blob: 61a29e1191a82d117d14e8e291722524b8d5e9c8 [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;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 ushort flags = 0;
1518
1519 first_item = itemq = trans->r_itemq;
1520 trans->r_itemq = NULL;
1521 do {
1522 itemq_next = itemq->ri_next;
1523 buf_f = (xfs_buf_log_format_t *)itemq->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524
1525 switch (ITEM_TYPE(itemq)) {
1526 case XFS_LI_BUF:
Eric Sandeen804195b2007-02-10 18:35:02 +11001527 flags = buf_f->blf_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 if (!(flags & XFS_BLI_CANCEL)) {
1529 xlog_recover_insert_item_frontq(&trans->r_itemq,
1530 itemq);
1531 break;
1532 }
1533 case XFS_LI_INODE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 case XFS_LI_DQUOT:
1535 case XFS_LI_QUOTAOFF:
1536 case XFS_LI_EFD:
1537 case XFS_LI_EFI:
1538 xlog_recover_insert_item_backq(&trans->r_itemq, itemq);
1539 break;
1540 default:
1541 xlog_warn(
1542 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1543 ASSERT(0);
1544 return XFS_ERROR(EIO);
1545 }
1546 itemq = itemq_next;
1547 } while (first_item != itemq);
1548 return 0;
1549}
1550
1551/*
1552 * Build up the table of buf cancel records so that we don't replay
1553 * cancelled data in the second pass. For buffer records that are
1554 * not cancel records, there is nothing to do here so we just return.
1555 *
1556 * If we get a cancel record which is already in the table, this indicates
1557 * that the buffer was cancelled multiple times. In order to ensure
1558 * that during pass 2 we keep the record in the table until we reach its
1559 * last occurrence in the log, we keep a reference count in the cancel
1560 * record in the table to tell us how many times we expect to see this
1561 * record during the second pass.
1562 */
1563STATIC void
1564xlog_recover_do_buffer_pass1(
1565 xlog_t *log,
1566 xfs_buf_log_format_t *buf_f)
1567{
1568 xfs_buf_cancel_t *bcp;
1569 xfs_buf_cancel_t *nextp;
1570 xfs_buf_cancel_t *prevp;
1571 xfs_buf_cancel_t **bucket;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 xfs_daddr_t blkno = 0;
1573 uint len = 0;
1574 ushort flags = 0;
1575
1576 switch (buf_f->blf_type) {
1577 case XFS_LI_BUF:
1578 blkno = buf_f->blf_blkno;
1579 len = buf_f->blf_len;
1580 flags = buf_f->blf_flags;
1581 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 }
1583
1584 /*
1585 * If this isn't a cancel buffer item, then just return.
1586 */
1587 if (!(flags & XFS_BLI_CANCEL))
1588 return;
1589
1590 /*
1591 * Insert an xfs_buf_cancel record into the hash table of
1592 * them. If there is already an identical record, bump
1593 * its reference count.
1594 */
1595 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1596 XLOG_BC_TABLE_SIZE];
1597 /*
1598 * If the hash bucket is empty then just insert a new record into
1599 * the bucket.
1600 */
1601 if (*bucket == NULL) {
1602 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1603 KM_SLEEP);
1604 bcp->bc_blkno = blkno;
1605 bcp->bc_len = len;
1606 bcp->bc_refcount = 1;
1607 bcp->bc_next = NULL;
1608 *bucket = bcp;
1609 return;
1610 }
1611
1612 /*
1613 * The hash bucket is not empty, so search for duplicates of our
1614 * record. If we find one them just bump its refcount. If not
1615 * then add us at the end of the list.
1616 */
1617 prevp = NULL;
1618 nextp = *bucket;
1619 while (nextp != NULL) {
1620 if (nextp->bc_blkno == blkno && nextp->bc_len == len) {
1621 nextp->bc_refcount++;
1622 return;
1623 }
1624 prevp = nextp;
1625 nextp = nextp->bc_next;
1626 }
1627 ASSERT(prevp != NULL);
1628 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1629 KM_SLEEP);
1630 bcp->bc_blkno = blkno;
1631 bcp->bc_len = len;
1632 bcp->bc_refcount = 1;
1633 bcp->bc_next = NULL;
1634 prevp->bc_next = bcp;
1635}
1636
1637/*
1638 * Check to see whether the buffer being recovered has a corresponding
1639 * entry in the buffer cancel record table. If it does then return 1
1640 * so that it will be cancelled, otherwise return 0. If the buffer is
1641 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
1642 * the refcount on the entry in the table and remove it from the table
1643 * if this is the last reference.
1644 *
1645 * We remove the cancel record from the table when we encounter its
1646 * last occurrence in the log so that if the same buffer is re-used
1647 * again after its last cancellation we actually replay the changes
1648 * made at that point.
1649 */
1650STATIC int
1651xlog_check_buffer_cancelled(
1652 xlog_t *log,
1653 xfs_daddr_t blkno,
1654 uint len,
1655 ushort flags)
1656{
1657 xfs_buf_cancel_t *bcp;
1658 xfs_buf_cancel_t *prevp;
1659 xfs_buf_cancel_t **bucket;
1660
1661 if (log->l_buf_cancel_table == NULL) {
1662 /*
1663 * There is nothing in the table built in pass one,
1664 * so this buffer must not be cancelled.
1665 */
1666 ASSERT(!(flags & XFS_BLI_CANCEL));
1667 return 0;
1668 }
1669
1670 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1671 XLOG_BC_TABLE_SIZE];
1672 bcp = *bucket;
1673 if (bcp == NULL) {
1674 /*
1675 * There is no corresponding entry in the table built
1676 * in pass one, so this buffer has not been cancelled.
1677 */
1678 ASSERT(!(flags & XFS_BLI_CANCEL));
1679 return 0;
1680 }
1681
1682 /*
1683 * Search for an entry in the buffer cancel table that
1684 * matches our buffer.
1685 */
1686 prevp = NULL;
1687 while (bcp != NULL) {
1688 if (bcp->bc_blkno == blkno && bcp->bc_len == len) {
1689 /*
1690 * We've go a match, so return 1 so that the
1691 * recovery of this buffer is cancelled.
1692 * If this buffer is actually a buffer cancel
1693 * log item, then decrement the refcount on the
1694 * one in the table and remove it if this is the
1695 * last reference.
1696 */
1697 if (flags & XFS_BLI_CANCEL) {
1698 bcp->bc_refcount--;
1699 if (bcp->bc_refcount == 0) {
1700 if (prevp == NULL) {
1701 *bucket = bcp->bc_next;
1702 } else {
1703 prevp->bc_next = bcp->bc_next;
1704 }
1705 kmem_free(bcp,
1706 sizeof(xfs_buf_cancel_t));
1707 }
1708 }
1709 return 1;
1710 }
1711 prevp = bcp;
1712 bcp = bcp->bc_next;
1713 }
1714 /*
1715 * We didn't find a corresponding entry in the table, so
1716 * return 0 so that the buffer is NOT cancelled.
1717 */
1718 ASSERT(!(flags & XFS_BLI_CANCEL));
1719 return 0;
1720}
1721
1722STATIC int
1723xlog_recover_do_buffer_pass2(
1724 xlog_t *log,
1725 xfs_buf_log_format_t *buf_f)
1726{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 xfs_daddr_t blkno = 0;
1728 ushort flags = 0;
1729 uint len = 0;
1730
1731 switch (buf_f->blf_type) {
1732 case XFS_LI_BUF:
1733 blkno = buf_f->blf_blkno;
1734 flags = buf_f->blf_flags;
1735 len = buf_f->blf_len;
1736 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 }
1738
1739 return xlog_check_buffer_cancelled(log, blkno, len, flags);
1740}
1741
1742/*
1743 * Perform recovery for a buffer full of inodes. In these buffers,
1744 * the only data which should be recovered is that which corresponds
1745 * to the di_next_unlinked pointers in the on disk inode structures.
1746 * The rest of the data for the inodes is always logged through the
1747 * inodes themselves rather than the inode buffer and is recovered
1748 * in xlog_recover_do_inode_trans().
1749 *
1750 * The only time when buffers full of inodes are fully recovered is
1751 * when the buffer is full of newly allocated inodes. In this case
1752 * the buffer will not be marked as an inode buffer and so will be
1753 * sent to xlog_recover_do_reg_buffer() below during recovery.
1754 */
1755STATIC int
1756xlog_recover_do_inode_buffer(
1757 xfs_mount_t *mp,
1758 xlog_recover_item_t *item,
1759 xfs_buf_t *bp,
1760 xfs_buf_log_format_t *buf_f)
1761{
1762 int i;
1763 int item_index;
1764 int bit;
1765 int nbits;
1766 int reg_buf_offset;
1767 int reg_buf_bytes;
1768 int next_unlinked_offset;
1769 int inodes_per_buf;
1770 xfs_agino_t *logged_nextp;
1771 xfs_agino_t *buffer_nextp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 unsigned int *data_map = NULL;
1773 unsigned int map_size = 0;
1774
1775 switch (buf_f->blf_type) {
1776 case XFS_LI_BUF:
1777 data_map = buf_f->blf_data_map;
1778 map_size = buf_f->blf_map_size;
1779 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 }
1781 /*
1782 * Set the variables corresponding to the current region to
1783 * 0 so that we'll initialize them on the first pass through
1784 * the loop.
1785 */
1786 reg_buf_offset = 0;
1787 reg_buf_bytes = 0;
1788 bit = 0;
1789 nbits = 0;
1790 item_index = 0;
1791 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1792 for (i = 0; i < inodes_per_buf; i++) {
1793 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1794 offsetof(xfs_dinode_t, di_next_unlinked);
1795
1796 while (next_unlinked_offset >=
1797 (reg_buf_offset + reg_buf_bytes)) {
1798 /*
1799 * The next di_next_unlinked field is beyond
1800 * the current logged region. Find the next
1801 * logged region that contains or is beyond
1802 * the current di_next_unlinked field.
1803 */
1804 bit += nbits;
1805 bit = xfs_next_bit(data_map, map_size, bit);
1806
1807 /*
1808 * If there are no more logged regions in the
1809 * buffer, then we're done.
1810 */
1811 if (bit == -1) {
1812 return 0;
1813 }
1814
1815 nbits = xfs_contig_bits(data_map, map_size,
1816 bit);
1817 ASSERT(nbits > 0);
1818 reg_buf_offset = bit << XFS_BLI_SHIFT;
1819 reg_buf_bytes = nbits << XFS_BLI_SHIFT;
1820 item_index++;
1821 }
1822
1823 /*
1824 * If the current logged region starts after the current
1825 * di_next_unlinked field, then move on to the next
1826 * di_next_unlinked field.
1827 */
1828 if (next_unlinked_offset < reg_buf_offset) {
1829 continue;
1830 }
1831
1832 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1833 ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0);
1834 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1835
1836 /*
1837 * The current logged region contains a copy of the
1838 * current di_next_unlinked field. Extract its value
1839 * and copy it to the buffer copy.
