Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/vapier...
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / xfs / xfs_log_recover.c
blob47da2fb45377309728acbeb3b4351908d61d6bde
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
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
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_error.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_alloc.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_log_priv.h"
42 #include "xfs_buf_item.h"
43 #include "xfs_log_recover.h"
44 #include "xfs_extfree_item.h"
45 #include "xfs_trans_priv.h"
46 #include "xfs_quota.h"
47 #include "xfs_rw.h"
48 #include "xfs_utils.h"
50 STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *);
51 STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t);
52 STATIC void xlog_recover_insert_item_backq(xlog_recover_item_t **q,
53 xlog_recover_item_t *item);
54 #if defined(DEBUG)
55 STATIC void xlog_recover_check_summary(xlog_t *);
56 #else
57 #define xlog_recover_check_summary(log)
58 #endif
62 * Sector aligned buffer routines for buffer create/read/write/access
65 #define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \
66 ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \
67 ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) )
68 #define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask)
70 xfs_buf_t *
71 xlog_get_bp(
72 xlog_t *log,
73 int nbblks)
75 if (nbblks <= 0 || nbblks > log->l_logBBsize) {
76 xlog_warn("XFS: Invalid block length (0x%x) given for buffer", nbblks);
77 XFS_ERROR_REPORT("xlog_get_bp(1)",
78 XFS_ERRLEVEL_HIGH, log->l_mp);
79 return NULL;
82 if (log->l_sectbb_log) {
83 if (nbblks > 1)
84 nbblks += XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
85 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
87 return xfs_buf_get_noaddr(BBTOB(nbblks), log->l_mp->m_logdev_targp);
90 void
91 xlog_put_bp(
92 xfs_buf_t *bp)
94 xfs_buf_free(bp);
97 STATIC xfs_caddr_t
98 xlog_align(
99 xlog_t *log,
100 xfs_daddr_t blk_no,
101 int nbblks,
102 xfs_buf_t *bp)
104 xfs_caddr_t ptr;
106 if (!log->l_sectbb_log)
107 return XFS_BUF_PTR(bp);
109 ptr = XFS_BUF_PTR(bp) + BBTOB((int)blk_no & log->l_sectbb_mask);
110 ASSERT(XFS_BUF_SIZE(bp) >=
111 BBTOB(nbblks + (blk_no & log->l_sectbb_mask)));
112 return ptr;
117 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
119 STATIC int
120 xlog_bread_noalign(
121 xlog_t *log,
122 xfs_daddr_t blk_no,
123 int nbblks,
124 xfs_buf_t *bp)
126 int error;
128 if (nbblks <= 0 || nbblks > log->l_logBBsize) {
129 xlog_warn("XFS: Invalid block length (0x%x) given for buffer", nbblks);
130 XFS_ERROR_REPORT("xlog_bread(1)",
131 XFS_ERRLEVEL_HIGH, log->l_mp);
132 return EFSCORRUPTED;
135 if (log->l_sectbb_log) {
136 blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
137 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
140 ASSERT(nbblks > 0);
141 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
142 ASSERT(bp);
144 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
145 XFS_BUF_READ(bp);
146 XFS_BUF_BUSY(bp);
147 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
148 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
150 xfsbdstrat(log->l_mp, bp);
151 error = xfs_iowait(bp);
152 if (error)
153 xfs_ioerror_alert("xlog_bread", log->l_mp,
154 bp, XFS_BUF_ADDR(bp));
155 return error;
158 STATIC int
159 xlog_bread(
160 xlog_t *log,
161 xfs_daddr_t blk_no,
162 int nbblks,
163 xfs_buf_t *bp,
164 xfs_caddr_t *offset)
166 int error;
168 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
169 if (error)
170 return error;
172 *offset = xlog_align(log, blk_no, nbblks, bp);
173 return 0;
177 * Write out the buffer at the given block for the given number of blocks.
178 * The buffer is kept locked across the write and is returned locked.
179 * This can only be used for synchronous log writes.
181 STATIC int
182 xlog_bwrite(
183 xlog_t *log,
184 xfs_daddr_t blk_no,
185 int nbblks,
186 xfs_buf_t *bp)
188 int error;
190 if (nbblks <= 0 || nbblks > log->l_logBBsize) {
191 xlog_warn("XFS: Invalid block length (0x%x) given for buffer", nbblks);
192 XFS_ERROR_REPORT("xlog_bwrite(1)",
193 XFS_ERRLEVEL_HIGH, log->l_mp);
194 return EFSCORRUPTED;
197 if (log->l_sectbb_log) {
198 blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
199 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
202 ASSERT(nbblks > 0);
203 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
205 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
206 XFS_BUF_ZEROFLAGS(bp);
207 XFS_BUF_BUSY(bp);
208 XFS_BUF_HOLD(bp);
209 XFS_BUF_PSEMA(bp, PRIBIO);
210 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
211 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
213 if ((error = xfs_bwrite(log->l_mp, bp)))
214 xfs_ioerror_alert("xlog_bwrite", log->l_mp,
215 bp, XFS_BUF_ADDR(bp));
216 return error;
219 #ifdef DEBUG
221 * dump debug superblock and log record information
223 STATIC void
224 xlog_header_check_dump(
225 xfs_mount_t *mp,
226 xlog_rec_header_t *head)
228 int b;
230 cmn_err(CE_DEBUG, "%s: SB : uuid = ", __func__);
231 for (b = 0; b < 16; b++)
232 cmn_err(CE_DEBUG, "%02x", ((__uint8_t *)&mp->m_sb.sb_uuid)[b]);
233 cmn_err(CE_DEBUG, ", fmt = %d\n", XLOG_FMT);
234 cmn_err(CE_DEBUG, " log : uuid = ");
235 for (b = 0; b < 16; b++)
236 cmn_err(CE_DEBUG, "%02x", ((__uint8_t *)&head->h_fs_uuid)[b]);
237 cmn_err(CE_DEBUG, ", fmt = %d\n", be32_to_cpu(head->h_fmt));
239 #else
240 #define xlog_header_check_dump(mp, head)
241 #endif
244 * check log record header for recovery
246 STATIC int
247 xlog_header_check_recover(
248 xfs_mount_t *mp,
249 xlog_rec_header_t *head)
251 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
254 * IRIX doesn't write the h_fmt field and leaves it zeroed
255 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
256 * a dirty log created in IRIX.
258 if (unlikely(be32_to_cpu(head->h_fmt) != XLOG_FMT)) {
259 xlog_warn(
260 "XFS: dirty log written in incompatible format - can't recover");
261 xlog_header_check_dump(mp, head);
262 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
263 XFS_ERRLEVEL_HIGH, mp);
264 return XFS_ERROR(EFSCORRUPTED);
265 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
266 xlog_warn(
267 "XFS: dirty log entry has mismatched uuid - can't recover");
268 xlog_header_check_dump(mp, head);
269 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
270 XFS_ERRLEVEL_HIGH, mp);
271 return XFS_ERROR(EFSCORRUPTED);
273 return 0;
277 * read the head block of the log and check the header
279 STATIC int
280 xlog_header_check_mount(
281 xfs_mount_t *mp,
282 xlog_rec_header_t *head)
284 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
286 if (uuid_is_nil(&head->h_fs_uuid)) {
288 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
289 * h_fs_uuid is nil, we assume this log was last mounted
290 * by IRIX and continue.
292 xlog_warn("XFS: nil uuid in log - IRIX style log");
293 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
294 xlog_warn("XFS: log has mismatched uuid - can't recover");
295 xlog_header_check_dump(mp, head);
296 XFS_ERROR_REPORT("xlog_header_check_mount",
297 XFS_ERRLEVEL_HIGH, mp);
298 return XFS_ERROR(EFSCORRUPTED);
300 return 0;
303 STATIC void
304 xlog_recover_iodone(
305 struct xfs_buf *bp)
307 if (XFS_BUF_GETERROR(bp)) {
309 * We're not going to bother about retrying
310 * this during recovery. One strike!
312 xfs_ioerror_alert("xlog_recover_iodone",
313 bp->b_mount, bp, XFS_BUF_ADDR(bp));
314 xfs_force_shutdown(bp->b_mount, SHUTDOWN_META_IO_ERROR);
316 bp->b_mount = NULL;
317 XFS_BUF_CLR_IODONE_FUNC(bp);
318 xfs_biodone(bp);
322 * This routine finds (to an approximation) the first block in the physical
323 * log which contains the given cycle. It uses a binary search algorithm.
324 * Note that the algorithm can not be perfect because the disk will not
325 * necessarily be perfect.
327 STATIC int
328 xlog_find_cycle_start(
329 xlog_t *log,
330 xfs_buf_t *bp,
331 xfs_daddr_t first_blk,
332 xfs_daddr_t *last_blk,
333 uint cycle)
335 xfs_caddr_t offset;
336 xfs_daddr_t mid_blk;
337 uint mid_cycle;
338 int error;
340 mid_blk = BLK_AVG(first_blk, *last_blk);
341 while (mid_blk != first_blk && mid_blk != *last_blk) {
342 error = xlog_bread(log, mid_blk, 1, bp, &offset);
343 if (error)
344 return error;
345 mid_cycle = xlog_get_cycle(offset);
346 if (mid_cycle == cycle) {
347 *last_blk = mid_blk;
348 /* last_half_cycle == mid_cycle */
349 } else {
350 first_blk = mid_blk;
351 /* first_half_cycle == mid_cycle */
353 mid_blk = BLK_AVG(first_blk, *last_blk);
355 ASSERT((mid_blk == first_blk && mid_blk+1 == *last_blk) ||
356 (mid_blk == *last_blk && mid_blk-1 == first_blk));
358 return 0;
362 * Check that the range of blocks does not contain the cycle number
363 * given. The scan needs to occur from front to back and the ptr into the
364 * region must be updated since a later routine will need to perform another
365 * test. If the region is completely good, we end up returning the same
366 * last block number.
368 * Set blkno to -1 if we encounter no errors. This is an invalid block number
369 * since we don't ever expect logs to get this large.
371 STATIC int
372 xlog_find_verify_cycle(
373 xlog_t *log,
374 xfs_daddr_t start_blk,
375 int nbblks,
376 uint stop_on_cycle_no,
377 xfs_daddr_t *new_blk)
379 xfs_daddr_t i, j;
380 uint cycle;
381 xfs_buf_t *bp;
382 xfs_daddr_t bufblks;
383 xfs_caddr_t buf = NULL;
384 int error = 0;
386 bufblks = 1 << ffs(nbblks);
388 while (!(bp = xlog_get_bp(log, bufblks))) {
389 /* can't get enough memory to do everything in one big buffer */
390 bufblks >>= 1;
391 if (bufblks <= log->l_sectbb_log)
392 return ENOMEM;
395 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
396 int bcount;
398 bcount = min(bufblks, (start_blk + nbblks - i));
400 error = xlog_bread(log, i, bcount, bp, &buf);
401 if (error)
402 goto out;
404 for (j = 0; j < bcount; j++) {
405 cycle = xlog_get_cycle(buf);
406 if (cycle == stop_on_cycle_no) {
407 *new_blk = i+j;
408 goto out;
411 buf += BBSIZE;
415 *new_blk = -1;
417 out:
418 xlog_put_bp(bp);
419 return error;
423 * Potentially backup over partial log record write.
425 * In the typical case, last_blk is the number of the block directly after
426 * a good log record. Therefore, we subtract one to get the block number
427 * of the last block in the given buffer. extra_bblks contains the number
428 * of blocks we would have read on a previous read. This happens when the
429 * last log record is split over the end of the physical log.
431 * extra_bblks is the number of blocks potentially verified on a previous
432 * call to this routine.
434 STATIC int
435 xlog_find_verify_log_record(
436 xlog_t *log,
437 xfs_daddr_t start_blk,
438 xfs_daddr_t *last_blk,
439 int extra_bblks)
441 xfs_daddr_t i;
442 xfs_buf_t *bp;
443 xfs_caddr_t offset = NULL;
444 xlog_rec_header_t *head = NULL;
445 int error = 0;
446 int smallmem = 0;
447 int num_blks = *last_blk - start_blk;
448 int xhdrs;
450 ASSERT(start_blk != 0 || *last_blk != start_blk);
452 if (!(bp = xlog_get_bp(log, num_blks))) {
453 if (!(bp = xlog_get_bp(log, 1)))
454 return ENOMEM;
455 smallmem = 1;
456 } else {
457 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
458 if (error)
459 goto out;
460 offset += ((num_blks - 1) << BBSHIFT);
463 for (i = (*last_blk) - 1; i >= 0; i--) {
464 if (i < start_blk) {
465 /* valid log record not found */
466 xlog_warn(
467 "XFS: Log inconsistent (didn't find previous header)");
468 ASSERT(0);
469 error = XFS_ERROR(EIO);
470 goto out;
473 if (smallmem) {
474 error = xlog_bread(log, i, 1, bp, &offset);
475 if (error)
476 goto out;
479 head = (xlog_rec_header_t *)offset;
481 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(head->h_magicno))
482 break;
484 if (!smallmem)
485 offset -= BBSIZE;
489 * We hit the beginning of the physical log & still no header. Return
490 * to caller. If caller can handle a return of -1, then this routine
491 * will be called again for the end of the physical log.
493 if (i == -1) {
494 error = -1;
495 goto out;
499 * We have the final block of the good log (the first block
500 * of the log record _before_ the head. So we check the uuid.
502 if ((error = xlog_header_check_mount(log->l_mp, head)))
503 goto out;
506 * We may have found a log record header before we expected one.
507 * last_blk will be the 1st block # with a given cycle #. We may end
508 * up reading an entire log record. In this case, we don't want to
509 * reset last_blk. Only when last_blk points in the middle of a log
510 * record do we update last_blk.
512 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
513 uint h_size = be32_to_cpu(head->h_size);
515 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
516 if (h_size % XLOG_HEADER_CYCLE_SIZE)
517 xhdrs++;
518 } else {
519 xhdrs = 1;
522 if (*last_blk - i + extra_bblks !=
523 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
524 *last_blk = i;
526 out:
527 xlog_put_bp(bp);
528 return error;
532 * Head is defined to be the point of the log where the next log write
533 * write could go. This means that incomplete LR writes at the end are
534 * eliminated when calculating the head. We aren't guaranteed that previous
535 * LR have complete transactions. We only know that a cycle number of
536 * current cycle number -1 won't be present in the log if we start writing
537 * from our current block number.
539 * last_blk contains the block number of the first block with a given
540 * cycle number.
542 * Return: zero if normal, non-zero if error.
