2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_error.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_alloc.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_log_priv.h"
42 #include "xfs_buf_item.h"
43 #include "xfs_log_recover.h"
44 #include "xfs_extfree_item.h"
45 #include "xfs_trans_priv.h"
46 #include "xfs_quota.h"
48 #include "xfs_utils.h"
49 #include "xfs_trace.h"
51 STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *);
52 STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t);
53 STATIC void xlog_recover_insert_item_backq(xlog_recover_item_t **q,
54 xlog_recover_item_t *item);
56 STATIC void xlog_recover_check_summary(xlog_t *);
58 #define xlog_recover_check_summary(log)
63 * Sector aligned buffer routines for buffer create/read/write/access
66 #define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \
67 ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \
68 ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) )
69 #define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask)
76 if (nbblks <= 0 || nbblks > log->l_logBBsize) {
77 xlog_warn("XFS: Invalid block length (0x%x) given for buffer", nbblks);
78 XFS_ERROR_REPORT("xlog_get_bp(1)",
79 XFS_ERRLEVEL_HIGH, log->l_mp);
83 if (log->l_sectbb_log) {
85 nbblks += XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
86 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
88 return xfs_buf_get_noaddr(BBTOB(nbblks), log->l_mp->m_logdev_targp);
107 if (!log->l_sectbb_log)
108 return XFS_BUF_PTR(bp);
110 ptr = XFS_BUF_PTR(bp) + BBTOB((int)blk_no & log->l_sectbb_mask);
111 ASSERT(XFS_BUF_SIZE(bp) >=
112 BBTOB(nbblks + (blk_no & log->l_sectbb_mask)));
118 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
129 if (nbblks <= 0 || nbblks > log->l_logBBsize) {
130 xlog_warn("XFS: Invalid block length (0x%x) given for buffer", nbblks);
131 XFS_ERROR_REPORT("xlog_bread(1)",
132 XFS_ERRLEVEL_HIGH, log->l_mp);
136 if (log->l_sectbb_log) {
137 blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
138 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
142 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
145 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
148 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
149 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
151 xfsbdstrat(log->l_mp, bp);
152 error = xfs_iowait(bp);
154 xfs_ioerror_alert("xlog_bread", log->l_mp,
155 bp, XFS_BUF_ADDR(bp));
169 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
173 *offset = xlog_align(log, blk_no, nbblks, bp);
178 * Write out the buffer at the given block for the given number of blocks.
179 * The buffer is kept locked across the write and is returned locked.
180 * This can only be used for synchronous log writes.
191 if (nbblks <= 0 || nbblks > log->l_logBBsize) {
192 xlog_warn("XFS: Invalid block length (0x%x) given for buffer", nbblks);
193 XFS_ERROR_REPORT("xlog_bwrite(1)",
194 XFS_ERRLEVEL_HIGH, log->l_mp);
198 if (log->l_sectbb_log) {
199 blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
200 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
204 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
206 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
207 XFS_BUF_ZEROFLAGS(bp);
210 XFS_BUF_PSEMA(bp, PRIBIO);
211 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
212 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
214 if ((error = xfs_bwrite(log->l_mp, bp)))
215 xfs_ioerror_alert("xlog_bwrite", log->l_mp,
216 bp, XFS_BUF_ADDR(bp));
222 * dump debug superblock and log record information
225 xlog_header_check_dump(
227 xlog_rec_header_t *head)
231 cmn_err(CE_DEBUG, "%s: SB : uuid = ", __func__);
232 for (b = 0; b < 16; b++)
233 cmn_err(CE_DEBUG, "%02x", ((__uint8_t *)&mp->m_sb.sb_uuid)[b]);
234 cmn_err(CE_DEBUG, ", fmt = %d\n", XLOG_FMT);
235 cmn_err(CE_DEBUG, " log : uuid = ");
236 for (b = 0; b < 16; b++)
237 cmn_err(CE_DEBUG, "%02x", ((__uint8_t *)&head->h_fs_uuid)[b]);
238 cmn_err(CE_DEBUG, ", fmt = %d\n", be32_to_cpu(head->h_fmt));
241 #define xlog_header_check_dump(mp, head)
245 * check log record header for recovery
248 xlog_header_check_recover(
250 xlog_rec_header_t *head)
252 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
255 * IRIX doesn't write the h_fmt field and leaves it zeroed
256 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
257 * a dirty log created in IRIX.
259 if (unlikely(be32_to_cpu(head->h_fmt) != XLOG_FMT)) {
261 "XFS: dirty log written in incompatible format - can't recover");
262 xlog_header_check_dump(mp, head);
263 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
264 XFS_ERRLEVEL_HIGH, mp);
265 return XFS_ERROR(EFSCORRUPTED);
266 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
268 "XFS: dirty log entry has mismatched uuid - can't recover");
269 xlog_header_check_dump(mp, head);
270 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
271 XFS_ERRLEVEL_HIGH, mp);
272 return XFS_ERROR(EFSCORRUPTED);
278 * read the head block of the log and check the header
281 xlog_header_check_mount(
283 xlog_rec_header_t *head)
285 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
287 if (uuid_is_nil(&head->h_fs_uuid)) {
289 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
290 * h_fs_uuid is nil, we assume this log was last mounted
291 * by IRIX and continue.
293 xlog_warn("XFS: nil uuid in log - IRIX style log");
294 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
295 xlog_warn("XFS: log has mismatched uuid - can't recover");
296 xlog_header_check_dump(mp, head);
297 XFS_ERROR_REPORT("xlog_header_check_mount",
298 XFS_ERRLEVEL_HIGH, mp);
299 return XFS_ERROR(EFSCORRUPTED);
308 if (XFS_BUF_GETERROR(bp)) {
310 * We're not going to bother about retrying
311 * this during recovery. One strike!
313 xfs_ioerror_alert("xlog_recover_iodone",
314 bp->b_mount, bp, XFS_BUF_ADDR(bp));
315 xfs_force_shutdown(bp->b_mount, SHUTDOWN_META_IO_ERROR);
318 XFS_BUF_CLR_IODONE_FUNC(bp);
323 * This routine finds (to an approximation) the first block in the physical
324 * log which contains the given cycle. It uses a binary search algorithm.
325 * Note that the algorithm can not be perfect because the disk will not
326 * necessarily be perfect.
329 xlog_find_cycle_start(
332 xfs_daddr_t first_blk,
333 xfs_daddr_t *last_blk,
341 mid_blk = BLK_AVG(first_blk, *last_blk);
342 while (mid_blk != first_blk && mid_blk != *last_blk) {
343 error = xlog_bread(log, mid_blk, 1, bp, &offset);
346 mid_cycle = xlog_get_cycle(offset);
347 if (mid_cycle == cycle) {
349 /* last_half_cycle == mid_cycle */
352 /* first_half_cycle == mid_cycle */
354 mid_blk = BLK_AVG(first_blk, *last_blk);
356 ASSERT((mid_blk == first_blk && mid_blk+1 == *last_blk) ||
357 (mid_blk == *last_blk && mid_blk-1 == first_blk));
363 * Check that the range of blocks does not contain the cycle number
364 * given. The scan needs to occur from front to back and the ptr into the
365 * region must be updated since a later routine will need to perform another
366 * test. If the region is completely good, we end up returning the same
369 * Set blkno to -1 if we encounter no errors. This is an invalid block number
370 * since we don't ever expect logs to get this large.
373 xlog_find_verify_cycle(
375 xfs_daddr_t start_blk,
377 uint stop_on_cycle_no,
378 xfs_daddr_t *new_blk)
384 xfs_caddr_t buf = NULL;
387 bufblks = 1 << ffs(nbblks);
389 while (!(bp = xlog_get_bp(log, bufblks))) {
390 /* can't get enough memory to do everything in one big buffer */
392 if (bufblks <= log->l_sectbb_log)
396 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
399 bcount = min(bufblks, (start_blk + nbblks - i));
401 error = xlog_bread(log, i, bcount, bp, &buf);
405 for (j = 0; j < bcount; j++) {
406 cycle = xlog_get_cycle(buf);
407 if (cycle == stop_on_cycle_no) {
424 * Potentially backup over partial log record write.
426 * In the typical case, last_blk is the number of the block directly after
427 * a good log record. Therefore, we subtract one to get the block number
428 * of the last block in the given buffer. extra_bblks contains the number
429 * of blocks we would have read on a previous read. This happens when the
430 * last log record is split over the end of the physical log.
432 * extra_bblks is the number of blocks potentially verified on a previous
433 * call to this routine.
436 xlog_find_verify_log_record(
438 xfs_daddr_t start_blk,
439 xfs_daddr_t *last_blk,
444 xfs_caddr_t offset = NULL;
445 xlog_rec_header_t *head = NULL;
448 int num_blks = *last_blk - start_blk;
451 ASSERT(start_blk != 0 || *last_blk != start_blk);
453 if (!(bp = xlog_get_bp(log, num_blks))) {
454 if (!(bp = xlog_get_bp(log, 1)))
458 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
461 offset += ((num_blks - 1) << BBSHIFT);
464 for (i = (*last_blk) - 1; i >= 0; i--) {
466 /* valid log record not found */
468 "XFS: Log inconsistent (didn't find previous header)");
470 error = XFS_ERROR(EIO);
475 error = xlog_bread(log, i, 1, bp, &offset);
480 head = (xlog_rec_header_t *)offset;
482 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(head->h_magicno))
490 * We hit the beginning of the physical log & still no header. Return
491 * to caller. If caller can handle a return of -1, then this routine
492 * will be called again for the end of the physical log.
500 * We have the final block of the good log (the first block
501 * of the log record _before_ the head. So we check the uuid.
503 if ((error = xlog_header_check_mount(log->l_mp, head)))
507 * We may have found a log record header before we expected one.
508 * last_blk will be the 1st block # with a given cycle #. We may end
509 * up reading an entire log record. In this case, we don't want to
510 * reset last_blk. Only when last_blk points in the middle of a log
511 * record do we update last_blk.
513 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
514 uint h_size = be32_to_cpu(head->h_size);
516 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
517 if (h_size % XLOG_HEADER_CYCLE_SIZE)
523 if (*last_blk - i + extra_bblks !=
524 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
533 * Head is defined to be the point of the log where the next log write
534 * write could go. This means that incomplete LR writes at the end are
535 * eliminated when calculating the head. We aren't guaranteed that previous
536 * LR have complete transactions. We only know that a cycle number of
537 * current cycle number -1 won't be present in the log if we start writing
538 * from our current block number.
540 * last_blk contains the block number of the first block with a given
543 * Return: zero if normal, non-zero if error.
548 xfs_daddr_t *return_head_blk)
552 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
554 uint first_half_cycle, last_half_cycle;
556 int error, log_bbnum = log->l_logBBsize;
558 /* Is the end of the log device zeroed? */
559 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
560 *return_head_blk = first_blk;
562 /* Is the whole lot zeroed? */
564 /* Linux XFS shouldn't generate totally zeroed logs -
565 * mkfs etc write a dummy unmount record to a fresh
566 * log so we can store the uuid in there
568 xlog_warn("XFS: totally zeroed log");
573 xlog_warn("XFS: empty log check failed");
577 first_blk = 0; /* get cycle # of 1st block */
578 bp = xlog_get_bp(log, 1);
582 error = xlog_bread(log, 0, 1, bp, &offset);
586 first_half_cycle = xlog_get_cycle(offset);
588 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
589 error = xlog_bread(log, last_blk, 1, bp, &offset);
593 last_half_cycle = xlog_get_cycle(offset);
594 ASSERT(last_half_cycle != 0);
597 * If the 1st half cycle number is equal to the last half cycle number,
598 * then the entire log is stamped with the same cycle number. In this
599 * case, head_blk can't be set to zero (which makes sense). The below
600 * math doesn't work out properly with head_blk equal to zero. Instead,
601 * we set it to log_bbnum which is an invalid block number, but this
602 * value makes the math correct. If head_blk doesn't changed through
603 * all the tests below, *head_blk is set to zero at the very end rather
604 * than log_bbnum. In a sense, log_bbnum and zero are the same block
605 * in a circular file.
