2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
27 #include "xfs_mount.h"
28 #include "xfs_error.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_alloc.h"
36 #include "xfs_ialloc.h"
37 #include "xfs_log_priv.h"
38 #include "xfs_buf_item.h"
39 #include "xfs_log_recover.h"
40 #include "xfs_extfree_item.h"
41 #include "xfs_trans_priv.h"
42 #include "xfs_quota.h"
44 #include "xfs_utils.h"
45 #include "xfs_trace.h"
47 STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *);
48 STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t);
50 STATIC void xlog_recover_check_summary(xlog_t *);
52 #define xlog_recover_check_summary(log)
56 * This structure is used during recovery to record the buf log items which
57 * have been canceled and should not be replayed.
59 struct xfs_buf_cancel {
63 struct list_head bc_list;
67 * Sector aligned buffer routines for buffer create/read/write/access
71 * Verify the given count of basic blocks is valid number of blocks
72 * to specify for an operation involving the given XFS log buffer.
73 * Returns nonzero if the count is valid, 0 otherwise.
77 xlog_buf_bbcount_valid(
81 return bbcount > 0 && bbcount <= log->l_logBBsize;
85 * Allocate a buffer to hold log data. The buffer needs to be able
86 * to map to a range of nbblks basic blocks at any valid (basic
87 * block) offset within the log.
94 if (!xlog_buf_bbcount_valid(log, nbblks)) {
95 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
97 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
102 * We do log I/O in units of log sectors (a power-of-2
103 * multiple of the basic block size), so we round up the
104 * requested size to accommodate the basic blocks required
105 * for complete log sectors.
107 * In addition, the buffer may be used for a non-sector-
108 * aligned block offset, in which case an I/O of the
109 * requested size could extend beyond the end of the
110 * buffer. If the requested size is only 1 basic block it
111 * will never straddle a sector boundary, so this won't be
112 * an issue. Nor will this be a problem if the log I/O is
113 * done in basic blocks (sector size 1). But otherwise we
114 * extend the buffer by one extra log sector to ensure
115 * there's space to accommodate this possibility.
117 if (nbblks > 1 && log->l_sectBBsize > 1)
118 nbblks += log->l_sectBBsize;
119 nbblks = round_up(nbblks, log->l_sectBBsize);
121 return xfs_buf_get_uncached(log->l_mp->m_logdev_targp,
133 * Return the address of the start of the given block number's data
134 * in a log buffer. The buffer covers a log sector-aligned region.
143 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
145 ASSERT(BBTOB(offset + nbblks) <= XFS_BUF_SIZE(bp));
146 return XFS_BUF_PTR(bp) + BBTOB(offset);
151 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
162 if (!xlog_buf_bbcount_valid(log, nbblks)) {
163 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
165 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
169 blk_no = round_down(blk_no, log->l_sectBBsize);
170 nbblks = round_up(nbblks, log->l_sectBBsize);
173 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
175 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
178 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
179 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
181 xfsbdstrat(log->l_mp, bp);
182 error = xfs_buf_iowait(bp);
184 xfs_ioerror_alert("xlog_bread", log->l_mp,
185 bp, XFS_BUF_ADDR(bp));
199 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
203 *offset = xlog_align(log, blk_no, nbblks, bp);
208 * Read at an offset into the buffer. Returns with the buffer in it's original
209 * state regardless of the result of the read.
214 xfs_daddr_t blk_no, /* block to read from */
215 int nbblks, /* blocks to read */
219 xfs_caddr_t orig_offset = XFS_BUF_PTR(bp);
220 int orig_len = bp->b_buffer_length;
223 error = XFS_BUF_SET_PTR(bp, offset, BBTOB(nbblks));
227 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
229 /* must reset buffer pointer even on error */
230 error2 = XFS_BUF_SET_PTR(bp, orig_offset, orig_len);
237 * Write out the buffer at the given block for the given number of blocks.
238 * The buffer is kept locked across the write and is returned locked.
239 * This can only be used for synchronous log writes.
250 if (!xlog_buf_bbcount_valid(log, nbblks)) {
251 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
253 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
257 blk_no = round_down(blk_no, log->l_sectBBsize);
258 nbblks = round_up(nbblks, log->l_sectBBsize);
261 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
263 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
264 XFS_BUF_ZEROFLAGS(bp);
267 XFS_BUF_PSEMA(bp, PRIBIO);
268 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
269 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
271 if ((error = xfs_bwrite(log->l_mp, bp)))
272 xfs_ioerror_alert("xlog_bwrite", log->l_mp,
273 bp, XFS_BUF_ADDR(bp));
279 * dump debug superblock and log record information
282 xlog_header_check_dump(
284 xlog_rec_header_t *head)
286 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d\n",
287 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
288 xfs_debug(mp, " log : uuid = %pU, fmt = %d\n",
289 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
292 #define xlog_header_check_dump(mp, head)
296 * check log record header for recovery
299 xlog_header_check_recover(
301 xlog_rec_header_t *head)
303 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
306 * IRIX doesn't write the h_fmt field and leaves it zeroed
307 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
308 * a dirty log created in IRIX.
310 if (unlikely(be32_to_cpu(head->h_fmt) != XLOG_FMT)) {
312 "dirty log written in incompatible format - can't recover");
313 xlog_header_check_dump(mp, head);
314 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
315 XFS_ERRLEVEL_HIGH, mp);
316 return XFS_ERROR(EFSCORRUPTED);
317 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
319 "dirty log entry has mismatched uuid - can't recover");
320 xlog_header_check_dump(mp, head);
321 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
322 XFS_ERRLEVEL_HIGH, mp);
323 return XFS_ERROR(EFSCORRUPTED);
329 * read the head block of the log and check the header
332 xlog_header_check_mount(
334 xlog_rec_header_t *head)
336 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
338 if (uuid_is_nil(&head->h_fs_uuid)) {
340 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
341 * h_fs_uuid is nil, we assume this log was last mounted
342 * by IRIX and continue.
344 xfs_warn(mp, "nil uuid in log - IRIX style log");
345 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
346 xfs_warn(mp, "log has mismatched uuid - can't recover");
347 xlog_header_check_dump(mp, head);
348 XFS_ERROR_REPORT("xlog_header_check_mount",
349 XFS_ERRLEVEL_HIGH, mp);
350 return XFS_ERROR(EFSCORRUPTED);
359 if (XFS_BUF_GETERROR(bp)) {
361 * We're not going to bother about retrying
362 * this during recovery. One strike!
364 xfs_ioerror_alert("xlog_recover_iodone",
365 bp->b_target->bt_mount, bp,
367 xfs_force_shutdown(bp->b_target->bt_mount,
368 SHUTDOWN_META_IO_ERROR);
370 XFS_BUF_CLR_IODONE_FUNC(bp);
371 xfs_buf_ioend(bp, 0);
375 * This routine finds (to an approximation) the first block in the physical
376 * log which contains the given cycle. It uses a binary search algorithm.
377 * Note that the algorithm can not be perfect because the disk will not
378 * necessarily be perfect.
381 xlog_find_cycle_start(
384 xfs_daddr_t first_blk,
385 xfs_daddr_t *last_blk,
395 mid_blk = BLK_AVG(first_blk, end_blk);
396 while (mid_blk != first_blk && mid_blk != end_blk) {
397 error = xlog_bread(log, mid_blk, 1, bp, &offset);
400 mid_cycle = xlog_get_cycle(offset);
401 if (mid_cycle == cycle)
402 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
404 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
405 mid_blk = BLK_AVG(first_blk, end_blk);
407 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
408 (mid_blk == end_blk && mid_blk-1 == first_blk));
416 * Check that a range of blocks does not contain stop_on_cycle_no.
417 * Fill in *new_blk with the block offset where such a block is
418 * found, or with -1 (an invalid block number) if there is no such
419 * block in the range. The scan needs to occur from front to back
420 * and the pointer into the region must be updated since a later
421 * routine will need to perform another test.
424 xlog_find_verify_cycle(
426 xfs_daddr_t start_blk,
428 uint stop_on_cycle_no,
429 xfs_daddr_t *new_blk)
435 xfs_caddr_t buf = NULL;
439 * Greedily allocate a buffer big enough to handle the full
440 * range of basic blocks we'll be examining. If that fails,
441 * try a smaller size. We need to be able to read at least
442 * a log sector, or we're out of luck.
444 bufblks = 1 << ffs(nbblks);
445 while (!(bp = xlog_get_bp(log, bufblks))) {
447 if (bufblks < log->l_sectBBsize)
451 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
454 bcount = min(bufblks, (start_blk + nbblks - i));
456 error = xlog_bread(log, i, bcount, bp, &buf);
460 for (j = 0; j < bcount; j++) {
461 cycle = xlog_get_cycle(buf);
462 if (cycle == stop_on_cycle_no) {
479 * Potentially backup over partial log record write.
481 * In the typical case, last_blk is the number of the block directly after
482 * a good log record. Therefore, we subtract one to get the block number
483 * of the last block in the given buffer. extra_bblks contains the number
484 * of blocks we would have read on a previous read. This happens when the
485 * last log record is split over the end of the physical log.
487 * extra_bblks is the number of blocks potentially verified on a previous
488 * call to this routine.
491 xlog_find_verify_log_record(
493 xfs_daddr_t start_blk,
494 xfs_daddr_t *last_blk,
499 xfs_caddr_t offset = NULL;
500 xlog_rec_header_t *head = NULL;
503 int num_blks = *last_blk - start_blk;
506 ASSERT(start_blk != 0 || *last_blk != start_blk);
508 if (!(bp = xlog_get_bp(log, num_blks))) {
509 if (!(bp = xlog_get_bp(log, 1)))
513 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
516 offset += ((num_blks - 1) << BBSHIFT);
519 for (i = (*last_blk) - 1; i >= 0; i--) {
521 /* valid log record not found */
523 "Log inconsistent (didn't find previous header)");
525 error = XFS_ERROR(EIO);
530 error = xlog_bread(log, i, 1, bp, &offset);
535 head = (xlog_rec_header_t *)offset;
537 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(head->h_magicno))
545 * We hit the beginning of the physical log & still no header. Return
546 * to caller. If caller can handle a return of -1, then this routine
547 * will be called again for the end of the physical log.
555 * We have the final block of the good log (the first block
556 * of the log record _before_ the head. So we check the uuid.
558 if ((error = xlog_header_check_mount(log->l_mp, head)))
562 * We may have found a log record header before we expected one.
563 * last_blk will be the 1st block # with a given cycle #. We may end
564 * up reading an entire log record. In this case, we don't want to
565 * reset last_blk. Only when last_blk points in the middle of a log
566 * record do we update last_blk.
568 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
569 uint h_size = be32_to_cpu(head->h_size);
571 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
572 if (h_size % XLOG_HEADER_CYCLE_SIZE)
578 if (*last_blk - i + extra_bblks !=
579 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
588 * Head is defined to be the point of the log where the next log write
589 * write could go. This means that incomplete LR writes at the end are
590 * eliminated when calculating the head. We aren't guaranteed that previous
591 * LR have complete transactions. We only know that a cycle number of
592 * current cycle number -1 won't be present in the log if we start writing
593 * from our current block number.
