]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. | |
3 | * All Rights Reserved. | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or | |
6 | * modify it under the terms of the GNU General Public License as | |
7 | * published by the Free Software Foundation. | |
8 | * | |
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. | |
13 | * | |
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 | |
17 | */ | |
18 | #include "xfs.h" | |
19 | #include "xfs_fs.h" | |
20 | #include "xfs_shared.h" | |
21 | #include "xfs_format.h" | |
22 | #include "xfs_log_format.h" | |
23 | #include "xfs_trans_resv.h" | |
24 | #include "xfs_bit.h" | |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_mount.h" | |
27 | #include "xfs_da_format.h" | |
28 | #include "xfs_da_btree.h" | |
29 | #include "xfs_inode.h" | |
30 | #include "xfs_trans.h" | |
31 | #include "xfs_log.h" | |
32 | #include "xfs_log_priv.h" | |
33 | #include "xfs_log_recover.h" | |
34 | #include "xfs_inode_item.h" | |
35 | #include "xfs_extfree_item.h" | |
36 | #include "xfs_trans_priv.h" | |
37 | #include "xfs_alloc.h" | |
38 | #include "xfs_ialloc.h" | |
39 | #include "xfs_quota.h" | |
40 | #include "xfs_cksum.h" | |
41 | #include "xfs_trace.h" | |
42 | #include "xfs_icache.h" | |
43 | #include "xfs_bmap_btree.h" | |
44 | #include "xfs_error.h" | |
45 | #include "xfs_dir2.h" | |
46 | ||
47 | #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1) | |
48 | ||
49 | STATIC int | |
50 | xlog_find_zeroed( | |
51 | struct xlog *, | |
52 | xfs_daddr_t *); | |
53 | STATIC int | |
54 | xlog_clear_stale_blocks( | |
55 | struct xlog *, | |
56 | xfs_lsn_t); | |
57 | #if defined(DEBUG) | |
58 | STATIC void | |
59 | xlog_recover_check_summary( | |
60 | struct xlog *); | |
61 | #else | |
62 | #define xlog_recover_check_summary(log) | |
63 | #endif | |
64 | STATIC int | |
65 | xlog_do_recovery_pass( | |
66 | struct xlog *, xfs_daddr_t, xfs_daddr_t, int, xfs_daddr_t *); | |
67 | ||
68 | /* | |
69 | * This structure is used during recovery to record the buf log items which | |
70 | * have been canceled and should not be replayed. | |
71 | */ | |
72 | struct xfs_buf_cancel { | |
73 | xfs_daddr_t bc_blkno; | |
74 | uint bc_len; | |
75 | int bc_refcount; | |
76 | struct list_head bc_list; | |
77 | }; | |
78 | ||
79 | /* | |
80 | * Sector aligned buffer routines for buffer create/read/write/access | |
81 | */ | |
82 | ||
83 | /* | |
84 | * Verify the given count of basic blocks is valid number of blocks | |
85 | * to specify for an operation involving the given XFS log buffer. | |
86 | * Returns nonzero if the count is valid, 0 otherwise. | |
87 | */ | |
88 | ||
89 | static inline int | |
90 | xlog_buf_bbcount_valid( | |
91 | struct xlog *log, | |
92 | int bbcount) | |
93 | { | |
94 | return bbcount > 0 && bbcount <= log->l_logBBsize; | |
95 | } | |
96 | ||
97 | /* | |
98 | * Allocate a buffer to hold log data. The buffer needs to be able | |
99 | * to map to a range of nbblks basic blocks at any valid (basic | |
100 | * block) offset within the log. | |
101 | */ | |
102 | STATIC xfs_buf_t * | |
103 | xlog_get_bp( | |
104 | struct xlog *log, | |
105 | int nbblks) | |
106 | { | |
107 | struct xfs_buf *bp; | |
108 | ||
109 | if (!xlog_buf_bbcount_valid(log, nbblks)) { | |
110 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", | |
111 | nbblks); | |
112 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
113 | return NULL; | |
114 | } | |
115 | ||
116 | /* | |
117 | * We do log I/O in units of log sectors (a power-of-2 | |
118 | * multiple of the basic block size), so we round up the | |
119 | * requested size to accommodate the basic blocks required | |
120 | * for complete log sectors. | |
121 | * | |
122 | * In addition, the buffer may be used for a non-sector- | |
123 | * aligned block offset, in which case an I/O of the | |
124 | * requested size could extend beyond the end of the | |
125 | * buffer. If the requested size is only 1 basic block it | |
126 | * will never straddle a sector boundary, so this won't be | |
127 | * an issue. Nor will this be a problem if the log I/O is | |
128 | * done in basic blocks (sector size 1). But otherwise we | |
129 | * extend the buffer by one extra log sector to ensure | |
130 | * there's space to accommodate this possibility. | |
131 | */ | |
132 | if (nbblks > 1 && log->l_sectBBsize > 1) | |
133 | nbblks += log->l_sectBBsize; | |
134 | nbblks = round_up(nbblks, log->l_sectBBsize); | |
135 | ||
136 | bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0); | |
137 | if (bp) | |
138 | xfs_buf_unlock(bp); | |
139 | return bp; | |
140 | } | |
141 | ||
142 | STATIC void | |
143 | xlog_put_bp( | |
144 | xfs_buf_t *bp) | |
145 | { | |
146 | xfs_buf_free(bp); | |
147 | } | |
148 | ||
149 | /* | |
150 | * Return the address of the start of the given block number's data | |
151 | * in a log buffer. The buffer covers a log sector-aligned region. | |
152 | */ | |
153 | STATIC char * | |
154 | xlog_align( | |
155 | struct xlog *log, | |
156 | xfs_daddr_t blk_no, | |
157 | int nbblks, | |
158 | struct xfs_buf *bp) | |
159 | { | |
160 | xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1); | |
161 | ||
162 | ASSERT(offset + nbblks <= bp->b_length); | |
163 | return bp->b_addr + BBTOB(offset); | |
164 | } | |
165 | ||
166 | ||
167 | /* | |
168 | * nbblks should be uint, but oh well. Just want to catch that 32-bit length. | |
169 | */ | |
170 | STATIC int | |
171 | xlog_bread_noalign( | |
172 | struct xlog *log, | |
173 | xfs_daddr_t blk_no, | |
174 | int nbblks, | |
175 | struct xfs_buf *bp) | |
176 | { | |
177 | int error; | |
178 | ||
179 | if (!xlog_buf_bbcount_valid(log, nbblks)) { | |
180 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", | |
181 | nbblks); | |
182 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
183 | return -EFSCORRUPTED; | |
184 | } | |
185 | ||
186 | blk_no = round_down(blk_no, log->l_sectBBsize); | |
187 | nbblks = round_up(nbblks, log->l_sectBBsize); | |
188 | ||
189 | ASSERT(nbblks > 0); | |
190 | ASSERT(nbblks <= bp->b_length); | |
191 | ||
192 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); | |
193 | bp->b_flags |= XBF_READ; | |
194 | bp->b_io_length = nbblks; | |
195 | bp->b_error = 0; | |
196 | ||
197 | error = xfs_buf_submit_wait(bp); | |
198 | if (error && !XFS_FORCED_SHUTDOWN(log->l_mp)) | |
199 | xfs_buf_ioerror_alert(bp, __func__); | |
200 | return error; | |
201 | } | |
202 | ||
203 | STATIC int | |
204 | xlog_bread( | |
205 | struct xlog *log, | |
206 | xfs_daddr_t blk_no, | |
207 | int nbblks, | |
208 | struct xfs_buf *bp, | |
209 | char **offset) | |
210 | { | |
211 | int error; | |
212 | ||
213 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); | |
214 | if (error) | |
215 | return error; | |
216 | ||
217 | *offset = xlog_align(log, blk_no, nbblks, bp); | |
218 | return 0; | |
219 | } | |
220 | ||
221 | /* | |
222 | * Read at an offset into the buffer. Returns with the buffer in it's original | |
223 | * state regardless of the result of the read. | |
224 | */ | |
225 | STATIC int | |
226 | xlog_bread_offset( | |
227 | struct xlog *log, | |
228 | xfs_daddr_t blk_no, /* block to read from */ | |
229 | int nbblks, /* blocks to read */ | |
230 | struct xfs_buf *bp, | |
231 | char *offset) | |
232 | { | |
233 | char *orig_offset = bp->b_addr; | |
234 | int orig_len = BBTOB(bp->b_length); | |
235 | int error, error2; | |
236 | ||
237 | error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks)); | |
238 | if (error) | |
239 | return error; | |
240 | ||
241 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); | |
242 | ||
243 | /* must reset buffer pointer even on error */ | |
244 | error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len); | |
245 | if (error) | |
246 | return error; | |
247 | return error2; | |
248 | } | |
249 | ||
250 | /* | |
251 | * Write out the buffer at the given block for the given number of blocks. | |
252 | * The buffer is kept locked across the write and is returned locked. | |
253 | * This can only be used for synchronous log writes. | |
254 | */ | |
255 | STATIC int | |
256 | xlog_bwrite( | |
257 | struct xlog *log, | |
258 | xfs_daddr_t blk_no, | |
259 | int nbblks, | |
260 | struct xfs_buf *bp) | |
261 | { | |
262 | int error; | |
263 | ||
264 | if (!xlog_buf_bbcount_valid(log, nbblks)) { | |
265 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", | |
266 | nbblks); | |
267 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); | |
268 | return -EFSCORRUPTED; | |
269 | } | |
270 | ||
271 | blk_no = round_down(blk_no, log->l_sectBBsize); | |
272 | nbblks = round_up(nbblks, log->l_sectBBsize); | |
273 | ||
274 | ASSERT(nbblks > 0); | |
275 | ASSERT(nbblks <= bp->b_length); | |
276 | ||
277 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); | |
278 | xfs_buf_hold(bp); | |
279 | xfs_buf_lock(bp); | |
280 | bp->b_io_length = nbblks; | |
281 | bp->b_error = 0; | |
282 | ||
283 | error = xfs_bwrite(bp); | |
284 | if (error) | |
285 | xfs_buf_ioerror_alert(bp, __func__); | |
286 | xfs_buf_relse(bp); | |
287 | return error; | |
288 | } | |
289 | ||
290 | #ifdef DEBUG | |
291 | /* | |
292 | * dump debug superblock and log record information | |
293 | */ | |
294 | STATIC void | |
295 | xlog_header_check_dump( | |
296 | xfs_mount_t *mp, | |
297 | xlog_rec_header_t *head) | |
298 | { | |
299 | xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d", | |
300 | __func__, &mp->m_sb.sb_uuid, XLOG_FMT); | |
301 | xfs_debug(mp, " log : uuid = %pU, fmt = %d", | |
302 | &head->h_fs_uuid, be32_to_cpu(head->h_fmt)); | |
303 | } | |
304 | #else | |
305 | #define xlog_header_check_dump(mp, head) | |
306 | #endif | |
307 | ||
308 | /* | |
309 | * check log record header for recovery | |
310 | */ | |
311 | STATIC int | |
312 | xlog_header_check_recover( | |
313 | xfs_mount_t *mp, | |
314 | xlog_rec_header_t *head) | |
315 | { | |
316 | ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)); | |
317 | ||
318 | /* | |
319 | * IRIX doesn't write the h_fmt field and leaves it zeroed | |
320 | * (XLOG_FMT_UNKNOWN). This stops us from trying to recover | |
321 | * a dirty log created in IRIX. | |
322 | */ | |
323 | if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) { | |
324 | xfs_warn(mp, | |
325 | "dirty log written in incompatible format - can't recover"); | |
326 | xlog_header_check_dump(mp, head); | |
327 | XFS_ERROR_REPORT("xlog_header_check_recover(1)", | |
328 | XFS_ERRLEVEL_HIGH, mp); | |
329 | return -EFSCORRUPTED; | |
330 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { | |
331 | xfs_warn(mp, | |
332 | "dirty log entry has mismatched uuid - can't recover"); | |
333 | xlog_header_check_dump(mp, head); | |
334 | XFS_ERROR_REPORT("xlog_header_check_recover(2)", | |
335 | XFS_ERRLEVEL_HIGH, mp); | |
336 | return -EFSCORRUPTED; | |
337 | } | |
338 | return 0; | |
339 | } | |
340 | ||
341 | /* | |
342 | * read the head block of the log and check the header | |
343 | */ | |
344 | STATIC int | |
345 | xlog_header_check_mount( | |
346 | xfs_mount_t *mp, | |
347 | xlog_rec_header_t *head) | |
348 | { | |
349 | ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)); | |
350 | ||
351 | if (uuid_is_nil(&head->h_fs_uuid)) { | |
352 | /* | |
353 | * IRIX doesn't write the h_fs_uuid or h_fmt fields. If | |
354 | * h_fs_uuid is nil, we assume this log was last mounted | |
355 | * by IRIX and continue. | |
356 | */ | |
357 | xfs_warn(mp, "nil uuid in log - IRIX style log"); | |
358 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { | |
359 | xfs_warn(mp, "log has mismatched uuid - can't recover"); | |
360 | xlog_header_check_dump(mp, head); | |
361 | XFS_ERROR_REPORT("xlog_header_check_mount", | |
362 | XFS_ERRLEVEL_HIGH, mp); | |
363 | return -EFSCORRUPTED; | |
364 | } | |
365 | return 0; | |
366 | } | |
367 | ||
368 | STATIC void | |
369 | xlog_recover_iodone( | |
370 | struct xfs_buf *bp) | |
371 | { | |
372 | if (bp->b_error) { | |
373 | /* | |
374 | * We're not going to bother about retrying | |
375 | * this during recovery. One strike! | |
376 | */ | |
377 | if (!XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { | |
378 | xfs_buf_ioerror_alert(bp, __func__); | |
379 | xfs_force_shutdown(bp->b_target->bt_mount, | |
380 | SHUTDOWN_META_IO_ERROR); | |
381 | } | |
382 | } | |
383 | bp->b_iodone = NULL; | |
384 | xfs_buf_ioend(bp); | |
385 | } | |
386 | ||
387 | /* | |
388 | * This routine finds (to an approximation) the first block in the physical | |
389 | * log which contains the given cycle. It uses a binary search algorithm. | |
390 | * Note that the algorithm can not be perfect because the disk will not | |
391 | * necessarily be perfect. | |
392 | */ | |
393 | STATIC int | |
394 | xlog_find_cycle_start( | |
395 | struct xlog *log, | |
396 | struct xfs_buf *bp, | |
397 | xfs_daddr_t first_blk, | |
398 | xfs_daddr_t *last_blk, | |
399 | uint cycle) | |
400 | { | |
401 | char *offset; | |
402 | xfs_daddr_t mid_blk; | |
403 | xfs_daddr_t end_blk; | |
404 | uint mid_cycle; | |
405 | int error; | |
406 | ||
407 | end_blk = *last_blk; | |
408 | mid_blk = BLK_AVG(first_blk, end_blk); | |
409 | while (mid_blk != first_blk && mid_blk != end_blk) { | |
410 | error = xlog_bread(log, mid_blk, 1, bp, &offset); | |
411 | if (error) | |
412 | return error; | |
413 | mid_cycle = xlog_get_cycle(offset); | |
414 | if (mid_cycle == cycle) | |
415 | end_blk = mid_blk; /* last_half_cycle == mid_cycle */ | |
416 | else | |
417 | first_blk = mid_blk; /* first_half_cycle == mid_cycle */ | |
418 | mid_blk = BLK_AVG(first_blk, end_blk); | |
419 | } | |
420 | ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) || | |
421 | (mid_blk == end_blk && mid_blk-1 == first_blk)); | |
422 | ||
423 | *last_blk = end_blk; | |
424 | ||
425 | return 0; | |
426 | } | |
427 | ||
428 | /* | |
429 | * Check that a range of blocks does not contain stop_on_cycle_no. | |
430 | * Fill in *new_blk with the block offset where such a block is | |
431 | * found, or with -1 (an invalid block number) if there is no such | |
432 | * block in the range. The scan needs to occur from front to back | |
433 | * and the pointer into the region must be updated since a later | |
434 | * routine will need to perform another test. | |
435 | */ | |
436 | STATIC int | |
437 | xlog_find_verify_cycle( | |
438 | struct xlog *log, | |
439 | xfs_daddr_t start_blk, | |
440 | int nbblks, | |
441 | uint stop_on_cycle_no, | |
442 | xfs_daddr_t *new_blk) | |
443 | { | |
444 | xfs_daddr_t i, j; | |
445 | uint cycle; | |
446 | xfs_buf_t *bp; | |
447 | xfs_daddr_t bufblks; | |
448 | char *buf = NULL; | |
449 | int error = 0; | |
450 | ||
451 | /* | |
452 | * Greedily allocate a buffer big enough to handle the full | |
453 | * range of basic blocks we'll be examining. If that fails, | |
454 | * try a smaller size. We need to be able to read at least | |
455 | * a log sector, or we're out of luck. | |
456 | */ | |
457 | bufblks = 1 << ffs(nbblks); | |
458 | while (bufblks > log->l_logBBsize) | |
459 | bufblks >>= 1; | |
460 | while (!(bp = xlog_get_bp(log, bufblks))) { | |
461 | bufblks >>= 1; | |
462 | if (bufblks < log->l_sectBBsize) | |
463 | return -ENOMEM; | |
464 | } | |
465 | ||
466 | for (i = start_blk; i < start_blk + nbblks; i += bufblks) { | |
467 | int bcount; | |
468 | ||
469 | bcount = min(bufblks, (start_blk + nbblks - i)); | |
470 | ||
471 | error = xlog_bread(log, i, bcount, bp, &buf); | |
472 | if (error) | |
473 | goto out; | |
474 | ||
475 | for (j = 0; j < bcount; j++) { | |
476 | cycle = xlog_get_cycle(buf); | |
477 | if (cycle == stop_on_cycle_no) { | |
478 | *new_blk = i+j; | |
479 | goto out; | |
480 | } | |
481 | ||
482 | buf += BBSIZE; | |
483 | } | |
484 | } | |
485 | ||
486 | *new_blk = -1; | |
487 | ||
488 | out: | |
489 | xlog_put_bp(bp); | |
490 | return error; | |
491 | } | |
492 | ||
493 | /* | |
494 | * Potentially backup over partial log record write. | |
495 | * | |
496 | * In the typical case, last_blk is the number of the block directly after | |
497 | * a good log record. Therefore, we subtract one to get the block number | |
498 | * of the last block in the given buffer. extra_bblks contains the number | |
499 | * of blocks we would have read on a previous read. This happens when the | |
500 | * last log record is split over the end of the physical log. | |
501 | * | |
502 | * extra_bblks is the number of blocks potentially verified on a previous | |
503 | * call to this routine. | |
504 | */ | |
505 | STATIC int | |
506 | xlog_find_verify_log_record( | |
507 | struct xlog *log, | |
508 | xfs_daddr_t start_blk, | |
509 | xfs_daddr_t *last_blk, | |
510 | int extra_bblks) | |
511 | { | |
512 | xfs_daddr_t i; | |
513 | xfs_buf_t *bp; | |
514 | char *offset = NULL; | |
515 | xlog_rec_header_t *head = NULL; | |
516 | int error = 0; | |
517 | int smallmem = 0; | |
518 | int num_blks = *last_blk - start_blk; | |
519 | int xhdrs; | |
520 | ||
521 | ASSERT(start_blk != 0 || *last_blk != start_blk); | |
522 | ||
523 | if (!(bp = xlog_get_bp(log, num_blks))) { | |
524 | if (!(bp = xlog_get_bp(log, 1))) | |
525 | return -ENOMEM; | |
526 | smallmem = 1; | |
527 | } else { | |
528 | error = xlog_bread(log, start_blk, num_blks, bp, &offset); | |
529 | if (error) | |
530 | goto out; | |
531 | offset += ((num_blks - 1) << BBSHIFT); | |
532 | } | |
533 | ||
534 | for (i = (*last_blk) - 1; i >= 0; i--) { | |
535 | if (i < start_blk) { | |
536 | /* valid log record not found */ | |
537 | xfs_warn(log->l_mp, | |
538 | "Log inconsistent (didn't find previous header)"); | |
539 | ASSERT(0); | |
540 | error = -EIO; | |
541 | goto out; | |
542 | } | |
543 | ||
544 | if (smallmem) { | |
545 | error = xlog_bread(log, i, 1, bp, &offset); | |
546 | if (error) | |
547 | goto out; | |
548 | } | |
549 | ||
550 | head = (xlog_rec_header_t *)offset; | |
551 | ||
552 | if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) | |
553 | break; | |
554 | ||
555 | if (!smallmem) | |
556 | offset -= BBSIZE; | |
557 | } | |
558 | ||
559 | /* | |
560 | * We hit the beginning of the physical log & still no header. Return | |
561 | * to caller. If caller can handle a return of -1, then this routine | |
562 | * will be called again for the end of the physical log. | |
563 | */ | |
564 | if (i == -1) { | |
565 | error = 1; | |
566 | goto out; | |
567 | } | |
568 | ||
569 | /* | |
570 | * We have the final block of the good log (the first block | |
571 | * of the log record _before_ the head. So we check the uuid. | |
572 | */ | |
573 | if ((error = xlog_header_check_mount(log->l_mp, head))) | |
574 | goto out; | |
575 | ||
576 | /* | |
577 | * We may have found a log record header before we expected one. | |
578 | * last_blk will be the 1st block # with a given cycle #. We may end | |
579 | * up reading an entire log record. In this case, we don't want to | |
580 | * reset last_blk. Only when last_blk points in the middle of a log | |
581 | * record do we update last_blk. | |
582 | */ | |
583 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
584 | uint h_size = be32_to_cpu(head->h_size); | |
585 | ||
586 | xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE; | |
587 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
588 | xhdrs++; | |
589 | } else { | |
590 | xhdrs = 1; | |
591 | } | |
592 | ||
593 | if (*last_blk - i + extra_bblks != | |
594 | BTOBB(be32_to_cpu(head->h_len)) + xhdrs) | |
595 | *last_blk = i; | |
596 | ||
597 | out: | |
598 | xlog_put_bp(bp); | |
599 | return error; | |
600 | } | |
601 | ||
602 | /* | |
603 | * Head is defined to be the point of the log where the next log write | |
604 | * could go. This means that incomplete LR writes at the end are | |
605 | * eliminated when calculating the head. We aren't guaranteed that previous | |
606 | * LR have complete transactions. We only know that a cycle number of | |
607 | * current cycle number -1 won't be present in the log if we start writing | |
608 | * from our current block number. | |
609 | * | |
610 | * last_blk contains the block number of the first block with a given | |
611 | * cycle number. | |
612 | * | |
613 | * Return: zero if normal, non-zero if error. | |
614 | */ | |
615 | STATIC int | |
616 | xlog_find_head( | |
617 | struct xlog *log, | |
618 | xfs_daddr_t *return_head_blk) | |
619 | { | |
620 | xfs_buf_t *bp; | |
621 | char *offset; | |
622 | xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk; | |
623 | int num_scan_bblks; | |
624 | uint first_half_cycle, last_half_cycle; | |
625 | uint stop_on_cycle; | |
626 | int error, log_bbnum = log->l_logBBsize; | |
627 | ||
628 | /* Is the end of the log device zeroed? */ | |
629 | error = xlog_find_zeroed(log, &first_blk); | |
630 | if (error < 0) { | |
631 | xfs_warn(log->l_mp, "empty log check failed"); | |
632 | return error; | |
633 | } | |
634 | if (error == 1) { | |
635 | *return_head_blk = first_blk; | |
636 | ||
637 | /* Is the whole lot zeroed? */ | |
638 | if (!first_blk) { | |
639 | /* Linux XFS shouldn't generate totally zeroed logs - | |
640 | * mkfs etc write a dummy unmount record to a fresh | |
641 | * log so we can store the uuid in there | |
642 | */ | |
