2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_trans.h"
32 #include "xfs_log_priv.h"
33 #include "xfs_log_recover.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_extfree_item.h"
36 #include "xfs_trans_priv.h"
37 #include "xfs_alloc.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_quota.h"
40 #include "xfs_cksum.h"
41 #include "xfs_trace.h"
42 #include "xfs_icache.h"
43 #include "xfs_bmap_btree.h"
44 #include "xfs_error.h"
47 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
54 xlog_clear_stale_blocks(
59 xlog_recover_check_summary(
62 #define xlog_recover_check_summary(log)
66 * This structure is used during recovery to record the buf log items which
67 * have been canceled and should not be replayed.
69 struct xfs_buf_cancel {
73 struct list_head bc_list;
77 * Sector aligned buffer routines for buffer create/read/write/access
81 * Verify the given count of basic blocks is valid number of blocks
82 * to specify for an operation involving the given XFS log buffer.
83 * Returns nonzero if the count is valid, 0 otherwise.
87 xlog_buf_bbcount_valid(
91 return bbcount > 0 && bbcount <= log->l_logBBsize;
95 * Allocate a buffer to hold log data. The buffer needs to be able
96 * to map to a range of nbblks basic blocks at any valid (basic
97 * block) offset within the log.
106 if (!xlog_buf_bbcount_valid(log, nbblks)) {
107 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
109 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
114 * We do log I/O in units of log sectors (a power-of-2
115 * multiple of the basic block size), so we round up the
116 * requested size to accommodate the basic blocks required
117 * for complete log sectors.
119 * In addition, the buffer may be used for a non-sector-
120 * aligned block offset, in which case an I/O of the
121 * requested size could extend beyond the end of the
122 * buffer. If the requested size is only 1 basic block it
123 * will never straddle a sector boundary, so this won't be
124 * an issue. Nor will this be a problem if the log I/O is
125 * done in basic blocks (sector size 1). But otherwise we
126 * extend the buffer by one extra log sector to ensure
127 * there's space to accommodate this possibility.
129 if (nbblks > 1 && log->l_sectBBsize > 1)
130 nbblks += log->l_sectBBsize;
131 nbblks = round_up(nbblks, log->l_sectBBsize);
133 bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0);
147 * Return the address of the start of the given block number's data
148 * in a log buffer. The buffer covers a log sector-aligned region.
157 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
159 ASSERT(offset + nbblks <= bp->b_length);
160 return bp->b_addr + BBTOB(offset);
165 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
176 if (!xlog_buf_bbcount_valid(log, nbblks)) {
177 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
179 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
180 return -EFSCORRUPTED;
183 blk_no = round_down(blk_no, log->l_sectBBsize);
184 nbblks = round_up(nbblks, log->l_sectBBsize);
187 ASSERT(nbblks <= bp->b_length);
189 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
191 bp->b_io_length = nbblks;
194 error = xfs_buf_submit_wait(bp);
195 if (error && !XFS_FORCED_SHUTDOWN(log->l_mp))
196 xfs_buf_ioerror_alert(bp, __func__);
210 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
214 *offset = xlog_align(log, blk_no, nbblks, bp);
219 * Read at an offset into the buffer. Returns with the buffer in it's original
220 * state regardless of the result of the read.
225 xfs_daddr_t blk_no, /* block to read from */
226 int nbblks, /* blocks to read */
230 char *orig_offset = bp->b_addr;
231 int orig_len = BBTOB(bp->b_length);
234 error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks));
238 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
240 /* must reset buffer pointer even on error */
241 error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len);
248 * Write out the buffer at the given block for the given number of blocks.
249 * The buffer is kept locked across the write and is returned locked.
250 * This can only be used for synchronous log writes.
261 if (!xlog_buf_bbcount_valid(log, nbblks)) {
262 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
264 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
265 return -EFSCORRUPTED;
268 blk_no = round_down(blk_no, log->l_sectBBsize);
269 nbblks = round_up(nbblks, log->l_sectBBsize);
272 ASSERT(nbblks <= bp->b_length);
274 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
275 XFS_BUF_ZEROFLAGS(bp);
278 bp->b_io_length = nbblks;
281 error = xfs_bwrite(bp);
283 xfs_buf_ioerror_alert(bp, __func__);
290 * dump debug superblock and log record information
293 xlog_header_check_dump(
295 xlog_rec_header_t *head)
297 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d",
298 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
299 xfs_debug(mp, " log : uuid = %pU, fmt = %d",
300 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
303 #define xlog_header_check_dump(mp, head)
307 * check log record header for recovery
310 xlog_header_check_recover(
312 xlog_rec_header_t *head)
314 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
317 * IRIX doesn't write the h_fmt field and leaves it zeroed
318 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
319 * a dirty log created in IRIX.
321 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
323 "dirty log written in incompatible format - can't recover");
324 xlog_header_check_dump(mp, head);
325 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
326 XFS_ERRLEVEL_HIGH, mp);
327 return -EFSCORRUPTED;
328 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
330 "dirty log entry has mismatched uuid - can't recover");
331 xlog_header_check_dump(mp, head);
332 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
333 XFS_ERRLEVEL_HIGH, mp);
334 return -EFSCORRUPTED;
340 * read the head block of the log and check the header
343 xlog_header_check_mount(
345 xlog_rec_header_t *head)
347 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
349 if (uuid_is_nil(&head->h_fs_uuid)) {
351 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
352 * h_fs_uuid is nil, we assume this log was last mounted
353 * by IRIX and continue.
355 xfs_warn(mp, "nil uuid in log - IRIX style log");
356 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
357 xfs_warn(mp, "log has mismatched uuid - can't recover");
358 xlog_header_check_dump(mp, head);
359 XFS_ERROR_REPORT("xlog_header_check_mount",
360 XFS_ERRLEVEL_HIGH, mp);
361 return -EFSCORRUPTED;
372 * We're not going to bother about retrying
373 * this during recovery. One strike!
375 if (!XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
376 xfs_buf_ioerror_alert(bp, __func__);
377 xfs_force_shutdown(bp->b_target->bt_mount,
378 SHUTDOWN_META_IO_ERROR);
386 * This routine finds (to an approximation) the first block in the physical
387 * log which contains the given cycle. It uses a binary search algorithm.
388 * Note that the algorithm can not be perfect because the disk will not
389 * necessarily be perfect.
392 xlog_find_cycle_start(
395 xfs_daddr_t first_blk,
396 xfs_daddr_t *last_blk,
406 mid_blk = BLK_AVG(first_blk, end_blk);
407 while (mid_blk != first_blk && mid_blk != end_blk) {
408 error = xlog_bread(log, mid_blk, 1, bp, &offset);
411 mid_cycle = xlog_get_cycle(offset);
412 if (mid_cycle == cycle)
413 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
415 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
416 mid_blk = BLK_AVG(first_blk, end_blk);
418 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
419 (mid_blk == end_blk && mid_blk-1 == first_blk));
427 * Check that a range of blocks does not contain stop_on_cycle_no.
428 * Fill in *new_blk with the block offset where such a block is
429 * found, or with -1 (an invalid block number) if there is no such
430 * block in the range. The scan needs to occur from front to back
431 * and the pointer into the region must be updated since a later
432 * routine will need to perform another test.
435 xlog_find_verify_cycle(
437 xfs_daddr_t start_blk,
439 uint stop_on_cycle_no,
440 xfs_daddr_t *new_blk)
450 * Greedily allocate a buffer big enough to handle the full
451 * range of basic blocks we'll be examining. If that fails,
452 * try a smaller size. We need to be able to read at least
453 * a log sector, or we're out of luck.
455 bufblks = 1 << ffs(nbblks);
456 while (bufblks > log->l_logBBsize)
458 while (!(bp = xlog_get_bp(log, bufblks))) {
460 if (bufblks < log->l_sectBBsize)
464 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
467 bcount = min(bufblks, (start_blk + nbblks - i));
469 error = xlog_bread(log, i, bcount, bp, &buf);
473 for (j = 0; j < bcount; j++) {
474 cycle = xlog_get_cycle(buf);
475 if (cycle == stop_on_cycle_no) {
492 * Potentially backup over partial log record write.
494 * In the typical case, last_blk is the number of the block directly after
495 * a good log record. Therefore, we subtract one to get the block number
496 * of the last block in the given buffer. extra_bblks contains the number
497 * of blocks we would have read on a previous read. This happens when the
498 * last log record is split over the end of the physical log.
500 * extra_bblks is the number of blocks potentially verified on a previous
501 * call to this routine.
504 xlog_find_verify_log_record(
506 xfs_daddr_t start_blk,
507 xfs_daddr_t *last_blk,
513 xlog_rec_header_t *head = NULL;
516 int num_blks = *last_blk - start_blk;
519 ASSERT(start_blk != 0 || *last_blk != start_blk);
521 if (!(bp = xlog_get_bp(log, num_blks))) {
522 if (!(bp = xlog_get_bp(log, 1)))
526 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
529 offset += ((num_blks - 1) << BBSHIFT);
532 for (i = (*last_blk) - 1; i >= 0; i--) {
534 /* valid log record not found */
536 "Log inconsistent (didn't find previous header)");
543 error = xlog_bread(log, i, 1, bp, &offset);
548 head = (xlog_rec_header_t *)offset;
550 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
558 * We hit the beginning of the physical log & still no header. Return
559 * to caller. If caller can handle a return of -1, then this routine
560 * will be called again for the end of the physical log.
568 * We have the final block of the good log (the first block
569 * of the log record _before_ the head. So we check the uuid.
571 if ((error = xlog_header_check_mount(log->l_mp, head)))
575 * We may have found a log record header before we expected one.
576 * last_blk will be the 1st block # with a given cycle #. We may end
577 * up reading an entire log record. In this case, we don't want to
578 * reset last_blk. Only when last_blk points in the middle of a log
579 * record do we update last_blk.
581 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
582 uint h_size = be32_to_cpu(head->h_size);
584 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
585 if (h_size % XLOG_HEADER_CYCLE_SIZE)
591 if (*last_blk - i + extra_bblks !=
592 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
601 * Head is defined to be the point of the log where the next log write
602 * could go. This means that incomplete LR writes at the end are
603 * eliminated when calculating the head. We aren't guaranteed that previous
604 * LR have complete transactions. We only know that a cycle number of
605 * current cycle number -1 won't be present in the log if we start writing
606 * from our current block number.
608 * last_blk contains the block number of the first block with a given
611 * Return: zero if normal, non-zero if error.
616 xfs_daddr_t *return_head_blk)
620 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
622 uint first_half_cycle, last_half_cycle;
624 int error, log_bbnum = log->l_logBBsize;
626 /* Is the end of the log device zeroed? */
627 error = xlog_find_zeroed(log, &first_blk);
629 xfs_warn(log->l_mp, "empty log check failed");
633 *return_head_blk = first_blk;
635 /* Is the whole lot zeroed? */
637 /* Linux XFS shouldn't generate totally zeroed logs -
638 * mkfs etc write a dummy unmount record to a fresh
639 * log so we can store the uuid in there
641 xfs_warn(log->l_mp, "totally zeroed log");
647 first_blk = 0; /* get cycle # of 1st block */
648 bp = xlog_get_bp(log, 1);
652 error = xlog_bread(log, 0, 1, bp, &offset);
656 first_half_cycle = xlog_get_cycle(offset);
658 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
659 error = xlog_bread(log, last_blk, 1, bp, &offset);
663 last_half_cycle = xlog_get_cycle(offset);
664 ASSERT(last_half_cycle != 0);
667 * If the 1st half cycle number is equal to the last half cycle number,
668 * then the entire log is stamped with the same cycle number. In this
669 * case, head_blk can't be set to zero (which makes sense). The below
670 * math doesn't work out properly with head_blk equal to zero. Instead,
671 * we set it to log_bbnum which is an invalid block number, but this
672 * value makes the math correct. If head_blk doesn't changed through
673 * all the tests below, *head_blk is set to zero at the very end rather
674 * than log_bbnum. In a sense, log_bbnum and zero are the same block
675 * in a circular file.
677 if (first_half_cycle == last_half_cycle) {
679 * In this case we believe that the entire log should have
680 * cycle number last_half_cycle. We need to scan backwards
681 * from the end verifying that there are no holes still
682 * containing last_half_cycle - 1. If we find such a hole,
683 * then the start of that hole will be the new head. The
684 * simple case looks like
685 * x | x ... | x - 1 | x
686 * Another case that fits this picture would be
687 * x | x + 1 | x ... | x
688 * In this case the head really is somewhere at the end of the
689 * log, as one of the latest writes at the beginning was
692 * x | x + 1 | x ... | x - 1 | x
693 * This is really the combination of the above two cases, and
694 * the head has to end up at the start of the x-1 hole at the
697 * In the 256k log case, we will read from the beginning to the
698 * end of the log and search for cycle numbers equal to x-1.
699 * We don't worry about the x+1 blocks that we encounter,
700 * because we know that they cannot be the head since the log
703 head_blk = log_bbnum;
704 stop_on_cycle = last_half_cycle - 1;
707 * In this case we want to find the first block with cycle
708 * number matching last_half_cycle. We expect the log to be
710 * x + 1 ... | x ... | x
711 * The first block with cycle number x (last_half_cycle) will
712 * be where the new head belongs. First we do a binary search
713 * for the first occurrence of last_half_cycle. The binary
714 * search may not be totally accurate, so then we scan back
715 * from there looking for occurrences of last_half_cycle before
716 * us. If that backwards scan wraps around the beginning of
717 * the log, then we look for occurrences of last_half_cycle - 1
718 * at the end of the log. The cases we're looking for look
720 * v binary search stopped here
721 * x + 1 ... | x | x + 1 | x ... | x
722 * ^ but we want to locate this spot
724 * <---------> less than scan distance
725 * x + 1 ... | x ... | x - 1 | x
726 * ^ we want to locate this spot
728 stop_on_cycle = last_half_cycle;
729 if ((error = xlog_find_cycle_start(log, bp, first_blk,
730 &head_blk, last_half_cycle)))
735 * Now validate the answer. Scan back some number of maximum possible
736 * blocks and make sure each one has the expected cycle number. The
737 * maximum is determined by the total possible amount of buffering
738 * in the in-core log. The following number can be made tighter if
739 * we actually look at the block size of the filesystem.
