2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_error.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_inode_item.h"
40 #include "xfs_alloc.h"
41 #include "xfs_ialloc.h"
42 #include "xfs_log_priv.h"
43 #include "xfs_buf_item.h"
44 #include "xfs_log_recover.h"
45 #include "xfs_extfree_item.h"
46 #include "xfs_trans_priv.h"
47 #include "xfs_quota.h"
49 #include "xfs_utils.h"
51 STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *);
52 STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t);
53 STATIC void xlog_recover_insert_item_backq(xlog_recover_item_t **q,
54 xlog_recover_item_t *item);
56 STATIC void xlog_recover_check_summary(xlog_t *);
57 STATIC void xlog_recover_check_ail(xfs_mount_t *, xfs_log_item_t *, int);
59 #define xlog_recover_check_summary(log)
60 #define xlog_recover_check_ail(mp, lip, gen)
65 * Sector aligned buffer routines for buffer create/read/write/access
68 #define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \
69 ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \
70 ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) )
71 #define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask)
78 ASSERT(num_bblks > 0);
80 if (log->l_sectbb_log) {
82 num_bblks += XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
83 num_bblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, num_bblks);
85 return xfs_buf_get_noaddr(BBTOB(num_bblks), log->l_mp->m_logdev_targp);
97 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
108 if (log->l_sectbb_log) {
109 blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
110 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
114 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
117 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
120 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
121 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
123 xfsbdstrat(log->l_mp, bp);
124 if ((error = xfs_iowait(bp)))
125 xfs_ioerror_alert("xlog_bread", log->l_mp,
126 bp, XFS_BUF_ADDR(bp));
131 * Write out the buffer at the given block for the given number of blocks.
132 * The buffer is kept locked across the write and is returned locked.
133 * This can only be used for synchronous log writes.
144 if (log->l_sectbb_log) {
145 blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
146 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
150 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
152 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
153 XFS_BUF_ZEROFLAGS(bp);
156 XFS_BUF_PSEMA(bp, PRIBIO);
157 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
158 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
160 if ((error = xfs_bwrite(log->l_mp, bp)))
161 xfs_ioerror_alert("xlog_bwrite", log->l_mp,
162 bp, XFS_BUF_ADDR(bp));
175 if (!log->l_sectbb_log)
176 return XFS_BUF_PTR(bp);
178 ptr = XFS_BUF_PTR(bp) + BBTOB((int)blk_no & log->l_sectbb_mask);
179 ASSERT(XFS_BUF_SIZE(bp) >=
180 BBTOB(nbblks + (blk_no & log->l_sectbb_mask)));
186 * dump debug superblock and log record information
189 xlog_header_check_dump(
191 xlog_rec_header_t *head)
195 cmn_err(CE_DEBUG, "%s: SB : uuid = ", __FUNCTION__);
196 for (b = 0; b < 16; b++)
197 cmn_err(CE_DEBUG, "%02x", ((uchar_t *)&mp->m_sb.sb_uuid)[b]);
198 cmn_err(CE_DEBUG, ", fmt = %d\n", XLOG_FMT);
199 cmn_err(CE_DEBUG, " log : uuid = ");
200 for (b = 0; b < 16; b++)
201 cmn_err(CE_DEBUG, "%02x",((uchar_t *)&head->h_fs_uuid)[b]);
202 cmn_err(CE_DEBUG, ", fmt = %d\n", be32_to_cpu(head->h_fmt));
205 #define xlog_header_check_dump(mp, head)
209 * check log record header for recovery
212 xlog_header_check_recover(
214 xlog_rec_header_t *head)
216 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
219 * IRIX doesn't write the h_fmt field and leaves it zeroed
220 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
221 * a dirty log created in IRIX.
223 if (unlikely(be32_to_cpu(head->h_fmt) != XLOG_FMT)) {
225 "XFS: dirty log written in incompatible format - can't recover");
226 xlog_header_check_dump(mp, head);
227 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
228 XFS_ERRLEVEL_HIGH, mp);
229 return XFS_ERROR(EFSCORRUPTED);
230 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
232 "XFS: dirty log entry has mismatched uuid - can't recover");
233 xlog_header_check_dump(mp, head);
234 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
235 XFS_ERRLEVEL_HIGH, mp);
236 return XFS_ERROR(EFSCORRUPTED);
242 * read the head block of the log and check the header
245 xlog_header_check_mount(
247 xlog_rec_header_t *head)
249 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
251 if (uuid_is_nil(&head->h_fs_uuid)) {
253 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
254 * h_fs_uuid is nil, we assume this log was last mounted
255 * by IRIX and continue.
257 xlog_warn("XFS: nil uuid in log - IRIX style log");
258 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
259 xlog_warn("XFS: log has mismatched uuid - can't recover");
260 xlog_header_check_dump(mp, head);
261 XFS_ERROR_REPORT("xlog_header_check_mount",
262 XFS_ERRLEVEL_HIGH, mp);
263 return XFS_ERROR(EFSCORRUPTED);
274 ASSERT(XFS_BUF_FSPRIVATE(bp, void *));
276 if (XFS_BUF_GETERROR(bp)) {
278 * We're not going to bother about retrying
279 * this during recovery. One strike!
281 mp = XFS_BUF_FSPRIVATE(bp, xfs_mount_t *);
282 xfs_ioerror_alert("xlog_recover_iodone",
283 mp, bp, XFS_BUF_ADDR(bp));
284 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
286 XFS_BUF_SET_FSPRIVATE(bp, NULL);
287 XFS_BUF_CLR_IODONE_FUNC(bp);
292 * This routine finds (to an approximation) the first block in the physical
293 * log which contains the given cycle. It uses a binary search algorithm.
294 * Note that the algorithm can not be perfect because the disk will not
295 * necessarily be perfect.
298 xlog_find_cycle_start(
301 xfs_daddr_t first_blk,
302 xfs_daddr_t *last_blk,
310 mid_blk = BLK_AVG(first_blk, *last_blk);
311 while (mid_blk != first_blk && mid_blk != *last_blk) {
312 if ((error = xlog_bread(log, mid_blk, 1, bp)))
314 offset = xlog_align(log, mid_blk, 1, bp);
315 mid_cycle = xlog_get_cycle(offset);
316 if (mid_cycle == cycle) {
318 /* last_half_cycle == mid_cycle */
321 /* first_half_cycle == mid_cycle */
323 mid_blk = BLK_AVG(first_blk, *last_blk);
325 ASSERT((mid_blk == first_blk && mid_blk+1 == *last_blk) ||
326 (mid_blk == *last_blk && mid_blk-1 == first_blk));
332 * Check that the range of blocks does not contain the cycle number
333 * given. The scan needs to occur from front to back and the ptr into the
334 * region must be updated since a later routine will need to perform another
335 * test. If the region is completely good, we end up returning the same
338 * Set blkno to -1 if we encounter no errors. This is an invalid block number
339 * since we don't ever expect logs to get this large.
342 xlog_find_verify_cycle(
344 xfs_daddr_t start_blk,
346 uint stop_on_cycle_no,
347 xfs_daddr_t *new_blk)
353 xfs_caddr_t buf = NULL;
356 bufblks = 1 << ffs(nbblks);
358 while (!(bp = xlog_get_bp(log, bufblks))) {
359 /* can't get enough memory to do everything in one big buffer */
361 if (bufblks <= log->l_sectbb_log)
365 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
368 bcount = min(bufblks, (start_blk + nbblks - i));
370 if ((error = xlog_bread(log, i, bcount, bp)))
373 buf = xlog_align(log, i, bcount, bp);
374 for (j = 0; j < bcount; j++) {
375 cycle = xlog_get_cycle(buf);
376 if (cycle == stop_on_cycle_no) {
393 * Potentially backup over partial log record write.
395 * In the typical case, last_blk is the number of the block directly after
396 * a good log record. Therefore, we subtract one to get the block number
397 * of the last block in the given buffer. extra_bblks contains the number
398 * of blocks we would have read on a previous read. This happens when the
399 * last log record is split over the end of the physical log.
401 * extra_bblks is the number of blocks potentially verified on a previous
402 * call to this routine.
405 xlog_find_verify_log_record(
407 xfs_daddr_t start_blk,
408 xfs_daddr_t *last_blk,
413 xfs_caddr_t offset = NULL;
414 xlog_rec_header_t *head = NULL;
417 int num_blks = *last_blk - start_blk;
420 ASSERT(start_blk != 0 || *last_blk != start_blk);
422 if (!(bp = xlog_get_bp(log, num_blks))) {
423 if (!(bp = xlog_get_bp(log, 1)))
427 if ((error = xlog_bread(log, start_blk, num_blks, bp)))
429 offset = xlog_align(log, start_blk, num_blks, bp);
430 offset += ((num_blks - 1) << BBSHIFT);
433 for (i = (*last_blk) - 1; i >= 0; i--) {
435 /* valid log record not found */
437 "XFS: Log inconsistent (didn't find previous header)");
439 error = XFS_ERROR(EIO);
444 if ((error = xlog_bread(log, i, 1, bp)))
446 offset = xlog_align(log, i, 1, bp);
449 head = (xlog_rec_header_t *)offset;
451 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(head->h_magicno))
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 = be32_to_cpu(head->h_size);
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(be32_to_cpu(head->h_len)) + 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 == be32_to_cpu(*(__be32 *)offset)) {
832 * If we haven't found the log record header block, start looking
833 * again from the end of the physical log. XXXmiken: There should be
834 * a check here to make sure we didn't search more than N blocks in
838 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
839 if ((error = xlog_bread(log, i, 1, bp)))
841 offset = xlog_align(log, i, 1, bp);
842 if (XLOG_HEADER_MAGIC_NUM ==
843 be32_to_cpu(*(__be32 *)offset)) {
850 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
852 return XFS_ERROR(EIO);
855 /* find blk_no of tail of log */
856 rhead = (xlog_rec_header_t *)offset;
857 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
860 * Reset log values according to the state of the log when we
861 * crashed. In the case where head_blk == 0, we bump curr_cycle
862 * one because the next write starts a new cycle rather than
863 * continuing the cycle of the last good log record. At this
864 * point we have guaranteed that all partial log records have been
865 * accounted for. Therefore, we know that the last good log record
866 * written was complete and ended exactly on the end boundary
867 * of the physical log.
869 log->l_prev_block = i;
870 log->l_curr_block = (int)*head_blk;
871 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
874 log->l_tail_lsn = be64_to_cpu(rhead->h_tail_lsn);
875 log->l_last_sync_lsn = be64_to_cpu(rhead->h_lsn);
876 log->l_grant_reserve_cycle = log->l_curr_cycle;
877 log->l_grant_reserve_bytes = BBTOB(log->l_curr_block);
878 log->l_grant_write_cycle = log->l_curr_cycle;
879 log->l_grant_write_bytes = BBTOB(log->l_curr_block);
882 * Look for unmount record. If we find it, then we know there
883 * was a clean unmount. Since 'i' could be the last block in
884 * the physical log, we convert to a log block before comparing
887 * Save the current tail lsn to use to pass to
888 * xlog_clear_stale_blocks() below. We won't want to clear the
889 * unmount record if there is one, so we pass the lsn of the
890 * unmount record rather than the block after it.