643 | xfs_warn(log->l_mp, "totally zeroed log"); | |
644 | } | |
645 | ||
646 | return 0; | |
647 | } | |
648 | ||
649 | first_blk = 0; /* get cycle # of 1st block */ | |
650 | bp = xlog_get_bp(log, 1); | |
651 | if (!bp) | |
652 | return -ENOMEM; | |
653 | ||
654 | error = xlog_bread(log, 0, 1, bp, &offset); | |
655 | if (error) | |
656 | goto bp_err; | |
657 | ||
658 | first_half_cycle = xlog_get_cycle(offset); | |
659 | ||
660 | last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */ | |
661 | error = xlog_bread(log, last_blk, 1, bp, &offset); | |
662 | if (error) | |
663 | goto bp_err; | |
664 | ||
665 | last_half_cycle = xlog_get_cycle(offset); | |
666 | ASSERT(last_half_cycle != 0); | |
667 | ||
668 | /* | |
669 | * If the 1st half cycle number is equal to the last half cycle number, | |
670 | * then the entire log is stamped with the same cycle number. In this | |
671 | * case, head_blk can't be set to zero (which makes sense). The below | |
672 | * math doesn't work out properly with head_blk equal to zero. Instead, | |
673 | * we set it to log_bbnum which is an invalid block number, but this | |
674 | * value makes the math correct. If head_blk doesn't changed through | |
675 | * all the tests below, *head_blk is set to zero at the very end rather | |
676 | * than log_bbnum. In a sense, log_bbnum and zero are the same block | |
677 | * in a circular file. | |
678 | */ | |
679 | if (first_half_cycle == last_half_cycle) { | |
680 | /* | |
681 | * In this case we believe that the entire log should have | |
682 | * cycle number last_half_cycle. We need to scan backwards | |
683 | * from the end verifying that there are no holes still | |
684 | * containing last_half_cycle - 1. If we find such a hole, | |
685 | * then the start of that hole will be the new head. The | |
686 | * simple case looks like | |
687 | * x | x ... | x - 1 | x | |
688 | * Another case that fits this picture would be | |
689 | * x | x + 1 | x ... | x | |
690 | * In this case the head really is somewhere at the end of the | |
691 | * log, as one of the latest writes at the beginning was | |
692 | * incomplete. | |
693 | * One more case is | |
694 | * x | x + 1 | x ... | x - 1 | x | |
695 | * This is really the combination of the above two cases, and | |
696 | * the head has to end up at the start of the x-1 hole at the | |
697 | * end of the log. | |
698 | * | |
699 | * In the 256k log case, we will read from the beginning to the | |
700 | * end of the log and search for cycle numbers equal to x-1. | |
701 | * We don't worry about the x+1 blocks that we encounter, | |
702 | * because we know that they cannot be the head since the log | |
703 | * started with x. | |
704 | */ | |
705 | head_blk = log_bbnum; | |
706 | stop_on_cycle = last_half_cycle - 1; | |
707 | } else { | |
708 | /* | |
709 | * In this case we want to find the first block with cycle | |
710 | * number matching last_half_cycle. We expect the log to be | |
711 | * some variation on | |
712 | * x + 1 ... | x ... | x | |
713 | * The first block with cycle number x (last_half_cycle) will | |
714 | * be where the new head belongs. First we do a binary search | |
715 | * for the first occurrence of last_half_cycle. The binary | |
716 | * search may not be totally accurate, so then we scan back | |
717 | * from there looking for occurrences of last_half_cycle before | |
718 | * us. If that backwards scan wraps around the beginning of | |
719 | * the log, then we look for occurrences of last_half_cycle - 1 | |
720 | * at the end of the log. The cases we're looking for look | |
721 | * like | |
722 | * v binary search stopped here | |
723 | * x + 1 ... | x | x + 1 | x ... | x | |
724 | * ^ but we want to locate this spot | |
725 | * or | |
726 | * <---------> less than scan distance | |
727 | * x + 1 ... | x ... | x - 1 | x | |
728 | * ^ we want to locate this spot | |
729 | */ | |
730 | stop_on_cycle = last_half_cycle; | |
731 | if ((error = xlog_find_cycle_start(log, bp, first_blk, | |
732 | &head_blk, last_half_cycle))) | |
733 | goto bp_err; | |
734 | } | |
735 | ||
736 | /* | |
737 | * Now validate the answer. Scan back some number of maximum possible | |
738 | * blocks and make sure each one has the expected cycle number. The | |
739 | * maximum is determined by the total possible amount of buffering | |
740 | * in the in-core log. The following number can be made tighter if | |
741 | * we actually look at the block size of the filesystem. | |
742 | */ | |
743 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); | |
744 | if (head_blk >= num_scan_bblks) { | |
745 | /* | |
746 | * We are guaranteed that the entire check can be performed | |
747 | * in one buffer. | |
748 | */ | |
749 | start_blk = head_blk - num_scan_bblks; | |
750 | if ((error = xlog_find_verify_cycle(log, | |
751 | start_blk, num_scan_bblks, | |
752 | stop_on_cycle, &new_blk))) | |
753 | goto bp_err; | |
754 | if (new_blk != -1) | |
755 | head_blk = new_blk; | |
756 | } else { /* need to read 2 parts of log */ | |
757 | /* | |
758 | * We are going to scan backwards in the log in two parts. | |
759 | * First we scan the physical end of the log. In this part | |
760 | * of the log, we are looking for blocks with cycle number | |
761 | * last_half_cycle - 1. | |
762 | * If we find one, then we know that the log starts there, as | |
763 | * we've found a hole that didn't get written in going around | |
764 | * the end of the physical log. The simple case for this is | |
765 | * x + 1 ... | x ... | x - 1 | x | |
766 | * <---------> less than scan distance | |
767 | * If all of the blocks at the end of the log have cycle number | |
768 | * last_half_cycle, then we check the blocks at the start of | |
769 | * the log looking for occurrences of last_half_cycle. If we | |
770 | * find one, then our current estimate for the location of the | |
771 | * first occurrence of last_half_cycle is wrong and we move | |
772 | * back to the hole we've found. This case looks like | |
773 | * x + 1 ... | x | x + 1 | x ... | |
774 | * ^ binary search stopped here | |
775 | * Another case we need to handle that only occurs in 256k | |
776 | * logs is | |
777 | * x + 1 ... | x ... | x+1 | x ... | |
778 | * ^ binary search stops here | |
779 | * In a 256k log, the scan at the end of the log will see the | |
780 | * x + 1 blocks. We need to skip past those since that is | |
781 | * certainly not the head of the log. By searching for | |
782 | * last_half_cycle-1 we accomplish that. | |
783 | */ | |
784 | ASSERT(head_blk <= INT_MAX && | |
785 | (xfs_daddr_t) num_scan_bblks >= head_blk); | |
786 | start_blk = log_bbnum - (num_scan_bblks - head_blk); | |
787 | if ((error = xlog_find_verify_cycle(log, start_blk, | |
788 | num_scan_bblks - (int)head_blk, | |
789 | (stop_on_cycle - 1), &new_blk))) | |
790 | goto bp_err; | |
791 | if (new_blk != -1) { | |
792 | head_blk = new_blk; | |
793 | goto validate_head; | |
794 | } | |
795 | ||
796 | /* | |
797 | * Scan beginning of log now. The last part of the physical | |
798 | * log is good. This scan needs to verify that it doesn't find | |
799 | * the last_half_cycle. | |
800 | */ | |
801 | start_blk = 0; | |
802 | ASSERT(head_blk <= INT_MAX); | |
803 | if ((error = xlog_find_verify_cycle(log, | |
804 | start_blk, (int)head_blk, | |
805 | stop_on_cycle, &new_blk))) | |
806 | goto bp_err; | |
807 | if (new_blk != -1) | |
808 | head_blk = new_blk; | |
809 | } | |
810 | ||
811 | validate_head: | |
812 | /* | |
813 | * Now we need to make sure head_blk is not pointing to a block in | |
814 | * the middle of a log record. | |
815 | */ | |
816 | num_scan_bblks = XLOG_REC_SHIFT(log); | |
817 | if (head_blk >= num_scan_bblks) { | |
818 | start_blk = head_blk - num_scan_bblks; /* don't read head_blk */ | |
819 | ||
820 | /* start ptr at last block ptr before head_blk */ | |
821 | error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0); | |
822 | if (error == 1) | |
823 | error = -EIO; | |
824 | if (error) | |
825 | goto bp_err; | |
826 | } else { | |
827 | start_blk = 0; | |
828 | ASSERT(head_blk <= INT_MAX); | |
829 | error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0); | |
830 | if (error < 0) | |
831 | goto bp_err; | |
832 | if (error == 1) { | |
833 | /* We hit the beginning of the log during our search */ | |
834 | start_blk = log_bbnum - (num_scan_bblks - head_blk); | |
835 | new_blk = log_bbnum; | |
836 | ASSERT(start_blk <= INT_MAX && | |
837 | (xfs_daddr_t) log_bbnum-start_blk >= 0); | |
838 | ASSERT(head_blk <= INT_MAX); | |
839 | error = xlog_find_verify_log_record(log, start_blk, | |
840 | &new_blk, (int)head_blk); | |
841 | if (error == 1) | |
842 | error = -EIO; | |
843 | if (error) | |
844 | goto bp_err; | |
845 | if (new_blk != log_bbnum) | |
846 | head_blk = new_blk; | |
847 | } else if (error) | |
848 | goto bp_err; | |
849 | } | |
850 | ||
851 | xlog_put_bp(bp); | |
852 | if (head_blk == log_bbnum) | |
853 | *return_head_blk = 0; | |
854 | else | |
855 | *return_head_blk = head_blk; | |
856 | /* | |
857 | * When returning here, we have a good block number. Bad block | |
858 | * means that during a previous crash, we didn't have a clean break | |
859 | * from cycle number N to cycle number N-1. In this case, we need | |
860 | * to find the first block with cycle number N-1. | |
861 | */ | |
862 | return 0; | |
863 | ||
864 | bp_err: | |
865 | xlog_put_bp(bp); | |
866 | ||
867 | if (error) | |
868 | xfs_warn(log->l_mp, "failed to find log head"); | |
869 | return error; | |
870 | } | |
871 | ||
872 | /* | |
873 | * Seek backwards in the log for log record headers. | |
874 | * | |
875 | * Given a starting log block, walk backwards until we find the provided number | |
876 | * of records or hit the provided tail block. The return value is the number of | |
877 | * records encountered or a negative error code. The log block and buffer | |
878 | * pointer of the last record seen are returned in rblk and rhead respectively. | |
879 | */ | |
880 | STATIC int | |
881 | xlog_rseek_logrec_hdr( | |
882 | struct xlog *log, | |
883 | xfs_daddr_t head_blk, | |
884 | xfs_daddr_t tail_blk, | |
885 | int count, | |
886 | struct xfs_buf *bp, | |
887 | xfs_daddr_t *rblk, | |
888 | struct xlog_rec_header **rhead, | |
889 | bool *wrapped) | |
890 | { | |
891 | int i; | |
892 | int error; | |
893 | int found = 0; | |
894 | char *offset = NULL; | |
895 | xfs_daddr_t end_blk; | |
896 | ||
897 | *wrapped = false; | |
898 | ||
899 | /* | |
900 | * Walk backwards from the head block until we hit the tail or the first | |
901 | * block in the log. | |
902 | */ | |
903 | end_blk = head_blk > tail_blk ? tail_blk : 0; | |
904 | for (i = (int) head_blk - 1; i >= end_blk; i--) { | |
905 | error = xlog_bread(log, i, 1, bp, &offset); | |
906 | if (error) | |
907 | goto out_error; | |
908 | ||
909 | if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { | |
910 | *rblk = i; | |
911 | *rhead = (struct xlog_rec_header *) offset; | |
912 | if (++found == count) | |
913 | break; | |
914 | } | |
915 | } | |
916 | ||
917 | /* | |
918 | * If we haven't hit the tail block or the log record header count, | |
919 | * start looking again from the end of the physical log. Note that | |
920 | * callers can pass head == tail if the tail is not yet known. | |
921 | */ | |
922 | if (tail_blk >= head_blk && found != count) { | |
923 | for (i = log->l_logBBsize - 1; i >= (int) tail_blk; i--) { | |
924 | error = xlog_bread(log, i, 1, bp, &offset); | |
925 | if (error) | |
926 | goto out_error; | |
927 | ||
928 | if (*(__be32 *)offset == | |
929 | cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { | |
930 | *wrapped = true; | |
931 | *rblk = i; | |
932 | *rhead = (struct xlog_rec_header *) offset; | |
933 | if (++found == count) | |
934 | break; | |
935 | } | |
936 | } | |
937 | } | |
938 | ||
939 | return found; | |
940 | ||
941 | out_error: | |
942 | return error; | |
943 | } | |
944 | ||
945 | /* | |
946 | * Seek forward in the log for log record headers. | |
947 | * | |
948 | * Given head and tail blocks, walk forward from the tail block until we find | |
949 | * the provided number of records or hit the head block. The return value is the | |
950 | * number of records encountered or a negative error code. The log block and | |
951 | * buffer pointer of the last record seen are returned in rblk and rhead | |
952 | * respectively. | |
953 | */ | |
954 | STATIC int | |
955 | xlog_seek_logrec_hdr( | |
956 | struct xlog *log, | |
957 | xfs_daddr_t head_blk, | |
958 | xfs_daddr_t tail_blk, | |
959 | int count, | |
960 | struct xfs_buf *bp, | |
961 | xfs_daddr_t *rblk, | |
962 | struct xlog_rec_header **rhead, | |
963 | bool *wrapped) | |
964 | { | |
965 | int i; | |
966 | int error; | |
967 | int found = 0; | |
968 | char *offset = NULL; | |
969 | xfs_daddr_t end_blk; | |
970 | ||
971 | *wrapped = false; | |
972 | ||
973 | /* | |
974 | * Walk forward from the tail block until we hit the head or the last | |
975 | * block in the log. | |
976 | */ | |
977 | end_blk = head_blk > tail_blk ? head_blk : log->l_logBBsize - 1; | |
978 | for (i = (int) tail_blk; i <= end_blk; i++) { | |
979 | error = xlog_bread(log, i, 1, bp, &offset); | |
980 | if (error) | |
981 | goto out_error; | |
982 | ||
983 | if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { | |
984 | *rblk = i; | |
985 | *rhead = (struct xlog_rec_header *) offset; | |
986 | if (++found == count) | |
987 | break; | |
988 | } | |
989 | } | |
990 | ||
991 | /* | |
992 | * If we haven't hit the head block or the log record header count, | |
993 | * start looking again from the start of the physical log. | |
994 | */ | |
995 | if (tail_blk > head_blk && found != count) { | |
996 | for (i = 0; i < (int) head_blk; i++) { | |
997 | error = xlog_bread(log, i, 1, bp, &offset); | |
998 | if (error) | |
999 | goto out_error; | |
1000 | ||
1001 | if (*(__be32 *)offset == | |
1002 | cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { | |
1003 | *wrapped = true; | |
1004 | *rblk = i; | |
1005 | *rhead = (struct xlog_rec_header *) offset; | |
1006 | if (++found == count) | |
1007 | break; | |
1008 | } | |
1009 | } | |
1010 | } | |
1011 | ||
1012 | return found; | |
1013 | ||
1014 | out_error: | |
1015 | return error; | |
1016 | } | |
1017 | ||
1018 | /* | |
1019 | * Check the log tail for torn writes. This is required when torn writes are | |
1020 | * detected at the head and the head had to be walked back to a previous record. | |
1021 | * The tail of the previous record must now be verified to ensure the torn | |
1022 | * writes didn't corrupt the previous tail. | |
1023 | * | |
1024 | * Return an error if CRC verification fails as recovery cannot proceed. | |
1025 | */ | |
1026 | STATIC int | |
1027 | xlog_verify_tail( | |
1028 | struct xlog *log, | |
1029 | xfs_daddr_t head_blk, | |
1030 | xfs_daddr_t tail_blk) | |
1031 | { | |
1032 | struct xlog_rec_header *thead; | |
1033 | struct xfs_buf *bp; | |
1034 | xfs_daddr_t first_bad; | |
1035 | int count; | |
1036 | int error = 0; | |
1037 | bool wrapped; | |
1038 | xfs_daddr_t tmp_head; | |
1039 | ||
1040 | bp = xlog_get_bp(log, 1); | |
1041 | if (!bp) | |
1042 | return -ENOMEM; | |
1043 | ||
1044 | /* | |
1045 | * Seek XLOG_MAX_ICLOGS + 1 records past the current tail record to get | |
1046 | * a temporary head block that points after the last possible | |
1047 | * concurrently written record of the tail. | |
1048 | */ | |
1049 | count = xlog_seek_logrec_hdr(log, head_blk, tail_blk, | |
1050 | XLOG_MAX_ICLOGS + 1, bp, &tmp_head, &thead, | |
1051 | &wrapped); | |
1052 | if (count < 0) { | |
1053 | error = count; | |
1054 | goto out; | |
1055 | } | |
1056 | ||
1057 | /* | |
1058 | * If the call above didn't find XLOG_MAX_ICLOGS + 1 records, we ran | |
1059 | * into the actual log head. tmp_head points to the start of the record | |
1060 | * so update it to the actual head block. | |
1061 | */ | |
1062 | if (count < XLOG_MAX_ICLOGS + 1) | |
1063 | tmp_head = head_blk; | |
1064 | ||
1065 | /* | |
1066 | * We now have a tail and temporary head block that covers at least | |
1067 | * XLOG_MAX_ICLOGS records from the tail. We need to verify that these | |
1068 | * records were completely written. Run a CRC verification pass from | |
1069 | * tail to head and return the result. | |
1070 | */ | |
1071 | error = xlog_do_recovery_pass(log, tmp_head, tail_blk, | |
1072 | XLOG_RECOVER_CRCPASS, &first_bad); | |
1073 | ||
1074 | out: | |
1075 | xlog_put_bp(bp); | |
1076 | return error; | |
1077 | } | |
1078 | ||
1079 | /* | |
1080 | * Detect and trim torn writes from the head of the log. | |
1081 | * | |
1082 | * Storage without sector atomicity guarantees can result in torn writes in the | |
1083 | * log in the event of a crash. Our only means to detect this scenario is via | |
1084 | * CRC verification. While we can't always be certain that CRC verification | |
1085 | * failure is due to a torn write vs. an unrelated corruption, we do know that | |
1086 | * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at | |
1087 | * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of | |
1088 | * the log and treat failures in this range as torn writes as a matter of | |
1089 | * policy. In the event of CRC failure, the head is walked back to the last good | |
1090 | * record in the log and the tail is updated from that record and verified. | |
1091 | */ | |
1092 | STATIC int | |
1093 | xlog_verify_head( | |
1094 | struct xlog *log, | |
1095 | xfs_daddr_t *head_blk, /* in/out: unverified head */ | |
1096 | xfs_daddr_t *tail_blk, /* out: tail block */ | |
1097 | struct xfs_buf *bp, | |
1098 | xfs_daddr_t *rhead_blk, /* start blk of last record */ | |
1099 | struct xlog_rec_header **rhead, /* ptr to last record */ | |
1100 | bool *wrapped) /* last rec. wraps phys. log */ | |
1101 | { | |
1102 | struct xlog_rec_header *tmp_rhead; | |
1103 | struct xfs_buf *tmp_bp; | |
1104 | xfs_daddr_t first_bad; | |
1105 | xfs_daddr_t tmp_rhead_blk; | |
1106 | int found; | |
1107 | int error; | |
1108 | bool tmp_wrapped; | |
1109 | ||
1110 | /* | |
1111 | * Check the head of the log for torn writes. Search backwards from the | |
1112 | * head until we hit the tail or the maximum number of log record I/Os | |
1113 | * that could have been in flight at one time. Use a temporary buffer so | |
1114 | * we don't trash the rhead/bp pointers from the caller. | |
1115 | */ | |
1116 | tmp_bp = xlog_get_bp(log, 1); | |
1117 | if (!tmp_bp) | |
1118 | return -ENOMEM; | |
1119 | error = xlog_rseek_logrec_hdr(log, *head_blk, *tail_blk, | |
1120 | XLOG_MAX_ICLOGS, tmp_bp, &tmp_rhead_blk, | |
1121 | &tmp_rhead, &tmp_wrapped); | |
1122 | xlog_put_bp(tmp_bp); | |
1123 | if (error < 0) | |
1124 | return error; | |
1125 | ||
1126 | /* | |
1127 | * Now run a CRC verification pass over the records starting at the | |
1128 | * block found above to the current head. If a CRC failure occurs, the | |
1129 | * log block of the first bad record is saved in first_bad. | |
1130 | */ | |
1131 | error = xlog_do_recovery_pass(log, *head_blk, tmp_rhead_blk, | |
1132 | XLOG_RECOVER_CRCPASS, &first_bad); | |
1133 | if (error == -EFSBADCRC) { | |
1134 | /* | |
1135 | * We've hit a potential torn write. Reset the error and warn | |
1136 | * about it. | |
1137 | */ | |
1138 | error = 0; | |
1139 | xfs_warn(log->l_mp, | |
1140 | "Torn write (CRC failure) detected at log block 0x%llx. Truncating head block from 0x%llx.", | |
1141 | first_bad, *head_blk); | |
1142 | ||
1143 | /* | |
1144 | * Get the header block and buffer pointer for the last good | |
1145 | * record before the bad record. | |
1146 | * | |
1147 | * Note that xlog_find_tail() clears the blocks at the new head | |
1148 | * (i.e., the records with invalid CRC) if the cycle number | |
1149 | * matches the the current cycle. | |
1150 | */ | |
1151 | found = xlog_rseek_logrec_hdr(log, first_bad, *tail_blk, 1, bp, | |
1152 | rhead_blk, rhead, wrapped); | |
1153 | if (found < 0) | |
1154 | return found; | |
1155 | if (found == 0) /* XXX: right thing to do here? */ | |
1156 | return -EIO; | |
1157 | ||
1158 | /* | |
1159 | * Reset the head block to the starting block of the first bad | |
1160 | * log record and set the tail block based on the last good | |
1161 | * record. | |
1162 | * | |
1163 | * Bail out if the updated head/tail match as this indicates | |
1164 | * possible corruption outside of the acceptable | |
1165 | * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair... | |
1166 | */ | |
1167 | *head_blk = first_bad; | |
1168 | *tail_blk = BLOCK_LSN(be64_to_cpu((*rhead)->h_tail_lsn)); | |
1169 | if (*head_blk == *tail_blk) { | |
1170 | ASSERT(0); | |
1171 | return 0; | |
1172 | } | |
1173 | ||
1174 | /* | |
1175 | * Now verify the tail based on the updated head. This is | |
1176 | * required because the torn writes trimmed from the head could | |
1177 | * have been written over the tail of a previous record. Return | |
1178 | * any errors since recovery cannot proceed if the tail is | |
1179 | * corrupt. | |
1180 | * | |
1181 | * XXX: This leaves a gap in truly robust protection from torn | |
1182 | * writes in the log. If the head is behind the tail, the tail | |
