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.
96 if (!xlog_buf_bbcount_valid(log, nbblks)) {
97 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
99 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
104 * We do log I/O in units of log sectors (a power-of-2
105 * multiple of the basic block size), so we round up the
106 * requested size to accommodate the basic blocks required
107 * for complete log sectors.
109 * In addition, the buffer may be used for a non-sector-
110 * aligned block offset, in which case an I/O of the
111 * requested size could extend beyond the end of the
112 * buffer. If the requested size is only 1 basic block it
113 * will never straddle a sector boundary, so this won't be
114 * an issue. Nor will this be a problem if the log I/O is
115 * done in basic blocks (sector size 1). But otherwise we
116 * extend the buffer by one extra log sector to ensure
117 * there's space to accommodate this possibility.
119 if (nbblks > 1 && log->l_sectBBsize > 1)
120 nbblks += log->l_sectBBsize;
121 nbblks = round_up(nbblks, log->l_sectBBsize);
123 bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, BBTOB(nbblks), 0);
137 * Return the address of the start of the given block number's data
138 * in a log buffer. The buffer covers a log sector-aligned region.
147 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
149 ASSERT(BBTOB(offset + nbblks) <= XFS_BUF_SIZE(bp));
150 return bp->b_addr + BBTOB(offset);
155 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
166 if (!xlog_buf_bbcount_valid(log, nbblks)) {
167 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
169 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
173 blk_no = round_down(blk_no, log->l_sectBBsize);
174 nbblks = round_up(nbblks, log->l_sectBBsize);
177 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
179 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
181 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
184 xfsbdstrat(log->l_mp, bp);
185 error = xfs_buf_iowait(bp);
187 xfs_buf_ioerror_alert(bp, __func__);
201 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
205 *offset = xlog_align(log, blk_no, nbblks, bp);
210 * Read at an offset into the buffer. Returns with the buffer in it's original
211 * state regardless of the result of the read.
216 xfs_daddr_t blk_no, /* block to read from */
217 int nbblks, /* blocks to read */
221 xfs_caddr_t orig_offset = bp->b_addr;
222 int orig_len = bp->b_buffer_length;
225 error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks));
229 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
231 /* must reset buffer pointer even on error */
232 error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len);
239 * Write out the buffer at the given block for the given number of blocks.
240 * The buffer is kept locked across the write and is returned locked.
241 * This can only be used for synchronous log writes.
252 if (!xlog_buf_bbcount_valid(log, nbblks)) {
253 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
255 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
259 blk_no = round_down(blk_no, log->l_sectBBsize);
260 nbblks = round_up(nbblks, log->l_sectBBsize);
263 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
265 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
266 XFS_BUF_ZEROFLAGS(bp);
269 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
272 error = xfs_bwrite(bp);
274 xfs_buf_ioerror_alert(bp, __func__);
281 * dump debug superblock and log record information
284 xlog_header_check_dump(
286 xlog_rec_header_t *head)
288 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d\n",
289 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
290 xfs_debug(mp, " log : uuid = %pU, fmt = %d\n",
291 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
294 #define xlog_header_check_dump(mp, head)
298 * check log record header for recovery
301 xlog_header_check_recover(
303 xlog_rec_header_t *head)
305 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
308 * IRIX doesn't write the h_fmt field and leaves it zeroed
309 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
310 * a dirty log created in IRIX.
312 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
314 "dirty log written in incompatible format - can't recover");
315 xlog_header_check_dump(mp, head);
316 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
317 XFS_ERRLEVEL_HIGH, mp);
318 return XFS_ERROR(EFSCORRUPTED);
319 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
321 "dirty log entry has mismatched uuid - can't recover");
322 xlog_header_check_dump(mp, head);
323 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
324 XFS_ERRLEVEL_HIGH, mp);
325 return XFS_ERROR(EFSCORRUPTED);
331 * read the head block of the log and check the header
334 xlog_header_check_mount(
336 xlog_rec_header_t *head)
338 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
340 if (uuid_is_nil(&head->h_fs_uuid)) {
342 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
343 * h_fs_uuid is nil, we assume this log was last mounted
344 * by IRIX and continue.
346 xfs_warn(mp, "nil uuid in log - IRIX style log");
347 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
348 xfs_warn(mp, "log has mismatched uuid - can't recover");
349 xlog_header_check_dump(mp, head);
350 XFS_ERROR_REPORT("xlog_header_check_mount",
351 XFS_ERRLEVEL_HIGH, mp);
352 return XFS_ERROR(EFSCORRUPTED);
363 * We're not going to bother about retrying
364 * this during recovery. One strike!
366 xfs_buf_ioerror_alert(bp, __func__);
367 xfs_force_shutdown(bp->b_target->bt_mount,
368 SHUTDOWN_META_IO_ERROR);
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 (bufblks > log->l_logBBsize)
447 while (!(bp = xlog_get_bp(log, bufblks))) {
449 if (bufblks < log->l_sectBBsize)
453 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
456 bcount = min(bufblks, (start_blk + nbblks - i));
458 error = xlog_bread(log, i, bcount, bp, &buf);
462 for (j = 0; j < bcount; j++) {
463 cycle = xlog_get_cycle(buf);
464 if (cycle == stop_on_cycle_no) {
481 * Potentially backup over partial log record write.
483 * In the typical case, last_blk is the number of the block directly after
484 * a good log record. Therefore, we subtract one to get the block number
485 * of the last block in the given buffer. extra_bblks contains the number
486 * of blocks we would have read on a previous read. This happens when the
487 * last log record is split over the end of the physical log.
489 * extra_bblks is the number of blocks potentially verified on a previous
490 * call to this routine.
493 xlog_find_verify_log_record(
495 xfs_daddr_t start_blk,
496 xfs_daddr_t *last_blk,
501 xfs_caddr_t offset = NULL;
502 xlog_rec_header_t *head = NULL;
505 int num_blks = *last_blk - start_blk;
508 ASSERT(start_blk != 0 || *last_blk != start_blk);
510 if (!(bp = xlog_get_bp(log, num_blks))) {
511 if (!(bp = xlog_get_bp(log, 1)))
515 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
518 offset += ((num_blks - 1) << BBSHIFT);
521 for (i = (*last_blk) - 1; i >= 0; i--) {
523 /* valid log record not found */
525 "Log inconsistent (didn't find previous header)");
527 error = XFS_ERROR(EIO);
532 error = xlog_bread(log, i, 1, bp, &offset);
537 head = (xlog_rec_header_t *)offset;
539 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
547 * We hit the beginning of the physical log & still no header. Return
548 * to caller. If caller can handle a return of -1, then this routine
549 * will be called again for the end of the physical log.
557 * We have the final block of the good log (the first block
558 * of the log record _before_ the head. So we check the uuid.
560 if ((error = xlog_header_check_mount(log->l_mp, head)))
564 * We may have found a log record header before we expected one.
565 * last_blk will be the 1st block # with a given cycle #. We may end
566 * up reading an entire log record. In this case, we don't want to
567 * reset last_blk. Only when last_blk points in the middle of a log
568 * record do we update last_blk.
570 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
571 uint h_size = be32_to_cpu(head->h_size);
573 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
574 if (h_size % XLOG_HEADER_CYCLE_SIZE)
580 if (*last_blk - i + extra_bblks !=
581 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
590 * Head is defined to be the point of the log where the next log write
591 * write could go. This means that incomplete LR writes at the end are
592 * eliminated when calculating the head. We aren't guaranteed that previous
593 * LR have complete transactions. We only know that a cycle number of
594 * current cycle number -1 won't be present in the log if we start writing
595 * from our current block number.
597 * last_blk contains the block number of the first block with a given
600 * Return: zero if normal, non-zero if error.
605 xfs_daddr_t *return_head_blk)
609 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
611 uint first_half_cycle, last_half_cycle;
613 int error, log_bbnum = log->l_logBBsize;
615 /* Is the end of the log device zeroed? */
616 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
617 *return_head_blk = first_blk;
619 /* Is the whole lot zeroed? */
621 /* Linux XFS shouldn't generate totally zeroed logs -
622 * mkfs etc write a dummy unmount record to a fresh
623 * log so we can store the uuid in there
625 xfs_warn(log->l_mp, "totally zeroed log");
630 xfs_warn(log->l_mp, "empty log check failed");
634 first_blk = 0; /* get cycle # of 1st block */
635 bp = xlog_get_bp(log, 1);
639 error = xlog_bread(log, 0, 1, bp, &offset);
643 first_half_cycle = xlog_get_cycle(offset);
645 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
646 error = xlog_bread(log, last_blk, 1, bp, &offset);
650 last_half_cycle = xlog_get_cycle(offset);
651 ASSERT(last_half_cycle != 0);
654 * If the 1st half cycle number is equal to the last half cycle number,
655 * then the entire log is stamped with the same cycle number. In this
656 * case, head_blk can't be set to zero (which makes sense). The below
657 * math doesn't work out properly with head_blk equal to zero. Instead,
658 * we set it to log_bbnum which is an invalid block number, but this
659 * value makes the math correct. If head_blk doesn't changed through
660 * all the tests below, *head_blk is set to zero at the very end rather
661 * than log_bbnum. In a sense, log_bbnum and zero are the same block
662 * in a circular file.
664 if (first_half_cycle == last_half_cycle) {
666 * In this case we believe that the entire log should have
667 * cycle number last_half_cycle. We need to scan backwards
668 * from the end verifying that there are no holes still
669 * containing last_half_cycle - 1. If we find such a hole,
670 * then the start of that hole will be the new head. The
671 * simple case looks like
672 * x | x ... | x - 1 | x
673 * Another case that fits this picture would be
674 * x | x + 1 | x ... | x
675 * In this case the head really is somewhere at the end of the
676 * log, as one of the latest writes at the beginning was
679 * x | x + 1 | x ... | x - 1 | x
680 * This is really the combination of the above two cases, and
681 * the head has to end up at the start of the x-1 hole at the
684 * In the 256k log case, we will read from the beginning to the
685 * end of the log and search for cycle numbers equal to x-1.
