2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_format.h"
30 #include "xfs_inode.h"
31 #include "xfs_trans.h"
33 #include "xfs_log_priv.h"
34 #include "xfs_log_recover.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_extfree_item.h"
37 #include "xfs_trans_priv.h"
38 #include "xfs_alloc.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_quota.h"
41 #include "xfs_cksum.h"
42 #include "xfs_trace.h"
43 #include "xfs_icache.h"
44 #include "xfs_bmap_btree.h"
45 #include "xfs_dinode.h"
46 #include "xfs_error.h"
49 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
56 xlog_clear_stale_blocks(
61 xlog_recover_check_summary(
64 #define xlog_recover_check_summary(log)
68 * This structure is used during recovery to record the buf log items which
69 * have been canceled and should not be replayed.
71 struct xfs_buf_cancel {
75 struct list_head bc_list;
79 * Sector aligned buffer routines for buffer create/read/write/access
83 * Verify the given count of basic blocks is valid number of blocks
84 * to specify for an operation involving the given XFS log buffer.
85 * Returns nonzero if the count is valid, 0 otherwise.
89 xlog_buf_bbcount_valid(
93 return bbcount > 0 && bbcount <= log->l_logBBsize;
97 * Allocate a buffer to hold log data. The buffer needs to be able
98 * to map to a range of nbblks basic blocks at any valid (basic
99 * block) offset within the log.
108 if (!xlog_buf_bbcount_valid(log, nbblks)) {
109 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
111 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
116 * We do log I/O in units of log sectors (a power-of-2
117 * multiple of the basic block size), so we round up the
118 * requested size to accommodate the basic blocks required
119 * for complete log sectors.
121 * In addition, the buffer may be used for a non-sector-
122 * aligned block offset, in which case an I/O of the
123 * requested size could extend beyond the end of the
124 * buffer. If the requested size is only 1 basic block it
125 * will never straddle a sector boundary, so this won't be
126 * an issue. Nor will this be a problem if the log I/O is
127 * done in basic blocks (sector size 1). But otherwise we
128 * extend the buffer by one extra log sector to ensure
129 * there's space to accommodate this possibility.
131 if (nbblks > 1 && log->l_sectBBsize > 1)
132 nbblks += log->l_sectBBsize;
133 nbblks = round_up(nbblks, log->l_sectBBsize);
135 bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0);
149 * Return the address of the start of the given block number's data
150 * in a log buffer. The buffer covers a log sector-aligned region.
159 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
161 ASSERT(offset + nbblks <= bp->b_length);
162 return bp->b_addr + BBTOB(offset);
167 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
178 if (!xlog_buf_bbcount_valid(log, nbblks)) {
179 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
181 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
185 blk_no = round_down(blk_no, log->l_sectBBsize);
186 nbblks = round_up(nbblks, log->l_sectBBsize);
189 ASSERT(nbblks <= bp->b_length);
191 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
193 bp->b_io_length = nbblks;
196 xfsbdstrat(log->l_mp, bp);
197 error = xfs_buf_iowait(bp);
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 xfs_caddr_t 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);
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 XFS_ERROR(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 XFS_ERROR(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 XFS_ERROR(EFSCORRUPTED);
375 * We're not going to bother about retrying
376 * this during recovery. One strike!
378 xfs_buf_ioerror_alert(bp, __func__);
379 xfs_force_shutdown(bp->b_target->bt_mount,
380 SHUTDOWN_META_IO_ERROR);
383 xfs_buf_ioend(bp, 0);
387 * This routine finds (to an approximation) the first block in the physical
388 * log which contains the given cycle. It uses a binary search algorithm.
389 * Note that the algorithm can not be perfect because the disk will not
390 * necessarily be perfect.
393 xlog_find_cycle_start(
396 xfs_daddr_t first_blk,
397 xfs_daddr_t *last_blk,
407 mid_blk = BLK_AVG(first_blk, end_blk);
408 while (mid_blk != first_blk && mid_blk != end_blk) {
409 error = xlog_bread(log, mid_blk, 1, bp, &offset);
412 mid_cycle = xlog_get_cycle(offset);
413 if (mid_cycle == cycle)
414 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
416 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
417 mid_blk = BLK_AVG(first_blk, end_blk);
419 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
420 (mid_blk == end_blk && mid_blk-1 == first_blk));
428 * Check that a range of blocks does not contain stop_on_cycle_no.
429 * Fill in *new_blk with the block offset where such a block is
430 * found, or with -1 (an invalid block number) if there is no such
431 * block in the range. The scan needs to occur from front to back
432 * and the pointer into the region must be updated since a later
433 * routine will need to perform another test.
436 xlog_find_verify_cycle(
438 xfs_daddr_t start_blk,
440 uint stop_on_cycle_no,
441 xfs_daddr_t *new_blk)
447 xfs_caddr_t buf = NULL;
451 * Greedily allocate a buffer big enough to handle the full
452 * range of basic blocks we'll be examining. If that fails,
453 * try a smaller size. We need to be able to read at least
454 * a log sector, or we're out of luck.
456 bufblks = 1 << ffs(nbblks);
457 while (bufblks > log->l_logBBsize)
459 while (!(bp = xlog_get_bp(log, bufblks))) {
461 if (bufblks < log->l_sectBBsize)
465 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
468 bcount = min(bufblks, (start_blk + nbblks - i));
470 error = xlog_bread(log, i, bcount, bp, &buf);
474 for (j = 0; j < bcount; j++) {
475 cycle = xlog_get_cycle(buf);
476 if (cycle == stop_on_cycle_no) {
493 * Potentially backup over partial log record write.
495 * In the typical case, last_blk is the number of the block directly after
496 * a good log record. Therefore, we subtract one to get the block number
497 * of the last block in the given buffer. extra_bblks contains the number
498 * of blocks we would have read on a previous read. This happens when the
499 * last log record is split over the end of the physical log.
501 * extra_bblks is the number of blocks potentially verified on a previous
502 * call to this routine.
505 xlog_find_verify_log_record(
507 xfs_daddr_t start_blk,
508 xfs_daddr_t *last_blk,
513 xfs_caddr_t offset = NULL;
514 xlog_rec_header_t *head = NULL;
517 int num_blks = *last_blk - start_blk;
520 ASSERT(start_blk != 0 || *last_blk != start_blk);
522 if (!(bp = xlog_get_bp(log, num_blks))) {
523 if (!(bp = xlog_get_bp(log, 1)))
527 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
530 offset += ((num_blks - 1) << BBSHIFT);
533 for (i = (*last_blk) - 1; i >= 0; i--) {
535 /* valid log record not found */
537 "Log inconsistent (didn't find previous header)");
539 error = XFS_ERROR(EIO);
544 error = xlog_bread(log, i, 1, bp, &offset);
549 head = (xlog_rec_header_t *)offset;
551 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
559 * We hit the beginning of the physical log & still no header. Return
560 * to caller. If caller can handle a return of -1, then this routine
561 * will be called again for the end of the physical log.
569 * We have the final block of the good log (the first block
570 * of the log record _before_ the head. So we check the uuid.
572 if ((error = xlog_header_check_mount(log->l_mp, head)))
576 * We may have found a log record header before we expected one.
577 * last_blk will be the 1st block # with a given cycle #. We may end
578 * up reading an entire log record. In this case, we don't want to
579 * reset last_blk. Only when last_blk points in the middle of a log
580 * record do we update last_blk.
582 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
583 uint h_size = be32_to_cpu(head->h_size);
585 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
586 if (h_size % XLOG_HEADER_CYCLE_SIZE)
592 if (*last_blk - i + extra_bblks !=
593 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
602 * Head is defined to be the point of the log where the next log write
603 * could go. This means that incomplete LR writes at the end are
604 * eliminated when calculating the head. We aren't guaranteed that previous
605 * LR have complete transactions. We only know that a cycle number of
606 * current cycle number -1 won't be present in the log if we start writing
607 * from our current block number.
609 * last_blk contains the block number of the first block with a given
612 * Return: zero if normal, non-zero if error.
617 xfs_daddr_t *return_head_blk)
621 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
623 uint first_half_cycle, last_half_cycle;
625 int error, log_bbnum = log->l_logBBsize;
627 /* Is the end of the log device zeroed? */
628 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
629 *return_head_blk = first_blk;
631 /* Is the whole lot zeroed? */
633 /* Linux XFS shouldn't generate totally zeroed logs -
634 * mkfs etc write a dummy unmount record to a fresh
635 * log so we can store the uuid in there
637 xfs_warn(log->l_mp, "totally zeroed log");
642 xfs_warn(log->l_mp, "empty log check failed");
646 first_blk = 0; /* get cycle # of 1st block */
647 bp = xlog_get_bp(log, 1);
651 error = xlog_bread(log, 0, 1, bp, &offset);
655 first_half_cycle = xlog_get_cycle(offset);
657 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
658 error = xlog_bread(log, last_blk, 1, bp, &offset);
662 last_half_cycle = xlog_get_cycle(offset);
663 ASSERT(last_half_cycle != 0);
666 * If the 1st half cycle number is equal to the last half cycle number,
667 * then the entire log is stamped with the same cycle number. In this
668 * case, head_blk can't be set to zero (which makes sense). The below
669 * math doesn't work out properly with head_blk equal to zero. Instead,
670 * we set it to log_bbnum which is an invalid block number, but this
671 * value makes the math correct. If head_blk doesn't changed through
672 * all the tests below, *head_blk is set to zero at the very end rather
673 * than log_bbnum. In a sense, log_bbnum and zero are the same block
674 * in a circular file.
676 if (first_half_cycle == last_half_cycle) {
678 * In this case we believe that the entire log should have
679 * cycle number last_half_cycle. We need to scan backwards
680 * from the end verifying that there are no holes still
681 * containing last_half_cycle - 1. If we find such a hole,
682 * then the start of that hole will be the new head. The
683 * simple case looks like
684 * x | x ... | x - 1 | x
685 * Another case that fits this picture would be
686 * x | x + 1 | x ... | x
687 * In this case the head really is somewhere at the end of the
688 * log, as one of the latest writes at the beginning was
691 * x | x + 1 | x ... | x - 1 | x
692 * This is really the combination of the above two cases, and
693 * the head has to end up at the start of the x-1 hole at the
696 * In the 256k log case, we will read from the beginning to the
697 * end of the log and search for cycle numbers equal to x-1.
698 * We don't worry about the x+1 blocks that we encounter,
699 * because we know that they cannot be the head since the log
702 head_blk = log_bbnum;
703 stop_on_cycle = last_half_cycle - 1;
706 * In this case we want to find the first block with cycle
707 * number matching last_half_cycle. We expect the log to be
709 * x + 1 ... | x ... | x
710 * The first block with cycle number x (last_half_cycle) will
711 * be where the new head belongs. First we do a binary search
712 * for the first occurrence of last_half_cycle. The binary
713 * search may not be totally accurate, so then we scan back
714 * from there looking for occurrences of last_half_cycle before
715 * us. If that backwards scan wraps around the beginning of
716 * the log, then we look for occurrences of last_half_cycle - 1
717 * at the end of the log. The cases we're looking for look
719 * v binary search stopped here
720 * x + 1 ... | x | x + 1 | x ... | x
721 * ^ but we want to locate this spot
723 * <---------> less than scan distance
724 * x + 1 ... | x ... | x - 1 | x
725 * ^ we want to locate this spot
727 stop_on_cycle = last_half_cycle;
728 if ((error = xlog_find_cycle_start(log, bp, first_blk,
729 &head_blk, last_half_cycle)))
734 * Now validate the answer. Scan back some number of maximum possible
735 * blocks and make sure each one has the expected cycle number. The
736 * maximum is determined by the total possible amount of buffering
737 * in the in-core log. The following number can be made tighter if
738 * we actually look at the block size of the filesystem.
740 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
741 if (head_blk >= num_scan_bblks) {
743 * We are guaranteed that the entire check can be performed
746 start_blk = head_blk - num_scan_bblks;
747 if ((error = xlog_find_verify_cycle(log,
748 start_blk, num_scan_bblks,
749 stop_on_cycle, &new_blk)))
753 } else { /* need to read 2 parts of log */
755 * We are going to scan backwards in the log in two parts.
756 * First we scan the physical end of the log. In this part
757 * of the log, we are looking for blocks with cycle number
758 * last_half_cycle - 1.
759 * If we find one, then we know that the log starts there, as
760 * we've found a hole that didn't get written in going around
761 * the end of the physical log. The simple case for this is
762 * x + 1 ... | x ... | x - 1 | x
763 * <---------> less than scan distance
764 * If all of the blocks at the end of the log have cycle number
765 * last_half_cycle, then we check the blocks at the start of
766 * the log looking for occurrences of last_half_cycle. If we
767 * find one, then our current estimate for the location of the
768 * first occurrence of last_half_cycle is wrong and we move
769 * back to the hole we've found. This case looks like
770 * x + 1 ... | x | x + 1 | x ...
771 * ^ binary search stopped here
772 * Another case we need to handle that only occurs in 256k
774 * x + 1 ... | x ... | x+1 | x ...
