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", INT_GET(head->h_fmt, ARCH_CONVERT));
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(INT_GET(head->h_magicno, ARCH_CONVERT) == 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(INT_GET(head->h_fmt, ARCH_CONVERT) != 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(INT_GET(head->h_magicno, ARCH_CONVERT) == 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 ==
451 INT_GET(head->h_magicno, ARCH_CONVERT))
459 * We hit the beginning of the physical log & still no header. Return
460 * to caller. If caller can handle a return of -1, then this routine
461 * will be called again for the end of the physical log.
469 * We have the final block of the good log (the first block
470 * of the log record _before_ the head. So we check the uuid.
472 if ((error = xlog_header_check_mount(log->l_mp, head)))
476 * We may have found a log record header before we expected one.
477 * last_blk will be the 1st block # with a given cycle #. We may end
478 * up reading an entire log record. In this case, we don't want to
479 * reset last_blk. Only when last_blk points in the middle of a log
480 * record do we update last_blk.
482 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
483 uint h_size = INT_GET(head->h_size, ARCH_CONVERT);
485 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
486 if (h_size % XLOG_HEADER_CYCLE_SIZE)
492 if (*last_blk - i + extra_bblks
493 != BTOBB(INT_GET(head->h_len, ARCH_CONVERT)) + xhdrs)
502 * Head is defined to be the point of the log where the next log write
503 * write could go. This means that incomplete LR writes at the end are
504 * eliminated when calculating the head. We aren't guaranteed that previous
505 * LR have complete transactions. We only know that a cycle number of
506 * current cycle number -1 won't be present in the log if we start writing
507 * from our current block number.
509 * last_blk contains the block number of the first block with a given
512 * Return: zero if normal, non-zero if error.
517 xfs_daddr_t *return_head_blk)
521 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
523 uint first_half_cycle, last_half_cycle;
525 int error, log_bbnum = log->l_logBBsize;
527 /* Is the end of the log device zeroed? */
528 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
529 *return_head_blk = first_blk;
531 /* Is the whole lot zeroed? */
533 /* Linux XFS shouldn't generate totally zeroed logs -
534 * mkfs etc write a dummy unmount record to a fresh
535 * log so we can store the uuid in there
537 xlog_warn("XFS: totally zeroed log");
542 xlog_warn("XFS: empty log check failed");
546 first_blk = 0; /* get cycle # of 1st block */
547 bp = xlog_get_bp(log, 1);
550 if ((error = xlog_bread(log, 0, 1, bp)))
552 offset = xlog_align(log, 0, 1, bp);
553 first_half_cycle = xlog_get_cycle(offset);
555 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
556 if ((error = xlog_bread(log, last_blk, 1, bp)))
558 offset = xlog_align(log, last_blk, 1, bp);
559 last_half_cycle = xlog_get_cycle(offset);
560 ASSERT(last_half_cycle != 0);
563 * If the 1st half cycle number is equal to the last half cycle number,
564 * then the entire log is stamped with the same cycle number. In this
565 * case, head_blk can't be set to zero (which makes sense). The below
566 * math doesn't work out properly with head_blk equal to zero. Instead,
567 * we set it to log_bbnum which is an invalid block number, but this
568 * value makes the math correct. If head_blk doesn't changed through
569 * all the tests below, *head_blk is set to zero at the very end rather
570 * than log_bbnum. In a sense, log_bbnum and zero are the same block
571 * in a circular file.
573 if (first_half_cycle == last_half_cycle) {
575 * In this case we believe that the entire log should have
576 * cycle number last_half_cycle. We need to scan backwards
577 * from the end verifying that there are no holes still
578 * containing last_half_cycle - 1. If we find such a hole,
579 * then the start of that hole will be the new head. The
580 * simple case looks like
581 * x | x ... | x - 1 | x
582 * Another case that fits this picture would be
583 * x | x + 1 | x ... | x
584 * In this case the head really is somewhere at the end of the
585 * log, as one of the latest writes at the beginning was
588 * x | x + 1 | x ... | x - 1 | x
589 * This is really the combination of the above two cases, and
590 * the head has to end up at the start of the x-1 hole at the
593 * In the 256k log case, we will read from the beginning to the
594 * end of the log and search for cycle numbers equal to x-1.
595 * We don't worry about the x+1 blocks that we encounter,
596 * because we know that they cannot be the head since the log
599 head_blk = log_bbnum;
600 stop_on_cycle = last_half_cycle - 1;
603 * In this case we want to find the first block with cycle
604 * number matching last_half_cycle. We expect the log to be
607 * The first block with cycle number x (last_half_cycle) will
608 * be where the new head belongs. First we do a binary search
609 * for the first occurrence of last_half_cycle. The binary
610 * search may not be totally accurate, so then we scan back
611 * from there looking for occurrences of last_half_cycle before
612 * us. If that backwards scan wraps around the beginning of
613 * the log, then we look for occurrences of last_half_cycle - 1
614 * at the end of the log. The cases we're looking for look
616 * x + 1 ... | x | x + 1 | x ...
617 * ^ binary search stopped here
619 * x + 1 ... | x ... | x - 1 | x
620 * <---------> less than scan distance
622 stop_on_cycle = last_half_cycle;
623 if ((error = xlog_find_cycle_start(log, bp, first_blk,
624 &head_blk, last_half_cycle)))
629 * Now validate the answer. Scan back some number of maximum possible
630 * blocks and make sure each one has the expected cycle number. The
631 * maximum is determined by the total possible amount of buffering
632 * in the in-core log. The following number can be made tighter if
633 * we actually look at the block size of the filesystem.
635 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
636 if (head_blk >= num_scan_bblks) {
638 * We are guaranteed that the entire check can be performed
641 start_blk = head_blk - num_scan_bblks;
642 if ((error = xlog_find_verify_cycle(log,
643 start_blk, num_scan_bblks,
644 stop_on_cycle, &new_blk)))
648 } else { /* need to read 2 parts of log */
650 * We are going to scan backwards in the log in two parts.
651 * First we scan the physical end of the log. In this part
652 * of the log, we are looking for blocks with cycle number
653 * last_half_cycle - 1.
654 * If we find one, then we know that the log starts there, as
655 * we've found a hole that didn't get written in going around
656 * the end of the physical log. The simple case for this is
657 * x + 1 ... | x ... | x - 1 | x
658 * <---------> less than scan distance
659 * If all of the blocks at the end of the log have cycle number
660 * last_half_cycle, then we check the blocks at the start of
661 * the log looking for occurrences of last_half_cycle. If we
662 * find one, then our current estimate for the location of the
663 * first occurrence of last_half_cycle is wrong and we move
664 * back to the hole we've found. This case looks like
665 * x + 1 ... | x | x + 1 | x ...
666 * ^ binary search stopped here
667 * Another case we need to handle that only occurs in 256k
669 * x + 1 ... | x ... | x+1 | x ...
670 * ^ binary search stops here
671 * In a 256k log, the scan at the end of the log will see the
672 * x + 1 blocks. We need to skip past those since that is
673 * certainly not the head of the log. By searching for
674 * last_half_cycle-1 we accomplish that.
676 start_blk = log_bbnum - num_scan_bblks + head_blk;
677 ASSERT(head_blk <= INT_MAX &&
678 (xfs_daddr_t) num_scan_bblks - head_blk >= 0);
679 if ((error = xlog_find_verify_cycle(log, start_blk,
680 num_scan_bblks - (int)head_blk,
681 (stop_on_cycle - 1), &new_blk)))
689 * Scan beginning of log now. The last part of the physical
690 * log is good. This scan needs to verify that it doesn't find
691 * the last_half_cycle.
694 ASSERT(head_blk <= INT_MAX);
695 if ((error = xlog_find_verify_cycle(log,
696 start_blk, (int)head_blk,
697 stop_on_cycle, &new_blk)))
705 * Now we need to make sure head_blk is not pointing to a block in
706 * the middle of a log record.
708 num_scan_bblks = XLOG_REC_SHIFT(log);
709 if (head_blk >= num_scan_bblks) {
710 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
712 /* start ptr at last block ptr before head_blk */
713 if ((error = xlog_find_verify_log_record(log, start_blk,
714 &head_blk, 0)) == -1) {
715 error = XFS_ERROR(EIO);
721 ASSERT(head_blk <= INT_MAX);
722 if ((error = xlog_find_verify_log_record(log, start_blk,
723 &head_blk, 0)) == -1) {
724 /* We hit the beginning of the log during our search */
725 start_blk = log_bbnum - num_scan_bblks + head_blk;
727 ASSERT(start_blk <= INT_MAX &&
728 (xfs_daddr_t) log_bbnum-start_blk >= 0);
729 ASSERT(head_blk <= INT_MAX);
730 if ((error = xlog_find_verify_log_record(log,
732 (int)head_blk)) == -1) {
733 error = XFS_ERROR(EIO);
737 if (new_blk != log_bbnum)
744 if (head_blk == log_bbnum)
745 *return_head_blk = 0;
747 *return_head_blk = head_blk;
749 * When returning here, we have a good block number. Bad block
750 * means that during a previous crash, we didn't have a clean break
751 * from cycle number N to cycle number N-1. In this case, we need
752 * to find the first block with cycle number N-1.
760 xlog_warn("XFS: failed to find log head");
765 * Find the sync block number or the tail of the log.
767 * This will be the block number of the last record to have its
768 * associated buffers synced to disk. Every log record header has
769 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
770 * to get a sync block number. The only concern is to figure out which
771 * log record header to believe.
773 * The following algorithm uses the log record header with the largest
774 * lsn. The entire log record does not need to be valid. We only care
775 * that the header is valid.
777 * We could speed up search by using current head_blk buffer, but it is not
783 xfs_daddr_t *head_blk,
784 xfs_daddr_t *tail_blk)
786 xlog_rec_header_t *rhead;
787 xlog_op_header_t *op_head;
788 xfs_caddr_t offset = NULL;
791 xfs_daddr_t umount_data_blk;
792 xfs_daddr_t after_umount_blk;
799 * Find previous log record
801 if ((error = xlog_find_head(log, head_blk)))
804 bp = xlog_get_bp(log, 1);
807 if (*head_blk == 0) { /* special case */
808 if ((error = xlog_bread(log, 0, 1, bp)))
810 offset = xlog_align(log, 0, 1, bp);
811 if (xlog_get_cycle(offset) == 0) {
813 /* leave all other log inited values alone */
819 * Search backwards looking for log record header block
821 ASSERT(*head_blk < INT_MAX);
822 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
823 if ((error = xlog_bread(log, i, 1, bp)))
825 offset = xlog_align(log, i, 1, bp);
826 if (XLOG_HEADER_MAGIC_NUM ==
827 INT_GET(*(uint *)offset, ARCH_CONVERT)) {
833 * If we haven't found the log record header block, start looking
834 * again from the end of the physical log. XXXmiken: There should be
835 * a check here to make sure we didn't search more than N blocks in
839 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
840 if ((error = xlog_bread(log, i, 1, bp)))
842 offset = xlog_align(log, i, 1, bp);
843 if (XLOG_HEADER_MAGIC_NUM ==
844 INT_GET(*(uint*)offset, ARCH_CONVERT)) {
851 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
853 return XFS_ERROR(EIO);
856 /* find blk_no of tail of log */
857 rhead = (xlog_rec_header_t *)offset;
858 *tail_blk = BLOCK_LSN(INT_GET(rhead->h_tail_lsn, ARCH_CONVERT));
861 * Reset log values according to the state of the log when we
862 * crashed. In the case where head_blk == 0, we bump curr_cycle
863 * one because the next write starts a new cycle rather than
864 * continuing the cycle of the last good log record. At this
865 * point we have guaranteed that all partial log records have been
866 * accounted for. Therefore, we know that the last good log record
867 * written was complete and ended exactly on the end boundary
868 * of the physical log.
