2 * Copyright (c) 2000-2003,2005 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"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_error.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir_sf.h"
36 #include "xfs_dir2_sf.h"
37 #include "xfs_attr_sf.h"
38 #include "xfs_dinode.h"
39 #include "xfs_inode.h"
40 #include "xfs_inode_item.h"
42 #include "xfs_alloc.h"
43 #include "xfs_ialloc.h"
44 #include "xfs_log_priv.h"
45 #include "xfs_buf_item.h"
46 #include "xfs_log_recover.h"
47 #include "xfs_extfree_item.h"
48 #include "xfs_trans_priv.h"
49 #include "xfs_quota.h"
52 STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *);
53 STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t);
54 STATIC void xlog_recover_insert_item_backq(xlog_recover_item_t **q,
55 xlog_recover_item_t *item);
57 STATIC void xlog_recover_check_summary(xlog_t *);
58 STATIC void xlog_recover_check_ail(xfs_mount_t *, xfs_log_item_t *, int);
60 #define xlog_recover_check_summary(log)
61 #define xlog_recover_check_ail(mp, lip, gen)
66 * Sector aligned buffer routines for buffer create/read/write/access
69 #define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \
70 ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \
71 ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) )
72 #define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask)
79 ASSERT(num_bblks > 0);
81 if (log->l_sectbb_log) {
83 num_bblks += XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
84 num_bblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, num_bblks);
86 return xfs_buf_get_noaddr(BBTOB(num_bblks), log->l_mp->m_logdev_targp);
98 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
109 if (log->l_sectbb_log) {
110 blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
111 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
115 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
118 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
121 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
122 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
124 xfsbdstrat(log->l_mp, bp);
125 if ((error = xfs_iowait(bp)))
126 xfs_ioerror_alert("xlog_bread", log->l_mp,
127 bp, XFS_BUF_ADDR(bp));
132 * Write out the buffer at the given block for the given number of blocks.
133 * The buffer is kept locked across the write and is returned locked.
134 * This can only be used for synchronous log writes.
145 if (log->l_sectbb_log) {
146 blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
147 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
151 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
153 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
154 XFS_BUF_ZEROFLAGS(bp);
157 XFS_BUF_PSEMA(bp, PRIBIO);
158 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
159 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
161 if ((error = xfs_bwrite(log->l_mp, bp)))
162 xfs_ioerror_alert("xlog_bwrite", log->l_mp,
163 bp, XFS_BUF_ADDR(bp));
176 if (!log->l_sectbb_log)
177 return XFS_BUF_PTR(bp);
179 ptr = XFS_BUF_PTR(bp) + BBTOB((int)blk_no & log->l_sectbb_mask);
180 ASSERT(XFS_BUF_SIZE(bp) >=
181 BBTOB(nbblks + (blk_no & log->l_sectbb_mask)));
187 * dump debug superblock and log record information
190 xlog_header_check_dump(
192 xlog_rec_header_t *head)
196 printk("%s: SB : uuid = ", __FUNCTION__);
197 for (b = 0; b < 16; b++)
198 printk("%02x",((unsigned char *)&mp->m_sb.sb_uuid)[b]);
199 printk(", fmt = %d\n", XLOG_FMT);
200 printk(" log : uuid = ");
201 for (b = 0; b < 16; b++)
202 printk("%02x",((unsigned char *)&head->h_fs_uuid)[b]);
203 printk(", fmt = %d\n", INT_GET(head->h_fmt, ARCH_CONVERT));
206 #define xlog_header_check_dump(mp, head)
210 * check log record header for recovery
213 xlog_header_check_recover(
215 xlog_rec_header_t *head)
217 ASSERT(INT_GET(head->h_magicno, ARCH_CONVERT) == XLOG_HEADER_MAGIC_NUM);
220 * IRIX doesn't write the h_fmt field and leaves it zeroed
221 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
222 * a dirty log created in IRIX.
224 if (unlikely(INT_GET(head->h_fmt, ARCH_CONVERT) != XLOG_FMT)) {
226 "XFS: dirty log written in incompatible format - can't recover");
227 xlog_header_check_dump(mp, head);
228 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
229 XFS_ERRLEVEL_HIGH, mp);
230 return XFS_ERROR(EFSCORRUPTED);
231 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
233 "XFS: dirty log entry has mismatched uuid - can't recover");
234 xlog_header_check_dump(mp, head);
235 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
236 XFS_ERRLEVEL_HIGH, mp);
237 return XFS_ERROR(EFSCORRUPTED);
243 * read the head block of the log and check the header
246 xlog_header_check_mount(
248 xlog_rec_header_t *head)
250 ASSERT(INT_GET(head->h_magicno, ARCH_CONVERT) == XLOG_HEADER_MAGIC_NUM);
252 if (uuid_is_nil(&head->h_fs_uuid)) {
254 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
255 * h_fs_uuid is nil, we assume this log was last mounted
256 * by IRIX and continue.
258 xlog_warn("XFS: nil uuid in log - IRIX style log");
259 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
260 xlog_warn("XFS: log has mismatched uuid - can't recover");
261 xlog_header_check_dump(mp, head);
262 XFS_ERROR_REPORT("xlog_header_check_mount",
263 XFS_ERRLEVEL_HIGH, mp);
264 return XFS_ERROR(EFSCORRUPTED);
275 ASSERT(XFS_BUF_FSPRIVATE(bp, void *));
277 if (XFS_BUF_GETERROR(bp)) {
279 * We're not going to bother about retrying
280 * this during recovery. One strike!
282 mp = XFS_BUF_FSPRIVATE(bp, xfs_mount_t *);
283 xfs_ioerror_alert("xlog_recover_iodone",
284 mp, bp, XFS_BUF_ADDR(bp));
285 xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
287 XFS_BUF_SET_FSPRIVATE(bp, NULL);
288 XFS_BUF_CLR_IODONE_FUNC(bp);
293 * This routine finds (to an approximation) the first block in the physical
294 * log which contains the given cycle. It uses a binary search algorithm.
295 * Note that the algorithm can not be perfect because the disk will not
296 * necessarily be perfect.
299 xlog_find_cycle_start(
302 xfs_daddr_t first_blk,
303 xfs_daddr_t *last_blk,
311 mid_blk = BLK_AVG(first_blk, *last_blk);
312 while (mid_blk != first_blk && mid_blk != *last_blk) {
313 if ((error = xlog_bread(log, mid_blk, 1, bp)))
315 offset = xlog_align(log, mid_blk, 1, bp);
316 mid_cycle = GET_CYCLE(offset, ARCH_CONVERT);
317 if (mid_cycle == cycle) {
319 /* last_half_cycle == mid_cycle */
322 /* first_half_cycle == mid_cycle */
324 mid_blk = BLK_AVG(first_blk, *last_blk);
326 ASSERT((mid_blk == first_blk && mid_blk+1 == *last_blk) ||
327 (mid_blk == *last_blk && mid_blk-1 == first_blk));
333 * Check that the range of blocks does not contain the cycle number
334 * given. The scan needs to occur from front to back and the ptr into the
335 * region must be updated since a later routine will need to perform another
336 * test. If the region is completely good, we end up returning the same
339 * Set blkno to -1 if we encounter no errors. This is an invalid block number
340 * since we don't ever expect logs to get this large.
343 xlog_find_verify_cycle(
345 xfs_daddr_t start_blk,
347 uint stop_on_cycle_no,
348 xfs_daddr_t *new_blk)
354 xfs_caddr_t buf = NULL;
357 bufblks = 1 << ffs(nbblks);
359 while (!(bp = xlog_get_bp(log, bufblks))) {
360 /* can't get enough memory to do everything in one big buffer */
362 if (bufblks <= log->l_sectbb_log)
366 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
369 bcount = min(bufblks, (start_blk + nbblks - i));
371 if ((error = xlog_bread(log, i, bcount, bp)))
374 buf = xlog_align(log, i, bcount, bp);
375 for (j = 0; j < bcount; j++) {
376 cycle = GET_CYCLE(buf, ARCH_CONVERT);
377 if (cycle == stop_on_cycle_no) {
394 * Potentially backup over partial log record write.
396 * In the typical case, last_blk is the number of the block directly after
397 * a good log record. Therefore, we subtract one to get the block number
398 * of the last block in the given buffer. extra_bblks contains the number
399 * of blocks we would have read on a previous read. This happens when the
400 * last log record is split over the end of the physical log.
402 * extra_bblks is the number of blocks potentially verified on a previous
403 * call to this routine.
406 xlog_find_verify_log_record(
408 xfs_daddr_t start_blk,
409 xfs_daddr_t *last_blk,
414 xfs_caddr_t offset = NULL;
415 xlog_rec_header_t *head = NULL;
418 int num_blks = *last_blk - start_blk;
421 ASSERT(start_blk != 0 || *last_blk != start_blk);
423 if (!(bp = xlog_get_bp(log, num_blks))) {
424 if (!(bp = xlog_get_bp(log, 1)))
428 if ((error = xlog_bread(log, start_blk, num_blks, bp)))
430 offset = xlog_align(log, start_blk, num_blks, bp);
431 offset += ((num_blks - 1) << BBSHIFT);
434 for (i = (*last_blk) - 1; i >= 0; i--) {
436 /* valid log record not found */
438 "XFS: Log inconsistent (didn't find previous header)");
440 error = XFS_ERROR(EIO);
445 if ((error = xlog_bread(log, i, 1, bp)))
447 offset = xlog_align(log, i, 1, bp);
450 head = (xlog_rec_header_t *)offset;
452 if (XLOG_HEADER_MAGIC_NUM ==
453 INT_GET(head->h_magicno, ARCH_CONVERT))
461 * We hit the beginning of the physical log & still no header. Return
462 * to caller. If caller can handle a return of -1, then this routine
463 * will be called again for the end of the physical log.
471 * We have the final block of the good log (the first block
472 * of the log record _before_ the head. So we check the uuid.
474 if ((error = xlog_header_check_mount(log->l_mp, head)))
478 * We may have found a log record header before we expected one.
479 * last_blk will be the 1st block # with a given cycle #. We may end
480 * up reading an entire log record. In this case, we don't want to
481 * reset last_blk. Only when last_blk points in the middle of a log
482 * record do we update last_blk.
484 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
485 uint h_size = INT_GET(head->h_size, ARCH_CONVERT);
487 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
488 if (h_size % XLOG_HEADER_CYCLE_SIZE)
494 if (*last_blk - i + extra_bblks
495 != BTOBB(INT_GET(head->h_len, ARCH_CONVERT)) + xhdrs)
504 * Head is defined to be the point of the log where the next log write
505 * write could go. This means that incomplete LR writes at the end are
506 * eliminated when calculating the head. We aren't guaranteed that previous
507 * LR have complete transactions. We only know that a cycle number of
508 * current cycle number -1 won't be present in the log if we start writing
509 * from our current block number.
511 * last_blk contains the block number of the first block with a given
514 * Return: zero if normal, non-zero if error.
519 xfs_daddr_t *return_head_blk)
523 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
525 uint first_half_cycle, last_half_cycle;
527 int error, log_bbnum = log->l_logBBsize;
529 /* Is the end of the log device zeroed? */
530 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
531 *return_head_blk = first_blk;
533 /* Is the whole lot zeroed? */
535 /* Linux XFS shouldn't generate totally zeroed logs -
536 * mkfs etc write a dummy unmount record to a fresh
537 * log so we can store the uuid in there
539 xlog_warn("XFS: totally zeroed log");
544 xlog_warn("XFS: empty log check failed");
548 first_blk = 0; /* get cycle # of 1st block */
549 bp = xlog_get_bp(log, 1);
552 if ((error = xlog_bread(log, 0, 1, bp)))
554 offset = xlog_align(log, 0, 1, bp);
555 first_half_cycle = GET_CYCLE(offset, ARCH_CONVERT);
557 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
558 if ((error = xlog_bread(log, last_blk, 1, bp)))
560 offset = xlog_align(log, last_blk, 1, bp);
561 last_half_cycle = GET_CYCLE(offset, ARCH_CONVERT);
562 ASSERT(last_half_cycle != 0);
565 * If the 1st half cycle number is equal to the last half cycle number,
566 * then the entire log is stamped with the same cycle number. In this
567 * case, head_blk can't be set to zero (which makes sense). The below
568 * math doesn't work out properly with head_blk equal to zero. Instead,
569 * we set it to log_bbnum which is an invalid block number, but this
570 * value makes the math correct. If head_blk doesn't changed through
571 * all the tests below, *head_blk is set to zero at the very end rather
572 * than log_bbnum. In a sense, log_bbnum and zero are the same block
573 * in a circular file.
575 if (first_half_cycle == last_half_cycle) {
577 * In this case we believe that the entire log should have
578 * cycle number last_half_cycle. We need to scan backwards
579 * from the end verifying that there are no holes still
580 * containing last_half_cycle - 1. If we find such a hole,
581 * then the start of that hole will be the new head. The
582 * simple case looks like
583 * x | x ... | x - 1 | x
584 * Another case that fits this picture would be
585 * x | x + 1 | x ... | x
586 * In this case the head really is somwhere at the end of the
587 * log, as one of the latest writes at the beginning was
590 * x | x + 1 | x ... | x - 1 | x
591 * This is really the combination of the above two cases, and
592 * the head has to end up at the start of the x-1 hole at the
595 * In the 256k log case, we will read from the beginning to the
596 * end of the log and search for cycle numbers equal to x-1.
