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"
40 #include "xfs_alloc.h"
41 #include "xfs_ialloc.h"
42 #include "xfs_log_priv.h"
43 #include "xfs_buf_item.h"
44 #include "xfs_log_recover.h"
45 #include "xfs_extfree_item.h"
46 #include "xfs_trans_priv.h"
47 #include "xfs_quota.h"
49 #include "xfs_utils.h"
51 STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *);
52 STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t);
53 STATIC void xlog_recover_insert_item_backq(xlog_recover_item_t **q,
54 xlog_recover_item_t *item);
56 STATIC void xlog_recover_check_summary(xlog_t *);
58 #define xlog_recover_check_summary(log)
63 * Sector aligned buffer routines for buffer create/read/write/access
66 #define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \
67 ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \
68 ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) )
69 #define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask)
76 ASSERT(num_bblks > 0);
78 if (log->l_sectbb_log) {
80 num_bblks += XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
81 num_bblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, num_bblks);
83 return xfs_buf_get_noaddr(BBTOB(num_bblks), log->l_mp->m_logdev_targp);
95 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
106 if (log->l_sectbb_log) {
107 blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
108 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
112 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
115 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
118 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
119 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
121 xfsbdstrat(log->l_mp, bp);
122 error = xfs_iowait(bp);
124 xfs_ioerror_alert("xlog_bread", log->l_mp,
125 bp, XFS_BUF_ADDR(bp));
130 * Write out the buffer at the given block for the given number of blocks.
131 * The buffer is kept locked across the write and is returned locked.
132 * This can only be used for synchronous log writes.
143 if (log->l_sectbb_log) {
144 blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
145 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
149 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
151 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
152 XFS_BUF_ZEROFLAGS(bp);
155 XFS_BUF_PSEMA(bp, PRIBIO);
156 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
157 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
159 if ((error = xfs_bwrite(log->l_mp, bp)))
160 xfs_ioerror_alert("xlog_bwrite", log->l_mp,
161 bp, XFS_BUF_ADDR(bp));
174 if (!log->l_sectbb_log)
175 return XFS_BUF_PTR(bp);
177 ptr = XFS_BUF_PTR(bp) + BBTOB((int)blk_no & log->l_sectbb_mask);
178 ASSERT(XFS_BUF_SIZE(bp) >=
179 BBTOB(nbblks + (blk_no & log->l_sectbb_mask)));
185 * dump debug superblock and log record information
188 xlog_header_check_dump(
190 xlog_rec_header_t *head)
194 cmn_err(CE_DEBUG, "%s: SB : uuid = ", __func__);
195 for (b = 0; b < 16; b++)
196 cmn_err(CE_DEBUG, "%02x", ((uchar_t *)&mp->m_sb.sb_uuid)[b]);
197 cmn_err(CE_DEBUG, ", fmt = %d\n", XLOG_FMT);
198 cmn_err(CE_DEBUG, " log : uuid = ");
199 for (b = 0; b < 16; b++)
200 cmn_err(CE_DEBUG, "%02x",((uchar_t *)&head->h_fs_uuid)[b]);
201 cmn_err(CE_DEBUG, ", fmt = %d\n", be32_to_cpu(head->h_fmt));
204 #define xlog_header_check_dump(mp, head)
208 * check log record header for recovery
211 xlog_header_check_recover(
213 xlog_rec_header_t *head)
215 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
218 * IRIX doesn't write the h_fmt field and leaves it zeroed
219 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
220 * a dirty log created in IRIX.
222 if (unlikely(be32_to_cpu(head->h_fmt) != XLOG_FMT)) {
224 "XFS: dirty log written in incompatible format - can't recover");
225 xlog_header_check_dump(mp, head);
226 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
227 XFS_ERRLEVEL_HIGH, mp);
228 return XFS_ERROR(EFSCORRUPTED);
229 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
231 "XFS: dirty log entry has mismatched uuid - can't recover");
232 xlog_header_check_dump(mp, head);
233 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
234 XFS_ERRLEVEL_HIGH, mp);
235 return XFS_ERROR(EFSCORRUPTED);
241 * read the head block of the log and check the header
244 xlog_header_check_mount(
246 xlog_rec_header_t *head)
248 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
250 if (uuid_is_nil(&head->h_fs_uuid)) {
252 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
253 * h_fs_uuid is nil, we assume this log was last mounted
254 * by IRIX and continue.
256 xlog_warn("XFS: nil uuid in log - IRIX style log");
257 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
258 xlog_warn("XFS: log has mismatched uuid - can't recover");
259 xlog_header_check_dump(mp, head);
260 XFS_ERROR_REPORT("xlog_header_check_mount",
261 XFS_ERRLEVEL_HIGH, mp);
262 return XFS_ERROR(EFSCORRUPTED);
273 ASSERT(XFS_BUF_FSPRIVATE(bp, void *));
275 if (XFS_BUF_GETERROR(bp)) {
277 * We're not going to bother about retrying
278 * this during recovery. One strike!
280 mp = XFS_BUF_FSPRIVATE(bp, xfs_mount_t *);
281 xfs_ioerror_alert("xlog_recover_iodone",
282 mp, bp, XFS_BUF_ADDR(bp));
283 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
285 XFS_BUF_SET_FSPRIVATE(bp, NULL);
286 XFS_BUF_CLR_IODONE_FUNC(bp);
291 * This routine finds (to an approximation) the first block in the physical
292 * log which contains the given cycle. It uses a binary search algorithm.
293 * Note that the algorithm can not be perfect because the disk will not
294 * necessarily be perfect.
297 xlog_find_cycle_start(
300 xfs_daddr_t first_blk,
301 xfs_daddr_t *last_blk,
309 mid_blk = BLK_AVG(first_blk, *last_blk);
310 while (mid_blk != first_blk && mid_blk != *last_blk) {
311 if ((error = xlog_bread(log, mid_blk, 1, bp)))
313 offset = xlog_align(log, mid_blk, 1, bp);
314 mid_cycle = xlog_get_cycle(offset);
315 if (mid_cycle == cycle) {
317 /* last_half_cycle == mid_cycle */
320 /* first_half_cycle == mid_cycle */
322 mid_blk = BLK_AVG(first_blk, *last_blk);
324 ASSERT((mid_blk == first_blk && mid_blk+1 == *last_blk) ||
325 (mid_blk == *last_blk && mid_blk-1 == first_blk));
331 * Check that the range of blocks does not contain the cycle number
332 * given. The scan needs to occur from front to back and the ptr into the
333 * region must be updated since a later routine will need to perform another
334 * test. If the region is completely good, we end up returning the same
337 * Set blkno to -1 if we encounter no errors. This is an invalid block number
338 * since we don't ever expect logs to get this large.
341 xlog_find_verify_cycle(
343 xfs_daddr_t start_blk,
345 uint stop_on_cycle_no,
346 xfs_daddr_t *new_blk)
352 xfs_caddr_t buf = NULL;
355 bufblks = 1 << ffs(nbblks);
357 while (!(bp = xlog_get_bp(log, bufblks))) {
358 /* can't get enough memory to do everything in one big buffer */
360 if (bufblks <= log->l_sectbb_log)
364 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
367 bcount = min(bufblks, (start_blk + nbblks - i));
369 if ((error = xlog_bread(log, i, bcount, bp)))
372 buf = xlog_align(log, i, bcount, bp);
373 for (j = 0; j < bcount; j++) {
374 cycle = xlog_get_cycle(buf);
375 if (cycle == stop_on_cycle_no) {
392 * Potentially backup over partial log record write.
394 * In the typical case, last_blk is the number of the block directly after
395 * a good log record. Therefore, we subtract one to get the block number
396 * of the last block in the given buffer. extra_bblks contains the number
397 * of blocks we would have read on a previous read. This happens when the
398 * last log record is split over the end of the physical log.
400 * extra_bblks is the number of blocks potentially verified on a previous
401 * call to this routine.
404 xlog_find_verify_log_record(
406 xfs_daddr_t start_blk,
407 xfs_daddr_t *last_blk,
412 xfs_caddr_t offset = NULL;
413 xlog_rec_header_t *head = NULL;
416 int num_blks = *last_blk - start_blk;
419 ASSERT(start_blk != 0 || *last_blk != start_blk);
421 if (!(bp = xlog_get_bp(log, num_blks))) {
422 if (!(bp = xlog_get_bp(log, 1)))
426 if ((error = xlog_bread(log, start_blk, num_blks, bp)))
428 offset = xlog_align(log, start_blk, num_blks, bp);
429 offset += ((num_blks - 1) << BBSHIFT);
432 for (i = (*last_blk) - 1; i >= 0; i--) {
434 /* valid log record not found */
436 "XFS: Log inconsistent (didn't find previous header)");
438 error = XFS_ERROR(EIO);
443 if ((error = xlog_bread(log, i, 1, bp)))
445 offset = xlog_align(log, i, 1, bp);
448 head = (xlog_rec_header_t *)offset;
450 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(head->h_magicno))
458 * We hit the beginning of the physical log & still no header. Return
459 * to caller. If caller can handle a return of -1, then this routine
460 * will be called again for the end of the physical log.
468 * We have the final block of the good log (the first block
469 * of the log record _before_ the head. So we check the uuid.
471 if ((error = xlog_header_check_mount(log->l_mp, head)))
475 * We may have found a log record header before we expected one.
476 * last_blk will be the 1st block # with a given cycle #. We may end
477 * up reading an entire log record. In this case, we don't want to
478 * reset last_blk. Only when last_blk points in the middle of a log
479 * record do we update last_blk.
481 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
482 uint h_size = be32_to_cpu(head->h_size);
484 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
485 if (h_size % XLOG_HEADER_CYCLE_SIZE)
491 if (*last_blk - i + extra_bblks !=
492 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
501 * Head is defined to be the point of the log where the next log write
502 * write could go. This means that incomplete LR writes at the end are
503 * eliminated when calculating the head. We aren't guaranteed that previous
504 * LR have complete transactions. We only know that a cycle number of
505 * current cycle number -1 won't be present in the log if we start writing
506 * from our current block number.
508 * last_blk contains the block number of the first block with a given
511 * Return: zero if normal, non-zero if error.
516 xfs_daddr_t *return_head_blk)
520 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
522 uint first_half_cycle, last_half_cycle;
524 int error, log_bbnum = log->l_logBBsize;
526 /* Is the end of the log device zeroed? */
527 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
528 *return_head_blk = first_blk;
530 /* Is the whole lot zeroed? */
532 /* Linux XFS shouldn't generate totally zeroed logs -
533 * mkfs etc write a dummy unmount record to a fresh
534 * log so we can store the uuid in there
536 xlog_warn("XFS: totally zeroed log");
541 xlog_warn("XFS: empty log check failed");
545 first_blk = 0; /* get cycle # of 1st block */
546 bp = xlog_get_bp(log, 1);
549 if ((error = xlog_bread(log, 0, 1, bp)))
551 offset = xlog_align(log, 0, 1, bp);
552 first_half_cycle = xlog_get_cycle(offset);
554 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
555 if ((error = xlog_bread(log, last_blk, 1, bp)))
557 offset = xlog_align(log, last_blk, 1, bp);
558 last_half_cycle = xlog_get_cycle(offset);
559 ASSERT(last_half_cycle != 0);
562 * If the 1st half cycle number is equal to the last half cycle number,
563 * then the entire log is stamped with the same cycle number. In this
564 * case, head_blk can't be set to zero (which makes sense). The below
565 * math doesn't work out properly with head_blk equal to zero. Instead,
566 * we set it to log_bbnum which is an invalid block number, but this
567 * value makes the math correct. If head_blk doesn't changed through
568 * all the tests below, *head_blk is set to zero at the very end rather
569 * than log_bbnum. In a sense, log_bbnum and zero are the same block
570 * in a circular file.
572 if (first_half_cycle == last_half_cycle) {
574 * In this case we believe that the entire log should have
575 * cycle number last_half_cycle. We need to scan backwards
576 * from the end verifying that there are no holes still
577 * containing last_half_cycle - 1. If we find such a hole,
578 * then the start of that hole will be the new head. The
579 * simple case looks like
580 * x | x ... | x - 1 | x
581 * Another case that fits this picture would be
582 * x | x + 1 | x ... | x
583 * In this case the head really is somewhere at the end of the
584 * log, as one of the latest writes at the beginning was
587 * x | x + 1 | x ... | x - 1 | x
588 * This is really the combination of the above two cases, and
589 * the head has to end up at the start of the x-1 hole at the
592 * In the 256k log case, we will read from the beginning to the
593 * end of the log and search for cycle numbers equal to x-1.
