log in the event of a crash. Our only means to detect this scenario is via
1085 * CRC verification. While we can't always be certain that CRC verification
1086 * failure is due to a torn write vs. an unrelated corruption, we do know that
1087 * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at
1088 * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of
1089 * the log and treat failures in this range as torn writes as a matter of
1090 * policy. In the event of CRC failure, the head is walked back to the last good
1091 * record in the log and the tail is updated from that record and verified.
1096 xfs_daddr_t *head_blk, /* in/out: unverified head */
1097 xfs_daddr_t *tail_blk, /* out: tail block */
1099 xfs_daddr_t *rhead_blk, /* start blk of last record */
1100 struct xlog_rec_header **rhead, /* ptr to last record */
1101 bool *wrapped) /* last rec. wraps phys. log */
1103 struct xlog_rec_header *tmp_rhead;
1104 struct xfs_buf *tmp_bp;
1105 xfs_daddr_t first_bad;
1106 xfs_daddr_t tmp_rhead_blk;
1112 * Check the head of the log for torn writes. Search backwards from the
1113 * head until we hit the tail or the maximum number of log record I/Os
1114 * that could have been in flight at one time. Use a temporary buffer so
1115 * we don't trash the rhead/bp pointers from the caller.
1117 tmp_bp = xlog_get_bp(log, 1);
1120 error = xlog_rseek_logrec_hdr(log, *head_blk, *tail_blk,
1121 XLOG_MAX_ICLOGS, tmp_bp, &tmp_rhead_blk,
1122 &tmp_rhead, &tmp_wrapped);
1123 xlog_put_bp(tmp_bp);
1128 * Now run a CRC verification pass over the records starting at the
1129 * block found above to the current head. If a CRC failure occurs, the
1130 * log block of the first bad record is saved in first_bad.
1132 error = xlog_do_recovery_pass(log, *head_blk, tmp_rhead_blk,
1133 XLOG_RECOVER_CRCPASS, &first_bad);
1134 if (error == -EFSBADCRC) {
1136 * We've hit a potential torn write. Reset the error and warn
1141 "Torn write (CRC failure) detected at log block 0x%llx. Truncating head block from 0x%llx.",
1142 first_bad, *head_blk);
1145 * Get the header block and buffer pointer for the last good
1146 * record before the bad record.
1148 * Note that xlog_find_tail() clears the blocks at the new head
1149 * (i.e., the records with invalid CRC) if the cycle number
1150 * matches the the current cycle.
1152 found = xlog_rseek_logrec_hdr(log, first_bad, *tail_blk, 1, bp,
1153 rhead_blk, rhead, wrapped);
1156 if (found == 0) /* XXX: right thing to do here? */
1160 * Reset the head block to the starting block of the first bad
1161 * log record and set the tail block based on the last good
1164 * Bail out if the updated head/tail match as this indicates
1165 * possible corruption outside of the acceptable
1166 * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair...
1168 *head_blk = first_bad;
1169 *tail_blk = BLOCK_LSN(be64_to_cpu((*rhead)->h_tail_lsn));
1170 if (*head_blk == *tail_blk) {
1176 * Now verify the tail based on the updated head. This is
1177 * required because the torn writes trimmed from the head could
1178 * have been written over the tail of a previous record. Return
1179 * any errors since recovery cannot proceed if the tail is
1182 * XXX: This leaves a gap in truly robust protection from torn
1183 * writes in the log. If the head is behind the tail, the tail
1184 * pushes forward to create some space and then a crash occurs
1185 * causing the writes into the previous record's tail region to
1186 * tear, log recovery isn't able to recover.
1188 * How likely is this to occur? If possible, can we do something
1189 * more intelligent here? Is it safe to push the tail forward if
1190 * we can determine that the tail is within the range of the
1191 * torn write (e.g., the kernel can only overwrite the tail if
1192 * it has actually been pushed forward)? Alternatively, could we
1193 * somehow prevent this condition at runtime?
1195 error = xlog_verify_tail(log, *head_blk, *tail_blk);
1202 * Check whether the head of the log points to an unmount record. In other
1203 * words, determine whether the log is clean. If so, update the in-core state
1207 xlog_check_unmount_rec(
1209 xfs_daddr_t *head_blk,
1210 xfs_daddr_t *tail_blk,
1211 struct xlog_rec_header *rhead,
1212 xfs_daddr_t rhead_blk,
1216 struct xlog_op_header *op_head;
1217 xfs_daddr_t umount_data_blk;
1218 xfs_daddr_t after_umount_blk;
1226 * Look for unmount record. If we find it, then we know there was a
1227 * clean unmount. Since 'i' could be the last block in the physical
1228 * log, we convert to a log block before comparing to the head_blk.
1230 * Save the current tail lsn to use to pass to xlog_clear_stale_blocks()
1231 * below. We won't want to clear the unmount record if there is one, so
1232 * we pass the lsn of the unmount record rather than the block after it.
1234 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1235 int h_size = be32_to_cpu(rhead->h_size);
1236 int h_version = be32_to_cpu(rhead->h_version);
1238 if ((h_version & XLOG_VERSION_2) &&
1239 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
1240 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
1241 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1249 after_umount_blk = rhead_blk + hblks + BTOBB(be32_to_cpu(rhead->h_len));
1250 after_umount_blk = do_mod(after_umount_blk, log->l_logBBsize);
1251 if (*head_blk == after_umount_blk &&
1252 be32_to_cpu(rhead->h_num_logops) == 1) {
1253 umount_data_blk = rhead_blk + hblks;
1254 umount_data_blk = do_mod(umount_data_blk, log->l_logBBsize);
1255 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1259 op_head = (struct xlog_op_header *)offset;
1260 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1262 * Set tail and last sync so that newly written log
1263 * records will point recovery to after the current
1266 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1267 log->l_curr_cycle, after_umount_blk);
1268 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1269 log->l_curr_cycle, after_umount_blk);
1270 *tail_blk = after_umount_blk;
1280 * Find the sync block number or the tail of the log.
1282 * This will be the block number of the last record to have its
1283 * associated buffers synced to disk. Every log record header has
1284 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
1285 * to get a sync block number. The only concern is to figure out which
1286 * log record header to believe.
1288 * The following algorithm uses the log record header with the largest
1289 * lsn. The entire log record does not need to be valid. We only care
1290 * that the header is valid.
1292 * We could speed up search by using current head_blk buffer, but it is not
1298 xfs_daddr_t *head_blk,
1299 xfs_daddr_t *tail_blk)
1301 xlog_rec_header_t *rhead;
1302 char *offset = NULL;
1305 xfs_daddr_t rhead_blk;
1307 bool wrapped = false;
1311 * Find previous log record
1313 if ((error = xlog_find_head(log, head_blk)))
1315 ASSERT(*head_blk < INT_MAX);
1317 bp = xlog_get_bp(log, 1);
1320 if (*head_blk == 0) { /* special case */
1321 error = xlog_bread(log, 0, 1, bp, &offset);
1325 if (xlog_get_cycle(offset) == 0) {
1327 /* leave all other log inited values alone */
1333 * Search backwards through the log looking for the log record header
1334 * block. This wraps all the way back around to the head so something is
1335 * seriously wrong if we can't find it.
1337 error = xlog_rseek_logrec_hdr(log, *head_blk, *head_blk, 1, bp,
1338 &rhead_blk, &rhead, &wrapped);
1342 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
1345 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
1348 * Trim the head block back to skip over torn records. We can have
1349 * multiple log I/Os in flight at any time, so we assume CRC failures
1350 * back through the previous several records are torn writes and skip
1353 error = xlog_verify_head(log, head_blk, tail_blk, bp, &rhead_blk,
1359 * Reset log values according to the state of the log when we
1360 * crashed. In the case where head_blk == 0, we bump curr_cycle
1361 * one because the next write starts a new cycle rather than
1362 * continuing the cycle of the last good log record. At this
1363 * point we have guaranteed that all partial log records have been
1364 * accounted for. Therefore, we know that the last good log record
1365 * written was complete and ended exactly on the end boundary
1366 * of the physical log.
1368 log->l_prev_block = rhead_blk;
1369 log->l_curr_block = (int)*head_blk;
1370 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
1372 log->l_curr_cycle++;
1373 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
1374 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
1375 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
1376 BBTOB(log->l_curr_block));
1377 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
1378 BBTOB(log->l_curr_block));
1379 tail_lsn = atomic64_read(&log->l_tail_lsn);
1382 * Look for an unmount record at the head of the log. This sets the log
1383 * state to determine whether recovery is necessary.
1385 error = xlog_check_unmount_rec(log, head_blk, tail_blk, rhead,
1386 rhead_blk, bp, &clean);
1391 * Note that the unmount was clean. If the unmount was not clean, we
1392 * need to know this to rebuild the superblock counters from the perag
1393 * headers if we have a filesystem using non-persistent counters.
1396 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1399 * Make sure that there are no blocks in front of the head
1400 * with the same cycle number as the head. This can happen
1401 * because we allow multiple outstanding log writes concurrently,
1402 * and the later writes might make it out before earlier ones.
1404 * We use the lsn from before modifying it so that we'll never
1405 * overwrite the unmount record after a clean unmount.
1407 * Do this only if we are going to recover the filesystem
1409 * NOTE: This used to say "if (!readonly)"
1410 * However on Linux, we can & do recover a read-only filesystem.
1411 * We only skip recovery if NORECOVERY is specified on mount,
1412 * in which case we would not be here.
1414 * But... if the -device- itself is readonly, just skip this.
1415 * We can't recover this device anyway, so it won't matter.
1417 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
1418 error = xlog_clear_stale_blocks(log, tail_lsn);
1424 xfs_warn(log->l_mp, "failed to locate log tail");
1429 * Is the log zeroed at all?
1431 * The last binary search should be changed to perform an X block read
1432 * once X becomes small enough. You can then search linearly through
1433 * the X blocks. This will cut down on the number of reads we need to do.
1435 * If the log is partially zeroed, this routine will pass back the blkno
1436 * of the first block with cycle number 0. It won't have a complete LR
1440 * 0 => the log is completely written to
1441 * 1 => use *blk_no as the first block of the log
1442 * <0 => error has occurred
1447 xfs_daddr_t *blk_no)
1451 uint first_cycle, last_cycle;
1452 xfs_daddr_t new_blk, last_blk, start_blk;
1453 xfs_daddr_t num_scan_bblks;
1454 int error, log_bbnum = log->l_logBBsize;
1458 /* check totally zeroed log */
1459 bp = xlog_get_bp(log, 1);
1462 error = xlog_bread(log, 0, 1, bp, &offset);
1466 first_cycle = xlog_get_cycle(offset);
1467 if (first_cycle == 0) { /* completely zeroed log */
1473 /* check partially zeroed log */
1474 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1478 last_cycle = xlog_get_cycle(offset);
1479 if (last_cycle != 0) { /* log completely written to */
1482 } else if (first_cycle != 1) {
1484 * If the cycle of the last block is zero, the cycle of
1485 * the first block must be 1. If it's not, maybe we're
1486 * not looking at a log... Bail out.
1489 "Log inconsistent or not a log (last==0, first!=1)");
1494 /* we have a partially zeroed log */
1495 last_blk = log_bbnum-1;
1496 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1500 * Validate the answer. Because there is no way to guarantee that
1501 * the entire log is made up of log records which are the same size,
1502 * we scan over the defined maximum blocks. At this point, the maximum
1503 * is not chosen to mean anything special. XXXmiken
1505 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1506 ASSERT(num_scan_bblks <= INT_MAX);
1508 if (last_blk < num_scan_bblks)
1509 num_scan_bblks = last_blk;
1510 start_blk = last_blk - num_scan_bblks;
1513 * We search for any instances of cycle number 0 that occur before
1514 * our current estimate of the head. What we're trying to detect is
1515 * 1 ... | 0 | 1 | 0...
1516 * ^ binary search ends here
1518 if ((error = xlog_find_verify_cycle(log, start_blk,
1519 (int)num_scan_bblks, 0, &new_blk)))
1525 * Potentially backup over partial log record write. We don't need
1526 * to search the end of the log because we know it is zero.
1528 error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0);
1543 * These are simple subroutines used by xlog_clear_stale_blocks() below
1544 * to initialize a buffer full of empty log record headers and write
1545 * them into the log.
1556 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1558 memset(buf, 0, BBSIZE);
1559 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1560 recp->h_cycle = cpu_to_be32(cycle);
1561 recp->h_version = cpu_to_be32(
1562 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1563 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1564 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1565 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1566 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1570 xlog_write_log_records(
1581 int sectbb = log->l_sectBBsize;
1582 int end_block = start_block + blocks;
1588 * Greedily allocate a buffer big enough to handle the full
1589 * range of basic blocks to be written. If that fails, try
1590 * a smaller size. We need to be able to write at least a
1591 * log sector, or we're out of luck.
