blob: 3cb678e3a132997804998f7dbfea787b4a75a15a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Tim Shimmin87c199c2006-06-09 14:56:16 +10002 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
Nathan Scott7b718762005-11-02 14:58:39 +11003 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
Nathan Scotta844f452005-11-02 14:38:42 +110025#include "xfs_sb.h"
26#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#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"
Nathan Scotta844f452005-11-02 14:38:42 +110032#include "xfs_alloc_btree.h"
33#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_dir2_sf.h"
Nathan Scotta844f452005-11-02 14:38:42 +110035#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110038#include "xfs_inode_item.h"
39#include "xfs_imap.h"
40#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include "xfs_ialloc.h"
42#include "xfs_log_priv.h"
43#include "xfs_buf_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include "xfs_log_recover.h"
45#include "xfs_extfree_item.h"
46#include "xfs_trans_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include "xfs_quota.h"
48#include "xfs_rw.h"
49
50STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *);
51STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t);
52STATIC void xlog_recover_insert_item_backq(xlog_recover_item_t **q,
53 xlog_recover_item_t *item);
54#if defined(DEBUG)
55STATIC void xlog_recover_check_summary(xlog_t *);
56STATIC void xlog_recover_check_ail(xfs_mount_t *, xfs_log_item_t *, int);
57#else
58#define xlog_recover_check_summary(log)
59#define xlog_recover_check_ail(mp, lip, gen)
60#endif
61
62
63/*
64 * Sector aligned buffer routines for buffer create/read/write/access
65 */
66
67#define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \
68 ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \
69 ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) )
70#define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask)
71
72xfs_buf_t *
73xlog_get_bp(
74 xlog_t *log,
75 int num_bblks)
76{
77 ASSERT(num_bblks > 0);
78
79 if (log->l_sectbb_log) {
80 if (num_bblks > 1)
81 num_bblks += XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
82 num_bblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, num_bblks);
83 }
84 return xfs_buf_get_noaddr(BBTOB(num_bblks), log->l_mp->m_logdev_targp);
85}
86
87void
88xlog_put_bp(
89 xfs_buf_t *bp)
90{
91 xfs_buf_free(bp);
92}
93
94
95/*
96 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
97 */
98int
99xlog_bread(
100 xlog_t *log,
101 xfs_daddr_t blk_no,
102 int nbblks,
103 xfs_buf_t *bp)
104{
105 int error;
106
107 if (log->l_sectbb_log) {
108 blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
109 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
110 }
111
112 ASSERT(nbblks > 0);
113 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
114 ASSERT(bp);
115
116 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
117 XFS_BUF_READ(bp);
118 XFS_BUF_BUSY(bp);
119 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
120 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
121
122 xfsbdstrat(log->l_mp, bp);
123 if ((error = xfs_iowait(bp)))
124 xfs_ioerror_alert("xlog_bread", log->l_mp,
125 bp, XFS_BUF_ADDR(bp));
126 return error;
127}
128
129/*
130 * Write out the buffer at the given block for the given number of blocks.
131 * The buffer is kept locked across the write and is returned locked.
132 * This can only be used for synchronous log writes.
133 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000134STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135xlog_bwrite(
136 xlog_t *log,
137 xfs_daddr_t blk_no,
138 int nbblks,
139 xfs_buf_t *bp)
140{
141 int error;
142
143 if (log->l_sectbb_log) {
144 blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
145 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
146 }
147
148 ASSERT(nbblks > 0);
149 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
150
151 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
152 XFS_BUF_ZEROFLAGS(bp);
153 XFS_BUF_BUSY(bp);
154 XFS_BUF_HOLD(bp);
155 XFS_BUF_PSEMA(bp, PRIBIO);
156 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
157 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
158
159 if ((error = xfs_bwrite(log->l_mp, bp)))
160 xfs_ioerror_alert("xlog_bwrite", log->l_mp,
161 bp, XFS_BUF_ADDR(bp));
162 return error;
163}
164
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000165STATIC xfs_caddr_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166xlog_align(
167 xlog_t *log,
168 xfs_daddr_t blk_no,
169 int nbblks,
170 xfs_buf_t *bp)
171{
172 xfs_caddr_t ptr;
173
174 if (!log->l_sectbb_log)
175 return XFS_BUF_PTR(bp);
176
177 ptr = XFS_BUF_PTR(bp) + BBTOB((int)blk_no & log->l_sectbb_mask);
178 ASSERT(XFS_BUF_SIZE(bp) >=
179 BBTOB(nbblks + (blk_no & log->l_sectbb_mask)));
180 return ptr;
181}
182
183#ifdef DEBUG
184/*
185 * dump debug superblock and log record information
186 */
187STATIC void
188xlog_header_check_dump(
189 xfs_mount_t *mp,
190 xlog_rec_header_t *head)
191{
192 int b;
193
Nathan Scottb6574522006-06-09 15:29:40 +1000194 cmn_err(CE_DEBUG, "%s: SB : uuid = ", __FUNCTION__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 for (b = 0; b < 16; b++)
Nathan Scottb6574522006-06-09 15:29:40 +1000196 cmn_err(CE_DEBUG, "%02x", ((uchar_t *)&mp->m_sb.sb_uuid)[b]);
197 cmn_err(CE_DEBUG, ", fmt = %d\n", XLOG_FMT);
198 cmn_err(CE_DEBUG, " log : uuid = ");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 for (b = 0; b < 16; b++)
Nathan Scottb6574522006-06-09 15:29:40 +1000200 cmn_err(CE_DEBUG, "%02x",((uchar_t *)&head->h_fs_uuid)[b]);
201 cmn_err(CE_DEBUG, ", fmt = %d\n", INT_GET(head->h_fmt, ARCH_CONVERT));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202}
203#else
204#define xlog_header_check_dump(mp, head)
205#endif
206
207/*
208 * check log record header for recovery
209 */
210STATIC int
211xlog_header_check_recover(
212 xfs_mount_t *mp,
213 xlog_rec_header_t *head)
214{
215 ASSERT(INT_GET(head->h_magicno, ARCH_CONVERT) == XLOG_HEADER_MAGIC_NUM);
216
217 /*
218 * IRIX doesn't write the h_fmt field and leaves it zeroed
219 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
220 * a dirty log created in IRIX.
221 */
222 if (unlikely(INT_GET(head->h_fmt, ARCH_CONVERT) != XLOG_FMT)) {
223 xlog_warn(
224 "XFS: dirty log written in incompatible format - can't recover");
225 xlog_header_check_dump(mp, head);
226 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
227 XFS_ERRLEVEL_HIGH, mp);
228 return XFS_ERROR(EFSCORRUPTED);
229 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
230 xlog_warn(
231 "XFS: dirty log entry has mismatched uuid - can't recover");
232 xlog_header_check_dump(mp, head);
233 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
234 XFS_ERRLEVEL_HIGH, mp);
235 return XFS_ERROR(EFSCORRUPTED);
236 }
237 return 0;
238}
239
240/*
241 * read the head block of the log and check the header
242 */
243STATIC int
244xlog_header_check_mount(
245 xfs_mount_t *mp,
246 xlog_rec_header_t *head)
247{
248 ASSERT(INT_GET(head->h_magicno, ARCH_CONVERT) == XLOG_HEADER_MAGIC_NUM);
249
250 if (uuid_is_nil(&head->h_fs_uuid)) {
251 /*
252 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
253 * h_fs_uuid is nil, we assume this log was last mounted
254 * by IRIX and continue.
255 */
256 xlog_warn("XFS: nil uuid in log - IRIX style log");
257 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
258 xlog_warn("XFS: log has mismatched uuid - can't recover");
259 xlog_header_check_dump(mp, head);
260 XFS_ERROR_REPORT("xlog_header_check_mount",
261 XFS_ERRLEVEL_HIGH, mp);
262 return XFS_ERROR(EFSCORRUPTED);
263 }
264 return 0;
265}
266
267STATIC void
268xlog_recover_iodone(
269 struct xfs_buf *bp)
270{
271 xfs_mount_t *mp;
272
273 ASSERT(XFS_BUF_FSPRIVATE(bp, void *));
274
275 if (XFS_BUF_GETERROR(bp)) {
276 /*
277 * We're not going to bother about retrying
278 * this during recovery. One strike!
279 */
280 mp = XFS_BUF_FSPRIVATE(bp, xfs_mount_t *);
281 xfs_ioerror_alert("xlog_recover_iodone",
282 mp, bp, XFS_BUF_ADDR(bp));
Nathan Scott7d04a332006-06-09 14:58:38 +1000283 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 }
285 XFS_BUF_SET_FSPRIVATE(bp, NULL);
286 XFS_BUF_CLR_IODONE_FUNC(bp);
287 xfs_biodone(bp);
288}
289
290/*
291 * This routine finds (to an approximation) the first block in the physical
292 * log which contains the given cycle. It uses a binary search algorithm.
293 * Note that the algorithm can not be perfect because the disk will not
294 * necessarily be perfect.
295 */
296int
297xlog_find_cycle_start(
298 xlog_t *log,
299 xfs_buf_t *bp,
300 xfs_daddr_t first_blk,
301 xfs_daddr_t *last_blk,
302 uint cycle)
303{
304 xfs_caddr_t offset;
305 xfs_daddr_t mid_blk;
306 uint mid_cycle;
307 int error;
308
309 mid_blk = BLK_AVG(first_blk, *last_blk);
310 while (mid_blk != first_blk && mid_blk != *last_blk) {
311 if ((error = xlog_bread(log, mid_blk, 1, bp)))
312 return error;
313 offset = xlog_align(log, mid_blk, 1, bp);
314 mid_cycle = GET_CYCLE(offset, ARCH_CONVERT);
315 if (mid_cycle == cycle) {
316 *last_blk = mid_blk;
317 /* last_half_cycle == mid_cycle */
318 } else {
319 first_blk = mid_blk;
320 /* first_half_cycle == mid_cycle */
321 }
322 mid_blk = BLK_AVG(first_blk, *last_blk);
323 }
324 ASSERT((mid_blk == first_blk && mid_blk+1 == *last_blk) ||
325 (mid_blk == *last_blk && mid_blk-1 == first_blk));
326
327 return 0;
328}
329
330/*
331 * Check that the range of blocks does not contain the cycle number
332 * given. The scan needs to occur from front to back and the ptr into the
333 * region must be updated since a later routine will need to perform another
334 * test. If the region is completely good, we end up returning the same
335 * last block number.
336 *
337 * Set blkno to -1 if we encounter no errors. This is an invalid block number
338 * since we don't ever expect logs to get this large.
339 */
340STATIC int
341xlog_find_verify_cycle(
342 xlog_t *log,
343 xfs_daddr_t start_blk,
344 int nbblks,
345 uint stop_on_cycle_no,
346 xfs_daddr_t *new_blk)
347{
348 xfs_daddr_t i, j;
349 uint cycle;
350 xfs_buf_t *bp;
351 xfs_daddr_t bufblks;
352 xfs_caddr_t buf = NULL;
353 int error = 0;
354
355 bufblks = 1 << ffs(nbblks);
356
357 while (!(bp = xlog_get_bp(log, bufblks))) {
358 /* can't get enough memory to do everything in one big buffer */
359 bufblks >>= 1;
360 if (bufblks <= log->l_sectbb_log)
361 return ENOMEM;
362 }
363
364 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
365 int bcount;
366
367 bcount = min(bufblks, (start_blk + nbblks - i));
368
369 if ((error = xlog_bread(log, i, bcount, bp)))
370 goto out;
371
372 buf = xlog_align(log, i, bcount, bp);
373 for (j = 0; j < bcount; j++) {
374 cycle = GET_CYCLE(buf, ARCH_CONVERT);
375 if (cycle == stop_on_cycle_no) {
376 *new_blk = i+j;
377 goto out;
378 }
379
380 buf += BBSIZE;
381 }
382 }
383
384 *new_blk = -1;
385
386out:
387 xlog_put_bp(bp);
388 return error;
389}
390
391/*
392 * Potentially backup over partial log record write.
393 *
394 * In the typical case, last_blk is the number of the block directly after
395 * a good log record. Therefore, we subtract one to get the block number
396 * of the last block in the given buffer. extra_bblks contains the number
397 * of blocks we would have read on a previous read. This happens when the
398 * last log record is split over the end of the physical log.
399 *
400 * extra_bblks is the number of blocks potentially verified on a previous
401 * call to this routine.
402 */
403STATIC int
404xlog_find_verify_log_record(
405 xlog_t *log,
406 xfs_daddr_t start_blk,
407 xfs_daddr_t *last_blk,
408 int extra_bblks)
409{
410 xfs_daddr_t i;
411 xfs_buf_t *bp;
412 xfs_caddr_t offset = NULL;
413 xlog_rec_header_t *head = NULL;
414 int error = 0;
415 int smallmem = 0;
416 int num_blks = *last_blk - start_blk;
417 int xhdrs;
418
419 ASSERT(start_blk != 0 || *last_blk != start_blk);
420
421 if (!(bp = xlog_get_bp(log, num_blks))) {
422 if (!(bp = xlog_get_bp(log, 1)))
423 return ENOMEM;
424 smallmem = 1;
425 } else {
426 if ((error = xlog_bread(log, start_blk, num_blks, bp)))
427 goto out;
428 offset = xlog_align(log, start_blk, num_blks, bp);
429 offset += ((num_blks - 1) << BBSHIFT);
430 }
431
432 for (i = (*last_blk) - 1; i >= 0; i--) {
433 if (i < start_blk) {
434 /* valid log record not found */
435 xlog_warn(
436 "XFS: Log inconsistent (didn't find previous header)");
437 ASSERT(0);
438 error = XFS_ERROR(EIO);
439 goto out;
440 }
441
442 if (smallmem) {
443 if ((error = xlog_bread(log, i, 1, bp)))
444 goto out;
445 offset = xlog_align(log, i, 1, bp);
446 }
447
448 head = (xlog_rec_header_t *)offset;
449
450 if (XLOG_HEADER_MAGIC_NUM ==
451 INT_GET(head->h_magicno, ARCH_CONVERT))
452 break;
453
454 if (!smallmem)
455 offset -= BBSIZE;
456 }
457
458 /*
459 * We hit the beginning of the physical log & still no header. Return
460 * to caller. If caller can handle a return of -1, then this routine
461 * will be called again for the end of the physical log.
462 */
463 if (i == -1) {
464 error = -1;
465 goto out;
466 }
467
468 /*
469 * We have the final block of the good log (the first block
470 * of the log record _before_ the head. So we check the uuid.
471 */
472 if ((error = xlog_header_check_mount(log->l_mp, head)))
473 goto out;
474
475 /*
476 * We may have found a log record header before we expected one.
477 * last_blk will be the 1st block # with a given cycle #. We may end
478 * up reading an entire log record. In this case, we don't want to
479 * reset last_blk. Only when last_blk points in the middle of a log
480 * record do we update last_blk.
481 */
482 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
483 uint h_size = INT_GET(head->h_size, ARCH_CONVERT);
484
485 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
486 if (h_size % XLOG_HEADER_CYCLE_SIZE)
487 xhdrs++;
488 } else {
489 xhdrs = 1;
490 }
491
492 if (*last_blk - i + extra_bblks
493 != BTOBB(INT_GET(head->h_len, ARCH_CONVERT)) + xhdrs)
494 *last_blk = i;
495
496out:
497 xlog_put_bp(bp);
498 return error;
499}
500
501/*
502 * Head is defined to be the point of the log where the next log write
503 * write could go. This means that incomplete LR writes at the end are
504 * eliminated when calculating the head. We aren't guaranteed that previous
505 * LR have complete transactions. We only know that a cycle number of
506 * current cycle number -1 won't be present in the log if we start writing
507 * from our current block number.
508 *
509 * last_blk contains the block number of the first block with a given
510 * cycle number.
511 *
512 * Return: zero if normal, non-zero if error.
513 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000514STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515xlog_find_head(
516 xlog_t *log,
517 xfs_daddr_t *return_head_blk)
518{
519 xfs_buf_t *bp;
520 xfs_caddr_t offset;
521 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
522 int num_scan_bblks;
523 uint first_half_cycle, last_half_cycle;
524 uint stop_on_cycle;
525 int error, log_bbnum = log->l_logBBsize;
526
527 /* Is the end of the log device zeroed? */
528 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
529 *return_head_blk = first_blk;
530
531 /* Is the whole lot zeroed? */
532 if (!first_blk) {
533 /* Linux XFS shouldn't generate totally zeroed logs -
534 * mkfs etc write a dummy unmount record to a fresh
535 * log so we can store the uuid in there
536 */
537 xlog_warn("XFS: totally zeroed log");
538 }
539
540 return 0;
541 } else if (error) {
542 xlog_warn("XFS: empty log check failed");
543 return error;
544 }
545
546 first_blk = 0; /* get cycle # of 1st block */
547 bp = xlog_get_bp(log, 1);
548 if (!bp)
549 return ENOMEM;
550 if ((error = xlog_bread(log, 0, 1, bp)))
551 goto bp_err;
552 offset = xlog_align(log, 0, 1, bp);
553 first_half_cycle = GET_CYCLE(offset, ARCH_CONVERT);
554
555 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
556 if ((error = xlog_bread(log, last_blk, 1, bp)))
557 goto bp_err;
558 offset = xlog_align(log, last_blk, 1, bp);
559 last_half_cycle = GET_CYCLE(offset, ARCH_CONVERT);
560 ASSERT(last_half_cycle != 0);
561
562 /*
563 * If the 1st half cycle number is equal to the last half cycle number,
564 * then the entire log is stamped with the same cycle number. In this
565 * case, head_blk can't be set to zero (which makes sense). The below
566 * math doesn't work out properly with head_blk equal to zero. Instead,
567 * we set it to log_bbnum which is an invalid block number, but this
568 * value makes the math correct. If head_blk doesn't changed through
569 * all the tests below, *head_blk is set to zero at the very end rather
570 * than log_bbnum. In a sense, log_bbnum and zero are the same block
571 * in a circular file.