595 * last_blk contains the block number of the first block with a given
598 * Return: zero if normal, non-zero if error.
603 xfs_daddr_t *return_head_blk)
607 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
609 uint first_half_cycle, last_half_cycle;
611 int error, log_bbnum = log->l_logBBsize;
613 /* Is the end of the log device zeroed? */
614 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
615 *return_head_blk = first_blk;
617 /* Is the whole lot zeroed? */
619 /* Linux XFS shouldn't generate totally zeroed logs -
620 * mkfs etc write a dummy unmount record to a fresh
621 * log so we can store the uuid in there
623 xfs_warn(log->l_mp, "totally zeroed log");
628 xfs_warn(log->l_mp, "empty log check failed");
632 first_blk = 0; /* get cycle # of 1st block */
633 bp = xlog_get_bp(log, 1);
637 error = xlog_bread(log, 0, 1, bp, &offset);
641 first_half_cycle = xlog_get_cycle(offset);
643 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
644 error = xlog_bread(log, last_blk, 1, bp, &offset);
648 last_half_cycle = xlog_get_cycle(offset);
649 ASSERT(last_half_cycle != 0);
652 * If the 1st half cycle number is equal to the last half cycle number,
653 * then the entire log is stamped with the same cycle number. In this
654 * case, head_blk can't be set to zero (which makes sense). The below
655 * math doesn't work out properly with head_blk equal to zero. Instead,
656 * we set it to log_bbnum which is an invalid block number, but this
657 * value makes the math correct. If head_blk doesn't changed through
658 * all the tests below, *head_blk is set to zero at the very end rather
659 * than log_bbnum. In a sense, log_bbnum and zero are the same block
660 * in a circular file.
662 if (first_half_cycle == last_half_cycle) {
664 * In this case we believe that the entire log should have
665 * cycle number last_half_cycle. We need to scan backwards
666 * from the end verifying that there are no holes still
667 * containing last_half_cycle - 1. If we find such a hole,
668 * then the start of that hole will be the new head. The
669 * simple case looks like
670 * x | x ... | x - 1 | x
671 * Another case that fits this picture would be
672 * x | x + 1 | x ... | x
673 * In this case the head really is somewhere at the end of the
674 * log, as one of the latest writes at the beginning was
677 * x | x + 1 | x ... | x - 1 | x
678 * This is really the combination of the above two cases, and
679 * the head has to end up at the start of the x-1 hole at the
682 * In the 256k log case, we will read from the beginning to the
683 * end of the log and search for cycle numbers equal to x-1.
684 * We don't worry about the x+1 blocks that we encounter,
685 * because we know that they cannot be the head since the log
688 head_blk = log_bbnum;
689 stop_on_cycle = last_half_cycle - 1;
692 * In this case we want to find the first block with cycle
693 * number matching last_half_cycle. We expect the log to be
695 * x + 1 ... | x ... | x
696 * The first block with cycle number x (last_half_cycle) will
697 * be where the new head belongs. First we do a binary search
698 * for the first occurrence of last_half_cycle. The binary
699 * search may not be totally accurate, so then we scan back
700 * from there looking for occurrences of last_half_cycle before
701 * us. If that backwards scan wraps around the beginning of
702 * the log, then we look for occurrences of last_half_cycle - 1
703 * at the end of the log. The cases we're looking for look
705 * v binary search stopped here
706 * x + 1 ... | x | x + 1 | x ... | x
707 * ^ but we want to locate this spot
709 * <---------> less than scan distance
710 * x + 1 ... | x ... | x - 1 | x
711 * ^ we want to locate this spot
713 stop_on_cycle = last_half_cycle;
714 if ((error = xlog_find_cycle_start(log, bp, first_blk,
715 &head_blk, last_half_cycle)))
720 * Now validate the answer. Scan back some number of maximum possible
721 * blocks and make sure each one has the expected cycle number. The
722 * maximum is determined by the total possible amount of buffering
723 * in the in-core log. The following number can be made tighter if
724 * we actually look at the block size of the filesystem.
726 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
727 if (head_blk >= num_scan_bblks) {
729 * We are guaranteed that the entire check can be performed
732 start_blk = head_blk - num_scan_bblks;
733 if ((error = xlog_find_verify_cycle(log,
734 start_blk, num_scan_bblks,
735 stop_on_cycle, &new_blk)))
739 } else { /* need to read 2 parts of log */
741 * We are going to scan backwards in the log in two parts.
742 * First we scan the physical end of the log. In this part
743 * of the log, we are looking for blocks with cycle number
744 * last_half_cycle - 1.
745 * If we find one, then we know that the log starts there, as
746 * we've found a hole that didn't get written in going around
747 * the end of the physical log. The simple case for this is
748 * x + 1 ... | x ... | x - 1 | x
749 * <---------> less than scan distance
750 * If all of the blocks at the end of the log have cycle number
751 * last_half_cycle, then we check the blocks at the start of
752 * the log looking for occurrences of last_half_cycle. If we
753 * find one, then our current estimate for the location of the
754 * first occurrence of last_half_cycle is wrong and we move
755 * back to the hole we've found. This case looks like
756 * x + 1 ... | x | x + 1 | x ...
757 * ^ binary search stopped here
758 * Another case we need to handle that only occurs in 256k
760 * x + 1 ... | x ... | x+1 | x ...
761 * ^ binary search stops here
762 * In a 256k log, the scan at the end of the log will see the
763 * x + 1 blocks. We need to skip past those since that is
764 * certainly not the head of the log. By searching for
765 * last_half_cycle-1 we accomplish that.
767 ASSERT(head_blk <= INT_MAX &&
768 (xfs_daddr_t) num_scan_bblks >= head_blk);
769 start_blk = log_bbnum - (num_scan_bblks - head_blk);
770 if ((error = xlog_find_verify_cycle(log, start_blk,
771 num_scan_bblks - (int)head_blk,
772 (stop_on_cycle - 1), &new_blk)))
780 * Scan beginning of log now. The last part of the physical
781 * log is good. This scan needs to verify that it doesn't find
782 * the last_half_cycle.
785 ASSERT(head_blk <= INT_MAX);
786 if ((error = xlog_find_verify_cycle(log,
787 start_blk, (int)head_blk,
788 stop_on_cycle, &new_blk)))
796 * Now we need to make sure head_blk is not pointing to a block in
797 * the middle of a log record.
799 num_scan_bblks = XLOG_REC_SHIFT(log);
800 if (head_blk >= num_scan_bblks) {
801 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
803 /* start ptr at last block ptr before head_blk */
804 if ((error = xlog_find_verify_log_record(log, start_blk,
805 &head_blk, 0)) == -1) {
806 error = XFS_ERROR(EIO);
812 ASSERT(head_blk <= INT_MAX);
813 if ((error = xlog_find_verify_log_record(log, start_blk,
814 &head_blk, 0)) == -1) {
815 /* We hit the beginning of the log during our search */
816 start_blk = log_bbnum - (num_scan_bblks - head_blk);
818 ASSERT(start_blk <= INT_MAX &&
819 (xfs_daddr_t) log_bbnum-start_blk >= 0);
820 ASSERT(head_blk <= INT_MAX);
821 if ((error = xlog_find_verify_log_record(log,
823 (int)head_blk)) == -1) {
824 error = XFS_ERROR(EIO);
828 if (new_blk != log_bbnum)
835 if (head_blk == log_bbnum)
836 *return_head_blk = 0;
838 *return_head_blk = head_blk;
840 * When returning here, we have a good block number. Bad block
841 * means that during a previous crash, we didn't have a clean break
842 * from cycle number N to cycle number N-1. In this case, we need
843 * to find the first block with cycle number N-1.
851 xfs_warn(log->l_mp, "failed to find log head");
856 * Find the sync block number or the tail of the log.
858 * This will be the block number of the last record to have its
859 * associated buffers synced to disk. Every log record header has
860 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
861 * to get a sync block number. The only concern is to figure out which
862 * log record header to believe.
864 * The following algorithm uses the log record header with the largest
865 * lsn. The entire log record does not need to be valid. We only care
866 * that the header is valid.
868 * We could speed up search by using current head_blk buffer, but it is not
874 xfs_daddr_t *head_blk,
875 xfs_daddr_t *tail_blk)
877 xlog_rec_header_t *rhead;
878 xlog_op_header_t *op_head;
879 xfs_caddr_t offset = NULL;
882 xfs_daddr_t umount_data_blk;
883 xfs_daddr_t after_umount_blk;
890 * Find previous log record
892 if ((error = xlog_find_head(log, head_blk)))
895 bp = xlog_get_bp(log, 1);
898 if (*head_blk == 0) { /* special case */
899 error = xlog_bread(log, 0, 1, bp, &offset);
903 if (xlog_get_cycle(offset) == 0) {
905 /* leave all other log inited values alone */
911 * Search backwards looking for log record header block
913 ASSERT(*head_blk < INT_MAX);
914 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
915 error = xlog_bread(log, i, 1, bp, &offset);
919 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(*(__be32 *)offset)) {
925 * If we haven't found the log record header block, start looking
926 * again from the end of the physical log. XXXmiken: There should be
927 * a check here to make sure we didn't search more than N blocks in
931 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
932 error = xlog_bread(log, i, 1, bp, &offset);
936 if (XLOG_HEADER_MAGIC_NUM ==
937 be32_to_cpu(*(__be32 *)offset)) {
944 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
946 return XFS_ERROR(EIO);
949 /* find blk_no of tail of log */
950 rhead = (xlog_rec_header_t *)offset;
951 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
954 * Reset log values according to the state of the log when we
955 * crashed. In the case where head_blk == 0, we bump curr_cycle
956 * one because the next write starts a new cycle rather than
957 * continuing the cycle of the last good log record. At this
958 * point we have guaranteed that all partial log records have been
959 * accounted for. Therefore, we know that the last good log record
960 * written was complete and ended exactly on the end boundary
961 * of the physical log.
963 log->l_prev_block = i;
964 log->l_curr_block = (int)*head_blk;
965 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
968 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
969 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
970 xlog_assign_grant_head(&log->l_grant_reserve_head, log->l_curr_cycle,
971 BBTOB(log->l_curr_block));
972 xlog_assign_grant_head(&log->l_grant_write_head, log->l_curr_cycle,
973 BBTOB(log->l_curr_block));
976 * Look for unmount record. If we find it, then we know there
977 * was a clean unmount. Since 'i' could be the last block in
978 * the physical log, we convert to a log block before comparing
981 * Save the current tail lsn to use to pass to
982 * xlog_clear_stale_blocks() below. We won't want to clear the
983 * unmount record if there is one, so we pass the lsn of the
984 * unmount record rather than the block after it.