607 if (first_half_cycle == last_half_cycle) {
609 * In this case we believe that the entire log should have
610 * cycle number last_half_cycle. We need to scan backwards
611 * from the end verifying that there are no holes still
612 * containing last_half_cycle - 1. If we find such a hole,
613 * then the start of that hole will be the new head. The
614 * simple case looks like
615 * x | x ... | x - 1 | x
616 * Another case that fits this picture would be
617 * x | x + 1 | x ... | x
618 * In this case the head really is somewhere at the end of the
619 * log, as one of the latest writes at the beginning was
622 * x | x + 1 | x ... | x - 1 | x
623 * This is really the combination of the above two cases, and
624 * the head has to end up at the start of the x-1 hole at the
627 * In the 256k log case, we will read from the beginning to the
628 * end of the log and search for cycle numbers equal to x-1.
629 * We don't worry about the x+1 blocks that we encounter,
630 * because we know that they cannot be the head since the log
633 head_blk = log_bbnum;
634 stop_on_cycle = last_half_cycle - 1;
637 * In this case we want to find the first block with cycle
638 * number matching last_half_cycle. We expect the log to be
641 * The first block with cycle number x (last_half_cycle) will
642 * be where the new head belongs. First we do a binary search
643 * for the first occurrence of last_half_cycle. The binary
644 * search may not be totally accurate, so then we scan back
645 * from there looking for occurrences of last_half_cycle before
646 * us. If that backwards scan wraps around the beginning of
647 * the log, then we look for occurrences of last_half_cycle - 1
648 * at the end of the log. The cases we're looking for look
650 * x + 1 ... | x | x + 1 | x ...
651 * ^ binary search stopped here
653 * x + 1 ... | x ... | x - 1 | x
654 * <---------> less than scan distance
656 stop_on_cycle = last_half_cycle;
657 if ((error = xlog_find_cycle_start(log, bp, first_blk,
658 &head_blk, last_half_cycle)))
663 * Now validate the answer. Scan back some number of maximum possible
664 * blocks and make sure each one has the expected cycle number. The
665 * maximum is determined by the total possible amount of buffering
666 * in the in-core log. The following number can be made tighter if
667 * we actually look at the block size of the filesystem.
669 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
670 if (head_blk >= num_scan_bblks) {
672 * We are guaranteed that the entire check can be performed
675 start_blk = head_blk - num_scan_bblks;
676 if ((error = xlog_find_verify_cycle(log,
677 start_blk, num_scan_bblks,
678 stop_on_cycle, &new_blk)))
682 } else { /* need to read 2 parts of log */
684 * We are going to scan backwards in the log in two parts.
685 * First we scan the physical end of the log. In this part
686 * of the log, we are looking for blocks with cycle number
687 * last_half_cycle - 1.
688 * If we find one, then we know that the log starts there, as
689 * we've found a hole that didn't get written in going around
690 * the end of the physical log. The simple case for this is
691 * x + 1 ... | x ... | x - 1 | x
692 * <---------> less than scan distance
693 * If all of the blocks at the end of the log have cycle number
694 * last_half_cycle, then we check the blocks at the start of
695 * the log looking for occurrences of last_half_cycle. If we
696 * find one, then our current estimate for the location of the
697 * first occurrence of last_half_cycle is wrong and we move
698 * back to the hole we've found. This case looks like
699 * x + 1 ... | x | x + 1 | x ...
700 * ^ binary search stopped here
701 * Another case we need to handle that only occurs in 256k
703 * x + 1 ... | x ... | x+1 | x ...
704 * ^ binary search stops here
705 * In a 256k log, the scan at the end of the log will see the
706 * x + 1 blocks. We need to skip past those since that is
707 * certainly not the head of the log. By searching for
708 * last_half_cycle-1 we accomplish that.
710 start_blk = log_bbnum - num_scan_bblks + head_blk;
711 ASSERT(head_blk <= INT_MAX &&
712 (xfs_daddr_t) num_scan_bblks - head_blk >= 0);
713 if ((error = xlog_find_verify_cycle(log, start_blk,
714 num_scan_bblks - (int)head_blk,
715 (stop_on_cycle - 1), &new_blk)))
723 * Scan beginning of log now. The last part of the physical
724 * log is good. This scan needs to verify that it doesn't find
725 * the last_half_cycle.
728 ASSERT(head_blk <= INT_MAX);
729 if ((error = xlog_find_verify_cycle(log,
730 start_blk, (int)head_blk,
731 stop_on_cycle, &new_blk)))
739 * Now we need to make sure head_blk is not pointing to a block in
740 * the middle of a log record.
742 num_scan_bblks = XLOG_REC_SHIFT(log);
743 if (head_blk >= num_scan_bblks) {
744 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
746 /* start ptr at last block ptr before head_blk */
747 if ((error = xlog_find_verify_log_record(log, start_blk,
748 &head_blk, 0)) == -1) {
749 error = XFS_ERROR(EIO);
755 ASSERT(head_blk <= INT_MAX);
756 if ((error = xlog_find_verify_log_record(log, start_blk,
757 &head_blk, 0)) == -1) {
758 /* We hit the beginning of the log during our search */
759 start_blk = log_bbnum - num_scan_bblks + head_blk;
761 ASSERT(start_blk <= INT_MAX &&
762 (xfs_daddr_t) log_bbnum-start_blk >= 0);
763 ASSERT(head_blk <= INT_MAX);
764 if ((error = xlog_find_verify_log_record(log,
766 (int)head_blk)) == -1) {
767 error = XFS_ERROR(EIO);
771 if (new_blk != log_bbnum)
778 if (head_blk == log_bbnum)
779 *return_head_blk = 0;
781 *return_head_blk = head_blk;
783 * When returning here, we have a good block number. Bad block
784 * means that during a previous crash, we didn't have a clean break
785 * from cycle number N to cycle number N-1. In this case, we need
786 * to find the first block with cycle number N-1.
794 xlog_warn("XFS: failed to find log head");
799 * Find the sync block number or the tail of the log.
801 * This will be the block number of the last record to have its
802 * associated buffers synced to disk. Every log record header has
803 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
804 * to get a sync block number. The only concern is to figure out which
805 * log record header to believe.
807 * The following algorithm uses the log record header with the largest
808 * lsn. The entire log record does not need to be valid. We only care
809 * that the header is valid.
811 * We could speed up search by using current head_blk buffer, but it is not
817 xfs_daddr_t *head_blk,
818 xfs_daddr_t *tail_blk)
820 xlog_rec_header_t *rhead;
821 xlog_op_header_t *op_head;
822 xfs_caddr_t offset = NULL;
825 xfs_daddr_t umount_data_blk;
826 xfs_daddr_t after_umount_blk;
833 * Find previous log record
835 if ((error = xlog_find_head(log, head_blk)))
838 bp = xlog_get_bp(log, 1);
841 if (*head_blk == 0) { /* special case */
842 error = xlog_bread(log, 0, 1, bp, &offset);
846 if (xlog_get_cycle(offset) == 0) {
848 /* leave all other log inited values alone */
854 * Search backwards looking for log record header block
856 ASSERT(*head_blk < INT_MAX);
857 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
858 error = xlog_bread(log, i, 1, bp, &offset);
862 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(*(__be32 *)offset)) {
868 * If we haven't found the log record header block, start looking
869 * again from the end of the physical log. XXXmiken: There should be
870 * a check here to make sure we didn't search more than N blocks in
874 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
875 error = xlog_bread(log, i, 1, bp, &offset);
879 if (XLOG_HEADER_MAGIC_NUM ==
880 be32_to_cpu(*(__be32 *)offset)) {
887 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
889 return XFS_ERROR(EIO);
892 /* find blk_no of tail of log */
893 rhead = (xlog_rec_header_t *)offset;
894 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
897 * Reset log values according to the state of the log when we
898 * crashed. In the case where head_blk == 0, we bump curr_cycle
899 * one because the next write starts a new cycle rather than
900 * continuing the cycle of the last good log record. At this
901 * point we have guaranteed that all partial log records have been
902 * accounted for. Therefore, we know that the last good log record
903 * written was complete and ended exactly on the end boundary
904 * of the physical log.
906 log->l_prev_block = i;
907 log->l_curr_block = (int)*head_blk;
908 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
911 log->l_tail_lsn = be64_to_cpu(rhead->h_tail_lsn);
912 log->l_last_sync_lsn = be64_to_cpu(rhead->h_lsn);
913 log->l_grant_reserve_cycle = log->l_curr_cycle;
914 log->l_grant_reserve_bytes = BBTOB(log->l_curr_block);
915 log->l_grant_write_cycle = log->l_curr_cycle;
916 log->l_grant_write_bytes = BBTOB(log->l_curr_block);
919 * Look for unmount record. If we find it, then we know there
920 * was a clean unmount. Since 'i' could be the last block in
921 * the physical log, we convert to a log block before comparing
924 * Save the current tail lsn to use to pass to
925 * xlog_clear_stale_blocks() below. We won't want to clear the
926 * unmount record if there is one, so we pass the lsn of the
927 * unmount record rather than the block after it.
929 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
930 int h_size = be32_to_cpu(rhead->h_size);
931 int h_version = be32_to_cpu(rhead->h_version);
933 if ((h_version & XLOG_VERSION_2) &&
934 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
935 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
936 if (h_size % XLOG_HEADER_CYCLE_SIZE)
944 after_umount_blk = (i + hblks + (int)
945 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
946 tail_lsn = log->l_tail_lsn;
947 if (*head_blk == after_umount_blk &&
948 be32_to_cpu(rhead->h_num_logops) == 1) {
949 umount_data_blk = (i + hblks) % log->l_logBBsize;
950 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
954 op_head = (xlog_op_header_t *)offset;
955 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
957 * Set tail and last sync so that newly written
958 * log records will point recovery to after the
959 * current unmount record.
962 xlog_assign_lsn(log->l_curr_cycle,
964 log->l_last_sync_lsn =
965 xlog_assign_lsn(log->l_curr_cycle,
967 *tail_blk = after_umount_blk;
970 * Note that the unmount was clean. If the unmount
971 * was not clean, we need to know this to rebuild the
972 * superblock counters from the perag headers if we
973 * have a filesystem using non-persistent counters.
975 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
980 * Make sure that there are no blocks in front of the head
981 * with the same cycle number as the head. This can happen
982 * because we allow multiple outstanding log writes concurrently,
983 * and the later writes might make it out before earlier ones.
985 * We use the lsn from before modifying it so that we'll never
986 * overwrite the unmount record after a clean unmount.
988 * Do this only if we are going to recover the filesystem
990 * NOTE: This used to say "if (!readonly)"
991 * However on Linux, we can & do recover a read-only filesystem.
992 * We only skip recovery if NORECOVERY is specified on mount,
993 * in which case we would not be here.
995 * But... if the -device- itself is readonly, just skip this.
996 * We can't recover this device anyway, so it won't matter.
998 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) {
999 error = xlog_clear_stale_blocks(log, tail_lsn);
1007 xlog_warn("XFS: failed to locate log tail");
1012 * Is the log zeroed at all?
1014 * The last binary search should be changed to perform an X block read
1015 * once X becomes small enough. You can then search linearly through
1016 * the X blocks. This will cut down on the number of reads we need to do.