597 * We don't worry about the x+1 blocks that we encounter,
598 * because we know that they cannot be the head since the log
601 head_blk = log_bbnum;
602 stop_on_cycle = last_half_cycle - 1;
605 * In this case we want to find the first block with cycle
606 * number matching last_half_cycle. We expect the log to be
609 * The first block with cycle number x (last_half_cycle) will
610 * be where the new head belongs. First we do a binary search
611 * for the first occurrence of last_half_cycle. The binary
612 * search may not be totally accurate, so then we scan back
613 * from there looking for occurrences of last_half_cycle before
614 * us. If that backwards scan wraps around the beginning of
615 * the log, then we look for occurrences of last_half_cycle - 1
616 * at the end of the log. The cases we're looking for look
618 * x + 1 ... | x | x + 1 | x ...
619 * ^ binary search stopped here
621 * x + 1 ... | x ... | x - 1 | x
622 * <---------> less than scan distance
624 stop_on_cycle = last_half_cycle;
625 if ((error = xlog_find_cycle_start(log, bp, first_blk,
626 &head_blk, last_half_cycle)))
631 * Now validate the answer. Scan back some number of maximum possible
632 * blocks and make sure each one has the expected cycle number. The
633 * maximum is determined by the total possible amount of buffering
634 * in the in-core log. The following number can be made tighter if
635 * we actually look at the block size of the filesystem.
637 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
638 if (head_blk >= num_scan_bblks) {
640 * We are guaranteed that the entire check can be performed
643 start_blk = head_blk - num_scan_bblks;
644 if ((error = xlog_find_verify_cycle(log,
645 start_blk, num_scan_bblks,
646 stop_on_cycle, &new_blk)))
650 } else { /* need to read 2 parts of log */
652 * We are going to scan backwards in the log in two parts.
653 * First we scan the physical end of the log. In this part
654 * of the log, we are looking for blocks with cycle number
655 * last_half_cycle - 1.
656 * If we find one, then we know that the log starts there, as
657 * we've found a hole that didn't get written in going around
658 * the end of the physical log. The simple case for this is
659 * x + 1 ... | x ... | x - 1 | x
660 * <---------> less than scan distance
661 * If all of the blocks at the end of the log have cycle number
662 * last_half_cycle, then we check the blocks at the start of
663 * the log looking for occurrences of last_half_cycle. If we
664 * find one, then our current estimate for the location of the
665 * first occurrence of last_half_cycle is wrong and we move
666 * back to the hole we've found. This case looks like
667 * x + 1 ... | x | x + 1 | x ...
668 * ^ binary search stopped here
669 * Another case we need to handle that only occurs in 256k
671 * x + 1 ... | x ... | x+1 | x ...
672 * ^ binary search stops here
673 * In a 256k log, the scan at the end of the log will see the
674 * x + 1 blocks. We need to skip past those since that is
675 * certainly not the head of the log. By searching for
676 * last_half_cycle-1 we accomplish that.
678 start_blk = log_bbnum - num_scan_bblks + head_blk;
679 ASSERT(head_blk <= INT_MAX &&
680 (xfs_daddr_t) num_scan_bblks - head_blk >= 0);
681 if ((error = xlog_find_verify_cycle(log, start_blk,
682 num_scan_bblks - (int)head_blk,
683 (stop_on_cycle - 1), &new_blk)))
691 * Scan beginning of log now. The last part of the physical
692 * log is good. This scan needs to verify that it doesn't find
693 * the last_half_cycle.
696 ASSERT(head_blk <= INT_MAX);
697 if ((error = xlog_find_verify_cycle(log,
698 start_blk, (int)head_blk,
699 stop_on_cycle, &new_blk)))
707 * Now we need to make sure head_blk is not pointing to a block in
708 * the middle of a log record.
710 num_scan_bblks = XLOG_REC_SHIFT(log);
711 if (head_blk >= num_scan_bblks) {
712 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
714 /* start ptr at last block ptr before head_blk */
715 if ((error = xlog_find_verify_log_record(log, start_blk,
716 &head_blk, 0)) == -1) {
717 error = XFS_ERROR(EIO);
723 ASSERT(head_blk <= INT_MAX);
724 if ((error = xlog_find_verify_log_record(log, start_blk,
725 &head_blk, 0)) == -1) {
726 /* We hit the beginning of the log during our search */
727 start_blk = log_bbnum - num_scan_bblks + head_blk;
729 ASSERT(start_blk <= INT_MAX &&
730 (xfs_daddr_t) log_bbnum-start_blk >= 0);
731 ASSERT(head_blk <= INT_MAX);
732 if ((error = xlog_find_verify_log_record(log,
734 (int)head_blk)) == -1) {
735 error = XFS_ERROR(EIO);
739 if (new_blk != log_bbnum)
746 if (head_blk == log_bbnum)
747 *return_head_blk = 0;
749 *return_head_blk = head_blk;
751 * When returning here, we have a good block number. Bad block
752 * means that during a previous crash, we didn't have a clean break
753 * from cycle number N to cycle number N-1. In this case, we need
754 * to find the first block with cycle number N-1.
762 xlog_warn("XFS: failed to find log head");
767 * Find the sync block number or the tail of the log.
769 * This will be the block number of the last record to have its
770 * associated buffers synced to disk. Every log record header has
771 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
772 * to get a sync block number. The only concern is to figure out which
773 * log record header to believe.
775 * The following algorithm uses the log record header with the largest
776 * lsn. The entire log record does not need to be valid. We only care
777 * that the header is valid.
779 * We could speed up search by using current head_blk buffer, but it is not
785 xfs_daddr_t *head_blk,
786 xfs_daddr_t *tail_blk,
789 xlog_rec_header_t *rhead;
790 xlog_op_header_t *op_head;
791 xfs_caddr_t offset = NULL;
794 xfs_daddr_t umount_data_blk;
795 xfs_daddr_t after_umount_blk;
802 * Find previous log record
804 if ((error = xlog_find_head(log, head_blk)))
807 bp = xlog_get_bp(log, 1);
810 if (*head_blk == 0) { /* special case */
811 if ((error = xlog_bread(log, 0, 1, bp)))
813 offset = xlog_align(log, 0, 1, bp);
814 if (GET_CYCLE(offset, ARCH_CONVERT) == 0) {
816 /* leave all other log inited values alone */
822 * Search backwards looking for log record header block
824 ASSERT(*head_blk < INT_MAX);
825 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
826 if ((error = xlog_bread(log, i, 1, bp)))
828 offset = xlog_align(log, i, 1, bp);
829 if (XLOG_HEADER_MAGIC_NUM ==
830 INT_GET(*(uint *)offset, ARCH_CONVERT)) {
836 * If we haven't found the log record header block, start looking
837 * again from the end of the physical log. XXXmiken: There should be
838 * a check here to make sure we didn't search more than N blocks in
842 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
843 if ((error = xlog_bread(log, i, 1, bp)))
845 offset = xlog_align(log, i, 1, bp);
846 if (XLOG_HEADER_MAGIC_NUM ==
847 INT_GET(*(uint*)offset, ARCH_CONVERT)) {
854 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
856 return XFS_ERROR(EIO);
859 /* find blk_no of tail of log */
860 rhead = (xlog_rec_header_t *)offset;
861 *tail_blk = BLOCK_LSN(INT_GET(rhead->h_tail_lsn, ARCH_CONVERT));
864 * Reset log values according to the state of the log when we
865 * crashed. In the case where head_blk == 0, we bump curr_cycle
866 * one because the next write starts a new cycle rather than
867 * continuing the cycle of the last good log record. At this
868 * point we have guaranteed that all partial log records have been
869 * accounted for. Therefore, we know that the last good log record
870 * written was complete and ended exactly on the end boundary
871 * of the physical log.
873 log->l_prev_block = i;
874 log->l_curr_block = (int)*head_blk;
875 log->l_curr_cycle = INT_GET(rhead->h_cycle, ARCH_CONVERT);
878 log->l_tail_lsn = INT_GET(rhead->h_tail_lsn, ARCH_CONVERT);
879 log->l_last_sync_lsn = INT_GET(rhead->h_lsn, ARCH_CONVERT);
880 log->l_grant_reserve_cycle = log->l_curr_cycle;
881 log->l_grant_reserve_bytes = BBTOB(log->l_curr_block);
882 log->l_grant_write_cycle = log->l_curr_cycle;
883 log->l_grant_write_bytes = BBTOB(log->l_curr_block);
886 * Look for unmount record. If we find it, then we know there
887 * was a clean unmount. Since 'i' could be the last block in
888 * the physical log, we convert to a log block before comparing
891 * Save the current tail lsn to use to pass to
892 * xlog_clear_stale_blocks() below. We won't want to clear the
893 * unmount record if there is one, so we pass the lsn of the
894 * unmount record rather than the block after it.
896 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
897 int h_size = INT_GET(rhead->h_size, ARCH_CONVERT);
898 int h_version = INT_GET(rhead->h_version, ARCH_CONVERT);
900 if ((h_version & XLOG_VERSION_2) &&
901 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
902 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
903 if (h_size % XLOG_HEADER_CYCLE_SIZE)
911 after_umount_blk = (i + hblks + (int)
912 BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT))) % log->l_logBBsize;
913 tail_lsn = log->l_tail_lsn;
914 if (*head_blk == after_umount_blk &&
915 INT_GET(rhead->h_num_logops, ARCH_CONVERT) == 1) {
916 umount_data_blk = (i + hblks) % log->l_logBBsize;
917 if ((error = xlog_bread(log, umount_data_blk, 1, bp))) {
920 offset = xlog_align(log, umount_data_blk, 1, bp);
921 op_head = (xlog_op_header_t *)offset;
922 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
924 * Set tail and last sync so that newly written
925 * log records will point recovery to after the
926 * current unmount record.
928 ASSIGN_ANY_LSN_HOST(log->l_tail_lsn, log->l_curr_cycle,
930 ASSIGN_ANY_LSN_HOST(log->l_last_sync_lsn, log->l_curr_cycle,
932 *tail_blk = after_umount_blk;
937 * Make sure that there are no blocks in front of the head
938 * with the same cycle number as the head. This can happen
939 * because we allow multiple outstanding log writes concurrently,
940 * and the later writes might make it out before earlier ones.
942 * We use the lsn from before modifying it so that we'll never
943 * overwrite the unmount record after a clean unmount.
945 * Do this only if we are going to recover the filesystem
947 * NOTE: This used to say "if (!readonly)"
948 * However on Linux, we can & do recover a read-only filesystem.
949 * We only skip recovery if NORECOVERY is specified on mount,
950 * in which case we would not be here.
952 * But... if the -device- itself is readonly, just skip this.
953 * We can't recover this device anyway, so it won't matter.
955 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) {
956 error = xlog_clear_stale_blocks(log, tail_lsn);
964 xlog_warn("XFS: failed to locate log tail");
969 * Is the log zeroed at all?
971 * The last binary search should be changed to perform an X block read
972 * once X becomes small enough. You can then search linearly through
973 * the X blocks. This will cut down on the number of reads we need to do.
975 * If the log is partially zeroed, this routine will pass back the blkno
976 * of the first block with cycle number 0. It won't have a complete LR
980 * 0 => the log is completely written to
981 * -1 => use *blk_no as the first block of the log
982 * >0 => error has occurred
991 uint first_cycle, last_cycle;
992 xfs_daddr_t new_blk, last_blk, start_blk;
993 xfs_daddr_t num_scan_bblks;
994 int error, log_bbnum = log->l_logBBsize;
996 /* check totally zeroed log */
997 bp = xlog_get_bp(log, 1);
1000 if ((error = xlog_bread(log, 0, 1, bp)))
1002 offset = xlog_align(log, 0, 1, bp);
1003 first_cycle = GET_CYCLE(offset, ARCH_CONVERT);
1004 if (first_cycle == 0) { /* completely zeroed log */
1010 /* check partially zeroed log */
1011 if ((error = xlog_bread(log, log_bbnum-1, 1, bp)))
1013 offset = xlog_align(log, log_bbnum-1, 1, bp);
1014 last_cycle = GET_CYCLE(offset, ARCH_CONVERT);
1015 if (last_cycle != 0) { /* log completely written to */
1018 } else if (first_cycle != 1) {
1020 * If the cycle of the last block is zero, the cycle of
1021 * the first block must be 1. If it's not, maybe we're
1022 * not looking at a log... Bail out.
1024 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1025 return XFS_ERROR(EINVAL);
1028 /* we have a partially zeroed log */
1029 last_blk = log_bbnum-1;
1030 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1034 * Validate the answer. Because there is no way to guarantee that
1035 * the entire log is made up of log records which are the same size,
1036 * we scan over the defined maximum blocks. At this point, the maximum
1037 * is not chosen to mean anything special. XXXmiken
1039 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1040 ASSERT(num_scan_bblks <= INT_MAX);
1042 if (last_blk < num_scan_bblks)
1043 num_scan_bblks = last_blk;
1044 start_blk = last_blk - num_scan_bblks;
1047 * We search for any instances of cycle number 0 that occur before
1048 * our current estimate of the head. What we're trying to detect is
1049 * 1 ... | 0 | 1 | 0...
