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"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_error.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_alloc.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_log_priv.h"
42 #include "xfs_buf_item.h"
43 #include "xfs_log_recover.h"
44 #include "xfs_extfree_item.h"
45 #include "xfs_trans_priv.h"
46 #include "xfs_quota.h"
48 #include "xfs_utils.h"
49 #include "xfs_trace.h"
51 STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *);
52 STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t);
54 STATIC void xlog_recover_check_summary(xlog_t *);
56 #define xlog_recover_check_summary(log)
60 * Sector aligned buffer routines for buffer create/read/write/access
63 /* Number of basic blocks in a log sector */
64 #define xlog_sectbb(log) (1 << (log)->l_sectbb_log)
67 * Verify the given count of basic blocks is valid number of blocks
68 * to specify for an operation involving the given XFS log buffer.
69 * Returns nonzero if the count is valid, 0 otherwise.
73 xlog_buf_bbcount_valid(
77 return bbcount > 0 && bbcount <= log->l_logBBsize;
81 * Allocate a buffer to hold log data. The buffer needs to be able
82 * to map to a range of nbblks basic blocks at any valid (basic
83 * block) offset within the log.
90 if (!xlog_buf_bbcount_valid(log, nbblks)) {
91 xlog_warn("XFS: Invalid block length (0x%x) given for buffer",
93 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
98 * We do log I/O in units of log sectors (a power-of-2
99 * multiple of the basic block size), so we round up the
100 * requested size to acommodate the basic blocks required
101 * for complete log sectors.
103 * In addition, the buffer may be used for a non-sector-
104 * aligned block offset, in which case an I/O of the
105 * requested size could extend beyond the end of the
106 * buffer. If the requested size is only 1 basic block it
107 * will never straddle a sector boundary, so this won't be
108 * an issue. Nor will this be a problem if the log I/O is
109 * done in basic blocks (sector size 1). But otherwise we
110 * extend the buffer by one extra log sector to ensure
111 * there's space to accomodate this possiblility.
113 if (nbblks > 1 && log->l_sectbb_log)
114 nbblks += xlog_sectbb(log);
115 nbblks = round_up(nbblks, xlog_sectbb(log));
117 return xfs_buf_get_noaddr(BBTOB(nbblks), log->l_mp->m_logdev_targp);
136 if (!log->l_sectbb_log)
137 return XFS_BUF_PTR(bp);
139 ptr = XFS_BUF_PTR(bp) + BBTOB((int)blk_no & log->l_sectbb_mask);
140 ASSERT(XFS_BUF_SIZE(bp) >=
141 BBTOB(nbblks + (blk_no & log->l_sectbb_mask)));
147 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
158 if (!xlog_buf_bbcount_valid(log, nbblks)) {
159 xlog_warn("XFS: Invalid block length (0x%x) given for buffer",
161 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
165 blk_no = round_down(blk_no, xlog_sectbb(log));
166 nbblks = round_up(nbblks, xlog_sectbb(log));
169 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
171 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
174 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
175 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
177 xfsbdstrat(log->l_mp, bp);
178 error = xfs_iowait(bp);
180 xfs_ioerror_alert("xlog_bread", log->l_mp,
181 bp, XFS_BUF_ADDR(bp));
195 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
199 *offset = xlog_align(log, blk_no, nbblks, bp);
204 * Write out the buffer at the given block for the given number of blocks.
205 * The buffer is kept locked across the write and is returned locked.
206 * This can only be used for synchronous log writes.
217 if (!xlog_buf_bbcount_valid(log, nbblks)) {
218 xlog_warn("XFS: Invalid block length (0x%x) given for buffer",
220 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
224 blk_no = round_down(blk_no, xlog_sectbb(log));
225 nbblks = round_up(nbblks, xlog_sectbb(log));
228 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
230 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
231 XFS_BUF_ZEROFLAGS(bp);
234 XFS_BUF_PSEMA(bp, PRIBIO);
235 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
236 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
238 if ((error = xfs_bwrite(log->l_mp, bp)))
239 xfs_ioerror_alert("xlog_bwrite", log->l_mp,
240 bp, XFS_BUF_ADDR(bp));
246 * dump debug superblock and log record information
249 xlog_header_check_dump(
251 xlog_rec_header_t *head)
253 cmn_err(CE_DEBUG, "%s: SB : uuid = %pU, fmt = %d\n",
254 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
255 cmn_err(CE_DEBUG, " log : uuid = %pU, fmt = %d\n",
256 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
259 #define xlog_header_check_dump(mp, head)
263 * check log record header for recovery
266 xlog_header_check_recover(
268 xlog_rec_header_t *head)
270 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
273 * IRIX doesn't write the h_fmt field and leaves it zeroed
274 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
275 * a dirty log created in IRIX.
277 if (unlikely(be32_to_cpu(head->h_fmt) != XLOG_FMT)) {
279 "XFS: dirty log written in incompatible format - can't recover");
280 xlog_header_check_dump(mp, head);
281 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
282 XFS_ERRLEVEL_HIGH, mp);
283 return XFS_ERROR(EFSCORRUPTED);
284 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
286 "XFS: dirty log entry has mismatched uuid - can't recover");
287 xlog_header_check_dump(mp, head);
288 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
289 XFS_ERRLEVEL_HIGH, mp);
290 return XFS_ERROR(EFSCORRUPTED);
296 * read the head block of the log and check the header
299 xlog_header_check_mount(
301 xlog_rec_header_t *head)
303 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
305 if (uuid_is_nil(&head->h_fs_uuid)) {
307 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
308 * h_fs_uuid is nil, we assume this log was last mounted
309 * by IRIX and continue.
311 xlog_warn("XFS: nil uuid in log - IRIX style log");
312 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
313 xlog_warn("XFS: log has mismatched uuid - can't recover");
314 xlog_header_check_dump(mp, head);
315 XFS_ERROR_REPORT("xlog_header_check_mount",
316 XFS_ERRLEVEL_HIGH, mp);
317 return XFS_ERROR(EFSCORRUPTED);
326 if (XFS_BUF_GETERROR(bp)) {
328 * We're not going to bother about retrying
329 * this during recovery. One strike!
331 xfs_ioerror_alert("xlog_recover_iodone",
332 bp->b_mount, bp, XFS_BUF_ADDR(bp));
333 xfs_force_shutdown(bp->b_mount, SHUTDOWN_META_IO_ERROR);
336 XFS_BUF_CLR_IODONE_FUNC(bp);
341 * This routine finds (to an approximation) the first block in the physical
342 * log which contains the given cycle. It uses a binary search algorithm.
343 * Note that the algorithm can not be perfect because the disk will not
344 * necessarily be perfect.
347 xlog_find_cycle_start(
350 xfs_daddr_t first_blk,
351 xfs_daddr_t *last_blk,
361 mid_blk = BLK_AVG(first_blk, end_blk);
362 while (mid_blk != first_blk && mid_blk != end_blk) {
363 error = xlog_bread(log, mid_blk, 1, bp, &offset);
366 mid_cycle = xlog_get_cycle(offset);
367 if (mid_cycle == cycle)
368 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
370 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
371 mid_blk = BLK_AVG(first_blk, end_blk);
373 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
374 (mid_blk == end_blk && mid_blk-1 == first_blk));
382 * Check that a range of blocks does not contain stop_on_cycle_no.
383 * Fill in *new_blk with the block offset where such a block is
384 * found, or with -1 (an invalid block number) if there is no such
385 * block in the range. The scan needs to occur from front to back
386 * and the pointer into the region must be updated since a later
387 * routine will need to perform another test.
390 xlog_find_verify_cycle(
392 xfs_daddr_t start_blk,
394 uint stop_on_cycle_no,
395 xfs_daddr_t *new_blk)
401 xfs_caddr_t buf = NULL;
405 * Greedily allocate a buffer big enough to handle the full
406 * range of basic blocks we'll be examining. If that fails,
407 * try a smaller size. We need to be able to read at least
408 * a log sector, or we're out of luck.
410 bufblks = 1 << ffs(nbblks);
411 while (!(bp = xlog_get_bp(log, bufblks))) {
413 if (bufblks < xlog_sectbb(log))
417 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
420 bcount = min(bufblks, (start_blk + nbblks - i));
422 error = xlog_bread(log, i, bcount, bp, &buf);
426 for (j = 0; j < bcount; j++) {
427 cycle = xlog_get_cycle(buf);
428 if (cycle == stop_on_cycle_no) {
445 * Potentially backup over partial log record write.
447 * In the typical case, last_blk is the number of the block directly after
448 * a good log record. Therefore, we subtract one to get the block number
449 * of the last block in the given buffer. extra_bblks contains the number
450 * of blocks we would have read on a previous read. This happens when the
451 * last log record is split over the end of the physical log.
453 * extra_bblks is the number of blocks potentially verified on a previous
454 * call to this routine.
457 xlog_find_verify_log_record(
459 xfs_daddr_t start_blk,
460 xfs_daddr_t *last_blk,
465 xfs_caddr_t offset = NULL;
466 xlog_rec_header_t *head = NULL;
469 int num_blks = *last_blk - start_blk;
472 ASSERT(start_blk != 0 || *last_blk != start_blk);
474 if (!(bp = xlog_get_bp(log, num_blks))) {
475 if (!(bp = xlog_get_bp(log, 1)))
479 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
482 offset += ((num_blks - 1) << BBSHIFT);
485 for (i = (*last_blk) - 1; i >= 0; i--) {
487 /* valid log record not found */
489 "XFS: Log inconsistent (didn't find previous header)");
491 error = XFS_ERROR(EIO);
496 error = xlog_bread(log, i, 1, bp, &offset);
501 head = (xlog_rec_header_t *)offset;
503 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(head->h_magicno))
511 * We hit the beginning of the physical log & still no header. Return
512 * to caller. If caller can handle a return of -1, then this routine
513 * will be called again for the end of the physical log.
521 * We have the final block of the good log (the first block
522 * of the log record _before_ the head. So we check the uuid.
524 if ((error = xlog_header_check_mount(log->l_mp, head)))
528 * We may have found a log record header before we expected one.
529 * last_blk will be the 1st block # with a given cycle #. We may end
530 * up reading an entire log record. In this case, we don't want to
531 * reset last_blk. Only when last_blk points in the middle of a log
532 * record do we update last_blk.
534 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
535 uint h_size = be32_to_cpu(head->h_size);
537 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
538 if (h_size % XLOG_HEADER_CYCLE_SIZE)
544 if (*last_blk - i + extra_bblks !=
545 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
554 * Head is defined to be the point of the log where the next log write
555 * write could go. This means that incomplete LR writes at the end are
556 * eliminated when calculating the head. We aren't guaranteed that previous
557 * LR have complete transactions. We only know that a cycle number of
558 * current cycle number -1 won't be present in the log if we start writing
559 * from our current block number.
561 * last_blk contains the block number of the first block with a given
564 * Return: zero if normal, non-zero if error.
569 xfs_daddr_t *return_head_blk)
573 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
575 uint first_half_cycle, last_half_cycle;
577 int error, log_bbnum = log->l_logBBsize;
579 /* Is the end of the log device zeroed? */
580 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
581 *return_head_blk = first_blk;
583 /* Is the whole lot zeroed? */
585 /* Linux XFS shouldn't generate totally zeroed logs -
586 * mkfs etc write a dummy unmount record to a fresh
587 * log so we can store the uuid in there
589 xlog_warn("XFS: totally zeroed log");
594 xlog_warn("XFS: empty log check failed");
598 first_blk = 0; /* get cycle # of 1st block */
599 bp = xlog_get_bp(log, 1);
603 error = xlog_bread(log, 0, 1, bp, &offset);
607 first_half_cycle = xlog_get_cycle(offset);
609 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
610 error = xlog_bread(log, last_blk, 1, bp, &offset);
614 last_half_cycle = xlog_get_cycle(offset);
615 ASSERT(last_half_cycle != 0);
618 * If the 1st half cycle number is equal to the last half cycle number,
619 * then the entire log is stamped with the same cycle number. In this
620 * case, head_blk can't be set to zero (which makes sense). The below
621 * math doesn't work out properly with head_blk equal to zero. Instead,
622 * we set it to log_bbnum which is an invalid block number, but this
623 * value makes the math correct. If head_blk doesn't changed through
624 * all the tests below, *head_blk is set to zero at the very end rather
625 * than log_bbnum. In a sense, log_bbnum and zero are the same block
626 * in a circular file.