741 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
742 if (head_blk >= num_scan_bblks) {
744 * We are guaranteed that the entire check can be performed
747 start_blk = head_blk - num_scan_bblks;
748 if ((error = xlog_find_verify_cycle(log,
749 start_blk, num_scan_bblks,
750 stop_on_cycle, &new_blk)))
754 } else { /* need to read 2 parts of log */
756 * We are going to scan backwards in the log in two parts.
757 * First we scan the physical end of the log. In this part
758 * of the log, we are looking for blocks with cycle number
759 * last_half_cycle - 1.
760 * If we find one, then we know that the log starts there, as
761 * we've found a hole that didn't get written in going around
762 * the end of the physical log. The simple case for this is
763 * x + 1 ... | x ... | x - 1 | x
764 * <---------> less than scan distance
765 * If all of the blocks at the end of the log have cycle number
766 * last_half_cycle, then we check the blocks at the start of
767 * the log looking for occurrences of last_half_cycle. If we
768 * find one, then our current estimate for the location of the
769 * first occurrence of last_half_cycle is wrong and we move
770 * back to the hole we've found. This case looks like
771 * x + 1 ... | x | x + 1 | x ...
772 * ^ binary search stopped here
773 * Another case we need to handle that only occurs in 256k
775 * x + 1 ... | x ... | x+1 | x ...
776 * ^ binary search stops here
777 * In a 256k log, the scan at the end of the log will see the
778 * x + 1 blocks. We need to skip past those since that is
779 * certainly not the head of the log. By searching for
780 * last_half_cycle-1 we accomplish that.
782 ASSERT(head_blk <= INT_MAX &&
783 (xfs_daddr_t) num_scan_bblks >= head_blk);
784 start_blk = log_bbnum - (num_scan_bblks - head_blk);
785 if ((error = xlog_find_verify_cycle(log, start_blk,
786 num_scan_bblks - (int)head_blk,
787 (stop_on_cycle - 1), &new_blk)))
795 * Scan beginning of log now. The last part of the physical
796 * log is good. This scan needs to verify that it doesn't find
797 * the last_half_cycle.
800 ASSERT(head_blk <= INT_MAX);
801 if ((error = xlog_find_verify_cycle(log,
802 start_blk, (int)head_blk,
803 stop_on_cycle, &new_blk)))
811 * Now we need to make sure head_blk is not pointing to a block in
812 * the middle of a log record.
814 num_scan_bblks = XLOG_REC_SHIFT(log);
815 if (head_blk >= num_scan_bblks) {
816 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
818 /* start ptr at last block ptr before head_blk */
819 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
826 ASSERT(head_blk <= INT_MAX);
827 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
831 /* We hit the beginning of the log during our search */
832 start_blk = log_bbnum - (num_scan_bblks - head_blk);
834 ASSERT(start_blk <= INT_MAX &&
835 (xfs_daddr_t) log_bbnum-start_blk >= 0);
836 ASSERT(head_blk <= INT_MAX);
837 error = xlog_find_verify_log_record(log, start_blk,
838 &new_blk, (int)head_blk);
843 if (new_blk != log_bbnum)
850 if (head_blk == log_bbnum)
851 *return_head_blk = 0;
853 *return_head_blk = head_blk;
855 * When returning here, we have a good block number. Bad block
856 * means that during a previous crash, we didn't have a clean break
857 * from cycle number N to cycle number N-1. In this case, we need
858 * to find the first block with cycle number N-1.
866 xfs_warn(log->l_mp, "failed to find log head");
871 * Seek backwards in the log for log record headers.
873 * Given a starting log block, walk backwards until we find the provided number
874 * of records or hit the provided tail block. The return value is the number of
875 * records encountered or a negative error code. The log block and buffer
876 * pointer of the last record seen are returned in rblk and rhead respectively.
879 xlog_rseek_logrec_hdr(
881 xfs_daddr_t head_blk,
882 xfs_daddr_t tail_blk,
886 struct xlog_rec_header **rhead,
898 * Walk backwards from the head block until we hit the tail or the first
901 end_blk = head_blk > tail_blk ? tail_blk : 0;
902 for (i = (int) head_blk - 1; i >= end_blk; i--) {
903 error = xlog_bread(log, i, 1, bp, &offset);
907 if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
909 *rhead = (struct xlog_rec_header *) offset;
910 if (++found == count)
916 * If we haven't hit the tail block or the log record header count,
917 * start looking again from the end of the physical log. Note that
918 * callers can pass head == tail if the tail is not yet known.
920 if (tail_blk >= head_blk && found != count) {
921 for (i = log->l_logBBsize - 1; i >= (int) tail_blk; i--) {
922 error = xlog_bread(log, i, 1, bp, &offset);
926 if (*(__be32 *)offset ==
927 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
930 *rhead = (struct xlog_rec_header *) offset;
931 if (++found == count)
944 * Find the sync block number or the tail of the log.
946 * This will be the block number of the last record to have its
947 * associated buffers synced to disk. Every log record header has
948 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
949 * to get a sync block number. The only concern is to figure out which
950 * log record header to believe.
952 * The following algorithm uses the log record header with the largest
953 * lsn. The entire log record does not need to be valid. We only care
954 * that the header is valid.
956 * We could speed up search by using current head_blk buffer, but it is not
962 xfs_daddr_t *head_blk,
963 xfs_daddr_t *tail_blk)
965 xlog_rec_header_t *rhead;
966 xlog_op_header_t *op_head;
970 xfs_daddr_t umount_data_blk;
971 xfs_daddr_t after_umount_blk;
974 bool wrapped = false;
977 * Find previous log record
979 if ((error = xlog_find_head(log, head_blk)))
982 bp = xlog_get_bp(log, 1);
985 if (*head_blk == 0) { /* special case */
986 error = xlog_bread(log, 0, 1, bp, &offset);
990 if (xlog_get_cycle(offset) == 0) {
992 /* leave all other log inited values alone */
998 * Search backwards through the log looking for the log record header
999 * block. This wraps all the way back around to the head so something is
1000 * seriously wrong if we can't find it.
1002 ASSERT(*head_blk < INT_MAX);
1003 found = xlog_rseek_logrec_hdr(log, *head_blk, *head_blk, 1, bp, &i,
1010 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
1015 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
1018 * Reset log values according to the state of the log when we
1019 * crashed. In the case where head_blk == 0, we bump curr_cycle
1020 * one because the next write starts a new cycle rather than
1021 * continuing the cycle of the last good log record. At this
1022 * point we have guaranteed that all partial log records have been
1023 * accounted for. Therefore, we know that the last good log record
1024 * written was complete and ended exactly on the end boundary
1025 * of the physical log.
1027 log->l_prev_block = i;
1028 log->l_curr_block = (int)*head_blk;
1029 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
1031 log->l_curr_cycle++;
1032 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
1033 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
1034 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
1035 BBTOB(log->l_curr_block));
1036 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
1037 BBTOB(log->l_curr_block));
1040 * Look for unmount record. If we find it, then we know there
1041 * was a clean unmount. Since 'i' could be the last block in
1042 * the physical log, we convert to a log block before comparing
1045 * Save the current tail lsn to use to pass to
1046 * xlog_clear_stale_blocks() below. We won't want to clear the
1047 * unmount record if there is one, so we pass the lsn of the
1048 * unmount record rather than the block after it.
1050 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1051 int h_size = be32_to_cpu(rhead->h_size);
1052 int h_version = be32_to_cpu(rhead->h_version);
1054 if ((h_version & XLOG_VERSION_2) &&
1055 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
1056 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
1057 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1065 after_umount_blk = (i + hblks + (int)
1066 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
1067 tail_lsn = atomic64_read(&log->l_tail_lsn);
1068 if (*head_blk == after_umount_blk &&
1069 be32_to_cpu(rhead->h_num_logops) == 1) {
1070 umount_data_blk = (i + hblks) % log->l_logBBsize;
1071 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1075 op_head = (xlog_op_header_t *)offset;
1076 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1078 * Set tail and last sync so that newly written
1079 * log records will point recovery to after the
1080 * current unmount record.
1082 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1083 log->l_curr_cycle, after_umount_blk);
1084 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1085 log->l_curr_cycle, after_umount_blk);
1086 *tail_blk = after_umount_blk;
1089 * Note that the unmount was clean. If the unmount
1090 * was not clean, we need to know this to rebuild the
1091 * superblock counters from the perag headers if we
1092 * have a filesystem using non-persistent counters.
1094 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1099 * Make sure that there are no blocks in front of the head
1100 * with the same cycle number as the head. This can happen
1101 * because we allow multiple outstanding log writes concurrently,
1102 * and the later writes might make it out before earlier ones.
1104 * We use the lsn from before modifying it so that we'll never
1105 * overwrite the unmount record after a clean unmount.
1107 * Do this only if we are going to recover the filesystem
1109 * NOTE: This used to say "if (!readonly)"
1110 * However on Linux, we can & do recover a read-only filesystem.
1111 * We only skip recovery if NORECOVERY is specified on mount,
1112 * in which case we would not be here.
1114 * But... if the -device- itself is readonly, just skip this.
1115 * We can't recover this device anyway, so it won't matter.
1117 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
1118 error = xlog_clear_stale_blocks(log, tail_lsn);
1124 xfs_warn(log->l_mp, "failed to locate log tail");
1129 * Is the log zeroed at all?
1131 * The last binary search should be changed to perform an X block read
1132 * once X becomes small enough. You can then search linearly through
1133 * the X blocks. This will cut down on the number of reads we need to do.
1135 * If the log is partially zeroed, this routine will pass back the blkno
1136 * of the first block with cycle number 0. It won't have a complete LR
1140 * 0 => the log is completely written to
1141 * 1 => use *blk_no as the first block of the log
1142 * <0 => error has occurred
1147 xfs_daddr_t *blk_no)
1151 uint first_cycle, last_cycle;
1152 xfs_daddr_t new_blk, last_blk, start_blk;
1153 xfs_daddr_t num_scan_bblks;
1154 int error, log_bbnum = log->l_logBBsize;
1158 /* check totally zeroed log */
1159 bp = xlog_get_bp(log, 1);
1162 error = xlog_bread(log, 0, 1, bp, &offset);
1166 first_cycle = xlog_get_cycle(offset);
1167 if (first_cycle == 0) { /* completely zeroed log */
1173 /* check partially zeroed log */
1174 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1178 last_cycle = xlog_get_cycle(offset);
1179 if (last_cycle != 0) { /* log completely written to */
1182 } else if (first_cycle != 1) {
1184 * If the cycle of the last block is zero, the cycle of
1185 * the first block must be 1. If it's not, maybe we're
1186 * not looking at a log... Bail out.
1189 "Log inconsistent or not a log (last==0, first!=1)");
1194 /* we have a partially zeroed log */
1195 last_blk = log_bbnum-1;
1196 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1200 * Validate the answer. Because there is no way to guarantee that
1201 * the entire log is made up of log records which are the same size,
1202 * we scan over the defined maximum blocks. At this point, the maximum
1203 * is not chosen to mean anything special. XXXmiken
1205 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1206 ASSERT(num_scan_bblks <= INT_MAX);
1208 if (last_blk < num_scan_bblks)
1209 num_scan_bblks = last_blk;
1210 start_blk = last_blk - num_scan_bblks;
1213 * We search for any instances of cycle number 0 that occur before
1214 * our current estimate of the head. What we're trying to detect is
1215 * 1 ... | 0 | 1 | 0...
1216 * ^ binary search ends here
1218 if ((error = xlog_find_verify_cycle(log, start_blk,
1219 (int)num_scan_bblks, 0, &new_blk)))
1225 * Potentially backup over partial log record write. We don't need
1226 * to search the end of the log because we know it is zero.
1228 error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0);
1243 * These are simple subroutines used by xlog_clear_stale_blocks() below
1244 * to initialize a buffer full of empty log record headers and write
1245 * them into the log.
1256 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1258 memset(buf, 0, BBSIZE);
1259 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1260 recp->h_cycle = cpu_to_be32(cycle);
1261 recp->h_version = cpu_to_be32(
1262 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1263 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1264 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1265 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1266 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1270 xlog_write_log_records(
1281 int sectbb = log->l_sectBBsize;
1282 int end_block = start_block + blocks;
1288 * Greedily allocate a buffer big enough to handle the full
1289 * range of basic blocks to be written. If that fails, try
1290 * a smaller size. We need to be able to write at least a
1291 * log sector, or we're out of luck.
1293 bufblks = 1 << ffs(blocks);
1294 while (bufblks > log->l_logBBsize)
1296 while (!(bp = xlog_get_bp(log, bufblks))) {
1298 if (bufblks < sectbb)
1302 /* We may need to do a read at the start to fill in part of
1303 * the buffer in the starting sector not covered by the first
1306 balign = round_down(start_block, sectbb);
1307 if (balign != start_block) {
1308 error = xlog_bread_noalign(log, start_block, 1, bp);
1312 j = start_block - balign;
1315 for (i = start_block; i < end_block; i += bufblks) {
1316 int bcount, endcount;
1318 bcount = min(bufblks, end_block - start_block);
1319 endcount = bcount - j;
1321 /* We may need to do a read at the end to fill in part of
1322 * the buffer in the final sector not covered by the write.
1323 * If this is the same sector as the above read, skip it.
1325 ealign = round_down(end_block, sectbb);
1326 if (j == 0 && (start_block + endcount > ealign)) {
1327 offset = bp->b_addr + BBTOB(ealign - start_block);
1328 error = xlog_bread_offset(log, ealign, sectbb,
1335 offset = xlog_align(log, start_block, endcount, bp);
1336 for (; j < endcount; j++) {
1337 xlog_add_record(log, offset, cycle, i+j,
1338 tail_cycle, tail_block);
1341 error = xlog_bwrite(log, start_block, endcount, bp);
1344 start_block += endcount;
1354 * This routine is called to blow away any incomplete log writes out
1355 * in front of the log head. We do this so that we won't become confused
1356 * if we come up, write only a little bit more, and then crash again.
1357 * If we leave the partial log records out there, this situation could
1358 * cause us to think those partial writes are valid blocks since they
1359 * have the current cycle number. We get rid of them by overwriting them
1360 * with empty log records with the old cycle number rather than the
1363 * The tail lsn is passed in rather than taken from
1364 * the log so that we will not write over the unmount record after a
1365 * clean unmount in a 512 block log. Doing so would leave the log without
1366 * any valid log records in it until a new one was written. If we crashed
1367 * during that time we would not be able to recover.
