]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - fs/xfs/xfs_log_recover.c
Fix common misspellings
[mirror_ubuntu-zesty-kernel.git] / fs / xfs / xfs_log_recover.c
CommitLineData
1da177e4 1/*
87c199c2 2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
7b718769 3 * All Rights Reserved.
1da177e4 4 *
7b718769
NS
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
1da177e4
LT
7 * published by the Free Software Foundation.
8 *
7b718769
NS
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
1da177e4 13 *
7b718769
NS
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
1da177e4 17 */
1da177e4 18#include "xfs.h"
a844f451 19#include "xfs_fs.h"
1da177e4 20#include "xfs_types.h"
a844f451 21#include "xfs_bit.h"
1da177e4 22#include "xfs_log.h"
a844f451 23#include "xfs_inum.h"
1da177e4 24#include "xfs_trans.h"
a844f451
NS
25#include "xfs_sb.h"
26#include "xfs_ag.h"
1da177e4
LT
27#include "xfs_mount.h"
28#include "xfs_error.h"
29#include "xfs_bmap_btree.h"
a844f451
NS
30#include "xfs_alloc_btree.h"
31#include "xfs_ialloc_btree.h"
1da177e4 32#include "xfs_dinode.h"
1da177e4 33#include "xfs_inode.h"
a844f451 34#include "xfs_inode_item.h"
a844f451 35#include "xfs_alloc.h"
1da177e4
LT
36#include "xfs_ialloc.h"
37#include "xfs_log_priv.h"
38#include "xfs_buf_item.h"
1da177e4
LT
39#include "xfs_log_recover.h"
40#include "xfs_extfree_item.h"
41#include "xfs_trans_priv.h"
1da177e4
LT
42#include "xfs_quota.h"
43#include "xfs_rw.h"
43355099 44#include "xfs_utils.h"
0b1b213f 45#include "xfs_trace.h"
1da177e4
LT
46
47STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *);
48STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t);
1da177e4
LT
49#if defined(DEBUG)
50STATIC void xlog_recover_check_summary(xlog_t *);
1da177e4
LT
51#else
52#define xlog_recover_check_summary(log)
1da177e4
LT
53#endif
54
d5689eaa
CH
55/*
56 * This structure is used during recovery to record the buf log items which
57 * have been canceled and should not be replayed.
58 */
59struct xfs_buf_cancel {
60 xfs_daddr_t bc_blkno;
61 uint bc_len;
62 int bc_refcount;
63 struct list_head bc_list;
64};
65
1da177e4
LT
66/*
67 * Sector aligned buffer routines for buffer create/read/write/access
68 */
69
ff30a622
AE
70/*
71 * Verify the given count of basic blocks is valid number of blocks
72 * to specify for an operation involving the given XFS log buffer.
73 * Returns nonzero if the count is valid, 0 otherwise.
74 */
75
76static inline int
77xlog_buf_bbcount_valid(
78 xlog_t *log,
79 int bbcount)
80{
81 return bbcount > 0 && bbcount <= log->l_logBBsize;
82}
83
36adecff
AE
84/*
85 * Allocate a buffer to hold log data. The buffer needs to be able
86 * to map to a range of nbblks basic blocks at any valid (basic
87 * block) offset within the log.
88 */
5d77c0dc 89STATIC xfs_buf_t *
1da177e4
LT
90xlog_get_bp(
91 xlog_t *log,
3228149c 92 int nbblks)
1da177e4 93{
ff30a622 94 if (!xlog_buf_bbcount_valid(log, nbblks)) {
a0fa2b67 95 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
ff30a622
AE
96 nbblks);
97 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
3228149c
DC
98 return NULL;
99 }
1da177e4 100
36adecff
AE
101 /*
102 * We do log I/O in units of log sectors (a power-of-2
103 * multiple of the basic block size), so we round up the
25985edc 104 * requested size to accommodate the basic blocks required
36adecff
AE
105 * for complete log sectors.
106 *
107 * In addition, the buffer may be used for a non-sector-
108 * aligned block offset, in which case an I/O of the
109 * requested size could extend beyond the end of the
110 * buffer. If the requested size is only 1 basic block it
111 * will never straddle a sector boundary, so this won't be
112 * an issue. Nor will this be a problem if the log I/O is
113 * done in basic blocks (sector size 1). But otherwise we
114 * extend the buffer by one extra log sector to ensure
25985edc 115 * there's space to accommodate this possibility.
36adecff 116 */
69ce58f0
AE
117 if (nbblks > 1 && log->l_sectBBsize > 1)
118 nbblks += log->l_sectBBsize;
119 nbblks = round_up(nbblks, log->l_sectBBsize);
36adecff 120
686865f7
DC
121 return xfs_buf_get_uncached(log->l_mp->m_logdev_targp,
122 BBTOB(nbblks), 0);
1da177e4
LT
123}
124
5d77c0dc 125STATIC void
1da177e4
LT
126xlog_put_bp(
127 xfs_buf_t *bp)
128{
129 xfs_buf_free(bp);
130}
131
48389ef1
AE
132/*
133 * Return the address of the start of the given block number's data
134 * in a log buffer. The buffer covers a log sector-aligned region.
135 */
076e6acb
CH
136STATIC xfs_caddr_t
137xlog_align(
138 xlog_t *log,
139 xfs_daddr_t blk_no,
140 int nbblks,
141 xfs_buf_t *bp)
142{
fdc07f44 143 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
076e6acb 144
fdc07f44
CH
145 ASSERT(BBTOB(offset + nbblks) <= XFS_BUF_SIZE(bp));
146 return XFS_BUF_PTR(bp) + BBTOB(offset);
076e6acb
CH
147}
148
1da177e4
LT
149
150/*
151 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
152 */
076e6acb
CH
153STATIC int
154xlog_bread_noalign(
1da177e4
LT
155 xlog_t *log,
156 xfs_daddr_t blk_no,
157 int nbblks,
158 xfs_buf_t *bp)
159{
160 int error;
161
ff30a622 162 if (!xlog_buf_bbcount_valid(log, nbblks)) {
a0fa2b67 163 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
ff30a622
AE
164 nbblks);
165 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
3228149c
DC
166 return EFSCORRUPTED;
167 }
168
69ce58f0
AE
169 blk_no = round_down(blk_no, log->l_sectBBsize);
170 nbblks = round_up(nbblks, log->l_sectBBsize);
1da177e4
LT
171
172 ASSERT(nbblks > 0);
173 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
1da177e4
LT
174
175 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
176 XFS_BUF_READ(bp);
177 XFS_BUF_BUSY(bp);
178 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
179 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
180
181 xfsbdstrat(log->l_mp, bp);
1a1a3e97 182 error = xfs_buf_iowait(bp);
d64e31a2 183 if (error)
1da177e4
LT
184 xfs_ioerror_alert("xlog_bread", log->l_mp,
185 bp, XFS_BUF_ADDR(bp));
186 return error;
187}
188
076e6acb
CH
189STATIC int
190xlog_bread(
191 xlog_t *log,
192 xfs_daddr_t blk_no,
193 int nbblks,
194 xfs_buf_t *bp,
195 xfs_caddr_t *offset)
196{
197 int error;
198
199 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
200 if (error)
201 return error;
202
203 *offset = xlog_align(log, blk_no, nbblks, bp);
204 return 0;
205}
206
1da177e4
LT
207/*
208 * Write out the buffer at the given block for the given number of blocks.
209 * The buffer is kept locked across the write and is returned locked.
210 * This can only be used for synchronous log writes.
211 */
ba0f32d4 212STATIC int
1da177e4
LT
213xlog_bwrite(
214 xlog_t *log,
215 xfs_daddr_t blk_no,
216 int nbblks,
217 xfs_buf_t *bp)
218{
219 int error;
220
ff30a622 221 if (!xlog_buf_bbcount_valid(log, nbblks)) {
a0fa2b67 222 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
ff30a622
AE
223 nbblks);
224 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
3228149c
DC
225 return EFSCORRUPTED;
226 }
227
69ce58f0
AE
228 blk_no = round_down(blk_no, log->l_sectBBsize);
229 nbblks = round_up(nbblks, log->l_sectBBsize);
1da177e4
LT
230
231 ASSERT(nbblks > 0);
232 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
233
234 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
235 XFS_BUF_ZEROFLAGS(bp);
236 XFS_BUF_BUSY(bp);
237 XFS_BUF_HOLD(bp);
238 XFS_BUF_PSEMA(bp, PRIBIO);
239 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
240 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
241
242 if ((error = xfs_bwrite(log->l_mp, bp)))
243 xfs_ioerror_alert("xlog_bwrite", log->l_mp,
244 bp, XFS_BUF_ADDR(bp));
245 return error;
246}
247
1da177e4
LT
248#ifdef DEBUG
249/*
250 * dump debug superblock and log record information
251 */
252STATIC void
253xlog_header_check_dump(
254 xfs_mount_t *mp,
255 xlog_rec_header_t *head)
256{
a0fa2b67 257 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d\n",
03daa57c 258 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
a0fa2b67 259 xfs_debug(mp, " log : uuid = %pU, fmt = %d\n",
03daa57c 260 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
1da177e4
LT
261}
262#else
263#define xlog_header_check_dump(mp, head)
264#endif
265
266/*
267 * check log record header for recovery
268 */
269STATIC int
270xlog_header_check_recover(
271 xfs_mount_t *mp,
272 xlog_rec_header_t *head)
273{
b53e675d 274 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
1da177e4
LT
275
276 /*
277 * IRIX doesn't write the h_fmt field and leaves it zeroed
278 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
279 * a dirty log created in IRIX.
280 */
b53e675d 281 if (unlikely(be32_to_cpu(head->h_fmt) != XLOG_FMT)) {
a0fa2b67
DC
282 xfs_warn(mp,
283 "dirty log written in incompatible format - can't recover");
1da177e4
LT
284 xlog_header_check_dump(mp, head);
285 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
286 XFS_ERRLEVEL_HIGH, mp);
287 return XFS_ERROR(EFSCORRUPTED);
288 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
a0fa2b67
DC
289 xfs_warn(mp,
290 "dirty log entry has mismatched uuid - can't recover");
1da177e4
LT
291 xlog_header_check_dump(mp, head);
292 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
293 XFS_ERRLEVEL_HIGH, mp);
294 return XFS_ERROR(EFSCORRUPTED);
295 }
296 return 0;
297}
298
299/*
300 * read the head block of the log and check the header
301 */
302STATIC int
303xlog_header_check_mount(
304 xfs_mount_t *mp,
305 xlog_rec_header_t *head)
306{
b53e675d 307 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
1da177e4
LT
308
309 if (uuid_is_nil(&head->h_fs_uuid)) {
310 /*
311 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
312 * h_fs_uuid is nil, we assume this log was last mounted
313 * by IRIX and continue.
314 */
a0fa2b67 315 xfs_warn(mp, "nil uuid in log - IRIX style log");
1da177e4 316 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
a0fa2b67 317 xfs_warn(mp, "log has mismatched uuid - can't recover");
1da177e4
LT
318 xlog_header_check_dump(mp, head);
319 XFS_ERROR_REPORT("xlog_header_check_mount",
320 XFS_ERRLEVEL_HIGH, mp);
321 return XFS_ERROR(EFSCORRUPTED);
322 }
323 return 0;
324}
325
326STATIC void
327xlog_recover_iodone(
328 struct xfs_buf *bp)
329{
1da177e4
LT
330 if (XFS_BUF_GETERROR(bp)) {
331 /*
332 * We're not going to bother about retrying
333 * this during recovery. One strike!
334 */
1da177e4 335 xfs_ioerror_alert("xlog_recover_iodone",
ebad861b
DC
336 bp->b_target->bt_mount, bp,
337 XFS_BUF_ADDR(bp));
338 xfs_force_shutdown(bp->b_target->bt_mount,
339 SHUTDOWN_META_IO_ERROR);
1da177e4 340 }
1da177e4 341 XFS_BUF_CLR_IODONE_FUNC(bp);
1a1a3e97 342 xfs_buf_ioend(bp, 0);
1da177e4
LT
343}
344
345/*
346 * This routine finds (to an approximation) the first block in the physical
347 * log which contains the given cycle. It uses a binary search algorithm.
348 * Note that the algorithm can not be perfect because the disk will not
349 * necessarily be perfect.
350 */
a8272ce0 351STATIC int
1da177e4
LT
352xlog_find_cycle_start(
353 xlog_t *log,
354 xfs_buf_t *bp,
355 xfs_daddr_t first_blk,
356 xfs_daddr_t *last_blk,
357 uint cycle)
358{
359 xfs_caddr_t offset;
360 xfs_daddr_t mid_blk;
e3bb2e30 361 xfs_daddr_t end_blk;
1da177e4
LT
362 uint mid_cycle;
363 int error;
364
e3bb2e30
AE
365 end_blk = *last_blk;
366 mid_blk = BLK_AVG(first_blk, end_blk);
367 while (mid_blk != first_blk && mid_blk != end_blk) {
076e6acb
CH
368 error = xlog_bread(log, mid_blk, 1, bp, &offset);
369 if (error)
1da177e4 370 return error;
03bea6fe 371 mid_cycle = xlog_get_cycle(offset);
e3bb2e30
AE
372 if (mid_cycle == cycle)
373 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
374 else
375 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
376 mid_blk = BLK_AVG(first_blk, end_blk);
1da177e4 377 }
e3bb2e30
AE
378 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
379 (mid_blk == end_blk && mid_blk-1 == first_blk));
380
381 *last_blk = end_blk;
1da177e4
LT
382
383 return 0;
384}
385
386/*
3f943d85
AE
387 * Check that a range of blocks does not contain stop_on_cycle_no.
388 * Fill in *new_blk with the block offset where such a block is
389 * found, or with -1 (an invalid block number) if there is no such
390 * block in the range. The scan needs to occur from front to back
391 * and the pointer into the region must be updated since a later
392 * routine will need to perform another test.
1da177e4
LT
393 */
394STATIC int
395xlog_find_verify_cycle(
396 xlog_t *log,
397 xfs_daddr_t start_blk,
398 int nbblks,
399 uint stop_on_cycle_no,
400 xfs_daddr_t *new_blk)
401{
402 xfs_daddr_t i, j;
403 uint cycle;
404 xfs_buf_t *bp;
405 xfs_daddr_t bufblks;
406 xfs_caddr_t buf = NULL;
407 int error = 0;
408
6881a229
AE
409 /*
410 * Greedily allocate a buffer big enough to handle the full
411 * range of basic blocks we'll be examining. If that fails,
412 * try a smaller size. We need to be able to read at least
413 * a log sector, or we're out of luck.
414 */
1da177e4 415 bufblks = 1 << ffs(nbblks);
1da177e4 416 while (!(bp = xlog_get_bp(log, bufblks))) {
1da177e4 417 bufblks >>= 1;
69ce58f0 418 if (bufblks < log->l_sectBBsize)
1da177e4
LT
419 return ENOMEM;
420 }
421
422 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
423 int bcount;
424
425 bcount = min(bufblks, (start_blk + nbblks - i));
426
076e6acb
CH
427 error = xlog_bread(log, i, bcount, bp, &buf);
428 if (error)
1da177e4
LT
429 goto out;
430
1da177e4 431 for (j = 0; j < bcount; j++) {
03bea6fe 432 cycle = xlog_get_cycle(buf);
1da177e4
LT
433 if (cycle == stop_on_cycle_no) {
434 *new_blk = i+j;
435 goto out;
436 }
437
438 buf += BBSIZE;
439 }
440 }
441
442 *new_blk = -1;
443
444out:
445 xlog_put_bp(bp);
446 return error;
447}
448
449/*
450 * Potentially backup over partial log record write.
451 *
452 * In the typical case, last_blk is the number of the block directly after
453 * a good log record. Therefore, we subtract one to get the block number
454 * of the last block in the given buffer. extra_bblks contains the number
455 * of blocks we would have read on a previous read. This happens when the
456 * last log record is split over the end of the physical log.
457 *
458 * extra_bblks is the number of blocks potentially verified on a previous
459 * call to this routine.
460 */
461STATIC int
462xlog_find_verify_log_record(
463 xlog_t *log,
464 xfs_daddr_t start_blk,
465 xfs_daddr_t *last_blk,
466 int extra_bblks)
467{
468 xfs_daddr_t i;
469 xfs_buf_t *bp;
470 xfs_caddr_t offset = NULL;
471 xlog_rec_header_t *head = NULL;
472 int error = 0;
473 int smallmem = 0;
474 int num_blks = *last_blk - start_blk;
475 int xhdrs;
476
477 ASSERT(start_blk != 0 || *last_blk != start_blk);
478
479 if (!(bp = xlog_get_bp(log, num_blks))) {
480 if (!(bp = xlog_get_bp(log, 1)))
481 return ENOMEM;
482 smallmem = 1;
483 } else {
076e6acb
CH
484 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
485 if (error)
1da177e4 486 goto out;
1da177e4
LT
487 offset += ((num_blks - 1) << BBSHIFT);
488 }
489
490 for (i = (*last_blk) - 1; i >= 0; i--) {
491 if (i < start_blk) {
492 /* valid log record not found */
a0fa2b67
DC
493 xfs_warn(log->l_mp,
494 "Log inconsistent (didn't find previous header)");
1da177e4
LT
495 ASSERT(0);
496 error = XFS_ERROR(EIO);
497 goto out;
498 }
499
500 if (smallmem) {
076e6acb
CH
501 error = xlog_bread(log, i, 1, bp, &offset);
502 if (error)
1da177e4 503 goto out;
1da177e4
LT
504 }
505
506 head = (xlog_rec_header_t *)offset;
507
b53e675d 508 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(head->h_magicno))
1da177e4
LT
509 break;
510
511 if (!smallmem)
512 offset -= BBSIZE;
513 }
514
515 /*
516 * We hit the beginning of the physical log & still no header. Return
517 * to caller. If caller can handle a return of -1, then this routine
518 * will be called again for the end of the physical log.