1050 * ^ binary search ends here
1052 if ((error = xlog_find_verify_cycle(log, start_blk,
1053 (int)num_scan_bblks, 0, &new_blk)))
1059 * Potentially backup over partial log record write. We don't need
1060 * to search the end of the log because we know it is zero.
1062 if ((error = xlog_find_verify_log_record(log, start_blk,
1063 &last_blk, 0)) == -1) {
1064 error = XFS_ERROR(EIO);
1078 * These are simple subroutines used by xlog_clear_stale_blocks() below
1079 * to initialize a buffer full of empty log record headers and write
1080 * them into the log.
1091 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1093 memset(buf, 0, BBSIZE);
1094 INT_SET(recp->h_magicno, ARCH_CONVERT, XLOG_HEADER_MAGIC_NUM);
1095 INT_SET(recp->h_cycle, ARCH_CONVERT, cycle);
1096 INT_SET(recp->h_version, ARCH_CONVERT,
1097 XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? 2 : 1);
1098 ASSIGN_ANY_LSN_DISK(recp->h_lsn, cycle, block);
1099 ASSIGN_ANY_LSN_DISK(recp->h_tail_lsn, tail_cycle, tail_block);
1100 INT_SET(recp->h_fmt, ARCH_CONVERT, XLOG_FMT);
1101 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1105 xlog_write_log_records(
1116 int sectbb = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
1117 int end_block = start_block + blocks;
1122 bufblks = 1 << ffs(blocks);
1123 while (!(bp = xlog_get_bp(log, bufblks))) {
1125 if (bufblks <= log->l_sectbb_log)
1129 /* We may need to do a read at the start to fill in part of
1130 * the buffer in the starting sector not covered by the first
1133 balign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, start_block);
1134 if (balign != start_block) {
1135 if ((error = xlog_bread(log, start_block, 1, bp))) {
1139 j = start_block - balign;
1142 for (i = start_block; i < end_block; i += bufblks) {
1143 int bcount, endcount;
1145 bcount = min(bufblks, end_block - start_block);
1146 endcount = bcount - j;
1148 /* We may need to do a read at the end to fill in part of
1149 * the buffer in the final sector not covered by the write.
1150 * If this is the same sector as the above read, skip it.
1152 ealign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, end_block);
1153 if (j == 0 && (start_block + endcount > ealign)) {
1154 offset = XFS_BUF_PTR(bp);
1155 balign = BBTOB(ealign - start_block);
1156 XFS_BUF_SET_PTR(bp, offset + balign, BBTOB(sectbb));
1157 if ((error = xlog_bread(log, ealign, sectbb, bp)))
1159 XFS_BUF_SET_PTR(bp, offset, bufblks);
1162 offset = xlog_align(log, start_block, endcount, bp);
1163 for (; j < endcount; j++) {
1164 xlog_add_record(log, offset, cycle, i+j,
1165 tail_cycle, tail_block);
1168 error = xlog_bwrite(log, start_block, endcount, bp);
1171 start_block += endcount;
1179 * This routine is called to blow away any incomplete log writes out
1180 * in front of the log head. We do this so that we won't become confused
1181 * if we come up, write only a little bit more, and then crash again.
1182 * If we leave the partial log records out there, this situation could
1183 * cause us to think those partial writes are valid blocks since they
1184 * have the current cycle number. We get rid of them by overwriting them
1185 * with empty log records with the old cycle number rather than the
1188 * The tail lsn is passed in rather than taken from
1189 * the log so that we will not write over the unmount record after a
1190 * clean unmount in a 512 block log. Doing so would leave the log without
1191 * any valid log records in it until a new one was written. If we crashed
1192 * during that time we would not be able to recover.
1195 xlog_clear_stale_blocks(
1199 int tail_cycle, head_cycle;
1200 int tail_block, head_block;
1201 int tail_distance, max_distance;
1205 tail_cycle = CYCLE_LSN(tail_lsn);
1206 tail_block = BLOCK_LSN(tail_lsn);
1207 head_cycle = log->l_curr_cycle;
1208 head_block = log->l_curr_block;
1211 * Figure out the distance between the new head of the log
1212 * and the tail. We want to write over any blocks beyond the
1213 * head that we may have written just before the crash, but
1214 * we don't want to overwrite the tail of the log.
1216 if (head_cycle == tail_cycle) {
1218 * The tail is behind the head in the physical log,
1219 * so the distance from the head to the tail is the
1220 * distance from the head to the end of the log plus
1221 * the distance from the beginning of the log to the
1224 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1225 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1226 XFS_ERRLEVEL_LOW, log->l_mp);
1227 return XFS_ERROR(EFSCORRUPTED);
1229 tail_distance = tail_block + (log->l_logBBsize - head_block);
1232 * The head is behind the tail in the physical log,
1233 * so the distance from the head to the tail is just
1234 * the tail block minus the head block.
1236 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1237 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1238 XFS_ERRLEVEL_LOW, log->l_mp);
1239 return XFS_ERROR(EFSCORRUPTED);
1241 tail_distance = tail_block - head_block;
1245 * If the head is right up against the tail, we can't clear
1248 if (tail_distance <= 0) {
1249 ASSERT(tail_distance == 0);
1253 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1255 * Take the smaller of the maximum amount of outstanding I/O
1256 * we could have and the distance to the tail to clear out.
1257 * We take the smaller so that we don't overwrite the tail and
1258 * we don't waste all day writing from the head to the tail
1261 max_distance = MIN(max_distance, tail_distance);
1263 if ((head_block + max_distance) <= log->l_logBBsize) {
1265 * We can stomp all the blocks we need to without
1266 * wrapping around the end of the log. Just do it
1267 * in a single write. Use the cycle number of the
1268 * current cycle minus one so that the log will look like:
1271 error = xlog_write_log_records(log, (head_cycle - 1),
1272 head_block, max_distance, tail_cycle,
1278 * We need to wrap around the end of the physical log in
1279 * order to clear all the blocks. Do it in two separate
1280 * I/Os. The first write should be from the head to the
1281 * end of the physical log, and it should use the current
1282 * cycle number minus one just like above.
1284 distance = log->l_logBBsize - head_block;
1285 error = xlog_write_log_records(log, (head_cycle - 1),
1286 head_block, distance, tail_cycle,
1293 * Now write the blocks at the start of the physical log.
1294 * This writes the remainder of the blocks we want to clear.
1295 * It uses the current cycle number since we're now on the
1296 * same cycle as the head so that we get:
1297 * n ... n ... | n - 1 ...
1298 * ^^^^^ blocks we're writing
1300 distance = max_distance - (log->l_logBBsize - head_block);
1301 error = xlog_write_log_records(log, head_cycle, 0, distance,
1302 tail_cycle, tail_block);
1310 /******************************************************************************
1312 * Log recover routines
1314 ******************************************************************************
1317 STATIC xlog_recover_t *
1318 xlog_recover_find_tid(
1322 xlog_recover_t *p = q;
1325 if (p->r_log_tid == tid)
1333 xlog_recover_put_hashq(
1335 xlog_recover_t *trans)
1342 xlog_recover_add_item(
1343 xlog_recover_item_t **itemq)
1345 xlog_recover_item_t *item;
1347 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1348 xlog_recover_insert_item_backq(itemq, item);
1352 xlog_recover_add_to_cont_trans(
1353 xlog_recover_t *trans,
1357 xlog_recover_item_t *item;
1358 xfs_caddr_t ptr, old_ptr;
1361 item = trans->r_itemq;
1363 /* finish copying rest of trans header */
1364 xlog_recover_add_item(&trans->r_itemq);
1365 ptr = (xfs_caddr_t) &trans->r_theader +
1366 sizeof(xfs_trans_header_t) - len;
1367 memcpy(ptr, dp, len); /* d, s, l */
1370 item = item->ri_prev;
1372 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1373 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1375 ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
1376 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1377 item->ri_buf[item->ri_cnt-1].i_len += len;
1378 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1383 * The next region to add is the start of a new region. It could be
1384 * a whole region or it could be the first part of a new region. Because
1385 * of this, the assumption here is that the type and size fields of all
1386 * format structures fit into the first 32 bits of the structure.
1388 * This works because all regions must be 32 bit aligned. Therefore, we
1389 * either have both fields or we have neither field. In the case we have
1390 * neither field, the data part of the region is zero length. We only have
1391 * a log_op_header and can throw away the header since a new one will appear
1392 * later. If we have at least 4 bytes, then we can determine how many regions
1393 * will appear in the current log item.
1396 xlog_recover_add_to_trans(
1397 xlog_recover_t *trans,
1401 xfs_inode_log_format_t *in_f; /* any will do */
1402 xlog_recover_item_t *item;
1407 item = trans->r_itemq;
1409 ASSERT(*(uint *)dp == XFS_TRANS_HEADER_MAGIC);
1410 if (len == sizeof(xfs_trans_header_t))
1411 xlog_recover_add_item(&trans->r_itemq);
1412 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1416 ptr = kmem_alloc(len, KM_SLEEP);
1417 memcpy(ptr, dp, len);
1418 in_f = (xfs_inode_log_format_t *)ptr;
1420 if (item->ri_prev->ri_total != 0 &&
1421 item->ri_prev->ri_total == item->ri_prev->ri_cnt) {
1422 xlog_recover_add_item(&trans->r_itemq);
1424 item = trans->r_itemq;
1425 item = item->ri_prev;
1427 if (item->ri_total == 0) { /* first region to be added */
1428 item->ri_total = in_f->ilf_size;
1429 ASSERT(item->ri_total <= XLOG_MAX_REGIONS_IN_ITEM);
1430 item->ri_buf = kmem_zalloc((item->ri_total *
1431 sizeof(xfs_log_iovec_t)), KM_SLEEP);
1433 ASSERT(item->ri_total > item->ri_cnt);
1434 /* Description region is ri_buf[0] */
1435 item->ri_buf[item->ri_cnt].i_addr = ptr;
1436 item->ri_buf[item->ri_cnt].i_len = len;
1442 xlog_recover_new_tid(
1447 xlog_recover_t *trans;
1449 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1450 trans->r_log_tid = tid;
1452 xlog_recover_put_hashq(q, trans);
1456 xlog_recover_unlink_tid(
1458 xlog_recover_t *trans)
1469 if (tp->r_next == trans) {
1477 "XFS: xlog_recover_unlink_tid: trans not found");
1479 return XFS_ERROR(EIO);
1481 tp->r_next = tp->r_next->r_next;
1487 xlog_recover_insert_item_backq(
1488 xlog_recover_item_t **q,
1489 xlog_recover_item_t *item)
1492 item->ri_prev = item->ri_next = item;
1496 item->ri_prev = (*q)->ri_prev;
1497 (*q)->ri_prev = item;
1498 item->ri_prev->ri_next = item;
1503 xlog_recover_insert_item_frontq(
1504 xlog_recover_item_t **q,
1505 xlog_recover_item_t *item)
1507 xlog_recover_insert_item_backq(q, item);
1512 xlog_recover_reorder_trans(
1514 xlog_recover_t *trans)
1516 xlog_recover_item_t *first_item, *itemq, *itemq_next;
1517 xfs_buf_log_format_t *buf_f;
1518 xfs_buf_log_format_v1_t *obuf_f;
1521 first_item = itemq = trans->r_itemq;
1522 trans->r_itemq = NULL;
1524 itemq_next = itemq->ri_next;
1525 buf_f = (xfs_buf_log_format_t *)itemq->ri_buf[0].i_addr;
1526 switch (ITEM_TYPE(itemq)) {
1528 flags = buf_f->blf_flags;
1530 case XFS_LI_6_1_BUF:
1531 case XFS_LI_5_3_BUF:
1532 obuf_f = (xfs_buf_log_format_v1_t*)buf_f;
1533 flags = obuf_f->blf_flags;
1537 switch (ITEM_TYPE(itemq)) {
1539 case XFS_LI_6_1_BUF:
1540 case XFS_LI_5_3_BUF:
1541 if (!(flags & XFS_BLI_CANCEL)) {
1542 xlog_recover_insert_item_frontq(&trans->r_itemq,
1547 case XFS_LI_6_1_INODE:
1548 case XFS_LI_5_3_INODE:
1550 case XFS_LI_QUOTAOFF:
1553 xlog_recover_insert_item_backq(&trans->r_itemq, itemq);
1557 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1559 return XFS_ERROR(EIO);
1562 } while (first_item != itemq);
1567 * Build up the table of buf cancel records so that we don't replay
1568 * cancelled data in the second pass. For buffer records that are
1569 * not cancel records, there is nothing to do here so we just return.
1571 * If we get a cancel record which is already in the table, this indicates
1572 * that the buffer was cancelled multiple times. In order to ensure
1573 * that during pass 2 we keep the record in the table until we reach its
1574 * last occurrence in the log, we keep a reference count in the cancel
1575 * record in the table to tell us how many times we expect to see this
1576 * record during the second pass.
1579 xlog_recover_do_buffer_pass1(
1581 xfs_buf_log_format_t *buf_f)
1583 xfs_buf_cancel_t *bcp;
1584 xfs_buf_cancel_t *nextp;
1585 xfs_buf_cancel_t *prevp;
1586 xfs_buf_cancel_t **bucket;
1587 xfs_buf_log_format_v1_t *obuf_f;
1588 xfs_daddr_t blkno = 0;
1592 switch (buf_f->blf_type) {
1594 blkno = buf_f->blf_blkno;
1595 len = buf_f->blf_len;
1596 flags = buf_f->blf_flags;
1598 case XFS_LI_6_1_BUF:
1599 case XFS_LI_5_3_BUF:
1600 obuf_f = (xfs_buf_log_format_v1_t*)buf_f;
1601 blkno = (xfs_daddr_t) obuf_f->blf_blkno;
1602 len = obuf_f->blf_len;
1603 flags = obuf_f->blf_flags;
1608 * If this isn't a cancel buffer item, then just return.