1593 bufblks = 1 << ffs(blocks);
1594 while (bufblks > log->l_logBBsize)
1596 while (!(bp = xlog_get_bp(log, bufblks))) {
1598 if (bufblks < sectbb)
1602 /* We may need to do a read at the start to fill in part of
1603 * the buffer in the starting sector not covered by the first
1606 balign = round_down(start_block, sectbb);
1607 if (balign != start_block) {
1608 error = xlog_bread_noalign(log, start_block, 1, bp);
1612 j = start_block - balign;
1615 for (i = start_block; i < end_block; i += bufblks) {
1616 int bcount, endcount;
1618 bcount = min(bufblks, end_block - start_block);
1619 endcount = bcount - j;
1621 /* We may need to do a read at the end to fill in part of
1622 * the buffer in the final sector not covered by the write.
1623 * If this is the same sector as the above read, skip it.
1625 ealign = round_down(end_block, sectbb);
1626 if (j == 0 && (start_block + endcount > ealign)) {
1627 offset = bp->b_addr + BBTOB(ealign - start_block);
1628 error = xlog_bread_offset(log, ealign, sectbb,
1635 offset = xlog_align(log, start_block, endcount, bp);
1636 for (; j < endcount; j++) {
1637 xlog_add_record(log, offset, cycle, i+j,
1638 tail_cycle, tail_block);
1641 error = xlog_bwrite(log, start_block, endcount, bp);
1644 start_block += endcount;
1654 * This routine is called to blow away any incomplete log writes out
1655 * in front of the log head. We do this so that we won't become confused
1656 * if we come up, write only a little bit more, and then crash again.
1657 * If we leave the partial log records out there, this situation could
1658 * cause us to think those partial writes are valid blocks since they
1659 * have the current cycle number. We get rid of them by overwriting them
1660 * with empty log records with the old cycle number rather than the
1663 * The tail lsn is passed in rather than taken from
1664 * the log so that we will not write over the unmount record after a
1665 * clean unmount in a 512 block log. Doing so would leave the log without
1666 * any valid log records in it until a new one was written. If we crashed
1667 * during that time we would not be able to recover.
1670 xlog_clear_stale_blocks(
1674 int tail_cycle, head_cycle;
1675 int tail_block, head_block;
1676 int tail_distance, max_distance;
1680 tail_cycle = CYCLE_LSN(tail_lsn);
1681 tail_block = BLOCK_LSN(tail_lsn);
1682 head_cycle = log->l_curr_cycle;
1683 head_block = log->l_curr_block;
1686 * Figure out the distance between the new head of the log
1687 * and the tail. We want to write over any blocks beyond the
1688 * head that we may have written just before the crash, but
1689 * we don't want to overwrite the tail of the log.
1691 if (head_cycle == tail_cycle) {
1693 * The tail is behind the head in the physical log,
1694 * so the distance from the head to the tail is the
1695 * distance from the head to the end of the log plus
1696 * the distance from the beginning of the log to the
1699 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1700 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1701 XFS_ERRLEVEL_LOW, log->l_mp);
1702 return -EFSCORRUPTED;
1704 tail_distance = tail_block + (log->l_logBBsize - head_block);
1707 * The head is behind the tail in the physical log,
1708 * so the distance from the head to the tail is just
1709 * the tail block minus the head block.
1711 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1712 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1713 XFS_ERRLEVEL_LOW, log->l_mp);
1714 return -EFSCORRUPTED;
1716 tail_distance = tail_block - head_block;
1720 * If the head is right up against the tail, we can't clear
1723 if (tail_distance <= 0) {
1724 ASSERT(tail_distance == 0);
1728 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1730 * Take the smaller of the maximum amount of outstanding I/O
1731 * we could have and the distance to the tail to clear out.
1732 * We take the smaller so that we don't overwrite the tail and
1733 * we don't waste all day writing from the head to the tail
1736 max_distance = MIN(max_distance, tail_distance);
1738 if ((head_block + max_distance) <= log->l_logBBsize) {
1740 * We can stomp all the blocks we need to without
1741 * wrapping around the end of the log. Just do it
1742 * in a single write. Use the cycle number of the
1743 * current cycle minus one so that the log will look like:
1746 error = xlog_write_log_records(log, (head_cycle - 1),
1747 head_block, max_distance, tail_cycle,
1753 * We need to wrap around the end of the physical log in
1754 * order to clear all the blocks. Do it in two separate
1755 * I/Os. The first write should be from the head to the
1756 * end of the physical log, and it should use the current
1757 * cycle number minus one just like above.
1759 distance = log->l_logBBsize - head_block;
1760 error = xlog_write_log_records(log, (head_cycle - 1),
1761 head_block, distance, tail_cycle,
1768 * Now write the blocks at the start of the physical log.
1769 * This writes the remainder of the blocks we want to clear.
1770 * It uses the current cycle number since we're now on the
1771 * same cycle as the head so that we get:
1772 * n ... n ... | n - 1 ...
1773 * ^^^^^ blocks we're writing
1775 distance = max_distance - (log->l_logBBsize - head_block);
1776 error = xlog_write_log_records(log, head_cycle, 0, distance,
1777 tail_cycle, tail_block);
1785 /******************************************************************************
1787 * Log recover routines
1789 ******************************************************************************
1793 * Sort the log items in the transaction.
1795 * The ordering constraints are defined by the inode allocation and unlink
1796 * behaviour. The rules are:
1798 * 1. Every item is only logged once in a given transaction. Hence it
1799 * represents the last logged state of the item. Hence ordering is
1800 * dependent on the order in which operations need to be performed so
1801 * required initial conditions are always met.
1803 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1804 * there's nothing to replay from them so we can simply cull them
1805 * from the transaction. However, we can't do that until after we've
1806 * replayed all the other items because they may be dependent on the
1807 * cancelled buffer and replaying the cancelled buffer can remove it
1808 * form the cancelled buffer table. Hence they have tobe done last.
1810 * 3. Inode allocation buffers must be replayed before inode items that
1811 * read the buffer and replay changes into it. For filesystems using the
1812 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1813 * treated the same as inode allocation buffers as they create and
1814 * initialise the buffers directly.
1816 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1817 * This ensures that inodes are completely flushed to the inode buffer
1818 * in a "free" state before we remove the unlinked inode list pointer.
1820 * Hence the ordering needs to be inode allocation buffers first, inode items
1821 * second, inode unlink buffers third and cancelled buffers last.
1823 * But there's a problem with that - we can't tell an inode allocation buffer
1824 * apart from a regular buffer, so we can't separate them. We can, however,
1825 * tell an inode unlink buffer from the others, and so we can separate them out
1826 * from all the other buffers and move them to last.
1828 * Hence, 4 lists, in order from head to tail:
1829 * - buffer_list for all buffers except cancelled/inode unlink buffers
1830 * - item_list for all non-buffer items
1831 * - inode_buffer_list for inode unlink buffers
1832 * - cancel_list for the cancelled buffers
1834 * Note that we add objects to the tail of the lists so that first-to-last
1835 * ordering is preserved within the lists. Adding objects to the head of the
1836 * list means when we traverse from the head we walk them in last-to-first
1837 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1838 * but for all other items there may be specific ordering that we need to
1842 xlog_recover_reorder_trans(
1844 struct xlog_recover *trans,
1847 xlog_recover_item_t *item, *n;
1849 LIST_HEAD(sort_list);
1850 LIST_HEAD(cancel_list);
1851 LIST_HEAD(buffer_list);
1852 LIST_HEAD(inode_buffer_list);
1853 LIST_HEAD(inode_list);
1855 list_splice_init(&trans->r_itemq, &sort_list);
1856 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1857 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1859 switch (ITEM_TYPE(item)) {
1860 case XFS_LI_ICREATE:
1861 list_move_tail(&item->ri_list, &buffer_list);
1864 if (buf_f->blf_flags & XFS_BLF_CANCEL) {
1865 trace_xfs_log_recover_item_reorder_head(log,
1867 list_move(&item->ri_list, &cancel_list);
1870 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1871 list_move(&item->ri_list, &inode_buffer_list);
1874 list_move_tail(&item->ri_list, &buffer_list);
1878 case XFS_LI_QUOTAOFF:
1881 trace_xfs_log_recover_item_reorder_tail(log,
1883 list_move_tail(&item->ri_list, &inode_list);
1887 "%s: unrecognized type of log operation",
1891 * return the remaining items back to the transaction
1892 * item list so they can be freed in caller.
1894 if (!list_empty(&sort_list))
1895 list_splice_init(&sort_list, &trans->r_itemq);
1901 ASSERT(list_empty(&sort_list));
1902 if (!list_empty(&buffer_list))
1903 list_splice(&buffer_list, &trans->r_itemq);
1904 if (!list_empty(&inode_list))
1905 list_splice_tail(&inode_list, &trans->r_itemq);
1906 if (!list_empty(&inode_buffer_list))
1907 list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1908 if (!list_empty(&cancel_list))
1909 list_splice_tail(&cancel_list, &trans->r_itemq);
1914 * Build up the table of buf cancel records so that we don't replay
1915 * cancelled data in the second pass. For buffer records that are
1916 * not cancel records, there is nothing to do here so we just return.
1918 * If we get a cancel record which is already in the table, this indicates
1919 * that the buffer was cancelled multiple times. In order to ensure
1920 * that during pass 2 we keep the record in the table until we reach its
1921 * last occurrence in the log, we keep a reference count in the cancel
1922 * record in the table to tell us how many times we expect to see this
1923 * record during the second pass.
1926 xlog_recover_buffer_pass1(
1928 struct xlog_recover_item *item)
1930 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1931 struct list_head *bucket;
1932 struct xfs_buf_cancel *bcp;
1935 * If this isn't a cancel buffer item, then just return.
1937 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1938 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
1943 * Insert an xfs_buf_cancel record into the hash table of them.
1944 * If there is already an identical record, bump its reference count.
1946 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1947 list_for_each_entry(bcp, bucket, bc_list) {
1948 if (bcp->bc_blkno == buf_f->blf_blkno &&
1949 bcp->bc_len == buf_f->blf_len) {
1951 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
1956 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1957 bcp->bc_blkno = buf_f->blf_blkno;
1958 bcp->bc_len = buf_f->blf_len;
1959 bcp->bc_refcount = 1;
1960 list_add_tail(&bcp->bc_list, bucket);
1962 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
1967 * Check to see whether the buffer being recovered has a corresponding
1968 * entry in the buffer cancel record table. If it is, return the cancel
1969 * buffer structure to the caller.
1971 STATIC struct xfs_buf_cancel *
1972 xlog_peek_buffer_cancelled(
1978 struct list_head *bucket;
1979 struct xfs_buf_cancel *bcp;
1981 if (!log->l_buf_cancel_table) {
1982 /* empty table means no cancelled buffers in the log */
1983 ASSERT(!(flags & XFS_BLF_CANCEL));
1987 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1988 list_for_each_entry(bcp, bucket, bc_list) {
1989 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
1994 * We didn't find a corresponding entry in the table, so return 0 so
1995 * that the buffer is NOT cancelled.
1997 ASSERT(!(flags & XFS_BLF_CANCEL));
2002 * If the buffer is being cancelled then return 1 so that it will be cancelled,
2003 * otherwise return 0. If the buffer is actually a buffer cancel item
2004 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
2005 * table and remove it from the table if this is the last reference.
2007 * We remove the cancel record from the table when we encounter its last
2008 * occurrence in the log so that if the same buffer is re-used again after its
2009 * last cancellation we actually replay the changes made at that point.
2012 xlog_check_buffer_cancelled(
2018 struct xfs_buf_cancel *bcp;
2020 bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags);
2025 * We've go a match, so return 1 so that the recovery of this buffer
2026 * is cancelled. If this buffer is actually a buffer cancel log
2027 * item, then decrement the refcount on the one in the table and
2028 * remove it if this is the last reference.
2030 if (flags & XFS_BLF_CANCEL) {
2031 if (--bcp->bc_refcount == 0) {
2032 list_del(&bcp->bc_list);
2040 * Perform recovery for a buffer full of inodes. In these buffers, the only
2041 * data which should be recovered is that which corresponds to the
2042 * di_next_unlinked pointers in the on disk inode structures. The rest of the
2043 * data for the inodes is always logged through the inodes themselves rather
2044 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
2046 * The only time when buffers full of inodes are fully recovered is when the
2047 * buffer is full of newly allocated inodes. In this case the buffer will
2048 * not be marked as an inode buffer and so will be sent to
2049 * xlog_recover_do_reg_buffer() below during recovery.
