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"
50 STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *);
51 STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t);
52 STATIC void xlog_recover_insert_item_backq(xlog_recover_item_t **q,
53 xlog_recover_item_t *item);
55 STATIC void xlog_recover_check_summary(xlog_t *);
56 STATIC void xlog_recover_check_ail(xfs_mount_t *, xfs_log_item_t *, int);
58 #define xlog_recover_check_summary(log)
59 #define xlog_recover_check_ail(mp, lip, gen)
64 * Sector aligned buffer routines for buffer create/read/write/access
67 #define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \
68 ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \
69 ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) )
70 #define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask)
77 ASSERT(num_bblks > 0);
79 if (log->l_sectbb_log) {
81 num_bblks += XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
82 num_bblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, num_bblks);
84 return xfs_buf_get_noaddr(BBTOB(num_bblks), log->l_mp->m_logdev_targp);
96 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
107 if (log->l_sectbb_log) {
108 blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
109 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
113 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
116 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
119 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
120 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
122 xfsbdstrat(log->l_mp, bp);
123 if ((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 = ", __FUNCTION__);
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 XFS_BUF_SET_PTR(bp, offset + balign, BBTOB(sectbb));
1164 if ((error = xlog_bread(log, ealign, sectbb, bp)))
1166 XFS_BUF_SET_PTR(bp, offset, bufblks);
1169 offset = xlog_align(log, start_block, endcount, bp);
1170 for (; j < endcount; j++) {
1171 xlog_add_record(log, offset, cycle, i+j,
1172 tail_cycle, tail_block);
1175 error = xlog_bwrite(log, start_block, endcount, bp);
1178 start_block += endcount;
1186 * This routine is called to blow away any incomplete log writes out
1187 * in front of the log head. We do this so that we won't become confused
1188 * if we come up, write only a little bit more, and then crash again.
1189 * If we leave the partial log records out there, this situation could
1190 * cause us to think those partial writes are valid blocks since they
1191 * have the current cycle number. We get rid of them by overwriting them
1192 * with empty log records with the old cycle number rather than the
1195 * The tail lsn is passed in rather than taken from
1196 * the log so that we will not write over the unmount record after a
1197 * clean unmount in a 512 block log. Doing so would leave the log without
1198 * any valid log records in it until a new one was written. If we crashed
1199 * during that time we would not be able to recover.
1202 xlog_clear_stale_blocks(
1206 int tail_cycle, head_cycle;
1207 int tail_block, head_block;
1208 int tail_distance, max_distance;
1212 tail_cycle = CYCLE_LSN(tail_lsn);
1213 tail_block = BLOCK_LSN(tail_lsn);
1214 head_cycle = log->l_curr_cycle;
1215 head_block = log->l_curr_block;
1218 * Figure out the distance between the new head of the log
1219 * and the tail. We want to write over any blocks beyond the
1220 * head that we may have written just before the crash, but
1221 * we don't want to overwrite the tail of the log.
1223 if (head_cycle == tail_cycle) {
1225 * The tail is behind the head in the physical log,
1226 * so the distance from the head to the tail is the
1227 * distance from the head to the end of the log plus
1228 * the distance from the beginning of the log to the
1231 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1232 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1233 XFS_ERRLEVEL_LOW, log->l_mp);
1234 return XFS_ERROR(EFSCORRUPTED);
1236 tail_distance = tail_block + (log->l_logBBsize - head_block);
1239 * The head is behind the tail in the physical log,
1240 * so the distance from the head to the tail is just
1241 * the tail block minus the head block.
1243 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1244 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1245 XFS_ERRLEVEL_LOW, log->l_mp);
1246 return XFS_ERROR(EFSCORRUPTED);
1248 tail_distance = tail_block - head_block;
1252 * If the head is right up against the tail, we can't clear
1255 if (tail_distance <= 0) {
1256 ASSERT(tail_distance == 0);
1260 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1262 * Take the smaller of the maximum amount of outstanding I/O
1263 * we could have and the distance to the tail to clear out.
1264 * We take the smaller so that we don't overwrite the tail and
1265 * we don't waste all day writing from the head to the tail
1268 max_distance = MIN(max_distance, tail_distance);
1270 if ((head_block + max_distance) <= log->l_logBBsize) {
1272 * We can stomp all the blocks we need to without
1273 * wrapping around the end of the log. Just do it
1274 * in a single write. Use the cycle number of the
1275 * current cycle minus one so that the log will look like:
1278 error = xlog_write_log_records(log, (head_cycle - 1),
1279 head_block, max_distance, tail_cycle,
1285 * We need to wrap around the end of the physical log in
1286 * order to clear all the blocks. Do it in two separate
1287 * I/Os. The first write should be from the head to the
1288 * end of the physical log, and it should use the current
1289 * cycle number minus one just like above.
1291 distance = log->l_logBBsize - head_block;
1292 error = xlog_write_log_records(log, (head_cycle - 1),
1293 head_block, distance, tail_cycle,
1300 * Now write the blocks at the start of the physical log.
1301 * This writes the remainder of the blocks we want to clear.
1302 * It uses the current cycle number since we're now on the
1303 * same cycle as the head so that we get:
1304 * n ... n ... | n - 1 ...
1305 * ^^^^^ blocks we're writing
1307 distance = max_distance - (log->l_logBBsize - head_block);
1308 error = xlog_write_log_records(log, head_cycle, 0, distance,
1309 tail_cycle, tail_block);
1317 /******************************************************************************
1319 * Log recover routines
1321 ******************************************************************************
1324 STATIC xlog_recover_t *
1325 xlog_recover_find_tid(
1329 xlog_recover_t *p = q;
1332 if (p->r_log_tid == tid)
1340 xlog_recover_put_hashq(
1342 xlog_recover_t *trans)
1349 xlog_recover_add_item(
1350 xlog_recover_item_t **itemq)
1352 xlog_recover_item_t *item;
1354 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1355 xlog_recover_insert_item_backq(itemq, item);
1359 xlog_recover_add_to_cont_trans(
1360 xlog_recover_t *trans,
1364 xlog_recover_item_t *item;
1365 xfs_caddr_t ptr, old_ptr;
1368 item = trans->r_itemq;
1370 /* finish copying rest of trans header */
1371 xlog_recover_add_item(&trans->r_itemq);
1372 ptr = (xfs_caddr_t) &trans->r_theader +
1373 sizeof(xfs_trans_header_t) - len;
1374 memcpy(ptr, dp, len); /* d, s, l */
1377 item = item->ri_prev;
1379 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1380 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1382 ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
1383 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1384 item->ri_buf[item->ri_cnt-1].i_len += len;
1385 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1390 * The next region to add is the start of a new region. It could be
1391 * a whole region or it could be the first part of a new region. Because
1392 * of this, the assumption here is that the type and size fields of all
1393 * format structures fit into the first 32 bits of the structure.
1395 * This works because all regions must be 32 bit aligned. Therefore, we
1396 * either have both fields or we have neither field. In the case we have
1397 * neither field, the data part of the region is zero length. We only have
1398 * a log_op_header and can throw away the header since a new one will appear
1399 * later. If we have at least 4 bytes, then we can determine how many regions
1400 * will appear in the current log item.
1403 xlog_recover_add_to_trans(
1404 xlog_recover_t *trans,
1408 xfs_inode_log_format_t *in_f; /* any will do */
1409 xlog_recover_item_t *item;
1414 item = trans->r_itemq;
1416 ASSERT(*(uint *)dp == XFS_TRANS_HEADER_MAGIC);
1417 if (len == sizeof(xfs_trans_header_t))
1418 xlog_recover_add_item(&trans->r_itemq);
1419 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1423 ptr = kmem_alloc(len, KM_SLEEP);
1424 memcpy(ptr, dp, len);
1425 in_f = (xfs_inode_log_format_t *)ptr;
1427 if (item->ri_prev->ri_total != 0 &&
1428 item->ri_prev->ri_total == item->ri_prev->ri_cnt) {
1429 xlog_recover_add_item(&trans->r_itemq);
1431 item = trans->r_itemq;
1432 item = item->ri_prev;
1434 if (item->ri_total == 0) { /* first region to be added */
1435 item->ri_total = in_f->ilf_size;
1436 ASSERT(item->ri_total <= XLOG_MAX_REGIONS_IN_ITEM);
1437 item->ri_buf = kmem_zalloc((item->ri_total *
1438 sizeof(xfs_log_iovec_t)), KM_SLEEP);
1440 ASSERT(item->ri_total > item->ri_cnt);
1441 /* Description region is ri_buf[0] */
1442 item->ri_buf[item->ri_cnt].i_addr = ptr;
1443 item->ri_buf[item->ri_cnt].i_len = len;
1449 xlog_recover_new_tid(
1454 xlog_recover_t *trans;
1456 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1457 trans->r_log_tid = tid;
1459 xlog_recover_put_hashq(q, trans);
1463 xlog_recover_unlink_tid(
1465 xlog_recover_t *trans)
1470 ASSERT(trans != NULL);
1476 if (tp->r_next == trans) {
1484 "XFS: xlog_recover_unlink_tid: trans not found");
1486 return XFS_ERROR(EIO);
1488 tp->r_next = tp->r_next->r_next;
1494 xlog_recover_insert_item_backq(
1495 xlog_recover_item_t **q,
1496 xlog_recover_item_t *item)
1499 item->ri_prev = item->ri_next = item;
1503 item->ri_prev = (*q)->ri_prev;
1504 (*q)->ri_prev = item;
1505 item->ri_prev->ri_next = item;
1510 xlog_recover_insert_item_frontq(
1511 xlog_recover_item_t **q,
1512 xlog_recover_item_t *item)
1514 xlog_recover_insert_item_backq(q, item);
1519 xlog_recover_reorder_trans(
1520 xlog_recover_t *trans)
1522 xlog_recover_item_t *first_item, *itemq, *itemq_next;
1523 xfs_buf_log_format_t *buf_f;
1526 first_item = itemq = trans->r_itemq;
1527 trans->r_itemq = NULL;
1529 itemq_next = itemq->ri_next;
1530 buf_f = (xfs_buf_log_format_t *)itemq->ri_buf[0].i_addr;
1532 switch (ITEM_TYPE(itemq)) {
1534 flags = buf_f->blf_flags;
1535 if (!(flags & XFS_BLI_CANCEL)) {
1536 xlog_recover_insert_item_frontq(&trans->r_itemq,
1542 case XFS_LI_QUOTAOFF:
1545 xlog_recover_insert_item_backq(&trans->r_itemq, itemq);
1549 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1551 return XFS_ERROR(EIO);
1554 } while (first_item != itemq);
1559 * Build up the table of buf cancel records so that we don't replay
1560 * cancelled data in the second pass. For buffer records that are
1561 * not cancel records, there is nothing to do here so we just return.