775 * ^ binary search stops here
776 * In a 256k log, the scan at the end of the log will see the
777 * x + 1 blocks. We need to skip past those since that is
778 * certainly not the head of the log. By searching for
779 * last_half_cycle-1 we accomplish that.
781 ASSERT(head_blk <= INT_MAX &&
782 (xfs_daddr_t) num_scan_bblks >= head_blk);
783 start_blk = log_bbnum - (num_scan_bblks - head_blk);
784 if ((error = xlog_find_verify_cycle(log, start_blk,
785 num_scan_bblks - (int)head_blk,
786 (stop_on_cycle - 1), &new_blk)))
794 * Scan beginning of log now. The last part of the physical
795 * log is good. This scan needs to verify that it doesn't find
796 * the last_half_cycle.
799 ASSERT(head_blk <= INT_MAX);
800 if ((error = xlog_find_verify_cycle(log,
801 start_blk, (int)head_blk,
802 stop_on_cycle, &new_blk)))
810 * Now we need to make sure head_blk is not pointing to a block in
811 * the middle of a log record.
813 num_scan_bblks = XLOG_REC_SHIFT(log);
814 if (head_blk >= num_scan_bblks) {
815 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
817 /* start ptr at last block ptr before head_blk */
818 if ((error = xlog_find_verify_log_record(log, start_blk,
819 &head_blk, 0)) == -1) {
820 error = XFS_ERROR(EIO);
826 ASSERT(head_blk <= INT_MAX);
827 if ((error = xlog_find_verify_log_record(log, start_blk,
828 &head_blk, 0)) == -1) {
829 /* We hit the beginning of the log during our search */
830 start_blk = log_bbnum - (num_scan_bblks - head_blk);
832 ASSERT(start_blk <= INT_MAX &&
833 (xfs_daddr_t) log_bbnum-start_blk >= 0);
834 ASSERT(head_blk <= INT_MAX);
835 if ((error = xlog_find_verify_log_record(log,
837 (int)head_blk)) == -1) {
838 error = XFS_ERROR(EIO);
842 if (new_blk != log_bbnum)
849 if (head_blk == log_bbnum)
850 *return_head_blk = 0;
852 *return_head_blk = head_blk;
854 * When returning here, we have a good block number. Bad block
855 * means that during a previous crash, we didn't have a clean break
856 * from cycle number N to cycle number N-1. In this case, we need
857 * to find the first block with cycle number N-1.
865 xfs_warn(log->l_mp, "failed to find log head");
870 * Find the sync block number or the tail of the log.
872 * This will be the block number of the last record to have its
873 * associated buffers synced to disk. Every log record header has
874 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
875 * to get a sync block number. The only concern is to figure out which
876 * log record header to believe.
878 * The following algorithm uses the log record header with the largest
879 * lsn. The entire log record does not need to be valid. We only care
880 * that the header is valid.
882 * We could speed up search by using current head_blk buffer, but it is not
888 xfs_daddr_t *head_blk,
889 xfs_daddr_t *tail_blk)
891 xlog_rec_header_t *rhead;
892 xlog_op_header_t *op_head;
893 xfs_caddr_t offset = NULL;
896 xfs_daddr_t umount_data_blk;
897 xfs_daddr_t after_umount_blk;
904 * Find previous log record
906 if ((error = xlog_find_head(log, head_blk)))
909 bp = xlog_get_bp(log, 1);
912 if (*head_blk == 0) { /* special case */
913 error = xlog_bread(log, 0, 1, bp, &offset);
917 if (xlog_get_cycle(offset) == 0) {
919 /* leave all other log inited values alone */
925 * Search backwards looking for log record header block
927 ASSERT(*head_blk < INT_MAX);
928 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
929 error = xlog_bread(log, i, 1, bp, &offset);
933 if (*(__be32 *)offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
939 * If we haven't found the log record header block, start looking
940 * again from the end of the physical log. XXXmiken: There should be
941 * a check here to make sure we didn't search more than N blocks in
945 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
946 error = xlog_bread(log, i, 1, bp, &offset);
950 if (*(__be32 *)offset ==
951 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
958 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
961 return XFS_ERROR(EIO);
964 /* find blk_no of tail of log */
965 rhead = (xlog_rec_header_t *)offset;
966 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
969 * Reset log values according to the state of the log when we
970 * crashed. In the case where head_blk == 0, we bump curr_cycle
971 * one because the next write starts a new cycle rather than
972 * continuing the cycle of the last good log record. At this
973 * point we have guaranteed that all partial log records have been
974 * accounted for. Therefore, we know that the last good log record
975 * written was complete and ended exactly on the end boundary
976 * of the physical log.
978 log->l_prev_block = i;
979 log->l_curr_block = (int)*head_blk;
980 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
983 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
984 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
985 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
986 BBTOB(log->l_curr_block));
987 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
988 BBTOB(log->l_curr_block));
991 * Look for unmount record. If we find it, then we know there
992 * was a clean unmount. Since 'i' could be the last block in
993 * the physical log, we convert to a log block before comparing
996 * Save the current tail lsn to use to pass to
997 * xlog_clear_stale_blocks() below. We won't want to clear the
998 * unmount record if there is one, so we pass the lsn of the
999 * unmount record rather than the block after it.
1001 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1002 int h_size = be32_to_cpu(rhead->h_size);
1003 int h_version = be32_to_cpu(rhead->h_version);
1005 if ((h_version & XLOG_VERSION_2) &&
1006 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
1007 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
1008 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1016 after_umount_blk = (i + hblks + (int)
1017 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
1018 tail_lsn = atomic64_read(&log->l_tail_lsn);
1019 if (*head_blk == after_umount_blk &&
1020 be32_to_cpu(rhead->h_num_logops) == 1) {
1021 umount_data_blk = (i + hblks) % log->l_logBBsize;
1022 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1026 op_head = (xlog_op_header_t *)offset;
1027 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1029 * Set tail and last sync so that newly written
1030 * log records will point recovery to after the
1031 * current unmount record.
1033 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1034 log->l_curr_cycle, after_umount_blk);
1035 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1036 log->l_curr_cycle, after_umount_blk);
1037 *tail_blk = after_umount_blk;
1040 * Note that the unmount was clean. If the unmount
1041 * was not clean, we need to know this to rebuild the
1042 * superblock counters from the perag headers if we
1043 * have a filesystem using non-persistent counters.
1045 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1050 * Make sure that there are no blocks in front of the head
1051 * with the same cycle number as the head. This can happen
1052 * because we allow multiple outstanding log writes concurrently,
1053 * and the later writes might make it out before earlier ones.
1055 * We use the lsn from before modifying it so that we'll never
1056 * overwrite the unmount record after a clean unmount.
1058 * Do this only if we are going to recover the filesystem
1060 * NOTE: This used to say "if (!readonly)"
1061 * However on Linux, we can & do recover a read-only filesystem.
1062 * We only skip recovery if NORECOVERY is specified on mount,
1063 * in which case we would not be here.
1065 * But... if the -device- itself is readonly, just skip this.
1066 * We can't recover this device anyway, so it won't matter.
1068 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
1069 error = xlog_clear_stale_blocks(log, tail_lsn);
1075 xfs_warn(log->l_mp, "failed to locate log tail");
1080 * Is the log zeroed at all?
1082 * The last binary search should be changed to perform an X block read
1083 * once X becomes small enough. You can then search linearly through
1084 * the X blocks. This will cut down on the number of reads we need to do.
1086 * If the log is partially zeroed, this routine will pass back the blkno
1087 * of the first block with cycle number 0. It won't have a complete LR
1091 * 0 => the log is completely written to
1092 * -1 => use *blk_no as the first block of the log
1093 * >0 => error has occurred
1098 xfs_daddr_t *blk_no)
1102 uint first_cycle, last_cycle;
1103 xfs_daddr_t new_blk, last_blk, start_blk;
1104 xfs_daddr_t num_scan_bblks;
1105 int error, log_bbnum = log->l_logBBsize;
1109 /* check totally zeroed log */
1110 bp = xlog_get_bp(log, 1);
1113 error = xlog_bread(log, 0, 1, bp, &offset);
1117 first_cycle = xlog_get_cycle(offset);
1118 if (first_cycle == 0) { /* completely zeroed log */
1124 /* check partially zeroed log */
1125 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1129 last_cycle = xlog_get_cycle(offset);
1130 if (last_cycle != 0) { /* log completely written to */
1133 } else if (first_cycle != 1) {
1135 * If the cycle of the last block is zero, the cycle of
1136 * the first block must be 1. If it's not, maybe we're
1137 * not looking at a log... Bail out.
1140 "Log inconsistent or not a log (last==0, first!=1)");
1141 error = XFS_ERROR(EINVAL);
1145 /* we have a partially zeroed log */
1146 last_blk = log_bbnum-1;
1147 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1151 * Validate the answer. Because there is no way to guarantee that
1152 * the entire log is made up of log records which are the same size,
1153 * we scan over the defined maximum blocks. At this point, the maximum
1154 * is not chosen to mean anything special. XXXmiken
1156 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1157 ASSERT(num_scan_bblks <= INT_MAX);
1159 if (last_blk < num_scan_bblks)
1160 num_scan_bblks = last_blk;
1161 start_blk = last_blk - num_scan_bblks;
1164 * We search for any instances of cycle number 0 that occur before
1165 * our current estimate of the head. What we're trying to detect is
1166 * 1 ... | 0 | 1 | 0...
1167 * ^ binary search ends here
1169 if ((error = xlog_find_verify_cycle(log, start_blk,
1170 (int)num_scan_bblks, 0, &new_blk)))
1176 * Potentially backup over partial log record write. We don't need
1177 * to search the end of the log because we know it is zero.
1179 if ((error = xlog_find_verify_log_record(log, start_blk,
1180 &last_blk, 0)) == -1) {
1181 error = XFS_ERROR(EIO);
1195 * These are simple subroutines used by xlog_clear_stale_blocks() below
1196 * to initialize a buffer full of empty log record headers and write
1197 * them into the log.
1208 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1210 memset(buf, 0, BBSIZE);
1211 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1212 recp->h_cycle = cpu_to_be32(cycle);
1213 recp->h_version = cpu_to_be32(
1214 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1215 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1216 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1217 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1218 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1222 xlog_write_log_records(
1233 int sectbb = log->l_sectBBsize;
1234 int end_block = start_block + blocks;
1240 * Greedily allocate a buffer big enough to handle the full
1241 * range of basic blocks to be written. If that fails, try
1242 * a smaller size. We need to be able to write at least a
1243 * log sector, or we're out of luck.
1245 bufblks = 1 << ffs(blocks);
1246 while (bufblks > log->l_logBBsize)
1248 while (!(bp = xlog_get_bp(log, bufblks))) {
1250 if (bufblks < sectbb)
1254 /* We may need to do a read at the start to fill in part of
1255 * the buffer in the starting sector not covered by the first
1258 balign = round_down(start_block, sectbb);
1259 if (balign != start_block) {
1260 error = xlog_bread_noalign(log, start_block, 1, bp);
1264 j = start_block - balign;
1267 for (i = start_block; i < end_block; i += bufblks) {
1268 int bcount, endcount;
1270 bcount = min(bufblks, end_block - start_block);
1271 endcount = bcount - j;
1273 /* We may need to do a read at the end to fill in part of
1274 * the buffer in the final sector not covered by the write.
1275 * If this is the same sector as the above read, skip it.
1277 ealign = round_down(end_block, sectbb);
1278 if (j == 0 && (start_block + endcount > ealign)) {
1279 offset = bp->b_addr + BBTOB(ealign - start_block);
1280 error = xlog_bread_offset(log, ealign, sectbb,
1287 offset = xlog_align(log, start_block, endcount, bp);
1288 for (; j < endcount; j++) {
1289 xlog_add_record(log, offset, cycle, i+j,
1290 tail_cycle, tail_block);
1293 error = xlog_bwrite(log, start_block, endcount, bp);
1296 start_block += endcount;
1306 * This routine is called to blow away any incomplete log writes out
1307 * in front of the log head. We do this so that we won't become confused
1308 * if we come up, write only a little bit more, and then crash again.
1309 * If we leave the partial log records out there, this situation could
1310 * cause us to think those partial writes are valid blocks since they
1311 * have the current cycle number. We get rid of them by overwriting them
1312 * with empty log records with the old cycle number rather than the
1315 * The tail lsn is passed in rather than taken from
1316 * the log so that we will not write over the unmount record after a
1317 * clean unmount in a 512 block log. Doing so would leave the log without
1318 * any valid log records in it until a new one was written. If we crashed
1319 * during that time we would not be able to recover.
1322 xlog_clear_stale_blocks(
1326 int tail_cycle, head_cycle;
1327 int tail_block, head_block;
1328 int tail_distance, max_distance;
1332 tail_cycle = CYCLE_LSN(tail_lsn);
1333 tail_block = BLOCK_LSN(tail_lsn);
1334 head_cycle = log->l_curr_cycle;
1335 head_block = log->l_curr_block;
1338 * Figure out the distance between the new head of the log
1339 * and the tail. We want to write over any blocks beyond the
1340 * head that we may have written just before the crash, but
1341 * we don't want to overwrite the tail of the log.