686 * We don't worry about the x+1 blocks that we encounter,
687 * because we know that they cannot be the head since the log
690 head_blk = log_bbnum;
691 stop_on_cycle = last_half_cycle - 1;
694 * In this case we want to find the first block with cycle
695 * number matching last_half_cycle. We expect the log to be
697 * x + 1 ... | x ... | x
698 * The first block with cycle number x (last_half_cycle) will
699 * be where the new head belongs. First we do a binary search
700 * for the first occurrence of last_half_cycle. The binary
701 * search may not be totally accurate, so then we scan back
702 * from there looking for occurrences of last_half_cycle before
703 * us. If that backwards scan wraps around the beginning of
704 * the log, then we look for occurrences of last_half_cycle - 1
705 * at the end of the log. The cases we're looking for look
707 * v binary search stopped here
708 * x + 1 ... | x | x + 1 | x ... | x
709 * ^ but we want to locate this spot
711 * <---------> less than scan distance
712 * x + 1 ... | x ... | x - 1 | x
713 * ^ we want to locate this spot
715 stop_on_cycle = last_half_cycle;
716 if ((error = xlog_find_cycle_start(log, bp, first_blk,
717 &head_blk, last_half_cycle)))
722 * Now validate the answer. Scan back some number of maximum possible
723 * blocks and make sure each one has the expected cycle number. The
724 * maximum is determined by the total possible amount of buffering
725 * in the in-core log. The following number can be made tighter if
726 * we actually look at the block size of the filesystem.
728 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
729 if (head_blk >= num_scan_bblks) {
731 * We are guaranteed that the entire check can be performed
734 start_blk = head_blk - num_scan_bblks;
735 if ((error = xlog_find_verify_cycle(log,
736 start_blk, num_scan_bblks,
737 stop_on_cycle, &new_blk)))
741 } else { /* need to read 2 parts of log */
743 * We are going to scan backwards in the log in two parts.
744 * First we scan the physical end of the log. In this part
745 * of the log, we are looking for blocks with cycle number
746 * last_half_cycle - 1.
747 * If we find one, then we know that the log starts there, as
748 * we've found a hole that didn't get written in going around
749 * the end of the physical log. The simple case for this is
750 * x + 1 ... | x ... | x - 1 | x
751 * <---------> less than scan distance
752 * If all of the blocks at the end of the log have cycle number
753 * last_half_cycle, then we check the blocks at the start of
754 * the log looking for occurrences of last_half_cycle. If we
755 * find one, then our current estimate for the location of the
756 * first occurrence of last_half_cycle is wrong and we move
757 * back to the hole we've found. This case looks like
758 * x + 1 ... | x | x + 1 | x ...
759 * ^ binary search stopped here
760 * Another case we need to handle that only occurs in 256k
762 * x + 1 ... | x ... | x+1 | x ...
763 * ^ binary search stops here
764 * In a 256k log, the scan at the end of the log will see the
765 * x + 1 blocks. We need to skip past those since that is
766 * certainly not the head of the log. By searching for
767 * last_half_cycle-1 we accomplish that.
769 ASSERT(head_blk <= INT_MAX &&
770 (xfs_daddr_t) num_scan_bblks >= head_blk);
771 start_blk = log_bbnum - (num_scan_bblks - head_blk);
772 if ((error = xlog_find_verify_cycle(log, start_blk,
773 num_scan_bblks - (int)head_blk,
774 (stop_on_cycle - 1), &new_blk)))
782 * Scan beginning of log now. The last part of the physical
783 * log is good. This scan needs to verify that it doesn't find
784 * the last_half_cycle.
787 ASSERT(head_blk <= INT_MAX);
788 if ((error = xlog_find_verify_cycle(log,
789 start_blk, (int)head_blk,
790 stop_on_cycle, &new_blk)))
798 * Now we need to make sure head_blk is not pointing to a block in
799 * the middle of a log record.
801 num_scan_bblks = XLOG_REC_SHIFT(log);
802 if (head_blk >= num_scan_bblks) {
803 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
805 /* start ptr at last block ptr before head_blk */
806 if ((error = xlog_find_verify_log_record(log, start_blk,
807 &head_blk, 0)) == -1) {
808 error = XFS_ERROR(EIO);
814 ASSERT(head_blk <= INT_MAX);
815 if ((error = xlog_find_verify_log_record(log, start_blk,
816 &head_blk, 0)) == -1) {
817 /* We hit the beginning of the log during our search */
818 start_blk = log_bbnum - (num_scan_bblks - head_blk);
820 ASSERT(start_blk <= INT_MAX &&
821 (xfs_daddr_t) log_bbnum-start_blk >= 0);
822 ASSERT(head_blk <= INT_MAX);
823 if ((error = xlog_find_verify_log_record(log,
825 (int)head_blk)) == -1) {
826 error = XFS_ERROR(EIO);
830 if (new_blk != log_bbnum)
837 if (head_blk == log_bbnum)
838 *return_head_blk = 0;
840 *return_head_blk = head_blk;
842 * When returning here, we have a good block number. Bad block
843 * means that during a previous crash, we didn't have a clean break
844 * from cycle number N to cycle number N-1. In this case, we need
845 * to find the first block with cycle number N-1.
853 xfs_warn(log->l_mp, "failed to find log head");
858 * Find the sync block number or the tail of the log.
860 * This will be the block number of the last record to have its
861 * associated buffers synced to disk. Every log record header has
862 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
863 * to get a sync block number. The only concern is to figure out which
864 * log record header to believe.
866 * The following algorithm uses the log record header with the largest
867 * lsn. The entire log record does not need to be valid. We only care
868 * that the header is valid.
870 * We could speed up search by using current head_blk buffer, but it is not
876 xfs_daddr_t *head_blk,
877 xfs_daddr_t *tail_blk)
879 xlog_rec_header_t *rhead;
880 xlog_op_header_t *op_head;
881 xfs_caddr_t offset = NULL;
884 xfs_daddr_t umount_data_blk;
885 xfs_daddr_t after_umount_blk;
892 * Find previous log record
894 if ((error = xlog_find_head(log, head_blk)))
897 bp = xlog_get_bp(log, 1);
900 if (*head_blk == 0) { /* special case */
901 error = xlog_bread(log, 0, 1, bp, &offset);
905 if (xlog_get_cycle(offset) == 0) {
907 /* leave all other log inited values alone */
913 * Search backwards looking for log record header block
915 ASSERT(*head_blk < INT_MAX);
916 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
917 error = xlog_bread(log, i, 1, bp, &offset);
921 if (*(__be32 *)offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
927 * If we haven't found the log record header block, start looking
928 * again from the end of the physical log. XXXmiken: There should be
929 * a check here to make sure we didn't search more than N blocks in
933 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
934 error = xlog_bread(log, i, 1, bp, &offset);
938 if (*(__be32 *)offset ==
939 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
946 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
948 return XFS_ERROR(EIO);
951 /* find blk_no of tail of log */
952 rhead = (xlog_rec_header_t *)offset;
953 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
956 * Reset log values according to the state of the log when we
957 * crashed. In the case where head_blk == 0, we bump curr_cycle
958 * one because the next write starts a new cycle rather than
959 * continuing the cycle of the last good log record. At this
960 * point we have guaranteed that all partial log records have been
961 * accounted for. Therefore, we know that the last good log record
962 * written was complete and ended exactly on the end boundary
963 * of the physical log.
965 log->l_prev_block = i;
966 log->l_curr_block = (int)*head_blk;
967 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
970 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
971 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
972 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
973 BBTOB(log->l_curr_block));
974 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
975 BBTOB(log->l_curr_block));
978 * Look for unmount record. If we find it, then we know there
979 * was a clean unmount. Since 'i' could be the last block in
980 * the physical log, we convert to a log block before comparing
983 * Save the current tail lsn to use to pass to
984 * xlog_clear_stale_blocks() below. We won't want to clear the
985 * unmount record if there is one, so we pass the lsn of the
986 * unmount record rather than the block after it.
988 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
989 int h_size = be32_to_cpu(rhead->h_size);
990 int h_version = be32_to_cpu(rhead->h_version);
992 if ((h_version & XLOG_VERSION_2) &&
993 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
994 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
995 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1003 after_umount_blk = (i + hblks + (int)
1004 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
1005 tail_lsn = atomic64_read(&log->l_tail_lsn);
1006 if (*head_blk == after_umount_blk &&
1007 be32_to_cpu(rhead->h_num_logops) == 1) {
1008 umount_data_blk = (i + hblks) % log->l_logBBsize;
1009 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1013 op_head = (xlog_op_header_t *)offset;
1014 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1016 * Set tail and last sync so that newly written
1017 * log records will point recovery to after the
1018 * current unmount record.
1020 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1021 log->l_curr_cycle, after_umount_blk);
1022 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1023 log->l_curr_cycle, after_umount_blk);
1024 *tail_blk = after_umount_blk;
1027 * Note that the unmount was clean. If the unmount
1028 * was not clean, we need to know this to rebuild the
1029 * superblock counters from the perag headers if we
1030 * have a filesystem using non-persistent counters.
1032 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1037 * Make sure that there are no blocks in front of the head
1038 * with the same cycle number as the head. This can happen
1039 * because we allow multiple outstanding log writes concurrently,
1040 * and the later writes might make it out before earlier ones.
1042 * We use the lsn from before modifying it so that we'll never
1043 * overwrite the unmount record after a clean unmount.
1045 * Do this only if we are going to recover the filesystem
1047 * NOTE: This used to say "if (!readonly)"
1048 * However on Linux, we can & do recover a read-only filesystem.
1049 * We only skip recovery if NORECOVERY is specified on mount,
1050 * in which case we would not be here.
1052 * But... if the -device- itself is readonly, just skip this.
1053 * We can't recover this device anyway, so it won't matter.
1055 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
1056 error = xlog_clear_stale_blocks(log, tail_lsn);
1062 xfs_warn(log->l_mp, "failed to locate log tail");
1067 * Is the log zeroed at all?
1069 * The last binary search should be changed to perform an X block read
1070 * once X becomes small enough. You can then search linearly through
1071 * the X blocks. This will cut down on the number of reads we need to do.
1073 * If the log is partially zeroed, this routine will pass back the blkno
1074 * of the first block with cycle number 0. It won't have a complete LR
1078 * 0 => the log is completely written to
1079 * -1 => use *blk_no as the first block of the log
1080 * >0 => error has occurred
1085 xfs_daddr_t *blk_no)
1089 uint first_cycle, last_cycle;
1090 xfs_daddr_t new_blk, last_blk, start_blk;
1091 xfs_daddr_t num_scan_bblks;
1092 int error, log_bbnum = log->l_logBBsize;
1096 /* check totally zeroed log */
1097 bp = xlog_get_bp(log, 1);
1100 error = xlog_bread(log, 0, 1, bp, &offset);
1104 first_cycle = xlog_get_cycle(offset);
1105 if (first_cycle == 0) { /* completely zeroed log */
1111 /* check partially zeroed log */
1112 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1116 last_cycle = xlog_get_cycle(offset);
1117 if (last_cycle != 0) { /* log completely written to */
1120 } else if (first_cycle != 1) {
1122 * If the cycle of the last block is zero, the cycle of
1123 * the first block must be 1. If it's not, maybe we're
1124 * not looking at a log... Bail out.