594 * We don't worry about the x+1 blocks that we encounter,
595 * because we know that they cannot be the head since the log
598 head_blk = log_bbnum;
599 stop_on_cycle = last_half_cycle - 1;
602 * In this case we want to find the first block with cycle
603 * number matching last_half_cycle. We expect the log to be
606 * The first block with cycle number x (last_half_cycle) will
607 * be where the new head belongs. First we do a binary search
608 * for the first occurrence of last_half_cycle. The binary
609 * search may not be totally accurate, so then we scan back
610 * from there looking for occurrences of last_half_cycle before
611 * us. If that backwards scan wraps around the beginning of
612 * the log, then we look for occurrences of last_half_cycle - 1
613 * at the end of the log. The cases we're looking for look
615 * x + 1 ... | x | x + 1 | x ...
616 * ^ binary search stopped here
618 * x + 1 ... | x ... | x - 1 | x
619 * <---------> less than scan distance
621 stop_on_cycle = last_half_cycle;
622 if ((error = xlog_find_cycle_start(log, bp, first_blk,
623 &head_blk, last_half_cycle)))
628 * Now validate the answer. Scan back some number of maximum possible
629 * blocks and make sure each one has the expected cycle number. The
630 * maximum is determined by the total possible amount of buffering
631 * in the in-core log. The following number can be made tighter if
632 * we actually look at the block size of the filesystem.
634 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
635 if (head_blk >= num_scan_bblks) {
637 * We are guaranteed that the entire check can be performed
640 start_blk = head_blk - num_scan_bblks;
641 if ((error = xlog_find_verify_cycle(log,
642 start_blk, num_scan_bblks,
643 stop_on_cycle, &new_blk)))
647 } else { /* need to read 2 parts of log */
649 * We are going to scan backwards in the log in two parts.
650 * First we scan the physical end of the log. In this part
651 * of the log, we are looking for blocks with cycle number
652 * last_half_cycle - 1.
653 * If we find one, then we know that the log starts there, as
654 * we've found a hole that didn't get written in going around
655 * the end of the physical log. The simple case for this is
656 * x + 1 ... | x ... | x - 1 | x
657 * <---------> less than scan distance
658 * If all of the blocks at the end of the log have cycle number
659 * last_half_cycle, then we check the blocks at the start of
660 * the log looking for occurrences of last_half_cycle. If we
661 * find one, then our current estimate for the location of the
662 * first occurrence of last_half_cycle is wrong and we move
663 * back to the hole we've found. This case looks like
664 * x + 1 ... | x | x + 1 | x ...
665 * ^ binary search stopped here
666 * Another case we need to handle that only occurs in 256k
668 * x + 1 ... | x ... | x+1 | x ...
669 * ^ binary search stops here
670 * In a 256k log, the scan at the end of the log will see the
671 * x + 1 blocks. We need to skip past those since that is
672 * certainly not the head of the log. By searching for
673 * last_half_cycle-1 we accomplish that.
675 start_blk = log_bbnum - num_scan_bblks + head_blk;
676 ASSERT(head_blk <= INT_MAX &&
677 (xfs_daddr_t) num_scan_bblks - head_blk >= 0);
678 if ((error = xlog_find_verify_cycle(log, start_blk,
679 num_scan_bblks - (int)head_blk,
680 (stop_on_cycle - 1), &new_blk)))
688 * Scan beginning of log now. The last part of the physical
689 * log is good. This scan needs to verify that it doesn't find
690 * the last_half_cycle.
693 ASSERT(head_blk <= INT_MAX);
694 if ((error = xlog_find_verify_cycle(log,
695 start_blk, (int)head_blk,
696 stop_on_cycle, &new_blk)))
704 * Now we need to make sure head_blk is not pointing to a block in
705 * the middle of a log record.
707 num_scan_bblks = XLOG_REC_SHIFT(log);
708 if (head_blk >= num_scan_bblks) {
709 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
711 /* start ptr at last block ptr before head_blk */
712 if ((error = xlog_find_verify_log_record(log, start_blk,
713 &head_blk, 0)) == -1) {
714 error = XFS_ERROR(EIO);
720 ASSERT(head_blk <= INT_MAX);
721 if ((error = xlog_find_verify_log_record(log, start_blk,
722 &head_blk, 0)) == -1) {
723 /* We hit the beginning of the log during our search */
724 start_blk = log_bbnum - num_scan_bblks + head_blk;
726 ASSERT(start_blk <= INT_MAX &&
727 (xfs_daddr_t) log_bbnum-start_blk >= 0);
728 ASSERT(head_blk <= INT_MAX);
729 if ((error = xlog_find_verify_log_record(log,
731 (int)head_blk)) == -1) {
732 error = XFS_ERROR(EIO);
736 if (new_blk != log_bbnum)
743 if (head_blk == log_bbnum)
744 *return_head_blk = 0;
746 *return_head_blk = head_blk;
748 * When returning here, we have a good block number. Bad block
749 * means that during a previous crash, we didn't have a clean break
750 * from cycle number N to cycle number N-1. In this case, we need
751 * to find the first block with cycle number N-1.
759 xlog_warn("XFS: failed to find log head");
764 * Find the sync block number or the tail of the log.
766 * This will be the block number of the last record to have its
767 * associated buffers synced to disk. Every log record header has
768 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
769 * to get a sync block number. The only concern is to figure out which
770 * log record header to believe.
772 * The following algorithm uses the log record header with the largest
773 * lsn. The entire log record does not need to be valid. We only care
774 * that the header is valid.
776 * We could speed up search by using current head_blk buffer, but it is not
782 xfs_daddr_t *head_blk,
783 xfs_daddr_t *tail_blk)
785 xlog_rec_header_t *rhead;
786 xlog_op_header_t *op_head;
787 xfs_caddr_t offset = NULL;
790 xfs_daddr_t umount_data_blk;
791 xfs_daddr_t after_umount_blk;
798 * Find previous log record
800 if ((error = xlog_find_head(log, head_blk)))
803 bp = xlog_get_bp(log, 1);
806 if (*head_blk == 0) { /* special case */
807 if ((error = xlog_bread(log, 0, 1, bp)))
809 offset = xlog_align(log, 0, 1, bp);
810 if (xlog_get_cycle(offset) == 0) {
812 /* leave all other log inited values alone */
818 * Search backwards looking for log record header block
820 ASSERT(*head_blk < INT_MAX);
821 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
822 if ((error = xlog_bread(log, i, 1, bp)))
824 offset = xlog_align(log, i, 1, bp);
825 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(*(__be32 *)offset)) {
831 * If we haven't found the log record header block, start looking
832 * again from the end of the physical log. XXXmiken: There should be
833 * a check here to make sure we didn't search more than N blocks in
837 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
838 if ((error = xlog_bread(log, i, 1, bp)))
840 offset = xlog_align(log, i, 1, bp);
841 if (XLOG_HEADER_MAGIC_NUM ==
842 be32_to_cpu(*(__be32 *)offset)) {
849 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
851 return XFS_ERROR(EIO);
854 /* find blk_no of tail of log */
855 rhead = (xlog_rec_header_t *)offset;
856 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
859 * Reset log values according to the state of the log when we
860 * crashed. In the case where head_blk == 0, we bump curr_cycle
861 * one because the next write starts a new cycle rather than
862 * continuing the cycle of the last good log record. At this
863 * point we have guaranteed that all partial log records have been
864 * accounted for. Therefore, we know that the last good log record
865 * written was complete and ended exactly on the end boundary
866 * of the physical log.
868 log->l_prev_block = i;
869 log->l_curr_block = (int)*head_blk;
870 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
873 log->l_tail_lsn = be64_to_cpu(rhead->h_tail_lsn);
874 log->l_last_sync_lsn = be64_to_cpu(rhead->h_lsn);
875 log->l_grant_reserve_cycle = log->l_curr_cycle;
876 log->l_grant_reserve_bytes = BBTOB(log->l_curr_block);
877 log->l_grant_write_cycle = log->l_curr_cycle;
878 log->l_grant_write_bytes = BBTOB(log->l_curr_block);
881 * Look for unmount record. If we find it, then we know there
882 * was a clean unmount. Since 'i' could be the last block in
883 * the physical log, we convert to a log block before comparing
886 * Save the current tail lsn to use to pass to
887 * xlog_clear_stale_blocks() below. We won't want to clear the
888 * unmount record if there is one, so we pass the lsn of the
889 * unmount record rather than the block after it.
891 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
892 int h_size = be32_to_cpu(rhead->h_size);
893 int h_version = be32_to_cpu(rhead->h_version);
895 if ((h_version & XLOG_VERSION_2) &&
896 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
897 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
898 if (h_size % XLOG_HEADER_CYCLE_SIZE)
906 after_umount_blk = (i + hblks + (int)
907 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
908 tail_lsn = log->l_tail_lsn;
909 if (*head_blk == after_umount_blk &&
910 be32_to_cpu(rhead->h_num_logops) == 1) {
911 umount_data_blk = (i + hblks) % log->l_logBBsize;
912 if ((error = xlog_bread(log, umount_data_blk, 1, bp))) {
915 offset = xlog_align(log, umount_data_blk, 1, bp);
916 op_head = (xlog_op_header_t *)offset;
917 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
919 * Set tail and last sync so that newly written
920 * log records will point recovery to after the
921 * current unmount record.
924 xlog_assign_lsn(log->l_curr_cycle,
926 log->l_last_sync_lsn =
927 xlog_assign_lsn(log->l_curr_cycle,
929 *tail_blk = after_umount_blk;
932 * Note that the unmount was clean. If the unmount
933 * was not clean, we need to know this to rebuild the
934 * superblock counters from the perag headers if we
935 * have a filesystem using non-persistent counters.
937 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
942 * Make sure that there are no blocks in front of the head
943 * with the same cycle number as the head. This can happen
944 * because we allow multiple outstanding log writes concurrently,
945 * and the later writes might make it out before earlier ones.
947 * We use the lsn from before modifying it so that we'll never
948 * overwrite the unmount record after a clean unmount.
950 * Do this only if we are going to recover the filesystem
952 * NOTE: This used to say "if (!readonly)"
953 * However on Linux, we can & do recover a read-only filesystem.
954 * We only skip recovery if NORECOVERY is specified on mount,
955 * in which case we would not be here.
957 * But... if the -device- itself is readonly, just skip this.
958 * We can't recover this device anyway, so it won't matter.
960 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) {
961 error = xlog_clear_stale_blocks(log, tail_lsn);
969 xlog_warn("XFS: failed to locate log tail");
974 * Is the log zeroed at all?
976 * The last binary search should be changed to perform an X block read
977 * once X becomes small enough. You can then search linearly through
978 * the X blocks. This will cut down on the number of reads we need to do.
980 * If the log is partially zeroed, this routine will pass back the blkno
981 * of the first block with cycle number 0. It won't have a complete LR
985 * 0 => the log is completely written to
986 * -1 => use *blk_no as the first block of the log
987 * >0 => error has occurred
996 uint first_cycle, last_cycle;
997 xfs_daddr_t new_blk, last_blk, start_blk;
998 xfs_daddr_t num_scan_bblks;
999 int error, log_bbnum = log->l_logBBsize;
1003 /* check totally zeroed log */
1004 bp = xlog_get_bp(log, 1);
1007 if ((error = xlog_bread(log, 0, 1, bp)))
1009 offset = xlog_align(log, 0, 1, bp);
1010 first_cycle = xlog_get_cycle(offset);
1011 if (first_cycle == 0) { /* completely zeroed log */
1017 /* check partially zeroed log */
1018 if ((error = xlog_bread(log, log_bbnum-1, 1, bp)))
1020 offset = xlog_align(log, log_bbnum-1, 1, bp);
1021 last_cycle = xlog_get_cycle(offset);
1022 if (last_cycle != 0) { /* log completely written to */
1025 } else if (first_cycle != 1) {
1027 * If the cycle of the last block is zero, the cycle of
1028 * the first block must be 1. If it's not, maybe we're
1029 * not looking at a log... Bail out.