519 */
520 if (i == -1) {
521 error = -1;
522 goto out;
523 }
524
525 /*
526 * We have the final block of the good log (the first block
527 * of the log record _before_ the head. So we check the uuid.
528 */
529 if ((error = xlog_header_check_mount(log->l_mp, head)))
530 goto out;
531
532 /*
533 * We may have found a log record header before we expected one.
534 * last_blk will be the 1st block # with a given cycle #. We may end
535 * up reading an entire log record. In this case, we don't want to
536 * reset last_blk. Only when last_blk points in the middle of a log
537 * record do we update last_blk.
538 */
62118709 539 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
b53e675d 540 uint h_size = be32_to_cpu(head->h_size);
1da177e4
LT
541
542 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
543 if (h_size % XLOG_HEADER_CYCLE_SIZE)
544 xhdrs++;
545 } else {
546 xhdrs = 1;
547 }
548
b53e675d
CH
549 if (*last_blk - i + extra_bblks !=
550 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
1da177e4
LT
551 *last_blk = i;
552
553out:
554 xlog_put_bp(bp);
555 return error;
556}
557
558/*
559 * Head is defined to be the point of the log where the next log write
560 * write could go. This means that incomplete LR writes at the end are
561 * eliminated when calculating the head. We aren't guaranteed that previous
562 * LR have complete transactions. We only know that a cycle number of
563 * current cycle number -1 won't be present in the log if we start writing
564 * from our current block number.
565 *
566 * last_blk contains the block number of the first block with a given
567 * cycle number.
568 *
569 * Return: zero if normal, non-zero if error.
570 */
ba0f32d4 571STATIC int
1da177e4
LT
572xlog_find_head(
573 xlog_t *log,
574 xfs_daddr_t *return_head_blk)
575{
576 xfs_buf_t *bp;
577 xfs_caddr_t offset;
578 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
579 int num_scan_bblks;
580 uint first_half_cycle, last_half_cycle;
581 uint stop_on_cycle;
582 int error, log_bbnum = log->l_logBBsize;
583
584 /* Is the end of the log device zeroed? */
585 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
586 *return_head_blk = first_blk;
587
588 /* Is the whole lot zeroed? */
589 if (!first_blk) {
590 /* Linux XFS shouldn't generate totally zeroed logs -
591 * mkfs etc write a dummy unmount record to a fresh
592 * log so we can store the uuid in there
593 */
a0fa2b67 594 xfs_warn(log->l_mp, "totally zeroed log");
1da177e4
LT
595 }
596
597 return 0;
598 } else if (error) {
a0fa2b67 599 xfs_warn(log->l_mp, "empty log check failed");
1da177e4
LT
600 return error;
601 }
602
603 first_blk = 0; /* get cycle # of 1st block */
604 bp = xlog_get_bp(log, 1);
605 if (!bp)
606 return ENOMEM;
076e6acb
CH
607
608 error = xlog_bread(log, 0, 1, bp, &offset);
609 if (error)
1da177e4 610 goto bp_err;
076e6acb 611
03bea6fe 612 first_half_cycle = xlog_get_cycle(offset);
1da177e4
LT
613
614 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
076e6acb
CH
615 error = xlog_bread(log, last_blk, 1, bp, &offset);
616 if (error)
1da177e4 617 goto bp_err;
076e6acb 618
03bea6fe 619 last_half_cycle = xlog_get_cycle(offset);
1da177e4
LT
620 ASSERT(last_half_cycle != 0);
621
622 /*
623 * If the 1st half cycle number is equal to the last half cycle number,
624 * then the entire log is stamped with the same cycle number. In this
625 * case, head_blk can't be set to zero (which makes sense). The below
626 * math doesn't work out properly with head_blk equal to zero. Instead,
627 * we set it to log_bbnum which is an invalid block number, but this
628 * value makes the math correct. If head_blk doesn't changed through
629 * all the tests below, *head_blk is set to zero at the very end rather
630 * than log_bbnum. In a sense, log_bbnum and zero are the same block
631 * in a circular file.
632 */
633 if (first_half_cycle == last_half_cycle) {
634 /*
635 * In this case we believe that the entire log should have
636 * cycle number last_half_cycle. We need to scan backwards
637 * from the end verifying that there are no holes still
638 * containing last_half_cycle - 1. If we find such a hole,
639 * then the start of that hole will be the new head. The
640 * simple case looks like
641 * x | x ... | x - 1 | x
642 * Another case that fits this picture would be
643 * x | x + 1 | x ... | x
c41564b5 644 * In this case the head really is somewhere at the end of the
1da177e4
LT
645 * log, as one of the latest writes at the beginning was
646 * incomplete.
647 * One more case is
648 * x | x + 1 | x ... | x - 1 | x
649 * This is really the combination of the above two cases, and
650 * the head has to end up at the start of the x-1 hole at the
651 * end of the log.
652 *
653 * In the 256k log case, we will read from the beginning to the
654 * end of the log and search for cycle numbers equal to x-1.
655 * We don't worry about the x+1 blocks that we encounter,
656 * because we know that they cannot be the head since the log
657 * started with x.
658 */
659 head_blk = log_bbnum;
660 stop_on_cycle = last_half_cycle - 1;
661 } else {
662 /*
663 * In this case we want to find the first block with cycle
664 * number matching last_half_cycle. We expect the log to be
665 * some variation on
3f943d85 666 * x + 1 ... | x ... | x
1da177e4
LT
667 * The first block with cycle number x (last_half_cycle) will
668 * be where the new head belongs. First we do a binary search
669 * for the first occurrence of last_half_cycle. The binary
670 * search may not be totally accurate, so then we scan back
671 * from there looking for occurrences of last_half_cycle before
672 * us. If that backwards scan wraps around the beginning of
673 * the log, then we look for occurrences of last_half_cycle - 1
674 * at the end of the log. The cases we're looking for look
675 * like
3f943d85
AE
676 * v binary search stopped here
677 * x + 1 ... | x | x + 1 | x ... | x
678 * ^ but we want to locate this spot
1da177e4 679 * or
1da177e4 680 * <---------> less than scan distance
3f943d85
AE
681 * x + 1 ... | x ... | x - 1 | x
682 * ^ we want to locate this spot
1da177e4
LT
683 */
684 stop_on_cycle = last_half_cycle;
685 if ((error = xlog_find_cycle_start(log, bp, first_blk,
686 &head_blk, last_half_cycle)))
687 goto bp_err;
688 }
689
690 /*
691 * Now validate the answer. Scan back some number of maximum possible
692 * blocks and make sure each one has the expected cycle number. The
693 * maximum is determined by the total possible amount of buffering
694 * in the in-core log. The following number can be made tighter if
695 * we actually look at the block size of the filesystem.
696 */
697 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
698 if (head_blk >= num_scan_bblks) {
699 /*
700 * We are guaranteed that the entire check can be performed
701 * in one buffer.
702 */
703 start_blk = head_blk - num_scan_bblks;
704 if ((error = xlog_find_verify_cycle(log,
705 start_blk, num_scan_bblks,
706 stop_on_cycle, &new_blk)))
707 goto bp_err;
708 if (new_blk != -1)
709 head_blk = new_blk;
710 } else { /* need to read 2 parts of log */
711 /*
712 * We are going to scan backwards in the log in two parts.
713 * First we scan the physical end of the log. In this part
714 * of the log, we are looking for blocks with cycle number
715 * last_half_cycle - 1.
716 * If we find one, then we know that the log starts there, as
717 * we've found a hole that didn't get written in going around
718 * the end of the physical log. The simple case for this is
719 * x + 1 ... | x ... | x - 1 | x
720 * <---------> less than scan distance
721 * If all of the blocks at the end of the log have cycle number
722 * last_half_cycle, then we check the blocks at the start of
723 * the log looking for occurrences of last_half_cycle. If we
724 * find one, then our current estimate for the location of the
725 * first occurrence of last_half_cycle is wrong and we move
726 * back to the hole we've found. This case looks like
727 * x + 1 ... | x | x + 1 | x ...
728 * ^ binary search stopped here
729 * Another case we need to handle that only occurs in 256k
730 * logs is
731 * x + 1 ... | x ... | x+1 | x ...
732 * ^ binary search stops here
733 * In a 256k log, the scan at the end of the log will see the
734 * x + 1 blocks. We need to skip past those since that is
735 * certainly not the head of the log. By searching for
736 * last_half_cycle-1 we accomplish that.
737 */
1da177e4 738 ASSERT(head_blk <= INT_MAX &&
3f943d85
AE
739 (xfs_daddr_t) num_scan_bblks >= head_blk);
740 start_blk = log_bbnum - (num_scan_bblks - head_blk);
1da177e4
LT
741 if ((error = xlog_find_verify_cycle(log, start_blk,
742 num_scan_bblks - (int)head_blk,
743 (stop_on_cycle - 1), &new_blk)))
744 goto bp_err;
745 if (new_blk != -1) {
746 head_blk = new_blk;
9db127ed 747 goto validate_head;
1da177e4
LT
748 }
749
750 /*
751 * Scan beginning of log now. The last part of the physical
752 * log is good. This scan needs to verify that it doesn't find
753 * the last_half_cycle.
754 */
755 start_blk = 0;
756 ASSERT(head_blk <= INT_MAX);
757 if ((error = xlog_find_verify_cycle(log,
758 start_blk, (int)head_blk,
759 stop_on_cycle, &new_blk)))
760 goto bp_err;
761 if (new_blk != -1)
762 head_blk = new_blk;
763 }
764
9db127ed 765validate_head:
1da177e4
LT
766 /*
767 * Now we need to make sure head_blk is not pointing to a block in
768 * the middle of a log record.
769 */
770 num_scan_bblks = XLOG_REC_SHIFT(log);
771 if (head_blk >= num_scan_bblks) {
772 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
773
774 /* start ptr at last block ptr before head_blk */
775 if ((error = xlog_find_verify_log_record(log, start_blk,
776 &head_blk, 0)) == -1) {
777 error = XFS_ERROR(EIO);
778 goto bp_err;
779 } else if (error)
780 goto bp_err;
781 } else {
782 start_blk = 0;
783 ASSERT(head_blk <= INT_MAX);
784 if ((error = xlog_find_verify_log_record(log, start_blk,
785 &head_blk, 0)) == -1) {
786 /* We hit the beginning of the log during our search */
3f943d85 787 start_blk = log_bbnum - (num_scan_bblks - head_blk);
1da177e4
LT
788 new_blk = log_bbnum;
789 ASSERT(start_blk <= INT_MAX &&
790 (xfs_daddr_t) log_bbnum-start_blk >= 0);
791 ASSERT(head_blk <= INT_MAX);
792 if ((error = xlog_find_verify_log_record(log,
793 start_blk, &new_blk,
794 (int)head_blk)) == -1) {
795 error = XFS_ERROR(EIO);
796 goto bp_err;
797 } else if (error)
798 goto bp_err;
799 if (new_blk != log_bbnum)
800 head_blk = new_blk;
801 } else if (error)
802 goto bp_err;
803 }
804
805 xlog_put_bp(bp);
806 if (head_blk == log_bbnum)
807 *return_head_blk = 0;
808 else
809 *return_head_blk = head_blk;
810 /*
811 * When returning here, we have a good block number. Bad block
812 * means that during a previous crash, we didn't have a clean break
813 * from cycle number N to cycle number N-1. In this case, we need
814 * to find the first block with cycle number N-1.
815 */
816 return 0;
817
818 bp_err:
819 xlog_put_bp(bp);
820
821 if (error)
a0fa2b67 822 xfs_warn(log->l_mp, "failed to find log head");
1da177e4
LT
823 return error;
824}
825
826/*
827 * Find the sync block number or the tail of the log.
828 *
829 * This will be the block number of the last record to have its
830 * associated buffers synced to disk. Every log record header has
831 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
832 * to get a sync block number. The only concern is to figure out which
833 * log record header to believe.
834 *
835 * The following algorithm uses the log record header with the largest
836 * lsn. The entire log record does not need to be valid. We only care
837 * that the header is valid.
838 *
839 * We could speed up search by using current head_blk buffer, but it is not
840 * available.
841 */
5d77c0dc 842STATIC int
1da177e4
LT
843xlog_find_tail(
844 xlog_t *log,
845 xfs_daddr_t *head_blk,
65be6054 846 xfs_daddr_t *tail_blk)
1da177e4
LT
847{
848 xlog_rec_header_t *rhead;
849 xlog_op_header_t *op_head;
850 xfs_caddr_t offset = NULL;
851 xfs_buf_t *bp;
852 int error, i, found;
853 xfs_daddr_t umount_data_blk;
854 xfs_daddr_t after_umount_blk;
855 xfs_lsn_t tail_lsn;
856 int hblks;
857
858 found = 0;
859
860 /*
861 * Find previous log record
862 */
863 if ((error = xlog_find_head(log, head_blk)))
864 return error;
865
866 bp = xlog_get_bp(log, 1);
867 if (!bp)
868 return ENOMEM;
869 if (*head_blk == 0) { /* special case */
076e6acb
CH
870 error = xlog_bread(log, 0, 1, bp, &offset);
871 if (error)
9db127ed 872 goto done;
076e6acb 873
03bea6fe 874 if (xlog_get_cycle(offset) == 0) {
1da177e4
LT
875 *tail_blk = 0;
876 /* leave all other log inited values alone */
9db127ed 877 goto done;
1da177e4
LT
878 }
879 }
880
881 /*
882 * Search backwards looking for log record header block
883 */
884 ASSERT(*head_blk < INT_MAX);
885 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
076e6acb
CH
886 error = xlog_bread(log, i, 1, bp, &offset);
887 if (error)
9db127ed 888 goto done;
076e6acb 889
b53e675d 890 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(*(__be32 *)offset)) {
1da177e4
LT
891 found = 1;
892 break;
893 }
894 }
895 /*
896 * If we haven't found the log record header block, start looking
897 * again from the end of the physical log. XXXmiken: There should be
898 * a check here to make sure we didn't search more than N blocks in
899 * the previous code.
900 */
901 if (!found) {
902 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
076e6acb
CH
903 error = xlog_bread(log, i, 1, bp, &offset);
904 if (error)
9db127ed 905 goto done;
076e6acb 906
1da177e4 907 if (XLOG_HEADER_MAGIC_NUM ==
b53e675d 908 be32_to_cpu(*(__be32 *)offset)) {
1da177e4
LT
909 found = 2;
910 break;
911 }
912 }
913 }
914 if (!found) {
a0fa2b67 915 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
1da177e4
LT
916 ASSERT(0);
917 return XFS_ERROR(EIO);
918 }
919
920 /* find blk_no of tail of log */
921 rhead = (xlog_rec_header_t *)offset;
b53e675d 922 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
1da177e4
LT
923
924 /*
925 * Reset log values according to the state of the log when we
926 * crashed. In the case where head_blk == 0, we bump curr_cycle
927 * one because the next write starts a new cycle rather than
928 * continuing the cycle of the last good log record. At this
929 * point we have guaranteed that all partial log records have been
930 * accounted for. Therefore, we know that the last good log record
931 * written was complete and ended exactly on the end boundary
932 * of the physical log.
933 */
934 log->l_prev_block = i;
935 log->l_curr_block = (int)*head_blk;
b53e675d 936 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
1da177e4
LT
937 if (found == 2)
938 log->l_curr_cycle++;
1c3cb9ec 939 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
84f3c683 940 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
a69ed03c
DC
941 xlog_assign_grant_head(&log->l_grant_reserve_head, log->l_curr_cycle,
942 BBTOB(log->l_curr_block));
943 xlog_assign_grant_head(&log->l_grant_write_head, log->l_curr_cycle,
944 BBTOB(log->l_curr_block));
1da177e4
LT
945
946 /*
947 * Look for unmount record. If we find it, then we know there
948 * was a clean unmount. Since 'i' could be the last block in
949 * the physical log, we convert to a log block before comparing
950 * to the head_blk.
951 *
952 * Save the current tail lsn to use to pass to
953 * xlog_clear_stale_blocks() below. We won't want to clear the
954 * unmount record if there is one, so we pass the lsn of the
955 * unmount record rather than the block after it.
956 */
62118709 957 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
b53e675d
CH
958 int h_size = be32_to_cpu(rhead->h_size);
959 int h_version = be32_to_cpu(rhead->h_version);
1da177e4
LT
960
961 if ((h_version & XLOG_VERSION_2) &&
962 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
963 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
964 if (h_size % XLOG_HEADER_CYCLE_SIZE)
965 hblks++;
966 } else {
967 hblks = 1;
968 }
969 } else {
970 hblks = 1;
971 }
972 after_umount_blk = (i + hblks + (int)
b53e675d 973 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
1c3cb9ec 974 tail_lsn = atomic64_read(&log->l_tail_lsn);
1da177e4 975 if (*head_blk == after_umount_blk &&
b53e675d 976 be32_to_cpu(rhead->h_num_logops) == 1) {
1da177e4 977 umount_data_blk = (i + hblks) % log->l_logBBsize;
076e6acb
CH
978 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
979 if (error)
9db127ed 980 goto done;
076e6acb 981
1da177e4
LT
982 op_head = (xlog_op_header_t *)offset;
983 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
984 /*
985 * Set tail and last sync so that newly written
986 * log records will point recovery to after the
987 * current unmount record.
988 */
1c3cb9ec
DC
989 xlog_assign_atomic_lsn(&log->l_tail_lsn,
990 log->l_curr_cycle, after_umount_blk);
991 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
992 log->l_curr_cycle, after_umount_blk);
1da177e4 993 *tail_blk = after_umount_blk;
92821e2b
DC
994
995 /*
996 * Note that the unmount was clean. If the unmount
997 * was not clean, we need to know this to rebuild the
998 * superblock counters from the perag headers if we
999 * have a filesystem using non-persistent counters.