2052 xlog_recover_do_inode_buffer(
2053 struct xfs_mount *mp,
2054 xlog_recover_item_t *item,
2056 xfs_buf_log_format_t *buf_f)
2062 int reg_buf_offset = 0;
2063 int reg_buf_bytes = 0;
2064 int next_unlinked_offset;
2066 xfs_agino_t *logged_nextp;
2067 xfs_agino_t *buffer_nextp;
2069 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
2072 * Post recovery validation only works properly on CRC enabled
2075 if (xfs_sb_version_hascrc(&mp->m_sb))
2076 bp->b_ops = &xfs_inode_buf_ops;
2078 inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog;
2079 for (i = 0; i < inodes_per_buf; i++) {
2080 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
2081 offsetof(xfs_dinode_t, di_next_unlinked);
2083 while (next_unlinked_offset >=
2084 (reg_buf_offset + reg_buf_bytes)) {
2086 * The next di_next_unlinked field is beyond
2087 * the current logged region. Find the next
2088 * logged region that contains or is beyond
2089 * the current di_next_unlinked field.
2092 bit = xfs_next_bit(buf_f->blf_data_map,
2093 buf_f->blf_map_size, bit);
2096 * If there are no more logged regions in the
2097 * buffer, then we're done.
2102 nbits = xfs_contig_bits(buf_f->blf_data_map,
2103 buf_f->blf_map_size, bit);
2105 reg_buf_offset = bit << XFS_BLF_SHIFT;
2106 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
2111 * If the current logged region starts after the current
2112 * di_next_unlinked field, then move on to the next
2113 * di_next_unlinked field.
2115 if (next_unlinked_offset < reg_buf_offset)
2118 ASSERT(item->ri_buf[item_index].i_addr != NULL);
2119 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
2120 ASSERT((reg_buf_offset + reg_buf_bytes) <=
2121 BBTOB(bp->b_io_length));
2124 * The current logged region contains a copy of the
2125 * current di_next_unlinked field. Extract its value
2126 * and copy it to the buffer copy.
2128 logged_nextp = item->ri_buf[item_index].i_addr +
2129 next_unlinked_offset - reg_buf_offset;
2130 if (unlikely(*logged_nextp == 0)) {
2132 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
2133 "Trying to replay bad (0) inode di_next_unlinked field.",
2135 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
2136 XFS_ERRLEVEL_LOW, mp);
2137 return -EFSCORRUPTED;
2140 buffer_nextp = xfs_buf_offset(bp, next_unlinked_offset);
2141 *buffer_nextp = *logged_nextp;
2144 * If necessary, recalculate the CRC in the on-disk inode. We
2145 * have to leave the inode in a consistent state for whoever
2148 xfs_dinode_calc_crc(mp,
2149 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
2157 * V5 filesystems know the age of the buffer on disk being recovered. We can
2158 * have newer objects on disk than we are replaying, and so for these cases we
2159 * don't want to replay the current change as that will make the buffer contents
2160 * temporarily invalid on disk.
2162 * The magic number might not match the buffer type we are going to recover
2163 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
2164 * extract the LSN of the existing object in the buffer based on it's current
2165 * magic number. If we don't recognise the magic number in the buffer, then
2166 * return a LSN of -1 so that the caller knows it was an unrecognised block and
2167 * so can recover the buffer.
2169 * Note: we cannot rely solely on magic number matches to determine that the
2170 * buffer has a valid LSN - we also need to verify that it belongs to this
2171 * filesystem, so we need to extract the object's LSN and compare it to that
2172 * which we read from the superblock. If the UUIDs don't match, then we've got a
2173 * stale metadata block from an old filesystem instance that we need to recover
2177 xlog_recover_get_buf_lsn(
2178 struct xfs_mount *mp,
2184 void *blk = bp->b_addr;
2188 /* v4 filesystems always recover immediately */
2189 if (!xfs_sb_version_hascrc(&mp->m_sb))
2190 goto recover_immediately;
2192 magic32 = be32_to_cpu(*(__be32 *)blk);
2194 case XFS_ABTB_CRC_MAGIC:
2195 case XFS_ABTC_CRC_MAGIC:
2196 case XFS_ABTB_MAGIC:
2197 case XFS_ABTC_MAGIC:
2198 case XFS_IBT_CRC_MAGIC:
2199 case XFS_IBT_MAGIC: {
2200 struct xfs_btree_block *btb = blk;
2202 lsn = be64_to_cpu(btb->bb_u.s.bb_lsn);
2203 uuid = &btb->bb_u.s.bb_uuid;
2206 case XFS_BMAP_CRC_MAGIC:
2207 case XFS_BMAP_MAGIC: {
2208 struct xfs_btree_block *btb = blk;
2210 lsn = be64_to_cpu(btb->bb_u.l.bb_lsn);
2211 uuid = &btb->bb_u.l.bb_uuid;
2215 lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn);
2216 uuid = &((struct xfs_agf *)blk)->agf_uuid;
2218 case XFS_AGFL_MAGIC:
2219 lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn);
2220 uuid = &((struct xfs_agfl *)blk)->agfl_uuid;
2223 lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn);
2224 uuid = &((struct xfs_agi *)blk)->agi_uuid;
2226 case XFS_SYMLINK_MAGIC:
2227 lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn);
2228 uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid;
2230 case XFS_DIR3_BLOCK_MAGIC:
2231 case XFS_DIR3_DATA_MAGIC:
2232 case XFS_DIR3_FREE_MAGIC:
2233 lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn);
2234 uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid;
2236 case XFS_ATTR3_RMT_MAGIC:
2238 * Remote attr blocks are written synchronously, rather than
2239 * being logged. That means they do not contain a valid LSN
2240 * (i.e. transactionally ordered) in them, and hence any time we
2241 * see a buffer to replay over the top of a remote attribute
2242 * block we should simply do so.
2244 goto recover_immediately;
2247 * superblock uuids are magic. We may or may not have a
2248 * sb_meta_uuid on disk, but it will be set in the in-core
2249 * superblock. We set the uuid pointer for verification
2250 * according to the superblock feature mask to ensure we check
2251 * the relevant UUID in the superblock.
2253 lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn);
2254 if (xfs_sb_version_hasmetauuid(&mp->m_sb))
2255 uuid = &((struct xfs_dsb *)blk)->sb_meta_uuid;
2257 uuid = &((struct xfs_dsb *)blk)->sb_uuid;
2263 if (lsn != (xfs_lsn_t)-1) {
2264 if (!uuid_equal(&mp->m_sb.sb_meta_uuid, uuid))
2265 goto recover_immediately;
2269 magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic);
2271 case XFS_DIR3_LEAF1_MAGIC:
2272 case XFS_DIR3_LEAFN_MAGIC:
2273 case XFS_DA3_NODE_MAGIC:
2274 lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn);
2275 uuid = &((struct xfs_da3_blkinfo *)blk)->uuid;
2281 if (lsn != (xfs_lsn_t)-1) {
2282 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
2283 goto recover_immediately;
2288 * We do individual object checks on dquot and inode buffers as they
2289 * have their own individual LSN records. Also, we could have a stale
2290 * buffer here, so we have to at least recognise these buffer types.
2292 * A notd complexity here is inode unlinked list processing - it logs
2293 * the inode directly in the buffer, but we don't know which inodes have
2294 * been modified, and there is no global buffer LSN. Hence we need to
2295 * recover all inode buffer types immediately. This problem will be
2296 * fixed by logical logging of the unlinked list modifications.
2298 magic16 = be16_to_cpu(*(__be16 *)blk);
2300 case XFS_DQUOT_MAGIC:
2301 case XFS_DINODE_MAGIC:
2302 goto recover_immediately;
2307 /* unknown buffer contents, recover immediately */
2309 recover_immediately:
2310 return (xfs_lsn_t)-1;
2315 * Validate the recovered buffer is of the correct type and attach the
2316 * appropriate buffer operations to them for writeback. Magic numbers are in a
2318 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2319 * the first 32 bits of the buffer (most blocks),
2320 * inside a struct xfs_da_blkinfo at the start of the buffer.
2323 xlog_recover_validate_buf_type(
2324 struct xfs_mount *mp,
2326 xfs_buf_log_format_t *buf_f)
2328 struct xfs_da_blkinfo *info = bp->b_addr;
2334 * We can only do post recovery validation on items on CRC enabled
2335 * fielsystems as we need to know when the buffer was written to be able
2336 * to determine if we should have replayed the item. If we replay old
2337 * metadata over a newer buffer, then it will enter a temporarily
2338 * inconsistent state resulting in verification failures. Hence for now
2339 * just avoid the verification stage for non-crc filesystems
2341 if (!xfs_sb_version_hascrc(&mp->m_sb))
2344 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
2345 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
2346 magicda = be16_to_cpu(info->magic);
2347 switch (xfs_blft_from_flags(buf_f)) {
2348 case XFS_BLFT_BTREE_BUF:
2350 case XFS_ABTB_CRC_MAGIC:
2351 case XFS_ABTC_CRC_MAGIC:
2352 case XFS_ABTB_MAGIC:
2353 case XFS_ABTC_MAGIC:
2354 bp->b_ops = &xfs_allocbt_buf_ops;
2356 case XFS_IBT_CRC_MAGIC:
2357 case XFS_FIBT_CRC_MAGIC:
2359 case XFS_FIBT_MAGIC:
2360 bp->b_ops = &xfs_inobt_buf_ops;
2362 case XFS_BMAP_CRC_MAGIC:
2363 case XFS_BMAP_MAGIC:
2364 bp->b_ops = &xfs_bmbt_buf_ops;
2367 xfs_warn(mp, "Bad btree block magic!");
2372 case XFS_BLFT_AGF_BUF:
2373 if (magic32 != XFS_AGF_MAGIC) {
2374 xfs_warn(mp, "Bad AGF block magic!");
2378 bp->b_ops = &xfs_agf_buf_ops;
2380 case XFS_BLFT_AGFL_BUF:
2381 if (magic32 != XFS_AGFL_MAGIC) {
2382 xfs_warn(mp, "Bad AGFL block magic!");
2386 bp->b_ops = &xfs_agfl_buf_ops;
2388 case XFS_BLFT_AGI_BUF:
2389 if (magic32 != XFS_AGI_MAGIC) {
2390 xfs_warn(mp, "Bad AGI block magic!");
2394 bp->b_ops = &xfs_agi_buf_ops;
2396 case XFS_BLFT_UDQUOT_BUF:
2397 case XFS_BLFT_PDQUOT_BUF:
2398 case XFS_BLFT_GDQUOT_BUF:
2399 #ifdef CONFIG_XFS_QUOTA
2400 if (magic16 != XFS_DQUOT_MAGIC) {
2401 xfs_warn(mp, "Bad DQUOT block magic!");
2405 bp->b_ops = &xfs_dquot_buf_ops;
2408 "Trying to recover dquots without QUOTA support built in!");
2412 case XFS_BLFT_DINO_BUF:
2413 if (magic16 != XFS_DINODE_MAGIC) {
2414 xfs_warn(mp, "Bad INODE block magic!");
2418 bp->b_ops = &xfs_inode_buf_ops;
2420 case XFS_BLFT_SYMLINK_BUF:
2421 if (magic32 != XFS_SYMLINK_MAGIC) {
2422 xfs_warn(mp, "Bad symlink block magic!");
2426 bp->b_ops = &xfs_symlink_buf_ops;
2428 case XFS_BLFT_DIR_BLOCK_BUF:
2429 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2430 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2431 xfs_warn(mp, "Bad dir block magic!");
2435 bp->b_ops = &xfs_dir3_block_buf_ops;
2437 case XFS_BLFT_DIR_DATA_BUF:
2438 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2439 magic32 != XFS_DIR3_DATA_MAGIC) {
2440 xfs_warn(mp, "Bad dir data magic!");
2444 bp->b_ops = &xfs_dir3_data_buf_ops;
2446 case XFS_BLFT_DIR_FREE_BUF:
2447 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2448 magic32 != XFS_DIR3_FREE_MAGIC) {
2449 xfs_warn(mp, "Bad dir3 free magic!");
2453 bp->b_ops = &xfs_dir3_free_buf_ops;
2455 case XFS_BLFT_DIR_LEAF1_BUF:
2456 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2457 magicda != XFS_DIR3_LEAF1_MAGIC) {
2458 xfs_warn(mp, "Bad dir leaf1 magic!");
2462 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2464 case XFS_BLFT_DIR_LEAFN_BUF:
2465 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2466 magicda != XFS_DIR3_LEAFN_MAGIC) {
2467 xfs_warn(mp, "Bad dir leafn magic!");
2471 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2473 case XFS_BLFT_DA_NODE_BUF:
2474 if (magicda != XFS_DA_NODE_MAGIC &&
2475 magicda != XFS_DA3_NODE_MAGIC) {
2476 xfs_warn(mp, "Bad da node magic!");
2480 bp->b_ops = &xfs_da3_node_buf_ops;
2482 case XFS_BLFT_ATTR_LEAF_BUF:
2483 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2484 magicda != XFS_ATTR3_LEAF_MAGIC) {
2485 xfs_warn(mp, "Bad attr leaf magic!");
2489 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2491 case XFS_BLFT_ATTR_RMT_BUF:
2492 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
2493 xfs_warn(mp, "Bad attr remote magic!");
2497 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2499 case XFS_BLFT_SB_BUF:
2500 if (magic32 != XFS_SB_MAGIC) {
2501 xfs_warn(mp, "Bad SB block magic!");
2505 bp->b_ops = &xfs_sb_buf_ops;
2508 xfs_warn(mp, "Unknown buffer type %d!",
2509 xfs_blft_from_flags(buf_f));
2515 * Perform a 'normal' buffer recovery. Each logged region of the
2516 * buffer should be copied over the corresponding region in the
2517 * given buffer. The bitmap in the buf log format structure indicates
2518 * where to place the logged data.