1840 */
1841 logged_nextp = (xfs_agino_t *)
1842 ((char *)(item->ri_buf[item_index].i_addr) +
1843 (next_unlinked_offset - reg_buf_offset));
1844 if (unlikely(*logged_nextp == 0)) {
1845 xfs_fs_cmn_err(CE_ALERT, mp,
1846 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1847 item, bp);
1848 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1849 XFS_ERRLEVEL_LOW, mp);
1850 return XFS_ERROR(EFSCORRUPTED);
1851 }
1852
1853 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1854 next_unlinked_offset);
Tim Shimmin87c199c2006-06-09 14:56:16 +10001855 *buffer_nextp = *logged_nextp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 }
1857
1858 return 0;
1859}
1860
1861/*
1862 * Perform a 'normal' buffer recovery. Each logged region of the
1863 * buffer should be copied over the corresponding region in the
1864 * given buffer. The bitmap in the buf log format structure indicates
1865 * where to place the logged data.
1866 */
1867/*ARGSUSED*/
1868STATIC void
1869xlog_recover_do_reg_buffer(
1870 xfs_mount_t *mp,
1871 xlog_recover_item_t *item,
1872 xfs_buf_t *bp,
1873 xfs_buf_log_format_t *buf_f)
1874{
1875 int i;
1876 int bit;
1877 int nbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 unsigned int *data_map = NULL;
1879 unsigned int map_size = 0;
1880 int error;
1881
1882 switch (buf_f->blf_type) {
1883 case XFS_LI_BUF:
1884 data_map = buf_f->blf_data_map;
1885 map_size = buf_f->blf_map_size;
1886 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 }
1888 bit = 0;
1889 i = 1; /* 0 is the buf format structure */
1890 while (1) {
1891 bit = xfs_next_bit(data_map, map_size, bit);
1892 if (bit == -1)
1893 break;
1894 nbits = xfs_contig_bits(data_map, map_size, bit);
1895 ASSERT(nbits > 0);
1896 ASSERT(item->ri_buf[i].i_addr != 0);
1897 ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0);
1898 ASSERT(XFS_BUF_COUNT(bp) >=
1899 ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT));
1900
1901 /*
1902 * Do a sanity check if this is a dquot buffer. Just checking
1903 * the first dquot in the buffer should do. XXXThis is
1904 * probably a good thing to do for other buf types also.
1905 */
1906 error = 0;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001907 if (buf_f->blf_flags &
1908 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 error = xfs_qm_dqcheck((xfs_disk_dquot_t *)
1910 item->ri_buf[i].i_addr,
1911 -1, 0, XFS_QMOPT_DOWARN,
1912 "dquot_buf_recover");
1913 }
1914 if (!error)
1915 memcpy(xfs_buf_offset(bp,
1916 (uint)bit << XFS_BLI_SHIFT), /* dest */
1917 item->ri_buf[i].i_addr, /* source */
1918 nbits<<XFS_BLI_SHIFT); /* length */
1919 i++;
1920 bit += nbits;
1921 }
1922
1923 /* Shouldn't be any more regions */
1924 ASSERT(i == item->ri_total);
1925}
1926
1927/*
1928 * Do some primitive error checking on ondisk dquot data structures.
1929 */
1930int
1931xfs_qm_dqcheck(
1932 xfs_disk_dquot_t *ddq,
1933 xfs_dqid_t id,
1934 uint type, /* used only when IO_dorepair is true */
1935 uint flags,
1936 char *str)
1937{
1938 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
1939 int errs = 0;
1940
1941 /*
1942 * We can encounter an uninitialized dquot buffer for 2 reasons:
1943 * 1. If we crash while deleting the quotainode(s), and those blks got
1944 * used for user data. This is because we take the path of regular
1945 * file deletion; however, the size field of quotainodes is never
1946 * updated, so all the tricks that we play in itruncate_finish
1947 * don't quite matter.
1948 *
1949 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1950 * But the allocation will be replayed so we'll end up with an
1951 * uninitialized quota block.
1952 *
1953 * This is all fine; things are still consistent, and we haven't lost
1954 * any quota information. Just don't complain about bad dquot blks.
1955 */
Christoph Hellwig1149d962005-11-02 15:01:12 +11001956 if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 if (flags & XFS_QMOPT_DOWARN)
1958 cmn_err(CE_ALERT,
1959 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11001960 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 errs++;
1962 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11001963 if (ddq->d_version != XFS_DQUOT_VERSION) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 if (flags & XFS_QMOPT_DOWARN)
1965 cmn_err(CE_ALERT,
1966 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11001967 str, id, ddq->d_version, XFS_DQUOT_VERSION);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 errs++;
1969 }
1970
Christoph Hellwig1149d962005-11-02 15:01:12 +11001971 if (ddq->d_flags != XFS_DQ_USER &&
1972 ddq->d_flags != XFS_DQ_PROJ &&
1973 ddq->d_flags != XFS_DQ_GROUP) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 if (flags & XFS_QMOPT_DOWARN)
1975 cmn_err(CE_ALERT,
1976 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11001977 str, id, ddq->d_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 errs++;
1979 }
1980
Christoph Hellwig1149d962005-11-02 15:01:12 +11001981 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 if (flags & XFS_QMOPT_DOWARN)
1983 cmn_err(CE_ALERT,
1984 "%s : ondisk-dquot 0x%p, ID mismatch: "
1985 "0x%x expected, found id 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11001986 str, ddq, id, be32_to_cpu(ddq->d_id));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 errs++;
1988 }
1989
1990 if (!errs && ddq->d_id) {
Christoph Hellwig1149d962005-11-02 15:01:12 +11001991 if (ddq->d_blk_softlimit &&
1992 be64_to_cpu(ddq->d_bcount) >=
1993 be64_to_cpu(ddq->d_blk_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 if (!ddq->d_btimer) {
1995 if (flags & XFS_QMOPT_DOWARN)
1996 cmn_err(CE_ALERT,
1997 "%s : Dquot ID 0x%x (0x%p) "
1998 "BLK TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11001999 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 errs++;
2001 }
2002 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11002003 if (ddq->d_ino_softlimit &&
2004 be64_to_cpu(ddq->d_icount) >=
2005 be64_to_cpu(ddq->d_ino_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 if (!ddq->d_itimer) {
2007 if (flags & XFS_QMOPT_DOWARN)
2008 cmn_err(CE_ALERT,
2009 "%s : Dquot ID 0x%x (0x%p) "
2010 "INODE TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002011 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 errs++;
2013 }
2014 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11002015 if (ddq->d_rtb_softlimit &&
2016 be64_to_cpu(ddq->d_rtbcount) >=
2017 be64_to_cpu(ddq->d_rtb_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 if (!ddq->d_rtbtimer) {
2019 if (flags & XFS_QMOPT_DOWARN)
2020 cmn_err(CE_ALERT,
2021 "%s : Dquot ID 0x%x (0x%p) "
2022 "RTBLK TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002023 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 errs++;
2025 }
2026 }
2027 }
2028
2029 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2030 return errs;
2031
2032 if (flags & XFS_QMOPT_DOWARN)
2033 cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id);
2034
2035 /*
2036 * Typically, a repair is only requested by quotacheck.
2037 */
2038 ASSERT(id != -1);
2039 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2040 memset(d, 0, sizeof(xfs_dqblk_t));
Christoph Hellwig1149d962005-11-02 15:01:12 +11002041
2042 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2043 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2044 d->dd_diskdq.d_flags = type;
2045 d->dd_diskdq.d_id = cpu_to_be32(id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046
2047 return errs;
2048}
2049
2050/*
2051 * Perform a dquot buffer recovery.
2052 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2053 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2054 * Else, treat it as a regular buffer and do recovery.
2055 */
2056STATIC void
2057xlog_recover_do_dquot_buffer(
2058 xfs_mount_t *mp,
2059 xlog_t *log,
2060 xlog_recover_item_t *item,
2061 xfs_buf_t *bp,
2062 xfs_buf_log_format_t *buf_f)
2063{
2064 uint type;
2065
2066 /*
2067 * Filesystems are required to send in quota flags at mount time.
2068 */
2069 if (mp->m_qflags == 0) {
2070 return;
2071 }
2072
2073 type = 0;
2074 if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF)
2075 type |= XFS_DQ_USER;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002076 if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF)
2077 type |= XFS_DQ_PROJ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF)
2079 type |= XFS_DQ_GROUP;
2080 /*
2081 * This type of quotas was turned off, so ignore this buffer
2082 */
2083 if (log->l_quotaoffs_flag & type)
2084 return;
2085
2086 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2087}
2088
2089/*
2090 * This routine replays a modification made to a buffer at runtime.
2091 * There are actually two types of buffer, regular and inode, which
2092 * are handled differently. Inode buffers are handled differently
2093 * in that we only recover a specific set of data from them, namely
2094 * the inode di_next_unlinked fields. This is because all other inode
2095 * data is actually logged via inode records and any data we replay
2096 * here which overlaps that may be stale.
2097 *
2098 * When meta-data buffers are freed at run time we log a buffer item
2099 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
2100 * of the buffer in the log should not be replayed at recovery time.
2101 * This is so that if the blocks covered by the buffer are reused for
2102 * file data before we crash we don't end up replaying old, freed
2103 * meta-data into a user's file.
2104 *
2105 * To handle the cancellation of buffer log items, we make two passes
2106 * over the log during recovery. During the first we build a table of
2107 * those buffers which have been cancelled, and during the second we
2108 * only replay those buffers which do not have corresponding cancel
2109 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2110 * for more details on the implementation of the table of cancel records.
2111 */
2112STATIC int
2113xlog_recover_do_buffer_trans(
2114 xlog_t *log,
2115 xlog_recover_item_t *item,
2116 int pass)
2117{
2118 xfs_buf_log_format_t *buf_f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 xfs_mount_t *mp;
2120 xfs_buf_t *bp;
2121 int error;
2122 int cancel;
2123 xfs_daddr_t blkno;
2124 int len;
2125 ushort flags;
2126
2127 buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr;
2128
2129 if (pass == XLOG_RECOVER_PASS1) {
2130 /*
2131 * In this pass we're only looking for buf items
2132 * with the XFS_BLI_CANCEL bit set.
2133 */
2134 xlog_recover_do_buffer_pass1(log, buf_f);
2135 return 0;
2136 } else {
2137 /*
2138 * In this pass we want to recover all the buffers
2139 * which have not been cancelled and are not
2140 * cancellation buffers themselves. The routine
2141 * we call here will tell us whether or not to
2142 * continue with the replay of this buffer.
2143 */
2144 cancel = xlog_recover_do_buffer_pass2(log, buf_f);
2145 if (cancel) {
2146 return 0;
2147 }
2148 }
2149 switch (buf_f->blf_type) {
2150 case XFS_LI_BUF:
2151 blkno = buf_f->blf_blkno;
2152 len = buf_f->blf_len;
2153 flags = buf_f->blf_flags;
2154 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 default:
2156 xfs_fs_cmn_err(CE_ALERT, log->l_mp,
Nathan Scottfc1f8c12005-11-02 11:44:33 +11002157 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2158 buf_f->blf_type, log->l_mp->m_logname ?