1610 if (!(flags & XFS_BLI_CANCEL))
1614 * Insert an xfs_buf_cancel record into the hash table of
1615 * them. If there is already an identical record, bump
1616 * its reference count.
1618 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1619 XLOG_BC_TABLE_SIZE];
1621 * If the hash bucket is empty then just insert a new record into
1624 if (*bucket == NULL) {
1625 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1627 bcp->bc_blkno = blkno;
1629 bcp->bc_refcount = 1;
1630 bcp->bc_next = NULL;
1636 * The hash bucket is not empty, so search for duplicates of our
1637 * record. If we find one them just bump its refcount. If not
1638 * then add us at the end of the list.
1642 while (nextp != NULL) {
1643 if (nextp->bc_blkno == blkno && nextp->bc_len == len) {
1644 nextp->bc_refcount++;
1648 nextp = nextp->bc_next;
1650 ASSERT(prevp != NULL);
1651 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1653 bcp->bc_blkno = blkno;
1655 bcp->bc_refcount = 1;
1656 bcp->bc_next = NULL;
1657 prevp->bc_next = bcp;
1661 * Check to see whether the buffer being recovered has a corresponding
1662 * entry in the buffer cancel record table. If it does then return 1
1663 * so that it will be cancelled, otherwise return 0. If the buffer is
1664 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
1665 * the refcount on the entry in the table and remove it from the table
1666 * if this is the last reference.
1668 * We remove the cancel record from the table when we encounter its
1669 * last occurrence in the log so that if the same buffer is re-used
1670 * again after its last cancellation we actually replay the changes
1671 * made at that point.
1674 xlog_check_buffer_cancelled(
1680 xfs_buf_cancel_t *bcp;
1681 xfs_buf_cancel_t *prevp;
1682 xfs_buf_cancel_t **bucket;
1684 if (log->l_buf_cancel_table == NULL) {
1686 * There is nothing in the table built in pass one,
1687 * so this buffer must not be cancelled.
1689 ASSERT(!(flags & XFS_BLI_CANCEL));
1693 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1694 XLOG_BC_TABLE_SIZE];
1698 * There is no corresponding entry in the table built
1699 * in pass one, so this buffer has not been cancelled.
1701 ASSERT(!(flags & XFS_BLI_CANCEL));
1706 * Search for an entry in the buffer cancel table that
1707 * matches our buffer.
1710 while (bcp != NULL) {
1711 if (bcp->bc_blkno == blkno && bcp->bc_len == len) {
1713 * We've go a match, so return 1 so that the
1714 * recovery of this buffer is cancelled.
1715 * If this buffer is actually a buffer cancel
1716 * log item, then decrement the refcount on the
1717 * one in the table and remove it if this is the
1720 if (flags & XFS_BLI_CANCEL) {
1722 if (bcp->bc_refcount == 0) {
1723 if (prevp == NULL) {
1724 *bucket = bcp->bc_next;
1726 prevp->bc_next = bcp->bc_next;
1729 sizeof(xfs_buf_cancel_t));
1738 * We didn't find a corresponding entry in the table, so
1739 * return 0 so that the buffer is NOT cancelled.
1741 ASSERT(!(flags & XFS_BLI_CANCEL));
1746 xlog_recover_do_buffer_pass2(
1748 xfs_buf_log_format_t *buf_f)
1750 xfs_buf_log_format_v1_t *obuf_f;
1751 xfs_daddr_t blkno = 0;
1755 switch (buf_f->blf_type) {
1757 blkno = buf_f->blf_blkno;
1758 flags = buf_f->blf_flags;
1759 len = buf_f->blf_len;
1761 case XFS_LI_6_1_BUF:
1762 case XFS_LI_5_3_BUF:
1763 obuf_f = (xfs_buf_log_format_v1_t*)buf_f;
1764 blkno = (xfs_daddr_t) obuf_f->blf_blkno;
1765 flags = obuf_f->blf_flags;
1766 len = (xfs_daddr_t) obuf_f->blf_len;
1770 return xlog_check_buffer_cancelled(log, blkno, len, flags);
1774 * Perform recovery for a buffer full of inodes. In these buffers,
1775 * the only data which should be recovered is that which corresponds
1776 * to the di_next_unlinked pointers in the on disk inode structures.
1777 * The rest of the data for the inodes is always logged through the
1778 * inodes themselves rather than the inode buffer and is recovered
1779 * in xlog_recover_do_inode_trans().
1781 * The only time when buffers full of inodes are fully recovered is
1782 * when the buffer is full of newly allocated inodes. In this case
1783 * the buffer will not be marked as an inode buffer and so will be
1784 * sent to xlog_recover_do_reg_buffer() below during recovery.
1787 xlog_recover_do_inode_buffer(
1789 xlog_recover_item_t *item,
1791 xfs_buf_log_format_t *buf_f)
1799 int next_unlinked_offset;
1801 xfs_agino_t *logged_nextp;
1802 xfs_agino_t *buffer_nextp;
1803 xfs_buf_log_format_v1_t *obuf_f;
1804 unsigned int *data_map = NULL;
1805 unsigned int map_size = 0;
1807 switch (buf_f->blf_type) {
1809 data_map = buf_f->blf_data_map;
1810 map_size = buf_f->blf_map_size;
1812 case XFS_LI_6_1_BUF:
1813 case XFS_LI_5_3_BUF:
1814 obuf_f = (xfs_buf_log_format_v1_t*)buf_f;
1815 data_map = obuf_f->blf_data_map;
1816 map_size = obuf_f->blf_map_size;
1820 * Set the variables corresponding to the current region to
1821 * 0 so that we'll initialize them on the first pass through
1829 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1830 for (i = 0; i < inodes_per_buf; i++) {
1831 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1832 offsetof(xfs_dinode_t, di_next_unlinked);
1834 while (next_unlinked_offset >=
1835 (reg_buf_offset + reg_buf_bytes)) {
1837 * The next di_next_unlinked field is beyond
1838 * the current logged region. Find the next
1839 * logged region that contains or is beyond
1840 * the current di_next_unlinked field.
1843 bit = xfs_next_bit(data_map, map_size, bit);
1846 * If there are no more logged regions in the
1847 * buffer, then we're done.
1853 nbits = xfs_contig_bits(data_map, map_size,
1856 reg_buf_offset = bit << XFS_BLI_SHIFT;
1857 reg_buf_bytes = nbits << XFS_BLI_SHIFT;
1862 * If the current logged region starts after the current
1863 * di_next_unlinked field, then move on to the next
1864 * di_next_unlinked field.
1866 if (next_unlinked_offset < reg_buf_offset) {
1870 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1871 ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0);
1872 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1875 * The current logged region contains a copy of the
1876 * current di_next_unlinked field. Extract its value
1877 * and copy it to the buffer copy.
1879 logged_nextp = (xfs_agino_t *)
1880 ((char *)(item->ri_buf[item_index].i_addr) +
1881 (next_unlinked_offset - reg_buf_offset));
1882 if (unlikely(*logged_nextp == 0)) {
1883 xfs_fs_cmn_err(CE_ALERT, mp,
1884 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1886 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1887 XFS_ERRLEVEL_LOW, mp);
1888 return XFS_ERROR(EFSCORRUPTED);
1891 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1892 next_unlinked_offset);
1893 INT_SET(*buffer_nextp, ARCH_CONVERT, *logged_nextp);
1900 * Perform a 'normal' buffer recovery. Each logged region of the
1901 * buffer should be copied over the corresponding region in the
1902 * given buffer. The bitmap in the buf log format structure indicates
1903 * where to place the logged data.
1907 xlog_recover_do_reg_buffer(
1909 xlog_recover_item_t *item,
1911 xfs_buf_log_format_t *buf_f)
1916 xfs_buf_log_format_v1_t *obuf_f;
1917 unsigned int *data_map = NULL;
1918 unsigned int map_size = 0;
1921 switch (buf_f->blf_type) {
1923 data_map = buf_f->blf_data_map;
1924 map_size = buf_f->blf_map_size;
1926 case XFS_LI_6_1_BUF:
1927 case XFS_LI_5_3_BUF:
1928 obuf_f = (xfs_buf_log_format_v1_t*)buf_f;
1929 data_map = obuf_f->blf_data_map;
1930 map_size = obuf_f->blf_map_size;
1934 i = 1; /* 0 is the buf format structure */
1936 bit = xfs_next_bit(data_map, map_size, bit);
1939 nbits = xfs_contig_bits(data_map, map_size, bit);
1941 ASSERT(item->ri_buf[i].i_addr != 0);
1942 ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0);
1943 ASSERT(XFS_BUF_COUNT(bp) >=
1944 ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT));
1947 * Do a sanity check if this is a dquot buffer. Just checking
1948 * the first dquot in the buffer should do. XXXThis is
1949 * probably a good thing to do for other buf types also.
1952 if (buf_f->blf_flags &
1953 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
1954 error = xfs_qm_dqcheck((xfs_disk_dquot_t *)
1955 item->ri_buf[i].i_addr,
1956 -1, 0, XFS_QMOPT_DOWARN,
1957 "dquot_buf_recover");
1960 memcpy(xfs_buf_offset(bp,
1961 (uint)bit << XFS_BLI_SHIFT), /* dest */
1962 item->ri_buf[i].i_addr, /* source */
1963 nbits<<XFS_BLI_SHIFT); /* length */
1968 /* Shouldn't be any more regions */
1969 ASSERT(i == item->ri_total);
1973 * Do some primitive error checking on ondisk dquot data structures.
1977 xfs_disk_dquot_t *ddq,
1979 uint type, /* used only when IO_dorepair is true */
1983 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
1987 * We can encounter an uninitialized dquot buffer for 2 reasons:
1988 * 1. If we crash while deleting the quotainode(s), and those blks got
1989 * used for user data. This is because we take the path of regular
1990 * file deletion; however, the size field of quotainodes is never
1991 * updated, so all the tricks that we play in itruncate_finish
1992 * don't quite matter.
1994 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1995 * But the allocation will be replayed so we'll end up with an
1996 * uninitialized quota block.
1998 * This is all fine; things are still consistent, and we haven't lost
1999 * any quota information. Just don't complain about bad dquot blks.
2001 if (INT_GET(ddq->d_magic, ARCH_CONVERT) != XFS_DQUOT_MAGIC) {
2002 if (flags & XFS_QMOPT_DOWARN)
2004 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
2006 INT_GET(ddq->d_magic, ARCH_CONVERT), XFS_DQUOT_MAGIC);
2009 if (INT_GET(ddq->d_version, ARCH_CONVERT) != XFS_DQUOT_VERSION) {
2010 if (flags & XFS_QMOPT_DOWARN)
2012 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
2014 INT_GET(ddq->d_magic, ARCH_CONVERT), XFS_DQUOT_VERSION);
2018 if (INT_GET(ddq->d_flags, ARCH_CONVERT) != XFS_DQ_USER &&
2019 INT_GET(ddq->d_flags, ARCH_CONVERT) != XFS_DQ_PROJ &&
2020 INT_GET(ddq->d_flags, ARCH_CONVERT) != XFS_DQ_GROUP) {
2021 if (flags & XFS_QMOPT_DOWARN)
2023 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
2024 str, id, INT_GET(ddq->d_flags, ARCH_CONVERT));
2028 if (id != -1 && id != INT_GET(ddq->d_id, ARCH_CONVERT)) {
2029 if (flags & XFS_QMOPT_DOWARN)
2031 "%s : ondisk-dquot 0x%p, ID mismatch: "
2032 "0x%x expected, found id 0x%x",
2033 str, ddq, id, INT_GET(ddq->d_id, ARCH_CONVERT));
2037 if (!errs && ddq->d_id) {
2038 if (INT_GET(ddq->d_blk_softlimit, ARCH_CONVERT) &&
2039 INT_GET(ddq->d_bcount, ARCH_CONVERT) >=
2040 INT_GET(ddq->d_blk_softlimit, ARCH_CONVERT)) {
2041 if (!ddq->d_btimer) {
2042 if (flags & XFS_QMOPT_DOWARN)
2044 "%s : Dquot ID 0x%x (0x%p) "
2045 "BLK TIMER NOT STARTED",
2047 INT_GET(ddq->d_id, ARCH_CONVERT), ddq);
2051 if (INT_GET(ddq->d_ino_softlimit, ARCH_CONVERT) &&
2052 INT_GET(ddq->d_icount, ARCH_CONVERT) >=
2053 INT_GET(ddq->d_ino_softlimit, ARCH_CONVERT)) {
2054 if (!ddq->d_itimer) {
2055 if (flags & XFS_QMOPT_DOWARN)
2057 "%s : Dquot ID 0x%x (0x%p) "
2058 "INODE TIMER NOT STARTED",
2060 INT_GET(ddq->d_id, ARCH_CONVERT), ddq);
2064 if (INT_GET(ddq->d_rtb_softlimit, ARCH_CONVERT) &&
2065 INT_GET(ddq->d_rtbcount, ARCH_CONVERT) >=
2066 INT_GET(ddq->d_rtb_softlimit, ARCH_CONVERT)) {
2067 if (!ddq->d_rtbtimer) {
2068 if (flags & XFS_QMOPT_DOWARN)
2070 "%s : Dquot ID 0x%x (0x%p) "
2071 "RTBLK TIMER NOT STARTED",
2073 INT_GET(ddq->d_id, ARCH_CONVERT), ddq);
2079 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2082 if (flags & XFS_QMOPT_DOWARN)
2083 cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id);
2086 * Typically, a repair is only requested by quotacheck.