544 STATIC int
545 xlog_find_head(
546 xlog_t *log,
547 xfs_daddr_t *return_head_blk)
549 xfs_buf_t *bp;
550 xfs_caddr_t offset;
551 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
552 int num_scan_bblks;
553 uint first_half_cycle, last_half_cycle;
554 uint stop_on_cycle;
555 int error, log_bbnum = log->l_logBBsize;
557 /* Is the end of the log device zeroed? */
558 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
559 *return_head_blk = first_blk;
561 /* Is the whole lot zeroed? */
562 if (!first_blk) {
563 /* Linux XFS shouldn't generate totally zeroed logs -
564 * mkfs etc write a dummy unmount record to a fresh
565 * log so we can store the uuid in there
567 xlog_warn("XFS: totally zeroed log");
570 return 0;
571 } else if (error) {
572 xlog_warn("XFS: empty log check failed");
573 return error;
576 first_blk = 0; /* get cycle # of 1st block */
577 bp = xlog_get_bp(log, 1);
578 if (!bp)
579 return ENOMEM;
581 error = xlog_bread(log, 0, 1, bp, &offset);
582 if (error)
583 goto bp_err;
585 first_half_cycle = xlog_get_cycle(offset);
587 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
588 error = xlog_bread(log, last_blk, 1, bp, &offset);
589 if (error)
590 goto bp_err;
592 last_half_cycle = xlog_get_cycle(offset);
593 ASSERT(last_half_cycle != 0);
596 * If the 1st half cycle number is equal to the last half cycle number,
597 * then the entire log is stamped with the same cycle number. In this
598 * case, head_blk can't be set to zero (which makes sense). The below
599 * math doesn't work out properly with head_blk equal to zero. Instead,
600 * we set it to log_bbnum which is an invalid block number, but this
601 * value makes the math correct. If head_blk doesn't changed through
602 * all the tests below, *head_blk is set to zero at the very end rather
603 * than log_bbnum. In a sense, log_bbnum and zero are the same block
604 * in a circular file.
606 if (first_half_cycle == last_half_cycle) {
608 * In this case we believe that the entire log should have
609 * cycle number last_half_cycle. We need to scan backwards
610 * from the end verifying that there are no holes still
611 * containing last_half_cycle - 1. If we find such a hole,
612 * then the start of that hole will be the new head. The
613 * simple case looks like
614 * x | x ... | x - 1 | x
615 * Another case that fits this picture would be
616 * x | x + 1 | x ... | x
617 * In this case the head really is somewhere at the end of the
618 * log, as one of the latest writes at the beginning was
619 * incomplete.
620 * One more case is
621 * x | x + 1 | x ... | x - 1 | x
622 * This is really the combination of the above two cases, and
623 * the head has to end up at the start of the x-1 hole at the
624 * end of the log.
626 * In the 256k log case, we will read from the beginning to the
627 * end of the log and search for cycle numbers equal to x-1.
628 * We don't worry about the x+1 blocks that we encounter,
629 * because we know that they cannot be the head since the log
630 * started with x.
632 head_blk = log_bbnum;
633 stop_on_cycle = last_half_cycle - 1;
634 } else {
636 * In this case we want to find the first block with cycle
637 * number matching last_half_cycle. We expect the log to be
638 * some variation on
639 * x + 1 ... | x ...
640 * The first block with cycle number x (last_half_cycle) will
641 * be where the new head belongs. First we do a binary search
642 * for the first occurrence of last_half_cycle. The binary
643 * search may not be totally accurate, so then we scan back
644 * from there looking for occurrences of last_half_cycle before
645 * us. If that backwards scan wraps around the beginning of
646 * the log, then we look for occurrences of last_half_cycle - 1
647 * at the end of the log. The cases we're looking for look
648 * like
649 * x + 1 ... | x | x + 1 | x ...
650 * ^ binary search stopped here
651 * or
652 * x + 1 ... | x ... | x - 1 | x
653 * <---------> less than scan distance
655 stop_on_cycle = last_half_cycle;
656 if ((error = xlog_find_cycle_start(log, bp, first_blk,
657 &head_blk, last_half_cycle)))
658 goto bp_err;
662 * Now validate the answer. Scan back some number of maximum possible
663 * blocks and make sure each one has the expected cycle number. The
664 * maximum is determined by the total possible amount of buffering
665 * in the in-core log. The following number can be made tighter if
666 * we actually look at the block size of the filesystem.
668 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
669 if (head_blk >= num_scan_bblks) {
671 * We are guaranteed that the entire check can be performed
672 * in one buffer.
674 start_blk = head_blk - num_scan_bblks;
675 if ((error = xlog_find_verify_cycle(log,
676 start_blk, num_scan_bblks,
677 stop_on_cycle, &new_blk)))
678 goto bp_err;
679 if (new_blk != -1)
680 head_blk = new_blk;
681 } else { /* need to read 2 parts of log */
683 * We are going to scan backwards in the log in two parts.
684 * First we scan the physical end of the log. In this part
685 * of the log, we are looking for blocks with cycle number
686 * last_half_cycle - 1.
687 * If we find one, then we know that the log starts there, as
688 * we've found a hole that didn't get written in going around
689 * the end of the physical log. The simple case for this is
690 * x + 1 ... | x ... | x - 1 | x
691 * <---------> less than scan distance
692 * If all of the blocks at the end of the log have cycle number
693 * last_half_cycle, then we check the blocks at the start of
694 * the log looking for occurrences of last_half_cycle. If we
695 * find one, then our current estimate for the location of the
696 * first occurrence of last_half_cycle is wrong and we move
697 * back to the hole we've found. This case looks like
698 * x + 1 ... | x | x + 1 | x ...
699 * ^ binary search stopped here
700 * Another case we need to handle that only occurs in 256k
701 * logs is
702 * x + 1 ... | x ... | x+1 | x ...
703 * ^ binary search stops here
704 * In a 256k log, the scan at the end of the log will see the
705 * x + 1 blocks. We need to skip past those since that is
706 * certainly not the head of the log. By searching for
707 * last_half_cycle-1 we accomplish that.
709 start_blk = log_bbnum - num_scan_bblks + head_blk;
710 ASSERT(head_blk <= INT_MAX &&
711 (xfs_daddr_t) num_scan_bblks - head_blk >= 0);
712 if ((error = xlog_find_verify_cycle(log, start_blk,
713 num_scan_bblks - (int)head_blk,
714 (stop_on_cycle - 1), &new_blk)))
715 goto bp_err;
716 if (new_blk != -1) {
717 head_blk = new_blk;
718 goto bad_blk;
722 * Scan beginning of log now. The last part of the physical
723 * log is good. This scan needs to verify that it doesn't find
724 * the last_half_cycle.
726 start_blk = 0;
727 ASSERT(head_blk <= INT_MAX);
728 if ((error = xlog_find_verify_cycle(log,
729 start_blk, (int)head_blk,
730 stop_on_cycle, &new_blk)))
731 goto bp_err;
732 if (new_blk != -1)
733 head_blk = new_blk;
736 bad_blk:
738 * Now we need to make sure head_blk is not pointing to a block in
739 * the middle of a log record.
741 num_scan_bblks = XLOG_REC_SHIFT(log);
742 if (head_blk >= num_scan_bblks) {
743 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
745 /* start ptr at last block ptr before head_blk */
746 if ((error = xlog_find_verify_log_record(log, start_blk,
747 &head_blk, 0)) == -1) {
748 error = XFS_ERROR(EIO);
749 goto bp_err;
750 } else if (error)
751 goto bp_err;
752 } else {
753 start_blk = 0;
754 ASSERT(head_blk <= INT_MAX);
755 if ((error = xlog_find_verify_log_record(log, start_blk,
756 &head_blk, 0)) == -1) {
757 /* We hit the beginning of the log during our search */
758 start_blk = log_bbnum - num_scan_bblks + head_blk;
759 new_blk = log_bbnum;
760 ASSERT(start_blk <= INT_MAX &&
761 (xfs_daddr_t) log_bbnum-start_blk >= 0);
762 ASSERT(head_blk <= INT_MAX);
763 if ((error = xlog_find_verify_log_record(log,
764 start_blk, &new_blk,
765 (int)head_blk)) == -1) {
766 error = XFS_ERROR(EIO);
767 goto bp_err;
768 } else if (error)
769 goto bp_err;
770 if (new_blk != log_bbnum)
771 head_blk = new_blk;
772 } else if (error)
773 goto bp_err;
776 xlog_put_bp(bp);
777 if (head_blk == log_bbnum)
778 *return_head_blk = 0;
779 else
780 *return_head_blk = head_blk;
782 * When returning here, we have a good block number. Bad block
783 * means that during a previous crash, we didn't have a clean break
784 * from cycle number N to cycle number N-1. In this case, we need
785 * to find the first block with cycle number N-1.
787 return 0;
789 bp_err:
790 xlog_put_bp(bp);
792 if (error)
793 xlog_warn("XFS: failed to find log head");
794 return error;
798 * Find the sync block number or the tail of the log.
800 * This will be the block number of the last record to have its
801 * associated buffers synced to disk. Every log record header has
802 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
803 * to get a sync block number. The only concern is to figure out which
804 * log record header to believe.
806 * The following algorithm uses the log record header with the largest
807 * lsn. The entire log record does not need to be valid. We only care
808 * that the header is valid.
810 * We could speed up search by using current head_blk buffer, but it is not
811 * available.
814 xlog_find_tail(
815 xlog_t *log,
816 xfs_daddr_t *head_blk,
817 xfs_daddr_t *tail_blk)
819 xlog_rec_header_t *rhead;
820 xlog_op_header_t *op_head;
821 xfs_caddr_t offset = NULL;
822 xfs_buf_t *bp;
823 int error, i, found;
824 xfs_daddr_t umount_data_blk;
825 xfs_daddr_t after_umount_blk;
826 xfs_lsn_t tail_lsn;
827 int hblks;
829 found = 0;
832 * Find previous log record
834 if ((error = xlog_find_head(log, head_blk)))
835 return error;
837 bp = xlog_get_bp(log, 1);
838 if (!bp)
839 return ENOMEM;
840 if (*head_blk == 0) { /* special case */
841 error = xlog_bread(log, 0, 1, bp, &offset);
842 if (error)
843 goto bread_err;
845 if (xlog_get_cycle(offset) == 0) {
846 *tail_blk = 0;
847 /* leave all other log inited values alone */
848 goto exit;
853 * Search backwards looking for log record header block
855 ASSERT(*head_blk < INT_MAX);
856 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
857 error = xlog_bread(log, i, 1, bp, &offset);
858 if (error)
859 goto bread_err;
861 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(*(__be32 *)offset)) {
862 found = 1;
863 break;
867 * If we haven't found the log record header block, start looking
868 * again from the end of the physical log. XXXmiken: There should be
869 * a check here to make sure we didn't search more than N blocks in
870 * the previous code.
872 if (!found) {
873 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
874 error = xlog_bread(log, i, 1, bp, &offset);
875 if (error)
876 goto bread_err;
878 if (XLOG_HEADER_MAGIC_NUM ==
879 be32_to_cpu(*(__be32 *)offset)) {
880 found = 2;
881 break;
885 if (!found) {
886 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
887 ASSERT(0);
888 return XFS_ERROR(EIO);
891 /* find blk_no of tail of log */
892 rhead = (xlog_rec_header_t *)offset;
893 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
896 * Reset log values according to the state of the log when we
897 * crashed. In the case where head_blk == 0, we bump curr_cycle
898 * one because the next write starts a new cycle rather than
899 * continuing the cycle of the last good log record. At this
900 * point we have guaranteed that all partial log records have been
901 * accounted for. Therefore, we know that the last good log record
902 * written was complete and ended exactly on the end boundary
903 * of the physical log.
905 log->l_prev_block = i;
906 log->l_curr_block = (int)*head_blk;
907 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
908 if (found == 2)
909 log->l_curr_cycle++;
910 log->l_tail_lsn = be64_to_cpu(rhead->h_tail_lsn);
911 log->l_last_sync_lsn = be64_to_cpu(rhead->h_lsn);
912 log->l_grant_reserve_cycle = log->l_curr_cycle;
913 log->l_grant_reserve_bytes = BBTOB(log->l_curr_block);
914 log->l_grant_write_cycle = log->l_curr_cycle;
915 log->l_grant_write_bytes = BBTOB(log->l_curr_block);
918 * Look for unmount record. If we find it, then we know there
919 * was a clean unmount. Since 'i' could be the last block in
920 * the physical log, we convert to a log block before comparing
921 * to the head_blk.
923 * Save the current tail lsn to use to pass to
924 * xlog_clear_stale_blocks() below. We won't want to clear the
925 * unmount record if there is one, so we pass the lsn of the
926 * unmount record rather than the block after it.
928 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
929 int h_size = be32_to_cpu(rhead->h_size);
930 int h_version = be32_to_cpu(rhead->h_version);
932 if ((h_version & XLOG_VERSION_2) &&
933 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
934 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
935 if (h_size % XLOG_HEADER_CYCLE_SIZE)
936 hblks++;
937 } else {
938 hblks = 1;
940 } else {
941 hblks = 1;
943 after_umount_blk = (i + hblks + (int)
944 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
945 tail_lsn = log->l_tail_lsn;
946 if (*head_blk == after_umount_blk &&
947 be32_to_cpu(rhead->h_num_logops) == 1) {
948 umount_data_blk = (i + hblks) % log->l_logBBsize;
949 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
950 if (error)
951 goto bread_err;
953 op_head = (xlog_op_header_t *)offset;
954 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
956 * Set tail and last sync so that newly written
957 * log records will point recovery to after the
958 * current unmount record.
960 log->l_tail_lsn =
961 xlog_assign_lsn(log->l_curr_cycle,
962 after_umount_blk);
963 log->l_last_sync_lsn =
964 xlog_assign_lsn(log->l_curr_cycle,
965 after_umount_blk);
966 *tail_blk = after_umount_blk;
969 * Note that the unmount was clean. If the unmount
970 * was not clean, we need to know this to rebuild the
971 * superblock counters from the perag headers if we
972 * have a filesystem using non-persistent counters.
974 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
979 * Make sure that there are no blocks in front of the head
980 * with the same cycle number as the head. This can happen
981 * because we allow multiple outstanding log writes concurrently,
982 * and the later writes might make it out before earlier ones.
984 * We use the lsn from before modifying it so that we'll never
985 * overwrite the unmount record after a clean unmount.
987 * Do this only if we are going to recover the filesystem
989 * NOTE: This used to say "if (!readonly)"
990 * However on Linux, we can & do recover a read-only filesystem.
991 * We only skip recovery if NORECOVERY is specified on mount,
992 * in which case we would not be here.
994 * But... if the -device- itself is readonly, just skip this.
995 * We can't recover this device anyway, so it won't matter.
997 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) {
998 error = xlog_clear_stale_blocks(log, tail_lsn);
1001 bread_err:
1002 exit:
1003 xlog_put_bp(bp);
1005 if (error)
1006 xlog_warn("XFS: failed to locate log tail");
1007 return error;
1011 * Is the log zeroed at all?
1013 * The last binary search should be changed to perform an X block read
1014 * once X becomes small enough. You can then search linearly through
1015 * the X blocks. This will cut down on the number of reads we need to do.
1017 * If the log is partially zeroed, this routine will pass back the blkno
1018 * of the first block with cycle number 0. It won't have a complete LR
1019 * preceding it.