572 */
573 if (first_half_cycle == last_half_cycle) {
574 /*
575 * In this case we believe that the entire log should have
576 * cycle number last_half_cycle. We need to scan backwards
577 * from the end verifying that there are no holes still
578 * containing last_half_cycle - 1. If we find such a hole,
579 * then the start of that hole will be the new head. The
580 * simple case looks like
581 * x | x ... | x - 1 | x
582 * Another case that fits this picture would be
583 * x | x + 1 | x ... | x
Nathan Scottc41564b2006-03-29 08:55:14 +1000584 * In this case the head really is somewhere at the end of the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 * log, as one of the latest writes at the beginning was
586 * incomplete.
587 * One more case is
588 * x | x + 1 | x ... | x - 1 | x
589 * This is really the combination of the above two cases, and
590 * the head has to end up at the start of the x-1 hole at the
591 * end of the log.
592 *
593 * In the 256k log case, we will read from the beginning to the
594 * end of the log and search for cycle numbers equal to x-1.
595 * We don't worry about the x+1 blocks that we encounter,
596 * because we know that they cannot be the head since the log
597 * started with x.
598 */
599 head_blk = log_bbnum;
600 stop_on_cycle = last_half_cycle - 1;
601 } else {
602 /*
603 * In this case we want to find the first block with cycle
604 * number matching last_half_cycle. We expect the log to be
605 * some variation on
606 * x + 1 ... | x ...
607 * The first block with cycle number x (last_half_cycle) will
608 * be where the new head belongs. First we do a binary search
609 * for the first occurrence of last_half_cycle. The binary
610 * search may not be totally accurate, so then we scan back
611 * from there looking for occurrences of last_half_cycle before
612 * us. If that backwards scan wraps around the beginning of
613 * the log, then we look for occurrences of last_half_cycle - 1
614 * at the end of the log. The cases we're looking for look
615 * like
616 * x + 1 ... | x | x + 1 | x ...
617 * ^ binary search stopped here
618 * or
619 * x + 1 ... | x ... | x - 1 | x
620 * <---------> less than scan distance
621 */
622 stop_on_cycle = last_half_cycle;
623 if ((error = xlog_find_cycle_start(log, bp, first_blk,
624 &head_blk, last_half_cycle)))
625 goto bp_err;
626 }
627
628 /*
629 * Now validate the answer. Scan back some number of maximum possible
630 * blocks and make sure each one has the expected cycle number. The
631 * maximum is determined by the total possible amount of buffering
632 * in the in-core log. The following number can be made tighter if
633 * we actually look at the block size of the filesystem.
634 */
635 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
636 if (head_blk >= num_scan_bblks) {
637 /*
638 * We are guaranteed that the entire check can be performed
639 * in one buffer.
640 */
641 start_blk = head_blk - num_scan_bblks;
642 if ((error = xlog_find_verify_cycle(log,
643 start_blk, num_scan_bblks,
644 stop_on_cycle, &new_blk)))
645 goto bp_err;
646 if (new_blk != -1)
647 head_blk = new_blk;
648 } else { /* need to read 2 parts of log */
649 /*
650 * We are going to scan backwards in the log in two parts.
651 * First we scan the physical end of the log. In this part
652 * of the log, we are looking for blocks with cycle number
653 * last_half_cycle - 1.
654 * If we find one, then we know that the log starts there, as
655 * we've found a hole that didn't get written in going around
656 * the end of the physical log. The simple case for this is
657 * x + 1 ... | x ... | x - 1 | x
658 * <---------> less than scan distance
659 * If all of the blocks at the end of the log have cycle number
660 * last_half_cycle, then we check the blocks at the start of
661 * the log looking for occurrences of last_half_cycle. If we
662 * find one, then our current estimate for the location of the
663 * first occurrence of last_half_cycle is wrong and we move
664 * back to the hole we've found. This case looks like
665 * x + 1 ... | x | x + 1 | x ...
666 * ^ binary search stopped here
667 * Another case we need to handle that only occurs in 256k
668 * logs is
669 * x + 1 ... | x ... | x+1 | x ...
670 * ^ binary search stops here
671 * In a 256k log, the scan at the end of the log will see the
672 * x + 1 blocks. We need to skip past those since that is
673 * certainly not the head of the log. By searching for
674 * last_half_cycle-1 we accomplish that.
675 */
676 start_blk = log_bbnum - num_scan_bblks + head_blk;
677 ASSERT(head_blk <= INT_MAX &&
678 (xfs_daddr_t) num_scan_bblks - head_blk >= 0);
679 if ((error = xlog_find_verify_cycle(log, start_blk,
680 num_scan_bblks - (int)head_blk,
681 (stop_on_cycle - 1), &new_blk)))
682 goto bp_err;
683 if (new_blk != -1) {
684 head_blk = new_blk;
685 goto bad_blk;
686 }
687
688 /*
689 * Scan beginning of log now. The last part of the physical
690 * log is good. This scan needs to verify that it doesn't find
691 * the last_half_cycle.
692 */
693 start_blk = 0;
694 ASSERT(head_blk <= INT_MAX);
695 if ((error = xlog_find_verify_cycle(log,
696 start_blk, (int)head_blk,
697 stop_on_cycle, &new_blk)))
698 goto bp_err;
699 if (new_blk != -1)
700 head_blk = new_blk;
701 }
702
703 bad_blk:
704 /*
705 * Now we need to make sure head_blk is not pointing to a block in
706 * the middle of a log record.
707 */
708 num_scan_bblks = XLOG_REC_SHIFT(log);
709 if (head_blk >= num_scan_bblks) {
710 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
711
712 /* start ptr at last block ptr before head_blk */
713 if ((error = xlog_find_verify_log_record(log, start_blk,
714 &head_blk, 0)) == -1) {
715 error = XFS_ERROR(EIO);
716 goto bp_err;
717 } else if (error)
718 goto bp_err;
719 } else {
720 start_blk = 0;
721 ASSERT(head_blk <= INT_MAX);
722 if ((error = xlog_find_verify_log_record(log, start_blk,
723 &head_blk, 0)) == -1) {
724 /* We hit the beginning of the log during our search */
725 start_blk = log_bbnum - num_scan_bblks + head_blk;
726 new_blk = log_bbnum;
727 ASSERT(start_blk <= INT_MAX &&
728 (xfs_daddr_t) log_bbnum-start_blk >= 0);
729 ASSERT(head_blk <= INT_MAX);
730 if ((error = xlog_find_verify_log_record(log,
731 start_blk, &new_blk,
732 (int)head_blk)) == -1) {
733 error = XFS_ERROR(EIO);
734 goto bp_err;
735 } else if (error)
736 goto bp_err;
737 if (new_blk != log_bbnum)
738 head_blk = new_blk;
739 } else if (error)
740 goto bp_err;
741 }
742
743 xlog_put_bp(bp);
744 if (head_blk == log_bbnum)
745 *return_head_blk = 0;
746 else
747 *return_head_blk = head_blk;
748 /*
749 * When returning here, we have a good block number. Bad block
750 * means that during a previous crash, we didn't have a clean break
751 * from cycle number N to cycle number N-1. In this case, we need
752 * to find the first block with cycle number N-1.
753 */
754 return 0;
755
756 bp_err:
757 xlog_put_bp(bp);
758
759 if (error)
760 xlog_warn("XFS: failed to find log head");
761 return error;
762}
763
764/*
765 * Find the sync block number or the tail of the log.
766 *
767 * This will be the block number of the last record to have its
768 * associated buffers synced to disk. Every log record header has
769 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
770 * to get a sync block number. The only concern is to figure out which
771 * log record header to believe.
772 *
773 * The following algorithm uses the log record header with the largest
774 * lsn. The entire log record does not need to be valid. We only care
775 * that the header is valid.
776 *
777 * We could speed up search by using current head_blk buffer, but it is not
778 * available.
779 */
780int
781xlog_find_tail(
782 xlog_t *log,
783 xfs_daddr_t *head_blk,
Eric Sandeen65be6052006-01-11 15:34:19 +1100784 xfs_daddr_t *tail_blk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785{
786 xlog_rec_header_t *rhead;
787 xlog_op_header_t *op_head;
788 xfs_caddr_t offset = NULL;
789 xfs_buf_t *bp;
790 int error, i, found;
791 xfs_daddr_t umount_data_blk;
792 xfs_daddr_t after_umount_blk;
793 xfs_lsn_t tail_lsn;
794 int hblks;
795
796 found = 0;
797
798 /*
799 * Find previous log record
800 */
801 if ((error = xlog_find_head(log, head_blk)))
802 return error;
803
804 bp = xlog_get_bp(log, 1);
805 if (!bp)
806 return ENOMEM;
807 if (*head_blk == 0) { /* special case */
808 if ((error = xlog_bread(log, 0, 1, bp)))
809 goto bread_err;
810 offset = xlog_align(log, 0, 1, bp);
811 if (GET_CYCLE(offset, ARCH_CONVERT) == 0) {
812 *tail_blk = 0;
813 /* leave all other log inited values alone */
814 goto exit;
815 }
816 }
817
818 /*
819 * Search backwards looking for log record header block
820 */
821 ASSERT(*head_blk < INT_MAX);
822 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
823 if ((error = xlog_bread(log, i, 1, bp)))
824 goto bread_err;
825 offset = xlog_align(log, i, 1, bp);
826 if (XLOG_HEADER_MAGIC_NUM ==
827 INT_GET(*(uint *)offset, ARCH_CONVERT)) {
828 found = 1;
829 break;
830 }
831 }
832 /*
833 * If we haven't found the log record header block, start looking
834 * again from the end of the physical log. XXXmiken: There should be
835 * a check here to make sure we didn't search more than N blocks in
836 * the previous code.
837 */
838 if (!found) {
839 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
840 if ((error = xlog_bread(log, i, 1, bp)))
841 goto bread_err;
842 offset = xlog_align(log, i, 1, bp);
843 if (XLOG_HEADER_MAGIC_NUM ==
844 INT_GET(*(uint*)offset, ARCH_CONVERT)) {
845 found = 2;
846 break;
847 }
848 }
849 }
850 if (!found) {
851 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
852 ASSERT(0);
853 return XFS_ERROR(EIO);
854 }
855
856 /* find blk_no of tail of log */
857 rhead = (xlog_rec_header_t *)offset;
858 *tail_blk = BLOCK_LSN(INT_GET(rhead->h_tail_lsn, ARCH_CONVERT));
859
860 /*
861 * Reset log values according to the state of the log when we
862 * crashed. In the case where head_blk == 0, we bump curr_cycle
863 * one because the next write starts a new cycle rather than
864 * continuing the cycle of the last good log record. At this
865 * point we have guaranteed that all partial log records have been
866 * accounted for. Therefore, we know that the last good log record
867 * written was complete and ended exactly on the end boundary
868 * of the physical log.
869 */
870 log->l_prev_block = i;
871 log->l_curr_block = (int)*head_blk;
872 log->l_curr_cycle = INT_GET(rhead->h_cycle, ARCH_CONVERT);
873 if (found == 2)
874 log->l_curr_cycle++;
875 log->l_tail_lsn = INT_GET(rhead->h_tail_lsn, ARCH_CONVERT);
876 log->l_last_sync_lsn = INT_GET(rhead->h_lsn, ARCH_CONVERT);
877 log->l_grant_reserve_cycle = log->l_curr_cycle;
878 log->l_grant_reserve_bytes = BBTOB(log->l_curr_block);
879 log->l_grant_write_cycle = log->l_curr_cycle;
880 log->l_grant_write_bytes = BBTOB(log->l_curr_block);
881
882 /*
883 * Look for unmount record. If we find it, then we know there
884 * was a clean unmount. Since 'i' could be the last block in
885 * the physical log, we convert to a log block before comparing
886 * to the head_blk.
887 *
888 * Save the current tail lsn to use to pass to
889 * xlog_clear_stale_blocks() below. We won't want to clear the
890 * unmount record if there is one, so we pass the lsn of the
891 * unmount record rather than the block after it.
892 */
893 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
894 int h_size = INT_GET(rhead->h_size, ARCH_CONVERT);
895 int h_version = INT_GET(rhead->h_version, ARCH_CONVERT);
896
897 if ((h_version & XLOG_VERSION_2) &&
898 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
899 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
900 if (h_size % XLOG_HEADER_CYCLE_SIZE)
901 hblks++;
902 } else {
903 hblks = 1;
904 }
905 } else {
906 hblks = 1;
907 }
908 after_umount_blk = (i + hblks + (int)
909 BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT))) % log->l_logBBsize;
910 tail_lsn = log->l_tail_lsn;
911 if (*head_blk == after_umount_blk &&
912 INT_GET(rhead->h_num_logops, ARCH_CONVERT) == 1) {
913 umount_data_blk = (i + hblks) % log->l_logBBsize;
914 if ((error = xlog_bread(log, umount_data_blk, 1, bp))) {
915 goto bread_err;
916 }
917 offset = xlog_align(log, umount_data_blk, 1, bp);
918 op_head = (xlog_op_header_t *)offset;
919 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
920 /*
921 * Set tail and last sync so that newly written
922 * log records will point recovery to after the
923 * current unmount record.
924 */
925 ASSIGN_ANY_LSN_HOST(log->l_tail_lsn, log->l_curr_cycle,
926 after_umount_blk);
927 ASSIGN_ANY_LSN_HOST(log->l_last_sync_lsn, log->l_curr_cycle,
928 after_umount_blk);
929 *tail_blk = after_umount_blk;
930 }
931 }
932
933 /*
934 * Make sure that there are no blocks in front of the head
935 * with the same cycle number as the head. This can happen
936 * because we allow multiple outstanding log writes concurrently,
937 * and the later writes might make it out before earlier ones.
938 *
939 * We use the lsn from before modifying it so that we'll never
940 * overwrite the unmount record after a clean unmount.
941 *
942 * Do this only if we are going to recover the filesystem
943 *
944 * NOTE: This used to say "if (!readonly)"
945 * However on Linux, we can & do recover a read-only filesystem.
946 * We only skip recovery if NORECOVERY is specified on mount,
947 * in which case we would not be here.
948 *
949 * But... if the -device- itself is readonly, just skip this.
950 * We can't recover this device anyway, so it won't matter.
951 */
952 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) {
953 error = xlog_clear_stale_blocks(log, tail_lsn);
954 }
955
956bread_err:
957exit:
958 xlog_put_bp(bp);
959
960 if (error)
961 xlog_warn("XFS: failed to locate log tail");
962 return error;
963}
964
965/*
966 * Is the log zeroed at all?
967 *
968 * The last binary search should be changed to perform an X block read
969 * once X becomes small enough. You can then search linearly through
970 * the X blocks. This will cut down on the number of reads we need to do.
971 *
972 * If the log is partially zeroed, this routine will pass back the blkno
973 * of the first block with cycle number 0. It won't have a complete LR
974 * preceding it.
975 *
976 * Return:
977 * 0 => the log is completely written to
978 * -1 => use *blk_no as the first block of the log
979 * >0 => error has occurred
980 */
981int
982xlog_find_zeroed(
983 xlog_t *log,
984 xfs_daddr_t *blk_no)
985{
986 xfs_buf_t *bp;
987 xfs_caddr_t offset;
988 uint first_cycle, last_cycle;
989 xfs_daddr_t new_blk, last_blk, start_blk;
990 xfs_daddr_t num_scan_bblks;
991 int error, log_bbnum = log->l_logBBsize;
992
Nathan Scott6fdf8cc2006-06-28 10:13:52 +1000993 *blk_no = 0;
994
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 /* check totally zeroed log */
996 bp = xlog_get_bp(log, 1);
997 if (!bp)
998 return ENOMEM;
999 if ((error = xlog_bread(log, 0, 1, bp)))
1000 goto bp_err;
1001 offset = xlog_align(log, 0, 1, bp);
1002 first_cycle = GET_CYCLE(offset, ARCH_CONVERT);
1003 if (first_cycle == 0) { /* completely zeroed log */
1004 *blk_no = 0;
1005 xlog_put_bp(bp);
1006 return -1;
1007 }
1008
1009 /* check partially zeroed log */
1010 if ((error = xlog_bread(log, log_bbnum-1, 1, bp)))
1011 goto bp_err;
1012 offset = xlog_align(log, log_bbnum-1, 1, bp);
1013 last_cycle = GET_CYCLE(offset, ARCH_CONVERT);
1014 if (last_cycle != 0) { /* log completely written to */
1015 xlog_put_bp(bp);
1016 return 0;
1017 } else if (first_cycle != 1) {
1018 /*
1019 * If the cycle of the last block is zero, the cycle of
1020 * the first block must be 1. If it's not, maybe we're
1021 * not looking at a log... Bail out.