986 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
987 int h_size = be32_to_cpu(rhead->h_size);
988 int h_version = be32_to_cpu(rhead->h_version);
990 if ((h_version & XLOG_VERSION_2) &&
991 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
992 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
993 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1001 after_umount_blk = (i + hblks + (int)
1002 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
1003 tail_lsn = atomic64_read(&log->l_tail_lsn);
1004 if (*head_blk == after_umount_blk &&
1005 be32_to_cpu(rhead->h_num_logops) == 1) {
1006 umount_data_blk = (i + hblks) % log->l_logBBsize;
1007 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1011 op_head = (xlog_op_header_t *)offset;
1012 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1014 * Set tail and last sync so that newly written
1015 * log records will point recovery to after the
1016 * current unmount record.
1018 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1019 log->l_curr_cycle, after_umount_blk);
1020 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1021 log->l_curr_cycle, after_umount_blk);
1022 *tail_blk = after_umount_blk;
1025 * Note that the unmount was clean. If the unmount
1026 * was not clean, we need to know this to rebuild the
1027 * superblock counters from the perag headers if we
1028 * have a filesystem using non-persistent counters.
1030 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1035 * Make sure that there are no blocks in front of the head
1036 * with the same cycle number as the head. This can happen
1037 * because we allow multiple outstanding log writes concurrently,
1038 * and the later writes might make it out before earlier ones.
1040 * We use the lsn from before modifying it so that we'll never
1041 * overwrite the unmount record after a clean unmount.
1043 * Do this only if we are going to recover the filesystem
1045 * NOTE: This used to say "if (!readonly)"
1046 * However on Linux, we can & do recover a read-only filesystem.
1047 * We only skip recovery if NORECOVERY is specified on mount,
1048 * in which case we would not be here.
1050 * But... if the -device- itself is readonly, just skip this.
1051 * We can't recover this device anyway, so it won't matter.
1053 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
1054 error = xlog_clear_stale_blocks(log, tail_lsn);
1060 xfs_warn(log->l_mp, "failed to locate log tail");
1065 * Is the log zeroed at all?
1067 * The last binary search should be changed to perform an X block read
1068 * once X becomes small enough. You can then search linearly through
1069 * the X blocks. This will cut down on the number of reads we need to do.
1071 * If the log is partially zeroed, this routine will pass back the blkno
1072 * of the first block with cycle number 0. It won't have a complete LR
1076 * 0 => the log is completely written to
1077 * -1 => use *blk_no as the first block of the log
1078 * >0 => error has occurred
1083 xfs_daddr_t *blk_no)
1087 uint first_cycle, last_cycle;
1088 xfs_daddr_t new_blk, last_blk, start_blk;
1089 xfs_daddr_t num_scan_bblks;
1090 int error, log_bbnum = log->l_logBBsize;
1094 /* check totally zeroed log */
1095 bp = xlog_get_bp(log, 1);
1098 error = xlog_bread(log, 0, 1, bp, &offset);
1102 first_cycle = xlog_get_cycle(offset);
1103 if (first_cycle == 0) { /* completely zeroed log */
1109 /* check partially zeroed log */
1110 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1114 last_cycle = xlog_get_cycle(offset);
1115 if (last_cycle != 0) { /* log completely written to */
1118 } else if (first_cycle != 1) {
1120 * If the cycle of the last block is zero, the cycle of
1121 * the first block must be 1. If it's not, maybe we're
1122 * not looking at a log... Bail out.
1125 "Log inconsistent or not a log (last==0, first!=1)");
1126 return XFS_ERROR(EINVAL);
1129 /* we have a partially zeroed log */
1130 last_blk = log_bbnum-1;
1131 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1135 * Validate the answer. Because there is no way to guarantee that
1136 * the entire log is made up of log records which are the same size,
1137 * we scan over the defined maximum blocks. At this point, the maximum
1138 * is not chosen to mean anything special. XXXmiken
1140 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1141 ASSERT(num_scan_bblks <= INT_MAX);
1143 if (last_blk < num_scan_bblks)
1144 num_scan_bblks = last_blk;
1145 start_blk = last_blk - num_scan_bblks;
1148 * We search for any instances of cycle number 0 that occur before
1149 * our current estimate of the head. What we're trying to detect is
1150 * 1 ... | 0 | 1 | 0...
1151 * ^ binary search ends here
1153 if ((error = xlog_find_verify_cycle(log, start_blk,
1154 (int)num_scan_bblks, 0, &new_blk)))
1160 * Potentially backup over partial log record write. We don't need
1161 * to search the end of the log because we know it is zero.
1163 if ((error = xlog_find_verify_log_record(log, start_blk,
1164 &last_blk, 0)) == -1) {
1165 error = XFS_ERROR(EIO);
1179 * These are simple subroutines used by xlog_clear_stale_blocks() below
1180 * to initialize a buffer full of empty log record headers and write
1181 * them into the log.
1192 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1194 memset(buf, 0, BBSIZE);
1195 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1196 recp->h_cycle = cpu_to_be32(cycle);
1197 recp->h_version = cpu_to_be32(
1198 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1199 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1200 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1201 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1202 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1206 xlog_write_log_records(
1217 int sectbb = log->l_sectBBsize;
1218 int end_block = start_block + blocks;
1224 * Greedily allocate a buffer big enough to handle the full
1225 * range of basic blocks to be written. If that fails, try
1226 * a smaller size. We need to be able to write at least a
1227 * log sector, or we're out of luck.
1229 bufblks = 1 << ffs(blocks);
1230 while (!(bp = xlog_get_bp(log, bufblks))) {
1232 if (bufblks < sectbb)
1236 /* We may need to do a read at the start to fill in part of
1237 * the buffer in the starting sector not covered by the first
1240 balign = round_down(start_block, sectbb);
1241 if (balign != start_block) {
1242 error = xlog_bread_noalign(log, start_block, 1, bp);
1246 j = start_block - balign;
1249 for (i = start_block; i < end_block; i += bufblks) {
1250 int bcount, endcount;
1252 bcount = min(bufblks, end_block - start_block);
1253 endcount = bcount - j;
1255 /* We may need to do a read at the end to fill in part of
1256 * the buffer in the final sector not covered by the write.
1257 * If this is the same sector as the above read, skip it.
1259 ealign = round_down(end_block, sectbb);
1260 if (j == 0 && (start_block + endcount > ealign)) {
1261 offset = XFS_BUF_PTR(bp) + BBTOB(ealign - start_block);
1262 error = xlog_bread_offset(log, ealign, sectbb,
1269 offset = xlog_align(log, start_block, endcount, bp);
1270 for (; j < endcount; j++) {
1271 xlog_add_record(log, offset, cycle, i+j,
1272 tail_cycle, tail_block);
1275 error = xlog_bwrite(log, start_block, endcount, bp);
1278 start_block += endcount;
1288 * This routine is called to blow away any incomplete log writes out
1289 * in front of the log head. We do this so that we won't become confused
1290 * if we come up, write only a little bit more, and then crash again.
1291 * If we leave the partial log records out there, this situation could
1292 * cause us to think those partial writes are valid blocks since they
1293 * have the current cycle number. We get rid of them by overwriting them
1294 * with empty log records with the old cycle number rather than the
1297 * The tail lsn is passed in rather than taken from
1298 * the log so that we will not write over the unmount record after a
1299 * clean unmount in a 512 block log. Doing so would leave the log without
1300 * any valid log records in it until a new one was written. If we crashed
1301 * during that time we would not be able to recover.
1304 xlog_clear_stale_blocks(
1308 int tail_cycle, head_cycle;
1309 int tail_block, head_block;
1310 int tail_distance, max_distance;
1314 tail_cycle = CYCLE_LSN(tail_lsn);
1315 tail_block = BLOCK_LSN(tail_lsn);
1316 head_cycle = log->l_curr_cycle;
1317 head_block = log->l_curr_block;
1320 * Figure out the distance between the new head of the log
1321 * and the tail. We want to write over any blocks beyond the
1322 * head that we may have written just before the crash, but
1323 * we don't want to overwrite the tail of the log.
1325 if (head_cycle == tail_cycle) {
1327 * The tail is behind the head in the physical log,
1328 * so the distance from the head to the tail is the
1329 * distance from the head to the end of the log plus
1330 * the distance from the beginning of the log to the
1333 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1334 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1335 XFS_ERRLEVEL_LOW, log->l_mp);
1336 return XFS_ERROR(EFSCORRUPTED);
1338 tail_distance = tail_block + (log->l_logBBsize - head_block);
1341 * The head is behind the tail in the physical log,
1342 * so the distance from the head to the tail is just
1343 * the tail block minus the head block.
1345 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1346 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1347 XFS_ERRLEVEL_LOW, log->l_mp);
1348 return XFS_ERROR(EFSCORRUPTED);
1350 tail_distance = tail_block - head_block;
1354 * If the head is right up against the tail, we can't clear
1357 if (tail_distance <= 0) {
1358 ASSERT(tail_distance == 0);
1362 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1364 * Take the smaller of the maximum amount of outstanding I/O
1365 * we could have and the distance to the tail to clear out.
1366 * We take the smaller so that we don't overwrite the tail and
1367 * we don't waste all day writing from the head to the tail
1370 max_distance = MIN(max_distance, tail_distance);
1372 if ((head_block + max_distance) <= log->l_logBBsize) {
1374 * We can stomp all the blocks we need to without
1375 * wrapping around the end of the log. Just do it
1376 * in a single write. Use the cycle number of the
1377 * current cycle minus one so that the log will look like:
1380 error = xlog_write_log_records(log, (head_cycle - 1),
1381 head_block, max_distance, tail_cycle,
1387 * We need to wrap around the end of the physical log in
1388 * order to clear all the blocks. Do it in two separate
1389 * I/Os. The first write should be from the head to the
1390 * end of the physical log, and it should use the current
1391 * cycle number minus one just like above.
1393 distance = log->l_logBBsize - head_block;
1394 error = xlog_write_log_records(log, (head_cycle - 1),
1395 head_block, distance, tail_cycle,
1402 * Now write the blocks at the start of the physical log.
1403 * This writes the remainder of the blocks we want to clear.
1404 * It uses the current cycle number since we're now on the
1405 * same cycle as the head so that we get:
1406 * n ... n ... | n - 1 ...