1183 | * pushes forward to create some space and then a crash occurs | |
1184 | * causing the writes into the previous record's tail region to | |
1185 | * tear, log recovery isn't able to recover. | |
1186 | * | |
1187 | * How likely is this to occur? If possible, can we do something | |
1188 | * more intelligent here? Is it safe to push the tail forward if | |
1189 | * we can determine that the tail is within the range of the | |
1190 | * torn write (e.g., the kernel can only overwrite the tail if | |
1191 | * it has actually been pushed forward)? Alternatively, could we | |
1192 | * somehow prevent this condition at runtime? | |
1193 | */ | |
1194 | error = xlog_verify_tail(log, *head_blk, *tail_blk); | |
1195 | } | |
1196 | ||
1197 | return error; | |
1198 | } | |
1199 | ||
1200 | /* | |
1201 | * Check whether the head of the log points to an unmount record. In other | |
1202 | * words, determine whether the log is clean. If so, update the in-core state | |
1203 | * appropriately. | |
1204 | */ | |
1205 | static int | |
1206 | xlog_check_unmount_rec( | |
1207 | struct xlog *log, | |
1208 | xfs_daddr_t *head_blk, | |
1209 | xfs_daddr_t *tail_blk, | |
1210 | struct xlog_rec_header *rhead, | |
1211 | xfs_daddr_t rhead_blk, | |
1212 | struct xfs_buf *bp, | |
1213 | bool *clean) | |
1214 | { | |
1215 | struct xlog_op_header *op_head; | |
1216 | xfs_daddr_t umount_data_blk; | |
1217 | xfs_daddr_t after_umount_blk; | |
1218 | int hblks; | |
1219 | int error; | |
1220 | char *offset; | |
1221 | ||
1222 | *clean = false; | |
1223 | ||
1224 | /* | |
1225 | * Look for unmount record. If we find it, then we know there was a | |
1226 | * clean unmount. Since 'i' could be the last block in the physical | |
1227 | * log, we convert to a log block before comparing to the head_blk. | |
1228 | * | |
1229 | * Save the current tail lsn to use to pass to xlog_clear_stale_blocks() | |
1230 | * below. We won't want to clear the unmount record if there is one, so | |
1231 | * we pass the lsn of the unmount record rather than the block after it. | |
1232 | */ | |
1233 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1234 | int h_size = be32_to_cpu(rhead->h_size); | |
1235 | int h_version = be32_to_cpu(rhead->h_version); | |
1236 | ||
1237 | if ((h_version & XLOG_VERSION_2) && | |
1238 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { | |
1239 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; | |
1240 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
1241 | hblks++; | |
1242 | } else { | |
1243 | hblks = 1; | |
1244 | } | |
1245 | } else { | |
1246 | hblks = 1; | |
1247 | } | |
1248 | after_umount_blk = rhead_blk + hblks + BTOBB(be32_to_cpu(rhead->h_len)); | |
1249 | after_umount_blk = do_mod(after_umount_blk, log->l_logBBsize); | |
1250 | if (*head_blk == after_umount_blk && | |
1251 | be32_to_cpu(rhead->h_num_logops) == 1) { | |
1252 | umount_data_blk = rhead_blk + hblks; | |
1253 | umount_data_blk = do_mod(umount_data_blk, log->l_logBBsize); | |
1254 | error = xlog_bread(log, umount_data_blk, 1, bp, &offset); | |
1255 | if (error) | |
1256 | return error; | |
1257 | ||
1258 | op_head = (struct xlog_op_header *)offset; | |
1259 | if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) { | |
1260 | /* | |
1261 | * Set tail and last sync so that newly written log | |
1262 | * records will point recovery to after the current | |
1263 | * unmount record. | |
1264 | */ | |
1265 | xlog_assign_atomic_lsn(&log->l_tail_lsn, | |
1266 | log->l_curr_cycle, after_umount_blk); | |
1267 | xlog_assign_atomic_lsn(&log->l_last_sync_lsn, | |
1268 | log->l_curr_cycle, after_umount_blk); | |
1269 | *tail_blk = after_umount_blk; | |
1270 | ||
1271 | *clean = true; | |
1272 | } | |
1273 | } | |
1274 | ||
1275 | return 0; | |
1276 | } | |
1277 | ||
1278 | static void | |
1279 | xlog_set_state( | |
1280 | struct xlog *log, | |
1281 | xfs_daddr_t head_blk, | |
1282 | struct xlog_rec_header *rhead, | |
1283 | xfs_daddr_t rhead_blk, | |
1284 | bool bump_cycle) | |
1285 | { | |
1286 | /* | |
1287 | * Reset log values according to the state of the log when we | |
1288 | * crashed. In the case where head_blk == 0, we bump curr_cycle | |
1289 | * one because the next write starts a new cycle rather than | |
1290 | * continuing the cycle of the last good log record. At this | |
1291 | * point we have guaranteed that all partial log records have been | |
1292 | * accounted for. Therefore, we know that the last good log record | |
1293 | * written was complete and ended exactly on the end boundary | |
1294 | * of the physical log. | |
1295 | */ | |
1296 | log->l_prev_block = rhead_blk; | |
1297 | log->l_curr_block = (int)head_blk; | |
1298 | log->l_curr_cycle = be32_to_cpu(rhead->h_cycle); | |
1299 | if (bump_cycle) | |
1300 | log->l_curr_cycle++; | |
1301 | atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn)); | |
1302 | atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn)); | |
1303 | xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle, | |
1304 | BBTOB(log->l_curr_block)); | |
1305 | xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle, | |
1306 | BBTOB(log->l_curr_block)); | |
1307 | } | |
1308 | ||
1309 | /* | |
1310 | * Find the sync block number or the tail of the log. | |
1311 | * | |
1312 | * This will be the block number of the last record to have its | |
1313 | * associated buffers synced to disk. Every log record header has | |
1314 | * a sync lsn embedded in it. LSNs hold block numbers, so it is easy | |
1315 | * to get a sync block number. The only concern is to figure out which | |
1316 | * log record header to believe. | |
1317 | * | |
1318 | * The following algorithm uses the log record header with the largest | |
1319 | * lsn. The entire log record does not need to be valid. We only care | |
1320 | * that the header is valid. | |
1321 | * | |
1322 | * We could speed up search by using current head_blk buffer, but it is not | |
1323 | * available. | |
1324 | */ | |
1325 | STATIC int | |
1326 | xlog_find_tail( | |
1327 | struct xlog *log, | |
1328 | xfs_daddr_t *head_blk, | |
1329 | xfs_daddr_t *tail_blk) | |
1330 | { | |
1331 | xlog_rec_header_t *rhead; | |
1332 | char *offset = NULL; | |
1333 | xfs_buf_t *bp; | |
1334 | int error; | |
1335 | xfs_daddr_t rhead_blk; | |
1336 | xfs_lsn_t tail_lsn; | |
1337 | bool wrapped = false; | |
1338 | bool clean = false; | |
1339 | ||
1340 | /* | |
1341 | * Find previous log record | |
1342 | */ | |
1343 | if ((error = xlog_find_head(log, head_blk))) | |
1344 | return error; | |
1345 | ASSERT(*head_blk < INT_MAX); | |
1346 | ||
1347 | bp = xlog_get_bp(log, 1); | |
1348 | if (!bp) | |
1349 | return -ENOMEM; | |
1350 | if (*head_blk == 0) { /* special case */ | |
1351 | error = xlog_bread(log, 0, 1, bp, &offset); | |
1352 | if (error) | |
1353 | goto done; | |
1354 | ||
1355 | if (xlog_get_cycle(offset) == 0) { | |
1356 | *tail_blk = 0; | |
1357 | /* leave all other log inited values alone */ | |
1358 | goto done; | |
1359 | } | |
1360 | } | |
1361 | ||
1362 | /* | |
1363 | * Search backwards through the log looking for the log record header | |
1364 | * block. This wraps all the way back around to the head so something is | |
1365 | * seriously wrong if we can't find it. | |
1366 | */ | |
1367 | error = xlog_rseek_logrec_hdr(log, *head_blk, *head_blk, 1, bp, | |
1368 | &rhead_blk, &rhead, &wrapped); | |
1369 | if (error < 0) | |
1370 | return error; | |
1371 | if (!error) { | |
1372 | xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__); | |
1373 | return -EIO; | |
1374 | } | |
1375 | *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn)); | |
1376 | ||
1377 | /* | |
1378 | * Set the log state based on the current head record. | |
1379 | */ | |
1380 | xlog_set_state(log, *head_blk, rhead, rhead_blk, wrapped); | |
1381 | tail_lsn = atomic64_read(&log->l_tail_lsn); | |
1382 | ||
1383 | /* | |
1384 | * Look for an unmount record at the head of the log. This sets the log | |
1385 | * state to determine whether recovery is necessary. | |
1386 | */ | |
1387 | error = xlog_check_unmount_rec(log, head_blk, tail_blk, rhead, | |
1388 | rhead_blk, bp, &clean); | |
1389 | if (error) | |
1390 | goto done; | |
1391 | ||
1392 | /* | |
1393 | * Verify the log head if the log is not clean (e.g., we have anything | |
1394 | * but an unmount record at the head). This uses CRC verification to | |
1395 | * detect and trim torn writes. If discovered, CRC failures are | |
1396 | * considered torn writes and the log head is trimmed accordingly. | |
1397 | * | |
1398 | * Note that we can only run CRC verification when the log is dirty | |
1399 | * because there's no guarantee that the log data behind an unmount | |
1400 | * record is compatible with the current architecture. | |
1401 | */ | |
1402 | if (!clean) { | |
1403 | xfs_daddr_t orig_head = *head_blk; | |
1404 | ||
1405 | error = xlog_verify_head(log, head_blk, tail_blk, bp, | |
1406 | &rhead_blk, &rhead, &wrapped); | |
1407 | if (error) | |
1408 | goto done; | |
1409 | ||
1410 | /* update in-core state again if the head changed */ | |
1411 | if (*head_blk != orig_head) { | |
1412 | xlog_set_state(log, *head_blk, rhead, rhead_blk, | |
1413 | wrapped); | |
1414 | tail_lsn = atomic64_read(&log->l_tail_lsn); | |
1415 | error = xlog_check_unmount_rec(log, head_blk, tail_blk, | |
1416 | rhead, rhead_blk, bp, | |
1417 | &clean); | |
1418 | if (error) | |
1419 | goto done; | |
1420 | } | |
1421 | } | |
1422 | ||
1423 | /* | |
1424 | * Note that the unmount was clean. If the unmount was not clean, we | |
1425 | * need to know this to rebuild the superblock counters from the perag | |
1426 | * headers if we have a filesystem using non-persistent counters. | |
1427 | */ | |
1428 | if (clean) | |
1429 | log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN; | |
1430 | ||
1431 | /* | |
1432 | * Make sure that there are no blocks in front of the head | |
1433 | * with the same cycle number as the head. This can happen | |
1434 | * because we allow multiple outstanding log writes concurrently, | |
1435 | * and the later writes might make it out before earlier ones. | |
1436 | * | |
1437 | * We use the lsn from before modifying it so that we'll never | |
1438 | * overwrite the unmount record after a clean unmount. | |
1439 | * | |
1440 | * Do this only if we are going to recover the filesystem | |
1441 | * | |
1442 | * NOTE: This used to say "if (!readonly)" | |
1443 | * However on Linux, we can & do recover a read-only filesystem. | |
1444 | * We only skip recovery if NORECOVERY is specified on mount, | |
1445 | * in which case we would not be here. | |
1446 | * | |
1447 | * But... if the -device- itself is readonly, just skip this. | |
1448 | * We can't recover this device anyway, so it won't matter. | |
1449 | */ | |
1450 | if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) | |
1451 | error = xlog_clear_stale_blocks(log, tail_lsn); | |
1452 | ||
1453 | done: | |
1454 | xlog_put_bp(bp); | |
1455 | ||
1456 | if (error) | |
1457 | xfs_warn(log->l_mp, "failed to locate log tail"); | |
1458 | return error; | |
1459 | } | |
1460 | ||
1461 | /* | |
1462 | * Is the log zeroed at all? | |
1463 | * | |
1464 | * The last binary search should be changed to perform an X block read | |
1465 | * once X becomes small enough. You can then search linearly through | |
1466 | * the X blocks. This will cut down on the number of reads we need to do. | |
1467 | * | |
1468 | * If the log is partially zeroed, this routine will pass back the blkno | |
1469 | * of the first block with cycle number 0. It won't have a complete LR | |
1470 | * preceding it. | |
1471 | * | |
1472 | * Return: | |
1473 | * 0 => the log is completely written to | |
1474 | * 1 => use *blk_no as the first block of the log | |
1475 | * <0 => error has occurred | |
1476 | */ | |
1477 | STATIC int | |
1478 | xlog_find_zeroed( | |
1479 | struct xlog *log, | |
1480 | xfs_daddr_t *blk_no) | |
1481 | { | |
1482 | xfs_buf_t *bp; | |
1483 | char *offset; | |
1484 | uint first_cycle, last_cycle; | |
1485 | xfs_daddr_t new_blk, last_blk, start_blk; | |
1486 | xfs_daddr_t num_scan_bblks; | |
1487 | int error, log_bbnum = log->l_logBBsize; | |
1488 | ||
1489 | *blk_no = 0; | |
1490 | ||
1491 | /* check totally zeroed log */ | |
1492 | bp = xlog_get_bp(log, 1); | |
1493 | if (!bp) | |
1494 | return -ENOMEM; | |
1495 | error = xlog_bread(log, 0, 1, bp, &offset); | |
1496 | if (error) | |
1497 | goto bp_err; | |
1498 | ||
1499 | first_cycle = xlog_get_cycle(offset); | |
1500 | if (first_cycle == 0) { /* completely zeroed log */ | |
1501 | *blk_no = 0; | |
1502 | xlog_put_bp(bp); | |
1503 | return 1; | |
1504 | } | |
1505 | ||
1506 | /* check partially zeroed log */ | |
1507 | error = xlog_bread(log, log_bbnum-1, 1, bp, &offset); | |
1508 | if (error) | |
1509 | goto bp_err; | |
1510 | ||
1511 | last_cycle = xlog_get_cycle(offset); | |
1512 | if (last_cycle != 0) { /* log completely written to */ | |
1513 | xlog_put_bp(bp); | |
1514 | return 0; | |
1515 | } else if (first_cycle != 1) { | |
1516 | /* | |
1517 | * If the cycle of the last block is zero, the cycle of | |
1518 | * the first block must be 1. If it's not, maybe we're | |
1519 | * not looking at a log... Bail out. | |
1520 | */ | |
1521 | xfs_warn(log->l_mp, | |
1522 | "Log inconsistent or not a log (last==0, first!=1)"); | |
1523 | error = -EINVAL; | |
1524 | goto bp_err; | |
1525 | } | |
1526 | ||
1527 | /* we have a partially zeroed log */ | |
1528 | last_blk = log_bbnum-1; | |
1529 | if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0))) | |
1530 | goto bp_err; | |
1531 | ||
1532 | /* | |
1533 | * Validate the answer. Because there is no way to guarantee that | |
1534 | * the entire log is made up of log records which are the same size, | |
1535 | * we scan over the defined maximum blocks. At this point, the maximum | |
1536 | * is not chosen to mean anything special. XXXmiken | |
1537 | */ | |
1538 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); | |
1539 | ASSERT(num_scan_bblks <= INT_MAX); | |
1540 | ||
1541 | if (last_blk < num_scan_bblks) | |
1542 | num_scan_bblks = last_blk; | |
1543 | start_blk = last_blk - num_scan_bblks; | |
1544 | ||
1545 | /* | |
1546 | * We search for any instances of cycle number 0 that occur before | |
1547 | * our current estimate of the head. What we're trying to detect is | |
1548 | * 1 ... | 0 | 1 | 0... | |
1549 | * ^ binary search ends here | |
1550 | */ | |
1551 | if ((error = xlog_find_verify_cycle(log, start_blk, | |
1552 | (int)num_scan_bblks, 0, &new_blk))) | |
1553 | goto bp_err; | |
1554 | if (new_blk != -1) | |
1555 | last_blk = new_blk; | |
1556 | ||
1557 | /* | |
1558 | * Potentially backup over partial log record write. We don't need | |
1559 | * to search the end of the log because we know it is zero. | |
1560 | */ | |
1561 | error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0); | |
1562 | if (error == 1) | |
1563 | error = -EIO; | |
1564 | if (error) | |
1565 | goto bp_err; | |
1566 | ||
1567 | *blk_no = last_blk; | |
1568 | bp_err: | |
1569 | xlog_put_bp(bp); | |
1570 | if (error) | |
1571 | return error; | |
1572 | return 1; | |
1573 | } | |
1574 | ||
1575 | /* | |
1576 | * These are simple subroutines used by xlog_clear_stale_blocks() below | |
1577 | * to initialize a buffer full of empty log record headers and write | |
1578 | * them into the log. | |
1579 | */ | |
1580 | STATIC void | |
1581 | xlog_add_record( | |
1582 | struct xlog *log, | |
1583 | char *buf, | |
1584 | int cycle, | |
1585 | int block, | |
1586 | int tail_cycle, | |
1587 | int tail_block) | |
1588 | { | |
1589 | xlog_rec_header_t *recp = (xlog_rec_header_t *)buf; | |
1590 | ||
1591 | memset(buf, 0, BBSIZE); | |
1592 | recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); | |
1593 | recp->h_cycle = cpu_to_be32(cycle); | |
1594 | recp->h_version = cpu_to_be32( | |
1595 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); | |
1596 | recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block)); | |
1597 | recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block)); | |
1598 | recp->h_fmt = cpu_to_be32(XLOG_FMT); | |
1599 | memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t)); | |
1600 | } | |
1601 | ||
1602 | STATIC int | |
1603 | xlog_write_log_records( | |
1604 | struct xlog *log, | |
1605 | int cycle, | |
1606 | int start_block, | |
1607 | int blocks, | |
1608 | int tail_cycle, | |
1609 | int tail_block) | |
1610 | { | |
1611 | char *offset; | |
1612 | xfs_buf_t *bp; | |
1613 | int balign, ealign; | |
1614 | int sectbb = log->l_sectBBsize; | |
1615 | int end_block = start_block + blocks; | |
1616 | int bufblks; | |
1617 | int error = 0; | |
1618 | int i, j = 0; | |
1619 | ||
1620 | /* | |
1621 | * Greedily allocate a buffer big enough to handle the full | |
1622 | * range of basic blocks to be written. If that fails, try | |
1623 | * a smaller size. We need to be able to write at least a | |
1624 | * log sector, or we're out of luck. | |
1625 | */ | |
1626 | bufblks = 1 << ffs(blocks); | |
1627 | while (bufblks > log->l_logBBsize) | |
1628 | bufblks >>= 1; | |
1629 | while (!(bp = xlog_get_bp(log, bufblks))) { | |
1630 | bufblks >>= 1; | |
1631 | if (bufblks < sectbb) | |
1632 | return -ENOMEM; | |
1633 | } | |
1634 | ||
1635 | /* We may need to do a read at the start to fill in part of | |
1636 | * the buffer in the starting sector not covered by the first | |
1637 | * write below. | |
1638 | */ | |
1639 | balign = round_down(start_block, sectbb); | |
1640 | if (balign != start_block) { | |
1641 | error = xlog_bread_noalign(log, start_block, 1, bp); | |
1642 | if (error) | |
1643 | goto out_put_bp; | |
1644 | ||
1645 | j = start_block - balign; | |
1646 | } | |
1647 | ||
1648 | for (i = start_block; i < end_block; i += bufblks) { | |
1649 | int bcount, endcount; | |
1650 | ||
1651 | bcount = min(bufblks, end_block - start_block); | |
1652 | endcount = bcount - j; | |
1653 | ||
1654 | /* We may need to do a read at the end to fill in part of | |
1655 | * the buffer in the final sector not covered by the write. | |
1656 | * If this is the same sector as the above read, skip it. | |
1657 | */ | |
1658 | ealign = round_down(end_block, sectbb); | |
1659 | if (j == 0 && (start_block + endcount > ealign)) { | |
1660 | offset = bp->b_addr + BBTOB(ealign - start_block); | |
1661 | error = xlog_bread_offset(log, ealign, sectbb, | |
1662 | bp, offset); | |
1663 | if (error) | |
1664 | break; | |
1665 | ||
1666 | } | |
1667 | ||
1668 | offset = xlog_align(log, start_block, endcount, bp); | |
1669 | for (; j < endcount; j++) { | |
1670 | xlog_add_record(log, offset, cycle, i+j, | |
1671 | tail_cycle, tail_block); | |
1672 | offset += BBSIZE; | |
1673 | } | |
1674 | error = xlog_bwrite(log, start_block, endcount, bp); | |
1675 | if (error) | |
1676 | break; | |
1677 | start_block += endcount; | |
1678 | j = 0; | |
1679 | } | |
1680 | ||
1681 | out_put_bp: | |
1682 | xlog_put_bp(bp); | |
1683 | return error; | |
1684 | } | |
1685 | ||
1686 | /* | |
1687 | * This routine is called to blow away any incomplete log writes out | |
1688 | * in front of the log head. We do this so that we won't become confused | |
1689 | * if we come up, write only a little bit more, and then crash again. | |
1690 | * If we leave the partial log records out there, this situation could | |
1691 | * cause us to think those partial writes are valid blocks since they | |
1692 | * have the current cycle number. We get rid of them by overwriting them | |
1693 | * with empty log records with the old cycle number rather than the | |
1694 | * current one. | |
1695 | * | |
1696 | * The tail lsn is passed in rather than taken from | |
1697 | * the log so that we will not write over the unmount record after a | |
1698 | * clean unmount in a 512 block log. Doing so would leave the log without | |
1699 | * any valid log records in it until a new one was written. If we crashed | |
1700 | * during that time we would not be able to recover. | |
1701 | */ | |
1702 | STATIC int | |
1703 | xlog_clear_stale_blocks( | |
1704 | struct xlog *log, | |
1705 | xfs_lsn_t tail_lsn) | |
1706 | { | |
1707 | int tail_cycle, head_cycle; | |
1708 | int tail_block, head_block; | |
1709 | int tail_distance, max_distance; | |
1710 | int distance; | |
1711 | int error; | |
1712 | ||
1713 | tail_cycle = CYCLE_LSN(tail_lsn); | |
1714 | tail_block = BLOCK_LSN(tail_lsn); | |
1715 | head_cycle = log->l_curr_cycle; | |
1716 | head_block = log->l_curr_block; | |
1717 | ||
1718 | /* | |
1719 | * Figure out the distance between the new head of the log | |
1720 | * and the tail. We want to write over any blocks beyond the | |
1721 | * head that we may have written just before the crash, but | |
1722 | * we don't want to overwrite the tail of the log. | |
1723 | */ | |
1724 | if (head_cycle == tail_cycle) { | |
1725 | /* | |
1726 | * The tail is behind the head in the physical log, | |
1727 | * so the distance from the head to the tail is the | |
1728 | * distance from the head to the end of the log plus | |
1729 | * the distance from the beginning of the log to the | |
1730 | * tail. | |