628 if (first_half_cycle == last_half_cycle) {
630 * In this case we believe that the entire log should have
631 * cycle number last_half_cycle. We need to scan backwards
632 * from the end verifying that there are no holes still
633 * containing last_half_cycle - 1. If we find such a hole,
634 * then the start of that hole will be the new head. The
635 * simple case looks like
636 * x | x ... | x - 1 | x
637 * Another case that fits this picture would be
638 * x | x + 1 | x ... | x
639 * In this case the head really is somewhere at the end of the
640 * log, as one of the latest writes at the beginning was
643 * x | x + 1 | x ... | x - 1 | x
644 * This is really the combination of the above two cases, and
645 * the head has to end up at the start of the x-1 hole at the
648 * In the 256k log case, we will read from the beginning to the
649 * end of the log and search for cycle numbers equal to x-1.
650 * We don't worry about the x+1 blocks that we encounter,
651 * because we know that they cannot be the head since the log
654 head_blk = log_bbnum;
655 stop_on_cycle = last_half_cycle - 1;
658 * In this case we want to find the first block with cycle
659 * number matching last_half_cycle. We expect the log to be
661 * x + 1 ... | x ... | x
662 * The first block with cycle number x (last_half_cycle) will
663 * be where the new head belongs. First we do a binary search
664 * for the first occurrence of last_half_cycle. The binary
665 * search may not be totally accurate, so then we scan back
666 * from there looking for occurrences of last_half_cycle before
667 * us. If that backwards scan wraps around the beginning of
668 * the log, then we look for occurrences of last_half_cycle - 1
669 * at the end of the log. The cases we're looking for look
671 * v binary search stopped here
672 * x + 1 ... | x | x + 1 | x ... | x
673 * ^ but we want to locate this spot
675 * <---------> less than scan distance
676 * x + 1 ... | x ... | x - 1 | x
677 * ^ we want to locate this spot
679 stop_on_cycle = last_half_cycle;
680 if ((error = xlog_find_cycle_start(log, bp, first_blk,
681 &head_blk, last_half_cycle)))
686 * Now validate the answer. Scan back some number of maximum possible
687 * blocks and make sure each one has the expected cycle number. The
688 * maximum is determined by the total possible amount of buffering
689 * in the in-core log. The following number can be made tighter if
690 * we actually look at the block size of the filesystem.
692 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
693 if (head_blk >= num_scan_bblks) {
695 * We are guaranteed that the entire check can be performed
698 start_blk = head_blk - num_scan_bblks;
699 if ((error = xlog_find_verify_cycle(log,
700 start_blk, num_scan_bblks,
701 stop_on_cycle, &new_blk)))
705 } else { /* need to read 2 parts of log */
707 * We are going to scan backwards in the log in two parts.
708 * First we scan the physical end of the log. In this part
709 * of the log, we are looking for blocks with cycle number
710 * last_half_cycle - 1.
711 * If we find one, then we know that the log starts there, as
712 * we've found a hole that didn't get written in going around
713 * the end of the physical log. The simple case for this is
714 * x + 1 ... | x ... | x - 1 | x
715 * <---------> less than scan distance
716 * If all of the blocks at the end of the log have cycle number
717 * last_half_cycle, then we check the blocks at the start of
718 * the log looking for occurrences of last_half_cycle. If we
719 * find one, then our current estimate for the location of the
720 * first occurrence of last_half_cycle is wrong and we move
721 * back to the hole we've found. This case looks like
722 * x + 1 ... | x | x + 1 | x ...
723 * ^ binary search stopped here
724 * Another case we need to handle that only occurs in 256k
726 * x + 1 ... | x ... | x+1 | x ...
727 * ^ binary search stops here
728 * In a 256k log, the scan at the end of the log will see the
729 * x + 1 blocks. We need to skip past those since that is
730 * certainly not the head of the log. By searching for
731 * last_half_cycle-1 we accomplish that.
733 ASSERT(head_blk <= INT_MAX &&
734 (xfs_daddr_t) num_scan_bblks >= head_blk);
735 start_blk = log_bbnum - (num_scan_bblks - head_blk);
736 if ((error = xlog_find_verify_cycle(log, start_blk,
737 num_scan_bblks - (int)head_blk,
738 (stop_on_cycle - 1), &new_blk)))
746 * Scan beginning of log now. The last part of the physical
747 * log is good. This scan needs to verify that it doesn't find
748 * the last_half_cycle.
751 ASSERT(head_blk <= INT_MAX);
752 if ((error = xlog_find_verify_cycle(log,
753 start_blk, (int)head_blk,
754 stop_on_cycle, &new_blk)))
762 * Now we need to make sure head_blk is not pointing to a block in
763 * the middle of a log record.
765 num_scan_bblks = XLOG_REC_SHIFT(log);
766 if (head_blk >= num_scan_bblks) {
767 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
769 /* start ptr at last block ptr before head_blk */
770 if ((error = xlog_find_verify_log_record(log, start_blk,
771 &head_blk, 0)) == -1) {
772 error = XFS_ERROR(EIO);
778 ASSERT(head_blk <= INT_MAX);
779 if ((error = xlog_find_verify_log_record(log, start_blk,
780 &head_blk, 0)) == -1) {
781 /* We hit the beginning of the log during our search */
782 start_blk = log_bbnum - (num_scan_bblks - head_blk);
784 ASSERT(start_blk <= INT_MAX &&
785 (xfs_daddr_t) log_bbnum-start_blk >= 0);
786 ASSERT(head_blk <= INT_MAX);
787 if ((error = xlog_find_verify_log_record(log,
789 (int)head_blk)) == -1) {
790 error = XFS_ERROR(EIO);
794 if (new_blk != log_bbnum)
801 if (head_blk == log_bbnum)
802 *return_head_blk = 0;
804 *return_head_blk = head_blk;
806 * When returning here, we have a good block number. Bad block
807 * means that during a previous crash, we didn't have a clean break
808 * from cycle number N to cycle number N-1. In this case, we need
809 * to find the first block with cycle number N-1.
817 xlog_warn("XFS: failed to find log head");
822 * Find the sync block number or the tail of the log.
824 * This will be the block number of the last record to have its
825 * associated buffers synced to disk. Every log record header has
826 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
827 * to get a sync block number. The only concern is to figure out which
828 * log record header to believe.
830 * The following algorithm uses the log record header with the largest
831 * lsn. The entire log record does not need to be valid. We only care
832 * that the header is valid.
834 * We could speed up search by using current head_blk buffer, but it is not
840 xfs_daddr_t *head_blk,
841 xfs_daddr_t *tail_blk)
843 xlog_rec_header_t *rhead;
844 xlog_op_header_t *op_head;
845 xfs_caddr_t offset = NULL;
848 xfs_daddr_t umount_data_blk;
849 xfs_daddr_t after_umount_blk;
856 * Find previous log record
858 if ((error = xlog_find_head(log, head_blk)))
861 bp = xlog_get_bp(log, 1);
864 if (*head_blk == 0) { /* special case */
865 error = xlog_bread(log, 0, 1, bp, &offset);
869 if (xlog_get_cycle(offset) == 0) {
871 /* leave all other log inited values alone */
877 * Search backwards looking for log record header block
879 ASSERT(*head_blk < INT_MAX);
880 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
881 error = xlog_bread(log, i, 1, bp, &offset);
885 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(*(__be32 *)offset)) {
891 * If we haven't found the log record header block, start looking
892 * again from the end of the physical log. XXXmiken: There should be
893 * a check here to make sure we didn't search more than N blocks in
897 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
898 error = xlog_bread(log, i, 1, bp, &offset);
902 if (XLOG_HEADER_MAGIC_NUM ==
903 be32_to_cpu(*(__be32 *)offset)) {
910 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
912 return XFS_ERROR(EIO);
915 /* find blk_no of tail of log */
916 rhead = (xlog_rec_header_t *)offset;
917 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
920 * Reset log values according to the state of the log when we
921 * crashed. In the case where head_blk == 0, we bump curr_cycle
922 * one because the next write starts a new cycle rather than
923 * continuing the cycle of the last good log record. At this
924 * point we have guaranteed that all partial log records have been
925 * accounted for. Therefore, we know that the last good log record
926 * written was complete and ended exactly on the end boundary
927 * of the physical log.
929 log->l_prev_block = i;
930 log->l_curr_block = (int)*head_blk;
931 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
934 log->l_tail_lsn = be64_to_cpu(rhead->h_tail_lsn);
935 log->l_last_sync_lsn = be64_to_cpu(rhead->h_lsn);
936 log->l_grant_reserve_cycle = log->l_curr_cycle;
937 log->l_grant_reserve_bytes = BBTOB(log->l_curr_block);
938 log->l_grant_write_cycle = log->l_curr_cycle;
939 log->l_grant_write_bytes = BBTOB(log->l_curr_block);
942 * Look for unmount record. If we find it, then we know there
943 * was a clean unmount. Since 'i' could be the last block in
944 * the physical log, we convert to a log block before comparing
947 * Save the current tail lsn to use to pass to
948 * xlog_clear_stale_blocks() below. We won't want to clear the
949 * unmount record if there is one, so we pass the lsn of the
950 * unmount record rather than the block after it.
952 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
953 int h_size = be32_to_cpu(rhead->h_size);
954 int h_version = be32_to_cpu(rhead->h_version);
956 if ((h_version & XLOG_VERSION_2) &&
957 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
958 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
959 if (h_size % XLOG_HEADER_CYCLE_SIZE)
967 after_umount_blk = (i + hblks + (int)
968 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
969 tail_lsn = log->l_tail_lsn;
970 if (*head_blk == after_umount_blk &&
971 be32_to_cpu(rhead->h_num_logops) == 1) {
972 umount_data_blk = (i + hblks) % log->l_logBBsize;
973 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
977 op_head = (xlog_op_header_t *)offset;
978 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
980 * Set tail and last sync so that newly written
981 * log records will point recovery to after the
982 * current unmount record.
985 xlog_assign_lsn(log->l_curr_cycle,
987 log->l_last_sync_lsn =
988 xlog_assign_lsn(log->l_curr_cycle,
990 *tail_blk = after_umount_blk;
993 * Note that the unmount was clean. If the unmount
994 * was not clean, we need to know this to rebuild the
995 * superblock counters from the perag headers if we
996 * have a filesystem using non-persistent counters.
998 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1003 * Make sure that there are no blocks in front of the head
1004 * with the same cycle number as the head. This can happen
1005 * because we allow multiple outstanding log writes concurrently,
1006 * and the later writes might make it out before earlier ones.
1008 * We use the lsn from before modifying it so that we'll never
1009 * overwrite the unmount record after a clean unmount.
1011 * Do this only if we are going to recover the filesystem
1013 * NOTE: This used to say "if (!readonly)"
1014 * However on Linux, we can & do recover a read-only filesystem.
1015 * We only skip recovery if NORECOVERY is specified on mount,
1016 * in which case we would not be here.
1018 * But... if the -device- itself is readonly, just skip this.
1019 * We can't recover this device anyway, so it won't matter.
1021 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
1022 error = xlog_clear_stale_blocks(log, tail_lsn);
1028 xlog_warn("XFS: failed to locate log tail");
1033 * Is the log zeroed at all?