892 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
893 int h_size = be32_to_cpu(rhead->h_size);
894 int h_version = be32_to_cpu(rhead->h_version);
896 if ((h_version & XLOG_VERSION_2) &&
897 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
898 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
899 if (h_size % XLOG_HEADER_CYCLE_SIZE)
907 after_umount_blk = (i + hblks + (int)
908 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
909 tail_lsn = log->l_tail_lsn;
910 if (*head_blk == after_umount_blk &&
911 be32_to_cpu(rhead->h_num_logops) == 1) {
912 umount_data_blk = (i + hblks) % log->l_logBBsize;
913 if ((error = xlog_bread(log, umount_data_blk, 1, bp))) {
916 offset = xlog_align(log, umount_data_blk, 1, bp);
917 op_head = (xlog_op_header_t *)offset;
918 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
920 * Set tail and last sync so that newly written
921 * log records will point recovery to after the
922 * current unmount record.
925 xlog_assign_lsn(log->l_curr_cycle,
927 log->l_last_sync_lsn =
928 xlog_assign_lsn(log->l_curr_cycle,
930 *tail_blk = after_umount_blk;
933 * Note that the unmount was clean. If the unmount
934 * was not clean, we need to know this to rebuild the
935 * superblock counters from the perag headers if we
936 * have a filesystem using non-persistent counters.
938 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
943 * Make sure that there are no blocks in front of the head
944 * with the same cycle number as the head. This can happen
945 * because we allow multiple outstanding log writes concurrently,
946 * and the later writes might make it out before earlier ones.
948 * We use the lsn from before modifying it so that we'll never
949 * overwrite the unmount record after a clean unmount.
951 * Do this only if we are going to recover the filesystem
953 * NOTE: This used to say "if (!readonly)"
954 * However on Linux, we can & do recover a read-only filesystem.
955 * We only skip recovery if NORECOVERY is specified on mount,
956 * in which case we would not be here.
958 * But... if the -device- itself is readonly, just skip this.
959 * We can't recover this device anyway, so it won't matter.
961 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) {
962 error = xlog_clear_stale_blocks(log, tail_lsn);
970 xlog_warn("XFS: failed to locate log tail");
975 * Is the log zeroed at all?
977 * The last binary search should be changed to perform an X block read
978 * once X becomes small enough. You can then search linearly through
979 * the X blocks. This will cut down on the number of reads we need to do.
981 * If the log is partially zeroed, this routine will pass back the blkno
982 * of the first block with cycle number 0. It won't have a complete LR
986 * 0 => the log is completely written to
987 * -1 => use *blk_no as the first block of the log
988 * >0 => error has occurred
997 uint first_cycle, last_cycle;
998 xfs_daddr_t new_blk, last_blk, start_blk;
999 xfs_daddr_t num_scan_bblks;
1000 int error, log_bbnum = log->l_logBBsize;
1004 /* check totally zeroed log */
1005 bp = xlog_get_bp(log, 1);
1008 if ((error = xlog_bread(log, 0, 1, bp)))
1010 offset = xlog_align(log, 0, 1, bp);
1011 first_cycle = xlog_get_cycle(offset);
1012 if (first_cycle == 0) { /* completely zeroed log */
1018 /* check partially zeroed log */
1019 if ((error = xlog_bread(log, log_bbnum-1, 1, bp)))
1021 offset = xlog_align(log, log_bbnum-1, 1, bp);
1022 last_cycle = xlog_get_cycle(offset);
1023 if (last_cycle != 0) { /* log completely written to */
1026 } else if (first_cycle != 1) {
1028 * If the cycle of the last block is zero, the cycle of
1029 * the first block must be 1. If it's not, maybe we're
1030 * not looking at a log... Bail out.
1032 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1033 return XFS_ERROR(EINVAL);
1036 /* we have a partially zeroed log */
1037 last_blk = log_bbnum-1;
1038 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1042 * Validate the answer. Because there is no way to guarantee that
1043 * the entire log is made up of log records which are the same size,
1044 * we scan over the defined maximum blocks. At this point, the maximum
1045 * is not chosen to mean anything special. XXXmiken
1047 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1048 ASSERT(num_scan_bblks <= INT_MAX);
1050 if (last_blk < num_scan_bblks)
1051 num_scan_bblks = last_blk;
1052 start_blk = last_blk - num_scan_bblks;
1055 * We search for any instances of cycle number 0 that occur before
1056 * our current estimate of the head. What we're trying to detect is
1057 * 1 ... | 0 | 1 | 0...
1058 * ^ binary search ends here
1060 if ((error = xlog_find_verify_cycle(log, start_blk,
1061 (int)num_scan_bblks, 0, &new_blk)))
1067 * Potentially backup over partial log record write. We don't need
1068 * to search the end of the log because we know it is zero.
1070 if ((error = xlog_find_verify_log_record(log, start_blk,
1071 &last_blk, 0)) == -1) {
1072 error = XFS_ERROR(EIO);
1086 * These are simple subroutines used by xlog_clear_stale_blocks() below
1087 * to initialize a buffer full of empty log record headers and write
1088 * them into the log.
1099 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1101 memset(buf, 0, BBSIZE);
1102 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1103 recp->h_cycle = cpu_to_be32(cycle);
1104 recp->h_version = cpu_to_be32(
1105 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1106 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1107 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1108 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1109 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1113 xlog_write_log_records(
1124 int sectbb = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
1125 int end_block = start_block + blocks;
1130 bufblks = 1 << ffs(blocks);
1131 while (!(bp = xlog_get_bp(log, bufblks))) {
1133 if (bufblks <= log->l_sectbb_log)
1137 /* We may need to do a read at the start to fill in part of
1138 * the buffer in the starting sector not covered by the first
1141 balign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, start_block);
1142 if (balign != start_block) {
1143 if ((error = xlog_bread(log, start_block, 1, bp))) {
1147 j = start_block - balign;
1150 for (i = start_block; i < end_block; i += bufblks) {
1151 int bcount, endcount;
1153 bcount = min(bufblks, end_block - start_block);
1154 endcount = bcount - j;
1156 /* We may need to do a read at the end to fill in part of
1157 * the buffer in the final sector not covered by the write.
1158 * If this is the same sector as the above read, skip it.
1160 ealign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, end_block);
1161 if (j == 0 && (start_block + endcount > ealign)) {
1162 offset = XFS_BUF_PTR(bp);
1163 balign = BBTOB(ealign - start_block);
1164 XFS_BUF_SET_PTR(bp, offset + balign, BBTOB(sectbb));
1165 if ((error = xlog_bread(log, ealign, sectbb, bp)))
1167 XFS_BUF_SET_PTR(bp, offset, bufblks);
1170 offset = xlog_align(log, start_block, endcount, bp);
1171 for (; j < endcount; j++) {
1172 xlog_add_record(log, offset, cycle, i+j,
1173 tail_cycle, tail_block);
1176 error = xlog_bwrite(log, start_block, endcount, bp);
1179 start_block += endcount;
1187 * This routine is called to blow away any incomplete log writes out
1188 * in front of the log head. We do this so that we won't become confused
1189 * if we come up, write only a little bit more, and then crash again.
1190 * If we leave the partial log records out there, this situation could
1191 * cause us to think those partial writes are valid blocks since they
1192 * have the current cycle number. We get rid of them by overwriting them
1193 * with empty log records with the old cycle number rather than the
1196 * The tail lsn is passed in rather than taken from
1197 * the log so that we will not write over the unmount record after a
1198 * clean unmount in a 512 block log. Doing so would leave the log without
1199 * any valid log records in it until a new one was written. If we crashed
1200 * during that time we would not be able to recover.
1203 xlog_clear_stale_blocks(
1207 int tail_cycle, head_cycle;
1208 int tail_block, head_block;
1209 int tail_distance, max_distance;
1213 tail_cycle = CYCLE_LSN(tail_lsn);
1214 tail_block = BLOCK_LSN(tail_lsn);
1215 head_cycle = log->l_curr_cycle;
1216 head_block = log->l_curr_block;
1219 * Figure out the distance between the new head of the log
1220 * and the tail. We want to write over any blocks beyond the
1221 * head that we may have written just before the crash, but
1222 * we don't want to overwrite the tail of the log.
1224 if (head_cycle == tail_cycle) {
1226 * The tail is behind the head in the physical log,
1227 * so the distance from the head to the tail is the
1228 * distance from the head to the end of the log plus
1229 * the distance from the beginning of the log to the
1232 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1233 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1234 XFS_ERRLEVEL_LOW, log->l_mp);
1235 return XFS_ERROR(EFSCORRUPTED);
1237 tail_distance = tail_block + (log->l_logBBsize - head_block);
1240 * The head is behind the tail in the physical log,
1241 * so the distance from the head to the tail is just
1242 * the tail block minus the head block.
1244 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1245 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1246 XFS_ERRLEVEL_LOW, log->l_mp);
1247 return XFS_ERROR(EFSCORRUPTED);
1249 tail_distance = tail_block - head_block;
1253 * If the head is right up against the tail, we can't clear
1256 if (tail_distance <= 0) {
1257 ASSERT(tail_distance == 0);
1261 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1263 * Take the smaller of the maximum amount of outstanding I/O
1264 * we could have and the distance to the tail to clear out.
1265 * We take the smaller so that we don't overwrite the tail and
1266 * we don't waste all day writing from the head to the tail
1269 max_distance = MIN(max_distance, tail_distance);
1271 if ((head_block + max_distance) <= log->l_logBBsize) {
1273 * We can stomp all the blocks we need to without
1274 * wrapping around the end of the log. Just do it
1275 * in a single write. Use the cycle number of the
1276 * current cycle minus one so that the log will look like:
1279 error = xlog_write_log_records(log, (head_cycle - 1),
1280 head_block, max_distance, tail_cycle,
1286 * We need to wrap around the end of the physical log in
1287 * order to clear all the blocks. Do it in two separate
1288 * I/Os. The first write should be from the head to the
1289 * end of the physical log, and it should use the current
1290 * cycle number minus one just like above.
1292 distance = log->l_logBBsize - head_block;
1293 error = xlog_write_log_records(log, (head_cycle - 1),
1294 head_block, distance, tail_cycle,
1301 * Now write the blocks at the start of the physical log.
1302 * This writes the remainder of the blocks we want to clear.
1303 * It uses the current cycle number since we're now on the
1304 * same cycle as the head so that we get:
1305 * n ... n ... | n - 1 ...
1306 * ^^^^^ blocks we're writing
1308 distance = max_distance - (log->l_logBBsize - head_block);
1309 error = xlog_write_log_records(log, head_cycle, 0, distance,
1310 tail_cycle, tail_block);
1318 /******************************************************************************
1320 * Log recover routines
1322 ******************************************************************************
1325 STATIC xlog_recover_t *
1326 xlog_recover_find_tid(
1330 xlog_recover_t *p = q;
1333 if (p->r_log_tid == tid)
1341 xlog_recover_put_hashq(
1343 xlog_recover_t *trans)
1350 xlog_recover_add_item(
1351 xlog_recover_item_t **itemq)
1353 xlog_recover_item_t *item;
1355 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1356 xlog_recover_insert_item_backq(itemq, item);
1360 xlog_recover_add_to_cont_trans(
1361 xlog_recover_t *trans,
1365 xlog_recover_item_t *item;
1366 xfs_caddr_t ptr, old_ptr;
1369 item = trans->r_itemq;
1371 /* finish copying rest of trans header */
1372 xlog_recover_add_item(&trans->r_itemq);
1373 ptr = (xfs_caddr_t) &trans->r_theader +
1374 sizeof(xfs_trans_header_t) - len;
1375 memcpy(ptr, dp, len); /* d, s, l */
1378 item = item->ri_prev;
1380 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1381 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1383 ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
1384 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1385 item->ri_buf[item->ri_cnt-1].i_len += len;
1386 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1391 * The next region to add is the start of a new region. It could be
1392 * a whole region or it could be the first part of a new region. Because
1393 * of this, the assumption here is that the type and size fields of all
1394 * format structures fit into the first 32 bits of the structure.