1127 "Log inconsistent or not a log (last==0, first!=1)");
1128 return XFS_ERROR(EINVAL);
1131 /* we have a partially zeroed log */
1132 last_blk = log_bbnum-1;
1133 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1137 * Validate the answer. Because there is no way to guarantee that
1138 * the entire log is made up of log records which are the same size,
1139 * we scan over the defined maximum blocks. At this point, the maximum
1140 * is not chosen to mean anything special. XXXmiken
1142 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1143 ASSERT(num_scan_bblks <= INT_MAX);
1145 if (last_blk < num_scan_bblks)
1146 num_scan_bblks = last_blk;
1147 start_blk = last_blk - num_scan_bblks;
1150 * We search for any instances of cycle number 0 that occur before
1151 * our current estimate of the head. What we're trying to detect is
1152 * 1 ... | 0 | 1 | 0...
1153 * ^ binary search ends here
1155 if ((error = xlog_find_verify_cycle(log, start_blk,
1156 (int)num_scan_bblks, 0, &new_blk)))
1162 * Potentially backup over partial log record write. We don't need
1163 * to search the end of the log because we know it is zero.
1165 if ((error = xlog_find_verify_log_record(log, start_blk,
1166 &last_blk, 0)) == -1) {
1167 error = XFS_ERROR(EIO);
1181 * These are simple subroutines used by xlog_clear_stale_blocks() below
1182 * to initialize a buffer full of empty log record headers and write
1183 * them into the log.
1194 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1196 memset(buf, 0, BBSIZE);
1197 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1198 recp->h_cycle = cpu_to_be32(cycle);
1199 recp->h_version = cpu_to_be32(
1200 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1201 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1202 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1203 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1204 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1208 xlog_write_log_records(
1219 int sectbb = log->l_sectBBsize;
1220 int end_block = start_block + blocks;
1226 * Greedily allocate a buffer big enough to handle the full
1227 * range of basic blocks to be written. If that fails, try
1228 * a smaller size. We need to be able to write at least a
1229 * log sector, or we're out of luck.
1231 bufblks = 1 << ffs(blocks);
1232 while (bufblks > log->l_logBBsize)
1234 while (!(bp = xlog_get_bp(log, bufblks))) {
1236 if (bufblks < sectbb)
1240 /* We may need to do a read at the start to fill in part of
1241 * the buffer in the starting sector not covered by the first
1244 balign = round_down(start_block, sectbb);
1245 if (balign != start_block) {
1246 error = xlog_bread_noalign(log, start_block, 1, bp);
1250 j = start_block - balign;
1253 for (i = start_block; i < end_block; i += bufblks) {
1254 int bcount, endcount;
1256 bcount = min(bufblks, end_block - start_block);
1257 endcount = bcount - j;
1259 /* We may need to do a read at the end to fill in part of
1260 * the buffer in the final sector not covered by the write.
1261 * If this is the same sector as the above read, skip it.
1263 ealign = round_down(end_block, sectbb);
1264 if (j == 0 && (start_block + endcount > ealign)) {
1265 offset = bp->b_addr + BBTOB(ealign - start_block);
1266 error = xlog_bread_offset(log, ealign, sectbb,
1273 offset = xlog_align(log, start_block, endcount, bp);
1274 for (; j < endcount; j++) {
1275 xlog_add_record(log, offset, cycle, i+j,
1276 tail_cycle, tail_block);
1279 error = xlog_bwrite(log, start_block, endcount, bp);
1282 start_block += endcount;
1292 * This routine is called to blow away any incomplete log writes out
1293 * in front of the log head. We do this so that we won't become confused
1294 * if we come up, write only a little bit more, and then crash again.
1295 * If we leave the partial log records out there, this situation could
1296 * cause us to think those partial writes are valid blocks since they
1297 * have the current cycle number. We get rid of them by overwriting them
1298 * with empty log records with the old cycle number rather than the
1301 * The tail lsn is passed in rather than taken from
1302 * the log so that we will not write over the unmount record after a
1303 * clean unmount in a 512 block log. Doing so would leave the log without
1304 * any valid log records in it until a new one was written. If we crashed
1305 * during that time we would not be able to recover.
1308 xlog_clear_stale_blocks(
1312 int tail_cycle, head_cycle;
1313 int tail_block, head_block;
1314 int tail_distance, max_distance;
1318 tail_cycle = CYCLE_LSN(tail_lsn);
1319 tail_block = BLOCK_LSN(tail_lsn);
1320 head_cycle = log->l_curr_cycle;
1321 head_block = log->l_curr_block;
1324 * Figure out the distance between the new head of the log
1325 * and the tail. We want to write over any blocks beyond the
1326 * head that we may have written just before the crash, but
1327 * we don't want to overwrite the tail of the log.
1329 if (head_cycle == tail_cycle) {
1331 * The tail is behind the head in the physical log,
1332 * so the distance from the head to the tail is the
1333 * distance from the head to the end of the log plus
1334 * the distance from the beginning of the log to the
1337 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1338 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1339 XFS_ERRLEVEL_LOW, log->l_mp);
1340 return XFS_ERROR(EFSCORRUPTED);
1342 tail_distance = tail_block + (log->l_logBBsize - head_block);
1345 * The head is behind the tail in the physical log,
1346 * so the distance from the head to the tail is just
1347 * the tail block minus the head block.
1349 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1350 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1351 XFS_ERRLEVEL_LOW, log->l_mp);
1352 return XFS_ERROR(EFSCORRUPTED);
1354 tail_distance = tail_block - head_block;
1358 * If the head is right up against the tail, we can't clear
1361 if (tail_distance <= 0) {
1362 ASSERT(tail_distance == 0);
1366 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1368 * Take the smaller of the maximum amount of outstanding I/O
1369 * we could have and the distance to the tail to clear out.
1370 * We take the smaller so that we don't overwrite the tail and
1371 * we don't waste all day writing from the head to the tail
1374 max_distance = MIN(max_distance, tail_distance);
1376 if ((head_block + max_distance) <= log->l_logBBsize) {
1378 * We can stomp all the blocks we need to without
1379 * wrapping around the end of the log. Just do it
1380 * in a single write. Use the cycle number of the
1381 * current cycle minus one so that the log will look like:
1384 error = xlog_write_log_records(log, (head_cycle - 1),
1385 head_block, max_distance, tail_cycle,
1391 * We need to wrap around the end of the physical log in
1392 * order to clear all the blocks. Do it in two separate
1393 * I/Os. The first write should be from the head to the
1394 * end of the physical log, and it should use the current
1395 * cycle number minus one just like above.
1397 distance = log->l_logBBsize - head_block;
1398 error = xlog_write_log_records(log, (head_cycle - 1),
1399 head_block, distance, tail_cycle,
1406 * Now write the blocks at the start of the physical log.
1407 * This writes the remainder of the blocks we want to clear.
1408 * It uses the current cycle number since we're now on the
1409 * same cycle as the head so that we get:
1410 * n ... n ... | n - 1 ...
1411 * ^^^^^ blocks we're writing
1413 distance = max_distance - (log->l_logBBsize - head_block);
1414 error = xlog_write_log_records(log, head_cycle, 0, distance,
1415 tail_cycle, tail_block);
1423 /******************************************************************************
1425 * Log recover routines
1427 ******************************************************************************
1430 STATIC xlog_recover_t *
1431 xlog_recover_find_tid(
1432 struct hlist_head *head,
1435 xlog_recover_t *trans;
1436 struct hlist_node *n;
1438 hlist_for_each_entry(trans, n, head, r_list) {
1439 if (trans->r_log_tid == tid)
1446 xlog_recover_new_tid(
1447 struct hlist_head *head,
1451 xlog_recover_t *trans;
1453 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1454 trans->r_log_tid = tid;
1456 INIT_LIST_HEAD(&trans->r_itemq);
1458 INIT_HLIST_NODE(&trans->r_list);
1459 hlist_add_head(&trans->r_list, head);
1463 xlog_recover_add_item(
1464 struct list_head *head)
1466 xlog_recover_item_t *item;
1468 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1469 INIT_LIST_HEAD(&item->ri_list);
1470 list_add_tail(&item->ri_list, head);
1474 xlog_recover_add_to_cont_trans(
1476 xlog_recover_t *trans,
1480 xlog_recover_item_t *item;
1481 xfs_caddr_t ptr, old_ptr;
1484 if (list_empty(&trans->r_itemq)) {
1485 /* finish copying rest of trans header */
1486 xlog_recover_add_item(&trans->r_itemq);
1487 ptr = (xfs_caddr_t) &trans->r_theader +
1488 sizeof(xfs_trans_header_t) - len;
1489 memcpy(ptr, dp, len); /* d, s, l */
1492 /* take the tail entry */
1493 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1495 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1496 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1498 ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP);
1499 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1500 item->ri_buf[item->ri_cnt-1].i_len += len;
1501 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1502 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
1507 * The next region to add is the start of a new region. It could be
1508 * a whole region or it could be the first part of a new region. Because
1509 * of this, the assumption here is that the type and size fields of all
1510 * format structures fit into the first 32 bits of the structure.
1512 * This works because all regions must be 32 bit aligned. Therefore, we
1513 * either have both fields or we have neither field. In the case we have
1514 * neither field, the data part of the region is zero length. We only have
1515 * a log_op_header and can throw away the header since a new one will appear
1516 * later. If we have at least 4 bytes, then we can determine how many regions
1517 * will appear in the current log item.