1563 * If we get a cancel record which is already in the table, this indicates
1564 * that the buffer was cancelled multiple times. In order to ensure
1565 * that during pass 2 we keep the record in the table until we reach its
1566 * last occurrence in the log, we keep a reference count in the cancel
1567 * record in the table to tell us how many times we expect to see this
1568 * record during the second pass.
1571 xlog_recover_do_buffer_pass1(
1573 xfs_buf_log_format_t *buf_f)
1575 xfs_buf_cancel_t *bcp;
1576 xfs_buf_cancel_t *nextp;
1577 xfs_buf_cancel_t *prevp;
1578 xfs_buf_cancel_t **bucket;
1579 xfs_daddr_t blkno = 0;
1583 switch (buf_f->blf_type) {
1585 blkno = buf_f->blf_blkno;
1586 len = buf_f->blf_len;
1587 flags = buf_f->blf_flags;
1592 * If this isn't a cancel buffer item, then just return.
1594 if (!(flags & XFS_BLI_CANCEL))
1598 * Insert an xfs_buf_cancel record into the hash table of
1599 * them. If there is already an identical record, bump
1600 * its reference count.
1602 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1603 XLOG_BC_TABLE_SIZE];
1605 * If the hash bucket is empty then just insert a new record into
1608 if (*bucket == NULL) {
1609 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1611 bcp->bc_blkno = blkno;
1613 bcp->bc_refcount = 1;
1614 bcp->bc_next = NULL;
1620 * The hash bucket is not empty, so search for duplicates of our
1621 * record. If we find one them just bump its refcount. If not
1622 * then add us at the end of the list.
1626 while (nextp != NULL) {
1627 if (nextp->bc_blkno == blkno && nextp->bc_len == len) {
1628 nextp->bc_refcount++;
1632 nextp = nextp->bc_next;
1634 ASSERT(prevp != NULL);
1635 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1637 bcp->bc_blkno = blkno;
1639 bcp->bc_refcount = 1;
1640 bcp->bc_next = NULL;
1641 prevp->bc_next = bcp;
1645 * Check to see whether the buffer being recovered has a corresponding
1646 * entry in the buffer cancel record table. If it does then return 1
1647 * so that it will be cancelled, otherwise return 0. If the buffer is
1648 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
1649 * the refcount on the entry in the table and remove it from the table
1650 * if this is the last reference.
1652 * We remove the cancel record from the table when we encounter its
1653 * last occurrence in the log so that if the same buffer is re-used
1654 * again after its last cancellation we actually replay the changes
1655 * made at that point.
1658 xlog_check_buffer_cancelled(
1664 xfs_buf_cancel_t *bcp;
1665 xfs_buf_cancel_t *prevp;
1666 xfs_buf_cancel_t **bucket;
1668 if (log->l_buf_cancel_table == NULL) {
1670 * There is nothing in the table built in pass one,
1671 * so this buffer must not be cancelled.
1673 ASSERT(!(flags & XFS_BLI_CANCEL));
1677 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1678 XLOG_BC_TABLE_SIZE];
1682 * There is no corresponding entry in the table built
1683 * in pass one, so this buffer has not been cancelled.
1685 ASSERT(!(flags & XFS_BLI_CANCEL));
1690 * Search for an entry in the buffer cancel table that
1691 * matches our buffer.
1694 while (bcp != NULL) {
1695 if (bcp->bc_blkno == blkno && bcp->bc_len == len) {
1697 * We've go a match, so return 1 so that the
1698 * recovery of this buffer is cancelled.
1699 * If this buffer is actually a buffer cancel
1700 * log item, then decrement the refcount on the
1701 * one in the table and remove it if this is the
1704 if (flags & XFS_BLI_CANCEL) {
1706 if (bcp->bc_refcount == 0) {
1707 if (prevp == NULL) {
1708 *bucket = bcp->bc_next;
1710 prevp->bc_next = bcp->bc_next;
1713 sizeof(xfs_buf_cancel_t));
1722 * We didn't find a corresponding entry in the table, so
1723 * return 0 so that the buffer is NOT cancelled.
1725 ASSERT(!(flags & XFS_BLI_CANCEL));
1730 xlog_recover_do_buffer_pass2(
1732 xfs_buf_log_format_t *buf_f)
1734 xfs_daddr_t blkno = 0;
1738 switch (buf_f->blf_type) {
1740 blkno = buf_f->blf_blkno;
1741 flags = buf_f->blf_flags;
1742 len = buf_f->blf_len;
1746 return xlog_check_buffer_cancelled(log, blkno, len, flags);
1750 * Perform recovery for a buffer full of inodes. In these buffers,
1751 * the only data which should be recovered is that which corresponds
1752 * to the di_next_unlinked pointers in the on disk inode structures.
1753 * The rest of the data for the inodes is always logged through the
1754 * inodes themselves rather than the inode buffer and is recovered
1755 * in xlog_recover_do_inode_trans().
1757 * The only time when buffers full of inodes are fully recovered is
1758 * when the buffer is full of newly allocated inodes. In this case
1759 * the buffer will not be marked as an inode buffer and so will be
1760 * sent to xlog_recover_do_reg_buffer() below during recovery.
1763 xlog_recover_do_inode_buffer(
1765 xlog_recover_item_t *item,
1767 xfs_buf_log_format_t *buf_f)
1775 int next_unlinked_offset;
1777 xfs_agino_t *logged_nextp;
1778 xfs_agino_t *buffer_nextp;
1779 unsigned int *data_map = NULL;
1780 unsigned int map_size = 0;
1782 switch (buf_f->blf_type) {
1784 data_map = buf_f->blf_data_map;
1785 map_size = buf_f->blf_map_size;
1789 * Set the variables corresponding to the current region to
1790 * 0 so that we'll initialize them on the first pass through
1798 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1799 for (i = 0; i < inodes_per_buf; i++) {
1800 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1801 offsetof(xfs_dinode_t, di_next_unlinked);
1803 while (next_unlinked_offset >=
1804 (reg_buf_offset + reg_buf_bytes)) {
1806 * The next di_next_unlinked field is beyond
1807 * the current logged region. Find the next
1808 * logged region that contains or is beyond
1809 * the current di_next_unlinked field.
1812 bit = xfs_next_bit(data_map, map_size, bit);
1815 * If there are no more logged regions in the
1816 * buffer, then we're done.
1822 nbits = xfs_contig_bits(data_map, map_size,
1825 reg_buf_offset = bit << XFS_BLI_SHIFT;
1826 reg_buf_bytes = nbits << XFS_BLI_SHIFT;
1831 * If the current logged region starts after the current
1832 * di_next_unlinked field, then move on to the next
1833 * di_next_unlinked field.
1835 if (next_unlinked_offset < reg_buf_offset) {
1839 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1840 ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0);
1841 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1844 * The current logged region contains a copy of the
1845 * current di_next_unlinked field. Extract its value
1846 * and copy it to the buffer copy.
1848 logged_nextp = (xfs_agino_t *)
1849 ((char *)(item->ri_buf[item_index].i_addr) +
1850 (next_unlinked_offset - reg_buf_offset));
1851 if (unlikely(*logged_nextp == 0)) {
1852 xfs_fs_cmn_err(CE_ALERT, mp,
1853 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1855 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1856 XFS_ERRLEVEL_LOW, mp);
1857 return XFS_ERROR(EFSCORRUPTED);
1860 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1861 next_unlinked_offset);
1862 *buffer_nextp = *logged_nextp;
1869 * Perform a 'normal' buffer recovery. Each logged region of the
1870 * buffer should be copied over the corresponding region in the
1871 * given buffer. The bitmap in the buf log format structure indicates
1872 * where to place the logged data.