1343 if (head_cycle == tail_cycle) {
1345 * The tail is behind the head in the physical log,
1346 * so the distance from the head to the tail is the
1347 * distance from the head to the end of the log plus
1348 * the distance from the beginning of the log to the
1351 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1352 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1353 XFS_ERRLEVEL_LOW, log->l_mp);
1354 return XFS_ERROR(EFSCORRUPTED);
1356 tail_distance = tail_block + (log->l_logBBsize - head_block);
1359 * The head is behind the tail in the physical log,
1360 * so the distance from the head to the tail is just
1361 * the tail block minus the head block.
1363 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1364 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1365 XFS_ERRLEVEL_LOW, log->l_mp);
1366 return XFS_ERROR(EFSCORRUPTED);
1368 tail_distance = tail_block - head_block;
1372 * If the head is right up against the tail, we can't clear
1375 if (tail_distance <= 0) {
1376 ASSERT(tail_distance == 0);
1380 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1382 * Take the smaller of the maximum amount of outstanding I/O
1383 * we could have and the distance to the tail to clear out.
1384 * We take the smaller so that we don't overwrite the tail and
1385 * we don't waste all day writing from the head to the tail
1388 max_distance = MIN(max_distance, tail_distance);
1390 if ((head_block + max_distance) <= log->l_logBBsize) {
1392 * We can stomp all the blocks we need to without
1393 * wrapping around the end of the log. Just do it
1394 * in a single write. Use the cycle number of the
1395 * current cycle minus one so that the log will look like:
1398 error = xlog_write_log_records(log, (head_cycle - 1),
1399 head_block, max_distance, tail_cycle,
1405 * We need to wrap around the end of the physical log in
1406 * order to clear all the blocks. Do it in two separate
1407 * I/Os. The first write should be from the head to the
1408 * end of the physical log, and it should use the current
1409 * cycle number minus one just like above.
1411 distance = log->l_logBBsize - head_block;
1412 error = xlog_write_log_records(log, (head_cycle - 1),
1413 head_block, distance, tail_cycle,
1420 * Now write the blocks at the start of the physical log.
1421 * This writes the remainder of the blocks we want to clear.
1422 * It uses the current cycle number since we're now on the
1423 * same cycle as the head so that we get:
1424 * n ... n ... | n - 1 ...
1425 * ^^^^^ blocks we're writing
1427 distance = max_distance - (log->l_logBBsize - head_block);
1428 error = xlog_write_log_records(log, head_cycle, 0, distance,
1429 tail_cycle, tail_block);
1437 /******************************************************************************
1439 * Log recover routines
1441 ******************************************************************************
1444 STATIC xlog_recover_t *
1445 xlog_recover_find_tid(
1446 struct hlist_head *head,
1449 xlog_recover_t *trans;
1451 hlist_for_each_entry(trans, head, r_list) {
1452 if (trans->r_log_tid == tid)
1459 xlog_recover_new_tid(
1460 struct hlist_head *head,
1464 xlog_recover_t *trans;
1466 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1467 trans->r_log_tid = tid;
1469 INIT_LIST_HEAD(&trans->r_itemq);
1471 INIT_HLIST_NODE(&trans->r_list);
1472 hlist_add_head(&trans->r_list, head);
1476 xlog_recover_add_item(
1477 struct list_head *head)
1479 xlog_recover_item_t *item;
1481 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1482 INIT_LIST_HEAD(&item->ri_list);
1483 list_add_tail(&item->ri_list, head);
1487 xlog_recover_add_to_cont_trans(
1489 struct xlog_recover *trans,
1493 xlog_recover_item_t *item;
1494 xfs_caddr_t ptr, old_ptr;
1497 if (list_empty(&trans->r_itemq)) {
1498 /* finish copying rest of trans header */
1499 xlog_recover_add_item(&trans->r_itemq);
1500 ptr = (xfs_caddr_t) &trans->r_theader +
1501 sizeof(xfs_trans_header_t) - len;
1502 memcpy(ptr, dp, len); /* d, s, l */
1505 /* take the tail entry */
1506 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1508 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1509 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1511 ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP);
1512 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1513 item->ri_buf[item->ri_cnt-1].i_len += len;
1514 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1515 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
1520 * The next region to add is the start of a new region. It could be
1521 * a whole region or it could be the first part of a new region. Because
1522 * of this, the assumption here is that the type and size fields of all
1523 * format structures fit into the first 32 bits of the structure.
1525 * This works because all regions must be 32 bit aligned. Therefore, we
1526 * either have both fields or we have neither field. In the case we have
1527 * neither field, the data part of the region is zero length. We only have
1528 * a log_op_header and can throw away the header since a new one will appear
1529 * later. If we have at least 4 bytes, then we can determine how many regions
1530 * will appear in the current log item.
1533 xlog_recover_add_to_trans(
1535 struct xlog_recover *trans,
1539 xfs_inode_log_format_t *in_f; /* any will do */
1540 xlog_recover_item_t *item;
1545 if (list_empty(&trans->r_itemq)) {
1546 /* we need to catch log corruptions here */
1547 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
1548 xfs_warn(log->l_mp, "%s: bad header magic number",
1551 return XFS_ERROR(EIO);
1553 if (len == sizeof(xfs_trans_header_t))
1554 xlog_recover_add_item(&trans->r_itemq);
1555 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1559 ptr = kmem_alloc(len, KM_SLEEP);
1560 memcpy(ptr, dp, len);
1561 in_f = (xfs_inode_log_format_t *)ptr;
1563 /* take the tail entry */
1564 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1565 if (item->ri_total != 0 &&
1566 item->ri_total == item->ri_cnt) {
1567 /* tail item is in use, get a new one */
1568 xlog_recover_add_item(&trans->r_itemq);
1569 item = list_entry(trans->r_itemq.prev,
1570 xlog_recover_item_t, ri_list);
1573 if (item->ri_total == 0) { /* first region to be added */
1574 if (in_f->ilf_size == 0 ||
1575 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
1577 "bad number of regions (%d) in inode log format",
1581 return XFS_ERROR(EIO);
1584 item->ri_total = in_f->ilf_size;
1586 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1589 ASSERT(item->ri_total > item->ri_cnt);
1590 /* Description region is ri_buf[0] */
1591 item->ri_buf[item->ri_cnt].i_addr = ptr;
1592 item->ri_buf[item->ri_cnt].i_len = len;
1594 trace_xfs_log_recover_item_add(log, trans, item, 0);
1599 * Sort the log items in the transaction.
1601 * The ordering constraints are defined by the inode allocation and unlink
1602 * behaviour. The rules are:
1604 * 1. Every item is only logged once in a given transaction. Hence it
1605 * represents the last logged state of the item. Hence ordering is
1606 * dependent on the order in which operations need to be performed so
1607 * required initial conditions are always met.
1609 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1610 * there's nothing to replay from them so we can simply cull them
1611 * from the transaction. However, we can't do that until after we've
1612 * replayed all the other items because they may be dependent on the
1613 * cancelled buffer and replaying the cancelled buffer can remove it
1614 * form the cancelled buffer table. Hence they have tobe done last.
1616 * 3. Inode allocation buffers must be replayed before inode items that
1617 * read the buffer and replay changes into it. For filesystems using the
1618 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1619 * treated the same as inode allocation buffers as they create and
1620 * initialise the buffers directly.
1622 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1623 * This ensures that inodes are completely flushed to the inode buffer
1624 * in a "free" state before we remove the unlinked inode list pointer.
1626 * Hence the ordering needs to be inode allocation buffers first, inode items
1627 * second, inode unlink buffers third and cancelled buffers last.
1629 * But there's a problem with that - we can't tell an inode allocation buffer
1630 * apart from a regular buffer, so we can't separate them. We can, however,
1631 * tell an inode unlink buffer from the others, and so we can separate them out
1632 * from all the other buffers and move them to last.
1634 * Hence, 4 lists, in order from head to tail:
1635 * - buffer_list for all buffers except cancelled/inode unlink buffers
1636 * - item_list for all non-buffer items
1637 * - inode_buffer_list for inode unlink buffers
1638 * - cancel_list for the cancelled buffers
1640 * Note that we add objects to the tail of the lists so that first-to-last
1641 * ordering is preserved within the lists. Adding objects to the head of the
1642 * list means when we traverse from the head we walk them in last-to-first
1643 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1644 * but for all other items there may be specific ordering that we need to
1648 xlog_recover_reorder_trans(
1650 struct xlog_recover *trans,
1653 xlog_recover_item_t *item, *n;
1654 LIST_HEAD(sort_list);
1655 LIST_HEAD(cancel_list);
1656 LIST_HEAD(buffer_list);
1657 LIST_HEAD(inode_buffer_list);
1658 LIST_HEAD(inode_list);
1660 list_splice_init(&trans->r_itemq, &sort_list);
1661 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1662 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1664 switch (ITEM_TYPE(item)) {
1665 case XFS_LI_ICREATE:
1666 list_move_tail(&item->ri_list, &buffer_list);
1669 if (buf_f->blf_flags & XFS_BLF_CANCEL) {
1670 trace_xfs_log_recover_item_reorder_head(log,
1672 list_move(&item->ri_list, &cancel_list);
1675 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1676 list_move(&item->ri_list, &inode_buffer_list);
1679 list_move_tail(&item->ri_list, &buffer_list);
1683 case XFS_LI_QUOTAOFF:
1686 trace_xfs_log_recover_item_reorder_tail(log,
1688 list_move_tail(&item->ri_list, &inode_list);
1692 "%s: unrecognized type of log operation",
1695 return XFS_ERROR(EIO);
1698 ASSERT(list_empty(&sort_list));
1699 if (!list_empty(&buffer_list))
1700 list_splice(&buffer_list, &trans->r_itemq);
1701 if (!list_empty(&inode_list))
1702 list_splice_tail(&inode_list, &trans->r_itemq);
1703 if (!list_empty(&inode_buffer_list))
1704 list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1705 if (!list_empty(&cancel_list))
1706 list_splice_tail(&cancel_list, &trans->r_itemq);
1711 * Build up the table of buf cancel records so that we don't replay
1712 * cancelled data in the second pass. For buffer records that are
1713 * not cancel records, there is nothing to do here so we just return.
1715 * If we get a cancel record which is already in the table, this indicates
1716 * that the buffer was cancelled multiple times. In order to ensure
1717 * that during pass 2 we keep the record in the table until we reach its
1718 * last occurrence in the log, we keep a reference count in the cancel
1719 * record in the table to tell us how many times we expect to see this
1720 * record during the second pass.
1723 xlog_recover_buffer_pass1(
1725 struct xlog_recover_item *item)
1727 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1728 struct list_head *bucket;
1729 struct xfs_buf_cancel *bcp;
1732 * If this isn't a cancel buffer item, then just return.
1734 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1735 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
1740 * Insert an xfs_buf_cancel record into the hash table of them.
1741 * If there is already an identical record, bump its reference count.
1743 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1744 list_for_each_entry(bcp, bucket, bc_list) {
1745 if (bcp->bc_blkno == buf_f->blf_blkno &&
1746 bcp->bc_len == buf_f->blf_len) {
1748 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
1753 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1754 bcp->bc_blkno = buf_f->blf_blkno;
1755 bcp->bc_len = buf_f->blf_len;
1756 bcp->bc_refcount = 1;
1757 list_add_tail(&bcp->bc_list, bucket);
1759 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
1764 * Check to see whether the buffer being recovered has a corresponding
1765 * entry in the buffer cancel record table. If it is, return the cancel
1766 * buffer structure to the caller.
1768 STATIC struct xfs_buf_cancel *
1769 xlog_peek_buffer_cancelled(
1775 struct list_head *bucket;
1776 struct xfs_buf_cancel *bcp;
1778 if (!log->l_buf_cancel_table) {
1779 /* empty table means no cancelled buffers in the log */
1780 ASSERT(!(flags & XFS_BLF_CANCEL));
1784 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1785 list_for_each_entry(bcp, bucket, bc_list) {
1786 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
1791 * We didn't find a corresponding entry in the table, so return 0 so
1792 * that the buffer is NOT cancelled.
1794 ASSERT(!(flags & XFS_BLF_CANCEL));
1799 * If the buffer is being cancelled then return 1 so that it will be cancelled,
1800 * otherwise return 0. If the buffer is actually a buffer cancel item
1801 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
1802 * table and remove it from the table if this is the last reference.
1804 * We remove the cancel record from the table when we encounter its last
1805 * occurrence in the log so that if the same buffer is re-used again after its
1806 * last cancellation we actually replay the changes made at that point.
1809 xlog_check_buffer_cancelled(
1815 struct xfs_buf_cancel *bcp;
1817 bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags);
1822 * We've go a match, so return 1 so that the recovery of this buffer
1823 * is cancelled. If this buffer is actually a buffer cancel log
1824 * item, then decrement the refcount on the one in the table and
1825 * remove it if this is the last reference.