2089 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2090 memset(d, 0, sizeof(xfs_dqblk_t));
2091 INT_SET(d->dd_diskdq.d_magic, ARCH_CONVERT, XFS_DQUOT_MAGIC);
2092 INT_SET(d->dd_diskdq.d_version, ARCH_CONVERT, XFS_DQUOT_VERSION);
2093 INT_SET(d->dd_diskdq.d_id, ARCH_CONVERT, id);
2094 INT_SET(d->dd_diskdq.d_flags, ARCH_CONVERT, type);
2100 * Perform a dquot buffer recovery.
2101 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2102 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2103 * Else, treat it as a regular buffer and do recovery.
2106 xlog_recover_do_dquot_buffer(
2109 xlog_recover_item_t *item,
2111 xfs_buf_log_format_t *buf_f)
2116 * Filesystems are required to send in quota flags at mount time.
2118 if (mp->m_qflags == 0) {
2123 if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF)
2124 type |= XFS_DQ_USER;
2125 if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF)
2126 type |= XFS_DQ_PROJ;
2127 if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF)
2128 type |= XFS_DQ_GROUP;
2130 * This type of quotas was turned off, so ignore this buffer
2132 if (log->l_quotaoffs_flag & type)
2135 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2139 * This routine replays a modification made to a buffer at runtime.
2140 * There are actually two types of buffer, regular and inode, which
2141 * are handled differently. Inode buffers are handled differently
2142 * in that we only recover a specific set of data from them, namely
2143 * the inode di_next_unlinked fields. This is because all other inode
2144 * data is actually logged via inode records and any data we replay
2145 * here which overlaps that may be stale.
2147 * When meta-data buffers are freed at run time we log a buffer item
2148 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
2149 * of the buffer in the log should not be replayed at recovery time.
2150 * This is so that if the blocks covered by the buffer are reused for
2151 * file data before we crash we don't end up replaying old, freed
2152 * meta-data into a user's file.
2154 * To handle the cancellation of buffer log items, we make two passes
2155 * over the log during recovery. During the first we build a table of
2156 * those buffers which have been cancelled, and during the second we
2157 * only replay those buffers which do not have corresponding cancel
2158 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2159 * for more details on the implementation of the table of cancel records.
2162 xlog_recover_do_buffer_trans(
2164 xlog_recover_item_t *item,
2167 xfs_buf_log_format_t *buf_f;
2168 xfs_buf_log_format_v1_t *obuf_f;
2177 buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr;
2179 if (pass == XLOG_RECOVER_PASS1) {
2181 * In this pass we're only looking for buf items
2182 * with the XFS_BLI_CANCEL bit set.
2184 xlog_recover_do_buffer_pass1(log, buf_f);
2188 * In this pass we want to recover all the buffers
2189 * which have not been cancelled and are not
2190 * cancellation buffers themselves. The routine
2191 * we call here will tell us whether or not to
2192 * continue with the replay of this buffer.
2194 cancel = xlog_recover_do_buffer_pass2(log, buf_f);
2199 switch (buf_f->blf_type) {
2201 blkno = buf_f->blf_blkno;
2202 len = buf_f->blf_len;
2203 flags = buf_f->blf_flags;
2205 case XFS_LI_6_1_BUF:
2206 case XFS_LI_5_3_BUF:
2207 obuf_f = (xfs_buf_log_format_v1_t*)buf_f;
2208 blkno = obuf_f->blf_blkno;
2209 len = obuf_f->blf_len;
2210 flags = obuf_f->blf_flags;
2213 xfs_fs_cmn_err(CE_ALERT, log->l_mp,
2214 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2215 buf_f->blf_type, log->l_mp->m_logname ?
2216 log->l_mp->m_logname : "internal");
2217 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2218 XFS_ERRLEVEL_LOW, log->l_mp);
2219 return XFS_ERROR(EFSCORRUPTED);
2223 if (flags & XFS_BLI_INODE_BUF) {
2224 bp = xfs_buf_read_flags(mp->m_ddev_targp, blkno, len,
2227 bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, 0);
2229 if (XFS_BUF_ISERROR(bp)) {
2230 xfs_ioerror_alert("xlog_recover_do..(read#1)", log->l_mp,
2232 error = XFS_BUF_GETERROR(bp);
2238 if (flags & XFS_BLI_INODE_BUF) {
2239 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2241 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
2242 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2244 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2247 return XFS_ERROR(error);
2250 * Perform delayed write on the buffer. Asynchronous writes will be
2251 * slower when taking into account all the buffers to be flushed.
2253 * Also make sure that only inode buffers with good sizes stay in
2254 * the buffer cache. The kernel moves inodes in buffers of 1 block
2255 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2256 * buffers in the log can be a different size if the log was generated
2257 * by an older kernel using unclustered inode buffers or a newer kernel
2258 * running with a different inode cluster size. Regardless, if the
2259 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2260 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2261 * the buffer out of the buffer cache so that the buffer won't
2262 * overlap with future reads of those inodes.
2264 if (XFS_DINODE_MAGIC ==
2265 INT_GET(*((__uint16_t *)(xfs_buf_offset(bp, 0))), ARCH_CONVERT) &&
2266 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2267 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2269 error = xfs_bwrite(mp, bp);
2271 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2272 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2273 XFS_BUF_SET_FSPRIVATE(bp, mp);
2274 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2275 xfs_bdwrite(mp, bp);
2282 xlog_recover_do_inode_trans(
2284 xlog_recover_item_t *item,
2287 xfs_inode_log_format_t *in_f;
2299 xfs_dinode_core_t *dicp;
2301 if (pass == XLOG_RECOVER_PASS1) {
2305 in_f = (xfs_inode_log_format_t *)item->ri_buf[0].i_addr;
2306 ino = in_f->ilf_ino;
2308 if (ITEM_TYPE(item) == XFS_LI_INODE) {
2309 imap.im_blkno = (xfs_daddr_t)in_f->ilf_blkno;
2310 imap.im_len = in_f->ilf_len;
2311 imap.im_boffset = in_f->ilf_boffset;
2314 * It's an old inode format record. We don't know where
2315 * its cluster is located on disk, and we can't allow
2316 * xfs_imap() to figure it out because the inode btrees
2317 * are not ready to be used. Therefore do not pass the
2318 * XFS_IMAP_LOOKUP flag to xfs_imap(). This will give
2319 * us only the single block in which the inode lives
2320 * rather than its cluster, so we must make sure to
2321 * invalidate the buffer when we write it out below.
2324 xfs_imap(log->l_mp, NULL, ino, &imap, 0);
2328 * Inode buffers can be freed, look out for it,
2329 * and do not replay the inode.
2331 if (xlog_check_buffer_cancelled(log, imap.im_blkno, imap.im_len, 0))
2334 bp = xfs_buf_read_flags(mp->m_ddev_targp, imap.im_blkno, imap.im_len,
2336 if (XFS_BUF_ISERROR(bp)) {
2337 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
2339 error = XFS_BUF_GETERROR(bp);
2344 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2345 dip = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
2348 * Make sure the place we're flushing out to really looks
2351 if (unlikely(INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC)) {
2353 xfs_fs_cmn_err(CE_ALERT, mp,
2354 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2356 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2357 XFS_ERRLEVEL_LOW, mp);
2358 return XFS_ERROR(EFSCORRUPTED);
2360 dicp = (xfs_dinode_core_t*)(item->ri_buf[1].i_addr);
2361 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2363 xfs_fs_cmn_err(CE_ALERT, mp,
2364 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2366 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2367 XFS_ERRLEVEL_LOW, mp);
2368 return XFS_ERROR(EFSCORRUPTED);
2371 /* Skip replay when the on disk inode is newer than the log one */
2372 if (dicp->di_flushiter <
2373 INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT)) {
2375 * Deal with the wrap case, DI_MAX_FLUSH is less
2376 * than smaller numbers
2378 if ((INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT)
2380 (dicp->di_flushiter < (DI_MAX_FLUSH>>1))) {
2387 /* Take the opportunity to reset the flush iteration count */
2388 dicp->di_flushiter = 0;
2390 if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) {
2391 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2392 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2393 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2394 XFS_ERRLEVEL_LOW, mp, dicp);
2396 xfs_fs_cmn_err(CE_ALERT, mp,
2397 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2398 item, dip, bp, ino);
2399 return XFS_ERROR(EFSCORRUPTED);
2401 } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) {
2402 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2403 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2404 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2405 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2406 XFS_ERRLEVEL_LOW, mp, dicp);
2408 xfs_fs_cmn_err(CE_ALERT, mp,
2409 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2410 item, dip, bp, ino);
2411 return XFS_ERROR(EFSCORRUPTED);
2414 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2415 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2416 XFS_ERRLEVEL_LOW, mp, dicp);
2418 xfs_fs_cmn_err(CE_ALERT, mp,
2419 "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",
2421 dicp->di_nextents + dicp->di_anextents,
2423 return XFS_ERROR(EFSCORRUPTED);
2425 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2426 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2427 XFS_ERRLEVEL_LOW, mp, dicp);
2429 xfs_fs_cmn_err(CE_ALERT, mp,
2430 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2431 item, dip, bp, ino, dicp->di_forkoff);
2432 return XFS_ERROR(EFSCORRUPTED);
2434 if (unlikely(item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t))) {
2435 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2436 XFS_ERRLEVEL_LOW, mp, dicp);
2438 xfs_fs_cmn_err(CE_ALERT, mp,
2439 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2440 item->ri_buf[1].i_len, item);
2441 return XFS_ERROR(EFSCORRUPTED);
2444 /* The core is in in-core format */
2445 xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
2446 (xfs_dinode_core_t*)item->ri_buf[1].i_addr, -1);
2448 /* the rest is in on-disk format */
2449 if (item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t)) {
2450 memcpy((xfs_caddr_t) dip + sizeof(xfs_dinode_core_t),
2451 item->ri_buf[1].i_addr + sizeof(xfs_dinode_core_t),
2452 item->ri_buf[1].i_len - sizeof(xfs_dinode_core_t));
2455 fields = in_f->ilf_fields;
2456 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2458 INT_SET(dip->di_u.di_dev, ARCH_CONVERT, in_f->ilf_u.ilfu_rdev);
2462 dip->di_u.di_muuid = in_f->ilf_u.ilfu_uuid;
2466 if (in_f->ilf_size == 2)
2467 goto write_inode_buffer;
2468 len = item->ri_buf[2].i_len;
2469 src = item->ri_buf[2].i_addr;
2470 ASSERT(in_f->ilf_size <= 4);
2471 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2472 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2473 (len == in_f->ilf_dsize));
2475 switch (fields & XFS_ILOG_DFORK) {
2476 case XFS_ILOG_DDATA:
2478 memcpy(&dip->di_u, src, len);
2481 case XFS_ILOG_DBROOT:
2482 xfs_bmbt_to_bmdr((xfs_bmbt_block_t *)src, len,
2483 &(dip->di_u.di_bmbt),
2484 XFS_DFORK_DSIZE(dip, mp));
2489 * There are no data fork flags set.
2491 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2496 * If we logged any attribute data, recover it. There may or
2497 * may not have been any other non-core data logged in this
2500 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2501 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2506 len = item->ri_buf[attr_index].i_len;
2507 src = item->ri_buf[attr_index].i_addr;
2508 ASSERT(len == in_f->ilf_asize);
2510 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2511 case XFS_ILOG_ADATA:
2513 dest = XFS_DFORK_APTR(dip);
2514 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2515 memcpy(dest, src, len);
2518 case XFS_ILOG_ABROOT:
2519 dest = XFS_DFORK_APTR(dip);
2520 xfs_bmbt_to_bmdr((xfs_bmbt_block_t *)src, len,
2521 (xfs_bmdr_block_t*)dest,
2522 XFS_DFORK_ASIZE(dip, mp));
2526 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2529 return XFS_ERROR(EIO);
2534 if (ITEM_TYPE(item) == XFS_LI_INODE) {
2535 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2536 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2537 XFS_BUF_SET_FSPRIVATE(bp, mp);
2538 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2539 xfs_bdwrite(mp, bp);
2542 error = xfs_bwrite(mp, bp);
2549 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2550 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2554 xlog_recover_do_quotaoff_trans(
2556 xlog_recover_item_t *item,
2559 xfs_qoff_logformat_t *qoff_f;
2561 if (pass == XLOG_RECOVER_PASS2) {
2565 qoff_f = (xfs_qoff_logformat_t *)item->ri_buf[0].i_addr;
2569 * The logitem format's flag tells us if this was user quotaoff,
2570 * group quotaoff or both.