1876 xlog_recover_do_reg_buffer(
1877 xlog_recover_item_t *item,
1879 xfs_buf_log_format_t *buf_f)
1884 unsigned int *data_map = NULL;
1885 unsigned int map_size = 0;
1888 switch (buf_f->blf_type) {
1890 data_map = buf_f->blf_data_map;
1891 map_size = buf_f->blf_map_size;
1895 i = 1; /* 0 is the buf format structure */
1897 bit = xfs_next_bit(data_map, map_size, bit);
1900 nbits = xfs_contig_bits(data_map, map_size, bit);
1902 ASSERT(item->ri_buf[i].i_addr != NULL);
1903 ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0);
1904 ASSERT(XFS_BUF_COUNT(bp) >=
1905 ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT));
1908 * Do a sanity check if this is a dquot buffer. Just checking
1909 * the first dquot in the buffer should do. XXXThis is
1910 * probably a good thing to do for other buf types also.
1913 if (buf_f->blf_flags &
1914 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
1915 error = xfs_qm_dqcheck((xfs_disk_dquot_t *)
1916 item->ri_buf[i].i_addr,
1917 -1, 0, XFS_QMOPT_DOWARN,
1918 "dquot_buf_recover");
1921 memcpy(xfs_buf_offset(bp,
1922 (uint)bit << XFS_BLI_SHIFT), /* dest */
1923 item->ri_buf[i].i_addr, /* source */
1924 nbits<<XFS_BLI_SHIFT); /* length */
1929 /* Shouldn't be any more regions */
1930 ASSERT(i == item->ri_total);
1934 * Do some primitive error checking on ondisk dquot data structures.
1938 xfs_disk_dquot_t *ddq,
1940 uint type, /* used only when IO_dorepair is true */
1944 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
1948 * We can encounter an uninitialized dquot buffer for 2 reasons:
1949 * 1. If we crash while deleting the quotainode(s), and those blks got
1950 * used for user data. This is because we take the path of regular
1951 * file deletion; however, the size field of quotainodes is never
1952 * updated, so all the tricks that we play in itruncate_finish
1953 * don't quite matter.
1955 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1956 * But the allocation will be replayed so we'll end up with an
1957 * uninitialized quota block.
1959 * This is all fine; things are still consistent, and we haven't lost
1960 * any quota information. Just don't complain about bad dquot blks.
1962 if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) {
1963 if (flags & XFS_QMOPT_DOWARN)
1965 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
1966 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
1969 if (ddq->d_version != XFS_DQUOT_VERSION) {
1970 if (flags & XFS_QMOPT_DOWARN)
1972 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
1973 str, id, ddq->d_version, XFS_DQUOT_VERSION);
1977 if (ddq->d_flags != XFS_DQ_USER &&
1978 ddq->d_flags != XFS_DQ_PROJ &&
1979 ddq->d_flags != XFS_DQ_GROUP) {
1980 if (flags & XFS_QMOPT_DOWARN)
1982 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
1983 str, id, ddq->d_flags);
1987 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
1988 if (flags & XFS_QMOPT_DOWARN)
1990 "%s : ondisk-dquot 0x%p, ID mismatch: "
1991 "0x%x expected, found id 0x%x",
1992 str, ddq, id, be32_to_cpu(ddq->d_id));
1996 if (!errs && ddq->d_id) {
1997 if (ddq->d_blk_softlimit &&
1998 be64_to_cpu(ddq->d_bcount) >=
1999 be64_to_cpu(ddq->d_blk_softlimit)) {
2000 if (!ddq->d_btimer) {
2001 if (flags & XFS_QMOPT_DOWARN)
2003 "%s : Dquot ID 0x%x (0x%p) "
2004 "BLK TIMER NOT STARTED",
2005 str, (int)be32_to_cpu(ddq->d_id), ddq);
2009 if (ddq->d_ino_softlimit &&
2010 be64_to_cpu(ddq->d_icount) >=
2011 be64_to_cpu(ddq->d_ino_softlimit)) {
2012 if (!ddq->d_itimer) {
2013 if (flags & XFS_QMOPT_DOWARN)
2015 "%s : Dquot ID 0x%x (0x%p) "
2016 "INODE TIMER NOT STARTED",
2017 str, (int)be32_to_cpu(ddq->d_id), ddq);
2021 if (ddq->d_rtb_softlimit &&
2022 be64_to_cpu(ddq->d_rtbcount) >=
2023 be64_to_cpu(ddq->d_rtb_softlimit)) {
2024 if (!ddq->d_rtbtimer) {
2025 if (flags & XFS_QMOPT_DOWARN)
2027 "%s : Dquot ID 0x%x (0x%p) "
2028 "RTBLK TIMER NOT STARTED",
2029 str, (int)be32_to_cpu(ddq->d_id), ddq);
2035 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2038 if (flags & XFS_QMOPT_DOWARN)
2039 cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id);
2042 * Typically, a repair is only requested by quotacheck.
2045 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2046 memset(d, 0, sizeof(xfs_dqblk_t));
2048 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2049 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2050 d->dd_diskdq.d_flags = type;
2051 d->dd_diskdq.d_id = cpu_to_be32(id);
2057 * Perform a dquot buffer recovery.
2058 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2059 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2060 * Else, treat it as a regular buffer and do recovery.
2063 xlog_recover_do_dquot_buffer(
2066 xlog_recover_item_t *item,
2068 xfs_buf_log_format_t *buf_f)
2073 * Filesystems are required to send in quota flags at mount time.
2075 if (mp->m_qflags == 0) {
2080 if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF)
2081 type |= XFS_DQ_USER;
2082 if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF)
2083 type |= XFS_DQ_PROJ;
2084 if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF)
2085 type |= XFS_DQ_GROUP;
2087 * This type of quotas was turned off, so ignore this buffer
2089 if (log->l_quotaoffs_flag & type)
2092 xlog_recover_do_reg_buffer(item, bp, buf_f);
2096 * This routine replays a modification made to a buffer at runtime.
2097 * There are actually two types of buffer, regular and inode, which
2098 * are handled differently. Inode buffers are handled differently
2099 * in that we only recover a specific set of data from them, namely
2100 * the inode di_next_unlinked fields. This is because all other inode
2101 * data is actually logged via inode records and any data we replay
2102 * here which overlaps that may be stale.
2104 * When meta-data buffers are freed at run time we log a buffer item
2105 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
2106 * of the buffer in the log should not be replayed at recovery time.
2107 * This is so that if the blocks covered by the buffer are reused for
2108 * file data before we crash we don't end up replaying old, freed
2109 * meta-data into a user's file.
2111 * To handle the cancellation of buffer log items, we make two passes
2112 * over the log during recovery. During the first we build a table of
2113 * those buffers which have been cancelled, and during the second we
2114 * only replay those buffers which do not have corresponding cancel
2115 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2116 * for more details on the implementation of the table of cancel records.
2119 xlog_recover_do_buffer_trans(
2121 xlog_recover_item_t *item,
2124 xfs_buf_log_format_t *buf_f;
2133 buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr;
2135 if (pass == XLOG_RECOVER_PASS1) {
2137 * In this pass we're only looking for buf items
2138 * with the XFS_BLI_CANCEL bit set.
2140 xlog_recover_do_buffer_pass1(log, buf_f);
2144 * In this pass we want to recover all the buffers
2145 * which have not been cancelled and are not
2146 * cancellation buffers themselves. The routine
2147 * we call here will tell us whether or not to
2148 * continue with the replay of this buffer.
2150 cancel = xlog_recover_do_buffer_pass2(log, buf_f);
2155 switch (buf_f->blf_type) {
2157 blkno = buf_f->blf_blkno;
2158 len = buf_f->blf_len;
2159 flags = buf_f->blf_flags;
2162 xfs_fs_cmn_err(CE_ALERT, log->l_mp,
2163 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2164 buf_f->blf_type, log->l_mp->m_logname ?
2165 log->l_mp->m_logname : "internal");
2166 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2167 XFS_ERRLEVEL_LOW, log->l_mp);
2168 return XFS_ERROR(EFSCORRUPTED);
2172 if (flags & XFS_BLI_INODE_BUF) {
2173 bp = xfs_buf_read_flags(mp->m_ddev_targp, blkno, len,
2176 bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, 0);
2178 if (XFS_BUF_ISERROR(bp)) {
2179 xfs_ioerror_alert("xlog_recover_do..(read#1)", log->l_mp,
2181 error = XFS_BUF_GETERROR(bp);
2187 if (flags & XFS_BLI_INODE_BUF) {
2188 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2190 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
2191 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2193 xlog_recover_do_reg_buffer(item, bp, buf_f);
2196 return XFS_ERROR(error);
2199 * Perform delayed write on the buffer. Asynchronous writes will be
2200 * slower when taking into account all the buffers to be flushed.
2202 * Also make sure that only inode buffers with good sizes stay in
2203 * the buffer cache. The kernel moves inodes in buffers of 1 block
2204 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2205 * buffers in the log can be a different size if the log was generated
2206 * by an older kernel using unclustered inode buffers or a newer kernel
2207 * running with a different inode cluster size. Regardless, if the
2208 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2209 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2210 * the buffer out of the buffer cache so that the buffer won't
2211 * overlap with future reads of those inodes.