1035 * The last binary search should be changed to perform an X block read
1036 * once X becomes small enough. You can then search linearly through
1037 * the X blocks. This will cut down on the number of reads we need to do.
1039 * If the log is partially zeroed, this routine will pass back the blkno
1040 * of the first block with cycle number 0. It won't have a complete LR
1044 * 0 => the log is completely written to
1045 * -1 => use *blk_no as the first block of the log
1046 * >0 => error has occurred
1051 xfs_daddr_t *blk_no)
1055 uint first_cycle, last_cycle;
1056 xfs_daddr_t new_blk, last_blk, start_blk;
1057 xfs_daddr_t num_scan_bblks;
1058 int error, log_bbnum = log->l_logBBsize;
1062 /* check totally zeroed log */
1063 bp = xlog_get_bp(log, 1);
1066 error = xlog_bread(log, 0, 1, bp, &offset);
1070 first_cycle = xlog_get_cycle(offset);
1071 if (first_cycle == 0) { /* completely zeroed log */
1077 /* check partially zeroed log */
1078 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1082 last_cycle = xlog_get_cycle(offset);
1083 if (last_cycle != 0) { /* log completely written to */
1086 } else if (first_cycle != 1) {
1088 * If the cycle of the last block is zero, the cycle of
1089 * the first block must be 1. If it's not, maybe we're
1090 * not looking at a log... Bail out.
1092 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1093 return XFS_ERROR(EINVAL);
1096 /* we have a partially zeroed log */
1097 last_blk = log_bbnum-1;
1098 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1102 * Validate the answer. Because there is no way to guarantee that
1103 * the entire log is made up of log records which are the same size,
1104 * we scan over the defined maximum blocks. At this point, the maximum
1105 * is not chosen to mean anything special. XXXmiken
1107 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1108 ASSERT(num_scan_bblks <= INT_MAX);
1110 if (last_blk < num_scan_bblks)
1111 num_scan_bblks = last_blk;
1112 start_blk = last_blk - num_scan_bblks;
1115 * We search for any instances of cycle number 0 that occur before
1116 * our current estimate of the head. What we're trying to detect is
1117 * 1 ... | 0 | 1 | 0...
1118 * ^ binary search ends here
1120 if ((error = xlog_find_verify_cycle(log, start_blk,
1121 (int)num_scan_bblks, 0, &new_blk)))
1127 * Potentially backup over partial log record write. We don't need
1128 * to search the end of the log because we know it is zero.
1130 if ((error = xlog_find_verify_log_record(log, start_blk,
1131 &last_blk, 0)) == -1) {
1132 error = XFS_ERROR(EIO);
1146 * These are simple subroutines used by xlog_clear_stale_blocks() below
1147 * to initialize a buffer full of empty log record headers and write
1148 * them into the log.
1159 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1161 memset(buf, 0, BBSIZE);
1162 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1163 recp->h_cycle = cpu_to_be32(cycle);
1164 recp->h_version = cpu_to_be32(
1165 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1166 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1167 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1168 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1169 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1173 xlog_write_log_records(
1184 int sectbb = xlog_sectbb(log);
1185 int end_block = start_block + blocks;
1191 * Greedily allocate a buffer big enough to handle the full
1192 * range of basic blocks to be written. If that fails, try
1193 * a smaller size. We need to be able to write at least a
1194 * log sector, or we're out of luck.
1196 bufblks = 1 << ffs(blocks);
1197 while (!(bp = xlog_get_bp(log, bufblks))) {
1199 if (bufblks < xlog_sectbb(log))
1203 /* We may need to do a read at the start to fill in part of
1204 * the buffer in the starting sector not covered by the first
1207 balign = round_down(start_block, sectbb);
1208 if (balign != start_block) {
1209 error = xlog_bread_noalign(log, start_block, 1, bp);
1213 j = start_block - balign;
1216 for (i = start_block; i < end_block; i += bufblks) {
1217 int bcount, endcount;
1219 bcount = min(bufblks, end_block - start_block);
1220 endcount = bcount - j;
1222 /* We may need to do a read at the end to fill in part of
1223 * the buffer in the final sector not covered by the write.
1224 * If this is the same sector as the above read, skip it.
1226 ealign = round_down(end_block, sectbb);
1227 if (j == 0 && (start_block + endcount > ealign)) {
1228 offset = XFS_BUF_PTR(bp);
1229 balign = BBTOB(ealign - start_block);
1230 error = XFS_BUF_SET_PTR(bp, offset + balign,
1235 error = xlog_bread_noalign(log, ealign, sectbb, bp);
1239 error = XFS_BUF_SET_PTR(bp, offset, bufblks);
1244 offset = xlog_align(log, start_block, endcount, bp);
1245 for (; j < endcount; j++) {
1246 xlog_add_record(log, offset, cycle, i+j,
1247 tail_cycle, tail_block);
1250 error = xlog_bwrite(log, start_block, endcount, bp);
1253 start_block += endcount;
1263 * This routine is called to blow away any incomplete log writes out
1264 * in front of the log head. We do this so that we won't become confused
1265 * if we come up, write only a little bit more, and then crash again.
1266 * If we leave the partial log records out there, this situation could
1267 * cause us to think those partial writes are valid blocks since they
1268 * have the current cycle number. We get rid of them by overwriting them
1269 * with empty log records with the old cycle number rather than the
1272 * The tail lsn is passed in rather than taken from
1273 * the log so that we will not write over the unmount record after a
1274 * clean unmount in a 512 block log. Doing so would leave the log without
1275 * any valid log records in it until a new one was written. If we crashed
1276 * during that time we would not be able to recover.
1279 xlog_clear_stale_blocks(
1283 int tail_cycle, head_cycle;
1284 int tail_block, head_block;
1285 int tail_distance, max_distance;
1289 tail_cycle = CYCLE_LSN(tail_lsn);
1290 tail_block = BLOCK_LSN(tail_lsn);
1291 head_cycle = log->l_curr_cycle;
1292 head_block = log->l_curr_block;
1295 * Figure out the distance between the new head of the log
1296 * and the tail. We want to write over any blocks beyond the
1297 * head that we may have written just before the crash, but
1298 * we don't want to overwrite the tail of the log.
1300 if (head_cycle == tail_cycle) {
1302 * The tail is behind the head in the physical log,
1303 * so the distance from the head to the tail is the
1304 * distance from the head to the end of the log plus
1305 * the distance from the beginning of the log to the
1308 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1309 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1310 XFS_ERRLEVEL_LOW, log->l_mp);
1311 return XFS_ERROR(EFSCORRUPTED);
1313 tail_distance = tail_block + (log->l_logBBsize - head_block);
1316 * The head is behind the tail in the physical log,
1317 * so the distance from the head to the tail is just
1318 * the tail block minus the head block.
1320 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1321 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1322 XFS_ERRLEVEL_LOW, log->l_mp);
1323 return XFS_ERROR(EFSCORRUPTED);
1325 tail_distance = tail_block - head_block;
1329 * If the head is right up against the tail, we can't clear
1332 if (tail_distance <= 0) {
1333 ASSERT(tail_distance == 0);
1337 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1339 * Take the smaller of the maximum amount of outstanding I/O
1340 * we could have and the distance to the tail to clear out.
1341 * We take the smaller so that we don't overwrite the tail and
1342 * we don't waste all day writing from the head to the tail
1345 max_distance = MIN(max_distance, tail_distance);
1347 if ((head_block + max_distance) <= log->l_logBBsize) {
1349 * We can stomp all the blocks we need to without
1350 * wrapping around the end of the log. Just do it
1351 * in a single write. Use the cycle number of the
1352 * current cycle minus one so that the log will look like:
1355 error = xlog_write_log_records(log, (head_cycle - 1),
1356 head_block, max_distance, tail_cycle,
1362 * We need to wrap around the end of the physical log in
1363 * order to clear all the blocks. Do it in two separate
1364 * I/Os. The first write should be from the head to the
1365 * end of the physical log, and it should use the current
1366 * cycle number minus one just like above.
1368 distance = log->l_logBBsize - head_block;
1369 error = xlog_write_log_records(log, (head_cycle - 1),
1370 head_block, distance, tail_cycle,
1377 * Now write the blocks at the start of the physical log.
1378 * This writes the remainder of the blocks we want to clear.
1379 * It uses the current cycle number since we're now on the
1380 * same cycle as the head so that we get:
1381 * n ... n ... | n - 1 ...
1382 * ^^^^^ blocks we're writing
1384 distance = max_distance - (log->l_logBBsize - head_block);
1385 error = xlog_write_log_records(log, head_cycle, 0, distance,
1386 tail_cycle, tail_block);
1394 /******************************************************************************
1396 * Log recover routines
1398 ******************************************************************************
1401 STATIC xlog_recover_t *
1402 xlog_recover_find_tid(
1403 struct hlist_head *head,
1406 xlog_recover_t *trans;
1407 struct hlist_node *n;
1409 hlist_for_each_entry(trans, n, head, r_list) {
1410 if (trans->r_log_tid == tid)
1417 xlog_recover_new_tid(
1418 struct hlist_head *head,
1422 xlog_recover_t *trans;
1424 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1425 trans->r_log_tid = tid;
1427 INIT_LIST_HEAD(&trans->r_itemq);
1429 INIT_HLIST_NODE(&trans->r_list);
1430 hlist_add_head(&trans->r_list, head);
1434 xlog_recover_add_item(
1435 struct list_head *head)
1437 xlog_recover_item_t *item;
1439 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1440 INIT_LIST_HEAD(&item->ri_list);
1441 list_add_tail(&item->ri_list, head);
1445 xlog_recover_add_to_cont_trans(
1447 xlog_recover_t *trans,
1451 xlog_recover_item_t *item;
1452 xfs_caddr_t ptr, old_ptr;
1455 if (list_empty(&trans->r_itemq)) {
1456 /* finish copying rest of trans header */
1457 xlog_recover_add_item(&trans->r_itemq);
1458 ptr = (xfs_caddr_t) &trans->r_theader +
1459 sizeof(xfs_trans_header_t) - len;
1460 memcpy(ptr, dp, len); /* d, s, l */
1463 /* take the tail entry */
1464 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1466 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1467 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1469 ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
1470 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1471 item->ri_buf[item->ri_cnt-1].i_len += len;
1472 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1473 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
1478 * The next region to add is the start of a new region. It could be
1479 * a whole region or it could be the first part of a new region. Because
1480 * of this, the assumption here is that the type and size fields of all
1481 * format structures fit into the first 32 bits of the structure.
1483 * This works because all regions must be 32 bit aligned. Therefore, we
1484 * either have both fields or we have neither field. In the case we have
1485 * neither field, the data part of the region is zero length. We only have
1486 * a log_op_header and can throw away the header since a new one will appear
1487 * later. If we have at least 4 bytes, then we can determine how many regions
1488 * will appear in the current log item.
1491 xlog_recover_add_to_trans(
1493 xlog_recover_t *trans,
1497 xfs_inode_log_format_t *in_f; /* any will do */
1498 xlog_recover_item_t *item;
1503 if (list_empty(&trans->r_itemq)) {
1504 /* we need to catch log corruptions here */
1505 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
1506 xlog_warn("XFS: xlog_recover_add_to_trans: "
1507 "bad header magic number");
1509 return XFS_ERROR(EIO);
1511 if (len == sizeof(xfs_trans_header_t))
1512 xlog_recover_add_item(&trans->r_itemq);
1513 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1517 ptr = kmem_alloc(len, KM_SLEEP);
1518 memcpy(ptr, dp, len);
1519 in_f = (xfs_inode_log_format_t *)ptr;
1521 /* take the tail entry */
1522 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1523 if (item->ri_total != 0 &&
1524 item->ri_total == item->ri_cnt) {
1525 /* tail item is in use, get a new one */
1526 xlog_recover_add_item(&trans->r_itemq);
1527 item = list_entry(trans->r_itemq.prev,
1528 xlog_recover_item_t, ri_list);
1531 if (item->ri_total == 0) { /* first region to be added */
1532 if (in_f->ilf_size == 0 ||
1533 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
1535 "XFS: bad number of regions (%d) in inode log format",
1538 return XFS_ERROR(EIO);
1541 item->ri_total = in_f->ilf_size;
1543 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1546 ASSERT(item->ri_total > item->ri_cnt);
1547 /* Description region is ri_buf[0] */
1548 item->ri_buf[item->ri_cnt].i_addr = ptr;
1549 item->ri_buf[item->ri_cnt].i_len = len;
1551 trace_xfs_log_recover_item_add(log, trans, item, 0);
1556 * Sort the log items in the transaction. Cancelled buffers need
1557 * to be put first so they are processed before any items that might
1558 * modify the buffers. If they are cancelled, then the modifications
1559 * don't need to be replayed.