1731 | */ | |
1732 | if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) { | |
1733 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)", | |
1734 | XFS_ERRLEVEL_LOW, log->l_mp); | |
1735 | return -EFSCORRUPTED; | |
1736 | } | |
1737 | tail_distance = tail_block + (log->l_logBBsize - head_block); | |
1738 | } else { | |
1739 | /* | |
1740 | * The head is behind the tail in the physical log, | |
1741 | * so the distance from the head to the tail is just | |
1742 | * the tail block minus the head block. | |
1743 | */ | |
1744 | if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){ | |
1745 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)", | |
1746 | XFS_ERRLEVEL_LOW, log->l_mp); | |
1747 | return -EFSCORRUPTED; | |
1748 | } | |
1749 | tail_distance = tail_block - head_block; | |
1750 | } | |
1751 | ||
1752 | /* | |
1753 | * If the head is right up against the tail, we can't clear | |
1754 | * anything. | |
1755 | */ | |
1756 | if (tail_distance <= 0) { | |
1757 | ASSERT(tail_distance == 0); | |
1758 | return 0; | |
1759 | } | |
1760 | ||
1761 | max_distance = XLOG_TOTAL_REC_SHIFT(log); | |
1762 | /* | |
1763 | * Take the smaller of the maximum amount of outstanding I/O | |
1764 | * we could have and the distance to the tail to clear out. | |
1765 | * We take the smaller so that we don't overwrite the tail and | |
1766 | * we don't waste all day writing from the head to the tail | |
1767 | * for no reason. | |
1768 | */ | |
1769 | max_distance = MIN(max_distance, tail_distance); | |
1770 | ||
1771 | if ((head_block + max_distance) <= log->l_logBBsize) { | |
1772 | /* | |
1773 | * We can stomp all the blocks we need to without | |
1774 | * wrapping around the end of the log. Just do it | |
1775 | * in a single write. Use the cycle number of the | |
1776 | * current cycle minus one so that the log will look like: | |
1777 | * n ... | n - 1 ... | |
1778 | */ | |
1779 | error = xlog_write_log_records(log, (head_cycle - 1), | |
1780 | head_block, max_distance, tail_cycle, | |
1781 | tail_block); | |
1782 | if (error) | |
1783 | return error; | |
1784 | } else { | |
1785 | /* | |
1786 | * We need to wrap around the end of the physical log in | |
1787 | * order to clear all the blocks. Do it in two separate | |
1788 | * I/Os. The first write should be from the head to the | |
1789 | * end of the physical log, and it should use the current | |
1790 | * cycle number minus one just like above. | |
1791 | */ | |
1792 | distance = log->l_logBBsize - head_block; | |
1793 | error = xlog_write_log_records(log, (head_cycle - 1), | |
1794 | head_block, distance, tail_cycle, | |
1795 | tail_block); | |
1796 | ||
1797 | if (error) | |
1798 | return error; | |
1799 | ||
1800 | /* | |
1801 | * Now write the blocks at the start of the physical log. | |
1802 | * This writes the remainder of the blocks we want to clear. | |
1803 | * It uses the current cycle number since we're now on the | |
1804 | * same cycle as the head so that we get: | |
1805 | * n ... n ... | n - 1 ... | |
1806 | * ^^^^^ blocks we're writing | |
1807 | */ | |
1808 | distance = max_distance - (log->l_logBBsize - head_block); | |
1809 | error = xlog_write_log_records(log, head_cycle, 0, distance, | |
1810 | tail_cycle, tail_block); | |
1811 | if (error) | |
1812 | return error; | |
1813 | } | |
1814 | ||
1815 | return 0; | |
1816 | } | |
1817 | ||
1818 | /****************************************************************************** | |
1819 | * | |
1820 | * Log recover routines | |
1821 | * | |
1822 | ****************************************************************************** | |
1823 | */ | |
1824 | ||
1825 | /* | |
1826 | * Sort the log items in the transaction. | |
1827 | * | |
1828 | * The ordering constraints are defined by the inode allocation and unlink | |
1829 | * behaviour. The rules are: | |
1830 | * | |
1831 | * 1. Every item is only logged once in a given transaction. Hence it | |
1832 | * represents the last logged state of the item. Hence ordering is | |
1833 | * dependent on the order in which operations need to be performed so | |
1834 | * required initial conditions are always met. | |
1835 | * | |
1836 | * 2. Cancelled buffers are recorded in pass 1 in a separate table and | |
1837 | * there's nothing to replay from them so we can simply cull them | |
1838 | * from the transaction. However, we can't do that until after we've | |
1839 | * replayed all the other items because they may be dependent on the | |
1840 | * cancelled buffer and replaying the cancelled buffer can remove it | |
1841 | * form the cancelled buffer table. Hence they have tobe done last. | |
1842 | * | |
1843 | * 3. Inode allocation buffers must be replayed before inode items that | |
1844 | * read the buffer and replay changes into it. For filesystems using the | |
1845 | * ICREATE transactions, this means XFS_LI_ICREATE objects need to get | |
1846 | * treated the same as inode allocation buffers as they create and | |
1847 | * initialise the buffers directly. | |
1848 | * | |
1849 | * 4. Inode unlink buffers must be replayed after inode items are replayed. | |
1850 | * This ensures that inodes are completely flushed to the inode buffer | |
1851 | * in a "free" state before we remove the unlinked inode list pointer. | |
1852 | * | |
1853 | * Hence the ordering needs to be inode allocation buffers first, inode items | |
1854 | * second, inode unlink buffers third and cancelled buffers last. | |
1855 | * | |
1856 | * But there's a problem with that - we can't tell an inode allocation buffer | |
1857 | * apart from a regular buffer, so we can't separate them. We can, however, | |
1858 | * tell an inode unlink buffer from the others, and so we can separate them out | |
1859 | * from all the other buffers and move them to last. | |
1860 | * | |
1861 | * Hence, 4 lists, in order from head to tail: | |
1862 | * - buffer_list for all buffers except cancelled/inode unlink buffers | |
1863 | * - item_list for all non-buffer items | |
1864 | * - inode_buffer_list for inode unlink buffers | |
1865 | * - cancel_list for the cancelled buffers | |
1866 | * | |
1867 | * Note that we add objects to the tail of the lists so that first-to-last | |
1868 | * ordering is preserved within the lists. Adding objects to the head of the | |
1869 | * list means when we traverse from the head we walk them in last-to-first | |
1870 | * order. For cancelled buffers and inode unlink buffers this doesn't matter, | |
1871 | * but for all other items there may be specific ordering that we need to | |
1872 | * preserve. | |
1873 | */ | |
1874 | STATIC int | |
1875 | xlog_recover_reorder_trans( | |
1876 | struct xlog *log, | |
1877 | struct xlog_recover *trans, | |
1878 | int pass) | |
1879 | { | |
1880 | xlog_recover_item_t *item, *n; | |
1881 | int error = 0; | |
1882 | LIST_HEAD(sort_list); | |
1883 | LIST_HEAD(cancel_list); | |
1884 | LIST_HEAD(buffer_list); | |
1885 | LIST_HEAD(inode_buffer_list); | |
1886 | LIST_HEAD(inode_list); | |
1887 | ||
1888 | list_splice_init(&trans->r_itemq, &sort_list); | |
1889 | list_for_each_entry_safe(item, n, &sort_list, ri_list) { | |
1890 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; | |
1891 | ||
1892 | switch (ITEM_TYPE(item)) { | |
1893 | case XFS_LI_ICREATE: | |
1894 | list_move_tail(&item->ri_list, &buffer_list); | |
1895 | break; | |
1896 | case XFS_LI_BUF: | |
1897 | if (buf_f->blf_flags & XFS_BLF_CANCEL) { | |
1898 | trace_xfs_log_recover_item_reorder_head(log, | |
1899 | trans, item, pass); | |
1900 | list_move(&item->ri_list, &cancel_list); | |
1901 | break; | |
1902 | } | |
1903 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) { | |
1904 | list_move(&item->ri_list, &inode_buffer_list); | |
1905 | break; | |
1906 | } | |
1907 | list_move_tail(&item->ri_list, &buffer_list); | |
1908 | break; | |
1909 | case XFS_LI_INODE: | |
1910 | case XFS_LI_DQUOT: | |
1911 | case XFS_LI_QUOTAOFF: | |
1912 | case XFS_LI_EFD: | |
1913 | case XFS_LI_EFI: | |
1914 | trace_xfs_log_recover_item_reorder_tail(log, | |
1915 | trans, item, pass); | |
1916 | list_move_tail(&item->ri_list, &inode_list); | |
1917 | break; | |
1918 | default: | |
1919 | xfs_warn(log->l_mp, | |
1920 | "%s: unrecognized type of log operation", | |
1921 | __func__); | |
1922 | ASSERT(0); | |
1923 | /* | |
1924 | * return the remaining items back to the transaction | |
1925 | * item list so they can be freed in caller. | |
1926 | */ | |
1927 | if (!list_empty(&sort_list)) | |
1928 | list_splice_init(&sort_list, &trans->r_itemq); | |
1929 | error = -EIO; | |
1930 | goto out; | |
1931 | } | |
1932 | } | |
1933 | out: | |
1934 | ASSERT(list_empty(&sort_list)); | |
1935 | if (!list_empty(&buffer_list)) | |
1936 | list_splice(&buffer_list, &trans->r_itemq); | |
1937 | if (!list_empty(&inode_list)) | |
1938 | list_splice_tail(&inode_list, &trans->r_itemq); | |
1939 | if (!list_empty(&inode_buffer_list)) | |
1940 | list_splice_tail(&inode_buffer_list, &trans->r_itemq); | |
1941 | if (!list_empty(&cancel_list)) | |
1942 | list_splice_tail(&cancel_list, &trans->r_itemq); | |
1943 | return error; | |
1944 | } | |
1945 | ||
1946 | /* | |
1947 | * Build up the table of buf cancel records so that we don't replay | |
1948 | * cancelled data in the second pass. For buffer records that are | |
1949 | * not cancel records, there is nothing to do here so we just return. | |
1950 | * | |
1951 | * If we get a cancel record which is already in the table, this indicates | |
1952 | * that the buffer was cancelled multiple times. In order to ensure | |
1953 | * that during pass 2 we keep the record in the table until we reach its | |
1954 | * last occurrence in the log, we keep a reference count in the cancel | |
1955 | * record in the table to tell us how many times we expect to see this | |
1956 | * record during the second pass. | |
1957 | */ | |
1958 | STATIC int | |
1959 | xlog_recover_buffer_pass1( | |
1960 | struct xlog *log, | |
1961 | struct xlog_recover_item *item) | |
1962 | { | |
1963 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; | |
1964 | struct list_head *bucket; | |
1965 | struct xfs_buf_cancel *bcp; | |
1966 | ||
1967 | /* | |
1968 | * If this isn't a cancel buffer item, then just return. | |
1969 | */ | |
1970 | if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) { | |
1971 | trace_xfs_log_recover_buf_not_cancel(log, buf_f); | |
1972 | return 0; | |
1973 | } | |
1974 | ||
1975 | /* | |
1976 | * Insert an xfs_buf_cancel record into the hash table of them. | |
1977 | * If there is already an identical record, bump its reference count. | |
1978 | */ | |
1979 | bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno); | |
1980 | list_for_each_entry(bcp, bucket, bc_list) { | |
1981 | if (bcp->bc_blkno == buf_f->blf_blkno && | |
1982 | bcp->bc_len == buf_f->blf_len) { | |
1983 | bcp->bc_refcount++; | |
1984 | trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f); | |
1985 | return 0; | |
1986 | } | |
1987 | } | |
1988 | ||
1989 | bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP); | |
1990 | bcp->bc_blkno = buf_f->blf_blkno; | |
1991 | bcp->bc_len = buf_f->blf_len; | |
1992 | bcp->bc_refcount = 1; | |
1993 | list_add_tail(&bcp->bc_list, bucket); | |
1994 | ||
1995 | trace_xfs_log_recover_buf_cancel_add(log, buf_f); | |
1996 | return 0; | |
1997 | } | |
1998 | ||
1999 | /* | |
2000 | * Check to see whether the buffer being recovered has a corresponding | |
2001 | * entry in the buffer cancel record table. If it is, return the cancel | |
2002 | * buffer structure to the caller. | |
2003 | */ | |
2004 | STATIC struct xfs_buf_cancel * | |
2005 | xlog_peek_buffer_cancelled( | |
2006 | struct xlog *log, | |
2007 | xfs_daddr_t blkno, | |
2008 | uint len, | |
2009 | ushort flags) | |
2010 | { | |
2011 | struct list_head *bucket; | |
2012 | struct xfs_buf_cancel *bcp; | |
2013 | ||
2014 | if (!log->l_buf_cancel_table) { | |
2015 | /* empty table means no cancelled buffers in the log */ | |
2016 | ASSERT(!(flags & XFS_BLF_CANCEL)); | |
2017 | return NULL; | |
2018 | } | |
2019 | ||
2020 | bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno); | |
2021 | list_for_each_entry(bcp, bucket, bc_list) { | |
2022 | if (bcp->bc_blkno == blkno && bcp->bc_len == len) | |
2023 | return bcp; | |
2024 | } | |
2025 | ||
2026 | /* | |
2027 | * We didn't find a corresponding entry in the table, so return 0 so | |
2028 | * that the buffer is NOT cancelled. | |
2029 | */ | |
2030 | ASSERT(!(flags & XFS_BLF_CANCEL)); | |
2031 | return NULL; | |
2032 | } | |
2033 | ||
2034 | /* | |
2035 | * If the buffer is being cancelled then return 1 so that it will be cancelled, | |
2036 | * otherwise return 0. If the buffer is actually a buffer cancel item | |
2037 | * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the | |
2038 | * table and remove it from the table if this is the last reference. | |
2039 | * | |
2040 | * We remove the cancel record from the table when we encounter its last | |
2041 | * occurrence in the log so that if the same buffer is re-used again after its | |
2042 | * last cancellation we actually replay the changes made at that point. | |
2043 | */ | |
2044 | STATIC int | |
2045 | xlog_check_buffer_cancelled( | |
2046 | struct xlog *log, | |
2047 | xfs_daddr_t blkno, | |
2048 | uint len, | |
2049 | ushort flags) | |
2050 | { | |
2051 | struct xfs_buf_cancel *bcp; | |
2052 | ||
2053 | bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags); | |
2054 | if (!bcp) | |
2055 | return 0; | |
2056 | ||
2057 | /* | |
2058 | * We've go a match, so return 1 so that the recovery of this buffer | |
2059 | * is cancelled. If this buffer is actually a buffer cancel log | |
2060 | * item, then decrement the refcount on the one in the table and | |
2061 | * remove it if this is the last reference. | |
2062 | */ | |
2063 | if (flags & XFS_BLF_CANCEL) { | |
2064 | if (--bcp->bc_refcount == 0) { | |
2065 | list_del(&bcp->bc_list); | |
2066 | kmem_free(bcp); | |
2067 | } | |
2068 | } | |
2069 | return 1; | |
2070 | } | |
2071 | ||
2072 | /* | |
2073 | * Perform recovery for a buffer full of inodes. In these buffers, the only | |
2074 | * data which should be recovered is that which corresponds to the | |
2075 | * di_next_unlinked pointers in the on disk inode structures. The rest of the | |
2076 | * data for the inodes is always logged through the inodes themselves rather | |
2077 | * than the inode buffer and is recovered in xlog_recover_inode_pass2(). | |
2078 | * | |
2079 | * The only time when buffers full of inodes are fully recovered is when the | |
2080 | * buffer is full of newly allocated inodes. In this case the buffer will | |
2081 | * not be marked as an inode buffer and so will be sent to | |
2082 | * xlog_recover_do_reg_buffer() below during recovery. | |
2083 | */ | |
2084 | STATIC int | |
2085 | xlog_recover_do_inode_buffer( | |
2086 | struct xfs_mount *mp, | |
2087 | xlog_recover_item_t *item, | |
2088 | struct xfs_buf *bp, | |
2089 | xfs_buf_log_format_t *buf_f) | |
2090 | { | |
2091 | int i; | |
2092 | int item_index = 0; | |
2093 | int bit = 0; | |
2094 | int nbits = 0; | |
2095 | int reg_buf_offset = 0; | |
2096 | int reg_buf_bytes = 0; | |
2097 | int next_unlinked_offset; | |
2098 | int inodes_per_buf; | |
2099 | xfs_agino_t *logged_nextp; | |
2100 | xfs_agino_t *buffer_nextp; | |
2101 | ||
2102 | trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f); | |
2103 | ||
2104 | /* | |
2105 | * Post recovery validation only works properly on CRC enabled | |
2106 | * filesystems. | |
2107 | */ | |
2108 | if (xfs_sb_version_hascrc(&mp->m_sb)) | |
2109 | bp->b_ops = &xfs_inode_buf_ops; | |
2110 | ||
2111 | inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog; | |
2112 | for (i = 0; i < inodes_per_buf; i++) { | |
2113 | next_unlinked_offset = (i * mp->m_sb.sb_inodesize) + | |
2114 | offsetof(xfs_dinode_t, di_next_unlinked); | |
2115 | ||
2116 | while (next_unlinked_offset >= | |
2117 | (reg_buf_offset + reg_buf_bytes)) { | |
2118 | /* | |
2119 | * The next di_next_unlinked field is beyond | |
2120 | * the current logged region. Find the next | |
2121 | * logged region that contains or is beyond | |
2122 | * the current di_next_unlinked field. | |
2123 | */ | |
2124 | bit += nbits; | |
2125 | bit = xfs_next_bit(buf_f->blf_data_map, | |
2126 | buf_f->blf_map_size, bit); | |
2127 | ||
2128 | /* | |
2129 | * If there are no more logged regions in the | |
2130 | * buffer, then we're done. | |
2131 | */ | |
2132 | if (bit == -1) | |
2133 | return 0; | |
2134 | ||
2135 | nbits = xfs_contig_bits(buf_f->blf_data_map, | |
2136 | buf_f->blf_map_size, bit); | |
2137 | ASSERT(nbits > 0); | |
2138 | reg_buf_offset = bit << XFS_BLF_SHIFT; | |
2139 | reg_buf_bytes = nbits << XFS_BLF_SHIFT; | |
2140 | item_index++; | |
2141 | } | |
2142 | ||
2143 | /* | |
2144 | * If the current logged region starts after the current | |
2145 | * di_next_unlinked field, then move on to the next | |
2146 | * di_next_unlinked field. | |
2147 | */ | |
2148 | if (next_unlinked_offset < reg_buf_offset) | |
2149 | continue; | |
2150 | ||
2151 | ASSERT(item->ri_buf[item_index].i_addr != NULL); | |
2152 | ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0); | |
2153 | ASSERT((reg_buf_offset + reg_buf_bytes) <= | |
2154 | BBTOB(bp->b_io_length)); | |
2155 | ||
2156 | /* | |
2157 | * The current logged region contains a copy of the | |
2158 | * current di_next_unlinked field. Extract its value | |
2159 | * and copy it to the buffer copy. | |
2160 | */ | |
2161 | logged_nextp = item->ri_buf[item_index].i_addr + | |
2162 | next_unlinked_offset - reg_buf_offset; | |
2163 | if (unlikely(*logged_nextp == 0)) { | |
2164 | xfs_alert(mp, | |
2165 | "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). " | |
2166 | "Trying to replay bad (0) inode di_next_unlinked field.", | |
2167 | item, bp); | |
2168 | XFS_ERROR_REPORT("xlog_recover_do_inode_buf", | |
2169 | XFS_ERRLEVEL_LOW, mp); | |
2170 | return -EFSCORRUPTED; | |
2171 | } | |
2172 | ||
2173 | buffer_nextp = xfs_buf_offset(bp, next_unlinked_offset); | |
2174 | *buffer_nextp = *logged_nextp; | |
2175 | ||
2176 | /* | |
2177 | * If necessary, recalculate the CRC in the on-disk inode. We | |
2178 | * have to leave the inode in a consistent state for whoever | |
2179 | * reads it next.... | |
2180 | */ | |
2181 | xfs_dinode_calc_crc(mp, | |
2182 | xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize)); | |
2183 | ||
2184 | } | |
2185 | ||
2186 | return 0; | |
2187 | } | |
2188 | ||
2189 | /* | |
2190 | * V5 filesystems know the age of the buffer on disk being recovered. We can | |
2191 | * have newer objects on disk than we are replaying, and so for these cases we | |
2192 | * don't want to replay the current change as that will make the buffer contents | |
2193 | * temporarily invalid on disk. | |
2194 | * | |
2195 | * The magic number might not match the buffer type we are going to recover | |
2196 | * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence | |
2197 | * extract the LSN of the existing object in the buffer based on it's current | |
2198 | * magic number. If we don't recognise the magic number in the buffer, then | |
2199 | * return a LSN of -1 so that the caller knows it was an unrecognised block and | |
2200 | * so can recover the buffer. | |
2201 | * | |
2202 | * Note: we cannot rely solely on magic number matches to determine that the | |
2203 | * buffer has a valid LSN - we also need to verify that it belongs to this | |
2204 | * filesystem, so we need to extract the object's LSN and compare it to that | |
2205 | * which we read from the superblock. If the UUIDs don't match, then we've got a | |
2206 | * stale metadata block from an old filesystem instance that we need to recover | |
2207 | * over the top of. | |
2208 | */ | |
2209 | static xfs_lsn_t | |
2210 | xlog_recover_get_buf_lsn( | |
2211 | struct xfs_mount *mp, | |
2212 | struct xfs_buf *bp) | |
2213 | { | |
2214 | __uint32_t magic32; | |
2215 | __uint16_t magic16; | |
2216 | __uint16_t magicda; | |
2217 | void *blk = bp->b_addr; | |
2218 | uuid_t *uuid; | |
2219 | xfs_lsn_t lsn = -1; | |
2220 | ||
2221 | /* v4 filesystems always recover immediately */ | |
2222 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2223 | goto recover_immediately; | |
2224 | ||
2225 | magic32 = be32_to_cpu(*(__be32 *)blk); | |
2226 | switch (magic32) { | |
2227 | case XFS_ABTB_CRC_MAGIC: | |
2228 | case XFS_ABTC_CRC_MAGIC: | |
2229 | case XFS_ABTB_MAGIC: | |
2230 | case XFS_ABTC_MAGIC: | |
2231 | case XFS_IBT_CRC_MAGIC: | |
2232 | case XFS_IBT_MAGIC: { | |
2233 | struct xfs_btree_block *btb = blk; | |
2234 | ||
2235 | lsn = be64_to_cpu(btb->bb_u.s.bb_lsn); | |
2236 | uuid = &btb->bb_u.s.bb_uuid; | |
2237 | break; | |
2238 | } | |
2239 | case XFS_BMAP_CRC_MAGIC: | |
2240 | case XFS_BMAP_MAGIC: { | |
2241 | struct xfs_btree_block *btb = blk; | |
2242 | ||
2243 | lsn = be64_to_cpu(btb->bb_u.l.bb_lsn); | |
2244 | uuid = &btb->bb_u.l.bb_uuid; | |
2245 | break; | |
2246 | } | |
2247 | case XFS_AGF_MAGIC: | |
2248 | lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn); | |
2249 | uuid = &((struct xfs_agf *)blk)->agf_uuid; | |