1520 xlog_recover_add_to_trans(
1522 xlog_recover_t *trans,
1526 xfs_inode_log_format_t *in_f; /* any will do */
1527 xlog_recover_item_t *item;
1532 if (list_empty(&trans->r_itemq)) {
1533 /* we need to catch log corruptions here */
1534 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
1535 xfs_warn(log->l_mp, "%s: bad header magic number",
1538 return XFS_ERROR(EIO);
1540 if (len == sizeof(xfs_trans_header_t))
1541 xlog_recover_add_item(&trans->r_itemq);
1542 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1546 ptr = kmem_alloc(len, KM_SLEEP);
1547 memcpy(ptr, dp, len);
1548 in_f = (xfs_inode_log_format_t *)ptr;
1550 /* take the tail entry */
1551 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1552 if (item->ri_total != 0 &&
1553 item->ri_total == item->ri_cnt) {
1554 /* tail item is in use, get a new one */
1555 xlog_recover_add_item(&trans->r_itemq);
1556 item = list_entry(trans->r_itemq.prev,
1557 xlog_recover_item_t, ri_list);
1560 if (item->ri_total == 0) { /* first region to be added */
1561 if (in_f->ilf_size == 0 ||
1562 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
1564 "bad number of regions (%d) in inode log format",
1567 return XFS_ERROR(EIO);
1570 item->ri_total = in_f->ilf_size;
1572 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1575 ASSERT(item->ri_total > item->ri_cnt);
1576 /* Description region is ri_buf[0] */
1577 item->ri_buf[item->ri_cnt].i_addr = ptr;
1578 item->ri_buf[item->ri_cnt].i_len = len;
1580 trace_xfs_log_recover_item_add(log, trans, item, 0);
1585 * Sort the log items in the transaction. Cancelled buffers need
1586 * to be put first so they are processed before any items that might
1587 * modify the buffers. If they are cancelled, then the modifications
1588 * don't need to be replayed.
1591 xlog_recover_reorder_trans(
1593 xlog_recover_t *trans,
1596 xlog_recover_item_t *item, *n;
1597 LIST_HEAD(sort_list);
1599 list_splice_init(&trans->r_itemq, &sort_list);
1600 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1601 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1603 switch (ITEM_TYPE(item)) {
1605 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1606 trace_xfs_log_recover_item_reorder_head(log,
1608 list_move(&item->ri_list, &trans->r_itemq);
1613 case XFS_LI_QUOTAOFF:
1616 trace_xfs_log_recover_item_reorder_tail(log,
1618 list_move_tail(&item->ri_list, &trans->r_itemq);
1622 "%s: unrecognized type of log operation",
1625 return XFS_ERROR(EIO);
1628 ASSERT(list_empty(&sort_list));
1633 * Build up the table of buf cancel records so that we don't replay
1634 * cancelled data in the second pass. For buffer records that are
1635 * not cancel records, there is nothing to do here so we just return.
1637 * If we get a cancel record which is already in the table, this indicates
1638 * that the buffer was cancelled multiple times. In order to ensure
1639 * that during pass 2 we keep the record in the table until we reach its
1640 * last occurrence in the log, we keep a reference count in the cancel
1641 * record in the table to tell us how many times we expect to see this
1642 * record during the second pass.
1645 xlog_recover_buffer_pass1(
1647 xlog_recover_item_t *item)
1649 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1650 struct list_head *bucket;
1651 struct xfs_buf_cancel *bcp;
1654 * If this isn't a cancel buffer item, then just return.
1656 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1657 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
1662 * Insert an xfs_buf_cancel record into the hash table of them.
1663 * If there is already an identical record, bump its reference count.
1665 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1666 list_for_each_entry(bcp, bucket, bc_list) {
1667 if (bcp->bc_blkno == buf_f->blf_blkno &&
1668 bcp->bc_len == buf_f->blf_len) {
1670 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
1675 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1676 bcp->bc_blkno = buf_f->blf_blkno;
1677 bcp->bc_len = buf_f->blf_len;
1678 bcp->bc_refcount = 1;
1679 list_add_tail(&bcp->bc_list, bucket);
1681 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
1686 * Check to see whether the buffer being recovered has a corresponding
1687 * entry in the buffer cancel record table. If it does then return 1
1688 * so that it will be cancelled, otherwise return 0. If the buffer is
1689 * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement
1690 * the refcount on the entry in the table and remove it from the table
1691 * if this is the last reference.
1693 * We remove the cancel record from the table when we encounter its
1694 * last occurrence in the log so that if the same buffer is re-used
1695 * again after its last cancellation we actually replay the changes
1696 * made at that point.
1699 xlog_check_buffer_cancelled(
1705 struct list_head *bucket;
1706 struct xfs_buf_cancel *bcp;
1708 if (log->l_buf_cancel_table == NULL) {
1710 * There is nothing in the table built in pass one,
1711 * so this buffer must not be cancelled.
1713 ASSERT(!(flags & XFS_BLF_CANCEL));
1718 * Search for an entry in the cancel table that matches our buffer.
1720 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1721 list_for_each_entry(bcp, bucket, bc_list) {
1722 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
1727 * We didn't find a corresponding entry in the table, so return 0 so
1728 * that the buffer is NOT cancelled.
1730 ASSERT(!(flags & XFS_BLF_CANCEL));
1735 * We've go a match, so return 1 so that the recovery of this buffer
1736 * is cancelled. If this buffer is actually a buffer cancel log
1737 * item, then decrement the refcount on the one in the table and
1738 * remove it if this is the last reference.
1740 if (flags & XFS_BLF_CANCEL) {
1741 if (--bcp->bc_refcount == 0) {
1742 list_del(&bcp->bc_list);
1750 * Perform recovery for a buffer full of inodes. In these buffers, the only
1751 * data which should be recovered is that which corresponds to the
1752 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1753 * data for the inodes is always logged through the inodes themselves rather
1754 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1756 * The only time when buffers full of inodes are fully recovered is when the
1757 * buffer is full of newly allocated inodes. In this case the buffer will
1758 * not be marked as an inode buffer and so will be sent to
1759 * xlog_recover_do_reg_buffer() below during recovery.
1762 xlog_recover_do_inode_buffer(
1763 struct xfs_mount *mp,
1764 xlog_recover_item_t *item,
1766 xfs_buf_log_format_t *buf_f)
1772 int reg_buf_offset = 0;
1773 int reg_buf_bytes = 0;
1774 int next_unlinked_offset;
1776 xfs_agino_t *logged_nextp;
1777 xfs_agino_t *buffer_nextp;
1779 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
1781 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1782 for (i = 0; i < inodes_per_buf; i++) {
1783 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1784 offsetof(xfs_dinode_t, di_next_unlinked);
1786 while (next_unlinked_offset >=
1787 (reg_buf_offset + reg_buf_bytes)) {
1789 * The next di_next_unlinked field is beyond
1790 * the current logged region. Find the next
1791 * logged region that contains or is beyond
1792 * the current di_next_unlinked field.
1795 bit = xfs_next_bit(buf_f->blf_data_map,
1796 buf_f->blf_map_size, bit);
1799 * If there are no more logged regions in the
1800 * buffer, then we're done.
1805 nbits = xfs_contig_bits(buf_f->blf_data_map,
1806 buf_f->blf_map_size, bit);
1808 reg_buf_offset = bit << XFS_BLF_SHIFT;
1809 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
1814 * If the current logged region starts after the current
1815 * di_next_unlinked field, then move on to the next
1816 * di_next_unlinked field.
1818 if (next_unlinked_offset < reg_buf_offset)
1821 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1822 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
1823 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1826 * The current logged region contains a copy of the
1827 * current di_next_unlinked field. Extract its value
1828 * and copy it to the buffer copy.
1830 logged_nextp = item->ri_buf[item_index].i_addr +
1831 next_unlinked_offset - reg_buf_offset;
1832 if (unlikely(*logged_nextp == 0)) {
1834 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1835 "Trying to replay bad (0) inode di_next_unlinked field.",
1837 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1838 XFS_ERRLEVEL_LOW, mp);
1839 return XFS_ERROR(EFSCORRUPTED);
1842 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1843 next_unlinked_offset);
1844 *buffer_nextp = *logged_nextp;
1851 * Perform a 'normal' buffer recovery. Each logged region of the
1852 * buffer should be copied over the corresponding region in the
1853 * given buffer. The bitmap in the buf log format structure indicates
1854 * where to place the logged data.
1857 xlog_recover_do_reg_buffer(
1858 struct xfs_mount *mp,
1859 xlog_recover_item_t *item,
1861 xfs_buf_log_format_t *buf_f)
1868 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
1871 i = 1; /* 0 is the buf format structure */
1873 bit = xfs_next_bit(buf_f->blf_data_map,
1874 buf_f->blf_map_size, bit);
1877 nbits = xfs_contig_bits(buf_f->blf_data_map,
1878 buf_f->blf_map_size, bit);
1880 ASSERT(item->ri_buf[i].i_addr != NULL);
1881 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
1882 ASSERT(XFS_BUF_COUNT(bp) >=
1883 ((uint)bit << XFS_BLF_SHIFT)+(nbits<<XFS_BLF_SHIFT));
1886 * Do a sanity check if this is a dquot buffer. Just checking
1887 * the first dquot in the buffer should do. XXXThis is
1888 * probably a good thing to do for other buf types also.
1891 if (buf_f->blf_flags &
1892 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
1893 if (item->ri_buf[i].i_addr == NULL) {
1895 "XFS: NULL dquot in %s.", __func__);
1898 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
1900 "XFS: dquot too small (%d) in %s.",
1901 item->ri_buf[i].i_len, __func__);
1904 error = xfs_qm_dqcheck(mp, item->ri_buf[i].i_addr,
1905 -1, 0, XFS_QMOPT_DOWARN,
1906 "dquot_buf_recover");
1911 memcpy(xfs_buf_offset(bp,
1912 (uint)bit << XFS_BLF_SHIFT), /* dest */
1913 item->ri_buf[i].i_addr, /* source */
1914 nbits<<XFS_BLF_SHIFT); /* length */
1920 /* Shouldn't be any more regions */
1921 ASSERT(i == item->ri_total);
1925 * Do some primitive error checking on ondisk dquot data structures.
1929 struct xfs_mount *mp,
1930 xfs_disk_dquot_t *ddq,
1932 uint type, /* used only when IO_dorepair is true */
1936 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
1940 * We can encounter an uninitialized dquot buffer for 2 reasons:
1941 * 1. If we crash while deleting the quotainode(s), and those blks got
1942 * used for user data. This is because we take the path of regular
1943 * file deletion; however, the size field of quotainodes is never
1944 * updated, so all the tricks that we play in itruncate_finish
1945 * don't quite matter.
1947 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1948 * But the allocation will be replayed so we'll end up with an
1949 * uninitialized quota block.
1951 * This is all fine; things are still consistent, and we haven't lost
1952 * any quota information. Just don't complain about bad dquot blks.