1370 xlog_clear_stale_blocks(
1374 int tail_cycle, head_cycle;
1375 int tail_block, head_block;
1376 int tail_distance, max_distance;
1380 tail_cycle = CYCLE_LSN(tail_lsn);
1381 tail_block = BLOCK_LSN(tail_lsn);
1382 head_cycle = log->l_curr_cycle;
1383 head_block = log->l_curr_block;
1386 * Figure out the distance between the new head of the log
1387 * and the tail. We want to write over any blocks beyond the
1388 * head that we may have written just before the crash, but
1389 * we don't want to overwrite the tail of the log.
1391 if (head_cycle == tail_cycle) {
1393 * The tail is behind the head in the physical log,
1394 * so the distance from the head to the tail is the
1395 * distance from the head to the end of the log plus
1396 * the distance from the beginning of the log to the
1399 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1400 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1401 XFS_ERRLEVEL_LOW, log->l_mp);
1402 return -EFSCORRUPTED;
1404 tail_distance = tail_block + (log->l_logBBsize - head_block);
1407 * The head is behind the tail in the physical log,
1408 * so the distance from the head to the tail is just
1409 * the tail block minus the head block.
1411 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1412 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1413 XFS_ERRLEVEL_LOW, log->l_mp);
1414 return -EFSCORRUPTED;
1416 tail_distance = tail_block - head_block;
1420 * If the head is right up against the tail, we can't clear
1423 if (tail_distance <= 0) {
1424 ASSERT(tail_distance == 0);
1428 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1430 * Take the smaller of the maximum amount of outstanding I/O
1431 * we could have and the distance to the tail to clear out.
1432 * We take the smaller so that we don't overwrite the tail and
1433 * we don't waste all day writing from the head to the tail
1436 max_distance = MIN(max_distance, tail_distance);
1438 if ((head_block + max_distance) <= log->l_logBBsize) {
1440 * We can stomp all the blocks we need to without
1441 * wrapping around the end of the log. Just do it
1442 * in a single write. Use the cycle number of the
1443 * current cycle minus one so that the log will look like:
1446 error = xlog_write_log_records(log, (head_cycle - 1),
1447 head_block, max_distance, tail_cycle,
1453 * We need to wrap around the end of the physical log in
1454 * order to clear all the blocks. Do it in two separate
1455 * I/Os. The first write should be from the head to the
1456 * end of the physical log, and it should use the current
1457 * cycle number minus one just like above.
1459 distance = log->l_logBBsize - head_block;
1460 error = xlog_write_log_records(log, (head_cycle - 1),
1461 head_block, distance, tail_cycle,
1468 * Now write the blocks at the start of the physical log.
1469 * This writes the remainder of the blocks we want to clear.
1470 * It uses the current cycle number since we're now on the
1471 * same cycle as the head so that we get:
1472 * n ... n ... | n - 1 ...
1473 * ^^^^^ blocks we're writing
1475 distance = max_distance - (log->l_logBBsize - head_block);
1476 error = xlog_write_log_records(log, head_cycle, 0, distance,
1477 tail_cycle, tail_block);
1485 /******************************************************************************
1487 * Log recover routines
1489 ******************************************************************************
1493 * Sort the log items in the transaction.
1495 * The ordering constraints are defined by the inode allocation and unlink
1496 * behaviour. The rules are:
1498 * 1. Every item is only logged once in a given transaction. Hence it
1499 * represents the last logged state of the item. Hence ordering is
1500 * dependent on the order in which operations need to be performed so
1501 * required initial conditions are always met.
1503 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1504 * there's nothing to replay from them so we can simply cull them
1505 * from the transaction. However, we can't do that until after we've
1506 * replayed all the other items because they may be dependent on the
1507 * cancelled buffer and replaying the cancelled buffer can remove it
1508 * form the cancelled buffer table. Hence they have tobe done last.
1510 * 3. Inode allocation buffers must be replayed before inode items that
1511 * read the buffer and replay changes into it. For filesystems using the
1512 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1513 * treated the same as inode allocation buffers as they create and
1514 * initialise the buffers directly.
1516 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1517 * This ensures that inodes are completely flushed to the inode buffer
1518 * in a "free" state before we remove the unlinked inode list pointer.
1520 * Hence the ordering needs to be inode allocation buffers first, inode items
1521 * second, inode unlink buffers third and cancelled buffers last.
1523 * But there's a problem with that - we can't tell an inode allocation buffer
1524 * apart from a regular buffer, so we can't separate them. We can, however,
1525 * tell an inode unlink buffer from the others, and so we can separate them out
1526 * from all the other buffers and move them to last.
1528 * Hence, 4 lists, in order from head to tail:
1529 * - buffer_list for all buffers except cancelled/inode unlink buffers
1530 * - item_list for all non-buffer items
1531 * - inode_buffer_list for inode unlink buffers
1532 * - cancel_list for the cancelled buffers
1534 * Note that we add objects to the tail of the lists so that first-to-last
1535 * ordering is preserved within the lists. Adding objects to the head of the
1536 * list means when we traverse from the head we walk them in last-to-first
1537 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1538 * but for all other items there may be specific ordering that we need to
1542 xlog_recover_reorder_trans(
1544 struct xlog_recover *trans,
1547 xlog_recover_item_t *item, *n;
1549 LIST_HEAD(sort_list);
1550 LIST_HEAD(cancel_list);
1551 LIST_HEAD(buffer_list);
1552 LIST_HEAD(inode_buffer_list);
1553 LIST_HEAD(inode_list);
1555 list_splice_init(&trans->r_itemq, &sort_list);
1556 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1557 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1559 switch (ITEM_TYPE(item)) {
1560 case XFS_LI_ICREATE:
1561 list_move_tail(&item->ri_list, &buffer_list);
1564 if (buf_f->blf_flags & XFS_BLF_CANCEL) {
1565 trace_xfs_log_recover_item_reorder_head(log,
1567 list_move(&item->ri_list, &cancel_list);
1570 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1571 list_move(&item->ri_list, &inode_buffer_list);
1574 list_move_tail(&item->ri_list, &buffer_list);
1578 case XFS_LI_QUOTAOFF:
1581 trace_xfs_log_recover_item_reorder_tail(log,
1583 list_move_tail(&item->ri_list, &inode_list);
1587 "%s: unrecognized type of log operation",
1591 * return the remaining items back to the transaction
1592 * item list so they can be freed in caller.
1594 if (!list_empty(&sort_list))
1595 list_splice_init(&sort_list, &trans->r_itemq);
1601 ASSERT(list_empty(&sort_list));
1602 if (!list_empty(&buffer_list))
1603 list_splice(&buffer_list, &trans->r_itemq);
1604 if (!list_empty(&inode_list))
1605 list_splice_tail(&inode_list, &trans->r_itemq);
1606 if (!list_empty(&inode_buffer_list))
1607 list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1608 if (!list_empty(&cancel_list))
1609 list_splice_tail(&cancel_list, &trans->r_itemq);
1614 * Build up the table of buf cancel records so that we don't replay
1615 * cancelled data in the second pass. For buffer records that are
1616 * not cancel records, there is nothing to do here so we just return.
1618 * If we get a cancel record which is already in the table, this indicates
1619 * that the buffer was cancelled multiple times. In order to ensure
1620 * that during pass 2 we keep the record in the table until we reach its
1621 * last occurrence in the log, we keep a reference count in the cancel
1622 * record in the table to tell us how many times we expect to see this
1623 * record during the second pass.
1626 xlog_recover_buffer_pass1(
1628 struct xlog_recover_item *item)
1630 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1631 struct list_head *bucket;
1632 struct xfs_buf_cancel *bcp;
1635 * If this isn't a cancel buffer item, then just return.
1637 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1638 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
1643 * Insert an xfs_buf_cancel record into the hash table of them.
1644 * If there is already an identical record, bump its reference count.
1646 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1647 list_for_each_entry(bcp, bucket, bc_list) {
1648 if (bcp->bc_blkno == buf_f->blf_blkno &&
1649 bcp->bc_len == buf_f->blf_len) {
1651 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
1656 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1657 bcp->bc_blkno = buf_f->blf_blkno;
1658 bcp->bc_len = buf_f->blf_len;
1659 bcp->bc_refcount = 1;
1660 list_add_tail(&bcp->bc_list, bucket);
1662 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
1667 * Check to see whether the buffer being recovered has a corresponding
1668 * entry in the buffer cancel record table. If it is, return the cancel
1669 * buffer structure to the caller.
1671 STATIC struct xfs_buf_cancel *
1672 xlog_peek_buffer_cancelled(
1678 struct list_head *bucket;
1679 struct xfs_buf_cancel *bcp;
1681 if (!log->l_buf_cancel_table) {
1682 /* empty table means no cancelled buffers in the log */
1683 ASSERT(!(flags & XFS_BLF_CANCEL));
1687 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1688 list_for_each_entry(bcp, bucket, bc_list) {
1689 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
1694 * We didn't find a corresponding entry in the table, so return 0 so
1695 * that the buffer is NOT cancelled.
1697 ASSERT(!(flags & XFS_BLF_CANCEL));
1702 * If the buffer is being cancelled then return 1 so that it will be cancelled,
1703 * otherwise return 0. If the buffer is actually a buffer cancel item
1704 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
1705 * table and remove it from the table if this is the last reference.
1707 * We remove the cancel record from the table when we encounter its last
1708 * occurrence in the log so that if the same buffer is re-used again after its
1709 * last cancellation we actually replay the changes made at that point.
1712 xlog_check_buffer_cancelled(
1718 struct xfs_buf_cancel *bcp;
1720 bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags);
1725 * We've go a match, so return 1 so that the recovery of this buffer
1726 * is cancelled. If this buffer is actually a buffer cancel log
1727 * item, then decrement the refcount on the one in the table and
1728 * remove it if this is the last reference.
1730 if (flags & XFS_BLF_CANCEL) {
1731 if (--bcp->bc_refcount == 0) {
1732 list_del(&bcp->bc_list);
1740 * Perform recovery for a buffer full of inodes. In these buffers, the only
1741 * data which should be recovered is that which corresponds to the
1742 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1743 * data for the inodes is always logged through the inodes themselves rather
1744 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1746 * The only time when buffers full of inodes are fully recovered is when the
1747 * buffer is full of newly allocated inodes. In this case the buffer will
1748 * not be marked as an inode buffer and so will be sent to
1749 * xlog_recover_do_reg_buffer() below during recovery.
1752 xlog_recover_do_inode_buffer(
1753 struct xfs_mount *mp,
1754 xlog_recover_item_t *item,
1756 xfs_buf_log_format_t *buf_f)
1762 int reg_buf_offset = 0;
1763 int reg_buf_bytes = 0;
1764 int next_unlinked_offset;
1766 xfs_agino_t *logged_nextp;
1767 xfs_agino_t *buffer_nextp;
1769 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
1772 * Post recovery validation only works properly on CRC enabled
1775 if (xfs_sb_version_hascrc(&mp->m_sb))
1776 bp->b_ops = &xfs_inode_buf_ops;
1778 inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog;
1779 for (i = 0; i < inodes_per_buf; i++) {
1780 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1781 offsetof(xfs_dinode_t, di_next_unlinked);
1783 while (next_unlinked_offset >=
1784 (reg_buf_offset + reg_buf_bytes)) {
1786 * The next di_next_unlinked field is beyond
1787 * the current logged region. Find the next
1788 * logged region that contains or is beyond
1789 * the current di_next_unlinked field.
1792 bit = xfs_next_bit(buf_f->blf_data_map,
1793 buf_f->blf_map_size, bit);
1796 * If there are no more logged regions in the
1797 * buffer, then we're done.
1802 nbits = xfs_contig_bits(buf_f->blf_data_map,
1803 buf_f->blf_map_size, bit);
1805 reg_buf_offset = bit << XFS_BLF_SHIFT;
1806 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
1811 * If the current logged region starts after the current
1812 * di_next_unlinked field, then move on to the next
1813 * di_next_unlinked field.
1815 if (next_unlinked_offset < reg_buf_offset)
1818 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1819 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
1820 ASSERT((reg_buf_offset + reg_buf_bytes) <=
1821 BBTOB(bp->b_io_length));
1824 * The current logged region contains a copy of the
1825 * current di_next_unlinked field. Extract its value
1826 * and copy it to the buffer copy.
1828 logged_nextp = item->ri_buf[item_index].i_addr +
1829 next_unlinked_offset - reg_buf_offset;
1830 if (unlikely(*logged_nextp == 0)) {
1832 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1833 "Trying to replay bad (0) inode di_next_unlinked field.",
1835 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1836 XFS_ERRLEVEL_LOW, mp);
1837 return -EFSCORRUPTED;
1840 buffer_nextp = xfs_buf_offset(bp, next_unlinked_offset);
1841 *buffer_nextp = *logged_nextp;
1844 * If necessary, recalculate the CRC in the on-disk inode. We
1845 * have to leave the inode in a consistent state for whoever
1848 xfs_dinode_calc_crc(mp,
1849 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
1857 * V5 filesystems know the age of the buffer on disk being recovered. We can
1858 * have newer objects on disk than we are replaying, and so for these cases we
1859 * don't want to replay the current change as that will make the buffer contents
1860 * temporarily invalid on disk.
1862 * The magic number might not match the buffer type we are going to recover
1863 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
1864 * extract the LSN of the existing object in the buffer based on it's current
1865 * magic number. If we don't recognise the magic number in the buffer, then
1866 * return a LSN of -1 so that the caller knows it was an unrecognised block and
1867 * so can recover the buffer.