1562 xlog_recover_reorder_trans(
1564 xlog_recover_t *trans,
1567 xlog_recover_item_t *item, *n;
1568 LIST_HEAD(sort_list);
1570 list_splice_init(&trans->r_itemq, &sort_list);
1571 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1572 xfs_buf_log_format_t *buf_f;
1574 buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr;
1576 switch (ITEM_TYPE(item)) {
1578 if (!(buf_f->blf_flags & XFS_BLI_CANCEL)) {
1579 trace_xfs_log_recover_item_reorder_head(log,
1581 list_move(&item->ri_list, &trans->r_itemq);
1586 case XFS_LI_QUOTAOFF:
1589 trace_xfs_log_recover_item_reorder_tail(log,
1591 list_move_tail(&item->ri_list, &trans->r_itemq);
1595 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1597 return XFS_ERROR(EIO);
1600 ASSERT(list_empty(&sort_list));
1605 * Build up the table of buf cancel records so that we don't replay
1606 * cancelled data in the second pass. For buffer records that are
1607 * not cancel records, there is nothing to do here so we just return.
1609 * If we get a cancel record which is already in the table, this indicates
1610 * that the buffer was cancelled multiple times. In order to ensure
1611 * that during pass 2 we keep the record in the table until we reach its
1612 * last occurrence in the log, we keep a reference count in the cancel
1613 * record in the table to tell us how many times we expect to see this
1614 * record during the second pass.
1617 xlog_recover_do_buffer_pass1(
1619 xfs_buf_log_format_t *buf_f)
1621 xfs_buf_cancel_t *bcp;
1622 xfs_buf_cancel_t *nextp;
1623 xfs_buf_cancel_t *prevp;
1624 xfs_buf_cancel_t **bucket;
1625 xfs_daddr_t blkno = 0;
1629 switch (buf_f->blf_type) {
1631 blkno = buf_f->blf_blkno;
1632 len = buf_f->blf_len;
1633 flags = buf_f->blf_flags;
1638 * If this isn't a cancel buffer item, then just return.
1640 if (!(flags & XFS_BLI_CANCEL)) {
1641 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
1646 * Insert an xfs_buf_cancel record into the hash table of
1647 * them. If there is already an identical record, bump
1648 * its reference count.
1650 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1651 XLOG_BC_TABLE_SIZE];
1653 * If the hash bucket is empty then just insert a new record into
1656 if (*bucket == NULL) {
1657 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1659 bcp->bc_blkno = blkno;
1661 bcp->bc_refcount = 1;
1662 bcp->bc_next = NULL;
1668 * The hash bucket is not empty, so search for duplicates of our
1669 * record. If we find one them just bump its refcount. If not
1670 * then add us at the end of the list.
1674 while (nextp != NULL) {
1675 if (nextp->bc_blkno == blkno && nextp->bc_len == len) {
1676 nextp->bc_refcount++;
1677 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
1681 nextp = nextp->bc_next;
1683 ASSERT(prevp != NULL);
1684 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1686 bcp->bc_blkno = blkno;
1688 bcp->bc_refcount = 1;
1689 bcp->bc_next = NULL;
1690 prevp->bc_next = bcp;
1691 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
1695 * Check to see whether the buffer being recovered has a corresponding
1696 * entry in the buffer cancel record table. If it does then return 1
1697 * so that it will be cancelled, otherwise return 0. If the buffer is
1698 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
1699 * the refcount on the entry in the table and remove it from the table
1700 * if this is the last reference.
1702 * We remove the cancel record from the table when we encounter its
1703 * last occurrence in the log so that if the same buffer is re-used
1704 * again after its last cancellation we actually replay the changes
1705 * made at that point.
1708 xlog_check_buffer_cancelled(
1714 xfs_buf_cancel_t *bcp;
1715 xfs_buf_cancel_t *prevp;
1716 xfs_buf_cancel_t **bucket;
1718 if (log->l_buf_cancel_table == NULL) {
1720 * There is nothing in the table built in pass one,
1721 * so this buffer must not be cancelled.
1723 ASSERT(!(flags & XFS_BLI_CANCEL));
1727 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1728 XLOG_BC_TABLE_SIZE];
1732 * There is no corresponding entry in the table built
1733 * in pass one, so this buffer has not been cancelled.
1735 ASSERT(!(flags & XFS_BLI_CANCEL));
1740 * Search for an entry in the buffer cancel table that
1741 * matches our buffer.
1744 while (bcp != NULL) {
1745 if (bcp->bc_blkno == blkno && bcp->bc_len == len) {
1747 * We've go a match, so return 1 so that the
1748 * recovery of this buffer is cancelled.
1749 * If this buffer is actually a buffer cancel
1750 * log item, then decrement the refcount on the
1751 * one in the table and remove it if this is the
1754 if (flags & XFS_BLI_CANCEL) {
1756 if (bcp->bc_refcount == 0) {
1757 if (prevp == NULL) {
1758 *bucket = bcp->bc_next;
1760 prevp->bc_next = bcp->bc_next;
1771 * We didn't find a corresponding entry in the table, so
1772 * return 0 so that the buffer is NOT cancelled.
1774 ASSERT(!(flags & XFS_BLI_CANCEL));
1779 xlog_recover_do_buffer_pass2(
1781 xfs_buf_log_format_t *buf_f)
1783 xfs_daddr_t blkno = 0;
1787 switch (buf_f->blf_type) {
1789 blkno = buf_f->blf_blkno;
1790 flags = buf_f->blf_flags;
1791 len = buf_f->blf_len;
1795 return xlog_check_buffer_cancelled(log, blkno, len, flags);
1799 * Perform recovery for a buffer full of inodes. In these buffers,
1800 * the only data which should be recovered is that which corresponds
1801 * to the di_next_unlinked pointers in the on disk inode structures.
1802 * The rest of the data for the inodes is always logged through the
1803 * inodes themselves rather than the inode buffer and is recovered
1804 * in xlog_recover_do_inode_trans().
1806 * The only time when buffers full of inodes are fully recovered is
1807 * when the buffer is full of newly allocated inodes. In this case
1808 * the buffer will not be marked as an inode buffer and so will be
1809 * sent to xlog_recover_do_reg_buffer() below during recovery.
1812 xlog_recover_do_inode_buffer(
1814 xlog_recover_item_t *item,
1816 xfs_buf_log_format_t *buf_f)
1824 int next_unlinked_offset;
1826 xfs_agino_t *logged_nextp;
1827 xfs_agino_t *buffer_nextp;
1828 unsigned int *data_map = NULL;
1829 unsigned int map_size = 0;
1831 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
1833 switch (buf_f->blf_type) {
1835 data_map = buf_f->blf_data_map;
1836 map_size = buf_f->blf_map_size;
1840 * Set the variables corresponding to the current region to
1841 * 0 so that we'll initialize them on the first pass through
1849 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1850 for (i = 0; i < inodes_per_buf; i++) {
1851 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1852 offsetof(xfs_dinode_t, di_next_unlinked);
1854 while (next_unlinked_offset >=
1855 (reg_buf_offset + reg_buf_bytes)) {
1857 * The next di_next_unlinked field is beyond
1858 * the current logged region. Find the next
1859 * logged region that contains or is beyond
1860 * the current di_next_unlinked field.
1863 bit = xfs_next_bit(data_map, map_size, bit);
1866 * If there are no more logged regions in the
1867 * buffer, then we're done.
1873 nbits = xfs_contig_bits(data_map, map_size,
1876 reg_buf_offset = bit << XFS_BLI_SHIFT;
1877 reg_buf_bytes = nbits << XFS_BLI_SHIFT;
1882 * If the current logged region starts after the current
1883 * di_next_unlinked field, then move on to the next
1884 * di_next_unlinked field.
1886 if (next_unlinked_offset < reg_buf_offset) {
1890 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1891 ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0);
1892 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1895 * The current logged region contains a copy of the
1896 * current di_next_unlinked field. Extract its value
1897 * and copy it to the buffer copy.
1899 logged_nextp = (xfs_agino_t *)
1900 ((char *)(item->ri_buf[item_index].i_addr) +
1901 (next_unlinked_offset - reg_buf_offset));
1902 if (unlikely(*logged_nextp == 0)) {
1903 xfs_fs_cmn_err(CE_ALERT, mp,
1904 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1906 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1907 XFS_ERRLEVEL_LOW, mp);
1908 return XFS_ERROR(EFSCORRUPTED);
1911 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1912 next_unlinked_offset);
1913 *buffer_nextp = *logged_nextp;
1920 * Perform a 'normal' buffer recovery. Each logged region of the
1921 * buffer should be copied over the corresponding region in the
1922 * given buffer. The bitmap in the buf log format structure indicates
1923 * where to place the logged data.
1927 xlog_recover_do_reg_buffer(
1928 struct xfs_mount *mp,
1929 xlog_recover_item_t *item,
1931 xfs_buf_log_format_t *buf_f)
1936 unsigned int *data_map = NULL;
1937 unsigned int map_size = 0;
1940 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
1942 switch (buf_f->blf_type) {
1944 data_map = buf_f->blf_data_map;
1945 map_size = buf_f->blf_map_size;
1949 i = 1; /* 0 is the buf format structure */
1951 bit = xfs_next_bit(data_map, map_size, bit);
1954 nbits = xfs_contig_bits(data_map, map_size, bit);
1956 ASSERT(item->ri_buf[i].i_addr != NULL);
1957 ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0);
1958 ASSERT(XFS_BUF_COUNT(bp) >=
1959 ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT));
1962 * Do a sanity check if this is a dquot buffer. Just checking
1963 * the first dquot in the buffer should do. XXXThis is
1964 * probably a good thing to do for other buf types also.
1967 if (buf_f->blf_flags &
1968 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
1969 if (item->ri_buf[i].i_addr == NULL) {
1971 "XFS: NULL dquot in %s.", __func__);
1974 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
1976 "XFS: dquot too small (%d) in %s.",
1977 item->ri_buf[i].i_len, __func__);
1980 error = xfs_qm_dqcheck((xfs_disk_dquot_t *)
1981 item->ri_buf[i].i_addr,
1982 -1, 0, XFS_QMOPT_DOWARN,
1983 "dquot_buf_recover");
1988 memcpy(xfs_buf_offset(bp,
1989 (uint)bit << XFS_BLI_SHIFT), /* dest */
1990 item->ri_buf[i].i_addr, /* source */
1991 nbits<<XFS_BLI_SHIFT); /* length */
1997 /* Shouldn't be any more regions */
1998 ASSERT(i == item->ri_total);
2002 * Do some primitive error checking on ondisk dquot data structures.
2006 xfs_disk_dquot_t *ddq,
2008 uint type, /* used only when IO_dorepair is true */
2012 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
2016 * We can encounter an uninitialized dquot buffer for 2 reasons:
2017 * 1. If we crash while deleting the quotainode(s), and those blks got
2018 * used for user data. This is because we take the path of regular
2019 * file deletion; however, the size field of quotainodes is never
2020 * updated, so all the tricks that we play in itruncate_finish
2021 * don't quite matter.