1022 */
1023 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1024 return XFS_ERROR(EINVAL);
1025 }
1026
1027 /* we have a partially zeroed log */
1028 last_blk = log_bbnum-1;
1029 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1030 goto bp_err;
1031
1032 /*
1033 * Validate the answer. Because there is no way to guarantee that
1034 * the entire log is made up of log records which are the same size,
1035 * we scan over the defined maximum blocks. At this point, the maximum
1036 * is not chosen to mean anything special. XXXmiken
1037 */
1038 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1039 ASSERT(num_scan_bblks <= INT_MAX);
1040
1041 if (last_blk < num_scan_bblks)
1042 num_scan_bblks = last_blk;
1043 start_blk = last_blk - num_scan_bblks;
1044
1045 /*
1046 * We search for any instances of cycle number 0 that occur before
1047 * our current estimate of the head. What we're trying to detect is
1048 * 1 ... | 0 | 1 | 0...
1049 * ^ binary search ends here
1050 */
1051 if ((error = xlog_find_verify_cycle(log, start_blk,
1052 (int)num_scan_bblks, 0, &new_blk)))
1053 goto bp_err;
1054 if (new_blk != -1)
1055 last_blk = new_blk;
1056
1057 /*
1058 * Potentially backup over partial log record write. We don't need
1059 * to search the end of the log because we know it is zero.
1060 */
1061 if ((error = xlog_find_verify_log_record(log, start_blk,
1062 &last_blk, 0)) == -1) {
1063 error = XFS_ERROR(EIO);
1064 goto bp_err;
1065 } else if (error)
1066 goto bp_err;
1067
1068 *blk_no = last_blk;
1069bp_err:
1070 xlog_put_bp(bp);
1071 if (error)
1072 return error;
1073 return -1;
1074}
1075
1076/*
1077 * These are simple subroutines used by xlog_clear_stale_blocks() below
1078 * to initialize a buffer full of empty log record headers and write
1079 * them into the log.
1080 */
1081STATIC void
1082xlog_add_record(
1083 xlog_t *log,
1084 xfs_caddr_t buf,
1085 int cycle,
1086 int block,
1087 int tail_cycle,
1088 int tail_block)
1089{
1090 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1091
1092 memset(buf, 0, BBSIZE);
1093 INT_SET(recp->h_magicno, ARCH_CONVERT, XLOG_HEADER_MAGIC_NUM);
1094 INT_SET(recp->h_cycle, ARCH_CONVERT, cycle);
1095 INT_SET(recp->h_version, ARCH_CONVERT,
1096 XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? 2 : 1);
1097 ASSIGN_ANY_LSN_DISK(recp->h_lsn, cycle, block);
1098 ASSIGN_ANY_LSN_DISK(recp->h_tail_lsn, tail_cycle, tail_block);
1099 INT_SET(recp->h_fmt, ARCH_CONVERT, XLOG_FMT);
1100 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1101}
1102
1103STATIC int
1104xlog_write_log_records(
1105 xlog_t *log,
1106 int cycle,
1107 int start_block,
1108 int blocks,
1109 int tail_cycle,
1110 int tail_block)
1111{
1112 xfs_caddr_t offset;
1113 xfs_buf_t *bp;
1114 int balign, ealign;
1115 int sectbb = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
1116 int end_block = start_block + blocks;
1117 int bufblks;
1118 int error = 0;
1119 int i, j = 0;
1120
1121 bufblks = 1 << ffs(blocks);
1122 while (!(bp = xlog_get_bp(log, bufblks))) {
1123 bufblks >>= 1;
1124 if (bufblks <= log->l_sectbb_log)
1125 return ENOMEM;
1126 }
1127
1128 /* We may need to do a read at the start to fill in part of
1129 * the buffer in the starting sector not covered by the first
1130 * write below.
1131 */
1132 balign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, start_block);
1133 if (balign != start_block) {
1134 if ((error = xlog_bread(log, start_block, 1, bp))) {
1135 xlog_put_bp(bp);
1136 return error;
1137 }
1138 j = start_block - balign;
1139 }
1140
1141 for (i = start_block; i < end_block; i += bufblks) {
1142 int bcount, endcount;
1143
1144 bcount = min(bufblks, end_block - start_block);
1145 endcount = bcount - j;
1146
1147 /* We may need to do a read at the end to fill in part of
1148 * the buffer in the final sector not covered by the write.
1149 * If this is the same sector as the above read, skip it.
1150 */
1151 ealign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, end_block);
1152 if (j == 0 && (start_block + endcount > ealign)) {
1153 offset = XFS_BUF_PTR(bp);
1154 balign = BBTOB(ealign - start_block);
1155 XFS_BUF_SET_PTR(bp, offset + balign, BBTOB(sectbb));
1156 if ((error = xlog_bread(log, ealign, sectbb, bp)))
1157 break;
1158 XFS_BUF_SET_PTR(bp, offset, bufblks);
1159 }
1160
1161 offset = xlog_align(log, start_block, endcount, bp);
1162 for (; j < endcount; j++) {
1163 xlog_add_record(log, offset, cycle, i+j,
1164 tail_cycle, tail_block);
1165 offset += BBSIZE;
1166 }
1167 error = xlog_bwrite(log, start_block, endcount, bp);
1168 if (error)
1169 break;
1170 start_block += endcount;
1171 j = 0;
1172 }
1173 xlog_put_bp(bp);
1174 return error;
1175}
1176
1177/*
1178 * This routine is called to blow away any incomplete log writes out
1179 * in front of the log head. We do this so that we won't become confused
1180 * if we come up, write only a little bit more, and then crash again.
1181 * If we leave the partial log records out there, this situation could
1182 * cause us to think those partial writes are valid blocks since they
1183 * have the current cycle number. We get rid of them by overwriting them
1184 * with empty log records with the old cycle number rather than the
1185 * current one.
1186 *
1187 * The tail lsn is passed in rather than taken from
1188 * the log so that we will not write over the unmount record after a
1189 * clean unmount in a 512 block log. Doing so would leave the log without
1190 * any valid log records in it until a new one was written. If we crashed
1191 * during that time we would not be able to recover.
1192 */
1193STATIC int
1194xlog_clear_stale_blocks(
1195 xlog_t *log,
1196 xfs_lsn_t tail_lsn)
1197{
1198 int tail_cycle, head_cycle;
1199 int tail_block, head_block;
1200 int tail_distance, max_distance;
1201 int distance;
1202 int error;
1203
1204 tail_cycle = CYCLE_LSN(tail_lsn);
1205 tail_block = BLOCK_LSN(tail_lsn);
1206 head_cycle = log->l_curr_cycle;
1207 head_block = log->l_curr_block;
1208
1209 /*
1210 * Figure out the distance between the new head of the log
1211 * and the tail. We want to write over any blocks beyond the
1212 * head that we may have written just before the crash, but
1213 * we don't want to overwrite the tail of the log.
1214 */
1215 if (head_cycle == tail_cycle) {
1216 /*
1217 * The tail is behind the head in the physical log,
1218 * so the distance from the head to the tail is the
1219 * distance from the head to the end of the log plus
1220 * the distance from the beginning of the log to the
1221 * tail.
1222 */
1223 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1224 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1225 XFS_ERRLEVEL_LOW, log->l_mp);
1226 return XFS_ERROR(EFSCORRUPTED);
1227 }
1228 tail_distance = tail_block + (log->l_logBBsize - head_block);
1229 } else {
1230 /*
1231 * The head is behind the tail in the physical log,
1232 * so the distance from the head to the tail is just
1233 * the tail block minus the head block.
1234 */
1235 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1236 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1237 XFS_ERRLEVEL_LOW, log->l_mp);
1238 return XFS_ERROR(EFSCORRUPTED);
1239 }
1240 tail_distance = tail_block - head_block;
1241 }
1242
1243 /*
1244 * If the head is right up against the tail, we can't clear
1245 * anything.
1246 */
1247 if (tail_distance <= 0) {
1248 ASSERT(tail_distance == 0);
1249 return 0;
1250 }
1251
1252 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1253 /*
1254 * Take the smaller of the maximum amount of outstanding I/O
1255 * we could have and the distance to the tail to clear out.
1256 * We take the smaller so that we don't overwrite the tail and
1257 * we don't waste all day writing from the head to the tail
1258 * for no reason.
1259 */
1260 max_distance = MIN(max_distance, tail_distance);
1261
1262 if ((head_block + max_distance) <= log->l_logBBsize) {
1263 /*
1264 * We can stomp all the blocks we need to without
1265 * wrapping around the end of the log. Just do it
1266 * in a single write. Use the cycle number of the
1267 * current cycle minus one so that the log will look like:
1268 * n ... | n - 1 ...
1269 */
1270 error = xlog_write_log_records(log, (head_cycle - 1),
1271 head_block, max_distance, tail_cycle,
1272 tail_block);
1273 if (error)
1274 return error;
1275 } else {
1276 /*
1277 * We need to wrap around the end of the physical log in
1278 * order to clear all the blocks. Do it in two separate
1279 * I/Os. The first write should be from the head to the
1280 * end of the physical log, and it should use the current
1281 * cycle number minus one just like above.
1282 */
1283 distance = log->l_logBBsize - head_block;
1284 error = xlog_write_log_records(log, (head_cycle - 1),
1285 head_block, distance, tail_cycle,
1286 tail_block);
1287
1288 if (error)
1289 return error;
1290
1291 /*
1292 * Now write the blocks at the start of the physical log.
1293 * This writes the remainder of the blocks we want to clear.
1294 * It uses the current cycle number since we're now on the
1295 * same cycle as the head so that we get:
1296 * n ... n ... | n - 1 ...
1297 * ^^^^^ blocks we're writing
1298 */
1299 distance = max_distance - (log->l_logBBsize - head_block);
1300 error = xlog_write_log_records(log, head_cycle, 0, distance,
1301 tail_cycle, tail_block);
1302 if (error)
1303 return error;
1304 }
1305
1306 return 0;
1307}
1308
1309/******************************************************************************
1310 *
1311 * Log recover routines
1312 *
1313 ******************************************************************************
1314 */
1315
1316STATIC xlog_recover_t *
1317xlog_recover_find_tid(
1318 xlog_recover_t *q,
1319 xlog_tid_t tid)
1320{
1321 xlog_recover_t *p = q;
1322
1323 while (p != NULL) {
1324 if (p->r_log_tid == tid)
1325 break;
1326 p = p->r_next;
1327 }
1328 return p;
1329}
1330
1331STATIC void
1332xlog_recover_put_hashq(
1333 xlog_recover_t **q,
1334 xlog_recover_t *trans)
1335{
1336 trans->r_next = *q;
1337 *q = trans;
1338}
1339
1340STATIC void
1341xlog_recover_add_item(
1342 xlog_recover_item_t **itemq)
1343{
1344 xlog_recover_item_t *item;
1345
1346 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1347 xlog_recover_insert_item_backq(itemq, item);
1348}
1349
1350STATIC int
1351xlog_recover_add_to_cont_trans(
1352 xlog_recover_t *trans,
1353 xfs_caddr_t dp,
1354 int len)
1355{
1356 xlog_recover_item_t *item;
1357 xfs_caddr_t ptr, old_ptr;
1358 int old_len;
1359
1360 item = trans->r_itemq;
1361 if (item == 0) {
1362 /* finish copying rest of trans header */
1363 xlog_recover_add_item(&trans->r_itemq);
1364 ptr = (xfs_caddr_t) &trans->r_theader +
1365 sizeof(xfs_trans_header_t) - len;
1366 memcpy(ptr, dp, len); /* d, s, l */
1367 return 0;
1368 }
1369 item = item->ri_prev;
1370
1371 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1372 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1373
Christoph Hellwig760dea62005-09-02 16:56:02 +10001374 ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1376 item->ri_buf[item->ri_cnt-1].i_len += len;
1377 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1378 return 0;
1379}
1380
1381/*
1382 * The next region to add is the start of a new region. It could be
1383 * a whole region or it could be the first part of a new region. Because
1384 * of this, the assumption here is that the type and size fields of all
1385 * format structures fit into the first 32 bits of the structure.
1386 *
1387 * This works because all regions must be 32 bit aligned. Therefore, we
1388 * either have both fields or we have neither field. In the case we have
1389 * neither field, the data part of the region is zero length. We only have
1390 * a log_op_header and can throw away the header since a new one will appear
1391 * later. If we have at least 4 bytes, then we can determine how many regions
1392 * will appear in the current log item.
1393 */
1394STATIC int
1395xlog_recover_add_to_trans(
1396 xlog_recover_t *trans,
1397 xfs_caddr_t dp,
1398 int len)
1399{
1400 xfs_inode_log_format_t *in_f; /* any will do */
1401 xlog_recover_item_t *item;
1402 xfs_caddr_t ptr;
1403
1404 if (!len)
1405 return 0;
1406 item = trans->r_itemq;
1407 if (item == 0) {
1408 ASSERT(*(uint *)dp == XFS_TRANS_HEADER_MAGIC);
1409 if (len == sizeof(xfs_trans_header_t))
1410 xlog_recover_add_item(&trans->r_itemq);
1411 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1412 return 0;
1413 }
1414
1415 ptr = kmem_alloc(len, KM_SLEEP);
1416 memcpy(ptr, dp, len);
1417 in_f = (xfs_inode_log_format_t *)ptr;
1418
1419 if (item->ri_prev->ri_total != 0 &&
1420 item->ri_prev->ri_total == item->ri_prev->ri_cnt) {
1421 xlog_recover_add_item(&trans->r_itemq);
1422 }
1423 item = trans->r_itemq;
1424 item = item->ri_prev;
1425
1426 if (item->ri_total == 0) { /* first region to be added */
1427 item->ri_total = in_f->ilf_size;
1428 ASSERT(item->ri_total <= XLOG_MAX_REGIONS_IN_ITEM);
1429 item->ri_buf = kmem_zalloc((item->ri_total *
1430 sizeof(xfs_log_iovec_t)), KM_SLEEP);
1431 }
1432 ASSERT(item->ri_total > item->ri_cnt);
1433 /* Description region is ri_buf[0] */
1434 item->ri_buf[item->ri_cnt].i_addr = ptr;
1435 item->ri_buf[item->ri_cnt].i_len = len;
1436 item->ri_cnt++;
1437 return 0;
1438}
1439
1440STATIC void
1441xlog_recover_new_tid(
1442 xlog_recover_t **q,
1443 xlog_tid_t tid,
1444 xfs_lsn_t lsn)
1445{
1446 xlog_recover_t *trans;
1447
1448 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1449 trans->r_log_tid = tid;
1450 trans->r_lsn = lsn;
1451 xlog_recover_put_hashq(q, trans);
1452}
1453
1454STATIC int
1455xlog_recover_unlink_tid(
1456 xlog_recover_t **q,
1457 xlog_recover_t *trans)
1458{
1459 xlog_recover_t *tp;
1460 int found = 0;
1461
1462 ASSERT(trans != 0);
1463 if (trans == *q) {
1464 *q = (*q)->r_next;
1465 } else {
1466 tp = *q;
1467 while (tp != 0) {
1468 if (tp->r_next == trans) {
1469 found = 1;
1470 break;
1471 }
1472 tp = tp->r_next;
1473 }
1474 if (!found) {
1475 xlog_warn(
1476 "XFS: xlog_recover_unlink_tid: trans not found");
1477 ASSERT(0);
1478 return XFS_ERROR(EIO);
1479 }
1480 tp->r_next = tp->r_next->r_next;
1481 }
1482 return 0;
1483}
1484
1485STATIC void
1486xlog_recover_insert_item_backq(
1487 xlog_recover_item_t **q,
1488 xlog_recover_item_t *item)
1489{
1490 if (*q == 0) {
1491 item->ri_prev = item->ri_next = item;
1492 *q = item;
1493 } else {
1494 item->ri_next = *q;
1495 item->ri_prev = (*q)->ri_prev;
1496 (*q)->ri_prev = item;
1497 item->ri_prev->ri_next = item;
1498 }
1499}
1500
1501STATIC void
1502xlog_recover_insert_item_frontq(
1503 xlog_recover_item_t **q,
1504 xlog_recover_item_t *item)
1505{
1506 xlog_recover_insert_item_backq(q, item);
1507 *q = item;
1508}
1509
1510STATIC int
1511xlog_recover_reorder_trans(
1512 xlog_t *log,
1513 xlog_recover_t *trans)
1514{
1515 xlog_recover_item_t *first_item, *itemq, *itemq_next;
1516 xfs_buf_log_format_t *buf_f;
1517 xfs_buf_log_format_v1_t *obuf_f;
1518 ushort flags = 0;
1519
1520 first_item = itemq = trans->r_itemq;
1521 trans->r_itemq = NULL;
1522 do {
1523 itemq_next = itemq->ri_next;
1524 buf_f = (xfs_buf_log_format_t *)itemq->ri_buf[0].i_addr;
1525 switch (ITEM_TYPE(itemq)) {
1526 case XFS_LI_BUF:
1527 flags = buf_f->blf_flags;
1528 break;
1529 case XFS_LI_6_1_BUF:
1530 case XFS_LI_5_3_BUF:
1531 obuf_f = (xfs_buf_log_format_v1_t*)buf_f;
1532 flags = obuf_f->blf_flags;
1533 break;
1534 }
1535
1536 switch (ITEM_TYPE(itemq)) {
1537 case XFS_LI_BUF:
1538 case XFS_LI_6_1_BUF:
1539 case XFS_LI_5_3_BUF:
1540 if (!(flags & XFS_BLI_CANCEL)) {
1541 xlog_recover_insert_item_frontq(&trans->r_itemq,
1542 itemq);
1543 break;
1544 }
1545 case XFS_LI_INODE:
1546 case XFS_LI_6_1_INODE:
1547 case XFS_LI_5_3_INODE:
1548 case XFS_LI_DQUOT:
1549 case XFS_LI_QUOTAOFF:
1550 case XFS_LI_EFD:
1551 case XFS_LI_EFI:
1552 xlog_recover_insert_item_backq(&trans->r_itemq, itemq);
1553 break;
1554 default:
1555 xlog_warn(
1556 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1557 ASSERT(0);
1558 return XFS_ERROR(EIO);
1559 }
1560 itemq = itemq_next;
1561 } while (first_item != itemq);
1562 return 0;
1563}
1564
1565/*
1566 * Build up the table of buf cancel records so that we don't replay
1567 * cancelled data in the second pass. For buffer records that are
1568 * not cancel records, there is nothing to do here so we just return.