1396 * This works because all regions must be 32 bit aligned. Therefore, we
1397 * either have both fields or we have neither field. In the case we have
1398 * neither field, the data part of the region is zero length. We only have
1399 * a log_op_header and can throw away the header since a new one will appear
1400 * later. If we have at least 4 bytes, then we can determine how many regions
1401 * will appear in the current log item.
1404 xlog_recover_add_to_trans(
1405 xlog_recover_t *trans,
1409 xfs_inode_log_format_t *in_f; /* any will do */
1410 xlog_recover_item_t *item;
1415 item = trans->r_itemq;
1417 ASSERT(*(uint *)dp == XFS_TRANS_HEADER_MAGIC);
1418 if (len == sizeof(xfs_trans_header_t))
1419 xlog_recover_add_item(&trans->r_itemq);
1420 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1424 ptr = kmem_alloc(len, KM_SLEEP);
1425 memcpy(ptr, dp, len);
1426 in_f = (xfs_inode_log_format_t *)ptr;
1428 if (item->ri_prev->ri_total != 0 &&
1429 item->ri_prev->ri_total == item->ri_prev->ri_cnt) {
1430 xlog_recover_add_item(&trans->r_itemq);
1432 item = trans->r_itemq;
1433 item = item->ri_prev;
1435 if (item->ri_total == 0) { /* first region to be added */
1436 item->ri_total = in_f->ilf_size;
1437 ASSERT(item->ri_total <= XLOG_MAX_REGIONS_IN_ITEM);
1438 item->ri_buf = kmem_zalloc((item->ri_total *
1439 sizeof(xfs_log_iovec_t)), KM_SLEEP);
1441 ASSERT(item->ri_total > item->ri_cnt);
1442 /* Description region is ri_buf[0] */
1443 item->ri_buf[item->ri_cnt].i_addr = ptr;
1444 item->ri_buf[item->ri_cnt].i_len = len;
1450 xlog_recover_new_tid(
1455 xlog_recover_t *trans;
1457 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1458 trans->r_log_tid = tid;
1460 xlog_recover_put_hashq(q, trans);
1464 xlog_recover_unlink_tid(
1466 xlog_recover_t *trans)
1471 ASSERT(trans != NULL);
1477 if (tp->r_next == trans) {
1485 "XFS: xlog_recover_unlink_tid: trans not found");
1487 return XFS_ERROR(EIO);
1489 tp->r_next = tp->r_next->r_next;
1495 xlog_recover_insert_item_backq(
1496 xlog_recover_item_t **q,
1497 xlog_recover_item_t *item)
1500 item->ri_prev = item->ri_next = item;
1504 item->ri_prev = (*q)->ri_prev;
1505 (*q)->ri_prev = item;
1506 item->ri_prev->ri_next = item;
1511 xlog_recover_insert_item_frontq(
1512 xlog_recover_item_t **q,
1513 xlog_recover_item_t *item)
1515 xlog_recover_insert_item_backq(q, item);
1520 xlog_recover_reorder_trans(
1521 xlog_recover_t *trans)
1523 xlog_recover_item_t *first_item, *itemq, *itemq_next;
1524 xfs_buf_log_format_t *buf_f;
1527 first_item = itemq = trans->r_itemq;
1528 trans->r_itemq = NULL;
1530 itemq_next = itemq->ri_next;
1531 buf_f = (xfs_buf_log_format_t *)itemq->ri_buf[0].i_addr;
1533 switch (ITEM_TYPE(itemq)) {
1535 flags = buf_f->blf_flags;
1536 if (!(flags & XFS_BLI_CANCEL)) {
1537 xlog_recover_insert_item_frontq(&trans->r_itemq,
1543 case XFS_LI_QUOTAOFF:
1546 xlog_recover_insert_item_backq(&trans->r_itemq, itemq);
1550 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1552 return XFS_ERROR(EIO);
1555 } while (first_item != itemq);
1560 * Build up the table of buf cancel records so that we don't replay
1561 * cancelled data in the second pass. For buffer records that are
1562 * not cancel records, there is nothing to do here so we just return.
1564 * If we get a cancel record which is already in the table, this indicates
1565 * that the buffer was cancelled multiple times. In order to ensure
1566 * that during pass 2 we keep the record in the table until we reach its
1567 * last occurrence in the log, we keep a reference count in the cancel
1568 * record in the table to tell us how many times we expect to see this
1569 * record during the second pass.
1572 xlog_recover_do_buffer_pass1(
1574 xfs_buf_log_format_t *buf_f)
1576 xfs_buf_cancel_t *bcp;
1577 xfs_buf_cancel_t *nextp;
1578 xfs_buf_cancel_t *prevp;
1579 xfs_buf_cancel_t **bucket;
1580 xfs_daddr_t blkno = 0;
1584 switch (buf_f->blf_type) {
1586 blkno = buf_f->blf_blkno;
1587 len = buf_f->blf_len;
1588 flags = buf_f->blf_flags;
1593 * If this isn't a cancel buffer item, then just return.
1595 if (!(flags & XFS_BLI_CANCEL))
1599 * Insert an xfs_buf_cancel record into the hash table of
1600 * them. If there is already an identical record, bump
1601 * its reference count.
1603 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1604 XLOG_BC_TABLE_SIZE];
1606 * If the hash bucket is empty then just insert a new record into
1609 if (*bucket == NULL) {
1610 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1612 bcp->bc_blkno = blkno;
1614 bcp->bc_refcount = 1;
1615 bcp->bc_next = NULL;
1621 * The hash bucket is not empty, so search for duplicates of our
1622 * record. If we find one them just bump its refcount. If not
1623 * then add us at the end of the list.
1627 while (nextp != NULL) {
1628 if (nextp->bc_blkno == blkno && nextp->bc_len == len) {
1629 nextp->bc_refcount++;
1633 nextp = nextp->bc_next;
1635 ASSERT(prevp != NULL);
1636 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1638 bcp->bc_blkno = blkno;
1640 bcp->bc_refcount = 1;
1641 bcp->bc_next = NULL;
1642 prevp->bc_next = bcp;
1646 * Check to see whether the buffer being recovered has a corresponding
1647 * entry in the buffer cancel record table. If it does then return 1
1648 * so that it will be cancelled, otherwise return 0. If the buffer is
1649 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
1650 * the refcount on the entry in the table and remove it from the table
1651 * if this is the last reference.
1653 * We remove the cancel record from the table when we encounter its
1654 * last occurrence in the log so that if the same buffer is re-used
1655 * again after its last cancellation we actually replay the changes
1656 * made at that point.
1659 xlog_check_buffer_cancelled(
1665 xfs_buf_cancel_t *bcp;
1666 xfs_buf_cancel_t *prevp;
1667 xfs_buf_cancel_t **bucket;
1669 if (log->l_buf_cancel_table == NULL) {
1671 * There is nothing in the table built in pass one,
1672 * so this buffer must not be cancelled.
1674 ASSERT(!(flags & XFS_BLI_CANCEL));
1678 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1679 XLOG_BC_TABLE_SIZE];
1683 * There is no corresponding entry in the table built
1684 * in pass one, so this buffer has not been cancelled.
1686 ASSERT(!(flags & XFS_BLI_CANCEL));
1691 * Search for an entry in the buffer cancel table that
1692 * matches our buffer.
1695 while (bcp != NULL) {
1696 if (bcp->bc_blkno == blkno && bcp->bc_len == len) {
1698 * We've go a match, so return 1 so that the
1699 * recovery of this buffer is cancelled.
1700 * If this buffer is actually a buffer cancel
1701 * log item, then decrement the refcount on the
1702 * one in the table and remove it if this is the
1705 if (flags & XFS_BLI_CANCEL) {
1707 if (bcp->bc_refcount == 0) {
1708 if (prevp == NULL) {
1709 *bucket = bcp->bc_next;
1711 prevp->bc_next = bcp->bc_next;
1714 sizeof(xfs_buf_cancel_t));
1723 * We didn't find a corresponding entry in the table, so
1724 * return 0 so that the buffer is NOT cancelled.
1726 ASSERT(!(flags & XFS_BLI_CANCEL));
1731 xlog_recover_do_buffer_pass2(
1733 xfs_buf_log_format_t *buf_f)
1735 xfs_daddr_t blkno = 0;
1739 switch (buf_f->blf_type) {
1741 blkno = buf_f->blf_blkno;
1742 flags = buf_f->blf_flags;
1743 len = buf_f->blf_len;
1747 return xlog_check_buffer_cancelled(log, blkno, len, flags);
1751 * Perform recovery for a buffer full of inodes. In these buffers,
1752 * the only data which should be recovered is that which corresponds
1753 * to the di_next_unlinked pointers in the on disk inode structures.
1754 * The rest of the data for the inodes is always logged through the
1755 * inodes themselves rather than the inode buffer and is recovered
1756 * in xlog_recover_do_inode_trans().
1758 * The only time when buffers full of inodes are fully recovered is
1759 * when the buffer is full of newly allocated inodes. In this case
1760 * the buffer will not be marked as an inode buffer and so will be
1761 * sent to xlog_recover_do_reg_buffer() below during recovery.
1764 xlog_recover_do_inode_buffer(
1766 xlog_recover_item_t *item,
1768 xfs_buf_log_format_t *buf_f)
1776 int next_unlinked_offset;
1778 xfs_agino_t *logged_nextp;
1779 xfs_agino_t *buffer_nextp;
1780 unsigned int *data_map = NULL;
1781 unsigned int map_size = 0;
1783 switch (buf_f->blf_type) {
1785 data_map = buf_f->blf_data_map;
1786 map_size = buf_f->blf_map_size;
1790 * Set the variables corresponding to the current region to
1791 * 0 so that we'll initialize them on the first pass through
1799 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1800 for (i = 0; i < inodes_per_buf; i++) {
1801 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1802 offsetof(xfs_dinode_t, di_next_unlinked);
1804 while (next_unlinked_offset >=
1805 (reg_buf_offset + reg_buf_bytes)) {
1807 * The next di_next_unlinked field is beyond
1808 * the current logged region. Find the next
1809 * logged region that contains or is beyond
1810 * the current di_next_unlinked field.
1813 bit = xfs_next_bit(data_map, map_size, bit);
1816 * If there are no more logged regions in the
1817 * buffer, then we're done.
1823 nbits = xfs_contig_bits(data_map, map_size,
1826 reg_buf_offset = bit << XFS_BLI_SHIFT;
1827 reg_buf_bytes = nbits << XFS_BLI_SHIFT;
1832 * If the current logged region starts after the current
1833 * di_next_unlinked field, then move on to the next
1834 * di_next_unlinked field.