2213 if (XFS_DINODE_MAGIC ==
2214 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2215 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2216 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2218 error = xfs_bwrite(mp, bp);
2220 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2221 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2222 XFS_BUF_SET_FSPRIVATE(bp, mp);
2223 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2224 xfs_bdwrite(mp, bp);
2231 xlog_recover_do_inode_trans(
2233 xlog_recover_item_t *item,
2236 xfs_inode_log_format_t *in_f;
2248 xfs_icdinode_t *dicp;
2251 if (pass == XLOG_RECOVER_PASS1) {
2255 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2256 in_f = (xfs_inode_log_format_t *)item->ri_buf[0].i_addr;
2258 in_f = (xfs_inode_log_format_t *)kmem_alloc(
2259 sizeof(xfs_inode_log_format_t), KM_SLEEP);
2261 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2265 ino = in_f->ilf_ino;
2267 if (ITEM_TYPE(item) == XFS_LI_INODE) {
2268 imap.im_blkno = (xfs_daddr_t)in_f->ilf_blkno;
2269 imap.im_len = in_f->ilf_len;
2270 imap.im_boffset = in_f->ilf_boffset;
2273 * It's an old inode format record. We don't know where
2274 * its cluster is located on disk, and we can't allow
2275 * xfs_imap() to figure it out because the inode btrees
2276 * are not ready to be used. Therefore do not pass the
2277 * XFS_IMAP_LOOKUP flag to xfs_imap(). This will give
2278 * us only the single block in which the inode lives
2279 * rather than its cluster, so we must make sure to
2280 * invalidate the buffer when we write it out below.
2283 xfs_imap(log->l_mp, NULL, ino, &imap, 0);
2287 * Inode buffers can be freed, look out for it,
2288 * and do not replay the inode.
2290 if (xlog_check_buffer_cancelled(log, imap.im_blkno, imap.im_len, 0)) {
2295 bp = xfs_buf_read_flags(mp->m_ddev_targp, imap.im_blkno, imap.im_len,
2297 if (XFS_BUF_ISERROR(bp)) {
2298 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
2300 error = XFS_BUF_GETERROR(bp);
2305 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2306 dip = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
2309 * Make sure the place we're flushing out to really looks
2312 if (unlikely(be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC)) {
2314 xfs_fs_cmn_err(CE_ALERT, mp,
2315 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2317 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2318 XFS_ERRLEVEL_LOW, mp);
2319 error = EFSCORRUPTED;
2322 dicp = (xfs_icdinode_t *)(item->ri_buf[1].i_addr);
2323 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2325 xfs_fs_cmn_err(CE_ALERT, mp,
2326 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2328 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2329 XFS_ERRLEVEL_LOW, mp);
2330 error = EFSCORRUPTED;
2334 /* Skip replay when the on disk inode is newer than the log one */
2335 if (dicp->di_flushiter < be16_to_cpu(dip->di_core.di_flushiter)) {
2337 * Deal with the wrap case, DI_MAX_FLUSH is less
2338 * than smaller numbers
2340 if (be16_to_cpu(dip->di_core.di_flushiter) == DI_MAX_FLUSH &&
2341 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2349 /* Take the opportunity to reset the flush iteration count */
2350 dicp->di_flushiter = 0;
2352 if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) {
2353 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2354 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2355 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2356 XFS_ERRLEVEL_LOW, mp, dicp);
2358 xfs_fs_cmn_err(CE_ALERT, mp,
2359 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2360 item, dip, bp, ino);
2361 error = EFSCORRUPTED;
2364 } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) {
2365 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2366 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2367 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2368 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2369 XFS_ERRLEVEL_LOW, mp, dicp);
2371 xfs_fs_cmn_err(CE_ALERT, mp,
2372 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2373 item, dip, bp, ino);
2374 error = EFSCORRUPTED;
2378 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2379 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2380 XFS_ERRLEVEL_LOW, mp, dicp);
2382 xfs_fs_cmn_err(CE_ALERT, mp,
2383 "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",
2385 dicp->di_nextents + dicp->di_anextents,
2387 error = EFSCORRUPTED;
2390 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2391 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2392 XFS_ERRLEVEL_LOW, mp, dicp);
2394 xfs_fs_cmn_err(CE_ALERT, mp,
2395 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2396 item, dip, bp, ino, dicp->di_forkoff);
2397 error = EFSCORRUPTED;
2400 if (unlikely(item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t))) {
2401 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2402 XFS_ERRLEVEL_LOW, mp, dicp);
2404 xfs_fs_cmn_err(CE_ALERT, mp,
2405 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2406 item->ri_buf[1].i_len, item);
2407 error = EFSCORRUPTED;
2411 /* The core is in in-core format */
2412 xfs_dinode_to_disk(&dip->di_core,
2413 (xfs_icdinode_t *)item->ri_buf[1].i_addr);
2415 /* the rest is in on-disk format */
2416 if (item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t)) {
2417 memcpy((xfs_caddr_t) dip + sizeof(xfs_dinode_core_t),
2418 item->ri_buf[1].i_addr + sizeof(xfs_dinode_core_t),
2419 item->ri_buf[1].i_len - sizeof(xfs_dinode_core_t));
2422 fields = in_f->ilf_fields;
2423 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2425 dip->di_u.di_dev = cpu_to_be32(in_f->ilf_u.ilfu_rdev);
2428 dip->di_u.di_muuid = in_f->ilf_u.ilfu_uuid;
2432 if (in_f->ilf_size == 2)
2433 goto write_inode_buffer;
2434 len = item->ri_buf[2].i_len;
2435 src = item->ri_buf[2].i_addr;
2436 ASSERT(in_f->ilf_size <= 4);
2437 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2438 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2439 (len == in_f->ilf_dsize));
2441 switch (fields & XFS_ILOG_DFORK) {
2442 case XFS_ILOG_DDATA:
2444 memcpy(&dip->di_u, src, len);
2447 case XFS_ILOG_DBROOT:
2448 xfs_bmbt_to_bmdr((xfs_bmbt_block_t *)src, len,
2449 &(dip->di_u.di_bmbt),
2450 XFS_DFORK_DSIZE(dip, mp));
2455 * There are no data fork flags set.
2457 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2462 * If we logged any attribute data, recover it. There may or
2463 * may not have been any other non-core data logged in this
2466 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2467 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2472 len = item->ri_buf[attr_index].i_len;
2473 src = item->ri_buf[attr_index].i_addr;
2474 ASSERT(len == in_f->ilf_asize);
2476 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2477 case XFS_ILOG_ADATA:
2479 dest = XFS_DFORK_APTR(dip);
2480 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2481 memcpy(dest, src, len);
2484 case XFS_ILOG_ABROOT:
2485 dest = XFS_DFORK_APTR(dip);
2486 xfs_bmbt_to_bmdr((xfs_bmbt_block_t *)src, len,
2487 (xfs_bmdr_block_t*)dest,
2488 XFS_DFORK_ASIZE(dip, mp));
2492 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2501 if (ITEM_TYPE(item) == XFS_LI_INODE) {
2502 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2503 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2504 XFS_BUF_SET_FSPRIVATE(bp, mp);
2505 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2506 xfs_bdwrite(mp, bp);
2509 error = xfs_bwrite(mp, bp);
2514 kmem_free(in_f, sizeof(*in_f));
2515 return XFS_ERROR(error);
2519 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2520 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2524 xlog_recover_do_quotaoff_trans(
2526 xlog_recover_item_t *item,
2529 xfs_qoff_logformat_t *qoff_f;
2531 if (pass == XLOG_RECOVER_PASS2) {
2535 qoff_f = (xfs_qoff_logformat_t *)item->ri_buf[0].i_addr;
2539 * The logitem format's flag tells us if this was user quotaoff,
2540 * group/project quotaoff or both.
2542 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2543 log->l_quotaoffs_flag |= XFS_DQ_USER;
2544 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2545 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2546 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2547 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2553 * Recover a dquot record
2556 xlog_recover_do_dquot_trans(
2558 xlog_recover_item_t *item,
2563 struct xfs_disk_dquot *ddq, *recddq;
2565 xfs_dq_logformat_t *dq_f;
2568 if (pass == XLOG_RECOVER_PASS1) {
2574 * Filesystems are required to send in quota flags at mount time.
2576 if (mp->m_qflags == 0)
2579 recddq = (xfs_disk_dquot_t *)item->ri_buf[1].i_addr;
2582 * This type of quotas was turned off, so ignore this record.
2584 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
2586 if (log->l_quotaoffs_flag & type)
2590 * At this point we know that quota was _not_ turned off.
2591 * Since the mount flags are not indicating to us otherwise, this
2592 * must mean that quota is on, and the dquot needs to be replayed.
2593 * Remember that we may not have fully recovered the superblock yet,
2594 * so we can't do the usual trick of looking at the SB quota bits.
2596 * The other possibility, of course, is that the quota subsystem was
2597 * removed since the last mount - ENOSYS.