2572 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2573 log->l_quotaoffs_flag |= XFS_DQ_USER;
2574 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2575 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2581 * Recover a dquot record
2584 xlog_recover_do_dquot_trans(
2586 xlog_recover_item_t *item,
2591 struct xfs_disk_dquot *ddq, *recddq;
2593 xfs_dq_logformat_t *dq_f;
2596 if (pass == XLOG_RECOVER_PASS1) {
2602 * Filesystems are required to send in quota flags at mount time.
2604 if (mp->m_qflags == 0)
2607 recddq = (xfs_disk_dquot_t *)item->ri_buf[1].i_addr;
2610 * This type of quotas was turned off, so ignore this record.
2612 type = INT_GET(recddq->d_flags, ARCH_CONVERT) &
2613 (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
2615 if (log->l_quotaoffs_flag & type)
2619 * At this point we know that quota was _not_ turned off.
2620 * Since the mount flags are not indicating to us otherwise, this
2621 * must mean that quota is on, and the dquot needs to be replayed.
2622 * Remember that we may not have fully recovered the superblock yet,
2623 * so we can't do the usual trick of looking at the SB quota bits.
2625 * The other possibility, of course, is that the quota subsystem was
2626 * removed since the last mount - ENOSYS.
2628 dq_f = (xfs_dq_logformat_t *)item->ri_buf[0].i_addr;
2630 if ((error = xfs_qm_dqcheck(recddq,
2632 0, XFS_QMOPT_DOWARN,
2633 "xlog_recover_do_dquot_trans (log copy)"))) {
2634 return XFS_ERROR(EIO);
2636 ASSERT(dq_f->qlf_len == 1);
2638 error = xfs_read_buf(mp, mp->m_ddev_targp,
2640 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2643 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
2644 bp, dq_f->qlf_blkno);
2648 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2651 * At least the magic num portion should be on disk because this
2652 * was among a chunk of dquots created earlier, and we did some
2653 * minimal initialization then.
2655 if (xfs_qm_dqcheck(ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2656 "xlog_recover_do_dquot_trans")) {
2658 return XFS_ERROR(EIO);
2661 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2663 ASSERT(dq_f->qlf_size == 2);
2664 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2665 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2666 XFS_BUF_SET_FSPRIVATE(bp, mp);
2667 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2668 xfs_bdwrite(mp, bp);
2674 * This routine is called to create an in-core extent free intent
2675 * item from the efi format structure which was logged on disk.
2676 * It allocates an in-core efi, copies the extents from the format
2677 * structure into it, and adds the efi to the AIL with the given
2681 xlog_recover_do_efi_trans(
2683 xlog_recover_item_t *item,
2688 xfs_efi_log_item_t *efip;
2689 xfs_efi_log_format_t *efi_formatp;
2692 if (pass == XLOG_RECOVER_PASS1) {
2696 efi_formatp = (xfs_efi_log_format_t *)item->ri_buf[0].i_addr;
2697 ASSERT(item->ri_buf[0].i_len ==
2698 (sizeof(xfs_efi_log_format_t) +
2699 ((efi_formatp->efi_nextents - 1) * sizeof(xfs_extent_t))));
2702 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
2703 memcpy((char *)&(efip->efi_format), (char *)efi_formatp,
2704 sizeof(xfs_efi_log_format_t) +
2705 ((efi_formatp->efi_nextents - 1) * sizeof(xfs_extent_t)));
2706 efip->efi_next_extent = efi_formatp->efi_nextents;
2707 efip->efi_flags |= XFS_EFI_COMMITTED;
2711 * xfs_trans_update_ail() drops the AIL lock.
2713 xfs_trans_update_ail(mp, (xfs_log_item_t *)efip, lsn, s);
2718 * This routine is called when an efd format structure is found in
2719 * a committed transaction in the log. It's purpose is to cancel
2720 * the corresponding efi if it was still in the log. To do this
2721 * it searches the AIL for the efi with an id equal to that in the
2722 * efd format structure. If we find it, we remove the efi from the
2726 xlog_recover_do_efd_trans(
2728 xlog_recover_item_t *item,
2732 xfs_efd_log_format_t *efd_formatp;
2733 xfs_efi_log_item_t *efip = NULL;
2734 xfs_log_item_t *lip;
2739 if (pass == XLOG_RECOVER_PASS1) {
2743 efd_formatp = (xfs_efd_log_format_t *)item->ri_buf[0].i_addr;
2744 ASSERT(item->ri_buf[0].i_len ==
2745 (sizeof(xfs_efd_log_format_t) +
2746 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_t))));
2747 efi_id = efd_formatp->efd_efi_id;
2750 * Search for the efi with the id in the efd format structure
2755 lip = xfs_trans_first_ail(mp, &gen);
2756 while (lip != NULL) {
2757 if (lip->li_type == XFS_LI_EFI) {
2758 efip = (xfs_efi_log_item_t *)lip;
2759 if (efip->efi_format.efi_id == efi_id) {
2761 * xfs_trans_delete_ail() drops the
2764 xfs_trans_delete_ail(mp, lip, s);
2768 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
2772 * If we found it, then free it up. If it wasn't there, it
2773 * must have been overwritten in the log. Oh well.
2776 xfs_efi_item_free(efip);
2783 * Perform the transaction
2785 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2786 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2789 xlog_recover_do_trans(
2791 xlog_recover_t *trans,
2795 xlog_recover_item_t *item, *first_item;
2797 if ((error = xlog_recover_reorder_trans(log, trans)))
2799 first_item = item = trans->r_itemq;
2802 * we don't need to worry about the block number being
2803 * truncated in > 1 TB buffers because in user-land,
2804 * we're now n32 or 64-bit so xfs_daddr_t is 64-bits so
2805 * the blkno's will get through the user-mode buffer
2806 * cache properly. The only bad case is o32 kernels
2807 * where xfs_daddr_t is 32-bits but mount will warn us
2808 * off a > 1 TB filesystem before we get here.
2810 if ((ITEM_TYPE(item) == XFS_LI_BUF) ||
2811 (ITEM_TYPE(item) == XFS_LI_6_1_BUF) ||
2812 (ITEM_TYPE(item) == XFS_LI_5_3_BUF)) {
2813 if ((error = xlog_recover_do_buffer_trans(log, item,
2816 } else if ((ITEM_TYPE(item) == XFS_LI_INODE) ||
2817 (ITEM_TYPE(item) == XFS_LI_6_1_INODE) ||
2818 (ITEM_TYPE(item) == XFS_LI_5_3_INODE)) {
2819 if ((error = xlog_recover_do_inode_trans(log, item,
2822 } else if (ITEM_TYPE(item) == XFS_LI_EFI) {
2823 xlog_recover_do_efi_trans(log, item, trans->r_lsn,
2825 } else if (ITEM_TYPE(item) == XFS_LI_EFD) {
2826 xlog_recover_do_efd_trans(log, item, pass);
2827 } else if (ITEM_TYPE(item) == XFS_LI_DQUOT) {
2828 if ((error = xlog_recover_do_dquot_trans(log, item,
2831 } else if ((ITEM_TYPE(item) == XFS_LI_QUOTAOFF)) {
2832 if ((error = xlog_recover_do_quotaoff_trans(log, item,
2836 xlog_warn("XFS: xlog_recover_do_trans");
2838 error = XFS_ERROR(EIO);
2841 item = item->ri_next;
2842 } while (first_item != item);
2848 * Free up any resources allocated by the transaction
2850 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2853 xlog_recover_free_trans(
2854 xlog_recover_t *trans)
2856 xlog_recover_item_t *first_item, *item, *free_item;
2859 item = first_item = trans->r_itemq;
2862 item = item->ri_next;
2863 /* Free the regions in the item. */
2864 for (i = 0; i < free_item->ri_cnt; i++) {
2865 kmem_free(free_item->ri_buf[i].i_addr,
2866 free_item->ri_buf[i].i_len);
2868 /* Free the item itself */
2869 kmem_free(free_item->ri_buf,
2870 (free_item->ri_total * sizeof(xfs_log_iovec_t)));
2871 kmem_free(free_item, sizeof(xlog_recover_item_t));
2872 } while (first_item != item);
2873 /* Free the transaction recover structure */
2874 kmem_free(trans, sizeof(xlog_recover_t));
2878 xlog_recover_commit_trans(
2881 xlog_recover_t *trans,
2886 if ((error = xlog_recover_unlink_tid(q, trans)))
2888 if ((error = xlog_recover_do_trans(log, trans, pass)))
2890 xlog_recover_free_trans(trans); /* no error */
2895 xlog_recover_unmount_trans(
2896 xlog_recover_t *trans)
2898 /* Do nothing now */
2899 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2904 * There are two valid states of the r_state field. 0 indicates that the
2905 * transaction structure is in a normal state. We have either seen the
2906 * start of the transaction or the last operation we added was not a partial
2907 * operation. If the last operation we added to the transaction was a
2908 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2910 * NOTE: skip LRs with 0 data length.
2913 xlog_recover_process_data(
2915 xlog_recover_t *rhash[],
2916 xlog_rec_header_t *rhead,
2922 xlog_op_header_t *ohead;
2923 xlog_recover_t *trans;
2929 lp = dp + INT_GET(rhead->h_len, ARCH_CONVERT);
2930 num_logops = INT_GET(rhead->h_num_logops, ARCH_CONVERT);
2932 /* check the log format matches our own - else we can't recover */
2933 if (xlog_header_check_recover(log->l_mp, rhead))
2934 return (XFS_ERROR(EIO));
2936 while ((dp < lp) && num_logops) {
2937 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2938 ohead = (xlog_op_header_t *)dp;
2939 dp += sizeof(xlog_op_header_t);
2940 if (ohead->oh_clientid != XFS_TRANSACTION &&
2941 ohead->oh_clientid != XFS_LOG) {
2943 "XFS: xlog_recover_process_data: bad clientid");
2945 return (XFS_ERROR(EIO));
2947 tid = INT_GET(ohead->oh_tid, ARCH_CONVERT);
2948 hash = XLOG_RHASH(tid);
2949 trans = xlog_recover_find_tid(rhash[hash], tid);
2950 if (trans == NULL) { /* not found; add new tid */
2951 if (ohead->oh_flags & XLOG_START_TRANS)
2952 xlog_recover_new_tid(&rhash[hash], tid,
2953 INT_GET(rhead->h_lsn, ARCH_CONVERT));
2955 ASSERT(dp+INT_GET(ohead->oh_len, ARCH_CONVERT) <= lp);
2956 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2957 if (flags & XLOG_WAS_CONT_TRANS)
2958 flags &= ~XLOG_CONTINUE_TRANS;
2960 case XLOG_COMMIT_TRANS:
2961 error = xlog_recover_commit_trans(log,
2962 &rhash[hash], trans, pass);
2964 case XLOG_UNMOUNT_TRANS:
2965 error = xlog_recover_unmount_trans(trans);
2967 case XLOG_WAS_CONT_TRANS:
2968 error = xlog_recover_add_to_cont_trans(trans,
2969 dp, INT_GET(ohead->oh_len,
2972 case XLOG_START_TRANS:
2974 "XFS: xlog_recover_process_data: bad transaction");
2976 error = XFS_ERROR(EIO);
2979 case XLOG_CONTINUE_TRANS:
2980 error = xlog_recover_add_to_trans(trans,
2981 dp, INT_GET(ohead->oh_len,
2986 "XFS: xlog_recover_process_data: bad flag");
2988 error = XFS_ERROR(EIO);
2994 dp += INT_GET(ohead->oh_len, ARCH_CONVERT);
3001 * Process an extent free intent item that was recovered from
3002 * the log. We need to free the extents that it describes.
3005 xlog_recover_process_efi(
3007 xfs_efi_log_item_t *efip)
3009 xfs_efd_log_item_t *efdp;
3013 xfs_fsblock_t startblock_fsb;
3015 ASSERT(!(efip->efi_flags & XFS_EFI_RECOVERED));
3018 * First check the validity of the extents described by the
3019 * EFI. If any are bad, then assume that all are bad and
3020 * just toss the EFI.
3022 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3023 extp = &(efip->efi_format.efi_extents[i]);
3024 startblock_fsb = XFS_BB_TO_FSB(mp,
3025 XFS_FSB_TO_DADDR(mp, extp->ext_start));
3026 if ((startblock_fsb == 0) ||
3027 (extp->ext_len == 0) ||
3028 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3029 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3031 * This will pull the EFI from the AIL and
3032 * free the memory associated with it.
3034 xfs_efi_release(efip, efip->efi_format.efi_nextents);
3039 tp = xfs_trans_alloc(mp, 0);
3040 xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
3041 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3043 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3044 extp = &(efip->efi_format.efi_extents[i]);
3045 xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3046 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3050 efip->efi_flags |= XFS_EFI_RECOVERED;
3051 xfs_trans_commit(tp, 0, NULL);
3055 * Verify that once we've encountered something other than an EFI
3056 * in the AIL that there are no more EFIs in the AIL.
3060 xlog_recover_check_ail(
3062 xfs_log_item_t *lip,
3068 ASSERT(lip->li_type != XFS_LI_EFI);
3069 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
3071 * The check will be bogus if we restart from the
3072 * beginning of the AIL, so ASSERT that we don't.
3073 * We never should since we're holding the AIL lock
3076 ASSERT(gen == orig_gen);
3077 } while (lip != NULL);
3082 * When this is called, all of the EFIs which did not have
3083 * corresponding EFDs should be in the AIL. What we do now
3084 * is free the extents associated with each one.
3086 * Since we process the EFIs in normal transactions, they
3087 * will be removed at some point after the commit. This prevents
3088 * us from just walking down the list processing each one.