2250 | break; | |
2251 | case XFS_AGFL_MAGIC: | |
2252 | lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn); | |
2253 | uuid = &((struct xfs_agfl *)blk)->agfl_uuid; | |
2254 | break; | |
2255 | case XFS_AGI_MAGIC: | |
2256 | lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn); | |
2257 | uuid = &((struct xfs_agi *)blk)->agi_uuid; | |
2258 | break; | |
2259 | case XFS_SYMLINK_MAGIC: | |
2260 | lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn); | |
2261 | uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid; | |
2262 | break; | |
2263 | case XFS_DIR3_BLOCK_MAGIC: | |
2264 | case XFS_DIR3_DATA_MAGIC: | |
2265 | case XFS_DIR3_FREE_MAGIC: | |
2266 | lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn); | |
2267 | uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid; | |
2268 | break; | |
2269 | case XFS_ATTR3_RMT_MAGIC: | |
2270 | /* | |
2271 | * Remote attr blocks are written synchronously, rather than | |
2272 | * being logged. That means they do not contain a valid LSN | |
2273 | * (i.e. transactionally ordered) in them, and hence any time we | |
2274 | * see a buffer to replay over the top of a remote attribute | |
2275 | * block we should simply do so. | |
2276 | */ | |
2277 | goto recover_immediately; | |
2278 | case XFS_SB_MAGIC: | |
2279 | /* | |
2280 | * superblock uuids are magic. We may or may not have a | |
2281 | * sb_meta_uuid on disk, but it will be set in the in-core | |
2282 | * superblock. We set the uuid pointer for verification | |
2283 | * according to the superblock feature mask to ensure we check | |
2284 | * the relevant UUID in the superblock. | |
2285 | */ | |
2286 | lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn); | |
2287 | if (xfs_sb_version_hasmetauuid(&mp->m_sb)) | |
2288 | uuid = &((struct xfs_dsb *)blk)->sb_meta_uuid; | |
2289 | else | |
2290 | uuid = &((struct xfs_dsb *)blk)->sb_uuid; | |
2291 | break; | |
2292 | default: | |
2293 | break; | |
2294 | } | |
2295 | ||
2296 | if (lsn != (xfs_lsn_t)-1) { | |
2297 | if (!uuid_equal(&mp->m_sb.sb_meta_uuid, uuid)) | |
2298 | goto recover_immediately; | |
2299 | return lsn; | |
2300 | } | |
2301 | ||
2302 | magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic); | |
2303 | switch (magicda) { | |
2304 | case XFS_DIR3_LEAF1_MAGIC: | |
2305 | case XFS_DIR3_LEAFN_MAGIC: | |
2306 | case XFS_DA3_NODE_MAGIC: | |
2307 | lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn); | |
2308 | uuid = &((struct xfs_da3_blkinfo *)blk)->uuid; | |
2309 | break; | |
2310 | default: | |
2311 | break; | |
2312 | } | |
2313 | ||
2314 | if (lsn != (xfs_lsn_t)-1) { | |
2315 | if (!uuid_equal(&mp->m_sb.sb_uuid, uuid)) | |
2316 | goto recover_immediately; | |
2317 | return lsn; | |
2318 | } | |
2319 | ||
2320 | /* | |
2321 | * We do individual object checks on dquot and inode buffers as they | |
2322 | * have their own individual LSN records. Also, we could have a stale | |
2323 | * buffer here, so we have to at least recognise these buffer types. | |
2324 | * | |
2325 | * A notd complexity here is inode unlinked list processing - it logs | |
2326 | * the inode directly in the buffer, but we don't know which inodes have | |
2327 | * been modified, and there is no global buffer LSN. Hence we need to | |
2328 | * recover all inode buffer types immediately. This problem will be | |
2329 | * fixed by logical logging of the unlinked list modifications. | |
2330 | */ | |
2331 | magic16 = be16_to_cpu(*(__be16 *)blk); | |
2332 | switch (magic16) { | |
2333 | case XFS_DQUOT_MAGIC: | |
2334 | case XFS_DINODE_MAGIC: | |
2335 | goto recover_immediately; | |
2336 | default: | |
2337 | break; | |
2338 | } | |
2339 | ||
2340 | /* unknown buffer contents, recover immediately */ | |
2341 | ||
2342 | recover_immediately: | |
2343 | return (xfs_lsn_t)-1; | |
2344 | ||
2345 | } | |
2346 | ||
2347 | /* | |
2348 | * Validate the recovered buffer is of the correct type and attach the | |
2349 | * appropriate buffer operations to them for writeback. Magic numbers are in a | |
2350 | * few places: | |
2351 | * the first 16 bits of the buffer (inode buffer, dquot buffer), | |
2352 | * the first 32 bits of the buffer (most blocks), | |
2353 | * inside a struct xfs_da_blkinfo at the start of the buffer. | |
2354 | */ | |
2355 | static void | |
2356 | xlog_recover_validate_buf_type( | |
2357 | struct xfs_mount *mp, | |
2358 | struct xfs_buf *bp, | |
2359 | xfs_buf_log_format_t *buf_f) | |
2360 | { | |
2361 | struct xfs_da_blkinfo *info = bp->b_addr; | |
2362 | __uint32_t magic32; | |
2363 | __uint16_t magic16; | |
2364 | __uint16_t magicda; | |
2365 | ||
2366 | /* | |
2367 | * We can only do post recovery validation on items on CRC enabled | |
2368 | * fielsystems as we need to know when the buffer was written to be able | |
2369 | * to determine if we should have replayed the item. If we replay old | |
2370 | * metadata over a newer buffer, then it will enter a temporarily | |
2371 | * inconsistent state resulting in verification failures. Hence for now | |
2372 | * just avoid the verification stage for non-crc filesystems | |
2373 | */ | |
2374 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2375 | return; | |
2376 | ||
2377 | magic32 = be32_to_cpu(*(__be32 *)bp->b_addr); | |
2378 | magic16 = be16_to_cpu(*(__be16*)bp->b_addr); | |
2379 | magicda = be16_to_cpu(info->magic); | |
2380 | switch (xfs_blft_from_flags(buf_f)) { | |
2381 | case XFS_BLFT_BTREE_BUF: | |
2382 | switch (magic32) { | |
2383 | case XFS_ABTB_CRC_MAGIC: | |
2384 | case XFS_ABTC_CRC_MAGIC: | |
2385 | case XFS_ABTB_MAGIC: | |
2386 | case XFS_ABTC_MAGIC: | |
2387 | bp->b_ops = &xfs_allocbt_buf_ops; | |
2388 | break; | |
2389 | case XFS_IBT_CRC_MAGIC: | |
2390 | case XFS_FIBT_CRC_MAGIC: | |
2391 | case XFS_IBT_MAGIC: | |
2392 | case XFS_FIBT_MAGIC: | |
2393 | bp->b_ops = &xfs_inobt_buf_ops; | |
2394 | break; | |
2395 | case XFS_BMAP_CRC_MAGIC: | |
2396 | case XFS_BMAP_MAGIC: | |
2397 | bp->b_ops = &xfs_bmbt_buf_ops; | |
2398 | break; | |
2399 | default: | |
2400 | xfs_warn(mp, "Bad btree block magic!"); | |
2401 | ASSERT(0); | |
2402 | break; | |
2403 | } | |
2404 | break; | |
2405 | case XFS_BLFT_AGF_BUF: | |
2406 | if (magic32 != XFS_AGF_MAGIC) { | |
2407 | xfs_warn(mp, "Bad AGF block magic!"); | |
2408 | ASSERT(0); | |
2409 | break; | |
2410 | } | |
2411 | bp->b_ops = &xfs_agf_buf_ops; | |
2412 | break; | |
2413 | case XFS_BLFT_AGFL_BUF: | |
2414 | if (magic32 != XFS_AGFL_MAGIC) { | |
2415 | xfs_warn(mp, "Bad AGFL block magic!"); | |
2416 | ASSERT(0); | |
2417 | break; | |
2418 | } | |
2419 | bp->b_ops = &xfs_agfl_buf_ops; | |
2420 | break; | |
2421 | case XFS_BLFT_AGI_BUF: | |
2422 | if (magic32 != XFS_AGI_MAGIC) { | |
2423 | xfs_warn(mp, "Bad AGI block magic!"); | |
2424 | ASSERT(0); | |
2425 | break; | |
2426 | } | |
2427 | bp->b_ops = &xfs_agi_buf_ops; | |
2428 | break; | |
2429 | case XFS_BLFT_UDQUOT_BUF: | |
2430 | case XFS_BLFT_PDQUOT_BUF: | |
2431 | case XFS_BLFT_GDQUOT_BUF: | |
2432 | #ifdef CONFIG_XFS_QUOTA | |
2433 | if (magic16 != XFS_DQUOT_MAGIC) { | |
2434 | xfs_warn(mp, "Bad DQUOT block magic!"); | |
2435 | ASSERT(0); | |
2436 | break; | |
2437 | } | |
2438 | bp->b_ops = &xfs_dquot_buf_ops; | |
2439 | #else | |
2440 | xfs_alert(mp, | |
2441 | "Trying to recover dquots without QUOTA support built in!"); | |
2442 | ASSERT(0); | |
2443 | #endif | |
2444 | break; | |
2445 | case XFS_BLFT_DINO_BUF: | |
2446 | if (magic16 != XFS_DINODE_MAGIC) { | |
2447 | xfs_warn(mp, "Bad INODE block magic!"); | |
2448 | ASSERT(0); | |
2449 | break; | |
2450 | } | |
2451 | bp->b_ops = &xfs_inode_buf_ops; | |
2452 | break; | |
2453 | case XFS_BLFT_SYMLINK_BUF: | |
2454 | if (magic32 != XFS_SYMLINK_MAGIC) { | |
2455 | xfs_warn(mp, "Bad symlink block magic!"); | |
2456 | ASSERT(0); | |
2457 | break; | |
2458 | } | |
2459 | bp->b_ops = &xfs_symlink_buf_ops; | |
2460 | break; | |
2461 | case XFS_BLFT_DIR_BLOCK_BUF: | |
2462 | if (magic32 != XFS_DIR2_BLOCK_MAGIC && | |
2463 | magic32 != XFS_DIR3_BLOCK_MAGIC) { | |
2464 | xfs_warn(mp, "Bad dir block magic!"); | |
2465 | ASSERT(0); | |
2466 | break; | |
2467 | } | |
2468 | bp->b_ops = &xfs_dir3_block_buf_ops; | |
2469 | break; | |
2470 | case XFS_BLFT_DIR_DATA_BUF: | |
2471 | if (magic32 != XFS_DIR2_DATA_MAGIC && | |
2472 | magic32 != XFS_DIR3_DATA_MAGIC) { | |
2473 | xfs_warn(mp, "Bad dir data magic!"); | |
2474 | ASSERT(0); | |
2475 | break; | |
2476 | } | |
2477 | bp->b_ops = &xfs_dir3_data_buf_ops; | |
2478 | break; | |
2479 | case XFS_BLFT_DIR_FREE_BUF: | |
2480 | if (magic32 != XFS_DIR2_FREE_MAGIC && | |
2481 | magic32 != XFS_DIR3_FREE_MAGIC) { | |
2482 | xfs_warn(mp, "Bad dir3 free magic!"); | |
2483 | ASSERT(0); | |
2484 | break; | |
2485 | } | |
2486 | bp->b_ops = &xfs_dir3_free_buf_ops; | |
2487 | break; | |
2488 | case XFS_BLFT_DIR_LEAF1_BUF: | |
2489 | if (magicda != XFS_DIR2_LEAF1_MAGIC && | |
2490 | magicda != XFS_DIR3_LEAF1_MAGIC) { | |
2491 | xfs_warn(mp, "Bad dir leaf1 magic!"); | |
2492 | ASSERT(0); | |
2493 | break; | |
2494 | } | |
2495 | bp->b_ops = &xfs_dir3_leaf1_buf_ops; | |
2496 | break; | |
2497 | case XFS_BLFT_DIR_LEAFN_BUF: | |
2498 | if (magicda != XFS_DIR2_LEAFN_MAGIC && | |
2499 | magicda != XFS_DIR3_LEAFN_MAGIC) { | |
2500 | xfs_warn(mp, "Bad dir leafn magic!"); | |
2501 | ASSERT(0); | |
2502 | break; | |
2503 | } | |
2504 | bp->b_ops = &xfs_dir3_leafn_buf_ops; | |
2505 | break; | |
2506 | case XFS_BLFT_DA_NODE_BUF: | |
2507 | if (magicda != XFS_DA_NODE_MAGIC && | |
2508 | magicda != XFS_DA3_NODE_MAGIC) { | |
2509 | xfs_warn(mp, "Bad da node magic!"); | |
2510 | ASSERT(0); | |
2511 | break; | |
2512 | } | |
2513 | bp->b_ops = &xfs_da3_node_buf_ops; | |
2514 | break; | |
2515 | case XFS_BLFT_ATTR_LEAF_BUF: | |
2516 | if (magicda != XFS_ATTR_LEAF_MAGIC && | |
2517 | magicda != XFS_ATTR3_LEAF_MAGIC) { | |
2518 | xfs_warn(mp, "Bad attr leaf magic!"); | |
2519 | ASSERT(0); | |
2520 | break; | |
2521 | } | |
2522 | bp->b_ops = &xfs_attr3_leaf_buf_ops; | |
2523 | break; | |
2524 | case XFS_BLFT_ATTR_RMT_BUF: | |
2525 | if (magic32 != XFS_ATTR3_RMT_MAGIC) { | |
2526 | xfs_warn(mp, "Bad attr remote magic!"); | |
2527 | ASSERT(0); | |
2528 | break; | |
2529 | } | |
2530 | bp->b_ops = &xfs_attr3_rmt_buf_ops; | |
2531 | break; | |
2532 | case XFS_BLFT_SB_BUF: | |
2533 | if (magic32 != XFS_SB_MAGIC) { | |
2534 | xfs_warn(mp, "Bad SB block magic!"); | |
2535 | ASSERT(0); | |
2536 | break; | |
2537 | } | |
2538 | bp->b_ops = &xfs_sb_buf_ops; | |
2539 | break; | |
2540 | #ifdef CONFIG_XFS_RT | |
2541 | case XFS_BLFT_RTBITMAP_BUF: | |
2542 | case XFS_BLFT_RTSUMMARY_BUF: | |
2543 | /* no magic numbers for verification of RT buffers */ | |
2544 | bp->b_ops = &xfs_rtbuf_ops; | |
2545 | break; | |
2546 | #endif /* CONFIG_XFS_RT */ | |
2547 | default: | |
2548 | xfs_warn(mp, "Unknown buffer type %d!", | |
2549 | xfs_blft_from_flags(buf_f)); | |
2550 | break; | |
2551 | } | |
2552 | } | |
2553 | ||
2554 | /* | |
2555 | * Perform a 'normal' buffer recovery. Each logged region of the | |
2556 | * buffer should be copied over the corresponding region in the | |
2557 | * given buffer. The bitmap in the buf log format structure indicates | |
2558 | * where to place the logged data. | |
2559 | */ | |
2560 | STATIC void | |
2561 | xlog_recover_do_reg_buffer( | |
2562 | struct xfs_mount *mp, | |
2563 | xlog_recover_item_t *item, | |
2564 | struct xfs_buf *bp, | |
2565 | xfs_buf_log_format_t *buf_f) | |
2566 | { | |
2567 | int i; | |
2568 | int bit; | |
2569 | int nbits; | |
2570 | int error; | |
2571 | ||
2572 | trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f); | |
2573 | ||
2574 | bit = 0; | |
2575 | i = 1; /* 0 is the buf format structure */ | |
2576 | while (1) { | |
2577 | bit = xfs_next_bit(buf_f->blf_data_map, | |
2578 | buf_f->blf_map_size, bit); | |
2579 | if (bit == -1) | |
2580 | break; | |
2581 | nbits = xfs_contig_bits(buf_f->blf_data_map, | |
2582 | buf_f->blf_map_size, bit); | |
2583 | ASSERT(nbits > 0); | |
2584 | ASSERT(item->ri_buf[i].i_addr != NULL); | |
2585 | ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0); | |
2586 | ASSERT(BBTOB(bp->b_io_length) >= | |
2587 | ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT)); | |
2588 | ||
2589 | /* | |
2590 | * The dirty regions logged in the buffer, even though | |
2591 | * contiguous, may span multiple chunks. This is because the | |
2592 | * dirty region may span a physical page boundary in a buffer | |
2593 | * and hence be split into two separate vectors for writing into | |
2594 | * the log. Hence we need to trim nbits back to the length of | |
2595 | * the current region being copied out of the log. | |
2596 | */ | |
2597 | if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT)) | |
2598 | nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT; | |
2599 | ||
2600 | /* | |
2601 | * Do a sanity check if this is a dquot buffer. Just checking | |
2602 | * the first dquot in the buffer should do. XXXThis is | |
2603 | * probably a good thing to do for other buf types also. | |
2604 | */ | |
2605 | error = 0; | |
2606 | if (buf_f->blf_flags & | |
2607 | (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) { | |
2608 | if (item->ri_buf[i].i_addr == NULL) { | |
2609 | xfs_alert(mp, | |
2610 | "XFS: NULL dquot in %s.", __func__); | |
2611 | goto next; | |
2612 | } | |
2613 | if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) { | |
2614 | xfs_alert(mp, | |
2615 | "XFS: dquot too small (%d) in %s.", | |
2616 | item->ri_buf[i].i_len, __func__); | |
2617 | goto next; | |
2618 | } | |
2619 | error = xfs_dqcheck(mp, item->ri_buf[i].i_addr, | |
2620 | -1, 0, XFS_QMOPT_DOWARN, | |
2621 | "dquot_buf_recover"); | |
2622 | if (error) | |
2623 | goto next; | |
2624 | } | |
2625 | ||
2626 | memcpy(xfs_buf_offset(bp, | |
2627 | (uint)bit << XFS_BLF_SHIFT), /* dest */ | |
2628 | item->ri_buf[i].i_addr, /* source */ | |
2629 | nbits<<XFS_BLF_SHIFT); /* length */ | |
2630 | next: | |
2631 | i++; | |
2632 | bit += nbits; | |
2633 | } | |
2634 | ||
2635 | /* Shouldn't be any more regions */ | |
2636 | ASSERT(i == item->ri_total); | |
2637 | ||
2638 | xlog_recover_validate_buf_type(mp, bp, buf_f); | |
2639 | } | |
2640 | ||
2641 | /* | |
2642 | * Perform a dquot buffer recovery. | |
2643 | * Simple algorithm: if we have found a QUOTAOFF log item of the same type | |
2644 | * (ie. USR or GRP), then just toss this buffer away; don't recover it. | |
2645 | * Else, treat it as a regular buffer and do recovery. | |
2646 | * | |
2647 | * Return false if the buffer was tossed and true if we recovered the buffer to | |
2648 | * indicate to the caller if the buffer needs writing. | |
2649 | */ | |
2650 | STATIC bool | |
2651 | xlog_recover_do_dquot_buffer( | |
2652 | struct xfs_mount *mp, | |
2653 | struct xlog *log, | |
2654 | struct xlog_recover_item *item, | |
2655 | struct xfs_buf *bp, | |
2656 | struct xfs_buf_log_format *buf_f) | |
2657 | { | |
2658 | uint type; | |
2659 | ||
2660 | trace_xfs_log_recover_buf_dquot_buf(log, buf_f); | |
2661 | ||
2662 | /* | |
2663 | * Filesystems are required to send in quota flags at mount time. | |
2664 | */ | |
2665 | if (!mp->m_qflags) | |
2666 | return false; | |
2667 | ||
2668 | type = 0; | |
2669 | if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF) | |
2670 | type |= XFS_DQ_USER; | |
2671 | if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF) | |
2672 | type |= XFS_DQ_PROJ; | |
2673 | if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF) | |
2674 | type |= XFS_DQ_GROUP; | |
2675 | /* | |
2676 | * This type of quotas was turned off, so ignore this buffer | |
2677 | */ | |
2678 | if (log->l_quotaoffs_flag & type) | |
2679 | return false; | |
2680 | ||
2681 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f); | |
2682 | return true; | |
2683 | } | |
2684 | ||
2685 | /* | |
2686 | * This routine replays a modification made to a buffer at runtime. | |
2687 | * There are actually two types of buffer, regular and inode, which | |
2688 | * are handled differently. Inode buffers are handled differently | |
2689 | * in that we only recover a specific set of data from them, namely | |
2690 | * the inode di_next_unlinked fields. This is because all other inode | |
2691 | * data is actually logged via inode records and any data we replay | |
2692 | * here which overlaps that may be stale. | |
2693 | * | |
2694 | * When meta-data buffers are freed at run time we log a buffer item | |
2695 | * with the XFS_BLF_CANCEL bit set to indicate that previous copies | |
2696 | * of the buffer in the log should not be replayed at recovery time. | |
2697 | * This is so that if the blocks covered by the buffer are reused for | |
2698 | * file data before we crash we don't end up replaying old, freed | |
2699 | * meta-data into a user's file. | |
2700 | * | |
2701 | * To handle the cancellation of buffer log items, we make two passes | |
2702 | * over the log during recovery. During the first we build a table of | |
2703 | * those buffers which have been cancelled, and during the second we | |
2704 | * only replay those buffers which do not have corresponding cancel | |
2705 | * records in the table. See xlog_recover_buffer_pass[1,2] above | |
2706 | * for more details on the implementation of the table of cancel records. | |
2707 | */ | |
2708 | STATIC int | |
2709 | xlog_recover_buffer_pass2( | |
2710 | struct xlog *log, | |
2711 | struct list_head *buffer_list, | |
2712 | struct xlog_recover_item *item, | |
2713 | xfs_lsn_t current_lsn) | |
2714 | { | |
2715 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; | |
2716 | xfs_mount_t *mp = log->l_mp; | |
2717 | xfs_buf_t *bp; | |
2718 | int error; | |
2719 | uint buf_flags; | |
2720 | xfs_lsn_t lsn; | |
2721 | ||
2722 | /* | |
2723 | * In this pass we only want to recover all the buffers which have | |
2724 | * not been cancelled and are not cancellation buffers themselves. | |
2725 | */ | |
2726 | if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno, | |
2727 | buf_f->blf_len, buf_f->blf_flags)) { | |
2728 | trace_xfs_log_recover_buf_cancel(log, buf_f); | |
2729 | return 0; | |
2730 | } | |
2731 | ||
2732 | trace_xfs_log_recover_buf_recover(log, buf_f); | |
2733 | ||
2734 | buf_flags = 0; | |
2735 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) | |
2736 | buf_flags |= XBF_UNMAPPED; | |
2737 | ||
2738 | bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len, | |
2739 | buf_flags, NULL); | |
2740 | if (!bp) | |
2741 | return -ENOMEM; | |
2742 | error = bp->b_error; | |
2743 | if (error) { | |
2744 | xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)"); | |
2745 | goto out_release; | |
2746 | } | |
2747 | ||
2748 | /* | |
2749 | * Recover the buffer only if we get an LSN from it and it's less than | |
2750 | * the lsn of the transaction we are replaying. | |
2751 | * | |
2752 | * Note that we have to be extremely careful of readahead here. | |
2753 | * Readahead does not attach verfiers to the buffers so if we don't | |
2754 | * actually do any replay after readahead because of the LSN we found | |
2755 | * in the buffer if more recent than that current transaction then we | |
2756 | * need to attach the verifier directly. Failure to do so can lead to | |
2757 | * future recovery actions (e.g. EFI and unlinked list recovery) can | |
2758 | * operate on the buffers and they won't get the verifier attached. This | |
2759 | * can lead to blocks on disk having the correct content but a stale | |
2760 | * CRC. | |
2761 | * | |
2762 | * It is safe to assume these clean buffers are currently up to date. | |
2763 | * If the buffer is dirtied by a later transaction being replayed, then | |
2764 | * the verifier will be reset to match whatever recover turns that | |
2765 | * buffer into. | |
2766 | */ | |
2767 | lsn = xlog_recover_get_buf_lsn(mp, bp); | |
2768 | if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) { | |
2769 | xlog_recover_validate_buf_type(mp, bp, buf_f); | |
2770 | goto out_release; | |
2771 | } | |
2772 | ||
2773 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) { | |
2774 | error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f); | |
2775 | if (error) | |
2776 | goto out_release; | |
2777 | } else if (buf_f->blf_flags & | |
2778 | (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) { | |
2779 | bool dirty; | |
2780 | ||
2781 | dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f); | |
2782 | if (!dirty) | |
2783 | goto out_release; | |
2784 | } else { | |
2785 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f); | |
2786 | } | |
2787 | ||
2788 | /* | |
2789 | * Perform delayed write on the buffer. Asynchronous writes will be | |
2790 | * slower when taking into account all the buffers to be flushed. | |
2791 | * | |
2792 | * Also make sure that only inode buffers with good sizes stay in | |
2793 | * the buffer cache. The kernel moves inodes in buffers of 1 block | |
2794 | * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode | |
2795 | * buffers in the log can be a different size if the log was generated | |
2796 | * by an older kernel using unclustered inode buffers or a newer kernel | |
2797 | * running with a different inode cluster size. Regardless, if the | |