1021 * Return:
1022 * 0 => the log is completely written to
1023 * -1 => use *blk_no as the first block of the log
1024 * >0 => error has occurred
1026 STATIC int
1027 xlog_find_zeroed(
1028 xlog_t *log,
1029 xfs_daddr_t *blk_no)
1031 xfs_buf_t *bp;
1032 xfs_caddr_t offset;
1033 uint first_cycle, last_cycle;
1034 xfs_daddr_t new_blk, last_blk, start_blk;
1035 xfs_daddr_t num_scan_bblks;
1036 int error, log_bbnum = log->l_logBBsize;
1038 *blk_no = 0;
1040 /* check totally zeroed log */
1041 bp = xlog_get_bp(log, 1);
1042 if (!bp)
1043 return ENOMEM;
1044 error = xlog_bread(log, 0, 1, bp, &offset);
1045 if (error)
1046 goto bp_err;
1048 first_cycle = xlog_get_cycle(offset);
1049 if (first_cycle == 0) { /* completely zeroed log */
1050 *blk_no = 0;
1051 xlog_put_bp(bp);
1052 return -1;
1055 /* check partially zeroed log */
1056 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1057 if (error)
1058 goto bp_err;
1060 last_cycle = xlog_get_cycle(offset);
1061 if (last_cycle != 0) { /* log completely written to */
1062 xlog_put_bp(bp);
1063 return 0;
1064 } else if (first_cycle != 1) {
1066 * If the cycle of the last block is zero, the cycle of
1067 * the first block must be 1. If it's not, maybe we're
1068 * not looking at a log... Bail out.
1070 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1071 return XFS_ERROR(EINVAL);
1074 /* we have a partially zeroed log */
1075 last_blk = log_bbnum-1;
1076 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1077 goto bp_err;
1080 * Validate the answer. Because there is no way to guarantee that
1081 * the entire log is made up of log records which are the same size,
1082 * we scan over the defined maximum blocks. At this point, the maximum
1083 * is not chosen to mean anything special. XXXmiken
1085 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1086 ASSERT(num_scan_bblks <= INT_MAX);
1088 if (last_blk < num_scan_bblks)
1089 num_scan_bblks = last_blk;
1090 start_blk = last_blk - num_scan_bblks;
1093 * We search for any instances of cycle number 0 that occur before
1094 * our current estimate of the head. What we're trying to detect is
1095 * 1 ... | 0 | 1 | 0...
1096 * ^ binary search ends here
1098 if ((error = xlog_find_verify_cycle(log, start_blk,
1099 (int)num_scan_bblks, 0, &new_blk)))
1100 goto bp_err;
1101 if (new_blk != -1)
1102 last_blk = new_blk;
1105 * Potentially backup over partial log record write. We don't need
1106 * to search the end of the log because we know it is zero.
1108 if ((error = xlog_find_verify_log_record(log, start_blk,
1109 &last_blk, 0)) == -1) {
1110 error = XFS_ERROR(EIO);
1111 goto bp_err;
1112 } else if (error)
1113 goto bp_err;
1115 *blk_no = last_blk;
1116 bp_err:
1117 xlog_put_bp(bp);
1118 if (error)
1119 return error;
1120 return -1;
1124 * These are simple subroutines used by xlog_clear_stale_blocks() below
1125 * to initialize a buffer full of empty log record headers and write
1126 * them into the log.
1128 STATIC void
1129 xlog_add_record(
1130 xlog_t *log,
1131 xfs_caddr_t buf,
1132 int cycle,
1133 int block,
1134 int tail_cycle,
1135 int tail_block)
1137 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1139 memset(buf, 0, BBSIZE);
1140 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1141 recp->h_cycle = cpu_to_be32(cycle);
1142 recp->h_version = cpu_to_be32(
1143 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1144 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1145 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1146 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1147 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1150 STATIC int
1151 xlog_write_log_records(
1152 xlog_t *log,
1153 int cycle,
1154 int start_block,
1155 int blocks,
1156 int tail_cycle,
1157 int tail_block)
1159 xfs_caddr_t offset;
1160 xfs_buf_t *bp;
1161 int balign, ealign;
1162 int sectbb = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
1163 int end_block = start_block + blocks;
1164 int bufblks;
1165 int error = 0;
1166 int i, j = 0;
1168 bufblks = 1 << ffs(blocks);
1169 while (!(bp = xlog_get_bp(log, bufblks))) {
1170 bufblks >>= 1;
1171 if (bufblks <= log->l_sectbb_log)
1172 return ENOMEM;
1175 /* We may need to do a read at the start to fill in part of
1176 * the buffer in the starting sector not covered by the first
1177 * write below.
1179 balign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, start_block);
1180 if (balign != start_block) {
1181 error = xlog_bread_noalign(log, start_block, 1, bp);
1182 if (error)
1183 goto out_put_bp;
1185 j = start_block - balign;
1188 for (i = start_block; i < end_block; i += bufblks) {
1189 int bcount, endcount;
1191 bcount = min(bufblks, end_block - start_block);
1192 endcount = bcount - j;
1194 /* We may need to do a read at the end to fill in part of
1195 * the buffer in the final sector not covered by the write.
1196 * If this is the same sector as the above read, skip it.
1198 ealign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, end_block);
1199 if (j == 0 && (start_block + endcount > ealign)) {
1200 offset = XFS_BUF_PTR(bp);
1201 balign = BBTOB(ealign - start_block);
1202 error = XFS_BUF_SET_PTR(bp, offset + balign,
1203 BBTOB(sectbb));
1204 if (error)
1205 break;
1207 error = xlog_bread_noalign(log, ealign, sectbb, bp);
1208 if (error)
1209 break;
1211 error = XFS_BUF_SET_PTR(bp, offset, bufblks);
1212 if (error)
1213 break;
1216 offset = xlog_align(log, start_block, endcount, bp);
1217 for (; j < endcount; j++) {
1218 xlog_add_record(log, offset, cycle, i+j,
1219 tail_cycle, tail_block);
1220 offset += BBSIZE;
1222 error = xlog_bwrite(log, start_block, endcount, bp);
1223 if (error)
1224 break;
1225 start_block += endcount;
1226 j = 0;
1229 out_put_bp:
1230 xlog_put_bp(bp);
1231 return error;
1235 * This routine is called to blow away any incomplete log writes out
1236 * in front of the log head. We do this so that we won't become confused
1237 * if we come up, write only a little bit more, and then crash again.
1238 * If we leave the partial log records out there, this situation could
1239 * cause us to think those partial writes are valid blocks since they
1240 * have the current cycle number. We get rid of them by overwriting them
1241 * with empty log records with the old cycle number rather than the
1242 * current one.
1244 * The tail lsn is passed in rather than taken from
1245 * the log so that we will not write over the unmount record after a
1246 * clean unmount in a 512 block log. Doing so would leave the log without
1247 * any valid log records in it until a new one was written. If we crashed
1248 * during that time we would not be able to recover.
1250 STATIC int
1251 xlog_clear_stale_blocks(
1252 xlog_t *log,
1253 xfs_lsn_t tail_lsn)
1255 int tail_cycle, head_cycle;
1256 int tail_block, head_block;
1257 int tail_distance, max_distance;
1258 int distance;
1259 int error;
1261 tail_cycle = CYCLE_LSN(tail_lsn);
1262 tail_block = BLOCK_LSN(tail_lsn);
1263 head_cycle = log->l_curr_cycle;
1264 head_block = log->l_curr_block;
1267 * Figure out the distance between the new head of the log
1268 * and the tail. We want to write over any blocks beyond the
1269 * head that we may have written just before the crash, but
1270 * we don't want to overwrite the tail of the log.
1272 if (head_cycle == tail_cycle) {
1274 * The tail is behind the head in the physical log,
1275 * so the distance from the head to the tail is the
1276 * distance from the head to the end of the log plus
1277 * the distance from the beginning of the log to the
1278 * tail.
1280 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1281 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1282 XFS_ERRLEVEL_LOW, log->l_mp);
1283 return XFS_ERROR(EFSCORRUPTED);
1285 tail_distance = tail_block + (log->l_logBBsize - head_block);
1286 } else {
1288 * The head is behind the tail in the physical log,
1289 * so the distance from the head to the tail is just
1290 * the tail block minus the head block.
1292 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1293 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1294 XFS_ERRLEVEL_LOW, log->l_mp);
1295 return XFS_ERROR(EFSCORRUPTED);
1297 tail_distance = tail_block - head_block;
1301 * If the head is right up against the tail, we can't clear
1302 * anything.
1304 if (tail_distance <= 0) {
1305 ASSERT(tail_distance == 0);
1306 return 0;
1309 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1311 * Take the smaller of the maximum amount of outstanding I/O
1312 * we could have and the distance to the tail to clear out.
1313 * We take the smaller so that we don't overwrite the tail and
1314 * we don't waste all day writing from the head to the tail
1315 * for no reason.
1317 max_distance = MIN(max_distance, tail_distance);
1319 if ((head_block + max_distance) <= log->l_logBBsize) {
1321 * We can stomp all the blocks we need to without
1322 * wrapping around the end of the log. Just do it
1323 * in a single write. Use the cycle number of the
1324 * current cycle minus one so that the log will look like:
1325 * n ... | n - 1 ...
1327 error = xlog_write_log_records(log, (head_cycle - 1),
1328 head_block, max_distance, tail_cycle,
1329 tail_block);
1330 if (error)
1331 return error;
1332 } else {
1334 * We need to wrap around the end of the physical log in
1335 * order to clear all the blocks. Do it in two separate
1336 * I/Os. The first write should be from the head to the
1337 * end of the physical log, and it should use the current
1338 * cycle number minus one just like above.
1340 distance = log->l_logBBsize - head_block;
1341 error = xlog_write_log_records(log, (head_cycle - 1),
1342 head_block, distance, tail_cycle,
1343 tail_block);
1345 if (error)
1346 return error;
1349 * Now write the blocks at the start of the physical log.
1350 * This writes the remainder of the blocks we want to clear.
1351 * It uses the current cycle number since we're now on the
1352 * same cycle as the head so that we get:
1353 * n ... n ... | n - 1 ...
1354 * ^^^^^ blocks we're writing
1356 distance = max_distance - (log->l_logBBsize - head_block);
1357 error = xlog_write_log_records(log, head_cycle, 0, distance,
1358 tail_cycle, tail_block);
1359 if (error)
1360 return error;
1363 return 0;
1366 /******************************************************************************
1368 * Log recover routines
1370 ******************************************************************************
1373 STATIC xlog_recover_t *
1374 xlog_recover_find_tid(
1375 xlog_recover_t *q,
1376 xlog_tid_t tid)
1378 xlog_recover_t *p = q;
1380 while (p != NULL) {
1381 if (p->r_log_tid == tid)
1382 break;
1383 p = p->r_next;
1385 return p;
1388 STATIC void
1389 xlog_recover_put_hashq(
1390 xlog_recover_t **q,
1391 xlog_recover_t *trans)
1393 trans->r_next = *q;
1394 *q = trans;
1397 STATIC void
1398 xlog_recover_add_item(
1399 xlog_recover_item_t **itemq)
1401 xlog_recover_item_t *item;
1403 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1404 xlog_recover_insert_item_backq(itemq, item);
1407 STATIC int
1408 xlog_recover_add_to_cont_trans(
1409 xlog_recover_t *trans,
1410 xfs_caddr_t dp,
1411 int len)
1413 xlog_recover_item_t *item;
1414 xfs_caddr_t ptr, old_ptr;
1415 int old_len;
1417 item = trans->r_itemq;
1418 if (item == NULL) {
1419 /* finish copying rest of trans header */
1420 xlog_recover_add_item(&trans->r_itemq);
1421 ptr = (xfs_caddr_t) &trans->r_theader +
1422 sizeof(xfs_trans_header_t) - len;
1423 memcpy(ptr, dp, len); /* d, s, l */
1424 return 0;
1426 item = item->ri_prev;
1428 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1429 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1431 ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
1432 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1433 item->ri_buf[item->ri_cnt-1].i_len += len;
1434 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1435 return 0;
1439 * The next region to add is the start of a new region. It could be
1440 * a whole region or it could be the first part of a new region. Because
1441 * of this, the assumption here is that the type and size fields of all
1442 * format structures fit into the first 32 bits of the structure.
1444 * This works because all regions must be 32 bit aligned. Therefore, we
1445 * either have both fields or we have neither field. In the case we have
1446 * neither field, the data part of the region is zero length. We only have
1447 * a log_op_header and can throw away the header since a new one will appear
1448 * later. If we have at least 4 bytes, then we can determine how many regions
1449 * will appear in the current log item.
1451 STATIC int
1452 xlog_recover_add_to_trans(
1453 xlog_recover_t *trans,
1454 xfs_caddr_t dp,
1455 int len)
1457 xfs_inode_log_format_t *in_f; /* any will do */
1458 xlog_recover_item_t *item;
1459 xfs_caddr_t ptr;
1461 if (!len)
1462 return 0;
1463 item = trans->r_itemq;
1464 if (item == NULL) {
1465 /* we need to catch log corruptions here */
1466 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
1467 xlog_warn("XFS: xlog_recover_add_to_trans: "
1468 "bad header magic number");
1469 ASSERT(0);
1470 return XFS_ERROR(EIO);
1472 if (len == sizeof(xfs_trans_header_t))
1473 xlog_recover_add_item(&trans->r_itemq);
1474 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1475 return 0;
1478 ptr = kmem_alloc(len, KM_SLEEP);
1479 memcpy(ptr, dp, len);
1480 in_f = (xfs_inode_log_format_t *)ptr;
1482 if (item->ri_prev->ri_total != 0 &&
1483 item->ri_prev->ri_total == item->ri_prev->ri_cnt) {
1484 xlog_recover_add_item(&trans->r_itemq);
1486 item = trans->r_itemq;
1487 item = item->ri_prev;
1489 if (item->ri_total == 0) { /* first region to be added */
1490 if (in_f->ilf_size == 0 ||
1491 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
1492 xlog_warn(
1493 "XFS: bad number of regions (%d) in inode log format",
1494 in_f->ilf_size);
1495 ASSERT(0);
1496 return XFS_ERROR(EIO);
1499 item->ri_total = in_f->ilf_size;
1500 item->ri_buf =
1501 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1502 KM_SLEEP);
1504 ASSERT(item->ri_total > item->ri_cnt);
1505 /* Description region is ri_buf[0] */
1506 item->ri_buf[item->ri_cnt].i_addr = ptr;
1507 item->ri_buf[item->ri_cnt].i_len = len;
1508 item->ri_cnt++;
1509 return 0;
1512 STATIC void
1513 xlog_recover_new_tid(
1514 xlog_recover_t **q,
1515 xlog_tid_t tid,
1516 xfs_lsn_t lsn)
1518 xlog_recover_t *trans;
1520 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1521 trans->r_log_tid = tid;
1522 trans->r_lsn = lsn;
1523 xlog_recover_put_hashq(q, trans);
1526 STATIC int
1527 xlog_recover_unlink_tid(
1528 xlog_recover_t **q,
1529 xlog_recover_t *trans)
1531 xlog_recover_t *tp;
1532 int found = 0;
1534 ASSERT(trans != NULL);
1535 if (trans == *q) {
1536 *q = (*q)->r_next;
1537 } else {
1538 tp = *q;
1539 while (tp) {
1540 if (tp->r_next == trans) {
1541 found = 1;
1542 break;
1544 tp = tp->r_next;
1546 if (!found) {
1547 xlog_warn(
1548 "XFS: xlog_recover_unlink_tid: trans not found");
1549 ASSERT(0);
1550 return XFS_ERROR(EIO);
1552 tp->r_next = tp->r_next->r_next;
1554 return 0;
1557 STATIC void
1558 xlog_recover_insert_item_backq(
1559 xlog_recover_item_t **q,
1560 xlog_recover_item_t *item)
1562 if (*q == NULL) {
1563 item->ri_prev = item->ri_next = item;
1564 *q = item;
1565 } else {
1566 item->ri_next = *q;
1567 item->ri_prev = (*q)->ri_prev;
1568 (*q)->ri_prev = item;
1569 item->ri_prev->ri_next = item;
1573 STATIC void
1574 xlog_recover_insert_item_frontq(
1575 xlog_recover_item_t **q,
1576 xlog_recover_item_t *item)
1578 xlog_recover_insert_item_backq(q, item);
1579 *q = item;
1582 STATIC int
1583 xlog_recover_reorder_trans(
1584 xlog_recover_t *trans)
1586 xlog_recover_item_t *first_item, *itemq, *itemq_next;
1587 xfs_buf_log_format_t *buf_f;
1588 ushort flags = 0;
1590 first_item = itemq = trans->r_itemq;
1591 trans->r_itemq = NULL;
1592 do {
1593 itemq_next = itemq->ri_next;
1594 buf_f = (xfs_buf_log_format_t *)itemq->ri_buf[0].i_addr;
1596 switch (ITEM_TYPE(itemq)) {
1597 case XFS_LI_BUF:
1598 flags = buf_f->blf_flags;
1599 if (!(flags & XFS_BLI_CANCEL)) {
1600 xlog_recover_insert_item_frontq(&trans->r_itemq,
1601 itemq);
1602 break;
1604 case XFS_LI_INODE:
1605 case XFS_LI_DQUOT:
1606 case XFS_LI_QUOTAOFF:
1607 case XFS_LI_EFD:
1608 case XFS_LI_EFI:
1609 xlog_recover_insert_item_backq(&trans->r_itemq, itemq);
1610 break;
1611 default:
1612 xlog_warn(
1613 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1614 ASSERT(0);
1615 return XFS_ERROR(EIO);
1617 itemq = itemq_next;
1618 } while (first_item != itemq);
1619 return 0;
1623 * Build up the table of buf cancel records so that we don't replay
1624 * cancelled data in the second pass. For buffer records that are
1625 * not cancel records, there is nothing to do here so we just return.