1000 */
1001 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1da177e4
LT
1002 }
1003 }
1004
1005 /*
1006 * Make sure that there are no blocks in front of the head
1007 * with the same cycle number as the head. This can happen
1008 * because we allow multiple outstanding log writes concurrently,
1009 * and the later writes might make it out before earlier ones.
1010 *
1011 * We use the lsn from before modifying it so that we'll never
1012 * overwrite the unmount record after a clean unmount.
1013 *
1014 * Do this only if we are going to recover the filesystem
1015 *
1016 * NOTE: This used to say "if (!readonly)"
1017 * However on Linux, we can & do recover a read-only filesystem.
1018 * We only skip recovery if NORECOVERY is specified on mount,
1019 * in which case we would not be here.
1020 *
1021 * But... if the -device- itself is readonly, just skip this.
1022 * We can't recover this device anyway, so it won't matter.
1023 */
9db127ed 1024 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
1da177e4 1025 error = xlog_clear_stale_blocks(log, tail_lsn);
1da177e4 1026
9db127ed 1027done:
1da177e4
LT
1028 xlog_put_bp(bp);
1029
1030 if (error)
a0fa2b67 1031 xfs_warn(log->l_mp, "failed to locate log tail");
1da177e4
LT
1032 return error;
1033}
1034
1035/*
1036 * Is the log zeroed at all?
1037 *
1038 * The last binary search should be changed to perform an X block read
1039 * once X becomes small enough. You can then search linearly through
1040 * the X blocks. This will cut down on the number of reads we need to do.
1041 *
1042 * If the log is partially zeroed, this routine will pass back the blkno
1043 * of the first block with cycle number 0. It won't have a complete LR
1044 * preceding it.
1045 *
1046 * Return:
1047 * 0 => the log is completely written to
1048 * -1 => use *blk_no as the first block of the log
1049 * >0 => error has occurred
1050 */
a8272ce0 1051STATIC int
1da177e4
LT
1052xlog_find_zeroed(
1053 xlog_t *log,
1054 xfs_daddr_t *blk_no)
1055{
1056 xfs_buf_t *bp;
1057 xfs_caddr_t offset;
1058 uint first_cycle, last_cycle;
1059 xfs_daddr_t new_blk, last_blk, start_blk;
1060 xfs_daddr_t num_scan_bblks;
1061 int error, log_bbnum = log->l_logBBsize;
1062
6fdf8ccc
NS
1063 *blk_no = 0;
1064
1da177e4
LT
1065 /* check totally zeroed log */
1066 bp = xlog_get_bp(log, 1);
1067 if (!bp)
1068 return ENOMEM;
076e6acb
CH
1069 error = xlog_bread(log, 0, 1, bp, &offset);
1070 if (error)
1da177e4 1071 goto bp_err;
076e6acb 1072
03bea6fe 1073 first_cycle = xlog_get_cycle(offset);
1da177e4
LT
1074 if (first_cycle == 0) { /* completely zeroed log */
1075 *blk_no = 0;
1076 xlog_put_bp(bp);
1077 return -1;
1078 }
1079
1080 /* check partially zeroed log */
076e6acb
CH
1081 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1082 if (error)
1da177e4 1083 goto bp_err;
076e6acb 1084
03bea6fe 1085 last_cycle = xlog_get_cycle(offset);
1da177e4
LT
1086 if (last_cycle != 0) { /* log completely written to */
1087 xlog_put_bp(bp);
1088 return 0;
1089 } else if (first_cycle != 1) {
1090 /*
1091 * If the cycle of the last block is zero, the cycle of
1092 * the first block must be 1. If it's not, maybe we're
1093 * not looking at a log... Bail out.
1094 */
a0fa2b67
DC
1095 xfs_warn(log->l_mp,
1096 "Log inconsistent or not a log (last==0, first!=1)");
1da177e4
LT
1097 return XFS_ERROR(EINVAL);
1098 }
1099
1100 /* we have a partially zeroed log */
1101 last_blk = log_bbnum-1;
1102 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1103 goto bp_err;
1104
1105 /*
1106 * Validate the answer. Because there is no way to guarantee that
1107 * the entire log is made up of log records which are the same size,
1108 * we scan over the defined maximum blocks. At this point, the maximum
1109 * is not chosen to mean anything special. XXXmiken
1110 */
1111 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1112 ASSERT(num_scan_bblks <= INT_MAX);
1113
1114 if (last_blk < num_scan_bblks)
1115 num_scan_bblks = last_blk;
1116 start_blk = last_blk - num_scan_bblks;
1117
1118 /*
1119 * We search for any instances of cycle number 0 that occur before
1120 * our current estimate of the head. What we're trying to detect is
1121 * 1 ... | 0 | 1 | 0...
1122 * ^ binary search ends here
1123 */
1124 if ((error = xlog_find_verify_cycle(log, start_blk,
1125 (int)num_scan_bblks, 0, &new_blk)))
1126 goto bp_err;
1127 if (new_blk != -1)
1128 last_blk = new_blk;
1129
1130 /*
1131 * Potentially backup over partial log record write. We don't need
1132 * to search the end of the log because we know it is zero.
1133 */
1134 if ((error = xlog_find_verify_log_record(log, start_blk,
1135 &last_blk, 0)) == -1) {
1136 error = XFS_ERROR(EIO);
1137 goto bp_err;
1138 } else if (error)
1139 goto bp_err;
1140
1141 *blk_no = last_blk;
1142bp_err:
1143 xlog_put_bp(bp);
1144 if (error)
1145 return error;
1146 return -1;
1147}
1148
1149/*
1150 * These are simple subroutines used by xlog_clear_stale_blocks() below
1151 * to initialize a buffer full of empty log record headers and write
1152 * them into the log.
1153 */
1154STATIC void
1155xlog_add_record(
1156 xlog_t *log,
1157 xfs_caddr_t buf,
1158 int cycle,
1159 int block,
1160 int tail_cycle,
1161 int tail_block)
1162{
1163 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1164
1165 memset(buf, 0, BBSIZE);
b53e675d
CH
1166 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1167 recp->h_cycle = cpu_to_be32(cycle);
1168 recp->h_version = cpu_to_be32(
62118709 1169 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
b53e675d
CH
1170 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1171 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1172 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1da177e4
LT
1173 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1174}
1175
1176STATIC int
1177xlog_write_log_records(
1178 xlog_t *log,
1179 int cycle,
1180 int start_block,
1181 int blocks,
1182 int tail_cycle,
1183 int tail_block)
1184{
1185 xfs_caddr_t offset;
1186 xfs_buf_t *bp;
1187 int balign, ealign;
69ce58f0 1188 int sectbb = log->l_sectBBsize;
1da177e4
LT
1189 int end_block = start_block + blocks;
1190 int bufblks;
1191 int error = 0;
1192 int i, j = 0;
1193
6881a229
AE
1194 /*
1195 * Greedily allocate a buffer big enough to handle the full
1196 * range of basic blocks to be written. If that fails, try
1197 * a smaller size. We need to be able to write at least a
1198 * log sector, or we're out of luck.
1199 */
1da177e4
LT
1200 bufblks = 1 << ffs(blocks);
1201 while (!(bp = xlog_get_bp(log, bufblks))) {
1202 bufblks >>= 1;
69ce58f0 1203 if (bufblks < sectbb)
1da177e4
LT
1204 return ENOMEM;
1205 }
1206
1207 /* We may need to do a read at the start to fill in part of
1208 * the buffer in the starting sector not covered by the first
1209 * write below.
1210 */
5c17f533 1211 balign = round_down(start_block, sectbb);
1da177e4 1212 if (balign != start_block) {
076e6acb
CH
1213 error = xlog_bread_noalign(log, start_block, 1, bp);
1214 if (error)
1215 goto out_put_bp;
1216
1da177e4
LT
1217 j = start_block - balign;
1218 }
1219
1220 for (i = start_block; i < end_block; i += bufblks) {
1221 int bcount, endcount;
1222
1223 bcount = min(bufblks, end_block - start_block);
1224 endcount = bcount - j;
1225
1226 /* We may need to do a read at the end to fill in part of
1227 * the buffer in the final sector not covered by the write.
1228 * If this is the same sector as the above read, skip it.
1229 */
5c17f533 1230 ealign = round_down(end_block, sectbb);
1da177e4
LT
1231 if (j == 0 && (start_block + endcount > ealign)) {
1232 offset = XFS_BUF_PTR(bp);
1233 balign = BBTOB(ealign - start_block);
234f56ac
DC
1234 error = XFS_BUF_SET_PTR(bp, offset + balign,
1235 BBTOB(sectbb));
076e6acb
CH
1236 if (error)
1237 break;
1238
1239 error = xlog_bread_noalign(log, ealign, sectbb, bp);
1240 if (error)
1241 break;
1242
1243 error = XFS_BUF_SET_PTR(bp, offset, bufblks);
234f56ac 1244 if (error)
1da177e4 1245 break;
1da177e4
LT
1246 }
1247
1248 offset = xlog_align(log, start_block, endcount, bp);
1249 for (; j < endcount; j++) {
1250 xlog_add_record(log, offset, cycle, i+j,
1251 tail_cycle, tail_block);
1252 offset += BBSIZE;
1253 }
1254 error = xlog_bwrite(log, start_block, endcount, bp);
1255 if (error)
1256 break;
1257 start_block += endcount;
1258 j = 0;
1259 }
076e6acb
CH
1260
1261 out_put_bp:
1da177e4
LT
1262 xlog_put_bp(bp);
1263 return error;
1264}
1265
1266/*
1267 * This routine is called to blow away any incomplete log writes out
1268 * in front of the log head. We do this so that we won't become confused
1269 * if we come up, write only a little bit more, and then crash again.
1270 * If we leave the partial log records out there, this situation could
1271 * cause us to think those partial writes are valid blocks since they
1272 * have the current cycle number. We get rid of them by overwriting them
1273 * with empty log records with the old cycle number rather than the
1274 * current one.
1275 *
1276 * The tail lsn is passed in rather than taken from
1277 * the log so that we will not write over the unmount record after a
1278 * clean unmount in a 512 block log. Doing so would leave the log without
1279 * any valid log records in it until a new one was written. If we crashed
1280 * during that time we would not be able to recover.
1281 */
1282STATIC int
1283xlog_clear_stale_blocks(
1284 xlog_t *log,
1285 xfs_lsn_t tail_lsn)
1286{
1287 int tail_cycle, head_cycle;
1288 int tail_block, head_block;
1289 int tail_distance, max_distance;
1290 int distance;
1291 int error;
1292
1293 tail_cycle = CYCLE_LSN(tail_lsn);
1294 tail_block = BLOCK_LSN(tail_lsn);
1295 head_cycle = log->l_curr_cycle;
1296 head_block = log->l_curr_block;
1297
1298 /*
1299 * Figure out the distance between the new head of the log
1300 * and the tail. We want to write over any blocks beyond the
1301 * head that we may have written just before the crash, but
1302 * we don't want to overwrite the tail of the log.
1303 */
1304 if (head_cycle == tail_cycle) {
1305 /*
1306 * The tail is behind the head in the physical log,
1307 * so the distance from the head to the tail is the
1308 * distance from the head to the end of the log plus
1309 * the distance from the beginning of the log to the
1310 * tail.
1311 */
1312 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1313 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1314 XFS_ERRLEVEL_LOW, log->l_mp);
1315 return XFS_ERROR(EFSCORRUPTED);
1316 }
1317 tail_distance = tail_block + (log->l_logBBsize - head_block);
1318 } else {
1319 /*
1320 * The head is behind the tail in the physical log,
1321 * so the distance from the head to the tail is just
1322 * the tail block minus the head block.
1323 */
1324 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1325 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1326 XFS_ERRLEVEL_LOW, log->l_mp);
1327 return XFS_ERROR(EFSCORRUPTED);
1328 }
1329 tail_distance = tail_block - head_block;
1330 }
1331
1332 /*
1333 * If the head is right up against the tail, we can't clear
1334 * anything.
1335 */
1336 if (tail_distance <= 0) {
1337 ASSERT(tail_distance == 0);
1338 return 0;
1339 }
1340
1341 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1342 /*
1343 * Take the smaller of the maximum amount of outstanding I/O
1344 * we could have and the distance to the tail to clear out.
1345 * We take the smaller so that we don't overwrite the tail and
1346 * we don't waste all day writing from the head to the tail
1347 * for no reason.
1348 */
1349 max_distance = MIN(max_distance, tail_distance);
1350
1351 if ((head_block + max_distance) <= log->l_logBBsize) {
1352 /*
1353 * We can stomp all the blocks we need to without
1354 * wrapping around the end of the log. Just do it
1355 * in a single write. Use the cycle number of the
1356 * current cycle minus one so that the log will look like:
1357 * n ... | n - 1 ...
1358 */
1359 error = xlog_write_log_records(log, (head_cycle - 1),
1360 head_block, max_distance, tail_cycle,
1361 tail_block);
1362 if (error)
1363 return error;
1364 } else {
1365 /*
1366 * We need to wrap around the end of the physical log in
1367 * order to clear all the blocks. Do it in two separate
1368 * I/Os. The first write should be from the head to the
1369 * end of the physical log, and it should use the current
1370 * cycle number minus one just like above.
1371 */
1372 distance = log->l_logBBsize - head_block;
1373 error = xlog_write_log_records(log, (head_cycle - 1),
1374 head_block, distance, tail_cycle,
1375 tail_block);
1376
1377 if (error)
1378 return error;
1379
1380 /*
1381 * Now write the blocks at the start of the physical log.
1382 * This writes the remainder of the blocks we want to clear.
1383 * It uses the current cycle number since we're now on the
1384 * same cycle as the head so that we get:
1385 * n ... n ... | n - 1 ...
1386 * ^^^^^ blocks we're writing
1387 */
1388 distance = max_distance - (log->l_logBBsize - head_block);
1389 error = xlog_write_log_records(log, head_cycle, 0, distance,
1390 tail_cycle, tail_block);
1391 if (error)
1392 return error;
1393 }
1394
1395 return 0;
1396}
1397
1398/******************************************************************************
1399 *
1400 * Log recover routines
1401 *
1402 ******************************************************************************
1403 */
1404
1405STATIC xlog_recover_t *
1406xlog_recover_find_tid(
f0a76953 1407 struct hlist_head *head,
1da177e4
LT
1408 xlog_tid_t tid)
1409{
f0a76953
DC
1410 xlog_recover_t *trans;
1411 struct hlist_node *n;
1da177e4 1412
f0a76953
DC
1413 hlist_for_each_entry(trans, n, head, r_list) {
1414 if (trans->r_log_tid == tid)
1415 return trans;
1da177e4 1416 }
f0a76953 1417 return NULL;
1da177e4
LT
1418}
1419
1420STATIC void
f0a76953
DC
1421xlog_recover_new_tid(
1422 struct hlist_head *head,
1423 xlog_tid_t tid,
1424 xfs_lsn_t lsn)
1da177e4 1425{
f0a76953
DC
1426 xlog_recover_t *trans;
1427
1428 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1429 trans->r_log_tid = tid;
1430 trans->r_lsn = lsn;
1431 INIT_LIST_HEAD(&trans->r_itemq);
1432
1433 INIT_HLIST_NODE(&trans->r_list);
1434 hlist_add_head(&trans->r_list, head);
1da177e4
LT
1435}
1436
1437STATIC void
1438xlog_recover_add_item(
f0a76953 1439 struct list_head *head)
1da177e4
LT
1440{
1441 xlog_recover_item_t *item;
1442
1443 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
f0a76953
DC
1444 INIT_LIST_HEAD(&item->ri_list);
1445 list_add_tail(&item->ri_list, head);
1da177e4
LT
1446}
1447
1448STATIC int
1449xlog_recover_add_to_cont_trans(
9abbc539 1450 struct log *log,
1da177e4
LT
1451 xlog_recover_t *trans,
1452 xfs_caddr_t dp,
1453 int len)
1454{
1455 xlog_recover_item_t *item;
1456 xfs_caddr_t ptr, old_ptr;
1457 int old_len;
1458
f0a76953 1459 if (list_empty(&trans->r_itemq)) {
1da177e4
LT
1460 /* finish copying rest of trans header */
1461 xlog_recover_add_item(&trans->r_itemq);
1462 ptr = (xfs_caddr_t) &trans->r_theader +
1463 sizeof(xfs_trans_header_t) - len;
1464 memcpy(ptr, dp, len); /* d, s, l */
1465 return 0;
1466 }
f0a76953
DC
1467 /* take the tail entry */
1468 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1da177e4
LT
1469
1470 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1471 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1472
760dea67 1473 ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
1da177e4
LT
1474 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1475 item->ri_buf[item->ri_cnt-1].i_len += len;
1476 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
9abbc539 1477 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
1da177e4
LT
1478 return 0;
1479}
1480
1481/*
1482 * The next region to add is the start of a new region. It could be
1483 * a whole region or it could be the first part of a new region. Because
1484 * of this, the assumption here is that the type and size fields of all
1485 * format structures fit into the first 32 bits of the structure.
1486 *
1487 * This works because all regions must be 32 bit aligned. Therefore, we
1488 * either have both fields or we have neither field. In the case we have
1489 * neither field, the data part of the region is zero length. We only have
1490 * a log_op_header and can throw away the header since a new one will appear
1491 * later. If we have at least 4 bytes, then we can determine how many regions
1492 * will appear in the current log item.