1407 * ^^^^^ blocks we're writing
1409 distance = max_distance - (log->l_logBBsize - head_block);
1410 error = xlog_write_log_records(log, head_cycle, 0, distance,
1411 tail_cycle, tail_block);
1419 /******************************************************************************
1421 * Log recover routines
1423 ******************************************************************************
1426 STATIC xlog_recover_t *
1427 xlog_recover_find_tid(
1428 struct hlist_head *head,
1431 xlog_recover_t *trans;
1432 struct hlist_node *n;
1434 hlist_for_each_entry(trans, n, head, r_list) {
1435 if (trans->r_log_tid == tid)
1442 xlog_recover_new_tid(
1443 struct hlist_head *head,
1447 xlog_recover_t *trans;
1449 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1450 trans->r_log_tid = tid;
1452 INIT_LIST_HEAD(&trans->r_itemq);
1454 INIT_HLIST_NODE(&trans->r_list);
1455 hlist_add_head(&trans->r_list, head);
1459 xlog_recover_add_item(
1460 struct list_head *head)
1462 xlog_recover_item_t *item;
1464 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1465 INIT_LIST_HEAD(&item->ri_list);
1466 list_add_tail(&item->ri_list, head);
1470 xlog_recover_add_to_cont_trans(
1472 xlog_recover_t *trans,
1476 xlog_recover_item_t *item;
1477 xfs_caddr_t ptr, old_ptr;
1480 if (list_empty(&trans->r_itemq)) {
1481 /* finish copying rest of trans header */
1482 xlog_recover_add_item(&trans->r_itemq);
1483 ptr = (xfs_caddr_t) &trans->r_theader +
1484 sizeof(xfs_trans_header_t) - len;
1485 memcpy(ptr, dp, len); /* d, s, l */
1488 /* take the tail entry */
1489 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1491 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1492 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1494 ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
1495 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1496 item->ri_buf[item->ri_cnt-1].i_len += len;
1497 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1498 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
1503 * The next region to add is the start of a new region. It could be
1504 * a whole region or it could be the first part of a new region. Because
1505 * of this, the assumption here is that the type and size fields of all
1506 * format structures fit into the first 32 bits of the structure.
1508 * This works because all regions must be 32 bit aligned. Therefore, we
1509 * either have both fields or we have neither field. In the case we have
1510 * neither field, the data part of the region is zero length. We only have
1511 * a log_op_header and can throw away the header since a new one will appear
1512 * later. If we have at least 4 bytes, then we can determine how many regions
1513 * will appear in the current log item.
1516 xlog_recover_add_to_trans(
1518 xlog_recover_t *trans,
1522 xfs_inode_log_format_t *in_f; /* any will do */
1523 xlog_recover_item_t *item;
1528 if (list_empty(&trans->r_itemq)) {
1529 /* we need to catch log corruptions here */
1530 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
1531 xfs_warn(log->l_mp, "%s: bad header magic number",
1534 return XFS_ERROR(EIO);
1536 if (len == sizeof(xfs_trans_header_t))
1537 xlog_recover_add_item(&trans->r_itemq);
1538 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1542 ptr = kmem_alloc(len, KM_SLEEP);
1543 memcpy(ptr, dp, len);
1544 in_f = (xfs_inode_log_format_t *)ptr;
1546 /* take the tail entry */
1547 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1548 if (item->ri_total != 0 &&
1549 item->ri_total == item->ri_cnt) {
1550 /* tail item is in use, get a new one */
1551 xlog_recover_add_item(&trans->r_itemq);
1552 item = list_entry(trans->r_itemq.prev,
1553 xlog_recover_item_t, ri_list);
1556 if (item->ri_total == 0) { /* first region to be added */
1557 if (in_f->ilf_size == 0 ||
1558 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
1560 "bad number of regions (%d) in inode log format",
1563 return XFS_ERROR(EIO);
1566 item->ri_total = in_f->ilf_size;
1568 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1571 ASSERT(item->ri_total > item->ri_cnt);
1572 /* Description region is ri_buf[0] */
1573 item->ri_buf[item->ri_cnt].i_addr = ptr;
1574 item->ri_buf[item->ri_cnt].i_len = len;
1576 trace_xfs_log_recover_item_add(log, trans, item, 0);
1581 * Sort the log items in the transaction. Cancelled buffers need
1582 * to be put first so they are processed before any items that might
1583 * modify the buffers. If they are cancelled, then the modifications
1584 * don't need to be replayed.
1587 xlog_recover_reorder_trans(
1589 xlog_recover_t *trans,
1592 xlog_recover_item_t *item, *n;
1593 LIST_HEAD(sort_list);
1595 list_splice_init(&trans->r_itemq, &sort_list);
1596 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1597 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1599 switch (ITEM_TYPE(item)) {
1601 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1602 trace_xfs_log_recover_item_reorder_head(log,
1604 list_move(&item->ri_list, &trans->r_itemq);
1609 case XFS_LI_QUOTAOFF:
1612 trace_xfs_log_recover_item_reorder_tail(log,
1614 list_move_tail(&item->ri_list, &trans->r_itemq);
1618 "%s: unrecognized type of log operation",
1621 return XFS_ERROR(EIO);
1624 ASSERT(list_empty(&sort_list));
1629 * Build up the table of buf cancel records so that we don't replay
1630 * cancelled data in the second pass. For buffer records that are
1631 * not cancel records, there is nothing to do here so we just return.
1633 * If we get a cancel record which is already in the table, this indicates
1634 * that the buffer was cancelled multiple times. In order to ensure
1635 * that during pass 2 we keep the record in the table until we reach its
1636 * last occurrence in the log, we keep a reference count in the cancel
1637 * record in the table to tell us how many times we expect to see this
1638 * record during the second pass.
1641 xlog_recover_buffer_pass1(
1643 xlog_recover_item_t *item)
1645 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1646 struct list_head *bucket;
1647 struct xfs_buf_cancel *bcp;
1650 * If this isn't a cancel buffer item, then just return.
1652 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1653 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
1658 * Insert an xfs_buf_cancel record into the hash table of them.
1659 * If there is already an identical record, bump its reference count.
1661 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1662 list_for_each_entry(bcp, bucket, bc_list) {
1663 if (bcp->bc_blkno == buf_f->blf_blkno &&
1664 bcp->bc_len == buf_f->blf_len) {
1666 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
1671 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1672 bcp->bc_blkno = buf_f->blf_blkno;
1673 bcp->bc_len = buf_f->blf_len;
1674 bcp->bc_refcount = 1;
1675 list_add_tail(&bcp->bc_list, bucket);
1677 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
1682 * Check to see whether the buffer being recovered has a corresponding
1683 * entry in the buffer cancel record table. If it does then return 1
1684 * so that it will be cancelled, otherwise return 0. If the buffer is
1685 * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement
1686 * the refcount on the entry in the table and remove it from the table
1687 * if this is the last reference.
1689 * We remove the cancel record from the table when we encounter its
1690 * last occurrence in the log so that if the same buffer is re-used
1691 * again after its last cancellation we actually replay the changes
1692 * made at that point.
1695 xlog_check_buffer_cancelled(
1701 struct list_head *bucket;
1702 struct xfs_buf_cancel *bcp;
1704 if (log->l_buf_cancel_table == NULL) {
1706 * There is nothing in the table built in pass one,
1707 * so this buffer must not be cancelled.
1709 ASSERT(!(flags & XFS_BLF_CANCEL));
1714 * Search for an entry in the cancel table that matches our buffer.
1716 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1717 list_for_each_entry(bcp, bucket, bc_list) {
1718 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
1723 * We didn't find a corresponding entry in the table, so return 0 so
1724 * that the buffer is NOT cancelled.
1726 ASSERT(!(flags & XFS_BLF_CANCEL));
1731 * We've go a match, so return 1 so that the recovery of this buffer
1732 * is cancelled. If this buffer is actually a buffer cancel log
1733 * item, then decrement the refcount on the one in the table and
1734 * remove it if this is the last reference.
1736 if (flags & XFS_BLF_CANCEL) {
1737 if (--bcp->bc_refcount == 0) {
1738 list_del(&bcp->bc_list);
1746 * Perform recovery for a buffer full of inodes. In these buffers, the only
1747 * data which should be recovered is that which corresponds to the
1748 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1749 * data for the inodes is always logged through the inodes themselves rather
1750 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1752 * The only time when buffers full of inodes are fully recovered is when the
1753 * buffer is full of newly allocated inodes. In this case the buffer will
1754 * not be marked as an inode buffer and so will be sent to
1755 * xlog_recover_do_reg_buffer() below during recovery.
1758 xlog_recover_do_inode_buffer(
1759 struct xfs_mount *mp,
1760 xlog_recover_item_t *item,
1762 xfs_buf_log_format_t *buf_f)
1768 int reg_buf_offset = 0;
1769 int reg_buf_bytes = 0;
1770 int next_unlinked_offset;
1772 xfs_agino_t *logged_nextp;
1773 xfs_agino_t *buffer_nextp;
1775 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
1777 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1778 for (i = 0; i < inodes_per_buf; i++) {
1779 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1780 offsetof(xfs_dinode_t, di_next_unlinked);
1782 while (next_unlinked_offset >=
1783 (reg_buf_offset + reg_buf_bytes)) {
1785 * The next di_next_unlinked field is beyond
1786 * the current logged region. Find the next
1787 * logged region that contains or is beyond
1788 * the current di_next_unlinked field.
1791 bit = xfs_next_bit(buf_f->blf_data_map,
1792 buf_f->blf_map_size, bit);
1795 * If there are no more logged regions in the
1796 * buffer, then we're done.
1801 nbits = xfs_contig_bits(buf_f->blf_data_map,
1802 buf_f->blf_map_size, bit);
1804 reg_buf_offset = bit << XFS_BLF_SHIFT;
1805 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
1810 * If the current logged region starts after the current
1811 * di_next_unlinked field, then move on to the next
1812 * di_next_unlinked field.
1814 if (next_unlinked_offset < reg_buf_offset)
1817 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1818 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
1819 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1822 * The current logged region contains a copy of the
1823 * current di_next_unlinked field. Extract its value
1824 * and copy it to the buffer copy.
1826 logged_nextp = item->ri_buf[item_index].i_addr +
1827 next_unlinked_offset - reg_buf_offset;
1828 if (unlikely(*logged_nextp == 0)) {
1830 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1831 "Trying to replay bad (0) inode di_next_unlinked field.",
1833 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1834 XFS_ERRLEVEL_LOW, mp);
1835 return XFS_ERROR(EFSCORRUPTED);
1838 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1839 next_unlinked_offset);
1840 *buffer_nextp = *logged_nextp;
1847 * Perform a 'normal' buffer recovery. Each logged region of the
1848 * buffer should be copied over the corresponding region in the
1849 * given buffer. The bitmap in the buf log format structure indicates
1850 * where to place the logged data.
1853 xlog_recover_do_reg_buffer(
1854 struct xfs_mount *mp,
1855 xlog_recover_item_t *item,
1857 xfs_buf_log_format_t *buf_f)
1864 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
1867 i = 1; /* 0 is the buf format structure */
1869 bit = xfs_next_bit(buf_f->blf_data_map,
1870 buf_f->blf_map_size, bit);
1873 nbits = xfs_contig_bits(buf_f->blf_data_map,
1874 buf_f->blf_map_size, bit);
1876 ASSERT(item->ri_buf[i].i_addr != NULL);
1877 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
1878 ASSERT(XFS_BUF_COUNT(bp) >=
1879 ((uint)bit << XFS_BLF_SHIFT)+(nbits<<XFS_BLF_SHIFT));
1882 * Do a sanity check if this is a dquot buffer. Just checking
1883 * the first dquot in the buffer should do. XXXThis is
1884 * probably a good thing to do for other buf types also.