1031 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1032 return XFS_ERROR(EINVAL);
1035 /* we have a partially zeroed log */
1036 last_blk = log_bbnum-1;
1037 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1041 * Validate the answer. Because there is no way to guarantee that
1042 * the entire log is made up of log records which are the same size,
1043 * we scan over the defined maximum blocks. At this point, the maximum
1044 * is not chosen to mean anything special. XXXmiken
1046 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1047 ASSERT(num_scan_bblks <= INT_MAX);
1049 if (last_blk < num_scan_bblks)
1050 num_scan_bblks = last_blk;
1051 start_blk = last_blk - num_scan_bblks;
1054 * We search for any instances of cycle number 0 that occur before
1055 * our current estimate of the head. What we're trying to detect is
1056 * 1 ... | 0 | 1 | 0...
1057 * ^ binary search ends here
1059 if ((error = xlog_find_verify_cycle(log, start_blk,
1060 (int)num_scan_bblks, 0, &new_blk)))
1066 * Potentially backup over partial log record write. We don't need
1067 * to search the end of the log because we know it is zero.
1069 if ((error = xlog_find_verify_log_record(log, start_blk,
1070 &last_blk, 0)) == -1) {
1071 error = XFS_ERROR(EIO);
1085 * These are simple subroutines used by xlog_clear_stale_blocks() below
1086 * to initialize a buffer full of empty log record headers and write
1087 * them into the log.
1098 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1100 memset(buf, 0, BBSIZE);
1101 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1102 recp->h_cycle = cpu_to_be32(cycle);
1103 recp->h_version = cpu_to_be32(
1104 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1105 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1106 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1107 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1108 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1112 xlog_write_log_records(
1123 int sectbb = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
1124 int end_block = start_block + blocks;
1129 bufblks = 1 << ffs(blocks);
1130 while (!(bp = xlog_get_bp(log, bufblks))) {
1132 if (bufblks <= log->l_sectbb_log)
1136 /* We may need to do a read at the start to fill in part of
1137 * the buffer in the starting sector not covered by the first
1140 balign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, start_block);
1141 if (balign != start_block) {
1142 if ((error = xlog_bread(log, start_block, 1, bp))) {
1146 j = start_block - balign;
1149 for (i = start_block; i < end_block; i += bufblks) {
1150 int bcount, endcount;
1152 bcount = min(bufblks, end_block - start_block);
1153 endcount = bcount - j;
1155 /* We may need to do a read at the end to fill in part of
1156 * the buffer in the final sector not covered by the write.
1157 * If this is the same sector as the above read, skip it.
1159 ealign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, end_block);
1160 if (j == 0 && (start_block + endcount > ealign)) {
1161 offset = XFS_BUF_PTR(bp);
1162 balign = BBTOB(ealign - start_block);
1163 error = XFS_BUF_SET_PTR(bp, offset + balign,
1166 error = xlog_bread(log, ealign, sectbb, bp);
1168 error = XFS_BUF_SET_PTR(bp, offset, bufblks);
1173 offset = xlog_align(log, start_block, endcount, bp);
1174 for (; j < endcount; j++) {
1175 xlog_add_record(log, offset, cycle, i+j,
1176 tail_cycle, tail_block);
1179 error = xlog_bwrite(log, start_block, endcount, bp);
1182 start_block += endcount;
1190 * This routine is called to blow away any incomplete log writes out
1191 * in front of the log head. We do this so that we won't become confused
1192 * if we come up, write only a little bit more, and then crash again.
1193 * If we leave the partial log records out there, this situation could
1194 * cause us to think those partial writes are valid blocks since they
1195 * have the current cycle number. We get rid of them by overwriting them
1196 * with empty log records with the old cycle number rather than the
1199 * The tail lsn is passed in rather than taken from
1200 * the log so that we will not write over the unmount record after a
1201 * clean unmount in a 512 block log. Doing so would leave the log without
1202 * any valid log records in it until a new one was written. If we crashed
1203 * during that time we would not be able to recover.
1206 xlog_clear_stale_blocks(
1210 int tail_cycle, head_cycle;
1211 int tail_block, head_block;
1212 int tail_distance, max_distance;
1216 tail_cycle = CYCLE_LSN(tail_lsn);
1217 tail_block = BLOCK_LSN(tail_lsn);
1218 head_cycle = log->l_curr_cycle;
1219 head_block = log->l_curr_block;
1222 * Figure out the distance between the new head of the log
1223 * and the tail. We want to write over any blocks beyond the
1224 * head that we may have written just before the crash, but
1225 * we don't want to overwrite the tail of the log.
1227 if (head_cycle == tail_cycle) {
1229 * The tail is behind the head in the physical log,
1230 * so the distance from the head to the tail is the
1231 * distance from the head to the end of the log plus
1232 * the distance from the beginning of the log to the
1235 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1236 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1237 XFS_ERRLEVEL_LOW, log->l_mp);
1238 return XFS_ERROR(EFSCORRUPTED);
1240 tail_distance = tail_block + (log->l_logBBsize - head_block);
1243 * The head is behind the tail in the physical log,
1244 * so the distance from the head to the tail is just
1245 * the tail block minus the head block.
1247 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1248 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1249 XFS_ERRLEVEL_LOW, log->l_mp);
1250 return XFS_ERROR(EFSCORRUPTED);
1252 tail_distance = tail_block - head_block;
1256 * If the head is right up against the tail, we can't clear
1259 if (tail_distance <= 0) {
1260 ASSERT(tail_distance == 0);
1264 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1266 * Take the smaller of the maximum amount of outstanding I/O
1267 * we could have and the distance to the tail to clear out.
1268 * We take the smaller so that we don't overwrite the tail and
1269 * we don't waste all day writing from the head to the tail
1272 max_distance = MIN(max_distance, tail_distance);
1274 if ((head_block + max_distance) <= log->l_logBBsize) {
1276 * We can stomp all the blocks we need to without
1277 * wrapping around the end of the log. Just do it
1278 * in a single write. Use the cycle number of the
1279 * current cycle minus one so that the log will look like:
1282 error = xlog_write_log_records(log, (head_cycle - 1),
1283 head_block, max_distance, tail_cycle,
1289 * We need to wrap around the end of the physical log in
1290 * order to clear all the blocks. Do it in two separate
1291 * I/Os. The first write should be from the head to the
1292 * end of the physical log, and it should use the current
1293 * cycle number minus one just like above.
1295 distance = log->l_logBBsize - head_block;
1296 error = xlog_write_log_records(log, (head_cycle - 1),
1297 head_block, distance, tail_cycle,
1304 * Now write the blocks at the start of the physical log.
1305 * This writes the remainder of the blocks we want to clear.
1306 * It uses the current cycle number since we're now on the
1307 * same cycle as the head so that we get:
1308 * n ... n ... | n - 1 ...
1309 * ^^^^^ blocks we're writing
1311 distance = max_distance - (log->l_logBBsize - head_block);
1312 error = xlog_write_log_records(log, head_cycle, 0, distance,
1313 tail_cycle, tail_block);
1321 /******************************************************************************
1323 * Log recover routines
1325 ******************************************************************************
1328 STATIC xlog_recover_t *
1329 xlog_recover_find_tid(
1333 xlog_recover_t *p = q;
1336 if (p->r_log_tid == tid)
1344 xlog_recover_put_hashq(
1346 xlog_recover_t *trans)
1353 xlog_recover_add_item(
1354 xlog_recover_item_t **itemq)
1356 xlog_recover_item_t *item;
1358 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1359 xlog_recover_insert_item_backq(itemq, item);
1363 xlog_recover_add_to_cont_trans(
1364 xlog_recover_t *trans,
1368 xlog_recover_item_t *item;
1369 xfs_caddr_t ptr, old_ptr;
1372 item = trans->r_itemq;
1374 /* finish copying rest of trans header */
1375 xlog_recover_add_item(&trans->r_itemq);
1376 ptr = (xfs_caddr_t) &trans->r_theader +
1377 sizeof(xfs_trans_header_t) - len;
1378 memcpy(ptr, dp, len); /* d, s, l */
1381 item = item->ri_prev;
1383 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1384 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1386 ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
1387 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1388 item->ri_buf[item->ri_cnt-1].i_len += len;
1389 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1394 * The next region to add is the start of a new region. It could be
1395 * a whole region or it could be the first part of a new region. Because
1396 * of this, the assumption here is that the type and size fields of all
1397 * format structures fit into the first 32 bits of the structure.
1399 * This works because all regions must be 32 bit aligned. Therefore, we
1400 * either have both fields or we have neither field. In the case we have
1401 * neither field, the data part of the region is zero length. We only have
1402 * a log_op_header and can throw away the header since a new one will appear
1403 * later. If we have at least 4 bytes, then we can determine how many regions
1404 * will appear in the current log item.
1407 xlog_recover_add_to_trans(
1408 xlog_recover_t *trans,
1412 xfs_inode_log_format_t *in_f; /* any will do */
1413 xlog_recover_item_t *item;
1418 item = trans->r_itemq;
1420 ASSERT(*(uint *)dp == XFS_TRANS_HEADER_MAGIC);
1421 if (len == sizeof(xfs_trans_header_t))
1422 xlog_recover_add_item(&trans->r_itemq);
1423 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1427 ptr = kmem_alloc(len, KM_SLEEP);
1428 memcpy(ptr, dp, len);
1429 in_f = (xfs_inode_log_format_t *)ptr;
1431 if (item->ri_prev->ri_total != 0 &&
1432 item->ri_prev->ri_total == item->ri_prev->ri_cnt) {
1433 xlog_recover_add_item(&trans->r_itemq);
1435 item = trans->r_itemq;
1436 item = item->ri_prev;
1438 if (item->ri_total == 0) { /* first region to be added */
1439 item->ri_total = in_f->ilf_size;
1440 ASSERT(item->ri_total <= XLOG_MAX_REGIONS_IN_ITEM);
1441 item->ri_buf = kmem_zalloc((item->ri_total *
1442 sizeof(xfs_log_iovec_t)), KM_SLEEP);
1444 ASSERT(item->ri_total > item->ri_cnt);
1445 /* Description region is ri_buf[0] */
1446 item->ri_buf[item->ri_cnt].i_addr = ptr;
1447 item->ri_buf[item->ri_cnt].i_len = len;
1453 xlog_recover_new_tid(
1458 xlog_recover_t *trans;
1460 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1461 trans->r_log_tid = tid;
1463 xlog_recover_put_hashq(q, trans);
1467 xlog_recover_unlink_tid(
1469 xlog_recover_t *trans)
1474 ASSERT(trans != NULL);
1480 if (tp->r_next == trans) {
1488 "XFS: xlog_recover_unlink_tid: trans not found");
1490 return XFS_ERROR(EIO);
1492 tp->r_next = tp->r_next->r_next;
1498 xlog_recover_insert_item_backq(
1499 xlog_recover_item_t **q,
1500 xlog_recover_item_t *item)
1503 item->ri_prev = item->ri_next = item;
1507 item->ri_prev = (*q)->ri_prev;
1508 (*q)->ri_prev = item;
1509 item->ri_prev->ri_next = item;
1514 xlog_recover_insert_item_frontq(
1515 xlog_recover_item_t **q,
1516 xlog_recover_item_t *item)
1518 xlog_recover_insert_item_backq(q, item);
1523 xlog_recover_reorder_trans(
1524 xlog_recover_t *trans)
1526 xlog_recover_item_t *first_item, *itemq, *itemq_next;
1527 xfs_buf_log_format_t *buf_f;
1530 first_item = itemq = trans->r_itemq;
1531 trans->r_itemq = NULL;
1533 itemq_next = itemq->ri_next;
1534 buf_f = (xfs_buf_log_format_t *)itemq->ri_buf[0].i_addr;
1536 switch (ITEM_TYPE(itemq)) {
1538 flags = buf_f->blf_flags;
1539 if (!(flags & XFS_BLI_CANCEL)) {
1540 xlog_recover_insert_item_frontq(&trans->r_itemq,
1546 case XFS_LI_QUOTAOFF:
1549 xlog_recover_insert_item_backq(&trans->r_itemq, itemq);
1553 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1555 return XFS_ERROR(EIO);
1558 } while (first_item != itemq);
1563 * Build up the table of buf cancel records so that we don't replay
1564 * cancelled data in the second pass. For buffer records that are
1565 * not cancel records, there is nothing to do here so we just return.
1567 * If we get a cancel record which is already in the table, this indicates
1568 * that the buffer was cancelled multiple times. In order to ensure
1569 * that during pass 2 we keep the record in the table until we reach its
1570 * last occurrence in the log, we keep a reference count in the cancel
1571 * record in the table to tell us how many times we expect to see this
1572 * record during the second pass.