1954 if (ddq->d_magic != cpu_to_be16(XFS_DQUOT_MAGIC)) {
1955 if (flags & XFS_QMOPT_DOWARN)
1957 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
1958 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
1961 if (ddq->d_version != XFS_DQUOT_VERSION) {
1962 if (flags & XFS_QMOPT_DOWARN)
1964 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
1965 str, id, ddq->d_version, XFS_DQUOT_VERSION);
1969 if (ddq->d_flags != XFS_DQ_USER &&
1970 ddq->d_flags != XFS_DQ_PROJ &&
1971 ddq->d_flags != XFS_DQ_GROUP) {
1972 if (flags & XFS_QMOPT_DOWARN)
1974 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
1975 str, id, ddq->d_flags);
1979 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
1980 if (flags & XFS_QMOPT_DOWARN)
1982 "%s : ondisk-dquot 0x%p, ID mismatch: "
1983 "0x%x expected, found id 0x%x",
1984 str, ddq, id, be32_to_cpu(ddq->d_id));
1988 if (!errs && ddq->d_id) {
1989 if (ddq->d_blk_softlimit &&
1990 be64_to_cpu(ddq->d_bcount) >
1991 be64_to_cpu(ddq->d_blk_softlimit)) {
1992 if (!ddq->d_btimer) {
1993 if (flags & XFS_QMOPT_DOWARN)
1995 "%s : Dquot ID 0x%x (0x%p) BLK TIMER NOT STARTED",
1996 str, (int)be32_to_cpu(ddq->d_id), ddq);
2000 if (ddq->d_ino_softlimit &&
2001 be64_to_cpu(ddq->d_icount) >
2002 be64_to_cpu(ddq->d_ino_softlimit)) {
2003 if (!ddq->d_itimer) {
2004 if (flags & XFS_QMOPT_DOWARN)
2006 "%s : Dquot ID 0x%x (0x%p) INODE TIMER NOT STARTED",
2007 str, (int)be32_to_cpu(ddq->d_id), ddq);
2011 if (ddq->d_rtb_softlimit &&
2012 be64_to_cpu(ddq->d_rtbcount) >
2013 be64_to_cpu(ddq->d_rtb_softlimit)) {
2014 if (!ddq->d_rtbtimer) {
2015 if (flags & XFS_QMOPT_DOWARN)
2017 "%s : Dquot ID 0x%x (0x%p) RTBLK TIMER NOT STARTED",
2018 str, (int)be32_to_cpu(ddq->d_id), ddq);
2024 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2027 if (flags & XFS_QMOPT_DOWARN)
2028 xfs_notice(mp, "Re-initializing dquot ID 0x%x", id);
2031 * Typically, a repair is only requested by quotacheck.
2034 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2035 memset(d, 0, sizeof(xfs_dqblk_t));
2037 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2038 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2039 d->dd_diskdq.d_flags = type;
2040 d->dd_diskdq.d_id = cpu_to_be32(id);
2046 * Perform a dquot buffer recovery.
2047 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2048 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2049 * Else, treat it as a regular buffer and do recovery.
2052 xlog_recover_do_dquot_buffer(
2055 xlog_recover_item_t *item,
2057 xfs_buf_log_format_t *buf_f)
2061 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2064 * Filesystems are required to send in quota flags at mount time.
2066 if (mp->m_qflags == 0) {
2071 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2072 type |= XFS_DQ_USER;
2073 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2074 type |= XFS_DQ_PROJ;
2075 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2076 type |= XFS_DQ_GROUP;
2078 * This type of quotas was turned off, so ignore this buffer
2080 if (log->l_quotaoffs_flag & type)
2083 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2087 * This routine replays a modification made to a buffer at runtime.
2088 * There are actually two types of buffer, regular and inode, which
2089 * are handled differently. Inode buffers are handled differently
2090 * in that we only recover a specific set of data from them, namely
2091 * the inode di_next_unlinked fields. This is because all other inode
2092 * data is actually logged via inode records and any data we replay
2093 * here which overlaps that may be stale.
2095 * When meta-data buffers are freed at run time we log a buffer item
2096 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2097 * of the buffer in the log should not be replayed at recovery time.
2098 * This is so that if the blocks covered by the buffer are reused for
2099 * file data before we crash we don't end up replaying old, freed
2100 * meta-data into a user's file.
2102 * To handle the cancellation of buffer log items, we make two passes
2103 * over the log during recovery. During the first we build a table of
2104 * those buffers which have been cancelled, and during the second we
2105 * only replay those buffers which do not have corresponding cancel
2106 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2107 * for more details on the implementation of the table of cancel records.
2110 xlog_recover_buffer_pass2(
2112 struct list_head *buffer_list,
2113 xlog_recover_item_t *item)
2115 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2116 xfs_mount_t *mp = log->l_mp;
2122 * In this pass we only want to recover all the buffers which have
2123 * not been cancelled and are not cancellation buffers themselves.
2125 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2126 buf_f->blf_len, buf_f->blf_flags)) {
2127 trace_xfs_log_recover_buf_cancel(log, buf_f);
2131 trace_xfs_log_recover_buf_recover(log, buf_f);
2133 buf_flags = XBF_LOCK;
2134 if (!(buf_f->blf_flags & XFS_BLF_INODE_BUF))
2135 buf_flags |= XBF_MAPPED;
2137 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2140 return XFS_ERROR(ENOMEM);
2141 error = bp->b_error;
2143 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
2148 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
2149 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2150 } else if (buf_f->blf_flags &
2151 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2152 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2154 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2157 return XFS_ERROR(error);
2160 * Perform delayed write on the buffer. Asynchronous writes will be
2161 * slower when taking into account all the buffers to be flushed.
2163 * Also make sure that only inode buffers with good sizes stay in
2164 * the buffer cache. The kernel moves inodes in buffers of 1 block
2165 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2166 * buffers in the log can be a different size if the log was generated
2167 * by an older kernel using unclustered inode buffers or a newer kernel
2168 * running with a different inode cluster size. Regardless, if the
2169 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2170 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2171 * the buffer out of the buffer cache so that the buffer won't
2172 * overlap with future reads of those inodes.
2174 if (XFS_DINODE_MAGIC ==
2175 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2176 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2177 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2179 error = xfs_bwrite(bp);
2181 ASSERT(bp->b_target->bt_mount == mp);
2182 bp->b_iodone = xlog_recover_iodone;
2183 xfs_buf_delwri_queue(bp, buffer_list);
2191 xlog_recover_inode_pass2(
2193 struct list_head *buffer_list,
2194 xlog_recover_item_t *item)
2196 xfs_inode_log_format_t *in_f;
2197 xfs_mount_t *mp = log->l_mp;
2206 xfs_icdinode_t *dicp;
2209 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2210 in_f = item->ri_buf[0].i_addr;
2212 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
2214 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2220 * Inode buffers can be freed, look out for it,
2221 * and do not replay the inode.
2223 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2224 in_f->ilf_len, 0)) {
2226 trace_xfs_log_recover_inode_cancel(log, in_f);
2229 trace_xfs_log_recover_inode_recover(log, in_f);
2231 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len,
2237 error = bp->b_error;
2239 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
2243 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2244 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
2247 * Make sure the place we're flushing out to really looks
2250 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
2253 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2254 __func__, dip, bp, in_f->ilf_ino);
2255 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2256 XFS_ERRLEVEL_LOW, mp);
2257 error = EFSCORRUPTED;
2260 dicp = item->ri_buf[1].i_addr;
2261 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2264 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2265 __func__, item, in_f->ilf_ino);
2266 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2267 XFS_ERRLEVEL_LOW, mp);
2268 error = EFSCORRUPTED;
2272 /* Skip replay when the on disk inode is newer than the log one */
2273 if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
2275 * Deal with the wrap case, DI_MAX_FLUSH is less
2276 * than smaller numbers
2278 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
2279 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2283 trace_xfs_log_recover_inode_skip(log, in_f);
2288 /* Take the opportunity to reset the flush iteration count */
2289 dicp->di_flushiter = 0;
2291 if (unlikely(S_ISREG(dicp->di_mode))) {
2292 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2293 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2294 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2295 XFS_ERRLEVEL_LOW, mp, dicp);
2298 "%s: Bad regular inode log record, rec ptr 0x%p, "
2299 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2300 __func__, item, dip, bp, in_f->ilf_ino);
2301 error = EFSCORRUPTED;
2304 } else if (unlikely(S_ISDIR(dicp->di_mode))) {
2305 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2306 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2307 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2308 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2309 XFS_ERRLEVEL_LOW, mp, dicp);
2312 "%s: Bad dir inode log record, rec ptr 0x%p, "
2313 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2314 __func__, item, dip, bp, in_f->ilf_ino);
2315 error = EFSCORRUPTED;
2319 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2320 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2321 XFS_ERRLEVEL_LOW, mp, dicp);
2324 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2325 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2326 __func__, item, dip, bp, in_f->ilf_ino,
2327 dicp->di_nextents + dicp->di_anextents,
2329 error = EFSCORRUPTED;
2332 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2333 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2334 XFS_ERRLEVEL_LOW, mp, dicp);
2337 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2338 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
2339 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
2340 error = EFSCORRUPTED;
2343 if (unlikely(item->ri_buf[1].i_len > sizeof(struct xfs_icdinode))) {
2344 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
2345 XFS_ERRLEVEL_LOW, mp, dicp);
2348 "%s: Bad inode log record length %d, rec ptr 0x%p",
2349 __func__, item->ri_buf[1].i_len, item);
2350 error = EFSCORRUPTED;
2354 /* The core is in in-core format */
2355 xfs_dinode_to_disk(dip, item->ri_buf[1].i_addr);
2357 /* the rest is in on-disk format */
2358 if (item->ri_buf[1].i_len > sizeof(struct xfs_icdinode)) {
2359 memcpy((xfs_caddr_t) dip + sizeof(struct xfs_icdinode),
2360 item->ri_buf[1].i_addr + sizeof(struct xfs_icdinode),
2361 item->ri_buf[1].i_len - sizeof(struct xfs_icdinode));
2364 fields = in_f->ilf_fields;
2365 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2367 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
2370 memcpy(XFS_DFORK_DPTR(dip),
2371 &in_f->ilf_u.ilfu_uuid,
2376 if (in_f->ilf_size == 2)
2377 goto write_inode_buffer;
2378 len = item->ri_buf[2].i_len;
2379 src = item->ri_buf[2].i_addr;
2380 ASSERT(in_f->ilf_size <= 4);
2381 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2382 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2383 (len == in_f->ilf_dsize));
2385 switch (fields & XFS_ILOG_DFORK) {
2386 case XFS_ILOG_DDATA:
2388 memcpy(XFS_DFORK_DPTR(dip), src, len);
2391 case XFS_ILOG_DBROOT:
2392 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
2393 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
2394 XFS_DFORK_DSIZE(dip, mp));
2399 * There are no data fork flags set.