1569 *
1570 * If we get a cancel record which is already in the table, this indicates
1571 * that the buffer was cancelled multiple times. In order to ensure
1572 * that during pass 2 we keep the record in the table until we reach its
1573 * last occurrence in the log, we keep a reference count in the cancel
1574 * record in the table to tell us how many times we expect to see this
1575 * record during the second pass.
1576 */
1577STATIC void
1578xlog_recover_do_buffer_pass1(
1579 xlog_t *log,
1580 xfs_buf_log_format_t *buf_f)
1581{
1582 xfs_buf_cancel_t *bcp;
1583 xfs_buf_cancel_t *nextp;
1584 xfs_buf_cancel_t *prevp;
1585 xfs_buf_cancel_t **bucket;
1586 xfs_buf_log_format_v1_t *obuf_f;
1587 xfs_daddr_t blkno = 0;
1588 uint len = 0;
1589 ushort flags = 0;
1590
1591 switch (buf_f->blf_type) {
1592 case XFS_LI_BUF:
1593 blkno = buf_f->blf_blkno;
1594 len = buf_f->blf_len;
1595 flags = buf_f->blf_flags;
1596 break;
1597 case XFS_LI_6_1_BUF:
1598 case XFS_LI_5_3_BUF:
1599 obuf_f = (xfs_buf_log_format_v1_t*)buf_f;
1600 blkno = (xfs_daddr_t) obuf_f->blf_blkno;
1601 len = obuf_f->blf_len;
1602 flags = obuf_f->blf_flags;
1603 break;
1604 }
1605
1606 /*
1607 * If this isn't a cancel buffer item, then just return.
1608 */
1609 if (!(flags & XFS_BLI_CANCEL))
1610 return;
1611
1612 /*
1613 * Insert an xfs_buf_cancel record into the hash table of
1614 * them. If there is already an identical record, bump
1615 * its reference count.
1616 */
1617 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1618 XLOG_BC_TABLE_SIZE];
1619 /*
1620 * If the hash bucket is empty then just insert a new record into
1621 * the bucket.
1622 */
1623 if (*bucket == NULL) {
1624 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1625 KM_SLEEP);
1626 bcp->bc_blkno = blkno;
1627 bcp->bc_len = len;
1628 bcp->bc_refcount = 1;
1629 bcp->bc_next = NULL;
1630 *bucket = bcp;
1631 return;
1632 }
1633
1634 /*
1635 * The hash bucket is not empty, so search for duplicates of our
1636 * record. If we find one them just bump its refcount. If not
1637 * then add us at the end of the list.
1638 */
1639 prevp = NULL;
1640 nextp = *bucket;
1641 while (nextp != NULL) {
1642 if (nextp->bc_blkno == blkno && nextp->bc_len == len) {
1643 nextp->bc_refcount++;
1644 return;
1645 }
1646 prevp = nextp;
1647 nextp = nextp->bc_next;
1648 }
1649 ASSERT(prevp != NULL);
1650 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1651 KM_SLEEP);
1652 bcp->bc_blkno = blkno;
1653 bcp->bc_len = len;
1654 bcp->bc_refcount = 1;
1655 bcp->bc_next = NULL;
1656 prevp->bc_next = bcp;
1657}
1658
1659/*
1660 * Check to see whether the buffer being recovered has a corresponding
1661 * entry in the buffer cancel record table. If it does then return 1
1662 * so that it will be cancelled, otherwise return 0. If the buffer is
1663 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
1664 * the refcount on the entry in the table and remove it from the table
1665 * if this is the last reference.
1666 *
1667 * We remove the cancel record from the table when we encounter its
1668 * last occurrence in the log so that if the same buffer is re-used
1669 * again after its last cancellation we actually replay the changes
1670 * made at that point.
1671 */
1672STATIC int
1673xlog_check_buffer_cancelled(
1674 xlog_t *log,
1675 xfs_daddr_t blkno,
1676 uint len,
1677 ushort flags)
1678{
1679 xfs_buf_cancel_t *bcp;
1680 xfs_buf_cancel_t *prevp;
1681 xfs_buf_cancel_t **bucket;
1682
1683 if (log->l_buf_cancel_table == NULL) {
1684 /*
1685 * There is nothing in the table built in pass one,
1686 * so this buffer must not be cancelled.
1687 */
1688 ASSERT(!(flags & XFS_BLI_CANCEL));
1689 return 0;
1690 }
1691
1692 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1693 XLOG_BC_TABLE_SIZE];
1694 bcp = *bucket;
1695 if (bcp == NULL) {
1696 /*
1697 * There is no corresponding entry in the table built
1698 * in pass one, so this buffer has not been cancelled.
1699 */
1700 ASSERT(!(flags & XFS_BLI_CANCEL));
1701 return 0;
1702 }
1703
1704 /*
1705 * Search for an entry in the buffer cancel table that
1706 * matches our buffer.
1707 */
1708 prevp = NULL;
1709 while (bcp != NULL) {
1710 if (bcp->bc_blkno == blkno && bcp->bc_len == len) {
1711 /*
1712 * We've go a match, so return 1 so that the
1713 * recovery of this buffer is cancelled.
1714 * If this buffer is actually a buffer cancel
1715 * log item, then decrement the refcount on the
1716 * one in the table and remove it if this is the
1717 * last reference.
1718 */
1719 if (flags & XFS_BLI_CANCEL) {
1720 bcp->bc_refcount--;
1721 if (bcp->bc_refcount == 0) {
1722 if (prevp == NULL) {
1723 *bucket = bcp->bc_next;
1724 } else {
1725 prevp->bc_next = bcp->bc_next;
1726 }
1727 kmem_free(bcp,
1728 sizeof(xfs_buf_cancel_t));
1729 }
1730 }
1731 return 1;
1732 }
1733 prevp = bcp;
1734 bcp = bcp->bc_next;
1735 }
1736 /*
1737 * We didn't find a corresponding entry in the table, so
1738 * return 0 so that the buffer is NOT cancelled.
1739 */
1740 ASSERT(!(flags & XFS_BLI_CANCEL));
1741 return 0;
1742}
1743
1744STATIC int
1745xlog_recover_do_buffer_pass2(
1746 xlog_t *log,
1747 xfs_buf_log_format_t *buf_f)
1748{
1749 xfs_buf_log_format_v1_t *obuf_f;
1750 xfs_daddr_t blkno = 0;
1751 ushort flags = 0;
1752 uint len = 0;
1753
1754 switch (buf_f->blf_type) {
1755 case XFS_LI_BUF:
1756 blkno = buf_f->blf_blkno;
1757 flags = buf_f->blf_flags;
1758 len = buf_f->blf_len;
1759 break;
1760 case XFS_LI_6_1_BUF:
1761 case XFS_LI_5_3_BUF:
1762 obuf_f = (xfs_buf_log_format_v1_t*)buf_f;
1763 blkno = (xfs_daddr_t) obuf_f->blf_blkno;
1764 flags = obuf_f->blf_flags;
1765 len = (xfs_daddr_t) obuf_f->blf_len;
1766 break;
1767 }
1768
1769 return xlog_check_buffer_cancelled(log, blkno, len, flags);
1770}
1771
1772/*
1773 * Perform recovery for a buffer full of inodes. In these buffers,
1774 * the only data which should be recovered is that which corresponds
1775 * to the di_next_unlinked pointers in the on disk inode structures.
1776 * The rest of the data for the inodes is always logged through the
1777 * inodes themselves rather than the inode buffer and is recovered
1778 * in xlog_recover_do_inode_trans().
1779 *
1780 * The only time when buffers full of inodes are fully recovered is
1781 * when the buffer is full of newly allocated inodes. In this case
1782 * the buffer will not be marked as an inode buffer and so will be
1783 * sent to xlog_recover_do_reg_buffer() below during recovery.
1784 */
1785STATIC int
1786xlog_recover_do_inode_buffer(
1787 xfs_mount_t *mp,
1788 xlog_recover_item_t *item,
1789 xfs_buf_t *bp,
1790 xfs_buf_log_format_t *buf_f)
1791{
1792 int i;
1793 int item_index;
1794 int bit;
1795 int nbits;
1796 int reg_buf_offset;
1797 int reg_buf_bytes;
1798 int next_unlinked_offset;
1799 int inodes_per_buf;
1800 xfs_agino_t *logged_nextp;
1801 xfs_agino_t *buffer_nextp;
1802 xfs_buf_log_format_v1_t *obuf_f;
1803 unsigned int *data_map = NULL;
1804 unsigned int map_size = 0;
1805
1806 switch (buf_f->blf_type) {
1807 case XFS_LI_BUF:
1808 data_map = buf_f->blf_data_map;
1809 map_size = buf_f->blf_map_size;
1810 break;
1811 case XFS_LI_6_1_BUF:
1812 case XFS_LI_5_3_BUF:
1813 obuf_f = (xfs_buf_log_format_v1_t*)buf_f;
1814 data_map = obuf_f->blf_data_map;
1815 map_size = obuf_f->blf_map_size;
1816 break;
1817 }
1818 /*
1819 * Set the variables corresponding to the current region to
1820 * 0 so that we'll initialize them on the first pass through
1821 * the loop.
1822 */
1823 reg_buf_offset = 0;
1824 reg_buf_bytes = 0;
1825 bit = 0;
1826 nbits = 0;
1827 item_index = 0;
1828 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1829 for (i = 0; i < inodes_per_buf; i++) {
1830 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1831 offsetof(xfs_dinode_t, di_next_unlinked);
1832
1833 while (next_unlinked_offset >=
1834 (reg_buf_offset + reg_buf_bytes)) {
1835 /*
1836 * The next di_next_unlinked field is beyond
1837 * the current logged region. Find the next
1838 * logged region that contains or is beyond
1839 * the current di_next_unlinked field.
1840 */
1841 bit += nbits;
1842 bit = xfs_next_bit(data_map, map_size, bit);
1843
1844 /*
1845 * If there are no more logged regions in the
1846 * buffer, then we're done.
1847 */
1848 if (bit == -1) {
1849 return 0;
1850 }
1851
1852 nbits = xfs_contig_bits(data_map, map_size,
1853 bit);
1854 ASSERT(nbits > 0);
1855 reg_buf_offset = bit << XFS_BLI_SHIFT;
1856 reg_buf_bytes = nbits << XFS_BLI_SHIFT;
1857 item_index++;
1858 }
1859
1860 /*
1861 * If the current logged region starts after the current
1862 * di_next_unlinked field, then move on to the next
1863 * di_next_unlinked field.
1864 */
1865 if (next_unlinked_offset < reg_buf_offset) {
1866 continue;
1867 }
1868
1869 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1870 ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0);
1871 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1872
1873 /*
1874 * The current logged region contains a copy of the
1875 * current di_next_unlinked field. Extract its value
1876 * and copy it to the buffer copy.
1877 */
1878 logged_nextp = (xfs_agino_t *)
1879 ((char *)(item->ri_buf[item_index].i_addr) +
1880 (next_unlinked_offset - reg_buf_offset));
1881 if (unlikely(*logged_nextp == 0)) {
1882 xfs_fs_cmn_err(CE_ALERT, mp,
1883 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1884 item, bp);
1885 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1886 XFS_ERRLEVEL_LOW, mp);
1887 return XFS_ERROR(EFSCORRUPTED);
1888 }
1889
1890 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1891 next_unlinked_offset);
Tim Shimmin87c199c2006-06-09 14:56:16 +10001892 *buffer_nextp = *logged_nextp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 }
1894
1895 return 0;
1896}
1897
1898/*
1899 * Perform a 'normal' buffer recovery. Each logged region of the
1900 * buffer should be copied over the corresponding region in the
1901 * given buffer. The bitmap in the buf log format structure indicates
1902 * where to place the logged data.
1903 */
1904/*ARGSUSED*/
1905STATIC void
1906xlog_recover_do_reg_buffer(
1907 xfs_mount_t *mp,
1908 xlog_recover_item_t *item,
1909 xfs_buf_t *bp,
1910 xfs_buf_log_format_t *buf_f)
1911{
1912 int i;
1913 int bit;
1914 int nbits;
1915 xfs_buf_log_format_v1_t *obuf_f;
1916 unsigned int *data_map = NULL;
1917 unsigned int map_size = 0;
1918 int error;
1919
1920 switch (buf_f->blf_type) {
1921 case XFS_LI_BUF:
1922 data_map = buf_f->blf_data_map;
1923 map_size = buf_f->blf_map_size;
1924 break;
1925 case XFS_LI_6_1_BUF:
1926 case XFS_LI_5_3_BUF:
1927 obuf_f = (xfs_buf_log_format_v1_t*)buf_f;
1928 data_map = obuf_f->blf_data_map;
1929 map_size = obuf_f->blf_map_size;
1930 break;
1931 }
1932 bit = 0;
1933 i = 1; /* 0 is the buf format structure */
1934 while (1) {
1935 bit = xfs_next_bit(data_map, map_size, bit);
1936 if (bit == -1)
1937 break;
1938 nbits = xfs_contig_bits(data_map, map_size, bit);
1939 ASSERT(nbits > 0);
1940 ASSERT(item->ri_buf[i].i_addr != 0);
1941 ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0);
1942 ASSERT(XFS_BUF_COUNT(bp) >=
1943 ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT));
1944
1945 /*
1946 * Do a sanity check if this is a dquot buffer. Just checking
1947 * the first dquot in the buffer should do. XXXThis is
1948 * probably a good thing to do for other buf types also.
1949 */
1950 error = 0;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001951 if (buf_f->blf_flags &
1952 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 error = xfs_qm_dqcheck((xfs_disk_dquot_t *)
1954 item->ri_buf[i].i_addr,
1955 -1, 0, XFS_QMOPT_DOWARN,
1956 "dquot_buf_recover");
1957 }
1958 if (!error)
1959 memcpy(xfs_buf_offset(bp,
1960 (uint)bit << XFS_BLI_SHIFT), /* dest */
1961 item->ri_buf[i].i_addr, /* source */
1962 nbits<<XFS_BLI_SHIFT); /* length */
1963 i++;
1964 bit += nbits;
1965 }
1966
1967 /* Shouldn't be any more regions */
1968 ASSERT(i == item->ri_total);
1969}
1970
1971/*
1972 * Do some primitive error checking on ondisk dquot data structures.
1973 */
1974int
1975xfs_qm_dqcheck(
1976 xfs_disk_dquot_t *ddq,
1977 xfs_dqid_t id,
1978 uint type, /* used only when IO_dorepair is true */
1979 uint flags,
1980 char *str)
1981{
1982 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
1983 int errs = 0;
1984
1985 /*
1986 * We can encounter an uninitialized dquot buffer for 2 reasons:
1987 * 1. If we crash while deleting the quotainode(s), and those blks got
1988 * used for user data. This is because we take the path of regular
1989 * file deletion; however, the size field of quotainodes is never
1990 * updated, so all the tricks that we play in itruncate_finish
1991 * don't quite matter.
1992 *
1993 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1994 * But the allocation will be replayed so we'll end up with an
1995 * uninitialized quota block.
1996 *
1997 * This is all fine; things are still consistent, and we haven't lost
1998 * any quota information. Just don't complain about bad dquot blks.
1999 */
Christoph Hellwig1149d962005-11-02 15:01:12 +11002000 if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 if (flags & XFS_QMOPT_DOWARN)
2002 cmn_err(CE_ALERT,
2003 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002004 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 errs++;
2006 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11002007 if (ddq->d_version != XFS_DQUOT_VERSION) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 if (flags & XFS_QMOPT_DOWARN)
2009 cmn_err(CE_ALERT,
2010 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002011 str, id, ddq->d_version, XFS_DQUOT_VERSION);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 errs++;
2013 }
2014
Christoph Hellwig1149d962005-11-02 15:01:12 +11002015 if (ddq->d_flags != XFS_DQ_USER &&
2016 ddq->d_flags != XFS_DQ_PROJ &&
2017 ddq->d_flags != XFS_DQ_GROUP) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 if (flags & XFS_QMOPT_DOWARN)
2019 cmn_err(CE_ALERT,
2020 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002021 str, id, ddq->d_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 errs++;
2023 }
2024
Christoph Hellwig1149d962005-11-02 15:01:12 +11002025 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 if (flags & XFS_QMOPT_DOWARN)
2027 cmn_err(CE_ALERT,
2028 "%s : ondisk-dquot 0x%p, ID mismatch: "
2029 "0x%x expected, found id 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002030 str, ddq, id, be32_to_cpu(ddq->d_id));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 errs++;
2032 }
2033
2034 if (!errs && ddq->d_id) {
Christoph Hellwig1149d962005-11-02 15:01:12 +11002035 if (ddq->d_blk_softlimit &&
2036 be64_to_cpu(ddq->d_bcount) >=
2037 be64_to_cpu(ddq->d_blk_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 if (!ddq->d_btimer) {
2039 if (flags & XFS_QMOPT_DOWARN)
2040 cmn_err(CE_ALERT,
2041 "%s : Dquot ID 0x%x (0x%p) "
2042 "BLK TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002043 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 errs++;
2045 }
2046 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11002047 if (ddq->d_ino_softlimit &&
2048 be64_to_cpu(ddq->d_icount) >=
2049 be64_to_cpu(ddq->d_ino_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 if (!ddq->d_itimer) {
2051 if (flags & XFS_QMOPT_DOWARN)
2052 cmn_err(CE_ALERT,
2053 "%s : Dquot ID 0x%x (0x%p) "
2054 "INODE TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002055 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 errs++;
2057 }
2058 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11002059 if (ddq->d_rtb_softlimit &&
2060 be64_to_cpu(ddq->d_rtbcount) >=
2061 be64_to_cpu(ddq->d_rtb_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 if (!ddq->d_rtbtimer) {
2063 if (flags & XFS_QMOPT_DOWARN)
2064 cmn_err(CE_ALERT,
2065 "%s : Dquot ID 0x%x (0x%p) "
2066 "RTBLK TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002067 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 errs++;
2069 }
2070 }
2071 }
2072
2073 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2074 return errs;
2075
2076 if (flags & XFS_QMOPT_DOWARN)
2077 cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id);
2078
2079 /*
2080 * Typically, a repair is only requested by quotacheck.