2159 log->l_mp->m_logname : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2161 XFS_ERRLEVEL_LOW, log->l_mp);
2162 return XFS_ERROR(EFSCORRUPTED);
2163 }
2164
2165 mp = log->l_mp;
2166 if (flags & XFS_BLI_INODE_BUF) {
2167 bp = xfs_buf_read_flags(mp->m_ddev_targp, blkno, len,
2168 XFS_BUF_LOCK);
2169 } else {
2170 bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, 0);
2171 }
2172 if (XFS_BUF_ISERROR(bp)) {
2173 xfs_ioerror_alert("xlog_recover_do..(read#1)", log->l_mp,
2174 bp, blkno);
2175 error = XFS_BUF_GETERROR(bp);
2176 xfs_buf_relse(bp);
2177 return error;
2178 }
2179
2180 error = 0;
2181 if (flags & XFS_BLI_INODE_BUF) {
2182 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002183 } else if (flags &
2184 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2186 } else {
2187 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2188 }
2189 if (error)
2190 return XFS_ERROR(error);
2191
2192 /*
2193 * Perform delayed write on the buffer. Asynchronous writes will be
2194 * slower when taking into account all the buffers to be flushed.
2195 *
2196 * Also make sure that only inode buffers with good sizes stay in
2197 * the buffer cache. The kernel moves inodes in buffers of 1 block
2198 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2199 * buffers in the log can be a different size if the log was generated
2200 * by an older kernel using unclustered inode buffers or a newer kernel
2201 * running with a different inode cluster size. Regardless, if the
2202 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2203 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2204 * the buffer out of the buffer cache so that the buffer won't
2205 * overlap with future reads of those inodes.
2206 */
2207 if (XFS_DINODE_MAGIC ==
2208 INT_GET(*((__uint16_t *)(xfs_buf_offset(bp, 0))), ARCH_CONVERT) &&
2209 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2210 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2211 XFS_BUF_STALE(bp);
2212 error = xfs_bwrite(mp, bp);
2213 } else {
2214 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2215 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2216 XFS_BUF_SET_FSPRIVATE(bp, mp);
2217 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2218 xfs_bdwrite(mp, bp);
2219 }
2220
2221 return (error);
2222}
2223
2224STATIC int
2225xlog_recover_do_inode_trans(
2226 xlog_t *log,
2227 xlog_recover_item_t *item,
2228 int pass)
2229{
2230 xfs_inode_log_format_t *in_f;
2231 xfs_mount_t *mp;
2232 xfs_buf_t *bp;
2233 xfs_imap_t imap;
2234 xfs_dinode_t *dip;
2235 xfs_ino_t ino;
2236 int len;
2237 xfs_caddr_t src;
2238 xfs_caddr_t dest;
2239 int error;
2240 int attr_index;
2241 uint fields;
2242 xfs_dinode_core_t *dicp;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002243 int need_free = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244
2245 if (pass == XLOG_RECOVER_PASS1) {
2246 return 0;
2247 }
2248
Tim Shimmin6d192a92006-06-09 14:55:38 +10002249 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2250 in_f = (xfs_inode_log_format_t *)item->ri_buf[0].i_addr;
2251 } else {
2252 in_f = (xfs_inode_log_format_t *)kmem_alloc(
2253 sizeof(xfs_inode_log_format_t), KM_SLEEP);
2254 need_free = 1;
2255 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2256 if (error)
2257 goto error;
2258 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 ino = in_f->ilf_ino;
2260 mp = log->l_mp;
2261 if (ITEM_TYPE(item) == XFS_LI_INODE) {
2262 imap.im_blkno = (xfs_daddr_t)in_f->ilf_blkno;
2263 imap.im_len = in_f->ilf_len;
2264 imap.im_boffset = in_f->ilf_boffset;
2265 } else {
2266 /*
2267 * It's an old inode format record. We don't know where
2268 * its cluster is located on disk, and we can't allow
2269 * xfs_imap() to figure it out because the inode btrees
2270 * are not ready to be used. Therefore do not pass the
2271 * XFS_IMAP_LOOKUP flag to xfs_imap(). This will give
2272 * us only the single block in which the inode lives
2273 * rather than its cluster, so we must make sure to
2274 * invalidate the buffer when we write it out below.
2275 */
2276 imap.im_blkno = 0;
2277 xfs_imap(log->l_mp, NULL, ino, &imap, 0);
2278 }
2279
2280 /*
2281 * Inode buffers can be freed, look out for it,
2282 * and do not replay the inode.
2283 */
Tim Shimmin6d192a92006-06-09 14:55:38 +10002284 if (xlog_check_buffer_cancelled(log, imap.im_blkno, imap.im_len, 0)) {
2285 error = 0;
2286 goto error;
2287 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288
2289 bp = xfs_buf_read_flags(mp->m_ddev_targp, imap.im_blkno, imap.im_len,
2290 XFS_BUF_LOCK);
2291 if (XFS_BUF_ISERROR(bp)) {
2292 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
2293 bp, imap.im_blkno);
2294 error = XFS_BUF_GETERROR(bp);
2295 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002296 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 }
2298 error = 0;
2299 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2300 dip = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
2301
2302 /*
2303 * Make sure the place we're flushing out to really looks
2304 * like an inode!
2305 */
2306 if (unlikely(INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC)) {
2307 xfs_buf_relse(bp);
2308 xfs_fs_cmn_err(CE_ALERT, mp,
2309 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2310 dip, bp, ino);
2311 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2312 XFS_ERRLEVEL_LOW, mp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002313 error = EFSCORRUPTED;
2314 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 }
2316 dicp = (xfs_dinode_core_t*)(item->ri_buf[1].i_addr);
2317 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2318 xfs_buf_relse(bp);
2319 xfs_fs_cmn_err(CE_ALERT, mp,
2320 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2321 item, ino);
2322 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2323 XFS_ERRLEVEL_LOW, mp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002324 error = EFSCORRUPTED;
2325 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 }
2327
2328 /* Skip replay when the on disk inode is newer than the log one */
2329 if (dicp->di_flushiter <
2330 INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT)) {
2331 /*
2332 * Deal with the wrap case, DI_MAX_FLUSH is less
2333 * than smaller numbers
2334 */
2335 if ((INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT)
2336 == DI_MAX_FLUSH) &&
2337 (dicp->di_flushiter < (DI_MAX_FLUSH>>1))) {
2338 /* do nothing */
2339 } else {
2340 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002341 error = 0;
2342 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 }
2344 }
2345 /* Take the opportunity to reset the flush iteration count */
2346 dicp->di_flushiter = 0;
2347
2348 if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) {
2349 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2350 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2351 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2352 XFS_ERRLEVEL_LOW, mp, dicp);
2353 xfs_buf_relse(bp);
2354 xfs_fs_cmn_err(CE_ALERT, mp,
2355 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2356 item, dip, bp, ino);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002357 error = EFSCORRUPTED;
2358 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 }
2360 } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) {
2361 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2362 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2363 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2364 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2365 XFS_ERRLEVEL_LOW, mp, dicp);
2366 xfs_buf_relse(bp);
2367 xfs_fs_cmn_err(CE_ALERT, mp,
2368 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2369 item, dip, bp, ino);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002370 error = EFSCORRUPTED;
2371 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 }
2373 }
2374 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2375 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2376 XFS_ERRLEVEL_LOW, mp, dicp);
2377 xfs_buf_relse(bp);
2378 xfs_fs_cmn_err(CE_ALERT, mp,
2379 "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",
2380 item, dip, bp, ino,
2381 dicp->di_nextents + dicp->di_anextents,
2382 dicp->di_nblocks);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002383 error = EFSCORRUPTED;
2384 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 }
2386 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2387 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2388 XFS_ERRLEVEL_LOW, mp, dicp);
2389 xfs_buf_relse(bp);
2390 xfs_fs_cmn_err(CE_ALERT, mp,
2391 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2392 item, dip, bp, ino, dicp->di_forkoff);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002393 error = EFSCORRUPTED;
2394 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 }
2396 if (unlikely(item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t))) {
2397 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2398 XFS_ERRLEVEL_LOW, mp, dicp);
2399 xfs_buf_relse(bp);
2400 xfs_fs_cmn_err(CE_ALERT, mp,
2401 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2402 item->ri_buf[1].i_len, item);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002403 error = EFSCORRUPTED;
2404 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 }
2406
2407 /* The core is in in-core format */
2408 xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
2409 (xfs_dinode_core_t*)item->ri_buf[1].i_addr, -1);
2410
2411 /* the rest is in on-disk format */
2412 if (item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t)) {
2413 memcpy((xfs_caddr_t) dip + sizeof(xfs_dinode_core_t),
2414 item->ri_buf[1].i_addr + sizeof(xfs_dinode_core_t),
2415 item->ri_buf[1].i_len - sizeof(xfs_dinode_core_t));
2416 }
2417
2418 fields = in_f->ilf_fields;
2419 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2420 case XFS_ILOG_DEV:
2421 INT_SET(dip->di_u.di_dev, ARCH_CONVERT, in_f->ilf_u.ilfu_rdev);
2422
2423 break;
2424 case XFS_ILOG_UUID:
2425 dip->di_u.di_muuid = in_f->ilf_u.ilfu_uuid;
2426 break;
2427 }
2428
2429 if (in_f->ilf_size == 2)
2430 goto write_inode_buffer;
2431 len = item->ri_buf[2].i_len;
2432 src = item->ri_buf[2].i_addr;
2433 ASSERT(in_f->ilf_size <= 4);
2434 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2435 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2436 (len == in_f->ilf_dsize));
2437
2438 switch (fields & XFS_ILOG_DFORK) {
2439 case XFS_ILOG_DDATA:
2440 case XFS_ILOG_DEXT:
2441 memcpy(&dip->di_u, src, len);
2442 break;
2443
2444 case XFS_ILOG_DBROOT:
2445 xfs_bmbt_to_bmdr((xfs_bmbt_block_t *)src, len,
2446 &(dip->di_u.di_bmbt),
2447 XFS_DFORK_DSIZE(dip, mp));
2448 break;
2449
2450 default:
2451 /*
2452 * There are no data fork flags set.
2453 */
2454 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2455 break;
2456 }
2457
2458 /*
2459 * If we logged any attribute data, recover it. There may or
2460 * may not have been any other non-core data logged in this
2461 * transaction.