2521 xlog_recover_do_reg_buffer(
2522 struct xfs_mount *mp,
2523 xlog_recover_item_t *item,
2525 xfs_buf_log_format_t *buf_f)
2532 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2535 i = 1; /* 0 is the buf format structure */
2537 bit = xfs_next_bit(buf_f->blf_data_map,
2538 buf_f->blf_map_size, bit);
2541 nbits = xfs_contig_bits(buf_f->blf_data_map,
2542 buf_f->blf_map_size, bit);
2544 ASSERT(item->ri_buf[i].i_addr != NULL);
2545 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
2546 ASSERT(BBTOB(bp->b_io_length) >=
2547 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
2550 * The dirty regions logged in the buffer, even though
2551 * contiguous, may span multiple chunks. This is because the
2552 * dirty region may span a physical page boundary in a buffer
2553 * and hence be split into two separate vectors for writing into
2554 * the log. Hence we need to trim nbits back to the length of
2555 * the current region being copied out of the log.
2557 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2558 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2561 * Do a sanity check if this is a dquot buffer. Just checking
2562 * the first dquot in the buffer should do. XXXThis is
2563 * probably a good thing to do for other buf types also.
2566 if (buf_f->blf_flags &
2567 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2568 if (item->ri_buf[i].i_addr == NULL) {
2570 "XFS: NULL dquot in %s.", __func__);
2573 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
2575 "XFS: dquot too small (%d) in %s.",
2576 item->ri_buf[i].i_len, __func__);
2579 error = xfs_dqcheck(mp, item->ri_buf[i].i_addr,
2580 -1, 0, XFS_QMOPT_DOWARN,
2581 "dquot_buf_recover");
2586 memcpy(xfs_buf_offset(bp,
2587 (uint)bit << XFS_BLF_SHIFT), /* dest */
2588 item->ri_buf[i].i_addr, /* source */
2589 nbits<<XFS_BLF_SHIFT); /* length */
2595 /* Shouldn't be any more regions */
2596 ASSERT(i == item->ri_total);
2598 xlog_recover_validate_buf_type(mp, bp, buf_f);
2602 * Perform a dquot buffer recovery.
2603 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2604 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2605 * Else, treat it as a regular buffer and do recovery.
2607 * Return false if the buffer was tossed and true if we recovered the buffer to
2608 * indicate to the caller if the buffer needs writing.
2611 xlog_recover_do_dquot_buffer(
2612 struct xfs_mount *mp,
2614 struct xlog_recover_item *item,
2616 struct xfs_buf_log_format *buf_f)
2620 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2623 * Filesystems are required to send in quota flags at mount time.
2629 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2630 type |= XFS_DQ_USER;
2631 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2632 type |= XFS_DQ_PROJ;
2633 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2634 type |= XFS_DQ_GROUP;
2636 * This type of quotas was turned off, so ignore this buffer
2638 if (log->l_quotaoffs_flag & type)
2641 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2646 * This routine replays a modification made to a buffer at runtime.
2647 * There are actually two types of buffer, regular and inode, which
2648 * are handled differently. Inode buffers are handled differently
2649 * in that we only recover a specific set of data from them, namely
2650 * the inode di_next_unlinked fields. This is because all other inode
2651 * data is actually logged via inode records and any data we replay
2652 * here which overlaps that may be stale.
2654 * When meta-data buffers are freed at run time we log a buffer item
2655 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2656 * of the buffer in the log should not be replayed at recovery time.
2657 * This is so that if the blocks covered by the buffer are reused for
2658 * file data before we crash we don't end up replaying old, freed
2659 * meta-data into a user's file.
2661 * To handle the cancellation of buffer log items, we make two passes
2662 * over the log during recovery. During the first we build a table of
2663 * those buffers which have been cancelled, and during the second we
2664 * only replay those buffers which do not have corresponding cancel
2665 * records in the table. See xlog_recover_buffer_pass[1,2] above
2666 * for more details on the implementation of the table of cancel records.
2669 xlog_recover_buffer_pass2(
2671 struct list_head *buffer_list,
2672 struct xlog_recover_item *item,
2673 xfs_lsn_t current_lsn)
2675 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2676 xfs_mount_t *mp = log->l_mp;
2683 * In this pass we only want to recover all the buffers which have
2684 * not been cancelled and are not cancellation buffers themselves.
2686 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2687 buf_f->blf_len, buf_f->blf_flags)) {
2688 trace_xfs_log_recover_buf_cancel(log, buf_f);
2692 trace_xfs_log_recover_buf_recover(log, buf_f);
2695 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2696 buf_flags |= XBF_UNMAPPED;
2698 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2702 error = bp->b_error;
2704 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
2709 * Recover the buffer only if we get an LSN from it and it's less than
2710 * the lsn of the transaction we are replaying.
2712 * Note that we have to be extremely careful of readahead here.
2713 * Readahead does not attach verfiers to the buffers so if we don't
2714 * actually do any replay after readahead because of the LSN we found
2715 * in the buffer if more recent than that current transaction then we
2716 * need to attach the verifier directly. Failure to do so can lead to
2717 * future recovery actions (e.g. EFI and unlinked list recovery) can
2718 * operate on the buffers and they won't get the verifier attached. This
2719 * can lead to blocks on disk having the correct content but a stale
2722 * It is safe to assume these clean buffers are currently up to date.
2723 * If the buffer is dirtied by a later transaction being replayed, then
2724 * the verifier will be reset to match whatever recover turns that
2727 lsn = xlog_recover_get_buf_lsn(mp, bp);
2728 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2729 xlog_recover_validate_buf_type(mp, bp, buf_f);
2733 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
2734 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2737 } else if (buf_f->blf_flags &
2738 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2741 dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2745 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2749 * Perform delayed write on the buffer. Asynchronous writes will be
2750 * slower when taking into account all the buffers to be flushed.
2752 * Also make sure that only inode buffers with good sizes stay in
2753 * the buffer cache. The kernel moves inodes in buffers of 1 block
2754 * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode
2755 * buffers in the log can be a different size if the log was generated
2756 * by an older kernel using unclustered inode buffers or a newer kernel
2757 * running with a different inode cluster size. Regardless, if the
2758 * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size)
2759 * for *our* value of mp->m_inode_cluster_size, then we need to keep
2760 * the buffer out of the buffer cache so that the buffer won't
2761 * overlap with future reads of those inodes.
2763 if (XFS_DINODE_MAGIC ==
2764 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2765 (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize,
2766 (__uint32_t)log->l_mp->m_inode_cluster_size))) {
2768 error = xfs_bwrite(bp);
2770 ASSERT(bp->b_target->bt_mount == mp);
2771 bp->b_iodone = xlog_recover_iodone;
2772 xfs_buf_delwri_queue(bp, buffer_list);
2781 * Inode fork owner changes
2783 * If we have been told that we have to reparent the inode fork, it's because an
2784 * extent swap operation on a CRC enabled filesystem has been done and we are
2785 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2788 * The complexity here is that we don't have an inode context to work with, so
2789 * after we've replayed the inode we need to instantiate one. This is where the
2792 * We are in the middle of log recovery, so we can't run transactions. That
2793 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2794 * that will result in the corresponding iput() running the inode through
2795 * xfs_inactive(). If we've just replayed an inode core that changes the link
2796 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2797 * transactions (bad!).
2799 * So, to avoid this, we instantiate an inode directly from the inode core we've
2800 * just recovered. We have the buffer still locked, and all we really need to
2801 * instantiate is the inode core and the forks being modified. We can do this
2802 * manually, then run the inode btree owner change, and then tear down the
2803 * xfs_inode without having to run any transactions at all.
2805 * Also, because we don't have a transaction context available here but need to
2806 * gather all the buffers we modify for writeback so we pass the buffer_list
2807 * instead for the operation to use.
2811 xfs_recover_inode_owner_change(
2812 struct xfs_mount *mp,
2813 struct xfs_dinode *dip,
2814 struct xfs_inode_log_format *in_f,
2815 struct list_head *buffer_list)
2817 struct xfs_inode *ip;
2820 ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER));
2822 ip = xfs_inode_alloc(mp, in_f->ilf_ino);
2826 /* instantiate the inode */
2827 xfs_dinode_from_disk(&ip->i_d, dip);
2828 ASSERT(ip->i_d.di_version >= 3);
2830 error = xfs_iformat_fork(ip, dip);
2835 if (in_f->ilf_fields & XFS_ILOG_DOWNER) {
2836 ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT);
2837 error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK,
2838 ip->i_ino, buffer_list);
2843 if (in_f->ilf_fields & XFS_ILOG_AOWNER) {
2844 ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT);
2845 error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK,
2846 ip->i_ino, buffer_list);
2857 xlog_recover_inode_pass2(
2859 struct list_head *buffer_list,
2860 struct xlog_recover_item *item,
2861 xfs_lsn_t current_lsn)
2863 xfs_inode_log_format_t *in_f;
2864 xfs_mount_t *mp = log->l_mp;
2873 xfs_icdinode_t *dicp;
2877 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2878 in_f = item->ri_buf[0].i_addr;
2880 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
2882 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2888 * Inode buffers can be freed, look out for it,
2889 * and do not replay the inode.
2891 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2892 in_f->ilf_len, 0)) {
2894 trace_xfs_log_recover_inode_cancel(log, in_f);
2897 trace_xfs_log_recover_inode_recover(log, in_f);
2899 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
2900 &xfs_inode_buf_ops);
2905 error = bp->b_error;
2907 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
2910 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2911 dip = xfs_buf_offset(bp, in_f->ilf_boffset);
2914 * Make sure the place we're flushing out to really looks
2917 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
2919 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2920 __func__, dip, bp, in_f->ilf_ino);
2921 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2922 XFS_ERRLEVEL_LOW, mp);
2923 error = -EFSCORRUPTED;
2926 dicp = item->ri_buf[1].i_addr;
2927 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2929 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2930 __func__, item, in_f->ilf_ino);
2931 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2932 XFS_ERRLEVEL_LOW, mp);
2933 error = -EFSCORRUPTED;
2938 * If the inode has an LSN in it, recover the inode only if it's less
2939 * than the lsn of the transaction we are replaying. Note: we still
2940 * need to replay an owner change even though the inode is more recent
2941 * than the transaction as there is no guarantee that all the btree
2942 * blocks are more recent than this transaction, too.