1827 if (flags & XFS_BLF_CANCEL) {
1828 if (--bcp->bc_refcount == 0) {
1829 list_del(&bcp->bc_list);
1837 * Perform recovery for a buffer full of inodes. In these buffers, the only
1838 * data which should be recovered is that which corresponds to the
1839 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1840 * data for the inodes is always logged through the inodes themselves rather
1841 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1843 * The only time when buffers full of inodes are fully recovered is when the
1844 * buffer is full of newly allocated inodes. In this case the buffer will
1845 * not be marked as an inode buffer and so will be sent to
1846 * xlog_recover_do_reg_buffer() below during recovery.
1849 xlog_recover_do_inode_buffer(
1850 struct xfs_mount *mp,
1851 xlog_recover_item_t *item,
1853 xfs_buf_log_format_t *buf_f)
1859 int reg_buf_offset = 0;
1860 int reg_buf_bytes = 0;
1861 int next_unlinked_offset;
1863 xfs_agino_t *logged_nextp;
1864 xfs_agino_t *buffer_nextp;
1866 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
1869 * Post recovery validation only works properly on CRC enabled
1872 if (xfs_sb_version_hascrc(&mp->m_sb))
1873 bp->b_ops = &xfs_inode_buf_ops;
1875 inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog;
1876 for (i = 0; i < inodes_per_buf; i++) {
1877 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1878 offsetof(xfs_dinode_t, di_next_unlinked);
1880 while (next_unlinked_offset >=
1881 (reg_buf_offset + reg_buf_bytes)) {
1883 * The next di_next_unlinked field is beyond
1884 * the current logged region. Find the next
1885 * logged region that contains or is beyond
1886 * the current di_next_unlinked field.
1889 bit = xfs_next_bit(buf_f->blf_data_map,
1890 buf_f->blf_map_size, bit);
1893 * If there are no more logged regions in the
1894 * buffer, then we're done.
1899 nbits = xfs_contig_bits(buf_f->blf_data_map,
1900 buf_f->blf_map_size, bit);
1902 reg_buf_offset = bit << XFS_BLF_SHIFT;
1903 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
1908 * If the current logged region starts after the current
1909 * di_next_unlinked field, then move on to the next
1910 * di_next_unlinked field.
1912 if (next_unlinked_offset < reg_buf_offset)
1915 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1916 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
1917 ASSERT((reg_buf_offset + reg_buf_bytes) <=
1918 BBTOB(bp->b_io_length));
1921 * The current logged region contains a copy of the
1922 * current di_next_unlinked field. Extract its value
1923 * and copy it to the buffer copy.
1925 logged_nextp = item->ri_buf[item_index].i_addr +
1926 next_unlinked_offset - reg_buf_offset;
1927 if (unlikely(*logged_nextp == 0)) {
1929 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1930 "Trying to replay bad (0) inode di_next_unlinked field.",
1932 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1933 XFS_ERRLEVEL_LOW, mp);
1934 return XFS_ERROR(EFSCORRUPTED);
1937 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1938 next_unlinked_offset);
1939 *buffer_nextp = *logged_nextp;
1942 * If necessary, recalculate the CRC in the on-disk inode. We
1943 * have to leave the inode in a consistent state for whoever
1946 xfs_dinode_calc_crc(mp, (struct xfs_dinode *)
1947 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
1955 * V5 filesystems know the age of the buffer on disk being recovered. We can
1956 * have newer objects on disk than we are replaying, and so for these cases we
1957 * don't want to replay the current change as that will make the buffer contents
1958 * temporarily invalid on disk.
1960 * The magic number might not match the buffer type we are going to recover
1961 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
1962 * extract the LSN of the existing object in the buffer based on it's current
1963 * magic number. If we don't recognise the magic number in the buffer, then
1964 * return a LSN of -1 so that the caller knows it was an unrecognised block and
1965 * so can recover the buffer.
1967 * Note: we cannot rely solely on magic number matches to determine that the
1968 * buffer has a valid LSN - we also need to verify that it belongs to this
1969 * filesystem, so we need to extract the object's LSN and compare it to that
1970 * which we read from the superblock. If the UUIDs don't match, then we've got a
1971 * stale metadata block from an old filesystem instance that we need to recover
1975 xlog_recover_get_buf_lsn(
1976 struct xfs_mount *mp,
1982 void *blk = bp->b_addr;
1986 /* v4 filesystems always recover immediately */
1987 if (!xfs_sb_version_hascrc(&mp->m_sb))
1988 goto recover_immediately;
1990 magic32 = be32_to_cpu(*(__be32 *)blk);
1992 case XFS_ABTB_CRC_MAGIC:
1993 case XFS_ABTC_CRC_MAGIC:
1994 case XFS_ABTB_MAGIC:
1995 case XFS_ABTC_MAGIC:
1996 case XFS_IBT_CRC_MAGIC:
1997 case XFS_IBT_MAGIC: {
1998 struct xfs_btree_block *btb = blk;
2000 lsn = be64_to_cpu(btb->bb_u.s.bb_lsn);
2001 uuid = &btb->bb_u.s.bb_uuid;
2004 case XFS_BMAP_CRC_MAGIC:
2005 case XFS_BMAP_MAGIC: {
2006 struct xfs_btree_block *btb = blk;
2008 lsn = be64_to_cpu(btb->bb_u.l.bb_lsn);
2009 uuid = &btb->bb_u.l.bb_uuid;
2013 lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn);
2014 uuid = &((struct xfs_agf *)blk)->agf_uuid;
2016 case XFS_AGFL_MAGIC:
2017 lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn);
2018 uuid = &((struct xfs_agfl *)blk)->agfl_uuid;
2021 lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn);
2022 uuid = &((struct xfs_agi *)blk)->agi_uuid;
2024 case XFS_SYMLINK_MAGIC:
2025 lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn);
2026 uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid;
2028 case XFS_DIR3_BLOCK_MAGIC:
2029 case XFS_DIR3_DATA_MAGIC:
2030 case XFS_DIR3_FREE_MAGIC:
2031 lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn);
2032 uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid;
2034 case XFS_ATTR3_RMT_MAGIC:
2035 lsn = be64_to_cpu(((struct xfs_attr3_rmt_hdr *)blk)->rm_lsn);
2036 uuid = &((struct xfs_attr3_rmt_hdr *)blk)->rm_uuid;
2039 lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn);
2040 uuid = &((struct xfs_dsb *)blk)->sb_uuid;
2046 if (lsn != (xfs_lsn_t)-1) {
2047 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
2048 goto recover_immediately;
2052 magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic);
2054 case XFS_DIR3_LEAF1_MAGIC:
2055 case XFS_DIR3_LEAFN_MAGIC:
2056 case XFS_DA3_NODE_MAGIC:
2057 lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn);
2058 uuid = &((struct xfs_da3_blkinfo *)blk)->uuid;
2064 if (lsn != (xfs_lsn_t)-1) {
2065 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
2066 goto recover_immediately;
2071 * We do individual object checks on dquot and inode buffers as they
2072 * have their own individual LSN records. Also, we could have a stale
2073 * buffer here, so we have to at least recognise these buffer types.
2075 * A notd complexity here is inode unlinked list processing - it logs
2076 * the inode directly in the buffer, but we don't know which inodes have
2077 * been modified, and there is no global buffer LSN. Hence we need to
2078 * recover all inode buffer types immediately. This problem will be
2079 * fixed by logical logging of the unlinked list modifications.
2081 magic16 = be16_to_cpu(*(__be16 *)blk);
2083 case XFS_DQUOT_MAGIC:
2084 case XFS_DINODE_MAGIC:
2085 goto recover_immediately;
2090 /* unknown buffer contents, recover immediately */
2092 recover_immediately:
2093 return (xfs_lsn_t)-1;
2098 * Validate the recovered buffer is of the correct type and attach the
2099 * appropriate buffer operations to them for writeback. Magic numbers are in a
2101 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2102 * the first 32 bits of the buffer (most blocks),
2103 * inside a struct xfs_da_blkinfo at the start of the buffer.
2106 xlog_recover_validate_buf_type(
2107 struct xfs_mount *mp,
2109 xfs_buf_log_format_t *buf_f)
2111 struct xfs_da_blkinfo *info = bp->b_addr;
2116 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
2117 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
2118 magicda = be16_to_cpu(info->magic);
2119 switch (xfs_blft_from_flags(buf_f)) {
2120 case XFS_BLFT_BTREE_BUF:
2122 case XFS_ABTB_CRC_MAGIC:
2123 case XFS_ABTC_CRC_MAGIC:
2124 case XFS_ABTB_MAGIC:
2125 case XFS_ABTC_MAGIC:
2126 bp->b_ops = &xfs_allocbt_buf_ops;
2128 case XFS_IBT_CRC_MAGIC:
2130 bp->b_ops = &xfs_inobt_buf_ops;
2132 case XFS_BMAP_CRC_MAGIC:
2133 case XFS_BMAP_MAGIC:
2134 bp->b_ops = &xfs_bmbt_buf_ops;
2137 xfs_warn(mp, "Bad btree block magic!");
2142 case XFS_BLFT_AGF_BUF:
2143 if (magic32 != XFS_AGF_MAGIC) {
2144 xfs_warn(mp, "Bad AGF block magic!");
2148 bp->b_ops = &xfs_agf_buf_ops;
2150 case XFS_BLFT_AGFL_BUF:
2151 if (!xfs_sb_version_hascrc(&mp->m_sb))
2153 if (magic32 != XFS_AGFL_MAGIC) {
2154 xfs_warn(mp, "Bad AGFL block magic!");
2158 bp->b_ops = &xfs_agfl_buf_ops;
2160 case XFS_BLFT_AGI_BUF:
2161 if (magic32 != XFS_AGI_MAGIC) {
2162 xfs_warn(mp, "Bad AGI block magic!");
2166 bp->b_ops = &xfs_agi_buf_ops;
2168 case XFS_BLFT_UDQUOT_BUF:
2169 case XFS_BLFT_PDQUOT_BUF:
2170 case XFS_BLFT_GDQUOT_BUF:
2171 #ifdef CONFIG_XFS_QUOTA
2172 if (magic16 != XFS_DQUOT_MAGIC) {
2173 xfs_warn(mp, "Bad DQUOT block magic!");
2177 bp->b_ops = &xfs_dquot_buf_ops;
2180 "Trying to recover dquots without QUOTA support built in!");
2184 case XFS_BLFT_DINO_BUF:
2186 * we get here with inode allocation buffers, not buffers that
2187 * track unlinked list changes.
2189 if (magic16 != XFS_DINODE_MAGIC) {
2190 xfs_warn(mp, "Bad INODE block magic!");
2194 bp->b_ops = &xfs_inode_buf_ops;
2196 case XFS_BLFT_SYMLINK_BUF:
2197 if (magic32 != XFS_SYMLINK_MAGIC) {
2198 xfs_warn(mp, "Bad symlink block magic!");
2202 bp->b_ops = &xfs_symlink_buf_ops;
2204 case XFS_BLFT_DIR_BLOCK_BUF:
2205 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2206 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2207 xfs_warn(mp, "Bad dir block magic!");
2211 bp->b_ops = &xfs_dir3_block_buf_ops;
2213 case XFS_BLFT_DIR_DATA_BUF:
2214 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2215 magic32 != XFS_DIR3_DATA_MAGIC) {
2216 xfs_warn(mp, "Bad dir data magic!");
2220 bp->b_ops = &xfs_dir3_data_buf_ops;
2222 case XFS_BLFT_DIR_FREE_BUF:
2223 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2224 magic32 != XFS_DIR3_FREE_MAGIC) {
2225 xfs_warn(mp, "Bad dir3 free magic!");
2229 bp->b_ops = &xfs_dir3_free_buf_ops;
2231 case XFS_BLFT_DIR_LEAF1_BUF:
2232 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2233 magicda != XFS_DIR3_LEAF1_MAGIC) {
2234 xfs_warn(mp, "Bad dir leaf1 magic!");
2238 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2240 case XFS_BLFT_DIR_LEAFN_BUF:
2241 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2242 magicda != XFS_DIR3_LEAFN_MAGIC) {
2243 xfs_warn(mp, "Bad dir leafn magic!");
2247 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2249 case XFS_BLFT_DA_NODE_BUF:
2250 if (magicda != XFS_DA_NODE_MAGIC &&
2251 magicda != XFS_DA3_NODE_MAGIC) {
2252 xfs_warn(mp, "Bad da node magic!");
2256 bp->b_ops = &xfs_da3_node_buf_ops;
2258 case XFS_BLFT_ATTR_LEAF_BUF:
2259 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2260 magicda != XFS_ATTR3_LEAF_MAGIC) {
2261 xfs_warn(mp, "Bad attr leaf magic!");
2265 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2267 case XFS_BLFT_ATTR_RMT_BUF:
2268 if (!xfs_sb_version_hascrc(&mp->m_sb))
2270 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
2271 xfs_warn(mp, "Bad attr remote magic!");
2275 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2277 case XFS_BLFT_SB_BUF:
2278 if (magic32 != XFS_SB_MAGIC) {
2279 xfs_warn(mp, "Bad SB block magic!");
2283 bp->b_ops = &xfs_sb_buf_ops;
2286 xfs_warn(mp, "Unknown buffer type %d!",
2287 xfs_blft_from_flags(buf_f));
2293 * Perform a 'normal' buffer recovery. Each logged region of the
2294 * buffer should be copied over the corresponding region in the
2295 * given buffer. The bitmap in the buf log format structure indicates
2296 * where to place the logged data.