870 log->l_prev_block = i;
871 log->l_curr_block = (int)*head_blk;
872 log->l_curr_cycle = INT_GET(rhead->h_cycle, ARCH_CONVERT);
875 log->l_tail_lsn = INT_GET(rhead->h_tail_lsn, ARCH_CONVERT);
876 log->l_last_sync_lsn = INT_GET(rhead->h_lsn, ARCH_CONVERT);
877 log->l_grant_reserve_cycle = log->l_curr_cycle;
878 log->l_grant_reserve_bytes = BBTOB(log->l_curr_block);
879 log->l_grant_write_cycle = log->l_curr_cycle;
880 log->l_grant_write_bytes = BBTOB(log->l_curr_block);
883 * Look for unmount record. If we find it, then we know there
884 * was a clean unmount. Since 'i' could be the last block in
885 * the physical log, we convert to a log block before comparing
888 * Save the current tail lsn to use to pass to
889 * xlog_clear_stale_blocks() below. We won't want to clear the
890 * unmount record if there is one, so we pass the lsn of the
891 * unmount record rather than the block after it.
893 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
894 int h_size = INT_GET(rhead->h_size, ARCH_CONVERT);
895 int h_version = INT_GET(rhead->h_version, ARCH_CONVERT);
897 if ((h_version & XLOG_VERSION_2) &&
898 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
899 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
900 if (h_size % XLOG_HEADER_CYCLE_SIZE)
908 after_umount_blk = (i + hblks + (int)
909 BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT))) % log->l_logBBsize;
910 tail_lsn = log->l_tail_lsn;
911 if (*head_blk == after_umount_blk &&
912 INT_GET(rhead->h_num_logops, ARCH_CONVERT) == 1) {
913 umount_data_blk = (i + hblks) % log->l_logBBsize;
914 if ((error = xlog_bread(log, umount_data_blk, 1, bp))) {
917 offset = xlog_align(log, umount_data_blk, 1, bp);
918 op_head = (xlog_op_header_t *)offset;
919 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
921 * Set tail and last sync so that newly written
922 * log records will point recovery to after the
923 * current unmount record.
926 xlog_assign_lsn(log->l_curr_cycle,
928 log->l_last_sync_lsn =
929 xlog_assign_lsn(log->l_curr_cycle,
931 *tail_blk = after_umount_blk;
934 * Note that the unmount was clean. If the unmount
935 * was not clean, we need to know this to rebuild the
936 * superblock counters from the perag headers if we
937 * have a filesystem using non-persistent counters.
939 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
944 * Make sure that there are no blocks in front of the head
945 * with the same cycle number as the head. This can happen
946 * because we allow multiple outstanding log writes concurrently,
947 * and the later writes might make it out before earlier ones.
949 * We use the lsn from before modifying it so that we'll never
950 * overwrite the unmount record after a clean unmount.
952 * Do this only if we are going to recover the filesystem
954 * NOTE: This used to say "if (!readonly)"
955 * However on Linux, we can & do recover a read-only filesystem.
956 * We only skip recovery if NORECOVERY is specified on mount,
957 * in which case we would not be here.
959 * But... if the -device- itself is readonly, just skip this.
960 * We can't recover this device anyway, so it won't matter.
962 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) {
963 error = xlog_clear_stale_blocks(log, tail_lsn);
971 xlog_warn("XFS: failed to locate log tail");
976 * Is the log zeroed at all?
978 * The last binary search should be changed to perform an X block read
979 * once X becomes small enough. You can then search linearly through
980 * the X blocks. This will cut down on the number of reads we need to do.
982 * If the log is partially zeroed, this routine will pass back the blkno
983 * of the first block with cycle number 0. It won't have a complete LR
987 * 0 => the log is completely written to
988 * -1 => use *blk_no as the first block of the log
989 * >0 => error has occurred
998 uint first_cycle, last_cycle;
999 xfs_daddr_t new_blk, last_blk, start_blk;
1000 xfs_daddr_t num_scan_bblks;
1001 int error, log_bbnum = log->l_logBBsize;
1005 /* check totally zeroed log */
1006 bp = xlog_get_bp(log, 1);
1009 if ((error = xlog_bread(log, 0, 1, bp)))
1011 offset = xlog_align(log, 0, 1, bp);
1012 first_cycle = xlog_get_cycle(offset);
1013 if (first_cycle == 0) { /* completely zeroed log */
1019 /* check partially zeroed log */
1020 if ((error = xlog_bread(log, log_bbnum-1, 1, bp)))
1022 offset = xlog_align(log, log_bbnum-1, 1, bp);
1023 last_cycle = xlog_get_cycle(offset);
1024 if (last_cycle != 0) { /* log completely written to */
1027 } else if (first_cycle != 1) {
1029 * If the cycle of the last block is zero, the cycle of
1030 * the first block must be 1. If it's not, maybe we're
1031 * not looking at a log... Bail out.
1033 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1034 return XFS_ERROR(EINVAL);
1037 /* we have a partially zeroed log */
1038 last_blk = log_bbnum-1;
1039 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1043 * Validate the answer. Because there is no way to guarantee that
1044 * the entire log is made up of log records which are the same size,
1045 * we scan over the defined maximum blocks. At this point, the maximum
1046 * is not chosen to mean anything special. XXXmiken
1048 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1049 ASSERT(num_scan_bblks <= INT_MAX);
1051 if (last_blk < num_scan_bblks)
1052 num_scan_bblks = last_blk;
1053 start_blk = last_blk - num_scan_bblks;
1056 * We search for any instances of cycle number 0 that occur before
1057 * our current estimate of the head. What we're trying to detect is
1058 * 1 ... | 0 | 1 | 0...
1059 * ^ binary search ends here
1061 if ((error = xlog_find_verify_cycle(log, start_blk,
1062 (int)num_scan_bblks, 0, &new_blk)))
1068 * Potentially backup over partial log record write. We don't need
1069 * to search the end of the log because we know it is zero.
1071 if ((error = xlog_find_verify_log_record(log, start_blk,
1072 &last_blk, 0)) == -1) {
1073 error = XFS_ERROR(EIO);
1087 * These are simple subroutines used by xlog_clear_stale_blocks() below
1088 * to initialize a buffer full of empty log record headers and write
1089 * them into the log.
1100 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1102 memset(buf, 0, BBSIZE);
1103 INT_SET(recp->h_magicno, ARCH_CONVERT, XLOG_HEADER_MAGIC_NUM);
1104 INT_SET(recp->h_cycle, ARCH_CONVERT, cycle);
1105 INT_SET(recp->h_version, ARCH_CONVERT,
1106 XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? 2 : 1);
1107 INT_SET(recp->h_lsn, ARCH_CONVERT, xlog_assign_lsn(cycle, block));
1108 INT_SET(recp->h_tail_lsn, ARCH_CONVERT,
1109 xlog_assign_lsn(tail_cycle, tail_block));
1110 INT_SET(recp->h_fmt, ARCH_CONVERT, XLOG_FMT);
1111 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1115 xlog_write_log_records(
1126 int sectbb = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
1127 int end_block = start_block + blocks;
1132 bufblks = 1 << ffs(blocks);
1133 while (!(bp = xlog_get_bp(log, bufblks))) {
1135 if (bufblks <= log->l_sectbb_log)
1139 /* We may need to do a read at the start to fill in part of
1140 * the buffer in the starting sector not covered by the first
1143 balign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, start_block);
1144 if (balign != start_block) {
1145 if ((error = xlog_bread(log, start_block, 1, bp))) {
1149 j = start_block - balign;
1152 for (i = start_block; i < end_block; i += bufblks) {
1153 int bcount, endcount;
1155 bcount = min(bufblks, end_block - start_block);
1156 endcount = bcount - j;
1158 /* We may need to do a read at the end to fill in part of
1159 * the buffer in the final sector not covered by the write.
1160 * If this is the same sector as the above read, skip it.
1162 ealign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, end_block);
1163 if (j == 0 && (start_block + endcount > ealign)) {
1164 offset = XFS_BUF_PTR(bp);
1165 balign = BBTOB(ealign - start_block);
1166 XFS_BUF_SET_PTR(bp, offset + balign, BBTOB(sectbb));
1167 if ((error = xlog_bread(log, ealign, sectbb, bp)))
1169 XFS_BUF_SET_PTR(bp, offset, bufblks);
1172 offset = xlog_align(log, start_block, endcount, bp);
1173 for (; j < endcount; j++) {
1174 xlog_add_record(log, offset, cycle, i+j,
1175 tail_cycle, tail_block);
1178 error = xlog_bwrite(log, start_block, endcount, bp);
1181 start_block += endcount;
1189 * This routine is called to blow away any incomplete log writes out
1190 * in front of the log head. We do this so that we won't become confused
1191 * if we come up, write only a little bit more, and then crash again.
1192 * If we leave the partial log records out there, this situation could
1193 * cause us to think those partial writes are valid blocks since they
1194 * have the current cycle number. We get rid of them by overwriting them
1195 * with empty log records with the old cycle number rather than the
1198 * The tail lsn is passed in rather than taken from
1199 * the log so that we will not write over the unmount record after a
1200 * clean unmount in a 512 block log. Doing so would leave the log without
1201 * any valid log records in it until a new one was written. If we crashed
1202 * during that time we would not be able to recover.
1205 xlog_clear_stale_blocks(
1209 int tail_cycle, head_cycle;
1210 int tail_block, head_block;
1211 int tail_distance, max_distance;
1215 tail_cycle = CYCLE_LSN(tail_lsn);
1216 tail_block = BLOCK_LSN(tail_lsn);
1217 head_cycle = log->l_curr_cycle;
1218 head_block = log->l_curr_block;
1221 * Figure out the distance between the new head of the log
1222 * and the tail. We want to write over any blocks beyond the
1223 * head that we may have written just before the crash, but
1224 * we don't want to overwrite the tail of the log.
1226 if (head_cycle == tail_cycle) {
1228 * The tail is behind the head in the physical log,
1229 * so the distance from the head to the tail is the
1230 * distance from the head to the end of the log plus
1231 * the distance from the beginning of the log to the
1234 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1235 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1236 XFS_ERRLEVEL_LOW, log->l_mp);
1237 return XFS_ERROR(EFSCORRUPTED);
1239 tail_distance = tail_block + (log->l_logBBsize - head_block);
1242 * The head is behind the tail in the physical log,
1243 * so the distance from the head to the tail is just
1244 * the tail block minus the head block.
1246 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1247 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1248 XFS_ERRLEVEL_LOW, log->l_mp);
1249 return XFS_ERROR(EFSCORRUPTED);
1251 tail_distance = tail_block - head_block;
1255 * If the head is right up against the tail, we can't clear
1258 if (tail_distance <= 0) {
1259 ASSERT(tail_distance == 0);
1263 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1265 * Take the smaller of the maximum amount of outstanding I/O
1266 * we could have and the distance to the tail to clear out.
1267 * We take the smaller so that we don't overwrite the tail and
1268 * we don't waste all day writing from the head to the tail
1271 max_distance = MIN(max_distance, tail_distance);
1273 if ((head_block + max_distance) <= log->l_logBBsize) {
1275 * We can stomp all the blocks we need to without
1276 * wrapping around the end of the log. Just do it
1277 * in a single write. Use the cycle number of the
1278 * current cycle minus one so that the log will look like:
1281 error = xlog_write_log_records(log, (head_cycle - 1),
1282 head_block, max_distance, tail_cycle,
1288 * We need to wrap around the end of the physical log in
1289 * order to clear all the blocks. Do it in two separate
1290 * I/Os. The first write should be from the head to the
1291 * end of the physical log, and it should use the current
1292 * cycle number minus one just like above.