1887 if (buf_f->blf_flags &
1888 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
1889 if (item->ri_buf[i].i_addr == NULL) {
1891 "XFS: NULL dquot in %s.", __func__);
1894 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
1896 "XFS: dquot too small (%d) in %s.",
1897 item->ri_buf[i].i_len, __func__);
1900 error = xfs_qm_dqcheck(mp, item->ri_buf[i].i_addr,
1901 -1, 0, XFS_QMOPT_DOWARN,
1902 "dquot_buf_recover");
1907 memcpy(xfs_buf_offset(bp,
1908 (uint)bit << XFS_BLF_SHIFT), /* dest */
1909 item->ri_buf[i].i_addr, /* source */
1910 nbits<<XFS_BLF_SHIFT); /* length */
1916 /* Shouldn't be any more regions */
1917 ASSERT(i == item->ri_total);
1921 * Do some primitive error checking on ondisk dquot data structures.
1925 struct xfs_mount *mp,
1926 xfs_disk_dquot_t *ddq,
1928 uint type, /* used only when IO_dorepair is true */
1932 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
1936 * We can encounter an uninitialized dquot buffer for 2 reasons:
1937 * 1. If we crash while deleting the quotainode(s), and those blks got
1938 * used for user data. This is because we take the path of regular
1939 * file deletion; however, the size field of quotainodes is never
1940 * updated, so all the tricks that we play in itruncate_finish
1941 * don't quite matter.
1943 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1944 * But the allocation will be replayed so we'll end up with an
1945 * uninitialized quota block.
1947 * This is all fine; things are still consistent, and we haven't lost
1948 * any quota information. Just don't complain about bad dquot blks.
1950 if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) {
1951 if (flags & XFS_QMOPT_DOWARN)
1953 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
1954 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
1957 if (ddq->d_version != XFS_DQUOT_VERSION) {
1958 if (flags & XFS_QMOPT_DOWARN)
1960 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
1961 str, id, ddq->d_version, XFS_DQUOT_VERSION);
1965 if (ddq->d_flags != XFS_DQ_USER &&
1966 ddq->d_flags != XFS_DQ_PROJ &&
1967 ddq->d_flags != XFS_DQ_GROUP) {
1968 if (flags & XFS_QMOPT_DOWARN)
1970 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
1971 str, id, ddq->d_flags);
1975 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
1976 if (flags & XFS_QMOPT_DOWARN)
1978 "%s : ondisk-dquot 0x%p, ID mismatch: "
1979 "0x%x expected, found id 0x%x",
1980 str, ddq, id, be32_to_cpu(ddq->d_id));
1984 if (!errs && ddq->d_id) {
1985 if (ddq->d_blk_softlimit &&
1986 be64_to_cpu(ddq->d_bcount) >=
1987 be64_to_cpu(ddq->d_blk_softlimit)) {
1988 if (!ddq->d_btimer) {
1989 if (flags & XFS_QMOPT_DOWARN)
1991 "%s : Dquot ID 0x%x (0x%p) BLK TIMER NOT STARTED",
1992 str, (int)be32_to_cpu(ddq->d_id), ddq);
1996 if (ddq->d_ino_softlimit &&
1997 be64_to_cpu(ddq->d_icount) >=
1998 be64_to_cpu(ddq->d_ino_softlimit)) {
1999 if (!ddq->d_itimer) {
2000 if (flags & XFS_QMOPT_DOWARN)
2002 "%s : Dquot ID 0x%x (0x%p) INODE TIMER NOT STARTED",
2003 str, (int)be32_to_cpu(ddq->d_id), ddq);
2007 if (ddq->d_rtb_softlimit &&
2008 be64_to_cpu(ddq->d_rtbcount) >=
2009 be64_to_cpu(ddq->d_rtb_softlimit)) {
2010 if (!ddq->d_rtbtimer) {
2011 if (flags & XFS_QMOPT_DOWARN)
2013 "%s : Dquot ID 0x%x (0x%p) RTBLK TIMER NOT STARTED",
2014 str, (int)be32_to_cpu(ddq->d_id), ddq);
2020 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2023 if (flags & XFS_QMOPT_DOWARN)
2024 xfs_notice(mp, "Re-initializing dquot ID 0x%x", id);
2027 * Typically, a repair is only requested by quotacheck.
2030 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2031 memset(d, 0, sizeof(xfs_dqblk_t));
2033 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2034 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2035 d->dd_diskdq.d_flags = type;
2036 d->dd_diskdq.d_id = cpu_to_be32(id);
2042 * Perform a dquot buffer recovery.
2043 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2044 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2045 * Else, treat it as a regular buffer and do recovery.
2048 xlog_recover_do_dquot_buffer(
2051 xlog_recover_item_t *item,
2053 xfs_buf_log_format_t *buf_f)
2057 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2060 * Filesystems are required to send in quota flags at mount time.
2062 if (mp->m_qflags == 0) {
2067 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2068 type |= XFS_DQ_USER;
2069 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2070 type |= XFS_DQ_PROJ;
2071 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2072 type |= XFS_DQ_GROUP;
2074 * This type of quotas was turned off, so ignore this buffer
2076 if (log->l_quotaoffs_flag & type)
2079 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2083 * This routine replays a modification made to a buffer at runtime.
2084 * There are actually two types of buffer, regular and inode, which
2085 * are handled differently. Inode buffers are handled differently
2086 * in that we only recover a specific set of data from them, namely
2087 * the inode di_next_unlinked fields. This is because all other inode
2088 * data is actually logged via inode records and any data we replay
2089 * here which overlaps that may be stale.
2091 * When meta-data buffers are freed at run time we log a buffer item
2092 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2093 * of the buffer in the log should not be replayed at recovery time.
2094 * This is so that if the blocks covered by the buffer are reused for
2095 * file data before we crash we don't end up replaying old, freed
2096 * meta-data into a user's file.
2098 * To handle the cancellation of buffer log items, we make two passes
2099 * over the log during recovery. During the first we build a table of
2100 * those buffers which have been cancelled, and during the second we
2101 * only replay those buffers which do not have corresponding cancel
2102 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2103 * for more details on the implementation of the table of cancel records.
2106 xlog_recover_buffer_pass2(
2108 xlog_recover_item_t *item)
2110 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2111 xfs_mount_t *mp = log->l_mp;
2117 * In this pass we only want to recover all the buffers which have
2118 * not been cancelled and are not cancellation buffers themselves.
2120 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2121 buf_f->blf_len, buf_f->blf_flags)) {
2122 trace_xfs_log_recover_buf_cancel(log, buf_f);
2126 trace_xfs_log_recover_buf_recover(log, buf_f);
2128 buf_flags = XBF_LOCK;
2129 if (!(buf_f->blf_flags & XFS_BLF_INODE_BUF))
2130 buf_flags |= XBF_MAPPED;
2132 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2134 if (XFS_BUF_ISERROR(bp)) {
2135 xfs_ioerror_alert("xlog_recover_do..(read#1)", mp,
2136 bp, buf_f->blf_blkno);
2137 error = XFS_BUF_GETERROR(bp);
2143 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
2144 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2145 } else if (buf_f->blf_flags &
2146 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2147 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2149 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2152 return XFS_ERROR(error);
2155 * Perform delayed write on the buffer. Asynchronous writes will be
2156 * slower when taking into account all the buffers to be flushed.
2158 * Also make sure that only inode buffers with good sizes stay in
2159 * the buffer cache. The kernel moves inodes in buffers of 1 block
2160 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2161 * buffers in the log can be a different size if the log was generated
2162 * by an older kernel using unclustered inode buffers or a newer kernel
2163 * running with a different inode cluster size. Regardless, if the
2164 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2165 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2166 * the buffer out of the buffer cache so that the buffer won't
2167 * overlap with future reads of those inodes.
2169 if (XFS_DINODE_MAGIC ==
2170 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2171 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2172 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2174 error = xfs_bwrite(mp, bp);
2176 ASSERT(bp->b_target->bt_mount == mp);
2177 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2178 xfs_bdwrite(mp, bp);
2185 xlog_recover_inode_pass2(
2187 xlog_recover_item_t *item)
2189 xfs_inode_log_format_t *in_f;
2190 xfs_mount_t *mp = log->l_mp;
2199 xfs_icdinode_t *dicp;
2202 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2203 in_f = item->ri_buf[0].i_addr;
2205 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
2207 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2213 * Inode buffers can be freed, look out for it,
2214 * and do not replay the inode.
2216 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2217 in_f->ilf_len, 0)) {
2219 trace_xfs_log_recover_inode_cancel(log, in_f);
2222 trace_xfs_log_recover_inode_recover(log, in_f);
2224 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len,
2226 if (XFS_BUF_ISERROR(bp)) {
2227 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
2228 bp, in_f->ilf_blkno);
2229 error = XFS_BUF_GETERROR(bp);
2234 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2235 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
2238 * Make sure the place we're flushing out to really looks
2241 if (unlikely(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC)) {
2244 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2245 __func__, dip, bp, in_f->ilf_ino);
2246 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2247 XFS_ERRLEVEL_LOW, mp);
2248 error = EFSCORRUPTED;
2251 dicp = item->ri_buf[1].i_addr;
2252 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2255 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2256 __func__, item, in_f->ilf_ino);
2257 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2258 XFS_ERRLEVEL_LOW, mp);
2259 error = EFSCORRUPTED;
2263 /* Skip replay when the on disk inode is newer than the log one */
2264 if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
2266 * Deal with the wrap case, DI_MAX_FLUSH is less
2267 * than smaller numbers
2269 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
2270 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2274 trace_xfs_log_recover_inode_skip(log, in_f);
2279 /* Take the opportunity to reset the flush iteration count */
2280 dicp->di_flushiter = 0;
2282 if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) {
2283 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2284 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2285 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2286 XFS_ERRLEVEL_LOW, mp, dicp);
2289 "%s: Bad regular inode log record, rec ptr 0x%p, "
2290 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2291 __func__, item, dip, bp, in_f->ilf_ino);
2292 error = EFSCORRUPTED;
2295 } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) {
2296 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2297 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2298 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2299 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2300 XFS_ERRLEVEL_LOW, mp, dicp);
2303 "%s: Bad dir inode log record, rec ptr 0x%p, "
2304 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2305 __func__, item, dip, bp, in_f->ilf_ino);
2306 error = EFSCORRUPTED;
2310 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2311 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2312 XFS_ERRLEVEL_LOW, mp, dicp);
2315 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2316 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2317 __func__, item, dip, bp, in_f->ilf_ino,
2318 dicp->di_nextents + dicp->di_anextents,
2320 error = EFSCORRUPTED;
2323 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2324 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2325 XFS_ERRLEVEL_LOW, mp, dicp);
2328 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2329 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
2330 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
2331 error = EFSCORRUPTED;
2334 if (unlikely(item->ri_buf[1].i_len > sizeof(struct xfs_icdinode))) {
2335 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
2336 XFS_ERRLEVEL_LOW, mp, dicp);
2339 "%s: Bad inode log record length %d, rec ptr 0x%p",
2340 __func__, item->ri_buf[1].i_len, item);
2341 error = EFSCORRUPTED;
2345 /* The core is in in-core format */
2346 xfs_dinode_to_disk(dip, item->ri_buf[1].i_addr);
2348 /* the rest is in on-disk format */
2349 if (item->ri_buf[1].i_len > sizeof(struct xfs_icdinode)) {
2350 memcpy((xfs_caddr_t) dip + sizeof(struct xfs_icdinode),
2351 item->ri_buf[1].i_addr + sizeof(struct xfs_icdinode),
2352 item->ri_buf[1].i_len - sizeof(struct xfs_icdinode));
2355 fields = in_f->ilf_fields;
2356 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2358 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
2361 memcpy(XFS_DFORK_DPTR(dip),
2362 &in_f->ilf_u.ilfu_uuid,
2367 if (in_f->ilf_size == 2)
2368 goto write_inode_buffer;
2369 len = item->ri_buf[2].i_len;
2370 src = item->ri_buf[2].i_addr;
2371 ASSERT(in_f->ilf_size <= 4);
2372 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2373 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2374 (len == in_f->ilf_dsize));
2376 switch (fields & XFS_ILOG_DFORK) {
2377 case XFS_ILOG_DDATA:
2379 memcpy(XFS_DFORK_DPTR(dip), src, len);
2382 case XFS_ILOG_DBROOT:
2383 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
2384 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
2385 XFS_DFORK_DSIZE(dip, mp));
2390 * There are no data fork flags set.