2798 | * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size) | |
2799 | * for *our* value of mp->m_inode_cluster_size, then we need to keep | |
2800 | * the buffer out of the buffer cache so that the buffer won't | |
2801 | * overlap with future reads of those inodes. | |
2802 | */ | |
2803 | if (XFS_DINODE_MAGIC == | |
2804 | be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) && | |
2805 | (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize, | |
2806 | (__uint32_t)log->l_mp->m_inode_cluster_size))) { | |
2807 | xfs_buf_stale(bp); | |
2808 | error = xfs_bwrite(bp); | |
2809 | } else { | |
2810 | ASSERT(bp->b_target->bt_mount == mp); | |
2811 | bp->b_iodone = xlog_recover_iodone; | |
2812 | xfs_buf_delwri_queue(bp, buffer_list); | |
2813 | } | |
2814 | ||
2815 | out_release: | |
2816 | xfs_buf_relse(bp); | |
2817 | return error; | |
2818 | } | |
2819 | ||
2820 | /* | |
2821 | * Inode fork owner changes | |
2822 | * | |
2823 | * If we have been told that we have to reparent the inode fork, it's because an | |
2824 | * extent swap operation on a CRC enabled filesystem has been done and we are | |
2825 | * replaying it. We need to walk the BMBT of the appropriate fork and change the | |
2826 | * owners of it. | |
2827 | * | |
2828 | * The complexity here is that we don't have an inode context to work with, so | |
2829 | * after we've replayed the inode we need to instantiate one. This is where the | |
2830 | * fun begins. | |
2831 | * | |
2832 | * We are in the middle of log recovery, so we can't run transactions. That | |
2833 | * means we cannot use cache coherent inode instantiation via xfs_iget(), as | |
2834 | * that will result in the corresponding iput() running the inode through | |
2835 | * xfs_inactive(). If we've just replayed an inode core that changes the link | |
2836 | * count to zero (i.e. it's been unlinked), then xfs_inactive() will run | |
2837 | * transactions (bad!). | |
2838 | * | |
2839 | * So, to avoid this, we instantiate an inode directly from the inode core we've | |
2840 | * just recovered. We have the buffer still locked, and all we really need to | |
2841 | * instantiate is the inode core and the forks being modified. We can do this | |
2842 | * manually, then run the inode btree owner change, and then tear down the | |
2843 | * xfs_inode without having to run any transactions at all. | |
2844 | * | |
2845 | * Also, because we don't have a transaction context available here but need to | |
2846 | * gather all the buffers we modify for writeback so we pass the buffer_list | |
2847 | * instead for the operation to use. | |
2848 | */ | |
2849 | ||
2850 | STATIC int | |
2851 | xfs_recover_inode_owner_change( | |
2852 | struct xfs_mount *mp, | |
2853 | struct xfs_dinode *dip, | |
2854 | struct xfs_inode_log_format *in_f, | |
2855 | struct list_head *buffer_list) | |
2856 | { | |
2857 | struct xfs_inode *ip; | |
2858 | int error; | |
2859 | ||
2860 | ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER)); | |
2861 | ||
2862 | ip = xfs_inode_alloc(mp, in_f->ilf_ino); | |
2863 | if (!ip) | |
2864 | return -ENOMEM; | |
2865 | ||
2866 | /* instantiate the inode */ | |
2867 | xfs_inode_from_disk(ip, dip); | |
2868 | ASSERT(ip->i_d.di_version >= 3); | |
2869 | ||
2870 | error = xfs_iformat_fork(ip, dip); | |
2871 | if (error) | |
2872 | goto out_free_ip; | |
2873 | ||
2874 | ||
2875 | if (in_f->ilf_fields & XFS_ILOG_DOWNER) { | |
2876 | ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT); | |
2877 | error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK, | |
2878 | ip->i_ino, buffer_list); | |
2879 | if (error) | |
2880 | goto out_free_ip; | |
2881 | } | |
2882 | ||
2883 | if (in_f->ilf_fields & XFS_ILOG_AOWNER) { | |
2884 | ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT); | |
2885 | error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK, | |
2886 | ip->i_ino, buffer_list); | |
2887 | if (error) | |
2888 | goto out_free_ip; | |
2889 | } | |
2890 | ||
2891 | out_free_ip: | |
2892 | xfs_inode_free(ip); | |
2893 | return error; | |
2894 | } | |
2895 | ||
2896 | STATIC int | |
2897 | xlog_recover_inode_pass2( | |
2898 | struct xlog *log, | |
2899 | struct list_head *buffer_list, | |
2900 | struct xlog_recover_item *item, | |
2901 | xfs_lsn_t current_lsn) | |
2902 | { | |
2903 | xfs_inode_log_format_t *in_f; | |
2904 | xfs_mount_t *mp = log->l_mp; | |
2905 | xfs_buf_t *bp; | |
2906 | xfs_dinode_t *dip; | |
2907 | int len; | |
2908 | char *src; | |
2909 | char *dest; | |
2910 | int error; | |
2911 | int attr_index; | |
2912 | uint fields; | |
2913 | struct xfs_log_dinode *ldip; | |
2914 | uint isize; | |
2915 | int need_free = 0; | |
2916 | ||
2917 | if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) { | |
2918 | in_f = item->ri_buf[0].i_addr; | |
2919 | } else { | |
2920 | in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP); | |
2921 | need_free = 1; | |
2922 | error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f); | |
2923 | if (error) | |
2924 | goto error; | |
2925 | } | |
2926 | ||
2927 | /* | |
2928 | * Inode buffers can be freed, look out for it, | |
2929 | * and do not replay the inode. | |
2930 | */ | |
2931 | if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno, | |
2932 | in_f->ilf_len, 0)) { | |
2933 | error = 0; | |
2934 | trace_xfs_log_recover_inode_cancel(log, in_f); | |
2935 | goto error; | |
2936 | } | |
2937 | trace_xfs_log_recover_inode_recover(log, in_f); | |
2938 | ||
2939 | bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0, | |
2940 | &xfs_inode_buf_ops); | |
2941 | if (!bp) { | |
2942 | error = -ENOMEM; | |
2943 | goto error; | |
2944 | } | |
2945 | error = bp->b_error; | |
2946 | if (error) { | |
2947 | xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)"); | |
2948 | goto out_release; | |
2949 | } | |
2950 | ASSERT(in_f->ilf_fields & XFS_ILOG_CORE); | |
2951 | dip = xfs_buf_offset(bp, in_f->ilf_boffset); | |
2952 | ||
2953 | /* | |
2954 | * Make sure the place we're flushing out to really looks | |
2955 | * like an inode! | |
2956 | */ | |
2957 | if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) { | |
2958 | xfs_alert(mp, | |
2959 | "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld", | |
2960 | __func__, dip, bp, in_f->ilf_ino); | |
2961 | XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)", | |
2962 | XFS_ERRLEVEL_LOW, mp); | |
2963 | error = -EFSCORRUPTED; | |
2964 | goto out_release; | |
2965 | } | |
2966 | ldip = item->ri_buf[1].i_addr; | |
2967 | if (unlikely(ldip->di_magic != XFS_DINODE_MAGIC)) { | |
2968 | xfs_alert(mp, | |
2969 | "%s: Bad inode log record, rec ptr 0x%p, ino %Ld", | |
2970 | __func__, item, in_f->ilf_ino); | |
2971 | XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)", | |
2972 | XFS_ERRLEVEL_LOW, mp); | |
2973 | error = -EFSCORRUPTED; | |
2974 | goto out_release; | |
2975 | } | |
2976 | ||
2977 | /* | |
2978 | * If the inode has an LSN in it, recover the inode only if it's less | |
2979 | * than the lsn of the transaction we are replaying. Note: we still | |
2980 | * need to replay an owner change even though the inode is more recent | |
2981 | * than the transaction as there is no guarantee that all the btree | |
2982 | * blocks are more recent than this transaction, too. | |
2983 | */ | |
2984 | if (dip->di_version >= 3) { | |
2985 | xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn); | |
2986 | ||
2987 | if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) { | |
2988 | trace_xfs_log_recover_inode_skip(log, in_f); | |
2989 | error = 0; | |
2990 | goto out_owner_change; | |
2991 | } | |
2992 | } | |
2993 | ||
2994 | /* | |
2995 | * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes | |
2996 | * are transactional and if ordering is necessary we can determine that | |
2997 | * more accurately by the LSN field in the V3 inode core. Don't trust | |
2998 | * the inode versions we might be changing them here - use the | |
2999 | * superblock flag to determine whether we need to look at di_flushiter | |
3000 | * to skip replay when the on disk inode is newer than the log one | |
3001 | */ | |
3002 | if (!xfs_sb_version_hascrc(&mp->m_sb) && | |
3003 | ldip->di_flushiter < be16_to_cpu(dip->di_flushiter)) { | |
3004 | /* | |
3005 | * Deal with the wrap case, DI_MAX_FLUSH is less | |
3006 | * than smaller numbers | |
3007 | */ | |
3008 | if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH && | |
3009 | ldip->di_flushiter < (DI_MAX_FLUSH >> 1)) { | |
3010 | /* do nothing */ | |
3011 | } else { | |
3012 | trace_xfs_log_recover_inode_skip(log, in_f); | |
3013 | error = 0; | |
3014 | goto out_release; | |
3015 | } | |
3016 | } | |
3017 | ||
3018 | /* Take the opportunity to reset the flush iteration count */ | |
3019 | ldip->di_flushiter = 0; | |
3020 | ||
3021 | if (unlikely(S_ISREG(ldip->di_mode))) { | |
3022 | if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) && | |
3023 | (ldip->di_format != XFS_DINODE_FMT_BTREE)) { | |
3024 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)", | |
3025 | XFS_ERRLEVEL_LOW, mp, ldip); | |
3026 | xfs_alert(mp, | |
3027 | "%s: Bad regular inode log record, rec ptr 0x%p, " | |
3028 | "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", | |
3029 | __func__, item, dip, bp, in_f->ilf_ino); | |
3030 | error = -EFSCORRUPTED; | |
3031 | goto out_release; | |
3032 | } | |
3033 | } else if (unlikely(S_ISDIR(ldip->di_mode))) { | |
3034 | if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) && | |
3035 | (ldip->di_format != XFS_DINODE_FMT_BTREE) && | |
3036 | (ldip->di_format != XFS_DINODE_FMT_LOCAL)) { | |
3037 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)", | |
3038 | XFS_ERRLEVEL_LOW, mp, ldip); | |
3039 | xfs_alert(mp, | |
3040 | "%s: Bad dir inode log record, rec ptr 0x%p, " | |
3041 | "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", | |
3042 | __func__, item, dip, bp, in_f->ilf_ino); | |
3043 | error = -EFSCORRUPTED; | |
3044 | goto out_release; | |
3045 | } | |
3046 | } | |
3047 | if (unlikely(ldip->di_nextents + ldip->di_anextents > ldip->di_nblocks)){ | |
3048 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)", | |
3049 | XFS_ERRLEVEL_LOW, mp, ldip); | |
3050 | xfs_alert(mp, | |
3051 | "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, " | |
3052 | "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld", | |
3053 | __func__, item, dip, bp, in_f->ilf_ino, | |
3054 | ldip->di_nextents + ldip->di_anextents, | |
3055 | ldip->di_nblocks); | |
3056 | error = -EFSCORRUPTED; | |
3057 | goto out_release; | |
3058 | } | |
3059 | if (unlikely(ldip->di_forkoff > mp->m_sb.sb_inodesize)) { | |
3060 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)", | |
3061 | XFS_ERRLEVEL_LOW, mp, ldip); | |
3062 | xfs_alert(mp, | |
3063 | "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, " | |
3064 | "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__, | |
3065 | item, dip, bp, in_f->ilf_ino, ldip->di_forkoff); | |
3066 | error = -EFSCORRUPTED; | |
3067 | goto out_release; | |
3068 | } | |
3069 | isize = xfs_log_dinode_size(ldip->di_version); | |
3070 | if (unlikely(item->ri_buf[1].i_len > isize)) { | |
3071 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)", | |
3072 | XFS_ERRLEVEL_LOW, mp, ldip); | |
3073 | xfs_alert(mp, | |
3074 | "%s: Bad inode log record length %d, rec ptr 0x%p", | |
3075 | __func__, item->ri_buf[1].i_len, item); | |
3076 | error = -EFSCORRUPTED; | |
3077 | goto out_release; | |
3078 | } | |
3079 | ||
3080 | /* recover the log dinode inode into the on disk inode */ | |
3081 | xfs_log_dinode_to_disk(ldip, dip); | |
3082 | ||
3083 | /* the rest is in on-disk format */ | |
3084 | if (item->ri_buf[1].i_len > isize) { | |
3085 | memcpy((char *)dip + isize, | |
3086 | item->ri_buf[1].i_addr + isize, | |
3087 | item->ri_buf[1].i_len - isize); | |
3088 | } | |
3089 | ||
3090 | fields = in_f->ilf_fields; | |
3091 | switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) { | |
3092 | case XFS_ILOG_DEV: | |
3093 | xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev); | |
3094 | break; | |
3095 | case XFS_ILOG_UUID: | |
3096 | memcpy(XFS_DFORK_DPTR(dip), | |
3097 | &in_f->ilf_u.ilfu_uuid, | |
3098 | sizeof(uuid_t)); | |
3099 | break; | |
3100 | } | |
3101 | ||
3102 | if (in_f->ilf_size == 2) | |
3103 | goto out_owner_change; | |
3104 | len = item->ri_buf[2].i_len; | |
3105 | src = item->ri_buf[2].i_addr; | |
3106 | ASSERT(in_f->ilf_size <= 4); | |
3107 | ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK)); | |
3108 | ASSERT(!(fields & XFS_ILOG_DFORK) || | |
3109 | (len == in_f->ilf_dsize)); | |
3110 | ||
3111 | switch (fields & XFS_ILOG_DFORK) { | |
3112 | case XFS_ILOG_DDATA: | |
3113 | case XFS_ILOG_DEXT: | |
3114 | memcpy(XFS_DFORK_DPTR(dip), src, len); | |
3115 | break; | |
3116 | ||
3117 | case XFS_ILOG_DBROOT: | |
3118 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len, | |
3119 | (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip), | |
3120 | XFS_DFORK_DSIZE(dip, mp)); | |
3121 | break; | |
3122 | ||
3123 | default: | |
3124 | /* | |
3125 | * There are no data fork flags set. | |
3126 | */ | |
3127 | ASSERT((fields & XFS_ILOG_DFORK) == 0); | |
3128 | break; | |
3129 | } | |
3130 | ||
3131 | /* | |
3132 | * If we logged any attribute data, recover it. There may or | |
3133 | * may not have been any other non-core data logged in this | |
3134 | * transaction. | |
3135 | */ | |
3136 | if (in_f->ilf_fields & XFS_ILOG_AFORK) { | |
3137 | if (in_f->ilf_fields & XFS_ILOG_DFORK) { | |
3138 | attr_index = 3; | |
3139 | } else { | |
3140 | attr_index = 2; | |
3141 | } | |
3142 | len = item->ri_buf[attr_index].i_len; | |
3143 | src = item->ri_buf[attr_index].i_addr; | |
3144 | ASSERT(len == in_f->ilf_asize); | |
3145 | ||
3146 | switch (in_f->ilf_fields & XFS_ILOG_AFORK) { | |
3147 | case XFS_ILOG_ADATA: | |
3148 | case XFS_ILOG_AEXT: | |
3149 | dest = XFS_DFORK_APTR(dip); | |
3150 | ASSERT(len <= XFS_DFORK_ASIZE(dip, mp)); | |
3151 | memcpy(dest, src, len); | |
3152 | break; | |
3153 | ||
3154 | case XFS_ILOG_ABROOT: | |
3155 | dest = XFS_DFORK_APTR(dip); | |
3156 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, | |
3157 | len, (xfs_bmdr_block_t*)dest, | |
3158 | XFS_DFORK_ASIZE(dip, mp)); | |
3159 | break; | |
3160 | ||
3161 | default: | |
3162 | xfs_warn(log->l_mp, "%s: Invalid flag", __func__); | |
3163 | ASSERT(0); | |
3164 | error = -EIO; | |
3165 | goto out_release; | |
3166 | } | |
3167 | } | |
3168 | ||
3169 | out_owner_change: | |
3170 | if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER)) | |
3171 | error = xfs_recover_inode_owner_change(mp, dip, in_f, | |
3172 | buffer_list); | |
3173 | /* re-generate the checksum. */ | |
3174 | xfs_dinode_calc_crc(log->l_mp, dip); | |
3175 | ||
3176 | ASSERT(bp->b_target->bt_mount == mp); | |
3177 | bp->b_iodone = xlog_recover_iodone; | |
3178 | xfs_buf_delwri_queue(bp, buffer_list); | |
3179 | ||
3180 | out_release: | |
3181 | xfs_buf_relse(bp); | |
3182 | error: | |
3183 | if (need_free) | |
3184 | kmem_free(in_f); | |
3185 | return error; | |
3186 | } | |
3187 | ||
3188 | /* | |
3189 | * Recover QUOTAOFF records. We simply make a note of it in the xlog | |
3190 | * structure, so that we know not to do any dquot item or dquot buffer recovery, | |
3191 | * of that type. | |
3192 | */ | |
3193 | STATIC int | |
3194 | xlog_recover_quotaoff_pass1( | |
3195 | struct xlog *log, | |
3196 | struct xlog_recover_item *item) | |
3197 | { | |
3198 | xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr; | |
3199 | ASSERT(qoff_f); | |
3200 | ||
3201 | /* | |
3202 | * The logitem format's flag tells us if this was user quotaoff, | |
3203 | * group/project quotaoff or both. | |
3204 | */ | |
3205 | if (qoff_f->qf_flags & XFS_UQUOTA_ACCT) | |
3206 | log->l_quotaoffs_flag |= XFS_DQ_USER; | |
3207 | if (qoff_f->qf_flags & XFS_PQUOTA_ACCT) | |
3208 | log->l_quotaoffs_flag |= XFS_DQ_PROJ; | |
3209 | if (qoff_f->qf_flags & XFS_GQUOTA_ACCT) | |
3210 | log->l_quotaoffs_flag |= XFS_DQ_GROUP; | |
3211 | ||
3212 | return 0; | |
3213 | } | |
3214 | ||
3215 | /* | |
3216 | * Recover a dquot record | |
3217 | */ | |
3218 | STATIC int | |
3219 | xlog_recover_dquot_pass2( | |
3220 | struct xlog *log, | |
3221 | struct list_head *buffer_list, | |
3222 | struct xlog_recover_item *item, | |
3223 | xfs_lsn_t current_lsn) | |
3224 | { | |
3225 | xfs_mount_t *mp = log->l_mp; | |
3226 | xfs_buf_t *bp; | |
3227 | struct xfs_disk_dquot *ddq, *recddq; | |
3228 | int error; | |
3229 | xfs_dq_logformat_t *dq_f; | |
3230 | uint type; | |
3231 | ||
3232 | ||
3233 | /* | |
3234 | * Filesystems are required to send in quota flags at mount time. | |
3235 | */ | |
3236 | if (mp->m_qflags == 0) | |
3237 | return 0; | |
3238 | ||
3239 | recddq = item->ri_buf[1].i_addr; | |
3240 | if (recddq == NULL) { | |
3241 | xfs_alert(log->l_mp, "NULL dquot in %s.", __func__); | |
3242 | return -EIO; | |
3243 | } | |
3244 | if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) { | |
3245 | xfs_alert(log->l_mp, "dquot too small (%d) in %s.", | |
3246 | item->ri_buf[1].i_len, __func__); | |
3247 | return -EIO; | |
3248 | } | |
3249 | ||
3250 | /* | |
3251 | * This type of quotas was turned off, so ignore this record. | |
3252 | */ | |
3253 | type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP); | |
3254 | ASSERT(type); | |
3255 | if (log->l_quotaoffs_flag & type) | |
3256 | return 0; | |
3257 | ||
3258 | /* | |
3259 | * At this point we know that quota was _not_ turned off. | |
3260 | * Since the mount flags are not indicating to us otherwise, this | |
3261 | * must mean that quota is on, and the dquot needs to be replayed. | |
3262 | * Remember that we may not have fully recovered the superblock yet, | |
3263 | * so we can't do the usual trick of looking at the SB quota bits. | |
3264 | * | |
3265 | * The other possibility, of course, is that the quota subsystem was | |
3266 | * removed since the last mount - ENOSYS. | |
3267 | */ | |
3268 | dq_f = item->ri_buf[0].i_addr; | |
3269 | ASSERT(dq_f); | |
3270 | error = xfs_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN, | |
3271 | "xlog_recover_dquot_pass2 (log copy)"); | |
3272 | if (error) | |
3273 | return -EIO; | |
3274 | ASSERT(dq_f->qlf_len == 1); | |
3275 | ||
3276 | /* | |
3277 | * At this point we are assuming that the dquots have been allocated | |
3278 | * and hence the buffer has valid dquots stamped in it. It should, | |
3279 | * therefore, pass verifier validation. If the dquot is bad, then the | |
3280 | * we'll return an error here, so we don't need to specifically check | |
3281 | * the dquot in the buffer after the verifier has run. | |
3282 | */ | |
3283 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno, | |
3284 | XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp, | |
3285 | &xfs_dquot_buf_ops); | |
3286 | if (error) | |
3287 | return error; | |
3288 | ||
3289 | ASSERT(bp); | |
3290 | ddq = xfs_buf_offset(bp, dq_f->qlf_boffset); | |
3291 | ||
3292 | /* | |
3293 | * If the dquot has an LSN in it, recover the dquot only if it's less | |
3294 | * than the lsn of the transaction we are replaying. | |
3295 | */ | |
3296 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
3297 | struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq; | |
3298 | xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn); | |
3299 | ||
3300 | if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) { | |
3301 | goto out_release; | |
3302 | } | |
3303 | } | |
3304 | ||
3305 | memcpy(ddq, recddq, item->ri_buf[1].i_len); | |
3306 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
3307 | xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk), | |
3308 | XFS_DQUOT_CRC_OFF); | |
3309 | } | |
3310 | ||
3311 | ASSERT(dq_f->qlf_size == 2); | |
3312 | ASSERT(bp->b_target->bt_mount == mp); | |
3313 | bp->b_iodone = xlog_recover_iodone; | |
3314 | xfs_buf_delwri_queue(bp, buffer_list); | |
3315 | ||
3316 | out_release: | |
3317 | xfs_buf_relse(bp); | |
3318 | return 0; | |
3319 | } | |
3320 | ||
3321 | /* | |
3322 | * This routine is called to create an in-core extent free intent | |
3323 | * item from the efi format structure which was logged on disk. | |
3324 | * It allocates an in-core efi, copies the extents from the format | |
3325 | * structure into it, and adds the efi to the AIL with the given | |
3326 | * LSN. | |
3327 | */ | |
3328 | STATIC int | |
3329 | xlog_recover_efi_pass2( | |
3330 | struct xlog *log, | |
3331 | struct xlog_recover_item *item, | |
3332 | xfs_lsn_t lsn) | |
3333 | { | |
3334 | int error; | |
3335 | struct xfs_mount *mp = log->l_mp; | |
3336 | struct xfs_efi_log_item *efip; | |
3337 | struct xfs_efi_log_format *efi_formatp; | |
3338 | ||
3339 | efi_formatp = item->ri_buf[0].i_addr; | |
3340 | ||
3341 | efip = xfs_efi_init(mp, efi_formatp->efi_nextents); | |
3342 | error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format); | |
3343 | if (error) { | |
3344 | xfs_efi_item_free(efip); | |
3345 | return error; | |
3346 | } | |
3347 | atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents); | |
3348 | ||
3349 | spin_lock(&log->l_ailp->xa_lock); | |