1575 xlog_recover_do_buffer_pass1(
1577 xfs_buf_log_format_t *buf_f)
1579 xfs_buf_cancel_t *bcp;
1580 xfs_buf_cancel_t *nextp;
1581 xfs_buf_cancel_t *prevp;
1582 xfs_buf_cancel_t **bucket;
1583 xfs_daddr_t blkno = 0;
1587 switch (buf_f->blf_type) {
1589 blkno = buf_f->blf_blkno;
1590 len = buf_f->blf_len;
1591 flags = buf_f->blf_flags;
1596 * If this isn't a cancel buffer item, then just return.
1598 if (!(flags & XFS_BLI_CANCEL))
1602 * Insert an xfs_buf_cancel record into the hash table of
1603 * them. If there is already an identical record, bump
1604 * its reference count.
1606 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1607 XLOG_BC_TABLE_SIZE];
1609 * If the hash bucket is empty then just insert a new record into
1612 if (*bucket == NULL) {
1613 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1615 bcp->bc_blkno = blkno;
1617 bcp->bc_refcount = 1;
1618 bcp->bc_next = NULL;
1624 * The hash bucket is not empty, so search for duplicates of our
1625 * record. If we find one them just bump its refcount. If not
1626 * then add us at the end of the list.
1630 while (nextp != NULL) {
1631 if (nextp->bc_blkno == blkno && nextp->bc_len == len) {
1632 nextp->bc_refcount++;
1636 nextp = nextp->bc_next;
1638 ASSERT(prevp != NULL);
1639 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1641 bcp->bc_blkno = blkno;
1643 bcp->bc_refcount = 1;
1644 bcp->bc_next = NULL;
1645 prevp->bc_next = bcp;
1649 * Check to see whether the buffer being recovered has a corresponding
1650 * entry in the buffer cancel record table. If it does then return 1
1651 * so that it will be cancelled, otherwise return 0. If the buffer is
1652 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
1653 * the refcount on the entry in the table and remove it from the table
1654 * if this is the last reference.
1656 * We remove the cancel record from the table when we encounter its
1657 * last occurrence in the log so that if the same buffer is re-used
1658 * again after its last cancellation we actually replay the changes
1659 * made at that point.
1662 xlog_check_buffer_cancelled(
1668 xfs_buf_cancel_t *bcp;
1669 xfs_buf_cancel_t *prevp;
1670 xfs_buf_cancel_t **bucket;
1672 if (log->l_buf_cancel_table == NULL) {
1674 * There is nothing in the table built in pass one,
1675 * so this buffer must not be cancelled.
1677 ASSERT(!(flags & XFS_BLI_CANCEL));
1681 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1682 XLOG_BC_TABLE_SIZE];
1686 * There is no corresponding entry in the table built
1687 * in pass one, so this buffer has not been cancelled.
1689 ASSERT(!(flags & XFS_BLI_CANCEL));
1694 * Search for an entry in the buffer cancel table that
1695 * matches our buffer.
1698 while (bcp != NULL) {
1699 if (bcp->bc_blkno == blkno && bcp->bc_len == len) {
1701 * We've go a match, so return 1 so that the
1702 * recovery of this buffer is cancelled.
1703 * If this buffer is actually a buffer cancel
1704 * log item, then decrement the refcount on the
1705 * one in the table and remove it if this is the
1708 if (flags & XFS_BLI_CANCEL) {
1710 if (bcp->bc_refcount == 0) {
1711 if (prevp == NULL) {
1712 *bucket = bcp->bc_next;
1714 prevp->bc_next = bcp->bc_next;
1725 * We didn't find a corresponding entry in the table, so
1726 * return 0 so that the buffer is NOT cancelled.
1728 ASSERT(!(flags & XFS_BLI_CANCEL));
1733 xlog_recover_do_buffer_pass2(
1735 xfs_buf_log_format_t *buf_f)
1737 xfs_daddr_t blkno = 0;
1741 switch (buf_f->blf_type) {
1743 blkno = buf_f->blf_blkno;
1744 flags = buf_f->blf_flags;
1745 len = buf_f->blf_len;
1749 return xlog_check_buffer_cancelled(log, blkno, len, flags);
1753 * Perform recovery for a buffer full of inodes. In these buffers,
1754 * the only data which should be recovered is that which corresponds
1755 * to the di_next_unlinked pointers in the on disk inode structures.
1756 * The rest of the data for the inodes is always logged through the
1757 * inodes themselves rather than the inode buffer and is recovered
1758 * in xlog_recover_do_inode_trans().
1760 * The only time when buffers full of inodes are fully recovered is
1761 * when the buffer is full of newly allocated inodes. In this case
1762 * the buffer will not be marked as an inode buffer and so will be
1763 * sent to xlog_recover_do_reg_buffer() below during recovery.
1766 xlog_recover_do_inode_buffer(
1768 xlog_recover_item_t *item,
1770 xfs_buf_log_format_t *buf_f)
1778 int next_unlinked_offset;
1780 xfs_agino_t *logged_nextp;
1781 xfs_agino_t *buffer_nextp;
1782 unsigned int *data_map = NULL;
1783 unsigned int map_size = 0;
1785 switch (buf_f->blf_type) {
1787 data_map = buf_f->blf_data_map;
1788 map_size = buf_f->blf_map_size;
1792 * Set the variables corresponding to the current region to
1793 * 0 so that we'll initialize them on the first pass through
1801 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1802 for (i = 0; i < inodes_per_buf; i++) {
1803 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1804 offsetof(xfs_dinode_t, di_next_unlinked);
1806 while (next_unlinked_offset >=
1807 (reg_buf_offset + reg_buf_bytes)) {
1809 * The next di_next_unlinked field is beyond
1810 * the current logged region. Find the next
1811 * logged region that contains or is beyond
1812 * the current di_next_unlinked field.
1815 bit = xfs_next_bit(data_map, map_size, bit);
1818 * If there are no more logged regions in the
1819 * buffer, then we're done.
1825 nbits = xfs_contig_bits(data_map, map_size,
1828 reg_buf_offset = bit << XFS_BLI_SHIFT;
1829 reg_buf_bytes = nbits << XFS_BLI_SHIFT;
1834 * If the current logged region starts after the current
1835 * di_next_unlinked field, then move on to the next
1836 * di_next_unlinked field.
1838 if (next_unlinked_offset < reg_buf_offset) {
1842 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1843 ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0);
1844 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1847 * The current logged region contains a copy of the
1848 * current di_next_unlinked field. Extract its value
1849 * and copy it to the buffer copy.
1851 logged_nextp = (xfs_agino_t *)
1852 ((char *)(item->ri_buf[item_index].i_addr) +
1853 (next_unlinked_offset - reg_buf_offset));
1854 if (unlikely(*logged_nextp == 0)) {
1855 xfs_fs_cmn_err(CE_ALERT, mp,
1856 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1858 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1859 XFS_ERRLEVEL_LOW, mp);
1860 return XFS_ERROR(EFSCORRUPTED);
1863 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1864 next_unlinked_offset);
1865 *buffer_nextp = *logged_nextp;
1872 * Perform a 'normal' buffer recovery. Each logged region of the
1873 * buffer should be copied over the corresponding region in the
1874 * given buffer. The bitmap in the buf log format structure indicates
1875 * where to place the logged data.
1879 xlog_recover_do_reg_buffer(
1880 xlog_recover_item_t *item,
1882 xfs_buf_log_format_t *buf_f)
1887 unsigned int *data_map = NULL;
1888 unsigned int map_size = 0;
1891 switch (buf_f->blf_type) {
1893 data_map = buf_f->blf_data_map;
1894 map_size = buf_f->blf_map_size;
1898 i = 1; /* 0 is the buf format structure */
1900 bit = xfs_next_bit(data_map, map_size, bit);
1903 nbits = xfs_contig_bits(data_map, map_size, bit);
1905 ASSERT(item->ri_buf[i].i_addr != NULL);
1906 ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0);
1907 ASSERT(XFS_BUF_COUNT(bp) >=
1908 ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT));
1911 * Do a sanity check if this is a dquot buffer. Just checking
1912 * the first dquot in the buffer should do. XXXThis is
1913 * probably a good thing to do for other buf types also.
1916 if (buf_f->blf_flags &
1917 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
1918 error = xfs_qm_dqcheck((xfs_disk_dquot_t *)
1919 item->ri_buf[i].i_addr,
1920 -1, 0, XFS_QMOPT_DOWARN,
1921 "dquot_buf_recover");
1924 memcpy(xfs_buf_offset(bp,
1925 (uint)bit << XFS_BLI_SHIFT), /* dest */
1926 item->ri_buf[i].i_addr, /* source */
1927 nbits<<XFS_BLI_SHIFT); /* length */
1932 /* Shouldn't be any more regions */
1933 ASSERT(i == item->ri_total);
1937 * Do some primitive error checking on ondisk dquot data structures.
1941 xfs_disk_dquot_t *ddq,
1943 uint type, /* used only when IO_dorepair is true */
1947 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
1951 * We can encounter an uninitialized dquot buffer for 2 reasons:
1952 * 1. If we crash while deleting the quotainode(s), and those blks got
1953 * used for user data. This is because we take the path of regular
1954 * file deletion; however, the size field of quotainodes is never
1955 * updated, so all the tricks that we play in itruncate_finish
1956 * don't quite matter.
1958 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1959 * But the allocation will be replayed so we'll end up with an
1960 * uninitialized quota block.
1962 * This is all fine; things are still consistent, and we haven't lost
1963 * any quota information. Just don't complain about bad dquot blks.
1965 if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) {
1966 if (flags & XFS_QMOPT_DOWARN)
1968 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
1969 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
1972 if (ddq->d_version != XFS_DQUOT_VERSION) {
1973 if (flags & XFS_QMOPT_DOWARN)
1975 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
1976 str, id, ddq->d_version, XFS_DQUOT_VERSION);
1980 if (ddq->d_flags != XFS_DQ_USER &&
1981 ddq->d_flags != XFS_DQ_PROJ &&
1982 ddq->d_flags != XFS_DQ_GROUP) {
1983 if (flags & XFS_QMOPT_DOWARN)
1985 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
1986 str, id, ddq->d_flags);
1990 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
1991 if (flags & XFS_QMOPT_DOWARN)
1993 "%s : ondisk-dquot 0x%p, ID mismatch: "
1994 "0x%x expected, found id 0x%x",
1995 str, ddq, id, be32_to_cpu(ddq->d_id));
1999 if (!errs && ddq->d_id) {
2000 if (ddq->d_blk_softlimit &&
2001 be64_to_cpu(ddq->d_bcount) >=
2002 be64_to_cpu(ddq->d_blk_softlimit)) {
2003 if (!ddq->d_btimer) {
2004 if (flags & XFS_QMOPT_DOWARN)
2006 "%s : Dquot ID 0x%x (0x%p) "
2007 "BLK TIMER NOT STARTED",
2008 str, (int)be32_to_cpu(ddq->d_id), ddq);
2012 if (ddq->d_ino_softlimit &&
2013 be64_to_cpu(ddq->d_icount) >=
2014 be64_to_cpu(ddq->d_ino_softlimit)) {
2015 if (!ddq->d_itimer) {
2016 if (flags & XFS_QMOPT_DOWARN)
2018 "%s : Dquot ID 0x%x (0x%p) "
2019 "INODE TIMER NOT STARTED",
2020 str, (int)be32_to_cpu(ddq->d_id), ddq);
2024 if (ddq->d_rtb_softlimit &&
2025 be64_to_cpu(ddq->d_rtbcount) >=
2026 be64_to_cpu(ddq->d_rtb_softlimit)) {
2027 if (!ddq->d_rtbtimer) {
2028 if (flags & XFS_QMOPT_DOWARN)
2030 "%s : Dquot ID 0x%x (0x%p) "
2031 "RTBLK TIMER NOT STARTED",
2032 str, (int)be32_to_cpu(ddq->d_id), ddq);
2038 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2041 if (flags & XFS_QMOPT_DOWARN)
2042 cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id);
2045 * Typically, a repair is only requested by quotacheck.
2048 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2049 memset(d, 0, sizeof(xfs_dqblk_t));
2051 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2052 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2053 d->dd_diskdq.d_flags = type;
2054 d->dd_diskdq.d_id = cpu_to_be32(id);
2060 * Perform a dquot buffer recovery.