1627 * If we get a cancel record which is already in the table, this indicates
1628 * that the buffer was cancelled multiple times. In order to ensure
1629 * that during pass 2 we keep the record in the table until we reach its
1630 * last occurrence in the log, we keep a reference count in the cancel
1631 * record in the table to tell us how many times we expect to see this
1632 * record during the second pass.
1634 STATIC void
1635 xlog_recover_do_buffer_pass1(
1636 xlog_t *log,
1637 xfs_buf_log_format_t *buf_f)
1639 xfs_buf_cancel_t *bcp;
1640 xfs_buf_cancel_t *nextp;
1641 xfs_buf_cancel_t *prevp;
1642 xfs_buf_cancel_t **bucket;
1643 xfs_daddr_t blkno = 0;
1644 uint len = 0;
1645 ushort flags = 0;
1647 switch (buf_f->blf_type) {
1648 case XFS_LI_BUF:
1649 blkno = buf_f->blf_blkno;
1650 len = buf_f->blf_len;
1651 flags = buf_f->blf_flags;
1652 break;
1656 * If this isn't a cancel buffer item, then just return.
1658 if (!(flags & XFS_BLI_CANCEL))
1659 return;
1662 * Insert an xfs_buf_cancel record into the hash table of
1663 * them. If there is already an identical record, bump
1664 * its reference count.
1666 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1667 XLOG_BC_TABLE_SIZE];
1669 * If the hash bucket is empty then just insert a new record into
1670 * the bucket.
1672 if (*bucket == NULL) {
1673 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1674 KM_SLEEP);
1675 bcp->bc_blkno = blkno;
1676 bcp->bc_len = len;
1677 bcp->bc_refcount = 1;
1678 bcp->bc_next = NULL;
1679 *bucket = bcp;
1680 return;
1684 * The hash bucket is not empty, so search for duplicates of our
1685 * record. If we find one them just bump its refcount. If not
1686 * then add us at the end of the list.
1688 prevp = NULL;
1689 nextp = *bucket;
1690 while (nextp != NULL) {
1691 if (nextp->bc_blkno == blkno && nextp->bc_len == len) {
1692 nextp->bc_refcount++;
1693 return;
1695 prevp = nextp;
1696 nextp = nextp->bc_next;
1698 ASSERT(prevp != NULL);
1699 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1700 KM_SLEEP);
1701 bcp->bc_blkno = blkno;
1702 bcp->bc_len = len;
1703 bcp->bc_refcount = 1;
1704 bcp->bc_next = NULL;
1705 prevp->bc_next = bcp;
1709 * Check to see whether the buffer being recovered has a corresponding
1710 * entry in the buffer cancel record table. If it does then return 1
1711 * so that it will be cancelled, otherwise return 0. If the buffer is
1712 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
1713 * the refcount on the entry in the table and remove it from the table
1714 * if this is the last reference.
1716 * We remove the cancel record from the table when we encounter its
1717 * last occurrence in the log so that if the same buffer is re-used
1718 * again after its last cancellation we actually replay the changes
1719 * made at that point.
1721 STATIC int
1722 xlog_check_buffer_cancelled(
1723 xlog_t *log,
1724 xfs_daddr_t blkno,
1725 uint len,
1726 ushort flags)
1728 xfs_buf_cancel_t *bcp;
1729 xfs_buf_cancel_t *prevp;
1730 xfs_buf_cancel_t **bucket;
1732 if (log->l_buf_cancel_table == NULL) {
1734 * There is nothing in the table built in pass one,
1735 * so this buffer must not be cancelled.
1737 ASSERT(!(flags & XFS_BLI_CANCEL));
1738 return 0;
1741 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1742 XLOG_BC_TABLE_SIZE];
1743 bcp = *bucket;
1744 if (bcp == NULL) {
1746 * There is no corresponding entry in the table built
1747 * in pass one, so this buffer has not been cancelled.
1749 ASSERT(!(flags & XFS_BLI_CANCEL));
1750 return 0;
1754 * Search for an entry in the buffer cancel table that
1755 * matches our buffer.
1757 prevp = NULL;
1758 while (bcp != NULL) {
1759 if (bcp->bc_blkno == blkno && bcp->bc_len == len) {
1761 * We've go a match, so return 1 so that the
1762 * recovery of this buffer is cancelled.
1763 * If this buffer is actually a buffer cancel
1764 * log item, then decrement the refcount on the
1765 * one in the table and remove it if this is the
1766 * last reference.
1768 if (flags & XFS_BLI_CANCEL) {
1769 bcp->bc_refcount--;
1770 if (bcp->bc_refcount == 0) {
1771 if (prevp == NULL) {
1772 *bucket = bcp->bc_next;
1773 } else {
1774 prevp->bc_next = bcp->bc_next;
1776 kmem_free(bcp);
1779 return 1;
1781 prevp = bcp;
1782 bcp = bcp->bc_next;
1785 * We didn't find a corresponding entry in the table, so
1786 * return 0 so that the buffer is NOT cancelled.
1788 ASSERT(!(flags & XFS_BLI_CANCEL));
1789 return 0;
1792 STATIC int
1793 xlog_recover_do_buffer_pass2(
1794 xlog_t *log,
1795 xfs_buf_log_format_t *buf_f)
1797 xfs_daddr_t blkno = 0;
1798 ushort flags = 0;
1799 uint len = 0;
1801 switch (buf_f->blf_type) {
1802 case XFS_LI_BUF:
1803 blkno = buf_f->blf_blkno;
1804 flags = buf_f->blf_flags;
1805 len = buf_f->blf_len;
1806 break;
1809 return xlog_check_buffer_cancelled(log, blkno, len, flags);
1813 * Perform recovery for a buffer full of inodes. In these buffers,
1814 * the only data which should be recovered is that which corresponds
1815 * to the di_next_unlinked pointers in the on disk inode structures.
1816 * The rest of the data for the inodes is always logged through the
1817 * inodes themselves rather than the inode buffer and is recovered
1818 * in xlog_recover_do_inode_trans().
1820 * The only time when buffers full of inodes are fully recovered is
1821 * when the buffer is full of newly allocated inodes. In this case
1822 * the buffer will not be marked as an inode buffer and so will be
1823 * sent to xlog_recover_do_reg_buffer() below during recovery.
1825 STATIC int
1826 xlog_recover_do_inode_buffer(
1827 xfs_mount_t *mp,
1828 xlog_recover_item_t *item,
1829 xfs_buf_t *bp,
1830 xfs_buf_log_format_t *buf_f)
1832 int i;
1833 int item_index;
1834 int bit;
1835 int nbits;
1836 int reg_buf_offset;
1837 int reg_buf_bytes;
1838 int next_unlinked_offset;
1839 int inodes_per_buf;
1840 xfs_agino_t *logged_nextp;
1841 xfs_agino_t *buffer_nextp;
1842 unsigned int *data_map = NULL;
1843 unsigned int map_size = 0;
1845 switch (buf_f->blf_type) {
1846 case XFS_LI_BUF:
1847 data_map = buf_f->blf_data_map;
1848 map_size = buf_f->blf_map_size;
1849 break;
1852 * Set the variables corresponding to the current region to
1853 * 0 so that we'll initialize them on the first pass through
1854 * the loop.
1856 reg_buf_offset = 0;
1857 reg_buf_bytes = 0;
1858 bit = 0;
1859 nbits = 0;
1860 item_index = 0;
1861 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1862 for (i = 0; i < inodes_per_buf; i++) {
1863 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1864 offsetof(xfs_dinode_t, di_next_unlinked);
1866 while (next_unlinked_offset >=
1867 (reg_buf_offset + reg_buf_bytes)) {
1869 * The next di_next_unlinked field is beyond
1870 * the current logged region. Find the next
1871 * logged region that contains or is beyond
1872 * the current di_next_unlinked field.
1874 bit += nbits;
1875 bit = xfs_next_bit(data_map, map_size, bit);
1878 * If there are no more logged regions in the
1879 * buffer, then we're done.
1881 if (bit == -1) {
1882 return 0;
1885 nbits = xfs_contig_bits(data_map, map_size,
1886 bit);
1887 ASSERT(nbits > 0);
1888 reg_buf_offset = bit << XFS_BLI_SHIFT;
1889 reg_buf_bytes = nbits << XFS_BLI_SHIFT;
1890 item_index++;
1894 * If the current logged region starts after the current
1895 * di_next_unlinked field, then move on to the next
1896 * di_next_unlinked field.
1898 if (next_unlinked_offset < reg_buf_offset) {
1899 continue;
1902 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1903 ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0);
1904 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1907 * The current logged region contains a copy of the
1908 * current di_next_unlinked field. Extract its value
1909 * and copy it to the buffer copy.
1911 logged_nextp = (xfs_agino_t *)
1912 ((char *)(item->ri_buf[item_index].i_addr) +
1913 (next_unlinked_offset - reg_buf_offset));
1914 if (unlikely(*logged_nextp == 0)) {
1915 xfs_fs_cmn_err(CE_ALERT, mp,
1916 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1917 item, bp);
1918 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1919 XFS_ERRLEVEL_LOW, mp);
1920 return XFS_ERROR(EFSCORRUPTED);
1923 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1924 next_unlinked_offset);
1925 *buffer_nextp = *logged_nextp;
1928 return 0;
1932 * Perform a 'normal' buffer recovery. Each logged region of the
1933 * buffer should be copied over the corresponding region in the
1934 * given buffer. The bitmap in the buf log format structure indicates
1935 * where to place the logged data.
1937 /*ARGSUSED*/
1938 STATIC void
1939 xlog_recover_do_reg_buffer(
1940 xlog_recover_item_t *item,
1941 xfs_buf_t *bp,
1942 xfs_buf_log_format_t *buf_f)
1944 int i;
1945 int bit;
1946 int nbits;
1947 unsigned int *data_map = NULL;
1948 unsigned int map_size = 0;
1949 int error;
1951 switch (buf_f->blf_type) {
1952 case XFS_LI_BUF:
1953 data_map = buf_f->blf_data_map;
1954 map_size = buf_f->blf_map_size;
1955 break;
1957 bit = 0;
1958 i = 1; /* 0 is the buf format structure */
1959 while (1) {
1960 bit = xfs_next_bit(data_map, map_size, bit);
1961 if (bit == -1)
1962 break;
1963 nbits = xfs_contig_bits(data_map, map_size, bit);
1964 ASSERT(nbits > 0);
1965 ASSERT(item->ri_buf[i].i_addr != NULL);
1966 ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0);
1967 ASSERT(XFS_BUF_COUNT(bp) >=
1968 ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT));
1971 * Do a sanity check if this is a dquot buffer. Just checking
1972 * the first dquot in the buffer should do. XXXThis is
1973 * probably a good thing to do for other buf types also.
1975 error = 0;
1976 if (buf_f->blf_flags &
1977 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
1978 if (item->ri_buf[i].i_addr == NULL) {
1979 cmn_err(CE_ALERT,
1980 "XFS: NULL dquot in %s.", __func__);
1981 goto next;
1983 if (item->ri_buf[i].i_len < sizeof(xfs_dqblk_t)) {
1984 cmn_err(CE_ALERT,
1985 "XFS: dquot too small (%d) in %s.",
1986 item->ri_buf[i].i_len, __func__);
1987 goto next;
1989 error = xfs_qm_dqcheck((xfs_disk_dquot_t *)
1990 item->ri_buf[i].i_addr,
1991 -1, 0, XFS_QMOPT_DOWARN,
1992 "dquot_buf_recover");
1993 if (error)
1994 goto next;
1997 memcpy(xfs_buf_offset(bp,
1998 (uint)bit << XFS_BLI_SHIFT), /* dest */
1999 item->ri_buf[i].i_addr, /* source */
2000 nbits<<XFS_BLI_SHIFT); /* length */
2001 next:
2002 i++;
2003 bit += nbits;
2006 /* Shouldn't be any more regions */
2007 ASSERT(i == item->ri_total);
2011 * Do some primitive error checking on ondisk dquot data structures.
2014 xfs_qm_dqcheck(
2015 xfs_disk_dquot_t *ddq,
2016 xfs_dqid_t id,
2017 uint type, /* used only when IO_dorepair is true */
2018 uint flags,
2019 char *str)
2021 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
2022 int errs = 0;
2025 * We can encounter an uninitialized dquot buffer for 2 reasons:
2026 * 1. If we crash while deleting the quotainode(s), and those blks got
2027 * used for user data. This is because we take the path of regular
2028 * file deletion; however, the size field of quotainodes is never
2029 * updated, so all the tricks that we play in itruncate_finish
2030 * don't quite matter.
2032 * 2. We don't play the quota buffers when there's a quotaoff logitem.
2033 * But the allocation will be replayed so we'll end up with an
2034 * uninitialized quota block.
2036 * This is all fine; things are still consistent, and we haven't lost
2037 * any quota information. Just don't complain about bad dquot blks.