2392 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2397 * If we logged any attribute data, recover it. There may or
2398 * may not have been any other non-core data logged in this
2401 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2402 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2407 len = item->ri_buf[attr_index].i_len;
2408 src = item->ri_buf[attr_index].i_addr;
2409 ASSERT(len == in_f->ilf_asize);
2411 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2412 case XFS_ILOG_ADATA:
2414 dest = XFS_DFORK_APTR(dip);
2415 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2416 memcpy(dest, src, len);
2419 case XFS_ILOG_ABROOT:
2420 dest = XFS_DFORK_APTR(dip);
2421 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2422 len, (xfs_bmdr_block_t*)dest,
2423 XFS_DFORK_ASIZE(dip, mp));
2427 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
2436 ASSERT(bp->b_target->bt_mount == mp);
2437 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2438 xfs_bdwrite(mp, bp);
2442 return XFS_ERROR(error);
2446 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2447 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2451 xlog_recover_quotaoff_pass1(
2453 xlog_recover_item_t *item)
2455 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
2459 * The logitem format's flag tells us if this was user quotaoff,
2460 * group/project quotaoff or both.
2462 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2463 log->l_quotaoffs_flag |= XFS_DQ_USER;
2464 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2465 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2466 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2467 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2473 * Recover a dquot record
2476 xlog_recover_dquot_pass2(
2478 xlog_recover_item_t *item)
2480 xfs_mount_t *mp = log->l_mp;
2482 struct xfs_disk_dquot *ddq, *recddq;
2484 xfs_dq_logformat_t *dq_f;
2489 * Filesystems are required to send in quota flags at mount time.
2491 if (mp->m_qflags == 0)
2494 recddq = item->ri_buf[1].i_addr;
2495 if (recddq == NULL) {
2496 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
2497 return XFS_ERROR(EIO);
2499 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
2500 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
2501 item->ri_buf[1].i_len, __func__);
2502 return XFS_ERROR(EIO);
2506 * This type of quotas was turned off, so ignore this record.
2508 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
2510 if (log->l_quotaoffs_flag & type)
2514 * At this point we know that quota was _not_ turned off.
2515 * Since the mount flags are not indicating to us otherwise, this
2516 * must mean that quota is on, and the dquot needs to be replayed.
2517 * Remember that we may not have fully recovered the superblock yet,
2518 * so we can't do the usual trick of looking at the SB quota bits.
2520 * The other possibility, of course, is that the quota subsystem was
2521 * removed since the last mount - ENOSYS.
2523 dq_f = item->ri_buf[0].i_addr;
2525 error = xfs_qm_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2526 "xlog_recover_dquot_pass2 (log copy)");
2528 return XFS_ERROR(EIO);
2529 ASSERT(dq_f->qlf_len == 1);
2531 error = xfs_read_buf(mp, mp->m_ddev_targp,
2533 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2536 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
2537 bp, dq_f->qlf_blkno);
2541 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2544 * At least the magic num portion should be on disk because this
2545 * was among a chunk of dquots created earlier, and we did some
2546 * minimal initialization then.
2548 error = xfs_qm_dqcheck(mp, ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2549 "xlog_recover_dquot_pass2");
2552 return XFS_ERROR(EIO);
2555 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2557 ASSERT(dq_f->qlf_size == 2);
2558 ASSERT(bp->b_target->bt_mount == mp);
2559 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2560 xfs_bdwrite(mp, bp);
2566 * This routine is called to create an in-core extent free intent
2567 * item from the efi format structure which was logged on disk.
2568 * It allocates an in-core efi, copies the extents from the format
2569 * structure into it, and adds the efi to the AIL with the given
2573 xlog_recover_efi_pass2(
2575 xlog_recover_item_t *item,
2579 xfs_mount_t *mp = log->l_mp;
2580 xfs_efi_log_item_t *efip;
2581 xfs_efi_log_format_t *efi_formatp;
2583 efi_formatp = item->ri_buf[0].i_addr;
2585 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
2586 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2587 &(efip->efi_format)))) {
2588 xfs_efi_item_free(efip);
2591 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
2593 spin_lock(&log->l_ailp->xa_lock);
2595 * xfs_trans_ail_update() drops the AIL lock.
2597 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
2603 * This routine is called when an efd format structure is found in
2604 * a committed transaction in the log. It's purpose is to cancel
2605 * the corresponding efi if it was still in the log. To do this
2606 * it searches the AIL for the efi with an id equal to that in the
2607 * efd format structure. If we find it, we remove the efi from the
2611 xlog_recover_efd_pass2(
2613 xlog_recover_item_t *item)
2615 xfs_efd_log_format_t *efd_formatp;
2616 xfs_efi_log_item_t *efip = NULL;
2617 xfs_log_item_t *lip;
2619 struct xfs_ail_cursor cur;
2620 struct xfs_ail *ailp = log->l_ailp;
2622 efd_formatp = item->ri_buf[0].i_addr;
2623 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2624 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2625 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2626 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
2627 efi_id = efd_formatp->efd_efi_id;
2630 * Search for the efi with the id in the efd format structure
2633 spin_lock(&ailp->xa_lock);
2634 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
2635 while (lip != NULL) {
2636 if (lip->li_type == XFS_LI_EFI) {
2637 efip = (xfs_efi_log_item_t *)lip;
2638 if (efip->efi_format.efi_id == efi_id) {
2640 * xfs_trans_ail_delete() drops the
2643 xfs_trans_ail_delete(ailp, lip);
2644 xfs_efi_item_free(efip);
2645 spin_lock(&ailp->xa_lock);
2649 lip = xfs_trans_ail_cursor_next(ailp, &cur);
2651 xfs_trans_ail_cursor_done(ailp, &cur);
2652 spin_unlock(&ailp->xa_lock);
2658 * Free up any resources allocated by the transaction
2660 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2663 xlog_recover_free_trans(
2664 struct xlog_recover *trans)
2666 xlog_recover_item_t *item, *n;
2669 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
2670 /* Free the regions in the item. */
2671 list_del(&item->ri_list);
2672 for (i = 0; i < item->ri_cnt; i++)
2673 kmem_free(item->ri_buf[i].i_addr);
2674 /* Free the item itself */
2675 kmem_free(item->ri_buf);
2678 /* Free the transaction recover structure */
2683 xlog_recover_commit_pass1(
2685 struct xlog_recover *trans,
2686 xlog_recover_item_t *item)
2688 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
2690 switch (ITEM_TYPE(item)) {
2692 return xlog_recover_buffer_pass1(log, item);
2693 case XFS_LI_QUOTAOFF:
2694 return xlog_recover_quotaoff_pass1(log, item);
2699 /* nothing to do in pass 1 */
2702 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
2703 __func__, ITEM_TYPE(item));
2705 return XFS_ERROR(EIO);
2710 xlog_recover_commit_pass2(
2712 struct xlog_recover *trans,
2713 xlog_recover_item_t *item)
2715 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
2717 switch (ITEM_TYPE(item)) {
2719 return xlog_recover_buffer_pass2(log, item);
2721 return xlog_recover_inode_pass2(log, item);
2723 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
2725 return xlog_recover_efd_pass2(log, item);
2727 return xlog_recover_dquot_pass2(log, item);
2728 case XFS_LI_QUOTAOFF:
2729 /* nothing to do in pass2 */
2732 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
2733 __func__, ITEM_TYPE(item));
2735 return XFS_ERROR(EIO);
2740 * Perform the transaction.
2742 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2743 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2746 xlog_recover_commit_trans(
2748 struct xlog_recover *trans,
2752 xlog_recover_item_t *item;
2754 hlist_del(&trans->r_list);
2756 error = xlog_recover_reorder_trans(log, trans, pass);
2760 list_for_each_entry(item, &trans->r_itemq, ri_list) {
2761 if (pass == XLOG_RECOVER_PASS1)
2762 error = xlog_recover_commit_pass1(log, trans, item);
2764 error = xlog_recover_commit_pass2(log, trans, item);
2769 xlog_recover_free_trans(trans);
2774 xlog_recover_unmount_trans(
2776 xlog_recover_t *trans)
2778 /* Do nothing now */
2779 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
2784 * There are two valid states of the r_state field. 0 indicates that the
2785 * transaction structure is in a normal state. We have either seen the
2786 * start of the transaction or the last operation we added was not a partial
2787 * operation. If the last operation we added to the transaction was a
2788 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2790 * NOTE: skip LRs with 0 data length.