1869 * Note: we cannot rely solely on magic number matches to determine that the
1870 * buffer has a valid LSN - we also need to verify that it belongs to this
1871 * filesystem, so we need to extract the object's LSN and compare it to that
1872 * which we read from the superblock. If the UUIDs don't match, then we've got a
1873 * stale metadata block from an old filesystem instance that we need to recover
1877 xlog_recover_get_buf_lsn(
1878 struct xfs_mount *mp,
1884 void *blk = bp->b_addr;
1888 /* v4 filesystems always recover immediately */
1889 if (!xfs_sb_version_hascrc(&mp->m_sb))
1890 goto recover_immediately;
1892 magic32 = be32_to_cpu(*(__be32 *)blk);
1894 case XFS_ABTB_CRC_MAGIC:
1895 case XFS_ABTC_CRC_MAGIC:
1896 case XFS_ABTB_MAGIC:
1897 case XFS_ABTC_MAGIC:
1898 case XFS_IBT_CRC_MAGIC:
1899 case XFS_IBT_MAGIC: {
1900 struct xfs_btree_block *btb = blk;
1902 lsn = be64_to_cpu(btb->bb_u.s.bb_lsn);
1903 uuid = &btb->bb_u.s.bb_uuid;
1906 case XFS_BMAP_CRC_MAGIC:
1907 case XFS_BMAP_MAGIC: {
1908 struct xfs_btree_block *btb = blk;
1910 lsn = be64_to_cpu(btb->bb_u.l.bb_lsn);
1911 uuid = &btb->bb_u.l.bb_uuid;
1915 lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn);
1916 uuid = &((struct xfs_agf *)blk)->agf_uuid;
1918 case XFS_AGFL_MAGIC:
1919 lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn);
1920 uuid = &((struct xfs_agfl *)blk)->agfl_uuid;
1923 lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn);
1924 uuid = &((struct xfs_agi *)blk)->agi_uuid;
1926 case XFS_SYMLINK_MAGIC:
1927 lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn);
1928 uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid;
1930 case XFS_DIR3_BLOCK_MAGIC:
1931 case XFS_DIR3_DATA_MAGIC:
1932 case XFS_DIR3_FREE_MAGIC:
1933 lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn);
1934 uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid;
1936 case XFS_ATTR3_RMT_MAGIC:
1938 * Remote attr blocks are written synchronously, rather than
1939 * being logged. That means they do not contain a valid LSN
1940 * (i.e. transactionally ordered) in them, and hence any time we
1941 * see a buffer to replay over the top of a remote attribute
1942 * block we should simply do so.
1944 goto recover_immediately;
1947 * superblock uuids are magic. We may or may not have a
1948 * sb_meta_uuid on disk, but it will be set in the in-core
1949 * superblock. We set the uuid pointer for verification
1950 * according to the superblock feature mask to ensure we check
1951 * the relevant UUID in the superblock.
1953 lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn);
1954 if (xfs_sb_version_hasmetauuid(&mp->m_sb))
1955 uuid = &((struct xfs_dsb *)blk)->sb_meta_uuid;
1957 uuid = &((struct xfs_dsb *)blk)->sb_uuid;
1963 if (lsn != (xfs_lsn_t)-1) {
1964 if (!uuid_equal(&mp->m_sb.sb_meta_uuid, uuid))
1965 goto recover_immediately;
1969 magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic);
1971 case XFS_DIR3_LEAF1_MAGIC:
1972 case XFS_DIR3_LEAFN_MAGIC:
1973 case XFS_DA3_NODE_MAGIC:
1974 lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn);
1975 uuid = &((struct xfs_da3_blkinfo *)blk)->uuid;
1981 if (lsn != (xfs_lsn_t)-1) {
1982 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
1983 goto recover_immediately;
1988 * We do individual object checks on dquot and inode buffers as they
1989 * have their own individual LSN records. Also, we could have a stale
1990 * buffer here, so we have to at least recognise these buffer types.
1992 * A notd complexity here is inode unlinked list processing - it logs
1993 * the inode directly in the buffer, but we don't know which inodes have
1994 * been modified, and there is no global buffer LSN. Hence we need to
1995 * recover all inode buffer types immediately. This problem will be
1996 * fixed by logical logging of the unlinked list modifications.
1998 magic16 = be16_to_cpu(*(__be16 *)blk);
2000 case XFS_DQUOT_MAGIC:
2001 case XFS_DINODE_MAGIC:
2002 goto recover_immediately;
2007 /* unknown buffer contents, recover immediately */
2009 recover_immediately:
2010 return (xfs_lsn_t)-1;
2015 * Validate the recovered buffer is of the correct type and attach the
2016 * appropriate buffer operations to them for writeback. Magic numbers are in a
2018 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2019 * the first 32 bits of the buffer (most blocks),
2020 * inside a struct xfs_da_blkinfo at the start of the buffer.
2023 xlog_recover_validate_buf_type(
2024 struct xfs_mount *mp,
2026 xfs_buf_log_format_t *buf_f)
2028 struct xfs_da_blkinfo *info = bp->b_addr;
2034 * We can only do post recovery validation on items on CRC enabled
2035 * fielsystems as we need to know when the buffer was written to be able
2036 * to determine if we should have replayed the item. If we replay old
2037 * metadata over a newer buffer, then it will enter a temporarily
2038 * inconsistent state resulting in verification failures. Hence for now
2039 * just avoid the verification stage for non-crc filesystems
2041 if (!xfs_sb_version_hascrc(&mp->m_sb))
2044 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
2045 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
2046 magicda = be16_to_cpu(info->magic);
2047 switch (xfs_blft_from_flags(buf_f)) {
2048 case XFS_BLFT_BTREE_BUF:
2050 case XFS_ABTB_CRC_MAGIC:
2051 case XFS_ABTC_CRC_MAGIC:
2052 case XFS_ABTB_MAGIC:
2053 case XFS_ABTC_MAGIC:
2054 bp->b_ops = &xfs_allocbt_buf_ops;
2056 case XFS_IBT_CRC_MAGIC:
2057 case XFS_FIBT_CRC_MAGIC:
2059 case XFS_FIBT_MAGIC:
2060 bp->b_ops = &xfs_inobt_buf_ops;
2062 case XFS_BMAP_CRC_MAGIC:
2063 case XFS_BMAP_MAGIC:
2064 bp->b_ops = &xfs_bmbt_buf_ops;
2067 xfs_warn(mp, "Bad btree block magic!");
2072 case XFS_BLFT_AGF_BUF:
2073 if (magic32 != XFS_AGF_MAGIC) {
2074 xfs_warn(mp, "Bad AGF block magic!");
2078 bp->b_ops = &xfs_agf_buf_ops;
2080 case XFS_BLFT_AGFL_BUF:
2081 if (magic32 != XFS_AGFL_MAGIC) {
2082 xfs_warn(mp, "Bad AGFL block magic!");
2086 bp->b_ops = &xfs_agfl_buf_ops;
2088 case XFS_BLFT_AGI_BUF:
2089 if (magic32 != XFS_AGI_MAGIC) {
2090 xfs_warn(mp, "Bad AGI block magic!");
2094 bp->b_ops = &xfs_agi_buf_ops;
2096 case XFS_BLFT_UDQUOT_BUF:
2097 case XFS_BLFT_PDQUOT_BUF:
2098 case XFS_BLFT_GDQUOT_BUF:
2099 #ifdef CONFIG_XFS_QUOTA
2100 if (magic16 != XFS_DQUOT_MAGIC) {
2101 xfs_warn(mp, "Bad DQUOT block magic!");
2105 bp->b_ops = &xfs_dquot_buf_ops;
2108 "Trying to recover dquots without QUOTA support built in!");
2112 case XFS_BLFT_DINO_BUF:
2113 if (magic16 != XFS_DINODE_MAGIC) {
2114 xfs_warn(mp, "Bad INODE block magic!");
2118 bp->b_ops = &xfs_inode_buf_ops;
2120 case XFS_BLFT_SYMLINK_BUF:
2121 if (magic32 != XFS_SYMLINK_MAGIC) {
2122 xfs_warn(mp, "Bad symlink block magic!");
2126 bp->b_ops = &xfs_symlink_buf_ops;
2128 case XFS_BLFT_DIR_BLOCK_BUF:
2129 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2130 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2131 xfs_warn(mp, "Bad dir block magic!");
2135 bp->b_ops = &xfs_dir3_block_buf_ops;
2137 case XFS_BLFT_DIR_DATA_BUF:
2138 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2139 magic32 != XFS_DIR3_DATA_MAGIC) {
2140 xfs_warn(mp, "Bad dir data magic!");
2144 bp->b_ops = &xfs_dir3_data_buf_ops;
2146 case XFS_BLFT_DIR_FREE_BUF:
2147 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2148 magic32 != XFS_DIR3_FREE_MAGIC) {
2149 xfs_warn(mp, "Bad dir3 free magic!");
2153 bp->b_ops = &xfs_dir3_free_buf_ops;
2155 case XFS_BLFT_DIR_LEAF1_BUF:
2156 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2157 magicda != XFS_DIR3_LEAF1_MAGIC) {
2158 xfs_warn(mp, "Bad dir leaf1 magic!");
2162 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2164 case XFS_BLFT_DIR_LEAFN_BUF:
2165 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2166 magicda != XFS_DIR3_LEAFN_MAGIC) {
2167 xfs_warn(mp, "Bad dir leafn magic!");
2171 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2173 case XFS_BLFT_DA_NODE_BUF:
2174 if (magicda != XFS_DA_NODE_MAGIC &&
2175 magicda != XFS_DA3_NODE_MAGIC) {
2176 xfs_warn(mp, "Bad da node magic!");
2180 bp->b_ops = &xfs_da3_node_buf_ops;
2182 case XFS_BLFT_ATTR_LEAF_BUF:
2183 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2184 magicda != XFS_ATTR3_LEAF_MAGIC) {
2185 xfs_warn(mp, "Bad attr leaf magic!");
2189 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2191 case XFS_BLFT_ATTR_RMT_BUF:
2192 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
2193 xfs_warn(mp, "Bad attr remote magic!");
2197 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2199 case XFS_BLFT_SB_BUF:
2200 if (magic32 != XFS_SB_MAGIC) {
2201 xfs_warn(mp, "Bad SB block magic!");
2205 bp->b_ops = &xfs_sb_buf_ops;
2208 xfs_warn(mp, "Unknown buffer type %d!",
2209 xfs_blft_from_flags(buf_f));
2215 * Perform a 'normal' buffer recovery. Each logged region of the
2216 * buffer should be copied over the corresponding region in the
2217 * given buffer. The bitmap in the buf log format structure indicates
2218 * where to place the logged data.
2221 xlog_recover_do_reg_buffer(
2222 struct xfs_mount *mp,
2223 xlog_recover_item_t *item,
2225 xfs_buf_log_format_t *buf_f)
2232 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2235 i = 1; /* 0 is the buf format structure */
2237 bit = xfs_next_bit(buf_f->blf_data_map,
2238 buf_f->blf_map_size, bit);
2241 nbits = xfs_contig_bits(buf_f->blf_data_map,
2242 buf_f->blf_map_size, bit);
2244 ASSERT(item->ri_buf[i].i_addr != NULL);
2245 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
2246 ASSERT(BBTOB(bp->b_io_length) >=
2247 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
2250 * The dirty regions logged in the buffer, even though
2251 * contiguous, may span multiple chunks. This is because the
2252 * dirty region may span a physical page boundary in a buffer
2253 * and hence be split into two separate vectors for writing into
2254 * the log. Hence we need to trim nbits back to the length of
2255 * the current region being copied out of the log.
2257 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2258 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2261 * Do a sanity check if this is a dquot buffer. Just checking
2262 * the first dquot in the buffer should do. XXXThis is
2263 * probably a good thing to do for other buf types also.
2266 if (buf_f->blf_flags &
2267 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2268 if (item->ri_buf[i].i_addr == NULL) {
2270 "XFS: NULL dquot in %s.", __func__);
2273 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
2275 "XFS: dquot too small (%d) in %s.",
2276 item->ri_buf[i].i_len, __func__);
2279 error = xfs_dqcheck(mp, item->ri_buf[i].i_addr,
2280 -1, 0, XFS_QMOPT_DOWARN,
2281 "dquot_buf_recover");
2286 memcpy(xfs_buf_offset(bp,
2287 (uint)bit << XFS_BLF_SHIFT), /* dest */
2288 item->ri_buf[i].i_addr, /* source */
2289 nbits<<XFS_BLF_SHIFT); /* length */
2295 /* Shouldn't be any more regions */
2296 ASSERT(i == item->ri_total);
2298 xlog_recover_validate_buf_type(mp, bp, buf_f);
2302 * Perform a dquot buffer recovery.
2303 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2304 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2305 * Else, treat it as a regular buffer and do recovery.
2307 * Return false if the buffer was tossed and true if we recovered the buffer to
2308 * indicate to the caller if the buffer needs writing.
2311 xlog_recover_do_dquot_buffer(
2312 struct xfs_mount *mp,
2314 struct xlog_recover_item *item,
2316 struct xfs_buf_log_format *buf_f)
2320 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2323 * Filesystems are required to send in quota flags at mount time.
2329 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2330 type |= XFS_DQ_USER;
2331 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2332 type |= XFS_DQ_PROJ;
2333 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2334 type |= XFS_DQ_GROUP;
2336 * This type of quotas was turned off, so ignore this buffer
2338 if (log->l_quotaoffs_flag & type)
2341 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2346 * This routine replays a modification made to a buffer at runtime.
2347 * There are actually two types of buffer, regular and inode, which
2348 * are handled differently. Inode buffers are handled differently
2349 * in that we only recover a specific set of data from them, namely
2350 * the inode di_next_unlinked fields. This is because all other inode
2351 * data is actually logged via inode records and any data we replay
2352 * here which overlaps that may be stale.
2354 * When meta-data buffers are freed at run time we log a buffer item
2355 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2356 * of the buffer in the log should not be replayed at recovery time.
2357 * This is so that if the blocks covered by the buffer are reused for
2358 * file data before we crash we don't end up replaying old, freed
2359 * meta-data into a user's file.
2361 * To handle the cancellation of buffer log items, we make two passes
2362 * over the log during recovery. During the first we build a table of
2363 * those buffers which have been cancelled, and during the second we
2364 * only replay those buffers which do not have corresponding cancel
2365 * records in the table. See xlog_recover_buffer_pass[1,2] above
2366 * for more details on the implementation of the table of cancel records.
2369 xlog_recover_buffer_pass2(
2371 struct list_head *buffer_list,
2372 struct xlog_recover_item *item,
2373 xfs_lsn_t current_lsn)
2375 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2376 xfs_mount_t *mp = log->l_mp;
2383 * In this pass we only want to recover all the buffers which have
2384 * not been cancelled and are not cancellation buffers themselves.