1836 if (next_unlinked_offset < reg_buf_offset) {
1840 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1841 ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0);
1842 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1845 * The current logged region contains a copy of the
1846 * current di_next_unlinked field. Extract its value
1847 * and copy it to the buffer copy.
1849 logged_nextp = (xfs_agino_t *)
1850 ((char *)(item->ri_buf[item_index].i_addr) +
1851 (next_unlinked_offset - reg_buf_offset));
1852 if (unlikely(*logged_nextp == 0)) {
1853 xfs_fs_cmn_err(CE_ALERT, mp,
1854 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1856 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1857 XFS_ERRLEVEL_LOW, mp);
1858 return XFS_ERROR(EFSCORRUPTED);
1861 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1862 next_unlinked_offset);
1863 *buffer_nextp = *logged_nextp;
1870 * Perform a 'normal' buffer recovery. Each logged region of the
1871 * buffer should be copied over the corresponding region in the
1872 * given buffer. The bitmap in the buf log format structure indicates
1873 * where to place the logged data.
1877 xlog_recover_do_reg_buffer(
1878 xlog_recover_item_t *item,
1880 xfs_buf_log_format_t *buf_f)
1885 unsigned int *data_map = NULL;
1886 unsigned int map_size = 0;
1889 switch (buf_f->blf_type) {
1891 data_map = buf_f->blf_data_map;
1892 map_size = buf_f->blf_map_size;
1896 i = 1; /* 0 is the buf format structure */
1898 bit = xfs_next_bit(data_map, map_size, bit);
1901 nbits = xfs_contig_bits(data_map, map_size, bit);
1903 ASSERT(item->ri_buf[i].i_addr != NULL);
1904 ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0);
1905 ASSERT(XFS_BUF_COUNT(bp) >=
1906 ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT));
1909 * Do a sanity check if this is a dquot buffer. Just checking
1910 * the first dquot in the buffer should do. XXXThis is
1911 * probably a good thing to do for other buf types also.
1914 if (buf_f->blf_flags &
1915 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
1916 error = xfs_qm_dqcheck((xfs_disk_dquot_t *)
1917 item->ri_buf[i].i_addr,
1918 -1, 0, XFS_QMOPT_DOWARN,
1919 "dquot_buf_recover");
1922 memcpy(xfs_buf_offset(bp,
1923 (uint)bit << XFS_BLI_SHIFT), /* dest */
1924 item->ri_buf[i].i_addr, /* source */
1925 nbits<<XFS_BLI_SHIFT); /* length */
1930 /* Shouldn't be any more regions */
1931 ASSERT(i == item->ri_total);
1935 * Do some primitive error checking on ondisk dquot data structures.
1939 xfs_disk_dquot_t *ddq,
1941 uint type, /* used only when IO_dorepair is true */
1945 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
1949 * We can encounter an uninitialized dquot buffer for 2 reasons:
1950 * 1. If we crash while deleting the quotainode(s), and those blks got
1951 * used for user data. This is because we take the path of regular
1952 * file deletion; however, the size field of quotainodes is never
1953 * updated, so all the tricks that we play in itruncate_finish
1954 * don't quite matter.
1956 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1957 * But the allocation will be replayed so we'll end up with an
1958 * uninitialized quota block.
1960 * This is all fine; things are still consistent, and we haven't lost
1961 * any quota information. Just don't complain about bad dquot blks.
1963 if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) {
1964 if (flags & XFS_QMOPT_DOWARN)
1966 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
1967 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
1970 if (ddq->d_version != XFS_DQUOT_VERSION) {
1971 if (flags & XFS_QMOPT_DOWARN)
1973 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
1974 str, id, ddq->d_version, XFS_DQUOT_VERSION);
1978 if (ddq->d_flags != XFS_DQ_USER &&
1979 ddq->d_flags != XFS_DQ_PROJ &&
1980 ddq->d_flags != XFS_DQ_GROUP) {
1981 if (flags & XFS_QMOPT_DOWARN)
1983 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
1984 str, id, ddq->d_flags);
1988 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
1989 if (flags & XFS_QMOPT_DOWARN)
1991 "%s : ondisk-dquot 0x%p, ID mismatch: "
1992 "0x%x expected, found id 0x%x",
1993 str, ddq, id, be32_to_cpu(ddq->d_id));
1997 if (!errs && ddq->d_id) {
1998 if (ddq->d_blk_softlimit &&
1999 be64_to_cpu(ddq->d_bcount) >=
2000 be64_to_cpu(ddq->d_blk_softlimit)) {
2001 if (!ddq->d_btimer) {
2002 if (flags & XFS_QMOPT_DOWARN)
2004 "%s : Dquot ID 0x%x (0x%p) "
2005 "BLK TIMER NOT STARTED",
2006 str, (int)be32_to_cpu(ddq->d_id), ddq);
2010 if (ddq->d_ino_softlimit &&
2011 be64_to_cpu(ddq->d_icount) >=
2012 be64_to_cpu(ddq->d_ino_softlimit)) {
2013 if (!ddq->d_itimer) {
2014 if (flags & XFS_QMOPT_DOWARN)
2016 "%s : Dquot ID 0x%x (0x%p) "
2017 "INODE TIMER NOT STARTED",
2018 str, (int)be32_to_cpu(ddq->d_id), ddq);
2022 if (ddq->d_rtb_softlimit &&
2023 be64_to_cpu(ddq->d_rtbcount) >=
2024 be64_to_cpu(ddq->d_rtb_softlimit)) {
2025 if (!ddq->d_rtbtimer) {
2026 if (flags & XFS_QMOPT_DOWARN)
2028 "%s : Dquot ID 0x%x (0x%p) "
2029 "RTBLK TIMER NOT STARTED",
2030 str, (int)be32_to_cpu(ddq->d_id), ddq);
2036 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2039 if (flags & XFS_QMOPT_DOWARN)
2040 cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id);
2043 * Typically, a repair is only requested by quotacheck.
2046 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2047 memset(d, 0, sizeof(xfs_dqblk_t));
2049 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2050 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2051 d->dd_diskdq.d_flags = type;
2052 d->dd_diskdq.d_id = cpu_to_be32(id);
2058 * Perform a dquot buffer recovery.
2059 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2060 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2061 * Else, treat it as a regular buffer and do recovery.
2064 xlog_recover_do_dquot_buffer(
2067 xlog_recover_item_t *item,
2069 xfs_buf_log_format_t *buf_f)
2074 * Filesystems are required to send in quota flags at mount time.
2076 if (mp->m_qflags == 0) {
2081 if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF)
2082 type |= XFS_DQ_USER;
2083 if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF)
2084 type |= XFS_DQ_PROJ;
2085 if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF)
2086 type |= XFS_DQ_GROUP;
2088 * This type of quotas was turned off, so ignore this buffer
2090 if (log->l_quotaoffs_flag & type)
2093 xlog_recover_do_reg_buffer(item, bp, buf_f);
2097 * This routine replays a modification made to a buffer at runtime.
2098 * There are actually two types of buffer, regular and inode, which
2099 * are handled differently. Inode buffers are handled differently
2100 * in that we only recover a specific set of data from them, namely
2101 * the inode di_next_unlinked fields. This is because all other inode
2102 * data is actually logged via inode records and any data we replay
2103 * here which overlaps that may be stale.
2105 * When meta-data buffers are freed at run time we log a buffer item
2106 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
2107 * of the buffer in the log should not be replayed at recovery time.
2108 * This is so that if the blocks covered by the buffer are reused for
2109 * file data before we crash we don't end up replaying old, freed
2110 * meta-data into a user's file.
2112 * To handle the cancellation of buffer log items, we make two passes
2113 * over the log during recovery. During the first we build a table of
2114 * those buffers which have been cancelled, and during the second we
2115 * only replay those buffers which do not have corresponding cancel
2116 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2117 * for more details on the implementation of the table of cancel records.
2120 xlog_recover_do_buffer_trans(
2122 xlog_recover_item_t *item,
2125 xfs_buf_log_format_t *buf_f;
2134 buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr;
2136 if (pass == XLOG_RECOVER_PASS1) {
2138 * In this pass we're only looking for buf items
2139 * with the XFS_BLI_CANCEL bit set.
2141 xlog_recover_do_buffer_pass1(log, buf_f);
2145 * In this pass we want to recover all the buffers
2146 * which have not been cancelled and are not
2147 * cancellation buffers themselves. The routine
2148 * we call here will tell us whether or not to
2149 * continue with the replay of this buffer.
2151 cancel = xlog_recover_do_buffer_pass2(log, buf_f);
2156 switch (buf_f->blf_type) {
2158 blkno = buf_f->blf_blkno;
2159 len = buf_f->blf_len;
2160 flags = buf_f->blf_flags;
2163 xfs_fs_cmn_err(CE_ALERT, log->l_mp,
2164 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2165 buf_f->blf_type, log->l_mp->m_logname ?
2166 log->l_mp->m_logname : "internal");
2167 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2168 XFS_ERRLEVEL_LOW, log->l_mp);
2169 return XFS_ERROR(EFSCORRUPTED);
2173 if (flags & XFS_BLI_INODE_BUF) {
2174 bp = xfs_buf_read_flags(mp->m_ddev_targp, blkno, len,
2177 bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, 0);
2179 if (XFS_BUF_ISERROR(bp)) {
2180 xfs_ioerror_alert("xlog_recover_do..(read#1)", log->l_mp,
2182 error = XFS_BUF_GETERROR(bp);
2188 if (flags & XFS_BLI_INODE_BUF) {
2189 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2191 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
2192 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2194 xlog_recover_do_reg_buffer(item, bp, buf_f);
2197 return XFS_ERROR(error);
2200 * Perform delayed write on the buffer. Asynchronous writes will be
2201 * slower when taking into account all the buffers to be flushed.
2203 * Also make sure that only inode buffers with good sizes stay in
2204 * the buffer cache. The kernel moves inodes in buffers of 1 block
2205 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2206 * buffers in the log can be a different size if the log was generated
2207 * by an older kernel using unclustered inode buffers or a newer kernel
2208 * running with a different inode cluster size. Regardless, if the
2209 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2210 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2211 * the buffer out of the buffer cache so that the buffer won't
2212 * overlap with future reads of those inodes.
2214 if (XFS_DINODE_MAGIC ==
2215 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2216 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2217 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2219 error = xfs_bwrite(mp, bp);
2221 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2222 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2223 XFS_BUF_SET_FSPRIVATE(bp, mp);
2224 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2225 xfs_bdwrite(mp, bp);
2232 xlog_recover_do_inode_trans(
2234 xlog_recover_item_t *item,
2237 xfs_inode_log_format_t *in_f;
2249 xfs_icdinode_t *dicp;
2252 if (pass == XLOG_RECOVER_PASS1) {
2256 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2257 in_f = (xfs_inode_log_format_t *)item->ri_buf[0].i_addr;
2259 in_f = (xfs_inode_log_format_t *)kmem_alloc(
2260 sizeof(xfs_inode_log_format_t), KM_SLEEP);
2262 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2266 ino = in_f->ilf_ino;
2268 if (ITEM_TYPE(item) == XFS_LI_INODE) {
2269 imap.im_blkno = (xfs_daddr_t)in_f->ilf_blkno;
2270 imap.im_len = in_f->ilf_len;
2271 imap.im_boffset = in_f->ilf_boffset;
2274 * It's an old inode format record. We don't know where
2275 * its cluster is located on disk, and we can't allow
2276 * xfs_imap() to figure it out because the inode btrees
2277 * are not ready to be used. Therefore do not pass the
2278 * XFS_IMAP_LOOKUP flag to xfs_imap(). This will give
2279 * us only the single block in which the inode lives
2280 * rather than its cluster, so we must make sure to
2281 * invalidate the buffer when we write it out below.