1018 * If the log is partially zeroed, this routine will pass back the blkno
1019 * of the first block with cycle number 0. It won't have a complete LR
1023 * 0 => the log is completely written to
1024 * -1 => use *blk_no as the first block of the log
1025 * >0 => error has occurred
1030 xfs_daddr_t *blk_no)
1034 uint first_cycle, last_cycle;
1035 xfs_daddr_t new_blk, last_blk, start_blk;
1036 xfs_daddr_t num_scan_bblks;
1037 int error, log_bbnum = log->l_logBBsize;
1041 /* check totally zeroed log */
1042 bp = xlog_get_bp(log, 1);
1045 error = xlog_bread(log, 0, 1, bp, &offset);
1049 first_cycle = xlog_get_cycle(offset);
1050 if (first_cycle == 0) { /* completely zeroed log */
1056 /* check partially zeroed log */
1057 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1061 last_cycle = xlog_get_cycle(offset);
1062 if (last_cycle != 0) { /* log completely written to */
1065 } else if (first_cycle != 1) {
1067 * If the cycle of the last block is zero, the cycle of
1068 * the first block must be 1. If it's not, maybe we're
1069 * not looking at a log... Bail out.
1071 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1072 return XFS_ERROR(EINVAL);
1075 /* we have a partially zeroed log */
1076 last_blk = log_bbnum-1;
1077 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1081 * Validate the answer. Because there is no way to guarantee that
1082 * the entire log is made up of log records which are the same size,
1083 * we scan over the defined maximum blocks. At this point, the maximum
1084 * is not chosen to mean anything special. XXXmiken
1086 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1087 ASSERT(num_scan_bblks <= INT_MAX);
1089 if (last_blk < num_scan_bblks)
1090 num_scan_bblks = last_blk;
1091 start_blk = last_blk - num_scan_bblks;
1094 * We search for any instances of cycle number 0 that occur before
1095 * our current estimate of the head. What we're trying to detect is
1096 * 1 ... | 0 | 1 | 0...
1097 * ^ binary search ends here
1099 if ((error = xlog_find_verify_cycle(log, start_blk,
1100 (int)num_scan_bblks, 0, &new_blk)))
1106 * Potentially backup over partial log record write. We don't need
1107 * to search the end of the log because we know it is zero.
1109 if ((error = xlog_find_verify_log_record(log, start_blk,
1110 &last_blk, 0)) == -1) {
1111 error = XFS_ERROR(EIO);
1125 * These are simple subroutines used by xlog_clear_stale_blocks() below
1126 * to initialize a buffer full of empty log record headers and write
1127 * them into the log.
1138 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1140 memset(buf, 0, BBSIZE);
1141 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1142 recp->h_cycle = cpu_to_be32(cycle);
1143 recp->h_version = cpu_to_be32(
1144 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1145 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1146 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1147 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1148 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1152 xlog_write_log_records(
1163 int sectbb = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
1164 int end_block = start_block + blocks;
1169 bufblks = 1 << ffs(blocks);
1170 while (!(bp = xlog_get_bp(log, bufblks))) {
1172 if (bufblks <= log->l_sectbb_log)
1176 /* We may need to do a read at the start to fill in part of
1177 * the buffer in the starting sector not covered by the first
1180 balign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, start_block);
1181 if (balign != start_block) {
1182 error = xlog_bread_noalign(log, start_block, 1, bp);
1186 j = start_block - balign;
1189 for (i = start_block; i < end_block; i += bufblks) {
1190 int bcount, endcount;
1192 bcount = min(bufblks, end_block - start_block);
1193 endcount = bcount - j;
1195 /* We may need to do a read at the end to fill in part of
1196 * the buffer in the final sector not covered by the write.
1197 * If this is the same sector as the above read, skip it.
1199 ealign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, end_block);
1200 if (j == 0 && (start_block + endcount > ealign)) {
1201 offset = XFS_BUF_PTR(bp);
1202 balign = BBTOB(ealign - start_block);
1203 error = XFS_BUF_SET_PTR(bp, offset + balign,
1208 error = xlog_bread_noalign(log, ealign, sectbb, bp);
1212 error = XFS_BUF_SET_PTR(bp, offset, bufblks);
1217 offset = xlog_align(log, start_block, endcount, bp);
1218 for (; j < endcount; j++) {
1219 xlog_add_record(log, offset, cycle, i+j,
1220 tail_cycle, tail_block);
1223 error = xlog_bwrite(log, start_block, endcount, bp);
1226 start_block += endcount;
1236 * This routine is called to blow away any incomplete log writes out
1237 * in front of the log head. We do this so that we won't become confused
1238 * if we come up, write only a little bit more, and then crash again.
1239 * If we leave the partial log records out there, this situation could
1240 * cause us to think those partial writes are valid blocks since they
1241 * have the current cycle number. We get rid of them by overwriting them
1242 * with empty log records with the old cycle number rather than the
1245 * The tail lsn is passed in rather than taken from
1246 * the log so that we will not write over the unmount record after a
1247 * clean unmount in a 512 block log. Doing so would leave the log without
1248 * any valid log records in it until a new one was written. If we crashed
1249 * during that time we would not be able to recover.
1252 xlog_clear_stale_blocks(
1256 int tail_cycle, head_cycle;
1257 int tail_block, head_block;
1258 int tail_distance, max_distance;
1262 tail_cycle = CYCLE_LSN(tail_lsn);
1263 tail_block = BLOCK_LSN(tail_lsn);
1264 head_cycle = log->l_curr_cycle;
1265 head_block = log->l_curr_block;
1268 * Figure out the distance between the new head of the log
1269 * and the tail. We want to write over any blocks beyond the
1270 * head that we may have written just before the crash, but
1271 * we don't want to overwrite the tail of the log.
1273 if (head_cycle == tail_cycle) {
1275 * The tail is behind the head in the physical log,
1276 * so the distance from the head to the tail is the
1277 * distance from the head to the end of the log plus
1278 * the distance from the beginning of the log to the
1281 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1282 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1283 XFS_ERRLEVEL_LOW, log->l_mp);
1284 return XFS_ERROR(EFSCORRUPTED);
1286 tail_distance = tail_block + (log->l_logBBsize - head_block);
1289 * The head is behind the tail in the physical log,
1290 * so the distance from the head to the tail is just
1291 * the tail block minus the head block.
1293 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1294 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1295 XFS_ERRLEVEL_LOW, log->l_mp);
1296 return XFS_ERROR(EFSCORRUPTED);
1298 tail_distance = tail_block - head_block;
1302 * If the head is right up against the tail, we can't clear
1305 if (tail_distance <= 0) {
1306 ASSERT(tail_distance == 0);
1310 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1312 * Take the smaller of the maximum amount of outstanding I/O
1313 * we could have and the distance to the tail to clear out.
1314 * We take the smaller so that we don't overwrite the tail and
1315 * we don't waste all day writing from the head to the tail
1318 max_distance = MIN(max_distance, tail_distance);
1320 if ((head_block + max_distance) <= log->l_logBBsize) {
1322 * We can stomp all the blocks we need to without
1323 * wrapping around the end of the log. Just do it
1324 * in a single write. Use the cycle number of the
1325 * current cycle minus one so that the log will look like:
1328 error = xlog_write_log_records(log, (head_cycle - 1),
1329 head_block, max_distance, tail_cycle,
1335 * We need to wrap around the end of the physical log in
1336 * order to clear all the blocks. Do it in two separate
1337 * I/Os. The first write should be from the head to the
1338 * end of the physical log, and it should use the current
1339 * cycle number minus one just like above.
1341 distance = log->l_logBBsize - head_block;
1342 error = xlog_write_log_records(log, (head_cycle - 1),
1343 head_block, distance, tail_cycle,
1350 * Now write the blocks at the start of the physical log.
1351 * This writes the remainder of the blocks we want to clear.
1352 * It uses the current cycle number since we're now on the
1353 * same cycle as the head so that we get:
1354 * n ... n ... | n - 1 ...
1355 * ^^^^^ blocks we're writing
1357 distance = max_distance - (log->l_logBBsize - head_block);
1358 error = xlog_write_log_records(log, head_cycle, 0, distance,
1359 tail_cycle, tail_block);
1367 /******************************************************************************
1369 * Log recover routines
1371 ******************************************************************************
1374 STATIC xlog_recover_t *
1375 xlog_recover_find_tid(
1379 xlog_recover_t *p = q;
1382 if (p->r_log_tid == tid)
1390 xlog_recover_put_hashq(
1392 xlog_recover_t *trans)
1399 xlog_recover_add_item(
1400 xlog_recover_item_t **itemq)
1402 xlog_recover_item_t *item;
1404 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1405 xlog_recover_insert_item_backq(itemq, item);
1409 xlog_recover_add_to_cont_trans(
1410 xlog_recover_t *trans,
1414 xlog_recover_item_t *item;
1415 xfs_caddr_t ptr, old_ptr;
1418 item = trans->r_itemq;
1420 /* finish copying rest of trans header */
1421 xlog_recover_add_item(&trans->r_itemq);
1422 ptr = (xfs_caddr_t) &trans->r_theader +
1423 sizeof(xfs_trans_header_t) - len;
1424 memcpy(ptr, dp, len); /* d, s, l */
1427 item = item->ri_prev;
1429 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1430 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1432 ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
1433 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1434 item->ri_buf[item->ri_cnt-1].i_len += len;
1435 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1440 * The next region to add is the start of a new region. It could be
1441 * a whole region or it could be the first part of a new region. Because
1442 * of this, the assumption here is that the type and size fields of all
1443 * format structures fit into the first 32 bits of the structure.
1445 * This works because all regions must be 32 bit aligned. Therefore, we
1446 * either have both fields or we have neither field. In the case we have
1447 * neither field, the data part of the region is zero length. We only have
1448 * a log_op_header and can throw away the header since a new one will appear
1449 * later. If we have at least 4 bytes, then we can determine how many regions
1450 * will appear in the current log item.
1453 xlog_recover_add_to_trans(
1454 xlog_recover_t *trans,
1458 xfs_inode_log_format_t *in_f; /* any will do */
1459 xlog_recover_item_t *item;
1464 item = trans->r_itemq;
1466 /* we need to catch log corruptions here */
1467 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
1468 xlog_warn("XFS: xlog_recover_add_to_trans: "
1469 "bad header magic number");
1471 return XFS_ERROR(EIO);
1473 if (len == sizeof(xfs_trans_header_t))
1474 xlog_recover_add_item(&trans->r_itemq);
1475 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1479 ptr = kmem_alloc(len, KM_SLEEP);
1480 memcpy(ptr, dp, len);
1481 in_f = (xfs_inode_log_format_t *)ptr;
1483 if (item->ri_prev->ri_total != 0 &&
1484 item->ri_prev->ri_total == item->ri_prev->ri_cnt) {
1485 xlog_recover_add_item(&trans->r_itemq);
1487 item = trans->r_itemq;
1488 item = item->ri_prev;
1490 if (item->ri_total == 0) { /* first region to be added */
1491 if (in_f->ilf_size == 0 ||
1492 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
1494 "XFS: bad number of regions (%d) in inode log format",
1497 return XFS_ERROR(EIO);
1500 item->ri_total = in_f->ilf_size;
1502 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1505 ASSERT(item->ri_total > item->ri_cnt);
1506 /* Description region is ri_buf[0] */
1507 item->ri_buf[item->ri_cnt].i_addr = ptr;
1508 item->ri_buf[item->ri_cnt].i_len = len;
1514 xlog_recover_new_tid(
1519 xlog_recover_t *trans;
1521 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1522 trans->r_log_tid = tid;
1524 xlog_recover_put_hashq(q, trans);
1528 xlog_recover_unlink_tid(
1530 xlog_recover_t *trans)
1535 ASSERT(trans != NULL);
1541 if (tp->r_next == trans) {
1549 "XFS: xlog_recover_unlink_tid: trans not found");
1551 return XFS_ERROR(EIO);
1553 tp->r_next = tp->r_next->r_next;
1559 xlog_recover_insert_item_backq(
1560 xlog_recover_item_t **q,
1561 xlog_recover_item_t *item)
1564 item->ri_prev = item->ri_next = item;
1568 item->ri_prev = (*q)->ri_prev;
1569 (*q)->ri_prev = item;
1570 item->ri_prev->ri_next = item;
1575 xlog_recover_insert_item_frontq(
1576 xlog_recover_item_t **q,
1577 xlog_recover_item_t *item)
1579 xlog_recover_insert_item_backq(q, item);
1584 xlog_recover_reorder_trans(
1585 xlog_recover_t *trans)
1587 xlog_recover_item_t *first_item, *itemq, *itemq_next;
1588 xfs_buf_log_format_t *buf_f;
1591 first_item = itemq = trans->r_itemq;
1592 trans->r_itemq = NULL;
1594 itemq_next = itemq->ri_next;
1595 buf_f = (xfs_buf_log_format_t *)itemq->ri_buf[0].i_addr;
1597 switch (ITEM_TYPE(itemq)) {
1599 flags = buf_f->blf_flags;
1600 if (!(flags & XFS_BLI_CANCEL)) {
1601 xlog_recover_insert_item_frontq(&trans->r_itemq,
1607 case XFS_LI_QUOTAOFF:
1610 xlog_recover_insert_item_backq(&trans->r_itemq, itemq);
1614 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1616 return XFS_ERROR(EIO);
1619 } while (first_item != itemq);
1624 * Build up the table of buf cancel records so that we don't replay
1625 * cancelled data in the second pass. For buffer records that are
1626 * not cancel records, there is nothing to do here so we just return.