2386 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2387 buf_f->blf_len, buf_f->blf_flags)) {
2388 trace_xfs_log_recover_buf_cancel(log, buf_f);
2392 trace_xfs_log_recover_buf_recover(log, buf_f);
2395 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2396 buf_flags |= XBF_UNMAPPED;
2398 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2402 error = bp->b_error;
2404 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
2409 * Recover the buffer only if we get an LSN from it and it's less than
2410 * the lsn of the transaction we are replaying.
2412 * Note that we have to be extremely careful of readahead here.
2413 * Readahead does not attach verfiers to the buffers so if we don't
2414 * actually do any replay after readahead because of the LSN we found
2415 * in the buffer if more recent than that current transaction then we
2416 * need to attach the verifier directly. Failure to do so can lead to
2417 * future recovery actions (e.g. EFI and unlinked list recovery) can
2418 * operate on the buffers and they won't get the verifier attached. This
2419 * can lead to blocks on disk having the correct content but a stale
2422 * It is safe to assume these clean buffers are currently up to date.
2423 * If the buffer is dirtied by a later transaction being replayed, then
2424 * the verifier will be reset to match whatever recover turns that
2427 lsn = xlog_recover_get_buf_lsn(mp, bp);
2428 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2429 xlog_recover_validate_buf_type(mp, bp, buf_f);
2433 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
2434 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2437 } else if (buf_f->blf_flags &
2438 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2441 dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2445 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2449 * Perform delayed write on the buffer. Asynchronous writes will be
2450 * slower when taking into account all the buffers to be flushed.
2452 * Also make sure that only inode buffers with good sizes stay in
2453 * the buffer cache. The kernel moves inodes in buffers of 1 block
2454 * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode
2455 * buffers in the log can be a different size if the log was generated
2456 * by an older kernel using unclustered inode buffers or a newer kernel
2457 * running with a different inode cluster size. Regardless, if the
2458 * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size)
2459 * for *our* value of mp->m_inode_cluster_size, then we need to keep
2460 * the buffer out of the buffer cache so that the buffer won't
2461 * overlap with future reads of those inodes.
2463 if (XFS_DINODE_MAGIC ==
2464 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2465 (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize,
2466 (__uint32_t)log->l_mp->m_inode_cluster_size))) {
2468 error = xfs_bwrite(bp);
2470 ASSERT(bp->b_target->bt_mount == mp);
2471 bp->b_iodone = xlog_recover_iodone;
2472 xfs_buf_delwri_queue(bp, buffer_list);
2481 * Inode fork owner changes
2483 * If we have been told that we have to reparent the inode fork, it's because an
2484 * extent swap operation on a CRC enabled filesystem has been done and we are
2485 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2488 * The complexity here is that we don't have an inode context to work with, so
2489 * after we've replayed the inode we need to instantiate one. This is where the
2492 * We are in the middle of log recovery, so we can't run transactions. That
2493 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2494 * that will result in the corresponding iput() running the inode through
2495 * xfs_inactive(). If we've just replayed an inode core that changes the link
2496 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2497 * transactions (bad!).
2499 * So, to avoid this, we instantiate an inode directly from the inode core we've
2500 * just recovered. We have the buffer still locked, and all we really need to
2501 * instantiate is the inode core and the forks being modified. We can do this
2502 * manually, then run the inode btree owner change, and then tear down the
2503 * xfs_inode without having to run any transactions at all.
2505 * Also, because we don't have a transaction context available here but need to
2506 * gather all the buffers we modify for writeback so we pass the buffer_list
2507 * instead for the operation to use.
2511 xfs_recover_inode_owner_change(
2512 struct xfs_mount *mp,
2513 struct xfs_dinode *dip,
2514 struct xfs_inode_log_format *in_f,
2515 struct list_head *buffer_list)
2517 struct xfs_inode *ip;
2520 ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER));
2522 ip = xfs_inode_alloc(mp, in_f->ilf_ino);
2526 /* instantiate the inode */
2527 xfs_dinode_from_disk(&ip->i_d, dip);
2528 ASSERT(ip->i_d.di_version >= 3);
2530 error = xfs_iformat_fork(ip, dip);
2535 if (in_f->ilf_fields & XFS_ILOG_DOWNER) {
2536 ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT);
2537 error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK,
2538 ip->i_ino, buffer_list);
2543 if (in_f->ilf_fields & XFS_ILOG_AOWNER) {
2544 ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT);
2545 error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK,
2546 ip->i_ino, buffer_list);
2557 xlog_recover_inode_pass2(
2559 struct list_head *buffer_list,
2560 struct xlog_recover_item *item,
2561 xfs_lsn_t current_lsn)
2563 xfs_inode_log_format_t *in_f;
2564 xfs_mount_t *mp = log->l_mp;
2573 xfs_icdinode_t *dicp;
2577 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2578 in_f = item->ri_buf[0].i_addr;
2580 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
2582 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2588 * Inode buffers can be freed, look out for it,
2589 * and do not replay the inode.
2591 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2592 in_f->ilf_len, 0)) {
2594 trace_xfs_log_recover_inode_cancel(log, in_f);
2597 trace_xfs_log_recover_inode_recover(log, in_f);
2599 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
2600 &xfs_inode_buf_ops);
2605 error = bp->b_error;
2607 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
2610 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2611 dip = xfs_buf_offset(bp, in_f->ilf_boffset);
2614 * Make sure the place we're flushing out to really looks
2617 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
2619 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2620 __func__, dip, bp, in_f->ilf_ino);
2621 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2622 XFS_ERRLEVEL_LOW, mp);
2623 error = -EFSCORRUPTED;
2626 dicp = item->ri_buf[1].i_addr;
2627 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2629 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2630 __func__, item, in_f->ilf_ino);
2631 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2632 XFS_ERRLEVEL_LOW, mp);
2633 error = -EFSCORRUPTED;
2638 * If the inode has an LSN in it, recover the inode only if it's less
2639 * than the lsn of the transaction we are replaying. Note: we still
2640 * need to replay an owner change even though the inode is more recent
2641 * than the transaction as there is no guarantee that all the btree
2642 * blocks are more recent than this transaction, too.
2644 if (dip->di_version >= 3) {
2645 xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn);
2647 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2648 trace_xfs_log_recover_inode_skip(log, in_f);
2650 goto out_owner_change;
2655 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2656 * are transactional and if ordering is necessary we can determine that
2657 * more accurately by the LSN field in the V3 inode core. Don't trust
2658 * the inode versions we might be changing them here - use the
2659 * superblock flag to determine whether we need to look at di_flushiter
2660 * to skip replay when the on disk inode is newer than the log one
2662 if (!xfs_sb_version_hascrc(&mp->m_sb) &&
2663 dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
2665 * Deal with the wrap case, DI_MAX_FLUSH is less
2666 * than smaller numbers
2668 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
2669 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2672 trace_xfs_log_recover_inode_skip(log, in_f);
2678 /* Take the opportunity to reset the flush iteration count */
2679 dicp->di_flushiter = 0;
2681 if (unlikely(S_ISREG(dicp->di_mode))) {
2682 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2683 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2684 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2685 XFS_ERRLEVEL_LOW, mp, dicp);
2687 "%s: Bad regular inode log record, rec ptr 0x%p, "
2688 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2689 __func__, item, dip, bp, in_f->ilf_ino);
2690 error = -EFSCORRUPTED;
2693 } else if (unlikely(S_ISDIR(dicp->di_mode))) {
2694 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2695 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2696 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2697 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2698 XFS_ERRLEVEL_LOW, mp, dicp);
2700 "%s: Bad dir inode log record, rec ptr 0x%p, "
2701 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2702 __func__, item, dip, bp, in_f->ilf_ino);
2703 error = -EFSCORRUPTED;
2707 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2708 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2709 XFS_ERRLEVEL_LOW, mp, dicp);
2711 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2712 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2713 __func__, item, dip, bp, in_f->ilf_ino,
2714 dicp->di_nextents + dicp->di_anextents,
2716 error = -EFSCORRUPTED;
2719 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2720 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2721 XFS_ERRLEVEL_LOW, mp, dicp);
2723 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2724 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
2725 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
2726 error = -EFSCORRUPTED;
2729 isize = xfs_icdinode_size(dicp->di_version);
2730 if (unlikely(item->ri_buf[1].i_len > isize)) {
2731 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
2732 XFS_ERRLEVEL_LOW, mp, dicp);
2734 "%s: Bad inode log record length %d, rec ptr 0x%p",
2735 __func__, item->ri_buf[1].i_len, item);
2736 error = -EFSCORRUPTED;
2740 /* The core is in in-core format */
2741 xfs_dinode_to_disk(dip, dicp);
2743 /* the rest is in on-disk format */
2744 if (item->ri_buf[1].i_len > isize) {
2745 memcpy((char *)dip + isize,
2746 item->ri_buf[1].i_addr + isize,
2747 item->ri_buf[1].i_len - isize);
2750 fields = in_f->ilf_fields;
2751 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2753 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
2756 memcpy(XFS_DFORK_DPTR(dip),
2757 &in_f->ilf_u.ilfu_uuid,
2762 if (in_f->ilf_size == 2)
2763 goto out_owner_change;
2764 len = item->ri_buf[2].i_len;
2765 src = item->ri_buf[2].i_addr;
2766 ASSERT(in_f->ilf_size <= 4);
2767 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2768 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2769 (len == in_f->ilf_dsize));
2771 switch (fields & XFS_ILOG_DFORK) {
2772 case XFS_ILOG_DDATA:
2774 memcpy(XFS_DFORK_DPTR(dip), src, len);
2777 case XFS_ILOG_DBROOT:
2778 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
2779 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
2780 XFS_DFORK_DSIZE(dip, mp));
2785 * There are no data fork flags set.
2787 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2792 * If we logged any attribute data, recover it. There may or
2793 * may not have been any other non-core data logged in this
2796 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2797 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2802 len = item->ri_buf[attr_index].i_len;
2803 src = item->ri_buf[attr_index].i_addr;
2804 ASSERT(len == in_f->ilf_asize);
2806 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2807 case XFS_ILOG_ADATA:
2809 dest = XFS_DFORK_APTR(dip);
2810 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2811 memcpy(dest, src, len);
2814 case XFS_ILOG_ABROOT:
2815 dest = XFS_DFORK_APTR(dip);
2816 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2817 len, (xfs_bmdr_block_t*)dest,
2818 XFS_DFORK_ASIZE(dip, mp));
2822 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
2830 if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER))
2831 error = xfs_recover_inode_owner_change(mp, dip, in_f,
2833 /* re-generate the checksum. */
2834 xfs_dinode_calc_crc(log->l_mp, dip);
2836 ASSERT(bp->b_target->bt_mount == mp);
2837 bp->b_iodone = xlog_recover_iodone;
2838 xfs_buf_delwri_queue(bp, buffer_list);
2849 * Recover QUOTAOFF records. We simply make a note of it in the xlog
2850 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2854 xlog_recover_quotaoff_pass1(
2856 struct xlog_recover_item *item)
2858 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
2862 * The logitem format's flag tells us if this was user quotaoff,
2863 * group/project quotaoff or both.
2865 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2866 log->l_quotaoffs_flag |= XFS_DQ_USER;
2867 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2868 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2869 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2870 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2876 * Recover a dquot record
2879 xlog_recover_dquot_pass2(
2881 struct list_head *buffer_list,
2882 struct xlog_recover_item *item,
2883 xfs_lsn_t current_lsn)
2885 xfs_mount_t *mp = log->l_mp;
2887 struct xfs_disk_dquot *ddq, *recddq;
2889 xfs_dq_logformat_t *dq_f;
2894 * Filesystems are required to send in quota flags at mount time.
2896 if (mp->m_qflags == 0)
2899 recddq = item->ri_buf[1].i_addr;
2900 if (recddq == NULL) {
2901 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
2904 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
2905 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
2906 item->ri_buf[1].i_len, __func__);
2911 * This type of quotas was turned off, so ignore this record.
2913 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
2915 if (log->l_quotaoffs_flag & type)
2919 * At this point we know that quota was _not_ turned off.
2920 * Since the mount flags are not indicating to us otherwise, this
2921 * must mean that quota is on, and the dquot needs to be replayed.
2922 * Remember that we may not have fully recovered the superblock yet,
2923 * so we can't do the usual trick of looking at the SB quota bits.
2925 * The other possibility, of course, is that the quota subsystem was
2926 * removed since the last mount - ENOSYS.
2928 dq_f = item->ri_buf[0].i_addr;
2930 error = xfs_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2931 "xlog_recover_dquot_pass2 (log copy)");
2934 ASSERT(dq_f->qlf_len == 1);
2937 * At this point we are assuming that the dquots have been allocated
2938 * and hence the buffer has valid dquots stamped in it. It should,
2939 * therefore, pass verifier validation. If the dquot is bad, then the
2940 * we'll return an error here, so we don't need to specifically check
2941 * the dquot in the buffer after the verifier has run.
2943 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
2944 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
2945 &xfs_dquot_buf_ops);
2950 ddq = xfs_buf_offset(bp, dq_f->qlf_boffset);
2953 * If the dquot has an LSN in it, recover the dquot only if it's less
2954 * than the lsn of the transaction we are replaying.
2956 if (xfs_sb_version_hascrc(&mp->m_sb)) {
2957 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq;
2958 xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn);
2960 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2965 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2966 if (xfs_sb_version_hascrc(&mp->m_sb)) {
2967 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
2971 ASSERT(dq_f->qlf_size == 2);
2972 ASSERT(bp->b_target->bt_mount == mp);
2973 bp->b_iodone = xlog_recover_iodone;
2974 xfs_buf_delwri_queue(bp, buffer_list);
2982 * This routine is called to create an in-core extent free intent
2983 * item from the efi format structure which was logged on disk.