2944 if (dip->di_version >= 3) {
2945 xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn);
2947 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2948 trace_xfs_log_recover_inode_skip(log, in_f);
2950 goto out_owner_change;
2955 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2956 * are transactional and if ordering is necessary we can determine that
2957 * more accurately by the LSN field in the V3 inode core. Don't trust
2958 * the inode versions we might be changing them here - use the
2959 * superblock flag to determine whether we need to look at di_flushiter
2960 * to skip replay when the on disk inode is newer than the log one
2962 if (!xfs_sb_version_hascrc(&mp->m_sb) &&
2963 dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
2965 * Deal with the wrap case, DI_MAX_FLUSH is less
2966 * than smaller numbers
2968 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
2969 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2972 trace_xfs_log_recover_inode_skip(log, in_f);
2978 /* Take the opportunity to reset the flush iteration count */
2979 dicp->di_flushiter = 0;
2981 if (unlikely(S_ISREG(dicp->di_mode))) {
2982 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2983 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2984 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2985 XFS_ERRLEVEL_LOW, mp, dicp);
2987 "%s: Bad regular inode log record, rec ptr 0x%p, "
2988 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2989 __func__, item, dip, bp, in_f->ilf_ino);
2990 error = -EFSCORRUPTED;
2993 } else if (unlikely(S_ISDIR(dicp->di_mode))) {
2994 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2995 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2996 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2997 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2998 XFS_ERRLEVEL_LOW, mp, dicp);
3000 "%s: Bad dir inode log record, rec ptr 0x%p, "
3001 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
3002 __func__, item, dip, bp, in_f->ilf_ino);
3003 error = -EFSCORRUPTED;
3007 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
3008 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
3009 XFS_ERRLEVEL_LOW, mp, dicp);
3011 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
3012 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
3013 __func__, item, dip, bp, in_f->ilf_ino,
3014 dicp->di_nextents + dicp->di_anextents,
3016 error = -EFSCORRUPTED;
3019 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
3020 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
3021 XFS_ERRLEVEL_LOW, mp, dicp);
3023 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
3024 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
3025 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
3026 error = -EFSCORRUPTED;
3029 isize = xfs_icdinode_size(dicp->di_version);
3030 if (unlikely(item->ri_buf[1].i_len > isize)) {
3031 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
3032 XFS_ERRLEVEL_LOW, mp, dicp);
3034 "%s: Bad inode log record length %d, rec ptr 0x%p",
3035 __func__, item->ri_buf[1].i_len, item);
3036 error = -EFSCORRUPTED;
3040 /* The core is in in-core format */
3041 xfs_dinode_to_disk(dip, dicp);
3043 /* the rest is in on-disk format */
3044 if (item->ri_buf[1].i_len > isize) {
3045 memcpy((char *)dip + isize,
3046 item->ri_buf[1].i_addr + isize,
3047 item->ri_buf[1].i_len - isize);
3050 fields = in_f->ilf_fields;
3051 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
3053 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
3056 memcpy(XFS_DFORK_DPTR(dip),
3057 &in_f->ilf_u.ilfu_uuid,
3062 if (in_f->ilf_size == 2)
3063 goto out_owner_change;
3064 len = item->ri_buf[2].i_len;
3065 src = item->ri_buf[2].i_addr;
3066 ASSERT(in_f->ilf_size <= 4);
3067 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
3068 ASSERT(!(fields & XFS_ILOG_DFORK) ||
3069 (len == in_f->ilf_dsize));
3071 switch (fields & XFS_ILOG_DFORK) {
3072 case XFS_ILOG_DDATA:
3074 memcpy(XFS_DFORK_DPTR(dip), src, len);
3077 case XFS_ILOG_DBROOT:
3078 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
3079 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
3080 XFS_DFORK_DSIZE(dip, mp));
3085 * There are no data fork flags set.
3087 ASSERT((fields & XFS_ILOG_DFORK) == 0);
3092 * If we logged any attribute data, recover it. There may or
3093 * may not have been any other non-core data logged in this
3096 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
3097 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
3102 len = item->ri_buf[attr_index].i_len;
3103 src = item->ri_buf[attr_index].i_addr;
3104 ASSERT(len == in_f->ilf_asize);
3106 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
3107 case XFS_ILOG_ADATA:
3109 dest = XFS_DFORK_APTR(dip);
3110 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
3111 memcpy(dest, src, len);
3114 case XFS_ILOG_ABROOT:
3115 dest = XFS_DFORK_APTR(dip);
3116 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
3117 len, (xfs_bmdr_block_t*)dest,
3118 XFS_DFORK_ASIZE(dip, mp));
3122 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
3130 if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER))
3131 error = xfs_recover_inode_owner_change(mp, dip, in_f,
3133 /* re-generate the checksum. */
3134 xfs_dinode_calc_crc(log->l_mp, dip);
3136 ASSERT(bp->b_target->bt_mount == mp);
3137 bp->b_iodone = xlog_recover_iodone;
3138 xfs_buf_delwri_queue(bp, buffer_list);
3149 * Recover QUOTAOFF records. We simply make a note of it in the xlog
3150 * structure, so that we know not to do any dquot item or dquot buffer recovery,
3154 xlog_recover_quotaoff_pass1(
3156 struct xlog_recover_item *item)
3158 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
3162 * The logitem format's flag tells us if this was user quotaoff,
3163 * group/project quotaoff or both.
3165 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
3166 log->l_quotaoffs_flag |= XFS_DQ_USER;
3167 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
3168 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
3169 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
3170 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
3176 * Recover a dquot record
3179 xlog_recover_dquot_pass2(
3181 struct list_head *buffer_list,
3182 struct xlog_recover_item *item,
3183 xfs_lsn_t current_lsn)
3185 xfs_mount_t *mp = log->l_mp;
3187 struct xfs_disk_dquot *ddq, *recddq;
3189 xfs_dq_logformat_t *dq_f;
3194 * Filesystems are required to send in quota flags at mount time.
3196 if (mp->m_qflags == 0)
3199 recddq = item->ri_buf[1].i_addr;
3200 if (recddq == NULL) {
3201 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
3204 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
3205 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
3206 item->ri_buf[1].i_len, __func__);
3211 * This type of quotas was turned off, so ignore this record.
3213 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3215 if (log->l_quotaoffs_flag & type)
3219 * At this point we know that quota was _not_ turned off.
3220 * Since the mount flags are not indicating to us otherwise, this
3221 * must mean that quota is on, and the dquot needs to be replayed.
3222 * Remember that we may not have fully recovered the superblock yet,
3223 * so we can't do the usual trick of looking at the SB quota bits.
3225 * The other possibility, of course, is that the quota subsystem was
3226 * removed since the last mount - ENOSYS.
3228 dq_f = item->ri_buf[0].i_addr;
3230 error = xfs_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
3231 "xlog_recover_dquot_pass2 (log copy)");
3234 ASSERT(dq_f->qlf_len == 1);
3237 * At this point we are assuming that the dquots have been allocated
3238 * and hence the buffer has valid dquots stamped in it. It should,
3239 * therefore, pass verifier validation. If the dquot is bad, then the
3240 * we'll return an error here, so we don't need to specifically check
3241 * the dquot in the buffer after the verifier has run.
3243 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
3244 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
3245 &xfs_dquot_buf_ops);
3250 ddq = xfs_buf_offset(bp, dq_f->qlf_boffset);
3253 * If the dquot has an LSN in it, recover the dquot only if it's less
3254 * than the lsn of the transaction we are replaying.
3256 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3257 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq;
3258 xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn);
3260 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3265 memcpy(ddq, recddq, item->ri_buf[1].i_len);
3266 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3267 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
3271 ASSERT(dq_f->qlf_size == 2);
3272 ASSERT(bp->b_target->bt_mount == mp);
3273 bp->b_iodone = xlog_recover_iodone;
3274 xfs_buf_delwri_queue(bp, buffer_list);
3282 * This routine is called to create an in-core extent free intent
3283 * item from the efi format structure which was logged on disk.
3284 * It allocates an in-core efi, copies the extents from the format
3285 * structure into it, and adds the efi to the AIL with the given
3289 xlog_recover_efi_pass2(
3291 struct xlog_recover_item *item,
3295 struct xfs_mount *mp = log->l_mp;
3296 struct xfs_efi_log_item *efip;
3297 struct xfs_efi_log_format *efi_formatp;
3299 efi_formatp = item->ri_buf[0].i_addr;
3301 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
3302 error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format);
3304 xfs_efi_item_free(efip);
3307 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
3309 spin_lock(&log->l_ailp->xa_lock);
3311 * The EFI has two references. One for the EFD and one for EFI to ensure
3312 * it makes it into the AIL. Insert the EFI into the AIL directly and
3313 * drop the EFI reference. Note that xfs_trans_ail_update() drops the
3316 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
3317 xfs_efi_release(efip);
3323 * This routine is called when an EFD format structure is found in a committed
3324 * transaction in the log. Its purpose is to cancel the corresponding EFI if it
3325 * was still in the log. To do this it searches the AIL for the EFI with an id
3326 * equal to that in the EFD format structure. If we find it we drop the EFD
3327 * reference, which removes the EFI from the AIL and frees it.
3330 xlog_recover_efd_pass2(
3332 struct xlog_recover_item *item)
3334 xfs_efd_log_format_t *efd_formatp;
3335 xfs_efi_log_item_t *efip = NULL;
3336 xfs_log_item_t *lip;
3338 struct xfs_ail_cursor cur;
3339 struct xfs_ail *ailp = log->l_ailp;
3341 efd_formatp = item->ri_buf[0].i_addr;
3342 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
3343 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
3344 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
3345 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
3346 efi_id = efd_formatp->efd_efi_id;
3349 * Search for the EFI with the id in the EFD format structure in the
3352 spin_lock(&ailp->xa_lock);
3353 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3354 while (lip != NULL) {
3355 if (lip->li_type == XFS_LI_EFI) {
3356 efip = (xfs_efi_log_item_t *)lip;
3357 if (efip->efi_format.efi_id == efi_id) {
3359 * Drop the EFD reference to the EFI. This
3360 * removes the EFI from the AIL and frees it.
3362 spin_unlock(&ailp->xa_lock);
3363 xfs_efi_release(efip);
3364 spin_lock(&ailp->xa_lock);
3368 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3371 xfs_trans_ail_cursor_done(&cur);
3372 spin_unlock(&ailp->xa_lock);
3378 * This routine is called when an inode create format structure is found in a
3379 * committed transaction in the log. It's purpose is to initialise the inodes
3380 * being allocated on disk. This requires us to get inode cluster buffers that
3381 * match the range to be intialised, stamped with inode templates and written
3382 * by delayed write so that subsequent modifications will hit the cached buffer
3383 * and only need writing out at the end of recovery.
3386 xlog_recover_do_icreate_pass2(
3388 struct list_head *buffer_list,
3389 xlog_recover_item_t *item)
3391 struct xfs_mount *mp = log->l_mp;
3392 struct xfs_icreate_log *icl;
3393 xfs_agnumber_t agno;
3394 xfs_agblock_t agbno;
3397 xfs_agblock_t length;
3398 int blks_per_cluster;
3404 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3405 if (icl->icl_type != XFS_LI_ICREATE) {
3406 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
3410 if (icl->icl_size != 1) {
3411 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
3415 agno = be32_to_cpu(icl->icl_ag);
3416 if (agno >= mp->m_sb.sb_agcount) {
3417 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
3420 agbno = be32_to_cpu(icl->icl_agbno);
3421 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3422 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
3425 isize = be32_to_cpu(icl->icl_isize);
3426 if (isize != mp->m_sb.sb_inodesize) {
3427 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
3430 count = be32_to_cpu(icl->icl_count);
3432 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
3435 length = be32_to_cpu(icl->icl_length);
3436 if (!length || length >= mp->m_sb.sb_agblocks) {
3437 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
3442 * The inode chunk is either full or sparse and we only support
3443 * m_ialloc_min_blks sized sparse allocations at this time.
3445 if (length != mp->m_ialloc_blks &&
3446 length != mp->m_ialloc_min_blks) {
3448 "%s: unsupported chunk length", __FUNCTION__);
3452 /* verify inode count is consistent with extent length */
3453 if ((count >> mp->m_sb.sb_inopblog) != length) {
3455 "%s: inconsistent inode count and chunk length",
3461 * The icreate transaction can cover multiple cluster buffers and these
3462 * buffers could have been freed and reused. Check the individual
3463 * buffers for cancellation so we don't overwrite anything written after
3466 blks_per_cluster = xfs_icluster_size_fsb(mp);
3467 bb_per_cluster = XFS_FSB_TO_BB(mp, blks_per_cluster);
3468 nbufs = length / blks_per_cluster;
3469 for (i = 0, cancel_count = 0; i < nbufs; i++) {
3472 daddr = XFS_AGB_TO_DADDR(mp, agno,
3473 agbno + i * blks_per_cluster);
3474 if (xlog_check_buffer_cancelled(log, daddr, bb_per_cluster, 0))
3479 * We currently only use icreate for a single allocation at a time. This
3480 * means we should expect either all or none of the buffers to be
3481 * cancelled. Be conservative and skip replay if at least one buffer is
3482 * cancelled, but warn the user that something is awry if the buffers
3483 * are not consistent.
3485 * XXX: This must be refined to only skip cancelled clusters once we use
3486 * icreate for multiple chunk allocations.