1294 distance = log->l_logBBsize - head_block;
1295 error = xlog_write_log_records(log, (head_cycle - 1),
1296 head_block, distance, tail_cycle,
1303 * Now write the blocks at the start of the physical log.
1304 * This writes the remainder of the blocks we want to clear.
1305 * It uses the current cycle number since we're now on the
1306 * same cycle as the head so that we get:
1307 * n ... n ... | n - 1 ...
1308 * ^^^^^ blocks we're writing
1310 distance = max_distance - (log->l_logBBsize - head_block);
1311 error = xlog_write_log_records(log, head_cycle, 0, distance,
1312 tail_cycle, tail_block);
1320 /******************************************************************************
1322 * Log recover routines
1324 ******************************************************************************
1327 STATIC xlog_recover_t *
1328 xlog_recover_find_tid(
1332 xlog_recover_t *p = q;
1335 if (p->r_log_tid == tid)
1343 xlog_recover_put_hashq(
1345 xlog_recover_t *trans)
1352 xlog_recover_add_item(
1353 xlog_recover_item_t **itemq)
1355 xlog_recover_item_t *item;
1357 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1358 xlog_recover_insert_item_backq(itemq, item);
1362 xlog_recover_add_to_cont_trans(
1363 xlog_recover_t *trans,
1367 xlog_recover_item_t *item;
1368 xfs_caddr_t ptr, old_ptr;
1371 item = trans->r_itemq;
1373 /* finish copying rest of trans header */
1374 xlog_recover_add_item(&trans->r_itemq);
1375 ptr = (xfs_caddr_t) &trans->r_theader +
1376 sizeof(xfs_trans_header_t) - len;
1377 memcpy(ptr, dp, len); /* d, s, l */
1380 item = item->ri_prev;
1382 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1383 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1385 ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
1386 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1387 item->ri_buf[item->ri_cnt-1].i_len += len;
1388 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1393 * The next region to add is the start of a new region. It could be
1394 * a whole region or it could be the first part of a new region. Because
1395 * of this, the assumption here is that the type and size fields of all
1396 * format structures fit into the first 32 bits of the structure.
1398 * This works because all regions must be 32 bit aligned. Therefore, we
1399 * either have both fields or we have neither field. In the case we have
1400 * neither field, the data part of the region is zero length. We only have
1401 * a log_op_header and can throw away the header since a new one will appear
1402 * later. If we have at least 4 bytes, then we can determine how many regions
1403 * will appear in the current log item.
1406 xlog_recover_add_to_trans(
1407 xlog_recover_t *trans,
1411 xfs_inode_log_format_t *in_f; /* any will do */
1412 xlog_recover_item_t *item;
1417 item = trans->r_itemq;
1419 ASSERT(*(uint *)dp == XFS_TRANS_HEADER_MAGIC);
1420 if (len == sizeof(xfs_trans_header_t))
1421 xlog_recover_add_item(&trans->r_itemq);
1422 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1426 ptr = kmem_alloc(len, KM_SLEEP);
1427 memcpy(ptr, dp, len);
1428 in_f = (xfs_inode_log_format_t *)ptr;
1430 if (item->ri_prev->ri_total != 0 &&
1431 item->ri_prev->ri_total == item->ri_prev->ri_cnt) {
1432 xlog_recover_add_item(&trans->r_itemq);
1434 item = trans->r_itemq;
1435 item = item->ri_prev;
1437 if (item->ri_total == 0) { /* first region to be added */
1438 item->ri_total = in_f->ilf_size;
1439 ASSERT(item->ri_total <= XLOG_MAX_REGIONS_IN_ITEM);
1440 item->ri_buf = kmem_zalloc((item->ri_total *
1441 sizeof(xfs_log_iovec_t)), KM_SLEEP);
1443 ASSERT(item->ri_total > item->ri_cnt);
1444 /* Description region is ri_buf[0] */
1445 item->ri_buf[item->ri_cnt].i_addr = ptr;
1446 item->ri_buf[item->ri_cnt].i_len = len;
1452 xlog_recover_new_tid(
1457 xlog_recover_t *trans;
1459 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1460 trans->r_log_tid = tid;
1462 xlog_recover_put_hashq(q, trans);
1466 xlog_recover_unlink_tid(
1468 xlog_recover_t *trans)
1473 ASSERT(trans != NULL);
1479 if (tp->r_next == trans) {
1487 "XFS: xlog_recover_unlink_tid: trans not found");
1489 return XFS_ERROR(EIO);
1491 tp->r_next = tp->r_next->r_next;
1497 xlog_recover_insert_item_backq(
1498 xlog_recover_item_t **q,
1499 xlog_recover_item_t *item)
1502 item->ri_prev = item->ri_next = item;
1506 item->ri_prev = (*q)->ri_prev;
1507 (*q)->ri_prev = item;
1508 item->ri_prev->ri_next = item;
1513 xlog_recover_insert_item_frontq(
1514 xlog_recover_item_t **q,
1515 xlog_recover_item_t *item)
1517 xlog_recover_insert_item_backq(q, item);
1522 xlog_recover_reorder_trans(
1523 xlog_recover_t *trans)
1525 xlog_recover_item_t *first_item, *itemq, *itemq_next;
1526 xfs_buf_log_format_t *buf_f;
1529 first_item = itemq = trans->r_itemq;
1530 trans->r_itemq = NULL;
1532 itemq_next = itemq->ri_next;
1533 buf_f = (xfs_buf_log_format_t *)itemq->ri_buf[0].i_addr;
1535 switch (ITEM_TYPE(itemq)) {
1537 flags = buf_f->blf_flags;
1538 if (!(flags & XFS_BLI_CANCEL)) {
1539 xlog_recover_insert_item_frontq(&trans->r_itemq,
1545 case XFS_LI_QUOTAOFF:
1548 xlog_recover_insert_item_backq(&trans->r_itemq, itemq);
1552 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1554 return XFS_ERROR(EIO);
1557 } while (first_item != itemq);
1562 * Build up the table of buf cancel records so that we don't replay
1563 * cancelled data in the second pass. For buffer records that are
1564 * not cancel records, there is nothing to do here so we just return.
1566 * If we get a cancel record which is already in the table, this indicates
1567 * that the buffer was cancelled multiple times. In order to ensure
1568 * that during pass 2 we keep the record in the table until we reach its
1569 * last occurrence in the log, we keep a reference count in the cancel
1570 * record in the table to tell us how many times we expect to see this
1571 * record during the second pass.
1574 xlog_recover_do_buffer_pass1(
1576 xfs_buf_log_format_t *buf_f)
1578 xfs_buf_cancel_t *bcp;
1579 xfs_buf_cancel_t *nextp;
1580 xfs_buf_cancel_t *prevp;
1581 xfs_buf_cancel_t **bucket;
1582 xfs_daddr_t blkno = 0;
1586 switch (buf_f->blf_type) {
1588 blkno = buf_f->blf_blkno;
1589 len = buf_f->blf_len;
1590 flags = buf_f->blf_flags;
1595 * If this isn't a cancel buffer item, then just return.
1597 if (!(flags & XFS_BLI_CANCEL))
1601 * Insert an xfs_buf_cancel record into the hash table of
1602 * them. If there is already an identical record, bump
1603 * its reference count.
1605 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1606 XLOG_BC_TABLE_SIZE];
1608 * If the hash bucket is empty then just insert a new record into
1611 if (*bucket == NULL) {
1612 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1614 bcp->bc_blkno = blkno;
1616 bcp->bc_refcount = 1;
1617 bcp->bc_next = NULL;
1623 * The hash bucket is not empty, so search for duplicates of our
1624 * record. If we find one them just bump its refcount. If not
1625 * then add us at the end of the list.
1629 while (nextp != NULL) {
1630 if (nextp->bc_blkno == blkno && nextp->bc_len == len) {
1631 nextp->bc_refcount++;
1635 nextp = nextp->bc_next;
1637 ASSERT(prevp != NULL);
1638 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1640 bcp->bc_blkno = blkno;
1642 bcp->bc_refcount = 1;
1643 bcp->bc_next = NULL;
1644 prevp->bc_next = bcp;
1648 * Check to see whether the buffer being recovered has a corresponding
1649 * entry in the buffer cancel record table. If it does then return 1
1650 * so that it will be cancelled, otherwise return 0. If the buffer is
1651 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
1652 * the refcount on the entry in the table and remove it from the table
1653 * if this is the last reference.
1655 * We remove the cancel record from the table when we encounter its
1656 * last occurrence in the log so that if the same buffer is re-used
1657 * again after its last cancellation we actually replay the changes
1658 * made at that point.
1661 xlog_check_buffer_cancelled(
1667 xfs_buf_cancel_t *bcp;
1668 xfs_buf_cancel_t *prevp;
1669 xfs_buf_cancel_t **bucket;
1671 if (log->l_buf_cancel_table == NULL) {
1673 * There is nothing in the table built in pass one,
1674 * so this buffer must not be cancelled.
1676 ASSERT(!(flags & XFS_BLI_CANCEL));
1680 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1681 XLOG_BC_TABLE_SIZE];
1685 * There is no corresponding entry in the table built
1686 * in pass one, so this buffer has not been cancelled.
1688 ASSERT(!(flags & XFS_BLI_CANCEL));
1693 * Search for an entry in the buffer cancel table that
1694 * matches our buffer.
1697 while (bcp != NULL) {
1698 if (bcp->bc_blkno == blkno && bcp->bc_len == len) {
1700 * We've go a match, so return 1 so that the
1701 * recovery of this buffer is cancelled.
1702 * If this buffer is actually a buffer cancel
1703 * log item, then decrement the refcount on the
1704 * one in the table and remove it if this is the
1707 if (flags & XFS_BLI_CANCEL) {
1709 if (bcp->bc_refcount == 0) {
1710 if (prevp == NULL) {
1711 *bucket = bcp->bc_next;
1713 prevp->bc_next = bcp->bc_next;
1716 sizeof(xfs_buf_cancel_t));
1725 * We didn't find a corresponding entry in the table, so
1726 * return 0 so that the buffer is NOT cancelled.
1728 ASSERT(!(flags & XFS_BLI_CANCEL));
1733 xlog_recover_do_buffer_pass2(
1735 xfs_buf_log_format_t *buf_f)
1737 xfs_daddr_t blkno = 0;
1741 switch (buf_f->blf_type) {
1743 blkno = buf_f->blf_blkno;
1744 flags = buf_f->blf_flags;
1745 len = buf_f->blf_len;
1749 return xlog_check_buffer_cancelled(log, blkno, len, flags);
1753 * Perform recovery for a buffer full of inodes. In these buffers,
1754 * the only data which should be recovered is that which corresponds
1755 * to the di_next_unlinked pointers in the on disk inode structures.
1756 * The rest of the data for the inodes is always logged through the
1757 * inodes themselves rather than the inode buffer and is recovered
1758 * in xlog_recover_do_inode_trans().
1760 * The only time when buffers full of inodes are fully recovered is
1761 * when the buffer is full of newly allocated inodes. In this case
1762 * the buffer will not be marked as an inode buffer and so will be
1763 * sent to xlog_recover_do_reg_buffer() below during recovery.