2984 * It allocates an in-core efi, copies the extents from the format
2985 * structure into it, and adds the efi to the AIL with the given
2989 xlog_recover_efi_pass2(
2991 struct xlog_recover_item *item,
2995 struct xfs_mount *mp = log->l_mp;
2996 struct xfs_efi_log_item *efip;
2997 struct xfs_efi_log_format *efi_formatp;
2999 efi_formatp = item->ri_buf[0].i_addr;
3001 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
3002 error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format);
3004 xfs_efi_item_free(efip);
3007 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
3009 spin_lock(&log->l_ailp->xa_lock);
3011 * The EFI has two references. One for the EFD and one for EFI to ensure
3012 * it makes it into the AIL. Insert the EFI into the AIL directly and
3013 * drop the EFI reference. Note that xfs_trans_ail_update() drops the
3016 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
3017 xfs_efi_release(efip);
3023 * This routine is called when an EFD format structure is found in a committed
3024 * transaction in the log. Its purpose is to cancel the corresponding EFI if it
3025 * was still in the log. To do this it searches the AIL for the EFI with an id
3026 * equal to that in the EFD format structure. If we find it we drop the EFD
3027 * reference, which removes the EFI from the AIL and frees it.
3030 xlog_recover_efd_pass2(
3032 struct xlog_recover_item *item)
3034 xfs_efd_log_format_t *efd_formatp;
3035 xfs_efi_log_item_t *efip = NULL;
3036 xfs_log_item_t *lip;
3038 struct xfs_ail_cursor cur;
3039 struct xfs_ail *ailp = log->l_ailp;
3041 efd_formatp = item->ri_buf[0].i_addr;
3042 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
3043 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
3044 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
3045 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
3046 efi_id = efd_formatp->efd_efi_id;
3049 * Search for the EFI with the id in the EFD format structure in the
3052 spin_lock(&ailp->xa_lock);
3053 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3054 while (lip != NULL) {
3055 if (lip->li_type == XFS_LI_EFI) {
3056 efip = (xfs_efi_log_item_t *)lip;
3057 if (efip->efi_format.efi_id == efi_id) {
3059 * Drop the EFD reference to the EFI. This
3060 * removes the EFI from the AIL and frees it.
3062 spin_unlock(&ailp->xa_lock);
3063 xfs_efi_release(efip);
3064 spin_lock(&ailp->xa_lock);
3068 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3071 xfs_trans_ail_cursor_done(&cur);
3072 spin_unlock(&ailp->xa_lock);
3078 * This routine is called when an inode create format structure is found in a
3079 * committed transaction in the log. It's purpose is to initialise the inodes
3080 * being allocated on disk. This requires us to get inode cluster buffers that
3081 * match the range to be intialised, stamped with inode templates and written
3082 * by delayed write so that subsequent modifications will hit the cached buffer
3083 * and only need writing out at the end of recovery.
3086 xlog_recover_do_icreate_pass2(
3088 struct list_head *buffer_list,
3089 xlog_recover_item_t *item)
3091 struct xfs_mount *mp = log->l_mp;
3092 struct xfs_icreate_log *icl;
3093 xfs_agnumber_t agno;
3094 xfs_agblock_t agbno;
3097 xfs_agblock_t length;
3098 int blks_per_cluster;
3104 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3105 if (icl->icl_type != XFS_LI_ICREATE) {
3106 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
3110 if (icl->icl_size != 1) {
3111 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
3115 agno = be32_to_cpu(icl->icl_ag);
3116 if (agno >= mp->m_sb.sb_agcount) {
3117 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
3120 agbno = be32_to_cpu(icl->icl_agbno);
3121 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3122 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
3125 isize = be32_to_cpu(icl->icl_isize);
3126 if (isize != mp->m_sb.sb_inodesize) {
3127 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
3130 count = be32_to_cpu(icl->icl_count);
3132 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
3135 length = be32_to_cpu(icl->icl_length);
3136 if (!length || length >= mp->m_sb.sb_agblocks) {
3137 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
3142 * The inode chunk is either full or sparse and we only support
3143 * m_ialloc_min_blks sized sparse allocations at this time.
3145 if (length != mp->m_ialloc_blks &&
3146 length != mp->m_ialloc_min_blks) {
3148 "%s: unsupported chunk length", __FUNCTION__);
3152 /* verify inode count is consistent with extent length */
3153 if ((count >> mp->m_sb.sb_inopblog) != length) {
3155 "%s: inconsistent inode count and chunk length",
3161 * The icreate transaction can cover multiple cluster buffers and these
3162 * buffers could have been freed and reused. Check the individual
3163 * buffers for cancellation so we don't overwrite anything written after
3166 blks_per_cluster = xfs_icluster_size_fsb(mp);
3167 bb_per_cluster = XFS_FSB_TO_BB(mp, blks_per_cluster);
3168 nbufs = length / blks_per_cluster;
3169 for (i = 0, cancel_count = 0; i < nbufs; i++) {
3172 daddr = XFS_AGB_TO_DADDR(mp, agno,
3173 agbno + i * blks_per_cluster);
3174 if (xlog_check_buffer_cancelled(log, daddr, bb_per_cluster, 0))
3179 * We currently only use icreate for a single allocation at a time. This
3180 * means we should expect either all or none of the buffers to be
3181 * cancelled. Be conservative and skip replay if at least one buffer is
3182 * cancelled, but warn the user that something is awry if the buffers
3183 * are not consistent.
3185 * XXX: This must be refined to only skip cancelled clusters once we use
3186 * icreate for multiple chunk allocations.
3188 ASSERT(!cancel_count || cancel_count == nbufs);
3190 if (cancel_count != nbufs)
3192 "WARNING: partial inode chunk cancellation, skipped icreate.");
3193 trace_xfs_log_recover_icreate_cancel(log, icl);
3197 trace_xfs_log_recover_icreate_recover(log, icl);
3198 return xfs_ialloc_inode_init(mp, NULL, buffer_list, count, agno, agbno,
3199 length, be32_to_cpu(icl->icl_gen));
3203 xlog_recover_buffer_ra_pass2(
3205 struct xlog_recover_item *item)
3207 struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr;
3208 struct xfs_mount *mp = log->l_mp;
3210 if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
3211 buf_f->blf_len, buf_f->blf_flags)) {
3215 xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
3216 buf_f->blf_len, NULL);
3220 xlog_recover_inode_ra_pass2(
3222 struct xlog_recover_item *item)
3224 struct xfs_inode_log_format ilf_buf;
3225 struct xfs_inode_log_format *ilfp;
3226 struct xfs_mount *mp = log->l_mp;
3229 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3230 ilfp = item->ri_buf[0].i_addr;
3233 memset(ilfp, 0, sizeof(*ilfp));
3234 error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
3239 if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
3242 xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
3243 ilfp->ilf_len, &xfs_inode_buf_ra_ops);
3247 xlog_recover_dquot_ra_pass2(
3249 struct xlog_recover_item *item)
3251 struct xfs_mount *mp = log->l_mp;
3252 struct xfs_disk_dquot *recddq;
3253 struct xfs_dq_logformat *dq_f;
3257 if (mp->m_qflags == 0)
3260 recddq = item->ri_buf[1].i_addr;
3263 if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
3266 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3268 if (log->l_quotaoffs_flag & type)
3271 dq_f = item->ri_buf[0].i_addr;
3273 ASSERT(dq_f->qlf_len == 1);
3275 xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno,
3276 XFS_FSB_TO_BB(mp, dq_f->qlf_len), NULL);
3280 xlog_recover_ra_pass2(
3282 struct xlog_recover_item *item)
3284 switch (ITEM_TYPE(item)) {
3286 xlog_recover_buffer_ra_pass2(log, item);
3289 xlog_recover_inode_ra_pass2(log, item);
3292 xlog_recover_dquot_ra_pass2(log, item);
3296 case XFS_LI_QUOTAOFF:
3303 xlog_recover_commit_pass1(
3305 struct xlog_recover *trans,
3306 struct xlog_recover_item *item)
3308 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
3310 switch (ITEM_TYPE(item)) {
3312 return xlog_recover_buffer_pass1(log, item);
3313 case XFS_LI_QUOTAOFF:
3314 return xlog_recover_quotaoff_pass1(log, item);
3319 case XFS_LI_ICREATE:
3320 /* nothing to do in pass 1 */
3323 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3324 __func__, ITEM_TYPE(item));
3331 xlog_recover_commit_pass2(
3333 struct xlog_recover *trans,
3334 struct list_head *buffer_list,
3335 struct xlog_recover_item *item)
3337 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
3339 switch (ITEM_TYPE(item)) {
3341 return xlog_recover_buffer_pass2(log, buffer_list, item,
3344 return xlog_recover_inode_pass2(log, buffer_list, item,
3347 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
3349 return xlog_recover_efd_pass2(log, item);
3351 return xlog_recover_dquot_pass2(log, buffer_list, item,
3353 case XFS_LI_ICREATE:
3354 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
3355 case XFS_LI_QUOTAOFF:
3356 /* nothing to do in pass2 */
3359 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3360 __func__, ITEM_TYPE(item));
3367 xlog_recover_items_pass2(
3369 struct xlog_recover *trans,
3370 struct list_head *buffer_list,
3371 struct list_head *item_list)
3373 struct xlog_recover_item *item;
3376 list_for_each_entry(item, item_list, ri_list) {
3377 error = xlog_recover_commit_pass2(log, trans,
3387 * Perform the transaction.
3389 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3390 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3393 xlog_recover_commit_trans(
3395 struct xlog_recover *trans,
3400 int items_queued = 0;
3401 struct xlog_recover_item *item;
3402 struct xlog_recover_item *next;
3403 LIST_HEAD (buffer_list);
3404 LIST_HEAD (ra_list);
3405 LIST_HEAD (done_list);
3407 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
3409 hlist_del(&trans->r_list);
3411 error = xlog_recover_reorder_trans(log, trans, pass);
3415 list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
3417 case XLOG_RECOVER_PASS1:
3418 error = xlog_recover_commit_pass1(log, trans, item);
3420 case XLOG_RECOVER_PASS2:
3421 xlog_recover_ra_pass2(log, item);
3422 list_move_tail(&item->ri_list, &ra_list);
3424 if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
3425 error = xlog_recover_items_pass2(log, trans,
3426 &buffer_list, &ra_list);
3427 list_splice_tail_init(&ra_list, &done_list);
3441 if (!list_empty(&ra_list)) {
3443 error = xlog_recover_items_pass2(log, trans,
3444 &buffer_list, &ra_list);
3445 list_splice_tail_init(&ra_list, &done_list);
3448 if (!list_empty(&done_list))
3449 list_splice_init(&done_list, &trans->r_itemq);
3451 error2 = xfs_buf_delwri_submit(&buffer_list);
3452 return error ? error : error2;
3456 xlog_recover_add_item(
3457 struct list_head *head)
3459 xlog_recover_item_t *item;
3461 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
3462 INIT_LIST_HEAD(&item->ri_list);
3463 list_add_tail(&item->ri_list, head);
3467 xlog_recover_add_to_cont_trans(
3469 struct xlog_recover *trans,
3473 xlog_recover_item_t *item;
3474 char *ptr, *old_ptr;
3478 * If the transaction is empty, the header was split across this and the
3479 * previous record. Copy the rest of the header.
3481 if (list_empty(&trans->r_itemq)) {
3482 ASSERT(len <= sizeof(struct xfs_trans_header));
3483 if (len > sizeof(struct xfs_trans_header)) {
3484 xfs_warn(log->l_mp, "%s: bad header length", __func__);
3488 xlog_recover_add_item(&trans->r_itemq);
3489 ptr = (char *)&trans->r_theader +
3490 sizeof(struct xfs_trans_header) - len;
3491 memcpy(ptr, dp, len);
3495 /* take the tail entry */
3496 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
3498 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
3499 old_len = item->ri_buf[item->ri_cnt-1].i_len;
3501 ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP);
3502 memcpy(&ptr[old_len], dp, len);
3503 item->ri_buf[item->ri_cnt-1].i_len += len;
3504 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
3505 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
3510 * The next region to add is the start of a new region. It could be
3511 * a whole region or it could be the first part of a new region. Because
3512 * of this, the assumption here is that the type and size fields of all
3513 * format structures fit into the first 32 bits of the structure.
3515 * This works because all regions must be 32 bit aligned. Therefore, we
3516 * either have both fields or we have neither field. In the case we have
3517 * neither field, the data part of the region is zero length. We only have
3518 * a log_op_header and can throw away the header since a new one will appear
3519 * later. If we have at least 4 bytes, then we can determine how many regions
3520 * will appear in the current log item.
3523 xlog_recover_add_to_trans(
3525 struct xlog_recover *trans,
3529 xfs_inode_log_format_t *in_f; /* any will do */
3530 xlog_recover_item_t *item;
3535 if (list_empty(&trans->r_itemq)) {
3536 /* we need to catch log corruptions here */
3537 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
3538 xfs_warn(log->l_mp, "%s: bad header magic number",
3544 if (len > sizeof(struct xfs_trans_header)) {
3545 xfs_warn(log->l_mp, "%s: bad header length", __func__);
3551 * The transaction header can be arbitrarily split across op
3552 * records. If we don't have the whole thing here, copy what we
3553 * do have and handle the rest in the next record.
3555 if (len == sizeof(struct xfs_trans_header))
3556 xlog_recover_add_item(&trans->r_itemq);
3557 memcpy(&trans->r_theader, dp, len);
3561 ptr = kmem_alloc(len, KM_SLEEP);
3562 memcpy(ptr, dp, len);
3563 in_f = (xfs_inode_log_format_t *)ptr;
3565 /* take the tail entry */
3566 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
3567 if (item->ri_total != 0 &&
3568 item->ri_total == item->ri_cnt) {
3569 /* tail item is in use, get a new one */
3570 xlog_recover_add_item(&trans->r_itemq);
3571 item = list_entry(trans->r_itemq.prev,
3572 xlog_recover_item_t, ri_list);
3575 if (item->ri_total == 0) { /* first region to be added */
3576 if (in_f->ilf_size == 0 ||
3577 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
3579 "bad number of regions (%d) in inode log format",
3586 item->ri_total = in_f->ilf_size;
3588 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
3591 ASSERT(item->ri_total > item->ri_cnt);
3592 /* Description region is ri_buf[0] */
3593 item->ri_buf[item->ri_cnt].i_addr = ptr;
3594 item->ri_buf[item->ri_cnt].i_len = len;
3596 trace_xfs_log_recover_item_add(log, trans, item, 0);
3601 * Free up any resources allocated by the transaction
3603 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3606 xlog_recover_free_trans(
3607 struct xlog_recover *trans)
3609 xlog_recover_item_t *item, *n;
3612 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
3613 /* Free the regions in the item. */
3614 list_del(&item->ri_list);
3615 for (i = 0; i < item->ri_cnt; i++)
3616 kmem_free(item->ri_buf[i].i_addr);
3617 /* Free the item itself */
3618 kmem_free(item->ri_buf);
3621 /* Free the transaction recover structure */
3626 * On error or completion, trans is freed.