2081 */
2082 ASSERT(id != -1);
2083 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2084 memset(d, 0, sizeof(xfs_dqblk_t));
Christoph Hellwig1149d962005-11-02 15:01:12 +11002085
2086 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2087 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2088 d->dd_diskdq.d_flags = type;
2089 d->dd_diskdq.d_id = cpu_to_be32(id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090
2091 return errs;
2092}
2093
2094/*
2095 * Perform a dquot buffer recovery.
2096 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2097 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2098 * Else, treat it as a regular buffer and do recovery.
2099 */
2100STATIC void
2101xlog_recover_do_dquot_buffer(
2102 xfs_mount_t *mp,
2103 xlog_t *log,
2104 xlog_recover_item_t *item,
2105 xfs_buf_t *bp,
2106 xfs_buf_log_format_t *buf_f)
2107{
2108 uint type;
2109
2110 /*
2111 * Filesystems are required to send in quota flags at mount time.
2112 */
2113 if (mp->m_qflags == 0) {
2114 return;
2115 }
2116
2117 type = 0;
2118 if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF)
2119 type |= XFS_DQ_USER;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002120 if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF)
2121 type |= XFS_DQ_PROJ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF)
2123 type |= XFS_DQ_GROUP;
2124 /*
2125 * This type of quotas was turned off, so ignore this buffer
2126 */
2127 if (log->l_quotaoffs_flag & type)
2128 return;
2129
2130 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2131}
2132
2133/*
2134 * This routine replays a modification made to a buffer at runtime.
2135 * There are actually two types of buffer, regular and inode, which
2136 * are handled differently. Inode buffers are handled differently
2137 * in that we only recover a specific set of data from them, namely
2138 * the inode di_next_unlinked fields. This is because all other inode
2139 * data is actually logged via inode records and any data we replay
2140 * here which overlaps that may be stale.
2141 *
2142 * When meta-data buffers are freed at run time we log a buffer item
2143 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
2144 * of the buffer in the log should not be replayed at recovery time.
2145 * This is so that if the blocks covered by the buffer are reused for
2146 * file data before we crash we don't end up replaying old, freed
2147 * meta-data into a user's file.
2148 *
2149 * To handle the cancellation of buffer log items, we make two passes
2150 * over the log during recovery. During the first we build a table of
2151 * those buffers which have been cancelled, and during the second we
2152 * only replay those buffers which do not have corresponding cancel
2153 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2154 * for more details on the implementation of the table of cancel records.
2155 */
2156STATIC int
2157xlog_recover_do_buffer_trans(
2158 xlog_t *log,
2159 xlog_recover_item_t *item,
2160 int pass)
2161{
2162 xfs_buf_log_format_t *buf_f;
2163 xfs_buf_log_format_v1_t *obuf_f;
2164 xfs_mount_t *mp;
2165 xfs_buf_t *bp;
2166 int error;
2167 int cancel;
2168 xfs_daddr_t blkno;
2169 int len;
2170 ushort flags;
2171
2172 buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr;
2173
2174 if (pass == XLOG_RECOVER_PASS1) {
2175 /*
2176 * In this pass we're only looking for buf items
2177 * with the XFS_BLI_CANCEL bit set.
2178 */
2179 xlog_recover_do_buffer_pass1(log, buf_f);
2180 return 0;
2181 } else {
2182 /*
2183 * In this pass we want to recover all the buffers
2184 * which have not been cancelled and are not
2185 * cancellation buffers themselves. The routine
2186 * we call here will tell us whether or not to
2187 * continue with the replay of this buffer.
2188 */
2189 cancel = xlog_recover_do_buffer_pass2(log, buf_f);
2190 if (cancel) {
2191 return 0;
2192 }
2193 }
2194 switch (buf_f->blf_type) {
2195 case XFS_LI_BUF:
2196 blkno = buf_f->blf_blkno;
2197 len = buf_f->blf_len;
2198 flags = buf_f->blf_flags;
2199 break;
2200 case XFS_LI_6_1_BUF:
2201 case XFS_LI_5_3_BUF:
2202 obuf_f = (xfs_buf_log_format_v1_t*)buf_f;
2203 blkno = obuf_f->blf_blkno;
2204 len = obuf_f->blf_len;
2205 flags = obuf_f->blf_flags;
2206 break;
2207 default:
2208 xfs_fs_cmn_err(CE_ALERT, log->l_mp,
Nathan Scottfc1f8c12005-11-02 11:44:33 +11002209 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2210 buf_f->blf_type, log->l_mp->m_logname ?
2211 log->l_mp->m_logname : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2213 XFS_ERRLEVEL_LOW, log->l_mp);
2214 return XFS_ERROR(EFSCORRUPTED);
2215 }
2216
2217 mp = log->l_mp;
2218 if (flags & XFS_BLI_INODE_BUF) {
2219 bp = xfs_buf_read_flags(mp->m_ddev_targp, blkno, len,
2220 XFS_BUF_LOCK);
2221 } else {
2222 bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, 0);
2223 }
2224 if (XFS_BUF_ISERROR(bp)) {
2225 xfs_ioerror_alert("xlog_recover_do..(read#1)", log->l_mp,
2226 bp, blkno);
2227 error = XFS_BUF_GETERROR(bp);
2228 xfs_buf_relse(bp);
2229 return error;
2230 }
2231
2232 error = 0;
2233 if (flags & XFS_BLI_INODE_BUF) {
2234 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002235 } else if (flags &
2236 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2238 } else {
2239 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2240 }
2241 if (error)
2242 return XFS_ERROR(error);
2243
2244 /*
2245 * Perform delayed write on the buffer. Asynchronous writes will be
2246 * slower when taking into account all the buffers to be flushed.
2247 *
2248 * Also make sure that only inode buffers with good sizes stay in
2249 * the buffer cache. The kernel moves inodes in buffers of 1 block
2250 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2251 * buffers in the log can be a different size if the log was generated
2252 * by an older kernel using unclustered inode buffers or a newer kernel
2253 * running with a different inode cluster size. Regardless, if the
2254 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2255 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2256 * the buffer out of the buffer cache so that the buffer won't
2257 * overlap with future reads of those inodes.
2258 */
2259 if (XFS_DINODE_MAGIC ==
2260 INT_GET(*((__uint16_t *)(xfs_buf_offset(bp, 0))), ARCH_CONVERT) &&
2261 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2262 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2263 XFS_BUF_STALE(bp);
2264 error = xfs_bwrite(mp, bp);
2265 } else {
2266 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2267 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2268 XFS_BUF_SET_FSPRIVATE(bp, mp);
2269 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2270 xfs_bdwrite(mp, bp);
2271 }
2272
2273 return (error);
2274}
2275
2276STATIC int
2277xlog_recover_do_inode_trans(
2278 xlog_t *log,
2279 xlog_recover_item_t *item,
2280 int pass)
2281{
2282 xfs_inode_log_format_t *in_f;
2283 xfs_mount_t *mp;
2284 xfs_buf_t *bp;
2285 xfs_imap_t imap;
2286 xfs_dinode_t *dip;
2287 xfs_ino_t ino;
2288 int len;
2289 xfs_caddr_t src;
2290 xfs_caddr_t dest;
2291 int error;
2292 int attr_index;
2293 uint fields;
2294 xfs_dinode_core_t *dicp;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002295 int need_free = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296
2297 if (pass == XLOG_RECOVER_PASS1) {
2298 return 0;
2299 }
2300
Tim Shimmin6d192a92006-06-09 14:55:38 +10002301 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2302 in_f = (xfs_inode_log_format_t *)item->ri_buf[0].i_addr;
2303 } else {
2304 in_f = (xfs_inode_log_format_t *)kmem_alloc(
2305 sizeof(xfs_inode_log_format_t), KM_SLEEP);
2306 need_free = 1;
2307 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2308 if (error)
2309 goto error;
2310 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 ino = in_f->ilf_ino;
2312 mp = log->l_mp;
2313 if (ITEM_TYPE(item) == XFS_LI_INODE) {
2314 imap.im_blkno = (xfs_daddr_t)in_f->ilf_blkno;
2315 imap.im_len = in_f->ilf_len;
2316 imap.im_boffset = in_f->ilf_boffset;
2317 } else {
2318 /*
2319 * It's an old inode format record. We don't know where
2320 * its cluster is located on disk, and we can't allow
2321 * xfs_imap() to figure it out because the inode btrees
2322 * are not ready to be used. Therefore do not pass the
2323 * XFS_IMAP_LOOKUP flag to xfs_imap(). This will give
2324 * us only the single block in which the inode lives
2325 * rather than its cluster, so we must make sure to
2326 * invalidate the buffer when we write it out below.
2327 */
2328 imap.im_blkno = 0;
2329 xfs_imap(log->l_mp, NULL, ino, &imap, 0);
2330 }
2331
2332 /*
2333 * Inode buffers can be freed, look out for it,
2334 * and do not replay the inode.
2335 */
Tim Shimmin6d192a92006-06-09 14:55:38 +10002336 if (xlog_check_buffer_cancelled(log, imap.im_blkno, imap.im_len, 0)) {
2337 error = 0;
2338 goto error;
2339 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340
2341 bp = xfs_buf_read_flags(mp->m_ddev_targp, imap.im_blkno, imap.im_len,
2342 XFS_BUF_LOCK);
2343 if (XFS_BUF_ISERROR(bp)) {
2344 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
2345 bp, imap.im_blkno);
2346 error = XFS_BUF_GETERROR(bp);
2347 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002348 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 }
2350 error = 0;
2351 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2352 dip = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
2353
2354 /*
2355 * Make sure the place we're flushing out to really looks
2356 * like an inode!
2357 */
2358 if (unlikely(INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC)) {
2359 xfs_buf_relse(bp);
2360 xfs_fs_cmn_err(CE_ALERT, mp,
2361 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2362 dip, bp, ino);
2363 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2364 XFS_ERRLEVEL_LOW, mp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002365 error = EFSCORRUPTED;
2366 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 }
2368 dicp = (xfs_dinode_core_t*)(item->ri_buf[1].i_addr);
2369 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2370 xfs_buf_relse(bp);
2371 xfs_fs_cmn_err(CE_ALERT, mp,
2372 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2373 item, ino);
2374 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2375 XFS_ERRLEVEL_LOW, mp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002376 error = EFSCORRUPTED;
2377 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 }
2379
2380 /* Skip replay when the on disk inode is newer than the log one */
2381 if (dicp->di_flushiter <
2382 INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT)) {
2383 /*
2384 * Deal with the wrap case, DI_MAX_FLUSH is less
2385 * than smaller numbers
2386 */
2387 if ((INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT)
2388 == DI_MAX_FLUSH) &&
2389 (dicp->di_flushiter < (DI_MAX_FLUSH>>1))) {
2390 /* do nothing */
2391 } else {
2392 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002393 error = 0;
2394 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 }
2396 }
2397 /* Take the opportunity to reset the flush iteration count */
2398 dicp->di_flushiter = 0;
2399
2400 if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) {
2401 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2402 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2403 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2404 XFS_ERRLEVEL_LOW, mp, dicp);
2405 xfs_buf_relse(bp);
2406 xfs_fs_cmn_err(CE_ALERT, mp,
2407 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2408 item, dip, bp, ino);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002409 error = EFSCORRUPTED;
2410 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 }
2412 } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) {
2413 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2414 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2415 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2416 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2417 XFS_ERRLEVEL_LOW, mp, dicp);
2418 xfs_buf_relse(bp);
2419 xfs_fs_cmn_err(CE_ALERT, mp,
2420 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2421 item, dip, bp, ino);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002422 error = EFSCORRUPTED;
2423 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 }
2425 }
2426 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2427 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2428 XFS_ERRLEVEL_LOW, mp, dicp);
2429 xfs_buf_relse(bp);
2430 xfs_fs_cmn_err(CE_ALERT, mp,
2431 "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",
2432 item, dip, bp, ino,
2433 dicp->di_nextents + dicp->di_anextents,
2434 dicp->di_nblocks);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002435 error = EFSCORRUPTED;
2436 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 }
2438 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2439 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2440 XFS_ERRLEVEL_LOW, mp, dicp);
2441 xfs_buf_relse(bp);
2442 xfs_fs_cmn_err(CE_ALERT, mp,
2443 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2444 item, dip, bp, ino, dicp->di_forkoff);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002445 error = EFSCORRUPTED;
2446 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 }
2448 if (unlikely(item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t))) {
2449 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
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 record length %d, rec ptr 0x%p",
2454 item->ri_buf[1].i_len, item);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002455 error = EFSCORRUPTED;
2456 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 }
2458
2459 /* The core is in in-core format */
2460 xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
2461 (xfs_dinode_core_t*)item->ri_buf[1].i_addr, -1);
2462
2463 /* the rest is in on-disk format */
2464 if (item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t)) {
2465 memcpy((xfs_caddr_t) dip + sizeof(xfs_dinode_core_t),
2466 item->ri_buf[1].i_addr + sizeof(xfs_dinode_core_t),
2467 item->ri_buf[1].i_len - sizeof(xfs_dinode_core_t));
2468 }
2469
2470 fields = in_f->ilf_fields;
2471 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2472 case XFS_ILOG_DEV:
2473 INT_SET(dip->di_u.di_dev, ARCH_CONVERT, in_f->ilf_u.ilfu_rdev);
2474
2475 break;
2476 case XFS_ILOG_UUID:
2477 dip->di_u.di_muuid = in_f->ilf_u.ilfu_uuid;
2478 break;
2479 }
2480
2481 if (in_f->ilf_size == 2)
2482 goto write_inode_buffer;
2483 len = item->ri_buf[2].i_len;
2484 src = item->ri_buf[2].i_addr;
2485 ASSERT(in_f->ilf_size <= 4);
2486 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2487 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2488 (len == in_f->ilf_dsize));
2489
2490 switch (fields & XFS_ILOG_DFORK) {
2491 case XFS_ILOG_DDATA:
2492 case XFS_ILOG_DEXT:
2493 memcpy(&dip->di_u, src, len);
2494 break;
2495
2496 case XFS_ILOG_DBROOT:
2497 xfs_bmbt_to_bmdr((xfs_bmbt_block_t *)src, len,
2498 &(dip->di_u.di_bmbt),
2499 XFS_DFORK_DSIZE(dip, mp));
2500 break;
2501
2502 default:
2503 /*
2504 * There are no data fork flags set.
2505 */
2506 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2507 break;
2508 }
2509
2510 /*
2511 * If we logged any attribute data, recover it. There may or
2512 * may not have been any other non-core data logged in this
2513 * transaction.
2514 */
2515 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2516 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2517 attr_index = 3;
2518 } else {
2519 attr_index = 2;
2520 }
2521 len = item->ri_buf[attr_index].i_len;
2522 src = item->ri_buf[attr_index].i_addr;
2523 ASSERT(len == in_f->ilf_asize);
2524
2525 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2526 case XFS_ILOG_ADATA:
2527 case XFS_ILOG_AEXT:
2528 dest = XFS_DFORK_APTR(dip);
2529 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2530 memcpy(dest, src, len);
2531 break;
2532
2533 case XFS_ILOG_ABROOT:
2534 dest = XFS_DFORK_APTR(dip);
2535 xfs_bmbt_to_bmdr((xfs_bmbt_block_t *)src, len,
2536 (xfs_bmdr_block_t*)dest,
2537 XFS_DFORK_ASIZE(dip, mp));
2538 break;
2539
2540 default:
2541 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2542 ASSERT(0);
2543 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002544 error = EIO;
2545 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 }
2547 }
2548
2549write_inode_buffer:
2550 if (ITEM_TYPE(item) == XFS_LI_INODE) {
2551 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2552 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2553 XFS_BUF_SET_FSPRIVATE(bp, mp);
2554 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2555 xfs_bdwrite(mp, bp);
2556 } else {
2557 XFS_BUF_STALE(bp);
2558 error = xfs_bwrite(mp, bp);
2559 }
2560
Tim Shimmin6d192a92006-06-09 14:55:38 +10002561error:
2562 if (need_free)
2563 kmem_free(in_f, sizeof(*in_f));
2564 return XFS_ERROR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565}
2566
2567/*
2568 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2569 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2570 * of that type.