2061 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2062 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2063 * Else, treat it as a regular buffer and do recovery.
2066 xlog_recover_do_dquot_buffer(
2069 xlog_recover_item_t *item,
2071 xfs_buf_log_format_t *buf_f)
2076 * Filesystems are required to send in quota flags at mount time.
2078 if (mp->m_qflags == 0) {
2083 if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF)
2084 type |= XFS_DQ_USER;
2085 if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF)
2086 type |= XFS_DQ_PROJ;
2087 if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF)
2088 type |= XFS_DQ_GROUP;
2090 * This type of quotas was turned off, so ignore this buffer
2092 if (log->l_quotaoffs_flag & type)
2095 xlog_recover_do_reg_buffer(item, bp, buf_f);
2099 * This routine replays a modification made to a buffer at runtime.
2100 * There are actually two types of buffer, regular and inode, which
2101 * are handled differently. Inode buffers are handled differently
2102 * in that we only recover a specific set of data from them, namely
2103 * the inode di_next_unlinked fields. This is because all other inode
2104 * data is actually logged via inode records and any data we replay
2105 * here which overlaps that may be stale.
2107 * When meta-data buffers are freed at run time we log a buffer item
2108 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
2109 * of the buffer in the log should not be replayed at recovery time.
2110 * This is so that if the blocks covered by the buffer are reused for
2111 * file data before we crash we don't end up replaying old, freed
2112 * meta-data into a user's file.
2114 * To handle the cancellation of buffer log items, we make two passes
2115 * over the log during recovery. During the first we build a table of
2116 * those buffers which have been cancelled, and during the second we
2117 * only replay those buffers which do not have corresponding cancel
2118 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2119 * for more details on the implementation of the table of cancel records.
2122 xlog_recover_do_buffer_trans(
2124 xlog_recover_item_t *item,
2127 xfs_buf_log_format_t *buf_f;
2136 buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr;
2138 if (pass == XLOG_RECOVER_PASS1) {
2140 * In this pass we're only looking for buf items
2141 * with the XFS_BLI_CANCEL bit set.
2143 xlog_recover_do_buffer_pass1(log, buf_f);
2147 * In this pass we want to recover all the buffers
2148 * which have not been cancelled and are not
2149 * cancellation buffers themselves. The routine
2150 * we call here will tell us whether or not to
2151 * continue with the replay of this buffer.
2153 cancel = xlog_recover_do_buffer_pass2(log, buf_f);
2158 switch (buf_f->blf_type) {
2160 blkno = buf_f->blf_blkno;
2161 len = buf_f->blf_len;
2162 flags = buf_f->blf_flags;
2165 xfs_fs_cmn_err(CE_ALERT, log->l_mp,
2166 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2167 buf_f->blf_type, log->l_mp->m_logname ?
2168 log->l_mp->m_logname : "internal");
2169 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2170 XFS_ERRLEVEL_LOW, log->l_mp);
2171 return XFS_ERROR(EFSCORRUPTED);
2175 if (flags & XFS_BLI_INODE_BUF) {
2176 bp = xfs_buf_read_flags(mp->m_ddev_targp, blkno, len,
2179 bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, 0);
2181 if (XFS_BUF_ISERROR(bp)) {
2182 xfs_ioerror_alert("xlog_recover_do..(read#1)", log->l_mp,
2184 error = XFS_BUF_GETERROR(bp);
2190 if (flags & XFS_BLI_INODE_BUF) {
2191 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2193 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
2194 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2196 xlog_recover_do_reg_buffer(item, bp, buf_f);
2199 return XFS_ERROR(error);
2202 * Perform delayed write on the buffer. Asynchronous writes will be
2203 * slower when taking into account all the buffers to be flushed.
2205 * Also make sure that only inode buffers with good sizes stay in
2206 * the buffer cache. The kernel moves inodes in buffers of 1 block
2207 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2208 * buffers in the log can be a different size if the log was generated
2209 * by an older kernel using unclustered inode buffers or a newer kernel
2210 * running with a different inode cluster size. Regardless, if the
2211 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2212 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2213 * the buffer out of the buffer cache so that the buffer won't
2214 * overlap with future reads of those inodes.
2216 if (XFS_DINODE_MAGIC ==
2217 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2218 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2219 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2221 error = xfs_bwrite(mp, bp);
2223 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2224 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2225 XFS_BUF_SET_FSPRIVATE(bp, mp);
2226 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2227 xfs_bdwrite(mp, bp);
2234 xlog_recover_do_inode_trans(
2236 xlog_recover_item_t *item,
2239 xfs_inode_log_format_t *in_f;
2251 xfs_icdinode_t *dicp;
2254 if (pass == XLOG_RECOVER_PASS1) {
2258 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2259 in_f = (xfs_inode_log_format_t *)item->ri_buf[0].i_addr;
2261 in_f = (xfs_inode_log_format_t *)kmem_alloc(
2262 sizeof(xfs_inode_log_format_t), KM_SLEEP);
2264 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2268 ino = in_f->ilf_ino;
2270 if (ITEM_TYPE(item) == XFS_LI_INODE) {
2271 imap.im_blkno = (xfs_daddr_t)in_f->ilf_blkno;
2272 imap.im_len = in_f->ilf_len;
2273 imap.im_boffset = in_f->ilf_boffset;
2276 * It's an old inode format record. We don't know where
2277 * its cluster is located on disk, and we can't allow
2278 * xfs_imap() to figure it out because the inode btrees
2279 * are not ready to be used. Therefore do not pass the
2280 * XFS_IMAP_LOOKUP flag to xfs_imap(). This will give
2281 * us only the single block in which the inode lives
2282 * rather than its cluster, so we must make sure to
2283 * invalidate the buffer when we write it out below.
2286 error = xfs_imap(log->l_mp, NULL, ino, &imap, 0);
2292 * Inode buffers can be freed, look out for it,
2293 * and do not replay the inode.
2295 if (xlog_check_buffer_cancelled(log, imap.im_blkno, imap.im_len, 0)) {
2300 bp = xfs_buf_read_flags(mp->m_ddev_targp, imap.im_blkno, imap.im_len,
2302 if (XFS_BUF_ISERROR(bp)) {
2303 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
2305 error = XFS_BUF_GETERROR(bp);
2310 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2311 dip = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
2314 * Make sure the place we're flushing out to really looks
2317 if (unlikely(be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC)) {
2319 xfs_fs_cmn_err(CE_ALERT, mp,
2320 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2322 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2323 XFS_ERRLEVEL_LOW, mp);
2324 error = EFSCORRUPTED;
2327 dicp = (xfs_icdinode_t *)(item->ri_buf[1].i_addr);
2328 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2330 xfs_fs_cmn_err(CE_ALERT, mp,
2331 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2333 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2334 XFS_ERRLEVEL_LOW, mp);
2335 error = EFSCORRUPTED;
2339 /* Skip replay when the on disk inode is newer than the log one */
2340 if (dicp->di_flushiter < be16_to_cpu(dip->di_core.di_flushiter)) {
2342 * Deal with the wrap case, DI_MAX_FLUSH is less
2343 * than smaller numbers
2345 if (be16_to_cpu(dip->di_core.di_flushiter) == DI_MAX_FLUSH &&
2346 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2354 /* Take the opportunity to reset the flush iteration count */
2355 dicp->di_flushiter = 0;
2357 if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) {
2358 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2359 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2360 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2361 XFS_ERRLEVEL_LOW, mp, dicp);
2363 xfs_fs_cmn_err(CE_ALERT, mp,
2364 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2365 item, dip, bp, ino);
2366 error = EFSCORRUPTED;
2369 } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) {
2370 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2371 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2372 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2373 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2374 XFS_ERRLEVEL_LOW, mp, dicp);
2376 xfs_fs_cmn_err(CE_ALERT, mp,
2377 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2378 item, dip, bp, ino);
2379 error = EFSCORRUPTED;
2383 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2384 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2385 XFS_ERRLEVEL_LOW, mp, dicp);
2387 xfs_fs_cmn_err(CE_ALERT, mp,
2388 "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",
2390 dicp->di_nextents + dicp->di_anextents,
2392 error = EFSCORRUPTED;
2395 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2396 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2397 XFS_ERRLEVEL_LOW, mp, dicp);
2399 xfs_fs_cmn_err(CE_ALERT, mp,
2400 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2401 item, dip, bp, ino, dicp->di_forkoff);
2402 error = EFSCORRUPTED;
2405 if (unlikely(item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t))) {
2406 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2407 XFS_ERRLEVEL_LOW, mp, dicp);
2409 xfs_fs_cmn_err(CE_ALERT, mp,
2410 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2411 item->ri_buf[1].i_len, item);
2412 error = EFSCORRUPTED;
2416 /* The core is in in-core format */
2417 xfs_dinode_to_disk(&dip->di_core,
2418 (xfs_icdinode_t *)item->ri_buf[1].i_addr);
2420 /* the rest is in on-disk format */
2421 if (item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t)) {
2422 memcpy((xfs_caddr_t) dip + sizeof(xfs_dinode_core_t),
2423 item->ri_buf[1].i_addr + sizeof(xfs_dinode_core_t),
2424 item->ri_buf[1].i_len - sizeof(xfs_dinode_core_t));
2427 fields = in_f->ilf_fields;
2428 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2430 dip->di_u.di_dev = cpu_to_be32(in_f->ilf_u.ilfu_rdev);
2433 dip->di_u.di_muuid = in_f->ilf_u.ilfu_uuid;
2437 if (in_f->ilf_size == 2)
2438 goto write_inode_buffer;
2439 len = item->ri_buf[2].i_len;
2440 src = item->ri_buf[2].i_addr;
2441 ASSERT(in_f->ilf_size <= 4);
2442 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2443 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2444 (len == in_f->ilf_dsize));
2446 switch (fields & XFS_ILOG_DFORK) {
2447 case XFS_ILOG_DDATA:
2449 memcpy(&dip->di_u, src, len);
2452 case XFS_ILOG_DBROOT:
2453 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
2455 XFS_DFORK_DSIZE(dip, mp));
2460 * There are no data fork flags set.
2462 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2467 * If we logged any attribute data, recover it. There may or
2468 * may not have been any other non-core data logged in this
2471 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2472 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2477 len = item->ri_buf[attr_index].i_len;
2478 src = item->ri_buf[attr_index].i_addr;
2479 ASSERT(len == in_f->ilf_asize);
2481 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2482 case XFS_ILOG_ADATA:
2484 dest = XFS_DFORK_APTR(dip);
2485 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2486 memcpy(dest, src, len);
2489 case XFS_ILOG_ABROOT:
2490 dest = XFS_DFORK_APTR(dip);
2491 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2492 len, (xfs_bmdr_block_t*)dest,
2493 XFS_DFORK_ASIZE(dip, mp));
2497 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2506 if (ITEM_TYPE(item) == XFS_LI_INODE) {
2507 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2508 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2509 XFS_BUF_SET_FSPRIVATE(bp, mp);
2510 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2511 xfs_bdwrite(mp, bp);
2514 error = xfs_bwrite(mp, bp);
2520 return XFS_ERROR(error);
2524 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2525 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2529 xlog_recover_do_quotaoff_trans(
2531 xlog_recover_item_t *item,
2534 xfs_qoff_logformat_t *qoff_f;
2536 if (pass == XLOG_RECOVER_PASS2) {
2540 qoff_f = (xfs_qoff_logformat_t *)item->ri_buf[0].i_addr;
2544 * The logitem format's flag tells us if this was user quotaoff,
2545 * group/project quotaoff or both.
2547 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2548 log->l_quotaoffs_flag |= XFS_DQ_USER;
2549 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2550 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2551 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2552 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2558 * Recover a dquot record
2561 xlog_recover_do_dquot_trans(
2563 xlog_recover_item_t *item,
2568 struct xfs_disk_dquot *ddq, *recddq;
2570 xfs_dq_logformat_t *dq_f;
2573 if (pass == XLOG_RECOVER_PASS1) {
2579 * Filesystems are required to send in quota flags at mount time.
2581 if (mp->m_qflags == 0)
2584 recddq = (xfs_disk_dquot_t *)item->ri_buf[1].i_addr;
2587 * This type of quotas was turned off, so ignore this record.
2589 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
2591 if (log->l_quotaoffs_flag & type)
2595 * At this point we know that quota was _not_ turned off.
2596 * Since the mount flags are not indicating to us otherwise, this
2597 * must mean that quota is on, and the dquot needs to be replayed.