2462 */
2463 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2464 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2465 attr_index = 3;
2466 } else {
2467 attr_index = 2;
2468 }
2469 len = item->ri_buf[attr_index].i_len;
2470 src = item->ri_buf[attr_index].i_addr;
2471 ASSERT(len == in_f->ilf_asize);
2472
2473 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2474 case XFS_ILOG_ADATA:
2475 case XFS_ILOG_AEXT:
2476 dest = XFS_DFORK_APTR(dip);
2477 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2478 memcpy(dest, src, len);
2479 break;
2480
2481 case XFS_ILOG_ABROOT:
2482 dest = XFS_DFORK_APTR(dip);
2483 xfs_bmbt_to_bmdr((xfs_bmbt_block_t *)src, len,
2484 (xfs_bmdr_block_t*)dest,
2485 XFS_DFORK_ASIZE(dip, mp));
2486 break;
2487
2488 default:
2489 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2490 ASSERT(0);
2491 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002492 error = EIO;
2493 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 }
2495 }
2496
2497write_inode_buffer:
2498 if (ITEM_TYPE(item) == XFS_LI_INODE) {
2499 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2500 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2501 XFS_BUF_SET_FSPRIVATE(bp, mp);
2502 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2503 xfs_bdwrite(mp, bp);
2504 } else {
2505 XFS_BUF_STALE(bp);
2506 error = xfs_bwrite(mp, bp);
2507 }
2508
Tim Shimmin6d192a92006-06-09 14:55:38 +10002509error:
2510 if (need_free)
2511 kmem_free(in_f, sizeof(*in_f));
2512 return XFS_ERROR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513}
2514
2515/*
2516 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2517 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2518 * of that type.
2519 */
2520STATIC int
2521xlog_recover_do_quotaoff_trans(
2522 xlog_t *log,
2523 xlog_recover_item_t *item,
2524 int pass)
2525{
2526 xfs_qoff_logformat_t *qoff_f;
2527
2528 if (pass == XLOG_RECOVER_PASS2) {
2529 return (0);
2530 }
2531
2532 qoff_f = (xfs_qoff_logformat_t *)item->ri_buf[0].i_addr;
2533 ASSERT(qoff_f);
2534
2535 /*
2536 * The logitem format's flag tells us if this was user quotaoff,
Nathan Scott77a7cce2006-01-11 15:35:57 +11002537 * group/project quotaoff or both.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 */
2539 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2540 log->l_quotaoffs_flag |= XFS_DQ_USER;
Nathan Scott77a7cce2006-01-11 15:35:57 +11002541 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2542 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2544 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2545
2546 return (0);
2547}
2548
2549/*
2550 * Recover a dquot record
2551 */
2552STATIC int
2553xlog_recover_do_dquot_trans(
2554 xlog_t *log,
2555 xlog_recover_item_t *item,
2556 int pass)
2557{
2558 xfs_mount_t *mp;
2559 xfs_buf_t *bp;
2560 struct xfs_disk_dquot *ddq, *recddq;
2561 int error;
2562 xfs_dq_logformat_t *dq_f;
2563 uint type;
2564
2565 if (pass == XLOG_RECOVER_PASS1) {
2566 return 0;
2567 }
2568 mp = log->l_mp;
2569
2570 /*
2571 * Filesystems are required to send in quota flags at mount time.
2572 */
2573 if (mp->m_qflags == 0)
2574 return (0);
2575
2576 recddq = (xfs_disk_dquot_t *)item->ri_buf[1].i_addr;
2577 ASSERT(recddq);
2578 /*
2579 * This type of quotas was turned off, so ignore this record.
2580 */
2581 type = INT_GET(recddq->d_flags, ARCH_CONVERT) &
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002582 (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 ASSERT(type);
2584 if (log->l_quotaoffs_flag & type)
2585 return (0);
2586
2587 /*
2588 * At this point we know that quota was _not_ turned off.
2589 * Since the mount flags are not indicating to us otherwise, this
2590 * must mean that quota is on, and the dquot needs to be replayed.
2591 * Remember that we may not have fully recovered the superblock yet,
2592 * so we can't do the usual trick of looking at the SB quota bits.
2593 *
2594 * The other possibility, of course, is that the quota subsystem was
2595 * removed since the last mount - ENOSYS.
2596 */
2597 dq_f = (xfs_dq_logformat_t *)item->ri_buf[0].i_addr;
2598 ASSERT(dq_f);
2599 if ((error = xfs_qm_dqcheck(recddq,
2600 dq_f->qlf_id,
2601 0, XFS_QMOPT_DOWARN,
2602 "xlog_recover_do_dquot_trans (log copy)"))) {
2603 return XFS_ERROR(EIO);
2604 }
2605 ASSERT(dq_f->qlf_len == 1);
2606
2607 error = xfs_read_buf(mp, mp->m_ddev_targp,
2608 dq_f->qlf_blkno,
2609 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2610 0, &bp);
2611 if (error) {
2612 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
2613 bp, dq_f->qlf_blkno);
2614 return error;
2615 }
2616 ASSERT(bp);
2617 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2618
2619 /*
2620 * At least the magic num portion should be on disk because this
2621 * was among a chunk of dquots created earlier, and we did some
2622 * minimal initialization then.
2623 */
2624 if (xfs_qm_dqcheck(ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2625 "xlog_recover_do_dquot_trans")) {
2626 xfs_buf_relse(bp);
2627 return XFS_ERROR(EIO);
2628 }
2629
2630 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2631
2632 ASSERT(dq_f->qlf_size == 2);
2633 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
2634 XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
2635 XFS_BUF_SET_FSPRIVATE(bp, mp);
2636 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2637 xfs_bdwrite(mp, bp);
2638
2639 return (0);
2640}
2641
2642/*
2643 * This routine is called to create an in-core extent free intent
2644 * item from the efi format structure which was logged on disk.
2645 * It allocates an in-core efi, copies the extents from the format
2646 * structure into it, and adds the efi to the AIL with the given
2647 * LSN.
2648 */
Tim Shimmin6d192a92006-06-09 14:55:38 +10002649STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650xlog_recover_do_efi_trans(
2651 xlog_t *log,
2652 xlog_recover_item_t *item,
2653 xfs_lsn_t lsn,
2654 int pass)
2655{
Tim Shimmin6d192a92006-06-09 14:55:38 +10002656 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 xfs_mount_t *mp;
2658 xfs_efi_log_item_t *efip;
2659 xfs_efi_log_format_t *efi_formatp;
2660 SPLDECL(s);
2661
2662 if (pass == XLOG_RECOVER_PASS1) {
Tim Shimmin6d192a92006-06-09 14:55:38 +10002663 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664 }
2665
2666 efi_formatp = (xfs_efi_log_format_t *)item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667
2668 mp = log->l_mp;
2669 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002670 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2671 &(efip->efi_format)))) {
2672 xfs_efi_item_free(efip);
2673 return error;
2674 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 efip->efi_next_extent = efi_formatp->efi_nextents;
2676 efip->efi_flags |= XFS_EFI_COMMITTED;
2677
2678 AIL_LOCK(mp,s);
2679 /*
2680 * xfs_trans_update_ail() drops the AIL lock.
2681 */
2682 xfs_trans_update_ail(mp, (xfs_log_item_t *)efip, lsn, s);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002683 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684}
2685
2686
2687/*
2688 * This routine is called when an efd format structure is found in
2689 * a committed transaction in the log. It's purpose is to cancel
2690 * the corresponding efi if it was still in the log. To do this
2691 * it searches the AIL for the efi with an id equal to that in the
2692 * efd format structure. If we find it, we remove the efi from the
2693 * AIL and free it.
2694 */
2695STATIC void
2696xlog_recover_do_efd_trans(
2697 xlog_t *log,
2698 xlog_recover_item_t *item,
2699 int pass)
2700{
2701 xfs_mount_t *mp;
2702 xfs_efd_log_format_t *efd_formatp;
2703 xfs_efi_log_item_t *efip = NULL;
2704 xfs_log_item_t *lip;
2705 int gen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 __uint64_t efi_id;
2707 SPLDECL(s);
2708
2709 if (pass == XLOG_RECOVER_PASS1) {
2710 return;
2711 }
2712
2713 efd_formatp = (xfs_efd_log_format_t *)item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002714 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2715 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2716 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2717 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 efi_id = efd_formatp->efd_efi_id;
2719
2720 /*
2721 * Search for the efi with the id in the efd format structure
2722 * in the AIL.
2723 */
2724 mp = log->l_mp;
2725 AIL_LOCK(mp,s);
2726 lip = xfs_trans_first_ail(mp, &gen);
2727 while (lip != NULL) {
2728 if (lip->li_type == XFS_LI_EFI) {
2729 efip = (xfs_efi_log_item_t *)lip;
2730 if (efip->efi_format.efi_id == efi_id) {
2731 /*
2732 * xfs_trans_delete_ail() drops the
2733 * AIL lock.
2734 */
2735 xfs_trans_delete_ail(mp, lip, s);
2736 break;
2737 }
2738 }
2739 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
2740 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741
2742 /*
2743 * If we found it, then free it up. If it wasn't there, it
2744 * must have been overwritten in the log. Oh well.
2745 */
2746 if (lip != NULL) {
Christoph Hellwig7d795ca2005-06-21 15:41:19 +10002747 xfs_efi_item_free(efip);
2748 } else {
2749 AIL_UNLOCK(mp, s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750 }
2751}
2752
2753/*
2754 * Perform the transaction
2755 *
2756 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2757 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2758 */
2759STATIC int
2760xlog_recover_do_trans(
2761 xlog_t *log,
2762 xlog_recover_t *trans,
2763 int pass)
2764{
2765 int error = 0;
2766 xlog_recover_item_t *item, *first_item;
2767
2768 if ((error = xlog_recover_reorder_trans(log, trans)))
2769 return error;
2770 first_item = item = trans->r_itemq;
2771 do {
2772 /*
2773 * we don't need to worry about the block number being
2774 * truncated in > 1 TB buffers because in user-land,
2775 * we're now n32 or 64-bit so xfs_daddr_t is 64-bits so
Nathan Scottc41564b2006-03-29 08:55:14 +10002776 * the blknos will get through the user-mode buffer
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777 * cache properly. The only bad case is o32 kernels
2778 * where xfs_daddr_t is 32-bits but mount will warn us
2779 * off a > 1 TB filesystem before we get here.