1493 */
1494STATIC int
1495xlog_recover_add_to_trans(
9abbc539 1496 struct log *log,
1da177e4
LT
1497 xlog_recover_t *trans,
1498 xfs_caddr_t dp,
1499 int len)
1500{
1501 xfs_inode_log_format_t *in_f; /* any will do */
1502 xlog_recover_item_t *item;
1503 xfs_caddr_t ptr;
1504
1505 if (!len)
1506 return 0;
f0a76953 1507 if (list_empty(&trans->r_itemq)) {
5a792c45
DC
1508 /* we need to catch log corruptions here */
1509 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
a0fa2b67
DC
1510 xfs_warn(log->l_mp, "%s: bad header magic number",
1511 __func__);
5a792c45
DC
1512 ASSERT(0);
1513 return XFS_ERROR(EIO);
1514 }
1da177e4
LT
1515 if (len == sizeof(xfs_trans_header_t))
1516 xlog_recover_add_item(&trans->r_itemq);
1517 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1518 return 0;
1519 }
1520
1521 ptr = kmem_alloc(len, KM_SLEEP);
1522 memcpy(ptr, dp, len);
1523 in_f = (xfs_inode_log_format_t *)ptr;
1524
f0a76953
DC
1525 /* take the tail entry */
1526 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1527 if (item->ri_total != 0 &&
1528 item->ri_total == item->ri_cnt) {
1529 /* tail item is in use, get a new one */
1da177e4 1530 xlog_recover_add_item(&trans->r_itemq);
f0a76953
DC
1531 item = list_entry(trans->r_itemq.prev,
1532 xlog_recover_item_t, ri_list);
1da177e4 1533 }
1da177e4
LT
1534
1535 if (item->ri_total == 0) { /* first region to be added */
e8fa6b48
CH
1536 if (in_f->ilf_size == 0 ||
1537 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
a0fa2b67
DC
1538 xfs_warn(log->l_mp,
1539 "bad number of regions (%d) in inode log format",
e8fa6b48
CH
1540 in_f->ilf_size);
1541 ASSERT(0);
1542 return XFS_ERROR(EIO);
1543 }
1544
1545 item->ri_total = in_f->ilf_size;
1546 item->ri_buf =
1547 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1548 KM_SLEEP);
1da177e4
LT
1549 }
1550 ASSERT(item->ri_total > item->ri_cnt);
1551 /* Description region is ri_buf[0] */
1552 item->ri_buf[item->ri_cnt].i_addr = ptr;
1553 item->ri_buf[item->ri_cnt].i_len = len;
1554 item->ri_cnt++;
9abbc539 1555 trace_xfs_log_recover_item_add(log, trans, item, 0);
1da177e4
LT
1556 return 0;
1557}
1558
f0a76953
DC
1559/*
1560 * Sort the log items in the transaction. Cancelled buffers need
1561 * to be put first so they are processed before any items that might
1562 * modify the buffers. If they are cancelled, then the modifications
1563 * don't need to be replayed.
1564 */
1da177e4
LT
1565STATIC int
1566xlog_recover_reorder_trans(
9abbc539
DC
1567 struct log *log,
1568 xlog_recover_t *trans,
1569 int pass)
1da177e4 1570{
f0a76953
DC
1571 xlog_recover_item_t *item, *n;
1572 LIST_HEAD(sort_list);
1573
1574 list_splice_init(&trans->r_itemq, &sort_list);
1575 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
4e0d5f92 1576 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1da177e4 1577
f0a76953 1578 switch (ITEM_TYPE(item)) {
1da177e4 1579 case XFS_LI_BUF:
c1155410 1580 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
9abbc539
DC
1581 trace_xfs_log_recover_item_reorder_head(log,
1582 trans, item, pass);
f0a76953 1583 list_move(&item->ri_list, &trans->r_itemq);
1da177e4
LT
1584 break;
1585 }
1586 case XFS_LI_INODE:
1da177e4
LT
1587 case XFS_LI_DQUOT:
1588 case XFS_LI_QUOTAOFF:
1589 case XFS_LI_EFD:
1590 case XFS_LI_EFI:
9abbc539
DC
1591 trace_xfs_log_recover_item_reorder_tail(log,
1592 trans, item, pass);
f0a76953 1593 list_move_tail(&item->ri_list, &trans->r_itemq);
1da177e4
LT
1594 break;
1595 default:
a0fa2b67
DC
1596 xfs_warn(log->l_mp,
1597 "%s: unrecognized type of log operation",
1598 __func__);
1da177e4
LT
1599 ASSERT(0);
1600 return XFS_ERROR(EIO);
1601 }
f0a76953
DC
1602 }
1603 ASSERT(list_empty(&sort_list));
1da177e4
LT
1604 return 0;
1605}
1606
1607/*
1608 * Build up the table of buf cancel records so that we don't replay
1609 * cancelled data in the second pass. For buffer records that are
1610 * not cancel records, there is nothing to do here so we just return.
1611 *
1612 * If we get a cancel record which is already in the table, this indicates
1613 * that the buffer was cancelled multiple times. In order to ensure
1614 * that during pass 2 we keep the record in the table until we reach its
1615 * last occurrence in the log, we keep a reference count in the cancel
1616 * record in the table to tell us how many times we expect to see this
1617 * record during the second pass.
1618 */
c9f71f5f
CH
1619STATIC int
1620xlog_recover_buffer_pass1(
d5689eaa 1621 struct log *log,
c9f71f5f 1622 xlog_recover_item_t *item)
1da177e4 1623{
c9f71f5f 1624 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
d5689eaa
CH
1625 struct list_head *bucket;
1626 struct xfs_buf_cancel *bcp;
1da177e4
LT
1627
1628 /*
1629 * If this isn't a cancel buffer item, then just return.
1630 */
e2714bf8 1631 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
9abbc539 1632 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
c9f71f5f 1633 return 0;
9abbc539 1634 }
1da177e4
LT
1635
1636 /*
d5689eaa
CH
1637 * Insert an xfs_buf_cancel record into the hash table of them.
1638 * If there is already an identical record, bump its reference count.
1da177e4 1639 */
d5689eaa
CH
1640 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1641 list_for_each_entry(bcp, bucket, bc_list) {
1642 if (bcp->bc_blkno == buf_f->blf_blkno &&
1643 bcp->bc_len == buf_f->blf_len) {
1644 bcp->bc_refcount++;
9abbc539 1645 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
c9f71f5f 1646 return 0;
1da177e4 1647 }
d5689eaa
CH
1648 }
1649
1650 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1651 bcp->bc_blkno = buf_f->blf_blkno;
1652 bcp->bc_len = buf_f->blf_len;
1da177e4 1653 bcp->bc_refcount = 1;
d5689eaa
CH
1654 list_add_tail(&bcp->bc_list, bucket);
1655
9abbc539 1656 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
c9f71f5f 1657 return 0;
1da177e4
LT
1658}
1659
1660/*
1661 * Check to see whether the buffer being recovered has a corresponding
1662 * entry in the buffer cancel record table. If it does then return 1
1663 * so that it will be cancelled, otherwise return 0. If the buffer is
c1155410 1664 * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement
1da177e4
LT
1665 * the refcount on the entry in the table and remove it from the table
1666 * if this is the last reference.
1667 *
1668 * We remove the cancel record from the table when we encounter its
1669 * last occurrence in the log so that if the same buffer is re-used
1670 * again after its last cancellation we actually replay the changes
1671 * made at that point.
1672 */
1673STATIC int
1674xlog_check_buffer_cancelled(
d5689eaa 1675 struct log *log,
1da177e4
LT
1676 xfs_daddr_t blkno,
1677 uint len,
1678 ushort flags)
1679{
d5689eaa
CH
1680 struct list_head *bucket;
1681 struct xfs_buf_cancel *bcp;
1da177e4
LT
1682
1683 if (log->l_buf_cancel_table == NULL) {
1684 /*
1685 * There is nothing in the table built in pass one,
1686 * so this buffer must not be cancelled.
1687 */
c1155410 1688 ASSERT(!(flags & XFS_BLF_CANCEL));
1da177e4
LT
1689 return 0;
1690 }
1691
1da177e4 1692 /*
d5689eaa 1693 * Search for an entry in the cancel table that matches our buffer.
1da177e4 1694 */
d5689eaa
CH
1695 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1696 list_for_each_entry(bcp, bucket, bc_list) {
1697 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
1698 goto found;
1da177e4 1699 }
d5689eaa 1700
1da177e4 1701 /*
d5689eaa
CH
1702 * We didn't find a corresponding entry in the table, so return 0 so
1703 * that the buffer is NOT cancelled.
1da177e4 1704 */
c1155410 1705 ASSERT(!(flags & XFS_BLF_CANCEL));
1da177e4 1706 return 0;
d5689eaa
CH
1707
1708found:
1709 /*
1710 * We've go a match, so return 1 so that the recovery of this buffer
1711 * is cancelled. If this buffer is actually a buffer cancel log
1712 * item, then decrement the refcount on the one in the table and
1713 * remove it if this is the last reference.
1714 */
1715 if (flags & XFS_BLF_CANCEL) {
1716 if (--bcp->bc_refcount == 0) {
1717 list_del(&bcp->bc_list);
1718 kmem_free(bcp);
1719 }
1720 }
1721 return 1;
1da177e4
LT
1722}
1723
1da177e4 1724/*
e2714bf8
CH
1725 * Perform recovery for a buffer full of inodes. In these buffers, the only
1726 * data which should be recovered is that which corresponds to the
1727 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1728 * data for the inodes is always logged through the inodes themselves rather
1729 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1da177e4 1730 *
e2714bf8
CH
1731 * The only time when buffers full of inodes are fully recovered is when the
1732 * buffer is full of newly allocated inodes. In this case the buffer will
1733 * not be marked as an inode buffer and so will be sent to
1734 * xlog_recover_do_reg_buffer() below during recovery.
1da177e4
LT
1735 */
1736STATIC int
1737xlog_recover_do_inode_buffer(
e2714bf8 1738 struct xfs_mount *mp,
1da177e4 1739 xlog_recover_item_t *item,
e2714bf8 1740 struct xfs_buf *bp,
1da177e4
LT
1741 xfs_buf_log_format_t *buf_f)
1742{
1743 int i;
e2714bf8
CH
1744 int item_index = 0;
1745 int bit = 0;
1746 int nbits = 0;
1747 int reg_buf_offset = 0;
1748 int reg_buf_bytes = 0;
1da177e4
LT
1749 int next_unlinked_offset;
1750 int inodes_per_buf;
1751 xfs_agino_t *logged_nextp;
1752 xfs_agino_t *buffer_nextp;
1da177e4 1753
9abbc539
DC
1754 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
1755
1da177e4
LT
1756 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1757 for (i = 0; i < inodes_per_buf; i++) {
1758 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1759 offsetof(xfs_dinode_t, di_next_unlinked);
1760
1761 while (next_unlinked_offset >=
1762 (reg_buf_offset + reg_buf_bytes)) {
1763 /*
1764 * The next di_next_unlinked field is beyond
1765 * the current logged region. Find the next
1766 * logged region that contains or is beyond
1767 * the current di_next_unlinked field.
1768 */
1769 bit += nbits;
e2714bf8
CH
1770 bit = xfs_next_bit(buf_f->blf_data_map,
1771 buf_f->blf_map_size, bit);
1da177e4
LT
1772
1773 /*
1774 * If there are no more logged regions in the
1775 * buffer, then we're done.
1776 */
e2714bf8 1777 if (bit == -1)
1da177e4 1778 return 0;
1da177e4 1779
e2714bf8
CH
1780 nbits = xfs_contig_bits(buf_f->blf_data_map,
1781 buf_f->blf_map_size, bit);
1da177e4 1782 ASSERT(nbits > 0);
c1155410
DC
1783 reg_buf_offset = bit << XFS_BLF_SHIFT;
1784 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
1da177e4
LT
1785 item_index++;
1786 }
1787
1788 /*
1789 * If the current logged region starts after the current
1790 * di_next_unlinked field, then move on to the next
1791 * di_next_unlinked field.
1792 */
e2714bf8 1793 if (next_unlinked_offset < reg_buf_offset)
1da177e4 1794 continue;
1da177e4
LT
1795
1796 ASSERT(item->ri_buf[item_index].i_addr != NULL);
c1155410 1797 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
1da177e4
LT
1798 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1799
1800 /*
1801 * The current logged region contains a copy of the
1802 * current di_next_unlinked field. Extract its value
1803 * and copy it to the buffer copy.
1804 */
4e0d5f92
CH
1805 logged_nextp = item->ri_buf[item_index].i_addr +
1806 next_unlinked_offset - reg_buf_offset;
1da177e4 1807 if (unlikely(*logged_nextp == 0)) {
a0fa2b67
DC
1808 xfs_alert(mp,
1809 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1810 "Trying to replay bad (0) inode di_next_unlinked field.",
1da177e4
LT
1811 item, bp);
1812 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1813 XFS_ERRLEVEL_LOW, mp);
1814 return XFS_ERROR(EFSCORRUPTED);
1815 }
1816
1817 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1818 next_unlinked_offset);
87c199c2 1819 *buffer_nextp = *logged_nextp;
1da177e4
LT
1820 }
1821
1822 return 0;
1823}
1824
1825/*
1826 * Perform a 'normal' buffer recovery. Each logged region of the
1827 * buffer should be copied over the corresponding region in the
1828 * given buffer. The bitmap in the buf log format structure indicates
1829 * where to place the logged data.
1830 */
1da177e4
LT
1831STATIC void
1832xlog_recover_do_reg_buffer(
9abbc539 1833 struct xfs_mount *mp,
1da177e4 1834 xlog_recover_item_t *item,
e2714bf8 1835 struct xfs_buf *bp,
1da177e4
LT
1836 xfs_buf_log_format_t *buf_f)
1837{
1838 int i;
1839 int bit;
1840 int nbits;
1da177e4
LT
1841 int error;
1842
9abbc539
DC
1843 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
1844
1da177e4
LT
1845 bit = 0;
1846 i = 1; /* 0 is the buf format structure */
1847 while (1) {
e2714bf8
CH
1848 bit = xfs_next_bit(buf_f->blf_data_map,
1849 buf_f->blf_map_size, bit);
1da177e4
LT
1850 if (bit == -1)
1851 break;
e2714bf8
CH
1852 nbits = xfs_contig_bits(buf_f->blf_data_map,
1853 buf_f->blf_map_size, bit);
1da177e4 1854 ASSERT(nbits > 0);
4b80916b 1855 ASSERT(item->ri_buf[i].i_addr != NULL);
c1155410 1856 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
1da177e4 1857 ASSERT(XFS_BUF_COUNT(bp) >=
c1155410 1858 ((uint)bit << XFS_BLF_SHIFT)+(nbits<<XFS_BLF_SHIFT));
1da177e4
LT
1859
1860 /*
1861 * Do a sanity check if this is a dquot buffer. Just checking
1862 * the first dquot in the buffer should do. XXXThis is
1863 * probably a good thing to do for other buf types also.
1864 */
1865 error = 0;
c8ad20ff 1866 if (buf_f->blf_flags &
c1155410 1867 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
0c5e1ce8 1868 if (item->ri_buf[i].i_addr == NULL) {
a0fa2b67 1869 xfs_alert(mp,
0c5e1ce8
CH
1870 "XFS: NULL dquot in %s.", __func__);
1871 goto next;
1872 }
8ec6dba2 1873 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
a0fa2b67 1874 xfs_alert(mp,
0c5e1ce8
CH
1875 "XFS: dquot too small (%d) in %s.",
1876 item->ri_buf[i].i_len, __func__);
1877 goto next;
1878 }
a0fa2b67 1879 error = xfs_qm_dqcheck(mp, item->ri_buf[i].i_addr,
1da177e4
LT
1880 -1, 0, XFS_QMOPT_DOWARN,
1881 "dquot_buf_recover");
0c5e1ce8
CH
1882 if (error)
1883 goto next;
1da177e4 1884 }
0c5e1ce8
CH
1885
1886 memcpy(xfs_buf_offset(bp,
c1155410 1887 (uint)bit << XFS_BLF_SHIFT), /* dest */
0c5e1ce8 1888 item->ri_buf[i].i_addr, /* source */
c1155410 1889 nbits<<XFS_BLF_SHIFT); /* length */
0c5e1ce8 1890 next:
1da177e4
LT
1891 i++;
1892 bit += nbits;
1893 }
1894
1895 /* Shouldn't be any more regions */
1896 ASSERT(i == item->ri_total);
1897}
1898
1899/*
1900 * Do some primitive error checking on ondisk dquot data structures.
1901 */
1902int
1903xfs_qm_dqcheck(
a0fa2b67 1904 struct xfs_mount *mp,
1da177e4
LT
1905 xfs_disk_dquot_t *ddq,
1906 xfs_dqid_t id,
1907 uint type, /* used only when IO_dorepair is true */
1908 uint flags,
1909 char *str)
1910{
1911 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
1912 int errs = 0;
1913
1914 /*
1915 * We can encounter an uninitialized dquot buffer for 2 reasons:
1916 * 1. If we crash while deleting the quotainode(s), and those blks got
1917 * used for user data. This is because we take the path of regular
1918 * file deletion; however, the size field of quotainodes is never
1919 * updated, so all the tricks that we play in itruncate_finish
1920 * don't quite matter.
1921 *
1922 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1923 * But the allocation will be replayed so we'll end up with an
1924 * uninitialized quota block.
1925 *
1926 * This is all fine; things are still consistent, and we haven't lost
1927 * any quota information. Just don't complain about bad dquot blks.