2284 xfs_imap(log->l_mp, NULL, ino, &imap, 0);
2288 * Inode buffers can be freed, look out for it,
2289 * and do not replay the inode.
2291 if (xlog_check_buffer_cancelled(log, imap.im_blkno, imap.im_len, 0)) {
2296 bp = xfs_buf_read_flags(mp->m_ddev_targp, imap.im_blkno, imap.im_len,
2298 if (XFS_BUF_ISERROR(bp)) {
2299 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
2301 error = XFS_BUF_GETERROR(bp);
2306 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2307 dip = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
2310 * Make sure the place we're flushing out to really looks
2313 if (unlikely(be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC)) {
2315 xfs_fs_cmn_err(CE_ALERT, mp,
2316 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2318 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2319 XFS_ERRLEVEL_LOW, mp);
2320 error = EFSCORRUPTED;
2323 dicp = (xfs_icdinode_t *)(item->ri_buf[1].i_addr);
2324 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2326 xfs_fs_cmn_err(CE_ALERT, mp,
2327 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2329 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2330 XFS_ERRLEVEL_LOW, mp);
2331 error = EFSCORRUPTED;
2335 /* Skip replay when the on disk inode is newer than the log one */
2336 if (dicp->di_flushiter < be16_to_cpu(dip->di_core.di_flushiter)) {
2338 * Deal with the wrap case, DI_MAX_FLUSH is less
2339 * than smaller numbers
2341 if (be16_to_cpu(dip->di_core.di_flushiter) == DI_MAX_FLUSH &&
2342 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2350 /* Take the opportunity to reset the flush iteration count */
2351 dicp->di_flushiter = 0;
2353 if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) {
2354 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2355 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2356 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2357 XFS_ERRLEVEL_LOW, mp, dicp);
2359 xfs_fs_cmn_err(CE_ALERT, mp,
2360 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2361 item, dip, bp, ino);
2362 error = EFSCORRUPTED;
2365 } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) {
2366 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2367 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2368 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2369 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2370 XFS_ERRLEVEL_LOW, mp, dicp);
2372 xfs_fs_cmn_err(CE_ALERT, mp,
2373 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2374 item, dip, bp, ino);
2375 error = EFSCORRUPTED;
2379 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2380 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2381 XFS_ERRLEVEL_LOW, mp, dicp);
2383 xfs_fs_cmn_err(CE_ALERT, mp,
2384 "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",
2386 dicp->di_nextents + dicp->di_anextents,
2388 error = EFSCORRUPTED;
2391 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2392 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2393 XFS_ERRLEVEL_LOW, mp, dicp);
2395 xfs_fs_cmn_err(CE_ALERT, mp,
2396 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2397 item, dip, bp, ino, dicp->di_forkoff);
2398 error = EFSCORRUPTED;
2401 if (unlikely(item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t))) {
2402 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2403 XFS_ERRLEVEL_LOW, mp, dicp);
2405 xfs_fs_cmn_err(CE_ALERT, mp,
2406 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2407 item->ri_buf[1].i_len, item);
2408 error = EFSCORRUPTED;
2412 /* The core is in in-core format */
2413 xfs_dinode_to_disk(&dip->di_core,
2414 (xfs_icdinode_t *)item->ri_buf[1].i_addr);
2416 /* the rest is in on-disk format */
2417 if (item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t)) {
2418 memcpy((xfs_caddr_t) dip + sizeof(xfs_dinode_core_t),
2419 item->ri_buf[1].i_addr + sizeof(xfs_dinode_core_t),
2420 item->ri_buf[1].i_len - sizeof(xfs_dinode_core_t));
2423 fields = in_f->ilf_fields;
2424 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2426 dip->di_u.di_dev = cpu_to_be32(in_f->ilf_u.ilfu_rdev);
2429 dip->di_u.di_muuid = in_f->ilf_u.ilfu_uuid;
2433 if (in_f->ilf_size == 2)
2434 goto write_inode_buffer;
2435 len = item->ri_buf[2].i_len;
2436 src = item->ri_buf[2].i_addr;
2437 ASSERT(in_f->ilf_size <= 4);
2438 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2439 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2440 (len == in_f->ilf_dsize));
2442 switch (fields & XFS_ILOG_DFORK) {
2443 case XFS_ILOG_DDATA:
2445 memcpy(&dip->di_u, src, len);
2448 case XFS_ILOG_DBROOT:
2449 xfs_bmbt_to_bmdr((xfs_bmbt_block_t *)src, len,
2450 &(dip->di_u.di_bmbt),
2451 XFS_DFORK_DSIZE(dip, mp));
2456 * There are no data fork flags set.
2458 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2463 * If we logged any attribute data, recover it. There may or
2464 * may not have been any other non-core data logged in this
2467 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2468 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2473 len = item->ri_buf[attr_index].i_len;
2474 src = item->ri_buf[attr_index].i_addr;
2475 ASSERT(len == in_f->ilf_asize);
2477 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2478 case XFS_ILOG_ADATA:
2480 dest = XFS_DFORK_APTR(dip);
2481 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2482 memcpy(dest, src, len);
2485 case XFS_ILOG_ABROOT:
2486 dest = XFS_DFORK_APTR(dip);
2487 xfs_bmbt_to_bmdr((xfs_bmbt_block_t *)src, len,
2488 (xfs_bmdr_block_t*)dest,
2489 XFS_DFORK_ASIZE(dip, mp));
2493 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2502 if (ITEM_TYPE(item) == XFS_LI_INODE) {
2503 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2504 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2505 XFS_BUF_SET_FSPRIVATE(bp, mp);
2506 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2507 xfs_bdwrite(mp, bp);
2510 error = xfs_bwrite(mp, bp);
2515 kmem_free(in_f, sizeof(*in_f));
2516 return XFS_ERROR(error);
2520 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2521 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2525 xlog_recover_do_quotaoff_trans(
2527 xlog_recover_item_t *item,
2530 xfs_qoff_logformat_t *qoff_f;
2532 if (pass == XLOG_RECOVER_PASS2) {
2536 qoff_f = (xfs_qoff_logformat_t *)item->ri_buf[0].i_addr;
2540 * The logitem format's flag tells us if this was user quotaoff,
2541 * group/project quotaoff or both.
2543 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2544 log->l_quotaoffs_flag |= XFS_DQ_USER;
2545 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2546 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2547 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2548 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2554 * Recover a dquot record
2557 xlog_recover_do_dquot_trans(
2559 xlog_recover_item_t *item,
2564 struct xfs_disk_dquot *ddq, *recddq;
2566 xfs_dq_logformat_t *dq_f;
2569 if (pass == XLOG_RECOVER_PASS1) {
2575 * Filesystems are required to send in quota flags at mount time.
2577 if (mp->m_qflags == 0)
2580 recddq = (xfs_disk_dquot_t *)item->ri_buf[1].i_addr;
2583 * This type of quotas was turned off, so ignore this record.
2585 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
2587 if (log->l_quotaoffs_flag & type)
2591 * At this point we know that quota was _not_ turned off.
2592 * Since the mount flags are not indicating to us otherwise, this
2593 * must mean that quota is on, and the dquot needs to be replayed.
2594 * Remember that we may not have fully recovered the superblock yet,
2595 * so we can't do the usual trick of looking at the SB quota bits.
2597 * The other possibility, of course, is that the quota subsystem was
2598 * removed since the last mount - ENOSYS.
2600 dq_f = (xfs_dq_logformat_t *)item->ri_buf[0].i_addr;
2602 if ((error = xfs_qm_dqcheck(recddq,
2604 0, XFS_QMOPT_DOWARN,
2605 "xlog_recover_do_dquot_trans (log copy)"))) {
2606 return XFS_ERROR(EIO);
2608 ASSERT(dq_f->qlf_len == 1);
2610 error = xfs_read_buf(mp, mp->m_ddev_targp,
2612 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2615 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
2616 bp, dq_f->qlf_blkno);
2620 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2623 * At least the magic num portion should be on disk because this
2624 * was among a chunk of dquots created earlier, and we did some
2625 * minimal initialization then.
2627 if (xfs_qm_dqcheck(ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2628 "xlog_recover_do_dquot_trans")) {
2630 return XFS_ERROR(EIO);
2633 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2635 ASSERT(dq_f->qlf_size == 2);
2636 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2637 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2638 XFS_BUF_SET_FSPRIVATE(bp, mp);
2639 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2640 xfs_bdwrite(mp, bp);
2646 * This routine is called to create an in-core extent free intent
2647 * item from the efi format structure which was logged on disk.
2648 * It allocates an in-core efi, copies the extents from the format
2649 * structure into it, and adds the efi to the AIL with the given
2653 xlog_recover_do_efi_trans(
2655 xlog_recover_item_t *item,
2661 xfs_efi_log_item_t *efip;
2662 xfs_efi_log_format_t *efi_formatp;
2664 if (pass == XLOG_RECOVER_PASS1) {
2668 efi_formatp = (xfs_efi_log_format_t *)item->ri_buf[0].i_addr;
2671 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
2672 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2673 &(efip->efi_format)))) {
2674 xfs_efi_item_free(efip);
2677 efip->efi_next_extent = efi_formatp->efi_nextents;
2678 efip->efi_flags |= XFS_EFI_COMMITTED;
2680 spin_lock(&mp->m_ail_lock);
2682 * xfs_trans_update_ail() drops the AIL lock.
2684 xfs_trans_update_ail(mp, (xfs_log_item_t *)efip, lsn);
2690 * This routine is called when an efd format structure is found in
2691 * a committed transaction in the log. It's purpose is to cancel
2692 * the corresponding efi if it was still in the log. To do this
2693 * it searches the AIL for the efi with an id equal to that in the
2694 * efd format structure. If we find it, we remove the efi from the
2698 xlog_recover_do_efd_trans(
2700 xlog_recover_item_t *item,
2704 xfs_efd_log_format_t *efd_formatp;
2705 xfs_efi_log_item_t *efip = NULL;
2706 xfs_log_item_t *lip;
2710 if (pass == XLOG_RECOVER_PASS1) {
2714 efd_formatp = (xfs_efd_log_format_t *)item->ri_buf[0].i_addr;
2715 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2716 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2717 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2718 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
2719 efi_id = efd_formatp->efd_efi_id;
2722 * Search for the efi with the id in the efd format structure
2726 spin_lock(&mp->m_ail_lock);
2727 lip = xfs_trans_first_ail(mp, &gen);
2728 while (lip != NULL) {
2729 if (lip->li_type == XFS_LI_EFI) {
2730 efip = (xfs_efi_log_item_t *)lip;
2731 if (efip->efi_format.efi_id == efi_id) {
2733 * xfs_trans_delete_ail() drops the
2736 xfs_trans_delete_ail(mp, lip);
2737 xfs_efi_item_free(efip);
2741 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
2743 spin_unlock(&mp->m_ail_lock);
2747 * Perform the transaction
2749 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2750 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2753 xlog_recover_do_trans(
2755 xlog_recover_t *trans,
2759 xlog_recover_item_t *item, *first_item;
2761 if ((error = xlog_recover_reorder_trans(trans)))
2763 first_item = item = trans->r_itemq;
2766 * we don't need to worry about the block number being
2767 * truncated in > 1 TB buffers because in user-land,
2768 * we're now n32 or 64-bit so xfs_daddr_t is 64-bits so
2769 * the blknos will get through the user-mode buffer
2770 * cache properly. The only bad case is o32 kernels
2771 * where xfs_daddr_t is 32-bits but mount will warn us
2772 * off a > 1 TB filesystem before we get here.