2780 */
Eric Sandeen804195b2007-02-10 18:35:02 +11002781 if ((ITEM_TYPE(item) == XFS_LI_BUF)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 if ((error = xlog_recover_do_buffer_trans(log, item,
2783 pass)))
2784 break;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002785 } else if ((ITEM_TYPE(item) == XFS_LI_INODE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 if ((error = xlog_recover_do_inode_trans(log, item,
2787 pass)))
2788 break;
2789 } else if (ITEM_TYPE(item) == XFS_LI_EFI) {
Tim Shimmin6d192a92006-06-09 14:55:38 +10002790 if ((error = xlog_recover_do_efi_trans(log, item, trans->r_lsn,
2791 pass)))
2792 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793 } else if (ITEM_TYPE(item) == XFS_LI_EFD) {
2794 xlog_recover_do_efd_trans(log, item, pass);
2795 } else if (ITEM_TYPE(item) == XFS_LI_DQUOT) {
2796 if ((error = xlog_recover_do_dquot_trans(log, item,
2797 pass)))
2798 break;
2799 } else if ((ITEM_TYPE(item) == XFS_LI_QUOTAOFF)) {
2800 if ((error = xlog_recover_do_quotaoff_trans(log, item,
2801 pass)))
2802 break;
2803 } else {
2804 xlog_warn("XFS: xlog_recover_do_trans");
2805 ASSERT(0);
2806 error = XFS_ERROR(EIO);
2807 break;
2808 }
2809 item = item->ri_next;
2810 } while (first_item != item);
2811
2812 return error;
2813}
2814
2815/*
2816 * Free up any resources allocated by the transaction
2817 *
2818 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2819 */
2820STATIC void
2821xlog_recover_free_trans(
2822 xlog_recover_t *trans)
2823{
2824 xlog_recover_item_t *first_item, *item, *free_item;
2825 int i;
2826
2827 item = first_item = trans->r_itemq;
2828 do {
2829 free_item = item;
2830 item = item->ri_next;
2831 /* Free the regions in the item. */
2832 for (i = 0; i < free_item->ri_cnt; i++) {
2833 kmem_free(free_item->ri_buf[i].i_addr,
2834 free_item->ri_buf[i].i_len);
2835 }
2836 /* Free the item itself */
2837 kmem_free(free_item->ri_buf,
2838 (free_item->ri_total * sizeof(xfs_log_iovec_t)));
2839 kmem_free(free_item, sizeof(xlog_recover_item_t));
2840 } while (first_item != item);
2841 /* Free the transaction recover structure */
2842 kmem_free(trans, sizeof(xlog_recover_t));
2843}
2844
2845STATIC int
2846xlog_recover_commit_trans(
2847 xlog_t *log,
2848 xlog_recover_t **q,
2849 xlog_recover_t *trans,
2850 int pass)
2851{
2852 int error;
2853
2854 if ((error = xlog_recover_unlink_tid(q, trans)))
2855 return error;
2856 if ((error = xlog_recover_do_trans(log, trans, pass)))
2857 return error;
2858 xlog_recover_free_trans(trans); /* no error */
2859 return 0;
2860}
2861
2862STATIC int
2863xlog_recover_unmount_trans(
2864 xlog_recover_t *trans)
2865{
2866 /* Do nothing now */
2867 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2868 return 0;
2869}
2870
2871/*
2872 * There are two valid states of the r_state field. 0 indicates that the
2873 * transaction structure is in a normal state. We have either seen the
2874 * start of the transaction or the last operation we added was not a partial
2875 * operation. If the last operation we added to the transaction was a
2876 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2877 *
2878 * NOTE: skip LRs with 0 data length.
2879 */
2880STATIC int
2881xlog_recover_process_data(
2882 xlog_t *log,
2883 xlog_recover_t *rhash[],
2884 xlog_rec_header_t *rhead,
2885 xfs_caddr_t dp,
2886 int pass)
2887{
2888 xfs_caddr_t lp;
2889 int num_logops;
2890 xlog_op_header_t *ohead;
2891 xlog_recover_t *trans;
2892 xlog_tid_t tid;
2893 int error;
2894 unsigned long hash;
2895 uint flags;
2896
2897 lp = dp + INT_GET(rhead->h_len, ARCH_CONVERT);
2898 num_logops = INT_GET(rhead->h_num_logops, ARCH_CONVERT);
2899
2900 /* check the log format matches our own - else we can't recover */
2901 if (xlog_header_check_recover(log->l_mp, rhead))
2902 return (XFS_ERROR(EIO));
2903
2904 while ((dp < lp) && num_logops) {
2905 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2906 ohead = (xlog_op_header_t *)dp;
2907 dp += sizeof(xlog_op_header_t);
2908 if (ohead->oh_clientid != XFS_TRANSACTION &&
2909 ohead->oh_clientid != XFS_LOG) {
2910 xlog_warn(
2911 "XFS: xlog_recover_process_data: bad clientid");
2912 ASSERT(0);
2913 return (XFS_ERROR(EIO));
2914 }
2915 tid = INT_GET(ohead->oh_tid, ARCH_CONVERT);
2916 hash = XLOG_RHASH(tid);
2917 trans = xlog_recover_find_tid(rhash[hash], tid);
2918 if (trans == NULL) { /* not found; add new tid */
2919 if (ohead->oh_flags & XLOG_START_TRANS)
2920 xlog_recover_new_tid(&rhash[hash], tid,
2921 INT_GET(rhead->h_lsn, ARCH_CONVERT));
2922 } else {
2923 ASSERT(dp+INT_GET(ohead->oh_len, ARCH_CONVERT) <= lp);
2924 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2925 if (flags & XLOG_WAS_CONT_TRANS)
2926 flags &= ~XLOG_CONTINUE_TRANS;
2927 switch (flags) {
2928 case XLOG_COMMIT_TRANS:
2929 error = xlog_recover_commit_trans(log,
2930 &rhash[hash], trans, pass);
2931 break;
2932 case XLOG_UNMOUNT_TRANS:
2933 error = xlog_recover_unmount_trans(trans);
2934 break;
2935 case XLOG_WAS_CONT_TRANS:
2936 error = xlog_recover_add_to_cont_trans(trans,
2937 dp, INT_GET(ohead->oh_len,
2938 ARCH_CONVERT));
2939 break;
2940 case XLOG_START_TRANS:
2941 xlog_warn(
2942 "XFS: xlog_recover_process_data: bad transaction");
2943 ASSERT(0);
2944 error = XFS_ERROR(EIO);
2945 break;
2946 case 0:
2947 case XLOG_CONTINUE_TRANS:
2948 error = xlog_recover_add_to_trans(trans,
2949 dp, INT_GET(ohead->oh_len,
2950 ARCH_CONVERT));
2951 break;
2952 default:
2953 xlog_warn(
2954 "XFS: xlog_recover_process_data: bad flag");
2955 ASSERT(0);
2956 error = XFS_ERROR(EIO);
2957 break;
2958 }
2959 if (error)
2960 return error;
2961 }
2962 dp += INT_GET(ohead->oh_len, ARCH_CONVERT);
2963 num_logops--;
2964 }
2965 return 0;
2966}
2967
2968/*
2969 * Process an extent free intent item that was recovered from
2970 * the log. We need to free the extents that it describes.
2971 */
2972STATIC void
2973xlog_recover_process_efi(
2974 xfs_mount_t *mp,
2975 xfs_efi_log_item_t *efip)
2976{
2977 xfs_efd_log_item_t *efdp;
2978 xfs_trans_t *tp;
2979 int i;
2980 xfs_extent_t *extp;
2981 xfs_fsblock_t startblock_fsb;
2982
2983 ASSERT(!(efip->efi_flags & XFS_EFI_RECOVERED));
2984
2985 /*
2986 * First check the validity of the extents described by the
2987 * EFI. If any are bad, then assume that all are bad and
2988 * just toss the EFI.
2989 */
2990 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2991 extp = &(efip->efi_format.efi_extents[i]);
2992 startblock_fsb = XFS_BB_TO_FSB(mp,
2993 XFS_FSB_TO_DADDR(mp, extp->ext_start));
2994 if ((startblock_fsb == 0) ||
2995 (extp->ext_len == 0) ||
2996 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
2997 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
2998 /*
2999 * This will pull the EFI from the AIL and
3000 * free the memory associated with it.
3001 */
3002 xfs_efi_release(efip, efip->efi_format.efi_nextents);
3003 return;
3004 }
3005 }
3006
3007 tp = xfs_trans_alloc(mp, 0);
3008 xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
3009 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3010
3011 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3012 extp = &(efip->efi_format.efi_extents[i]);
3013 xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3014 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3015 extp->ext_len);
3016 }
3017
3018 efip->efi_flags |= XFS_EFI_RECOVERED;
3019 xfs_trans_commit(tp, 0, NULL);
3020}
3021
3022/*
3023 * Verify that once we've encountered something other than an EFI
3024 * in the AIL that there are no more EFIs in the AIL.
3025 */
3026#if defined(DEBUG)
3027STATIC void
3028xlog_recover_check_ail(
3029 xfs_mount_t *mp,
3030 xfs_log_item_t *lip,
3031 int gen)
3032{
3033 int orig_gen = gen;
3034
3035 do {
3036 ASSERT(lip->li_type != XFS_LI_EFI);
3037 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
3038 /*
3039 * The check will be bogus if we restart from the
3040 * beginning of the AIL, so ASSERT that we don't.
3041 * We never should since we're holding the AIL lock
3042 * the entire time.
3043 */
3044 ASSERT(gen == orig_gen);
3045 } while (lip != NULL);
3046}
3047#endif /* DEBUG */
3048
3049/*
3050 * When this is called, all of the EFIs which did not have
3051 * corresponding EFDs should be in the AIL. What we do now
3052 * is free the extents associated with each one.
3053 *
3054 * Since we process the EFIs in normal transactions, they
3055 * will be removed at some point after the commit. This prevents
3056 * us from just walking down the list processing each one.
3057 * We'll use a flag in the EFI to skip those that we've already
3058 * processed and use the AIL iteration mechanism's generation
3059 * count to try to speed this up at least a bit.
3060 *
3061 * When we start, we know that the EFIs are the only things in
3062 * the AIL. As we process them, however, other items are added
3063 * to the AIL. Since everything added to the AIL must come after
3064 * everything already in the AIL, we stop processing as soon as
3065 * we see something other than an EFI in the AIL.
3066 */
3067STATIC void
3068xlog_recover_process_efis(
3069 xlog_t *log)
3070{
3071 xfs_log_item_t *lip;
3072 xfs_efi_log_item_t *efip;
3073 int gen;
3074 xfs_mount_t *mp;
3075 SPLDECL(s);
3076
3077 mp = log->l_mp;
3078 AIL_LOCK(mp,s);
3079
3080 lip = xfs_trans_first_ail(mp, &gen);
3081 while (lip != NULL) {
3082 /*
3083 * We're done when we see something other than an EFI.
3084 */
3085 if (lip->li_type != XFS_LI_EFI) {
3086 xlog_recover_check_ail(mp, lip, gen);
3087 break;
3088 }
3089
3090 /*
3091 * Skip EFIs that we've already processed.
3092 */
3093 efip = (xfs_efi_log_item_t *)lip;
3094 if (efip->efi_flags & XFS_EFI_RECOVERED) {
3095 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
3096 continue;
3097 }
3098
3099 AIL_UNLOCK(mp, s);
3100 xlog_recover_process_efi(mp, efip);
3101 AIL_LOCK(mp,s);
3102 lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
3103 }
3104 AIL_UNLOCK(mp, s);
3105}
3106
3107/*
3108 * This routine performs a transaction to null out a bad inode pointer
3109 * in an agi unlinked inode hash bucket.
3110 */
3111STATIC void
3112xlog_recover_clear_agi_bucket(
3113 xfs_mount_t *mp,
3114 xfs_agnumber_t agno,
3115 int bucket)
3116{
3117 xfs_trans_t *tp;
3118 xfs_agi_t *agi;
3119 xfs_buf_t *agibp;
3120 int offset;
3121 int error;
3122
3123 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3124 xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp), 0, 0, 0);
3125
3126 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
3127 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
3128 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
3129 if (error) {
3130 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3131 return;
3132 }
3133
3134 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11003135 if (be32_to_cpu(agi->agi_magicnum) != XFS_AGI_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3137 return;
3138 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139
Christoph Hellwig16259e72005-11-02 15:11:25 +11003140 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003141 offset = offsetof(xfs_agi_t, agi_unlinked) +
3142 (sizeof(xfs_agino_t) * bucket);
3143 xfs_trans_log_buf(tp, agibp, offset,
3144 (offset + sizeof(xfs_agino_t) - 1));
3145
3146 (void) xfs_trans_commit(tp, 0, NULL);
3147}
3148
3149/*
3150 * xlog_iunlink_recover
3151 *
3152 * This is called during recovery to process any inodes which
3153 * we unlinked but not freed when the system crashed. These
3154 * inodes will be on the lists in the AGI blocks. What we do
3155 * here is scan all the AGIs and fully truncate and free any
3156 * inodes found on the lists. Each inode is removed from the
3157 * lists when it has been fully truncated and is freed. The
3158 * freeing of the inode and its removal from the list must be
3159 * atomic.