2023 * 2. We don't play the quota buffers when there's a quotaoff logitem.
2024 * But the allocation will be replayed so we'll end up with an
2025 * uninitialized quota block.
2027 * This is all fine; things are still consistent, and we haven't lost
2028 * any quota information. Just don't complain about bad dquot blks.
2030 if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) {
2031 if (flags & XFS_QMOPT_DOWARN)
2033 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
2034 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
2037 if (ddq->d_version != XFS_DQUOT_VERSION) {
2038 if (flags & XFS_QMOPT_DOWARN)
2040 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
2041 str, id, ddq->d_version, XFS_DQUOT_VERSION);
2045 if (ddq->d_flags != XFS_DQ_USER &&
2046 ddq->d_flags != XFS_DQ_PROJ &&
2047 ddq->d_flags != XFS_DQ_GROUP) {
2048 if (flags & XFS_QMOPT_DOWARN)
2050 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
2051 str, id, ddq->d_flags);
2055 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
2056 if (flags & XFS_QMOPT_DOWARN)
2058 "%s : ondisk-dquot 0x%p, ID mismatch: "
2059 "0x%x expected, found id 0x%x",
2060 str, ddq, id, be32_to_cpu(ddq->d_id));
2064 if (!errs && ddq->d_id) {
2065 if (ddq->d_blk_softlimit &&
2066 be64_to_cpu(ddq->d_bcount) >=
2067 be64_to_cpu(ddq->d_blk_softlimit)) {
2068 if (!ddq->d_btimer) {
2069 if (flags & XFS_QMOPT_DOWARN)
2071 "%s : Dquot ID 0x%x (0x%p) "
2072 "BLK TIMER NOT STARTED",
2073 str, (int)be32_to_cpu(ddq->d_id), ddq);
2077 if (ddq->d_ino_softlimit &&
2078 be64_to_cpu(ddq->d_icount) >=
2079 be64_to_cpu(ddq->d_ino_softlimit)) {
2080 if (!ddq->d_itimer) {
2081 if (flags & XFS_QMOPT_DOWARN)
2083 "%s : Dquot ID 0x%x (0x%p) "
2084 "INODE TIMER NOT STARTED",
2085 str, (int)be32_to_cpu(ddq->d_id), ddq);
2089 if (ddq->d_rtb_softlimit &&
2090 be64_to_cpu(ddq->d_rtbcount) >=
2091 be64_to_cpu(ddq->d_rtb_softlimit)) {
2092 if (!ddq->d_rtbtimer) {
2093 if (flags & XFS_QMOPT_DOWARN)
2095 "%s : Dquot ID 0x%x (0x%p) "
2096 "RTBLK TIMER NOT STARTED",
2097 str, (int)be32_to_cpu(ddq->d_id), ddq);
2103 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2106 if (flags & XFS_QMOPT_DOWARN)
2107 cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id);
2110 * Typically, a repair is only requested by quotacheck.
2113 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2114 memset(d, 0, sizeof(xfs_dqblk_t));
2116 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2117 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2118 d->dd_diskdq.d_flags = type;
2119 d->dd_diskdq.d_id = cpu_to_be32(id);
2125 * Perform a dquot buffer recovery.
2126 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2127 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2128 * Else, treat it as a regular buffer and do recovery.
2131 xlog_recover_do_dquot_buffer(
2134 xlog_recover_item_t *item,
2136 xfs_buf_log_format_t *buf_f)
2140 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2143 * Filesystems are required to send in quota flags at mount time.
2145 if (mp->m_qflags == 0) {
2150 if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF)
2151 type |= XFS_DQ_USER;
2152 if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF)
2153 type |= XFS_DQ_PROJ;
2154 if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF)
2155 type |= XFS_DQ_GROUP;
2157 * This type of quotas was turned off, so ignore this buffer
2159 if (log->l_quotaoffs_flag & type)
2162 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2166 * This routine replays a modification made to a buffer at runtime.
2167 * There are actually two types of buffer, regular and inode, which
2168 * are handled differently. Inode buffers are handled differently
2169 * in that we only recover a specific set of data from them, namely
2170 * the inode di_next_unlinked fields. This is because all other inode
2171 * data is actually logged via inode records and any data we replay
2172 * here which overlaps that may be stale.
2174 * When meta-data buffers are freed at run time we log a buffer item
2175 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
2176 * of the buffer in the log should not be replayed at recovery time.
2177 * This is so that if the blocks covered by the buffer are reused for
2178 * file data before we crash we don't end up replaying old, freed
2179 * meta-data into a user's file.
2181 * To handle the cancellation of buffer log items, we make two passes
2182 * over the log during recovery. During the first we build a table of
2183 * those buffers which have been cancelled, and during the second we
2184 * only replay those buffers which do not have corresponding cancel
2185 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2186 * for more details on the implementation of the table of cancel records.
2189 xlog_recover_do_buffer_trans(
2191 xlog_recover_item_t *item,
2194 xfs_buf_log_format_t *buf_f;
2204 buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr;
2206 if (pass == XLOG_RECOVER_PASS1) {
2208 * In this pass we're only looking for buf items
2209 * with the XFS_BLI_CANCEL bit set.
2211 xlog_recover_do_buffer_pass1(log, buf_f);
2215 * In this pass we want to recover all the buffers
2216 * which have not been cancelled and are not
2217 * cancellation buffers themselves. The routine
2218 * we call here will tell us whether or not to
2219 * continue with the replay of this buffer.
2221 cancel = xlog_recover_do_buffer_pass2(log, buf_f);
2223 trace_xfs_log_recover_buf_cancel(log, buf_f);
2227 trace_xfs_log_recover_buf_recover(log, buf_f);
2228 switch (buf_f->blf_type) {
2230 blkno = buf_f->blf_blkno;
2231 len = buf_f->blf_len;
2232 flags = buf_f->blf_flags;
2235 xfs_fs_cmn_err(CE_ALERT, log->l_mp,
2236 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2237 buf_f->blf_type, log->l_mp->m_logname ?
2238 log->l_mp->m_logname : "internal");
2239 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2240 XFS_ERRLEVEL_LOW, log->l_mp);
2241 return XFS_ERROR(EFSCORRUPTED);
2245 buf_flags = XBF_LOCK;
2246 if (!(flags & XFS_BLI_INODE_BUF))
2247 buf_flags |= XBF_MAPPED;
2249 bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, buf_flags);
2250 if (XFS_BUF_ISERROR(bp)) {
2251 xfs_ioerror_alert("xlog_recover_do..(read#1)", log->l_mp,
2253 error = XFS_BUF_GETERROR(bp);
2259 if (flags & XFS_BLI_INODE_BUF) {
2260 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2262 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
2263 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2265 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2268 return XFS_ERROR(error);
2271 * Perform delayed write on the buffer. Asynchronous writes will be
2272 * slower when taking into account all the buffers to be flushed.
2274 * Also make sure that only inode buffers with good sizes stay in
2275 * the buffer cache. The kernel moves inodes in buffers of 1 block
2276 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2277 * buffers in the log can be a different size if the log was generated
2278 * by an older kernel using unclustered inode buffers or a newer kernel
2279 * running with a different inode cluster size. Regardless, if the
2280 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2281 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2282 * the buffer out of the buffer cache so that the buffer won't
2283 * overlap with future reads of those inodes.
2285 if (XFS_DINODE_MAGIC ==
2286 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2287 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2288 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2290 error = xfs_bwrite(mp, bp);
2292 ASSERT(bp->b_mount == NULL || bp->b_mount == mp);
2294 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2295 xfs_bdwrite(mp, bp);
2302 xlog_recover_do_inode_trans(
2304 xlog_recover_item_t *item,
2307 xfs_inode_log_format_t *in_f;
2318 xfs_icdinode_t *dicp;
2321 if (pass == XLOG_RECOVER_PASS1) {
2325 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2326 in_f = (xfs_inode_log_format_t *)item->ri_buf[0].i_addr;
2328 in_f = (xfs_inode_log_format_t *)kmem_alloc(
2329 sizeof(xfs_inode_log_format_t), KM_SLEEP);
2331 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2335 ino = in_f->ilf_ino;
2339 * Inode buffers can be freed, look out for it,
2340 * and do not replay the inode.
2342 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2343 in_f->ilf_len, 0)) {
2345 trace_xfs_log_recover_inode_cancel(log, in_f);
2348 trace_xfs_log_recover_inode_recover(log, in_f);
2350 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len,
2352 if (XFS_BUF_ISERROR(bp)) {
2353 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
2354 bp, in_f->ilf_blkno);
2355 error = XFS_BUF_GETERROR(bp);
2360 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2361 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
2364 * Make sure the place we're flushing out to really looks
2367 if (unlikely(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC)) {
2369 xfs_fs_cmn_err(CE_ALERT, mp,
2370 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2372 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2373 XFS_ERRLEVEL_LOW, mp);
2374 error = EFSCORRUPTED;
2377 dicp = (xfs_icdinode_t *)(item->ri_buf[1].i_addr);
2378 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2380 xfs_fs_cmn_err(CE_ALERT, mp,
2381 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2383 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2384 XFS_ERRLEVEL_LOW, mp);
2385 error = EFSCORRUPTED;
2389 /* Skip replay when the on disk inode is newer than the log one */
2390 if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
2392 * Deal with the wrap case, DI_MAX_FLUSH is less
2393 * than smaller numbers
2395 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
2396 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2400 trace_xfs_log_recover_inode_skip(log, in_f);
2405 /* Take the opportunity to reset the flush iteration count */
2406 dicp->di_flushiter = 0;
2408 if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) {
2409 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2410 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2411 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2412 XFS_ERRLEVEL_LOW, mp, dicp);
2414 xfs_fs_cmn_err(CE_ALERT, mp,
2415 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2416 item, dip, bp, ino);
2417 error = EFSCORRUPTED;
2420 } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) {
2421 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2422 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2423 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2424 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2425 XFS_ERRLEVEL_LOW, mp, dicp);
2427 xfs_fs_cmn_err(CE_ALERT, mp,
2428 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2429 item, dip, bp, ino);
2430 error = EFSCORRUPTED;
2434 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2435 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2436 XFS_ERRLEVEL_LOW, mp, dicp);
2438 xfs_fs_cmn_err(CE_ALERT, mp,
2439 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2441 dicp->di_nextents + dicp->di_anextents,
2443 error = EFSCORRUPTED;
2446 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2447 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2448 XFS_ERRLEVEL_LOW, mp, dicp);
2450 xfs_fs_cmn_err(CE_ALERT, mp,
2451 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2452 item, dip, bp, ino, dicp->di_forkoff);
2453 error = EFSCORRUPTED;
2456 if (unlikely(item->ri_buf[1].i_len > sizeof(struct xfs_icdinode))) {
2457 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2458 XFS_ERRLEVEL_LOW, mp, dicp);
2460 xfs_fs_cmn_err(CE_ALERT, mp,
2461 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2462 item->ri_buf[1].i_len, item);
2463 error = EFSCORRUPTED;
2467 /* The core is in in-core format */
2468 xfs_dinode_to_disk(dip, (xfs_icdinode_t *)item->ri_buf[1].i_addr);
2470 /* the rest is in on-disk format */
2471 if (item->ri_buf[1].i_len > sizeof(struct xfs_icdinode)) {
2472 memcpy((xfs_caddr_t) dip + sizeof(struct xfs_icdinode),
2473 item->ri_buf[1].i_addr + sizeof(struct xfs_icdinode),
2474 item->ri_buf[1].i_len - sizeof(struct xfs_icdinode));
2477 fields = in_f->ilf_fields;
2478 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2480 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
2483 memcpy(XFS_DFORK_DPTR(dip),
2484 &in_f->ilf_u.ilfu_uuid,
2489 if (in_f->ilf_size == 2)
2490 goto write_inode_buffer;
2491 len = item->ri_buf[2].i_len;
2492 src = item->ri_buf[2].i_addr;
2493 ASSERT(in_f->ilf_size <= 4);
2494 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2495 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2496 (len == in_f->ilf_dsize));
2498 switch (fields & XFS_ILOG_DFORK) {
2499 case XFS_ILOG_DDATA:
2501 memcpy(XFS_DFORK_DPTR(dip), src, len);
2504 case XFS_ILOG_DBROOT:
2505 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
2506 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
2507 XFS_DFORK_DSIZE(dip, mp));
2512 * There are no data fork flags set.