2598 * Remember that we may not have fully recovered the superblock yet,
2599 * so we can't do the usual trick of looking at the SB quota bits.
2601 * The other possibility, of course, is that the quota subsystem was
2602 * removed since the last mount - ENOSYS.
2604 dq_f = (xfs_dq_logformat_t *)item->ri_buf[0].i_addr;
2606 if ((error = xfs_qm_dqcheck(recddq,
2608 0, XFS_QMOPT_DOWARN,
2609 "xlog_recover_do_dquot_trans (log copy)"))) {
2610 return XFS_ERROR(EIO);
2612 ASSERT(dq_f->qlf_len == 1);
2614 error = xfs_read_buf(mp, mp->m_ddev_targp,
2616 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2619 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
2620 bp, dq_f->qlf_blkno);
2624 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2627 * At least the magic num portion should be on disk because this
2628 * was among a chunk of dquots created earlier, and we did some
2629 * minimal initialization then.
2631 if (xfs_qm_dqcheck(ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2632 "xlog_recover_do_dquot_trans")) {
2634 return XFS_ERROR(EIO);
2637 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2639 ASSERT(dq_f->qlf_size == 2);
2640 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2641 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2642 XFS_BUF_SET_FSPRIVATE(bp, mp);
2643 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2644 xfs_bdwrite(mp, bp);
2650 * This routine is called to create an in-core extent free intent
2651 * item from the efi format structure which was logged on disk.
2652 * It allocates an in-core efi, copies the extents from the format
2653 * structure into it, and adds the efi to the AIL with the given
2657 xlog_recover_do_efi_trans(
2659 xlog_recover_item_t *item,
2665 xfs_efi_log_item_t *efip;
2666 xfs_efi_log_format_t *efi_formatp;
2668 if (pass == XLOG_RECOVER_PASS1) {
2672 efi_formatp = (xfs_efi_log_format_t *)item->ri_buf[0].i_addr;
2675 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
2676 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2677 &(efip->efi_format)))) {
2678 xfs_efi_item_free(efip);
2681 efip->efi_next_extent = efi_formatp->efi_nextents;
2682 efip->efi_flags |= XFS_EFI_COMMITTED;
2684 spin_lock(&mp->m_ail->xa_lock);
2686 * xfs_trans_update_ail() drops the AIL lock.
2688 xfs_trans_update_ail(mp, (xfs_log_item_t *)efip, lsn);
2694 * This routine is called when an efd format structure is found in
2695 * a committed transaction in the log. It's purpose is to cancel
2696 * the corresponding efi if it was still in the log. To do this
2697 * it searches the AIL for the efi with an id equal to that in the
2698 * efd format structure. If we find it, we remove the efi from the
2702 xlog_recover_do_efd_trans(
2704 xlog_recover_item_t *item,
2708 xfs_efd_log_format_t *efd_formatp;
2709 xfs_efi_log_item_t *efip = NULL;
2710 xfs_log_item_t *lip;
2712 struct xfs_ail_cursor cur;
2714 if (pass == XLOG_RECOVER_PASS1) {
2718 efd_formatp = (xfs_efd_log_format_t *)item->ri_buf[0].i_addr;
2719 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2720 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2721 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2722 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
2723 efi_id = efd_formatp->efd_efi_id;
2726 * Search for the efi with the id in the efd format structure
2730 spin_lock(&mp->m_ail->xa_lock);
2731 lip = xfs_trans_ail_cursor_first(mp->m_ail, &cur, 0);
2732 while (lip != NULL) {
2733 if (lip->li_type == XFS_LI_EFI) {
2734 efip = (xfs_efi_log_item_t *)lip;
2735 if (efip->efi_format.efi_id == efi_id) {
2737 * xfs_trans_delete_ail() drops the
2740 xfs_trans_delete_ail(mp, lip);
2741 xfs_efi_item_free(efip);
2742 spin_lock(&mp->m_ail->xa_lock);
2746 lip = xfs_trans_ail_cursor_next(mp->m_ail, &cur);
2748 xfs_trans_ail_cursor_done(mp->m_ail, &cur);
2749 spin_unlock(&mp->m_ail->xa_lock);
2753 * Perform the transaction
2755 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2756 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2759 xlog_recover_do_trans(
2761 xlog_recover_t *trans,
2765 xlog_recover_item_t *item, *first_item;
2767 if ((error = xlog_recover_reorder_trans(trans)))
2769 first_item = item = trans->r_itemq;
2772 * we don't need to worry about the block number being
2773 * truncated in > 1 TB buffers because in user-land,
2774 * we're now n32 or 64-bit so xfs_daddr_t is 64-bits so
2775 * the blknos will get through the user-mode buffer
2776 * cache properly. The only bad case is o32 kernels
2777 * where xfs_daddr_t is 32-bits but mount will warn us
2778 * off a > 1 TB filesystem before we get here.
2780 if ((ITEM_TYPE(item) == XFS_LI_BUF)) {
2781 if ((error = xlog_recover_do_buffer_trans(log, item,
2784 } else if ((ITEM_TYPE(item) == XFS_LI_INODE)) {
2785 if ((error = xlog_recover_do_inode_trans(log, item,
2788 } else if (ITEM_TYPE(item) == XFS_LI_EFI) {
2789 if ((error = xlog_recover_do_efi_trans(log, item, trans->r_lsn,
2792 } else if (ITEM_TYPE(item) == XFS_LI_EFD) {
2793 xlog_recover_do_efd_trans(log, item, pass);
2794 } else if (ITEM_TYPE(item) == XFS_LI_DQUOT) {
2795 if ((error = xlog_recover_do_dquot_trans(log, item,
2798 } else if ((ITEM_TYPE(item) == XFS_LI_QUOTAOFF)) {
2799 if ((error = xlog_recover_do_quotaoff_trans(log, item,
2803 xlog_warn("XFS: xlog_recover_do_trans");
2805 error = XFS_ERROR(EIO);
2808 item = item->ri_next;
2809 } while (first_item != item);
2815 * Free up any resources allocated by the transaction
2817 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2820 xlog_recover_free_trans(
2821 xlog_recover_t *trans)
2823 xlog_recover_item_t *first_item, *item, *free_item;
2826 item = first_item = trans->r_itemq;
2829 item = item->ri_next;
2830 /* Free the regions in the item. */
2831 for (i = 0; i < free_item->ri_cnt; i++) {
2832 kmem_free(free_item->ri_buf[i].i_addr);
2834 /* Free the item itself */
2835 kmem_free(free_item->ri_buf);
2836 kmem_free(free_item);
2837 } while (first_item != item);
2838 /* Free the transaction recover structure */
2843 xlog_recover_commit_trans(
2846 xlog_recover_t *trans,
2851 if ((error = xlog_recover_unlink_tid(q, trans)))
2853 if ((error = xlog_recover_do_trans(log, trans, pass)))
2855 xlog_recover_free_trans(trans); /* no error */
2860 xlog_recover_unmount_trans(
2861 xlog_recover_t *trans)
2863 /* Do nothing now */
2864 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2869 * There are two valid states of the r_state field. 0 indicates that the
2870 * transaction structure is in a normal state. We have either seen the
2871 * start of the transaction or the last operation we added was not a partial
2872 * operation. If the last operation we added to the transaction was a
2873 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2875 * NOTE: skip LRs with 0 data length.
2878 xlog_recover_process_data(
2880 xlog_recover_t *rhash[],
2881 xlog_rec_header_t *rhead,
2887 xlog_op_header_t *ohead;
2888 xlog_recover_t *trans;
2894 lp = dp + be32_to_cpu(rhead->h_len);
2895 num_logops = be32_to_cpu(rhead->h_num_logops);
2897 /* check the log format matches our own - else we can't recover */
2898 if (xlog_header_check_recover(log->l_mp, rhead))
2899 return (XFS_ERROR(EIO));
2901 while ((dp < lp) && num_logops) {
2902 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2903 ohead = (xlog_op_header_t *)dp;
2904 dp += sizeof(xlog_op_header_t);
2905 if (ohead->oh_clientid != XFS_TRANSACTION &&
2906 ohead->oh_clientid != XFS_LOG) {
2908 "XFS: xlog_recover_process_data: bad clientid");
2910 return (XFS_ERROR(EIO));
2912 tid = be32_to_cpu(ohead->oh_tid);
2913 hash = XLOG_RHASH(tid);
2914 trans = xlog_recover_find_tid(rhash[hash], tid);
2915 if (trans == NULL) { /* not found; add new tid */
2916 if (ohead->oh_flags & XLOG_START_TRANS)
2917 xlog_recover_new_tid(&rhash[hash], tid,
2918 be64_to_cpu(rhead->h_lsn));
2920 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
2922 "XFS: xlog_recover_process_data: bad length");
2924 return (XFS_ERROR(EIO));
2926 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2927 if (flags & XLOG_WAS_CONT_TRANS)
2928 flags &= ~XLOG_CONTINUE_TRANS;
2930 case XLOG_COMMIT_TRANS:
2931 error = xlog_recover_commit_trans(log,
2932 &rhash[hash], trans, pass);
2934 case XLOG_UNMOUNT_TRANS:
2935 error = xlog_recover_unmount_trans(trans);
2937 case XLOG_WAS_CONT_TRANS:
2938 error = xlog_recover_add_to_cont_trans(trans,
2939 dp, be32_to_cpu(ohead->oh_len));
2941 case XLOG_START_TRANS:
2943 "XFS: xlog_recover_process_data: bad transaction");
2945 error = XFS_ERROR(EIO);
2948 case XLOG_CONTINUE_TRANS:
2949 error = xlog_recover_add_to_trans(trans,
2950 dp, be32_to_cpu(ohead->oh_len));
2954 "XFS: xlog_recover_process_data: bad flag");
2956 error = XFS_ERROR(EIO);
2962 dp += be32_to_cpu(ohead->oh_len);
2969 * Process an extent free intent item that was recovered from
2970 * the log. We need to free the extents that it describes.
2973 xlog_recover_process_efi(
2975 xfs_efi_log_item_t *efip)
2977 xfs_efd_log_item_t *efdp;
2982 xfs_fsblock_t startblock_fsb;
2984 ASSERT(!(efip->efi_flags & XFS_EFI_RECOVERED));
2987 * First check the validity of the extents described by the
2988 * EFI. If any are bad, then assume that all are bad and
2989 * just toss the EFI.
2991 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2992 extp = &(efip->efi_format.efi_extents[i]);
2993 startblock_fsb = XFS_BB_TO_FSB(mp,
2994 XFS_FSB_TO_DADDR(mp, extp->ext_start));
2995 if ((startblock_fsb == 0) ||
2996 (extp->ext_len == 0) ||
2997 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
2998 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3000 * This will pull the EFI from the AIL and
3001 * free the memory associated with it.
3003 xfs_efi_release(efip, efip->efi_format.efi_nextents);
3004 return XFS_ERROR(EIO);
3008 tp = xfs_trans_alloc(mp, 0);
3009 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
3012 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3014 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3015 extp = &(efip->efi_format.efi_extents[i]);
3016 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3019 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3023 efip->efi_flags |= XFS_EFI_RECOVERED;
3024 error = xfs_trans_commit(tp, 0);
3028 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3033 * When this is called, all of the EFIs which did not have
3034 * corresponding EFDs should be in the AIL. What we do now
3035 * is free the extents associated with each one.
3037 * Since we process the EFIs in normal transactions, they
3038 * will be removed at some point after the commit. This prevents
3039 * us from just walking down the list processing each one.
3040 * We'll use a flag in the EFI to skip those that we've already
3041 * processed and use the AIL iteration mechanism's generation
3042 * count to try to speed this up at least a bit.
3044 * When we start, we know that the EFIs are the only things in
3045 * the AIL. As we process them, however, other items are added
3046 * to the AIL. Since everything added to the AIL must come after
3047 * everything already in the AIL, we stop processing as soon as
3048 * we see something other than an EFI in the AIL.
3051 xlog_recover_process_efis(
3054 xfs_log_item_t *lip;
3055 xfs_efi_log_item_t *efip;
3058 struct xfs_ail_cursor cur;
3061 spin_lock(&mp->m_ail->xa_lock);
3063 lip = xfs_trans_ail_cursor_first(mp->m_ail, &cur, 0);
3064 while (lip != NULL) {
3066 * We're done when we see something other than an EFI.