3160 */
3161void
3162xlog_recover_process_iunlinks(
3163 xlog_t *log)
3164{
3165 xfs_mount_t *mp;
3166 xfs_agnumber_t agno;
3167 xfs_agi_t *agi;
3168 xfs_buf_t *agibp;
3169 xfs_buf_t *ibp;
3170 xfs_dinode_t *dip;
3171 xfs_inode_t *ip;
3172 xfs_agino_t agino;
3173 xfs_ino_t ino;
3174 int bucket;
3175 int error;
3176 uint mp_dmevmask;
3177
3178 mp = log->l_mp;
3179
3180 /*
3181 * Prevent any DMAPI event from being sent while in this function.
3182 */
3183 mp_dmevmask = mp->m_dmevmask;
3184 mp->m_dmevmask = 0;
3185
3186 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3187 /*
3188 * Find the agi for this ag.
3189 */
3190 agibp = xfs_buf_read(mp->m_ddev_targp,
3191 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
3192 XFS_FSS_TO_BB(mp, 1), 0);
3193 if (XFS_BUF_ISERROR(agibp)) {
3194 xfs_ioerror_alert("xlog_recover_process_iunlinks(#1)",
3195 log->l_mp, agibp,
3196 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)));
3197 }
3198 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11003199 ASSERT(XFS_AGI_MAGIC == be32_to_cpu(agi->agi_magicnum));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003200
3201 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3202
Christoph Hellwig16259e72005-11-02 15:11:25 +11003203 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204 while (agino != NULLAGINO) {
3205
3206 /*
3207 * Release the agi buffer so that it can
3208 * be acquired in the normal course of the
3209 * transaction to truncate and free the inode.
3210 */
3211 xfs_buf_relse(agibp);
3212
3213 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3214 error = xfs_iget(mp, NULL, ino, 0, 0, &ip, 0);
3215 ASSERT(error || (ip != NULL));
3216
3217 if (!error) {
3218 /*
3219 * Get the on disk inode to find the
3220 * next inode in the bucket.
3221 */
3222 error = xfs_itobp(mp, NULL, ip, &dip,
Nathan Scottb12dd342006-03-17 17:26:04 +11003223 &ibp, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224 ASSERT(error || (dip != NULL));
3225 }
3226
3227 if (!error) {
3228 ASSERT(ip->i_d.di_nlink == 0);
3229
3230 /* setup for the next pass */
3231 agino = INT_GET(dip->di_next_unlinked,
3232 ARCH_CONVERT);
3233 xfs_buf_relse(ibp);
3234 /*
3235 * Prevent any DMAPI event from
3236 * being sent when the
3237 * reference on the inode is
3238 * dropped.
3239 */
3240 ip->i_d.di_dmevmask = 0;
3241
3242 /*
3243 * If this is a new inode, handle
3244 * it specially. Otherwise,
3245 * just drop our reference to the
3246 * inode. If there are no
3247 * other references, this will
3248 * send the inode to
3249 * xfs_inactive() which will
3250 * truncate the file and free
3251 * the inode.
3252 */
3253 if (ip->i_d.di_mode == 0)
3254 xfs_iput_new(ip, 0);
3255 else
3256 VN_RELE(XFS_ITOV(ip));
3257 } else {
3258 /*
3259 * We can't read in the inode
3260 * this bucket points to, or
3261 * this inode is messed up. Just
3262 * ditch this bucket of inodes. We
3263 * will lose some inodes and space,
3264 * but at least we won't hang. Call
3265 * xlog_recover_clear_agi_bucket()
3266 * to perform a transaction to clear
3267 * the inode pointer in the bucket.
3268 */
3269 xlog_recover_clear_agi_bucket(mp, agno,
3270 bucket);
3271
3272 agino = NULLAGINO;
3273 }
3274
3275 /*
3276 * Reacquire the agibuffer and continue around
3277 * the loop.
3278 */
3279 agibp = xfs_buf_read(mp->m_ddev_targp,
3280 XFS_AG_DADDR(mp, agno,
3281 XFS_AGI_DADDR(mp)),
3282 XFS_FSS_TO_BB(mp, 1), 0);
3283 if (XFS_BUF_ISERROR(agibp)) {
3284 xfs_ioerror_alert(
3285 "xlog_recover_process_iunlinks(#2)",
3286 log->l_mp, agibp,
3287 XFS_AG_DADDR(mp, agno,
3288 XFS_AGI_DADDR(mp)));
3289 }
3290 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11003291 ASSERT(XFS_AGI_MAGIC == be32_to_cpu(
3292 agi->agi_magicnum));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003293 }
3294 }
3295
3296 /*
3297 * Release the buffer for the current agi so we can
3298 * go on to the next one.
3299 */
3300 xfs_buf_relse(agibp);
3301 }
3302
3303 mp->m_dmevmask = mp_dmevmask;
3304}
3305
3306
3307#ifdef DEBUG
3308STATIC void
3309xlog_pack_data_checksum(
3310 xlog_t *log,
3311 xlog_in_core_t *iclog,
3312 int size)
3313{
3314 int i;
3315 uint *up;
3316 uint chksum = 0;
3317
3318 up = (uint *)iclog->ic_datap;
3319 /* divide length by 4 to get # words */
3320 for (i = 0; i < (size >> 2); i++) {
3321 chksum ^= INT_GET(*up, ARCH_CONVERT);
3322 up++;
3323 }
3324 INT_SET(iclog->ic_header.h_chksum, ARCH_CONVERT, chksum);
3325}
3326#else
3327#define xlog_pack_data_checksum(log, iclog, size)
3328#endif
3329
3330/*
3331 * Stamp cycle number in every block
3332 */
3333void
3334xlog_pack_data(
3335 xlog_t *log,
3336 xlog_in_core_t *iclog,
3337 int roundoff)
3338{
3339 int i, j, k;
3340 int size = iclog->ic_offset + roundoff;
3341 uint cycle_lsn;
3342 xfs_caddr_t dp;
3343 xlog_in_core_2_t *xhdr;
3344
3345 xlog_pack_data_checksum(log, iclog, size);
3346
3347 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3348
3349 dp = iclog->ic_datap;
3350 for (i = 0; i < BTOBB(size) &&
3351 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3352 iclog->ic_header.h_cycle_data[i] = *(uint *)dp;
3353 *(uint *)dp = cycle_lsn;
3354 dp += BBSIZE;
3355 }
3356
3357 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3358 xhdr = (xlog_in_core_2_t *)&iclog->ic_header;
3359 for ( ; i < BTOBB(size); i++) {
3360 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3361 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3362 xhdr[j].hic_xheader.xh_cycle_data[k] = *(uint *)dp;
3363 *(uint *)dp = cycle_lsn;
3364 dp += BBSIZE;
3365 }
3366
3367 for (i = 1; i < log->l_iclog_heads; i++) {
3368 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3369 }
3370 }
3371}
3372
3373#if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
3374STATIC void
3375xlog_unpack_data_checksum(
3376 xlog_rec_header_t *rhead,
3377 xfs_caddr_t dp,
3378 xlog_t *log)
3379{
3380 uint *up = (uint *)dp;
3381 uint chksum = 0;
3382 int i;
3383
3384 /* divide length by 4 to get # words */
3385 for (i=0; i < INT_GET(rhead->h_len, ARCH_CONVERT) >> 2; i++) {
3386 chksum ^= INT_GET(*up, ARCH_CONVERT);
3387 up++;
3388 }
3389 if (chksum != INT_GET(rhead->h_chksum, ARCH_CONVERT)) {
3390 if (rhead->h_chksum ||
3391 ((log->l_flags & XLOG_CHKSUM_MISMATCH) == 0)) {
3392 cmn_err(CE_DEBUG,
Nathan Scottb6574522006-06-09 15:29:40 +10003393 "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003394 INT_GET(rhead->h_chksum, ARCH_CONVERT), chksum);
3395 cmn_err(CE_DEBUG,
3396"XFS: Disregard message if filesystem was created with non-DEBUG kernel");
3397 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3398 cmn_err(CE_DEBUG,
Nathan Scottb6574522006-06-09 15:29:40 +10003399 "XFS: LogR this is a LogV2 filesystem\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400 }
3401 log->l_flags |= XLOG_CHKSUM_MISMATCH;
3402 }
3403 }
3404}
3405#else
3406#define xlog_unpack_data_checksum(rhead, dp, log)
3407#endif
3408
3409STATIC void
3410xlog_unpack_data(
3411 xlog_rec_header_t *rhead,
3412 xfs_caddr_t dp,
3413 xlog_t *log)
3414{
3415 int i, j, k;
3416 xlog_in_core_2_t *xhdr;
3417
3418 for (i = 0; i < BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT)) &&
3419 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3420 *(uint *)dp = *(uint *)&rhead->h_cycle_data[i];
3421 dp += BBSIZE;
3422 }
3423
3424 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3425 xhdr = (xlog_in_core_2_t *)rhead;
3426 for ( ; i < BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT)); i++) {
3427 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3428 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3429 *(uint *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
3430 dp += BBSIZE;
3431 }
3432 }
3433
3434 xlog_unpack_data_checksum(rhead, dp, log);
3435}
3436
3437STATIC int
3438xlog_valid_rec_header(
3439 xlog_t *log,
3440 xlog_rec_header_t *rhead,
3441 xfs_daddr_t blkno)
3442{
3443 int hlen;
3444
3445 if (unlikely(
3446 (INT_GET(rhead->h_magicno, ARCH_CONVERT) !=
3447 XLOG_HEADER_MAGIC_NUM))) {
3448 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3449 XFS_ERRLEVEL_LOW, log->l_mp);
3450 return XFS_ERROR(EFSCORRUPTED);
3451 }
3452 if (unlikely(
3453 (!rhead->h_version ||
3454 (INT_GET(rhead->h_version, ARCH_CONVERT) &
3455 (~XLOG_VERSION_OKBITS)) != 0))) {
3456 xlog_warn("XFS: %s: unrecognised log version (%d).",
3457 __FUNCTION__, INT_GET(rhead->h_version, ARCH_CONVERT));
3458 return XFS_ERROR(EIO);
3459 }
3460
3461 /* LR body must have data or it wouldn't have been written */
3462 hlen = INT_GET(rhead->h_len, ARCH_CONVERT);
3463 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3464 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3465 XFS_ERRLEVEL_LOW, log->l_mp);
3466 return XFS_ERROR(EFSCORRUPTED);
3467 }
3468 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3469 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3470 XFS_ERRLEVEL_LOW, log->l_mp);
3471 return XFS_ERROR(EFSCORRUPTED);
3472 }
3473 return 0;
3474}
3475
3476/*
3477 * Read the log from tail to head and process the log records found.