3350 | /* | |
3351 | * The EFI has two references. One for the EFD and one for EFI to ensure | |
3352 | * it makes it into the AIL. Insert the EFI into the AIL directly and | |
3353 | * drop the EFI reference. Note that xfs_trans_ail_update() drops the | |
3354 | * AIL lock. | |
3355 | */ | |
3356 | xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn); | |
3357 | xfs_efi_release(efip); | |
3358 | return 0; | |
3359 | } | |
3360 | ||
3361 | ||
3362 | /* | |
3363 | * This routine is called when an EFD format structure is found in a committed | |
3364 | * transaction in the log. Its purpose is to cancel the corresponding EFI if it | |
3365 | * was still in the log. To do this it searches the AIL for the EFI with an id | |
3366 | * equal to that in the EFD format structure. If we find it we drop the EFD | |
3367 | * reference, which removes the EFI from the AIL and frees it. | |
3368 | */ | |
3369 | STATIC int | |
3370 | xlog_recover_efd_pass2( | |
3371 | struct xlog *log, | |
3372 | struct xlog_recover_item *item) | |
3373 | { | |
3374 | xfs_efd_log_format_t *efd_formatp; | |
3375 | xfs_efi_log_item_t *efip = NULL; | |
3376 | xfs_log_item_t *lip; | |
3377 | __uint64_t efi_id; | |
3378 | struct xfs_ail_cursor cur; | |
3379 | struct xfs_ail *ailp = log->l_ailp; | |
3380 | ||
3381 | efd_formatp = item->ri_buf[0].i_addr; | |
3382 | ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) + | |
3383 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) || | |
3384 | (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) + | |
3385 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t))))); | |
3386 | efi_id = efd_formatp->efd_efi_id; | |
3387 | ||
3388 | /* | |
3389 | * Search for the EFI with the id in the EFD format structure in the | |
3390 | * AIL. | |
3391 | */ | |
3392 | spin_lock(&ailp->xa_lock); | |
3393 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); | |
3394 | while (lip != NULL) { | |
3395 | if (lip->li_type == XFS_LI_EFI) { | |
3396 | efip = (xfs_efi_log_item_t *)lip; | |
3397 | if (efip->efi_format.efi_id == efi_id) { | |
3398 | /* | |
3399 | * Drop the EFD reference to the EFI. This | |
3400 | * removes the EFI from the AIL and frees it. | |
3401 | */ | |
3402 | spin_unlock(&ailp->xa_lock); | |
3403 | xfs_efi_release(efip); | |
3404 | spin_lock(&ailp->xa_lock); | |
3405 | break; | |
3406 | } | |
3407 | } | |
3408 | lip = xfs_trans_ail_cursor_next(ailp, &cur); | |
3409 | } | |
3410 | ||
3411 | xfs_trans_ail_cursor_done(&cur); | |
3412 | spin_unlock(&ailp->xa_lock); | |
3413 | ||
3414 | return 0; | |
3415 | } | |
3416 | ||
3417 | /* | |
3418 | * This routine is called when an inode create format structure is found in a | |
3419 | * committed transaction in the log. It's purpose is to initialise the inodes | |
3420 | * being allocated on disk. This requires us to get inode cluster buffers that | |
3421 | * match the range to be intialised, stamped with inode templates and written | |
3422 | * by delayed write so that subsequent modifications will hit the cached buffer | |
3423 | * and only need writing out at the end of recovery. | |
3424 | */ | |
3425 | STATIC int | |
3426 | xlog_recover_do_icreate_pass2( | |
3427 | struct xlog *log, | |
3428 | struct list_head *buffer_list, | |
3429 | xlog_recover_item_t *item) | |
3430 | { | |
3431 | struct xfs_mount *mp = log->l_mp; | |
3432 | struct xfs_icreate_log *icl; | |
3433 | xfs_agnumber_t agno; | |
3434 | xfs_agblock_t agbno; | |
3435 | unsigned int count; | |
3436 | unsigned int isize; | |
3437 | xfs_agblock_t length; | |
3438 | int blks_per_cluster; | |
3439 | int bb_per_cluster; | |
3440 | int cancel_count; | |
3441 | int nbufs; | |
3442 | int i; | |
3443 | ||
3444 | icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr; | |
3445 | if (icl->icl_type != XFS_LI_ICREATE) { | |
3446 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type"); | |
3447 | return -EINVAL; | |
3448 | } | |
3449 | ||
3450 | if (icl->icl_size != 1) { | |
3451 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size"); | |
3452 | return -EINVAL; | |
3453 | } | |
3454 | ||
3455 | agno = be32_to_cpu(icl->icl_ag); | |
3456 | if (agno >= mp->m_sb.sb_agcount) { | |
3457 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno"); | |
3458 | return -EINVAL; | |
3459 | } | |
3460 | agbno = be32_to_cpu(icl->icl_agbno); | |
3461 | if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) { | |
3462 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno"); | |
3463 | return -EINVAL; | |
3464 | } | |
3465 | isize = be32_to_cpu(icl->icl_isize); | |
3466 | if (isize != mp->m_sb.sb_inodesize) { | |
3467 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize"); | |
3468 | return -EINVAL; | |
3469 | } | |
3470 | count = be32_to_cpu(icl->icl_count); | |
3471 | if (!count) { | |
3472 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count"); | |
3473 | return -EINVAL; | |
3474 | } | |
3475 | length = be32_to_cpu(icl->icl_length); | |
3476 | if (!length || length >= mp->m_sb.sb_agblocks) { | |
3477 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length"); | |
3478 | return -EINVAL; | |
3479 | } | |
3480 | ||
3481 | /* | |
3482 | * The inode chunk is either full or sparse and we only support | |
3483 | * m_ialloc_min_blks sized sparse allocations at this time. | |
3484 | */ | |
3485 | if (length != mp->m_ialloc_blks && | |
3486 | length != mp->m_ialloc_min_blks) { | |
3487 | xfs_warn(log->l_mp, | |
3488 | "%s: unsupported chunk length", __FUNCTION__); | |
3489 | return -EINVAL; | |
3490 | } | |
3491 | ||
3492 | /* verify inode count is consistent with extent length */ | |
3493 | if ((count >> mp->m_sb.sb_inopblog) != length) { | |
3494 | xfs_warn(log->l_mp, | |
3495 | "%s: inconsistent inode count and chunk length", | |
3496 | __FUNCTION__); | |
3497 | return -EINVAL; | |
3498 | } | |
3499 | ||
3500 | /* | |
3501 | * The icreate transaction can cover multiple cluster buffers and these | |
3502 | * buffers could have been freed and reused. Check the individual | |
3503 | * buffers for cancellation so we don't overwrite anything written after | |
3504 | * a cancellation. | |
3505 | */ | |
3506 | blks_per_cluster = xfs_icluster_size_fsb(mp); | |
3507 | bb_per_cluster = XFS_FSB_TO_BB(mp, blks_per_cluster); | |
3508 | nbufs = length / blks_per_cluster; | |
3509 | for (i = 0, cancel_count = 0; i < nbufs; i++) { | |
3510 | xfs_daddr_t daddr; | |
3511 | ||
3512 | daddr = XFS_AGB_TO_DADDR(mp, agno, | |
3513 | agbno + i * blks_per_cluster); | |
3514 | if (xlog_check_buffer_cancelled(log, daddr, bb_per_cluster, 0)) | |
3515 | cancel_count++; | |
3516 | } | |
3517 | ||
3518 | /* | |
3519 | * We currently only use icreate for a single allocation at a time. This | |
3520 | * means we should expect either all or none of the buffers to be | |
3521 | * cancelled. Be conservative and skip replay if at least one buffer is | |
3522 | * cancelled, but warn the user that something is awry if the buffers | |
3523 | * are not consistent. | |
3524 | * | |
3525 | * XXX: This must be refined to only skip cancelled clusters once we use | |
3526 | * icreate for multiple chunk allocations. | |
3527 | */ | |
3528 | ASSERT(!cancel_count || cancel_count == nbufs); | |
3529 | if (cancel_count) { | |
3530 | if (cancel_count != nbufs) | |
3531 | xfs_warn(mp, | |
3532 | "WARNING: partial inode chunk cancellation, skipped icreate."); | |
3533 | trace_xfs_log_recover_icreate_cancel(log, icl); | |
3534 | return 0; | |
3535 | } | |
3536 | ||
3537 | trace_xfs_log_recover_icreate_recover(log, icl); | |
3538 | return xfs_ialloc_inode_init(mp, NULL, buffer_list, count, agno, agbno, | |
3539 | length, be32_to_cpu(icl->icl_gen)); | |
3540 | } | |
3541 | ||
3542 | STATIC void | |
3543 | xlog_recover_buffer_ra_pass2( | |
3544 | struct xlog *log, | |
3545 | struct xlog_recover_item *item) | |
3546 | { | |
3547 | struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr; | |
3548 | struct xfs_mount *mp = log->l_mp; | |
3549 | ||
3550 | if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno, | |
3551 | buf_f->blf_len, buf_f->blf_flags)) { | |
3552 | return; | |
3553 | } | |
3554 | ||
3555 | xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno, | |
3556 | buf_f->blf_len, NULL); | |
3557 | } | |
3558 | ||
3559 | STATIC void | |
3560 | xlog_recover_inode_ra_pass2( | |
3561 | struct xlog *log, | |
3562 | struct xlog_recover_item *item) | |
3563 | { | |
3564 | struct xfs_inode_log_format ilf_buf; | |
3565 | struct xfs_inode_log_format *ilfp; | |
3566 | struct xfs_mount *mp = log->l_mp; | |
3567 | int error; | |
3568 | ||
3569 | if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) { | |
3570 | ilfp = item->ri_buf[0].i_addr; | |
3571 | } else { | |
3572 | ilfp = &ilf_buf; | |
3573 | memset(ilfp, 0, sizeof(*ilfp)); | |
3574 | error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp); | |
3575 | if (error) | |
3576 | return; | |
3577 | } | |
3578 | ||
3579 | if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0)) | |
3580 | return; | |
3581 | ||
3582 | xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno, | |
3583 | ilfp->ilf_len, &xfs_inode_buf_ra_ops); | |
3584 | } | |
3585 | ||
3586 | STATIC void | |
3587 | xlog_recover_dquot_ra_pass2( | |
3588 | struct xlog *log, | |
3589 | struct xlog_recover_item *item) | |
3590 | { | |
3591 | struct xfs_mount *mp = log->l_mp; | |
3592 | struct xfs_disk_dquot *recddq; | |
3593 | struct xfs_dq_logformat *dq_f; | |
3594 | uint type; | |
3595 | int len; | |
3596 | ||
3597 | ||
3598 | if (mp->m_qflags == 0) | |
3599 | return; | |
3600 | ||
3601 | recddq = item->ri_buf[1].i_addr; | |
3602 | if (recddq == NULL) | |
3603 | return; | |
3604 | if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot)) | |
3605 | return; | |
3606 | ||
3607 | type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP); | |
3608 | ASSERT(type); | |
3609 | if (log->l_quotaoffs_flag & type) | |
3610 | return; | |
3611 | ||
3612 | dq_f = item->ri_buf[0].i_addr; | |
3613 | ASSERT(dq_f); | |
3614 | ASSERT(dq_f->qlf_len == 1); | |
3615 | ||
3616 | len = XFS_FSB_TO_BB(mp, dq_f->qlf_len); | |
3617 | if (xlog_peek_buffer_cancelled(log, dq_f->qlf_blkno, len, 0)) | |
3618 | return; | |
3619 | ||
3620 | xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno, len, | |
3621 | &xfs_dquot_buf_ra_ops); | |
3622 | } | |
3623 | ||
3624 | STATIC void | |
3625 | xlog_recover_ra_pass2( | |
3626 | struct xlog *log, | |
3627 | struct xlog_recover_item *item) | |
3628 | { | |
3629 | switch (ITEM_TYPE(item)) { | |
3630 | case XFS_LI_BUF: | |
3631 | xlog_recover_buffer_ra_pass2(log, item); | |
3632 | break; | |
3633 | case XFS_LI_INODE: | |
3634 | xlog_recover_inode_ra_pass2(log, item); | |
3635 | break; | |
3636 | case XFS_LI_DQUOT: | |
3637 | xlog_recover_dquot_ra_pass2(log, item); | |
3638 | break; | |
3639 | case XFS_LI_EFI: | |
3640 | case XFS_LI_EFD: | |
3641 | case XFS_LI_QUOTAOFF: | |
3642 | default: | |
3643 | break; | |
3644 | } | |
3645 | } | |
3646 | ||
3647 | STATIC int | |
3648 | xlog_recover_commit_pass1( | |
3649 | struct xlog *log, | |
3650 | struct xlog_recover *trans, | |
3651 | struct xlog_recover_item *item) | |
3652 | { | |
3653 | trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1); | |
3654 | ||
3655 | switch (ITEM_TYPE(item)) { | |
3656 | case XFS_LI_BUF: | |
3657 | return xlog_recover_buffer_pass1(log, item); | |
3658 | case XFS_LI_QUOTAOFF: | |
3659 | return xlog_recover_quotaoff_pass1(log, item); | |
3660 | case XFS_LI_INODE: | |
3661 | case XFS_LI_EFI: | |
3662 | case XFS_LI_EFD: | |
3663 | case XFS_LI_DQUOT: | |
3664 | case XFS_LI_ICREATE: | |
3665 | /* nothing to do in pass 1 */ | |
3666 | return 0; | |
3667 | default: | |
3668 | xfs_warn(log->l_mp, "%s: invalid item type (%d)", | |
3669 | __func__, ITEM_TYPE(item)); | |
3670 | ASSERT(0); | |
3671 | return -EIO; | |
3672 | } | |
3673 | } | |
3674 | ||
3675 | STATIC int | |
3676 | xlog_recover_commit_pass2( | |
3677 | struct xlog *log, | |
3678 | struct xlog_recover *trans, | |
3679 | struct list_head *buffer_list, | |
3680 | struct xlog_recover_item *item) | |
3681 | { | |
3682 | trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2); | |
3683 | ||
3684 | switch (ITEM_TYPE(item)) { | |
3685 | case XFS_LI_BUF: | |
3686 | return xlog_recover_buffer_pass2(log, buffer_list, item, | |
3687 | trans->r_lsn); | |
3688 | case XFS_LI_INODE: | |
3689 | return xlog_recover_inode_pass2(log, buffer_list, item, | |
3690 | trans->r_lsn); | |
3691 | case XFS_LI_EFI: | |
3692 | return xlog_recover_efi_pass2(log, item, trans->r_lsn); | |
3693 | case XFS_LI_EFD: | |
3694 | return xlog_recover_efd_pass2(log, item); | |
3695 | case XFS_LI_DQUOT: | |
3696 | return xlog_recover_dquot_pass2(log, buffer_list, item, | |
3697 | trans->r_lsn); | |
3698 | case XFS_LI_ICREATE: | |
3699 | return xlog_recover_do_icreate_pass2(log, buffer_list, item); | |
3700 | case XFS_LI_QUOTAOFF: | |
3701 | /* nothing to do in pass2 */ | |
3702 | return 0; | |
3703 | default: | |
3704 | xfs_warn(log->l_mp, "%s: invalid item type (%d)", | |
3705 | __func__, ITEM_TYPE(item)); | |
3706 | ASSERT(0); | |
3707 | return -EIO; | |
3708 | } | |
3709 | } | |
3710 | ||
3711 | STATIC int | |
3712 | xlog_recover_items_pass2( | |
3713 | struct xlog *log, | |
3714 | struct xlog_recover *trans, | |
3715 | struct list_head *buffer_list, | |
3716 | struct list_head *item_list) | |
3717 | { | |
3718 | struct xlog_recover_item *item; | |
3719 | int error = 0; | |
3720 | ||
3721 | list_for_each_entry(item, item_list, ri_list) { | |
3722 | error = xlog_recover_commit_pass2(log, trans, | |
3723 | buffer_list, item); | |
3724 | if (error) | |
3725 | return error; | |
3726 | } | |
3727 | ||
3728 | return error; | |
3729 | } | |
3730 | ||
3731 | /* | |
3732 | * Perform the transaction. | |
3733 | * | |
3734 | * If the transaction modifies a buffer or inode, do it now. Otherwise, | |
3735 | * EFIs and EFDs get queued up by adding entries into the AIL for them. | |
3736 | */ | |
3737 | STATIC int | |
3738 | xlog_recover_commit_trans( | |
3739 | struct xlog *log, | |
3740 | struct xlog_recover *trans, | |
3741 | int pass) | |
3742 | { | |
3743 | int error = 0; | |
3744 | int error2; | |
3745 | int items_queued = 0; | |
3746 | struct xlog_recover_item *item; | |
3747 | struct xlog_recover_item *next; | |
3748 | LIST_HEAD (buffer_list); | |
3749 | LIST_HEAD (ra_list); | |
3750 | LIST_HEAD (done_list); | |
3751 | ||
3752 | #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100 | |
3753 | ||
3754 | hlist_del(&trans->r_list); | |
3755 | ||
3756 | error = xlog_recover_reorder_trans(log, trans, pass); | |
3757 | if (error) | |
3758 | return error; | |
3759 | ||
3760 | list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) { | |
3761 | switch (pass) { | |
3762 | case XLOG_RECOVER_PASS1: | |
3763 | error = xlog_recover_commit_pass1(log, trans, item); | |
3764 | break; | |
3765 | case XLOG_RECOVER_PASS2: | |
3766 | xlog_recover_ra_pass2(log, item); | |
3767 | list_move_tail(&item->ri_list, &ra_list); | |
3768 | items_queued++; | |
3769 | if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) { | |
3770 | error = xlog_recover_items_pass2(log, trans, | |
3771 | &buffer_list, &ra_list); | |
3772 | list_splice_tail_init(&ra_list, &done_list); | |
3773 | items_queued = 0; | |
3774 | } | |
3775 | ||
3776 | break; | |
3777 | default: | |
3778 | ASSERT(0); | |
3779 | } | |
3780 | ||
3781 | if (error) | |
3782 | goto out; | |
3783 | } | |
3784 | ||
3785 | out: | |
3786 | if (!list_empty(&ra_list)) { | |
3787 | if (!error) | |
3788 | error = xlog_recover_items_pass2(log, trans, | |
3789 | &buffer_list, &ra_list); | |
3790 | list_splice_tail_init(&ra_list, &done_list); | |
3791 | } | |
3792 | ||
3793 | if (!list_empty(&done_list)) | |
3794 | list_splice_init(&done_list, &trans->r_itemq); | |
3795 | ||
3796 | error2 = xfs_buf_delwri_submit(&buffer_list); | |
3797 | return error ? error : error2; | |
3798 | } | |
3799 | ||
3800 | STATIC void | |
3801 | xlog_recover_add_item( | |
3802 | struct list_head *head) | |
3803 | { | |
3804 | xlog_recover_item_t *item; | |
3805 | ||
3806 | item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP); | |
3807 | INIT_LIST_HEAD(&item->ri_list); | |
3808 | list_add_tail(&item->ri_list, head); | |
3809 | } | |
3810 | ||
3811 | STATIC int | |
3812 | xlog_recover_add_to_cont_trans( | |
3813 | struct xlog *log, | |
3814 | struct xlog_recover *trans, | |
3815 | char *dp, | |
3816 | int len) | |
3817 | { | |
3818 | xlog_recover_item_t *item; | |
3819 | char *ptr, *old_ptr; | |
3820 | int old_len; | |
3821 | ||
3822 | /* | |
3823 | * If the transaction is empty, the header was split across this and the | |
3824 | * previous record. Copy the rest of the header. | |
3825 | */ | |
3826 | if (list_empty(&trans->r_itemq)) { | |
3827 | ASSERT(len <= sizeof(struct xfs_trans_header)); | |
3828 | if (len > sizeof(struct xfs_trans_header)) { | |
3829 | xfs_warn(log->l_mp, "%s: bad header length", __func__); | |
3830 | return -EIO; | |
3831 | } | |
3832 | ||
3833 | xlog_recover_add_item(&trans->r_itemq); | |
3834 | ptr = (char *)&trans->r_theader + | |
3835 | sizeof(struct xfs_trans_header) - len; | |
3836 | memcpy(ptr, dp, len); | |
3837 | return 0; | |
3838 | } | |
3839 | ||
3840 | /* take the tail entry */ | |
3841 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); | |
3842 | ||
3843 | old_ptr = item->ri_buf[item->ri_cnt-1].i_addr; | |
3844 | old_len = item->ri_buf[item->ri_cnt-1].i_len; | |
3845 | ||
3846 | ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP); | |
3847 | memcpy(&ptr[old_len], dp, len); | |
3848 | item->ri_buf[item->ri_cnt-1].i_len += len; | |
3849 | item->ri_buf[item->ri_cnt-1].i_addr = ptr; | |
3850 | trace_xfs_log_recover_item_add_cont(log, trans, item, 0); | |
3851 | return 0; | |
3852 | } | |
3853 | ||
3854 | /* | |
3855 | * The next region to add is the start of a new region. It could be | |
3856 | * a whole region or it could be the first part of a new region. Because | |
3857 | * of this, the assumption here is that the type and size fields of all | |
3858 | * format structures fit into the first 32 bits of the structure. | |
3859 | * | |
3860 | * This works because all regions must be 32 bit aligned. Therefore, we | |
3861 | * either have both fields or we have neither field. In the case we have | |
3862 | * neither field, the data part of the region is zero length. We only have | |
3863 | * a log_op_header and can throw away the header since a new one will appear | |
3864 | * later. If we have at least 4 bytes, then we can determine how many regions | |
3865 | * will appear in the current log item. | |
3866 | */ | |
3867 | STATIC int | |
3868 | xlog_recover_add_to_trans( | |
3869 | struct xlog *log, | |
3870 | struct xlog_recover *trans, | |
3871 | char *dp, | |
3872 | int len) | |
3873 | { | |
3874 | xfs_inode_log_format_t *in_f; /* any will do */ | |
3875 | xlog_recover_item_t *item; | |
3876 | char *ptr; | |
3877 | ||
3878 | if (!len) | |
3879 | return 0; | |
3880 | if (list_empty(&trans->r_itemq)) { | |
3881 | /* we need to catch log corruptions here */ | |
3882 | if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) { | |
3883 | xfs_warn(log->l_mp, "%s: bad header magic number", | |
3884 | __func__); | |
3885 | ASSERT(0); | |
3886 | return -EIO; | |
3887 | } | |
3888 | ||
3889 | if (len > sizeof(struct xfs_trans_header)) { | |
3890 | xfs_warn(log->l_mp, "%s: bad header length", __func__); | |
3891 | ASSERT(0); | |
3892 | return -EIO; | |
3893 | } | |
3894 | ||
3895 | /* | |
3896 | * The transaction header can be arbitrarily split across op | |
3897 | * records. If we don't have the whole thing here, copy what we | |
3898 | * do have and handle the rest in the next record. | |
3899 | */ | |
3900 | if (len == sizeof(struct xfs_trans_header)) | |
3901 | xlog_recover_add_item(&trans->r_itemq); | |
3902 | memcpy(&trans->r_theader, dp, len); | |
3903 | return 0; | |
3904 | } | |
3905 | ||
3906 | ptr = kmem_alloc(len, KM_SLEEP); | |
3907 | memcpy(ptr, dp, len); | |
3908 | in_f = (xfs_inode_log_format_t *)ptr; | |
3909 | ||
3910 | /* take the tail entry */ | |
3911 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); | |
3912 | if (item->ri_total != 0 && | |
3913 | item->ri_total == item->ri_cnt) { | |
3914 | /* tail item is in use, get a new one */ | |
3915 | xlog_recover_add_item(&trans->r_itemq); | |
3916 | item = list_entry(trans->r_itemq.prev, | |
3917 | xlog_recover_item_t, ri_list); | |
3918 | } | |
3919 | ||
3920 | if (item->ri_total == 0) { /* first region to be added */ | |
3921 | if (in_f->ilf_size == 0 || | |
3922 | in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) { | |
3923 | xfs_warn(log->l_mp, | |
3924 | "bad number of regions (%d) in inode log format", | |
3925 | in_f->ilf_size); | |
3926 | ASSERT(0); | |
3927 | kmem_free(ptr); | |
3928 | return -EIO; | |
3929 | } | |
3930 | ||
3931 | item->ri_total = in_f->ilf_size; | |
3932 | item->ri_buf = | |
3933 | kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t), | |
3934 | KM_SLEEP); | |
3935 | } | |
3936 | ASSERT(item->ri_total > item->ri_cnt); | |
3937 | /* Description region is ri_buf[0] */ | |
3938 | item->ri_buf[item->ri_cnt].i_addr = ptr; | |