2514 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2519 * If we logged any attribute data, recover it. There may or
2520 * may not have been any other non-core data logged in this
2523 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2524 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2529 len = item->ri_buf[attr_index].i_len;
2530 src = item->ri_buf[attr_index].i_addr;
2531 ASSERT(len == in_f->ilf_asize);
2533 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2534 case XFS_ILOG_ADATA:
2536 dest = XFS_DFORK_APTR(dip);
2537 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2538 memcpy(dest, src, len);
2541 case XFS_ILOG_ABROOT:
2542 dest = XFS_DFORK_APTR(dip);
2543 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2544 len, (xfs_bmdr_block_t*)dest,
2545 XFS_DFORK_ASIZE(dip, mp));
2549 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2558 ASSERT(bp->b_mount == NULL || bp->b_mount == mp);
2560 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2561 xfs_bdwrite(mp, bp);
2565 return XFS_ERROR(error);
2569 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2570 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2574 xlog_recover_do_quotaoff_trans(
2576 xlog_recover_item_t *item,
2579 xfs_qoff_logformat_t *qoff_f;
2581 if (pass == XLOG_RECOVER_PASS2) {
2585 qoff_f = (xfs_qoff_logformat_t *)item->ri_buf[0].i_addr;
2589 * The logitem format's flag tells us if this was user quotaoff,
2590 * group/project quotaoff or both.
2592 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2593 log->l_quotaoffs_flag |= XFS_DQ_USER;
2594 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2595 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2596 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2597 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2603 * Recover a dquot record
2606 xlog_recover_do_dquot_trans(
2608 xlog_recover_item_t *item,
2613 struct xfs_disk_dquot *ddq, *recddq;
2615 xfs_dq_logformat_t *dq_f;
2618 if (pass == XLOG_RECOVER_PASS1) {
2624 * Filesystems are required to send in quota flags at mount time.
2626 if (mp->m_qflags == 0)
2629 recddq = (xfs_disk_dquot_t *)item->ri_buf[1].i_addr;
2631 if (item->ri_buf[1].i_addr == NULL) {
2633 "XFS: NULL dquot in %s.", __func__);
2634 return XFS_ERROR(EIO);
2636 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
2638 "XFS: dquot too small (%d) in %s.",
2639 item->ri_buf[1].i_len, __func__);
2640 return XFS_ERROR(EIO);
2644 * This type of quotas was turned off, so ignore this record.
2646 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
2648 if (log->l_quotaoffs_flag & type)
2652 * At this point we know that quota was _not_ turned off.
2653 * Since the mount flags are not indicating to us otherwise, this
2654 * must mean that quota is on, and the dquot needs to be replayed.
2655 * Remember that we may not have fully recovered the superblock yet,
2656 * so we can't do the usual trick of looking at the SB quota bits.
2658 * The other possibility, of course, is that the quota subsystem was
2659 * removed since the last mount - ENOSYS.
2661 dq_f = (xfs_dq_logformat_t *)item->ri_buf[0].i_addr;
2663 if ((error = xfs_qm_dqcheck(recddq,
2665 0, XFS_QMOPT_DOWARN,
2666 "xlog_recover_do_dquot_trans (log copy)"))) {
2667 return XFS_ERROR(EIO);
2669 ASSERT(dq_f->qlf_len == 1);
2671 error = xfs_read_buf(mp, mp->m_ddev_targp,
2673 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2676 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
2677 bp, dq_f->qlf_blkno);
2681 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2684 * At least the magic num portion should be on disk because this
2685 * was among a chunk of dquots created earlier, and we did some
2686 * minimal initialization then.
2688 if (xfs_qm_dqcheck(ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2689 "xlog_recover_do_dquot_trans")) {
2691 return XFS_ERROR(EIO);
2694 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2696 ASSERT(dq_f->qlf_size == 2);
2697 ASSERT(bp->b_mount == NULL || bp->b_mount == mp);
2699 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2700 xfs_bdwrite(mp, bp);
2706 * This routine is called to create an in-core extent free intent
2707 * item from the efi format structure which was logged on disk.
2708 * It allocates an in-core efi, copies the extents from the format
2709 * structure into it, and adds the efi to the AIL with the given
2713 xlog_recover_do_efi_trans(
2715 xlog_recover_item_t *item,
2721 xfs_efi_log_item_t *efip;
2722 xfs_efi_log_format_t *efi_formatp;
2724 if (pass == XLOG_RECOVER_PASS1) {
2728 efi_formatp = (xfs_efi_log_format_t *)item->ri_buf[0].i_addr;
2731 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
2732 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2733 &(efip->efi_format)))) {
2734 xfs_efi_item_free(efip);
2737 efip->efi_next_extent = efi_formatp->efi_nextents;
2738 efip->efi_flags |= XFS_EFI_COMMITTED;
2740 spin_lock(&log->l_ailp->xa_lock);
2742 * xfs_trans_ail_update() drops the AIL lock.
2744 xfs_trans_ail_update(log->l_ailp, (xfs_log_item_t *)efip, lsn);
2750 * This routine is called when an efd format structure is found in
2751 * a committed transaction in the log. It's purpose is to cancel
2752 * the corresponding efi if it was still in the log. To do this
2753 * it searches the AIL for the efi with an id equal to that in the
2754 * efd format structure. If we find it, we remove the efi from the
2758 xlog_recover_do_efd_trans(
2760 xlog_recover_item_t *item,
2763 xfs_efd_log_format_t *efd_formatp;
2764 xfs_efi_log_item_t *efip = NULL;
2765 xfs_log_item_t *lip;
2767 struct xfs_ail_cursor cur;
2768 struct xfs_ail *ailp = log->l_ailp;
2770 if (pass == XLOG_RECOVER_PASS1) {
2774 efd_formatp = (xfs_efd_log_format_t *)item->ri_buf[0].i_addr;
2775 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2776 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2777 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2778 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
2779 efi_id = efd_formatp->efd_efi_id;
2782 * Search for the efi with the id in the efd format structure
2785 spin_lock(&ailp->xa_lock);
2786 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
2787 while (lip != NULL) {
2788 if (lip->li_type == XFS_LI_EFI) {
2789 efip = (xfs_efi_log_item_t *)lip;
2790 if (efip->efi_format.efi_id == efi_id) {
2792 * xfs_trans_ail_delete() drops the
2795 xfs_trans_ail_delete(ailp, lip);
2796 xfs_efi_item_free(efip);
2797 spin_lock(&ailp->xa_lock);
2801 lip = xfs_trans_ail_cursor_next(ailp, &cur);
2803 xfs_trans_ail_cursor_done(ailp, &cur);
2804 spin_unlock(&ailp->xa_lock);
2808 * Perform the transaction
2810 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2811 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2814 xlog_recover_do_trans(
2816 xlog_recover_t *trans,
2820 xlog_recover_item_t *item;
2822 error = xlog_recover_reorder_trans(log, trans, pass);
2826 list_for_each_entry(item, &trans->r_itemq, ri_list) {
2827 trace_xfs_log_recover_item_recover(log, trans, item, pass);
2828 switch (ITEM_TYPE(item)) {
2830 error = xlog_recover_do_buffer_trans(log, item, pass);
2833 error = xlog_recover_do_inode_trans(log, item, pass);
2836 error = xlog_recover_do_efi_trans(log, item,
2837 trans->r_lsn, pass);
2840 xlog_recover_do_efd_trans(log, item, pass);
2844 error = xlog_recover_do_dquot_trans(log, item, pass);
2846 case XFS_LI_QUOTAOFF:
2847 error = xlog_recover_do_quotaoff_trans(log, item,
2852 "XFS: invalid item type (%d) xlog_recover_do_trans", ITEM_TYPE(item));
2854 error = XFS_ERROR(EIO);
2866 * Free up any resources allocated by the transaction
2868 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2871 xlog_recover_free_trans(
2872 xlog_recover_t *trans)
2874 xlog_recover_item_t *item, *n;
2877 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
2878 /* Free the regions in the item. */
2879 list_del(&item->ri_list);
2880 for (i = 0; i < item->ri_cnt; i++)
2881 kmem_free(item->ri_buf[i].i_addr);
2882 /* Free the item itself */
2883 kmem_free(item->ri_buf);
2886 /* Free the transaction recover structure */
2891 xlog_recover_commit_trans(
2893 xlog_recover_t *trans,
2898 hlist_del(&trans->r_list);
2899 if ((error = xlog_recover_do_trans(log, trans, pass)))
2901 xlog_recover_free_trans(trans); /* no error */
2906 xlog_recover_unmount_trans(
2907 xlog_recover_t *trans)
2909 /* Do nothing now */
2910 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2915 * There are two valid states of the r_state field. 0 indicates that the
2916 * transaction structure is in a normal state. We have either seen the
2917 * start of the transaction or the last operation we added was not a partial
2918 * operation. If the last operation we added to the transaction was a
2919 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2921 * NOTE: skip LRs with 0 data length.
2924 xlog_recover_process_data(
2926 struct hlist_head rhash[],
2927 xlog_rec_header_t *rhead,
2933 xlog_op_header_t *ohead;
2934 xlog_recover_t *trans;
2940 lp = dp + be32_to_cpu(rhead->h_len);
2941 num_logops = be32_to_cpu(rhead->h_num_logops);
2943 /* check the log format matches our own - else we can't recover */
2944 if (xlog_header_check_recover(log->l_mp, rhead))
2945 return (XFS_ERROR(EIO));
2947 while ((dp < lp) && num_logops) {
2948 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2949 ohead = (xlog_op_header_t *)dp;
2950 dp += sizeof(xlog_op_header_t);
2951 if (ohead->oh_clientid != XFS_TRANSACTION &&
2952 ohead->oh_clientid != XFS_LOG) {
2954 "XFS: xlog_recover_process_data: bad clientid");
2956 return (XFS_ERROR(EIO));
2958 tid = be32_to_cpu(ohead->oh_tid);
2959 hash = XLOG_RHASH(tid);
2960 trans = xlog_recover_find_tid(&rhash[hash], tid);
2961 if (trans == NULL) { /* not found; add new tid */
2962 if (ohead->oh_flags & XLOG_START_TRANS)
2963 xlog_recover_new_tid(&rhash[hash], tid,
2964 be64_to_cpu(rhead->h_lsn));
2966 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
2968 "XFS: xlog_recover_process_data: bad length");
2970 return (XFS_ERROR(EIO));
2972 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2973 if (flags & XLOG_WAS_CONT_TRANS)
2974 flags &= ~XLOG_CONTINUE_TRANS;
2976 case XLOG_COMMIT_TRANS:
2977 error = xlog_recover_commit_trans(log,
2980 case XLOG_UNMOUNT_TRANS:
2981 error = xlog_recover_unmount_trans(trans);
2983 case XLOG_WAS_CONT_TRANS:
2984 error = xlog_recover_add_to_cont_trans(log,
2986 be32_to_cpu(ohead->oh_len));
2988 case XLOG_START_TRANS:
2990 "XFS: xlog_recover_process_data: bad transaction");
2992 error = XFS_ERROR(EIO);
2995 case XLOG_CONTINUE_TRANS:
2996 error = xlog_recover_add_to_trans(log, trans,
2997 dp, be32_to_cpu(ohead->oh_len));
3001 "XFS: xlog_recover_process_data: bad flag");
3003 error = XFS_ERROR(EIO);
3009 dp += be32_to_cpu(ohead->oh_len);
3016 * Process an extent free intent item that was recovered from
3017 * the log. We need to free the extents that it describes.