3478 * Handle the two cases where the tail and head are in the same cycle
3479 * and where the active portion of the log wraps around the end of
3480 * the physical log separately. The pass parameter is passed through
3481 * to the routines called to process the data and is not looked at
3482 * here.
3483 */
3484STATIC int
3485xlog_do_recovery_pass(
3486 xlog_t *log,
3487 xfs_daddr_t head_blk,
3488 xfs_daddr_t tail_blk,
3489 int pass)
3490{
3491 xlog_rec_header_t *rhead;
3492 xfs_daddr_t blk_no;
3493 xfs_caddr_t bufaddr, offset;
3494 xfs_buf_t *hbp, *dbp;
3495 int error = 0, h_size;
3496 int bblks, split_bblks;
3497 int hblks, split_hblks, wrapped_hblks;
3498 xlog_recover_t *rhash[XLOG_RHASH_SIZE];
3499
3500 ASSERT(head_blk != tail_blk);
3501
3502 /*
3503 * Read the header of the tail block and get the iclog buffer size from
3504 * h_size. Use this to tell how many sectors make up the log header.
3505 */
3506 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
3507 /*
3508 * When using variable length iclogs, read first sector of
3509 * iclog header and extract the header size from it. Get a
3510 * new hbp that is the correct size.
3511 */
3512 hbp = xlog_get_bp(log, 1);
3513 if (!hbp)
3514 return ENOMEM;
3515 if ((error = xlog_bread(log, tail_blk, 1, hbp)))
3516 goto bread_err1;
3517 offset = xlog_align(log, tail_blk, 1, hbp);
3518 rhead = (xlog_rec_header_t *)offset;
3519 error = xlog_valid_rec_header(log, rhead, tail_blk);
3520 if (error)
3521 goto bread_err1;
3522 h_size = INT_GET(rhead->h_size, ARCH_CONVERT);
3523 if ((INT_GET(rhead->h_version, ARCH_CONVERT)
3524 & XLOG_VERSION_2) &&
3525 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3526 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3527 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3528 hblks++;
3529 xlog_put_bp(hbp);
3530 hbp = xlog_get_bp(log, hblks);
3531 } else {
3532 hblks = 1;
3533 }
3534 } else {
3535 ASSERT(log->l_sectbb_log == 0);
3536 hblks = 1;
3537 hbp = xlog_get_bp(log, 1);
3538 h_size = XLOG_BIG_RECORD_BSIZE;
3539 }
3540
3541 if (!hbp)
3542 return ENOMEM;
3543 dbp = xlog_get_bp(log, BTOBB(h_size));
3544 if (!dbp) {
3545 xlog_put_bp(hbp);
3546 return ENOMEM;
3547 }
3548
3549 memset(rhash, 0, sizeof(rhash));
3550 if (tail_blk <= head_blk) {
3551 for (blk_no = tail_blk; blk_no < head_blk; ) {
3552 if ((error = xlog_bread(log, blk_no, hblks, hbp)))
3553 goto bread_err2;
3554 offset = xlog_align(log, blk_no, hblks, hbp);
3555 rhead = (xlog_rec_header_t *)offset;
3556 error = xlog_valid_rec_header(log, rhead, blk_no);
3557 if (error)
3558 goto bread_err2;
3559
3560 /* blocks in data section */
3561 bblks = (int)BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT));
3562 error = xlog_bread(log, blk_no + hblks, bblks, dbp);
3563 if (error)
3564 goto bread_err2;
3565 offset = xlog_align(log, blk_no + hblks, bblks, dbp);
3566 xlog_unpack_data(rhead, offset, log);
3567 if ((error = xlog_recover_process_data(log,
3568 rhash, rhead, offset, pass)))
3569 goto bread_err2;
3570 blk_no += bblks + hblks;
3571 }
3572 } else {
3573 /*
3574 * Perform recovery around the end of the physical log.
3575 * When the head is not on the same cycle number as the tail,
3576 * we can't do a sequential recovery as above.
3577 */
3578 blk_no = tail_blk;
3579 while (blk_no < log->l_logBBsize) {
3580 /*
3581 * Check for header wrapping around physical end-of-log
3582 */
3583 offset = NULL;
3584 split_hblks = 0;
3585 wrapped_hblks = 0;
3586 if (blk_no + hblks <= log->l_logBBsize) {
3587 /* Read header in one read */
3588 error = xlog_bread(log, blk_no, hblks, hbp);
3589 if (error)
3590 goto bread_err2;
3591 offset = xlog_align(log, blk_no, hblks, hbp);
3592 } else {
3593 /* This LR is split across physical log end */
3594 if (blk_no != log->l_logBBsize) {
3595 /* some data before physical log end */
3596 ASSERT(blk_no <= INT_MAX);
3597 split_hblks = log->l_logBBsize - (int)blk_no;
3598 ASSERT(split_hblks > 0);
3599 if ((error = xlog_bread(log, blk_no,
3600 split_hblks, hbp)))
3601 goto bread_err2;
3602 offset = xlog_align(log, blk_no,
3603 split_hblks, hbp);
3604 }
3605 /*
3606 * Note: this black magic still works with
3607 * large sector sizes (non-512) only because:
3608 * - we increased the buffer size originally
3609 * by 1 sector giving us enough extra space
3610 * for the second read;
3611 * - the log start is guaranteed to be sector
3612 * aligned;
3613 * - we read the log end (LR header start)
3614 * _first_, then the log start (LR header end)
3615 * - order is important.
3616 */
3617 bufaddr = XFS_BUF_PTR(hbp);
3618 XFS_BUF_SET_PTR(hbp,
3619 bufaddr + BBTOB(split_hblks),
3620 BBTOB(hblks - split_hblks));
3621 wrapped_hblks = hblks - split_hblks;
3622 error = xlog_bread(log, 0, wrapped_hblks, hbp);
3623 if (error)
3624 goto bread_err2;
3625 XFS_BUF_SET_PTR(hbp, bufaddr, BBTOB(hblks));
3626 if (!offset)
3627 offset = xlog_align(log, 0,
3628 wrapped_hblks, hbp);
3629 }
3630 rhead = (xlog_rec_header_t *)offset;
3631 error = xlog_valid_rec_header(log, rhead,
3632 split_hblks ? blk_no : 0);
3633 if (error)
3634 goto bread_err2;
3635
3636 bblks = (int)BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT));
3637 blk_no += hblks;
3638
3639 /* Read in data for log record */
3640 if (blk_no + bblks <= log->l_logBBsize) {
3641 error = xlog_bread(log, blk_no, bblks, dbp);
3642 if (error)
3643 goto bread_err2;
3644 offset = xlog_align(log, blk_no, bblks, dbp);
3645 } else {
3646 /* This log record is split across the
3647 * physical end of log */
3648 offset = NULL;
3649 split_bblks = 0;
3650 if (blk_no != log->l_logBBsize) {
3651 /* some data is before the physical
3652 * end of log */
3653 ASSERT(!wrapped_hblks);
3654 ASSERT(blk_no <= INT_MAX);
3655 split_bblks =
3656 log->l_logBBsize - (int)blk_no;
3657 ASSERT(split_bblks > 0);
3658 if ((error = xlog_bread(log, blk_no,
3659 split_bblks, dbp)))
3660 goto bread_err2;
3661 offset = xlog_align(log, blk_no,
3662 split_bblks, dbp);
3663 }
3664 /*
3665 * Note: this black magic still works with
3666 * large sector sizes (non-512) only because:
3667 * - we increased the buffer size originally
3668 * by 1 sector giving us enough extra space
3669 * for the second read;
3670 * - the log start is guaranteed to be sector
3671 * aligned;
3672 * - we read the log end (LR header start)
3673 * _first_, then the log start (LR header end)
3674 * - order is important.
3675 */
3676 bufaddr = XFS_BUF_PTR(dbp);
3677 XFS_BUF_SET_PTR(dbp,
3678 bufaddr + BBTOB(split_bblks),
3679 BBTOB(bblks - split_bblks));
3680 if ((error = xlog_bread(log, wrapped_hblks,
3681 bblks - split_bblks, dbp)))
3682 goto bread_err2;
3683 XFS_BUF_SET_PTR(dbp, bufaddr, h_size);
3684 if (!offset)
3685 offset = xlog_align(log, wrapped_hblks,
3686 bblks - split_bblks, dbp);
3687 }
3688 xlog_unpack_data(rhead, offset, log);
3689 if ((error = xlog_recover_process_data(log, rhash,
3690 rhead, offset, pass)))
3691 goto bread_err2;
3692 blk_no += bblks;
3693 }
3694
3695 ASSERT(blk_no >= log->l_logBBsize);
3696 blk_no -= log->l_logBBsize;
3697
3698 /* read first part of physical log */
3699 while (blk_no < head_blk) {
3700 if ((error = xlog_bread(log, blk_no, hblks, hbp)))
3701 goto bread_err2;
3702 offset = xlog_align(log, blk_no, hblks, hbp);
3703 rhead = (xlog_rec_header_t *)offset;
3704 error = xlog_valid_rec_header(log, rhead, blk_no);
3705 if (error)
3706 goto bread_err2;
3707 bblks = (int)BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT));
3708 if ((error = xlog_bread(log, blk_no+hblks, bblks, dbp)))
3709 goto bread_err2;
3710 offset = xlog_align(log, blk_no+hblks, bblks, dbp);
3711 xlog_unpack_data(rhead, offset, log);
3712 if ((error = xlog_recover_process_data(log, rhash,
3713 rhead, offset, pass)))
3714 goto bread_err2;
3715 blk_no += bblks + hblks;
3716 }
3717 }
3718
3719 bread_err2:
3720 xlog_put_bp(dbp);
3721 bread_err1:
3722 xlog_put_bp(hbp);
3723 return error;
3724}
3725
3726/*
3727 * Do the recovery of the log. We actually do this in two phases.
3728 * The two passes are necessary in order to implement the function
3729 * of cancelling a record written into the log. The first pass
3730 * determines those things which have been cancelled, and the
3731 * second pass replays log items normally except for those which
3732 * have been cancelled. The handling of the replay and cancellations
3733 * takes place in the log item type specific routines.
3734 *
3735 * The table of items which have cancel records in the log is allocated
3736 * and freed at this level, since only here do we know when all of
3737 * the log recovery has been completed.
3738 */
3739STATIC int
3740xlog_do_log_recovery(
3741 xlog_t *log,
3742 xfs_daddr_t head_blk,
3743 xfs_daddr_t tail_blk)
3744{
3745 int error;
3746
3747 ASSERT(head_blk != tail_blk);
3748
3749 /*
3750 * First do a pass to find all of the cancelled buf log items.
3751 * Store them in the buf_cancel_table for use in the second pass.