1628 * If we get a cancel record which is already in the table, this indicates
1629 * that the buffer was cancelled multiple times. In order to ensure
1630 * that during pass 2 we keep the record in the table until we reach its
1631 * last occurrence in the log, we keep a reference count in the cancel
1632 * record in the table to tell us how many times we expect to see this
1633 * record during the second pass.
1636 xlog_recover_do_buffer_pass1(
1638 xfs_buf_log_format_t *buf_f)
1640 xfs_buf_cancel_t *bcp;
1641 xfs_buf_cancel_t *nextp;
1642 xfs_buf_cancel_t *prevp;
1643 xfs_buf_cancel_t **bucket;
1644 xfs_daddr_t blkno = 0;
1648 switch (buf_f->blf_type) {
1650 blkno = buf_f->blf_blkno;
1651 len = buf_f->blf_len;
1652 flags = buf_f->blf_flags;
1657 * If this isn't a cancel buffer item, then just return.
1659 if (!(flags & XFS_BLI_CANCEL))
1663 * Insert an xfs_buf_cancel record into the hash table of
1664 * them. If there is already an identical record, bump
1665 * its reference count.
1667 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1668 XLOG_BC_TABLE_SIZE];
1670 * If the hash bucket is empty then just insert a new record into
1673 if (*bucket == NULL) {
1674 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1676 bcp->bc_blkno = blkno;
1678 bcp->bc_refcount = 1;
1679 bcp->bc_next = NULL;
1685 * The hash bucket is not empty, so search for duplicates of our
1686 * record. If we find one them just bump its refcount. If not
1687 * then add us at the end of the list.
1691 while (nextp != NULL) {
1692 if (nextp->bc_blkno == blkno && nextp->bc_len == len) {
1693 nextp->bc_refcount++;
1697 nextp = nextp->bc_next;
1699 ASSERT(prevp != NULL);
1700 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1702 bcp->bc_blkno = blkno;
1704 bcp->bc_refcount = 1;
1705 bcp->bc_next = NULL;
1706 prevp->bc_next = bcp;
1710 * Check to see whether the buffer being recovered has a corresponding
1711 * entry in the buffer cancel record table. If it does then return 1
1712 * so that it will be cancelled, otherwise return 0. If the buffer is
1713 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
1714 * the refcount on the entry in the table and remove it from the table
1715 * if this is the last reference.
1717 * We remove the cancel record from the table when we encounter its
1718 * last occurrence in the log so that if the same buffer is re-used
1719 * again after its last cancellation we actually replay the changes
1720 * made at that point.
1723 xlog_check_buffer_cancelled(
1729 xfs_buf_cancel_t *bcp;
1730 xfs_buf_cancel_t *prevp;
1731 xfs_buf_cancel_t **bucket;
1733 if (log->l_buf_cancel_table == NULL) {
1735 * There is nothing in the table built in pass one,
1736 * so this buffer must not be cancelled.
1738 ASSERT(!(flags & XFS_BLI_CANCEL));
1742 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1743 XLOG_BC_TABLE_SIZE];
1747 * There is no corresponding entry in the table built
1748 * in pass one, so this buffer has not been cancelled.
1750 ASSERT(!(flags & XFS_BLI_CANCEL));
1755 * Search for an entry in the buffer cancel table that
1756 * matches our buffer.
1759 while (bcp != NULL) {
1760 if (bcp->bc_blkno == blkno && bcp->bc_len == len) {
1762 * We've go a match, so return 1 so that the
1763 * recovery of this buffer is cancelled.
1764 * If this buffer is actually a buffer cancel
1765 * log item, then decrement the refcount on the
1766 * one in the table and remove it if this is the
1769 if (flags & XFS_BLI_CANCEL) {
1771 if (bcp->bc_refcount == 0) {
1772 if (prevp == NULL) {
1773 *bucket = bcp->bc_next;
1775 prevp->bc_next = bcp->bc_next;
1786 * We didn't find a corresponding entry in the table, so
1787 * return 0 so that the buffer is NOT cancelled.
1789 ASSERT(!(flags & XFS_BLI_CANCEL));
1794 xlog_recover_do_buffer_pass2(
1796 xfs_buf_log_format_t *buf_f)
1798 xfs_daddr_t blkno = 0;
1802 switch (buf_f->blf_type) {
1804 blkno = buf_f->blf_blkno;
1805 flags = buf_f->blf_flags;
1806 len = buf_f->blf_len;
1810 return xlog_check_buffer_cancelled(log, blkno, len, flags);
1814 * Perform recovery for a buffer full of inodes. In these buffers,
1815 * the only data which should be recovered is that which corresponds
1816 * to the di_next_unlinked pointers in the on disk inode structures.
1817 * The rest of the data for the inodes is always logged through the
1818 * inodes themselves rather than the inode buffer and is recovered
1819 * in xlog_recover_do_inode_trans().
1821 * The only time when buffers full of inodes are fully recovered is
1822 * when the buffer is full of newly allocated inodes. In this case
1823 * the buffer will not be marked as an inode buffer and so will be
1824 * sent to xlog_recover_do_reg_buffer() below during recovery.
1827 xlog_recover_do_inode_buffer(
1829 xlog_recover_item_t *item,
1831 xfs_buf_log_format_t *buf_f)
1839 int next_unlinked_offset;
1841 xfs_agino_t *logged_nextp;
1842 xfs_agino_t *buffer_nextp;
1843 unsigned int *data_map = NULL;
1844 unsigned int map_size = 0;
1846 switch (buf_f->blf_type) {
1848 data_map = buf_f->blf_data_map;
1849 map_size = buf_f->blf_map_size;
1853 * Set the variables corresponding to the current region to
1854 * 0 so that we'll initialize them on the first pass through
1862 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1863 for (i = 0; i < inodes_per_buf; i++) {
1864 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1865 offsetof(xfs_dinode_t, di_next_unlinked);
1867 while (next_unlinked_offset >=
1868 (reg_buf_offset + reg_buf_bytes)) {
1870 * The next di_next_unlinked field is beyond
1871 * the current logged region. Find the next
1872 * logged region that contains or is beyond
1873 * the current di_next_unlinked field.
1876 bit = xfs_next_bit(data_map, map_size, bit);
1879 * If there are no more logged regions in the
1880 * buffer, then we're done.
1886 nbits = xfs_contig_bits(data_map, map_size,
1889 reg_buf_offset = bit << XFS_BLI_SHIFT;
1890 reg_buf_bytes = nbits << XFS_BLI_SHIFT;
1895 * If the current logged region starts after the current
1896 * di_next_unlinked field, then move on to the next
1897 * di_next_unlinked field.
1899 if (next_unlinked_offset < reg_buf_offset) {
1903 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1904 ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0);
1905 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1908 * The current logged region contains a copy of the
1909 * current di_next_unlinked field. Extract its value
1910 * and copy it to the buffer copy.
1912 logged_nextp = (xfs_agino_t *)
1913 ((char *)(item->ri_buf[item_index].i_addr) +
1914 (next_unlinked_offset - reg_buf_offset));
1915 if (unlikely(*logged_nextp == 0)) {
1916 xfs_fs_cmn_err(CE_ALERT, mp,
1917 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1919 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1920 XFS_ERRLEVEL_LOW, mp);
1921 return XFS_ERROR(EFSCORRUPTED);
1924 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1925 next_unlinked_offset);
1926 *buffer_nextp = *logged_nextp;
1933 * Perform a 'normal' buffer recovery. Each logged region of the
1934 * buffer should be copied over the corresponding region in the
1935 * given buffer. The bitmap in the buf log format structure indicates
1936 * where to place the logged data.
1940 xlog_recover_do_reg_buffer(
1941 xlog_recover_item_t *item,
1943 xfs_buf_log_format_t *buf_f)
1948 unsigned int *data_map = NULL;
1949 unsigned int map_size = 0;
1952 switch (buf_f->blf_type) {
1954 data_map = buf_f->blf_data_map;
1955 map_size = buf_f->blf_map_size;
1959 i = 1; /* 0 is the buf format structure */
1961 bit = xfs_next_bit(data_map, map_size, bit);
1964 nbits = xfs_contig_bits(data_map, map_size, bit);
1966 ASSERT(item->ri_buf[i].i_addr != NULL);
1967 ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0);
1968 ASSERT(XFS_BUF_COUNT(bp) >=
1969 ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT));
1972 * Do a sanity check if this is a dquot buffer. Just checking
1973 * the first dquot in the buffer should do. XXXThis is
1974 * probably a good thing to do for other buf types also.
1977 if (buf_f->blf_flags &
1978 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
1979 if (item->ri_buf[i].i_addr == NULL) {
1981 "XFS: NULL dquot in %s.", __func__);
1984 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
1986 "XFS: dquot too small (%d) in %s.",
1987 item->ri_buf[i].i_len, __func__);
1990 error = xfs_qm_dqcheck((xfs_disk_dquot_t *)
1991 item->ri_buf[i].i_addr,
1992 -1, 0, XFS_QMOPT_DOWARN,
1993 "dquot_buf_recover");
1998 memcpy(xfs_buf_offset(bp,
1999 (uint)bit << XFS_BLI_SHIFT), /* dest */
2000 item->ri_buf[i].i_addr, /* source */
2001 nbits<<XFS_BLI_SHIFT); /* length */
2007 /* Shouldn't be any more regions */
2008 ASSERT(i == item->ri_total);
2012 * Do some primitive error checking on ondisk dquot data structures.
2016 xfs_disk_dquot_t *ddq,
2018 uint type, /* used only when IO_dorepair is true */
2022 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
2026 * We can encounter an uninitialized dquot buffer for 2 reasons:
2027 * 1. If we crash while deleting the quotainode(s), and those blks got
2028 * used for user data. This is because we take the path of regular
2029 * file deletion; however, the size field of quotainodes is never
2030 * updated, so all the tricks that we play in itruncate_finish
2031 * don't quite matter.
2033 * 2. We don't play the quota buffers when there's a quotaoff logitem.
2034 * But the allocation will be replayed so we'll end up with an
2035 * uninitialized quota block.
2037 * This is all fine; things are still consistent, and we haven't lost
2038 * any quota information. Just don't complain about bad dquot blks.