2299 xlog_recover_do_reg_buffer(
2300 struct xfs_mount *mp,
2301 xlog_recover_item_t *item,
2303 xfs_buf_log_format_t *buf_f)
2310 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2313 i = 1; /* 0 is the buf format structure */
2315 bit = xfs_next_bit(buf_f->blf_data_map,
2316 buf_f->blf_map_size, bit);
2319 nbits = xfs_contig_bits(buf_f->blf_data_map,
2320 buf_f->blf_map_size, bit);
2322 ASSERT(item->ri_buf[i].i_addr != NULL);
2323 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
2324 ASSERT(BBTOB(bp->b_io_length) >=
2325 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
2328 * The dirty regions logged in the buffer, even though
2329 * contiguous, may span multiple chunks. This is because the
2330 * dirty region may span a physical page boundary in a buffer
2331 * and hence be split into two separate vectors for writing into
2332 * the log. Hence we need to trim nbits back to the length of
2333 * the current region being copied out of the log.
2335 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2336 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2339 * Do a sanity check if this is a dquot buffer. Just checking
2340 * the first dquot in the buffer should do. XXXThis is
2341 * probably a good thing to do for other buf types also.
2344 if (buf_f->blf_flags &
2345 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2346 if (item->ri_buf[i].i_addr == NULL) {
2348 "XFS: NULL dquot in %s.", __func__);
2351 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
2353 "XFS: dquot too small (%d) in %s.",
2354 item->ri_buf[i].i_len, __func__);
2357 error = xfs_dqcheck(mp, item->ri_buf[i].i_addr,
2358 -1, 0, XFS_QMOPT_DOWARN,
2359 "dquot_buf_recover");
2364 memcpy(xfs_buf_offset(bp,
2365 (uint)bit << XFS_BLF_SHIFT), /* dest */
2366 item->ri_buf[i].i_addr, /* source */
2367 nbits<<XFS_BLF_SHIFT); /* length */
2373 /* Shouldn't be any more regions */
2374 ASSERT(i == item->ri_total);
2377 * We can only do post recovery validation on items on CRC enabled
2378 * fielsystems as we need to know when the buffer was written to be able
2379 * to determine if we should have replayed the item. If we replay old
2380 * metadata over a newer buffer, then it will enter a temporarily
2381 * inconsistent state resulting in verification failures. Hence for now
2382 * just avoid the verification stage for non-crc filesystems
2384 if (xfs_sb_version_hascrc(&mp->m_sb))
2385 xlog_recover_validate_buf_type(mp, bp, buf_f);
2389 * Perform a dquot buffer recovery.
2390 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2391 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2392 * Else, treat it as a regular buffer and do recovery.
2395 xlog_recover_do_dquot_buffer(
2396 struct xfs_mount *mp,
2398 struct xlog_recover_item *item,
2400 struct xfs_buf_log_format *buf_f)
2404 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2407 * Filesystems are required to send in quota flags at mount time.
2409 if (mp->m_qflags == 0) {
2414 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2415 type |= XFS_DQ_USER;
2416 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2417 type |= XFS_DQ_PROJ;
2418 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2419 type |= XFS_DQ_GROUP;
2421 * This type of quotas was turned off, so ignore this buffer
2423 if (log->l_quotaoffs_flag & type)
2426 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2430 * This routine replays a modification made to a buffer at runtime.
2431 * There are actually two types of buffer, regular and inode, which
2432 * are handled differently. Inode buffers are handled differently
2433 * in that we only recover a specific set of data from them, namely
2434 * the inode di_next_unlinked fields. This is because all other inode
2435 * data is actually logged via inode records and any data we replay
2436 * here which overlaps that may be stale.
2438 * When meta-data buffers are freed at run time we log a buffer item
2439 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2440 * of the buffer in the log should not be replayed at recovery time.
2441 * This is so that if the blocks covered by the buffer are reused for
2442 * file data before we crash we don't end up replaying old, freed
2443 * meta-data into a user's file.
2445 * To handle the cancellation of buffer log items, we make two passes
2446 * over the log during recovery. During the first we build a table of
2447 * those buffers which have been cancelled, and during the second we
2448 * only replay those buffers which do not have corresponding cancel
2449 * records in the table. See xlog_recover_buffer_pass[1,2] above
2450 * for more details on the implementation of the table of cancel records.
2453 xlog_recover_buffer_pass2(
2455 struct list_head *buffer_list,
2456 struct xlog_recover_item *item,
2457 xfs_lsn_t current_lsn)
2459 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2460 xfs_mount_t *mp = log->l_mp;
2467 * In this pass we only want to recover all the buffers which have
2468 * not been cancelled and are not cancellation buffers themselves.
2470 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2471 buf_f->blf_len, buf_f->blf_flags)) {
2472 trace_xfs_log_recover_buf_cancel(log, buf_f);
2476 trace_xfs_log_recover_buf_recover(log, buf_f);
2479 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2480 buf_flags |= XBF_UNMAPPED;
2482 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2485 return XFS_ERROR(ENOMEM);
2486 error = bp->b_error;
2488 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
2493 * recover the buffer only if we get an LSN from it and it's less than
2494 * the lsn of the transaction we are replaying.
2496 lsn = xlog_recover_get_buf_lsn(mp, bp);
2497 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0)
2500 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
2501 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2502 } else if (buf_f->blf_flags &
2503 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2504 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2506 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2512 * Perform delayed write on the buffer. Asynchronous writes will be
2513 * slower when taking into account all the buffers to be flushed.
2515 * Also make sure that only inode buffers with good sizes stay in
2516 * the buffer cache. The kernel moves inodes in buffers of 1 block
2517 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2518 * buffers in the log can be a different size if the log was generated
2519 * by an older kernel using unclustered inode buffers or a newer kernel
2520 * running with a different inode cluster size. Regardless, if the
2521 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2522 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2523 * the buffer out of the buffer cache so that the buffer won't
2524 * overlap with future reads of those inodes.
2526 if (XFS_DINODE_MAGIC ==
2527 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2528 (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize,
2529 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2531 error = xfs_bwrite(bp);
2533 ASSERT(bp->b_target->bt_mount == mp);
2534 bp->b_iodone = xlog_recover_iodone;
2535 xfs_buf_delwri_queue(bp, buffer_list);
2544 * Inode fork owner changes
2546 * If we have been told that we have to reparent the inode fork, it's because an
2547 * extent swap operation on a CRC enabled filesystem has been done and we are
2548 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2551 * The complexity here is that we don't have an inode context to work with, so
2552 * after we've replayed the inode we need to instantiate one. This is where the
2555 * We are in the middle of log recovery, so we can't run transactions. That
2556 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2557 * that will result in the corresponding iput() running the inode through
2558 * xfs_inactive(). If we've just replayed an inode core that changes the link
2559 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2560 * transactions (bad!).
2562 * So, to avoid this, we instantiate an inode directly from the inode core we've
2563 * just recovered. We have the buffer still locked, and all we really need to
2564 * instantiate is the inode core and the forks being modified. We can do this
2565 * manually, then run the inode btree owner change, and then tear down the
2566 * xfs_inode without having to run any transactions at all.
2568 * Also, because we don't have a transaction context available here but need to
2569 * gather all the buffers we modify for writeback so we pass the buffer_list
2570 * instead for the operation to use.
2574 xfs_recover_inode_owner_change(
2575 struct xfs_mount *mp,
2576 struct xfs_dinode *dip,
2577 struct xfs_inode_log_format *in_f,
2578 struct list_head *buffer_list)
2580 struct xfs_inode *ip;
2583 ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER));
2585 ip = xfs_inode_alloc(mp, in_f->ilf_ino);
2589 /* instantiate the inode */
2590 xfs_dinode_from_disk(&ip->i_d, dip);
2591 ASSERT(ip->i_d.di_version >= 3);
2593 error = xfs_iformat_fork(ip, dip);
2598 if (in_f->ilf_fields & XFS_ILOG_DOWNER) {
2599 ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT);
2600 error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK,
2601 ip->i_ino, buffer_list);
2606 if (in_f->ilf_fields & XFS_ILOG_AOWNER) {
2607 ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT);
2608 error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK,
2609 ip->i_ino, buffer_list);
2620 xlog_recover_inode_pass2(
2622 struct list_head *buffer_list,
2623 struct xlog_recover_item *item,
2624 xfs_lsn_t current_lsn)
2626 xfs_inode_log_format_t *in_f;
2627 xfs_mount_t *mp = log->l_mp;
2636 xfs_icdinode_t *dicp;
2640 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2641 in_f = item->ri_buf[0].i_addr;
2643 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
2645 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2651 * Inode buffers can be freed, look out for it,
2652 * and do not replay the inode.
2654 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2655 in_f->ilf_len, 0)) {
2657 trace_xfs_log_recover_inode_cancel(log, in_f);
2660 trace_xfs_log_recover_inode_recover(log, in_f);
2662 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
2663 &xfs_inode_buf_ops);
2668 error = bp->b_error;
2670 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
2673 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2674 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
2677 * Make sure the place we're flushing out to really looks
2680 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
2682 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2683 __func__, dip, bp, in_f->ilf_ino);
2684 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2685 XFS_ERRLEVEL_LOW, mp);
2686 error = EFSCORRUPTED;
2689 dicp = item->ri_buf[1].i_addr;
2690 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2692 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2693 __func__, item, in_f->ilf_ino);
2694 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2695 XFS_ERRLEVEL_LOW, mp);
2696 error = EFSCORRUPTED;
2701 * If the inode has an LSN in it, recover the inode only if it's less
2702 * than the lsn of the transaction we are replaying. Note: we still
2703 * need to replay an owner change even though the inode is more recent
2704 * than the transaction as there is no guarantee that all the btree
2705 * blocks are more recent than this transaction, too.
2707 if (dip->di_version >= 3) {
2708 xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn);
2710 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2711 trace_xfs_log_recover_inode_skip(log, in_f);
2713 goto out_owner_change;
2718 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2719 * are transactional and if ordering is necessary we can determine that
2720 * more accurately by the LSN field in the V3 inode core. Don't trust
2721 * the inode versions we might be changing them here - use the
2722 * superblock flag to determine whether we need to look at di_flushiter
2723 * to skip replay when the on disk inode is newer than the log one
2725 if (!xfs_sb_version_hascrc(&mp->m_sb) &&
2726 dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
2728 * Deal with the wrap case, DI_MAX_FLUSH is less
2729 * than smaller numbers
2731 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
2732 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2735 trace_xfs_log_recover_inode_skip(log, in_f);
2741 /* Take the opportunity to reset the flush iteration count */
2742 dicp->di_flushiter = 0;
2744 if (unlikely(S_ISREG(dicp->di_mode))) {
2745 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2746 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2747 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2748 XFS_ERRLEVEL_LOW, mp, dicp);
2750 "%s: Bad regular inode log record, rec ptr 0x%p, "
2751 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2752 __func__, item, dip, bp, in_f->ilf_ino);
2753 error = EFSCORRUPTED;
2756 } else if (unlikely(S_ISDIR(dicp->di_mode))) {
2757 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2758 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2759 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2760 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2761 XFS_ERRLEVEL_LOW, mp, dicp);
2763 "%s: Bad dir inode log record, rec ptr 0x%p, "
2764 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2765 __func__, item, dip, bp, in_f->ilf_ino);
2766 error = EFSCORRUPTED;
2770 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2771 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2772 XFS_ERRLEVEL_LOW, mp, dicp);
2774 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2775 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2776 __func__, item, dip, bp, in_f->ilf_ino,
2777 dicp->di_nextents + dicp->di_anextents,
2779 error = EFSCORRUPTED;
2782 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2783 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2784 XFS_ERRLEVEL_LOW, mp, dicp);
2786 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2787 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
2788 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
2789 error = EFSCORRUPTED;
2792 isize = xfs_icdinode_size(dicp->di_version);
2793 if (unlikely(item->ri_buf[1].i_len > isize)) {
2794 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
2795 XFS_ERRLEVEL_LOW, mp, dicp);
2797 "%s: Bad inode log record length %d, rec ptr 0x%p",
2798 __func__, item->ri_buf[1].i_len, item);
2799 error = EFSCORRUPTED;
2803 /* The core is in in-core format */
2804 xfs_dinode_to_disk(dip, dicp);
2806 /* the rest is in on-disk format */
2807 if (item->ri_buf[1].i_len > isize) {
2808 memcpy((char *)dip + isize,
2809 item->ri_buf[1].i_addr + isize,
2810 item->ri_buf[1].i_len - isize);
2813 fields = in_f->ilf_fields;
2814 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2816 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
2819 memcpy(XFS_DFORK_DPTR(dip),
2820 &in_f->ilf_u.ilfu_uuid,
2825 if (in_f->ilf_size == 2)
2826 goto out_owner_change;
2827 len = item->ri_buf[2].i_len;
2828 src = item->ri_buf[2].i_addr;
2829 ASSERT(in_f->ilf_size <= 4);
2830 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2831 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2832 (len == in_f->ilf_dsize));
2834 switch (fields & XFS_ILOG_DFORK) {
2835 case XFS_ILOG_DDATA:
2837 memcpy(XFS_DFORK_DPTR(dip), src, len);
2840 case XFS_ILOG_DBROOT:
2841 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
2842 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
2843 XFS_DFORK_DSIZE(dip, mp));
2848 * There are no data fork flags set.