1928 */
1149d96a 1929 if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) {
1da177e4 1930 if (flags & XFS_QMOPT_DOWARN)
a0fa2b67 1931 xfs_alert(mp,
1da177e4 1932 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
1149d96a 1933 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
1da177e4
LT
1934 errs++;
1935 }
1149d96a 1936 if (ddq->d_version != XFS_DQUOT_VERSION) {
1da177e4 1937 if (flags & XFS_QMOPT_DOWARN)
a0fa2b67 1938 xfs_alert(mp,
1da177e4 1939 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
1149d96a 1940 str, id, ddq->d_version, XFS_DQUOT_VERSION);
1da177e4
LT
1941 errs++;
1942 }
1943
1149d96a
CH
1944 if (ddq->d_flags != XFS_DQ_USER &&
1945 ddq->d_flags != XFS_DQ_PROJ &&
1946 ddq->d_flags != XFS_DQ_GROUP) {
1da177e4 1947 if (flags & XFS_QMOPT_DOWARN)
a0fa2b67 1948 xfs_alert(mp,
1da177e4 1949 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
1149d96a 1950 str, id, ddq->d_flags);
1da177e4
LT
1951 errs++;
1952 }
1953
1149d96a 1954 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
1da177e4 1955 if (flags & XFS_QMOPT_DOWARN)
a0fa2b67 1956 xfs_alert(mp,
1da177e4
LT
1957 "%s : ondisk-dquot 0x%p, ID mismatch: "
1958 "0x%x expected, found id 0x%x",
1149d96a 1959 str, ddq, id, be32_to_cpu(ddq->d_id));
1da177e4
LT
1960 errs++;
1961 }
1962
1963 if (!errs && ddq->d_id) {
1149d96a
CH
1964 if (ddq->d_blk_softlimit &&
1965 be64_to_cpu(ddq->d_bcount) >=
1966 be64_to_cpu(ddq->d_blk_softlimit)) {
1da177e4
LT
1967 if (!ddq->d_btimer) {
1968 if (flags & XFS_QMOPT_DOWARN)
a0fa2b67
DC
1969 xfs_alert(mp,
1970 "%s : Dquot ID 0x%x (0x%p) BLK TIMER NOT STARTED",
1149d96a 1971 str, (int)be32_to_cpu(ddq->d_id), ddq);
1da177e4
LT
1972 errs++;
1973 }
1974 }
1149d96a
CH
1975 if (ddq->d_ino_softlimit &&
1976 be64_to_cpu(ddq->d_icount) >=
1977 be64_to_cpu(ddq->d_ino_softlimit)) {
1da177e4
LT
1978 if (!ddq->d_itimer) {
1979 if (flags & XFS_QMOPT_DOWARN)
a0fa2b67
DC
1980 xfs_alert(mp,
1981 "%s : Dquot ID 0x%x (0x%p) INODE TIMER NOT STARTED",
1149d96a 1982 str, (int)be32_to_cpu(ddq->d_id), ddq);
1da177e4
LT
1983 errs++;
1984 }
1985 }
1149d96a
CH
1986 if (ddq->d_rtb_softlimit &&
1987 be64_to_cpu(ddq->d_rtbcount) >=
1988 be64_to_cpu(ddq->d_rtb_softlimit)) {
1da177e4
LT
1989 if (!ddq->d_rtbtimer) {
1990 if (flags & XFS_QMOPT_DOWARN)
a0fa2b67
DC
1991 xfs_alert(mp,
1992 "%s : Dquot ID 0x%x (0x%p) RTBLK TIMER NOT STARTED",
1149d96a 1993 str, (int)be32_to_cpu(ddq->d_id), ddq);
1da177e4
LT
1994 errs++;
1995 }
1996 }
1997 }
1998
1999 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2000 return errs;
2001
2002 if (flags & XFS_QMOPT_DOWARN)
a0fa2b67 2003 xfs_notice(mp, "Re-initializing dquot ID 0x%x", id);
1da177e4
LT
2004
2005 /*
2006 * Typically, a repair is only requested by quotacheck.
2007 */
2008 ASSERT(id != -1);
2009 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2010 memset(d, 0, sizeof(xfs_dqblk_t));
1149d96a
CH
2011
2012 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2013 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2014 d->dd_diskdq.d_flags = type;
2015 d->dd_diskdq.d_id = cpu_to_be32(id);
1da177e4
LT
2016
2017 return errs;
2018}
2019
2020/*
2021 * Perform a dquot buffer recovery.
2022 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2023 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2024 * Else, treat it as a regular buffer and do recovery.
2025 */
2026STATIC void
2027xlog_recover_do_dquot_buffer(
2028 xfs_mount_t *mp,
2029 xlog_t *log,
2030 xlog_recover_item_t *item,
2031 xfs_buf_t *bp,
2032 xfs_buf_log_format_t *buf_f)
2033{
2034 uint type;
2035
9abbc539
DC
2036 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2037
1da177e4
LT
2038 /*
2039 * Filesystems are required to send in quota flags at mount time.
2040 */
2041 if (mp->m_qflags == 0) {
2042 return;
2043 }
2044
2045 type = 0;
c1155410 2046 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
1da177e4 2047 type |= XFS_DQ_USER;
c1155410 2048 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
c8ad20ff 2049 type |= XFS_DQ_PROJ;
c1155410 2050 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
1da177e4
LT
2051 type |= XFS_DQ_GROUP;
2052 /*
2053 * This type of quotas was turned off, so ignore this buffer
2054 */
2055 if (log->l_quotaoffs_flag & type)
2056 return;
2057
9abbc539 2058 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
1da177e4
LT
2059}
2060
2061/*
2062 * This routine replays a modification made to a buffer at runtime.
2063 * There are actually two types of buffer, regular and inode, which
2064 * are handled differently. Inode buffers are handled differently
2065 * in that we only recover a specific set of data from them, namely
2066 * the inode di_next_unlinked fields. This is because all other inode
2067 * data is actually logged via inode records and any data we replay
2068 * here which overlaps that may be stale.
2069 *
2070 * When meta-data buffers are freed at run time we log a buffer item
c1155410 2071 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
1da177e4
LT
2072 * of the buffer in the log should not be replayed at recovery time.
2073 * This is so that if the blocks covered by the buffer are reused for
2074 * file data before we crash we don't end up replaying old, freed
2075 * meta-data into a user's file.
2076 *
2077 * To handle the cancellation of buffer log items, we make two passes
2078 * over the log during recovery. During the first we build a table of
2079 * those buffers which have been cancelled, and during the second we
2080 * only replay those buffers which do not have corresponding cancel
2081 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2082 * for more details on the implementation of the table of cancel records.
2083 */
2084STATIC int
c9f71f5f 2085xlog_recover_buffer_pass2(
1da177e4 2086 xlog_t *log,
c9f71f5f 2087 xlog_recover_item_t *item)
1da177e4 2088{
4e0d5f92 2089 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
e2714bf8 2090 xfs_mount_t *mp = log->l_mp;
1da177e4
LT
2091 xfs_buf_t *bp;
2092 int error;
6ad112bf 2093 uint buf_flags;
1da177e4 2094
c9f71f5f
CH
2095 /*
2096 * In this pass we only want to recover all the buffers which have
2097 * not been cancelled and are not cancellation buffers themselves.
2098 */
2099 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2100 buf_f->blf_len, buf_f->blf_flags)) {
2101 trace_xfs_log_recover_buf_cancel(log, buf_f);
1da177e4 2102 return 0;
1da177e4 2103 }
c9f71f5f 2104
9abbc539 2105 trace_xfs_log_recover_buf_recover(log, buf_f);
1da177e4 2106
0cadda1c 2107 buf_flags = XBF_LOCK;
e2714bf8 2108 if (!(buf_f->blf_flags & XFS_BLF_INODE_BUF))
0cadda1c 2109 buf_flags |= XBF_MAPPED;
6ad112bf 2110
e2714bf8
CH
2111 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2112 buf_flags);
1da177e4 2113 if (XFS_BUF_ISERROR(bp)) {
e2714bf8
CH
2114 xfs_ioerror_alert("xlog_recover_do..(read#1)", mp,
2115 bp, buf_f->blf_blkno);
1da177e4
LT
2116 error = XFS_BUF_GETERROR(bp);
2117 xfs_buf_relse(bp);
2118 return error;
2119 }
2120
2121 error = 0;
e2714bf8 2122 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1da177e4 2123 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
e2714bf8 2124 } else if (buf_f->blf_flags &
c1155410 2125 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
1da177e4
LT
2126 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2127 } else {
9abbc539 2128 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
1da177e4
LT
2129 }
2130 if (error)
2131 return XFS_ERROR(error);
2132
2133 /*
2134 * Perform delayed write on the buffer. Asynchronous writes will be
2135 * slower when taking into account all the buffers to be flushed.
2136 *
2137 * Also make sure that only inode buffers with good sizes stay in
2138 * the buffer cache. The kernel moves inodes in buffers of 1 block
2139 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2140 * buffers in the log can be a different size if the log was generated
2141 * by an older kernel using unclustered inode buffers or a newer kernel
2142 * running with a different inode cluster size. Regardless, if the
2143 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2144 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2145 * the buffer out of the buffer cache so that the buffer won't
2146 * overlap with future reads of those inodes.
2147 */
2148 if (XFS_DINODE_MAGIC ==
b53e675d 2149 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
1da177e4
LT
2150 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2151 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2152 XFS_BUF_STALE(bp);
2153 error = xfs_bwrite(mp, bp);
2154 } else {
ebad861b 2155 ASSERT(bp->b_target->bt_mount == mp);
1da177e4
LT
2156 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2157 xfs_bdwrite(mp, bp);
2158 }
2159
2160 return (error);
2161}
2162
2163STATIC int
c9f71f5f 2164xlog_recover_inode_pass2(
1da177e4 2165 xlog_t *log,
c9f71f5f 2166 xlog_recover_item_t *item)
1da177e4
LT
2167{
2168 xfs_inode_log_format_t *in_f;
c9f71f5f 2169 xfs_mount_t *mp = log->l_mp;
1da177e4 2170 xfs_buf_t *bp;
1da177e4 2171 xfs_dinode_t *dip;
1da177e4
LT
2172 int len;
2173 xfs_caddr_t src;
2174 xfs_caddr_t dest;
2175 int error;
2176 int attr_index;
2177 uint fields;
347d1c01 2178 xfs_icdinode_t *dicp;
6d192a9b 2179 int need_free = 0;
1da177e4 2180
6d192a9b 2181 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
4e0d5f92 2182 in_f = item->ri_buf[0].i_addr;
6d192a9b 2183 } else {
4e0d5f92 2184 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
6d192a9b
TS
2185 need_free = 1;
2186 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2187 if (error)
2188 goto error;
2189 }
1da177e4
LT
2190
2191 /*
2192 * Inode buffers can be freed, look out for it,
2193 * and do not replay the inode.
2194 */
a1941895
CH
2195 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2196 in_f->ilf_len, 0)) {
6d192a9b 2197 error = 0;
9abbc539 2198 trace_xfs_log_recover_inode_cancel(log, in_f);
6d192a9b
TS
2199 goto error;
2200 }
9abbc539 2201 trace_xfs_log_recover_inode_recover(log, in_f);
1da177e4 2202
6ad112bf 2203 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len,
0cadda1c 2204 XBF_LOCK);
1da177e4
LT
2205 if (XFS_BUF_ISERROR(bp)) {
2206 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
a1941895 2207 bp, in_f->ilf_blkno);
1da177e4
LT
2208 error = XFS_BUF_GETERROR(bp);
2209 xfs_buf_relse(bp);
6d192a9b 2210 goto error;
1da177e4
LT
2211 }
2212 error = 0;
2213 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
a1941895 2214 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
1da177e4
LT
2215
2216 /*
2217 * Make sure the place we're flushing out to really looks
2218 * like an inode!
2219 */
81591fe2 2220 if (unlikely(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC)) {
1da177e4 2221 xfs_buf_relse(bp);
a0fa2b67
DC
2222 xfs_alert(mp,
2223 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2224 __func__, dip, bp, in_f->ilf_ino);
c9f71f5f 2225 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
1da177e4 2226 XFS_ERRLEVEL_LOW, mp);
6d192a9b
TS
2227 error = EFSCORRUPTED;
2228 goto error;
1da177e4 2229 }
4e0d5f92 2230 dicp = item->ri_buf[1].i_addr;
1da177e4
LT
2231 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2232 xfs_buf_relse(bp);
a0fa2b67
DC
2233 xfs_alert(mp,
2234 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2235 __func__, item, in_f->ilf_ino);
c9f71f5f 2236 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
1da177e4 2237 XFS_ERRLEVEL_LOW, mp);
6d192a9b
TS
2238 error = EFSCORRUPTED;
2239 goto error;
1da177e4
LT
2240 }
2241
2242 /* Skip replay when the on disk inode is newer than the log one */
81591fe2 2243 if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
1da177e4
LT
2244 /*
2245 * Deal with the wrap case, DI_MAX_FLUSH is less
2246 * than smaller numbers
2247 */
81591fe2 2248 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
347d1c01 2249 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
1da177e4
LT
2250 /* do nothing */
2251 } else {
2252 xfs_buf_relse(bp);
9abbc539 2253 trace_xfs_log_recover_inode_skip(log, in_f);
6d192a9b
TS
2254 error = 0;
2255 goto error;
1da177e4
LT
2256 }
2257 }
2258 /* Take the opportunity to reset the flush iteration count */
2259 dicp->di_flushiter = 0;
2260
2261 if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) {
2262 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2263 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
c9f71f5f 2264 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
1da177e4
LT
2265 XFS_ERRLEVEL_LOW, mp, dicp);
2266 xfs_buf_relse(bp);
a0fa2b67
DC
2267 xfs_alert(mp,
2268 "%s: Bad regular inode log record, rec ptr 0x%p, "
2269 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2270 __func__, item, dip, bp, in_f->ilf_ino);
6d192a9b
TS
2271 error = EFSCORRUPTED;
2272 goto error;
1da177e4
LT
2273 }
2274 } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) {
2275 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2276 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2277 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
c9f71f5f 2278 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
1da177e4
LT
2279 XFS_ERRLEVEL_LOW, mp, dicp);
2280 xfs_buf_relse(bp);
a0fa2b67
DC
2281 xfs_alert(mp,
2282 "%s: Bad dir inode log record, rec ptr 0x%p, "
2283 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2284 __func__, item, dip, bp, in_f->ilf_ino);
6d192a9b
TS
2285 error = EFSCORRUPTED;
2286 goto error;
1da177e4
LT
2287 }
2288 }
2289 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
c9f71f5f 2290 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
1da177e4
LT
2291 XFS_ERRLEVEL_LOW, mp, dicp);
2292 xfs_buf_relse(bp);
a0fa2b67
DC
2293 xfs_alert(mp,
2294 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2295 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2296 __func__, item, dip, bp, in_f->ilf_ino,
1da177e4
LT
2297 dicp->di_nextents + dicp->di_anextents,
2298 dicp->di_nblocks);
6d192a9b
TS
2299 error = EFSCORRUPTED;
2300 goto error;
1da177e4
LT
2301 }
2302 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
c9f71f5f 2303 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
1da177e4
LT
2304 XFS_ERRLEVEL_LOW, mp, dicp);
2305 xfs_buf_relse(bp);
a0fa2b67
DC
2306 xfs_alert(mp,
2307 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2308 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
c9f71f5f 2309 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
6d192a9b
TS
2310 error = EFSCORRUPTED;
2311 goto error;
1da177e4 2312 }
81591fe2 2313 if (unlikely(item->ri_buf[1].i_len > sizeof(struct xfs_icdinode))) {
c9f71f5f 2314 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
1da177e4
LT
2315 XFS_ERRLEVEL_LOW, mp, dicp);
2316 xfs_buf_relse(bp);
a0fa2b67
DC
2317 xfs_alert(mp,
2318 "%s: Bad inode log record length %d, rec ptr 0x%p",
2319 __func__, item->ri_buf[1].i_len, item);
6d192a9b
TS
2320 error = EFSCORRUPTED;
2321 goto error;
1da177e4
LT
2322 }
2323
2324 /* The core is in in-core format */
4e0d5f92 2325 xfs_dinode_to_disk(dip, item->ri_buf[1].i_addr);
1da177e4
LT
2326
2327 /* the rest is in on-disk format */
81591fe2
CH
2328 if (item->ri_buf[1].i_len > sizeof(struct xfs_icdinode)) {
2329 memcpy((xfs_caddr_t) dip + sizeof(struct xfs_icdinode),
2330 item->ri_buf[1].i_addr + sizeof(struct xfs_icdinode),
2331 item->ri_buf[1].i_len - sizeof(struct xfs_icdinode));
1da177e4
LT
2332 }
2333
2334 fields = in_f->ilf_fields;
2335 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2336 case XFS_ILOG_DEV:
81591fe2 2337 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
1da177e4
LT
2338 break;
2339 case XFS_ILOG_UUID:
81591fe2
CH
2340 memcpy(XFS_DFORK_DPTR(dip),
2341 &in_f->ilf_u.ilfu_uuid,
2342 sizeof(uuid_t));
1da177e4
LT
2343 break;
2344 }
2345
2346 if (in_f->ilf_size == 2)
2347 goto write_inode_buffer;
2348 len = item->ri_buf[2].i_len;
2349 src = item->ri_buf[2].i_addr;
2350 ASSERT(in_f->ilf_size <= 4);
2351 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2352 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2353 (len == in_f->ilf_dsize));
2354
2355 switch (fields & XFS_ILOG_DFORK) {
2356 case XFS_ILOG_DDATA:
2357 case XFS_ILOG_DEXT:
81591fe2 2358 memcpy(XFS_DFORK_DPTR(dip), src, len);
1da177e4
LT
2359 break;
2360
2361 case XFS_ILOG_DBROOT:
7cc95a82 2362 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
81591fe2 2363 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
1da177e4
LT
2364 XFS_DFORK_DSIZE(dip, mp));
2365 break;
2366
2367 default:
2368 /*
2369 * There are no data fork flags set.
2370 */
2371 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2372 break;
2373 }
2374
2375 /*
2376 * If we logged any attribute data, recover it. There may or
2377 * may not have been any other non-core data logged in this
2378 * transaction.