2774 if ((ITEM_TYPE(item) == XFS_LI_BUF)) {
2775 if ((error = xlog_recover_do_buffer_trans(log, item,
2778 } else if ((ITEM_TYPE(item) == XFS_LI_INODE)) {
2779 if ((error = xlog_recover_do_inode_trans(log, item,
2782 } else if (ITEM_TYPE(item) == XFS_LI_EFI) {
2783 if ((error = xlog_recover_do_efi_trans(log, item, trans->r_lsn,
2786 } else if (ITEM_TYPE(item) == XFS_LI_EFD) {
2787 xlog_recover_do_efd_trans(log, item, pass);
2788 } else if (ITEM_TYPE(item) == XFS_LI_DQUOT) {
2789 if ((error = xlog_recover_do_dquot_trans(log, item,
2792 } else if ((ITEM_TYPE(item) == XFS_LI_QUOTAOFF)) {
2793 if ((error = xlog_recover_do_quotaoff_trans(log, item,
2797 xlog_warn("XFS: xlog_recover_do_trans");
2799 error = XFS_ERROR(EIO);
2802 item = item->ri_next;
2803 } while (first_item != item);
2809 * Free up any resources allocated by the transaction
2811 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2814 xlog_recover_free_trans(
2815 xlog_recover_t *trans)
2817 xlog_recover_item_t *first_item, *item, *free_item;
2820 item = first_item = trans->r_itemq;
2823 item = item->ri_next;
2824 /* Free the regions in the item. */
2825 for (i = 0; i < free_item->ri_cnt; i++) {
2826 kmem_free(free_item->ri_buf[i].i_addr,
2827 free_item->ri_buf[i].i_len);
2829 /* Free the item itself */
2830 kmem_free(free_item->ri_buf,
2831 (free_item->ri_total * sizeof(xfs_log_iovec_t)));
2832 kmem_free(free_item, sizeof(xlog_recover_item_t));
2833 } while (first_item != item);
2834 /* Free the transaction recover structure */
2835 kmem_free(trans, sizeof(xlog_recover_t));
2839 xlog_recover_commit_trans(
2842 xlog_recover_t *trans,
2847 if ((error = xlog_recover_unlink_tid(q, trans)))
2849 if ((error = xlog_recover_do_trans(log, trans, pass)))
2851 xlog_recover_free_trans(trans); /* no error */
2856 xlog_recover_unmount_trans(
2857 xlog_recover_t *trans)
2859 /* Do nothing now */
2860 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2865 * There are two valid states of the r_state field. 0 indicates that the
2866 * transaction structure is in a normal state. We have either seen the
2867 * start of the transaction or the last operation we added was not a partial
2868 * operation. If the last operation we added to the transaction was a
2869 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2871 * NOTE: skip LRs with 0 data length.
2874 xlog_recover_process_data(
2876 xlog_recover_t *rhash[],
2877 xlog_rec_header_t *rhead,
2883 xlog_op_header_t *ohead;
2884 xlog_recover_t *trans;
2890 lp = dp + be32_to_cpu(rhead->h_len);
2891 num_logops = be32_to_cpu(rhead->h_num_logops);
2893 /* check the log format matches our own - else we can't recover */
2894 if (xlog_header_check_recover(log->l_mp, rhead))
2895 return (XFS_ERROR(EIO));
2897 while ((dp < lp) && num_logops) {
2898 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2899 ohead = (xlog_op_header_t *)dp;
2900 dp += sizeof(xlog_op_header_t);
2901 if (ohead->oh_clientid != XFS_TRANSACTION &&
2902 ohead->oh_clientid != XFS_LOG) {
2904 "XFS: xlog_recover_process_data: bad clientid");
2906 return (XFS_ERROR(EIO));
2908 tid = be32_to_cpu(ohead->oh_tid);
2909 hash = XLOG_RHASH(tid);
2910 trans = xlog_recover_find_tid(rhash[hash], tid);
2911 if (trans == NULL) { /* not found; add new tid */
2912 if (ohead->oh_flags & XLOG_START_TRANS)
2913 xlog_recover_new_tid(&rhash[hash], tid,
2914 be64_to_cpu(rhead->h_lsn));
2916 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
2918 "XFS: xlog_recover_process_data: bad length");
2920 return (XFS_ERROR(EIO));
2922 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2923 if (flags & XLOG_WAS_CONT_TRANS)
2924 flags &= ~XLOG_CONTINUE_TRANS;
2926 case XLOG_COMMIT_TRANS:
2927 error = xlog_recover_commit_trans(log,
2928 &rhash[hash], trans, pass);
2930 case XLOG_UNMOUNT_TRANS:
2931 error = xlog_recover_unmount_trans(trans);
2933 case XLOG_WAS_CONT_TRANS:
2934 error = xlog_recover_add_to_cont_trans(trans,
2935 dp, be32_to_cpu(ohead->oh_len));
2937 case XLOG_START_TRANS:
2939 "XFS: xlog_recover_process_data: bad transaction");
2941 error = XFS_ERROR(EIO);
2944 case XLOG_CONTINUE_TRANS:
2945 error = xlog_recover_add_to_trans(trans,
2946 dp, be32_to_cpu(ohead->oh_len));
2950 "XFS: xlog_recover_process_data: bad flag");
2952 error = XFS_ERROR(EIO);
2958 dp += be32_to_cpu(ohead->oh_len);
2965 * Process an extent free intent item that was recovered from
2966 * the log. We need to free the extents that it describes.
2969 xlog_recover_process_efi(
2971 xfs_efi_log_item_t *efip)
2973 xfs_efd_log_item_t *efdp;
2977 xfs_fsblock_t startblock_fsb;
2979 ASSERT(!(efip->efi_flags & XFS_EFI_RECOVERED));
2982 * First check the validity of the extents described by the
2983 * EFI. If any are bad, then assume that all are bad and
2984 * just toss the EFI.
2986 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2987 extp = &(efip->efi_format.efi_extents[i]);
2988 startblock_fsb = XFS_BB_TO_FSB(mp,
2989 XFS_FSB_TO_DADDR(mp, extp->ext_start));
2990 if ((startblock_fsb == 0) ||
2991 (extp->ext_len == 0) ||
2992 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
2993 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
2995 * This will pull the EFI from the AIL and
2996 * free the memory associated with it.
2998 xfs_efi_release(efip, efip->efi_format.efi_nextents);
3003 tp = xfs_trans_alloc(mp, 0);
3004 xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
3005 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3007 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3008 extp = &(efip->efi_format.efi_extents[i]);
3009 xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3010 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3014 efip->efi_flags |= XFS_EFI_RECOVERED;
3015 xfs_trans_commit(tp, 0);
3019 * Verify that once we've encountered something other than an EFI
3020 * in the AIL that there are no more EFIs in the AIL.
3024 xlog_recover_check_ail(
3026 xfs_log_item_t *lip,
3032 ASSERT(lip->li_type != XFS_LI_EFI);
3033 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
3035 * The check will be bogus if we restart from the
3036 * beginning of the AIL, so ASSERT that we don't.
3037 * We never should since we're holding the AIL lock
3040 ASSERT(gen == orig_gen);
3041 } while (lip != NULL);
3046 * When this is called, all of the EFIs which did not have
3047 * corresponding EFDs should be in the AIL. What we do now
3048 * is free the extents associated with each one.
3050 * Since we process the EFIs in normal transactions, they
3051 * will be removed at some point after the commit. This prevents
3052 * us from just walking down the list processing each one.
3053 * We'll use a flag in the EFI to skip those that we've already
3054 * processed and use the AIL iteration mechanism's generation
3055 * count to try to speed this up at least a bit.
3057 * When we start, we know that the EFIs are the only things in
3058 * the AIL. As we process them, however, other items are added
3059 * to the AIL. Since everything added to the AIL must come after
3060 * everything already in the AIL, we stop processing as soon as
3061 * we see something other than an EFI in the AIL.
3064 xlog_recover_process_efis(
3067 xfs_log_item_t *lip;
3068 xfs_efi_log_item_t *efip;
3073 spin_lock(&mp->m_ail_lock);
3075 lip = xfs_trans_first_ail(mp, &gen);
3076 while (lip != NULL) {
3078 * We're done when we see something other than an EFI.
3080 if (lip->li_type != XFS_LI_EFI) {
3081 xlog_recover_check_ail(mp, lip, gen);
3086 * Skip EFIs that we've already processed.
3088 efip = (xfs_efi_log_item_t *)lip;
3089 if (efip->efi_flags & XFS_EFI_RECOVERED) {
3090 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
3094 spin_unlock(&mp->m_ail_lock);
3095 xlog_recover_process_efi(mp, efip);
3096 spin_lock(&mp->m_ail_lock);
3097 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
3099 spin_unlock(&mp->m_ail_lock);
3103 * This routine performs a transaction to null out a bad inode pointer
3104 * in an agi unlinked inode hash bucket.
3107 xlog_recover_clear_agi_bucket(
3109 xfs_agnumber_t agno,
3118 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3119 xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp), 0, 0, 0);
3121 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
3122 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
3123 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
3125 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3129 agi = XFS_BUF_TO_AGI(agibp);
3130 if (be32_to_cpu(agi->agi_magicnum) != XFS_AGI_MAGIC) {
3131 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3135 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
3136 offset = offsetof(xfs_agi_t, agi_unlinked) +
3137 (sizeof(xfs_agino_t) * bucket);
3138 xfs_trans_log_buf(tp, agibp, offset,
3139 (offset + sizeof(xfs_agino_t) - 1));
3141 (void) xfs_trans_commit(tp, 0);
3145 * xlog_iunlink_recover
3147 * This is called during recovery to process any inodes which
3148 * we unlinked but not freed when the system crashed. These
3149 * inodes will be on the lists in the AGI blocks. What we do
3150 * here is scan all the AGIs and fully truncate and free any
3151 * inodes found on the lists. Each inode is removed from the
3152 * lists when it has been fully truncated and is freed. The
3153 * freeing of the inode and its removal from the list must be
3157 xlog_recover_process_iunlinks(
3161 xfs_agnumber_t agno;
3176 * Prevent any DMAPI event from being sent while in this function.
3178 mp_dmevmask = mp->m_dmevmask;
3181 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3183 * Find the agi for this ag.