2040 if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) {
2041 if (flags & XFS_QMOPT_DOWARN)
2043 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
2044 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
2047 if (ddq->d_version != XFS_DQUOT_VERSION) {
2048 if (flags & XFS_QMOPT_DOWARN)
2050 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
2051 str, id, ddq->d_version, XFS_DQUOT_VERSION);
2055 if (ddq->d_flags != XFS_DQ_USER &&
2056 ddq->d_flags != XFS_DQ_PROJ &&
2057 ddq->d_flags != XFS_DQ_GROUP) {
2058 if (flags & XFS_QMOPT_DOWARN)
2060 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
2061 str, id, ddq->d_flags);
2065 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
2066 if (flags & XFS_QMOPT_DOWARN)
2068 "%s : ondisk-dquot 0x%p, ID mismatch: "
2069 "0x%x expected, found id 0x%x",
2070 str, ddq, id, be32_to_cpu(ddq->d_id));
2074 if (!errs && ddq->d_id) {
2075 if (ddq->d_blk_softlimit &&
2076 be64_to_cpu(ddq->d_bcount) >=
2077 be64_to_cpu(ddq->d_blk_softlimit)) {
2078 if (!ddq->d_btimer) {
2079 if (flags & XFS_QMOPT_DOWARN)
2081 "%s : Dquot ID 0x%x (0x%p) "
2082 "BLK TIMER NOT STARTED",
2083 str, (int)be32_to_cpu(ddq->d_id), ddq);
2087 if (ddq->d_ino_softlimit &&
2088 be64_to_cpu(ddq->d_icount) >=
2089 be64_to_cpu(ddq->d_ino_softlimit)) {
2090 if (!ddq->d_itimer) {
2091 if (flags & XFS_QMOPT_DOWARN)
2093 "%s : Dquot ID 0x%x (0x%p) "
2094 "INODE TIMER NOT STARTED",
2095 str, (int)be32_to_cpu(ddq->d_id), ddq);
2099 if (ddq->d_rtb_softlimit &&
2100 be64_to_cpu(ddq->d_rtbcount) >=
2101 be64_to_cpu(ddq->d_rtb_softlimit)) {
2102 if (!ddq->d_rtbtimer) {
2103 if (flags & XFS_QMOPT_DOWARN)
2105 "%s : Dquot ID 0x%x (0x%p) "
2106 "RTBLK TIMER NOT STARTED",
2107 str, (int)be32_to_cpu(ddq->d_id), ddq);
2113 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2116 if (flags & XFS_QMOPT_DOWARN)
2117 cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id);
2120 * Typically, a repair is only requested by quotacheck.
2123 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2124 memset(d, 0, sizeof(xfs_dqblk_t));
2126 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2127 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2128 d->dd_diskdq.d_flags = type;
2129 d->dd_diskdq.d_id = cpu_to_be32(id);
2135 * Perform a dquot buffer recovery.
2136 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2137 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2138 * Else, treat it as a regular buffer and do recovery.
2141 xlog_recover_do_dquot_buffer(
2144 xlog_recover_item_t *item,
2146 xfs_buf_log_format_t *buf_f)
2151 * Filesystems are required to send in quota flags at mount time.
2153 if (mp->m_qflags == 0) {
2158 if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF)
2159 type |= XFS_DQ_USER;
2160 if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF)
2161 type |= XFS_DQ_PROJ;
2162 if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF)
2163 type |= XFS_DQ_GROUP;
2165 * This type of quotas was turned off, so ignore this buffer
2167 if (log->l_quotaoffs_flag & type)
2170 xlog_recover_do_reg_buffer(item, bp, buf_f);
2174 * This routine replays a modification made to a buffer at runtime.
2175 * There are actually two types of buffer, regular and inode, which
2176 * are handled differently. Inode buffers are handled differently
2177 * in that we only recover a specific set of data from them, namely
2178 * the inode di_next_unlinked fields. This is because all other inode
2179 * data is actually logged via inode records and any data we replay
2180 * here which overlaps that may be stale.
2182 * When meta-data buffers are freed at run time we log a buffer item
2183 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
2184 * of the buffer in the log should not be replayed at recovery time.
2185 * This is so that if the blocks covered by the buffer are reused for
2186 * file data before we crash we don't end up replaying old, freed
2187 * meta-data into a user's file.
2189 * To handle the cancellation of buffer log items, we make two passes
2190 * over the log during recovery. During the first we build a table of
2191 * those buffers which have been cancelled, and during the second we
2192 * only replay those buffers which do not have corresponding cancel
2193 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2194 * for more details on the implementation of the table of cancel records.
2197 xlog_recover_do_buffer_trans(
2199 xlog_recover_item_t *item,
2202 xfs_buf_log_format_t *buf_f;
2212 buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr;
2214 if (pass == XLOG_RECOVER_PASS1) {
2216 * In this pass we're only looking for buf items
2217 * with the XFS_BLI_CANCEL bit set.
2219 xlog_recover_do_buffer_pass1(log, buf_f);
2223 * In this pass we want to recover all the buffers
2224 * which have not been cancelled and are not
2225 * cancellation buffers themselves. The routine
2226 * we call here will tell us whether or not to
2227 * continue with the replay of this buffer.
2229 cancel = xlog_recover_do_buffer_pass2(log, buf_f);
2234 switch (buf_f->blf_type) {
2236 blkno = buf_f->blf_blkno;
2237 len = buf_f->blf_len;
2238 flags = buf_f->blf_flags;
2241 xfs_fs_cmn_err(CE_ALERT, log->l_mp,
2242 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2243 buf_f->blf_type, log->l_mp->m_logname ?
2244 log->l_mp->m_logname : "internal");
2245 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2246 XFS_ERRLEVEL_LOW, log->l_mp);
2247 return XFS_ERROR(EFSCORRUPTED);
2251 buf_flags = XFS_BUF_LOCK;
2252 if (!(flags & XFS_BLI_INODE_BUF))
2253 buf_flags |= XFS_BUF_MAPPED;
2255 bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, buf_flags);
2256 if (XFS_BUF_ISERROR(bp)) {
2257 xfs_ioerror_alert("xlog_recover_do..(read#1)", log->l_mp,
2259 error = XFS_BUF_GETERROR(bp);
2265 if (flags & XFS_BLI_INODE_BUF) {
2266 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2268 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
2269 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2271 xlog_recover_do_reg_buffer(item, bp, buf_f);
2274 return XFS_ERROR(error);
2277 * Perform delayed write on the buffer. Asynchronous writes will be
2278 * slower when taking into account all the buffers to be flushed.
2280 * Also make sure that only inode buffers with good sizes stay in
2281 * the buffer cache. The kernel moves inodes in buffers of 1 block
2282 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2283 * buffers in the log can be a different size if the log was generated
2284 * by an older kernel using unclustered inode buffers or a newer kernel
2285 * running with a different inode cluster size. Regardless, if the
2286 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2287 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2288 * the buffer out of the buffer cache so that the buffer won't
2289 * overlap with future reads of those inodes.
2291 if (XFS_DINODE_MAGIC ==
2292 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2293 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2294 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2296 error = xfs_bwrite(mp, bp);
2298 ASSERT(bp->b_mount == NULL || bp->b_mount == mp);
2300 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2301 xfs_bdwrite(mp, bp);
2308 xlog_recover_do_inode_trans(
2310 xlog_recover_item_t *item,
2313 xfs_inode_log_format_t *in_f;
2324 xfs_icdinode_t *dicp;
2327 if (pass == XLOG_RECOVER_PASS1) {
2331 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2332 in_f = (xfs_inode_log_format_t *)item->ri_buf[0].i_addr;
2334 in_f = (xfs_inode_log_format_t *)kmem_alloc(
2335 sizeof(xfs_inode_log_format_t), KM_SLEEP);
2337 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2341 ino = in_f->ilf_ino;
2345 * Inode buffers can be freed, look out for it,
2346 * and do not replay the inode.
2348 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2349 in_f->ilf_len, 0)) {
2354 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len,
2356 if (XFS_BUF_ISERROR(bp)) {
2357 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
2358 bp, in_f->ilf_blkno);
2359 error = XFS_BUF_GETERROR(bp);
2364 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2365 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
2368 * Make sure the place we're flushing out to really looks
2371 if (unlikely(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC)) {
2373 xfs_fs_cmn_err(CE_ALERT, mp,
2374 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2376 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2377 XFS_ERRLEVEL_LOW, mp);
2378 error = EFSCORRUPTED;
2381 dicp = (xfs_icdinode_t *)(item->ri_buf[1].i_addr);
2382 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2384 xfs_fs_cmn_err(CE_ALERT, mp,
2385 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2387 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2388 XFS_ERRLEVEL_LOW, mp);
2389 error = EFSCORRUPTED;
2393 /* Skip replay when the on disk inode is newer than the log one */
2394 if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
2396 * Deal with the wrap case, DI_MAX_FLUSH is less
2397 * than smaller numbers
2399 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
2400 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2408 /* Take the opportunity to reset the flush iteration count */
2409 dicp->di_flushiter = 0;
2411 if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) {
2412 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2413 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2414 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2415 XFS_ERRLEVEL_LOW, mp, dicp);
2417 xfs_fs_cmn_err(CE_ALERT, mp,
2418 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2419 item, dip, bp, ino);
2420 error = EFSCORRUPTED;
2423 } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) {
2424 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2425 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2426 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2427 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2428 XFS_ERRLEVEL_LOW, mp, dicp);
2430 xfs_fs_cmn_err(CE_ALERT, mp,
2431 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2432 item, dip, bp, ino);
2433 error = EFSCORRUPTED;
2437 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2438 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2439 XFS_ERRLEVEL_LOW, mp, dicp);
2441 xfs_fs_cmn_err(CE_ALERT, mp,
2442 "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",
2444 dicp->di_nextents + dicp->di_anextents,
2446 error = EFSCORRUPTED;
2449 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2450 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2451 XFS_ERRLEVEL_LOW, mp, dicp);
2453 xfs_fs_cmn_err(CE_ALERT, mp,
2454 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2455 item, dip, bp, ino, dicp->di_forkoff);
2456 error = EFSCORRUPTED;
2459 if (unlikely(item->ri_buf[1].i_len > sizeof(struct xfs_icdinode))) {
2460 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2461 XFS_ERRLEVEL_LOW, mp, dicp);
2463 xfs_fs_cmn_err(CE_ALERT, mp,
2464 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2465 item->ri_buf[1].i_len, item);
2466 error = EFSCORRUPTED;
2470 /* The core is in in-core format */
2471 xfs_dinode_to_disk(dip, (xfs_icdinode_t *)item->ri_buf[1].i_addr);
2473 /* the rest is in on-disk format */
2474 if (item->ri_buf[1].i_len > sizeof(struct xfs_icdinode)) {
2475 memcpy((xfs_caddr_t) dip + sizeof(struct xfs_icdinode),
2476 item->ri_buf[1].i_addr + sizeof(struct xfs_icdinode),
2477 item->ri_buf[1].i_len - sizeof(struct xfs_icdinode));
2480 fields = in_f->ilf_fields;
2481 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2483 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
2486 memcpy(XFS_DFORK_DPTR(dip),
2487 &in_f->ilf_u.ilfu_uuid,
2492 if (in_f->ilf_size == 2)
2493 goto write_inode_buffer;
2494 len = item->ri_buf[2].i_len;
2495 src = item->ri_buf[2].i_addr;
2496 ASSERT(in_f->ilf_size <= 4);
2497 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2498 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2499 (len == in_f->ilf_dsize));
2501 switch (fields & XFS_ILOG_DFORK) {
2502 case XFS_ILOG_DDATA:
2504 memcpy(XFS_DFORK_DPTR(dip), src, len);
2507 case XFS_ILOG_DBROOT:
2508 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
2509 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
2510 XFS_DFORK_DSIZE(dip, mp));
2515 * There are no data fork flags set.