3089 * We'll use a flag in the EFI to skip those that we've already
3090 * processed and use the AIL iteration mechanism's generation
3091 * count to try to speed this up at least a bit.
3093 * When we start, we know that the EFIs are the only things in
3094 * the AIL. As we process them, however, other items are added
3095 * to the AIL. Since everything added to the AIL must come after
3096 * everything already in the AIL, we stop processing as soon as
3097 * we see something other than an EFI in the AIL.
3100 xlog_recover_process_efis(
3103 xfs_log_item_t *lip;
3104 xfs_efi_log_item_t *efip;
3112 lip = xfs_trans_first_ail(mp, &gen);
3113 while (lip != NULL) {
3115 * We're done when we see something other than an EFI.
3117 if (lip->li_type != XFS_LI_EFI) {
3118 xlog_recover_check_ail(mp, lip, gen);
3123 * Skip EFIs that we've already processed.
3125 efip = (xfs_efi_log_item_t *)lip;
3126 if (efip->efi_flags & XFS_EFI_RECOVERED) {
3127 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
3132 xlog_recover_process_efi(mp, efip);
3134 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
3140 * This routine performs a transaction to null out a bad inode pointer
3141 * in an agi unlinked inode hash bucket.
3144 xlog_recover_clear_agi_bucket(
3146 xfs_agnumber_t agno,
3155 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3156 xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp), 0, 0, 0);
3158 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
3159 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
3160 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
3162 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3166 agi = XFS_BUF_TO_AGI(agibp);
3167 if (INT_GET(agi->agi_magicnum, ARCH_CONVERT) != XFS_AGI_MAGIC) {
3168 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3171 ASSERT(INT_GET(agi->agi_magicnum, ARCH_CONVERT) == XFS_AGI_MAGIC);
3173 INT_SET(agi->agi_unlinked[bucket], ARCH_CONVERT, NULLAGINO);
3174 offset = offsetof(xfs_agi_t, agi_unlinked) +
3175 (sizeof(xfs_agino_t) * bucket);
3176 xfs_trans_log_buf(tp, agibp, offset,
3177 (offset + sizeof(xfs_agino_t) - 1));
3179 (void) xfs_trans_commit(tp, 0, NULL);
3183 * xlog_iunlink_recover
3185 * This is called during recovery to process any inodes which
3186 * we unlinked but not freed when the system crashed. These
3187 * inodes will be on the lists in the AGI blocks. What we do
3188 * here is scan all the AGIs and fully truncate and free any
3189 * inodes found on the lists. Each inode is removed from the
3190 * lists when it has been fully truncated and is freed. The
3191 * freeing of the inode and its removal from the list must be
3195 xlog_recover_process_iunlinks(
3199 xfs_agnumber_t agno;
3214 * Prevent any DMAPI event from being sent while in this function.
3216 mp_dmevmask = mp->m_dmevmask;
3219 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3221 * Find the agi for this ag.
3223 agibp = xfs_buf_read(mp->m_ddev_targp,
3224 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
3225 XFS_FSS_TO_BB(mp, 1), 0);
3226 if (XFS_BUF_ISERROR(agibp)) {
3227 xfs_ioerror_alert("xlog_recover_process_iunlinks(#1)",
3229 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)));
3231 agi = XFS_BUF_TO_AGI(agibp);
3232 ASSERT(XFS_AGI_MAGIC ==
3233 INT_GET(agi->agi_magicnum, ARCH_CONVERT));
3235 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3237 agino = INT_GET(agi->agi_unlinked[bucket], ARCH_CONVERT);
3238 while (agino != NULLAGINO) {
3241 * Release the agi buffer so that it can
3242 * be acquired in the normal course of the
3243 * transaction to truncate and free the inode.
3245 xfs_buf_relse(agibp);
3247 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3248 error = xfs_iget(mp, NULL, ino, 0, 0, &ip, 0);
3249 ASSERT(error || (ip != NULL));
3253 * Get the on disk inode to find the
3254 * next inode in the bucket.
3256 error = xfs_itobp(mp, NULL, ip, &dip,
3258 ASSERT(error || (dip != NULL));
3262 ASSERT(ip->i_d.di_nlink == 0);
3264 /* setup for the next pass */
3265 agino = INT_GET(dip->di_next_unlinked,
3269 * Prevent any DMAPI event from
3270 * being sent when the
3271 * reference on the inode is
3274 ip->i_d.di_dmevmask = 0;
3277 * If this is a new inode, handle
3278 * it specially. Otherwise,
3279 * just drop our reference to the
3280 * inode. If there are no
3281 * other references, this will
3283 * xfs_inactive() which will
3284 * truncate the file and free
3287 if (ip->i_d.di_mode == 0)
3288 xfs_iput_new(ip, 0);
3290 VN_RELE(XFS_ITOV(ip));
3293 * We can't read in the inode
3294 * this bucket points to, or
3295 * this inode is messed up. Just
3296 * ditch this bucket of inodes. We
3297 * will lose some inodes and space,
3298 * but at least we won't hang. Call
3299 * xlog_recover_clear_agi_bucket()
3300 * to perform a transaction to clear
3301 * the inode pointer in the bucket.
3303 xlog_recover_clear_agi_bucket(mp, agno,
3310 * Reacquire the agibuffer and continue around
3313 agibp = xfs_buf_read(mp->m_ddev_targp,
3314 XFS_AG_DADDR(mp, agno,
3316 XFS_FSS_TO_BB(mp, 1), 0);
3317 if (XFS_BUF_ISERROR(agibp)) {
3319 "xlog_recover_process_iunlinks(#2)",
3321 XFS_AG_DADDR(mp, agno,
3322 XFS_AGI_DADDR(mp)));
3324 agi = XFS_BUF_TO_AGI(agibp);
3325 ASSERT(XFS_AGI_MAGIC == INT_GET(
3326 agi->agi_magicnum, ARCH_CONVERT));
3331 * Release the buffer for the current agi so we can
3332 * go on to the next one.
3334 xfs_buf_relse(agibp);
3337 mp->m_dmevmask = mp_dmevmask;
3343 xlog_pack_data_checksum(
3345 xlog_in_core_t *iclog,
3352 up = (uint *)iclog->ic_datap;
3353 /* divide length by 4 to get # words */
3354 for (i = 0; i < (size >> 2); i++) {
3355 chksum ^= INT_GET(*up, ARCH_CONVERT);
3358 INT_SET(iclog->ic_header.h_chksum, ARCH_CONVERT, chksum);
3361 #define xlog_pack_data_checksum(log, iclog, size)
3365 * Stamp cycle number in every block
3370 xlog_in_core_t *iclog,
3374 int size = iclog->ic_offset + roundoff;
3377 xlog_in_core_2_t *xhdr;
3379 xlog_pack_data_checksum(log, iclog, size);
3381 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3383 dp = iclog->ic_datap;
3384 for (i = 0; i < BTOBB(size) &&
3385 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3386 iclog->ic_header.h_cycle_data[i] = *(uint *)dp;
3387 *(uint *)dp = cycle_lsn;
3391 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3392 xhdr = (xlog_in_core_2_t *)&iclog->ic_header;
3393 for ( ; i < BTOBB(size); i++) {
3394 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3395 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3396 xhdr[j].hic_xheader.xh_cycle_data[k] = *(uint *)dp;
3397 *(uint *)dp = cycle_lsn;
3401 for (i = 1; i < log->l_iclog_heads; i++) {
3402 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3407 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
3409 xlog_unpack_data_checksum(
3410 xlog_rec_header_t *rhead,
3414 uint *up = (uint *)dp;
3418 /* divide length by 4 to get # words */
3419 for (i=0; i < INT_GET(rhead->h_len, ARCH_CONVERT) >> 2; i++) {
3420 chksum ^= INT_GET(*up, ARCH_CONVERT);
3423 if (chksum != INT_GET(rhead->h_chksum, ARCH_CONVERT)) {
3424 if (rhead->h_chksum ||
3425 ((log->l_flags & XLOG_CHKSUM_MISMATCH) == 0)) {
3427 "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)",
3428 INT_GET(rhead->h_chksum, ARCH_CONVERT), chksum);
3430 "XFS: Disregard message if filesystem was created with non-DEBUG kernel");
3431 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3433 "XFS: LogR this is a LogV2 filesystem");
3435 log->l_flags |= XLOG_CHKSUM_MISMATCH;
3440 #define xlog_unpack_data_checksum(rhead, dp, log)
3445 xlog_rec_header_t *rhead,
3450 xlog_in_core_2_t *xhdr;
3452 for (i = 0; i < BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT)) &&
3453 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3454 *(uint *)dp = *(uint *)&rhead->h_cycle_data[i];
3458 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3459 xhdr = (xlog_in_core_2_t *)rhead;
3460 for ( ; i < BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT)); i++) {
3461 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3462 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3463 *(uint *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
3468 xlog_unpack_data_checksum(rhead, dp, log);
3472 xlog_valid_rec_header(
3474 xlog_rec_header_t *rhead,
3480 (INT_GET(rhead->h_magicno, ARCH_CONVERT) !=
3481 XLOG_HEADER_MAGIC_NUM))) {
3482 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3483 XFS_ERRLEVEL_LOW, log->l_mp);
3484 return XFS_ERROR(EFSCORRUPTED);
3487 (!rhead->h_version ||
3488 (INT_GET(rhead->h_version, ARCH_CONVERT) &
3489 (~XLOG_VERSION_OKBITS)) != 0))) {
3490 xlog_warn("XFS: %s: unrecognised log version (%d).",
3491 __FUNCTION__, INT_GET(rhead->h_version, ARCH_CONVERT));
3492 return XFS_ERROR(EIO);
3495 /* LR body must have data or it wouldn't have been written */
3496 hlen = INT_GET(rhead->h_len, ARCH_CONVERT);
3497 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3498 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3499 XFS_ERRLEVEL_LOW, log->l_mp);
3500 return XFS_ERROR(EFSCORRUPTED);
3502 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3503 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3504 XFS_ERRLEVEL_LOW, log->l_mp);
3505 return XFS_ERROR(EFSCORRUPTED);
3511 * Read the log from tail to head and process the log records found.
3512 * Handle the two cases where the tail and head are in the same cycle
3513 * and where the active portion of the log wraps around the end of
3514 * the physical log separately. The pass parameter is passed through
3515 * to the routines called to process the data and is not looked at
3519 xlog_do_recovery_pass(
3521 xfs_daddr_t head_blk,
3522 xfs_daddr_t tail_blk,
3525 xlog_rec_header_t *rhead;
3527 xfs_caddr_t bufaddr, offset;
3528 xfs_buf_t *hbp, *dbp;
3529 int error = 0, h_size;
3530 int bblks, split_bblks;
3531 int hblks, split_hblks, wrapped_hblks;
3532 xlog_recover_t *rhash[XLOG_RHASH_SIZE];
3534 ASSERT(head_blk != tail_blk);
3537 * Read the header of the tail block and get the iclog buffer size from
3538 * h_size. Use this to tell how many sectors make up the log header.
3540 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3542 * When using variable length iclogs, read first sector of
3543 * iclog header and extract the header size from it. Get a
3544 * new hbp that is the correct size.
3546 hbp = xlog_get_bp(log, 1);
3549 if ((error = xlog_bread(log, tail_blk, 1, hbp)))
3551 offset = xlog_align(log, tail_blk, 1, hbp);
3552 rhead = (xlog_rec_header_t *)offset;
3553 error = xlog_valid_rec_header(log, rhead, tail_blk);
3556 h_size = INT_GET(rhead->h_size, ARCH_CONVERT);
3557 if ((INT_GET(rhead->h_version, ARCH_CONVERT)
3558 & XLOG_VERSION_2) &&
3559 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3560 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3561 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3564 hbp = xlog_get_bp(log, hblks);
3569 ASSERT(log->l_sectbb_log == 0);
3571 hbp = xlog_get_bp(log, 1);
3572 h_size = XLOG_BIG_RECORD_BSIZE;
3577 dbp = xlog_get_bp(log, BTOBB(h_size));
3583 memset(rhash, 0, sizeof(rhash));
3584 if (tail_blk <= head_blk) {
3585 for (blk_no = tail_blk; blk_no < head_blk; ) {
3586 if ((error = xlog_bread(log, blk_no, hblks, hbp)))
3588 offset = xlog_align(log, blk_no, hblks, hbp);
3589 rhead = (xlog_rec_header_t *)offset;
3590 error = xlog_valid_rec_header(log, rhead, blk_no);
3594 /* blocks in data section */
3595 bblks = (int)BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT));
3596 error = xlog_bread(log, blk_no + hblks, bblks, dbp);
3599 offset = xlog_align(log, blk_no + hblks, bblks, dbp);
3600 xlog_unpack_data(rhead, offset, log);
3601 if ((error = xlog_recover_process_data(log,
3602 rhash, rhead, offset, pass)))
3604 blk_no += bblks + hblks;
3608 * Perform recovery around the end of the physical log.
3609 * When the head is not on the same cycle number as the tail,
3610 * we can't do a sequential recovery as above.