2571 */
2572STATIC int
2573xlog_recover_do_quotaoff_trans(
2574 xlog_t *log,
2575 xlog_recover_item_t *item,
2576 int pass)
2577{
2578 xfs_qoff_logformat_t *qoff_f;
2579
2580 if (pass == XLOG_RECOVER_PASS2) {
2581 return (0);
2582 }
2583
2584 qoff_f = (xfs_qoff_logformat_t *)item->ri_buf[0].i_addr;
2585 ASSERT(qoff_f);
2586
2587 /*
2588 * The logitem format's flag tells us if this was user quotaoff,
Nathan Scott77a7cce2006-01-11 15:35:57 +11002589 * group/project quotaoff or both.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 */
2591 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2592 log->l_quotaoffs_flag |= XFS_DQ_USER;
Nathan Scott77a7cce2006-01-11 15:35:57 +11002593 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2594 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2596 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2597
2598 return (0);
2599}
2600
2601/*
2602 * Recover a dquot record
2603 */
2604STATIC int
2605xlog_recover_do_dquot_trans(
2606 xlog_t *log,
2607 xlog_recover_item_t *item,
2608 int pass)
2609{
2610 xfs_mount_t *mp;
2611 xfs_buf_t *bp;
2612 struct xfs_disk_dquot *ddq, *recddq;
2613 int error;
2614 xfs_dq_logformat_t *dq_f;
2615 uint type;
2616
2617 if (pass == XLOG_RECOVER_PASS1) {
2618 return 0;
2619 }
2620 mp = log->l_mp;
2621
2622 /*
2623 * Filesystems are required to send in quota flags at mount time.
2624 */
2625 if (mp->m_qflags == 0)
2626 return (0);
2627
2628 recddq = (xfs_disk_dquot_t *)item->ri_buf[1].i_addr;
2629 ASSERT(recddq);
2630 /*
2631 * This type of quotas was turned off, so ignore this record.
2632 */
2633 type = INT_GET(recddq->d_flags, ARCH_CONVERT) &
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002634 (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 ASSERT(type);
2636 if (log->l_quotaoffs_flag & type)
2637 return (0);
2638
2639 /*
2640 * At this point we know that quota was _not_ turned off.
2641 * Since the mount flags are not indicating to us otherwise, this
2642 * must mean that quota is on, and the dquot needs to be replayed.
2643 * Remember that we may not have fully recovered the superblock yet,
2644 * so we can't do the usual trick of looking at the SB quota bits.
2645 *
2646 * The other possibility, of course, is that the quota subsystem was
2647 * removed since the last mount - ENOSYS.
2648 */
2649 dq_f = (xfs_dq_logformat_t *)item->ri_buf[0].i_addr;
2650 ASSERT(dq_f);
2651 if ((error = xfs_qm_dqcheck(recddq,
2652 dq_f->qlf_id,
2653 0, XFS_QMOPT_DOWARN,
2654 "xlog_recover_do_dquot_trans (log copy)"))) {
2655 return XFS_ERROR(EIO);
2656 }
2657 ASSERT(dq_f->qlf_len == 1);
2658
2659 error = xfs_read_buf(mp, mp->m_ddev_targp,
2660 dq_f->qlf_blkno,
2661 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2662 0, &bp);
2663 if (error) {
2664 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
2665 bp, dq_f->qlf_blkno);
2666 return error;
2667 }
2668 ASSERT(bp);
2669 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2670
2671 /*
2672 * At least the magic num portion should be on disk because this
2673 * was among a chunk of dquots created earlier, and we did some
2674 * minimal initialization then.
2675 */
2676 if (xfs_qm_dqcheck(ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2677 "xlog_recover_do_dquot_trans")) {
2678 xfs_buf_relse(bp);
2679 return XFS_ERROR(EIO);
2680 }
2681
2682 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2683
2684 ASSERT(dq_f->qlf_size == 2);
2685 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2686 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2687 XFS_BUF_SET_FSPRIVATE(bp, mp);
2688 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2689 xfs_bdwrite(mp, bp);
2690
2691 return (0);
2692}
2693
2694/*
2695 * This routine is called to create an in-core extent free intent
2696 * item from the efi format structure which was logged on disk.
2697 * It allocates an in-core efi, copies the extents from the format
2698 * structure into it, and adds the efi to the AIL with the given
2699 * LSN.
2700 */
Tim Shimmin6d192a92006-06-09 14:55:38 +10002701STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702xlog_recover_do_efi_trans(
2703 xlog_t *log,
2704 xlog_recover_item_t *item,
2705 xfs_lsn_t lsn,
2706 int pass)
2707{
Tim Shimmin6d192a92006-06-09 14:55:38 +10002708 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 xfs_mount_t *mp;
2710 xfs_efi_log_item_t *efip;
2711 xfs_efi_log_format_t *efi_formatp;
2712 SPLDECL(s);
2713
2714 if (pass == XLOG_RECOVER_PASS1) {
Tim Shimmin6d192a92006-06-09 14:55:38 +10002715 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 }
2717
2718 efi_formatp = (xfs_efi_log_format_t *)item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719
2720 mp = log->l_mp;
2721 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002722 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2723 &(efip->efi_format)))) {
2724 xfs_efi_item_free(efip);
2725 return error;
2726 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727 efip->efi_next_extent = efi_formatp->efi_nextents;
2728 efip->efi_flags |= XFS_EFI_COMMITTED;
2729
2730 AIL_LOCK(mp,s);
2731 /*
2732 * xfs_trans_update_ail() drops the AIL lock.
2733 */
2734 xfs_trans_update_ail(mp, (xfs_log_item_t *)efip, lsn, s);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002735 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736}
2737
2738
2739/*
2740 * This routine is called when an efd format structure is found in
2741 * a committed transaction in the log. It's purpose is to cancel
2742 * the corresponding efi if it was still in the log. To do this
2743 * it searches the AIL for the efi with an id equal to that in the
2744 * efd format structure. If we find it, we remove the efi from the
2745 * AIL and free it.
2746 */
2747STATIC void
2748xlog_recover_do_efd_trans(
2749 xlog_t *log,
2750 xlog_recover_item_t *item,
2751 int pass)
2752{
2753 xfs_mount_t *mp;
2754 xfs_efd_log_format_t *efd_formatp;
2755 xfs_efi_log_item_t *efip = NULL;
2756 xfs_log_item_t *lip;
2757 int gen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758 __uint64_t efi_id;
2759 SPLDECL(s);
2760
2761 if (pass == XLOG_RECOVER_PASS1) {
2762 return;
2763 }
2764
2765 efd_formatp = (xfs_efd_log_format_t *)item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002766 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2767 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2768 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2769 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 efi_id = efd_formatp->efd_efi_id;
2771
2772 /*
2773 * Search for the efi with the id in the efd format structure
2774 * in the AIL.
2775 */
2776 mp = log->l_mp;
2777 AIL_LOCK(mp,s);
2778 lip = xfs_trans_first_ail(mp, &gen);
2779 while (lip != NULL) {
2780 if (lip->li_type == XFS_LI_EFI) {
2781 efip = (xfs_efi_log_item_t *)lip;
2782 if (efip->efi_format.efi_id == efi_id) {
2783 /*
2784 * xfs_trans_delete_ail() drops the
2785 * AIL lock.
2786 */
2787 xfs_trans_delete_ail(mp, lip, s);
2788 break;
2789 }
2790 }
2791 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
2792 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793
2794 /*
2795 * If we found it, then free it up. If it wasn't there, it
2796 * must have been overwritten in the log. Oh well.
2797 */
2798 if (lip != NULL) {
Christoph Hellwig7d795ca2005-06-21 15:41:19 +10002799 xfs_efi_item_free(efip);
2800 } else {
2801 AIL_UNLOCK(mp, s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802 }
2803}
2804
2805/*
2806 * Perform the transaction
2807 *
2808 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2809 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2810 */
2811STATIC int
2812xlog_recover_do_trans(
2813 xlog_t *log,
2814 xlog_recover_t *trans,
2815 int pass)
2816{
2817 int error = 0;
2818 xlog_recover_item_t *item, *first_item;
2819
2820 if ((error = xlog_recover_reorder_trans(log, trans)))
2821 return error;
2822 first_item = item = trans->r_itemq;
2823 do {
2824 /*
2825 * we don't need to worry about the block number being
2826 * truncated in > 1 TB buffers because in user-land,
2827 * we're now n32 or 64-bit so xfs_daddr_t is 64-bits so
Nathan Scottc41564b2006-03-29 08:55:14 +10002828 * the blknos will get through the user-mode buffer
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829 * cache properly. The only bad case is o32 kernels
2830 * where xfs_daddr_t is 32-bits but mount will warn us
2831 * off a > 1 TB filesystem before we get here.
2832 */
2833 if ((ITEM_TYPE(item) == XFS_LI_BUF) ||
2834 (ITEM_TYPE(item) == XFS_LI_6_1_BUF) ||
2835 (ITEM_TYPE(item) == XFS_LI_5_3_BUF)) {
2836 if ((error = xlog_recover_do_buffer_trans(log, item,
2837 pass)))
2838 break;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002839 } else if ((ITEM_TYPE(item) == XFS_LI_INODE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840 if ((error = xlog_recover_do_inode_trans(log, item,
2841 pass)))
2842 break;
2843 } else if (ITEM_TYPE(item) == XFS_LI_EFI) {
Tim Shimmin6d192a92006-06-09 14:55:38 +10002844 if ((error = xlog_recover_do_efi_trans(log, item, trans->r_lsn,
2845 pass)))
2846 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847 } else if (ITEM_TYPE(item) == XFS_LI_EFD) {
2848 xlog_recover_do_efd_trans(log, item, pass);
2849 } else if (ITEM_TYPE(item) == XFS_LI_DQUOT) {
2850 if ((error = xlog_recover_do_dquot_trans(log, item,
2851 pass)))
2852 break;
2853 } else if ((ITEM_TYPE(item) == XFS_LI_QUOTAOFF)) {
2854 if ((error = xlog_recover_do_quotaoff_trans(log, item,
2855 pass)))
2856 break;
2857 } else {
2858 xlog_warn("XFS: xlog_recover_do_trans");
2859 ASSERT(0);
2860 error = XFS_ERROR(EIO);
2861 break;
2862 }
2863 item = item->ri_next;
2864 } while (first_item != item);
2865
2866 return error;
2867}
2868
2869/*
2870 * Free up any resources allocated by the transaction
2871 *
2872 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2873 */
2874STATIC void
2875xlog_recover_free_trans(
2876 xlog_recover_t *trans)
2877{
2878 xlog_recover_item_t *first_item, *item, *free_item;
2879 int i;
2880
2881 item = first_item = trans->r_itemq;
2882 do {
2883 free_item = item;
2884 item = item->ri_next;
2885 /* Free the regions in the item. */
2886 for (i = 0; i < free_item->ri_cnt; i++) {
2887 kmem_free(free_item->ri_buf[i].i_addr,
2888 free_item->ri_buf[i].i_len);
2889 }
2890 /* Free the item itself */
2891 kmem_free(free_item->ri_buf,
2892 (free_item->ri_total * sizeof(xfs_log_iovec_t)));
2893 kmem_free(free_item, sizeof(xlog_recover_item_t));
2894 } while (first_item != item);
2895 /* Free the transaction recover structure */
2896 kmem_free(trans, sizeof(xlog_recover_t));
2897}
2898
2899STATIC int
2900xlog_recover_commit_trans(
2901 xlog_t *log,
2902 xlog_recover_t **q,
2903 xlog_recover_t *trans,
2904 int pass)
2905{
2906 int error;
2907
2908 if ((error = xlog_recover_unlink_tid(q, trans)))
2909 return error;
2910 if ((error = xlog_recover_do_trans(log, trans, pass)))
2911 return error;
2912 xlog_recover_free_trans(trans); /* no error */
2913 return 0;
2914}
2915
2916STATIC int
2917xlog_recover_unmount_trans(
2918 xlog_recover_t *trans)
2919{
2920 /* Do nothing now */
2921 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2922 return 0;
2923}
2924
2925/*
2926 * There are two valid states of the r_state field. 0 indicates that the
2927 * transaction structure is in a normal state. We have either seen the
2928 * start of the transaction or the last operation we added was not a partial
2929 * operation. If the last operation we added to the transaction was a
2930 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2931 *
2932 * NOTE: skip LRs with 0 data length.
2933 */
2934STATIC int
2935xlog_recover_process_data(
2936 xlog_t *log,
2937 xlog_recover_t *rhash[],
2938 xlog_rec_header_t *rhead,
2939 xfs_caddr_t dp,
2940 int pass)
2941{
2942 xfs_caddr_t lp;
2943 int num_logops;
2944 xlog_op_header_t *ohead;
2945 xlog_recover_t *trans;
2946 xlog_tid_t tid;
2947 int error;
2948 unsigned long hash;
2949 uint flags;
2950
2951 lp = dp + INT_GET(rhead->h_len, ARCH_CONVERT);
2952 num_logops = INT_GET(rhead->h_num_logops, ARCH_CONVERT);
2953
2954 /* check the log format matches our own - else we can't recover */
2955 if (xlog_header_check_recover(log->l_mp, rhead))
2956 return (XFS_ERROR(EIO));
2957
2958 while ((dp < lp) && num_logops) {
2959 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2960 ohead = (xlog_op_header_t *)dp;
2961 dp += sizeof(xlog_op_header_t);
2962 if (ohead->oh_clientid != XFS_TRANSACTION &&
2963 ohead->oh_clientid != XFS_LOG) {
2964 xlog_warn(
2965 "XFS: xlog_recover_process_data: bad clientid");
2966 ASSERT(0);
2967 return (XFS_ERROR(EIO));
2968 }
2969 tid = INT_GET(ohead->oh_tid, ARCH_CONVERT);
2970 hash = XLOG_RHASH(tid);
2971 trans = xlog_recover_find_tid(rhash[hash], tid);
2972 if (trans == NULL) { /* not found; add new tid */
2973 if (ohead->oh_flags & XLOG_START_TRANS)
2974 xlog_recover_new_tid(&rhash[hash], tid,
2975 INT_GET(rhead->h_lsn, ARCH_CONVERT));
2976 } else {
2977 ASSERT(dp+INT_GET(ohead->oh_len, ARCH_CONVERT) <= lp);
2978 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2979 if (flags & XLOG_WAS_CONT_TRANS)
2980 flags &= ~XLOG_CONTINUE_TRANS;
2981 switch (flags) {
2982 case XLOG_COMMIT_TRANS:
2983 error = xlog_recover_commit_trans(log,
2984 &rhash[hash], trans, pass);
2985 break;
2986 case XLOG_UNMOUNT_TRANS:
2987 error = xlog_recover_unmount_trans(trans);
2988 break;
2989 case XLOG_WAS_CONT_TRANS:
2990 error = xlog_recover_add_to_cont_trans(trans,
2991 dp, INT_GET(ohead->oh_len,
2992 ARCH_CONVERT));
2993 break;
2994 case XLOG_START_TRANS:
2995 xlog_warn(
2996 "XFS: xlog_recover_process_data: bad transaction");
2997 ASSERT(0);
2998 error = XFS_ERROR(EIO);
2999 break;
3000 case 0:
3001 case XLOG_CONTINUE_TRANS:
3002 error = xlog_recover_add_to_trans(trans,
3003 dp, INT_GET(ohead->oh_len,
3004 ARCH_CONVERT));
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;
3012 }
3013 if (error)
3014 return error;
3015 }
3016 dp += INT_GET(ohead->oh_len, ARCH_CONVERT);
3017 num_logops--;
3018 }
3019 return 0;
3020}
3021
3022/*
3023 * Process an extent free intent item that was recovered from
3024 * the log. We need to free the extents that it describes.
3025 */
3026STATIC void
3027xlog_recover_process_efi(
3028 xfs_mount_t *mp,
3029 xfs_efi_log_item_t *efip)
3030{
3031 xfs_efd_log_item_t *efdp;
3032 xfs_trans_t *tp;
3033 int i;
3034 xfs_extent_t *extp;
3035 xfs_fsblock_t startblock_fsb;
3036
3037 ASSERT(!(efip->efi_flags & XFS_EFI_RECOVERED));
3038
3039 /*
3040 * First check the validity of the extents described by the
3041 * EFI. If any are bad, then assume that all are bad and
3042 * just toss the EFI.
3043 */
3044 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3045 extp = &(efip->efi_format.efi_extents[i]);
3046 startblock_fsb = XFS_BB_TO_FSB(mp,
3047 XFS_FSB_TO_DADDR(mp, extp->ext_start));
3048 if ((startblock_fsb == 0) ||
3049 (extp->ext_len == 0) ||
3050 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3051 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3052 /*
3053 * This will pull the EFI from the AIL and
3054 * free the memory associated with it.
3055 */
3056 xfs_efi_release(efip, efip->efi_format.efi_nextents);
3057 return;
3058 }
3059 }
3060
3061 tp = xfs_trans_alloc(mp, 0);
3062 xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
3063 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3064
3065 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3066 extp = &(efip->efi_format.efi_extents[i]);
3067 xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3068 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3069 extp->ext_len);
3070 }
3071
3072 efip->efi_flags |= XFS_EFI_RECOVERED;
3073 xfs_trans_commit(tp, 0, NULL);
3074}
3075
3076/*
3077 * Verify that once we've encountered something other than an EFI
3078 * in the AIL that there are no more EFIs in the AIL.
3079 */
3080#if defined(DEBUG)
3081STATIC void
3082xlog_recover_check_ail(
3083 xfs_mount_t *mp,
3084 xfs_log_item_t *lip,
3085 int gen)
3086{
3087 int orig_gen = gen;
3088
3089 do {
3090 ASSERT(lip->li_type != XFS_LI_EFI);
3091 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
3092 /*
3093 * The check will be bogus if we restart from the
3094 * beginning of the AIL, so ASSERT that we don't.