2517 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2522 * If we logged any attribute data, recover it. There may or
2523 * may not have been any other non-core data logged in this
2526 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2527 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2532 len = item->ri_buf[attr_index].i_len;
2533 src = item->ri_buf[attr_index].i_addr;
2534 ASSERT(len == in_f->ilf_asize);
2536 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2537 case XFS_ILOG_ADATA:
2539 dest = XFS_DFORK_APTR(dip);
2540 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2541 memcpy(dest, src, len);
2544 case XFS_ILOG_ABROOT:
2545 dest = XFS_DFORK_APTR(dip);
2546 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2547 len, (xfs_bmdr_block_t*)dest,
2548 XFS_DFORK_ASIZE(dip, mp));
2552 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2561 ASSERT(bp->b_mount == NULL || bp->b_mount == mp);
2563 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2564 xfs_bdwrite(mp, bp);
2568 return XFS_ERROR(error);
2572 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2573 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2577 xlog_recover_do_quotaoff_trans(
2579 xlog_recover_item_t *item,
2582 xfs_qoff_logformat_t *qoff_f;
2584 if (pass == XLOG_RECOVER_PASS2) {
2588 qoff_f = (xfs_qoff_logformat_t *)item->ri_buf[0].i_addr;
2592 * The logitem format's flag tells us if this was user quotaoff,
2593 * group/project quotaoff or both.
2595 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2596 log->l_quotaoffs_flag |= XFS_DQ_USER;
2597 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2598 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2599 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2600 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2606 * Recover a dquot record
2609 xlog_recover_do_dquot_trans(
2611 xlog_recover_item_t *item,
2616 struct xfs_disk_dquot *ddq, *recddq;
2618 xfs_dq_logformat_t *dq_f;
2621 if (pass == XLOG_RECOVER_PASS1) {
2627 * Filesystems are required to send in quota flags at mount time.
2629 if (mp->m_qflags == 0)
2632 recddq = (xfs_disk_dquot_t *)item->ri_buf[1].i_addr;
2634 if (item->ri_buf[1].i_addr == NULL) {
2636 "XFS: NULL dquot in %s.", __func__);
2637 return XFS_ERROR(EIO);
2639 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
2641 "XFS: dquot too small (%d) in %s.",
2642 item->ri_buf[1].i_len, __func__);
2643 return XFS_ERROR(EIO);
2647 * This type of quotas was turned off, so ignore this record.
2649 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
2651 if (log->l_quotaoffs_flag & type)
2655 * At this point we know that quota was _not_ turned off.
2656 * Since the mount flags are not indicating to us otherwise, this
2657 * must mean that quota is on, and the dquot needs to be replayed.
2658 * Remember that we may not have fully recovered the superblock yet,
2659 * so we can't do the usual trick of looking at the SB quota bits.
2661 * The other possibility, of course, is that the quota subsystem was
2662 * removed since the last mount - ENOSYS.
2664 dq_f = (xfs_dq_logformat_t *)item->ri_buf[0].i_addr;
2666 if ((error = xfs_qm_dqcheck(recddq,
2668 0, XFS_QMOPT_DOWARN,
2669 "xlog_recover_do_dquot_trans (log copy)"))) {
2670 return XFS_ERROR(EIO);
2672 ASSERT(dq_f->qlf_len == 1);
2674 error = xfs_read_buf(mp, mp->m_ddev_targp,
2676 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2679 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
2680 bp, dq_f->qlf_blkno);
2684 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2687 * At least the magic num portion should be on disk because this
2688 * was among a chunk of dquots created earlier, and we did some
2689 * minimal initialization then.
2691 if (xfs_qm_dqcheck(ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2692 "xlog_recover_do_dquot_trans")) {
2694 return XFS_ERROR(EIO);
2697 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2699 ASSERT(dq_f->qlf_size == 2);
2700 ASSERT(bp->b_mount == NULL || bp->b_mount == mp);
2702 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2703 xfs_bdwrite(mp, bp);
2709 * This routine is called to create an in-core extent free intent
2710 * item from the efi format structure which was logged on disk.
2711 * It allocates an in-core efi, copies the extents from the format
2712 * structure into it, and adds the efi to the AIL with the given
2716 xlog_recover_do_efi_trans(
2718 xlog_recover_item_t *item,
2724 xfs_efi_log_item_t *efip;
2725 xfs_efi_log_format_t *efi_formatp;
2727 if (pass == XLOG_RECOVER_PASS1) {
2731 efi_formatp = (xfs_efi_log_format_t *)item->ri_buf[0].i_addr;
2734 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
2735 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2736 &(efip->efi_format)))) {
2737 xfs_efi_item_free(efip);
2740 efip->efi_next_extent = efi_formatp->efi_nextents;
2741 efip->efi_flags |= XFS_EFI_COMMITTED;
2743 spin_lock(&log->l_ailp->xa_lock);
2745 * xfs_trans_ail_update() drops the AIL lock.
2747 xfs_trans_ail_update(log->l_ailp, (xfs_log_item_t *)efip, lsn);
2753 * This routine is called when an efd format structure is found in
2754 * a committed transaction in the log. It's purpose is to cancel
2755 * the corresponding efi if it was still in the log. To do this
2756 * it searches the AIL for the efi with an id equal to that in the
2757 * efd format structure. If we find it, we remove the efi from the
2761 xlog_recover_do_efd_trans(
2763 xlog_recover_item_t *item,
2766 xfs_efd_log_format_t *efd_formatp;
2767 xfs_efi_log_item_t *efip = NULL;
2768 xfs_log_item_t *lip;
2770 struct xfs_ail_cursor cur;
2771 struct xfs_ail *ailp = log->l_ailp;
2773 if (pass == XLOG_RECOVER_PASS1) {
2777 efd_formatp = (xfs_efd_log_format_t *)item->ri_buf[0].i_addr;
2778 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2779 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2780 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2781 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
2782 efi_id = efd_formatp->efd_efi_id;
2785 * Search for the efi with the id in the efd format structure
2788 spin_lock(&ailp->xa_lock);
2789 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
2790 while (lip != NULL) {
2791 if (lip->li_type == XFS_LI_EFI) {
2792 efip = (xfs_efi_log_item_t *)lip;
2793 if (efip->efi_format.efi_id == efi_id) {
2795 * xfs_trans_ail_delete() drops the
2798 xfs_trans_ail_delete(ailp, lip);
2799 xfs_efi_item_free(efip);
2800 spin_lock(&ailp->xa_lock);
2804 lip = xfs_trans_ail_cursor_next(ailp, &cur);
2806 xfs_trans_ail_cursor_done(ailp, &cur);
2807 spin_unlock(&ailp->xa_lock);
2811 * Perform the transaction
2813 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2814 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2817 xlog_recover_do_trans(
2819 xlog_recover_t *trans,
2823 xlog_recover_item_t *item, *first_item;
2825 error = xlog_recover_reorder_trans(trans);
2829 first_item = item = trans->r_itemq;
2831 switch (ITEM_TYPE(item)) {
2833 error = xlog_recover_do_buffer_trans(log, item, pass);
2836 error = xlog_recover_do_inode_trans(log, item, pass);
2839 error = xlog_recover_do_efi_trans(log, item,
2840 trans->r_lsn, pass);
2843 xlog_recover_do_efd_trans(log, item, pass);
2847 error = xlog_recover_do_dquot_trans(log, item, pass);
2849 case XFS_LI_QUOTAOFF:
2850 error = xlog_recover_do_quotaoff_trans(log, item,
2855 "XFS: invalid item type (%d) xlog_recover_do_trans", ITEM_TYPE(item));
2857 error = XFS_ERROR(EIO);
2863 item = item->ri_next;
2864 } while (first_item != item);
2870 * Free up any resources allocated by the transaction
2872 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2875 xlog_recover_free_trans(
2876 xlog_recover_t *trans)
2878 xlog_recover_item_t *first_item, *item, *free_item;
2881 item = first_item = trans->r_itemq;
2884 item = item->ri_next;
2885 /* Free the regions in the item. */
2886 for (i = 0; i < free_item->ri_cnt; i++) {
2887 kmem_free(free_item->ri_buf[i].i_addr);
2889 /* Free the item itself */
2890 kmem_free(free_item->ri_buf);
2891 kmem_free(free_item);
2892 } while (first_item != item);
2893 /* Free the transaction recover structure */
2898 xlog_recover_commit_trans(
2901 xlog_recover_t *trans,
2906 if ((error = xlog_recover_unlink_tid(q, trans)))
2908 if ((error = xlog_recover_do_trans(log, trans, pass)))
2910 xlog_recover_free_trans(trans); /* no error */
2915 xlog_recover_unmount_trans(
2916 xlog_recover_t *trans)
2918 /* Do nothing now */
2919 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2924 * There are two valid states of the r_state field. 0 indicates that the
2925 * transaction structure is in a normal state. We have either seen the
2926 * start of the transaction or the last operation we added was not a partial
2927 * operation. If the last operation we added to the transaction was a
2928 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2930 * NOTE: skip LRs with 0 data length.
2933 xlog_recover_process_data(
2935 xlog_recover_t *rhash[],
2936 xlog_rec_header_t *rhead,
2942 xlog_op_header_t *ohead;
2943 xlog_recover_t *trans;
2949 lp = dp + be32_to_cpu(rhead->h_len);
2950 num_logops = be32_to_cpu(rhead->h_num_logops);
2952 /* check the log format matches our own - else we can't recover */
2953 if (xlog_header_check_recover(log->l_mp, rhead))
2954 return (XFS_ERROR(EIO));
2956 while ((dp < lp) && num_logops) {
2957 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2958 ohead = (xlog_op_header_t *)dp;
2959 dp += sizeof(xlog_op_header_t);
2960 if (ohead->oh_clientid != XFS_TRANSACTION &&
2961 ohead->oh_clientid != XFS_LOG) {
2963 "XFS: xlog_recover_process_data: bad clientid");
2965 return (XFS_ERROR(EIO));
2967 tid = be32_to_cpu(ohead->oh_tid);
2968 hash = XLOG_RHASH(tid);
2969 trans = xlog_recover_find_tid(rhash[hash], tid);
2970 if (trans == NULL) { /* not found; add new tid */
2971 if (ohead->oh_flags & XLOG_START_TRANS)
2972 xlog_recover_new_tid(&rhash[hash], tid,
2973 be64_to_cpu(rhead->h_lsn));
2975 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
2977 "XFS: xlog_recover_process_data: bad length");
2979 return (XFS_ERROR(EIO));
2981 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2982 if (flags & XLOG_WAS_CONT_TRANS)
2983 flags &= ~XLOG_CONTINUE_TRANS;
2985 case XLOG_COMMIT_TRANS:
2986 error = xlog_recover_commit_trans(log,
2987 &rhash[hash], trans, pass);
2989 case XLOG_UNMOUNT_TRANS:
2990 error = xlog_recover_unmount_trans(trans);
2992 case XLOG_WAS_CONT_TRANS:
2993 error = xlog_recover_add_to_cont_trans(trans,
2994 dp, be32_to_cpu(ohead->oh_len));
2996 case XLOG_START_TRANS:
2998 "XFS: xlog_recover_process_data: bad transaction");
3000 error = XFS_ERROR(EIO);
3003 case XLOG_CONTINUE_TRANS:
3004 error = xlog_recover_add_to_trans(trans,
3005 dp, be32_to_cpu(ohead->oh_len));
3009 "XFS: xlog_recover_process_data: bad flag");
3011 error = XFS_ERROR(EIO);
3017 dp += be32_to_cpu(ohead->oh_len);
3024 * Process an extent free intent item that was recovered from
3025 * the log. We need to free the extents that it describes.