1766 xlog_recover_do_inode_buffer(
1768 xlog_recover_item_t *item,
1770 xfs_buf_log_format_t *buf_f)
1778 int next_unlinked_offset;
1780 xfs_agino_t *logged_nextp;
1781 xfs_agino_t *buffer_nextp;
1782 unsigned int *data_map = NULL;
1783 unsigned int map_size = 0;
1785 switch (buf_f->blf_type) {
1787 data_map = buf_f->blf_data_map;
1788 map_size = buf_f->blf_map_size;
1792 * Set the variables corresponding to the current region to
1793 * 0 so that we'll initialize them on the first pass through
1801 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1802 for (i = 0; i < inodes_per_buf; i++) {
1803 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1804 offsetof(xfs_dinode_t, di_next_unlinked);
1806 while (next_unlinked_offset >=
1807 (reg_buf_offset + reg_buf_bytes)) {
1809 * The next di_next_unlinked field is beyond
1810 * the current logged region. Find the next
1811 * logged region that contains or is beyond
1812 * the current di_next_unlinked field.
1815 bit = xfs_next_bit(data_map, map_size, bit);
1818 * If there are no more logged regions in the
1819 * buffer, then we're done.
1825 nbits = xfs_contig_bits(data_map, map_size,
1828 reg_buf_offset = bit << XFS_BLI_SHIFT;
1829 reg_buf_bytes = nbits << XFS_BLI_SHIFT;
1834 * If the current logged region starts after the current
1835 * di_next_unlinked field, then move on to the next
1836 * di_next_unlinked field.
1838 if (next_unlinked_offset < reg_buf_offset) {
1842 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1843 ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0);
1844 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1847 * The current logged region contains a copy of the
1848 * current di_next_unlinked field. Extract its value
1849 * and copy it to the buffer copy.
1851 logged_nextp = (xfs_agino_t *)
1852 ((char *)(item->ri_buf[item_index].i_addr) +
1853 (next_unlinked_offset - reg_buf_offset));
1854 if (unlikely(*logged_nextp == 0)) {
1855 xfs_fs_cmn_err(CE_ALERT, mp,
1856 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1858 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1859 XFS_ERRLEVEL_LOW, mp);
1860 return XFS_ERROR(EFSCORRUPTED);
1863 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1864 next_unlinked_offset);
1865 *buffer_nextp = *logged_nextp;
1872 * Perform a 'normal' buffer recovery. Each logged region of the
1873 * buffer should be copied over the corresponding region in the
1874 * given buffer. The bitmap in the buf log format structure indicates
1875 * where to place the logged data.
1879 xlog_recover_do_reg_buffer(
1880 xlog_recover_item_t *item,
1882 xfs_buf_log_format_t *buf_f)
1887 unsigned int *data_map = NULL;
1888 unsigned int map_size = 0;
1891 switch (buf_f->blf_type) {
1893 data_map = buf_f->blf_data_map;
1894 map_size = buf_f->blf_map_size;
1898 i = 1; /* 0 is the buf format structure */
1900 bit = xfs_next_bit(data_map, map_size, bit);
1903 nbits = xfs_contig_bits(data_map, map_size, bit);
1905 ASSERT(item->ri_buf[i].i_addr != NULL);
1906 ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0);
1907 ASSERT(XFS_BUF_COUNT(bp) >=
1908 ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT));
1911 * Do a sanity check if this is a dquot buffer. Just checking
1912 * the first dquot in the buffer should do. XXXThis is
1913 * probably a good thing to do for other buf types also.
1916 if (buf_f->blf_flags &
1917 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
1918 error = xfs_qm_dqcheck((xfs_disk_dquot_t *)
1919 item->ri_buf[i].i_addr,
1920 -1, 0, XFS_QMOPT_DOWARN,
1921 "dquot_buf_recover");
1924 memcpy(xfs_buf_offset(bp,
1925 (uint)bit << XFS_BLI_SHIFT), /* dest */
1926 item->ri_buf[i].i_addr, /* source */
1927 nbits<<XFS_BLI_SHIFT); /* length */
1932 /* Shouldn't be any more regions */
1933 ASSERT(i == item->ri_total);
1937 * Do some primitive error checking on ondisk dquot data structures.
1941 xfs_disk_dquot_t *ddq,
1943 uint type, /* used only when IO_dorepair is true */
1947 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
1951 * We can encounter an uninitialized dquot buffer for 2 reasons:
1952 * 1. If we crash while deleting the quotainode(s), and those blks got
1953 * used for user data. This is because we take the path of regular
1954 * file deletion; however, the size field of quotainodes is never
1955 * updated, so all the tricks that we play in itruncate_finish
1956 * don't quite matter.
1958 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1959 * But the allocation will be replayed so we'll end up with an
1960 * uninitialized quota block.
1962 * This is all fine; things are still consistent, and we haven't lost
1963 * any quota information. Just don't complain about bad dquot blks.
1965 if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) {
1966 if (flags & XFS_QMOPT_DOWARN)
1968 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
1969 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
1972 if (ddq->d_version != XFS_DQUOT_VERSION) {
1973 if (flags & XFS_QMOPT_DOWARN)
1975 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
1976 str, id, ddq->d_version, XFS_DQUOT_VERSION);
1980 if (ddq->d_flags != XFS_DQ_USER &&
1981 ddq->d_flags != XFS_DQ_PROJ &&
1982 ddq->d_flags != XFS_DQ_GROUP) {
1983 if (flags & XFS_QMOPT_DOWARN)
1985 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
1986 str, id, ddq->d_flags);
1990 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
1991 if (flags & XFS_QMOPT_DOWARN)
1993 "%s : ondisk-dquot 0x%p, ID mismatch: "
1994 "0x%x expected, found id 0x%x",
1995 str, ddq, id, be32_to_cpu(ddq->d_id));
1999 if (!errs && ddq->d_id) {
2000 if (ddq->d_blk_softlimit &&
2001 be64_to_cpu(ddq->d_bcount) >=
2002 be64_to_cpu(ddq->d_blk_softlimit)) {
2003 if (!ddq->d_btimer) {
2004 if (flags & XFS_QMOPT_DOWARN)
2006 "%s : Dquot ID 0x%x (0x%p) "
2007 "BLK TIMER NOT STARTED",
2008 str, (int)be32_to_cpu(ddq->d_id), ddq);
2012 if (ddq->d_ino_softlimit &&
2013 be64_to_cpu(ddq->d_icount) >=
2014 be64_to_cpu(ddq->d_ino_softlimit)) {
2015 if (!ddq->d_itimer) {
2016 if (flags & XFS_QMOPT_DOWARN)
2018 "%s : Dquot ID 0x%x (0x%p) "
2019 "INODE TIMER NOT STARTED",
2020 str, (int)be32_to_cpu(ddq->d_id), ddq);
2024 if (ddq->d_rtb_softlimit &&
2025 be64_to_cpu(ddq->d_rtbcount) >=
2026 be64_to_cpu(ddq->d_rtb_softlimit)) {
2027 if (!ddq->d_rtbtimer) {
2028 if (flags & XFS_QMOPT_DOWARN)
2030 "%s : Dquot ID 0x%x (0x%p) "
2031 "RTBLK TIMER NOT STARTED",
2032 str, (int)be32_to_cpu(ddq->d_id), ddq);
2038 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2041 if (flags & XFS_QMOPT_DOWARN)
2042 cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id);
2045 * Typically, a repair is only requested by quotacheck.
2048 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2049 memset(d, 0, sizeof(xfs_dqblk_t));
2051 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2052 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2053 d->dd_diskdq.d_flags = type;
2054 d->dd_diskdq.d_id = cpu_to_be32(id);
2060 * Perform a dquot buffer recovery.
2061 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2062 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2063 * Else, treat it as a regular buffer and do recovery.
2066 xlog_recover_do_dquot_buffer(
2069 xlog_recover_item_t *item,
2071 xfs_buf_log_format_t *buf_f)
2076 * Filesystems are required to send in quota flags at mount time.
2078 if (mp->m_qflags == 0) {
2083 if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF)
2084 type |= XFS_DQ_USER;
2085 if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF)
2086 type |= XFS_DQ_PROJ;
2087 if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF)
2088 type |= XFS_DQ_GROUP;
2090 * This type of quotas was turned off, so ignore this buffer
2092 if (log->l_quotaoffs_flag & type)
2095 xlog_recover_do_reg_buffer(item, bp, buf_f);
2099 * This routine replays a modification made to a buffer at runtime.
2100 * There are actually two types of buffer, regular and inode, which
2101 * are handled differently. Inode buffers are handled differently
2102 * in that we only recover a specific set of data from them, namely
2103 * the inode di_next_unlinked fields. This is because all other inode
2104 * data is actually logged via inode records and any data we replay
2105 * here which overlaps that may be stale.
2107 * When meta-data buffers are freed at run time we log a buffer item
2108 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
2109 * of the buffer in the log should not be replayed at recovery time.
2110 * This is so that if the blocks covered by the buffer are reused for
2111 * file data before we crash we don't end up replaying old, freed
2112 * meta-data into a user's file.
2114 * To handle the cancellation of buffer log items, we make two passes
2115 * over the log during recovery. During the first we build a table of
2116 * those buffers which have been cancelled, and during the second we
2117 * only replay those buffers which do not have corresponding cancel
2118 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2119 * for more details on the implementation of the table of cancel records.
2122 xlog_recover_do_buffer_trans(
2124 xlog_recover_item_t *item,
2127 xfs_buf_log_format_t *buf_f;
2136 buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr;
2138 if (pass == XLOG_RECOVER_PASS1) {
2140 * In this pass we're only looking for buf items
2141 * with the XFS_BLI_CANCEL bit set.
2143 xlog_recover_do_buffer_pass1(log, buf_f);
2147 * In this pass we want to recover all the buffers
2148 * which have not been cancelled and are not
2149 * cancellation buffers themselves. The routine
2150 * we call here will tell us whether or not to
2151 * continue with the replay of this buffer.
2153 cancel = xlog_recover_do_buffer_pass2(log, buf_f);
2158 switch (buf_f->blf_type) {
2160 blkno = buf_f->blf_blkno;
2161 len = buf_f->blf_len;
2162 flags = buf_f->blf_flags;
2165 xfs_fs_cmn_err(CE_ALERT, log->l_mp,
2166 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2167 buf_f->blf_type, log->l_mp->m_logname ?
2168 log->l_mp->m_logname : "internal");
2169 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2170 XFS_ERRLEVEL_LOW, log->l_mp);
2171 return XFS_ERROR(EFSCORRUPTED);
2175 if (flags & XFS_BLI_INODE_BUF) {
2176 bp = xfs_buf_read_flags(mp->m_ddev_targp, blkno, len,
2179 bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, 0);
2181 if (XFS_BUF_ISERROR(bp)) {
2182 xfs_ioerror_alert("xlog_recover_do..(read#1)", log->l_mp,
2184 error = XFS_BUF_GETERROR(bp);
2190 if (flags & XFS_BLI_INODE_BUF) {
2191 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2193 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
2194 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2196 xlog_recover_do_reg_buffer(item, bp, buf_f);
2199 return XFS_ERROR(error);
2202 * Perform delayed write on the buffer. Asynchronous writes will be
2203 * slower when taking into account all the buffers to be flushed.
2205 * Also make sure that only inode buffers with good sizes stay in
2206 * the buffer cache. The kernel moves inodes in buffers of 1 block
2207 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2208 * buffers in the log can be a different size if the log was generated
2209 * by an older kernel using unclustered inode buffers or a newer kernel
2210 * running with a different inode cluster size. Regardless, if the
2211 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2212 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2213 * the buffer out of the buffer cache so that the buffer won't
2214 * overlap with future reads of those inodes.