2379 */
2380 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2381 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2382 attr_index = 3;
2383 } else {
2384 attr_index = 2;
2385 }
2386 len = item->ri_buf[attr_index].i_len;
2387 src = item->ri_buf[attr_index].i_addr;
2388 ASSERT(len == in_f->ilf_asize);
2389
2390 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2391 case XFS_ILOG_ADATA:
2392 case XFS_ILOG_AEXT:
2393 dest = XFS_DFORK_APTR(dip);
2394 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2395 memcpy(dest, src, len);
2396 break;
2397
2398 case XFS_ILOG_ABROOT:
2399 dest = XFS_DFORK_APTR(dip);
7cc95a82
CH
2400 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2401 len, (xfs_bmdr_block_t*)dest,
1da177e4
LT
2402 XFS_DFORK_ASIZE(dip, mp));
2403 break;
2404
2405 default:
a0fa2b67 2406 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
1da177e4
LT
2407 ASSERT(0);
2408 xfs_buf_relse(bp);
6d192a9b
TS
2409 error = EIO;
2410 goto error;
1da177e4
LT
2411 }
2412 }
2413
2414write_inode_buffer:
ebad861b 2415 ASSERT(bp->b_target->bt_mount == mp);
dd0bbad8
CH
2416 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2417 xfs_bdwrite(mp, bp);
6d192a9b
TS
2418error:
2419 if (need_free)
f0e2d93c 2420 kmem_free(in_f);
6d192a9b 2421 return XFS_ERROR(error);
1da177e4
LT
2422}
2423
2424/*
2425 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2426 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2427 * of that type.
2428 */
2429STATIC int
c9f71f5f 2430xlog_recover_quotaoff_pass1(
1da177e4 2431 xlog_t *log,
c9f71f5f 2432 xlog_recover_item_t *item)
1da177e4 2433{
c9f71f5f 2434 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
1da177e4
LT
2435 ASSERT(qoff_f);
2436
2437 /*
2438 * The logitem format's flag tells us if this was user quotaoff,
77a7cce4 2439 * group/project quotaoff or both.
1da177e4
LT
2440 */
2441 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2442 log->l_quotaoffs_flag |= XFS_DQ_USER;
77a7cce4
NS
2443 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2444 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
1da177e4
LT
2445 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2446 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2447
2448 return (0);
2449}
2450
2451/*
2452 * Recover a dquot record
2453 */
2454STATIC int
c9f71f5f 2455xlog_recover_dquot_pass2(
1da177e4 2456 xlog_t *log,
c9f71f5f 2457 xlog_recover_item_t *item)
1da177e4 2458{
c9f71f5f 2459 xfs_mount_t *mp = log->l_mp;
1da177e4
LT
2460 xfs_buf_t *bp;
2461 struct xfs_disk_dquot *ddq, *recddq;
2462 int error;
2463 xfs_dq_logformat_t *dq_f;
2464 uint type;
2465
1da177e4
LT
2466
2467 /*
2468 * Filesystems are required to send in quota flags at mount time.
2469 */
2470 if (mp->m_qflags == 0)
2471 return (0);
2472
4e0d5f92
CH
2473 recddq = item->ri_buf[1].i_addr;
2474 if (recddq == NULL) {
a0fa2b67 2475 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
0c5e1ce8
CH
2476 return XFS_ERROR(EIO);
2477 }
8ec6dba2 2478 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
a0fa2b67 2479 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
0c5e1ce8
CH
2480 item->ri_buf[1].i_len, __func__);
2481 return XFS_ERROR(EIO);
2482 }
2483
1da177e4
LT
2484 /*
2485 * This type of quotas was turned off, so ignore this record.
2486 */
b53e675d 2487 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
1da177e4
LT
2488 ASSERT(type);
2489 if (log->l_quotaoffs_flag & type)
2490 return (0);
2491
2492 /*
2493 * At this point we know that quota was _not_ turned off.
2494 * Since the mount flags are not indicating to us otherwise, this
2495 * must mean that quota is on, and the dquot needs to be replayed.
2496 * Remember that we may not have fully recovered the superblock yet,
2497 * so we can't do the usual trick of looking at the SB quota bits.
2498 *
2499 * The other possibility, of course, is that the quota subsystem was
2500 * removed since the last mount - ENOSYS.
2501 */
4e0d5f92 2502 dq_f = item->ri_buf[0].i_addr;
1da177e4 2503 ASSERT(dq_f);
a0fa2b67
DC
2504 error = xfs_qm_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2505 "xlog_recover_dquot_pass2 (log copy)");
2506 if (error)
1da177e4 2507 return XFS_ERROR(EIO);
1da177e4
LT
2508 ASSERT(dq_f->qlf_len == 1);
2509
2510 error = xfs_read_buf(mp, mp->m_ddev_targp,
2511 dq_f->qlf_blkno,
2512 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2513 0, &bp);
2514 if (error) {
2515 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
2516 bp, dq_f->qlf_blkno);
2517 return error;
2518 }
2519 ASSERT(bp);
2520 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2521
2522 /*
2523 * At least the magic num portion should be on disk because this
2524 * was among a chunk of dquots created earlier, and we did some
2525 * minimal initialization then.
2526 */
a0fa2b67
DC
2527 error = xfs_qm_dqcheck(mp, ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2528 "xlog_recover_dquot_pass2");
2529 if (error) {
1da177e4
LT
2530 xfs_buf_relse(bp);
2531 return XFS_ERROR(EIO);
2532 }
2533
2534 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2535
2536 ASSERT(dq_f->qlf_size == 2);
ebad861b 2537 ASSERT(bp->b_target->bt_mount == mp);
1da177e4
LT
2538 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2539 xfs_bdwrite(mp, bp);
2540
2541 return (0);
2542}
2543
2544/*
2545 * This routine is called to create an in-core extent free intent
2546 * item from the efi format structure which was logged on disk.
2547 * It allocates an in-core efi, copies the extents from the format
2548 * structure into it, and adds the efi to the AIL with the given
2549 * LSN.
2550 */
6d192a9b 2551STATIC int
c9f71f5f 2552xlog_recover_efi_pass2(
1da177e4
LT
2553 xlog_t *log,
2554 xlog_recover_item_t *item,
c9f71f5f 2555 xfs_lsn_t lsn)
1da177e4 2556{
6d192a9b 2557 int error;
c9f71f5f 2558 xfs_mount_t *mp = log->l_mp;
1da177e4
LT
2559 xfs_efi_log_item_t *efip;
2560 xfs_efi_log_format_t *efi_formatp;
1da177e4 2561
4e0d5f92 2562 efi_formatp = item->ri_buf[0].i_addr;
1da177e4 2563
1da177e4 2564 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
6d192a9b
TS
2565 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2566 &(efip->efi_format)))) {
2567 xfs_efi_item_free(efip);
2568 return error;
2569 }
b199c8a4 2570 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
1da177e4 2571
a9c21c1b 2572 spin_lock(&log->l_ailp->xa_lock);
1da177e4 2573 /*
783a2f65 2574 * xfs_trans_ail_update() drops the AIL lock.
1da177e4 2575 */
e6059949 2576 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
6d192a9b 2577 return 0;
1da177e4
LT
2578}
2579
2580
2581/*
2582 * This routine is called when an efd format structure is found in
2583 * a committed transaction in the log. It's purpose is to cancel
2584 * the corresponding efi if it was still in the log. To do this
2585 * it searches the AIL for the efi with an id equal to that in the
2586 * efd format structure. If we find it, we remove the efi from the
2587 * AIL and free it.
2588 */
c9f71f5f
CH
2589STATIC int
2590xlog_recover_efd_pass2(
1da177e4 2591 xlog_t *log,
c9f71f5f 2592 xlog_recover_item_t *item)
1da177e4 2593{
1da177e4
LT
2594 xfs_efd_log_format_t *efd_formatp;
2595 xfs_efi_log_item_t *efip = NULL;
2596 xfs_log_item_t *lip;
1da177e4 2597 __uint64_t efi_id;
27d8d5fe 2598 struct xfs_ail_cursor cur;
783a2f65 2599 struct xfs_ail *ailp = log->l_ailp;
1da177e4 2600
4e0d5f92 2601 efd_formatp = item->ri_buf[0].i_addr;
6d192a9b
TS
2602 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2603 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2604 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2605 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
1da177e4
LT
2606 efi_id = efd_formatp->efd_efi_id;
2607
2608 /*
2609 * Search for the efi with the id in the efd format structure
2610 * in the AIL.
2611 */
a9c21c1b
DC
2612 spin_lock(&ailp->xa_lock);
2613 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
1da177e4
LT
2614 while (lip != NULL) {
2615 if (lip->li_type == XFS_LI_EFI) {
2616 efip = (xfs_efi_log_item_t *)lip;
2617 if (efip->efi_format.efi_id == efi_id) {
2618 /*
783a2f65 2619 * xfs_trans_ail_delete() drops the
1da177e4
LT
2620 * AIL lock.
2621 */
783a2f65 2622 xfs_trans_ail_delete(ailp, lip);
8ae2c0f6 2623 xfs_efi_item_free(efip);
a9c21c1b 2624 spin_lock(&ailp->xa_lock);
27d8d5fe 2625 break;
1da177e4
LT
2626 }
2627 }
a9c21c1b 2628 lip = xfs_trans_ail_cursor_next(ailp, &cur);
1da177e4 2629 }
a9c21c1b
DC
2630 xfs_trans_ail_cursor_done(ailp, &cur);
2631 spin_unlock(&ailp->xa_lock);
c9f71f5f
CH
2632
2633 return 0;
1da177e4
LT
2634}
2635
1da177e4
LT
2636/*
2637 * Free up any resources allocated by the transaction
2638 *
2639 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2640 */
2641STATIC void
2642xlog_recover_free_trans(
d0450948 2643 struct xlog_recover *trans)
1da177e4 2644{
f0a76953 2645 xlog_recover_item_t *item, *n;
1da177e4
LT
2646 int i;
2647
f0a76953
DC
2648 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
2649 /* Free the regions in the item. */
2650 list_del(&item->ri_list);
2651 for (i = 0; i < item->ri_cnt; i++)
2652 kmem_free(item->ri_buf[i].i_addr);
1da177e4 2653 /* Free the item itself */
f0a76953
DC
2654 kmem_free(item->ri_buf);
2655 kmem_free(item);
2656 }
1da177e4 2657 /* Free the transaction recover structure */
f0e2d93c 2658 kmem_free(trans);
1da177e4
LT
2659}
2660
d0450948 2661STATIC int
c9f71f5f 2662xlog_recover_commit_pass1(
d0450948
CH
2663 struct log *log,
2664 struct xlog_recover *trans,
c9f71f5f 2665 xlog_recover_item_t *item)
d0450948 2666{
c9f71f5f 2667 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
d0450948
CH
2668
2669 switch (ITEM_TYPE(item)) {
2670 case XFS_LI_BUF:
c9f71f5f
CH
2671 return xlog_recover_buffer_pass1(log, item);
2672 case XFS_LI_QUOTAOFF:
2673 return xlog_recover_quotaoff_pass1(log, item);
d0450948 2674 case XFS_LI_INODE:
d0450948 2675 case XFS_LI_EFI:
d0450948 2676 case XFS_LI_EFD:
c9f71f5f
CH
2677 case XFS_LI_DQUOT:
2678 /* nothing to do in pass 1 */
d0450948 2679 return 0;
c9f71f5f 2680 default:
a0fa2b67
DC
2681 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
2682 __func__, ITEM_TYPE(item));
c9f71f5f
CH
2683 ASSERT(0);
2684 return XFS_ERROR(EIO);
2685 }
2686}
2687
2688STATIC int
2689xlog_recover_commit_pass2(
2690 struct log *log,
2691 struct xlog_recover *trans,
2692 xlog_recover_item_t *item)
2693{
2694 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
2695
2696 switch (ITEM_TYPE(item)) {
2697 case XFS_LI_BUF:
2698 return xlog_recover_buffer_pass2(log, item);
2699 case XFS_LI_INODE:
2700 return xlog_recover_inode_pass2(log, item);
2701 case XFS_LI_EFI:
2702 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
2703 case XFS_LI_EFD:
2704 return xlog_recover_efd_pass2(log, item);
d0450948 2705 case XFS_LI_DQUOT:
c9f71f5f 2706 return xlog_recover_dquot_pass2(log, item);
d0450948 2707 case XFS_LI_QUOTAOFF:
c9f71f5f
CH
2708 /* nothing to do in pass2 */
2709 return 0;
d0450948 2710 default:
a0fa2b67
DC
2711 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
2712 __func__, ITEM_TYPE(item));
d0450948
CH
2713 ASSERT(0);
2714 return XFS_ERROR(EIO);
2715 }
2716}
2717
2718/*
2719 * Perform the transaction.
2720 *
2721 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2722 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2723 */
1da177e4
LT
2724STATIC int
2725xlog_recover_commit_trans(
d0450948
CH
2726 struct log *log,
2727 struct xlog_recover *trans,
1da177e4
LT
2728 int pass)
2729{
d0450948
CH
2730 int error = 0;
2731 xlog_recover_item_t *item;
1da177e4 2732
f0a76953 2733 hlist_del(&trans->r_list);
d0450948
CH
2734
2735 error = xlog_recover_reorder_trans(log, trans, pass);
2736 if (error)
1da177e4 2737 return error;
d0450948
CH
2738
2739 list_for_each_entry(item, &trans->r_itemq, ri_list) {
c9f71f5f
CH
2740 if (pass == XLOG_RECOVER_PASS1)
2741 error = xlog_recover_commit_pass1(log, trans, item);
2742 else
2743 error = xlog_recover_commit_pass2(log, trans, item);
d0450948
CH
2744 if (error)
2745 return error;
2746 }
2747
2748 xlog_recover_free_trans(trans);
1da177e4
LT
2749 return 0;
2750}
2751
2752STATIC int
2753xlog_recover_unmount_trans(
a0fa2b67 2754 struct log *log,
1da177e4
LT
2755 xlog_recover_t *trans)
2756{
2757 /* Do nothing now */
a0fa2b67 2758 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
1da177e4
LT
2759 return 0;
2760}
2761
2762/*
2763 * There are two valid states of the r_state field. 0 indicates that the
2764 * transaction structure is in a normal state. We have either seen the
2765 * start of the transaction or the last operation we added was not a partial
2766 * operation. If the last operation we added to the transaction was a
2767 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2768 *
2769 * NOTE: skip LRs with 0 data length.
2770 */
2771STATIC int
2772xlog_recover_process_data(
2773 xlog_t *log,
f0a76953 2774 struct hlist_head rhash[],
1da177e4
LT
2775 xlog_rec_header_t *rhead,
2776 xfs_caddr_t dp,
2777 int pass)
2778{
2779 xfs_caddr_t lp;
2780 int num_logops;
2781 xlog_op_header_t *ohead;
2782 xlog_recover_t *trans;
2783 xlog_tid_t tid;
2784 int error;
2785 unsigned long hash;
2786 uint flags;
2787
b53e675d
CH
2788 lp = dp + be32_to_cpu(rhead->h_len);
2789 num_logops = be32_to_cpu(rhead->h_num_logops);
1da177e4
LT
2790
2791 /* check the log format matches our own - else we can't recover */
2792 if (xlog_header_check_recover(log->l_mp, rhead))
2793 return (XFS_ERROR(EIO));
2794
2795 while ((dp < lp) && num_logops) {
2796 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2797 ohead = (xlog_op_header_t *)dp;
2798 dp += sizeof(xlog_op_header_t);
2799 if (ohead->oh_clientid != XFS_TRANSACTION &&
2800 ohead->oh_clientid != XFS_LOG) {
a0fa2b67
DC
2801 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
2802 __func__, ohead->oh_clientid);
1da177e4
LT
2803 ASSERT(0);
2804 return (XFS_ERROR(EIO));
2805 }
67fcb7bf 2806 tid = be32_to_cpu(ohead->oh_tid);
1da177e4 2807 hash = XLOG_RHASH(tid);
f0a76953 2808 trans = xlog_recover_find_tid(&rhash[hash], tid);
1da177e4
LT
2809 if (trans == NULL) { /* not found; add new tid */
2810 if (ohead->oh_flags & XLOG_START_TRANS)
2811 xlog_recover_new_tid(&rhash[hash], tid,
b53e675d 2812 be64_to_cpu(rhead->h_lsn));
1da177e4 2813 } else {
9742bb93 2814 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
a0fa2b67
DC
2815 xfs_warn(log->l_mp, "%s: bad length 0x%x",
2816 __func__, be32_to_cpu(ohead->oh_len));
9742bb93
LM
2817 WARN_ON(1);
2818 return (XFS_ERROR(EIO));
2819 }
1da177e4
LT
2820 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2821 if (flags & XLOG_WAS_CONT_TRANS)
2822 flags &= ~XLOG_CONTINUE_TRANS;
2823 switch (flags) {
2824 case XLOG_COMMIT_TRANS:
2825 error = xlog_recover_commit_trans(log,
f0a76953 2826 trans, pass);
1da177e4
LT
2827 break;
2828 case XLOG_UNMOUNT_TRANS:
a0fa2b67 2829 error = xlog_recover_unmount_trans(log, trans);
1da177e4
LT
2830 break;
2831 case XLOG_WAS_CONT_TRANS:
9abbc539
DC
2832 error = xlog_recover_add_to_cont_trans(log,
2833 trans, dp,
2834 be32_to_cpu(ohead->oh_len));
1da177e4
LT
2835 break;
2836 case XLOG_START_TRANS:
a0fa2b67
DC
2837 xfs_warn(log->l_mp, "%s: bad transaction",
2838 __func__);
1da177e4
LT
2839 ASSERT(0);
2840 error = XFS_ERROR(EIO);
2841 break;
2842 case 0:
2843 case XLOG_CONTINUE_TRANS:
9abbc539 2844 error = xlog_recover_add_to_trans(log, trans,
67fcb7bf 2845 dp, be32_to_cpu(ohead->oh_len));
1da177e4
LT
2846 break;
2847 default:
a0fa2b67
DC
2848 xfs_warn(log->l_mp, "%s: bad flag 0x%x",
2849 __func__, flags);
1da177e4
LT
2850 ASSERT(0);
2851 error = XFS_ERROR(EIO);
2852 break;
2853 }
2854 if (error)
2855 return error;
2856 }
67fcb7bf 2857 dp += be32_to_cpu(ohead->oh_len);
1da177e4
LT
2858 num_logops--;
2859 }
2860 return 0;
2861}
2862
2863/*
2864 * Process an extent free intent item that was recovered from
2865 * the log. We need to free the extents that it describes.