3020 xlog_recover_process_efi(
3022 xfs_efi_log_item_t *efip)
3024 xfs_efd_log_item_t *efdp;
3029 xfs_fsblock_t startblock_fsb;
3031 ASSERT(!(efip->efi_flags & XFS_EFI_RECOVERED));
3034 * First check the validity of the extents described by the
3035 * EFI. If any are bad, then assume that all are bad and
3036 * just toss the EFI.
3038 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3039 extp = &(efip->efi_format.efi_extents[i]);
3040 startblock_fsb = XFS_BB_TO_FSB(mp,
3041 XFS_FSB_TO_DADDR(mp, extp->ext_start));
3042 if ((startblock_fsb == 0) ||
3043 (extp->ext_len == 0) ||
3044 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3045 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3047 * This will pull the EFI from the AIL and
3048 * free the memory associated with it.
3050 xfs_efi_release(efip, efip->efi_format.efi_nextents);
3051 return XFS_ERROR(EIO);
3055 tp = xfs_trans_alloc(mp, 0);
3056 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
3059 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3061 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3062 extp = &(efip->efi_format.efi_extents[i]);
3063 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3066 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3070 efip->efi_flags |= XFS_EFI_RECOVERED;
3071 error = xfs_trans_commit(tp, 0);
3075 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3080 * When this is called, all of the EFIs which did not have
3081 * corresponding EFDs should be in the AIL. What we do now
3082 * is free the extents associated with each one.
3084 * Since we process the EFIs in normal transactions, they
3085 * will be removed at some point after the commit. This prevents
3086 * us from just walking down the list processing each one.
3087 * We'll use a flag in the EFI to skip those that we've already
3088 * processed and use the AIL iteration mechanism's generation
3089 * count to try to speed this up at least a bit.
3091 * When we start, we know that the EFIs are the only things in
3092 * the AIL. As we process them, however, other items are added
3093 * to the AIL. Since everything added to the AIL must come after
3094 * everything already in the AIL, we stop processing as soon as
3095 * we see something other than an EFI in the AIL.
3098 xlog_recover_process_efis(
3101 xfs_log_item_t *lip;
3102 xfs_efi_log_item_t *efip;
3104 struct xfs_ail_cursor cur;
3105 struct xfs_ail *ailp;
3108 spin_lock(&ailp->xa_lock);
3109 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3110 while (lip != NULL) {
3112 * We're done when we see something other than an EFI.
3113 * There should be no EFIs left in the AIL now.
3115 if (lip->li_type != XFS_LI_EFI) {
3117 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
3118 ASSERT(lip->li_type != XFS_LI_EFI);
3124 * Skip EFIs that we've already processed.
3126 efip = (xfs_efi_log_item_t *)lip;
3127 if (efip->efi_flags & XFS_EFI_RECOVERED) {
3128 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3132 spin_unlock(&ailp->xa_lock);
3133 error = xlog_recover_process_efi(log->l_mp, efip);
3134 spin_lock(&ailp->xa_lock);
3137 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3140 xfs_trans_ail_cursor_done(ailp, &cur);
3141 spin_unlock(&ailp->xa_lock);
3146 * This routine performs a transaction to null out a bad inode pointer
3147 * in an agi unlinked inode hash bucket.
3150 xlog_recover_clear_agi_bucket(
3152 xfs_agnumber_t agno,
3161 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3162 error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp),
3167 error = xfs_read_agi(mp, tp, agno, &agibp);
3171 agi = XFS_BUF_TO_AGI(agibp);
3172 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
3173 offset = offsetof(xfs_agi_t, agi_unlinked) +
3174 (sizeof(xfs_agino_t) * bucket);
3175 xfs_trans_log_buf(tp, agibp, offset,
3176 (offset + sizeof(xfs_agino_t) - 1));
3178 error = xfs_trans_commit(tp, 0);
3184 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3186 xfs_fs_cmn_err(CE_WARN, mp, "xlog_recover_clear_agi_bucket: "
3187 "failed to clear agi %d. Continuing.", agno);
3192 xlog_recover_process_one_iunlink(
3193 struct xfs_mount *mp,
3194 xfs_agnumber_t agno,
3198 struct xfs_buf *ibp;
3199 struct xfs_dinode *dip;
3200 struct xfs_inode *ip;
3204 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3205 error = xfs_iget(mp, NULL, ino, 0, 0, &ip, 0);
3210 * Get the on disk inode to find the next inode in the bucket.
3212 error = xfs_itobp(mp, NULL, ip, &dip, &ibp, XBF_LOCK);
3216 ASSERT(ip->i_d.di_nlink == 0);
3217 ASSERT(ip->i_d.di_mode != 0);
3219 /* setup for the next pass */
3220 agino = be32_to_cpu(dip->di_next_unlinked);
3224 * Prevent any DMAPI event from being sent when the reference on
3225 * the inode is dropped.
3227 ip->i_d.di_dmevmask = 0;
3236 * We can't read in the inode this bucket points to, or this inode
3237 * is messed up. Just ditch this bucket of inodes. We will lose
3238 * some inodes and space, but at least we won't hang.
3240 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3241 * clear the inode pointer in the bucket.
3243 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3248 * xlog_iunlink_recover
3250 * This is called during recovery to process any inodes which
3251 * we unlinked but not freed when the system crashed. These
3252 * inodes will be on the lists in the AGI blocks. What we do
3253 * here is scan all the AGIs and fully truncate and free any
3254 * inodes found on the lists. Each inode is removed from the
3255 * lists when it has been fully truncated and is freed. The
3256 * freeing of the inode and its removal from the list must be
3260 xlog_recover_process_iunlinks(
3264 xfs_agnumber_t agno;
3275 * Prevent any DMAPI event from being sent while in this function.
3277 mp_dmevmask = mp->m_dmevmask;
3280 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3282 * Find the agi for this ag.
3284 error = xfs_read_agi(mp, NULL, agno, &agibp);
3287 * AGI is b0rked. Don't process it.
3289 * We should probably mark the filesystem as corrupt
3290 * after we've recovered all the ag's we can....
3294 agi = XFS_BUF_TO_AGI(agibp);
3296 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3297 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3298 while (agino != NULLAGINO) {
3300 * Release the agi buffer so that it can
3301 * be acquired in the normal course of the
3302 * transaction to truncate and free the inode.
3304 xfs_buf_relse(agibp);
3306 agino = xlog_recover_process_one_iunlink(mp,
3307 agno, agino, bucket);
3310 * Reacquire the agibuffer and continue around
3311 * the loop. This should never fail as we know
3312 * the buffer was good earlier on.
3314 error = xfs_read_agi(mp, NULL, agno, &agibp);
3316 agi = XFS_BUF_TO_AGI(agibp);
3321 * Release the buffer for the current agi so we can
3322 * go on to the next one.
3324 xfs_buf_relse(agibp);
3327 mp->m_dmevmask = mp_dmevmask;
3333 xlog_pack_data_checksum(
3335 xlog_in_core_t *iclog,
3342 up = (__be32 *)iclog->ic_datap;
3343 /* divide length by 4 to get # words */
3344 for (i = 0; i < (size >> 2); i++) {
3345 chksum ^= be32_to_cpu(*up);
3348 iclog->ic_header.h_chksum = cpu_to_be32(chksum);
3351 #define xlog_pack_data_checksum(log, iclog, size)
3355 * Stamp cycle number in every block
3360 xlog_in_core_t *iclog,
3364 int size = iclog->ic_offset + roundoff;
3368 xlog_pack_data_checksum(log, iclog, size);
3370 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3372 dp = iclog->ic_datap;
3373 for (i = 0; i < BTOBB(size) &&
3374 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3375 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
3376 *(__be32 *)dp = cycle_lsn;
3380 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3381 xlog_in_core_2_t *xhdr = iclog->ic_data;
3383 for ( ; i < BTOBB(size); i++) {
3384 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3385 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3386 xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
3387 *(__be32 *)dp = cycle_lsn;
3391 for (i = 1; i < log->l_iclog_heads; i++) {
3392 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3397 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
3399 xlog_unpack_data_checksum(
3400 xlog_rec_header_t *rhead,
3404 __be32 *up = (__be32 *)dp;
3408 /* divide length by 4 to get # words */
3409 for (i=0; i < be32_to_cpu(rhead->h_len) >> 2; i++) {
3410 chksum ^= be32_to_cpu(*up);
3413 if (chksum != be32_to_cpu(rhead->h_chksum)) {
3414 if (rhead->h_chksum ||
3415 ((log->l_flags & XLOG_CHKSUM_MISMATCH) == 0)) {
3417 "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n",
3418 be32_to_cpu(rhead->h_chksum), chksum);
3420 "XFS: Disregard message if filesystem was created with non-DEBUG kernel");
3421 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3423 "XFS: LogR this is a LogV2 filesystem\n");
3425 log->l_flags |= XLOG_CHKSUM_MISMATCH;
3430 #define xlog_unpack_data_checksum(rhead, dp, log)
3435 xlog_rec_header_t *rhead,
3441 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
3442 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3443 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
3447 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3448 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
3449 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
3450 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3451 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3452 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
3457 xlog_unpack_data_checksum(rhead, dp, log);
3461 xlog_valid_rec_header(
3463 xlog_rec_header_t *rhead,
3468 if (unlikely(be32_to_cpu(rhead->h_magicno) != XLOG_HEADER_MAGIC_NUM)) {
3469 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3470 XFS_ERRLEVEL_LOW, log->l_mp);
3471 return XFS_ERROR(EFSCORRUPTED);
3474 (!rhead->h_version ||
3475 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
3476 xlog_warn("XFS: %s: unrecognised log version (%d).",
3477 __func__, be32_to_cpu(rhead->h_version));
3478 return XFS_ERROR(EIO);
3481 /* LR body must have data or it wouldn't have been written */
3482 hlen = be32_to_cpu(rhead->h_len);
3483 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3484 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3485 XFS_ERRLEVEL_LOW, log->l_mp);
3486 return XFS_ERROR(EFSCORRUPTED);
3488 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3489 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3490 XFS_ERRLEVEL_LOW, log->l_mp);
3491 return XFS_ERROR(EFSCORRUPTED);
3497 * Read the log from tail to head and process the log records found.
3498 * Handle the two cases where the tail and head are in the same cycle
3499 * and where the active portion of the log wraps around the end of
3500 * the physical log separately. The pass parameter is passed through
3501 * to the routines called to process the data and is not looked at
3505 xlog_do_recovery_pass(
3507 xfs_daddr_t head_blk,
3508 xfs_daddr_t tail_blk,
3511 xlog_rec_header_t *rhead;
3514 xfs_buf_t *hbp, *dbp;
3515 int error = 0, h_size;
3516 int bblks, split_bblks;
3517 int hblks, split_hblks, wrapped_hblks;
3518 struct hlist_head rhash[XLOG_RHASH_SIZE];
3520 ASSERT(head_blk != tail_blk);
3523 * Read the header of the tail block and get the iclog buffer size from
3524 * h_size. Use this to tell how many sectors make up the log header.
3526 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3528 * When using variable length iclogs, read first sector of
3529 * iclog header and extract the header size from it. Get a
3530 * new hbp that is the correct size.