2039 if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) {
2040 if (flags & XFS_QMOPT_DOWARN)
2041 cmn_err(CE_ALERT,
2042 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
2043 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
2044 errs++;
2046 if (ddq->d_version != XFS_DQUOT_VERSION) {
2047 if (flags & XFS_QMOPT_DOWARN)
2048 cmn_err(CE_ALERT,
2049 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
2050 str, id, ddq->d_version, XFS_DQUOT_VERSION);
2051 errs++;
2054 if (ddq->d_flags != XFS_DQ_USER &&
2055 ddq->d_flags != XFS_DQ_PROJ &&
2056 ddq->d_flags != XFS_DQ_GROUP) {
2057 if (flags & XFS_QMOPT_DOWARN)
2058 cmn_err(CE_ALERT,
2059 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
2060 str, id, ddq->d_flags);
2061 errs++;
2064 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
2065 if (flags & XFS_QMOPT_DOWARN)
2066 cmn_err(CE_ALERT,
2067 "%s : ondisk-dquot 0x%p, ID mismatch: "
2068 "0x%x expected, found id 0x%x",
2069 str, ddq, id, be32_to_cpu(ddq->d_id));
2070 errs++;
2073 if (!errs && ddq->d_id) {
2074 if (ddq->d_blk_softlimit &&
2075 be64_to_cpu(ddq->d_bcount) >=
2076 be64_to_cpu(ddq->d_blk_softlimit)) {
2077 if (!ddq->d_btimer) {
2078 if (flags & XFS_QMOPT_DOWARN)
2079 cmn_err(CE_ALERT,
2080 "%s : Dquot ID 0x%x (0x%p) "
2081 "BLK TIMER NOT STARTED",
2082 str, (int)be32_to_cpu(ddq->d_id), ddq);
2083 errs++;
2086 if (ddq->d_ino_softlimit &&
2087 be64_to_cpu(ddq->d_icount) >=
2088 be64_to_cpu(ddq->d_ino_softlimit)) {
2089 if (!ddq->d_itimer) {
2090 if (flags & XFS_QMOPT_DOWARN)
2091 cmn_err(CE_ALERT,
2092 "%s : Dquot ID 0x%x (0x%p) "
2093 "INODE TIMER NOT STARTED",
2094 str, (int)be32_to_cpu(ddq->d_id), ddq);
2095 errs++;
2098 if (ddq->d_rtb_softlimit &&
2099 be64_to_cpu(ddq->d_rtbcount) >=
2100 be64_to_cpu(ddq->d_rtb_softlimit)) {
2101 if (!ddq->d_rtbtimer) {
2102 if (flags & XFS_QMOPT_DOWARN)
2103 cmn_err(CE_ALERT,
2104 "%s : Dquot ID 0x%x (0x%p) "
2105 "RTBLK TIMER NOT STARTED",
2106 str, (int)be32_to_cpu(ddq->d_id), ddq);
2107 errs++;
2112 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2113 return errs;
2115 if (flags & XFS_QMOPT_DOWARN)
2116 cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id);
2119 * Typically, a repair is only requested by quotacheck.
2121 ASSERT(id != -1);
2122 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2123 memset(d, 0, sizeof(xfs_dqblk_t));
2125 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2126 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2127 d->dd_diskdq.d_flags = type;
2128 d->dd_diskdq.d_id = cpu_to_be32(id);
2130 return errs;
2134 * Perform a dquot buffer recovery.
2135 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2136 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2137 * Else, treat it as a regular buffer and do recovery.
2139 STATIC void
2140 xlog_recover_do_dquot_buffer(
2141 xfs_mount_t *mp,
2142 xlog_t *log,
2143 xlog_recover_item_t *item,
2144 xfs_buf_t *bp,
2145 xfs_buf_log_format_t *buf_f)
2147 uint type;
2150 * Filesystems are required to send in quota flags at mount time.
2152 if (mp->m_qflags == 0) {
2153 return;
2156 type = 0;
2157 if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF)
2158 type |= XFS_DQ_USER;
2159 if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF)
2160 type |= XFS_DQ_PROJ;
2161 if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF)
2162 type |= XFS_DQ_GROUP;
2164 * This type of quotas was turned off, so ignore this buffer
2166 if (log->l_quotaoffs_flag & type)
2167 return;
2169 xlog_recover_do_reg_buffer(item, bp, buf_f);
2173 * This routine replays a modification made to a buffer at runtime.
2174 * There are actually two types of buffer, regular and inode, which
2175 * are handled differently. Inode buffers are handled differently
2176 * in that we only recover a specific set of data from them, namely
2177 * the inode di_next_unlinked fields. This is because all other inode
2178 * data is actually logged via inode records and any data we replay
2179 * here which overlaps that may be stale.
2181 * When meta-data buffers are freed at run time we log a buffer item
2182 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
2183 * of the buffer in the log should not be replayed at recovery time.
2184 * This is so that if the blocks covered by the buffer are reused for
2185 * file data before we crash we don't end up replaying old, freed
2186 * meta-data into a user's file.
2188 * To handle the cancellation of buffer log items, we make two passes
2189 * over the log during recovery. During the first we build a table of
2190 * those buffers which have been cancelled, and during the second we
2191 * only replay those buffers which do not have corresponding cancel
2192 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2193 * for more details on the implementation of the table of cancel records.
2195 STATIC int
2196 xlog_recover_do_buffer_trans(
2197 xlog_t *log,
2198 xlog_recover_item_t *item,
2199 int pass)
2201 xfs_buf_log_format_t *buf_f;
2202 xfs_mount_t *mp;
2203 xfs_buf_t *bp;
2204 int error;
2205 int cancel;
2206 xfs_daddr_t blkno;
2207 int len;
2208 ushort flags;
2210 buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr;
2212 if (pass == XLOG_RECOVER_PASS1) {
2214 * In this pass we're only looking for buf items
2215 * with the XFS_BLI_CANCEL bit set.
2217 xlog_recover_do_buffer_pass1(log, buf_f);
2218 return 0;
2219 } else {
2221 * In this pass we want to recover all the buffers
2222 * which have not been cancelled and are not
2223 * cancellation buffers themselves. The routine
2224 * we call here will tell us whether or not to
2225 * continue with the replay of this buffer.
2227 cancel = xlog_recover_do_buffer_pass2(log, buf_f);
2228 if (cancel) {
2229 return 0;
2232 switch (buf_f->blf_type) {
2233 case XFS_LI_BUF:
2234 blkno = buf_f->blf_blkno;
2235 len = buf_f->blf_len;
2236 flags = buf_f->blf_flags;
2237 break;
2238 default:
2239 xfs_fs_cmn_err(CE_ALERT, log->l_mp,
2240 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2241 buf_f->blf_type, log->l_mp->m_logname ?
2242 log->l_mp->m_logname : "internal");
2243 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2244 XFS_ERRLEVEL_LOW, log->l_mp);
2245 return XFS_ERROR(EFSCORRUPTED);
2248 mp = log->l_mp;
2249 if (flags & XFS_BLI_INODE_BUF) {
2250 bp = xfs_buf_read_flags(mp->m_ddev_targp, blkno, len,
2251 XFS_BUF_LOCK);
2252 } else {
2253 bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, 0);
2255 if (XFS_BUF_ISERROR(bp)) {
2256 xfs_ioerror_alert("xlog_recover_do..(read#1)", log->l_mp,
2257 bp, blkno);
2258 error = XFS_BUF_GETERROR(bp);
2259 xfs_buf_relse(bp);
2260 return error;
2263 error = 0;
2264 if (flags & XFS_BLI_INODE_BUF) {
2265 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2266 } else if (flags &
2267 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
2268 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2269 } else {
2270 xlog_recover_do_reg_buffer(item, bp, buf_f);
2272 if (error)
2273 return XFS_ERROR(error);
2276 * Perform delayed write on the buffer. Asynchronous writes will be
2277 * slower when taking into account all the buffers to be flushed.
2279 * Also make sure that only inode buffers with good sizes stay in
2280 * the buffer cache. The kernel moves inodes in buffers of 1 block
2281 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2282 * buffers in the log can be a different size if the log was generated
2283 * by an older kernel using unclustered inode buffers or a newer kernel
2284 * running with a different inode cluster size. Regardless, if the
2285 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2286 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2287 * the buffer out of the buffer cache so that the buffer won't
2288 * overlap with future reads of those inodes.
2290 if (XFS_DINODE_MAGIC ==
2291 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2292 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2293 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2294 XFS_BUF_STALE(bp);
2295 error = xfs_bwrite(mp, bp);
2296 } else {
2297 ASSERT(bp->b_mount == NULL || bp->b_mount == mp);
2298 bp->b_mount = mp;
2299 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2300 xfs_bdwrite(mp, bp);
2303 return (error);
2306 STATIC int
2307 xlog_recover_do_inode_trans(
2308 xlog_t *log,
2309 xlog_recover_item_t *item,
2310 int pass)
2312 xfs_inode_log_format_t *in_f;
2313 xfs_mount_t *mp;
2314 xfs_buf_t *bp;
2315 xfs_dinode_t *dip;
2316 xfs_ino_t ino;
2317 int len;
2318 xfs_caddr_t src;
2319 xfs_caddr_t dest;
2320 int error;
2321 int attr_index;
2322 uint fields;
2323 xfs_icdinode_t *dicp;
2324 int need_free = 0;
2326 if (pass == XLOG_RECOVER_PASS1) {
2327 return 0;
2330 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2331 in_f = (xfs_inode_log_format_t *)item->ri_buf[0].i_addr;
2332 } else {
2333 in_f = (xfs_inode_log_format_t *)kmem_alloc(
2334 sizeof(xfs_inode_log_format_t), KM_SLEEP);
2335 need_free = 1;
2336 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2337 if (error)
2338 goto error;
2340 ino = in_f->ilf_ino;
2341 mp = log->l_mp;
2344 * Inode buffers can be freed, look out for it,
2345 * and do not replay the inode.
2347 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2348 in_f->ilf_len, 0)) {
2349 error = 0;
2350 goto error;
2353 bp = xfs_buf_read_flags(mp->m_ddev_targp, in_f->ilf_blkno,
2354 in_f->ilf_len, XFS_BUF_LOCK);
2355 if (XFS_BUF_ISERROR(bp)) {
2356 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
2357 bp, in_f->ilf_blkno);
2358 error = XFS_BUF_GETERROR(bp);
2359 xfs_buf_relse(bp);
2360 goto error;
2362 error = 0;
2363 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2364 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
2367 * Make sure the place we're flushing out to really looks
2368 * like an inode!
2370 if (unlikely(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC)) {
2371 xfs_buf_relse(bp);
2372 xfs_fs_cmn_err(CE_ALERT, mp,
2373 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2374 dip, bp, ino);
2375 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2376 XFS_ERRLEVEL_LOW, mp);
2377 error = EFSCORRUPTED;
2378 goto error;
2380 dicp = (xfs_icdinode_t *)(item->ri_buf[1].i_addr);
2381 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2382 xfs_buf_relse(bp);
2383 xfs_fs_cmn_err(CE_ALERT, mp,
2384 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2385 item, ino);
2386 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2387 XFS_ERRLEVEL_LOW, mp);
2388 error = EFSCORRUPTED;
2389 goto error;
2392 /* Skip replay when the on disk inode is newer than the log one */
2393 if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
2395 * Deal with the wrap case, DI_MAX_FLUSH is less
2396 * than smaller numbers
2398 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
2399 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2400 /* do nothing */
2401 } else {
2402 xfs_buf_relse(bp);
2403 error = 0;
2404 goto error;
2407 /* Take the opportunity to reset the flush iteration count */
2408 dicp->di_flushiter = 0;
2410 if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) {
2411 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2412 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2413 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2414 XFS_ERRLEVEL_LOW, mp, dicp);
2415 xfs_buf_relse(bp);
2416 xfs_fs_cmn_err(CE_ALERT, mp,
2417 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2418 item, dip, bp, ino);
2419 error = EFSCORRUPTED;
2420 goto error;
2422 } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) {
2423 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2424 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2425 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2426 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2427 XFS_ERRLEVEL_LOW, mp, dicp);
2428 xfs_buf_relse(bp);
2429 xfs_fs_cmn_err(CE_ALERT, mp,
2430 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2431 item, dip, bp, ino);
2432 error = EFSCORRUPTED;
2433 goto error;
2436 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2437 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2438 XFS_ERRLEVEL_LOW, mp, dicp);
2439 xfs_buf_relse(bp);
2440 xfs_fs_cmn_err(CE_ALERT, mp,
2441 "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",
2442 item, dip, bp, ino,
2443 dicp->di_nextents + dicp->di_anextents,
2444 dicp->di_nblocks);
2445 error = EFSCORRUPTED;
2446 goto error;
2448 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2449 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2450 XFS_ERRLEVEL_LOW, mp, dicp);
2451 xfs_buf_relse(bp);
2452 xfs_fs_cmn_err(CE_ALERT, mp,
2453 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2454 item, dip, bp, ino, dicp->di_forkoff);
2455 error = EFSCORRUPTED;
2456 goto error;
2458 if (unlikely(item->ri_buf[1].i_len > sizeof(struct xfs_icdinode))) {
2459 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2460 XFS_ERRLEVEL_LOW, mp, dicp);
2461 xfs_buf_relse(bp);
2462 xfs_fs_cmn_err(CE_ALERT, mp,
2463 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2464 item->ri_buf[1].i_len, item);
2465 error = EFSCORRUPTED;
2466 goto error;
2469 /* The core is in in-core format */
2470 xfs_dinode_to_disk(dip, (xfs_icdinode_t *)item->ri_buf[1].i_addr);
2472 /* the rest is in on-disk format */
2473 if (item->ri_buf[1].i_len > sizeof(struct xfs_icdinode)) {
2474 memcpy((xfs_caddr_t) dip + sizeof(struct xfs_icdinode),
2475 item->ri_buf[1].i_addr + sizeof(struct xfs_icdinode),
2476 item->ri_buf[1].i_len - sizeof(struct xfs_icdinode));
2479 fields = in_f->ilf_fields;
2480 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2481 case XFS_ILOG_DEV:
2482 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
2483 break;
2484 case XFS_ILOG_UUID:
2485 memcpy(XFS_DFORK_DPTR(dip),
2486 &in_f->ilf_u.ilfu_uuid,
2487 sizeof(uuid_t));
2488 break;
2491 if (in_f->ilf_size == 2)
2492 goto write_inode_buffer;
2493 len = item->ri_buf[2].i_len;
2494 src = item->ri_buf[2].i_addr;
2495 ASSERT(in_f->ilf_size <= 4);
2496 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2497 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2498 (len == in_f->ilf_dsize));
2500 switch (fields & XFS_ILOG_DFORK) {
2501 case XFS_ILOG_DDATA:
2502 case XFS_ILOG_DEXT:
2503 memcpy(XFS_DFORK_DPTR(dip), src, len);
2504 break;
2506 case XFS_ILOG_DBROOT:
2507 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
2508 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
2509 XFS_DFORK_DSIZE(dip, mp));
2510 break;
2512 default:
2514 * There are no data fork flags set.
2516 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2517 break;
2521 * If we logged any attribute data, recover it. There may or
2522 * may not have been any other non-core data logged in this
2523 * transaction.