3629 xlog_recovery_process_trans(
3631 struct xlog_recover *trans,
3638 bool freeit = false;
3640 /* mask off ophdr transaction container flags */
3641 flags &= ~XLOG_END_TRANS;
3642 if (flags & XLOG_WAS_CONT_TRANS)
3643 flags &= ~XLOG_CONTINUE_TRANS;
3646 * Callees must not free the trans structure. We'll decide if we need to
3647 * free it or not based on the operation being done and it's result.
3650 /* expected flag values */
3652 case XLOG_CONTINUE_TRANS:
3653 error = xlog_recover_add_to_trans(log, trans, dp, len);
3655 case XLOG_WAS_CONT_TRANS:
3656 error = xlog_recover_add_to_cont_trans(log, trans, dp, len);
3658 case XLOG_COMMIT_TRANS:
3659 error = xlog_recover_commit_trans(log, trans, pass);
3660 /* success or fail, we are now done with this transaction. */
3664 /* unexpected flag values */
3665 case XLOG_UNMOUNT_TRANS:
3666 /* just skip trans */
3667 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
3670 case XLOG_START_TRANS:
3672 xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags);
3677 if (error || freeit)
3678 xlog_recover_free_trans(trans);
3683 * Lookup the transaction recovery structure associated with the ID in the
3684 * current ophdr. If the transaction doesn't exist and the start flag is set in
3685 * the ophdr, then allocate a new transaction for future ID matches to find.
3686 * Either way, return what we found during the lookup - an existing transaction
3689 STATIC struct xlog_recover *
3690 xlog_recover_ophdr_to_trans(
3691 struct hlist_head rhash[],
3692 struct xlog_rec_header *rhead,
3693 struct xlog_op_header *ohead)
3695 struct xlog_recover *trans;
3697 struct hlist_head *rhp;
3699 tid = be32_to_cpu(ohead->oh_tid);
3700 rhp = &rhash[XLOG_RHASH(tid)];
3701 hlist_for_each_entry(trans, rhp, r_list) {
3702 if (trans->r_log_tid == tid)
3707 * skip over non-start transaction headers - we could be
3708 * processing slack space before the next transaction starts
3710 if (!(ohead->oh_flags & XLOG_START_TRANS))
3713 ASSERT(be32_to_cpu(ohead->oh_len) == 0);
3716 * This is a new transaction so allocate a new recovery container to
3717 * hold the recovery ops that will follow.
3719 trans = kmem_zalloc(sizeof(struct xlog_recover), KM_SLEEP);
3720 trans->r_log_tid = tid;
3721 trans->r_lsn = be64_to_cpu(rhead->h_lsn);
3722 INIT_LIST_HEAD(&trans->r_itemq);
3723 INIT_HLIST_NODE(&trans->r_list);
3724 hlist_add_head(&trans->r_list, rhp);
3727 * Nothing more to do for this ophdr. Items to be added to this new
3728 * transaction will be in subsequent ophdr containers.
3734 xlog_recover_process_ophdr(
3736 struct hlist_head rhash[],
3737 struct xlog_rec_header *rhead,
3738 struct xlog_op_header *ohead,
3743 struct xlog_recover *trans;
3746 /* Do we understand who wrote this op? */
3747 if (ohead->oh_clientid != XFS_TRANSACTION &&
3748 ohead->oh_clientid != XFS_LOG) {
3749 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
3750 __func__, ohead->oh_clientid);
3756 * Check the ophdr contains all the data it is supposed to contain.
3758 len = be32_to_cpu(ohead->oh_len);
3759 if (dp + len > end) {
3760 xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len);
3765 trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead);
3767 /* nothing to do, so skip over this ophdr */
3771 return xlog_recovery_process_trans(log, trans, dp, len,
3772 ohead->oh_flags, pass);
3776 * There are two valid states of the r_state field. 0 indicates that the
3777 * transaction structure is in a normal state. We have either seen the
3778 * start of the transaction or the last operation we added was not a partial
3779 * operation. If the last operation we added to the transaction was a
3780 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
3782 * NOTE: skip LRs with 0 data length.
3785 xlog_recover_process_data(
3787 struct hlist_head rhash[],
3788 struct xlog_rec_header *rhead,
3792 struct xlog_op_header *ohead;
3797 end = dp + be32_to_cpu(rhead->h_len);
3798 num_logops = be32_to_cpu(rhead->h_num_logops);
3800 /* check the log format matches our own - else we can't recover */
3801 if (xlog_header_check_recover(log->l_mp, rhead))
3804 while ((dp < end) && num_logops) {
3806 ohead = (struct xlog_op_header *)dp;
3807 dp += sizeof(*ohead);
3810 /* errors will abort recovery */
3811 error = xlog_recover_process_ophdr(log, rhash, rhead, ohead,
3816 dp += be32_to_cpu(ohead->oh_len);
3823 * Process an extent free intent item that was recovered from
3824 * the log. We need to free the extents that it describes.
3827 xlog_recover_process_efi(
3829 xfs_efi_log_item_t *efip)
3831 xfs_efd_log_item_t *efdp;
3836 xfs_fsblock_t startblock_fsb;
3838 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
3841 * First check the validity of the extents described by the
3842 * EFI. If any are bad, then assume that all are bad and
3843 * just toss the EFI.
3845 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3846 extp = &(efip->efi_format.efi_extents[i]);
3847 startblock_fsb = XFS_BB_TO_FSB(mp,
3848 XFS_FSB_TO_DADDR(mp, extp->ext_start));
3849 if ((startblock_fsb == 0) ||
3850 (extp->ext_len == 0) ||
3851 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3852 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3854 * This will pull the EFI from the AIL and
3855 * free the memory associated with it.
3857 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
3858 xfs_efi_release(efip);
3863 tp = xfs_trans_alloc(mp, 0);
3864 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
3867 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3869 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3870 extp = &(efip->efi_format.efi_extents[i]);
3871 error = xfs_trans_free_extent(tp, efdp, extp->ext_start,
3878 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
3879 error = xfs_trans_commit(tp);
3883 xfs_trans_cancel(tp);
3888 * When this is called, all of the EFIs which did not have
3889 * corresponding EFDs should be in the AIL. What we do now
3890 * is free the extents associated with each one.
3892 * Since we process the EFIs in normal transactions, they
3893 * will be removed at some point after the commit. This prevents
3894 * us from just walking down the list processing each one.
3895 * We'll use a flag in the EFI to skip those that we've already
3896 * processed and use the AIL iteration mechanism's generation
3897 * count to try to speed this up at least a bit.
3899 * When we start, we know that the EFIs are the only things in
3900 * the AIL. As we process them, however, other items are added
3901 * to the AIL. Since everything added to the AIL must come after
3902 * everything already in the AIL, we stop processing as soon as
3903 * we see something other than an EFI in the AIL.
3906 xlog_recover_process_efis(
3909 struct xfs_log_item *lip;
3910 struct xfs_efi_log_item *efip;
3912 struct xfs_ail_cursor cur;
3913 struct xfs_ail *ailp;
3916 spin_lock(&ailp->xa_lock);
3917 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3918 while (lip != NULL) {
3920 * We're done when we see something other than an EFI.
3921 * There should be no EFIs left in the AIL now.
3923 if (lip->li_type != XFS_LI_EFI) {
3925 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
3926 ASSERT(lip->li_type != XFS_LI_EFI);
3932 * Skip EFIs that we've already processed.
3934 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
3935 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
3936 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3940 spin_unlock(&ailp->xa_lock);
3941 error = xlog_recover_process_efi(log->l_mp, efip);
3942 spin_lock(&ailp->xa_lock);
3945 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3948 xfs_trans_ail_cursor_done(&cur);
3949 spin_unlock(&ailp->xa_lock);
3954 * A cancel occurs when the mount has failed and we're bailing out. Release all
3955 * pending EFIs so they don't pin the AIL.
3958 xlog_recover_cancel_efis(
3961 struct xfs_log_item *lip;
3962 struct xfs_efi_log_item *efip;
3964 struct xfs_ail_cursor cur;
3965 struct xfs_ail *ailp;
3968 spin_lock(&ailp->xa_lock);
3969 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3970 while (lip != NULL) {
3972 * We're done when we see something other than an EFI.
3973 * There should be no EFIs left in the AIL now.
3975 if (lip->li_type != XFS_LI_EFI) {
3977 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
3978 ASSERT(lip->li_type != XFS_LI_EFI);
3983 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
3985 spin_unlock(&ailp->xa_lock);
3986 xfs_efi_release(efip);
3987 spin_lock(&ailp->xa_lock);
3989 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3992 xfs_trans_ail_cursor_done(&cur);
3993 spin_unlock(&ailp->xa_lock);
3998 * This routine performs a transaction to null out a bad inode pointer
3999 * in an agi unlinked inode hash bucket.
4002 xlog_recover_clear_agi_bucket(
4004 xfs_agnumber_t agno,
4013 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
4014 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_clearagi, 0, 0);
4018 error = xfs_read_agi(mp, tp, agno, &agibp);
4022 agi = XFS_BUF_TO_AGI(agibp);
4023 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
4024 offset = offsetof(xfs_agi_t, agi_unlinked) +
4025 (sizeof(xfs_agino_t) * bucket);
4026 xfs_trans_log_buf(tp, agibp, offset,
4027 (offset + sizeof(xfs_agino_t) - 1));
4029 error = xfs_trans_commit(tp);
4035 xfs_trans_cancel(tp);
4037 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
4042 xlog_recover_process_one_iunlink(
4043 struct xfs_mount *mp,
4044 xfs_agnumber_t agno,
4048 struct xfs_buf *ibp;
4049 struct xfs_dinode *dip;
4050 struct xfs_inode *ip;
4054 ino = XFS_AGINO_TO_INO(mp, agno, agino);
4055 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
4060 * Get the on disk inode to find the next inode in the bucket.
4062 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
4066 ASSERT(ip->i_d.di_nlink == 0);
4067 ASSERT(ip->i_d.di_mode != 0);
4069 /* setup for the next pass */
4070 agino = be32_to_cpu(dip->di_next_unlinked);
4074 * Prevent any DMAPI event from being sent when the reference on
4075 * the inode is dropped.
4077 ip->i_d.di_dmevmask = 0;
4086 * We can't read in the inode this bucket points to, or this inode
4087 * is messed up. Just ditch this bucket of inodes. We will lose
4088 * some inodes and space, but at least we won't hang.
4090 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
4091 * clear the inode pointer in the bucket.
4093 xlog_recover_clear_agi_bucket(mp, agno, bucket);
4098 * xlog_iunlink_recover
4100 * This is called during recovery to process any inodes which
4101 * we unlinked but not freed when the system crashed. These
4102 * inodes will be on the lists in the AGI blocks. What we do
4103 * here is scan all the AGIs and fully truncate and free any
4104 * inodes found on the lists. Each inode is removed from the
4105 * lists when it has been fully truncated and is freed. The
4106 * freeing of the inode and its removal from the list must be
4110 xlog_recover_process_iunlinks(
4114 xfs_agnumber_t agno;
4125 * Prevent any DMAPI event from being sent while in this function.
4127 mp_dmevmask = mp->m_dmevmask;
4130 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
4132 * Find the agi for this ag.
4134 error = xfs_read_agi(mp, NULL, agno, &agibp);
4137 * AGI is b0rked. Don't process it.
4139 * We should probably mark the filesystem as corrupt
4140 * after we've recovered all the ag's we can....
4145 * Unlock the buffer so that it can be acquired in the normal
4146 * course of the transaction to truncate and free each inode.
4147 * Because we are not racing with anyone else here for the AGI
4148 * buffer, we don't even need to hold it locked to read the
4149 * initial unlinked bucket entries out of the buffer. We keep
4150 * buffer reference though, so that it stays pinned in memory
4151 * while we need the buffer.
4153 agi = XFS_BUF_TO_AGI(agibp);
4154 xfs_buf_unlock(agibp);
4156 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
4157 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
4158 while (agino != NULLAGINO) {
4159 agino = xlog_recover_process_one_iunlink(mp,
4160 agno, agino, bucket);
4163 xfs_buf_rele(agibp);
4166 mp->m_dmevmask = mp_dmevmask;
4171 struct xlog_rec_header *rhead,
4177 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
4178 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
4179 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
4183 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
4184 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
4185 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
4186 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
4187 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
4188 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
4197 * CRC check, unpack and process a log record.
4200 xlog_recover_process(
4202 struct hlist_head rhash[],
4203 struct xlog_rec_header *rhead,
4210 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
4213 * Nothing else to do if this is a CRC verification pass. Just return
4214 * if this a record with a non-zero crc. Unfortunately, mkfs always
4215 * sets h_crc to 0 so we must consider this valid even on v5 supers.
4216 * Otherwise, return EFSBADCRC on failure so the callers up the stack
4217 * know precisely what failed.
4219 if (pass == XLOG_RECOVER_CRCPASS) {
4220 if (rhead->h_crc && crc != le32_to_cpu(rhead->h_crc))
4226 * We're in the normal recovery path. Issue a warning if and only if the
4227 * CRC in the header is non-zero. This is an advisory warning and the
4228 * zero CRC check prevents warnings from being emitted when upgrading
4229 * the kernel from one that does not add CRCs by default.
4231 if (crc != le32_to_cpu(rhead->h_crc)) {
4232 if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
4233 xfs_alert(log->l_mp,
4234 "log record CRC mismatch: found 0x%x, expected 0x%x.",
4235 le32_to_cpu(rhead->h_crc),
4237 xfs_hex_dump(dp, 32);
4241 * If the filesystem is CRC enabled, this mismatch becomes a
4242 * fatal log corruption failure.