3488 ASSERT(!cancel_count || cancel_count == nbufs);
3490 if (cancel_count != nbufs)
3492 "WARNING: partial inode chunk cancellation, skipped icreate.");
3493 trace_xfs_log_recover_icreate_cancel(log, icl);
3497 trace_xfs_log_recover_icreate_recover(log, icl);
3498 return xfs_ialloc_inode_init(mp, NULL, buffer_list, count, agno, agbno,
3499 length, be32_to_cpu(icl->icl_gen));
3503 xlog_recover_buffer_ra_pass2(
3505 struct xlog_recover_item *item)
3507 struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr;
3508 struct xfs_mount *mp = log->l_mp;
3510 if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
3511 buf_f->blf_len, buf_f->blf_flags)) {
3515 xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
3516 buf_f->blf_len, NULL);
3520 xlog_recover_inode_ra_pass2(
3522 struct xlog_recover_item *item)
3524 struct xfs_inode_log_format ilf_buf;
3525 struct xfs_inode_log_format *ilfp;
3526 struct xfs_mount *mp = log->l_mp;
3529 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3530 ilfp = item->ri_buf[0].i_addr;
3533 memset(ilfp, 0, sizeof(*ilfp));
3534 error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
3539 if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
3542 xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
3543 ilfp->ilf_len, &xfs_inode_buf_ra_ops);
3547 xlog_recover_dquot_ra_pass2(
3549 struct xlog_recover_item *item)
3551 struct xfs_mount *mp = log->l_mp;
3552 struct xfs_disk_dquot *recddq;
3553 struct xfs_dq_logformat *dq_f;
3558 if (mp->m_qflags == 0)
3561 recddq = item->ri_buf[1].i_addr;
3564 if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
3567 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3569 if (log->l_quotaoffs_flag & type)
3572 dq_f = item->ri_buf[0].i_addr;
3574 ASSERT(dq_f->qlf_len == 1);
3576 len = XFS_FSB_TO_BB(mp, dq_f->qlf_len);
3577 if (xlog_peek_buffer_cancelled(log, dq_f->qlf_blkno, len, 0))
3580 xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno, len,
3581 &xfs_dquot_buf_ra_ops);
3585 xlog_recover_ra_pass2(
3587 struct xlog_recover_item *item)
3589 switch (ITEM_TYPE(item)) {
3591 xlog_recover_buffer_ra_pass2(log, item);
3594 xlog_recover_inode_ra_pass2(log, item);
3597 xlog_recover_dquot_ra_pass2(log, item);
3601 case XFS_LI_QUOTAOFF:
3608 xlog_recover_commit_pass1(
3610 struct xlog_recover *trans,
3611 struct xlog_recover_item *item)
3613 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
3615 switch (ITEM_TYPE(item)) {
3617 return xlog_recover_buffer_pass1(log, item);
3618 case XFS_LI_QUOTAOFF:
3619 return xlog_recover_quotaoff_pass1(log, item);
3624 case XFS_LI_ICREATE:
3625 /* nothing to do in pass 1 */
3628 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3629 __func__, ITEM_TYPE(item));
3636 xlog_recover_commit_pass2(
3638 struct xlog_recover *trans,
3639 struct list_head *buffer_list,
3640 struct xlog_recover_item *item)
3642 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
3644 switch (ITEM_TYPE(item)) {
3646 return xlog_recover_buffer_pass2(log, buffer_list, item,
3649 return xlog_recover_inode_pass2(log, buffer_list, item,
3652 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
3654 return xlog_recover_efd_pass2(log, item);
3656 return xlog_recover_dquot_pass2(log, buffer_list, item,
3658 case XFS_LI_ICREATE:
3659 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
3660 case XFS_LI_QUOTAOFF:
3661 /* nothing to do in pass2 */
3664 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3665 __func__, ITEM_TYPE(item));
3672 xlog_recover_items_pass2(
3674 struct xlog_recover *trans,
3675 struct list_head *buffer_list,
3676 struct list_head *item_list)
3678 struct xlog_recover_item *item;
3681 list_for_each_entry(item, item_list, ri_list) {
3682 error = xlog_recover_commit_pass2(log, trans,
3692 * Perform the transaction.
3694 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3695 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3698 xlog_recover_commit_trans(
3700 struct xlog_recover *trans,
3705 int items_queued = 0;
3706 struct xlog_recover_item *item;
3707 struct xlog_recover_item *next;
3708 LIST_HEAD (buffer_list);
3709 LIST_HEAD (ra_list);
3710 LIST_HEAD (done_list);
3712 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
3714 hlist_del(&trans->r_list);
3716 error = xlog_recover_reorder_trans(log, trans, pass);
3720 list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
3722 case XLOG_RECOVER_PASS1:
3723 error = xlog_recover_commit_pass1(log, trans, item);
3725 case XLOG_RECOVER_PASS2:
3726 xlog_recover_ra_pass2(log, item);
3727 list_move_tail(&item->ri_list, &ra_list);
3729 if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
3730 error = xlog_recover_items_pass2(log, trans,
3731 &buffer_list, &ra_list);
3732 list_splice_tail_init(&ra_list, &done_list);
3746 if (!list_empty(&ra_list)) {
3748 error = xlog_recover_items_pass2(log, trans,
3749 &buffer_list, &ra_list);
3750 list_splice_tail_init(&ra_list, &done_list);
3753 if (!list_empty(&done_list))
3754 list_splice_init(&done_list, &trans->r_itemq);
3756 error2 = xfs_buf_delwri_submit(&buffer_list);
3757 return error ? error : error2;
3761 xlog_recover_add_item(
3762 struct list_head *head)
3764 xlog_recover_item_t *item;
3766 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
3767 INIT_LIST_HEAD(&item->ri_list);
3768 list_add_tail(&item->ri_list, head);
3772 xlog_recover_add_to_cont_trans(
3774 struct xlog_recover *trans,
3778 xlog_recover_item_t *item;
3779 char *ptr, *old_ptr;
3783 * If the transaction is empty, the header was split across this and the
3784 * previous record. Copy the rest of the header.
3786 if (list_empty(&trans->r_itemq)) {
3787 ASSERT(len <= sizeof(struct xfs_trans_header));
3788 if (len > sizeof(struct xfs_trans_header)) {
3789 xfs_warn(log->l_mp, "%s: bad header length", __func__);
3793 xlog_recover_add_item(&trans->r_itemq);
3794 ptr = (char *)&trans->r_theader +
3795 sizeof(struct xfs_trans_header) - len;
3796 memcpy(ptr, dp, len);
3800 /* take the tail entry */
3801 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
3803 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
3804 old_len = item->ri_buf[item->ri_cnt-1].i_len;
3806 ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP);
3807 memcpy(&ptr[old_len], dp, len);
3808 item->ri_buf[item->ri_cnt-1].i_len += len;
3809 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
3810 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
3815 * The next region to add is the start of a new region. It could be
3816 * a whole region or it could be the first part of a new region. Because
3817 * of this, the assumption here is that the type and size fields of all
3818 * format structures fit into the first 32 bits of the structure.
3820 * This works because all regions must be 32 bit aligned. Therefore, we
3821 * either have both fields or we have neither field. In the case we have
3822 * neither field, the data part of the region is zero length. We only have
3823 * a log_op_header and can throw away the header since a new one will appear
3824 * later. If we have at least 4 bytes, then we can determine how many regions
3825 * will appear in the current log item.
3828 xlog_recover_add_to_trans(
3830 struct xlog_recover *trans,
3834 xfs_inode_log_format_t *in_f; /* any will do */
3835 xlog_recover_item_t *item;
3840 if (list_empty(&trans->r_itemq)) {
3841 /* we need to catch log corruptions here */
3842 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
3843 xfs_warn(log->l_mp, "%s: bad header magic number",
3849 if (len > sizeof(struct xfs_trans_header)) {
3850 xfs_warn(log->l_mp, "%s: bad header length", __func__);
3856 * The transaction header can be arbitrarily split across op
3857 * records. If we don't have the whole thing here, copy what we
3858 * do have and handle the rest in the next record.
3860 if (len == sizeof(struct xfs_trans_header))
3861 xlog_recover_add_item(&trans->r_itemq);
3862 memcpy(&trans->r_theader, dp, len);
3866 ptr = kmem_alloc(len, KM_SLEEP);
3867 memcpy(ptr, dp, len);
3868 in_f = (xfs_inode_log_format_t *)ptr;
3870 /* take the tail entry */
3871 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
3872 if (item->ri_total != 0 &&
3873 item->ri_total == item->ri_cnt) {
3874 /* tail item is in use, get a new one */
3875 xlog_recover_add_item(&trans->r_itemq);
3876 item = list_entry(trans->r_itemq.prev,
3877 xlog_recover_item_t, ri_list);
3880 if (item->ri_total == 0) { /* first region to be added */
3881 if (in_f->ilf_size == 0 ||
3882 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
3884 "bad number of regions (%d) in inode log format",
3891 item->ri_total = in_f->ilf_size;
3893 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
3896 ASSERT(item->ri_total > item->ri_cnt);
3897 /* Description region is ri_buf[0] */
3898 item->ri_buf[item->ri_cnt].i_addr = ptr;
3899 item->ri_buf[item->ri_cnt].i_len = len;
3901 trace_xfs_log_recover_item_add(log, trans, item, 0);
3906 * Free up any resources allocated by the transaction
3908 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3911 xlog_recover_free_trans(
3912 struct xlog_recover *trans)
3914 xlog_recover_item_t *item, *n;
3917 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
3918 /* Free the regions in the item. */
3919 list_del(&item->ri_list);
3920 for (i = 0; i < item->ri_cnt; i++)
3921 kmem_free(item->ri_buf[i].i_addr);
3922 /* Free the item itself */
3923 kmem_free(item->ri_buf);
3926 /* Free the transaction recover structure */
3931 * On error or completion, trans is freed.
3934 xlog_recovery_process_trans(
3936 struct xlog_recover *trans,
3943 bool freeit = false;
3945 /* mask off ophdr transaction container flags */
3946 flags &= ~XLOG_END_TRANS;
3947 if (flags & XLOG_WAS_CONT_TRANS)
3948 flags &= ~XLOG_CONTINUE_TRANS;
3951 * Callees must not free the trans structure. We'll decide if we need to
3952 * free it or not based on the operation being done and it's result.
3955 /* expected flag values */
3957 case XLOG_CONTINUE_TRANS:
3958 error = xlog_recover_add_to_trans(log, trans, dp, len);
3960 case XLOG_WAS_CONT_TRANS:
3961 error = xlog_recover_add_to_cont_trans(log, trans, dp, len);
3963 case XLOG_COMMIT_TRANS:
3964 error = xlog_recover_commit_trans(log, trans, pass);
3965 /* success or fail, we are now done with this transaction. */
3969 /* unexpected flag values */
3970 case XLOG_UNMOUNT_TRANS:
3971 /* just skip trans */
3972 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
3975 case XLOG_START_TRANS:
3977 xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags);
3982 if (error || freeit)
3983 xlog_recover_free_trans(trans);
3988 * Lookup the transaction recovery structure associated with the ID in the
3989 * current ophdr. If the transaction doesn't exist and the start flag is set in
3990 * the ophdr, then allocate a new transaction for future ID matches to find.
3991 * Either way, return what we found during the lookup - an existing transaction
3994 STATIC struct xlog_recover *
3995 xlog_recover_ophdr_to_trans(
3996 struct hlist_head rhash[],
3997 struct xlog_rec_header *rhead,
3998 struct xlog_op_header *ohead)
4000 struct xlog_recover *trans;
4002 struct hlist_head *rhp;
4004 tid = be32_to_cpu(ohead->oh_tid);
4005 rhp = &rhash[XLOG_RHASH(tid)];
4006 hlist_for_each_entry(trans, rhp, r_list) {
4007 if (trans->r_log_tid == tid)
4012 * skip over non-start transaction headers - we could be
4013 * processing slack space before the next transaction starts
4015 if (!(ohead->oh_flags & XLOG_START_TRANS))
4018 ASSERT(be32_to_cpu(ohead->oh_len) == 0);
4021 * This is a new transaction so allocate a new recovery container to
4022 * hold the recovery ops that will follow.
4024 trans = kmem_zalloc(sizeof(struct xlog_recover), KM_SLEEP);
4025 trans->r_log_tid = tid;
4026 trans->r_lsn = be64_to_cpu(rhead->h_lsn);
4027 INIT_LIST_HEAD(&trans->r_itemq);
4028 INIT_HLIST_NODE(&trans->r_list);
4029 hlist_add_head(&trans->r_list, rhp);
4032 * Nothing more to do for this ophdr. Items to be added to this new
4033 * transaction will be in subsequent ophdr containers.
4039 xlog_recover_process_ophdr(
4041 struct hlist_head rhash[],
4042 struct xlog_rec_header *rhead,
4043 struct xlog_op_header *ohead,
4048 struct xlog_recover *trans;
4051 /* Do we understand who wrote this op? */
4052 if (ohead->oh_clientid != XFS_TRANSACTION &&
4053 ohead->oh_clientid != XFS_LOG) {
4054 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
4055 __func__, ohead->oh_clientid);
4061 * Check the ophdr contains all the data it is supposed to contain.