3095 * We never should since we're holding the AIL lock
3096 * the entire time.
3097 */
3098 ASSERT(gen == orig_gen);
3099 } while (lip != NULL);
3100}
3101#endif /* DEBUG */
3102
3103/*
3104 * When this is called, all of the EFIs which did not have
3105 * corresponding EFDs should be in the AIL. What we do now
3106 * is free the extents associated with each one.
3107 *
3108 * Since we process the EFIs in normal transactions, they
3109 * will be removed at some point after the commit. This prevents
3110 * us from just walking down the list processing each one.
3111 * We'll use a flag in the EFI to skip those that we've already
3112 * processed and use the AIL iteration mechanism's generation
3113 * count to try to speed this up at least a bit.
3114 *
3115 * When we start, we know that the EFIs are the only things in
3116 * the AIL. As we process them, however, other items are added
3117 * to the AIL. Since everything added to the AIL must come after
3118 * everything already in the AIL, we stop processing as soon as
3119 * we see something other than an EFI in the AIL.
3120 */
3121STATIC void
3122xlog_recover_process_efis(
3123 xlog_t *log)
3124{
3125 xfs_log_item_t *lip;
3126 xfs_efi_log_item_t *efip;
3127 int gen;
3128 xfs_mount_t *mp;
3129 SPLDECL(s);
3130
3131 mp = log->l_mp;
3132 AIL_LOCK(mp,s);
3133
3134 lip = xfs_trans_first_ail(mp, &gen);
3135 while (lip != NULL) {
3136 /*
3137 * We're done when we see something other than an EFI.
3138 */
3139 if (lip->li_type != XFS_LI_EFI) {
3140 xlog_recover_check_ail(mp, lip, gen);
3141 break;
3142 }
3143
3144 /*
3145 * Skip EFIs that we've already processed.
3146 */
3147 efip = (xfs_efi_log_item_t *)lip;
3148 if (efip->efi_flags & XFS_EFI_RECOVERED) {
3149 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
3150 continue;
3151 }
3152
3153 AIL_UNLOCK(mp, s);
3154 xlog_recover_process_efi(mp, efip);
3155 AIL_LOCK(mp,s);
3156 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
3157 }
3158 AIL_UNLOCK(mp, s);
3159}
3160
3161/*
3162 * This routine performs a transaction to null out a bad inode pointer
3163 * in an agi unlinked inode hash bucket.
3164 */
3165STATIC void
3166xlog_recover_clear_agi_bucket(
3167 xfs_mount_t *mp,
3168 xfs_agnumber_t agno,
3169 int bucket)
3170{
3171 xfs_trans_t *tp;
3172 xfs_agi_t *agi;
3173 xfs_buf_t *agibp;
3174 int offset;
3175 int error;
3176
3177 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3178 xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp), 0, 0, 0);
3179
3180 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
3181 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
3182 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
3183 if (error) {
3184 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3185 return;
3186 }
3187
3188 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11003189 if (be32_to_cpu(agi->agi_magicnum) != XFS_AGI_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003190 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3191 return;
3192 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003193
Christoph Hellwig16259e72005-11-02 15:11:25 +11003194 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195 offset = offsetof(xfs_agi_t, agi_unlinked) +
3196 (sizeof(xfs_agino_t) * bucket);
3197 xfs_trans_log_buf(tp, agibp, offset,
3198 (offset + sizeof(xfs_agino_t) - 1));
3199
3200 (void) xfs_trans_commit(tp, 0, NULL);
3201}
3202
3203/*
3204 * xlog_iunlink_recover
3205 *
3206 * This is called during recovery to process any inodes which
3207 * we unlinked but not freed when the system crashed. These
3208 * inodes will be on the lists in the AGI blocks. What we do
3209 * here is scan all the AGIs and fully truncate and free any
3210 * inodes found on the lists. Each inode is removed from the
3211 * lists when it has been fully truncated and is freed. The
3212 * freeing of the inode and its removal from the list must be
3213 * atomic.
3214 */
3215void
3216xlog_recover_process_iunlinks(
3217 xlog_t *log)
3218{
3219 xfs_mount_t *mp;
3220 xfs_agnumber_t agno;
3221 xfs_agi_t *agi;
3222 xfs_buf_t *agibp;
3223 xfs_buf_t *ibp;
3224 xfs_dinode_t *dip;
3225 xfs_inode_t *ip;
3226 xfs_agino_t agino;
3227 xfs_ino_t ino;
3228 int bucket;
3229 int error;
3230 uint mp_dmevmask;
3231
3232 mp = log->l_mp;
3233
3234 /*
3235 * Prevent any DMAPI event from being sent while in this function.
3236 */
3237 mp_dmevmask = mp->m_dmevmask;
3238 mp->m_dmevmask = 0;
3239
3240 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3241 /*
3242 * Find the agi for this ag.
3243 */
3244 agibp = xfs_buf_read(mp->m_ddev_targp,
3245 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
3246 XFS_FSS_TO_BB(mp, 1), 0);
3247 if (XFS_BUF_ISERROR(agibp)) {
3248 xfs_ioerror_alert("xlog_recover_process_iunlinks(#1)",
3249 log->l_mp, agibp,
3250 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)));
3251 }
3252 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11003253 ASSERT(XFS_AGI_MAGIC == be32_to_cpu(agi->agi_magicnum));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003254
3255 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3256
Christoph Hellwig16259e72005-11-02 15:11:25 +11003257 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258 while (agino != NULLAGINO) {
3259
3260 /*
3261 * Release the agi buffer so that it can
3262 * be acquired in the normal course of the
3263 * transaction to truncate and free the inode.
3264 */
3265 xfs_buf_relse(agibp);
3266
3267 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3268 error = xfs_iget(mp, NULL, ino, 0, 0, &ip, 0);
3269 ASSERT(error || (ip != NULL));
3270
3271 if (!error) {
3272 /*
3273 * Get the on disk inode to find the
3274 * next inode in the bucket.
3275 */
3276 error = xfs_itobp(mp, NULL, ip, &dip,
Nathan Scottb12dd342006-03-17 17:26:04 +11003277 &ibp, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278 ASSERT(error || (dip != NULL));
3279 }
3280
3281 if (!error) {
3282 ASSERT(ip->i_d.di_nlink == 0);
3283
3284 /* setup for the next pass */
3285 agino = INT_GET(dip->di_next_unlinked,
3286 ARCH_CONVERT);
3287 xfs_buf_relse(ibp);
3288 /*
3289 * Prevent any DMAPI event from
3290 * being sent when the
3291 * reference on the inode is
3292 * dropped.
3293 */
3294 ip->i_d.di_dmevmask = 0;
3295
3296 /*
3297 * If this is a new inode, handle
3298 * it specially. Otherwise,
3299 * just drop our reference to the
3300 * inode. If there are no
3301 * other references, this will
3302 * send the inode to
3303 * xfs_inactive() which will
3304 * truncate the file and free
3305 * the inode.
3306 */
3307 if (ip->i_d.di_mode == 0)
3308 xfs_iput_new(ip, 0);
3309 else
3310 VN_RELE(XFS_ITOV(ip));
3311 } else {
3312 /*
3313 * We can't read in the inode
3314 * this bucket points to, or
3315 * this inode is messed up. Just
3316 * ditch this bucket of inodes. We
3317 * will lose some inodes and space,
3318 * but at least we won't hang. Call
3319 * xlog_recover_clear_agi_bucket()
3320 * to perform a transaction to clear
3321 * the inode pointer in the bucket.
3322 */
3323 xlog_recover_clear_agi_bucket(mp, agno,
3324 bucket);
3325
3326 agino = NULLAGINO;
3327 }
3328
3329 /*
3330 * Reacquire the agibuffer and continue around
3331 * the loop.
3332 */
3333 agibp = xfs_buf_read(mp->m_ddev_targp,
3334 XFS_AG_DADDR(mp, agno,
3335 XFS_AGI_DADDR(mp)),
3336 XFS_FSS_TO_BB(mp, 1), 0);
3337 if (XFS_BUF_ISERROR(agibp)) {
3338 xfs_ioerror_alert(
3339 "xlog_recover_process_iunlinks(#2)",
3340 log->l_mp, agibp,
3341 XFS_AG_DADDR(mp, agno,
3342 XFS_AGI_DADDR(mp)));
3343 }
3344 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11003345 ASSERT(XFS_AGI_MAGIC == be32_to_cpu(
3346 agi->agi_magicnum));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003347 }
3348 }
3349
3350 /*
3351 * Release the buffer for the current agi so we can
3352 * go on to the next one.
3353 */
3354 xfs_buf_relse(agibp);
3355 }
3356
3357 mp->m_dmevmask = mp_dmevmask;
3358}
3359
3360
3361#ifdef DEBUG
3362STATIC void
3363xlog_pack_data_checksum(
3364 xlog_t *log,
3365 xlog_in_core_t *iclog,
3366 int size)
3367{
3368 int i;
3369 uint *up;
3370 uint chksum = 0;
3371
3372 up = (uint *)iclog->ic_datap;
3373 /* divide length by 4 to get # words */
3374 for (i = 0; i < (size >> 2); i++) {
3375 chksum ^= INT_GET(*up, ARCH_CONVERT);
3376 up++;
3377 }
3378 INT_SET(iclog->ic_header.h_chksum, ARCH_CONVERT, chksum);
3379}
3380#else
3381#define xlog_pack_data_checksum(log, iclog, size)
3382#endif
3383
3384/*
3385 * Stamp cycle number in every block
3386 */
3387void
3388xlog_pack_data(
3389 xlog_t *log,
3390 xlog_in_core_t *iclog,
3391 int roundoff)
3392{
3393 int i, j, k;
3394 int size = iclog->ic_offset + roundoff;
3395 uint cycle_lsn;
3396 xfs_caddr_t dp;
3397 xlog_in_core_2_t *xhdr;
3398
3399 xlog_pack_data_checksum(log, iclog, size);
3400
3401 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3402
3403 dp = iclog->ic_datap;
3404 for (i = 0; i < BTOBB(size) &&
3405 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3406 iclog->ic_header.h_cycle_data[i] = *(uint *)dp;
3407 *(uint *)dp = cycle_lsn;
3408 dp += BBSIZE;
3409 }
3410
3411 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3412 xhdr = (xlog_in_core_2_t *)&iclog->ic_header;
3413 for ( ; i < BTOBB(size); i++) {
3414 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3415 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3416 xhdr[j].hic_xheader.xh_cycle_data[k] = *(uint *)dp;
3417 *(uint *)dp = cycle_lsn;
3418 dp += BBSIZE;
3419 }
3420
3421 for (i = 1; i < log->l_iclog_heads; i++) {
3422 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3423 }
3424 }
3425}
3426
3427#if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
3428STATIC void
3429xlog_unpack_data_checksum(
3430 xlog_rec_header_t *rhead,
3431 xfs_caddr_t dp,
3432 xlog_t *log)
3433{
3434 uint *up = (uint *)dp;
3435 uint chksum = 0;
3436 int i;
3437
3438 /* divide length by 4 to get # words */
3439 for (i=0; i < INT_GET(rhead->h_len, ARCH_CONVERT) >> 2; i++) {
3440 chksum ^= INT_GET(*up, ARCH_CONVERT);
3441 up++;
3442 }
3443 if (chksum != INT_GET(rhead->h_chksum, ARCH_CONVERT)) {
3444 if (rhead->h_chksum ||
3445 ((log->l_flags & XLOG_CHKSUM_MISMATCH) == 0)) {
3446 cmn_err(CE_DEBUG,
Nathan Scottb6574522006-06-09 15:29:40 +10003447 "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003448 INT_GET(rhead->h_chksum, ARCH_CONVERT), chksum);
3449 cmn_err(CE_DEBUG,
3450"XFS: Disregard message if filesystem was created with non-DEBUG kernel");
3451 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3452 cmn_err(CE_DEBUG,
Nathan Scottb6574522006-06-09 15:29:40 +10003453 "XFS: LogR this is a LogV2 filesystem\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454 }
3455 log->l_flags |= XLOG_CHKSUM_MISMATCH;
3456 }
3457 }
3458}
3459#else
3460#define xlog_unpack_data_checksum(rhead, dp, log)
3461#endif
3462
3463STATIC void
3464xlog_unpack_data(
3465 xlog_rec_header_t *rhead,
3466 xfs_caddr_t dp,
3467 xlog_t *log)
3468{
3469 int i, j, k;
3470 xlog_in_core_2_t *xhdr;
3471
3472 for (i = 0; i < BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT)) &&
3473 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3474 *(uint *)dp = *(uint *)&rhead->h_cycle_data[i];
3475 dp += BBSIZE;
3476 }
3477
3478 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3479 xhdr = (xlog_in_core_2_t *)rhead;
3480 for ( ; i < BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT)); i++) {
3481 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3482 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3483 *(uint *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
3484 dp += BBSIZE;
3485 }
3486 }
3487
3488 xlog_unpack_data_checksum(rhead, dp, log);
3489}
3490
3491STATIC int
3492xlog_valid_rec_header(
3493 xlog_t *log,
3494 xlog_rec_header_t *rhead,
3495 xfs_daddr_t blkno)
3496{
3497 int hlen;
3498
3499 if (unlikely(
3500 (INT_GET(rhead->h_magicno, ARCH_CONVERT) !=
3501 XLOG_HEADER_MAGIC_NUM))) {
3502 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3503 XFS_ERRLEVEL_LOW, log->l_mp);
3504 return XFS_ERROR(EFSCORRUPTED);
3505 }
3506 if (unlikely(
3507 (!rhead->h_version ||
3508 (INT_GET(rhead->h_version, ARCH_CONVERT) &
3509 (~XLOG_VERSION_OKBITS)) != 0))) {
3510 xlog_warn("XFS: %s: unrecognised log version (%d).",
3511 __FUNCTION__, INT_GET(rhead->h_version, ARCH_CONVERT));
3512 return XFS_ERROR(EIO);
3513 }
3514
3515 /* LR body must have data or it wouldn't have been written */
3516 hlen = INT_GET(rhead->h_len, ARCH_CONVERT);
3517 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3518 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3519 XFS_ERRLEVEL_LOW, log->l_mp);
3520 return XFS_ERROR(EFSCORRUPTED);
3521 }
3522 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3523 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3524 XFS_ERRLEVEL_LOW, log->l_mp);
3525 return XFS_ERROR(EFSCORRUPTED);
3526 }
3527 return 0;
3528}
3529
3530/*
3531 * Read the log from tail to head and process the log records found.
3532 * Handle the two cases where the tail and head are in the same cycle
3533 * and where the active portion of the log wraps around the end of
3534 * the physical log separately. The pass parameter is passed through
3535 * to the routines called to process the data and is not looked at
3536 * here.
3537 */
3538STATIC int
3539xlog_do_recovery_pass(
3540 xlog_t *log,
3541 xfs_daddr_t head_blk,
3542 xfs_daddr_t tail_blk,
3543 int pass)
3544{
3545 xlog_rec_header_t *rhead;
3546 xfs_daddr_t blk_no;
3547 xfs_caddr_t bufaddr, offset;
3548 xfs_buf_t *hbp, *dbp;
3549 int error = 0, h_size;
3550 int bblks, split_bblks;
3551 int hblks, split_hblks, wrapped_hblks;
3552 xlog_recover_t *rhash[XLOG_RHASH_SIZE];
3553
3554 ASSERT(head_blk != tail_blk);
3555
3556 /*
3557 * Read the header of the tail block and get the iclog buffer size from
3558 * h_size. Use this to tell how many sectors make up the log header.
3559 */
3560 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3561 /*
3562 * When using variable length iclogs, read first sector of
3563 * iclog header and extract the header size from it. Get a
3564 * new hbp that is the correct size.
3565 */
3566 hbp = xlog_get_bp(log, 1);
3567 if (!hbp)
3568 return ENOMEM;
3569 if ((error = xlog_bread(log, tail_blk, 1, hbp)))
3570 goto bread_err1;
3571 offset = xlog_align(log, tail_blk, 1, hbp);
3572 rhead = (xlog_rec_header_t *)offset;
3573 error = xlog_valid_rec_header(log, rhead, tail_blk);
3574 if (error)
3575 goto bread_err1;
3576 h_size = INT_GET(rhead->h_size, ARCH_CONVERT);
3577 if ((INT_GET(rhead->h_version, ARCH_CONVERT)
3578 & XLOG_VERSION_2) &&
3579 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3580 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3581 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3582 hblks++;
3583 xlog_put_bp(hbp);
3584 hbp = xlog_get_bp(log, hblks);
3585 } else {
3586 hblks = 1;
3587 }
3588 } else {
3589 ASSERT(log->l_sectbb_log == 0);
3590 hblks = 1;
3591 hbp = xlog_get_bp(log, 1);
3592 h_size = XLOG_BIG_RECORD_BSIZE;
3593 }
3594
3595 if (!hbp)
3596 return ENOMEM;
3597 dbp = xlog_get_bp(log, BTOBB(h_size));
3598 if (!dbp) {
3599 xlog_put_bp(hbp);
3600 return ENOMEM;
3601 }
3602
3603 memset(rhash, 0, sizeof(rhash));
3604 if (tail_blk <= head_blk) {
3605 for (blk_no = tail_blk; blk_no < head_blk; ) {
3606 if ((error = xlog_bread(log, blk_no, hblks, hbp)))
3607 goto bread_err2;
3608 offset = xlog_align(log, blk_no, hblks, hbp);
3609 rhead = (xlog_rec_header_t *)offset;
3610 error = xlog_valid_rec_header(log, rhead, blk_no);
3611 if (error)
3612 goto bread_err2;
3613
3614 /* blocks in data section */
3615 bblks = (int)BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT));
3616 error = xlog_bread(log, blk_no + hblks, bblks, dbp);
3617 if (error)
3618 goto bread_err2;
3619 offset = xlog_align(log, blk_no + hblks, bblks, dbp);
3620 xlog_unpack_data(rhead, offset, log);
3621 if ((error = xlog_recover_process_data(log,
3622 rhash, rhead, offset, pass)))
3623 goto bread_err2;
3624 blk_no += bblks + hblks;
3625 }
3626 } else {
3627 /*
3628 * Perform recovery around the end of the physical log.