3752 */
3753 log->l_buf_cancel_table =
3754 (xfs_buf_cancel_t **)kmem_zalloc(XLOG_BC_TABLE_SIZE *
3755 sizeof(xfs_buf_cancel_t*),
3756 KM_SLEEP);
3757 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3758 XLOG_RECOVER_PASS1);
3759 if (error != 0) {
3760 kmem_free(log->l_buf_cancel_table,
3761 XLOG_BC_TABLE_SIZE * sizeof(xfs_buf_cancel_t*));
3762 log->l_buf_cancel_table = NULL;
3763 return error;
3764 }
3765 /*
3766 * Then do a second pass to actually recover the items in the log.
3767 * When it is complete free the table of buf cancel items.
3768 */
3769 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3770 XLOG_RECOVER_PASS2);
3771#ifdef DEBUG
Tim Shimmin6d192a92006-06-09 14:55:38 +10003772 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003773 int i;
3774
3775 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3776 ASSERT(log->l_buf_cancel_table[i] == NULL);
3777 }
3778#endif /* DEBUG */
3779
3780 kmem_free(log->l_buf_cancel_table,
3781 XLOG_BC_TABLE_SIZE * sizeof(xfs_buf_cancel_t*));
3782 log->l_buf_cancel_table = NULL;
3783
3784 return error;
3785}
3786
3787/*
3788 * Do the actual recovery
3789 */
3790STATIC int
3791xlog_do_recover(
3792 xlog_t *log,
3793 xfs_daddr_t head_blk,
3794 xfs_daddr_t tail_blk)
3795{
3796 int error;
3797 xfs_buf_t *bp;
3798 xfs_sb_t *sbp;
3799
3800 /*
3801 * First replay the images in the log.
3802 */
3803 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3804 if (error) {
3805 return error;
3806 }
3807
3808 XFS_bflush(log->l_mp->m_ddev_targp);
3809
3810 /*
3811 * If IO errors happened during recovery, bail out.
3812 */
3813 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3814 return (EIO);
3815 }
3816
3817 /*
3818 * We now update the tail_lsn since much of the recovery has completed
3819 * and there may be space available to use. If there were no extent
3820 * or iunlinks, we can free up the entire log and set the tail_lsn to
3821 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3822 * lsn of the last known good LR on disk. If there are extent frees
3823 * or iunlinks they will have some entries in the AIL; so we look at
3824 * the AIL to determine how to set the tail_lsn.
3825 */
3826 xlog_assign_tail_lsn(log->l_mp);
3827
3828 /*
3829 * Now that we've finished replaying all buffer and inode
3830 * updates, re-read in the superblock.
3831 */
3832 bp = xfs_getsb(log->l_mp, 0);
3833 XFS_BUF_UNDONE(bp);
3834 XFS_BUF_READ(bp);
3835 xfsbdstrat(log->l_mp, bp);
3836 if ((error = xfs_iowait(bp))) {
3837 xfs_ioerror_alert("xlog_do_recover",
3838 log->l_mp, bp, XFS_BUF_ADDR(bp));
3839 ASSERT(0);
3840 xfs_buf_relse(bp);
3841 return error;
3842 }
3843
3844 /* Convert superblock from on-disk format */
3845 sbp = &log->l_mp->m_sb;
3846 xfs_xlatesb(XFS_BUF_TO_SBP(bp), sbp, 1, XFS_SB_ALL_BITS);
3847 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
3848 ASSERT(XFS_SB_GOOD_VERSION(sbp));
3849 xfs_buf_relse(bp);
3850
3851 xlog_recover_check_summary(log);
3852
3853 /* Normal transactions can now occur */
3854 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3855 return 0;
3856}
3857
3858/*
3859 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3860 *
3861 * Return error or zero.
3862 */
3863int
3864xlog_recover(
Eric Sandeen65be6052006-01-11 15:34:19 +11003865 xlog_t *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003866{
3867 xfs_daddr_t head_blk, tail_blk;
3868 int error;
3869
3870 /* find the tail of the log */
Eric Sandeen65be6052006-01-11 15:34:19 +11003871 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003872 return error;
3873
3874 if (tail_blk != head_blk) {
3875 /* There used to be a comment here:
3876 *
3877 * disallow recovery on read-only mounts. note -- mount
3878 * checks for ENOSPC and turns it into an intelligent
3879 * error message.
3880 * ...but this is no longer true. Now, unless you specify
3881 * NORECOVERY (in which case this function would never be
3882 * called), we just go ahead and recover. We do this all
3883 * under the vfs layer, so we can get away with it unless
3884 * the device itself is read-only, in which case we fail.
3885 */
3886 if ((error = xfs_dev_is_read_only(log->l_mp,
3887 "recovery required"))) {
3888 return error;
3889 }
3890
3891 cmn_err(CE_NOTE,
Nathan Scottfc1f8c12005-11-02 11:44:33 +11003892 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3893 log->l_mp->m_fsname, log->l_mp->m_logname ?
3894 log->l_mp->m_logname : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003895
3896 error = xlog_do_recover(log, head_blk, tail_blk);
3897 log->l_flags |= XLOG_RECOVERY_NEEDED;
3898 }
3899 return error;
3900}
3901
3902/*
3903 * In the first part of recovery we replay inodes and buffers and build
3904 * up the list of extent free items which need to be processed. Here
3905 * we process the extent free items and clean up the on disk unlinked
3906 * inode lists. This is separated from the first part of recovery so
3907 * that the root and real-time bitmap inodes can be read in from disk in
3908 * between the two stages. This is necessary so that we can free space
3909 * in the real-time portion of the file system.
3910 */
3911int
3912xlog_recover_finish(
3913 xlog_t *log,
3914 int mfsi_flags)
3915{
3916 /*
3917 * Now we're ready to do the transactions needed for the
3918 * rest of recovery. Start with completing all the extent
3919 * free intent records and then process the unlinked inode
3920 * lists. At this point, we essentially run in normal mode
3921 * except that we're still performing recovery actions
3922 * rather than accepting new requests.
3923 */
3924 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
3925 xlog_recover_process_efis(log);
3926 /*
3927 * Sync the log to get all the EFIs out of the AIL.
3928 * This isn't absolutely necessary, but it helps in
3929 * case the unlink transactions would have problems
3930 * pushing the EFIs out of the way.
3931 */
3932 xfs_log_force(log->l_mp, (xfs_lsn_t)0,
3933 (XFS_LOG_FORCE | XFS_LOG_SYNC));
3934
3935 if ( (mfsi_flags & XFS_MFSI_NOUNLINK) == 0 ) {
3936 xlog_recover_process_iunlinks(log);
3937 }
3938
3939 xlog_recover_check_summary(log);
3940
3941 cmn_err(CE_NOTE,
Nathan Scottfc1f8c12005-11-02 11:44:33 +11003942 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3943 log->l_mp->m_fsname, log->l_mp->m_logname ?
3944 log->l_mp->m_logname : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003945 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
3946 } else {
3947 cmn_err(CE_DEBUG,
Nathan Scottb6574522006-06-09 15:29:40 +10003948 "!Ending clean XFS mount for filesystem: %s\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003949 log->l_mp->m_fsname);
3950 }
3951 return 0;
3952}
3953
3954
3955#if defined(DEBUG)
3956/*
3957 * Read all of the agf and agi counters and check that they
3958 * are consistent with the superblock counters.
3959 */
3960void
3961xlog_recover_check_summary(
3962 xlog_t *log)
3963{
3964 xfs_mount_t *mp;
3965 xfs_agf_t *agfp;
3966 xfs_agi_t *agip;
3967 xfs_buf_t *agfbp;
3968 xfs_buf_t *agibp;
3969 xfs_daddr_t agfdaddr;
3970 xfs_daddr_t agidaddr;
3971 xfs_buf_t *sbbp;
3972#ifdef XFS_LOUD_RECOVERY
3973 xfs_sb_t *sbp;
3974#endif
3975 xfs_agnumber_t agno;
3976 __uint64_t freeblks;
3977 __uint64_t itotal;
3978 __uint64_t ifree;
3979
3980 mp = log->l_mp;
3981
3982 freeblks = 0LL;
3983 itotal = 0LL;
3984 ifree = 0LL;
3985 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3986 agfdaddr = XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp));
3987 agfbp = xfs_buf_read(mp->m_ddev_targp, agfdaddr,
3988 XFS_FSS_TO_BB(mp, 1), 0);
3989 if (XFS_BUF_ISERROR(agfbp)) {
3990 xfs_ioerror_alert("xlog_recover_check_summary(agf)",
3991 mp, agfbp, agfdaddr);
3992 }
3993 agfp = XFS_BUF_TO_AGF(agfbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11003994 ASSERT(XFS_AGF_MAGIC == be32_to_cpu(agfp->agf_magicnum));
3995 ASSERT(XFS_AGF_GOOD_VERSION(be32_to_cpu(agfp->agf_versionnum)));
3996 ASSERT(be32_to_cpu(agfp->agf_seqno) == agno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003997
Christoph Hellwig16259e72005-11-02 15:11:25 +11003998 freeblks += be32_to_cpu(agfp->agf_freeblks) +
3999 be32_to_cpu(agfp->agf_flcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004000 xfs_buf_relse(agfbp);
4001
4002 agidaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
4003 agibp = xfs_buf_read(mp->m_ddev_targp, agidaddr,
4004 XFS_FSS_TO_BB(mp, 1), 0);
4005 if (XFS_BUF_ISERROR(agibp)) {
4006 xfs_ioerror_alert("xlog_recover_check_summary(agi)",
4007 mp, agibp, agidaddr);
4008 }
4009 agip = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11004010 ASSERT(XFS_AGI_MAGIC == be32_to_cpu(agip->agi_magicnum));
4011 ASSERT(XFS_AGI_GOOD_VERSION(be32_to_cpu(agip->agi_versionnum)));
4012 ASSERT(be32_to_cpu(agip->agi_seqno) == agno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004013
Christoph Hellwig16259e72005-11-02 15:11:25 +11004014 itotal += be32_to_cpu(agip->agi_count);
4015 ifree += be32_to_cpu(agip->agi_freecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004016 xfs_buf_relse(agibp);
4017 }
4018
4019 sbbp = xfs_getsb(mp, 0);
4020#ifdef XFS_LOUD_RECOVERY
4021 sbp = &mp->m_sb;
4022 xfs_xlatesb(XFS_BUF_TO_SBP(sbbp), sbp, 1, XFS_SB_ALL_BITS);
4023 cmn_err(CE_NOTE,
4024 "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
4025 sbp->sb_icount, itotal);
4026 cmn_err(CE_NOTE,
4027 "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
4028 sbp->sb_ifree, ifree);
4029 cmn_err(CE_NOTE,
4030 "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
4031 sbp->sb_fdblocks, freeblks);
4032#if 0
4033 /*
4034 * This is turned off until I account for the allocation
4035 * btree blocks which live in free space.
4036 */
4037 ASSERT(sbp->sb_icount == itotal);
4038 ASSERT(sbp->sb_ifree == ifree);
4039 ASSERT(sbp->sb_fdblocks == freeblks);
4040#endif
4041#endif
4042 xfs_buf_relse(sbbp);
4043}
4044#endif /* DEBUG */