4063 len = be32_to_cpu(ohead->oh_len);
4064 if (dp + len > end) {
4065 xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len);
4070 trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead);
4072 /* nothing to do, so skip over this ophdr */
4076 return xlog_recovery_process_trans(log, trans, dp, len,
4077 ohead->oh_flags, pass);
4081 * There are two valid states of the r_state field. 0 indicates that the
4082 * transaction structure is in a normal state. We have either seen the
4083 * start of the transaction or the last operation we added was not a partial
4084 * operation. If the last operation we added to the transaction was a
4085 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
4087 * NOTE: skip LRs with 0 data length.
4090 xlog_recover_process_data(
4092 struct hlist_head rhash[],
4093 struct xlog_rec_header *rhead,
4097 struct xlog_op_header *ohead;
4102 end = dp + be32_to_cpu(rhead->h_len);
4103 num_logops = be32_to_cpu(rhead->h_num_logops);
4105 /* check the log format matches our own - else we can't recover */
4106 if (xlog_header_check_recover(log->l_mp, rhead))
4109 while ((dp < end) && num_logops) {
4111 ohead = (struct xlog_op_header *)dp;
4112 dp += sizeof(*ohead);
4115 /* errors will abort recovery */
4116 error = xlog_recover_process_ophdr(log, rhash, rhead, ohead,
4121 dp += be32_to_cpu(ohead->oh_len);
4128 * Process an extent free intent item that was recovered from
4129 * the log. We need to free the extents that it describes.
4132 xlog_recover_process_efi(
4134 xfs_efi_log_item_t *efip)
4136 xfs_efd_log_item_t *efdp;
4141 xfs_fsblock_t startblock_fsb;
4143 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
4146 * First check the validity of the extents described by the
4147 * EFI. If any are bad, then assume that all are bad and
4148 * just toss the EFI.
4150 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
4151 extp = &(efip->efi_format.efi_extents[i]);
4152 startblock_fsb = XFS_BB_TO_FSB(mp,
4153 XFS_FSB_TO_DADDR(mp, extp->ext_start));
4154 if ((startblock_fsb == 0) ||
4155 (extp->ext_len == 0) ||
4156 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
4157 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
4159 * This will pull the EFI from the AIL and
4160 * free the memory associated with it.
4162 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
4163 xfs_efi_release(efip);
4168 tp = xfs_trans_alloc(mp, 0);
4169 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
4172 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
4174 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
4175 extp = &(efip->efi_format.efi_extents[i]);
4176 error = xfs_trans_free_extent(tp, efdp, extp->ext_start,
4183 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
4184 error = xfs_trans_commit(tp);
4188 xfs_trans_cancel(tp);
4193 * When this is called, all of the EFIs which did not have
4194 * corresponding EFDs should be in the AIL. What we do now
4195 * is free the extents associated with each one.
4197 * Since we process the EFIs in normal transactions, they
4198 * will be removed at some point after the commit. This prevents
4199 * us from just walking down the list processing each one.
4200 * We'll use a flag in the EFI to skip those that we've already
4201 * processed and use the AIL iteration mechanism's generation
4202 * count to try to speed this up at least a bit.
4204 * When we start, we know that the EFIs are the only things in
4205 * the AIL. As we process them, however, other items are added
4206 * to the AIL. Since everything added to the AIL must come after
4207 * everything already in the AIL, we stop processing as soon as
4208 * we see something other than an EFI in the AIL.
4211 xlog_recover_process_efis(
4214 struct xfs_log_item *lip;
4215 struct xfs_efi_log_item *efip;
4217 struct xfs_ail_cursor cur;
4218 struct xfs_ail *ailp;
4221 spin_lock(&ailp->xa_lock);
4222 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
4223 while (lip != NULL) {
4225 * We're done when we see something other than an EFI.
4226 * There should be no EFIs left in the AIL now.
4228 if (lip->li_type != XFS_LI_EFI) {
4230 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
4231 ASSERT(lip->li_type != XFS_LI_EFI);
4237 * Skip EFIs that we've already processed.
4239 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
4240 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
4241 lip = xfs_trans_ail_cursor_next(ailp, &cur);
4245 spin_unlock(&ailp->xa_lock);
4246 error = xlog_recover_process_efi(log->l_mp, efip);
4247 spin_lock(&ailp->xa_lock);
4250 lip = xfs_trans_ail_cursor_next(ailp, &cur);
4253 xfs_trans_ail_cursor_done(&cur);
4254 spin_unlock(&ailp->xa_lock);
4259 * A cancel occurs when the mount has failed and we're bailing out. Release all
4260 * pending EFIs so they don't pin the AIL.
4263 xlog_recover_cancel_efis(
4266 struct xfs_log_item *lip;
4267 struct xfs_efi_log_item *efip;
4269 struct xfs_ail_cursor cur;
4270 struct xfs_ail *ailp;
4273 spin_lock(&ailp->xa_lock);
4274 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
4275 while (lip != NULL) {
4277 * We're done when we see something other than an EFI.
4278 * There should be no EFIs left in the AIL now.
4280 if (lip->li_type != XFS_LI_EFI) {
4282 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
4283 ASSERT(lip->li_type != XFS_LI_EFI);
4288 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
4290 spin_unlock(&ailp->xa_lock);
4291 xfs_efi_release(efip);
4292 spin_lock(&ailp->xa_lock);
4294 lip = xfs_trans_ail_cursor_next(ailp, &cur);
4297 xfs_trans_ail_cursor_done(&cur);
4298 spin_unlock(&ailp->xa_lock);
4303 * This routine performs a transaction to null out a bad inode pointer
4304 * in an agi unlinked inode hash bucket.
4307 xlog_recover_clear_agi_bucket(
4309 xfs_agnumber_t agno,
4318 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
4319 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_clearagi, 0, 0);
4323 error = xfs_read_agi(mp, tp, agno, &agibp);
4327 agi = XFS_BUF_TO_AGI(agibp);
4328 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
4329 offset = offsetof(xfs_agi_t, agi_unlinked) +
4330 (sizeof(xfs_agino_t) * bucket);
4331 xfs_trans_log_buf(tp, agibp, offset,
4332 (offset + sizeof(xfs_agino_t) - 1));
4334 error = xfs_trans_commit(tp);
4340 xfs_trans_cancel(tp);
4342 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
4347 xlog_recover_process_one_iunlink(
4348 struct xfs_mount *mp,
4349 xfs_agnumber_t agno,
4353 struct xfs_buf *ibp;
4354 struct xfs_dinode *dip;
4355 struct xfs_inode *ip;
4359 ino = XFS_AGINO_TO_INO(mp, agno, agino);
4360 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
4365 * Get the on disk inode to find the next inode in the bucket.
4367 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
4371 ASSERT(ip->i_d.di_nlink == 0);
4372 ASSERT(ip->i_d.di_mode != 0);
4374 /* setup for the next pass */
4375 agino = be32_to_cpu(dip->di_next_unlinked);
4379 * Prevent any DMAPI event from being sent when the reference on
4380 * the inode is dropped.
4382 ip->i_d.di_dmevmask = 0;
4391 * We can't read in the inode this bucket points to, or this inode
4392 * is messed up. Just ditch this bucket of inodes. We will lose
4393 * some inodes and space, but at least we won't hang.
4395 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
4396 * clear the inode pointer in the bucket.
4398 xlog_recover_clear_agi_bucket(mp, agno, bucket);
4403 * xlog_iunlink_recover
4405 * This is called during recovery to process any inodes which
4406 * we unlinked but not freed when the system crashed. These
4407 * inodes will be on the lists in the AGI blocks. What we do
4408 * here is scan all the AGIs and fully truncate and free any
4409 * inodes found on the lists. Each inode is removed from the
4410 * lists when it has been fully truncated and is freed. The
4411 * freeing of the inode and its removal from the list must be
4415 xlog_recover_process_iunlinks(
4419 xfs_agnumber_t agno;
4430 * Prevent any DMAPI event from being sent while in this function.
4432 mp_dmevmask = mp->m_dmevmask;
4435 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
4437 * Find the agi for this ag.
4439 error = xfs_read_agi(mp, NULL, agno, &agibp);
4442 * AGI is b0rked. Don't process it.
4444 * We should probably mark the filesystem as corrupt
4445 * after we've recovered all the ag's we can....
4450 * Unlock the buffer so that it can be acquired in the normal
4451 * course of the transaction to truncate and free each inode.
4452 * Because we are not racing with anyone else here for the AGI
4453 * buffer, we don't even need to hold it locked to read the
4454 * initial unlinked bucket entries out of the buffer. We keep
4455 * buffer reference though, so that it stays pinned in memory
4456 * while we need the buffer.
4458 agi = XFS_BUF_TO_AGI(agibp);
4459 xfs_buf_unlock(agibp);
4461 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
4462 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
4463 while (agino != NULLAGINO) {
4464 agino = xlog_recover_process_one_iunlink(mp,
4465 agno, agino, bucket);
4468 xfs_buf_rele(agibp);
4471 mp->m_dmevmask = mp_dmevmask;
4476 struct xlog_rec_header *rhead,
4482 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
4483 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
4484 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
4488 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
4489 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
4490 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
4491 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
4492 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
4493 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
4502 * CRC check, unpack and process a log record.
4505 xlog_recover_process(
4507 struct hlist_head rhash[],
4508 struct xlog_rec_header *rhead,
4515 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
4518 * Nothing else to do if this is a CRC verification pass. Just return
4519 * if this a record with a non-zero crc. Unfortunately, mkfs always
4520 * sets h_crc to 0 so we must consider this valid even on v5 supers.
4521 * Otherwise, return EFSBADCRC on failure so the callers up the stack
4522 * know precisely what failed.
4524 if (pass == XLOG_RECOVER_CRCPASS) {
4525 if (rhead->h_crc && crc != le32_to_cpu(rhead->h_crc))
4531 * We're in the normal recovery path. Issue a warning if and only if the
4532 * CRC in the header is non-zero. This is an advisory warning and the
4533 * zero CRC check prevents warnings from being emitted when upgrading
4534 * the kernel from one that does not add CRCs by default.
4536 if (crc != le32_to_cpu(rhead->h_crc)) {
4537 if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
4538 xfs_alert(log->l_mp,
4539 "log record CRC mismatch: found 0x%x, expected 0x%x.",
4540 le32_to_cpu(rhead->h_crc),
4542 xfs_hex_dump(dp, 32);
4546 * If the filesystem is CRC enabled, this mismatch becomes a
4547 * fatal log corruption failure.
4549 if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
4550 return -EFSCORRUPTED;
4553 error = xlog_unpack_data(rhead, dp, log);
4557 return xlog_recover_process_data(log, rhash, rhead, dp, pass);
4561 xlog_valid_rec_header(
4563 struct xlog_rec_header *rhead,
4568 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
4569 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
4570 XFS_ERRLEVEL_LOW, log->l_mp);
4571 return -EFSCORRUPTED;
4574 (!rhead->h_version ||
4575 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
4576 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
4577 __func__, be32_to_cpu(rhead->h_version));
4581 /* LR body must have data or it wouldn't have been written */
4582 hlen = be32_to_cpu(rhead->h_len);
4583 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
4584 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
4585 XFS_ERRLEVEL_LOW, log->l_mp);
4586 return -EFSCORRUPTED;
4588 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
4589 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
4590 XFS_ERRLEVEL_LOW, log->l_mp);
4591 return -EFSCORRUPTED;
4597 * Read the log from tail to head and process the log records found.
4598 * Handle the two cases where the tail and head are in the same cycle
4599 * and where the active portion of the log wraps around the end of
4600 * the physical log separately. The pass parameter is passed through
4601 * to the routines called to process the data and is not looked at
4605 xlog_do_recovery_pass(
4607 xfs_daddr_t head_blk,
4608 xfs_daddr_t tail_blk,
4610 xfs_daddr_t *first_bad) /* out: first bad log rec */
4612 xlog_rec_header_t *rhead;
4614 xfs_daddr_t rhead_blk;
4616 xfs_buf_t *hbp, *dbp;
4617 int error = 0, h_size, h_len;
4618 int bblks, split_bblks;
4619 int hblks, split_hblks, wrapped_hblks;
4620 struct hlist_head rhash[XLOG_RHASH_SIZE];
4622 ASSERT(head_blk != tail_blk);
4626 * Read the header of the tail block and get the iclog buffer size from
4627 * h_size. Use this to tell how many sectors make up the log header.
4629 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
4631 * When using variable length iclogs, read first sector of
4632 * iclog header and extract the header size from it. Get a
4633 * new hbp that is the correct size.