3067 * There should be no EFIs left in the AIL now.
3069 if (lip->li_type != XFS_LI_EFI) {
3072 lip = xfs_trans_ail_cursor_next(mp->m_ail, &cur))
3073 ASSERT(lip->li_type != XFS_LI_EFI);
3079 * Skip EFIs that we've already processed.
3081 efip = (xfs_efi_log_item_t *)lip;
3082 if (efip->efi_flags & XFS_EFI_RECOVERED) {
3083 lip = xfs_trans_ail_cursor_next(mp->m_ail, &cur);
3087 spin_unlock(&mp->m_ail->xa_lock);
3088 error = xlog_recover_process_efi(mp, efip);
3089 spin_lock(&mp->m_ail->xa_lock);
3092 lip = xfs_trans_ail_cursor_next(mp->m_ail, &cur);
3095 xfs_trans_ail_cursor_done(mp->m_ail, &cur);
3096 spin_unlock(&mp->m_ail->xa_lock);
3101 * This routine performs a transaction to null out a bad inode pointer
3102 * in an agi unlinked inode hash bucket.
3105 xlog_recover_clear_agi_bucket(
3107 xfs_agnumber_t agno,
3116 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3117 error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp), 0, 0, 0);
3119 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
3120 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
3121 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
3126 agi = XFS_BUF_TO_AGI(agibp);
3127 if (be32_to_cpu(agi->agi_magicnum) != XFS_AGI_MAGIC)
3130 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
3131 offset = offsetof(xfs_agi_t, agi_unlinked) +
3132 (sizeof(xfs_agino_t) * bucket);
3133 xfs_trans_log_buf(tp, agibp, offset,
3134 (offset + sizeof(xfs_agino_t) - 1));
3136 error = xfs_trans_commit(tp, 0);
3142 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3144 xfs_fs_cmn_err(CE_WARN, mp, "xlog_recover_clear_agi_bucket: "
3145 "failed to clear agi %d. Continuing.", agno);
3150 * xlog_iunlink_recover
3152 * This is called during recovery to process any inodes which
3153 * we unlinked but not freed when the system crashed. These
3154 * inodes will be on the lists in the AGI blocks. What we do
3155 * here is scan all the AGIs and fully truncate and free any
3156 * inodes found on the lists. Each inode is removed from the
3157 * lists when it has been fully truncated and is freed. The
3158 * freeing of the inode and its removal from the list must be
3162 xlog_recover_process_iunlinks(
3166 xfs_agnumber_t agno;
3181 * Prevent any DMAPI event from being sent while in this function.
3183 mp_dmevmask = mp->m_dmevmask;
3186 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3188 * Find the agi for this ag.
3190 agibp = xfs_buf_read(mp->m_ddev_targp,
3191 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
3192 XFS_FSS_TO_BB(mp, 1), 0);
3193 if (XFS_BUF_ISERROR(agibp)) {
3194 xfs_ioerror_alert("xlog_recover_process_iunlinks(#1)",
3196 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)));
3198 agi = XFS_BUF_TO_AGI(agibp);
3199 ASSERT(XFS_AGI_MAGIC == be32_to_cpu(agi->agi_magicnum));
3201 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3203 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3204 while (agino != NULLAGINO) {
3207 * Release the agi buffer so that it can
3208 * be acquired in the normal course of the
3209 * transaction to truncate and free the inode.
3211 xfs_buf_relse(agibp);
3213 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3214 error = xfs_iget(mp, NULL, ino, 0, 0, &ip, 0);
3215 ASSERT(error || (ip != NULL));
3219 * Get the on disk inode to find the
3220 * next inode in the bucket.
3222 error = xfs_itobp(mp, NULL, ip, &dip,
3225 ASSERT(error || (dip != NULL));
3229 ASSERT(ip->i_d.di_nlink == 0);
3231 /* setup for the next pass */
3232 agino = be32_to_cpu(
3233 dip->di_next_unlinked);
3236 * Prevent any DMAPI event from
3237 * being sent when the
3238 * reference on the inode is
3241 ip->i_d.di_dmevmask = 0;
3244 * If this is a new inode, handle
3245 * it specially. Otherwise,
3246 * just drop our reference to the
3247 * inode. If there are no
3248 * other references, this will
3250 * xfs_inactive() which will
3251 * truncate the file and free
3254 if (ip->i_d.di_mode == 0)
3255 xfs_iput_new(ip, 0);
3260 * We can't read in the inode
3261 * this bucket points to, or
3262 * this inode is messed up. Just
3263 * ditch this bucket of inodes. We
3264 * will lose some inodes and space,
3265 * but at least we won't hang. Call
3266 * xlog_recover_clear_agi_bucket()
3267 * to perform a transaction to clear
3268 * the inode pointer in the bucket.
3270 xlog_recover_clear_agi_bucket(mp, agno,
3277 * Reacquire the agibuffer and continue around
3280 agibp = xfs_buf_read(mp->m_ddev_targp,
3281 XFS_AG_DADDR(mp, agno,
3283 XFS_FSS_TO_BB(mp, 1), 0);
3284 if (XFS_BUF_ISERROR(agibp)) {
3286 "xlog_recover_process_iunlinks(#2)",
3288 XFS_AG_DADDR(mp, agno,
3289 XFS_AGI_DADDR(mp)));
3291 agi = XFS_BUF_TO_AGI(agibp);
3292 ASSERT(XFS_AGI_MAGIC == be32_to_cpu(
3293 agi->agi_magicnum));
3298 * Release the buffer for the current agi so we can
3299 * go on to the next one.
3301 xfs_buf_relse(agibp);
3304 mp->m_dmevmask = mp_dmevmask;
3310 xlog_pack_data_checksum(
3312 xlog_in_core_t *iclog,
3319 up = (__be32 *)iclog->ic_datap;
3320 /* divide length by 4 to get # words */
3321 for (i = 0; i < (size >> 2); i++) {
3322 chksum ^= be32_to_cpu(*up);
3325 iclog->ic_header.h_chksum = cpu_to_be32(chksum);
3328 #define xlog_pack_data_checksum(log, iclog, size)
3332 * Stamp cycle number in every block
3337 xlog_in_core_t *iclog,
3341 int size = iclog->ic_offset + roundoff;
3344 xlog_in_core_2_t *xhdr;
3346 xlog_pack_data_checksum(log, iclog, size);
3348 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3350 dp = iclog->ic_datap;
3351 for (i = 0; i < BTOBB(size) &&
3352 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3353 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
3354 *(__be32 *)dp = cycle_lsn;
3358 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3359 xhdr = (xlog_in_core_2_t *)&iclog->ic_header;
3360 for ( ; i < BTOBB(size); i++) {
3361 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3362 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3363 xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
3364 *(__be32 *)dp = cycle_lsn;
3368 for (i = 1; i < log->l_iclog_heads; i++) {
3369 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3374 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
3376 xlog_unpack_data_checksum(
3377 xlog_rec_header_t *rhead,
3381 __be32 *up = (__be32 *)dp;
3385 /* divide length by 4 to get # words */
3386 for (i=0; i < be32_to_cpu(rhead->h_len) >> 2; i++) {
3387 chksum ^= be32_to_cpu(*up);
3390 if (chksum != be32_to_cpu(rhead->h_chksum)) {
3391 if (rhead->h_chksum ||
3392 ((log->l_flags & XLOG_CHKSUM_MISMATCH) == 0)) {
3394 "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n",
3395 be32_to_cpu(rhead->h_chksum), chksum);
3397 "XFS: Disregard message if filesystem was created with non-DEBUG kernel");
3398 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3400 "XFS: LogR this is a LogV2 filesystem\n");
3402 log->l_flags |= XLOG_CHKSUM_MISMATCH;
3407 #define xlog_unpack_data_checksum(rhead, dp, log)
3412 xlog_rec_header_t *rhead,
3417 xlog_in_core_2_t *xhdr;
3419 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
3420 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3421 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
3425 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3426 xhdr = (xlog_in_core_2_t *)rhead;
3427 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
3428 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3429 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3430 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
3435 xlog_unpack_data_checksum(rhead, dp, log);
3439 xlog_valid_rec_header(
3441 xlog_rec_header_t *rhead,
3446 if (unlikely(be32_to_cpu(rhead->h_magicno) != XLOG_HEADER_MAGIC_NUM)) {
3447 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3448 XFS_ERRLEVEL_LOW, log->l_mp);
3449 return XFS_ERROR(EFSCORRUPTED);
3452 (!rhead->h_version ||
3453 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
3454 xlog_warn("XFS: %s: unrecognised log version (%d).",
3455 __func__, be32_to_cpu(rhead->h_version));
3456 return XFS_ERROR(EIO);
3459 /* LR body must have data or it wouldn't have been written */
3460 hlen = be32_to_cpu(rhead->h_len);
3461 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3462 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3463 XFS_ERRLEVEL_LOW, log->l_mp);
3464 return XFS_ERROR(EFSCORRUPTED);
3466 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3467 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3468 XFS_ERRLEVEL_LOW, log->l_mp);
3469 return XFS_ERROR(EFSCORRUPTED);
3475 * Read the log from tail to head and process the log records found.
3476 * Handle the two cases where the tail and head are in the same cycle
3477 * and where the active portion of the log wraps around the end of
3478 * the physical log separately. The pass parameter is passed through
3479 * to the routines called to process the data and is not looked at
3483 xlog_do_recovery_pass(
3485 xfs_daddr_t head_blk,
3486 xfs_daddr_t tail_blk,
3489 xlog_rec_header_t *rhead;
3491 xfs_caddr_t bufaddr, offset;
3492 xfs_buf_t *hbp, *dbp;
3493 int error = 0, h_size;
3494 int bblks, split_bblks;
3495 int hblks, split_hblks, wrapped_hblks;
3496 xlog_recover_t *rhash[XLOG_RHASH_SIZE];
3498 ASSERT(head_blk != tail_blk);
3501 * Read the header of the tail block and get the iclog buffer size from
3502 * h_size. Use this to tell how many sectors make up the log header.
3504 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3506 * When using variable length iclogs, read first sector of
3507 * iclog header and extract the header size from it. Get a
3508 * new hbp that is the correct size.
3510 hbp = xlog_get_bp(log, 1);
3513 if ((error = xlog_bread(log, tail_blk, 1, hbp)))
3515 offset = xlog_align(log, tail_blk, 1, hbp);
3516 rhead = (xlog_rec_header_t *)offset;
3517 error = xlog_valid_rec_header(log, rhead, tail_blk);
3520 h_size = be32_to_cpu(rhead->h_size);
3521 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
3522 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3523 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3524 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3527 hbp = xlog_get_bp(log, hblks);
3532 ASSERT(log->l_sectbb_log == 0);
3534 hbp = xlog_get_bp(log, 1);
3535 h_size = XLOG_BIG_RECORD_BSIZE;
3540 dbp = xlog_get_bp(log, BTOBB(h_size));
3546 memset(rhash, 0, sizeof(rhash));
3547 if (tail_blk <= head_blk) {
3548 for (blk_no = tail_blk; blk_no < head_blk; ) {
3549 if ((error = xlog_bread(log, blk_no, hblks, hbp)))
3551 offset = xlog_align(log, blk_no, hblks, hbp);
3552 rhead = (xlog_rec_header_t *)offset;
3553 error = xlog_valid_rec_header(log, rhead, blk_no);
3557 /* blocks in data section */
3558 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3559 error = xlog_bread(log, blk_no + hblks, bblks, dbp);
3562 offset = xlog_align(log, blk_no + hblks, bblks, dbp);
3563 xlog_unpack_data(rhead, offset, log);
3564 if ((error = xlog_recover_process_data(log,
3565 rhash, rhead, offset, pass)))
3567 blk_no += bblks + hblks;
3571 * Perform recovery around the end of the physical log.
3572 * When the head is not on the same cycle number as the tail,
3573 * we can't do a sequential recovery as above.