3028 xlog_recover_process_efi(
3030 xfs_efi_log_item_t *efip)
3032 xfs_efd_log_item_t *efdp;
3037 xfs_fsblock_t startblock_fsb;
3039 ASSERT(!(efip->efi_flags & XFS_EFI_RECOVERED));
3042 * First check the validity of the extents described by the
3043 * EFI. If any are bad, then assume that all are bad and
3044 * just toss the EFI.
3046 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3047 extp = &(efip->efi_format.efi_extents[i]);
3048 startblock_fsb = XFS_BB_TO_FSB(mp,
3049 XFS_FSB_TO_DADDR(mp, extp->ext_start));
3050 if ((startblock_fsb == 0) ||
3051 (extp->ext_len == 0) ||
3052 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3053 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3055 * This will pull the EFI from the AIL and
3056 * free the memory associated with it.
3058 xfs_efi_release(efip, efip->efi_format.efi_nextents);
3059 return XFS_ERROR(EIO);
3063 tp = xfs_trans_alloc(mp, 0);
3064 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
3067 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3069 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3070 extp = &(efip->efi_format.efi_extents[i]);
3071 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3074 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3078 efip->efi_flags |= XFS_EFI_RECOVERED;
3079 error = xfs_trans_commit(tp, 0);
3083 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3088 * When this is called, all of the EFIs which did not have
3089 * corresponding EFDs should be in the AIL. What we do now
3090 * is free the extents associated with each one.
3092 * Since we process the EFIs in normal transactions, they
3093 * will be removed at some point after the commit. This prevents
3094 * us from just walking down the list processing each one.
3095 * We'll use a flag in the EFI to skip those that we've already
3096 * processed and use the AIL iteration mechanism's generation
3097 * count to try to speed this up at least a bit.
3099 * When we start, we know that the EFIs are the only things in
3100 * the AIL. As we process them, however, other items are added
3101 * to the AIL. Since everything added to the AIL must come after
3102 * everything already in the AIL, we stop processing as soon as
3103 * we see something other than an EFI in the AIL.
3106 xlog_recover_process_efis(
3109 xfs_log_item_t *lip;
3110 xfs_efi_log_item_t *efip;
3112 struct xfs_ail_cursor cur;
3113 struct xfs_ail *ailp;
3116 spin_lock(&ailp->xa_lock);
3117 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3118 while (lip != NULL) {
3120 * We're done when we see something other than an EFI.
3121 * There should be no EFIs left in the AIL now.
3123 if (lip->li_type != XFS_LI_EFI) {
3125 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
3126 ASSERT(lip->li_type != XFS_LI_EFI);
3132 * Skip EFIs that we've already processed.
3134 efip = (xfs_efi_log_item_t *)lip;
3135 if (efip->efi_flags & XFS_EFI_RECOVERED) {
3136 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3140 spin_unlock(&ailp->xa_lock);
3141 error = xlog_recover_process_efi(log->l_mp, efip);
3142 spin_lock(&ailp->xa_lock);
3145 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3148 xfs_trans_ail_cursor_done(ailp, &cur);
3149 spin_unlock(&ailp->xa_lock);
3154 * This routine performs a transaction to null out a bad inode pointer
3155 * in an agi unlinked inode hash bucket.
3158 xlog_recover_clear_agi_bucket(
3160 xfs_agnumber_t agno,
3169 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3170 error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp),
3175 error = xfs_read_agi(mp, tp, agno, &agibp);
3179 agi = XFS_BUF_TO_AGI(agibp);
3180 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
3181 offset = offsetof(xfs_agi_t, agi_unlinked) +
3182 (sizeof(xfs_agino_t) * bucket);
3183 xfs_trans_log_buf(tp, agibp, offset,
3184 (offset + sizeof(xfs_agino_t) - 1));
3186 error = xfs_trans_commit(tp, 0);
3192 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3194 xfs_fs_cmn_err(CE_WARN, mp, "xlog_recover_clear_agi_bucket: "
3195 "failed to clear agi %d. Continuing.", agno);
3200 xlog_recover_process_one_iunlink(
3201 struct xfs_mount *mp,
3202 xfs_agnumber_t agno,
3206 struct xfs_buf *ibp;
3207 struct xfs_dinode *dip;
3208 struct xfs_inode *ip;
3212 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3213 error = xfs_iget(mp, NULL, ino, 0, 0, &ip, 0);
3218 * Get the on disk inode to find the next inode in the bucket.
3220 error = xfs_itobp(mp, NULL, ip, &dip, &ibp, XFS_BUF_LOCK);
3224 ASSERT(ip->i_d.di_nlink == 0);
3225 ASSERT(ip->i_d.di_mode != 0);
3227 /* setup for the next pass */
3228 agino = be32_to_cpu(dip->di_next_unlinked);
3232 * Prevent any DMAPI event from being sent when the reference on
3233 * the inode is dropped.
3235 ip->i_d.di_dmevmask = 0;
3244 * We can't read in the inode this bucket points to, or this inode
3245 * is messed up. Just ditch this bucket of inodes. We will lose
3246 * some inodes and space, but at least we won't hang.
3248 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3249 * clear the inode pointer in the bucket.
3251 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3256 * xlog_iunlink_recover
3258 * This is called during recovery to process any inodes which
3259 * we unlinked but not freed when the system crashed. These
3260 * inodes will be on the lists in the AGI blocks. What we do
3261 * here is scan all the AGIs and fully truncate and free any
3262 * inodes found on the lists. Each inode is removed from the
3263 * lists when it has been fully truncated and is freed. The
3264 * freeing of the inode and its removal from the list must be
3268 xlog_recover_process_iunlinks(
3272 xfs_agnumber_t agno;
3283 * Prevent any DMAPI event from being sent while in this function.
3285 mp_dmevmask = mp->m_dmevmask;
3288 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3290 * Find the agi for this ag.
3292 error = xfs_read_agi(mp, NULL, agno, &agibp);
3295 * AGI is b0rked. Don't process it.
3297 * We should probably mark the filesystem as corrupt
3298 * after we've recovered all the ag's we can....
3302 agi = XFS_BUF_TO_AGI(agibp);
3304 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3305 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3306 while (agino != NULLAGINO) {
3308 * Release the agi buffer so that it can
3309 * be acquired in the normal course of the
3310 * transaction to truncate and free the inode.
3312 xfs_buf_relse(agibp);
3314 agino = xlog_recover_process_one_iunlink(mp,
3315 agno, agino, bucket);
3318 * Reacquire the agibuffer and continue around
3319 * the loop. This should never fail as we know
3320 * the buffer was good earlier on.
3322 error = xfs_read_agi(mp, NULL, agno, &agibp);
3324 agi = XFS_BUF_TO_AGI(agibp);
3329 * Release the buffer for the current agi so we can
3330 * go on to the next one.
3332 xfs_buf_relse(agibp);
3335 mp->m_dmevmask = mp_dmevmask;
3341 xlog_pack_data_checksum(
3343 xlog_in_core_t *iclog,
3350 up = (__be32 *)iclog->ic_datap;
3351 /* divide length by 4 to get # words */
3352 for (i = 0; i < (size >> 2); i++) {
3353 chksum ^= be32_to_cpu(*up);
3356 iclog->ic_header.h_chksum = cpu_to_be32(chksum);
3359 #define xlog_pack_data_checksum(log, iclog, size)
3363 * Stamp cycle number in every block
3368 xlog_in_core_t *iclog,
3372 int size = iclog->ic_offset + roundoff;
3376 xlog_pack_data_checksum(log, iclog, size);
3378 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3380 dp = iclog->ic_datap;
3381 for (i = 0; i < BTOBB(size) &&
3382 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3383 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
3384 *(__be32 *)dp = cycle_lsn;
3388 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3389 xlog_in_core_2_t *xhdr = iclog->ic_data;
3391 for ( ; i < BTOBB(size); i++) {
3392 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3393 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3394 xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
3395 *(__be32 *)dp = cycle_lsn;
3399 for (i = 1; i < log->l_iclog_heads; i++) {
3400 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3405 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
3407 xlog_unpack_data_checksum(
3408 xlog_rec_header_t *rhead,
3412 __be32 *up = (__be32 *)dp;
3416 /* divide length by 4 to get # words */
3417 for (i=0; i < be32_to_cpu(rhead->h_len) >> 2; i++) {
3418 chksum ^= be32_to_cpu(*up);
3421 if (chksum != be32_to_cpu(rhead->h_chksum)) {
3422 if (rhead->h_chksum ||
3423 ((log->l_flags & XLOG_CHKSUM_MISMATCH) == 0)) {
3425 "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n",
3426 be32_to_cpu(rhead->h_chksum), chksum);
3428 "XFS: Disregard message if filesystem was created with non-DEBUG kernel");
3429 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3431 "XFS: LogR this is a LogV2 filesystem\n");
3433 log->l_flags |= XLOG_CHKSUM_MISMATCH;
3438 #define xlog_unpack_data_checksum(rhead, dp, log)
3443 xlog_rec_header_t *rhead,
3449 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
3450 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3451 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
3455 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3456 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
3457 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
3458 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3459 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3460 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
3465 xlog_unpack_data_checksum(rhead, dp, log);
3469 xlog_valid_rec_header(
3471 xlog_rec_header_t *rhead,
3476 if (unlikely(be32_to_cpu(rhead->h_magicno) != XLOG_HEADER_MAGIC_NUM)) {
3477 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3478 XFS_ERRLEVEL_LOW, log->l_mp);
3479 return XFS_ERROR(EFSCORRUPTED);
3482 (!rhead->h_version ||
3483 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
3484 xlog_warn("XFS: %s: unrecognised log version (%d).",
3485 __func__, be32_to_cpu(rhead->h_version));
3486 return XFS_ERROR(EIO);
3489 /* LR body must have data or it wouldn't have been written */
3490 hlen = be32_to_cpu(rhead->h_len);
3491 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3492 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3493 XFS_ERRLEVEL_LOW, log->l_mp);
3494 return XFS_ERROR(EFSCORRUPTED);
3496 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3497 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3498 XFS_ERRLEVEL_LOW, log->l_mp);
3499 return XFS_ERROR(EFSCORRUPTED);
3505 * Read the log from tail to head and process the log records found.
3506 * Handle the two cases where the tail and head are in the same cycle
3507 * and where the active portion of the log wraps around the end of
3508 * the physical log separately. The pass parameter is passed through
3509 * to the routines called to process the data and is not looked at
3513 xlog_do_recovery_pass(
3515 xfs_daddr_t head_blk,
3516 xfs_daddr_t tail_blk,
3519 xlog_rec_header_t *rhead;
3522 xfs_buf_t *hbp, *dbp;
3523 int error = 0, h_size;
3524 int bblks, split_bblks;
3525 int hblks, split_hblks, wrapped_hblks;
3526 xlog_recover_t *rhash[XLOG_RHASH_SIZE];
3528 ASSERT(head_blk != tail_blk);
3531 * Read the header of the tail block and get the iclog buffer size from
3532 * h_size. Use this to tell how many sectors make up the log header.
3534 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3536 * When using variable length iclogs, read first sector of
3537 * iclog header and extract the header size from it. Get a
3538 * new hbp that is the correct size.