3629 * When the head is not on the same cycle number as the tail,
3630 * we can't do a sequential recovery as above.
3631 */
3632 blk_no = tail_blk;
3633 while (blk_no < log->l_logBBsize) {
3634 /*
3635 * Check for header wrapping around physical end-of-log
3636 */
3637 offset = NULL;
3638 split_hblks = 0;
3639 wrapped_hblks = 0;
3640 if (blk_no + hblks <= log->l_logBBsize) {
3641 /* Read header in one read */
3642 error = xlog_bread(log, blk_no, hblks, hbp);
3643 if (error)
3644 goto bread_err2;
3645 offset = xlog_align(log, blk_no, hblks, hbp);
3646 } else {
3647 /* This LR is split across physical log end */
3648 if (blk_no != log->l_logBBsize) {
3649 /* some data before physical log end */
3650 ASSERT(blk_no <= INT_MAX);
3651 split_hblks = log->l_logBBsize - (int)blk_no;
3652 ASSERT(split_hblks > 0);
3653 if ((error = xlog_bread(log, blk_no,
3654 split_hblks, hbp)))
3655 goto bread_err2;
3656 offset = xlog_align(log, blk_no,
3657 split_hblks, hbp);
3658 }
3659 /*
3660 * Note: this black magic still works with
3661 * large sector sizes (non-512) only because:
3662 * - we increased the buffer size originally
3663 * by 1 sector giving us enough extra space
3664 * for the second read;
3665 * - the log start is guaranteed to be sector
3666 * aligned;
3667 * - we read the log end (LR header start)
3668 * _first_, then the log start (LR header end)
3669 * - order is important.
3670 */
3671 bufaddr = XFS_BUF_PTR(hbp);
3672 XFS_BUF_SET_PTR(hbp,
3673 bufaddr + BBTOB(split_hblks),
3674 BBTOB(hblks - split_hblks));
3675 wrapped_hblks = hblks - split_hblks;
3676 error = xlog_bread(log, 0, wrapped_hblks, hbp);
3677 if (error)
3678 goto bread_err2;
3679 XFS_BUF_SET_PTR(hbp, bufaddr, BBTOB(hblks));
3680 if (!offset)
3681 offset = xlog_align(log, 0,
3682 wrapped_hblks, hbp);
3683 }
3684 rhead = (xlog_rec_header_t *)offset;
3685 error = xlog_valid_rec_header(log, rhead,
3686 split_hblks ? blk_no : 0);
3687 if (error)
3688 goto bread_err2;
3689
3690 bblks = (int)BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT));
3691 blk_no += hblks;
3692
3693 /* Read in data for log record */
3694 if (blk_no + bblks <= log->l_logBBsize) {
3695 error = xlog_bread(log, blk_no, bblks, dbp);
3696 if (error)
3697 goto bread_err2;
3698 offset = xlog_align(log, blk_no, bblks, dbp);
3699 } else {
3700 /* This log record is split across the
3701 * physical end of log */
3702 offset = NULL;
3703 split_bblks = 0;
3704 if (blk_no != log->l_logBBsize) {
3705 /* some data is before the physical
3706 * end of log */
3707 ASSERT(!wrapped_hblks);
3708 ASSERT(blk_no <= INT_MAX);
3709 split_bblks =
3710 log->l_logBBsize - (int)blk_no;
3711 ASSERT(split_bblks > 0);
3712 if ((error = xlog_bread(log, blk_no,
3713 split_bblks, dbp)))
3714 goto bread_err2;
3715 offset = xlog_align(log, blk_no,
3716 split_bblks, dbp);
3717 }
3718 /*
3719 * Note: this black magic still works with
3720 * large sector sizes (non-512) only because:
3721 * - we increased the buffer size originally
3722 * by 1 sector giving us enough extra space
3723 * for the second read;
3724 * - the log start is guaranteed to be sector
3725 * aligned;
3726 * - we read the log end (LR header start)
3727 * _first_, then the log start (LR header end)
3728 * - order is important.
3729 */
3730 bufaddr = XFS_BUF_PTR(dbp);
3731 XFS_BUF_SET_PTR(dbp,
3732 bufaddr + BBTOB(split_bblks),
3733 BBTOB(bblks - split_bblks));
3734 if ((error = xlog_bread(log, wrapped_hblks,
3735 bblks - split_bblks, dbp)))
3736 goto bread_err2;
3737 XFS_BUF_SET_PTR(dbp, bufaddr, h_size);
3738 if (!offset)
3739 offset = xlog_align(log, wrapped_hblks,
3740 bblks - split_bblks, dbp);
3741 }
3742 xlog_unpack_data(rhead, offset, log);
3743 if ((error = xlog_recover_process_data(log, rhash,
3744 rhead, offset, pass)))
3745 goto bread_err2;
3746 blk_no += bblks;
3747 }
3748
3749 ASSERT(blk_no >= log->l_logBBsize);
3750 blk_no -= log->l_logBBsize;
3751
3752 /* read first part of physical log */
3753 while (blk_no < head_blk) {
3754 if ((error = xlog_bread(log, blk_no, hblks, hbp)))
3755 goto bread_err2;
3756 offset = xlog_align(log, blk_no, hblks, hbp);
3757 rhead = (xlog_rec_header_t *)offset;
3758 error = xlog_valid_rec_header(log, rhead, blk_no);
3759 if (error)
3760 goto bread_err2;
3761 bblks = (int)BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT));
3762 if ((error = xlog_bread(log, blk_no+hblks, bblks, dbp)))
3763 goto bread_err2;
3764 offset = xlog_align(log, blk_no+hblks, bblks, dbp);
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;
3770 }
3771 }
3772
3773 bread_err2:
3774 xlog_put_bp(dbp);
3775 bread_err1:
3776 xlog_put_bp(hbp);
3777 return error;
3778}
3779
3780/*
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.
3788 *
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.
3792 */
3793STATIC int
3794xlog_do_log_recovery(
3795 xlog_t *log,
3796 xfs_daddr_t head_blk,
3797 xfs_daddr_t tail_blk)
3798{
3799 int error;
3800
3801 ASSERT(head_blk != tail_blk);
3802
3803 /*
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.
3806 */
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 XLOG_BC_TABLE_SIZE * sizeof(xfs_buf_cancel_t*));
3816 log->l_buf_cancel_table = NULL;
3817 return error;
3818 }
3819 /*
3820 * Then do a second pass to actually recover the items in the log.
3821 * When it is complete free the table of buf cancel items.
3822 */
3823 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3824 XLOG_RECOVER_PASS2);
3825#ifdef DEBUG
Tim Shimmin6d192a92006-06-09 14:55:38 +10003826 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003827 int i;
3828
3829 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3830 ASSERT(log->l_buf_cancel_table[i] == NULL);
3831 }
3832#endif /* DEBUG */
3833
3834 kmem_free(log->l_buf_cancel_table,
3835 XLOG_BC_TABLE_SIZE * sizeof(xfs_buf_cancel_t*));
3836 log->l_buf_cancel_table = NULL;
3837
3838 return error;
3839}
3840
3841/*
3842 * Do the actual recovery
3843 */
3844STATIC int
3845xlog_do_recover(
3846 xlog_t *log,
3847 xfs_daddr_t head_blk,
3848 xfs_daddr_t tail_blk)
3849{
3850 int error;
3851 xfs_buf_t *bp;
3852 xfs_sb_t *sbp;
3853
3854 /*
3855 * First replay the images in the log.
3856 */
3857 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3858 if (error) {
3859 return error;
3860 }
3861
3862 XFS_bflush(log->l_mp->m_ddev_targp);
3863
3864 /*
3865 * If IO errors happened during recovery, bail out.
3866 */
3867 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3868 return (EIO);
3869 }
3870
3871 /*
3872 * We now update the tail_lsn since much of the recovery has completed
3873 * and there may be space available to use. If there were no extent
3874 * or iunlinks, we can free up the entire log and set the tail_lsn to
3875 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3876 * lsn of the last known good LR on disk. If there are extent frees
3877 * or iunlinks they will have some entries in the AIL; so we look at
3878 * the AIL to determine how to set the tail_lsn.
3879 */
3880 xlog_assign_tail_lsn(log->l_mp);
3881
3882 /*
3883 * Now that we've finished replaying all buffer and inode
3884 * updates, re-read in the superblock.
3885 */
3886 bp = xfs_getsb(log->l_mp, 0);
3887 XFS_BUF_UNDONE(bp);
3888 XFS_BUF_READ(bp);
3889 xfsbdstrat(log->l_mp, bp);
3890 if ((error = xfs_iowait(bp))) {
3891 xfs_ioerror_alert("xlog_do_recover",
3892 log->l_mp, bp, XFS_BUF_ADDR(bp));
3893 ASSERT(0);
3894 xfs_buf_relse(bp);
3895 return error;
3896 }
3897
3898 /* Convert superblock from on-disk format */
3899 sbp = &log->l_mp->m_sb;
3900 xfs_xlatesb(XFS_BUF_TO_SBP(bp), sbp, 1, XFS_SB_ALL_BITS);
3901 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
3902 ASSERT(XFS_SB_GOOD_VERSION(sbp));
3903 xfs_buf_relse(bp);
3904
3905 xlog_recover_check_summary(log);
3906
3907 /* Normal transactions can now occur */
3908 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3909 return 0;
3910}
3911
3912/*
3913 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3914 *
3915 * Return error or zero.
3916 */
3917int
3918xlog_recover(
Eric Sandeen65be6052006-01-11 15:34:19 +11003919 xlog_t *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003920{
3921 xfs_daddr_t head_blk, tail_blk;
3922 int error;
3923
3924 /* find the tail of the log */
Eric Sandeen65be6052006-01-11 15:34:19 +11003925 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003926 return error;
3927
3928 if (tail_blk != head_blk) {
3929 /* There used to be a comment here:
3930 *
3931 * disallow recovery on read-only mounts. note -- mount
3932 * checks for ENOSPC and turns it into an intelligent
3933 * error message.
3934 * ...but this is no longer true. Now, unless you specify
3935 * NORECOVERY (in which case this function would never be
3936 * called), we just go ahead and recover. We do this all
3937 * under the vfs layer, so we can get away with it unless
3938 * the device itself is read-only, in which case we fail.
3939 */
3940 if ((error = xfs_dev_is_read_only(log->l_mp,
3941 "recovery required"))) {
3942 return error;
3943 }
3944
3945 cmn_err(CE_NOTE,
Nathan Scottfc1f8c12005-11-02 11:44:33 +11003946 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3947 log->l_mp->m_fsname, log->l_mp->m_logname ?
3948 log->l_mp->m_logname : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003949
3950 error = xlog_do_recover(log, head_blk, tail_blk);
3951 log->l_flags |= XLOG_RECOVERY_NEEDED;
3952 }
3953 return error;
3954}
3955
3956/*
3957 * In the first part of recovery we replay inodes and buffers and build
3958 * up the list of extent free items which need to be processed. Here
3959 * we process the extent free items and clean up the on disk unlinked
3960 * inode lists. This is separated from the first part of recovery so
3961 * that the root and real-time bitmap inodes can be read in from disk in
3962 * between the two stages. This is necessary so that we can free space
3963 * in the real-time portion of the file system.
3964 */
3965int
3966xlog_recover_finish(
3967 xlog_t *log,
3968 int mfsi_flags)
3969{
3970 /*
3971 * Now we're ready to do the transactions needed for the
3972 * rest of recovery. Start with completing all the extent
3973 * free intent records and then process the unlinked inode
3974 * lists. At this point, we essentially run in normal mode
3975 * except that we're still performing recovery actions
3976 * rather than accepting new requests.
3977 */
3978 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
3979 xlog_recover_process_efis(log);
3980 /*
3981 * Sync the log to get all the EFIs out of the AIL.
3982 * This isn't absolutely necessary, but it helps in
3983 * case the unlink transactions would have problems
3984 * pushing the EFIs out of the way.
3985 */
3986 xfs_log_force(log->l_mp, (xfs_lsn_t)0,
3987 (XFS_LOG_FORCE | XFS_LOG_SYNC));
3988
3989 if ( (mfsi_flags & XFS_MFSI_NOUNLINK) == 0 ) {
3990 xlog_recover_process_iunlinks(log);
3991 }
3992
3993 xlog_recover_check_summary(log);
3994
3995 cmn_err(CE_NOTE,
Nathan Scottfc1f8c12005-11-02 11:44:33 +11003996 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3997 log->l_mp->m_fsname, log->l_mp->m_logname ?
3998 log->l_mp->m_logname : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003999 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
4000 } else {
4001 cmn_err(CE_DEBUG,
Nathan Scottb6574522006-06-09 15:29:40 +10004002 "!Ending clean XFS mount for filesystem: %s\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004003 log->l_mp->m_fsname);
4004 }
4005 return 0;
4006}
4007
4008
4009#if defined(DEBUG)
4010/*
4011 * Read all of the agf and agi counters and check that they
4012 * are consistent with the superblock counters.
4013 */
4014void
4015xlog_recover_check_summary(
4016 xlog_t *log)
4017{
4018 xfs_mount_t *mp;
4019 xfs_agf_t *agfp;
4020 xfs_agi_t *agip;
4021 xfs_buf_t *agfbp;
4022 xfs_buf_t *agibp;
4023 xfs_daddr_t agfdaddr;
4024 xfs_daddr_t agidaddr;
4025 xfs_buf_t *sbbp;
4026#ifdef XFS_LOUD_RECOVERY
4027 xfs_sb_t *sbp;
4028#endif
4029 xfs_agnumber_t agno;
4030 __uint64_t freeblks;
4031 __uint64_t itotal;
4032 __uint64_t ifree;
4033
4034 mp = log->l_mp;
4035
4036 freeblks = 0LL;
4037 itotal = 0LL;
4038 ifree = 0LL;
4039 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
4040 agfdaddr = XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp));
4041 agfbp = xfs_buf_read(mp->m_ddev_targp, agfdaddr,
4042 XFS_FSS_TO_BB(mp, 1), 0);
4043 if (XFS_BUF_ISERROR(agfbp)) {
4044 xfs_ioerror_alert("xlog_recover_check_summary(agf)",
4045 mp, agfbp, agfdaddr);
4046 }
4047 agfp = XFS_BUF_TO_AGF(agfbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11004048 ASSERT(XFS_AGF_MAGIC == be32_to_cpu(agfp->agf_magicnum));
4049 ASSERT(XFS_AGF_GOOD_VERSION(be32_to_cpu(agfp->agf_versionnum)));
4050 ASSERT(be32_to_cpu(agfp->agf_seqno) == agno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004051
Christoph Hellwig16259e72005-11-02 15:11:25 +11004052 freeblks += be32_to_cpu(agfp->agf_freeblks) +
4053 be32_to_cpu(agfp->agf_flcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004054 xfs_buf_relse(agfbp);
4055
4056 agidaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
4057 agibp = xfs_buf_read(mp->m_ddev_targp, agidaddr,
4058 XFS_FSS_TO_BB(mp, 1), 0);
4059 if (XFS_BUF_ISERROR(agibp)) {
4060 xfs_ioerror_alert("xlog_recover_check_summary(agi)",
4061 mp, agibp, agidaddr);
4062 }
4063 agip = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11004064 ASSERT(XFS_AGI_MAGIC == be32_to_cpu(agip->agi_magicnum));
4065 ASSERT(XFS_AGI_GOOD_VERSION(be32_to_cpu(agip->agi_versionnum)));
4066 ASSERT(be32_to_cpu(agip->agi_seqno) == agno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004067
Christoph Hellwig16259e72005-11-02 15:11:25 +11004068 itotal += be32_to_cpu(agip->agi_count);
4069 ifree += be32_to_cpu(agip->agi_freecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004070 xfs_buf_relse(agibp);
4071 }
4072
4073 sbbp = xfs_getsb(mp, 0);
4074#ifdef XFS_LOUD_RECOVERY
4075 sbp = &mp->m_sb;
4076 xfs_xlatesb(XFS_BUF_TO_SBP(sbbp), sbp, 1, XFS_SB_ALL_BITS);
4077 cmn_err(CE_NOTE,
4078 "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
4079 sbp->sb_icount, itotal);
4080 cmn_err(CE_NOTE,
4081 "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
4082 sbp->sb_ifree, ifree);
4083 cmn_err(CE_NOTE,
4084 "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
4085 sbp->sb_fdblocks, freeblks);
4086#if 0
4087 /*
4088 * This is turned off until I account for the allocation
4089 * btree blocks which live in free space.
4090 */
4091 ASSERT(sbp->sb_icount == itotal);
4092 ASSERT(sbp->sb_ifree == ifree);
4093 ASSERT(sbp->sb_fdblocks == freeblks);
4094#endif
4095#endif
4096 xfs_buf_relse(sbbp);
4097}
4098#endif /* DEBUG */