2599 dq_f = (xfs_dq_logformat_t *)item->ri_buf[0].i_addr;
2601 if ((error = xfs_qm_dqcheck(recddq,
2603 0, XFS_QMOPT_DOWARN,
2604 "xlog_recover_do_dquot_trans (log copy)"))) {
2605 return XFS_ERROR(EIO);
2607 ASSERT(dq_f->qlf_len == 1);
2609 error = xfs_read_buf(mp, mp->m_ddev_targp,
2611 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2614 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
2615 bp, dq_f->qlf_blkno);
2619 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2622 * At least the magic num portion should be on disk because this
2623 * was among a chunk of dquots created earlier, and we did some
2624 * minimal initialization then.
2626 if (xfs_qm_dqcheck(ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2627 "xlog_recover_do_dquot_trans")) {
2629 return XFS_ERROR(EIO);
2632 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2634 ASSERT(dq_f->qlf_size == 2);
2635 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2636 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2637 XFS_BUF_SET_FSPRIVATE(bp, mp);
2638 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2639 xfs_bdwrite(mp, bp);
2645 * This routine is called to create an in-core extent free intent
2646 * item from the efi format structure which was logged on disk.
2647 * It allocates an in-core efi, copies the extents from the format
2648 * structure into it, and adds the efi to the AIL with the given
2652 xlog_recover_do_efi_trans(
2654 xlog_recover_item_t *item,
2660 xfs_efi_log_item_t *efip;
2661 xfs_efi_log_format_t *efi_formatp;
2663 if (pass == XLOG_RECOVER_PASS1) {
2667 efi_formatp = (xfs_efi_log_format_t *)item->ri_buf[0].i_addr;
2670 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
2671 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2672 &(efip->efi_format)))) {
2673 xfs_efi_item_free(efip);
2676 efip->efi_next_extent = efi_formatp->efi_nextents;
2677 efip->efi_flags |= XFS_EFI_COMMITTED;
2679 spin_lock(&mp->m_ail_lock);
2681 * xfs_trans_update_ail() drops the AIL lock.
2683 xfs_trans_update_ail(mp, (xfs_log_item_t *)efip, lsn);
2689 * This routine is called when an efd format structure is found in
2690 * a committed transaction in the log. It's purpose is to cancel
2691 * the corresponding efi if it was still in the log. To do this
2692 * it searches the AIL for the efi with an id equal to that in the
2693 * efd format structure. If we find it, we remove the efi from the
2697 xlog_recover_do_efd_trans(
2699 xlog_recover_item_t *item,
2703 xfs_efd_log_format_t *efd_formatp;
2704 xfs_efi_log_item_t *efip = NULL;
2705 xfs_log_item_t *lip;
2709 if (pass == XLOG_RECOVER_PASS1) {
2713 efd_formatp = (xfs_efd_log_format_t *)item->ri_buf[0].i_addr;
2714 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2715 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2716 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2717 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
2718 efi_id = efd_formatp->efd_efi_id;
2721 * Search for the efi with the id in the efd format structure
2725 spin_lock(&mp->m_ail_lock);
2726 lip = xfs_trans_first_ail(mp, &gen);
2727 while (lip != NULL) {
2728 if (lip->li_type == XFS_LI_EFI) {
2729 efip = (xfs_efi_log_item_t *)lip;
2730 if (efip->efi_format.efi_id == efi_id) {
2732 * xfs_trans_delete_ail() drops the
2735 xfs_trans_delete_ail(mp, lip);
2736 xfs_efi_item_free(efip);
2740 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
2742 spin_unlock(&mp->m_ail_lock);
2746 * Perform the transaction
2748 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2749 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2752 xlog_recover_do_trans(
2754 xlog_recover_t *trans,
2758 xlog_recover_item_t *item, *first_item;
2760 if ((error = xlog_recover_reorder_trans(trans)))
2762 first_item = item = trans->r_itemq;
2765 * we don't need to worry about the block number being
2766 * truncated in > 1 TB buffers because in user-land,
2767 * we're now n32 or 64-bit so xfs_daddr_t is 64-bits so
2768 * the blknos will get through the user-mode buffer
2769 * cache properly. The only bad case is o32 kernels
2770 * where xfs_daddr_t is 32-bits but mount will warn us
2771 * off a > 1 TB filesystem before we get here.
2773 if ((ITEM_TYPE(item) == XFS_LI_BUF)) {
2774 if ((error = xlog_recover_do_buffer_trans(log, item,
2777 } else if ((ITEM_TYPE(item) == XFS_LI_INODE)) {
2778 if ((error = xlog_recover_do_inode_trans(log, item,
2781 } else if (ITEM_TYPE(item) == XFS_LI_EFI) {
2782 if ((error = xlog_recover_do_efi_trans(log, item, trans->r_lsn,
2785 } else if (ITEM_TYPE(item) == XFS_LI_EFD) {
2786 xlog_recover_do_efd_trans(log, item, pass);
2787 } else if (ITEM_TYPE(item) == XFS_LI_DQUOT) {
2788 if ((error = xlog_recover_do_dquot_trans(log, item,
2791 } else if ((ITEM_TYPE(item) == XFS_LI_QUOTAOFF)) {
2792 if ((error = xlog_recover_do_quotaoff_trans(log, item,
2796 xlog_warn("XFS: xlog_recover_do_trans");
2798 error = XFS_ERROR(EIO);
2801 item = item->ri_next;
2802 } while (first_item != item);
2808 * Free up any resources allocated by the transaction
2810 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2813 xlog_recover_free_trans(
2814 xlog_recover_t *trans)
2816 xlog_recover_item_t *first_item, *item, *free_item;
2819 item = first_item = trans->r_itemq;
2822 item = item->ri_next;
2823 /* Free the regions in the item. */
2824 for (i = 0; i < free_item->ri_cnt; i++) {
2825 kmem_free(free_item->ri_buf[i].i_addr,
2826 free_item->ri_buf[i].i_len);
2828 /* Free the item itself */
2829 kmem_free(free_item->ri_buf,
2830 (free_item->ri_total * sizeof(xfs_log_iovec_t)));
2831 kmem_free(free_item, sizeof(xlog_recover_item_t));
2832 } while (first_item != item);
2833 /* Free the transaction recover structure */
2834 kmem_free(trans, sizeof(xlog_recover_t));
2838 xlog_recover_commit_trans(
2841 xlog_recover_t *trans,
2846 if ((error = xlog_recover_unlink_tid(q, trans)))
2848 if ((error = xlog_recover_do_trans(log, trans, pass)))
2850 xlog_recover_free_trans(trans); /* no error */
2855 xlog_recover_unmount_trans(
2856 xlog_recover_t *trans)
2858 /* Do nothing now */
2859 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2864 * There are two valid states of the r_state field. 0 indicates that the
2865 * transaction structure is in a normal state. We have either seen the
2866 * start of the transaction or the last operation we added was not a partial
2867 * operation. If the last operation we added to the transaction was a
2868 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2870 * NOTE: skip LRs with 0 data length.
2873 xlog_recover_process_data(
2875 xlog_recover_t *rhash[],
2876 xlog_rec_header_t *rhead,
2882 xlog_op_header_t *ohead;
2883 xlog_recover_t *trans;
2889 lp = dp + be32_to_cpu(rhead->h_len);
2890 num_logops = be32_to_cpu(rhead->h_num_logops);
2892 /* check the log format matches our own - else we can't recover */
2893 if (xlog_header_check_recover(log->l_mp, rhead))
2894 return (XFS_ERROR(EIO));
2896 while ((dp < lp) && num_logops) {
2897 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2898 ohead = (xlog_op_header_t *)dp;
2899 dp += sizeof(xlog_op_header_t);
2900 if (ohead->oh_clientid != XFS_TRANSACTION &&
2901 ohead->oh_clientid != XFS_LOG) {
2903 "XFS: xlog_recover_process_data: bad clientid");
2905 return (XFS_ERROR(EIO));
2907 tid = be32_to_cpu(ohead->oh_tid);
2908 hash = XLOG_RHASH(tid);
2909 trans = xlog_recover_find_tid(rhash[hash], tid);
2910 if (trans == NULL) { /* not found; add new tid */
2911 if (ohead->oh_flags & XLOG_START_TRANS)
2912 xlog_recover_new_tid(&rhash[hash], tid,
2913 be64_to_cpu(rhead->h_lsn));
2915 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
2917 "XFS: xlog_recover_process_data: bad length");
2919 return (XFS_ERROR(EIO));
2921 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2922 if (flags & XLOG_WAS_CONT_TRANS)
2923 flags &= ~XLOG_CONTINUE_TRANS;
2925 case XLOG_COMMIT_TRANS:
2926 error = xlog_recover_commit_trans(log,
2927 &rhash[hash], trans, pass);
2929 case XLOG_UNMOUNT_TRANS:
2930 error = xlog_recover_unmount_trans(trans);
2932 case XLOG_WAS_CONT_TRANS:
2933 error = xlog_recover_add_to_cont_trans(trans,
2934 dp, be32_to_cpu(ohead->oh_len));
2936 case XLOG_START_TRANS:
2938 "XFS: xlog_recover_process_data: bad transaction");
2940 error = XFS_ERROR(EIO);
2943 case XLOG_CONTINUE_TRANS:
2944 error = xlog_recover_add_to_trans(trans,
2945 dp, be32_to_cpu(ohead->oh_len));
2949 "XFS: xlog_recover_process_data: bad flag");
2951 error = XFS_ERROR(EIO);
2957 dp += be32_to_cpu(ohead->oh_len);
2964 * Process an extent free intent item that was recovered from
2965 * the log. We need to free the extents that it describes.