3576 while (blk_no < log->l_logBBsize) {
3578 * Check for header wrapping around physical end-of-log
3583 if (blk_no + hblks <= log->l_logBBsize) {
3584 /* Read header in one read */
3585 error = xlog_bread(log, blk_no, hblks, hbp);
3588 offset = xlog_align(log, blk_no, hblks, hbp);
3590 /* This LR is split across physical log end */
3591 if (blk_no != log->l_logBBsize) {
3592 /* some data before physical log end */
3593 ASSERT(blk_no <= INT_MAX);
3594 split_hblks = log->l_logBBsize - (int)blk_no;
3595 ASSERT(split_hblks > 0);
3596 if ((error = xlog_bread(log, blk_no,
3599 offset = xlog_align(log, blk_no,
3603 * Note: this black magic still works with
3604 * large sector sizes (non-512) only because:
3605 * - we increased the buffer size originally
3606 * by 1 sector giving us enough extra space
3607 * for the second read;
3608 * - the log start is guaranteed to be sector
3610 * - we read the log end (LR header start)
3611 * _first_, then the log start (LR header end)
3612 * - order is important.
3614 wrapped_hblks = hblks - split_hblks;
3615 bufaddr = XFS_BUF_PTR(hbp);
3616 error = XFS_BUF_SET_PTR(hbp,
3617 bufaddr + BBTOB(split_hblks),
3618 BBTOB(hblks - split_hblks));
3620 error = xlog_bread(log, 0,
3621 wrapped_hblks, hbp);
3623 error = XFS_BUF_SET_PTR(hbp, bufaddr,
3628 offset = xlog_align(log, 0,
3629 wrapped_hblks, hbp);
3631 rhead = (xlog_rec_header_t *)offset;
3632 error = xlog_valid_rec_header(log, rhead,
3633 split_hblks ? blk_no : 0);
3637 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3640 /* Read in data for log record */
3641 if (blk_no + bblks <= log->l_logBBsize) {
3642 error = xlog_bread(log, blk_no, bblks, dbp);
3645 offset = xlog_align(log, blk_no, bblks, dbp);
3647 /* This log record is split across the
3648 * physical end of log */
3651 if (blk_no != log->l_logBBsize) {
3652 /* some data is before the physical
3654 ASSERT(!wrapped_hblks);
3655 ASSERT(blk_no <= INT_MAX);
3657 log->l_logBBsize - (int)blk_no;
3658 ASSERT(split_bblks > 0);
3659 if ((error = xlog_bread(log, blk_no,
3662 offset = xlog_align(log, blk_no,
3666 * Note: this black magic still works with
3667 * large sector sizes (non-512) only because:
3668 * - we increased the buffer size originally
3669 * by 1 sector giving us enough extra space
3670 * for the second read;
3671 * - the log start is guaranteed to be sector
3673 * - we read the log end (LR header start)
3674 * _first_, then the log start (LR header end)
3675 * - order is important.
3677 bufaddr = XFS_BUF_PTR(dbp);
3678 error = XFS_BUF_SET_PTR(dbp,
3679 bufaddr + BBTOB(split_bblks),
3680 BBTOB(bblks - split_bblks));
3682 error = xlog_bread(log, wrapped_hblks,
3683 bblks - split_bblks,
3686 error = XFS_BUF_SET_PTR(dbp, bufaddr,
3691 offset = xlog_align(log, wrapped_hblks,
3692 bblks - split_bblks, dbp);
3694 xlog_unpack_data(rhead, offset, log);
3695 if ((error = xlog_recover_process_data(log, rhash,
3696 rhead, offset, pass)))
3701 ASSERT(blk_no >= log->l_logBBsize);
3702 blk_no -= log->l_logBBsize;
3704 /* read first part of physical log */
3705 while (blk_no < head_blk) {
3706 if ((error = xlog_bread(log, blk_no, hblks, hbp)))
3708 offset = xlog_align(log, blk_no, hblks, hbp);
3709 rhead = (xlog_rec_header_t *)offset;
3710 error = xlog_valid_rec_header(log, rhead, blk_no);
3713 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3714 if ((error = xlog_bread(log, blk_no+hblks, bblks, dbp)))
3716 offset = xlog_align(log, blk_no+hblks, bblks, dbp);
3717 xlog_unpack_data(rhead, offset, log);
3718 if ((error = xlog_recover_process_data(log, rhash,
3719 rhead, offset, pass)))
3721 blk_no += bblks + hblks;
3733 * Do the recovery of the log. We actually do this in two phases.
3734 * The two passes are necessary in order to implement the function
3735 * of cancelling a record written into the log. The first pass
3736 * determines those things which have been cancelled, and the
3737 * second pass replays log items normally except for those which
3738 * have been cancelled. The handling of the replay and cancellations
3739 * takes place in the log item type specific routines.
3741 * The table of items which have cancel records in the log is allocated
3742 * and freed at this level, since only here do we know when all of
3743 * the log recovery has been completed.
3746 xlog_do_log_recovery(
3748 xfs_daddr_t head_blk,
3749 xfs_daddr_t tail_blk)
3753 ASSERT(head_blk != tail_blk);
3756 * First do a pass to find all of the cancelled buf log items.
3757 * Store them in the buf_cancel_table for use in the second pass.
3759 log->l_buf_cancel_table =
3760 (xfs_buf_cancel_t **)kmem_zalloc(XLOG_BC_TABLE_SIZE *
3761 sizeof(xfs_buf_cancel_t*),
3763 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3764 XLOG_RECOVER_PASS1);
3766 kmem_free(log->l_buf_cancel_table);
3767 log->l_buf_cancel_table = NULL;
3771 * Then do a second pass to actually recover the items in the log.
3772 * When it is complete free the table of buf cancel items.
3774 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3775 XLOG_RECOVER_PASS2);
3780 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3781 ASSERT(log->l_buf_cancel_table[i] == NULL);
3785 kmem_free(log->l_buf_cancel_table);
3786 log->l_buf_cancel_table = NULL;
3792 * Do the actual recovery
3797 xfs_daddr_t head_blk,
3798 xfs_daddr_t tail_blk)
3805 * First replay the images in the log.
3807 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3812 XFS_bflush(log->l_mp->m_ddev_targp);
3815 * If IO errors happened during recovery, bail out.
3817 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3822 * We now update the tail_lsn since much of the recovery has completed
3823 * and there may be space available to use. If there were no extent
3824 * or iunlinks, we can free up the entire log and set the tail_lsn to
3825 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3826 * lsn of the last known good LR on disk. If there are extent frees
3827 * or iunlinks they will have some entries in the AIL; so we look at
3828 * the AIL to determine how to set the tail_lsn.
3830 xlog_assign_tail_lsn(log->l_mp);
3833 * Now that we've finished replaying all buffer and inode
3834 * updates, re-read in the superblock.
3836 bp = xfs_getsb(log->l_mp, 0);
3838 ASSERT(!(XFS_BUF_ISWRITE(bp)));
3839 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
3841 XFS_BUF_UNASYNC(bp);
3842 xfsbdstrat(log->l_mp, bp);
3843 error = xfs_iowait(bp);
3845 xfs_ioerror_alert("xlog_do_recover",
3846 log->l_mp, bp, XFS_BUF_ADDR(bp));
3852 /* Convert superblock from on-disk format */
3853 sbp = &log->l_mp->m_sb;
3854 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
3855 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
3856 ASSERT(xfs_sb_good_version(sbp));
3859 /* We've re-read the superblock so re-initialize per-cpu counters */
3860 xfs_icsb_reinit_counters(log->l_mp);
3862 xlog_recover_check_summary(log);
3864 /* Normal transactions can now occur */
3865 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3870 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3872 * Return error or zero.
3878 xfs_daddr_t head_blk, tail_blk;
3881 /* find the tail of the log */
3882 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
3885 if (tail_blk != head_blk) {
3886 /* There used to be a comment here:
3888 * disallow recovery on read-only mounts. note -- mount
3889 * checks for ENOSPC and turns it into an intelligent
3891 * ...but this is no longer true. Now, unless you specify
3892 * NORECOVERY (in which case this function would never be
3893 * called), we just go ahead and recover. We do this all
3894 * under the vfs layer, so we can get away with it unless
3895 * the device itself is read-only, in which case we fail.
3897 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
3902 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3903 log->l_mp->m_fsname, log->l_mp->m_logname ?
3904 log->l_mp->m_logname : "internal");
3906 error = xlog_do_recover(log, head_blk, tail_blk);
3907 log->l_flags |= XLOG_RECOVERY_NEEDED;
3913 * In the first part of recovery we replay inodes and buffers and build
3914 * up the list of extent free items which need to be processed. Here
3915 * we process the extent free items and clean up the on disk unlinked
3916 * inode lists. This is separated from the first part of recovery so
3917 * that the root and real-time bitmap inodes can be read in from disk in
3918 * between the two stages. This is necessary so that we can free space
3919 * in the real-time portion of the file system.
3922 xlog_recover_finish(
3926 * Now we're ready to do the transactions needed for the
3927 * rest of recovery. Start with completing all the extent
3928 * free intent records and then process the unlinked inode
3929 * lists. At this point, we essentially run in normal mode
3930 * except that we're still performing recovery actions
3931 * rather than accepting new requests.
3933 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
3935 error = xlog_recover_process_efis(log);
3938 "Failed to recover EFIs on filesystem: %s",
3939 log->l_mp->m_fsname);
3943 * Sync the log to get all the EFIs out of the AIL.
3944 * This isn't absolutely necessary, but it helps in
3945 * case the unlink transactions would have problems
3946 * pushing the EFIs out of the way.
3948 xfs_log_force(log->l_mp, (xfs_lsn_t)0,
3949 (XFS_LOG_FORCE | XFS_LOG_SYNC));
3951 xlog_recover_process_iunlinks(log);
3953 xlog_recover_check_summary(log);
3956 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3957 log->l_mp->m_fsname, log->l_mp->m_logname ?
3958 log->l_mp->m_logname : "internal");
3959 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
3962 "!Ending clean XFS mount for filesystem: %s\n",
3963 log->l_mp->m_fsname);
3971 * Read all of the agf and agi counters and check that they
3972 * are consistent with the superblock counters.
3975 xlog_recover_check_summary(
3983 xfs_daddr_t agfdaddr;
3984 xfs_daddr_t agidaddr;
3986 #ifdef XFS_LOUD_RECOVERY
3989 xfs_agnumber_t agno;
3990 __uint64_t freeblks;
3999 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
4000 agfdaddr = XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp));
4001 agfbp = xfs_buf_read(mp->m_ddev_targp, agfdaddr,
4002 XFS_FSS_TO_BB(mp, 1), 0);
4003 if (XFS_BUF_ISERROR(agfbp)) {
4004 xfs_ioerror_alert("xlog_recover_check_summary(agf)",
4005 mp, agfbp, agfdaddr);
4007 agfp = XFS_BUF_TO_AGF(agfbp);
4008 ASSERT(XFS_AGF_MAGIC == be32_to_cpu(agfp->agf_magicnum));
4009 ASSERT(XFS_AGF_GOOD_VERSION(be32_to_cpu(agfp->agf_versionnum)));
4010 ASSERT(be32_to_cpu(agfp->agf_seqno) == agno);
4012 freeblks += be32_to_cpu(agfp->agf_freeblks) +
4013 be32_to_cpu(agfp->agf_flcount);
4014 xfs_buf_relse(agfbp);
4016 agidaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
4017 agibp = xfs_buf_read(mp->m_ddev_targp, agidaddr,
4018 XFS_FSS_TO_BB(mp, 1), 0);
4019 if (XFS_BUF_ISERROR(agibp)) {
4020 xfs_ioerror_alert("xlog_recover_check_summary(agi)",
4021 mp, agibp, agidaddr);
4023 agip = XFS_BUF_TO_AGI(agibp);
4024 ASSERT(XFS_AGI_MAGIC == be32_to_cpu(agip->agi_magicnum));
4025 ASSERT(XFS_AGI_GOOD_VERSION(be32_to_cpu(agip->agi_versionnum)));
4026 ASSERT(be32_to_cpu(agip->agi_seqno) == agno);
4028 itotal += be32_to_cpu(agip->agi_count);
4029 ifree += be32_to_cpu(agip->agi_freecount);
4030 xfs_buf_relse(agibp);
4033 sbbp = xfs_getsb(mp, 0);
4034 #ifdef XFS_LOUD_RECOVERY
4036 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(sbbp));
4038 "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
4039 sbp->sb_icount, itotal);
4041 "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
4042 sbp->sb_ifree, ifree);
4044 "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
4045 sbp->sb_fdblocks, freeblks);
4048 * This is turned off until I account for the allocation
4049 * btree blocks which live in free space.
4051 ASSERT(sbp->sb_icount == itotal);
4052 ASSERT(sbp->sb_ifree == ifree);
4053 ASSERT(sbp->sb_fdblocks == freeblks);
4056 xfs_buf_relse(sbbp);