4635 hbp = xlog_get_bp(log, 1);
4639 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
4643 rhead = (xlog_rec_header_t *)offset;
4644 error = xlog_valid_rec_header(log, rhead, tail_blk);
4649 * xfsprogs has a bug where record length is based on lsunit but
4650 * h_size (iclog size) is hardcoded to 32k. Now that we
4651 * unconditionally CRC verify the unmount record, this means the
4652 * log buffer can be too small for the record and cause an
4655 * Detect this condition here. Use lsunit for the buffer size as
4656 * long as this looks like the mkfs case. Otherwise, return an
4657 * error to avoid a buffer overrun.
4659 h_size = be32_to_cpu(rhead->h_size);
4660 h_len = be32_to_cpu(rhead->h_len);
4661 if (h_len > h_size) {
4662 if (h_len <= log->l_mp->m_logbsize &&
4663 be32_to_cpu(rhead->h_num_logops) == 1) {
4665 "invalid iclog size (%d bytes), using lsunit (%d bytes)",
4666 h_size, log->l_mp->m_logbsize);
4667 h_size = log->l_mp->m_logbsize;
4669 return -EFSCORRUPTED;
4672 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
4673 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
4674 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
4675 if (h_size % XLOG_HEADER_CYCLE_SIZE)
4678 hbp = xlog_get_bp(log, hblks);
4683 ASSERT(log->l_sectBBsize == 1);
4685 hbp = xlog_get_bp(log, 1);
4686 h_size = XLOG_BIG_RECORD_BSIZE;
4691 dbp = xlog_get_bp(log, BTOBB(h_size));
4697 memset(rhash, 0, sizeof(rhash));
4698 blk_no = rhead_blk = tail_blk;
4699 if (tail_blk > head_blk) {
4701 * Perform recovery around the end of the physical log.
4702 * When the head is not on the same cycle number as the tail,
4703 * we can't do a sequential recovery.
4705 while (blk_no < log->l_logBBsize) {
4707 * Check for header wrapping around physical end-of-log
4709 offset = hbp->b_addr;
4712 if (blk_no + hblks <= log->l_logBBsize) {
4713 /* Read header in one read */
4714 error = xlog_bread(log, blk_no, hblks, hbp,
4719 /* This LR is split across physical log end */
4720 if (blk_no != log->l_logBBsize) {
4721 /* some data before physical log end */
4722 ASSERT(blk_no <= INT_MAX);
4723 split_hblks = log->l_logBBsize - (int)blk_no;
4724 ASSERT(split_hblks > 0);
4725 error = xlog_bread(log, blk_no,
4733 * Note: this black magic still works with
4734 * large sector sizes (non-512) only because:
4735 * - we increased the buffer size originally
4736 * by 1 sector giving us enough extra space
4737 * for the second read;
4738 * - the log start is guaranteed to be sector
4740 * - we read the log end (LR header start)
4741 * _first_, then the log start (LR header end)
4742 * - order is important.
4744 wrapped_hblks = hblks - split_hblks;
4745 error = xlog_bread_offset(log, 0,
4747 offset + BBTOB(split_hblks));
4751 rhead = (xlog_rec_header_t *)offset;
4752 error = xlog_valid_rec_header(log, rhead,
4753 split_hblks ? blk_no : 0);
4757 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4760 /* Read in data for log record */
4761 if (blk_no + bblks <= log->l_logBBsize) {
4762 error = xlog_bread(log, blk_no, bblks, dbp,
4767 /* This log record is split across the
4768 * physical end of log */
4769 offset = dbp->b_addr;
4771 if (blk_no != log->l_logBBsize) {
4772 /* some data is before the physical
4774 ASSERT(!wrapped_hblks);
4775 ASSERT(blk_no <= INT_MAX);
4777 log->l_logBBsize - (int)blk_no;
4778 ASSERT(split_bblks > 0);
4779 error = xlog_bread(log, blk_no,
4787 * Note: this black magic still works with
4788 * large sector sizes (non-512) only because:
4789 * - we increased the buffer size originally
4790 * by 1 sector giving us enough extra space
4791 * for the second read;
4792 * - the log start is guaranteed to be sector
4794 * - we read the log end (LR header start)
4795 * _first_, then the log start (LR header end)
4796 * - order is important.
4798 error = xlog_bread_offset(log, 0,
4799 bblks - split_bblks, dbp,
4800 offset + BBTOB(split_bblks));
4805 error = xlog_recover_process(log, rhash, rhead, offset,
4814 ASSERT(blk_no >= log->l_logBBsize);
4815 blk_no -= log->l_logBBsize;
4819 /* read first part of physical log */
4820 while (blk_no < head_blk) {
4821 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
4825 rhead = (xlog_rec_header_t *)offset;
4826 error = xlog_valid_rec_header(log, rhead, blk_no);
4830 /* blocks in data section */
4831 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4832 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
4837 error = xlog_recover_process(log, rhash, rhead, offset, pass);
4841 blk_no += bblks + hblks;
4850 if (error && first_bad)
4851 *first_bad = rhead_blk;
4857 * Do the recovery of the log. We actually do this in two phases.
4858 * The two passes are necessary in order to implement the function
4859 * of cancelling a record written into the log. The first pass
4860 * determines those things which have been cancelled, and the
4861 * second pass replays log items normally except for those which
4862 * have been cancelled. The handling of the replay and cancellations
4863 * takes place in the log item type specific routines.
4865 * The table of items which have cancel records in the log is allocated
4866 * and freed at this level, since only here do we know when all of
4867 * the log recovery has been completed.
4870 xlog_do_log_recovery(
4872 xfs_daddr_t head_blk,
4873 xfs_daddr_t tail_blk)
4877 ASSERT(head_blk != tail_blk);
4880 * First do a pass to find all of the cancelled buf log items.
4881 * Store them in the buf_cancel_table for use in the second pass.
4883 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
4884 sizeof(struct list_head),
4886 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4887 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
4889 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4890 XLOG_RECOVER_PASS1, NULL);
4892 kmem_free(log->l_buf_cancel_table);
4893 log->l_buf_cancel_table = NULL;
4897 * Then do a second pass to actually recover the items in the log.
4898 * When it is complete free the table of buf cancel items.
4900 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4901 XLOG_RECOVER_PASS2, NULL);
4906 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4907 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
4911 kmem_free(log->l_buf_cancel_table);
4912 log->l_buf_cancel_table = NULL;
4918 * Do the actual recovery
4923 xfs_daddr_t head_blk,
4924 xfs_daddr_t tail_blk)
4931 * First replay the images in the log.
4933 error = xlog_do_log_recovery(log, head_blk, tail_blk);
4938 * If IO errors happened during recovery, bail out.
4940 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
4945 * We now update the tail_lsn since much of the recovery has completed
4946 * and there may be space available to use. If there were no extent
4947 * or iunlinks, we can free up the entire log and set the tail_lsn to
4948 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4949 * lsn of the last known good LR on disk. If there are extent frees
4950 * or iunlinks they will have some entries in the AIL; so we look at
4951 * the AIL to determine how to set the tail_lsn.
4953 xlog_assign_tail_lsn(log->l_mp);
4956 * Now that we've finished replaying all buffer and inode
4957 * updates, re-read in the superblock and reverify it.
4959 bp = xfs_getsb(log->l_mp, 0);
4961 ASSERT(!(XFS_BUF_ISWRITE(bp)));
4963 XFS_BUF_UNASYNC(bp);
4964 bp->b_ops = &xfs_sb_buf_ops;
4966 error = xfs_buf_submit_wait(bp);
4968 if (!XFS_FORCED_SHUTDOWN(log->l_mp)) {
4969 xfs_buf_ioerror_alert(bp, __func__);
4976 /* Convert superblock from on-disk format */
4977 sbp = &log->l_mp->m_sb;
4978 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
4979 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
4980 ASSERT(xfs_sb_good_version(sbp));
4981 xfs_reinit_percpu_counters(log->l_mp);
4986 xlog_recover_check_summary(log);
4988 /* Normal transactions can now occur */
4989 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
4994 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4996 * Return error or zero.
5002 xfs_daddr_t head_blk, tail_blk;
5005 /* find the tail of the log */
5006 error = xlog_find_tail(log, &head_blk, &tail_blk);
5011 * The superblock was read before the log was available and thus the LSN
5012 * could not be verified. Check the superblock LSN against the current
5013 * LSN now that it's known.
5015 if (xfs_sb_version_hascrc(&log->l_mp->m_sb) &&
5016 !xfs_log_check_lsn(log->l_mp, log->l_mp->m_sb.sb_lsn))
5019 if (tail_blk != head_blk) {
5020 /* There used to be a comment here:
5022 * disallow recovery on read-only mounts. note -- mount
5023 * checks for ENOSPC and turns it into an intelligent
5025 * ...but this is no longer true. Now, unless you specify
5026 * NORECOVERY (in which case this function would never be
5027 * called), we just go ahead and recover. We do this all
5028 * under the vfs layer, so we can get away with it unless
5029 * the device itself is read-only, in which case we fail.
5031 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
5036 * Version 5 superblock log feature mask validation. We know the
5037 * log is dirty so check if there are any unknown log features
5038 * in what we need to recover. If there are unknown features
5039 * (e.g. unsupported transactions, then simply reject the
5040 * attempt at recovery before touching anything.
5042 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
5043 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
5044 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
5046 "Superblock has unknown incompatible log features (0x%x) enabled.",
5047 (log->l_mp->m_sb.sb_features_log_incompat &
5048 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
5050 "The log can not be fully and/or safely recovered by this kernel.");
5052 "Please recover the log on a kernel that supports the unknown features.");
5057 * Delay log recovery if the debug hook is set. This is debug
5058 * instrumention to coordinate simulation of I/O failures with
5061 if (xfs_globals.log_recovery_delay) {
5062 xfs_notice(log->l_mp,
5063 "Delaying log recovery for %d seconds.",
5064 xfs_globals.log_recovery_delay);
5065 msleep(xfs_globals.log_recovery_delay * 1000);
5068 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
5069 log->l_mp->m_logname ? log->l_mp->m_logname
5072 error = xlog_do_recover(log, head_blk, tail_blk);
5073 log->l_flags |= XLOG_RECOVERY_NEEDED;
5079 * In the first part of recovery we replay inodes and buffers and build
5080 * up the list of extent free items which need to be processed. Here
5081 * we process the extent free items and clean up the on disk unlinked
5082 * inode lists. This is separated from the first part of recovery so
5083 * that the root and real-time bitmap inodes can be read in from disk in
5084 * between the two stages. This is necessary so that we can free space
5085 * in the real-time portion of the file system.
5088 xlog_recover_finish(
5092 * Now we're ready to do the transactions needed for the
5093 * rest of recovery. Start with completing all the extent
5094 * free intent records and then process the unlinked inode
5095 * lists. At this point, we essentially run in normal mode
5096 * except that we're still performing recovery actions
5097 * rather than accepting new requests.
5099 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
5101 error = xlog_recover_process_efis(log);
5103 xfs_alert(log->l_mp, "Failed to recover EFIs");
5107 * Sync the log to get all the EFIs out of the AIL.
5108 * This isn't absolutely necessary, but it helps in
5109 * case the unlink transactions would have problems
5110 * pushing the EFIs out of the way.
5112 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
5114 xlog_recover_process_iunlinks(log);
5116 xlog_recover_check_summary(log);
5118 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
5119 log->l_mp->m_logname ? log->l_mp->m_logname
5121 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
5123 xfs_info(log->l_mp, "Ending clean mount");
5129 xlog_recover_cancel(
5134 if (log->l_flags & XLOG_RECOVERY_NEEDED)
5135 error = xlog_recover_cancel_efis(log);
5142 * Read all of the agf and agi counters and check that they
5143 * are consistent with the superblock counters.
5146 xlog_recover_check_summary(
5153 xfs_agnumber_t agno;
5154 __uint64_t freeblks;
5164 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
5165 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
5167 xfs_alert(mp, "%s agf read failed agno %d error %d",
5168 __func__, agno, error);
5170 agfp = XFS_BUF_TO_AGF(agfbp);
5171 freeblks += be32_to_cpu(agfp->agf_freeblks) +
5172 be32_to_cpu(agfp->agf_flcount);
5173 xfs_buf_relse(agfbp);
5176 error = xfs_read_agi(mp, NULL, agno, &agibp);
5178 xfs_alert(mp, "%s agi read failed agno %d error %d",
5179 __func__, agno, error);
5181 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
5183 itotal += be32_to_cpu(agi->agi_count);
5184 ifree += be32_to_cpu(agi->agi_freecount);
5185 xfs_buf_relse(agibp);