2968 xlog_recover_process_efi(
2970 xfs_efi_log_item_t *efip)
2972 xfs_efd_log_item_t *efdp;
2976 xfs_fsblock_t startblock_fsb;
2978 ASSERT(!(efip->efi_flags & XFS_EFI_RECOVERED));
2981 * First check the validity of the extents described by the
2982 * EFI. If any are bad, then assume that all are bad and
2983 * just toss the EFI.
2985 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2986 extp = &(efip->efi_format.efi_extents[i]);
2987 startblock_fsb = XFS_BB_TO_FSB(mp,
2988 XFS_FSB_TO_DADDR(mp, extp->ext_start));
2989 if ((startblock_fsb == 0) ||
2990 (extp->ext_len == 0) ||
2991 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
2992 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
2994 * This will pull the EFI from the AIL and
2995 * free the memory associated with it.
2997 xfs_efi_release(efip, efip->efi_format.efi_nextents);
3002 tp = xfs_trans_alloc(mp, 0);
3003 xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
3004 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3006 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3007 extp = &(efip->efi_format.efi_extents[i]);
3008 xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3009 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3013 efip->efi_flags |= XFS_EFI_RECOVERED;
3014 xfs_trans_commit(tp, 0);
3018 * Verify that once we've encountered something other than an EFI
3019 * in the AIL that there are no more EFIs in the AIL.
3023 xlog_recover_check_ail(
3025 xfs_log_item_t *lip,
3031 ASSERT(lip->li_type != XFS_LI_EFI);
3032 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
3034 * The check will be bogus if we restart from the
3035 * beginning of the AIL, so ASSERT that we don't.
3036 * We never should since we're holding the AIL lock
3039 ASSERT(gen == orig_gen);
3040 } while (lip != NULL);
3045 * When this is called, all of the EFIs which did not have
3046 * corresponding EFDs should be in the AIL. What we do now
3047 * is free the extents associated with each one.
3049 * Since we process the EFIs in normal transactions, they
3050 * will be removed at some point after the commit. This prevents
3051 * us from just walking down the list processing each one.
3052 * We'll use a flag in the EFI to skip those that we've already
3053 * processed and use the AIL iteration mechanism's generation
3054 * count to try to speed this up at least a bit.
3056 * When we start, we know that the EFIs are the only things in
3057 * the AIL. As we process them, however, other items are added
3058 * to the AIL. Since everything added to the AIL must come after
3059 * everything already in the AIL, we stop processing as soon as
3060 * we see something other than an EFI in the AIL.
3063 xlog_recover_process_efis(
3066 xfs_log_item_t *lip;
3067 xfs_efi_log_item_t *efip;
3072 spin_lock(&mp->m_ail_lock);
3074 lip = xfs_trans_first_ail(mp, &gen);
3075 while (lip != NULL) {
3077 * We're done when we see something other than an EFI.
3079 if (lip->li_type != XFS_LI_EFI) {
3080 xlog_recover_check_ail(mp, lip, gen);
3085 * Skip EFIs that we've already processed.
3087 efip = (xfs_efi_log_item_t *)lip;
3088 if (efip->efi_flags & XFS_EFI_RECOVERED) {
3089 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
3093 spin_unlock(&mp->m_ail_lock);
3094 xlog_recover_process_efi(mp, efip);
3095 spin_lock(&mp->m_ail_lock);
3096 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
3098 spin_unlock(&mp->m_ail_lock);
3102 * This routine performs a transaction to null out a bad inode pointer
3103 * in an agi unlinked inode hash bucket.
3106 xlog_recover_clear_agi_bucket(
3108 xfs_agnumber_t agno,
3117 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3118 xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp), 0, 0, 0);
3120 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
3121 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
3122 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
3124 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3128 agi = XFS_BUF_TO_AGI(agibp);
3129 if (be32_to_cpu(agi->agi_magicnum) != XFS_AGI_MAGIC) {
3130 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3134 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
3135 offset = offsetof(xfs_agi_t, agi_unlinked) +
3136 (sizeof(xfs_agino_t) * bucket);
3137 xfs_trans_log_buf(tp, agibp, offset,
3138 (offset + sizeof(xfs_agino_t) - 1));
3140 (void) xfs_trans_commit(tp, 0);
3144 * xlog_iunlink_recover
3146 * This is called during recovery to process any inodes which
3147 * we unlinked but not freed when the system crashed. These
3148 * inodes will be on the lists in the AGI blocks. What we do
3149 * here is scan all the AGIs and fully truncate and free any
3150 * inodes found on the lists. Each inode is removed from the
3151 * lists when it has been fully truncated and is freed. The
3152 * freeing of the inode and its removal from the list must be
3156 xlog_recover_process_iunlinks(
3160 xfs_agnumber_t agno;
3175 * Prevent any DMAPI event from being sent while in this function.
3177 mp_dmevmask = mp->m_dmevmask;
3180 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3182 * Find the agi for this ag.
3184 agibp = xfs_buf_read(mp->m_ddev_targp,
3185 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
3186 XFS_FSS_TO_BB(mp, 1), 0);
3187 if (XFS_BUF_ISERROR(agibp)) {
3188 xfs_ioerror_alert("xlog_recover_process_iunlinks(#1)",
3190 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)));
3192 agi = XFS_BUF_TO_AGI(agibp);
3193 ASSERT(XFS_AGI_MAGIC == be32_to_cpu(agi->agi_magicnum));
3195 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3197 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3198 while (agino != NULLAGINO) {
3201 * Release the agi buffer so that it can
3202 * be acquired in the normal course of the
3203 * transaction to truncate and free the inode.
3205 xfs_buf_relse(agibp);
3207 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3208 error = xfs_iget(mp, NULL, ino, 0, 0, &ip, 0);
3209 ASSERT(error || (ip != NULL));
3213 * Get the on disk inode to find the
3214 * next inode in the bucket.
3216 error = xfs_itobp(mp, NULL, ip, &dip,
3218 ASSERT(error || (dip != NULL));
3222 ASSERT(ip->i_d.di_nlink == 0);
3224 /* setup for the next pass */
3225 agino = be32_to_cpu(
3226 dip->di_next_unlinked);
3229 * Prevent any DMAPI event from
3230 * being sent when the
3231 * reference on the inode is
3234 ip->i_d.di_dmevmask = 0;
3237 * If this is a new inode, handle
3238 * it specially. Otherwise,
3239 * just drop our reference to the
3240 * inode. If there are no
3241 * other references, this will
3243 * xfs_inactive() which will
3244 * truncate the file and free
3247 if (ip->i_d.di_mode == 0)
3248 xfs_iput_new(ip, 0);
3250 VN_RELE(XFS_ITOV(ip));
3253 * We can't read in the inode
3254 * this bucket points to, or
3255 * this inode is messed up. Just
3256 * ditch this bucket of inodes. We
3257 * will lose some inodes and space,
3258 * but at least we won't hang. Call
3259 * xlog_recover_clear_agi_bucket()
3260 * to perform a transaction to clear
3261 * the inode pointer in the bucket.
3263 xlog_recover_clear_agi_bucket(mp, agno,
3270 * Reacquire the agibuffer and continue around
3273 agibp = xfs_buf_read(mp->m_ddev_targp,
3274 XFS_AG_DADDR(mp, agno,
3276 XFS_FSS_TO_BB(mp, 1), 0);
3277 if (XFS_BUF_ISERROR(agibp)) {
3279 "xlog_recover_process_iunlinks(#2)",
3281 XFS_AG_DADDR(mp, agno,
3282 XFS_AGI_DADDR(mp)));
3284 agi = XFS_BUF_TO_AGI(agibp);
3285 ASSERT(XFS_AGI_MAGIC == be32_to_cpu(
3286 agi->agi_magicnum));
3291 * Release the buffer for the current agi so we can
3292 * go on to the next one.