2401 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2406 * If we logged any attribute data, recover it. There may or
2407 * may not have been any other non-core data logged in this
2410 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2411 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2416 len = item->ri_buf[attr_index].i_len;
2417 src = item->ri_buf[attr_index].i_addr;
2418 ASSERT(len == in_f->ilf_asize);
2420 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2421 case XFS_ILOG_ADATA:
2423 dest = XFS_DFORK_APTR(dip);
2424 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2425 memcpy(dest, src, len);
2428 case XFS_ILOG_ABROOT:
2429 dest = XFS_DFORK_APTR(dip);
2430 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2431 len, (xfs_bmdr_block_t*)dest,
2432 XFS_DFORK_ASIZE(dip, mp));
2436 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
2445 ASSERT(bp->b_target->bt_mount == mp);
2446 bp->b_iodone = xlog_recover_iodone;
2447 xfs_buf_delwri_queue(bp, buffer_list);
2452 return XFS_ERROR(error);
2456 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2457 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2461 xlog_recover_quotaoff_pass1(
2463 xlog_recover_item_t *item)
2465 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
2469 * The logitem format's flag tells us if this was user quotaoff,
2470 * group/project quotaoff or both.
2472 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2473 log->l_quotaoffs_flag |= XFS_DQ_USER;
2474 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2475 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2476 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2477 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2483 * Recover a dquot record
2486 xlog_recover_dquot_pass2(
2488 struct list_head *buffer_list,
2489 xlog_recover_item_t *item)
2491 xfs_mount_t *mp = log->l_mp;
2493 struct xfs_disk_dquot *ddq, *recddq;
2495 xfs_dq_logformat_t *dq_f;
2500 * Filesystems are required to send in quota flags at mount time.
2502 if (mp->m_qflags == 0)
2505 recddq = item->ri_buf[1].i_addr;
2506 if (recddq == NULL) {
2507 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
2508 return XFS_ERROR(EIO);
2510 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
2511 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
2512 item->ri_buf[1].i_len, __func__);
2513 return XFS_ERROR(EIO);
2517 * This type of quotas was turned off, so ignore this record.
2519 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
2521 if (log->l_quotaoffs_flag & type)
2525 * At this point we know that quota was _not_ turned off.
2526 * Since the mount flags are not indicating to us otherwise, this
2527 * must mean that quota is on, and the dquot needs to be replayed.
2528 * Remember that we may not have fully recovered the superblock yet,
2529 * so we can't do the usual trick of looking at the SB quota bits.
2531 * The other possibility, of course, is that the quota subsystem was
2532 * removed since the last mount - ENOSYS.
2534 dq_f = item->ri_buf[0].i_addr;
2536 error = xfs_qm_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2537 "xlog_recover_dquot_pass2 (log copy)");
2539 return XFS_ERROR(EIO);
2540 ASSERT(dq_f->qlf_len == 1);
2542 error = xfs_read_buf(mp, mp->m_ddev_targp,
2544 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2547 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#3)");
2551 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2554 * At least the magic num portion should be on disk because this
2555 * was among a chunk of dquots created earlier, and we did some
2556 * minimal initialization then.
2558 error = xfs_qm_dqcheck(mp, ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2559 "xlog_recover_dquot_pass2");
2562 return XFS_ERROR(EIO);
2565 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2567 ASSERT(dq_f->qlf_size == 2);
2568 ASSERT(bp->b_target->bt_mount == mp);
2569 bp->b_iodone = xlog_recover_iodone;
2570 xfs_buf_delwri_queue(bp, buffer_list);
2577 * This routine is called to create an in-core extent free intent
2578 * item from the efi format structure which was logged on disk.
2579 * It allocates an in-core efi, copies the extents from the format
2580 * structure into it, and adds the efi to the AIL with the given
2584 xlog_recover_efi_pass2(
2586 xlog_recover_item_t *item,
2590 xfs_mount_t *mp = log->l_mp;
2591 xfs_efi_log_item_t *efip;
2592 xfs_efi_log_format_t *efi_formatp;
2594 efi_formatp = item->ri_buf[0].i_addr;
2596 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
2597 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2598 &(efip->efi_format)))) {
2599 xfs_efi_item_free(efip);
2602 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
2604 spin_lock(&log->l_ailp->xa_lock);
2606 * xfs_trans_ail_update() drops the AIL lock.
2608 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
2614 * This routine is called when an efd format structure is found in
2615 * a committed transaction in the log. It's purpose is to cancel
2616 * the corresponding efi if it was still in the log. To do this
2617 * it searches the AIL for the efi with an id equal to that in the
2618 * efd format structure. If we find it, we remove the efi from the
2622 xlog_recover_efd_pass2(
2624 xlog_recover_item_t *item)
2626 xfs_efd_log_format_t *efd_formatp;
2627 xfs_efi_log_item_t *efip = NULL;
2628 xfs_log_item_t *lip;
2630 struct xfs_ail_cursor cur;
2631 struct xfs_ail *ailp = log->l_ailp;
2633 efd_formatp = item->ri_buf[0].i_addr;
2634 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2635 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2636 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2637 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
2638 efi_id = efd_formatp->efd_efi_id;
2641 * Search for the efi with the id in the efd format structure
2644 spin_lock(&ailp->xa_lock);
2645 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
2646 while (lip != NULL) {
2647 if (lip->li_type == XFS_LI_EFI) {
2648 efip = (xfs_efi_log_item_t *)lip;
2649 if (efip->efi_format.efi_id == efi_id) {
2651 * xfs_trans_ail_delete() drops the
2654 xfs_trans_ail_delete(ailp, lip,
2655 SHUTDOWN_CORRUPT_INCORE);
2656 xfs_efi_item_free(efip);
2657 spin_lock(&ailp->xa_lock);
2661 lip = xfs_trans_ail_cursor_next(ailp, &cur);
2663 xfs_trans_ail_cursor_done(ailp, &cur);
2664 spin_unlock(&ailp->xa_lock);
2670 * Free up any resources allocated by the transaction
2672 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2675 xlog_recover_free_trans(
2676 struct xlog_recover *trans)
2678 xlog_recover_item_t *item, *n;
2681 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
2682 /* Free the regions in the item. */
2683 list_del(&item->ri_list);
2684 for (i = 0; i < item->ri_cnt; i++)
2685 kmem_free(item->ri_buf[i].i_addr);
2686 /* Free the item itself */
2687 kmem_free(item->ri_buf);
2690 /* Free the transaction recover structure */
2695 xlog_recover_commit_pass1(
2697 struct xlog_recover *trans,
2698 xlog_recover_item_t *item)
2700 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
2702 switch (ITEM_TYPE(item)) {
2704 return xlog_recover_buffer_pass1(log, item);
2705 case XFS_LI_QUOTAOFF:
2706 return xlog_recover_quotaoff_pass1(log, item);
2711 /* nothing to do in pass 1 */
2714 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
2715 __func__, ITEM_TYPE(item));
2717 return XFS_ERROR(EIO);
2722 xlog_recover_commit_pass2(
2724 struct xlog_recover *trans,
2725 struct list_head *buffer_list,
2726 xlog_recover_item_t *item)
2728 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
2730 switch (ITEM_TYPE(item)) {
2732 return xlog_recover_buffer_pass2(log, buffer_list, item);
2734 return xlog_recover_inode_pass2(log, buffer_list, item);
2736 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
2738 return xlog_recover_efd_pass2(log, item);
2740 return xlog_recover_dquot_pass2(log, buffer_list, item);
2741 case XFS_LI_QUOTAOFF:
2742 /* nothing to do in pass2 */
2745 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
2746 __func__, ITEM_TYPE(item));
2748 return XFS_ERROR(EIO);
2753 * Perform the transaction.
2755 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2756 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2759 xlog_recover_commit_trans(
2761 struct xlog_recover *trans,
2764 int error = 0, error2;
2765 xlog_recover_item_t *item;
2766 LIST_HEAD (buffer_list);
2768 hlist_del(&trans->r_list);
2770 error = xlog_recover_reorder_trans(log, trans, pass);
2774 list_for_each_entry(item, &trans->r_itemq, ri_list) {
2776 case XLOG_RECOVER_PASS1:
2777 error = xlog_recover_commit_pass1(log, trans, item);
2779 case XLOG_RECOVER_PASS2:
2780 error = xlog_recover_commit_pass2(log, trans,
2781 &buffer_list, item);
2791 xlog_recover_free_trans(trans);
2794 error2 = xfs_buf_delwri_submit(&buffer_list);
2795 return error ? error : error2;
2799 xlog_recover_unmount_trans(
2801 xlog_recover_t *trans)
2803 /* Do nothing now */
2804 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
2809 * There are two valid states of the r_state field. 0 indicates that the
2810 * transaction structure is in a normal state. We have either seen the
2811 * start of the transaction or the last operation we added was not a partial
2812 * operation. If the last operation we added to the transaction was a
2813 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2815 * NOTE: skip LRs with 0 data length.
2818 xlog_recover_process_data(
2820 struct hlist_head rhash[],
2821 xlog_rec_header_t *rhead,
2827 xlog_op_header_t *ohead;
2828 xlog_recover_t *trans;
2834 lp = dp + be32_to_cpu(rhead->h_len);
2835 num_logops = be32_to_cpu(rhead->h_num_logops);
2837 /* check the log format matches our own - else we can't recover */
2838 if (xlog_header_check_recover(log->l_mp, rhead))
2839 return (XFS_ERROR(EIO));
2841 while ((dp < lp) && num_logops) {
2842 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2843 ohead = (xlog_op_header_t *)dp;
2844 dp += sizeof(xlog_op_header_t);
2845 if (ohead->oh_clientid != XFS_TRANSACTION &&
2846 ohead->oh_clientid != XFS_LOG) {
2847 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
2848 __func__, ohead->oh_clientid);
2850 return (XFS_ERROR(EIO));
2852 tid = be32_to_cpu(ohead->oh_tid);
2853 hash = XLOG_RHASH(tid);
2854 trans = xlog_recover_find_tid(&rhash[hash], tid);
2855 if (trans == NULL) { /* not found; add new tid */
2856 if (ohead->oh_flags & XLOG_START_TRANS)
2857 xlog_recover_new_tid(&rhash[hash], tid,
2858 be64_to_cpu(rhead->h_lsn));
2860 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
2861 xfs_warn(log->l_mp, "%s: bad length 0x%x",
2862 __func__, be32_to_cpu(ohead->oh_len));
2864 return (XFS_ERROR(EIO));
2866 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2867 if (flags & XLOG_WAS_CONT_TRANS)
2868 flags &= ~XLOG_CONTINUE_TRANS;
2870 case XLOG_COMMIT_TRANS:
2871 error = xlog_recover_commit_trans(log,
2874 case XLOG_UNMOUNT_TRANS:
2875 error = xlog_recover_unmount_trans(log, trans);
2877 case XLOG_WAS_CONT_TRANS:
2878 error = xlog_recover_add_to_cont_trans(log,
2880 be32_to_cpu(ohead->oh_len));
2882 case XLOG_START_TRANS:
2883 xfs_warn(log->l_mp, "%s: bad transaction",
2886 error = XFS_ERROR(EIO);
2889 case XLOG_CONTINUE_TRANS:
2890 error = xlog_recover_add_to_trans(log, trans,
2891 dp, be32_to_cpu(ohead->oh_len));
2894 xfs_warn(log->l_mp, "%s: bad flag 0x%x",
2897 error = XFS_ERROR(EIO);
2903 dp += be32_to_cpu(ohead->oh_len);
2910 * Process an extent free intent item that was recovered from
2911 * the log. We need to free the extents that it describes.