3185 agibp = xfs_buf_read(mp->m_ddev_targp,
3186 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
3187 XFS_FSS_TO_BB(mp, 1), 0);
3188 if (XFS_BUF_ISERROR(agibp)) {
3189 xfs_ioerror_alert("xlog_recover_process_iunlinks(#1)",
3191 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)));
3193 agi = XFS_BUF_TO_AGI(agibp);
3194 ASSERT(XFS_AGI_MAGIC == be32_to_cpu(agi->agi_magicnum));
3196 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3198 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3199 while (agino != NULLAGINO) {
3202 * Release the agi buffer so that it can
3203 * be acquired in the normal course of the
3204 * transaction to truncate and free the inode.
3206 xfs_buf_relse(agibp);
3208 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3209 error = xfs_iget(mp, NULL, ino, 0, 0, &ip, 0);
3210 ASSERT(error || (ip != NULL));
3214 * Get the on disk inode to find the
3215 * next inode in the bucket.
3217 error = xfs_itobp(mp, NULL, ip, &dip,
3220 ASSERT(error || (dip != NULL));
3224 ASSERT(ip->i_d.di_nlink == 0);
3226 /* setup for the next pass */
3227 agino = be32_to_cpu(
3228 dip->di_next_unlinked);
3231 * Prevent any DMAPI event from
3232 * being sent when the
3233 * reference on the inode is
3236 ip->i_d.di_dmevmask = 0;
3239 * If this is a new inode, handle
3240 * it specially. Otherwise,
3241 * just drop our reference to the
3242 * inode. If there are no
3243 * other references, this will
3245 * xfs_inactive() which will
3246 * truncate the file and free
3249 if (ip->i_d.di_mode == 0)
3250 xfs_iput_new(ip, 0);
3255 * We can't read in the inode
3256 * this bucket points to, or
3257 * this inode is messed up. Just
3258 * ditch this bucket of inodes. We
3259 * will lose some inodes and space,
3260 * but at least we won't hang. Call
3261 * xlog_recover_clear_agi_bucket()
3262 * to perform a transaction to clear
3263 * the inode pointer in the bucket.
3265 xlog_recover_clear_agi_bucket(mp, agno,
3272 * Reacquire the agibuffer and continue around
3275 agibp = xfs_buf_read(mp->m_ddev_targp,
3276 XFS_AG_DADDR(mp, agno,
3278 XFS_FSS_TO_BB(mp, 1), 0);
3279 if (XFS_BUF_ISERROR(agibp)) {
3281 "xlog_recover_process_iunlinks(#2)",
3283 XFS_AG_DADDR(mp, agno,
3284 XFS_AGI_DADDR(mp)));
3286 agi = XFS_BUF_TO_AGI(agibp);
3287 ASSERT(XFS_AGI_MAGIC == be32_to_cpu(
3288 agi->agi_magicnum));
3293 * Release the buffer for the current agi so we can
3294 * go on to the next one.
3296 xfs_buf_relse(agibp);
3299 mp->m_dmevmask = mp_dmevmask;
3305 xlog_pack_data_checksum(
3307 xlog_in_core_t *iclog,
3314 up = (__be32 *)iclog->ic_datap;
3315 /* divide length by 4 to get # words */
3316 for (i = 0; i < (size >> 2); i++) {
3317 chksum ^= be32_to_cpu(*up);
3320 iclog->ic_header.h_chksum = cpu_to_be32(chksum);
3323 #define xlog_pack_data_checksum(log, iclog, size)
3327 * Stamp cycle number in every block
3332 xlog_in_core_t *iclog,
3336 int size = iclog->ic_offset + roundoff;
3339 xlog_in_core_2_t *xhdr;
3341 xlog_pack_data_checksum(log, iclog, size);
3343 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3345 dp = iclog->ic_datap;
3346 for (i = 0; i < BTOBB(size) &&
3347 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3348 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
3349 *(__be32 *)dp = cycle_lsn;
3353 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3354 xhdr = (xlog_in_core_2_t *)&iclog->ic_header;
3355 for ( ; i < BTOBB(size); i++) {
3356 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3357 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3358 xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
3359 *(__be32 *)dp = cycle_lsn;
3363 for (i = 1; i < log->l_iclog_heads; i++) {
3364 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3369 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
3371 xlog_unpack_data_checksum(
3372 xlog_rec_header_t *rhead,
3376 __be32 *up = (__be32 *)dp;
3380 /* divide length by 4 to get # words */
3381 for (i=0; i < be32_to_cpu(rhead->h_len) >> 2; i++) {
3382 chksum ^= be32_to_cpu(*up);
3385 if (chksum != be32_to_cpu(rhead->h_chksum)) {
3386 if (rhead->h_chksum ||
3387 ((log->l_flags & XLOG_CHKSUM_MISMATCH) == 0)) {
3389 "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n",
3390 be32_to_cpu(rhead->h_chksum), chksum);
3392 "XFS: Disregard message if filesystem was created with non-DEBUG kernel");
3393 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3395 "XFS: LogR this is a LogV2 filesystem\n");
3397 log->l_flags |= XLOG_CHKSUM_MISMATCH;
3402 #define xlog_unpack_data_checksum(rhead, dp, log)
3407 xlog_rec_header_t *rhead,
3412 xlog_in_core_2_t *xhdr;
3414 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
3415 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3416 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
3420 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3421 xhdr = (xlog_in_core_2_t *)rhead;
3422 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
3423 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3424 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3425 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
3430 xlog_unpack_data_checksum(rhead, dp, log);
3434 xlog_valid_rec_header(
3436 xlog_rec_header_t *rhead,
3441 if (unlikely(be32_to_cpu(rhead->h_magicno) != XLOG_HEADER_MAGIC_NUM)) {
3442 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3443 XFS_ERRLEVEL_LOW, log->l_mp);
3444 return XFS_ERROR(EFSCORRUPTED);
3447 (!rhead->h_version ||
3448 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
3449 xlog_warn("XFS: %s: unrecognised log version (%d).",
3450 __FUNCTION__, be32_to_cpu(rhead->h_version));
3451 return XFS_ERROR(EIO);
3454 /* LR body must have data or it wouldn't have been written */
3455 hlen = be32_to_cpu(rhead->h_len);
3456 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3457 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3458 XFS_ERRLEVEL_LOW, log->l_mp);
3459 return XFS_ERROR(EFSCORRUPTED);
3461 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3462 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3463 XFS_ERRLEVEL_LOW, log->l_mp);
3464 return XFS_ERROR(EFSCORRUPTED);
3470 * Read the log from tail to head and process the log records found.
3471 * Handle the two cases where the tail and head are in the same cycle
3472 * and where the active portion of the log wraps around the end of
3473 * the physical log separately. The pass parameter is passed through
3474 * to the routines called to process the data and is not looked at
3478 xlog_do_recovery_pass(
3480 xfs_daddr_t head_blk,
3481 xfs_daddr_t tail_blk,
3484 xlog_rec_header_t *rhead;
3486 xfs_caddr_t bufaddr, offset;
3487 xfs_buf_t *hbp, *dbp;
3488 int error = 0, h_size;
3489 int bblks, split_bblks;
3490 int hblks, split_hblks, wrapped_hblks;
3491 xlog_recover_t *rhash[XLOG_RHASH_SIZE];
3493 ASSERT(head_blk != tail_blk);
3496 * Read the header of the tail block and get the iclog buffer size from
3497 * h_size. Use this to tell how many sectors make up the log header.
3499 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3501 * When using variable length iclogs, read first sector of
3502 * iclog header and extract the header size from it. Get a
3503 * new hbp that is the correct size.
3505 hbp = xlog_get_bp(log, 1);
3508 if ((error = xlog_bread(log, tail_blk, 1, hbp)))
3510 offset = xlog_align(log, tail_blk, 1, hbp);
3511 rhead = (xlog_rec_header_t *)offset;
3512 error = xlog_valid_rec_header(log, rhead, tail_blk);
3515 h_size = be32_to_cpu(rhead->h_size);
3516 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
3517 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3518 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3519 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3522 hbp = xlog_get_bp(log, hblks);
3527 ASSERT(log->l_sectbb_log == 0);
3529 hbp = xlog_get_bp(log, 1);
3530 h_size = XLOG_BIG_RECORD_BSIZE;
3535 dbp = xlog_get_bp(log, BTOBB(h_size));
3541 memset(rhash, 0, sizeof(rhash));
3542 if (tail_blk <= head_blk) {
3543 for (blk_no = tail_blk; blk_no < head_blk; ) {
3544 if ((error = xlog_bread(log, blk_no, hblks, hbp)))
3546 offset = xlog_align(log, blk_no, hblks, hbp);
3547 rhead = (xlog_rec_header_t *)offset;
3548 error = xlog_valid_rec_header(log, rhead, blk_no);
3552 /* blocks in data section */
3553 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3554 error = xlog_bread(log, blk_no + hblks, bblks, dbp);
3557 offset = xlog_align(log, blk_no + hblks, bblks, dbp);
3558 xlog_unpack_data(rhead, offset, log);
3559 if ((error = xlog_recover_process_data(log,
3560 rhash, rhead, offset, pass)))
3562 blk_no += bblks + hblks;
3566 * Perform recovery around the end of the physical log.
3567 * When the head is not on the same cycle number as the tail,
3568 * we can't do a sequential recovery as above.
3571 while (blk_no < log->l_logBBsize) {
3573 * Check for header wrapping around physical end-of-log
3578 if (blk_no + hblks <= log->l_logBBsize) {
3579 /* Read header in one read */
3580 error = xlog_bread(log, blk_no, hblks, hbp);
3583 offset = xlog_align(log, blk_no, hblks, hbp);
3585 /* This LR is split across physical log end */
3586 if (blk_no != log->l_logBBsize) {
3587 /* some data before physical log end */
3588 ASSERT(blk_no <= INT_MAX);
3589 split_hblks = log->l_logBBsize - (int)blk_no;
3590 ASSERT(split_hblks > 0);
3591 if ((error = xlog_bread(log, blk_no,
3594 offset = xlog_align(log, blk_no,
3598 * Note: this black magic still works with
3599 * large sector sizes (non-512) only because:
3600 * - we increased the buffer size originally
3601 * by 1 sector giving us enough extra space
3602 * for the second read;
3603 * - the log start is guaranteed to be sector
3605 * - we read the log end (LR header start)
3606 * _first_, then the log start (LR header end)
3607 * - order is important.