3532 hbp = xlog_get_bp(log, 1);
3536 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
3540 rhead = (xlog_rec_header_t *)offset;
3541 error = xlog_valid_rec_header(log, rhead, tail_blk);
3544 h_size = be32_to_cpu(rhead->h_size);
3545 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
3546 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3547 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3548 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3551 hbp = xlog_get_bp(log, hblks);
3556 ASSERT(log->l_sectbb_log == 0);
3558 hbp = xlog_get_bp(log, 1);
3559 h_size = XLOG_BIG_RECORD_BSIZE;
3564 dbp = xlog_get_bp(log, BTOBB(h_size));
3570 memset(rhash, 0, sizeof(rhash));
3571 if (tail_blk <= head_blk) {
3572 for (blk_no = tail_blk; blk_no < head_blk; ) {
3573 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3577 rhead = (xlog_rec_header_t *)offset;
3578 error = xlog_valid_rec_header(log, rhead, blk_no);
3582 /* blocks in data section */
3583 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3584 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
3589 xlog_unpack_data(rhead, offset, log);
3590 if ((error = xlog_recover_process_data(log,
3591 rhash, rhead, offset, pass)))
3593 blk_no += bblks + hblks;
3597 * Perform recovery around the end of the physical log.
3598 * When the head is not on the same cycle number as the tail,
3599 * we can't do a sequential recovery as above.
3602 while (blk_no < log->l_logBBsize) {
3604 * Check for header wrapping around physical end-of-log
3606 offset = XFS_BUF_PTR(hbp);
3609 if (blk_no + hblks <= log->l_logBBsize) {
3610 /* Read header in one read */
3611 error = xlog_bread(log, blk_no, hblks, hbp,
3616 /* This LR is split across physical log end */
3617 if (blk_no != log->l_logBBsize) {
3618 /* some data before physical log end */
3619 ASSERT(blk_no <= INT_MAX);
3620 split_hblks = log->l_logBBsize - (int)blk_no;
3621 ASSERT(split_hblks > 0);
3622 error = xlog_bread(log, blk_no,
3630 * Note: this black magic still works with
3631 * large sector sizes (non-512) only because:
3632 * - we increased the buffer size originally
3633 * by 1 sector giving us enough extra space
3634 * for the second read;
3635 * - the log start is guaranteed to be sector
3637 * - we read the log end (LR header start)
3638 * _first_, then the log start (LR header end)
3639 * - order is important.
3641 wrapped_hblks = hblks - split_hblks;
3642 error = XFS_BUF_SET_PTR(hbp,
3643 offset + BBTOB(split_hblks),
3644 BBTOB(hblks - split_hblks));
3648 error = xlog_bread_noalign(log, 0,
3649 wrapped_hblks, hbp);
3653 error = XFS_BUF_SET_PTR(hbp, offset,
3658 rhead = (xlog_rec_header_t *)offset;
3659 error = xlog_valid_rec_header(log, rhead,
3660 split_hblks ? blk_no : 0);
3664 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3667 /* Read in data for log record */
3668 if (blk_no + bblks <= log->l_logBBsize) {
3669 error = xlog_bread(log, blk_no, bblks, dbp,
3674 /* This log record is split across the
3675 * physical end of log */
3676 offset = XFS_BUF_PTR(dbp);
3678 if (blk_no != log->l_logBBsize) {
3679 /* some data is before the physical
3681 ASSERT(!wrapped_hblks);
3682 ASSERT(blk_no <= INT_MAX);
3684 log->l_logBBsize - (int)blk_no;
3685 ASSERT(split_bblks > 0);
3686 error = xlog_bread(log, blk_no,
3694 * Note: this black magic still works with
3695 * large sector sizes (non-512) only because:
3696 * - we increased the buffer size originally
3697 * by 1 sector giving us enough extra space
3698 * for the second read;
3699 * - the log start is guaranteed to be sector
3701 * - we read the log end (LR header start)
3702 * _first_, then the log start (LR header end)
3703 * - order is important.
3705 error = XFS_BUF_SET_PTR(dbp,
3706 offset + BBTOB(split_bblks),
3707 BBTOB(bblks - split_bblks));
3711 error = xlog_bread_noalign(log, wrapped_hblks,
3712 bblks - split_bblks,
3717 error = XFS_BUF_SET_PTR(dbp, offset, h_size);
3721 xlog_unpack_data(rhead, offset, log);
3722 if ((error = xlog_recover_process_data(log, rhash,
3723 rhead, offset, pass)))
3728 ASSERT(blk_no >= log->l_logBBsize);
3729 blk_no -= log->l_logBBsize;
3731 /* read first part of physical log */
3732 while (blk_no < head_blk) {
3733 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3737 rhead = (xlog_rec_header_t *)offset;
3738 error = xlog_valid_rec_header(log, rhead, blk_no);
3742 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3743 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
3748 xlog_unpack_data(rhead, offset, log);
3749 if ((error = xlog_recover_process_data(log, rhash,
3750 rhead, offset, pass)))
3752 blk_no += bblks + hblks;
3764 * Do the recovery of the log. We actually do this in two phases.
3765 * The two passes are necessary in order to implement the function
3766 * of cancelling a record written into the log. The first pass
3767 * determines those things which have been cancelled, and the
3768 * second pass replays log items normally except for those which
3769 * have been cancelled. The handling of the replay and cancellations
3770 * takes place in the log item type specific routines.
3772 * The table of items which have cancel records in the log is allocated
3773 * and freed at this level, since only here do we know when all of
3774 * the log recovery has been completed.
3777 xlog_do_log_recovery(
3779 xfs_daddr_t head_blk,
3780 xfs_daddr_t tail_blk)
3784 ASSERT(head_blk != tail_blk);
3787 * First do a pass to find all of the cancelled buf log items.
3788 * Store them in the buf_cancel_table for use in the second pass.
3790 log->l_buf_cancel_table =
3791 (xfs_buf_cancel_t **)kmem_zalloc(XLOG_BC_TABLE_SIZE *
3792 sizeof(xfs_buf_cancel_t*),
3794 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3795 XLOG_RECOVER_PASS1);
3797 kmem_free(log->l_buf_cancel_table);
3798 log->l_buf_cancel_table = NULL;
3802 * Then do a second pass to actually recover the items in the log.
3803 * When it is complete free the table of buf cancel items.
3805 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3806 XLOG_RECOVER_PASS2);
3811 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3812 ASSERT(log->l_buf_cancel_table[i] == NULL);
3816 kmem_free(log->l_buf_cancel_table);
3817 log->l_buf_cancel_table = NULL;
3823 * Do the actual recovery
3828 xfs_daddr_t head_blk,
3829 xfs_daddr_t tail_blk)
3836 * First replay the images in the log.
3838 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3843 XFS_bflush(log->l_mp->m_ddev_targp);
3846 * If IO errors happened during recovery, bail out.
3848 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3853 * We now update the tail_lsn since much of the recovery has completed
3854 * and there may be space available to use. If there were no extent
3855 * or iunlinks, we can free up the entire log and set the tail_lsn to
3856 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3857 * lsn of the last known good LR on disk. If there are extent frees
3858 * or iunlinks they will have some entries in the AIL; so we look at
3859 * the AIL to determine how to set the tail_lsn.
3861 xlog_assign_tail_lsn(log->l_mp);
3864 * Now that we've finished replaying all buffer and inode
3865 * updates, re-read in the superblock.
3867 bp = xfs_getsb(log->l_mp, 0);
3869 ASSERT(!(XFS_BUF_ISWRITE(bp)));
3870 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
3872 XFS_BUF_UNASYNC(bp);
3873 xfsbdstrat(log->l_mp, bp);
3874 error = xfs_iowait(bp);
3876 xfs_ioerror_alert("xlog_do_recover",
3877 log->l_mp, bp, XFS_BUF_ADDR(bp));
3883 /* Convert superblock from on-disk format */
3884 sbp = &log->l_mp->m_sb;
3885 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
3886 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
3887 ASSERT(xfs_sb_good_version(sbp));
3890 /* We've re-read the superblock so re-initialize per-cpu counters */
3891 xfs_icsb_reinit_counters(log->l_mp);
3893 xlog_recover_check_summary(log);
3895 /* Normal transactions can now occur */
3896 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3901 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3903 * Return error or zero.
3909 xfs_daddr_t head_blk, tail_blk;
3912 /* find the tail of the log */
3913 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
3916 if (tail_blk != head_blk) {
3917 /* There used to be a comment here:
3919 * disallow recovery on read-only mounts. note -- mount
3920 * checks for ENOSPC and turns it into an intelligent
3922 * ...but this is no longer true. Now, unless you specify
3923 * NORECOVERY (in which case this function would never be
3924 * called), we just go ahead and recover. We do this all
3925 * under the vfs layer, so we can get away with it unless
3926 * the device itself is read-only, in which case we fail.
3928 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
3933 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3934 log->l_mp->m_fsname, log->l_mp->m_logname ?
3935 log->l_mp->m_logname : "internal");
3937 error = xlog_do_recover(log, head_blk, tail_blk);
3938 log->l_flags |= XLOG_RECOVERY_NEEDED;
3944 * In the first part of recovery we replay inodes and buffers and build
3945 * up the list of extent free items which need to be processed. Here
3946 * we process the extent free items and clean up the on disk unlinked
3947 * inode lists. This is separated from the first part of recovery so
3948 * that the root and real-time bitmap inodes can be read in from disk in
3949 * between the two stages. This is necessary so that we can free space
3950 * in the real-time portion of the file system.
3953 xlog_recover_finish(
3957 * Now we're ready to do the transactions needed for the
3958 * rest of recovery. Start with completing all the extent
3959 * free intent records and then process the unlinked inode
3960 * lists. At this point, we essentially run in normal mode
3961 * except that we're still performing recovery actions
3962 * rather than accepting new requests.
3964 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
3966 error = xlog_recover_process_efis(log);
3969 "Failed to recover EFIs on filesystem: %s",
3970 log->l_mp->m_fsname);
3974 * Sync the log to get all the EFIs out of the AIL.
3975 * This isn't absolutely necessary, but it helps in
3976 * case the unlink transactions would have problems
3977 * pushing the EFIs out of the way.
3979 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
3981 xlog_recover_process_iunlinks(log);
3983 xlog_recover_check_summary(log);
3986 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3987 log->l_mp->m_fsname, log->l_mp->m_logname ?
3988 log->l_mp->m_logname : "internal");
3989 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
3992 "!Ending clean XFS mount for filesystem: %s\n",
3993 log->l_mp->m_fsname);
4001 * Read all of the agf and agi counters and check that they
4002 * are consistent with the superblock counters.
4005 xlog_recover_check_summary(
4013 #ifdef XFS_LOUD_RECOVERY
4016 xfs_agnumber_t agno;
4017 __uint64_t freeblks;
4027 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
4028 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
4030 xfs_fs_cmn_err(CE_ALERT, mp,
4031 "xlog_recover_check_summary(agf)"
4032 "agf read failed agno %d error %d",
4035 agfp = XFS_BUF_TO_AGF(agfbp);
4036 freeblks += be32_to_cpu(agfp->agf_freeblks) +
4037 be32_to_cpu(agfp->agf_flcount);
4038 xfs_buf_relse(agfbp);
4041 error = xfs_read_agi(mp, NULL, agno, &agibp);
4043 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
4045 itotal += be32_to_cpu(agi->agi_count);
4046 ifree += be32_to_cpu(agi->agi_freecount);
4047 xfs_buf_relse(agibp);
4051 sbbp = xfs_getsb(mp, 0);
4052 #ifdef XFS_LOUD_RECOVERY
4054 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(sbbp));
4056 "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
4057 sbp->sb_icount, itotal);
4059 "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
4060 sbp->sb_ifree, ifree);
4062 "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
4063 sbp->sb_fdblocks, freeblks);
4066 * This is turned off until I account for the allocation
4067 * btree blocks which live in free space.
4069 ASSERT(sbp->sb_icount == itotal);
4070 ASSERT(sbp->sb_ifree == ifree);
4071 ASSERT(sbp->sb_fdblocks == freeblks);
4074 xfs_buf_relse(sbbp);