2914 xlog_recover_process_efi(
2916 xfs_efi_log_item_t *efip)
2918 xfs_efd_log_item_t *efdp;
2923 xfs_fsblock_t startblock_fsb;
2925 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
2928 * First check the validity of the extents described by the
2929 * EFI. If any are bad, then assume that all are bad and
2930 * just toss the EFI.
2932 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2933 extp = &(efip->efi_format.efi_extents[i]);
2934 startblock_fsb = XFS_BB_TO_FSB(mp,
2935 XFS_FSB_TO_DADDR(mp, extp->ext_start));
2936 if ((startblock_fsb == 0) ||
2937 (extp->ext_len == 0) ||
2938 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
2939 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
2941 * This will pull the EFI from the AIL and
2942 * free the memory associated with it.
2944 xfs_efi_release(efip, efip->efi_format.efi_nextents);
2945 return XFS_ERROR(EIO);
2949 tp = xfs_trans_alloc(mp, 0);
2950 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
2953 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
2955 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2956 extp = &(efip->efi_format.efi_extents[i]);
2957 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
2960 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
2964 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
2965 error = xfs_trans_commit(tp, 0);
2969 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
2974 * When this is called, all of the EFIs which did not have
2975 * corresponding EFDs should be in the AIL. What we do now
2976 * is free the extents associated with each one.
2978 * Since we process the EFIs in normal transactions, they
2979 * will be removed at some point after the commit. This prevents
2980 * us from just walking down the list processing each one.
2981 * We'll use a flag in the EFI to skip those that we've already
2982 * processed and use the AIL iteration mechanism's generation
2983 * count to try to speed this up at least a bit.
2985 * When we start, we know that the EFIs are the only things in
2986 * the AIL. As we process them, however, other items are added
2987 * to the AIL. Since everything added to the AIL must come after
2988 * everything already in the AIL, we stop processing as soon as
2989 * we see something other than an EFI in the AIL.
2992 xlog_recover_process_efis(
2995 xfs_log_item_t *lip;
2996 xfs_efi_log_item_t *efip;
2998 struct xfs_ail_cursor cur;
2999 struct xfs_ail *ailp;
3002 spin_lock(&ailp->xa_lock);
3003 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3004 while (lip != NULL) {
3006 * We're done when we see something other than an EFI.
3007 * There should be no EFIs left in the AIL now.
3009 if (lip->li_type != XFS_LI_EFI) {
3011 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
3012 ASSERT(lip->li_type != XFS_LI_EFI);
3018 * Skip EFIs that we've already processed.
3020 efip = (xfs_efi_log_item_t *)lip;
3021 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
3022 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3026 spin_unlock(&ailp->xa_lock);
3027 error = xlog_recover_process_efi(log->l_mp, efip);
3028 spin_lock(&ailp->xa_lock);
3031 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3034 xfs_trans_ail_cursor_done(ailp, &cur);
3035 spin_unlock(&ailp->xa_lock);
3040 * This routine performs a transaction to null out a bad inode pointer
3041 * in an agi unlinked inode hash bucket.
3044 xlog_recover_clear_agi_bucket(
3046 xfs_agnumber_t agno,
3055 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3056 error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp),
3061 error = xfs_read_agi(mp, tp, agno, &agibp);
3065 agi = XFS_BUF_TO_AGI(agibp);
3066 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
3067 offset = offsetof(xfs_agi_t, agi_unlinked) +
3068 (sizeof(xfs_agino_t) * bucket);
3069 xfs_trans_log_buf(tp, agibp, offset,
3070 (offset + sizeof(xfs_agino_t) - 1));
3072 error = xfs_trans_commit(tp, 0);
3078 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3080 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
3085 xlog_recover_process_one_iunlink(
3086 struct xfs_mount *mp,
3087 xfs_agnumber_t agno,
3091 struct xfs_buf *ibp;
3092 struct xfs_dinode *dip;
3093 struct xfs_inode *ip;
3097 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3098 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
3103 * Get the on disk inode to find the next inode in the bucket.
3105 error = xfs_itobp(mp, NULL, ip, &dip, &ibp, XBF_LOCK);
3109 ASSERT(ip->i_d.di_nlink == 0);
3110 ASSERT(ip->i_d.di_mode != 0);
3112 /* setup for the next pass */
3113 agino = be32_to_cpu(dip->di_next_unlinked);
3117 * Prevent any DMAPI event from being sent when the reference on
3118 * the inode is dropped.
3120 ip->i_d.di_dmevmask = 0;
3129 * We can't read in the inode this bucket points to, or this inode
3130 * is messed up. Just ditch this bucket of inodes. We will lose
3131 * some inodes and space, but at least we won't hang.
3133 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3134 * clear the inode pointer in the bucket.
3136 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3141 * xlog_iunlink_recover
3143 * This is called during recovery to process any inodes which
3144 * we unlinked but not freed when the system crashed. These
3145 * inodes will be on the lists in the AGI blocks. What we do
3146 * here is scan all the AGIs and fully truncate and free any
3147 * inodes found on the lists. Each inode is removed from the
3148 * lists when it has been fully truncated and is freed. The
3149 * freeing of the inode and its removal from the list must be
3153 xlog_recover_process_iunlinks(
3157 xfs_agnumber_t agno;
3168 * Prevent any DMAPI event from being sent while in this function.
3170 mp_dmevmask = mp->m_dmevmask;
3173 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3175 * Find the agi for this ag.
3177 error = xfs_read_agi(mp, NULL, agno, &agibp);
3180 * AGI is b0rked. Don't process it.
3182 * We should probably mark the filesystem as corrupt
3183 * after we've recovered all the ag's we can....
3188 * Unlock the buffer so that it can be acquired in the normal
3189 * course of the transaction to truncate and free each inode.
3190 * Because we are not racing with anyone else here for the AGI
3191 * buffer, we don't even need to hold it locked to read the
3192 * initial unlinked bucket entries out of the buffer. We keep
3193 * buffer reference though, so that it stays pinned in memory
3194 * while we need the buffer.
3196 agi = XFS_BUF_TO_AGI(agibp);
3197 xfs_buf_unlock(agibp);
3199 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3200 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3201 while (agino != NULLAGINO) {
3202 agino = xlog_recover_process_one_iunlink(mp,
3203 agno, agino, bucket);
3206 xfs_buf_rele(agibp);
3209 mp->m_dmevmask = mp_dmevmask;
3215 xlog_pack_data_checksum(
3217 xlog_in_core_t *iclog,
3224 up = (__be32 *)iclog->ic_datap;
3225 /* divide length by 4 to get # words */
3226 for (i = 0; i < (size >> 2); i++) {
3227 chksum ^= be32_to_cpu(*up);
3230 iclog->ic_header.h_chksum = cpu_to_be32(chksum);
3233 #define xlog_pack_data_checksum(log, iclog, size)
3237 * Stamp cycle number in every block
3242 xlog_in_core_t *iclog,
3246 int size = iclog->ic_offset + roundoff;
3250 xlog_pack_data_checksum(log, iclog, size);
3252 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3254 dp = iclog->ic_datap;
3255 for (i = 0; i < BTOBB(size) &&
3256 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3257 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
3258 *(__be32 *)dp = cycle_lsn;
3262 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3263 xlog_in_core_2_t *xhdr = iclog->ic_data;
3265 for ( ; i < BTOBB(size); i++) {
3266 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3267 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3268 xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
3269 *(__be32 *)dp = cycle_lsn;
3273 for (i = 1; i < log->l_iclog_heads; i++) {
3274 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3281 xlog_rec_header_t *rhead,
3287 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
3288 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3289 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
3293 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3294 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
3295 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
3296 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3297 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3298 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
3305 xlog_valid_rec_header(
3307 xlog_rec_header_t *rhead,
3312 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
3313 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3314 XFS_ERRLEVEL_LOW, log->l_mp);
3315 return XFS_ERROR(EFSCORRUPTED);
3318 (!rhead->h_version ||
3319 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
3320 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
3321 __func__, be32_to_cpu(rhead->h_version));
3322 return XFS_ERROR(EIO);
3325 /* LR body must have data or it wouldn't have been written */
3326 hlen = be32_to_cpu(rhead->h_len);
3327 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3328 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3329 XFS_ERRLEVEL_LOW, log->l_mp);
3330 return XFS_ERROR(EFSCORRUPTED);
3332 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3333 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3334 XFS_ERRLEVEL_LOW, log->l_mp);
3335 return XFS_ERROR(EFSCORRUPTED);
3341 * Read the log from tail to head and process the log records found.
3342 * Handle the two cases where the tail and head are in the same cycle
3343 * and where the active portion of the log wraps around the end of
3344 * the physical log separately. The pass parameter is passed through
3345 * to the routines called to process the data and is not looked at
3349 xlog_do_recovery_pass(
3351 xfs_daddr_t head_blk,
3352 xfs_daddr_t tail_blk,
3355 xlog_rec_header_t *rhead;
3358 xfs_buf_t *hbp, *dbp;
3359 int error = 0, h_size;
3360 int bblks, split_bblks;
3361 int hblks, split_hblks, wrapped_hblks;
3362 struct hlist_head rhash[XLOG_RHASH_SIZE];
3364 ASSERT(head_blk != tail_blk);
3367 * Read the header of the tail block and get the iclog buffer size from
3368 * h_size. Use this to tell how many sectors make up the log header.
3370 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3372 * When using variable length iclogs, read first sector of
3373 * iclog header and extract the header size from it. Get a
3374 * new hbp that is the correct size.