2216 if (XFS_DINODE_MAGIC ==
2217 INT_GET(*((__uint16_t *)(xfs_buf_offset(bp, 0))), ARCH_CONVERT) &&
2218 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2219 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2221 error = xfs_bwrite(mp, bp);
2223 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2224 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2225 XFS_BUF_SET_FSPRIVATE(bp, mp);
2226 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2227 xfs_bdwrite(mp, bp);
2234 xlog_recover_do_inode_trans(
2236 xlog_recover_item_t *item,
2239 xfs_inode_log_format_t *in_f;
2251 xfs_icdinode_t *dicp;
2254 if (pass == XLOG_RECOVER_PASS1) {
2258 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2259 in_f = (xfs_inode_log_format_t *)item->ri_buf[0].i_addr;
2261 in_f = (xfs_inode_log_format_t *)kmem_alloc(
2262 sizeof(xfs_inode_log_format_t), KM_SLEEP);
2264 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2268 ino = in_f->ilf_ino;
2270 if (ITEM_TYPE(item) == XFS_LI_INODE) {
2271 imap.im_blkno = (xfs_daddr_t)in_f->ilf_blkno;
2272 imap.im_len = in_f->ilf_len;
2273 imap.im_boffset = in_f->ilf_boffset;
2276 * It's an old inode format record. We don't know where
2277 * its cluster is located on disk, and we can't allow
2278 * xfs_imap() to figure it out because the inode btrees
2279 * are not ready to be used. Therefore do not pass the
2280 * XFS_IMAP_LOOKUP flag to xfs_imap(). This will give
2281 * us only the single block in which the inode lives
2282 * rather than its cluster, so we must make sure to
2283 * invalidate the buffer when we write it out below.
2286 xfs_imap(log->l_mp, NULL, ino, &imap, 0);
2290 * Inode buffers can be freed, look out for it,
2291 * and do not replay the inode.
2293 if (xlog_check_buffer_cancelled(log, imap.im_blkno, imap.im_len, 0)) {
2298 bp = xfs_buf_read_flags(mp->m_ddev_targp, imap.im_blkno, imap.im_len,
2300 if (XFS_BUF_ISERROR(bp)) {
2301 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
2303 error = XFS_BUF_GETERROR(bp);
2308 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2309 dip = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
2312 * Make sure the place we're flushing out to really looks
2315 if (unlikely(be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC)) {
2317 xfs_fs_cmn_err(CE_ALERT, mp,
2318 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2320 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2321 XFS_ERRLEVEL_LOW, mp);
2322 error = EFSCORRUPTED;
2325 dicp = (xfs_icdinode_t *)(item->ri_buf[1].i_addr);
2326 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2328 xfs_fs_cmn_err(CE_ALERT, mp,
2329 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2331 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2332 XFS_ERRLEVEL_LOW, mp);
2333 error = EFSCORRUPTED;
2337 /* Skip replay when the on disk inode is newer than the log one */
2338 if (dicp->di_flushiter < be16_to_cpu(dip->di_core.di_flushiter)) {
2340 * Deal with the wrap case, DI_MAX_FLUSH is less
2341 * than smaller numbers
2343 if (be16_to_cpu(dip->di_core.di_flushiter) == DI_MAX_FLUSH &&
2344 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2352 /* Take the opportunity to reset the flush iteration count */
2353 dicp->di_flushiter = 0;
2355 if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) {
2356 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2357 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2358 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2359 XFS_ERRLEVEL_LOW, mp, dicp);
2361 xfs_fs_cmn_err(CE_ALERT, mp,
2362 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2363 item, dip, bp, ino);
2364 error = EFSCORRUPTED;
2367 } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) {
2368 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2369 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2370 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2371 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2372 XFS_ERRLEVEL_LOW, mp, dicp);
2374 xfs_fs_cmn_err(CE_ALERT, mp,
2375 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2376 item, dip, bp, ino);
2377 error = EFSCORRUPTED;
2381 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2382 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2383 XFS_ERRLEVEL_LOW, mp, dicp);
2385 xfs_fs_cmn_err(CE_ALERT, mp,
2386 "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",
2388 dicp->di_nextents + dicp->di_anextents,
2390 error = EFSCORRUPTED;
2393 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2394 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2395 XFS_ERRLEVEL_LOW, mp, dicp);
2397 xfs_fs_cmn_err(CE_ALERT, mp,
2398 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2399 item, dip, bp, ino, dicp->di_forkoff);
2400 error = EFSCORRUPTED;
2403 if (unlikely(item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t))) {
2404 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2405 XFS_ERRLEVEL_LOW, mp, dicp);
2407 xfs_fs_cmn_err(CE_ALERT, mp,
2408 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2409 item->ri_buf[1].i_len, item);
2410 error = EFSCORRUPTED;
2414 /* The core is in in-core format */
2415 xfs_dinode_to_disk(&dip->di_core,
2416 (xfs_icdinode_t *)item->ri_buf[1].i_addr);
2418 /* the rest is in on-disk format */
2419 if (item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t)) {
2420 memcpy((xfs_caddr_t) dip + sizeof(xfs_dinode_core_t),
2421 item->ri_buf[1].i_addr + sizeof(xfs_dinode_core_t),
2422 item->ri_buf[1].i_len - sizeof(xfs_dinode_core_t));
2425 fields = in_f->ilf_fields;
2426 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2428 dip->di_u.di_dev = cpu_to_be32(in_f->ilf_u.ilfu_rdev);
2431 dip->di_u.di_muuid = in_f->ilf_u.ilfu_uuid;
2435 if (in_f->ilf_size == 2)
2436 goto write_inode_buffer;
2437 len = item->ri_buf[2].i_len;
2438 src = item->ri_buf[2].i_addr;
2439 ASSERT(in_f->ilf_size <= 4);
2440 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2441 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2442 (len == in_f->ilf_dsize));
2444 switch (fields & XFS_ILOG_DFORK) {
2445 case XFS_ILOG_DDATA:
2447 memcpy(&dip->di_u, src, len);
2450 case XFS_ILOG_DBROOT:
2451 xfs_bmbt_to_bmdr((xfs_bmbt_block_t *)src, len,
2452 &(dip->di_u.di_bmbt),
2453 XFS_DFORK_DSIZE(dip, mp));
2458 * There are no data fork flags set.
2460 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2465 * If we logged any attribute data, recover it. There may or
2466 * may not have been any other non-core data logged in this
2469 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2470 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2475 len = item->ri_buf[attr_index].i_len;
2476 src = item->ri_buf[attr_index].i_addr;
2477 ASSERT(len == in_f->ilf_asize);
2479 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2480 case XFS_ILOG_ADATA:
2482 dest = XFS_DFORK_APTR(dip);
2483 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2484 memcpy(dest, src, len);
2487 case XFS_ILOG_ABROOT:
2488 dest = XFS_DFORK_APTR(dip);
2489 xfs_bmbt_to_bmdr((xfs_bmbt_block_t *)src, len,
2490 (xfs_bmdr_block_t*)dest,
2491 XFS_DFORK_ASIZE(dip, mp));
2495 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2504 if (ITEM_TYPE(item) == XFS_LI_INODE) {
2505 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2506 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2507 XFS_BUF_SET_FSPRIVATE(bp, mp);
2508 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2509 xfs_bdwrite(mp, bp);
2512 error = xfs_bwrite(mp, bp);
2517 kmem_free(in_f, sizeof(*in_f));
2518 return XFS_ERROR(error);
2522 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2523 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2527 xlog_recover_do_quotaoff_trans(
2529 xlog_recover_item_t *item,
2532 xfs_qoff_logformat_t *qoff_f;
2534 if (pass == XLOG_RECOVER_PASS2) {
2538 qoff_f = (xfs_qoff_logformat_t *)item->ri_buf[0].i_addr;
2542 * The logitem format's flag tells us if this was user quotaoff,
2543 * group/project quotaoff or both.
2545 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2546 log->l_quotaoffs_flag |= XFS_DQ_USER;
2547 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2548 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2549 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2550 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2556 * Recover a dquot record
2559 xlog_recover_do_dquot_trans(
2561 xlog_recover_item_t *item,
2566 struct xfs_disk_dquot *ddq, *recddq;
2568 xfs_dq_logformat_t *dq_f;
2571 if (pass == XLOG_RECOVER_PASS1) {
2577 * Filesystems are required to send in quota flags at mount time.
2579 if (mp->m_qflags == 0)
2582 recddq = (xfs_disk_dquot_t *)item->ri_buf[1].i_addr;
2585 * This type of quotas was turned off, so ignore this record.
2587 type = INT_GET(recddq->d_flags, ARCH_CONVERT) &
2588 (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
2590 if (log->l_quotaoffs_flag & type)
2594 * At this point we know that quota was _not_ turned off.
2595 * Since the mount flags are not indicating to us otherwise, this
2596 * must mean that quota is on, and the dquot needs to be replayed.
2597 * Remember that we may not have fully recovered the superblock yet,
2598 * so we can't do the usual trick of looking at the SB quota bits.
2600 * The other possibility, of course, is that the quota subsystem was
2601 * removed since the last mount - ENOSYS.
2603 dq_f = (xfs_dq_logformat_t *)item->ri_buf[0].i_addr;
2605 if ((error = xfs_qm_dqcheck(recddq,
2607 0, XFS_QMOPT_DOWARN,
2608 "xlog_recover_do_dquot_trans (log copy)"))) {
2609 return XFS_ERROR(EIO);
2611 ASSERT(dq_f->qlf_len == 1);
2613 error = xfs_read_buf(mp, mp->m_ddev_targp,
2615 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2618 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
2619 bp, dq_f->qlf_blkno);
2623 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2626 * At least the magic num portion should be on disk because this
2627 * was among a chunk of dquots created earlier, and we did some
2628 * minimal initialization then.
2630 if (xfs_qm_dqcheck(ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2631 "xlog_recover_do_dquot_trans")) {
2633 return XFS_ERROR(EIO);
2636 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2638 ASSERT(dq_f->qlf_size == 2);
2639 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2640 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2641 XFS_BUF_SET_FSPRIVATE(bp, mp);
2642 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2643 xfs_bdwrite(mp, bp);
2649 * This routine is called to create an in-core extent free intent
2650 * item from the efi format structure which was logged on disk.
2651 * It allocates an in-core efi, copies the extents from the format
2652 * structure into it, and adds the efi to the AIL with the given
2656 xlog_recover_do_efi_trans(
2658 xlog_recover_item_t *item,
2664 xfs_efi_log_item_t *efip;
2665 xfs_efi_log_format_t *efi_formatp;
2667 if (pass == XLOG_RECOVER_PASS1) {
2671 efi_formatp = (xfs_efi_log_format_t *)item->ri_buf[0].i_addr;
2674 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
2675 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2676 &(efip->efi_format)))) {
2677 xfs_efi_item_free(efip);
2680 efip->efi_next_extent = efi_formatp->efi_nextents;
2681 efip->efi_flags |= XFS_EFI_COMMITTED;
2683 spin_lock(&mp->m_ail_lock);
2685 * xfs_trans_update_ail() drops the AIL lock.
2687 xfs_trans_update_ail(mp, (xfs_log_item_t *)efip, lsn);
2693 * This routine is called when an efd format structure is found in
2694 * a committed transaction in the log. It's purpose is to cancel
2695 * the corresponding efi if it was still in the log. To do this
2696 * it searches the AIL for the efi with an id equal to that in the
2697 * efd format structure. If we find it, we remove the efi from the
2701 xlog_recover_do_efd_trans(
2703 xlog_recover_item_t *item,
2707 xfs_efd_log_format_t *efd_formatp;
2708 xfs_efi_log_item_t *efip = NULL;
2709 xfs_log_item_t *lip;
2713 if (pass == XLOG_RECOVER_PASS1) {
2717 efd_formatp = (xfs_efd_log_format_t *)item->ri_buf[0].i_addr;
2718 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2719 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2720 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2721 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
2722 efi_id = efd_formatp->efd_efi_id;
2725 * Search for the efi with the id in the efd format structure
2729 spin_lock(&mp->m_ail_lock);
2730 lip = xfs_trans_first_ail(mp, &gen);
2731 while (lip != NULL) {
2732 if (lip->li_type == XFS_LI_EFI) {
2733 efip = (xfs_efi_log_item_t *)lip;
2734 if (efip->efi_format.efi_id == efi_id) {
2736 * xfs_trans_delete_ail() drops the
2739 xfs_trans_delete_ail(mp, lip);
2743 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
2747 * If we found it, then free it up. If it wasn't there, it
2748 * must have been overwritten in the log. Oh well.