2866 */
3c1e2bbe 2867STATIC int
1da177e4
LT
2868xlog_recover_process_efi(
2869 xfs_mount_t *mp,
2870 xfs_efi_log_item_t *efip)
2871{
2872 xfs_efd_log_item_t *efdp;
2873 xfs_trans_t *tp;
2874 int i;
3c1e2bbe 2875 int error = 0;
1da177e4
LT
2876 xfs_extent_t *extp;
2877 xfs_fsblock_t startblock_fsb;
2878
b199c8a4 2879 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
1da177e4
LT
2880
2881 /*
2882 * First check the validity of the extents described by the
2883 * EFI. If any are bad, then assume that all are bad and
2884 * just toss the EFI.
2885 */
2886 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2887 extp = &(efip->efi_format.efi_extents[i]);
2888 startblock_fsb = XFS_BB_TO_FSB(mp,
2889 XFS_FSB_TO_DADDR(mp, extp->ext_start));
2890 if ((startblock_fsb == 0) ||
2891 (extp->ext_len == 0) ||
2892 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
2893 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
2894 /*
2895 * This will pull the EFI from the AIL and
2896 * free the memory associated with it.
2897 */
2898 xfs_efi_release(efip, efip->efi_format.efi_nextents);
3c1e2bbe 2899 return XFS_ERROR(EIO);
1da177e4
LT
2900 }
2901 }
2902
2903 tp = xfs_trans_alloc(mp, 0);
3c1e2bbe 2904 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
fc6149d8
DC
2905 if (error)
2906 goto abort_error;
1da177e4
LT
2907 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
2908
2909 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2910 extp = &(efip->efi_format.efi_extents[i]);
fc6149d8
DC
2911 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
2912 if (error)
2913 goto abort_error;
1da177e4
LT
2914 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
2915 extp->ext_len);
2916 }
2917
b199c8a4 2918 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
e5720eec 2919 error = xfs_trans_commit(tp, 0);
3c1e2bbe 2920 return error;
fc6149d8
DC
2921
2922abort_error:
2923 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
2924 return error;
1da177e4
LT
2925}
2926
1da177e4
LT
2927/*
2928 * When this is called, all of the EFIs which did not have
2929 * corresponding EFDs should be in the AIL. What we do now
2930 * is free the extents associated with each one.
2931 *
2932 * Since we process the EFIs in normal transactions, they
2933 * will be removed at some point after the commit. This prevents
2934 * us from just walking down the list processing each one.
2935 * We'll use a flag in the EFI to skip those that we've already
2936 * processed and use the AIL iteration mechanism's generation
2937 * count to try to speed this up at least a bit.
2938 *
2939 * When we start, we know that the EFIs are the only things in
2940 * the AIL. As we process them, however, other items are added
2941 * to the AIL. Since everything added to the AIL must come after
2942 * everything already in the AIL, we stop processing as soon as
2943 * we see something other than an EFI in the AIL.
2944 */
3c1e2bbe 2945STATIC int
1da177e4
LT
2946xlog_recover_process_efis(
2947 xlog_t *log)
2948{
2949 xfs_log_item_t *lip;
2950 xfs_efi_log_item_t *efip;
3c1e2bbe 2951 int error = 0;
27d8d5fe 2952 struct xfs_ail_cursor cur;
a9c21c1b 2953 struct xfs_ail *ailp;
1da177e4 2954
a9c21c1b
DC
2955 ailp = log->l_ailp;
2956 spin_lock(&ailp->xa_lock);
2957 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
1da177e4
LT
2958 while (lip != NULL) {
2959 /*
2960 * We're done when we see something other than an EFI.
27d8d5fe 2961 * There should be no EFIs left in the AIL now.
1da177e4
LT
2962 */
2963 if (lip->li_type != XFS_LI_EFI) {
27d8d5fe 2964#ifdef DEBUG
a9c21c1b 2965 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
27d8d5fe
DC
2966 ASSERT(lip->li_type != XFS_LI_EFI);
2967#endif
1da177e4
LT
2968 break;
2969 }
2970
2971 /*
2972 * Skip EFIs that we've already processed.
2973 */
2974 efip = (xfs_efi_log_item_t *)lip;
b199c8a4 2975 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
a9c21c1b 2976 lip = xfs_trans_ail_cursor_next(ailp, &cur);
1da177e4
LT
2977 continue;
2978 }
2979
a9c21c1b
DC
2980 spin_unlock(&ailp->xa_lock);
2981 error = xlog_recover_process_efi(log->l_mp, efip);
2982 spin_lock(&ailp->xa_lock);
27d8d5fe
DC
2983 if (error)
2984 goto out;
a9c21c1b 2985 lip = xfs_trans_ail_cursor_next(ailp, &cur);
1da177e4 2986 }
27d8d5fe 2987out:
a9c21c1b
DC
2988 xfs_trans_ail_cursor_done(ailp, &cur);
2989 spin_unlock(&ailp->xa_lock);
3c1e2bbe 2990 return error;
1da177e4
LT
2991}
2992
2993/*
2994 * This routine performs a transaction to null out a bad inode pointer
2995 * in an agi unlinked inode hash bucket.
2996 */
2997STATIC void
2998xlog_recover_clear_agi_bucket(
2999 xfs_mount_t *mp,
3000 xfs_agnumber_t agno,
3001 int bucket)
3002{
3003 xfs_trans_t *tp;
3004 xfs_agi_t *agi;
3005 xfs_buf_t *agibp;
3006 int offset;
3007 int error;
3008
3009 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
5e1be0fb
CH
3010 error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp),
3011 0, 0, 0);
e5720eec
DC
3012 if (error)
3013 goto out_abort;
1da177e4 3014
5e1be0fb
CH
3015 error = xfs_read_agi(mp, tp, agno, &agibp);
3016 if (error)
e5720eec 3017 goto out_abort;
1da177e4 3018
5e1be0fb 3019 agi = XFS_BUF_TO_AGI(agibp);
16259e7d 3020 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
1da177e4
LT
3021 offset = offsetof(xfs_agi_t, agi_unlinked) +
3022 (sizeof(xfs_agino_t) * bucket);
3023 xfs_trans_log_buf(tp, agibp, offset,
3024 (offset + sizeof(xfs_agino_t) - 1));
3025
e5720eec
DC
3026 error = xfs_trans_commit(tp, 0);
3027 if (error)
3028 goto out_error;
3029 return;
3030
3031out_abort:
3032 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3033out_error:
a0fa2b67 3034 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
e5720eec 3035 return;
1da177e4
LT
3036}
3037
23fac50f
CH
3038STATIC xfs_agino_t
3039xlog_recover_process_one_iunlink(
3040 struct xfs_mount *mp,
3041 xfs_agnumber_t agno,
3042 xfs_agino_t agino,
3043 int bucket)
3044{
3045 struct xfs_buf *ibp;
3046 struct xfs_dinode *dip;
3047 struct xfs_inode *ip;
3048 xfs_ino_t ino;
3049 int error;
3050
3051 ino = XFS_AGINO_TO_INO(mp, agno, agino);
7b6259e7 3052 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
23fac50f
CH
3053 if (error)
3054 goto fail;
3055
3056 /*
3057 * Get the on disk inode to find the next inode in the bucket.
3058 */
0cadda1c 3059 error = xfs_itobp(mp, NULL, ip, &dip, &ibp, XBF_LOCK);
23fac50f 3060 if (error)
0e446673 3061 goto fail_iput;
23fac50f 3062
23fac50f 3063 ASSERT(ip->i_d.di_nlink == 0);
0e446673 3064 ASSERT(ip->i_d.di_mode != 0);
23fac50f
CH
3065
3066 /* setup for the next pass */
3067 agino = be32_to_cpu(dip->di_next_unlinked);
3068 xfs_buf_relse(ibp);
3069
3070 /*
3071 * Prevent any DMAPI event from being sent when the reference on
3072 * the inode is dropped.
3073 */
3074 ip->i_d.di_dmevmask = 0;
3075
0e446673 3076 IRELE(ip);
23fac50f
CH
3077 return agino;
3078
0e446673
CH
3079 fail_iput:
3080 IRELE(ip);
23fac50f
CH
3081 fail:
3082 /*
3083 * We can't read in the inode this bucket points to, or this inode
3084 * is messed up. Just ditch this bucket of inodes. We will lose
3085 * some inodes and space, but at least we won't hang.
3086 *
3087 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3088 * clear the inode pointer in the bucket.
3089 */
3090 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3091 return NULLAGINO;
3092}
3093
1da177e4
LT
3094/*
3095 * xlog_iunlink_recover
3096 *
3097 * This is called during recovery to process any inodes which
3098 * we unlinked but not freed when the system crashed. These
3099 * inodes will be on the lists in the AGI blocks. What we do
3100 * here is scan all the AGIs and fully truncate and free any
3101 * inodes found on the lists. Each inode is removed from the
3102 * lists when it has been fully truncated and is freed. The
3103 * freeing of the inode and its removal from the list must be
3104 * atomic.
3105 */
d96f8f89 3106STATIC void
1da177e4
LT
3107xlog_recover_process_iunlinks(
3108 xlog_t *log)
3109{
3110 xfs_mount_t *mp;
3111 xfs_agnumber_t agno;
3112 xfs_agi_t *agi;
3113 xfs_buf_t *agibp;
1da177e4 3114 xfs_agino_t agino;
1da177e4
LT
3115 int bucket;
3116 int error;
3117 uint mp_dmevmask;
3118
3119 mp = log->l_mp;
3120
3121 /*
3122 * Prevent any DMAPI event from being sent while in this function.
3123 */
3124 mp_dmevmask = mp->m_dmevmask;
3125 mp->m_dmevmask = 0;
3126
3127 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3128 /*
3129 * Find the agi for this ag.
3130 */
5e1be0fb
CH
3131 error = xfs_read_agi(mp, NULL, agno, &agibp);
3132 if (error) {
3133 /*
3134 * AGI is b0rked. Don't process it.
3135 *
3136 * We should probably mark the filesystem as corrupt
3137 * after we've recovered all the ag's we can....
3138 */
3139 continue;
1da177e4
LT
3140 }
3141 agi = XFS_BUF_TO_AGI(agibp);
1da177e4
LT
3142
3143 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
16259e7d 3144 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
1da177e4 3145 while (agino != NULLAGINO) {
1da177e4
LT
3146 /*
3147 * Release the agi buffer so that it can
3148 * be acquired in the normal course of the
3149 * transaction to truncate and free the inode.
3150 */
3151 xfs_buf_relse(agibp);
3152
23fac50f
CH
3153 agino = xlog_recover_process_one_iunlink(mp,
3154 agno, agino, bucket);
1da177e4
LT
3155
3156 /*
3157 * Reacquire the agibuffer and continue around
5e1be0fb
CH
3158 * the loop. This should never fail as we know
3159 * the buffer was good earlier on.
1da177e4 3160 */
5e1be0fb
CH
3161 error = xfs_read_agi(mp, NULL, agno, &agibp);
3162 ASSERT(error == 0);
1da177e4 3163 agi = XFS_BUF_TO_AGI(agibp);
1da177e4
LT
3164 }
3165 }
3166
3167 /*
3168 * Release the buffer for the current agi so we can
3169 * go on to the next one.
3170 */
3171 xfs_buf_relse(agibp);
3172 }
3173
3174 mp->m_dmevmask = mp_dmevmask;
3175}
3176
3177
3178#ifdef DEBUG
3179STATIC void
3180xlog_pack_data_checksum(
3181 xlog_t *log,
3182 xlog_in_core_t *iclog,
3183 int size)
3184{
3185 int i;
b53e675d 3186 __be32 *up;
1da177e4
LT
3187 uint chksum = 0;
3188
b53e675d 3189 up = (__be32 *)iclog->ic_datap;
1da177e4
LT
3190 /* divide length by 4 to get # words */
3191 for (i = 0; i < (size >> 2); i++) {
b53e675d 3192 chksum ^= be32_to_cpu(*up);
1da177e4
LT
3193 up++;
3194 }
b53e675d 3195 iclog->ic_header.h_chksum = cpu_to_be32(chksum);
1da177e4
LT
3196}
3197#else
3198#define xlog_pack_data_checksum(log, iclog, size)
3199#endif
3200
3201/*
3202 * Stamp cycle number in every block
3203 */
3204void
3205xlog_pack_data(
3206 xlog_t *log,
3207 xlog_in_core_t *iclog,
3208 int roundoff)
3209{
3210 int i, j, k;
3211 int size = iclog->ic_offset + roundoff;
b53e675d 3212 __be32 cycle_lsn;
1da177e4 3213 xfs_caddr_t dp;
1da177e4
LT
3214
3215 xlog_pack_data_checksum(log, iclog, size);
3216
3217 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3218
3219 dp = iclog->ic_datap;
3220 for (i = 0; i < BTOBB(size) &&
3221 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
b53e675d
CH
3222 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
3223 *(__be32 *)dp = cycle_lsn;
1da177e4
LT
3224 dp += BBSIZE;
3225 }
3226
62118709 3227 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
b28708d6
CH
3228 xlog_in_core_2_t *xhdr = iclog->ic_data;
3229
1da177e4
LT
3230 for ( ; i < BTOBB(size); i++) {
3231 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3232 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
b53e675d
CH
3233 xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
3234 *(__be32 *)dp = cycle_lsn;
1da177e4
LT
3235 dp += BBSIZE;
3236 }
3237
3238 for (i = 1; i < log->l_iclog_heads; i++) {
3239 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3240 }
3241 }
3242}
3243
1da177e4
LT
3244STATIC void
3245xlog_unpack_data(
3246 xlog_rec_header_t *rhead,
3247 xfs_caddr_t dp,
3248 xlog_t *log)
3249{
3250 int i, j, k;
1da177e4 3251
b53e675d 3252 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
1da177e4 3253 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
b53e675d 3254 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
1da177e4
LT
3255 dp += BBSIZE;
3256 }
3257
62118709 3258 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
b28708d6 3259 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
b53e675d 3260 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
1da177e4
LT
3261 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3262 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
b53e675d 3263 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
1da177e4
LT
3264 dp += BBSIZE;
3265 }
3266 }
1da177e4
LT
3267}
3268
3269STATIC int
3270xlog_valid_rec_header(
3271 xlog_t *log,
3272 xlog_rec_header_t *rhead,
3273 xfs_daddr_t blkno)
3274{
3275 int hlen;
3276
b53e675d 3277 if (unlikely(be32_to_cpu(rhead->h_magicno) != XLOG_HEADER_MAGIC_NUM)) {
1da177e4
LT
3278 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3279 XFS_ERRLEVEL_LOW, log->l_mp);
3280 return XFS_ERROR(EFSCORRUPTED);
3281 }
3282 if (unlikely(
3283 (!rhead->h_version ||
b53e675d 3284 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
a0fa2b67 3285 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
34a622b2 3286 __func__, be32_to_cpu(rhead->h_version));
1da177e4
LT
3287 return XFS_ERROR(EIO);
3288 }
3289
3290 /* LR body must have data or it wouldn't have been written */
b53e675d 3291 hlen = be32_to_cpu(rhead->h_len);
1da177e4
LT
3292 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3293 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3294 XFS_ERRLEVEL_LOW, log->l_mp);
3295 return XFS_ERROR(EFSCORRUPTED);
3296 }
3297 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3298 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3299 XFS_ERRLEVEL_LOW, log->l_mp);
3300 return XFS_ERROR(EFSCORRUPTED);
3301 }
3302 return 0;
3303}
3304
3305/*
3306 * Read the log from tail to head and process the log records found.
3307 * Handle the two cases where the tail and head are in the same cycle
3308 * and where the active portion of the log wraps around the end of
3309 * the physical log separately. The pass parameter is passed through
3310 * to the routines called to process the data and is not looked at
3311 * here.
3312 */
3313STATIC int
3314xlog_do_recovery_pass(
3315 xlog_t *log,
3316 xfs_daddr_t head_blk,
3317 xfs_daddr_t tail_blk,
3318 int pass)
3319{
3320 xlog_rec_header_t *rhead;
3321 xfs_daddr_t blk_no;
fc5bc4c8 3322 xfs_caddr_t offset;
1da177e4
LT
3323 xfs_buf_t *hbp, *dbp;
3324 int error = 0, h_size;
3325 int bblks, split_bblks;
3326 int hblks, split_hblks, wrapped_hblks;
f0a76953 3327 struct hlist_head rhash[XLOG_RHASH_SIZE];
1da177e4
LT
3328
3329 ASSERT(head_blk != tail_blk);
3330
3331 /*
3332 * Read the header of the tail block and get the iclog buffer size from
3333 * h_size. Use this to tell how many sectors make up the log header.
3334 */
62118709 3335 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1da177e4
LT
3336 /*
3337 * When using variable length iclogs, read first sector of
3338 * iclog header and extract the header size from it. Get a
3339 * new hbp that is the correct size.