2525 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2526 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2527 attr_index = 3;
2528 } else {
2529 attr_index = 2;
2531 len = item->ri_buf[attr_index].i_len;
2532 src = item->ri_buf[attr_index].i_addr;
2533 ASSERT(len == in_f->ilf_asize);
2535 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2536 case XFS_ILOG_ADATA:
2537 case XFS_ILOG_AEXT:
2538 dest = XFS_DFORK_APTR(dip);
2539 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2540 memcpy(dest, src, len);
2541 break;
2543 case XFS_ILOG_ABROOT:
2544 dest = XFS_DFORK_APTR(dip);
2545 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2546 len, (xfs_bmdr_block_t*)dest,
2547 XFS_DFORK_ASIZE(dip, mp));
2548 break;
2550 default:
2551 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2552 ASSERT(0);
2553 xfs_buf_relse(bp);
2554 error = EIO;
2555 goto error;
2559 write_inode_buffer:
2560 ASSERT(bp->b_mount == NULL || bp->b_mount == mp);
2561 bp->b_mount = mp;
2562 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2563 xfs_bdwrite(mp, bp);
2564 error:
2565 if (need_free)
2566 kmem_free(in_f);
2567 return XFS_ERROR(error);
2571 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2572 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2573 * of that type.
2575 STATIC int
2576 xlog_recover_do_quotaoff_trans(
2577 xlog_t *log,
2578 xlog_recover_item_t *item,
2579 int pass)
2581 xfs_qoff_logformat_t *qoff_f;
2583 if (pass == XLOG_RECOVER_PASS2) {
2584 return (0);
2587 qoff_f = (xfs_qoff_logformat_t *)item->ri_buf[0].i_addr;
2588 ASSERT(qoff_f);
2591 * The logitem format's flag tells us if this was user quotaoff,
2592 * group/project quotaoff or both.
2594 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2595 log->l_quotaoffs_flag |= XFS_DQ_USER;
2596 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2597 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2598 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2599 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2601 return (0);
2605 * Recover a dquot record
2607 STATIC int
2608 xlog_recover_do_dquot_trans(
2609 xlog_t *log,
2610 xlog_recover_item_t *item,
2611 int pass)
2613 xfs_mount_t *mp;
2614 xfs_buf_t *bp;
2615 struct xfs_disk_dquot *ddq, *recddq;
2616 int error;
2617 xfs_dq_logformat_t *dq_f;
2618 uint type;
2620 if (pass == XLOG_RECOVER_PASS1) {
2621 return 0;
2623 mp = log->l_mp;
2626 * Filesystems are required to send in quota flags at mount time.
2628 if (mp->m_qflags == 0)
2629 return (0);
2631 recddq = (xfs_disk_dquot_t *)item->ri_buf[1].i_addr;
2633 if (item->ri_buf[1].i_addr == NULL) {
2634 cmn_err(CE_ALERT,
2635 "XFS: NULL dquot in %s.", __func__);
2636 return XFS_ERROR(EIO);
2638 if (item->ri_buf[1].i_len < sizeof(xfs_dqblk_t)) {
2639 cmn_err(CE_ALERT,
2640 "XFS: dquot too small (%d) in %s.",
2641 item->ri_buf[1].i_len, __func__);
2642 return XFS_ERROR(EIO);
2646 * This type of quotas was turned off, so ignore this record.
2648 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
2649 ASSERT(type);
2650 if (log->l_quotaoffs_flag & type)
2651 return (0);
2654 * At this point we know that quota was _not_ turned off.
2655 * Since the mount flags are not indicating to us otherwise, this
2656 * must mean that quota is on, and the dquot needs to be replayed.
2657 * Remember that we may not have fully recovered the superblock yet,
2658 * so we can't do the usual trick of looking at the SB quota bits.
2660 * The other possibility, of course, is that the quota subsystem was
2661 * removed since the last mount - ENOSYS.
2663 dq_f = (xfs_dq_logformat_t *)item->ri_buf[0].i_addr;
2664 ASSERT(dq_f);
2665 if ((error = xfs_qm_dqcheck(recddq,
2666 dq_f->qlf_id,
2667 0, XFS_QMOPT_DOWARN,
2668 "xlog_recover_do_dquot_trans (log copy)"))) {
2669 return XFS_ERROR(EIO);
2671 ASSERT(dq_f->qlf_len == 1);
2673 error = xfs_read_buf(mp, mp->m_ddev_targp,
2674 dq_f->qlf_blkno,
2675 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2676 0, &bp);
2677 if (error) {
2678 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
2679 bp, dq_f->qlf_blkno);
2680 return error;
2682 ASSERT(bp);
2683 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2686 * At least the magic num portion should be on disk because this
2687 * was among a chunk of dquots created earlier, and we did some
2688 * minimal initialization then.
2690 if (xfs_qm_dqcheck(ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2691 "xlog_recover_do_dquot_trans")) {
2692 xfs_buf_relse(bp);
2693 return XFS_ERROR(EIO);
2696 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2698 ASSERT(dq_f->qlf_size == 2);
2699 ASSERT(bp->b_mount == NULL || bp->b_mount == mp);
2700 bp->b_mount = mp;
2701 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2702 xfs_bdwrite(mp, bp);
2704 return (0);
2708 * This routine is called to create an in-core extent free intent
2709 * item from the efi format structure which was logged on disk.
2710 * It allocates an in-core efi, copies the extents from the format
2711 * structure into it, and adds the efi to the AIL with the given
2712 * LSN.
2714 STATIC int
2715 xlog_recover_do_efi_trans(
2716 xlog_t *log,
2717 xlog_recover_item_t *item,
2718 xfs_lsn_t lsn,
2719 int pass)
2721 int error;
2722 xfs_mount_t *mp;
2723 xfs_efi_log_item_t *efip;
2724 xfs_efi_log_format_t *efi_formatp;
2726 if (pass == XLOG_RECOVER_PASS1) {
2727 return 0;
2730 efi_formatp = (xfs_efi_log_format_t *)item->ri_buf[0].i_addr;
2732 mp = log->l_mp;
2733 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
2734 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2735 &(efip->efi_format)))) {
2736 xfs_efi_item_free(efip);
2737 return error;
2739 efip->efi_next_extent = efi_formatp->efi_nextents;
2740 efip->efi_flags |= XFS_EFI_COMMITTED;
2742 spin_lock(&log->l_ailp->xa_lock);
2744 * xfs_trans_ail_update() drops the AIL lock.
2746 xfs_trans_ail_update(log->l_ailp, (xfs_log_item_t *)efip, lsn);
2747 return 0;
2752 * This routine is called when an efd format structure is found in
2753 * a committed transaction in the log. It's purpose is to cancel
2754 * the corresponding efi if it was still in the log. To do this
2755 * it searches the AIL for the efi with an id equal to that in the
2756 * efd format structure. If we find it, we remove the efi from the
2757 * AIL and free it.
2759 STATIC void
2760 xlog_recover_do_efd_trans(
2761 xlog_t *log,
2762 xlog_recover_item_t *item,
2763 int pass)
2765 xfs_efd_log_format_t *efd_formatp;
2766 xfs_efi_log_item_t *efip = NULL;
2767 xfs_log_item_t *lip;
2768 __uint64_t efi_id;
2769 struct xfs_ail_cursor cur;
2770 struct xfs_ail *ailp = log->l_ailp;
2772 if (pass == XLOG_RECOVER_PASS1) {
2773 return;
2776 efd_formatp = (xfs_efd_log_format_t *)item->ri_buf[0].i_addr;
2777 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2778 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2779 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2780 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
2781 efi_id = efd_formatp->efd_efi_id;
2784 * Search for the efi with the id in the efd format structure
2785 * in the AIL.
2787 spin_lock(&ailp->xa_lock);
2788 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
2789 while (lip != NULL) {
2790 if (lip->li_type == XFS_LI_EFI) {
2791 efip = (xfs_efi_log_item_t *)lip;
2792 if (efip->efi_format.efi_id == efi_id) {
2794 * xfs_trans_ail_delete() drops the
2795 * AIL lock.
2797 xfs_trans_ail_delete(ailp, lip);
2798 xfs_efi_item_free(efip);
2799 spin_lock(&ailp->xa_lock);
2800 break;
2803 lip = xfs_trans_ail_cursor_next(ailp, &cur);
2805 xfs_trans_ail_cursor_done(ailp, &cur);
2806 spin_unlock(&ailp->xa_lock);
2810 * Perform the transaction
2812 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2813 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2815 STATIC int
2816 xlog_recover_do_trans(
2817 xlog_t *log,
2818 xlog_recover_t *trans,
2819 int pass)
2821 int error = 0;
2822 xlog_recover_item_t *item, *first_item;
2824 error = xlog_recover_reorder_trans(trans);
2825 if (error)
2826 return error;
2828 first_item = item = trans->r_itemq;
2829 do {
2830 switch (ITEM_TYPE(item)) {
2831 case XFS_LI_BUF:
2832 error = xlog_recover_do_buffer_trans(log, item, pass);
2833 break;
2834 case XFS_LI_INODE:
2835 error = xlog_recover_do_inode_trans(log, item, pass);
2836 break;
2837 case XFS_LI_EFI:
2838 error = xlog_recover_do_efi_trans(log, item,
2839 trans->r_lsn, pass);
2840 break;
2841 case XFS_LI_EFD:
2842 xlog_recover_do_efd_trans(log, item, pass);
2843 error = 0;
2844 break;
2845 case XFS_LI_DQUOT:
2846 error = xlog_recover_do_dquot_trans(log, item, pass);
2847 break;
2848 case XFS_LI_QUOTAOFF:
2849 error = xlog_recover_do_quotaoff_trans(log, item,
2850 pass);
2851 break;
2852 default:
2853 xlog_warn(
2854 "XFS: invalid item type (%d) xlog_recover_do_trans", ITEM_TYPE(item));
2855 ASSERT(0);
2856 error = XFS_ERROR(EIO);
2857 break;
2860 if (error)
2861 return error;
2862 item = item->ri_next;
2863 } while (first_item != item);
2865 return 0;
2869 * Free up any resources allocated by the transaction
2871 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2873 STATIC void
2874 xlog_recover_free_trans(
2875 xlog_recover_t *trans)
2877 xlog_recover_item_t *first_item, *item, *free_item;
2878 int i;
2880 item = first_item = trans->r_itemq;
2881 do {
2882 free_item = item;
2883 item = item->ri_next;
2884 /* Free the regions in the item. */
2885 for (i = 0; i < free_item->ri_cnt; i++) {
2886 kmem_free(free_item->ri_buf[i].i_addr);
2888 /* Free the item itself */
2889 kmem_free(free_item->ri_buf);
2890 kmem_free(free_item);
2891 } while (first_item != item);
2892 /* Free the transaction recover structure */
2893 kmem_free(trans);
2896 STATIC int
2897 xlog_recover_commit_trans(
2898 xlog_t *log,
2899 xlog_recover_t **q,
2900 xlog_recover_t *trans,
2901 int pass)
2903 int error;
2905 if ((error = xlog_recover_unlink_tid(q, trans)))
2906 return error;
2907 if ((error = xlog_recover_do_trans(log, trans, pass)))
2908 return error;
2909 xlog_recover_free_trans(trans); /* no error */
2910 return 0;
2913 STATIC int
2914 xlog_recover_unmount_trans(
2915 xlog_recover_t *trans)
2917 /* Do nothing now */
2918 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2919 return 0;
2923 * There are two valid states of the r_state field. 0 indicates that the
2924 * transaction structure is in a normal state. We have either seen the
2925 * start of the transaction or the last operation we added was not a partial
2926 * operation. If the last operation we added to the transaction was a
2927 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2929 * NOTE: skip LRs with 0 data length.
2931 STATIC int
2932 xlog_recover_process_data(
2933 xlog_t *log,
2934 xlog_recover_t *rhash[],
2935 xlog_rec_header_t *rhead,
2936 xfs_caddr_t dp,
2937 int pass)
2939 xfs_caddr_t lp;
2940 int num_logops;
2941 xlog_op_header_t *ohead;
2942 xlog_recover_t *trans;
2943 xlog_tid_t tid;
2944 int error;
2945 unsigned long hash;
2946 uint flags;
2948 lp = dp + be32_to_cpu(rhead->h_len);
2949 num_logops = be32_to_cpu(rhead->h_num_logops);
2951 /* check the log format matches our own - else we can't recover */
2952 if (xlog_header_check_recover(log->l_mp, rhead))
2953 return (XFS_ERROR(EIO));
2955 while ((dp < lp) && num_logops) {
2956 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2957 ohead = (xlog_op_header_t *)dp;
2958 dp += sizeof(xlog_op_header_t);
2959 if (ohead->oh_clientid != XFS_TRANSACTION &&
2960 ohead->oh_clientid != XFS_LOG) {
2961 xlog_warn(
2962 "XFS: xlog_recover_process_data: bad clientid");
2963 ASSERT(0);
2964 return (XFS_ERROR(EIO));
2966 tid = be32_to_cpu(ohead->oh_tid);
2967 hash = XLOG_RHASH(tid);
2968 trans = xlog_recover_find_tid(rhash[hash], tid);
2969 if (trans == NULL) { /* not found; add new tid */
2970 if (ohead->oh_flags & XLOG_START_TRANS)
2971 xlog_recover_new_tid(&rhash[hash], tid,
2972 be64_to_cpu(rhead->h_lsn));
2973 } else {
2974 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
2975 xlog_warn(
2976 "XFS: xlog_recover_process_data: bad length");
2977 WARN_ON(1);
2978 return (XFS_ERROR(EIO));
2980 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2981 if (flags & XLOG_WAS_CONT_TRANS)
2982 flags &= ~XLOG_CONTINUE_TRANS;
2983 switch (flags) {
2984 case XLOG_COMMIT_TRANS:
2985 error = xlog_recover_commit_trans(log,
2986 &rhash[hash], trans, pass);
2987 break;
2988 case XLOG_UNMOUNT_TRANS:
2989 error = xlog_recover_unmount_trans(trans);
2990 break;
2991 case XLOG_WAS_CONT_TRANS:
2992 error = xlog_recover_add_to_cont_trans(trans,
2993 dp, be32_to_cpu(ohead->oh_len));
2994 break;
2995 case XLOG_START_TRANS:
2996 xlog_warn(
2997 "XFS: xlog_recover_process_data: bad transaction");
2998 ASSERT(0);
2999 error = XFS_ERROR(EIO);
3000 break;
3001 case 0:
3002 case XLOG_CONTINUE_TRANS:
3003 error = xlog_recover_add_to_trans(trans,
3004 dp, be32_to_cpu(ohead->oh_len));
3005 break;
3006 default:
3007 xlog_warn(
3008 "XFS: xlog_recover_process_data: bad flag");
3009 ASSERT(0);
3010 error = XFS_ERROR(EIO);
3011 break;
3013 if (error)
3014 return error;
3016 dp += be32_to_cpu(ohead->oh_len);
3017 num_logops--;
3019 return 0;
3023 * Process an extent free intent item that was recovered from
3024 * the log. We need to free the extents that it describes.
3026 STATIC int
3027 xlog_recover_process_efi(
3028 xfs_mount_t *mp,
3029 xfs_efi_log_item_t *efip)
3031 xfs_efd_log_item_t *efdp;
3032 xfs_trans_t *tp;
3033 int i;
3034 int error = 0;
3035 xfs_extent_t *extp;
3036 xfs_fsblock_t startblock_fsb;
3038 ASSERT(!(efip->efi_flags & XFS_EFI_RECOVERED));
3041 * First check the validity of the extents described by the
3042 * EFI. If any are bad, then assume that all are bad and
3043 * just toss the EFI.
3045 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3046 extp = &(efip->efi_format.efi_extents[i]);
3047 startblock_fsb = XFS_BB_TO_FSB(mp,
3048 XFS_FSB_TO_DADDR(mp, extp->ext_start));
3049 if ((startblock_fsb == 0) ||
3050 (extp->ext_len == 0) ||
3051 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3052 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3054 * This will pull the EFI from the AIL and
3055 * free the memory associated with it.