3609 bufaddr = XFS_BUF_PTR(hbp);
3610 XFS_BUF_SET_PTR(hbp,
3611 bufaddr + BBTOB(split_hblks),
3612 BBTOB(hblks - split_hblks));
3613 wrapped_hblks = hblks - split_hblks;
3614 error = xlog_bread(log, 0, wrapped_hblks, hbp);
3617 XFS_BUF_SET_PTR(hbp, bufaddr, BBTOB(hblks));
3619 offset = xlog_align(log, 0,
3620 wrapped_hblks, hbp);
3622 rhead = (xlog_rec_header_t *)offset;
3623 error = xlog_valid_rec_header(log, rhead,
3624 split_hblks ? blk_no : 0);
3628 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3631 /* Read in data for log record */
3632 if (blk_no + bblks <= log->l_logBBsize) {
3633 error = xlog_bread(log, blk_no, bblks, dbp);
3636 offset = xlog_align(log, blk_no, bblks, dbp);
3638 /* This log record is split across the
3639 * physical end of log */
3642 if (blk_no != log->l_logBBsize) {
3643 /* some data is before the physical
3645 ASSERT(!wrapped_hblks);
3646 ASSERT(blk_no <= INT_MAX);
3648 log->l_logBBsize - (int)blk_no;
3649 ASSERT(split_bblks > 0);
3650 if ((error = xlog_bread(log, blk_no,
3653 offset = xlog_align(log, blk_no,
3657 * Note: this black magic still works with
3658 * large sector sizes (non-512) only because:
3659 * - we increased the buffer size originally
3660 * by 1 sector giving us enough extra space
3661 * for the second read;
3662 * - the log start is guaranteed to be sector
3664 * - we read the log end (LR header start)
3665 * _first_, then the log start (LR header end)
3666 * - order is important.
3668 bufaddr = XFS_BUF_PTR(dbp);
3669 XFS_BUF_SET_PTR(dbp,
3670 bufaddr + BBTOB(split_bblks),
3671 BBTOB(bblks - split_bblks));
3672 if ((error = xlog_bread(log, wrapped_hblks,
3673 bblks - split_bblks, dbp)))
3675 XFS_BUF_SET_PTR(dbp, bufaddr, h_size);
3677 offset = xlog_align(log, wrapped_hblks,
3678 bblks - split_bblks, dbp);
3680 xlog_unpack_data(rhead, offset, log);
3681 if ((error = xlog_recover_process_data(log, rhash,
3682 rhead, offset, pass)))
3687 ASSERT(blk_no >= log->l_logBBsize);
3688 blk_no -= log->l_logBBsize;
3690 /* read first part of physical log */
3691 while (blk_no < head_blk) {
3692 if ((error = xlog_bread(log, blk_no, hblks, hbp)))
3694 offset = xlog_align(log, blk_no, hblks, hbp);
3695 rhead = (xlog_rec_header_t *)offset;
3696 error = xlog_valid_rec_header(log, rhead, blk_no);
3699 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3700 if ((error = xlog_bread(log, blk_no+hblks, bblks, dbp)))
3702 offset = xlog_align(log, blk_no+hblks, bblks, dbp);
3703 xlog_unpack_data(rhead, offset, log);
3704 if ((error = xlog_recover_process_data(log, rhash,
3705 rhead, offset, pass)))
3707 blk_no += bblks + hblks;
3719 * Do the recovery of the log. We actually do this in two phases.
3720 * The two passes are necessary in order to implement the function
3721 * of cancelling a record written into the log. The first pass
3722 * determines those things which have been cancelled, and the
3723 * second pass replays log items normally except for those which
3724 * have been cancelled. The handling of the replay and cancellations
3725 * takes place in the log item type specific routines.
3727 * The table of items which have cancel records in the log is allocated
3728 * and freed at this level, since only here do we know when all of
3729 * the log recovery has been completed.
3732 xlog_do_log_recovery(
3734 xfs_daddr_t head_blk,
3735 xfs_daddr_t tail_blk)
3739 ASSERT(head_blk != tail_blk);
3742 * First do a pass to find all of the cancelled buf log items.
3743 * Store them in the buf_cancel_table for use in the second pass.
3745 log->l_buf_cancel_table =
3746 (xfs_buf_cancel_t **)kmem_zalloc(XLOG_BC_TABLE_SIZE *
3747 sizeof(xfs_buf_cancel_t*),
3749 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3750 XLOG_RECOVER_PASS1);
3752 kmem_free(log->l_buf_cancel_table,
3753 XLOG_BC_TABLE_SIZE * sizeof(xfs_buf_cancel_t*));
3754 log->l_buf_cancel_table = NULL;
3758 * Then do a second pass to actually recover the items in the log.
3759 * When it is complete free the table of buf cancel items.
3761 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3762 XLOG_RECOVER_PASS2);
3767 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3768 ASSERT(log->l_buf_cancel_table[i] == NULL);
3772 kmem_free(log->l_buf_cancel_table,
3773 XLOG_BC_TABLE_SIZE * sizeof(xfs_buf_cancel_t*));
3774 log->l_buf_cancel_table = NULL;
3780 * Do the actual recovery
3785 xfs_daddr_t head_blk,
3786 xfs_daddr_t tail_blk)
3793 * First replay the images in the log.
3795 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3800 XFS_bflush(log->l_mp->m_ddev_targp);
3803 * If IO errors happened during recovery, bail out.
3805 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3810 * We now update the tail_lsn since much of the recovery has completed
3811 * and there may be space available to use. If there were no extent
3812 * or iunlinks, we can free up the entire log and set the tail_lsn to
3813 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3814 * lsn of the last known good LR on disk. If there are extent frees
3815 * or iunlinks they will have some entries in the AIL; so we look at
3816 * the AIL to determine how to set the tail_lsn.
3818 xlog_assign_tail_lsn(log->l_mp);
3821 * Now that we've finished replaying all buffer and inode
3822 * updates, re-read in the superblock.
3824 bp = xfs_getsb(log->l_mp, 0);
3826 ASSERT(!(XFS_BUF_ISWRITE(bp)));
3827 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
3829 XFS_BUF_UNASYNC(bp);
3830 xfsbdstrat(log->l_mp, bp);
3831 if ((error = xfs_iowait(bp))) {
3832 xfs_ioerror_alert("xlog_do_recover",
3833 log->l_mp, bp, XFS_BUF_ADDR(bp));
3839 /* Convert superblock from on-disk format */
3840 sbp = &log->l_mp->m_sb;
3841 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
3842 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
3843 ASSERT(xfs_sb_good_version(sbp));
3846 /* We've re-read the superblock so re-initialize per-cpu counters */
3847 xfs_icsb_reinit_counters(log->l_mp);
3849 xlog_recover_check_summary(log);
3851 /* Normal transactions can now occur */
3852 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3857 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3859 * Return error or zero.
3865 xfs_daddr_t head_blk, tail_blk;
3868 /* find the tail of the log */
3869 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
3872 if (tail_blk != head_blk) {
3873 /* There used to be a comment here:
3875 * disallow recovery on read-only mounts. note -- mount
3876 * checks for ENOSPC and turns it into an intelligent
3878 * ...but this is no longer true. Now, unless you specify
3879 * NORECOVERY (in which case this function would never be
3880 * called), we just go ahead and recover. We do this all
3881 * under the vfs layer, so we can get away with it unless
3882 * the device itself is read-only, in which case we fail.
3884 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
3889 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3890 log->l_mp->m_fsname, log->l_mp->m_logname ?
3891 log->l_mp->m_logname : "internal");
3893 error = xlog_do_recover(log, head_blk, tail_blk);
3894 log->l_flags |= XLOG_RECOVERY_NEEDED;
3900 * In the first part of recovery we replay inodes and buffers and build
3901 * up the list of extent free items which need to be processed. Here
3902 * we process the extent free items and clean up the on disk unlinked
3903 * inode lists. This is separated from the first part of recovery so
3904 * that the root and real-time bitmap inodes can be read in from disk in
3905 * between the two stages. This is necessary so that we can free space
3906 * in the real-time portion of the file system.
3909 xlog_recover_finish(
3914 * Now we're ready to do the transactions needed for the
3915 * rest of recovery. Start with completing all the extent
3916 * free intent records and then process the unlinked inode
3917 * lists. At this point, we essentially run in normal mode
3918 * except that we're still performing recovery actions
3919 * rather than accepting new requests.
3921 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
3922 xlog_recover_process_efis(log);
3924 * Sync the log to get all the EFIs out of the AIL.
3925 * This isn't absolutely necessary, but it helps in
3926 * case the unlink transactions would have problems
3927 * pushing the EFIs out of the way.
3929 xfs_log_force(log->l_mp, (xfs_lsn_t)0,
3930 (XFS_LOG_FORCE | XFS_LOG_SYNC));
3932 if ( (mfsi_flags & XFS_MFSI_NOUNLINK) == 0 ) {
3933 xlog_recover_process_iunlinks(log);
3936 xlog_recover_check_summary(log);
3939 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3940 log->l_mp->m_fsname, log->l_mp->m_logname ?
3941 log->l_mp->m_logname : "internal");
3942 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
3945 "!Ending clean XFS mount for filesystem: %s\n",
3946 log->l_mp->m_fsname);
3954 * Read all of the agf and agi counters and check that they
3955 * are consistent with the superblock counters.
3958 xlog_recover_check_summary(
3966 xfs_daddr_t agfdaddr;
3967 xfs_daddr_t agidaddr;
3969 #ifdef XFS_LOUD_RECOVERY
3972 xfs_agnumber_t agno;
3973 __uint64_t freeblks;
3982 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3983 agfdaddr = XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp));
3984 agfbp = xfs_buf_read(mp->m_ddev_targp, agfdaddr,
3985 XFS_FSS_TO_BB(mp, 1), 0);
3986 if (XFS_BUF_ISERROR(agfbp)) {
3987 xfs_ioerror_alert("xlog_recover_check_summary(agf)",
3988 mp, agfbp, agfdaddr);
3990 agfp = XFS_BUF_TO_AGF(agfbp);
3991 ASSERT(XFS_AGF_MAGIC == be32_to_cpu(agfp->agf_magicnum));
3992 ASSERT(XFS_AGF_GOOD_VERSION(be32_to_cpu(agfp->agf_versionnum)));
3993 ASSERT(be32_to_cpu(agfp->agf_seqno) == agno);
3995 freeblks += be32_to_cpu(agfp->agf_freeblks) +
3996 be32_to_cpu(agfp->agf_flcount);
3997 xfs_buf_relse(agfbp);
3999 agidaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
4000 agibp = xfs_buf_read(mp->m_ddev_targp, agidaddr,
4001 XFS_FSS_TO_BB(mp, 1), 0);
4002 if (XFS_BUF_ISERROR(agibp)) {
4003 xfs_ioerror_alert("xlog_recover_check_summary(agi)",
4004 mp, agibp, agidaddr);
4006 agip = XFS_BUF_TO_AGI(agibp);
4007 ASSERT(XFS_AGI_MAGIC == be32_to_cpu(agip->agi_magicnum));
4008 ASSERT(XFS_AGI_GOOD_VERSION(be32_to_cpu(agip->agi_versionnum)));
4009 ASSERT(be32_to_cpu(agip->agi_seqno) == agno);
4011 itotal += be32_to_cpu(agip->agi_count);
4012 ifree += be32_to_cpu(agip->agi_freecount);
4013 xfs_buf_relse(agibp);
4016 sbbp = xfs_getsb(mp, 0);
4017 #ifdef XFS_LOUD_RECOVERY
4019 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(sbbp));
4021 "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
4022 sbp->sb_icount, itotal);
4024 "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
4025 sbp->sb_ifree, ifree);
4027 "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
4028 sbp->sb_fdblocks, freeblks);
4031 * This is turned off until I account for the allocation
4032 * btree blocks which live in free space.
4034 ASSERT(sbp->sb_icount == itotal);
4035 ASSERT(sbp->sb_ifree == ifree);
4036 ASSERT(sbp->sb_fdblocks == freeblks);
4039 xfs_buf_relse(sbbp);