2850 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2855 * If we logged any attribute data, recover it. There may or
2856 * may not have been any other non-core data logged in this
2859 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2860 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2865 len = item->ri_buf[attr_index].i_len;
2866 src = item->ri_buf[attr_index].i_addr;
2867 ASSERT(len == in_f->ilf_asize);
2869 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2870 case XFS_ILOG_ADATA:
2872 dest = XFS_DFORK_APTR(dip);
2873 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2874 memcpy(dest, src, len);
2877 case XFS_ILOG_ABROOT:
2878 dest = XFS_DFORK_APTR(dip);
2879 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2880 len, (xfs_bmdr_block_t*)dest,
2881 XFS_DFORK_ASIZE(dip, mp));
2885 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
2893 if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER))
2894 error = xfs_recover_inode_owner_change(mp, dip, in_f,
2896 /* re-generate the checksum. */
2897 xfs_dinode_calc_crc(log->l_mp, dip);
2899 ASSERT(bp->b_target->bt_mount == mp);
2900 bp->b_iodone = xlog_recover_iodone;
2901 xfs_buf_delwri_queue(bp, buffer_list);
2908 return XFS_ERROR(error);
2912 * Recover QUOTAOFF records. We simply make a note of it in the xlog
2913 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2917 xlog_recover_quotaoff_pass1(
2919 struct xlog_recover_item *item)
2921 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
2925 * The logitem format's flag tells us if this was user quotaoff,
2926 * group/project quotaoff or both.
2928 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2929 log->l_quotaoffs_flag |= XFS_DQ_USER;
2930 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2931 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2932 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2933 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2939 * Recover a dquot record
2942 xlog_recover_dquot_pass2(
2944 struct list_head *buffer_list,
2945 struct xlog_recover_item *item,
2946 xfs_lsn_t current_lsn)
2948 xfs_mount_t *mp = log->l_mp;
2950 struct xfs_disk_dquot *ddq, *recddq;
2952 xfs_dq_logformat_t *dq_f;
2957 * Filesystems are required to send in quota flags at mount time.
2959 if (mp->m_qflags == 0)
2962 recddq = item->ri_buf[1].i_addr;
2963 if (recddq == NULL) {
2964 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
2965 return XFS_ERROR(EIO);
2967 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
2968 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
2969 item->ri_buf[1].i_len, __func__);
2970 return XFS_ERROR(EIO);
2974 * This type of quotas was turned off, so ignore this record.
2976 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
2978 if (log->l_quotaoffs_flag & type)
2982 * At this point we know that quota was _not_ turned off.
2983 * Since the mount flags are not indicating to us otherwise, this
2984 * must mean that quota is on, and the dquot needs to be replayed.
2985 * Remember that we may not have fully recovered the superblock yet,
2986 * so we can't do the usual trick of looking at the SB quota bits.
2988 * The other possibility, of course, is that the quota subsystem was
2989 * removed since the last mount - ENOSYS.
2991 dq_f = item->ri_buf[0].i_addr;
2993 error = xfs_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2994 "xlog_recover_dquot_pass2 (log copy)");
2996 return XFS_ERROR(EIO);
2997 ASSERT(dq_f->qlf_len == 1);
2999 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
3000 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
3006 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
3009 * At least the magic num portion should be on disk because this
3010 * was among a chunk of dquots created earlier, and we did some
3011 * minimal initialization then.
3013 error = xfs_dqcheck(mp, ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
3014 "xlog_recover_dquot_pass2");
3017 return XFS_ERROR(EIO);
3021 * If the dquot has an LSN in it, recover the dquot only if it's less
3022 * than the lsn of the transaction we are replaying.
3024 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3025 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq;
3026 xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn);
3028 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3033 memcpy(ddq, recddq, item->ri_buf[1].i_len);
3034 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3035 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
3039 ASSERT(dq_f->qlf_size == 2);
3040 ASSERT(bp->b_target->bt_mount == mp);
3041 bp->b_iodone = xlog_recover_iodone;
3042 xfs_buf_delwri_queue(bp, buffer_list);
3050 * This routine is called to create an in-core extent free intent
3051 * item from the efi format structure which was logged on disk.
3052 * It allocates an in-core efi, copies the extents from the format
3053 * structure into it, and adds the efi to the AIL with the given
3057 xlog_recover_efi_pass2(
3059 struct xlog_recover_item *item,
3063 xfs_mount_t *mp = log->l_mp;
3064 xfs_efi_log_item_t *efip;
3065 xfs_efi_log_format_t *efi_formatp;
3067 efi_formatp = item->ri_buf[0].i_addr;
3069 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
3070 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
3071 &(efip->efi_format)))) {
3072 xfs_efi_item_free(efip);
3075 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
3077 spin_lock(&log->l_ailp->xa_lock);
3079 * xfs_trans_ail_update() drops the AIL lock.
3081 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
3087 * This routine is called when an efd format structure is found in
3088 * a committed transaction in the log. It's purpose is to cancel
3089 * the corresponding efi if it was still in the log. To do this
3090 * it searches the AIL for the efi with an id equal to that in the
3091 * efd format structure. If we find it, we remove the efi from the
3095 xlog_recover_efd_pass2(
3097 struct xlog_recover_item *item)
3099 xfs_efd_log_format_t *efd_formatp;
3100 xfs_efi_log_item_t *efip = NULL;
3101 xfs_log_item_t *lip;
3103 struct xfs_ail_cursor cur;
3104 struct xfs_ail *ailp = log->l_ailp;
3106 efd_formatp = item->ri_buf[0].i_addr;
3107 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
3108 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
3109 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
3110 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
3111 efi_id = efd_formatp->efd_efi_id;
3114 * Search for the efi with the id in the efd format structure
3117 spin_lock(&ailp->xa_lock);
3118 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3119 while (lip != NULL) {
3120 if (lip->li_type == XFS_LI_EFI) {
3121 efip = (xfs_efi_log_item_t *)lip;
3122 if (efip->efi_format.efi_id == efi_id) {
3124 * xfs_trans_ail_delete() drops the
3127 xfs_trans_ail_delete(ailp, lip,
3128 SHUTDOWN_CORRUPT_INCORE);
3129 xfs_efi_item_free(efip);
3130 spin_lock(&ailp->xa_lock);
3134 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3136 xfs_trans_ail_cursor_done(ailp, &cur);
3137 spin_unlock(&ailp->xa_lock);
3143 * This routine is called when an inode create format structure is found in a
3144 * committed transaction in the log. It's purpose is to initialise the inodes
3145 * being allocated on disk. This requires us to get inode cluster buffers that
3146 * match the range to be intialised, stamped with inode templates and written
3147 * by delayed write so that subsequent modifications will hit the cached buffer
3148 * and only need writing out at the end of recovery.
3151 xlog_recover_do_icreate_pass2(
3153 struct list_head *buffer_list,
3154 xlog_recover_item_t *item)
3156 struct xfs_mount *mp = log->l_mp;
3157 struct xfs_icreate_log *icl;
3158 xfs_agnumber_t agno;
3159 xfs_agblock_t agbno;
3162 xfs_agblock_t length;
3164 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3165 if (icl->icl_type != XFS_LI_ICREATE) {
3166 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
3170 if (icl->icl_size != 1) {
3171 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
3175 agno = be32_to_cpu(icl->icl_ag);
3176 if (agno >= mp->m_sb.sb_agcount) {
3177 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
3180 agbno = be32_to_cpu(icl->icl_agbno);
3181 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3182 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
3185 isize = be32_to_cpu(icl->icl_isize);
3186 if (isize != mp->m_sb.sb_inodesize) {
3187 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
3190 count = be32_to_cpu(icl->icl_count);
3192 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
3195 length = be32_to_cpu(icl->icl_length);
3196 if (!length || length >= mp->m_sb.sb_agblocks) {
3197 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
3201 /* existing allocation is fixed value */
3202 ASSERT(count == XFS_IALLOC_INODES(mp));
3203 ASSERT(length == XFS_IALLOC_BLOCKS(mp));
3204 if (count != XFS_IALLOC_INODES(mp) ||
3205 length != XFS_IALLOC_BLOCKS(mp)) {
3206 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count 2");
3211 * Inode buffers can be freed. Do not replay the inode initialisation as
3212 * we could be overwriting something written after this inode buffer was
3215 * XXX: we need to iterate all buffers and only init those that are not
3216 * cancelled. I think that a more fine grained factoring of
3217 * xfs_ialloc_inode_init may be appropriate here to enable this to be
3220 if (xlog_check_buffer_cancelled(log,
3221 XFS_AGB_TO_DADDR(mp, agno, agbno), length, 0))
3224 xfs_ialloc_inode_init(mp, NULL, buffer_list, agno, agbno, length,
3225 be32_to_cpu(icl->icl_gen));
3230 * Free up any resources allocated by the transaction
3232 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3235 xlog_recover_free_trans(
3236 struct xlog_recover *trans)
3238 xlog_recover_item_t *item, *n;
3241 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
3242 /* Free the regions in the item. */
3243 list_del(&item->ri_list);
3244 for (i = 0; i < item->ri_cnt; i++)
3245 kmem_free(item->ri_buf[i].i_addr);
3246 /* Free the item itself */
3247 kmem_free(item->ri_buf);
3250 /* Free the transaction recover structure */
3255 xlog_recover_buffer_ra_pass2(
3257 struct xlog_recover_item *item)
3259 struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr;
3260 struct xfs_mount *mp = log->l_mp;
3262 if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
3263 buf_f->blf_len, buf_f->blf_flags)) {
3267 xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
3268 buf_f->blf_len, NULL);
3272 xlog_recover_inode_ra_pass2(
3274 struct xlog_recover_item *item)
3276 struct xfs_inode_log_format ilf_buf;
3277 struct xfs_inode_log_format *ilfp;
3278 struct xfs_mount *mp = log->l_mp;
3281 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3282 ilfp = item->ri_buf[0].i_addr;
3285 memset(ilfp, 0, sizeof(*ilfp));
3286 error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
3291 if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
3294 xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
3295 ilfp->ilf_len, &xfs_inode_buf_ra_ops);
3299 xlog_recover_dquot_ra_pass2(
3301 struct xlog_recover_item *item)
3303 struct xfs_mount *mp = log->l_mp;
3304 struct xfs_disk_dquot *recddq;
3305 struct xfs_dq_logformat *dq_f;
3309 if (mp->m_qflags == 0)
3312 recddq = item->ri_buf[1].i_addr;
3315 if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
3318 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3320 if (log->l_quotaoffs_flag & type)
3323 dq_f = item->ri_buf[0].i_addr;
3325 ASSERT(dq_f->qlf_len == 1);
3327 xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno,
3328 XFS_FSB_TO_BB(mp, dq_f->qlf_len), NULL);
3332 xlog_recover_ra_pass2(
3334 struct xlog_recover_item *item)
3336 switch (ITEM_TYPE(item)) {
3338 xlog_recover_buffer_ra_pass2(log, item);
3341 xlog_recover_inode_ra_pass2(log, item);
3344 xlog_recover_dquot_ra_pass2(log, item);
3348 case XFS_LI_QUOTAOFF:
3355 xlog_recover_commit_pass1(
3357 struct xlog_recover *trans,
3358 struct xlog_recover_item *item)
3360 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
3362 switch (ITEM_TYPE(item)) {
3364 return xlog_recover_buffer_pass1(log, item);
3365 case XFS_LI_QUOTAOFF:
3366 return xlog_recover_quotaoff_pass1(log, item);
3371 case XFS_LI_ICREATE:
3372 /* nothing to do in pass 1 */
3375 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3376 __func__, ITEM_TYPE(item));
3378 return XFS_ERROR(EIO);
3383 xlog_recover_commit_pass2(
3385 struct xlog_recover *trans,
3386 struct list_head *buffer_list,
3387 struct xlog_recover_item *item)
3389 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
3391 switch (ITEM_TYPE(item)) {
3393 return xlog_recover_buffer_pass2(log, buffer_list, item,
3396 return xlog_recover_inode_pass2(log, buffer_list, item,
3399 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
3401 return xlog_recover_efd_pass2(log, item);
3403 return xlog_recover_dquot_pass2(log, buffer_list, item,
3405 case XFS_LI_ICREATE:
3406 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
3407 case XFS_LI_QUOTAOFF:
3408 /* nothing to do in pass2 */
3411 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3412 __func__, ITEM_TYPE(item));
3414 return XFS_ERROR(EIO);
3419 xlog_recover_items_pass2(
3421 struct xlog_recover *trans,
3422 struct list_head *buffer_list,
3423 struct list_head *item_list)
3425 struct xlog_recover_item *item;
3428 list_for_each_entry(item, item_list, ri_list) {
3429 error = xlog_recover_commit_pass2(log, trans,
3439 * Perform the transaction.