3057 xfs_efi_release(efip, efip->efi_format.efi_nextents);
3058 return XFS_ERROR(EIO);
3062 tp = xfs_trans_alloc(mp, 0);
3063 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
3064 if (error)
3065 goto abort_error;
3066 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3068 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3069 extp = &(efip->efi_format.efi_extents[i]);
3070 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3071 if (error)
3072 goto abort_error;
3073 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3074 extp->ext_len);
3077 efip->efi_flags |= XFS_EFI_RECOVERED;
3078 error = xfs_trans_commit(tp, 0);
3079 return error;
3081 abort_error:
3082 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3083 return error;
3087 * When this is called, all of the EFIs which did not have
3088 * corresponding EFDs should be in the AIL. What we do now
3089 * is free the extents associated with each one.
3091 * Since we process the EFIs in normal transactions, they
3092 * will be removed at some point after the commit. This prevents
3093 * us from just walking down the list processing each one.
3094 * We'll use a flag in the EFI to skip those that we've already
3095 * processed and use the AIL iteration mechanism's generation
3096 * count to try to speed this up at least a bit.
3098 * When we start, we know that the EFIs are the only things in
3099 * the AIL. As we process them, however, other items are added
3100 * to the AIL. Since everything added to the AIL must come after
3101 * everything already in the AIL, we stop processing as soon as
3102 * we see something other than an EFI in the AIL.
3104 STATIC int
3105 xlog_recover_process_efis(
3106 xlog_t *log)
3108 xfs_log_item_t *lip;
3109 xfs_efi_log_item_t *efip;
3110 int error = 0;
3111 struct xfs_ail_cursor cur;
3112 struct xfs_ail *ailp;
3114 ailp = log->l_ailp;
3115 spin_lock(&ailp->xa_lock);
3116 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3117 while (lip != NULL) {
3119 * We're done when we see something other than an EFI.
3120 * There should be no EFIs left in the AIL now.
3122 if (lip->li_type != XFS_LI_EFI) {
3123 #ifdef DEBUG
3124 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
3125 ASSERT(lip->li_type != XFS_LI_EFI);
3126 #endif
3127 break;
3131 * Skip EFIs that we've already processed.
3133 efip = (xfs_efi_log_item_t *)lip;
3134 if (efip->efi_flags & XFS_EFI_RECOVERED) {
3135 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3136 continue;
3139 spin_unlock(&ailp->xa_lock);
3140 error = xlog_recover_process_efi(log->l_mp, efip);
3141 spin_lock(&ailp->xa_lock);
3142 if (error)
3143 goto out;
3144 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3146 out:
3147 xfs_trans_ail_cursor_done(ailp, &cur);
3148 spin_unlock(&ailp->xa_lock);
3149 return error;
3153 * This routine performs a transaction to null out a bad inode pointer
3154 * in an agi unlinked inode hash bucket.
3156 STATIC void
3157 xlog_recover_clear_agi_bucket(
3158 xfs_mount_t *mp,
3159 xfs_agnumber_t agno,
3160 int bucket)
3162 xfs_trans_t *tp;
3163 xfs_agi_t *agi;
3164 xfs_buf_t *agibp;
3165 int offset;
3166 int error;
3168 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3169 error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp),
3170 0, 0, 0);
3171 if (error)
3172 goto out_abort;
3174 error = xfs_read_agi(mp, tp, agno, &agibp);
3175 if (error)
3176 goto out_abort;
3178 agi = XFS_BUF_TO_AGI(agibp);
3179 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
3180 offset = offsetof(xfs_agi_t, agi_unlinked) +
3181 (sizeof(xfs_agino_t) * bucket);
3182 xfs_trans_log_buf(tp, agibp, offset,
3183 (offset + sizeof(xfs_agino_t) - 1));
3185 error = xfs_trans_commit(tp, 0);
3186 if (error)
3187 goto out_error;
3188 return;
3190 out_abort:
3191 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3192 out_error:
3193 xfs_fs_cmn_err(CE_WARN, mp, "xlog_recover_clear_agi_bucket: "
3194 "failed to clear agi %d. Continuing.", agno);
3195 return;
3198 STATIC xfs_agino_t
3199 xlog_recover_process_one_iunlink(
3200 struct xfs_mount *mp,
3201 xfs_agnumber_t agno,
3202 xfs_agino_t agino,
3203 int bucket)
3205 struct xfs_buf *ibp;
3206 struct xfs_dinode *dip;
3207 struct xfs_inode *ip;
3208 xfs_ino_t ino;
3209 int error;
3211 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3212 error = xfs_iget(mp, NULL, ino, 0, 0, &ip, 0);
3213 if (error)
3214 goto fail;
3217 * Get the on disk inode to find the next inode in the bucket.
3219 error = xfs_itobp(mp, NULL, ip, &dip, &ibp, XFS_BUF_LOCK);
3220 if (error)
3221 goto fail_iput;
3223 ASSERT(ip->i_d.di_nlink == 0);
3224 ASSERT(ip->i_d.di_mode != 0);
3226 /* setup for the next pass */
3227 agino = be32_to_cpu(dip->di_next_unlinked);
3228 xfs_buf_relse(ibp);
3231 * Prevent any DMAPI event from being sent when the reference on
3232 * the inode is dropped.
3234 ip->i_d.di_dmevmask = 0;
3236 IRELE(ip);
3237 return agino;
3239 fail_iput:
3240 IRELE(ip);
3241 fail:
3243 * We can't read in the inode this bucket points to, or this inode
3244 * is messed up. Just ditch this bucket of inodes. We will lose
3245 * some inodes and space, but at least we won't hang.
3247 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3248 * clear the inode pointer in the bucket.
3250 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3251 return NULLAGINO;
3255 * xlog_iunlink_recover
3257 * This is called during recovery to process any inodes which
3258 * we unlinked but not freed when the system crashed. These
3259 * inodes will be on the lists in the AGI blocks. What we do
3260 * here is scan all the AGIs and fully truncate and free any
3261 * inodes found on the lists. Each inode is removed from the
3262 * lists when it has been fully truncated and is freed. The
3263 * freeing of the inode and its removal from the list must be
3264 * atomic.
3266 void
3267 xlog_recover_process_iunlinks(
3268 xlog_t *log)
3270 xfs_mount_t *mp;
3271 xfs_agnumber_t agno;
3272 xfs_agi_t *agi;
3273 xfs_buf_t *agibp;
3274 xfs_agino_t agino;
3275 int bucket;
3276 int error;
3277 uint mp_dmevmask;
3279 mp = log->l_mp;
3282 * Prevent any DMAPI event from being sent while in this function.
3284 mp_dmevmask = mp->m_dmevmask;
3285 mp->m_dmevmask = 0;
3287 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3289 * Find the agi for this ag.
3291 error = xfs_read_agi(mp, NULL, agno, &agibp);
3292 if (error) {
3294 * AGI is b0rked. Don't process it.
3296 * We should probably mark the filesystem as corrupt
3297 * after we've recovered all the ag's we can....
3299 continue;
3301 agi = XFS_BUF_TO_AGI(agibp);
3303 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3304 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3305 while (agino != NULLAGINO) {
3307 * Release the agi buffer so that it can
3308 * be acquired in the normal course of the
3309 * transaction to truncate and free the inode.
3311 xfs_buf_relse(agibp);
3313 agino = xlog_recover_process_one_iunlink(mp,
3314 agno, agino, bucket);
3317 * Reacquire the agibuffer and continue around
3318 * the loop. This should never fail as we know
3319 * the buffer was good earlier on.
3321 error = xfs_read_agi(mp, NULL, agno, &agibp);
3322 ASSERT(error == 0);
3323 agi = XFS_BUF_TO_AGI(agibp);
3328 * Release the buffer for the current agi so we can
3329 * go on to the next one.
3331 xfs_buf_relse(agibp);
3334 mp->m_dmevmask = mp_dmevmask;
3338 #ifdef DEBUG
3339 STATIC void
3340 xlog_pack_data_checksum(
3341 xlog_t *log,
3342 xlog_in_core_t *iclog,
3343 int size)
3345 int i;
3346 __be32 *up;
3347 uint chksum = 0;
3349 up = (__be32 *)iclog->ic_datap;
3350 /* divide length by 4 to get # words */
3351 for (i = 0; i < (size >> 2); i++) {
3352 chksum ^= be32_to_cpu(*up);
3353 up++;
3355 iclog->ic_header.h_chksum = cpu_to_be32(chksum);
3357 #else
3358 #define xlog_pack_data_checksum(log, iclog, size)
3359 #endif
3362 * Stamp cycle number in every block
3364 void
3365 xlog_pack_data(
3366 xlog_t *log,
3367 xlog_in_core_t *iclog,
3368 int roundoff)
3370 int i, j, k;
3371 int size = iclog->ic_offset + roundoff;
3372 __be32 cycle_lsn;
3373 xfs_caddr_t dp;
3375 xlog_pack_data_checksum(log, iclog, size);
3377 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3379 dp = iclog->ic_datap;
3380 for (i = 0; i < BTOBB(size) &&
3381 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3382 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
3383 *(__be32 *)dp = cycle_lsn;
3384 dp += BBSIZE;
3387 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3388 xlog_in_core_2_t *xhdr = iclog->ic_data;
3390 for ( ; i < BTOBB(size); i++) {
3391 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3392 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3393 xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
3394 *(__be32 *)dp = cycle_lsn;
3395 dp += BBSIZE;
3398 for (i = 1; i < log->l_iclog_heads; i++) {
3399 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3404 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
3405 STATIC void
3406 xlog_unpack_data_checksum(
3407 xlog_rec_header_t *rhead,
3408 xfs_caddr_t dp,
3409 xlog_t *log)
3411 __be32 *up = (__be32 *)dp;
3412 uint chksum = 0;
3413 int i;
3415 /* divide length by 4 to get # words */
3416 for (i=0; i < be32_to_cpu(rhead->h_len) >> 2; i++) {
3417 chksum ^= be32_to_cpu(*up);
3418 up++;
3420 if (chksum != be32_to_cpu(rhead->h_chksum)) {
3421 if (rhead->h_chksum ||
3422 ((log->l_flags & XLOG_CHKSUM_MISMATCH) == 0)) {
3423 cmn_err(CE_DEBUG,
3424 "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n",
3425 be32_to_cpu(rhead->h_chksum), chksum);
3426 cmn_err(CE_DEBUG,
3427 "XFS: Disregard message if filesystem was created with non-DEBUG kernel");
3428 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3429 cmn_err(CE_DEBUG,
3430 "XFS: LogR this is a LogV2 filesystem\n");
3432 log->l_flags |= XLOG_CHKSUM_MISMATCH;
3436 #else
3437 #define xlog_unpack_data_checksum(rhead, dp, log)
3438 #endif
3440 STATIC void
3441 xlog_unpack_data(
3442 xlog_rec_header_t *rhead,
3443 xfs_caddr_t dp,
3444 xlog_t *log)
3446 int i, j, k;
3448 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
3449 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3450 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
3451 dp += BBSIZE;
3454 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3455 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
3456 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
3457 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3458 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3459 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
3460 dp += BBSIZE;
3464 xlog_unpack_data_checksum(rhead, dp, log);
3467 STATIC int
3468 xlog_valid_rec_header(
3469 xlog_t *log,
3470 xlog_rec_header_t *rhead,
3471 xfs_daddr_t blkno)
3473 int hlen;
3475 if (unlikely(be32_to_cpu(rhead->h_magicno) != XLOG_HEADER_MAGIC_NUM)) {
3476 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3477 XFS_ERRLEVEL_LOW, log->l_mp);
3478 return XFS_ERROR(EFSCORRUPTED);
3480 if (unlikely(
3481 (!rhead->h_version ||
3482 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
3483 xlog_warn("XFS: %s: unrecognised log version (%d).",
3484 __func__, be32_to_cpu(rhead->h_version));
3485 return XFS_ERROR(EIO);
3488 /* LR body must have data or it wouldn't have been written */
3489 hlen = be32_to_cpu(rhead->h_len);
3490 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3491 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3492 XFS_ERRLEVEL_LOW, log->l_mp);
3493 return XFS_ERROR(EFSCORRUPTED);
3495 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3496 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3497 XFS_ERRLEVEL_LOW, log->l_mp);
3498 return XFS_ERROR(EFSCORRUPTED);
3500 return 0;
3504 * Read the log from tail to head and process the log records found.
3505 * Handle the two cases where the tail and head are in the same cycle
3506 * and where the active portion of the log wraps around the end of
3507 * the physical log separately. The pass parameter is passed through
3508 * to the routines called to process the data and is not looked at
3509 * here.
3511 STATIC int
3512 xlog_do_recovery_pass(
3513 xlog_t *log,
3514 xfs_daddr_t head_blk,
3515 xfs_daddr_t tail_blk,
3516 int pass)
3518 xlog_rec_header_t *rhead;
3519 xfs_daddr_t blk_no;
3520 xfs_caddr_t bufaddr, offset;
3521 xfs_buf_t *hbp, *dbp;
3522 int error = 0, h_size;
3523 int bblks, split_bblks;
3524 int hblks, split_hblks, wrapped_hblks;
3525 xlog_recover_t *rhash[XLOG_RHASH_SIZE];
3527 ASSERT(head_blk != tail_blk);
3530 * Read the header of the tail block and get the iclog buffer size from
3531 * h_size. Use this to tell how many sectors make up the log header.
3533 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3535 * When using variable length iclogs, read first sector of
3536 * iclog header and extract the header size from it. Get a
3537 * new hbp that is the correct size.
3539 hbp = xlog_get_bp(log, 1);
3540 if (!hbp)
3541 return ENOMEM;
3543 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
3544 if (error)
3545 goto bread_err1;
3547 rhead = (xlog_rec_header_t *)offset;
3548 error = xlog_valid_rec_header(log, rhead, tail_blk);
3549 if (error)
3550 goto bread_err1;
3551 h_size = be32_to_cpu(rhead->h_size);
3552 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
3553 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3554 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3555 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3556 hblks++;
3557 xlog_put_bp(hbp);
3558 hbp = xlog_get_bp(log, hblks);
3559 } else {
3560 hblks = 1;
3562 } else {
3563 ASSERT(log->l_sectbb_log == 0);
3564 hblks = 1;
3565 hbp = xlog_get_bp(log, 1);
3566 h_size = XLOG_BIG_RECORD_BSIZE;
3569 if (!hbp)
3570 return ENOMEM;
3571 dbp = xlog_get_bp(log, BTOBB(h_size));
3572 if (!dbp) {
3573 xlog_put_bp(hbp);
3574 return ENOMEM;
3577 memset(rhash, 0, sizeof(rhash));
3578 if (tail_blk <= head_blk) {
3579 for (blk_no = tail_blk; blk_no < head_blk; ) {
3580 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3581 if (error)
3582 goto bread_err2;
3584 rhead = (xlog_rec_header_t *)offset;
3585 error = xlog_valid_rec_header(log, rhead, blk_no);
3586 if (error)
3587 goto bread_err2;
3589 /* blocks in data section */
3590 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3591 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
3592 &offset);
3593 if (error)
3594 goto bread_err2;
3596 xlog_unpack_data(rhead, offset, log);
3597 if ((error = xlog_recover_process_data(log,
3598 rhash, rhead, offset, pass)))
3599 goto bread_err2;
3600 blk_no += bblks + hblks;
3602 } else {
3604 * Perform recovery around the end of the physical log.