2751 xfs_efi_item_free(efip);
2753 spin_unlock(&mp->m_ail_lock);
2758 * Perform the transaction
2760 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2761 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2764 xlog_recover_do_trans(
2766 xlog_recover_t *trans,
2770 xlog_recover_item_t *item, *first_item;
2772 if ((error = xlog_recover_reorder_trans(trans)))
2774 first_item = item = trans->r_itemq;
2777 * we don't need to worry about the block number being
2778 * truncated in > 1 TB buffers because in user-land,
2779 * we're now n32 or 64-bit so xfs_daddr_t is 64-bits so
2780 * the blknos will get through the user-mode buffer
2781 * cache properly. The only bad case is o32 kernels
2782 * where xfs_daddr_t is 32-bits but mount will warn us
2783 * off a > 1 TB filesystem before we get here.
2785 if ((ITEM_TYPE(item) == XFS_LI_BUF)) {
2786 if ((error = xlog_recover_do_buffer_trans(log, item,
2789 } else if ((ITEM_TYPE(item) == XFS_LI_INODE)) {
2790 if ((error = xlog_recover_do_inode_trans(log, item,
2793 } else if (ITEM_TYPE(item) == XFS_LI_EFI) {
2794 if ((error = xlog_recover_do_efi_trans(log, item, trans->r_lsn,
2797 } else if (ITEM_TYPE(item) == XFS_LI_EFD) {
2798 xlog_recover_do_efd_trans(log, item, pass);
2799 } else if (ITEM_TYPE(item) == XFS_LI_DQUOT) {
2800 if ((error = xlog_recover_do_dquot_trans(log, item,
2803 } else if ((ITEM_TYPE(item) == XFS_LI_QUOTAOFF)) {
2804 if ((error = xlog_recover_do_quotaoff_trans(log, item,
2808 xlog_warn("XFS: xlog_recover_do_trans");
2810 error = XFS_ERROR(EIO);
2813 item = item->ri_next;
2814 } while (first_item != item);
2820 * Free up any resources allocated by the transaction
2822 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2825 xlog_recover_free_trans(
2826 xlog_recover_t *trans)
2828 xlog_recover_item_t *first_item, *item, *free_item;
2831 item = first_item = trans->r_itemq;
2834 item = item->ri_next;
2835 /* Free the regions in the item. */
2836 for (i = 0; i < free_item->ri_cnt; i++) {
2837 kmem_free(free_item->ri_buf[i].i_addr,
2838 free_item->ri_buf[i].i_len);
2840 /* Free the item itself */
2841 kmem_free(free_item->ri_buf,
2842 (free_item->ri_total * sizeof(xfs_log_iovec_t)));
2843 kmem_free(free_item, sizeof(xlog_recover_item_t));
2844 } while (first_item != item);
2845 /* Free the transaction recover structure */
2846 kmem_free(trans, sizeof(xlog_recover_t));
2850 xlog_recover_commit_trans(
2853 xlog_recover_t *trans,
2858 if ((error = xlog_recover_unlink_tid(q, trans)))
2860 if ((error = xlog_recover_do_trans(log, trans, pass)))
2862 xlog_recover_free_trans(trans); /* no error */
2867 xlog_recover_unmount_trans(
2868 xlog_recover_t *trans)
2870 /* Do nothing now */
2871 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2876 * There are two valid states of the r_state field. 0 indicates that the
2877 * transaction structure is in a normal state. We have either seen the
2878 * start of the transaction or the last operation we added was not a partial
2879 * operation. If the last operation we added to the transaction was a
2880 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2882 * NOTE: skip LRs with 0 data length.
2885 xlog_recover_process_data(
2887 xlog_recover_t *rhash[],
2888 xlog_rec_header_t *rhead,
2894 xlog_op_header_t *ohead;
2895 xlog_recover_t *trans;
2901 lp = dp + INT_GET(rhead->h_len, ARCH_CONVERT);
2902 num_logops = INT_GET(rhead->h_num_logops, ARCH_CONVERT);
2904 /* check the log format matches our own - else we can't recover */
2905 if (xlog_header_check_recover(log->l_mp, rhead))
2906 return (XFS_ERROR(EIO));
2908 while ((dp < lp) && num_logops) {
2909 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2910 ohead = (xlog_op_header_t *)dp;
2911 dp += sizeof(xlog_op_header_t);
2912 if (ohead->oh_clientid != XFS_TRANSACTION &&
2913 ohead->oh_clientid != XFS_LOG) {
2915 "XFS: xlog_recover_process_data: bad clientid");
2917 return (XFS_ERROR(EIO));
2919 tid = be32_to_cpu(ohead->oh_tid);
2920 hash = XLOG_RHASH(tid);
2921 trans = xlog_recover_find_tid(rhash[hash], tid);
2922 if (trans == NULL) { /* not found; add new tid */
2923 if (ohead->oh_flags & XLOG_START_TRANS)
2924 xlog_recover_new_tid(&rhash[hash], tid,
2925 INT_GET(rhead->h_lsn, ARCH_CONVERT));
2927 ASSERT(dp + be32_to_cpu(ohead->oh_len) <= lp);
2928 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2929 if (flags & XLOG_WAS_CONT_TRANS)
2930 flags &= ~XLOG_CONTINUE_TRANS;
2932 case XLOG_COMMIT_TRANS:
2933 error = xlog_recover_commit_trans(log,
2934 &rhash[hash], trans, pass);
2936 case XLOG_UNMOUNT_TRANS:
2937 error = xlog_recover_unmount_trans(trans);
2939 case XLOG_WAS_CONT_TRANS:
2940 error = xlog_recover_add_to_cont_trans(trans,
2941 dp, be32_to_cpu(ohead->oh_len));
2943 case XLOG_START_TRANS:
2945 "XFS: xlog_recover_process_data: bad transaction");
2947 error = XFS_ERROR(EIO);
2950 case XLOG_CONTINUE_TRANS:
2951 error = xlog_recover_add_to_trans(trans,
2952 dp, be32_to_cpu(ohead->oh_len));
2956 "XFS: xlog_recover_process_data: bad flag");
2958 error = XFS_ERROR(EIO);
2964 dp += be32_to_cpu(ohead->oh_len);
2971 * Process an extent free intent item that was recovered from
2972 * the log. We need to free the extents that it describes.
2975 xlog_recover_process_efi(
2977 xfs_efi_log_item_t *efip)
2979 xfs_efd_log_item_t *efdp;
2983 xfs_fsblock_t startblock_fsb;
2985 ASSERT(!(efip->efi_flags & XFS_EFI_RECOVERED));
2988 * First check the validity of the extents described by the
2989 * EFI. If any are bad, then assume that all are bad and
2990 * just toss the EFI.
2992 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2993 extp = &(efip->efi_format.efi_extents[i]);
2994 startblock_fsb = XFS_BB_TO_FSB(mp,
2995 XFS_FSB_TO_DADDR(mp, extp->ext_start));
2996 if ((startblock_fsb == 0) ||
2997 (extp->ext_len == 0) ||
2998 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
2999 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3001 * This will pull the EFI from the AIL and
3002 * free the memory associated with it.
3004 xfs_efi_release(efip, efip->efi_format.efi_nextents);
3009 tp = xfs_trans_alloc(mp, 0);
3010 xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
3011 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3013 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3014 extp = &(efip->efi_format.efi_extents[i]);
3015 xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3016 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3020 efip->efi_flags |= XFS_EFI_RECOVERED;
3021 xfs_trans_commit(tp, 0);
3025 * Verify that once we've encountered something other than an EFI
3026 * in the AIL that there are no more EFIs in the AIL.
3030 xlog_recover_check_ail(
3032 xfs_log_item_t *lip,
3038 ASSERT(lip->li_type != XFS_LI_EFI);
3039 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
3041 * The check will be bogus if we restart from the
3042 * beginning of the AIL, so ASSERT that we don't.
3043 * We never should since we're holding the AIL lock
3046 ASSERT(gen == orig_gen);
3047 } while (lip != NULL);
3052 * When this is called, all of the EFIs which did not have
3053 * corresponding EFDs should be in the AIL. What we do now
3054 * is free the extents associated with each one.
3056 * Since we process the EFIs in normal transactions, they
3057 * will be removed at some point after the commit. This prevents
3058 * us from just walking down the list processing each one.
3059 * We'll use a flag in the EFI to skip those that we've already
3060 * processed and use the AIL iteration mechanism's generation
3061 * count to try to speed this up at least a bit.
3063 * When we start, we know that the EFIs are the only things in
3064 * the AIL. As we process them, however, other items are added
3065 * to the AIL. Since everything added to the AIL must come after
3066 * everything already in the AIL, we stop processing as soon as
3067 * we see something other than an EFI in the AIL.
3070 xlog_recover_process_efis(
3073 xfs_log_item_t *lip;
3074 xfs_efi_log_item_t *efip;
3079 spin_lock(&mp->m_ail_lock);
3081 lip = xfs_trans_first_ail(mp, &gen);
3082 while (lip != NULL) {
3084 * We're done when we see something other than an EFI.
3086 if (lip->li_type != XFS_LI_EFI) {
3087 xlog_recover_check_ail(mp, lip, gen);
3092 * Skip EFIs that we've already processed.
3094 efip = (xfs_efi_log_item_t *)lip;
3095 if (efip->efi_flags & XFS_EFI_RECOVERED) {
3096 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
3100 spin_unlock(&mp->m_ail_lock);
3101 xlog_recover_process_efi(mp, efip);
3102 spin_lock(&mp->m_ail_lock);
3103 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
3105 spin_unlock(&mp->m_ail_lock);
3109 * This routine performs a transaction to null out a bad inode pointer
3110 * in an agi unlinked inode hash bucket.
3113 xlog_recover_clear_agi_bucket(
3115 xfs_agnumber_t agno,
3124 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3125 xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp), 0, 0, 0);
3127 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
3128 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
3129 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
3131 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3135 agi = XFS_BUF_TO_AGI(agibp);
3136 if (be32_to_cpu(agi->agi_magicnum) != XFS_AGI_MAGIC) {
3137 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3141 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
3142 offset = offsetof(xfs_agi_t, agi_unlinked) +
3143 (sizeof(xfs_agino_t) * bucket);
3144 xfs_trans_log_buf(tp, agibp, offset,
3145 (offset + sizeof(xfs_agino_t) - 1));
3147 (void) xfs_trans_commit(tp, 0);
3151 * xlog_iunlink_recover
3153 * This is called during recovery to process any inodes which
3154 * we unlinked but not freed when the system crashed. These
3155 * inodes will be on the lists in the AGI blocks. What we do
3156 * here is scan all the AGIs and fully truncate and free any
3157 * inodes found on the lists. Each inode is removed from the
3158 * lists when it has been fully truncated and is freed. The
3159 * freeing of the inode and its removal from the list must be
3163 xlog_recover_process_iunlinks(
3167 xfs_agnumber_t agno;
3182 * Prevent any DMAPI event from being sent while in this function.
3184 mp_dmevmask = mp->m_dmevmask;
3187 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3189 * Find the agi for this ag.