2793 xlog_recover_process_data(
2795 struct hlist_head rhash[],
2796 xlog_rec_header_t *rhead,
2802 xlog_op_header_t *ohead;
2803 xlog_recover_t *trans;
2809 lp = dp + be32_to_cpu(rhead->h_len);
2810 num_logops = be32_to_cpu(rhead->h_num_logops);
2812 /* check the log format matches our own - else we can't recover */
2813 if (xlog_header_check_recover(log->l_mp, rhead))
2814 return (XFS_ERROR(EIO));
2816 while ((dp < lp) && num_logops) {
2817 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2818 ohead = (xlog_op_header_t *)dp;
2819 dp += sizeof(xlog_op_header_t);
2820 if (ohead->oh_clientid != XFS_TRANSACTION &&
2821 ohead->oh_clientid != XFS_LOG) {
2822 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
2823 __func__, ohead->oh_clientid);
2825 return (XFS_ERROR(EIO));
2827 tid = be32_to_cpu(ohead->oh_tid);
2828 hash = XLOG_RHASH(tid);
2829 trans = xlog_recover_find_tid(&rhash[hash], tid);
2830 if (trans == NULL) { /* not found; add new tid */
2831 if (ohead->oh_flags & XLOG_START_TRANS)
2832 xlog_recover_new_tid(&rhash[hash], tid,
2833 be64_to_cpu(rhead->h_lsn));
2835 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
2836 xfs_warn(log->l_mp, "%s: bad length 0x%x",
2837 __func__, be32_to_cpu(ohead->oh_len));
2839 return (XFS_ERROR(EIO));
2841 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2842 if (flags & XLOG_WAS_CONT_TRANS)
2843 flags &= ~XLOG_CONTINUE_TRANS;
2845 case XLOG_COMMIT_TRANS:
2846 error = xlog_recover_commit_trans(log,
2849 case XLOG_UNMOUNT_TRANS:
2850 error = xlog_recover_unmount_trans(log, trans);
2852 case XLOG_WAS_CONT_TRANS:
2853 error = xlog_recover_add_to_cont_trans(log,
2855 be32_to_cpu(ohead->oh_len));
2857 case XLOG_START_TRANS:
2858 xfs_warn(log->l_mp, "%s: bad transaction",
2861 error = XFS_ERROR(EIO);
2864 case XLOG_CONTINUE_TRANS:
2865 error = xlog_recover_add_to_trans(log, trans,
2866 dp, be32_to_cpu(ohead->oh_len));
2869 xfs_warn(log->l_mp, "%s: bad flag 0x%x",
2872 error = XFS_ERROR(EIO);
2878 dp += be32_to_cpu(ohead->oh_len);
2885 * Process an extent free intent item that was recovered from
2886 * the log. We need to free the extents that it describes.
2889 xlog_recover_process_efi(
2891 xfs_efi_log_item_t *efip)
2893 xfs_efd_log_item_t *efdp;
2898 xfs_fsblock_t startblock_fsb;
2900 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
2903 * First check the validity of the extents described by the
2904 * EFI. If any are bad, then assume that all are bad and
2905 * just toss the EFI.
2907 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2908 extp = &(efip->efi_format.efi_extents[i]);
2909 startblock_fsb = XFS_BB_TO_FSB(mp,
2910 XFS_FSB_TO_DADDR(mp, extp->ext_start));
2911 if ((startblock_fsb == 0) ||
2912 (extp->ext_len == 0) ||
2913 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
2914 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
2916 * This will pull the EFI from the AIL and
2917 * free the memory associated with it.
2919 xfs_efi_release(efip, efip->efi_format.efi_nextents);
2920 return XFS_ERROR(EIO);
2924 tp = xfs_trans_alloc(mp, 0);
2925 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
2928 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
2930 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2931 extp = &(efip->efi_format.efi_extents[i]);
2932 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
2935 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
2939 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
2940 error = xfs_trans_commit(tp, 0);
2944 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
2949 * When this is called, all of the EFIs which did not have
2950 * corresponding EFDs should be in the AIL. What we do now
2951 * is free the extents associated with each one.
2953 * Since we process the EFIs in normal transactions, they
2954 * will be removed at some point after the commit. This prevents
2955 * us from just walking down the list processing each one.
2956 * We'll use a flag in the EFI to skip those that we've already
2957 * processed and use the AIL iteration mechanism's generation
2958 * count to try to speed this up at least a bit.
2960 * When we start, we know that the EFIs are the only things in
2961 * the AIL. As we process them, however, other items are added
2962 * to the AIL. Since everything added to the AIL must come after
2963 * everything already in the AIL, we stop processing as soon as
2964 * we see something other than an EFI in the AIL.
2967 xlog_recover_process_efis(
2970 xfs_log_item_t *lip;
2971 xfs_efi_log_item_t *efip;
2973 struct xfs_ail_cursor cur;
2974 struct xfs_ail *ailp;
2977 spin_lock(&ailp->xa_lock);
2978 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
2979 while (lip != NULL) {
2981 * We're done when we see something other than an EFI.
2982 * There should be no EFIs left in the AIL now.
2984 if (lip->li_type != XFS_LI_EFI) {
2986 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
2987 ASSERT(lip->li_type != XFS_LI_EFI);
2993 * Skip EFIs that we've already processed.
2995 efip = (xfs_efi_log_item_t *)lip;
2996 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
2997 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3001 spin_unlock(&ailp->xa_lock);
3002 error = xlog_recover_process_efi(log->l_mp, efip);
3003 spin_lock(&ailp->xa_lock);
3006 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3009 xfs_trans_ail_cursor_done(ailp, &cur);
3010 spin_unlock(&ailp->xa_lock);
3015 * This routine performs a transaction to null out a bad inode pointer
3016 * in an agi unlinked inode hash bucket.
3019 xlog_recover_clear_agi_bucket(
3021 xfs_agnumber_t agno,
3030 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3031 error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp),
3036 error = xfs_read_agi(mp, tp, agno, &agibp);
3040 agi = XFS_BUF_TO_AGI(agibp);
3041 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
3042 offset = offsetof(xfs_agi_t, agi_unlinked) +
3043 (sizeof(xfs_agino_t) * bucket);
3044 xfs_trans_log_buf(tp, agibp, offset,
3045 (offset + sizeof(xfs_agino_t) - 1));
3047 error = xfs_trans_commit(tp, 0);
3053 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3055 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
3060 xlog_recover_process_one_iunlink(
3061 struct xfs_mount *mp,
3062 xfs_agnumber_t agno,
3066 struct xfs_buf *ibp;
3067 struct xfs_dinode *dip;
3068 struct xfs_inode *ip;
3072 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3073 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
3078 * Get the on disk inode to find the next inode in the bucket.
3080 error = xfs_itobp(mp, NULL, ip, &dip, &ibp, XBF_LOCK);
3084 ASSERT(ip->i_d.di_nlink == 0);
3085 ASSERT(ip->i_d.di_mode != 0);
3087 /* setup for the next pass */
3088 agino = be32_to_cpu(dip->di_next_unlinked);
3092 * Prevent any DMAPI event from being sent when the reference on
3093 * the inode is dropped.
3095 ip->i_d.di_dmevmask = 0;
3104 * We can't read in the inode this bucket points to, or this inode
3105 * is messed up. Just ditch this bucket of inodes. We will lose
3106 * some inodes and space, but at least we won't hang.
3108 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3109 * clear the inode pointer in the bucket.
3111 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3116 * xlog_iunlink_recover
3118 * This is called during recovery to process any inodes which
3119 * we unlinked but not freed when the system crashed. These
3120 * inodes will be on the lists in the AGI blocks. What we do
3121 * here is scan all the AGIs and fully truncate and free any
3122 * inodes found on the lists. Each inode is removed from the
3123 * lists when it has been fully truncated and is freed. The
3124 * freeing of the inode and its removal from the list must be
3128 xlog_recover_process_iunlinks(
3132 xfs_agnumber_t agno;
3143 * Prevent any DMAPI event from being sent while in this function.
3145 mp_dmevmask = mp->m_dmevmask;
3148 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3150 * Find the agi for this ag.
3152 error = xfs_read_agi(mp, NULL, agno, &agibp);
3155 * AGI is b0rked. Don't process it.
3157 * We should probably mark the filesystem as corrupt
3158 * after we've recovered all the ag's we can....
3162 agi = XFS_BUF_TO_AGI(agibp);
3164 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3165 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3166 while (agino != NULLAGINO) {
3168 * Release the agi buffer so that it can
3169 * be acquired in the normal course of the
3170 * transaction to truncate and free the inode.
3172 xfs_buf_relse(agibp);
3174 agino = xlog_recover_process_one_iunlink(mp,
3175 agno, agino, bucket);
3178 * Reacquire the agibuffer and continue around
3179 * the loop. This should never fail as we know
3180 * the buffer was good earlier on.
3182 error = xfs_read_agi(mp, NULL, agno, &agibp);
3184 agi = XFS_BUF_TO_AGI(agibp);
3189 * Release the buffer for the current agi so we can
3190 * go on to the next one.
3192 xfs_buf_relse(agibp);
3195 mp->m_dmevmask = mp_dmevmask;
3201 xlog_pack_data_checksum(
3203 xlog_in_core_t *iclog,
3210 up = (__be32 *)iclog->ic_datap;
3211 /* divide length by 4 to get # words */
3212 for (i = 0; i < (size >> 2); i++) {
3213 chksum ^= be32_to_cpu(*up);
3216 iclog->ic_header.h_chksum = cpu_to_be32(chksum);
3219 #define xlog_pack_data_checksum(log, iclog, size)
3223 * Stamp cycle number in every block
3228 xlog_in_core_t *iclog,
3232 int size = iclog->ic_offset + roundoff;
3236 xlog_pack_data_checksum(log, iclog, size);
3238 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3240 dp = iclog->ic_datap;
3241 for (i = 0; i < BTOBB(size) &&
3242 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3243 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
3244 *(__be32 *)dp = cycle_lsn;
3248 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3249 xlog_in_core_2_t *xhdr = iclog->ic_data;
3251 for ( ; i < BTOBB(size); i++) {
3252 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3253 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3254 xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
3255 *(__be32 *)dp = cycle_lsn;
3259 for (i = 1; i < log->l_iclog_heads; i++) {
3260 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3267 xlog_rec_header_t *rhead,
3273 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
3274 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3275 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
3279 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3280 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
3281 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
3282 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3283 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3284 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
3291 xlog_valid_rec_header(
3293 xlog_rec_header_t *rhead,
3298 if (unlikely(be32_to_cpu(rhead->h_magicno) != XLOG_HEADER_MAGIC_NUM)) {
3299 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3300 XFS_ERRLEVEL_LOW, log->l_mp);
3301 return XFS_ERROR(EFSCORRUPTED);
3304 (!rhead->h_version ||
3305 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
3306 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
3307 __func__, be32_to_cpu(rhead->h_version));
3308 return XFS_ERROR(EIO);
3311 /* LR body must have data or it wouldn't have been written */
3312 hlen = be32_to_cpu(rhead->h_len);
3313 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3314 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3315 XFS_ERRLEVEL_LOW, log->l_mp);
3316 return XFS_ERROR(EFSCORRUPTED);
3318 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3319 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3320 XFS_ERRLEVEL_LOW, log->l_mp);
3321 return XFS_ERROR(EFSCORRUPTED);
3327 * Read the log from tail to head and process the log records found.
3328 * Handle the two cases where the tail and head are in the same cycle
3329 * and where the active portion of the log wraps around the end of
3330 * the physical log separately. The pass parameter is passed through
3331 * to the routines called to process the data and is not looked at
3335 xlog_do_recovery_pass(
3337 xfs_daddr_t head_blk,
3338 xfs_daddr_t tail_blk,
3341 xlog_rec_header_t *rhead;
3344 xfs_buf_t *hbp, *dbp;
3345 int error = 0, h_size;
3346 int bblks, split_bblks;
3347 int hblks, split_hblks, wrapped_hblks;
3348 struct hlist_head rhash[XLOG_RHASH_SIZE];
3350 ASSERT(head_blk != tail_blk);
3353 * Read the header of the tail block and get the iclog buffer size from
3354 * h_size. Use this to tell how many sectors make up the log header.