3294 xfs_buf_relse(agibp);
3297 mp->m_dmevmask = mp_dmevmask;
3303 xlog_pack_data_checksum(
3305 xlog_in_core_t *iclog,
3312 up = (__be32 *)iclog->ic_datap;
3313 /* divide length by 4 to get # words */
3314 for (i = 0; i < (size >> 2); i++) {
3315 chksum ^= be32_to_cpu(*up);
3318 iclog->ic_header.h_chksum = cpu_to_be32(chksum);
3321 #define xlog_pack_data_checksum(log, iclog, size)
3325 * Stamp cycle number in every block
3330 xlog_in_core_t *iclog,
3334 int size = iclog->ic_offset + roundoff;
3337 xlog_in_core_2_t *xhdr;
3339 xlog_pack_data_checksum(log, iclog, size);
3341 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3343 dp = iclog->ic_datap;
3344 for (i = 0; i < BTOBB(size) &&
3345 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3346 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
3347 *(__be32 *)dp = cycle_lsn;
3351 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3352 xhdr = (xlog_in_core_2_t *)&iclog->ic_header;
3353 for ( ; i < BTOBB(size); i++) {
3354 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3355 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3356 xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
3357 *(__be32 *)dp = cycle_lsn;
3361 for (i = 1; i < log->l_iclog_heads; i++) {
3362 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3367 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
3369 xlog_unpack_data_checksum(
3370 xlog_rec_header_t *rhead,
3374 __be32 *up = (__be32 *)dp;
3378 /* divide length by 4 to get # words */
3379 for (i=0; i < be32_to_cpu(rhead->h_len) >> 2; i++) {
3380 chksum ^= be32_to_cpu(*up);
3383 if (chksum != be32_to_cpu(rhead->h_chksum)) {
3384 if (rhead->h_chksum ||
3385 ((log->l_flags & XLOG_CHKSUM_MISMATCH) == 0)) {
3387 "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n",
3388 be32_to_cpu(rhead->h_chksum), chksum);
3390 "XFS: Disregard message if filesystem was created with non-DEBUG kernel");
3391 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3393 "XFS: LogR this is a LogV2 filesystem\n");
3395 log->l_flags |= XLOG_CHKSUM_MISMATCH;
3400 #define xlog_unpack_data_checksum(rhead, dp, log)
3405 xlog_rec_header_t *rhead,
3410 xlog_in_core_2_t *xhdr;
3412 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
3413 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3414 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
3418 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3419 xhdr = (xlog_in_core_2_t *)rhead;
3420 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
3421 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3422 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3423 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
3428 xlog_unpack_data_checksum(rhead, dp, log);
3432 xlog_valid_rec_header(
3434 xlog_rec_header_t *rhead,
3439 if (unlikely(be32_to_cpu(rhead->h_magicno) != XLOG_HEADER_MAGIC_NUM)) {
3440 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3441 XFS_ERRLEVEL_LOW, log->l_mp);
3442 return XFS_ERROR(EFSCORRUPTED);
3445 (!rhead->h_version ||
3446 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
3447 xlog_warn("XFS: %s: unrecognised log version (%d).",
3448 __FUNCTION__, be32_to_cpu(rhead->h_version));
3449 return XFS_ERROR(EIO);
3452 /* LR body must have data or it wouldn't have been written */
3453 hlen = be32_to_cpu(rhead->h_len);
3454 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3455 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3456 XFS_ERRLEVEL_LOW, log->l_mp);
3457 return XFS_ERROR(EFSCORRUPTED);
3459 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3460 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3461 XFS_ERRLEVEL_LOW, log->l_mp);
3462 return XFS_ERROR(EFSCORRUPTED);
3468 * Read the log from tail to head and process the log records found.
3469 * Handle the two cases where the tail and head are in the same cycle
3470 * and where the active portion of the log wraps around the end of
3471 * the physical log separately. The pass parameter is passed through
3472 * to the routines called to process the data and is not looked at
3476 xlog_do_recovery_pass(
3478 xfs_daddr_t head_blk,
3479 xfs_daddr_t tail_blk,
3482 xlog_rec_header_t *rhead;
3484 xfs_caddr_t bufaddr, offset;
3485 xfs_buf_t *hbp, *dbp;
3486 int error = 0, h_size;
3487 int bblks, split_bblks;
3488 int hblks, split_hblks, wrapped_hblks;
3489 xlog_recover_t *rhash[XLOG_RHASH_SIZE];
3491 ASSERT(head_blk != tail_blk);
3494 * Read the header of the tail block and get the iclog buffer size from
3495 * h_size. Use this to tell how many sectors make up the log header.
3497 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3499 * When using variable length iclogs, read first sector of
3500 * iclog header and extract the header size from it. Get a
3501 * new hbp that is the correct size.
3503 hbp = xlog_get_bp(log, 1);
3506 if ((error = xlog_bread(log, tail_blk, 1, hbp)))
3508 offset = xlog_align(log, tail_blk, 1, hbp);
3509 rhead = (xlog_rec_header_t *)offset;
3510 error = xlog_valid_rec_header(log, rhead, tail_blk);
3513 h_size = be32_to_cpu(rhead->h_size);
3514 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
3515 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3516 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3517 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3520 hbp = xlog_get_bp(log, hblks);
3525 ASSERT(log->l_sectbb_log == 0);
3527 hbp = xlog_get_bp(log, 1);
3528 h_size = XLOG_BIG_RECORD_BSIZE;
3533 dbp = xlog_get_bp(log, BTOBB(h_size));
3539 memset(rhash, 0, sizeof(rhash));
3540 if (tail_blk <= head_blk) {
3541 for (blk_no = tail_blk; blk_no < head_blk; ) {
3542 if ((error = xlog_bread(log, blk_no, hblks, hbp)))
3544 offset = xlog_align(log, blk_no, hblks, hbp);
3545 rhead = (xlog_rec_header_t *)offset;
3546 error = xlog_valid_rec_header(log, rhead, blk_no);
3550 /* blocks in data section */
3551 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3552 error = xlog_bread(log, blk_no + hblks, bblks, dbp);
3555 offset = xlog_align(log, blk_no + hblks, bblks, dbp);
3556 xlog_unpack_data(rhead, offset, log);
3557 if ((error = xlog_recover_process_data(log,
3558 rhash, rhead, offset, pass)))
3560 blk_no += bblks + hblks;
3564 * Perform recovery around the end of the physical log.
3565 * When the head is not on the same cycle number as the tail,
3566 * we can't do a sequential recovery as above.
3569 while (blk_no < log->l_logBBsize) {
3571 * Check for header wrapping around physical end-of-log
3576 if (blk_no + hblks <= log->l_logBBsize) {
3577 /* Read header in one read */
3578 error = xlog_bread(log, blk_no, hblks, hbp);
3581 offset = xlog_align(log, blk_no, hblks, hbp);
3583 /* This LR is split across physical log end */
3584 if (blk_no != log->l_logBBsize) {
3585 /* some data before physical log end */
3586 ASSERT(blk_no <= INT_MAX);
3587 split_hblks = log->l_logBBsize - (int)blk_no;
3588 ASSERT(split_hblks > 0);
3589 if ((error = xlog_bread(log, blk_no,
3592 offset = xlog_align(log, blk_no,
3596 * Note: this black magic still works with
3597 * large sector sizes (non-512) only because:
3598 * - we increased the buffer size originally
3599 * by 1 sector giving us enough extra space
3600 * for the second read;
3601 * - the log start is guaranteed to be sector
3603 * - we read the log end (LR header start)
3604 * _first_, then the log start (LR header end)
3605 * - order is important.
3607 bufaddr = XFS_BUF_PTR(hbp);
3608 XFS_BUF_SET_PTR(hbp,
3609 bufaddr + BBTOB(split_hblks),
3610 BBTOB(hblks - split_hblks));
3611 wrapped_hblks = hblks - split_hblks;
3612 error = xlog_bread(log, 0, wrapped_hblks, hbp);
3615 XFS_BUF_SET_PTR(hbp, bufaddr, BBTOB(hblks));
3617 offset = xlog_align(log, 0,
3618 wrapped_hblks, hbp);
3620 rhead = (xlog_rec_header_t *)offset;
3621 error = xlog_valid_rec_header(log, rhead,
3622 split_hblks ? blk_no : 0);
3626 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3629 /* Read in data for log record */
3630 if (blk_no + bblks <= log->l_logBBsize) {
3631 error = xlog_bread(log, blk_no, bblks, dbp);
3634 offset = xlog_align(log, blk_no, bblks, dbp);
3636 /* This log record is split across the
3637 * physical end of log */
3640 if (blk_no != log->l_logBBsize) {
3641 /* some data is before the physical
3643 ASSERT(!wrapped_hblks);
3644 ASSERT(blk_no <= INT_MAX);
3646 log->l_logBBsize - (int)blk_no;
3647 ASSERT(split_bblks > 0);
3648 if ((error = xlog_bread(log, blk_no,
3651 offset = xlog_align(log, blk_no,
3655 * Note: this black magic still works with
3656 * large sector sizes (non-512) only because:
3657 * - we increased the buffer size originally
3658 * by 1 sector giving us enough extra space
3659 * for the second read;
3660 * - the log start is guaranteed to be sector
3662 * - we read the log end (LR header start)
3663 * _first_, then the log start (LR header end)
3664 * - order is important.
3666 bufaddr = XFS_BUF_PTR(dbp);
3667 XFS_BUF_SET_PTR(dbp,
3668 bufaddr + BBTOB(split_bblks),
3669 BBTOB(bblks - split_bblks));
3670 if ((error = xlog_bread(log, wrapped_hblks,
3671 bblks - split_bblks, dbp)))
3673 XFS_BUF_SET_PTR(dbp, bufaddr, h_size);
3675 offset = xlog_align(log, wrapped_hblks,
3676 bblks - split_bblks, dbp);
3678 xlog_unpack_data(rhead, offset, log);
3679 if ((error = xlog_recover_process_data(log, rhash,
3680 rhead, offset, pass)))
3685 ASSERT(blk_no >= log->l_logBBsize);
3686 blk_no -= log->l_logBBsize;
3688 /* read first part of physical log */
3689 while (blk_no < head_blk) {
3690 if ((error = xlog_bread(log, blk_no, hblks, hbp)))
3692 offset = xlog_align(log, blk_no, hblks, hbp);
3693 rhead = (xlog_rec_header_t *)offset;
3694 error = xlog_valid_rec_header(log, rhead, blk_no);
3697 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3698 if ((error = xlog_bread(log, blk_no+hblks, bblks, dbp)))
3700 offset = xlog_align(log, blk_no+hblks, bblks, dbp);
3701 xlog_unpack_data(rhead, offset, log);
3702 if ((error = xlog_recover_process_data(log, rhash,
3703 rhead, offset, pass)))
3705 blk_no += bblks + hblks;
3717 * Do the recovery of the log. We actually do this in two phases.