3191 agibp = xfs_buf_read(mp->m_ddev_targp,
3192 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
3193 XFS_FSS_TO_BB(mp, 1), 0);
3194 if (XFS_BUF_ISERROR(agibp)) {
3195 xfs_ioerror_alert("xlog_recover_process_iunlinks(#1)",
3197 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)));
3199 agi = XFS_BUF_TO_AGI(agibp);
3200 ASSERT(XFS_AGI_MAGIC == be32_to_cpu(agi->agi_magicnum));
3202 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3204 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3205 while (agino != NULLAGINO) {
3208 * Release the agi buffer so that it can
3209 * be acquired in the normal course of the
3210 * transaction to truncate and free the inode.
3212 xfs_buf_relse(agibp);
3214 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3215 error = xfs_iget(mp, NULL, ino, 0, 0, &ip, 0);
3216 ASSERT(error || (ip != NULL));
3220 * Get the on disk inode to find the
3221 * next inode in the bucket.
3223 error = xfs_itobp(mp, NULL, ip, &dip,
3225 ASSERT(error || (dip != NULL));
3229 ASSERT(ip->i_d.di_nlink == 0);
3231 /* setup for the next pass */
3232 agino = be32_to_cpu(
3233 dip->di_next_unlinked);
3236 * Prevent any DMAPI event from
3237 * being sent when the
3238 * reference on the inode is
3241 ip->i_d.di_dmevmask = 0;
3244 * If this is a new inode, handle
3245 * it specially. Otherwise,
3246 * just drop our reference to the
3247 * inode. If there are no
3248 * other references, this will
3250 * xfs_inactive() which will
3251 * truncate the file and free
3254 if (ip->i_d.di_mode == 0)
3255 xfs_iput_new(ip, 0);
3257 VN_RELE(XFS_ITOV(ip));
3260 * We can't read in the inode
3261 * this bucket points to, or
3262 * this inode is messed up. Just
3263 * ditch this bucket of inodes. We
3264 * will lose some inodes and space,
3265 * but at least we won't hang. Call
3266 * xlog_recover_clear_agi_bucket()
3267 * to perform a transaction to clear
3268 * the inode pointer in the bucket.
3270 xlog_recover_clear_agi_bucket(mp, agno,
3277 * Reacquire the agibuffer and continue around
3280 agibp = xfs_buf_read(mp->m_ddev_targp,
3281 XFS_AG_DADDR(mp, agno,
3283 XFS_FSS_TO_BB(mp, 1), 0);
3284 if (XFS_BUF_ISERROR(agibp)) {
3286 "xlog_recover_process_iunlinks(#2)",
3288 XFS_AG_DADDR(mp, agno,
3289 XFS_AGI_DADDR(mp)));
3291 agi = XFS_BUF_TO_AGI(agibp);
3292 ASSERT(XFS_AGI_MAGIC == be32_to_cpu(
3293 agi->agi_magicnum));
3298 * Release the buffer for the current agi so we can
3299 * go on to the next one.
3301 xfs_buf_relse(agibp);
3304 mp->m_dmevmask = mp_dmevmask;
3310 xlog_pack_data_checksum(
3312 xlog_in_core_t *iclog,
3319 up = (uint *)iclog->ic_datap;
3320 /* divide length by 4 to get # words */
3321 for (i = 0; i < (size >> 2); i++) {
3322 chksum ^= INT_GET(*up, ARCH_CONVERT);
3325 INT_SET(iclog->ic_header.h_chksum, ARCH_CONVERT, chksum);
3328 #define xlog_pack_data_checksum(log, iclog, size)
3332 * Stamp cycle number in every block
3337 xlog_in_core_t *iclog,
3341 int size = iclog->ic_offset + roundoff;
3344 xlog_in_core_2_t *xhdr;
3346 xlog_pack_data_checksum(log, iclog, size);
3348 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3350 dp = iclog->ic_datap;
3351 for (i = 0; i < BTOBB(size) &&
3352 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3353 iclog->ic_header.h_cycle_data[i] = *(uint *)dp;
3354 *(uint *)dp = cycle_lsn;
3358 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3359 xhdr = (xlog_in_core_2_t *)&iclog->ic_header;
3360 for ( ; i < BTOBB(size); i++) {
3361 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3362 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3363 xhdr[j].hic_xheader.xh_cycle_data[k] = *(uint *)dp;
3364 *(uint *)dp = cycle_lsn;
3368 for (i = 1; i < log->l_iclog_heads; i++) {
3369 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3374 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
3376 xlog_unpack_data_checksum(
3377 xlog_rec_header_t *rhead,
3381 uint *up = (uint *)dp;
3385 /* divide length by 4 to get # words */
3386 for (i=0; i < INT_GET(rhead->h_len, ARCH_CONVERT) >> 2; i++) {
3387 chksum ^= INT_GET(*up, ARCH_CONVERT);
3390 if (chksum != INT_GET(rhead->h_chksum, ARCH_CONVERT)) {
3391 if (rhead->h_chksum ||
3392 ((log->l_flags & XLOG_CHKSUM_MISMATCH) == 0)) {
3394 "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n",
3395 INT_GET(rhead->h_chksum, ARCH_CONVERT), chksum);
3397 "XFS: Disregard message if filesystem was created with non-DEBUG kernel");
3398 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3400 "XFS: LogR this is a LogV2 filesystem\n");
3402 log->l_flags |= XLOG_CHKSUM_MISMATCH;
3407 #define xlog_unpack_data_checksum(rhead, dp, log)
3412 xlog_rec_header_t *rhead,
3417 xlog_in_core_2_t *xhdr;
3419 for (i = 0; i < BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT)) &&
3420 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3421 *(uint *)dp = *(uint *)&rhead->h_cycle_data[i];
3425 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3426 xhdr = (xlog_in_core_2_t *)rhead;
3427 for ( ; i < BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT)); i++) {
3428 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3429 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3430 *(uint *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
3435 xlog_unpack_data_checksum(rhead, dp, log);
3439 xlog_valid_rec_header(
3441 xlog_rec_header_t *rhead,
3447 (INT_GET(rhead->h_magicno, ARCH_CONVERT) !=
3448 XLOG_HEADER_MAGIC_NUM))) {
3449 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3450 XFS_ERRLEVEL_LOW, log->l_mp);
3451 return XFS_ERROR(EFSCORRUPTED);
3454 (!rhead->h_version ||
3455 (INT_GET(rhead->h_version, ARCH_CONVERT) &
3456 (~XLOG_VERSION_OKBITS)) != 0))) {
3457 xlog_warn("XFS: %s: unrecognised log version (%d).",
3458 __FUNCTION__, INT_GET(rhead->h_version, ARCH_CONVERT));
3459 return XFS_ERROR(EIO);
3462 /* LR body must have data or it wouldn't have been written */
3463 hlen = INT_GET(rhead->h_len, ARCH_CONVERT);
3464 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3465 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3466 XFS_ERRLEVEL_LOW, log->l_mp);
3467 return XFS_ERROR(EFSCORRUPTED);
3469 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3470 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3471 XFS_ERRLEVEL_LOW, log->l_mp);
3472 return XFS_ERROR(EFSCORRUPTED);
3478 * Read the log from tail to head and process the log records found.
3479 * Handle the two cases where the tail and head are in the same cycle
3480 * and where the active portion of the log wraps around the end of
3481 * the physical log separately. The pass parameter is passed through
3482 * to the routines called to process the data and is not looked at
3486 xlog_do_recovery_pass(
3488 xfs_daddr_t head_blk,
3489 xfs_daddr_t tail_blk,
3492 xlog_rec_header_t *rhead;
3494 xfs_caddr_t bufaddr, offset;
3495 xfs_buf_t *hbp, *dbp;
3496 int error = 0, h_size;
3497 int bblks, split_bblks;
3498 int hblks, split_hblks, wrapped_hblks;
3499 xlog_recover_t *rhash[XLOG_RHASH_SIZE];
3501 ASSERT(head_blk != tail_blk);
3504 * Read the header of the tail block and get the iclog buffer size from
3505 * h_size. Use this to tell how many sectors make up the log header.
3507 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3509 * When using variable length iclogs, read first sector of
3510 * iclog header and extract the header size from it. Get a
3511 * new hbp that is the correct size.
3513 hbp = xlog_get_bp(log, 1);
3516 if ((error = xlog_bread(log, tail_blk, 1, hbp)))
3518 offset = xlog_align(log, tail_blk, 1, hbp);
3519 rhead = (xlog_rec_header_t *)offset;
3520 error = xlog_valid_rec_header(log, rhead, tail_blk);
3523 h_size = INT_GET(rhead->h_size, ARCH_CONVERT);
3524 if ((INT_GET(rhead->h_version, ARCH_CONVERT)
3525 & XLOG_VERSION_2) &&
3526 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3527 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3528 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3531 hbp = xlog_get_bp(log, hblks);
3536 ASSERT(log->l_sectbb_log == 0);
3538 hbp = xlog_get_bp(log, 1);
3539 h_size = XLOG_BIG_RECORD_BSIZE;
3544 dbp = xlog_get_bp(log, BTOBB(h_size));
3550 memset(rhash, 0, sizeof(rhash));
3551 if (tail_blk <= head_blk) {
3552 for (blk_no = tail_blk; blk_no < head_blk; ) {
3553 if ((error = xlog_bread(log, blk_no, hblks, hbp)))
3555 offset = xlog_align(log, blk_no, hblks, hbp);
3556 rhead = (xlog_rec_header_t *)offset;
3557 error = xlog_valid_rec_header(log, rhead, blk_no);
3561 /* blocks in data section */
3562 bblks = (int)BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT));
3563 error = xlog_bread(log, blk_no + hblks, bblks, dbp);
3566 offset = xlog_align(log, blk_no + hblks, bblks, dbp);
3567 xlog_unpack_data(rhead, offset, log);
3568 if ((error = xlog_recover_process_data(log,
3569 rhash, rhead, offset, pass)))
3571 blk_no += bblks + hblks;
3575 * Perform recovery around the end of the physical log.
3576 * When the head is not on the same cycle number as the tail,
3577 * we can't do a sequential recovery as above.
3580 while (blk_no < log->l_logBBsize) {
3582 * Check for header wrapping around physical end-of-log
3587 if (blk_no + hblks <= log->l_logBBsize) {
3588 /* Read header in one read */
3589 error = xlog_bread(log, blk_no, hblks, hbp);
3592 offset = xlog_align(log, blk_no, hblks, hbp);
3594 /* This LR is split across physical log end */
3595 if (blk_no != log->l_logBBsize) {
3596 /* some data before physical log end */
3597 ASSERT(blk_no <= INT_MAX);
3598 split_hblks = log->l_logBBsize - (int)blk_no;
3599 ASSERT(split_hblks > 0);
3600 if ((error = xlog_bread(log, blk_no,
3603 offset = xlog_align(log, blk_no,
3607 * Note: this black magic still works with
3608 * large sector sizes (non-512) only because:
3609 * - we increased the buffer size originally
3610 * by 1 sector giving us enough extra space
3611 * for the second read;
3612 * - the log start is guaranteed to be sector
3614 * - we read the log end (LR header start)
3615 * _first_, then the log start (LR header end)
3616 * - order is important.