3376 hbp = xlog_get_bp(log, 1);
3380 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
3384 rhead = (xlog_rec_header_t *)offset;
3385 error = xlog_valid_rec_header(log, rhead, tail_blk);
3388 h_size = be32_to_cpu(rhead->h_size);
3389 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
3390 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3391 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3392 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3395 hbp = xlog_get_bp(log, hblks);
3400 ASSERT(log->l_sectBBsize == 1);
3402 hbp = xlog_get_bp(log, 1);
3403 h_size = XLOG_BIG_RECORD_BSIZE;
3408 dbp = xlog_get_bp(log, BTOBB(h_size));
3414 memset(rhash, 0, sizeof(rhash));
3415 if (tail_blk <= head_blk) {
3416 for (blk_no = tail_blk; blk_no < head_blk; ) {
3417 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3421 rhead = (xlog_rec_header_t *)offset;
3422 error = xlog_valid_rec_header(log, rhead, blk_no);
3426 /* blocks in data section */
3427 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3428 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
3433 xlog_unpack_data(rhead, offset, log);
3434 if ((error = xlog_recover_process_data(log,
3435 rhash, rhead, offset, pass)))
3437 blk_no += bblks + hblks;
3441 * Perform recovery around the end of the physical log.
3442 * When the head is not on the same cycle number as the tail,
3443 * we can't do a sequential recovery as above.
3446 while (blk_no < log->l_logBBsize) {
3448 * Check for header wrapping around physical end-of-log
3450 offset = hbp->b_addr;
3453 if (blk_no + hblks <= log->l_logBBsize) {
3454 /* Read header in one read */
3455 error = xlog_bread(log, blk_no, hblks, hbp,
3460 /* This LR is split across physical log end */
3461 if (blk_no != log->l_logBBsize) {
3462 /* some data before physical log end */
3463 ASSERT(blk_no <= INT_MAX);
3464 split_hblks = log->l_logBBsize - (int)blk_no;
3465 ASSERT(split_hblks > 0);
3466 error = xlog_bread(log, blk_no,
3474 * Note: this black magic still works with
3475 * large sector sizes (non-512) only because:
3476 * - we increased the buffer size originally
3477 * by 1 sector giving us enough extra space
3478 * for the second read;
3479 * - the log start is guaranteed to be sector
3481 * - we read the log end (LR header start)
3482 * _first_, then the log start (LR header end)
3483 * - order is important.
3485 wrapped_hblks = hblks - split_hblks;
3486 error = xlog_bread_offset(log, 0,
3488 offset + BBTOB(split_hblks));
3492 rhead = (xlog_rec_header_t *)offset;
3493 error = xlog_valid_rec_header(log, rhead,
3494 split_hblks ? blk_no : 0);
3498 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3501 /* Read in data for log record */
3502 if (blk_no + bblks <= log->l_logBBsize) {
3503 error = xlog_bread(log, blk_no, bblks, dbp,
3508 /* This log record is split across the
3509 * physical end of log */
3510 offset = dbp->b_addr;
3512 if (blk_no != log->l_logBBsize) {
3513 /* some data is before the physical
3515 ASSERT(!wrapped_hblks);
3516 ASSERT(blk_no <= INT_MAX);
3518 log->l_logBBsize - (int)blk_no;
3519 ASSERT(split_bblks > 0);
3520 error = xlog_bread(log, blk_no,
3528 * Note: this black magic still works with
3529 * large sector sizes (non-512) only because:
3530 * - we increased the buffer size originally
3531 * by 1 sector giving us enough extra space
3532 * for the second read;
3533 * - the log start is guaranteed to be sector
3535 * - we read the log end (LR header start)
3536 * _first_, then the log start (LR header end)
3537 * - order is important.
3539 error = xlog_bread_offset(log, 0,
3540 bblks - split_bblks, hbp,
3541 offset + BBTOB(split_bblks));
3545 xlog_unpack_data(rhead, offset, log);
3546 if ((error = xlog_recover_process_data(log, rhash,
3547 rhead, offset, pass)))
3552 ASSERT(blk_no >= log->l_logBBsize);
3553 blk_no -= log->l_logBBsize;
3555 /* read first part of physical log */
3556 while (blk_no < head_blk) {
3557 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3561 rhead = (xlog_rec_header_t *)offset;
3562 error = xlog_valid_rec_header(log, rhead, blk_no);
3566 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3567 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
3572 xlog_unpack_data(rhead, offset, log);
3573 if ((error = xlog_recover_process_data(log, rhash,
3574 rhead, offset, pass)))
3576 blk_no += bblks + hblks;
3588 * Do the recovery of the log. We actually do this in two phases.
3589 * The two passes are necessary in order to implement the function
3590 * of cancelling a record written into the log. The first pass
3591 * determines those things which have been cancelled, and the
3592 * second pass replays log items normally except for those which
3593 * have been cancelled. The handling of the replay and cancellations
3594 * takes place in the log item type specific routines.
3596 * The table of items which have cancel records in the log is allocated
3597 * and freed at this level, since only here do we know when all of
3598 * the log recovery has been completed.
3601 xlog_do_log_recovery(
3603 xfs_daddr_t head_blk,
3604 xfs_daddr_t tail_blk)
3608 ASSERT(head_blk != tail_blk);
3611 * First do a pass to find all of the cancelled buf log items.
3612 * Store them in the buf_cancel_table for use in the second pass.
3614 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
3615 sizeof(struct list_head),
3617 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3618 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
3620 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3621 XLOG_RECOVER_PASS1);
3623 kmem_free(log->l_buf_cancel_table);
3624 log->l_buf_cancel_table = NULL;
3628 * Then do a second pass to actually recover the items in the log.
3629 * When it is complete free the table of buf cancel items.
3631 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3632 XLOG_RECOVER_PASS2);
3637 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3638 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
3642 kmem_free(log->l_buf_cancel_table);
3643 log->l_buf_cancel_table = NULL;
3649 * Do the actual recovery
3654 xfs_daddr_t head_blk,
3655 xfs_daddr_t tail_blk)
3662 * First replay the images in the log.
3664 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3669 * If IO errors happened during recovery, bail out.
3671 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3676 * We now update the tail_lsn since much of the recovery has completed
3677 * and there may be space available to use. If there were no extent
3678 * or iunlinks, we can free up the entire log and set the tail_lsn to
3679 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3680 * lsn of the last known good LR on disk. If there are extent frees
3681 * or iunlinks they will have some entries in the AIL; so we look at
3682 * the AIL to determine how to set the tail_lsn.
3684 xlog_assign_tail_lsn(log->l_mp);
3687 * Now that we've finished replaying all buffer and inode
3688 * updates, re-read in the superblock.
3690 bp = xfs_getsb(log->l_mp, 0);
3692 ASSERT(!(XFS_BUF_ISWRITE(bp)));
3694 XFS_BUF_UNASYNC(bp);
3695 xfsbdstrat(log->l_mp, bp);
3696 error = xfs_buf_iowait(bp);
3698 xfs_buf_ioerror_alert(bp, __func__);
3704 /* Convert superblock from on-disk format */
3705 sbp = &log->l_mp->m_sb;
3706 xfs_sb_from_disk(log->l_mp, XFS_BUF_TO_SBP(bp));
3707 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
3708 ASSERT(xfs_sb_good_version(sbp));
3711 /* We've re-read the superblock so re-initialize per-cpu counters */
3712 xfs_icsb_reinit_counters(log->l_mp);
3714 xlog_recover_check_summary(log);
3716 /* Normal transactions can now occur */
3717 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3722 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3724 * Return error or zero.
3730 xfs_daddr_t head_blk, tail_blk;
3733 /* find the tail of the log */
3734 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
3737 if (tail_blk != head_blk) {
3738 /* There used to be a comment here:
3740 * disallow recovery on read-only mounts. note -- mount
3741 * checks for ENOSPC and turns it into an intelligent
3743 * ...but this is no longer true. Now, unless you specify
3744 * NORECOVERY (in which case this function would never be
3745 * called), we just go ahead and recover. We do this all
3746 * under the vfs layer, so we can get away with it unless
3747 * the device itself is read-only, in which case we fail.
3749 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
3753 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
3754 log->l_mp->m_logname ? log->l_mp->m_logname
3757 error = xlog_do_recover(log, head_blk, tail_blk);
3758 log->l_flags |= XLOG_RECOVERY_NEEDED;
3764 * In the first part of recovery we replay inodes and buffers and build
3765 * up the list of extent free items which need to be processed. Here
3766 * we process the extent free items and clean up the on disk unlinked
3767 * inode lists. This is separated from the first part of recovery so
3768 * that the root and real-time bitmap inodes can be read in from disk in
3769 * between the two stages. This is necessary so that we can free space
3770 * in the real-time portion of the file system.
3773 xlog_recover_finish(
3777 * Now we're ready to do the transactions needed for the
3778 * rest of recovery. Start with completing all the extent
3779 * free intent records and then process the unlinked inode
3780 * lists. At this point, we essentially run in normal mode
3781 * except that we're still performing recovery actions
3782 * rather than accepting new requests.
3784 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
3786 error = xlog_recover_process_efis(log);
3788 xfs_alert(log->l_mp, "Failed to recover EFIs");
3792 * Sync the log to get all the EFIs out of the AIL.
3793 * This isn't absolutely necessary, but it helps in
3794 * case the unlink transactions would have problems
3795 * pushing the EFIs out of the way.
3797 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
3799 xlog_recover_process_iunlinks(log);
3801 xlog_recover_check_summary(log);
3803 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
3804 log->l_mp->m_logname ? log->l_mp->m_logname
3806 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
3808 xfs_info(log->l_mp, "Ending clean mount");
3816 * Read all of the agf and agi counters and check that they
3817 * are consistent with the superblock counters.
3820 xlog_recover_check_summary(
3827 xfs_agnumber_t agno;
3828 __uint64_t freeblks;
3838 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3839 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
3841 xfs_alert(mp, "%s agf read failed agno %d error %d",
3842 __func__, agno, error);
3844 agfp = XFS_BUF_TO_AGF(agfbp);
3845 freeblks += be32_to_cpu(agfp->agf_freeblks) +
3846 be32_to_cpu(agfp->agf_flcount);
3847 xfs_buf_relse(agfbp);
3850 error = xfs_read_agi(mp, NULL, agno, &agibp);
3852 xfs_alert(mp, "%s agi read failed agno %d error %d",
3853 __func__, agno, error);
3855 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
3857 itotal += be32_to_cpu(agi->agi_count);
3858 ifree += be32_to_cpu(agi->agi_freecount);
3859 xfs_buf_relse(agibp);