3340 */
3341 hbp = xlog_get_bp(log, 1);
3342 if (!hbp)
3343 return ENOMEM;
076e6acb
CH
3344
3345 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
3346 if (error)
1da177e4 3347 goto bread_err1;
076e6acb 3348
1da177e4
LT
3349 rhead = (xlog_rec_header_t *)offset;
3350 error = xlog_valid_rec_header(log, rhead, tail_blk);
3351 if (error)
3352 goto bread_err1;
b53e675d
CH
3353 h_size = be32_to_cpu(rhead->h_size);
3354 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
1da177e4
LT
3355 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3356 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3357 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3358 hblks++;
3359 xlog_put_bp(hbp);
3360 hbp = xlog_get_bp(log, hblks);
3361 } else {
3362 hblks = 1;
3363 }
3364 } else {
69ce58f0 3365 ASSERT(log->l_sectBBsize == 1);
1da177e4
LT
3366 hblks = 1;
3367 hbp = xlog_get_bp(log, 1);
3368 h_size = XLOG_BIG_RECORD_BSIZE;
3369 }
3370
3371 if (!hbp)
3372 return ENOMEM;
3373 dbp = xlog_get_bp(log, BTOBB(h_size));
3374 if (!dbp) {
3375 xlog_put_bp(hbp);
3376 return ENOMEM;
3377 }
3378
3379 memset(rhash, 0, sizeof(rhash));
3380 if (tail_blk <= head_blk) {
3381 for (blk_no = tail_blk; blk_no < head_blk; ) {
076e6acb
CH
3382 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3383 if (error)
1da177e4 3384 goto bread_err2;
076e6acb 3385
1da177e4
LT
3386 rhead = (xlog_rec_header_t *)offset;
3387 error = xlog_valid_rec_header(log, rhead, blk_no);
3388 if (error)
3389 goto bread_err2;
3390
3391 /* blocks in data section */
b53e675d 3392 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
076e6acb
CH
3393 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
3394 &offset);
1da177e4
LT
3395 if (error)
3396 goto bread_err2;
076e6acb 3397
1da177e4
LT
3398 xlog_unpack_data(rhead, offset, log);
3399 if ((error = xlog_recover_process_data(log,
3400 rhash, rhead, offset, pass)))
3401 goto bread_err2;
3402 blk_no += bblks + hblks;
3403 }
3404 } else {
3405 /*
3406 * Perform recovery around the end of the physical log.
3407 * When the head is not on the same cycle number as the tail,
3408 * we can't do a sequential recovery as above.
3409 */
3410 blk_no = tail_blk;
3411 while (blk_no < log->l_logBBsize) {
3412 /*
3413 * Check for header wrapping around physical end-of-log
3414 */
fc5bc4c8 3415 offset = XFS_BUF_PTR(hbp);
1da177e4
LT
3416 split_hblks = 0;
3417 wrapped_hblks = 0;
3418 if (blk_no + hblks <= log->l_logBBsize) {
3419 /* Read header in one read */
076e6acb
CH
3420 error = xlog_bread(log, blk_no, hblks, hbp,
3421 &offset);
1da177e4
LT
3422 if (error)
3423 goto bread_err2;
1da177e4
LT
3424 } else {
3425 /* This LR is split across physical log end */
3426 if (blk_no != log->l_logBBsize) {
3427 /* some data before physical log end */
3428 ASSERT(blk_no <= INT_MAX);
3429 split_hblks = log->l_logBBsize - (int)blk_no;
3430 ASSERT(split_hblks > 0);
076e6acb
CH
3431 error = xlog_bread(log, blk_no,
3432 split_hblks, hbp,
3433 &offset);
3434 if (error)
1da177e4 3435 goto bread_err2;
1da177e4 3436 }
076e6acb 3437
1da177e4
LT
3438 /*
3439 * Note: this black magic still works with
3440 * large sector sizes (non-512) only because:
3441 * - we increased the buffer size originally
3442 * by 1 sector giving us enough extra space
3443 * for the second read;
3444 * - the log start is guaranteed to be sector
3445 * aligned;
3446 * - we read the log end (LR header start)
3447 * _first_, then the log start (LR header end)
3448 * - order is important.
3449 */
234f56ac 3450 wrapped_hblks = hblks - split_hblks;
234f56ac 3451 error = XFS_BUF_SET_PTR(hbp,
fc5bc4c8 3452 offset + BBTOB(split_hblks),
1da177e4 3453 BBTOB(hblks - split_hblks));
076e6acb
CH
3454 if (error)
3455 goto bread_err2;
3456
3457 error = xlog_bread_noalign(log, 0,
3458 wrapped_hblks, hbp);
3459 if (error)
3460 goto bread_err2;
3461
fc5bc4c8 3462 error = XFS_BUF_SET_PTR(hbp, offset,
234f56ac 3463 BBTOB(hblks));
1da177e4
LT
3464 if (error)
3465 goto bread_err2;
1da177e4
LT
3466 }
3467 rhead = (xlog_rec_header_t *)offset;
3468 error = xlog_valid_rec_header(log, rhead,
3469 split_hblks ? blk_no : 0);
3470 if (error)
3471 goto bread_err2;
3472
b53e675d 3473 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
1da177e4
LT
3474 blk_no += hblks;
3475
3476 /* Read in data for log record */
3477 if (blk_no + bblks <= log->l_logBBsize) {
076e6acb
CH
3478 error = xlog_bread(log, blk_no, bblks, dbp,
3479 &offset);
1da177e4
LT
3480 if (error)
3481 goto bread_err2;
1da177e4
LT
3482 } else {
3483 /* This log record is split across the
3484 * physical end of log */
fc5bc4c8 3485 offset = XFS_BUF_PTR(dbp);
1da177e4
LT
3486 split_bblks = 0;
3487 if (blk_no != log->l_logBBsize) {
3488 /* some data is before the physical
3489 * end of log */
3490 ASSERT(!wrapped_hblks);
3491 ASSERT(blk_no <= INT_MAX);
3492 split_bblks =
3493 log->l_logBBsize - (int)blk_no;
3494 ASSERT(split_bblks > 0);
076e6acb
CH
3495 error = xlog_bread(log, blk_no,
3496 split_bblks, dbp,
3497 &offset);
3498 if (error)
1da177e4 3499 goto bread_err2;
1da177e4 3500 }
076e6acb 3501
1da177e4
LT
3502 /*
3503 * Note: this black magic still works with
3504 * large sector sizes (non-512) only because:
3505 * - we increased the buffer size originally
3506 * by 1 sector giving us enough extra space
3507 * for the second read;
3508 * - the log start is guaranteed to be sector
3509 * aligned;
3510 * - we read the log end (LR header start)
3511 * _first_, then the log start (LR header end)
3512 * - order is important.
3513 */
234f56ac 3514 error = XFS_BUF_SET_PTR(dbp,
fc5bc4c8 3515 offset + BBTOB(split_bblks),
1da177e4 3516 BBTOB(bblks - split_bblks));
234f56ac 3517 if (error)
1da177e4 3518 goto bread_err2;
076e6acb
CH
3519
3520 error = xlog_bread_noalign(log, wrapped_hblks,
3521 bblks - split_bblks,
3522 dbp);
3523 if (error)
3524 goto bread_err2;
3525
fc5bc4c8 3526 error = XFS_BUF_SET_PTR(dbp, offset, h_size);
076e6acb
CH
3527 if (error)
3528 goto bread_err2;
1da177e4
LT
3529 }
3530 xlog_unpack_data(rhead, offset, log);
3531 if ((error = xlog_recover_process_data(log, rhash,
3532 rhead, offset, pass)))
3533 goto bread_err2;
3534 blk_no += bblks;
3535 }
3536
3537 ASSERT(blk_no >= log->l_logBBsize);
3538 blk_no -= log->l_logBBsize;
3539
3540 /* read first part of physical log */
3541 while (blk_no < head_blk) {
076e6acb
CH
3542 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3543 if (error)
1da177e4 3544 goto bread_err2;
076e6acb 3545
1da177e4
LT
3546 rhead = (xlog_rec_header_t *)offset;
3547 error = xlog_valid_rec_header(log, rhead, blk_no);
3548 if (error)
3549 goto bread_err2;
076e6acb 3550
b53e675d 3551 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
076e6acb
CH
3552 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
3553 &offset);
3554 if (error)
1da177e4 3555 goto bread_err2;
076e6acb 3556
1da177e4
LT
3557 xlog_unpack_data(rhead, offset, log);
3558 if ((error = xlog_recover_process_data(log, rhash,
3559 rhead, offset, pass)))
3560 goto bread_err2;
3561 blk_no += bblks + hblks;
3562 }
3563 }
3564
3565 bread_err2:
3566 xlog_put_bp(dbp);
3567 bread_err1:
3568 xlog_put_bp(hbp);
3569 return error;
3570}
3571
3572/*
3573 * Do the recovery of the log. We actually do this in two phases.
3574 * The two passes are necessary in order to implement the function
3575 * of cancelling a record written into the log. The first pass
3576 * determines those things which have been cancelled, and the
3577 * second pass replays log items normally except for those which
3578 * have been cancelled. The handling of the replay and cancellations
3579 * takes place in the log item type specific routines.
3580 *
3581 * The table of items which have cancel records in the log is allocated
3582 * and freed at this level, since only here do we know when all of
3583 * the log recovery has been completed.
3584 */
3585STATIC int
3586xlog_do_log_recovery(
3587 xlog_t *log,
3588 xfs_daddr_t head_blk,
3589 xfs_daddr_t tail_blk)
3590{
d5689eaa 3591 int error, i;
1da177e4
LT
3592
3593 ASSERT(head_blk != tail_blk);
3594
3595 /*
3596 * First do a pass to find all of the cancelled buf log items.
3597 * Store them in the buf_cancel_table for use in the second pass.
3598 */
d5689eaa
CH
3599 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
3600 sizeof(struct list_head),
1da177e4 3601 KM_SLEEP);
d5689eaa
CH
3602 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3603 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
3604
1da177e4
LT
3605 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3606 XLOG_RECOVER_PASS1);
3607 if (error != 0) {
f0e2d93c 3608 kmem_free(log->l_buf_cancel_table);
1da177e4
LT
3609 log->l_buf_cancel_table = NULL;
3610 return error;
3611 }
3612 /*
3613 * Then do a second pass to actually recover the items in the log.
3614 * When it is complete free the table of buf cancel items.
3615 */
3616 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3617 XLOG_RECOVER_PASS2);
3618#ifdef DEBUG
6d192a9b 3619 if (!error) {
1da177e4
LT
3620 int i;
3621
3622 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
d5689eaa 3623 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
1da177e4
LT
3624 }
3625#endif /* DEBUG */
3626
f0e2d93c 3627 kmem_free(log->l_buf_cancel_table);
1da177e4
LT
3628 log->l_buf_cancel_table = NULL;
3629
3630 return error;
3631}
3632
3633/*
3634 * Do the actual recovery
3635 */
3636STATIC int
3637xlog_do_recover(
3638 xlog_t *log,
3639 xfs_daddr_t head_blk,
3640 xfs_daddr_t tail_blk)
3641{
3642 int error;
3643 xfs_buf_t *bp;
3644 xfs_sb_t *sbp;
3645
3646 /*
3647 * First replay the images in the log.
3648 */
3649 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3650 if (error) {
3651 return error;
3652 }
3653
3654 XFS_bflush(log->l_mp->m_ddev_targp);
3655
3656 /*
3657 * If IO errors happened during recovery, bail out.
3658 */
3659 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3660 return (EIO);
3661 }
3662
3663 /*
3664 * We now update the tail_lsn since much of the recovery has completed
3665 * and there may be space available to use. If there were no extent
3666 * or iunlinks, we can free up the entire log and set the tail_lsn to
3667 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3668 * lsn of the last known good LR on disk. If there are extent frees
3669 * or iunlinks they will have some entries in the AIL; so we look at
3670 * the AIL to determine how to set the tail_lsn.
3671 */
3672 xlog_assign_tail_lsn(log->l_mp);
3673
3674 /*
3675 * Now that we've finished replaying all buffer and inode
3676 * updates, re-read in the superblock.
3677 */
3678 bp = xfs_getsb(log->l_mp, 0);
3679 XFS_BUF_UNDONE(bp);
bebf963f
LM
3680 ASSERT(!(XFS_BUF_ISWRITE(bp)));
3681 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
1da177e4 3682 XFS_BUF_READ(bp);
bebf963f 3683 XFS_BUF_UNASYNC(bp);
1da177e4 3684 xfsbdstrat(log->l_mp, bp);
1a1a3e97 3685 error = xfs_buf_iowait(bp);
d64e31a2 3686 if (error) {
1da177e4
LT
3687 xfs_ioerror_alert("xlog_do_recover",
3688 log->l_mp, bp, XFS_BUF_ADDR(bp));
3689 ASSERT(0);
3690 xfs_buf_relse(bp);
3691 return error;
3692 }
3693
3694 /* Convert superblock from on-disk format */
3695 sbp = &log->l_mp->m_sb;
2bdf7cd0 3696 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
1da177e4 3697 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
62118709 3698 ASSERT(xfs_sb_good_version(sbp));
1da177e4
LT
3699 xfs_buf_relse(bp);
3700
5478eead
LM
3701 /* We've re-read the superblock so re-initialize per-cpu counters */
3702 xfs_icsb_reinit_counters(log->l_mp);
3703
1da177e4
LT
3704 xlog_recover_check_summary(log);
3705
3706 /* Normal transactions can now occur */
3707 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3708 return 0;
3709}
3710
3711/*
3712 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3713 *
3714 * Return error or zero.
3715 */
3716int
3717xlog_recover(
65be6054 3718 xlog_t *log)
1da177e4
LT
3719{
3720 xfs_daddr_t head_blk, tail_blk;
3721 int error;
3722
3723 /* find the tail of the log */
65be6054 3724 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
1da177e4
LT
3725 return error;
3726
3727 if (tail_blk != head_blk) {
3728 /* There used to be a comment here:
3729 *
3730 * disallow recovery on read-only mounts. note -- mount
3731 * checks for ENOSPC and turns it into an intelligent
3732 * error message.
3733 * ...but this is no longer true. Now, unless you specify
3734 * NORECOVERY (in which case this function would never be
3735 * called), we just go ahead and recover. We do this all
3736 * under the vfs layer, so we can get away with it unless
3737 * the device itself is read-only, in which case we fail.
3738 */
3a02ee18 3739 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
1da177e4
LT
3740 return error;
3741 }
3742
a0fa2b67
DC
3743 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
3744 log->l_mp->m_logname ? log->l_mp->m_logname
3745 : "internal");
1da177e4
LT
3746
3747 error = xlog_do_recover(log, head_blk, tail_blk);
3748 log->l_flags |= XLOG_RECOVERY_NEEDED;
3749 }
3750 return error;
3751}
3752
3753/*
3754 * In the first part of recovery we replay inodes and buffers and build
3755 * up the list of extent free items which need to be processed. Here
3756 * we process the extent free items and clean up the on disk unlinked
3757 * inode lists. This is separated from the first part of recovery so
3758 * that the root and real-time bitmap inodes can be read in from disk in
3759 * between the two stages. This is necessary so that we can free space
3760 * in the real-time portion of the file system.
3761 */
3762int
3763xlog_recover_finish(
4249023a 3764 xlog_t *log)
1da177e4
LT
3765{
3766 /*
3767 * Now we're ready to do the transactions needed for the
3768 * rest of recovery. Start with completing all the extent
3769 * free intent records and then process the unlinked inode
3770 * lists. At this point, we essentially run in normal mode
3771 * except that we're still performing recovery actions
3772 * rather than accepting new requests.
3773 */
3774 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
3c1e2bbe
DC
3775 int error;
3776 error = xlog_recover_process_efis(log);
3777 if (error) {
a0fa2b67 3778 xfs_alert(log->l_mp, "Failed to recover EFIs");
3c1e2bbe
DC
3779 return error;
3780 }
1da177e4
LT
3781 /*
3782 * Sync the log to get all the EFIs out of the AIL.
3783 * This isn't absolutely necessary, but it helps in
3784 * case the unlink transactions would have problems
3785 * pushing the EFIs out of the way.
3786 */
a14a348b 3787 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
1da177e4 3788
4249023a 3789 xlog_recover_process_iunlinks(log);
1da177e4
LT
3790
3791 xlog_recover_check_summary(log);
3792
a0fa2b67
DC
3793 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
3794 log->l_mp->m_logname ? log->l_mp->m_logname
3795 : "internal");
1da177e4
LT
3796 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
3797 } else {
a0fa2b67 3798 xfs_info(log->l_mp, "Ending clean mount");
1da177e4
LT
3799 }
3800 return 0;
3801}
3802
3803
3804#if defined(DEBUG)
3805/*
3806 * Read all of the agf and agi counters and check that they
3807 * are consistent with the superblock counters.
3808 */
3809void
3810xlog_recover_check_summary(
3811 xlog_t *log)
3812{
3813 xfs_mount_t *mp;
3814 xfs_agf_t *agfp;
1da177e4
LT
3815 xfs_buf_t *agfbp;
3816 xfs_buf_t *agibp;
1da177e4
LT
3817 xfs_agnumber_t agno;
3818 __uint64_t freeblks;
3819 __uint64_t itotal;
3820 __uint64_t ifree;
5e1be0fb 3821 int error;
1da177e4
LT
3822
3823 mp = log->l_mp;
3824
3825 freeblks = 0LL;
3826 itotal = 0LL;
3827 ifree = 0LL;
3828 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
4805621a
CH
3829 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
3830 if (error) {
a0fa2b67
DC
3831 xfs_alert(mp, "%s agf read failed agno %d error %d",
3832 __func__, agno, error);
4805621a
CH
3833 } else {
3834 agfp = XFS_BUF_TO_AGF(agfbp);
3835 freeblks += be32_to_cpu(agfp->agf_freeblks) +
3836 be32_to_cpu(agfp->agf_flcount);
3837 xfs_buf_relse(agfbp);
1da177e4 3838 }
1da177e4 3839
5e1be0fb 3840 error = xfs_read_agi(mp, NULL, agno, &agibp);
a0fa2b67
DC
3841 if (error) {
3842 xfs_alert(mp, "%s agi read failed agno %d error %d",
3843 __func__, agno, error);
3844 } else {
5e1be0fb 3845 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
16259e7d 3846
5e1be0fb
CH
3847 itotal += be32_to_cpu(agi->agi_count);
3848 ifree += be32_to_cpu(agi->agi_freecount);
3849 xfs_buf_relse(agibp);
3850 }
1da177e4 3851 }
1da177e4
LT
3852}
3853#endif /* DEBUG */