3441 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3442 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3445 xlog_recover_commit_trans(
3447 struct xlog_recover *trans,
3452 int items_queued = 0;
3453 struct xlog_recover_item *item;
3454 struct xlog_recover_item *next;
3455 LIST_HEAD (buffer_list);
3456 LIST_HEAD (ra_list);
3457 LIST_HEAD (done_list);
3459 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
3461 hlist_del(&trans->r_list);
3463 error = xlog_recover_reorder_trans(log, trans, pass);
3467 list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
3469 case XLOG_RECOVER_PASS1:
3470 error = xlog_recover_commit_pass1(log, trans, item);
3472 case XLOG_RECOVER_PASS2:
3473 xlog_recover_ra_pass2(log, item);
3474 list_move_tail(&item->ri_list, &ra_list);
3476 if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
3477 error = xlog_recover_items_pass2(log, trans,
3478 &buffer_list, &ra_list);
3479 list_splice_tail_init(&ra_list, &done_list);
3493 if (!list_empty(&ra_list)) {
3495 error = xlog_recover_items_pass2(log, trans,
3496 &buffer_list, &ra_list);
3497 list_splice_tail_init(&ra_list, &done_list);
3500 if (!list_empty(&done_list))
3501 list_splice_init(&done_list, &trans->r_itemq);
3503 xlog_recover_free_trans(trans);
3505 error2 = xfs_buf_delwri_submit(&buffer_list);
3506 return error ? error : error2;
3510 xlog_recover_unmount_trans(
3512 struct xlog_recover *trans)
3514 /* Do nothing now */
3515 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
3520 * There are two valid states of the r_state field. 0 indicates that the
3521 * transaction structure is in a normal state. We have either seen the
3522 * start of the transaction or the last operation we added was not a partial
3523 * operation. If the last operation we added to the transaction was a
3524 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
3526 * NOTE: skip LRs with 0 data length.
3529 xlog_recover_process_data(
3531 struct hlist_head rhash[],
3532 struct xlog_rec_header *rhead,
3538 xlog_op_header_t *ohead;
3539 xlog_recover_t *trans;
3545 lp = dp + be32_to_cpu(rhead->h_len);
3546 num_logops = be32_to_cpu(rhead->h_num_logops);
3548 /* check the log format matches our own - else we can't recover */
3549 if (xlog_header_check_recover(log->l_mp, rhead))
3550 return (XFS_ERROR(EIO));
3552 while ((dp < lp) && num_logops) {
3553 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
3554 ohead = (xlog_op_header_t *)dp;
3555 dp += sizeof(xlog_op_header_t);
3556 if (ohead->oh_clientid != XFS_TRANSACTION &&
3557 ohead->oh_clientid != XFS_LOG) {
3558 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
3559 __func__, ohead->oh_clientid);
3561 return (XFS_ERROR(EIO));
3563 tid = be32_to_cpu(ohead->oh_tid);
3564 hash = XLOG_RHASH(tid);
3565 trans = xlog_recover_find_tid(&rhash[hash], tid);
3566 if (trans == NULL) { /* not found; add new tid */
3567 if (ohead->oh_flags & XLOG_START_TRANS)
3568 xlog_recover_new_tid(&rhash[hash], tid,
3569 be64_to_cpu(rhead->h_lsn));
3571 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
3572 xfs_warn(log->l_mp, "%s: bad length 0x%x",
3573 __func__, be32_to_cpu(ohead->oh_len));
3575 return (XFS_ERROR(EIO));
3577 flags = ohead->oh_flags & ~XLOG_END_TRANS;
3578 if (flags & XLOG_WAS_CONT_TRANS)
3579 flags &= ~XLOG_CONTINUE_TRANS;
3581 case XLOG_COMMIT_TRANS:
3582 error = xlog_recover_commit_trans(log,
3585 case XLOG_UNMOUNT_TRANS:
3586 error = xlog_recover_unmount_trans(log, trans);
3588 case XLOG_WAS_CONT_TRANS:
3589 error = xlog_recover_add_to_cont_trans(log,
3591 be32_to_cpu(ohead->oh_len));
3593 case XLOG_START_TRANS:
3594 xfs_warn(log->l_mp, "%s: bad transaction",
3597 error = XFS_ERROR(EIO);
3600 case XLOG_CONTINUE_TRANS:
3601 error = xlog_recover_add_to_trans(log, trans,
3602 dp, be32_to_cpu(ohead->oh_len));
3605 xfs_warn(log->l_mp, "%s: bad flag 0x%x",
3608 error = XFS_ERROR(EIO);
3614 dp += be32_to_cpu(ohead->oh_len);
3621 * Process an extent free intent item that was recovered from
3622 * the log. We need to free the extents that it describes.
3625 xlog_recover_process_efi(
3627 xfs_efi_log_item_t *efip)
3629 xfs_efd_log_item_t *efdp;
3634 xfs_fsblock_t startblock_fsb;
3636 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
3639 * First check the validity of the extents described by the
3640 * EFI. If any are bad, then assume that all are bad and
3641 * just toss the EFI.
3643 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3644 extp = &(efip->efi_format.efi_extents[i]);
3645 startblock_fsb = XFS_BB_TO_FSB(mp,
3646 XFS_FSB_TO_DADDR(mp, extp->ext_start));
3647 if ((startblock_fsb == 0) ||
3648 (extp->ext_len == 0) ||
3649 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3650 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3652 * This will pull the EFI from the AIL and
3653 * free the memory associated with it.
3655 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
3656 xfs_efi_release(efip, efip->efi_format.efi_nextents);
3657 return XFS_ERROR(EIO);
3661 tp = xfs_trans_alloc(mp, 0);
3662 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
3665 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3667 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3668 extp = &(efip->efi_format.efi_extents[i]);
3669 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3672 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3676 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
3677 error = xfs_trans_commit(tp, 0);
3681 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3686 * When this is called, all of the EFIs which did not have
3687 * corresponding EFDs should be in the AIL. What we do now
3688 * is free the extents associated with each one.
3690 * Since we process the EFIs in normal transactions, they
3691 * will be removed at some point after the commit. This prevents
3692 * us from just walking down the list processing each one.
3693 * We'll use a flag in the EFI to skip those that we've already
3694 * processed and use the AIL iteration mechanism's generation
3695 * count to try to speed this up at least a bit.
3697 * When we start, we know that the EFIs are the only things in
3698 * the AIL. As we process them, however, other items are added
3699 * to the AIL. Since everything added to the AIL must come after
3700 * everything already in the AIL, we stop processing as soon as
3701 * we see something other than an EFI in the AIL.
3704 xlog_recover_process_efis(
3707 xfs_log_item_t *lip;
3708 xfs_efi_log_item_t *efip;
3710 struct xfs_ail_cursor cur;
3711 struct xfs_ail *ailp;
3714 spin_lock(&ailp->xa_lock);
3715 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3716 while (lip != NULL) {
3718 * We're done when we see something other than an EFI.
3719 * There should be no EFIs left in the AIL now.
3721 if (lip->li_type != XFS_LI_EFI) {
3723 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
3724 ASSERT(lip->li_type != XFS_LI_EFI);
3730 * Skip EFIs that we've already processed.
3732 efip = (xfs_efi_log_item_t *)lip;
3733 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
3734 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3738 spin_unlock(&ailp->xa_lock);
3739 error = xlog_recover_process_efi(log->l_mp, efip);
3740 spin_lock(&ailp->xa_lock);
3743 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3746 xfs_trans_ail_cursor_done(ailp, &cur);
3747 spin_unlock(&ailp->xa_lock);
3752 * This routine performs a transaction to null out a bad inode pointer
3753 * in an agi unlinked inode hash bucket.
3756 xlog_recover_clear_agi_bucket(
3758 xfs_agnumber_t agno,
3767 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3768 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_clearagi, 0, 0);
3772 error = xfs_read_agi(mp, tp, agno, &agibp);
3776 agi = XFS_BUF_TO_AGI(agibp);
3777 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
3778 offset = offsetof(xfs_agi_t, agi_unlinked) +
3779 (sizeof(xfs_agino_t) * bucket);
3780 xfs_trans_log_buf(tp, agibp, offset,
3781 (offset + sizeof(xfs_agino_t) - 1));
3783 error = xfs_trans_commit(tp, 0);
3789 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3791 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
3796 xlog_recover_process_one_iunlink(
3797 struct xfs_mount *mp,
3798 xfs_agnumber_t agno,
3802 struct xfs_buf *ibp;
3803 struct xfs_dinode *dip;
3804 struct xfs_inode *ip;
3808 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3809 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
3814 * Get the on disk inode to find the next inode in the bucket.
3816 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
3820 ASSERT(ip->i_d.di_nlink == 0);
3821 ASSERT(ip->i_d.di_mode != 0);
3823 /* setup for the next pass */
3824 agino = be32_to_cpu(dip->di_next_unlinked);
3828 * Prevent any DMAPI event from being sent when the reference on
3829 * the inode is dropped.
3831 ip->i_d.di_dmevmask = 0;
3840 * We can't read in the inode this bucket points to, or this inode
3841 * is messed up. Just ditch this bucket of inodes. We will lose
3842 * some inodes and space, but at least we won't hang.
3844 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3845 * clear the inode pointer in the bucket.
3847 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3852 * xlog_iunlink_recover
3854 * This is called during recovery to process any inodes which
3855 * we unlinked but not freed when the system crashed. These
3856 * inodes will be on the lists in the AGI blocks. What we do
3857 * here is scan all the AGIs and fully truncate and free any
3858 * inodes found on the lists. Each inode is removed from the
3859 * lists when it has been fully truncated and is freed. The
3860 * freeing of the inode and its removal from the list must be
3864 xlog_recover_process_iunlinks(
3868 xfs_agnumber_t agno;
3879 * Prevent any DMAPI event from being sent while in this function.
3881 mp_dmevmask = mp->m_dmevmask;
3884 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3886 * Find the agi for this ag.
3888 error = xfs_read_agi(mp, NULL, agno, &agibp);
3891 * AGI is b0rked. Don't process it.
3893 * We should probably mark the filesystem as corrupt
3894 * after we've recovered all the ag's we can....
3899 * Unlock the buffer so that it can be acquired in the normal
3900 * course of the transaction to truncate and free each inode.
3901 * Because we are not racing with anyone else here for the AGI
3902 * buffer, we don't even need to hold it locked to read the
3903 * initial unlinked bucket entries out of the buffer. We keep
3904 * buffer reference though, so that it stays pinned in memory
3905 * while we need the buffer.
3907 agi = XFS_BUF_TO_AGI(agibp);
3908 xfs_buf_unlock(agibp);
3910 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3911 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3912 while (agino != NULLAGINO) {
3913 agino = xlog_recover_process_one_iunlink(mp,
3914 agno, agino, bucket);
3917 xfs_buf_rele(agibp);
3920 mp->m_dmevmask = mp_dmevmask;
3924 * Upack the log buffer data and crc check it. If the check fails, issue a
3925 * warning if and only if the CRC in the header is non-zero. This makes the
3926 * check an advisory warning, and the zero CRC check will prevent failure
3927 * warnings from being emitted when upgrading the kernel from one that does not
3928 * add CRCs by default.
3930 * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log
3931 * corruption failure
3934 xlog_unpack_data_crc(
3935 struct xlog_rec_header *rhead,
3941 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
3942 if (crc != rhead->h_crc) {
3943 if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
3944 xfs_alert(log->l_mp,
3945 "log record CRC mismatch: found 0x%x, expected 0x%x.",
3946 le32_to_cpu(rhead->h_crc),
3948 xfs_hex_dump(dp, 32);
3952 * If we've detected a log record corruption, then we can't
3953 * recover past this point. Abort recovery if we are enforcing
3954 * CRC protection by punting an error back up the stack.