3718 * The two passes are necessary in order to implement the function
3719 * of cancelling a record written into the log. The first pass
3720 * determines those things which have been cancelled, and the
3721 * second pass replays log items normally except for those which
3722 * have been cancelled. The handling of the replay and cancellations
3723 * takes place in the log item type specific routines.
3725 * The table of items which have cancel records in the log is allocated
3726 * and freed at this level, since only here do we know when all of
3727 * the log recovery has been completed.
3730 xlog_do_log_recovery(
3732 xfs_daddr_t head_blk,
3733 xfs_daddr_t tail_blk)
3737 ASSERT(head_blk != tail_blk);
3740 * First do a pass to find all of the cancelled buf log items.
3741 * Store them in the buf_cancel_table for use in the second pass.
3743 log->l_buf_cancel_table =
3744 (xfs_buf_cancel_t **)kmem_zalloc(XLOG_BC_TABLE_SIZE *
3745 sizeof(xfs_buf_cancel_t*),
3747 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3748 XLOG_RECOVER_PASS1);
3750 kmem_free(log->l_buf_cancel_table,
3751 XLOG_BC_TABLE_SIZE * sizeof(xfs_buf_cancel_t*));
3752 log->l_buf_cancel_table = NULL;
3756 * Then do a second pass to actually recover the items in the log.
3757 * When it is complete free the table of buf cancel items.
3759 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3760 XLOG_RECOVER_PASS2);
3765 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3766 ASSERT(log->l_buf_cancel_table[i] == NULL);
3770 kmem_free(log->l_buf_cancel_table,
3771 XLOG_BC_TABLE_SIZE * sizeof(xfs_buf_cancel_t*));
3772 log->l_buf_cancel_table = NULL;
3778 * Do the actual recovery
3783 xfs_daddr_t head_blk,
3784 xfs_daddr_t tail_blk)
3791 * First replay the images in the log.
3793 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3798 XFS_bflush(log->l_mp->m_ddev_targp);
3801 * If IO errors happened during recovery, bail out.
3803 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3808 * We now update the tail_lsn since much of the recovery has completed
3809 * and there may be space available to use. If there were no extent
3810 * or iunlinks, we can free up the entire log and set the tail_lsn to
3811 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3812 * lsn of the last known good LR on disk. If there are extent frees
3813 * or iunlinks they will have some entries in the AIL; so we look at
3814 * the AIL to determine how to set the tail_lsn.
3816 xlog_assign_tail_lsn(log->l_mp);
3819 * Now that we've finished replaying all buffer and inode
3820 * updates, re-read in the superblock.
3822 bp = xfs_getsb(log->l_mp, 0);
3824 ASSERT(!(XFS_BUF_ISWRITE(bp)));
3825 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
3827 XFS_BUF_UNASYNC(bp);
3828 xfsbdstrat(log->l_mp, bp);
3829 if ((error = xfs_iowait(bp))) {
3830 xfs_ioerror_alert("xlog_do_recover",
3831 log->l_mp, bp, XFS_BUF_ADDR(bp));
3837 /* Convert superblock from on-disk format */
3838 sbp = &log->l_mp->m_sb;
3839 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
3840 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
3841 ASSERT(XFS_SB_GOOD_VERSION(sbp));
3844 /* We've re-read the superblock so re-initialize per-cpu counters */
3845 xfs_icsb_reinit_counters(log->l_mp);
3847 xlog_recover_check_summary(log);
3849 /* Normal transactions can now occur */
3850 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3855 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3857 * Return error or zero.
3863 xfs_daddr_t head_blk, tail_blk;
3866 /* find the tail of the log */
3867 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
3870 if (tail_blk != head_blk) {
3871 /* There used to be a comment here:
3873 * disallow recovery on read-only mounts. note -- mount
3874 * checks for ENOSPC and turns it into an intelligent
3876 * ...but this is no longer true. Now, unless you specify
3877 * NORECOVERY (in which case this function would never be
3878 * called), we just go ahead and recover. We do this all
3879 * under the vfs layer, so we can get away with it unless
3880 * the device itself is read-only, in which case we fail.
3882 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
3887 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3888 log->l_mp->m_fsname, log->l_mp->m_logname ?
3889 log->l_mp->m_logname : "internal");
3891 error = xlog_do_recover(log, head_blk, tail_blk);
3892 log->l_flags |= XLOG_RECOVERY_NEEDED;
3898 * In the first part of recovery we replay inodes and buffers and build
3899 * up the list of extent free items which need to be processed. Here
3900 * we process the extent free items and clean up the on disk unlinked
3901 * inode lists. This is separated from the first part of recovery so
3902 * that the root and real-time bitmap inodes can be read in from disk in
3903 * between the two stages. This is necessary so that we can free space
3904 * in the real-time portion of the file system.
3907 xlog_recover_finish(
3912 * Now we're ready to do the transactions needed for the
3913 * rest of recovery. Start with completing all the extent
3914 * free intent records and then process the unlinked inode
3915 * lists. At this point, we essentially run in normal mode
3916 * except that we're still performing recovery actions
3917 * rather than accepting new requests.
3919 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
3920 xlog_recover_process_efis(log);
3922 * Sync the log to get all the EFIs out of the AIL.
3923 * This isn't absolutely necessary, but it helps in
3924 * case the unlink transactions would have problems
3925 * pushing the EFIs out of the way.
3927 xfs_log_force(log->l_mp, (xfs_lsn_t)0,
3928 (XFS_LOG_FORCE | XFS_LOG_SYNC));
3930 if ( (mfsi_flags & XFS_MFSI_NOUNLINK) == 0 ) {
3931 xlog_recover_process_iunlinks(log);
3934 xlog_recover_check_summary(log);
3937 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3938 log->l_mp->m_fsname, log->l_mp->m_logname ?
3939 log->l_mp->m_logname : "internal");
3940 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
3943 "!Ending clean XFS mount for filesystem: %s\n",
3944 log->l_mp->m_fsname);
3952 * Read all of the agf and agi counters and check that they
3953 * are consistent with the superblock counters.
3956 xlog_recover_check_summary(
3964 xfs_daddr_t agfdaddr;
3965 xfs_daddr_t agidaddr;
3967 #ifdef XFS_LOUD_RECOVERY
3970 xfs_agnumber_t agno;
3971 __uint64_t freeblks;
3980 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3981 agfdaddr = XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp));
3982 agfbp = xfs_buf_read(mp->m_ddev_targp, agfdaddr,
3983 XFS_FSS_TO_BB(mp, 1), 0);
3984 if (XFS_BUF_ISERROR(agfbp)) {
3985 xfs_ioerror_alert("xlog_recover_check_summary(agf)",
3986 mp, agfbp, agfdaddr);
3988 agfp = XFS_BUF_TO_AGF(agfbp);
3989 ASSERT(XFS_AGF_MAGIC == be32_to_cpu(agfp->agf_magicnum));
3990 ASSERT(XFS_AGF_GOOD_VERSION(be32_to_cpu(agfp->agf_versionnum)));
3991 ASSERT(be32_to_cpu(agfp->agf_seqno) == agno);
3993 freeblks += be32_to_cpu(agfp->agf_freeblks) +
3994 be32_to_cpu(agfp->agf_flcount);
3995 xfs_buf_relse(agfbp);
3997 agidaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
3998 agibp = xfs_buf_read(mp->m_ddev_targp, agidaddr,
3999 XFS_FSS_TO_BB(mp, 1), 0);
4000 if (XFS_BUF_ISERROR(agibp)) {
4001 xfs_ioerror_alert("xlog_recover_check_summary(agi)",
4002 mp, agibp, agidaddr);
4004 agip = XFS_BUF_TO_AGI(agibp);
4005 ASSERT(XFS_AGI_MAGIC == be32_to_cpu(agip->agi_magicnum));
4006 ASSERT(XFS_AGI_GOOD_VERSION(be32_to_cpu(agip->agi_versionnum)));
4007 ASSERT(be32_to_cpu(agip->agi_seqno) == agno);
4009 itotal += be32_to_cpu(agip->agi_count);
4010 ifree += be32_to_cpu(agip->agi_freecount);
4011 xfs_buf_relse(agibp);
4014 sbbp = xfs_getsb(mp, 0);
4015 #ifdef XFS_LOUD_RECOVERY
4017 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(sbbp));
4019 "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
4020 sbp->sb_icount, itotal);
4022 "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
4023 sbp->sb_ifree, ifree);
4025 "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
4026 sbp->sb_fdblocks, freeblks);
4029 * This is turned off until I account for the allocation
4030 * btree blocks which live in free space.
4032 ASSERT(sbp->sb_icount == itotal);
4033 ASSERT(sbp->sb_ifree == ifree);
4034 ASSERT(sbp->sb_fdblocks == freeblks);
4037 xfs_buf_relse(sbbp);
projects / linux-2.6 / blob