3356 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3358 * When using variable length iclogs, read first sector of
3359 * iclog header and extract the header size from it. Get a
3360 * new hbp that is the correct size.
3362 hbp = xlog_get_bp(log, 1);
3366 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
3370 rhead = (xlog_rec_header_t *)offset;
3371 error = xlog_valid_rec_header(log, rhead, tail_blk);
3374 h_size = be32_to_cpu(rhead->h_size);
3375 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
3376 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3377 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3378 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3381 hbp = xlog_get_bp(log, hblks);
3386 ASSERT(log->l_sectBBsize == 1);
3388 hbp = xlog_get_bp(log, 1);
3389 h_size = XLOG_BIG_RECORD_BSIZE;
3394 dbp = xlog_get_bp(log, BTOBB(h_size));
3400 memset(rhash, 0, sizeof(rhash));
3401 if (tail_blk <= head_blk) {
3402 for (blk_no = tail_blk; blk_no < head_blk; ) {
3403 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3407 rhead = (xlog_rec_header_t *)offset;
3408 error = xlog_valid_rec_header(log, rhead, blk_no);
3412 /* blocks in data section */
3413 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3414 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
3419 xlog_unpack_data(rhead, offset, log);
3420 if ((error = xlog_recover_process_data(log,
3421 rhash, rhead, offset, pass)))
3423 blk_no += bblks + hblks;
3427 * Perform recovery around the end of the physical log.
3428 * When the head is not on the same cycle number as the tail,
3429 * we can't do a sequential recovery as above.
3432 while (blk_no < log->l_logBBsize) {
3434 * Check for header wrapping around physical end-of-log
3436 offset = XFS_BUF_PTR(hbp);
3439 if (blk_no + hblks <= log->l_logBBsize) {
3440 /* Read header in one read */
3441 error = xlog_bread(log, blk_no, hblks, hbp,
3446 /* This LR is split across physical log end */
3447 if (blk_no != log->l_logBBsize) {
3448 /* some data before physical log end */
3449 ASSERT(blk_no <= INT_MAX);
3450 split_hblks = log->l_logBBsize - (int)blk_no;
3451 ASSERT(split_hblks > 0);
3452 error = xlog_bread(log, blk_no,
3460 * Note: this black magic still works with
3461 * large sector sizes (non-512) only because:
3462 * - we increased the buffer size originally
3463 * by 1 sector giving us enough extra space
3464 * for the second read;
3465 * - the log start is guaranteed to be sector
3467 * - we read the log end (LR header start)
3468 * _first_, then the log start (LR header end)
3469 * - order is important.
3471 wrapped_hblks = hblks - split_hblks;
3472 error = xlog_bread_offset(log, 0,
3474 offset + BBTOB(split_hblks));
3478 rhead = (xlog_rec_header_t *)offset;
3479 error = xlog_valid_rec_header(log, rhead,
3480 split_hblks ? blk_no : 0);
3484 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3487 /* Read in data for log record */
3488 if (blk_no + bblks <= log->l_logBBsize) {
3489 error = xlog_bread(log, blk_no, bblks, dbp,
3494 /* This log record is split across the
3495 * physical end of log */
3496 offset = XFS_BUF_PTR(dbp);
3498 if (blk_no != log->l_logBBsize) {
3499 /* some data is before the physical
3501 ASSERT(!wrapped_hblks);
3502 ASSERT(blk_no <= INT_MAX);
3504 log->l_logBBsize - (int)blk_no;
3505 ASSERT(split_bblks > 0);
3506 error = xlog_bread(log, blk_no,
3514 * Note: this black magic still works with
3515 * large sector sizes (non-512) only because:
3516 * - we increased the buffer size originally
3517 * by 1 sector giving us enough extra space
3518 * for the second read;
3519 * - the log start is guaranteed to be sector
3521 * - we read the log end (LR header start)
3522 * _first_, then the log start (LR header end)
3523 * - order is important.
3525 error = xlog_bread_offset(log, 0,
3526 bblks - split_bblks, hbp,
3527 offset + BBTOB(split_bblks));
3531 xlog_unpack_data(rhead, offset, log);
3532 if ((error = xlog_recover_process_data(log, rhash,
3533 rhead, offset, pass)))
3538 ASSERT(blk_no >= log->l_logBBsize);
3539 blk_no -= log->l_logBBsize;
3541 /* read first part of physical log */
3542 while (blk_no < head_blk) {
3543 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3547 rhead = (xlog_rec_header_t *)offset;
3548 error = xlog_valid_rec_header(log, rhead, blk_no);
3552 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3553 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
3558 xlog_unpack_data(rhead, offset, log);
3559 if ((error = xlog_recover_process_data(log, rhash,
3560 rhead, offset, pass)))
3562 blk_no += bblks + hblks;
3574 * Do the recovery of the log. We actually do this in two phases.
3575 * The two passes are necessary in order to implement the function
3576 * of cancelling a record written into the log. The first pass
3577 * determines those things which have been cancelled, and the
3578 * second pass replays log items normally except for those which
3579 * have been cancelled. The handling of the replay and cancellations
3580 * takes place in the log item type specific routines.
3582 * The table of items which have cancel records in the log is allocated
3583 * and freed at this level, since only here do we know when all of
3584 * the log recovery has been completed.
3587 xlog_do_log_recovery(
3589 xfs_daddr_t head_blk,
3590 xfs_daddr_t tail_blk)
3594 ASSERT(head_blk != tail_blk);
3597 * First do a pass to find all of the cancelled buf log items.
3598 * Store them in the buf_cancel_table for use in the second pass.
3600 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
3601 sizeof(struct list_head),
3603 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3604 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
3606 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3607 XLOG_RECOVER_PASS1);
3609 kmem_free(log->l_buf_cancel_table);
3610 log->l_buf_cancel_table = NULL;
3614 * Then do a second pass to actually recover the items in the log.
3615 * When it is complete free the table of buf cancel items.
3617 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3618 XLOG_RECOVER_PASS2);
3623 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3624 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
3628 kmem_free(log->l_buf_cancel_table);
3629 log->l_buf_cancel_table = NULL;
3635 * Do the actual recovery
3640 xfs_daddr_t head_blk,
3641 xfs_daddr_t tail_blk)
3648 * First replay the images in the log.
3650 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3655 XFS_bflush(log->l_mp->m_ddev_targp);
3658 * If IO errors happened during recovery, bail out.
3660 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3665 * We now update the tail_lsn since much of the recovery has completed
3666 * and there may be space available to use. If there were no extent
3667 * or iunlinks, we can free up the entire log and set the tail_lsn to
3668 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3669 * lsn of the last known good LR on disk. If there are extent frees
3670 * or iunlinks they will have some entries in the AIL; so we look at
3671 * the AIL to determine how to set the tail_lsn.
3673 xlog_assign_tail_lsn(log->l_mp);
3676 * Now that we've finished replaying all buffer and inode
3677 * updates, re-read in the superblock.
3679 bp = xfs_getsb(log->l_mp, 0);
3681 ASSERT(!(XFS_BUF_ISWRITE(bp)));
3682 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
3684 XFS_BUF_UNASYNC(bp);
3685 xfsbdstrat(log->l_mp, bp);
3686 error = xfs_buf_iowait(bp);
3688 xfs_ioerror_alert("xlog_do_recover",
3689 log->l_mp, bp, XFS_BUF_ADDR(bp));
3695 /* Convert superblock from on-disk format */
3696 sbp = &log->l_mp->m_sb;
3697 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
3698 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
3699 ASSERT(xfs_sb_good_version(sbp));
3702 /* We've re-read the superblock so re-initialize per-cpu counters */
3703 xfs_icsb_reinit_counters(log->l_mp);
3705 xlog_recover_check_summary(log);
3707 /* Normal transactions can now occur */
3708 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3713 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3715 * Return error or zero.
3721 xfs_daddr_t head_blk, tail_blk;
3724 /* find the tail of the log */
3725 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
3728 if (tail_blk != head_blk) {
3729 /* There used to be a comment here:
3731 * disallow recovery on read-only mounts. note -- mount
3732 * checks for ENOSPC and turns it into an intelligent
3734 * ...but this is no longer true. Now, unless you specify
3735 * NORECOVERY (in which case this function would never be
3736 * called), we just go ahead and recover. We do this all
3737 * under the vfs layer, so we can get away with it unless
3738 * the device itself is read-only, in which case we fail.
3740 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
3744 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
3745 log->l_mp->m_logname ? log->l_mp->m_logname
3748 error = xlog_do_recover(log, head_blk, tail_blk);
3749 log->l_flags |= XLOG_RECOVERY_NEEDED;
3755 * In the first part of recovery we replay inodes and buffers and build
3756 * up the list of extent free items which need to be processed. Here
3757 * we process the extent free items and clean up the on disk unlinked
3758 * inode lists. This is separated from the first part of recovery so
3759 * that the root and real-time bitmap inodes can be read in from disk in
3760 * between the two stages. This is necessary so that we can free space
3761 * in the real-time portion of the file system.
3764 xlog_recover_finish(
3768 * Now we're ready to do the transactions needed for the
3769 * rest of recovery. Start with completing all the extent
3770 * free intent records and then process the unlinked inode
3771 * lists. At this point, we essentially run in normal mode
3772 * except that we're still performing recovery actions
3773 * rather than accepting new requests.
3775 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
3777 error = xlog_recover_process_efis(log);
3779 xfs_alert(log->l_mp, "Failed to recover EFIs");
3783 * Sync the log to get all the EFIs out of the AIL.
3784 * This isn't absolutely necessary, but it helps in
3785 * case the unlink transactions would have problems
3786 * pushing the EFIs out of the way.
3788 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
3790 xlog_recover_process_iunlinks(log);
3792 xlog_recover_check_summary(log);
3794 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
3795 log->l_mp->m_logname ? log->l_mp->m_logname
3797 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
3799 xfs_info(log->l_mp, "Ending clean mount");
3807 * Read all of the agf and agi counters and check that they
3808 * are consistent with the superblock counters.
3811 xlog_recover_check_summary(
3818 xfs_agnumber_t agno;
3819 __uint64_t freeblks;
3829 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3830 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
3832 xfs_alert(mp, "%s agf read failed agno %d error %d",
3833 __func__, agno, error);
3835 agfp = XFS_BUF_TO_AGF(agfbp);
3836 freeblks += be32_to_cpu(agfp->agf_freeblks) +
3837 be32_to_cpu(agfp->agf_flcount);
3838 xfs_buf_relse(agfbp);
3841 error = xfs_read_agi(mp, NULL, agno, &agibp);
3843 xfs_alert(mp, "%s agi read failed agno %d error %d",
3844 __func__, agno, error);
3846 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
3848 itotal += be32_to_cpu(agi->agi_count);
3849 ifree += be32_to_cpu(agi->agi_freecount);
3850 xfs_buf_relse(agibp);
git.openpandora.org / pandora-kernel.git / blob