3613 while (blk_no < log->l_logBBsize) {
3615 * Check for header wrapping around physical end-of-log
3620 if (blk_no + hblks <= log->l_logBBsize) {
3621 /* Read header in one read */
3622 error = xlog_bread(log, blk_no, hblks, hbp);
3625 offset = xlog_align(log, blk_no, hblks, hbp);
3627 /* This LR is split across physical log end */
3628 if (blk_no != log->l_logBBsize) {
3629 /* some data before physical log end */
3630 ASSERT(blk_no <= INT_MAX);
3631 split_hblks = log->l_logBBsize - (int)blk_no;
3632 ASSERT(split_hblks > 0);
3633 if ((error = xlog_bread(log, blk_no,
3636 offset = xlog_align(log, blk_no,
3640 * Note: this black magic still works with
3641 * large sector sizes (non-512) only because:
3642 * - we increased the buffer size originally
3643 * by 1 sector giving us enough extra space
3644 * for the second read;
3645 * - the log start is guaranteed to be sector
3647 * - we read the log end (LR header start)
3648 * _first_, then the log start (LR header end)
3649 * - order is important.
3651 bufaddr = XFS_BUF_PTR(hbp);
3652 XFS_BUF_SET_PTR(hbp,
3653 bufaddr + BBTOB(split_hblks),
3654 BBTOB(hblks - split_hblks));
3655 wrapped_hblks = hblks - split_hblks;
3656 error = xlog_bread(log, 0, wrapped_hblks, hbp);
3659 XFS_BUF_SET_PTR(hbp, bufaddr, BBTOB(hblks));
3661 offset = xlog_align(log, 0,
3662 wrapped_hblks, hbp);
3664 rhead = (xlog_rec_header_t *)offset;
3665 error = xlog_valid_rec_header(log, rhead,
3666 split_hblks ? blk_no : 0);
3670 bblks = (int)BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT));
3673 /* Read in data for log record */
3674 if (blk_no + bblks <= log->l_logBBsize) {
3675 error = xlog_bread(log, blk_no, bblks, dbp);
3678 offset = xlog_align(log, blk_no, bblks, dbp);
3680 /* This log record is split across the
3681 * physical end of log */
3684 if (blk_no != log->l_logBBsize) {
3685 /* some data is before the physical
3687 ASSERT(!wrapped_hblks);
3688 ASSERT(blk_no <= INT_MAX);
3690 log->l_logBBsize - (int)blk_no;
3691 ASSERT(split_bblks > 0);
3692 if ((error = xlog_bread(log, blk_no,
3695 offset = xlog_align(log, blk_no,
3699 * Note: this black magic still works with
3700 * large sector sizes (non-512) only because:
3701 * - we increased the buffer size originally
3702 * by 1 sector giving us enough extra space
3703 * for the second read;
3704 * - the log start is guaranteed to be sector
3706 * - we read the log end (LR header start)
3707 * _first_, then the log start (LR header end)
3708 * - order is important.
3710 bufaddr = XFS_BUF_PTR(dbp);
3711 XFS_BUF_SET_PTR(dbp,
3712 bufaddr + BBTOB(split_bblks),
3713 BBTOB(bblks - split_bblks));
3714 if ((error = xlog_bread(log, wrapped_hblks,
3715 bblks - split_bblks, dbp)))
3717 XFS_BUF_SET_PTR(dbp, bufaddr, h_size);
3719 offset = xlog_align(log, wrapped_hblks,
3720 bblks - split_bblks, dbp);
3722 xlog_unpack_data(rhead, offset, log);
3723 if ((error = xlog_recover_process_data(log, rhash,
3724 rhead, offset, pass)))
3729 ASSERT(blk_no >= log->l_logBBsize);
3730 blk_no -= log->l_logBBsize;
3732 /* read first part of physical log */
3733 while (blk_no < head_blk) {
3734 if ((error = xlog_bread(log, blk_no, hblks, hbp)))
3736 offset = xlog_align(log, blk_no, hblks, hbp);
3737 rhead = (xlog_rec_header_t *)offset;
3738 error = xlog_valid_rec_header(log, rhead, blk_no);
3741 bblks = (int)BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT));
3742 if ((error = xlog_bread(log, blk_no+hblks, bblks, dbp)))
3744 offset = xlog_align(log, blk_no+hblks, bblks, dbp);
3745 xlog_unpack_data(rhead, offset, log);
3746 if ((error = xlog_recover_process_data(log, rhash,
3747 rhead, offset, pass)))
3749 blk_no += bblks + hblks;
3761 * Do the recovery of the log. We actually do this in two phases.
3762 * The two passes are necessary in order to implement the function
3763 * of cancelling a record written into the log. The first pass
3764 * determines those things which have been cancelled, and the
3765 * second pass replays log items normally except for those which
3766 * have been cancelled. The handling of the replay and cancellations
3767 * takes place in the log item type specific routines.
3769 * The table of items which have cancel records in the log is allocated
3770 * and freed at this level, since only here do we know when all of
3771 * the log recovery has been completed.
3774 xlog_do_log_recovery(
3776 xfs_daddr_t head_blk,
3777 xfs_daddr_t tail_blk)
3781 ASSERT(head_blk != tail_blk);
3784 * First do a pass to find all of the cancelled buf log items.
3785 * Store them in the buf_cancel_table for use in the second pass.
3787 log->l_buf_cancel_table =
3788 (xfs_buf_cancel_t **)kmem_zalloc(XLOG_BC_TABLE_SIZE *
3789 sizeof(xfs_buf_cancel_t*),
3791 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3792 XLOG_RECOVER_PASS1);
3794 kmem_free(log->l_buf_cancel_table,
3795 XLOG_BC_TABLE_SIZE * sizeof(xfs_buf_cancel_t*));
3796 log->l_buf_cancel_table = NULL;
3800 * Then do a second pass to actually recover the items in the log.
3801 * When it is complete free the table of buf cancel items.
3803 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3804 XLOG_RECOVER_PASS2);
3809 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3810 ASSERT(log->l_buf_cancel_table[i] == NULL);
3814 kmem_free(log->l_buf_cancel_table,
3815 XLOG_BC_TABLE_SIZE * sizeof(xfs_buf_cancel_t*));
3816 log->l_buf_cancel_table = NULL;
3822 * Do the actual recovery
3827 xfs_daddr_t head_blk,
3828 xfs_daddr_t tail_blk)
3835 * First replay the images in the log.
3837 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3842 XFS_bflush(log->l_mp->m_ddev_targp);
3845 * If IO errors happened during recovery, bail out.
3847 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3852 * We now update the tail_lsn since much of the recovery has completed
3853 * and there may be space available to use. If there were no extent
3854 * or iunlinks, we can free up the entire log and set the tail_lsn to
3855 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3856 * lsn of the last known good LR on disk. If there are extent frees
3857 * or iunlinks they will have some entries in the AIL; so we look at
3858 * the AIL to determine how to set the tail_lsn.
3860 xlog_assign_tail_lsn(log->l_mp);
3863 * Now that we've finished replaying all buffer and inode
3864 * updates, re-read in the superblock.
3866 bp = xfs_getsb(log->l_mp, 0);
3869 xfsbdstrat(log->l_mp, bp);
3870 if ((error = xfs_iowait(bp))) {
3871 xfs_ioerror_alert("xlog_do_recover",
3872 log->l_mp, bp, XFS_BUF_ADDR(bp));
3878 /* Convert superblock from on-disk format */
3879 sbp = &log->l_mp->m_sb;
3880 xfs_xlatesb(XFS_BUF_TO_SBP(bp), sbp, 1, XFS_SB_ALL_BITS);
3881 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
3882 ASSERT(XFS_SB_GOOD_VERSION(sbp));
3885 xlog_recover_check_summary(log);
3887 /* Normal transactions can now occur */
3888 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3893 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3895 * Return error or zero.
3902 xfs_daddr_t head_blk, tail_blk;
3905 /* find the tail of the log */
3906 if ((error = xlog_find_tail(log, &head_blk, &tail_blk, readonly)))
3909 if (tail_blk != head_blk) {
3910 /* There used to be a comment here:
3912 * disallow recovery on read-only mounts. note -- mount
3913 * checks for ENOSPC and turns it into an intelligent
3915 * ...but this is no longer true. Now, unless you specify
3916 * NORECOVERY (in which case this function would never be
3917 * called), we just go ahead and recover. We do this all
3918 * under the vfs layer, so we can get away with it unless
3919 * the device itself is read-only, in which case we fail.
3921 if ((error = xfs_dev_is_read_only(log->l_mp,
3922 "recovery required"))) {
3927 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3928 log->l_mp->m_fsname, log->l_mp->m_logname ?
3929 log->l_mp->m_logname : "internal");
3931 error = xlog_do_recover(log, head_blk, tail_blk);
3932 log->l_flags |= XLOG_RECOVERY_NEEDED;
3938 * In the first part of recovery we replay inodes and buffers and build
3939 * up the list of extent free items which need to be processed. Here
3940 * we process the extent free items and clean up the on disk unlinked
3941 * inode lists. This is separated from the first part of recovery so
3942 * that the root and real-time bitmap inodes can be read in from disk in
3943 * between the two stages. This is necessary so that we can free space
3944 * in the real-time portion of the file system.
3947 xlog_recover_finish(
3952 * Now we're ready to do the transactions needed for the
3953 * rest of recovery. Start with completing all the extent
3954 * free intent records and then process the unlinked inode
3955 * lists. At this point, we essentially run in normal mode
3956 * except that we're still performing recovery actions
3957 * rather than accepting new requests.
3959 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
3960 xlog_recover_process_efis(log);
3962 * Sync the log to get all the EFIs out of the AIL.
3963 * This isn't absolutely necessary, but it helps in
3964 * case the unlink transactions would have problems
3965 * pushing the EFIs out of the way.
3967 xfs_log_force(log->l_mp, (xfs_lsn_t)0,
3968 (XFS_LOG_FORCE | XFS_LOG_SYNC));
3970 if ( (mfsi_flags & XFS_MFSI_NOUNLINK) == 0 ) {
3971 xlog_recover_process_iunlinks(log);
3974 xlog_recover_check_summary(log);
3977 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3978 log->l_mp->m_fsname, log->l_mp->m_logname ?
3979 log->l_mp->m_logname : "internal");
3980 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
3983 "!Ending clean XFS mount for filesystem: %s",
3984 log->l_mp->m_fsname);
3992 * Read all of the agf and agi counters and check that they
3993 * are consistent with the superblock counters.
3996 xlog_recover_check_summary(
4004 xfs_daddr_t agfdaddr;
4005 xfs_daddr_t agidaddr;
4007 #ifdef XFS_LOUD_RECOVERY
4010 xfs_agnumber_t agno;
4011 __uint64_t freeblks;
4020 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
4021 agfdaddr = XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp));
4022 agfbp = xfs_buf_read(mp->m_ddev_targp, agfdaddr,
4023 XFS_FSS_TO_BB(mp, 1), 0);
4024 if (XFS_BUF_ISERROR(agfbp)) {
4025 xfs_ioerror_alert("xlog_recover_check_summary(agf)",
4026 mp, agfbp, agfdaddr);
4028 agfp = XFS_BUF_TO_AGF(agfbp);
4029 ASSERT(XFS_AGF_MAGIC ==
4030 INT_GET(agfp->agf_magicnum, ARCH_CONVERT));
4031 ASSERT(XFS_AGF_GOOD_VERSION(
4032 INT_GET(agfp->agf_versionnum, ARCH_CONVERT)));
4033 ASSERT(INT_GET(agfp->agf_seqno, ARCH_CONVERT) == agno);
4035 freeblks += INT_GET(agfp->agf_freeblks, ARCH_CONVERT) +
4036 INT_GET(agfp->agf_flcount, ARCH_CONVERT);
4037 xfs_buf_relse(agfbp);
4039 agidaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
4040 agibp = xfs_buf_read(mp->m_ddev_targp, agidaddr,
4041 XFS_FSS_TO_BB(mp, 1), 0);
4042 if (XFS_BUF_ISERROR(agibp)) {
4043 xfs_ioerror_alert("xlog_recover_check_summary(agi)",
4044 mp, agibp, agidaddr);
4046 agip = XFS_BUF_TO_AGI(agibp);
4047 ASSERT(XFS_AGI_MAGIC ==
4048 INT_GET(agip->agi_magicnum, ARCH_CONVERT));
4049 ASSERT(XFS_AGI_GOOD_VERSION(
4050 INT_GET(agip->agi_versionnum, ARCH_CONVERT)));
4051 ASSERT(INT_GET(agip->agi_seqno, ARCH_CONVERT) == agno);
4053 itotal += INT_GET(agip->agi_count, ARCH_CONVERT);
4054 ifree += INT_GET(agip->agi_freecount, ARCH_CONVERT);
4055 xfs_buf_relse(agibp);
4058 sbbp = xfs_getsb(mp, 0);
4059 #ifdef XFS_LOUD_RECOVERY
4061 xfs_xlatesb(XFS_BUF_TO_SBP(sbbp), sbp, 1, XFS_SB_ALL_BITS);
4063 "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
4064 sbp->sb_icount, itotal);
4066 "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
4067 sbp->sb_ifree, ifree);
4069 "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
4070 sbp->sb_fdblocks, freeblks);
4073 * This is turned off until I account for the allocation
4074 * btree blocks which live in free space.
4076 ASSERT(sbp->sb_icount == itotal);
4077 ASSERT(sbp->sb_ifree == ifree);
4078 ASSERT(sbp->sb_fdblocks == freeblks);
4081 xfs_buf_relse(sbbp);