3605 * When the head is not on the same cycle number as the tail,
3606 * we can't do a sequential recovery as above.
3608 blk_no = tail_blk;
3609 while (blk_no < log->l_logBBsize) {
3611 * Check for header wrapping around physical end-of-log
3613 offset = NULL;
3614 split_hblks = 0;
3615 wrapped_hblks = 0;
3616 if (blk_no + hblks <= log->l_logBBsize) {
3617 /* Read header in one read */
3618 error = xlog_bread(log, blk_no, hblks, hbp,
3619 &offset);
3620 if (error)
3621 goto bread_err2;
3622 } else {
3623 /* This LR is split across physical log end */
3624 if (blk_no != log->l_logBBsize) {
3625 /* some data before physical log end */
3626 ASSERT(blk_no <= INT_MAX);
3627 split_hblks = log->l_logBBsize - (int)blk_no;
3628 ASSERT(split_hblks > 0);
3629 error = xlog_bread(log, blk_no,
3630 split_hblks, hbp,
3631 &offset);
3632 if (error)
3633 goto bread_err2;
3637 * Note: this black magic still works with
3638 * large sector sizes (non-512) only because:
3639 * - we increased the buffer size originally
3640 * by 1 sector giving us enough extra space
3641 * for the second read;
3642 * - the log start is guaranteed to be sector
3643 * aligned;
3644 * - we read the log end (LR header start)
3645 * _first_, then the log start (LR header end)
3646 * - order is important.
3648 wrapped_hblks = hblks - split_hblks;
3649 bufaddr = XFS_BUF_PTR(hbp);
3650 error = XFS_BUF_SET_PTR(hbp,
3651 bufaddr + BBTOB(split_hblks),
3652 BBTOB(hblks - split_hblks));
3653 if (error)
3654 goto bread_err2;
3656 error = xlog_bread_noalign(log, 0,
3657 wrapped_hblks, hbp);
3658 if (error)
3659 goto bread_err2;
3661 error = XFS_BUF_SET_PTR(hbp, bufaddr,
3662 BBTOB(hblks));
3663 if (error)
3664 goto bread_err2;
3666 if (!offset)
3667 offset = xlog_align(log, 0,
3668 wrapped_hblks, hbp);
3670 rhead = (xlog_rec_header_t *)offset;
3671 error = xlog_valid_rec_header(log, rhead,
3672 split_hblks ? blk_no : 0);
3673 if (error)
3674 goto bread_err2;
3676 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3677 blk_no += hblks;
3679 /* Read in data for log record */
3680 if (blk_no + bblks <= log->l_logBBsize) {
3681 error = xlog_bread(log, blk_no, bblks, dbp,
3682 &offset);
3683 if (error)
3684 goto bread_err2;
3685 } else {
3686 /* This log record is split across the
3687 * physical end of log */
3688 offset = NULL;
3689 split_bblks = 0;
3690 if (blk_no != log->l_logBBsize) {
3691 /* some data is before the physical
3692 * end of log */
3693 ASSERT(!wrapped_hblks);
3694 ASSERT(blk_no <= INT_MAX);
3695 split_bblks =
3696 log->l_logBBsize - (int)blk_no;
3697 ASSERT(split_bblks > 0);
3698 error = xlog_bread(log, blk_no,
3699 split_bblks, dbp,
3700 &offset);
3701 if (error)
3702 goto bread_err2;
3706 * Note: this black magic still works with
3707 * large sector sizes (non-512) only because:
3708 * - we increased the buffer size originally
3709 * by 1 sector giving us enough extra space
3710 * for the second read;
3711 * - the log start is guaranteed to be sector
3712 * aligned;
3713 * - we read the log end (LR header start)
3714 * _first_, then the log start (LR header end)
3715 * - order is important.
3717 bufaddr = XFS_BUF_PTR(dbp);
3718 error = XFS_BUF_SET_PTR(dbp,
3719 bufaddr + BBTOB(split_bblks),
3720 BBTOB(bblks - split_bblks));
3721 if (error)
3722 goto bread_err2;
3724 error = xlog_bread_noalign(log, wrapped_hblks,
3725 bblks - split_bblks,
3726 dbp);
3727 if (error)
3728 goto bread_err2;
3730 error = XFS_BUF_SET_PTR(dbp, bufaddr, h_size);
3731 if (error)
3732 goto bread_err2;
3734 if (!offset)
3735 offset = xlog_align(log, wrapped_hblks,
3736 bblks - split_bblks, dbp);
3738 xlog_unpack_data(rhead, offset, log);
3739 if ((error = xlog_recover_process_data(log, rhash,
3740 rhead, offset, pass)))
3741 goto bread_err2;
3742 blk_no += bblks;
3745 ASSERT(blk_no >= log->l_logBBsize);
3746 blk_no -= log->l_logBBsize;
3748 /* read first part of physical log */
3749 while (blk_no < head_blk) {
3750 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3751 if (error)
3752 goto bread_err2;
3754 rhead = (xlog_rec_header_t *)offset;
3755 error = xlog_valid_rec_header(log, rhead, blk_no);
3756 if (error)
3757 goto bread_err2;
3759 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3760 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
3761 &offset);
3762 if (error)
3763 goto bread_err2;
3765 xlog_unpack_data(rhead, offset, log);
3766 if ((error = xlog_recover_process_data(log, rhash,
3767 rhead, offset, pass)))
3768 goto bread_err2;
3769 blk_no += bblks + hblks;
3773 bread_err2:
3774 xlog_put_bp(dbp);
3775 bread_err1:
3776 xlog_put_bp(hbp);
3777 return error;
3781 * Do the recovery of the log. We actually do this in two phases.
3782 * The two passes are necessary in order to implement the function
3783 * of cancelling a record written into the log. The first pass
3784 * determines those things which have been cancelled, and the
3785 * second pass replays log items normally except for those which
3786 * have been cancelled. The handling of the replay and cancellations
3787 * takes place in the log item type specific routines.
3789 * The table of items which have cancel records in the log is allocated
3790 * and freed at this level, since only here do we know when all of
3791 * the log recovery has been completed.
3793 STATIC int
3794 xlog_do_log_recovery(
3795 xlog_t *log,
3796 xfs_daddr_t head_blk,
3797 xfs_daddr_t tail_blk)
3799 int error;
3801 ASSERT(head_blk != tail_blk);
3804 * First do a pass to find all of the cancelled buf log items.
3805 * Store them in the buf_cancel_table for use in the second pass.
3807 log->l_buf_cancel_table =
3808 (xfs_buf_cancel_t **)kmem_zalloc(XLOG_BC_TABLE_SIZE *
3809 sizeof(xfs_buf_cancel_t*),
3810 KM_SLEEP);
3811 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3812 XLOG_RECOVER_PASS1);
3813 if (error != 0) {
3814 kmem_free(log->l_buf_cancel_table);
3815 log->l_buf_cancel_table = NULL;
3816 return error;
3819 * Then do a second pass to actually recover the items in the log.
3820 * When it is complete free the table of buf cancel items.
3822 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3823 XLOG_RECOVER_PASS2);
3824 #ifdef DEBUG
3825 if (!error) {
3826 int i;
3828 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3829 ASSERT(log->l_buf_cancel_table[i] == NULL);
3831 #endif /* DEBUG */
3833 kmem_free(log->l_buf_cancel_table);
3834 log->l_buf_cancel_table = NULL;
3836 return error;
3840 * Do the actual recovery
3842 STATIC int
3843 xlog_do_recover(
3844 xlog_t *log,
3845 xfs_daddr_t head_blk,
3846 xfs_daddr_t tail_blk)
3848 int error;
3849 xfs_buf_t *bp;
3850 xfs_sb_t *sbp;
3853 * First replay the images in the log.
3855 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3856 if (error) {
3857 return error;
3860 XFS_bflush(log->l_mp->m_ddev_targp);
3863 * If IO errors happened during recovery, bail out.
3865 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3866 return (EIO);
3870 * We now update the tail_lsn since much of the recovery has completed
3871 * and there may be space available to use. If there were no extent
3872 * or iunlinks, we can free up the entire log and set the tail_lsn to
3873 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3874 * lsn of the last known good LR on disk. If there are extent frees
3875 * or iunlinks they will have some entries in the AIL; so we look at
3876 * the AIL to determine how to set the tail_lsn.
3878 xlog_assign_tail_lsn(log->l_mp);
3881 * Now that we've finished replaying all buffer and inode
3882 * updates, re-read in the superblock.
3884 bp = xfs_getsb(log->l_mp, 0);
3885 XFS_BUF_UNDONE(bp);
3886 ASSERT(!(XFS_BUF_ISWRITE(bp)));
3887 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
3888 XFS_BUF_READ(bp);
3889 XFS_BUF_UNASYNC(bp);
3890 xfsbdstrat(log->l_mp, bp);
3891 error = xfs_iowait(bp);
3892 if (error) {
3893 xfs_ioerror_alert("xlog_do_recover",
3894 log->l_mp, bp, XFS_BUF_ADDR(bp));
3895 ASSERT(0);
3896 xfs_buf_relse(bp);
3897 return error;
3900 /* Convert superblock from on-disk format */
3901 sbp = &log->l_mp->m_sb;
3902 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
3903 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
3904 ASSERT(xfs_sb_good_version(sbp));
3905 xfs_buf_relse(bp);
3907 /* We've re-read the superblock so re-initialize per-cpu counters */
3908 xfs_icsb_reinit_counters(log->l_mp);
3910 xlog_recover_check_summary(log);
3912 /* Normal transactions can now occur */
3913 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3914 return 0;
3918 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3920 * Return error or zero.
3923 xlog_recover(
3924 xlog_t *log)
3926 xfs_daddr_t head_blk, tail_blk;
3927 int error;
3929 /* find the tail of the log */
3930 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
3931 return error;
3933 if (tail_blk != head_blk) {
3934 /* There used to be a comment here:
3936 * disallow recovery on read-only mounts. note -- mount
3937 * checks for ENOSPC and turns it into an intelligent
3938 * error message.
3939 * ...but this is no longer true. Now, unless you specify
3940 * NORECOVERY (in which case this function would never be
3941 * called), we just go ahead and recover. We do this all
3942 * under the vfs layer, so we can get away with it unless
3943 * the device itself is read-only, in which case we fail.
3945 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
3946 return error;
3949 cmn_err(CE_NOTE,
3950 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3951 log->l_mp->m_fsname, log->l_mp->m_logname ?
3952 log->l_mp->m_logname : "internal");
3954 error = xlog_do_recover(log, head_blk, tail_blk);
3955 log->l_flags |= XLOG_RECOVERY_NEEDED;
3957 return error;
3961 * In the first part of recovery we replay inodes and buffers and build
3962 * up the list of extent free items which need to be processed. Here
3963 * we process the extent free items and clean up the on disk unlinked
3964 * inode lists. This is separated from the first part of recovery so
3965 * that the root and real-time bitmap inodes can be read in from disk in
3966 * between the two stages. This is necessary so that we can free space
3967 * in the real-time portion of the file system.
3970 xlog_recover_finish(
3971 xlog_t *log)
3974 * Now we're ready to do the transactions needed for the
3975 * rest of recovery. Start with completing all the extent
3976 * free intent records and then process the unlinked inode
3977 * lists. At this point, we essentially run in normal mode
3978 * except that we're still performing recovery actions
3979 * rather than accepting new requests.
3981 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
3982 int error;
3983 error = xlog_recover_process_efis(log);
3984 if (error) {
3985 cmn_err(CE_ALERT,
3986 "Failed to recover EFIs on filesystem: %s",
3987 log->l_mp->m_fsname);
3988 return error;
3991 * Sync the log to get all the EFIs out of the AIL.
3992 * This isn't absolutely necessary, but it helps in
3993 * case the unlink transactions would have problems
3994 * pushing the EFIs out of the way.
3996 xfs_log_force(log->l_mp, (xfs_lsn_t)0,
3997 (XFS_LOG_FORCE | XFS_LOG_SYNC));
3999 xlog_recover_process_iunlinks(log);
4001 xlog_recover_check_summary(log);
4003 cmn_err(CE_NOTE,
4004 "Ending XFS recovery on filesystem: %s (logdev: %s)",
4005 log->l_mp->m_fsname, log->l_mp->m_logname ?
4006 log->l_mp->m_logname : "internal");
4007 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
4008 } else {
4009 cmn_err(CE_DEBUG,
4010 "!Ending clean XFS mount for filesystem: %s\n",
4011 log->l_mp->m_fsname);
4013 return 0;
4017 #if defined(DEBUG)
4019 * Read all of the agf and agi counters and check that they
4020 * are consistent with the superblock counters.
4022 void
4023 xlog_recover_check_summary(
4024 xlog_t *log)
4026 xfs_mount_t *mp;
4027 xfs_agf_t *agfp;
4028 xfs_buf_t *agfbp;
4029 xfs_buf_t *agibp;
4030 xfs_buf_t *sbbp;
4031 #ifdef XFS_LOUD_RECOVERY
4032 xfs_sb_t *sbp;
4033 #endif
4034 xfs_agnumber_t agno;
4035 __uint64_t freeblks;
4036 __uint64_t itotal;
4037 __uint64_t ifree;
4038 int error;
4040 mp = log->l_mp;
4042 freeblks = 0LL;
4043 itotal = 0LL;
4044 ifree = 0LL;
4045 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
4046 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
4047 if (error) {
4048 xfs_fs_cmn_err(CE_ALERT, mp,
4049 "xlog_recover_check_summary(agf)"
4050 "agf read failed agno %d error %d",
4051 agno, error);
4052 } else {
4053 agfp = XFS_BUF_TO_AGF(agfbp);
4054 freeblks += be32_to_cpu(agfp->agf_freeblks) +
4055 be32_to_cpu(agfp->agf_flcount);
4056 xfs_buf_relse(agfbp);
4059 error = xfs_read_agi(mp, NULL, agno, &agibp);
4060 if (!error) {
4061 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
4063 itotal += be32_to_cpu(agi->agi_count);
4064 ifree += be32_to_cpu(agi->agi_freecount);
4065 xfs_buf_relse(agibp);
4069 sbbp = xfs_getsb(mp, 0);
4070 #ifdef XFS_LOUD_RECOVERY
4071 sbp = &mp->m_sb;
4072 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(sbbp));
4073 cmn_err(CE_NOTE,
4074 "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
4075 sbp->sb_icount, itotal);
4076 cmn_err(CE_NOTE,
4077 "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
4078 sbp->sb_ifree, ifree);
4079 cmn_err(CE_NOTE,
4080 "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
4081 sbp->sb_fdblocks, freeblks);
4082 #if 0
4084 * This is turned off until I account for the allocation
4085 * btree blocks which live in free space.
4087 ASSERT(sbp->sb_icount == itotal);
4088 ASSERT(sbp->sb_ifree == ifree);
4089 ASSERT(sbp->sb_fdblocks == freeblks);
4090 #endif
4091 #endif
4092 xfs_buf_relse(sbbp);
4094 #endif /* DEBUG */