3540 hbp = xlog_get_bp(log, 1);
3544 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
3548 rhead = (xlog_rec_header_t *)offset;
3549 error = xlog_valid_rec_header(log, rhead, tail_blk);
3552 h_size = be32_to_cpu(rhead->h_size);
3553 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
3554 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3555 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3556 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3559 hbp = xlog_get_bp(log, hblks);
3564 ASSERT(log->l_sectbb_log == 0);
3566 hbp = xlog_get_bp(log, 1);
3567 h_size = XLOG_BIG_RECORD_BSIZE;
3572 dbp = xlog_get_bp(log, BTOBB(h_size));
3578 memset(rhash, 0, sizeof(rhash));
3579 if (tail_blk <= head_blk) {
3580 for (blk_no = tail_blk; blk_no < head_blk; ) {
3581 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3585 rhead = (xlog_rec_header_t *)offset;
3586 error = xlog_valid_rec_header(log, rhead, blk_no);
3590 /* blocks in data section */
3591 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3592 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
3597 xlog_unpack_data(rhead, offset, log);
3598 if ((error = xlog_recover_process_data(log,
3599 rhash, rhead, offset, pass)))
3601 blk_no += bblks + hblks;
3605 * Perform recovery around the end of the physical log.
3606 * When the head is not on the same cycle number as the tail,
3607 * we can't do a sequential recovery as above.
3610 while (blk_no < log->l_logBBsize) {
3612 * Check for header wrapping around physical end-of-log
3614 offset = XFS_BUF_PTR(hbp);
3617 if (blk_no + hblks <= log->l_logBBsize) {
3618 /* Read header in one read */
3619 error = xlog_bread(log, blk_no, hblks, hbp,
3624 /* This LR is split across physical log end */
3625 if (blk_no != log->l_logBBsize) {
3626 /* some data before physical log end */
3627 ASSERT(blk_no <= INT_MAX);
3628 split_hblks = log->l_logBBsize - (int)blk_no;
3629 ASSERT(split_hblks > 0);
3630 error = xlog_bread(log, blk_no,
3638 * Note: this black magic still works with
3639 * large sector sizes (non-512) only because:
3640 * - we increased the buffer size originally
3641 * by 1 sector giving us enough extra space
3642 * for the second read;
3643 * - the log start is guaranteed to be sector
3645 * - we read the log end (LR header start)
3646 * _first_, then the log start (LR header end)
3647 * - order is important.
3649 wrapped_hblks = hblks - split_hblks;
3650 error = XFS_BUF_SET_PTR(hbp,
3651 offset + BBTOB(split_hblks),
3652 BBTOB(hblks - split_hblks));
3656 error = xlog_bread_noalign(log, 0,
3657 wrapped_hblks, hbp);
3661 error = XFS_BUF_SET_PTR(hbp, offset,
3666 rhead = (xlog_rec_header_t *)offset;
3667 error = xlog_valid_rec_header(log, rhead,
3668 split_hblks ? blk_no : 0);
3672 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3675 /* Read in data for log record */
3676 if (blk_no + bblks <= log->l_logBBsize) {
3677 error = xlog_bread(log, blk_no, bblks, dbp,
3682 /* This log record is split across the
3683 * physical end of log */
3684 offset = XFS_BUF_PTR(dbp);
3686 if (blk_no != log->l_logBBsize) {
3687 /* some data is before the physical
3689 ASSERT(!wrapped_hblks);
3690 ASSERT(blk_no <= INT_MAX);
3692 log->l_logBBsize - (int)blk_no;
3693 ASSERT(split_bblks > 0);
3694 error = xlog_bread(log, blk_no,
3702 * Note: this black magic still works with
3703 * large sector sizes (non-512) only because:
3704 * - we increased the buffer size originally
3705 * by 1 sector giving us enough extra space
3706 * for the second read;
3707 * - the log start is guaranteed to be sector
3709 * - we read the log end (LR header start)
3710 * _first_, then the log start (LR header end)
3711 * - order is important.
3713 error = XFS_BUF_SET_PTR(dbp,
3714 offset + BBTOB(split_bblks),
3715 BBTOB(bblks - split_bblks));
3719 error = xlog_bread_noalign(log, wrapped_hblks,
3720 bblks - split_bblks,
3725 error = XFS_BUF_SET_PTR(dbp, offset, h_size);
3729 xlog_unpack_data(rhead, offset, log);
3730 if ((error = xlog_recover_process_data(log, rhash,
3731 rhead, offset, pass)))
3736 ASSERT(blk_no >= log->l_logBBsize);
3737 blk_no -= log->l_logBBsize;
3739 /* read first part of physical log */
3740 while (blk_no < head_blk) {
3741 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3745 rhead = (xlog_rec_header_t *)offset;
3746 error = xlog_valid_rec_header(log, rhead, blk_no);
3750 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3751 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
3756 xlog_unpack_data(rhead, offset, log);
3757 if ((error = xlog_recover_process_data(log, rhash,
3758 rhead, offset, pass)))
3760 blk_no += bblks + hblks;
3772 * Do the recovery of the log. We actually do this in two phases.
3773 * The two passes are necessary in order to implement the function
3774 * of cancelling a record written into the log. The first pass
3775 * determines those things which have been cancelled, and the
3776 * second pass replays log items normally except for those which
3777 * have been cancelled. The handling of the replay and cancellations
3778 * takes place in the log item type specific routines.
3780 * The table of items which have cancel records in the log is allocated
3781 * and freed at this level, since only here do we know when all of
3782 * the log recovery has been completed.
3785 xlog_do_log_recovery(
3787 xfs_daddr_t head_blk,
3788 xfs_daddr_t tail_blk)
3792 ASSERT(head_blk != tail_blk);
3795 * First do a pass to find all of the cancelled buf log items.
3796 * Store them in the buf_cancel_table for use in the second pass.
3798 log->l_buf_cancel_table =
3799 (xfs_buf_cancel_t **)kmem_zalloc(XLOG_BC_TABLE_SIZE *
3800 sizeof(xfs_buf_cancel_t*),
3802 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3803 XLOG_RECOVER_PASS1);
3805 kmem_free(log->l_buf_cancel_table);
3806 log->l_buf_cancel_table = NULL;
3810 * Then do a second pass to actually recover the items in the log.
3811 * When it is complete free the table of buf cancel items.
3813 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3814 XLOG_RECOVER_PASS2);
3819 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3820 ASSERT(log->l_buf_cancel_table[i] == NULL);
3824 kmem_free(log->l_buf_cancel_table);
3825 log->l_buf_cancel_table = NULL;
3831 * Do the actual recovery
3836 xfs_daddr_t head_blk,
3837 xfs_daddr_t tail_blk)
3844 * First replay the images in the log.
3846 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3851 XFS_bflush(log->l_mp->m_ddev_targp);
3854 * If IO errors happened during recovery, bail out.
3856 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3861 * We now update the tail_lsn since much of the recovery has completed
3862 * and there may be space available to use. If there were no extent
3863 * or iunlinks, we can free up the entire log and set the tail_lsn to
3864 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3865 * lsn of the last known good LR on disk. If there are extent frees
3866 * or iunlinks they will have some entries in the AIL; so we look at
3867 * the AIL to determine how to set the tail_lsn.
3869 xlog_assign_tail_lsn(log->l_mp);
3872 * Now that we've finished replaying all buffer and inode
3873 * updates, re-read in the superblock.
3875 bp = xfs_getsb(log->l_mp, 0);
3877 ASSERT(!(XFS_BUF_ISWRITE(bp)));
3878 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
3880 XFS_BUF_UNASYNC(bp);
3881 xfsbdstrat(log->l_mp, bp);
3882 error = xfs_iowait(bp);
3884 xfs_ioerror_alert("xlog_do_recover",
3885 log->l_mp, bp, XFS_BUF_ADDR(bp));
3891 /* Convert superblock from on-disk format */
3892 sbp = &log->l_mp->m_sb;
3893 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
3894 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
3895 ASSERT(xfs_sb_good_version(sbp));
3898 /* We've re-read the superblock so re-initialize per-cpu counters */
3899 xfs_icsb_reinit_counters(log->l_mp);
3901 xlog_recover_check_summary(log);
3903 /* Normal transactions can now occur */
3904 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3909 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3911 * Return error or zero.
3917 xfs_daddr_t head_blk, tail_blk;
3920 /* find the tail of the log */
3921 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
3924 if (tail_blk != head_blk) {
3925 /* There used to be a comment here:
3927 * disallow recovery on read-only mounts. note -- mount
3928 * checks for ENOSPC and turns it into an intelligent
3930 * ...but this is no longer true. Now, unless you specify
3931 * NORECOVERY (in which case this function would never be
3932 * called), we just go ahead and recover. We do this all
3933 * under the vfs layer, so we can get away with it unless
3934 * the device itself is read-only, in which case we fail.
3936 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
3941 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3942 log->l_mp->m_fsname, log->l_mp->m_logname ?
3943 log->l_mp->m_logname : "internal");
3945 error = xlog_do_recover(log, head_blk, tail_blk);
3946 log->l_flags |= XLOG_RECOVERY_NEEDED;
3952 * In the first part of recovery we replay inodes and buffers and build
3953 * up the list of extent free items which need to be processed. Here
3954 * we process the extent free items and clean up the on disk unlinked
3955 * inode lists. This is separated from the first part of recovery so
3956 * that the root and real-time bitmap inodes can be read in from disk in
3957 * between the two stages. This is necessary so that we can free space
3958 * in the real-time portion of the file system.
3961 xlog_recover_finish(
3965 * Now we're ready to do the transactions needed for the
3966 * rest of recovery. Start with completing all the extent
3967 * free intent records and then process the unlinked inode
3968 * lists. At this point, we essentially run in normal mode
3969 * except that we're still performing recovery actions
3970 * rather than accepting new requests.
3972 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
3974 error = xlog_recover_process_efis(log);
3977 "Failed to recover EFIs on filesystem: %s",
3978 log->l_mp->m_fsname);
3982 * Sync the log to get all the EFIs out of the AIL.
3983 * This isn't absolutely necessary, but it helps in
3984 * case the unlink transactions would have problems
3985 * pushing the EFIs out of the way.
3987 xfs_log_force(log->l_mp, (xfs_lsn_t)0,
3988 (XFS_LOG_FORCE | XFS_LOG_SYNC));
3990 xlog_recover_process_iunlinks(log);
3992 xlog_recover_check_summary(log);
3995 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3996 log->l_mp->m_fsname, log->l_mp->m_logname ?
3997 log->l_mp->m_logname : "internal");
3998 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
4001 "!Ending clean XFS mount for filesystem: %s\n",
4002 log->l_mp->m_fsname);
4010 * Read all of the agf and agi counters and check that they
4011 * are consistent with the superblock counters.
4014 xlog_recover_check_summary(
4022 #ifdef XFS_LOUD_RECOVERY
4025 xfs_agnumber_t agno;
4026 __uint64_t freeblks;
4036 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
4037 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
4039 xfs_fs_cmn_err(CE_ALERT, mp,
4040 "xlog_recover_check_summary(agf)"
4041 "agf read failed agno %d error %d",
4044 agfp = XFS_BUF_TO_AGF(agfbp);
4045 freeblks += be32_to_cpu(agfp->agf_freeblks) +
4046 be32_to_cpu(agfp->agf_flcount);
4047 xfs_buf_relse(agfbp);
4050 error = xfs_read_agi(mp, NULL, agno, &agibp);
4052 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
4054 itotal += be32_to_cpu(agi->agi_count);
4055 ifree += be32_to_cpu(agi->agi_freecount);
4056 xfs_buf_relse(agibp);
4060 sbbp = xfs_getsb(mp, 0);
4061 #ifdef XFS_LOUD_RECOVERY
4063 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(sbbp));
4065 "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
4066 sbp->sb_icount, itotal);
4068 "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
4069 sbp->sb_ifree, ifree);
4071 "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
4072 sbp->sb_fdblocks, freeblks);
4075 * This is turned off until I account for the allocation
4076 * btree blocks which live in free space.
4078 ASSERT(sbp->sb_icount == itotal);
4079 ASSERT(sbp->sb_ifree == ifree);
4080 ASSERT(sbp->sb_fdblocks == freeblks);
4083 xfs_buf_relse(sbbp);