4244 if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
4245 return -EFSCORRUPTED;
4248 error = xlog_unpack_data(rhead, dp, log);
4252 return xlog_recover_process_data(log, rhash, rhead, dp, pass);
4256 xlog_valid_rec_header(
4258 struct xlog_rec_header *rhead,
4263 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
4264 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
4265 XFS_ERRLEVEL_LOW, log->l_mp);
4266 return -EFSCORRUPTED;
4269 (!rhead->h_version ||
4270 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
4271 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
4272 __func__, be32_to_cpu(rhead->h_version));
4276 /* LR body must have data or it wouldn't have been written */
4277 hlen = be32_to_cpu(rhead->h_len);
4278 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
4279 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
4280 XFS_ERRLEVEL_LOW, log->l_mp);
4281 return -EFSCORRUPTED;
4283 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
4284 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
4285 XFS_ERRLEVEL_LOW, log->l_mp);
4286 return -EFSCORRUPTED;
4292 * Read the log from tail to head and process the log records found.
4293 * Handle the two cases where the tail and head are in the same cycle
4294 * and where the active portion of the log wraps around the end of
4295 * the physical log separately. The pass parameter is passed through
4296 * to the routines called to process the data and is not looked at
4300 xlog_do_recovery_pass(
4302 xfs_daddr_t head_blk,
4303 xfs_daddr_t tail_blk,
4305 xfs_daddr_t *first_bad) /* out: first bad log rec */
4307 xlog_rec_header_t *rhead;
4309 xfs_daddr_t rhead_blk;
4311 xfs_buf_t *hbp, *dbp;
4312 int error = 0, h_size, h_len;
4313 int bblks, split_bblks;
4314 int hblks, split_hblks, wrapped_hblks;
4315 struct hlist_head rhash[XLOG_RHASH_SIZE];
4317 ASSERT(head_blk != tail_blk);
4321 * Read the header of the tail block and get the iclog buffer size from
4322 * h_size. Use this to tell how many sectors make up the log header.
4324 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
4326 * When using variable length iclogs, read first sector of
4327 * iclog header and extract the header size from it. Get a
4328 * new hbp that is the correct size.
4330 hbp = xlog_get_bp(log, 1);
4334 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
4338 rhead = (xlog_rec_header_t *)offset;
4339 error = xlog_valid_rec_header(log, rhead, tail_blk);
4344 * xfsprogs has a bug where record length is based on lsunit but
4345 * h_size (iclog size) is hardcoded to 32k. Now that we
4346 * unconditionally CRC verify the unmount record, this means the
4347 * log buffer can be too small for the record and cause an
4350 * Detect this condition here. Use lsunit for the buffer size as
4351 * long as this looks like the mkfs case. Otherwise, return an
4352 * error to avoid a buffer overrun.
4354 h_size = be32_to_cpu(rhead->h_size);
4355 h_len = be32_to_cpu(rhead->h_len);
4356 if (h_len > h_size) {
4357 if (h_len <= log->l_mp->m_logbsize &&
4358 be32_to_cpu(rhead->h_num_logops) == 1) {
4360 "invalid iclog size (%d bytes), using lsunit (%d bytes)",
4361 h_size, log->l_mp->m_logbsize);
4362 h_size = log->l_mp->m_logbsize;
4364 return -EFSCORRUPTED;
4367 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
4368 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
4369 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
4370 if (h_size % XLOG_HEADER_CYCLE_SIZE)
4373 hbp = xlog_get_bp(log, hblks);
4378 ASSERT(log->l_sectBBsize == 1);
4380 hbp = xlog_get_bp(log, 1);
4381 h_size = XLOG_BIG_RECORD_BSIZE;
4386 dbp = xlog_get_bp(log, BTOBB(h_size));
4392 memset(rhash, 0, sizeof(rhash));
4393 blk_no = rhead_blk = tail_blk;
4394 if (tail_blk > head_blk) {
4396 * Perform recovery around the end of the physical log.
4397 * When the head is not on the same cycle number as the tail,
4398 * we can't do a sequential recovery.
4400 while (blk_no < log->l_logBBsize) {
4402 * Check for header wrapping around physical end-of-log
4404 offset = hbp->b_addr;
4407 if (blk_no + hblks <= log->l_logBBsize) {
4408 /* Read header in one read */
4409 error = xlog_bread(log, blk_no, hblks, hbp,
4414 /* This LR is split across physical log end */
4415 if (blk_no != log->l_logBBsize) {
4416 /* some data before physical log end */
4417 ASSERT(blk_no <= INT_MAX);
4418 split_hblks = log->l_logBBsize - (int)blk_no;
4419 ASSERT(split_hblks > 0);
4420 error = xlog_bread(log, blk_no,
4428 * Note: this black magic still works with
4429 * large sector sizes (non-512) only because:
4430 * - we increased the buffer size originally
4431 * by 1 sector giving us enough extra space
4432 * for the second read;
4433 * - the log start is guaranteed to be sector
4435 * - we read the log end (LR header start)
4436 * _first_, then the log start (LR header end)
4437 * - order is important.
4439 wrapped_hblks = hblks - split_hblks;
4440 error = xlog_bread_offset(log, 0,
4442 offset + BBTOB(split_hblks));
4446 rhead = (xlog_rec_header_t *)offset;
4447 error = xlog_valid_rec_header(log, rhead,
4448 split_hblks ? blk_no : 0);
4452 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4455 /* Read in data for log record */
4456 if (blk_no + bblks <= log->l_logBBsize) {
4457 error = xlog_bread(log, blk_no, bblks, dbp,
4462 /* This log record is split across the
4463 * physical end of log */
4464 offset = dbp->b_addr;
4466 if (blk_no != log->l_logBBsize) {
4467 /* some data is before the physical
4469 ASSERT(!wrapped_hblks);
4470 ASSERT(blk_no <= INT_MAX);
4472 log->l_logBBsize - (int)blk_no;
4473 ASSERT(split_bblks > 0);
4474 error = xlog_bread(log, blk_no,
4482 * Note: this black magic still works with
4483 * large sector sizes (non-512) only because:
4484 * - we increased the buffer size originally
4485 * by 1 sector giving us enough extra space
4486 * for the second read;
4487 * - the log start is guaranteed to be sector
4489 * - we read the log end (LR header start)
4490 * _first_, then the log start (LR header end)
4491 * - order is important.
4493 error = xlog_bread_offset(log, 0,
4494 bblks - split_bblks, dbp,
4495 offset + BBTOB(split_bblks));
4500 error = xlog_recover_process(log, rhash, rhead, offset,
4509 ASSERT(blk_no >= log->l_logBBsize);
4510 blk_no -= log->l_logBBsize;
4514 /* read first part of physical log */
4515 while (blk_no < head_blk) {
4516 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
4520 rhead = (xlog_rec_header_t *)offset;
4521 error = xlog_valid_rec_header(log, rhead, blk_no);
4525 /* blocks in data section */
4526 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4527 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
4532 error = xlog_recover_process(log, rhash, rhead, offset, pass);
4536 blk_no += bblks + hblks;
4545 if (error && first_bad)
4546 *first_bad = rhead_blk;
4552 * Do the recovery of the log. We actually do this in two phases.
4553 * The two passes are necessary in order to implement the function
4554 * of cancelling a record written into the log. The first pass
4555 * determines those things which have been cancelled, and the
4556 * second pass replays log items normally except for those which
4557 * have been cancelled. The handling of the replay and cancellations
4558 * takes place in the log item type specific routines.
4560 * The table of items which have cancel records in the log is allocated
4561 * and freed at this level, since only here do we know when all of
4562 * the log recovery has been completed.
4565 xlog_do_log_recovery(
4567 xfs_daddr_t head_blk,
4568 xfs_daddr_t tail_blk)
4572 ASSERT(head_blk != tail_blk);
4575 * First do a pass to find all of the cancelled buf log items.
4576 * Store them in the buf_cancel_table for use in the second pass.
4578 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
4579 sizeof(struct list_head),
4581 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4582 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
4584 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4585 XLOG_RECOVER_PASS1, NULL);
4587 kmem_free(log->l_buf_cancel_table);
4588 log->l_buf_cancel_table = NULL;
4592 * Then do a second pass to actually recover the items in the log.
4593 * When it is complete free the table of buf cancel items.
4595 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4596 XLOG_RECOVER_PASS2, NULL);
4601 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4602 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
4606 kmem_free(log->l_buf_cancel_table);
4607 log->l_buf_cancel_table = NULL;
4613 * Do the actual recovery
4618 xfs_daddr_t head_blk,
4619 xfs_daddr_t tail_blk)
4626 * First replay the images in the log.
4628 error = xlog_do_log_recovery(log, head_blk, tail_blk);
4633 * If IO errors happened during recovery, bail out.
4635 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
4640 * We now update the tail_lsn since much of the recovery has completed
4641 * and there may be space available to use. If there were no extent
4642 * or iunlinks, we can free up the entire log and set the tail_lsn to
4643 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4644 * lsn of the last known good LR on disk. If there are extent frees
4645 * or iunlinks they will have some entries in the AIL; so we look at
4646 * the AIL to determine how to set the tail_lsn.
4648 xlog_assign_tail_lsn(log->l_mp);
4651 * Now that we've finished replaying all buffer and inode
4652 * updates, re-read in the superblock and reverify it.
4654 bp = xfs_getsb(log->l_mp, 0);
4656 ASSERT(!(XFS_BUF_ISWRITE(bp)));
4658 XFS_BUF_UNASYNC(bp);
4659 bp->b_ops = &xfs_sb_buf_ops;
4661 error = xfs_buf_submit_wait(bp);
4663 if (!XFS_FORCED_SHUTDOWN(log->l_mp)) {
4664 xfs_buf_ioerror_alert(bp, __func__);
4671 /* Convert superblock from on-disk format */
4672 sbp = &log->l_mp->m_sb;
4673 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
4674 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
4675 ASSERT(xfs_sb_good_version(sbp));
4676 xfs_reinit_percpu_counters(log->l_mp);
4681 xlog_recover_check_summary(log);
4683 /* Normal transactions can now occur */
4684 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
4689 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4691 * Return error or zero.
4697 xfs_daddr_t head_blk, tail_blk;
4700 /* find the tail of the log */
4701 error = xlog_find_tail(log, &head_blk, &tail_blk);
4706 * The superblock was read before the log was available and thus the LSN
4707 * could not be verified. Check the superblock LSN against the current
4708 * LSN now that it's known.
4710 if (xfs_sb_version_hascrc(&log->l_mp->m_sb) &&
4711 !xfs_log_check_lsn(log->l_mp, log->l_mp->m_sb.sb_lsn))
4714 if (tail_blk != head_blk) {
4715 /* There used to be a comment here:
4717 * disallow recovery on read-only mounts. note -- mount
4718 * checks for ENOSPC and turns it into an intelligent
4720 * ...but this is no longer true. Now, unless you specify
4721 * NORECOVERY (in which case this function would never be
4722 * called), we just go ahead and recover. We do this all
4723 * under the vfs layer, so we can get away with it unless
4724 * the device itself is read-only, in which case we fail.
4726 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
4731 * Version 5 superblock log feature mask validation. We know the
4732 * log is dirty so check if there are any unknown log features
4733 * in what we need to recover. If there are unknown features
4734 * (e.g. unsupported transactions, then simply reject the
4735 * attempt at recovery before touching anything.
4737 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
4738 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
4739 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
4741 "Superblock has unknown incompatible log features (0x%x) enabled.",
4742 (log->l_mp->m_sb.sb_features_log_incompat &
4743 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
4745 "The log can not be fully and/or safely recovered by this kernel.");
4747 "Please recover the log on a kernel that supports the unknown features.");
4752 * Delay log recovery if the debug hook is set. This is debug
4753 * instrumention to coordinate simulation of I/O failures with
4756 if (xfs_globals.log_recovery_delay) {
4757 xfs_notice(log->l_mp,
4758 "Delaying log recovery for %d seconds.",
4759 xfs_globals.log_recovery_delay);
4760 msleep(xfs_globals.log_recovery_delay * 1000);
4763 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
4764 log->l_mp->m_logname ? log->l_mp->m_logname
4767 error = xlog_do_recover(log, head_blk, tail_blk);
4768 log->l_flags |= XLOG_RECOVERY_NEEDED;
4774 * In the first part of recovery we replay inodes and buffers and build
4775 * up the list of extent free items which need to be processed. Here
4776 * we process the extent free items and clean up the on disk unlinked
4777 * inode lists. This is separated from the first part of recovery so
4778 * that the root and real-time bitmap inodes can be read in from disk in
4779 * between the two stages. This is necessary so that we can free space
4780 * in the real-time portion of the file system.
4783 xlog_recover_finish(
4787 * Now we're ready to do the transactions needed for the
4788 * rest of recovery. Start with completing all the extent
4789 * free intent records and then process the unlinked inode
4790 * lists. At this point, we essentially run in normal mode
4791 * except that we're still performing recovery actions
4792 * rather than accepting new requests.
4794 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
4796 error = xlog_recover_process_efis(log);
4798 xfs_alert(log->l_mp, "Failed to recover EFIs");
4802 * Sync the log to get all the EFIs out of the AIL.
4803 * This isn't absolutely necessary, but it helps in
4804 * case the unlink transactions would have problems
4805 * pushing the EFIs out of the way.
4807 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
4809 xlog_recover_process_iunlinks(log);
4811 xlog_recover_check_summary(log);
4813 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
4814 log->l_mp->m_logname ? log->l_mp->m_logname
4816 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
4818 xfs_info(log->l_mp, "Ending clean mount");
4824 xlog_recover_cancel(
4829 if (log->l_flags & XLOG_RECOVERY_NEEDED)
4830 error = xlog_recover_cancel_efis(log);
4837 * Read all of the agf and agi counters and check that they
4838 * are consistent with the superblock counters.
4841 xlog_recover_check_summary(
4848 xfs_agnumber_t agno;
4849 __uint64_t freeblks;
4859 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
4860 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
4862 xfs_alert(mp, "%s agf read failed agno %d error %d",
4863 __func__, agno, error);
4865 agfp = XFS_BUF_TO_AGF(agfbp);
4866 freeblks += be32_to_cpu(agfp->agf_freeblks) +
4867 be32_to_cpu(agfp->agf_flcount);
4868 xfs_buf_relse(agfbp);
4871 error = xfs_read_agi(mp, NULL, agno, &agibp);
4873 xfs_alert(mp, "%s agi read failed agno %d error %d",
4874 __func__, agno, error);
4876 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
4878 itotal += be32_to_cpu(agi->agi_count);
4879 ifree += be32_to_cpu(agi->agi_freecount);
4880 xfs_buf_relse(agibp);