3618 bufaddr = XFS_BUF_PTR(hbp);
3619 XFS_BUF_SET_PTR(hbp,
3620 bufaddr + BBTOB(split_hblks),
3621 BBTOB(hblks - split_hblks));
3622 wrapped_hblks = hblks - split_hblks;
3623 error = xlog_bread(log, 0, wrapped_hblks, hbp);
3626 XFS_BUF_SET_PTR(hbp, bufaddr, BBTOB(hblks));
3628 offset = xlog_align(log, 0,
3629 wrapped_hblks, hbp);
3631 rhead = (xlog_rec_header_t *)offset;
3632 error = xlog_valid_rec_header(log, rhead,
3633 split_hblks ? blk_no : 0);
3637 bblks = (int)BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT));
3640 /* Read in data for log record */
3641 if (blk_no + bblks <= log->l_logBBsize) {
3642 error = xlog_bread(log, blk_no, bblks, dbp);
3645 offset = xlog_align(log, blk_no, bblks, dbp);
3647 /* This log record is split across the
3648 * physical end of log */
3651 if (blk_no != log->l_logBBsize) {
3652 /* some data is before the physical
3654 ASSERT(!wrapped_hblks);
3655 ASSERT(blk_no <= INT_MAX);
3657 log->l_logBBsize - (int)blk_no;
3658 ASSERT(split_bblks > 0);
3659 if ((error = xlog_bread(log, blk_no,
3662 offset = xlog_align(log, blk_no,
3666 * Note: this black magic still works with
3667 * large sector sizes (non-512) only because:
3668 * - we increased the buffer size originally
3669 * by 1 sector giving us enough extra space
3670 * for the second read;
3671 * - the log start is guaranteed to be sector
3673 * - we read the log end (LR header start)
3674 * _first_, then the log start (LR header end)
3675 * - order is important.
3677 bufaddr = XFS_BUF_PTR(dbp);
3678 XFS_BUF_SET_PTR(dbp,
3679 bufaddr + BBTOB(split_bblks),
3680 BBTOB(bblks - split_bblks));
3681 if ((error = xlog_bread(log, wrapped_hblks,
3682 bblks - split_bblks, dbp)))
3684 XFS_BUF_SET_PTR(dbp, bufaddr, h_size);
3686 offset = xlog_align(log, wrapped_hblks,
3687 bblks - split_bblks, dbp);
3689 xlog_unpack_data(rhead, offset, log);
3690 if ((error = xlog_recover_process_data(log, rhash,
3691 rhead, offset, pass)))
3696 ASSERT(blk_no >= log->l_logBBsize);
3697 blk_no -= log->l_logBBsize;
3699 /* read first part of physical log */
3700 while (blk_no < head_blk) {
3701 if ((error = xlog_bread(log, blk_no, hblks, hbp)))
3703 offset = xlog_align(log, blk_no, hblks, hbp);
3704 rhead = (xlog_rec_header_t *)offset;
3705 error = xlog_valid_rec_header(log, rhead, blk_no);
3708 bblks = (int)BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT));
3709 if ((error = xlog_bread(log, blk_no+hblks, bblks, dbp)))
3711 offset = xlog_align(log, blk_no+hblks, bblks, dbp);
3712 xlog_unpack_data(rhead, offset, log);
3713 if ((error = xlog_recover_process_data(log, rhash,
3714 rhead, offset, pass)))
3716 blk_no += bblks + hblks;
3728 * Do the recovery of the log. We actually do this in two phases.
3729 * The two passes are necessary in order to implement the function
3730 * of cancelling a record written into the log. The first pass
3731 * determines those things which have been cancelled, and the
3732 * second pass replays log items normally except for those which
3733 * have been cancelled. The handling of the replay and cancellations
3734 * takes place in the log item type specific routines.
3736 * The table of items which have cancel records in the log is allocated
3737 * and freed at this level, since only here do we know when all of
3738 * the log recovery has been completed.
3741 xlog_do_log_recovery(
3743 xfs_daddr_t head_blk,
3744 xfs_daddr_t tail_blk)
3748 ASSERT(head_blk != tail_blk);
3751 * First do a pass to find all of the cancelled buf log items.
3752 * Store them in the buf_cancel_table for use in the second pass.
3754 log->l_buf_cancel_table =
3755 (xfs_buf_cancel_t **)kmem_zalloc(XLOG_BC_TABLE_SIZE *
3756 sizeof(xfs_buf_cancel_t*),
3758 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3759 XLOG_RECOVER_PASS1);
3761 kmem_free(log->l_buf_cancel_table,
3762 XLOG_BC_TABLE_SIZE * sizeof(xfs_buf_cancel_t*));
3763 log->l_buf_cancel_table = NULL;
3767 * Then do a second pass to actually recover the items in the log.
3768 * When it is complete free the table of buf cancel items.
3770 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3771 XLOG_RECOVER_PASS2);
3776 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3777 ASSERT(log->l_buf_cancel_table[i] == NULL);
3781 kmem_free(log->l_buf_cancel_table,
3782 XLOG_BC_TABLE_SIZE * sizeof(xfs_buf_cancel_t*));
3783 log->l_buf_cancel_table = NULL;
3789 * Do the actual recovery
3794 xfs_daddr_t head_blk,
3795 xfs_daddr_t tail_blk)
3802 * First replay the images in the log.
3804 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3809 XFS_bflush(log->l_mp->m_ddev_targp);
3812 * If IO errors happened during recovery, bail out.
3814 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3819 * We now update the tail_lsn since much of the recovery has completed
3820 * and there may be space available to use. If there were no extent
3821 * or iunlinks, we can free up the entire log and set the tail_lsn to
3822 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3823 * lsn of the last known good LR on disk. If there are extent frees
3824 * or iunlinks they will have some entries in the AIL; so we look at
3825 * the AIL to determine how to set the tail_lsn.
3827 xlog_assign_tail_lsn(log->l_mp);
3830 * Now that we've finished replaying all buffer and inode
3831 * updates, re-read in the superblock.
3833 bp = xfs_getsb(log->l_mp, 0);
3835 ASSERT(!(XFS_BUF_ISWRITE(bp)));
3836 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
3838 XFS_BUF_UNASYNC(bp);
3839 xfsbdstrat(log->l_mp, bp);
3840 if ((error = xfs_iowait(bp))) {
3841 xfs_ioerror_alert("xlog_do_recover",
3842 log->l_mp, bp, XFS_BUF_ADDR(bp));
3848 /* Convert superblock from on-disk format */
3849 sbp = &log->l_mp->m_sb;
3850 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
3851 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
3852 ASSERT(XFS_SB_GOOD_VERSION(sbp));
3855 /* We've re-read the superblock so re-initialize per-cpu counters */
3856 xfs_icsb_reinit_counters(log->l_mp);
3858 xlog_recover_check_summary(log);
3860 /* Normal transactions can now occur */
3861 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3866 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3868 * Return error or zero.
3874 xfs_daddr_t head_blk, tail_blk;
3877 /* find the tail of the log */
3878 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
3881 if (tail_blk != head_blk) {
3882 /* There used to be a comment here:
3884 * disallow recovery on read-only mounts. note -- mount
3885 * checks for ENOSPC and turns it into an intelligent
3887 * ...but this is no longer true. Now, unless you specify
3888 * NORECOVERY (in which case this function would never be
3889 * called), we just go ahead and recover. We do this all
3890 * under the vfs layer, so we can get away with it unless
3891 * the device itself is read-only, in which case we fail.
3893 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
3898 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3899 log->l_mp->m_fsname, log->l_mp->m_logname ?
3900 log->l_mp->m_logname : "internal");
3902 error = xlog_do_recover(log, head_blk, tail_blk);
3903 log->l_flags |= XLOG_RECOVERY_NEEDED;
3909 * In the first part of recovery we replay inodes and buffers and build
3910 * up the list of extent free items which need to be processed. Here
3911 * we process the extent free items and clean up the on disk unlinked
3912 * inode lists. This is separated from the first part of recovery so
3913 * that the root and real-time bitmap inodes can be read in from disk in
3914 * between the two stages. This is necessary so that we can free space
3915 * in the real-time portion of the file system.
3918 xlog_recover_finish(
3923 * Now we're ready to do the transactions needed for the
3924 * rest of recovery. Start with completing all the extent
3925 * free intent records and then process the unlinked inode
3926 * lists. At this point, we essentially run in normal mode
3927 * except that we're still performing recovery actions
3928 * rather than accepting new requests.
3930 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
3931 xlog_recover_process_efis(log);
3933 * Sync the log to get all the EFIs out of the AIL.
3934 * This isn't absolutely necessary, but it helps in
3935 * case the unlink transactions would have problems
3936 * pushing the EFIs out of the way.
3938 xfs_log_force(log->l_mp, (xfs_lsn_t)0,
3939 (XFS_LOG_FORCE | XFS_LOG_SYNC));
3941 if ( (mfsi_flags & XFS_MFSI_NOUNLINK) == 0 ) {
3942 xlog_recover_process_iunlinks(log);
3945 xlog_recover_check_summary(log);
3948 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3949 log->l_mp->m_fsname, log->l_mp->m_logname ?
3950 log->l_mp->m_logname : "internal");
3951 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
3954 "!Ending clean XFS mount for filesystem: %s\n",
3955 log->l_mp->m_fsname);
3963 * Read all of the agf and agi counters and check that they
3964 * are consistent with the superblock counters.
3967 xlog_recover_check_summary(
3975 xfs_daddr_t agfdaddr;
3976 xfs_daddr_t agidaddr;
3978 #ifdef XFS_LOUD_RECOVERY
3981 xfs_agnumber_t agno;
3982 __uint64_t freeblks;
3991 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3992 agfdaddr = XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp));
3993 agfbp = xfs_buf_read(mp->m_ddev_targp, agfdaddr,
3994 XFS_FSS_TO_BB(mp, 1), 0);
3995 if (XFS_BUF_ISERROR(agfbp)) {
3996 xfs_ioerror_alert("xlog_recover_check_summary(agf)",
3997 mp, agfbp, agfdaddr);
3999 agfp = XFS_BUF_TO_AGF(agfbp);
4000 ASSERT(XFS_AGF_MAGIC == be32_to_cpu(agfp->agf_magicnum));
4001 ASSERT(XFS_AGF_GOOD_VERSION(be32_to_cpu(agfp->agf_versionnum)));
4002 ASSERT(be32_to_cpu(agfp->agf_seqno) == agno);
4004 freeblks += be32_to_cpu(agfp->agf_freeblks) +
4005 be32_to_cpu(agfp->agf_flcount);
4006 xfs_buf_relse(agfbp);
4008 agidaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
4009 agibp = xfs_buf_read(mp->m_ddev_targp, agidaddr,
4010 XFS_FSS_TO_BB(mp, 1), 0);
4011 if (XFS_BUF_ISERROR(agibp)) {
4012 xfs_ioerror_alert("xlog_recover_check_summary(agi)",
4013 mp, agibp, agidaddr);
4015 agip = XFS_BUF_TO_AGI(agibp);
4016 ASSERT(XFS_AGI_MAGIC == be32_to_cpu(agip->agi_magicnum));
4017 ASSERT(XFS_AGI_GOOD_VERSION(be32_to_cpu(agip->agi_versionnum)));
4018 ASSERT(be32_to_cpu(agip->agi_seqno) == agno);
4020 itotal += be32_to_cpu(agip->agi_count);
4021 ifree += be32_to_cpu(agip->agi_freecount);
4022 xfs_buf_relse(agibp);
4025 sbbp = xfs_getsb(mp, 0);
4026 #ifdef XFS_LOUD_RECOVERY
4028 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(sbbp));
4030 "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
4031 sbp->sb_icount, itotal);
4033 "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
4034 sbp->sb_ifree, ifree);
4036 "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
4037 sbp->sb_fdblocks, freeblks);
4040 * This is turned off until I account for the allocation
4041 * btree blocks which live in free space.
4043 ASSERT(sbp->sb_icount == itotal);
4044 ASSERT(sbp->sb_ifree == ifree);
4045 ASSERT(sbp->sb_fdblocks == freeblks);
4048 xfs_buf_relse(sbbp);