3956 if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
3957 return EFSCORRUPTED;
3965 struct xlog_rec_header *rhead,
3972 error = xlog_unpack_data_crc(rhead, dp, log);
3976 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
3977 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3978 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
3982 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3983 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
3984 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
3985 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3986 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3987 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
3996 xlog_valid_rec_header(
3998 struct xlog_rec_header *rhead,
4003 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
4004 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
4005 XFS_ERRLEVEL_LOW, log->l_mp);
4006 return XFS_ERROR(EFSCORRUPTED);
4009 (!rhead->h_version ||
4010 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
4011 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
4012 __func__, be32_to_cpu(rhead->h_version));
4013 return XFS_ERROR(EIO);
4016 /* LR body must have data or it wouldn't have been written */
4017 hlen = be32_to_cpu(rhead->h_len);
4018 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
4019 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
4020 XFS_ERRLEVEL_LOW, log->l_mp);
4021 return XFS_ERROR(EFSCORRUPTED);
4023 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
4024 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
4025 XFS_ERRLEVEL_LOW, log->l_mp);
4026 return XFS_ERROR(EFSCORRUPTED);
4032 * Read the log from tail to head and process the log records found.
4033 * Handle the two cases where the tail and head are in the same cycle
4034 * and where the active portion of the log wraps around the end of
4035 * the physical log separately. The pass parameter is passed through
4036 * to the routines called to process the data and is not looked at
4040 xlog_do_recovery_pass(
4042 xfs_daddr_t head_blk,
4043 xfs_daddr_t tail_blk,
4046 xlog_rec_header_t *rhead;
4049 xfs_buf_t *hbp, *dbp;
4050 int error = 0, h_size;
4051 int bblks, split_bblks;
4052 int hblks, split_hblks, wrapped_hblks;
4053 struct hlist_head rhash[XLOG_RHASH_SIZE];
4055 ASSERT(head_blk != tail_blk);
4058 * Read the header of the tail block and get the iclog buffer size from
4059 * h_size. Use this to tell how many sectors make up the log header.
4061 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
4063 * When using variable length iclogs, read first sector of
4064 * iclog header and extract the header size from it. Get a
4065 * new hbp that is the correct size.
4067 hbp = xlog_get_bp(log, 1);
4071 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
4075 rhead = (xlog_rec_header_t *)offset;
4076 error = xlog_valid_rec_header(log, rhead, tail_blk);
4079 h_size = be32_to_cpu(rhead->h_size);
4080 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
4081 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
4082 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
4083 if (h_size % XLOG_HEADER_CYCLE_SIZE)
4086 hbp = xlog_get_bp(log, hblks);
4091 ASSERT(log->l_sectBBsize == 1);
4093 hbp = xlog_get_bp(log, 1);
4094 h_size = XLOG_BIG_RECORD_BSIZE;
4099 dbp = xlog_get_bp(log, BTOBB(h_size));
4105 memset(rhash, 0, sizeof(rhash));
4106 if (tail_blk <= head_blk) {
4107 for (blk_no = tail_blk; blk_no < head_blk; ) {
4108 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
4112 rhead = (xlog_rec_header_t *)offset;
4113 error = xlog_valid_rec_header(log, rhead, blk_no);
4117 /* blocks in data section */
4118 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4119 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
4124 error = xlog_unpack_data(rhead, offset, log);
4128 error = xlog_recover_process_data(log,
4129 rhash, rhead, offset, pass);
4132 blk_no += bblks + hblks;
4136 * Perform recovery around the end of the physical log.
4137 * When the head is not on the same cycle number as the tail,
4138 * we can't do a sequential recovery as above.
4141 while (blk_no < log->l_logBBsize) {
4143 * Check for header wrapping around physical end-of-log
4145 offset = hbp->b_addr;
4148 if (blk_no + hblks <= log->l_logBBsize) {
4149 /* Read header in one read */
4150 error = xlog_bread(log, blk_no, hblks, hbp,
4155 /* This LR is split across physical log end */
4156 if (blk_no != log->l_logBBsize) {
4157 /* some data before physical log end */
4158 ASSERT(blk_no <= INT_MAX);
4159 split_hblks = log->l_logBBsize - (int)blk_no;
4160 ASSERT(split_hblks > 0);
4161 error = xlog_bread(log, blk_no,
4169 * Note: this black magic still works with
4170 * large sector sizes (non-512) only because:
4171 * - we increased the buffer size originally
4172 * by 1 sector giving us enough extra space
4173 * for the second read;
4174 * - the log start is guaranteed to be sector
4176 * - we read the log end (LR header start)
4177 * _first_, then the log start (LR header end)
4178 * - order is important.
4180 wrapped_hblks = hblks - split_hblks;
4181 error = xlog_bread_offset(log, 0,
4183 offset + BBTOB(split_hblks));
4187 rhead = (xlog_rec_header_t *)offset;
4188 error = xlog_valid_rec_header(log, rhead,
4189 split_hblks ? blk_no : 0);
4193 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4196 /* Read in data for log record */
4197 if (blk_no + bblks <= log->l_logBBsize) {
4198 error = xlog_bread(log, blk_no, bblks, dbp,
4203 /* This log record is split across the
4204 * physical end of log */
4205 offset = dbp->b_addr;
4207 if (blk_no != log->l_logBBsize) {
4208 /* some data is before the physical
4210 ASSERT(!wrapped_hblks);
4211 ASSERT(blk_no <= INT_MAX);
4213 log->l_logBBsize - (int)blk_no;
4214 ASSERT(split_bblks > 0);
4215 error = xlog_bread(log, blk_no,
4223 * Note: this black magic still works with
4224 * large sector sizes (non-512) only because:
4225 * - we increased the buffer size originally
4226 * by 1 sector giving us enough extra space
4227 * for the second read;
4228 * - the log start is guaranteed to be sector
4230 * - we read the log end (LR header start)
4231 * _first_, then the log start (LR header end)
4232 * - order is important.
4234 error = xlog_bread_offset(log, 0,
4235 bblks - split_bblks, dbp,
4236 offset + BBTOB(split_bblks));
4241 error = xlog_unpack_data(rhead, offset, log);
4245 error = xlog_recover_process_data(log, rhash,
4246 rhead, offset, pass);
4252 ASSERT(blk_no >= log->l_logBBsize);
4253 blk_no -= log->l_logBBsize;
4255 /* read first part of physical log */
4256 while (blk_no < head_blk) {
4257 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
4261 rhead = (xlog_rec_header_t *)offset;
4262 error = xlog_valid_rec_header(log, rhead, blk_no);
4266 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4267 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
4272 error = xlog_unpack_data(rhead, offset, log);
4276 error = xlog_recover_process_data(log, rhash,
4277 rhead, offset, pass);
4280 blk_no += bblks + hblks;
4292 * Do the recovery of the log. We actually do this in two phases.
4293 * The two passes are necessary in order to implement the function
4294 * of cancelling a record written into the log. The first pass
4295 * determines those things which have been cancelled, and the
4296 * second pass replays log items normally except for those which
4297 * have been cancelled. The handling of the replay and cancellations
4298 * takes place in the log item type specific routines.
4300 * The table of items which have cancel records in the log is allocated
4301 * and freed at this level, since only here do we know when all of
4302 * the log recovery has been completed.
4305 xlog_do_log_recovery(
4307 xfs_daddr_t head_blk,
4308 xfs_daddr_t tail_blk)
4312 ASSERT(head_blk != tail_blk);
4315 * First do a pass to find all of the cancelled buf log items.
4316 * Store them in the buf_cancel_table for use in the second pass.
4318 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
4319 sizeof(struct list_head),
4321 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4322 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
4324 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4325 XLOG_RECOVER_PASS1);
4327 kmem_free(log->l_buf_cancel_table);
4328 log->l_buf_cancel_table = NULL;
4332 * Then do a second pass to actually recover the items in the log.
4333 * When it is complete free the table of buf cancel items.
4335 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4336 XLOG_RECOVER_PASS2);
4341 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4342 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
4346 kmem_free(log->l_buf_cancel_table);
4347 log->l_buf_cancel_table = NULL;
4353 * Do the actual recovery
4358 xfs_daddr_t head_blk,
4359 xfs_daddr_t tail_blk)
4366 * First replay the images in the log.
4368 error = xlog_do_log_recovery(log, head_blk, tail_blk);
4373 * If IO errors happened during recovery, bail out.
4375 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
4380 * We now update the tail_lsn since much of the recovery has completed
4381 * and there may be space available to use. If there were no extent
4382 * or iunlinks, we can free up the entire log and set the tail_lsn to
4383 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4384 * lsn of the last known good LR on disk. If there are extent frees
4385 * or iunlinks they will have some entries in the AIL; so we look at
4386 * the AIL to determine how to set the tail_lsn.
4388 xlog_assign_tail_lsn(log->l_mp);
4391 * Now that we've finished replaying all buffer and inode
4392 * updates, re-read in the superblock and reverify it.
4394 bp = xfs_getsb(log->l_mp, 0);
4396 ASSERT(!(XFS_BUF_ISWRITE(bp)));
4398 XFS_BUF_UNASYNC(bp);
4399 bp->b_ops = &xfs_sb_buf_ops;
4400 xfsbdstrat(log->l_mp, bp);
4401 error = xfs_buf_iowait(bp);
4403 xfs_buf_ioerror_alert(bp, __func__);
4409 /* Convert superblock from on-disk format */
4410 sbp = &log->l_mp->m_sb;
4411 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
4412 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
4413 ASSERT(xfs_sb_good_version(sbp));
4416 /* We've re-read the superblock so re-initialize per-cpu counters */
4417 xfs_icsb_reinit_counters(log->l_mp);
4419 xlog_recover_check_summary(log);
4421 /* Normal transactions can now occur */
4422 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
4427 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4429 * Return error or zero.
4435 xfs_daddr_t head_blk, tail_blk;
4438 /* find the tail of the log */
4439 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
4442 if (tail_blk != head_blk) {
4443 /* There used to be a comment here:
4445 * disallow recovery on read-only mounts. note -- mount
4446 * checks for ENOSPC and turns it into an intelligent
4448 * ...but this is no longer true. Now, unless you specify
4449 * NORECOVERY (in which case this function would never be
4450 * called), we just go ahead and recover. We do this all
4451 * under the vfs layer, so we can get away with it unless
4452 * the device itself is read-only, in which case we fail.
4454 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
4459 * Version 5 superblock log feature mask validation. We know the
4460 * log is dirty so check if there are any unknown log features
4461 * in what we need to recover. If there are unknown features
4462 * (e.g. unsupported transactions, then simply reject the
4463 * attempt at recovery before touching anything.
4465 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
4466 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
4467 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
4469 "Superblock has unknown incompatible log features (0x%x) enabled.\n"
4470 "The log can not be fully and/or safely recovered by this kernel.\n"
4471 "Please recover the log on a kernel that supports the unknown features.",
4472 (log->l_mp->m_sb.sb_features_log_incompat &
4473 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
4477 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
4478 log->l_mp->m_logname ? log->l_mp->m_logname
4481 error = xlog_do_recover(log, head_blk, tail_blk);
4482 log->l_flags |= XLOG_RECOVERY_NEEDED;
4488 * In the first part of recovery we replay inodes and buffers and build
4489 * up the list of extent free items which need to be processed. Here
4490 * we process the extent free items and clean up the on disk unlinked
4491 * inode lists. This is separated from the first part of recovery so
4492 * that the root and real-time bitmap inodes can be read in from disk in
4493 * between the two stages. This is necessary so that we can free space
4494 * in the real-time portion of the file system.
4497 xlog_recover_finish(
4501 * Now we're ready to do the transactions needed for the
4502 * rest of recovery. Start with completing all the extent
4503 * free intent records and then process the unlinked inode
4504 * lists. At this point, we essentially run in normal mode
4505 * except that we're still performing recovery actions
4506 * rather than accepting new requests.
4508 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
4510 error = xlog_recover_process_efis(log);
4512 xfs_alert(log->l_mp, "Failed to recover EFIs");
4516 * Sync the log to get all the EFIs out of the AIL.
4517 * This isn't absolutely necessary, but it helps in
4518 * case the unlink transactions would have problems
4519 * pushing the EFIs out of the way.
4521 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
4523 xlog_recover_process_iunlinks(log);
4525 xlog_recover_check_summary(log);
4527 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
4528 log->l_mp->m_logname ? log->l_mp->m_logname
4530 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
4532 xfs_info(log->l_mp, "Ending clean mount");
4540 * Read all of the agf and agi counters and check that they
4541 * are consistent with the superblock counters.
4544 xlog_recover_check_summary(
4551 xfs_agnumber_t agno;
4552 __uint64_t freeblks;
4562 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
4563 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
4565 xfs_alert(mp, "%s agf read failed agno %d error %d",
4566 __func__, agno, error);
4568 agfp = XFS_BUF_TO_AGF(agfbp);
4569 freeblks += be32_to_cpu(agfp->agf_freeblks) +
4570 be32_to_cpu(agfp->agf_flcount);
4571 xfs_buf_relse(agfbp);
4574 error = xfs_read_agi(mp, NULL, agno, &agibp);
4576 xfs_alert(mp, "%s agi read failed agno %d error %d",
4577 __func__, agno, error);
4579 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
4581 itotal += be32_to_cpu(agi->agi_count);
4582 ifree += be32_to_cpu(agi->agi_freecount);
4583 xfs_buf_relse(agibp);