3939 | item->ri_buf[item->ri_cnt].i_len = len; | |
3940 | item->ri_cnt++; | |
3941 | trace_xfs_log_recover_item_add(log, trans, item, 0); | |
3942 | return 0; | |
3943 | } | |
3944 | ||
3945 | /* | |
3946 | * Free up any resources allocated by the transaction | |
3947 | * | |
3948 | * Remember that EFIs, EFDs, and IUNLINKs are handled later. | |
3949 | */ | |
3950 | STATIC void | |
3951 | xlog_recover_free_trans( | |
3952 | struct xlog_recover *trans) | |
3953 | { | |
3954 | xlog_recover_item_t *item, *n; | |
3955 | int i; | |
3956 | ||
3957 | list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) { | |
3958 | /* Free the regions in the item. */ | |
3959 | list_del(&item->ri_list); | |
3960 | for (i = 0; i < item->ri_cnt; i++) | |
3961 | kmem_free(item->ri_buf[i].i_addr); | |
3962 | /* Free the item itself */ | |
3963 | kmem_free(item->ri_buf); | |
3964 | kmem_free(item); | |
3965 | } | |
3966 | /* Free the transaction recover structure */ | |
3967 | kmem_free(trans); | |
3968 | } | |
3969 | ||
3970 | /* | |
3971 | * On error or completion, trans is freed. | |
3972 | */ | |
3973 | STATIC int | |
3974 | xlog_recovery_process_trans( | |
3975 | struct xlog *log, | |
3976 | struct xlog_recover *trans, | |
3977 | char *dp, | |
3978 | unsigned int len, | |
3979 | unsigned int flags, | |
3980 | int pass) | |
3981 | { | |
3982 | int error = 0; | |
3983 | bool freeit = false; | |
3984 | ||
3985 | /* mask off ophdr transaction container flags */ | |
3986 | flags &= ~XLOG_END_TRANS; | |
3987 | if (flags & XLOG_WAS_CONT_TRANS) | |
3988 | flags &= ~XLOG_CONTINUE_TRANS; | |
3989 | ||
3990 | /* | |
3991 | * Callees must not free the trans structure. We'll decide if we need to | |
3992 | * free it or not based on the operation being done and it's result. | |
3993 | */ | |
3994 | switch (flags) { | |
3995 | /* expected flag values */ | |
3996 | case 0: | |
3997 | case XLOG_CONTINUE_TRANS: | |
3998 | error = xlog_recover_add_to_trans(log, trans, dp, len); | |
3999 | break; | |
4000 | case XLOG_WAS_CONT_TRANS: | |
4001 | error = xlog_recover_add_to_cont_trans(log, trans, dp, len); | |
4002 | break; | |
4003 | case XLOG_COMMIT_TRANS: | |
4004 | error = xlog_recover_commit_trans(log, trans, pass); | |
4005 | /* success or fail, we are now done with this transaction. */ | |
4006 | freeit = true; | |
4007 | break; | |
4008 | ||
4009 | /* unexpected flag values */ | |
4010 | case XLOG_UNMOUNT_TRANS: | |
4011 | /* just skip trans */ | |
4012 | xfs_warn(log->l_mp, "%s: Unmount LR", __func__); | |
4013 | freeit = true; | |
4014 | break; | |
4015 | case XLOG_START_TRANS: | |
4016 | default: | |
4017 | xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags); | |
4018 | ASSERT(0); | |
4019 | error = -EIO; | |
4020 | break; | |
4021 | } | |
4022 | if (error || freeit) | |
4023 | xlog_recover_free_trans(trans); | |
4024 | return error; | |
4025 | } | |
4026 | ||
4027 | /* | |
4028 | * Lookup the transaction recovery structure associated with the ID in the | |
4029 | * current ophdr. If the transaction doesn't exist and the start flag is set in | |
4030 | * the ophdr, then allocate a new transaction for future ID matches to find. | |
4031 | * Either way, return what we found during the lookup - an existing transaction | |
4032 | * or nothing. | |
4033 | */ | |
4034 | STATIC struct xlog_recover * | |
4035 | xlog_recover_ophdr_to_trans( | |
4036 | struct hlist_head rhash[], | |
4037 | struct xlog_rec_header *rhead, | |
4038 | struct xlog_op_header *ohead) | |
4039 | { | |
4040 | struct xlog_recover *trans; | |
4041 | xlog_tid_t tid; | |
4042 | struct hlist_head *rhp; | |
4043 | ||
4044 | tid = be32_to_cpu(ohead->oh_tid); | |
4045 | rhp = &rhash[XLOG_RHASH(tid)]; | |
4046 | hlist_for_each_entry(trans, rhp, r_list) { | |
4047 | if (trans->r_log_tid == tid) | |
4048 | return trans; | |
4049 | } | |
4050 | ||
4051 | /* | |
4052 | * skip over non-start transaction headers - we could be | |
4053 | * processing slack space before the next transaction starts | |
4054 | */ | |
4055 | if (!(ohead->oh_flags & XLOG_START_TRANS)) | |
4056 | return NULL; | |
4057 | ||
4058 | ASSERT(be32_to_cpu(ohead->oh_len) == 0); | |
4059 | ||
4060 | /* | |
4061 | * This is a new transaction so allocate a new recovery container to | |
4062 | * hold the recovery ops that will follow. | |
4063 | */ | |
4064 | trans = kmem_zalloc(sizeof(struct xlog_recover), KM_SLEEP); | |
4065 | trans->r_log_tid = tid; | |
4066 | trans->r_lsn = be64_to_cpu(rhead->h_lsn); | |
4067 | INIT_LIST_HEAD(&trans->r_itemq); | |
4068 | INIT_HLIST_NODE(&trans->r_list); | |
4069 | hlist_add_head(&trans->r_list, rhp); | |
4070 | ||
4071 | /* | |
4072 | * Nothing more to do for this ophdr. Items to be added to this new | |
4073 | * transaction will be in subsequent ophdr containers. | |
4074 | */ | |
4075 | return NULL; | |
4076 | } | |
4077 | ||
4078 | STATIC int | |
4079 | xlog_recover_process_ophdr( | |
4080 | struct xlog *log, | |
4081 | struct hlist_head rhash[], | |
4082 | struct xlog_rec_header *rhead, | |
4083 | struct xlog_op_header *ohead, | |
4084 | char *dp, | |
4085 | char *end, | |
4086 | int pass) | |
4087 | { | |
4088 | struct xlog_recover *trans; | |
4089 | unsigned int len; | |
4090 | ||
4091 | /* Do we understand who wrote this op? */ | |
4092 | if (ohead->oh_clientid != XFS_TRANSACTION && | |
4093 | ohead->oh_clientid != XFS_LOG) { | |
4094 | xfs_warn(log->l_mp, "%s: bad clientid 0x%x", | |
4095 | __func__, ohead->oh_clientid); | |
4096 | ASSERT(0); | |
4097 | return -EIO; | |
4098 | } | |
4099 | ||
4100 | /* | |
4101 | * Check the ophdr contains all the data it is supposed to contain. | |
4102 | */ | |
4103 | len = be32_to_cpu(ohead->oh_len); | |
4104 | if (dp + len > end) { | |
4105 | xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len); | |
4106 | WARN_ON(1); | |
4107 | return -EIO; | |
4108 | } | |
4109 | ||
4110 | trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead); | |
4111 | if (!trans) { | |
4112 | /* nothing to do, so skip over this ophdr */ | |
4113 | return 0; | |
4114 | } | |
4115 | ||
4116 | return xlog_recovery_process_trans(log, trans, dp, len, | |
4117 | ohead->oh_flags, pass); | |
4118 | } | |
4119 | ||
4120 | /* | |
4121 | * There are two valid states of the r_state field. 0 indicates that the | |
4122 | * transaction structure is in a normal state. We have either seen the | |
4123 | * start of the transaction or the last operation we added was not a partial | |
4124 | * operation. If the last operation we added to the transaction was a | |
4125 | * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS. | |
4126 | * | |
4127 | * NOTE: skip LRs with 0 data length. | |
4128 | */ | |
4129 | STATIC int | |
4130 | xlog_recover_process_data( | |
4131 | struct xlog *log, | |
4132 | struct hlist_head rhash[], | |
4133 | struct xlog_rec_header *rhead, | |
4134 | char *dp, | |
4135 | int pass) | |
4136 | { | |
4137 | struct xlog_op_header *ohead; | |
4138 | char *end; | |
4139 | int num_logops; | |
4140 | int error; | |
4141 | ||
4142 | end = dp + be32_to_cpu(rhead->h_len); | |
4143 | num_logops = be32_to_cpu(rhead->h_num_logops); | |
4144 | ||
4145 | /* check the log format matches our own - else we can't recover */ | |
4146 | if (xlog_header_check_recover(log->l_mp, rhead)) | |
4147 | return -EIO; | |
4148 | ||
4149 | while ((dp < end) && num_logops) { | |
4150 | ||
4151 | ohead = (struct xlog_op_header *)dp; | |
4152 | dp += sizeof(*ohead); | |
4153 | ASSERT(dp <= end); | |
4154 | ||
4155 | /* errors will abort recovery */ | |
4156 | error = xlog_recover_process_ophdr(log, rhash, rhead, ohead, | |
4157 | dp, end, pass); | |
4158 | if (error) | |
4159 | return error; | |
4160 | ||
4161 | dp += be32_to_cpu(ohead->oh_len); | |
4162 | num_logops--; | |
4163 | } | |
4164 | return 0; | |
4165 | } | |
4166 | ||
4167 | /* | |
4168 | * Process an extent free intent item that was recovered from | |
4169 | * the log. We need to free the extents that it describes. | |
4170 | */ | |
4171 | STATIC int | |
4172 | xlog_recover_process_efi( | |
4173 | xfs_mount_t *mp, | |
4174 | xfs_efi_log_item_t *efip) | |
4175 | { | |
4176 | xfs_efd_log_item_t *efdp; | |
4177 | xfs_trans_t *tp; | |
4178 | int i; | |
4179 | int error = 0; | |
4180 | xfs_extent_t *extp; | |
4181 | xfs_fsblock_t startblock_fsb; | |
4182 | ||
4183 | ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)); | |
4184 | ||
4185 | /* | |
4186 | * First check the validity of the extents described by the | |
4187 | * EFI. If any are bad, then assume that all are bad and | |
4188 | * just toss the EFI. | |
4189 | */ | |
4190 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { | |
4191 | extp = &(efip->efi_format.efi_extents[i]); | |
4192 | startblock_fsb = XFS_BB_TO_FSB(mp, | |
4193 | XFS_FSB_TO_DADDR(mp, extp->ext_start)); | |
4194 | if ((startblock_fsb == 0) || | |
4195 | (extp->ext_len == 0) || | |
4196 | (startblock_fsb >= mp->m_sb.sb_dblocks) || | |
4197 | (extp->ext_len >= mp->m_sb.sb_agblocks)) { | |
4198 | /* | |
4199 | * This will pull the EFI from the AIL and | |
4200 | * free the memory associated with it. | |
4201 | */ | |
4202 | set_bit(XFS_EFI_RECOVERED, &efip->efi_flags); | |
4203 | xfs_efi_release(efip); | |
4204 | return -EIO; | |
4205 | } | |
4206 | } | |
4207 | ||
4208 | tp = xfs_trans_alloc(mp, 0); | |
4209 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0); | |
4210 | if (error) | |
4211 | goto abort_error; | |
4212 | efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents); | |
4213 | ||
4214 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { | |
4215 | extp = &(efip->efi_format.efi_extents[i]); | |
4216 | error = xfs_trans_free_extent(tp, efdp, extp->ext_start, | |
4217 | extp->ext_len); | |
4218 | if (error) | |
4219 | goto abort_error; | |
4220 | ||
4221 | } | |
4222 | ||
4223 | set_bit(XFS_EFI_RECOVERED, &efip->efi_flags); | |
4224 | error = xfs_trans_commit(tp); | |
4225 | return error; | |
4226 | ||
4227 | abort_error: | |
4228 | xfs_trans_cancel(tp); | |
4229 | return error; | |
4230 | } | |
4231 | ||
4232 | /* | |
4233 | * When this is called, all of the EFIs which did not have | |
4234 | * corresponding EFDs should be in the AIL. What we do now | |
4235 | * is free the extents associated with each one. | |
4236 | * | |
4237 | * Since we process the EFIs in normal transactions, they | |
4238 | * will be removed at some point after the commit. This prevents | |
4239 | * us from just walking down the list processing each one. | |
4240 | * We'll use a flag in the EFI to skip those that we've already | |
4241 | * processed and use the AIL iteration mechanism's generation | |
4242 | * count to try to speed this up at least a bit. | |
4243 | * | |
4244 | * When we start, we know that the EFIs are the only things in | |
4245 | * the AIL. As we process them, however, other items are added | |
4246 | * to the AIL. Since everything added to the AIL must come after | |
4247 | * everything already in the AIL, we stop processing as soon as | |
4248 | * we see something other than an EFI in the AIL. | |
4249 | */ | |
4250 | STATIC int | |
4251 | xlog_recover_process_efis( | |
4252 | struct xlog *log) | |
4253 | { | |
4254 | struct xfs_log_item *lip; | |
4255 | struct xfs_efi_log_item *efip; | |
4256 | int error = 0; | |
4257 | struct xfs_ail_cursor cur; | |
4258 | struct xfs_ail *ailp; | |
4259 | ||
4260 | ailp = log->l_ailp; | |
4261 | spin_lock(&ailp->xa_lock); | |
4262 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); | |
4263 | while (lip != NULL) { | |
4264 | /* | |
4265 | * We're done when we see something other than an EFI. | |
4266 | * There should be no EFIs left in the AIL now. | |
4267 | */ | |
4268 | if (lip->li_type != XFS_LI_EFI) { | |
4269 | #ifdef DEBUG | |
4270 | for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) | |
4271 | ASSERT(lip->li_type != XFS_LI_EFI); | |
4272 | #endif | |
4273 | break; | |
4274 | } | |
4275 | ||
4276 | /* | |
4277 | * Skip EFIs that we've already processed. | |
4278 | */ | |
4279 | efip = container_of(lip, struct xfs_efi_log_item, efi_item); | |
4280 | if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) { | |
4281 | lip = xfs_trans_ail_cursor_next(ailp, &cur); | |
4282 | continue; | |
4283 | } | |
4284 | ||
4285 | spin_unlock(&ailp->xa_lock); | |
4286 | error = xlog_recover_process_efi(log->l_mp, efip); | |
4287 | spin_lock(&ailp->xa_lock); | |
4288 | if (error) | |
4289 | goto out; | |
4290 | lip = xfs_trans_ail_cursor_next(ailp, &cur); | |
4291 | } | |
4292 | out: | |
4293 | xfs_trans_ail_cursor_done(&cur); | |
4294 | spin_unlock(&ailp->xa_lock); | |
4295 | return error; | |
4296 | } | |
4297 | ||
4298 | /* | |
4299 | * A cancel occurs when the mount has failed and we're bailing out. Release all | |
4300 | * pending EFIs so they don't pin the AIL. | |
4301 | */ | |
4302 | STATIC int | |
4303 | xlog_recover_cancel_efis( | |
4304 | struct xlog *log) | |
4305 | { | |
4306 | struct xfs_log_item *lip; | |
4307 | struct xfs_efi_log_item *efip; | |
4308 | int error = 0; | |
4309 | struct xfs_ail_cursor cur; | |
4310 | struct xfs_ail *ailp; | |
4311 | ||
4312 | ailp = log->l_ailp; | |
4313 | spin_lock(&ailp->xa_lock); | |
4314 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); | |
4315 | while (lip != NULL) { | |
4316 | /* | |
4317 | * We're done when we see something other than an EFI. | |
4318 | * There should be no EFIs left in the AIL now. | |
4319 | */ | |
4320 | if (lip->li_type != XFS_LI_EFI) { | |
4321 | #ifdef DEBUG | |
4322 | for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) | |
4323 | ASSERT(lip->li_type != XFS_LI_EFI); | |
4324 | #endif | |
4325 | break; | |
4326 | } | |
4327 | ||
4328 | efip = container_of(lip, struct xfs_efi_log_item, efi_item); | |
4329 | ||
4330 | spin_unlock(&ailp->xa_lock); | |
4331 | xfs_efi_release(efip); | |
4332 | spin_lock(&ailp->xa_lock); | |
4333 | ||
4334 | lip = xfs_trans_ail_cursor_next(ailp, &cur); | |
4335 | } | |
4336 | ||
4337 | xfs_trans_ail_cursor_done(&cur); | |
4338 | spin_unlock(&ailp->xa_lock); | |
4339 | return error; | |
4340 | } | |
4341 | ||
4342 | /* | |
4343 | * This routine performs a transaction to null out a bad inode pointer | |
4344 | * in an agi unlinked inode hash bucket. | |
4345 | */ | |
4346 | STATIC void | |
4347 | xlog_recover_clear_agi_bucket( | |
4348 | xfs_mount_t *mp, | |
4349 | xfs_agnumber_t agno, | |
4350 | int bucket) | |
4351 | { | |
4352 | xfs_trans_t *tp; | |
4353 | xfs_agi_t *agi; | |
4354 | xfs_buf_t *agibp; | |
4355 | int offset; | |
4356 | int error; | |
4357 | ||
4358 | tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET); | |
4359 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_clearagi, 0, 0); | |
4360 | if (error) | |
4361 | goto out_abort; | |
4362 | ||
4363 | error = xfs_read_agi(mp, tp, agno, &agibp); | |
4364 | if (error) | |
4365 | goto out_abort; | |
4366 | ||
4367 | agi = XFS_BUF_TO_AGI(agibp); | |
4368 | agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO); | |
4369 | offset = offsetof(xfs_agi_t, agi_unlinked) + | |
4370 | (sizeof(xfs_agino_t) * bucket); | |
4371 | xfs_trans_log_buf(tp, agibp, offset, | |
4372 | (offset + sizeof(xfs_agino_t) - 1)); | |
4373 | ||
4374 | error = xfs_trans_commit(tp); | |
4375 | if (error) | |
4376 | goto out_error; | |
4377 | return; | |
4378 | ||
4379 | out_abort: | |
4380 | xfs_trans_cancel(tp); | |
4381 | out_error: | |
4382 | xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno); | |
4383 | return; | |
4384 | } | |
4385 | ||
4386 | STATIC xfs_agino_t | |
4387 | xlog_recover_process_one_iunlink( | |
4388 | struct xfs_mount *mp, | |
4389 | xfs_agnumber_t agno, | |
4390 | xfs_agino_t agino, | |
4391 | int bucket) | |
4392 | { | |
4393 | struct xfs_buf *ibp; | |
4394 | struct xfs_dinode *dip; | |
4395 | struct xfs_inode *ip; | |
4396 | xfs_ino_t ino; | |
4397 | int error; | |
4398 | ||
4399 | ino = XFS_AGINO_TO_INO(mp, agno, agino); | |
4400 | error = xfs_iget(mp, NULL, ino, 0, 0, &ip); | |
4401 | if (error) | |
4402 | goto fail; | |
4403 | ||
4404 | /* | |
4405 | * Get the on disk inode to find the next inode in the bucket. | |
4406 | */ | |
4407 | error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0); | |
4408 | if (error) | |
4409 | goto fail_iput; | |
4410 | ||
4411 | ASSERT(VFS_I(ip)->i_nlink == 0); | |
4412 | ASSERT(VFS_I(ip)->i_mode != 0); | |
4413 | ||
4414 | /* setup for the next pass */ | |
4415 | agino = be32_to_cpu(dip->di_next_unlinked); | |
4416 | xfs_buf_relse(ibp); | |
4417 | ||
4418 | /* | |
4419 | * Prevent any DMAPI event from being sent when the reference on | |
4420 | * the inode is dropped. | |
4421 | */ | |
4422 | ip->i_d.di_dmevmask = 0; | |
4423 | ||
4424 | IRELE(ip); | |
4425 | return agino; | |
4426 | ||
4427 | fail_iput: | |
4428 | IRELE(ip); | |
4429 | fail: | |
4430 | /* | |
4431 | * We can't read in the inode this bucket points to, or this inode | |
4432 | * is messed up. Just ditch this bucket of inodes. We will lose | |
4433 | * some inodes and space, but at least we won't hang. | |
4434 | * | |
4435 | * Call xlog_recover_clear_agi_bucket() to perform a transaction to | |
4436 | * clear the inode pointer in the bucket. | |
4437 | */ | |
4438 | xlog_recover_clear_agi_bucket(mp, agno, bucket); | |
4439 | return NULLAGINO; | |
4440 | } | |
4441 | ||
4442 | /* | |
4443 | * xlog_iunlink_recover | |
4444 | * | |
4445 | * This is called during recovery to process any inodes which | |
4446 | * we unlinked but not freed when the system crashed. These | |
4447 | * inodes will be on the lists in the AGI blocks. What we do | |
4448 | * here is scan all the AGIs and fully truncate and free any | |
4449 | * inodes found on the lists. Each inode is removed from the | |
4450 | * lists when it has been fully truncated and is freed. The | |
4451 | * freeing of the inode and its removal from the list must be | |
4452 | * atomic. | |
4453 | */ | |
4454 | STATIC void | |
4455 | xlog_recover_process_iunlinks( | |
4456 | struct xlog *log) | |
4457 | { | |
4458 | xfs_mount_t *mp; | |
4459 | xfs_agnumber_t agno; | |
4460 | xfs_agi_t *agi; | |
4461 | xfs_buf_t *agibp; | |
4462 | xfs_agino_t agino; | |
4463 | int bucket; | |
4464 | int error; | |
4465 | uint mp_dmevmask; | |
4466 | ||
4467 | mp = log->l_mp; | |
4468 | ||
4469 | /* | |
4470 | * Prevent any DMAPI event from being sent while in this function. | |
4471 | */ | |
4472 | mp_dmevmask = mp->m_dmevmask; | |
4473 | mp->m_dmevmask = 0; | |
4474 | ||
4475 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
4476 | /* | |
4477 | * Find the agi for this ag. | |
4478 | */ | |
4479 | error = xfs_read_agi(mp, NULL, agno, &agibp); | |
4480 | if (error) { | |
4481 | /* | |
4482 | * AGI is b0rked. Don't process it. | |
4483 | * | |
4484 | * We should probably mark the filesystem as corrupt | |
4485 | * after we've recovered all the ag's we can.... | |
4486 | */ | |
4487 | continue; | |
4488 | } | |
4489 | /* | |
4490 | * Unlock the buffer so that it can be acquired in the normal | |
4491 | * course of the transaction to truncate and free each inode. | |
4492 | * Because we are not racing with anyone else here for the AGI | |
4493 | * buffer, we don't even need to hold it locked to read the | |
4494 | * initial unlinked bucket entries out of the buffer. We keep | |
4495 | * buffer reference though, so that it stays pinned in memory | |
4496 | * while we need the buffer. | |
4497 | */ | |
4498 | agi = XFS_BUF_TO_AGI(agibp); | |
4499 | xfs_buf_unlock(agibp); | |
4500 | ||
4501 | for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) { | |
4502 | agino = be32_to_cpu(agi->agi_unlinked[bucket]); | |
4503 | while (agino != NULLAGINO) { | |
4504 | agino = xlog_recover_process_one_iunlink(mp, | |
4505 | agno, agino, bucket); | |
4506 | } | |
4507 | } | |
4508 | xfs_buf_rele(agibp); | |
4509 | } | |
4510 | ||
4511 | mp->m_dmevmask = mp_dmevmask; | |
4512 | } | |
4513 | ||
4514 | STATIC int | |
4515 | xlog_unpack_data( | |
4516 | struct xlog_rec_header *rhead, | |
4517 | char *dp, | |
4518 | struct xlog *log) | |
4519 | { | |
4520 | int i, j, k; | |
4521 | ||
4522 | for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) && | |
4523 | i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) { | |
4524 | *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i]; | |
4525 | dp += BBSIZE; | |
4526 | } | |
4527 | ||
4528 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
4529 | xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead; | |
4530 | for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) { | |
4531 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
4532 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
4533 | *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k]; | |
4534 | dp += BBSIZE; | |
4535 | } | |
4536 | } | |
4537 | ||
4538 | return 0; | |
4539 | } | |
4540 | ||
4541 | /* | |
4542 | * CRC check, unpack and process a log record. | |
4543 | */ | |
4544 | STATIC int | |
4545 | xlog_recover_process( | |
4546 | struct xlog *log, | |
4547 | struct hlist_head rhash[], | |
4548 | struct xlog_rec_header *rhead, | |
4549 | char *dp, | |
4550 | int pass) | |
4551 | { | |
4552 | int error; | |
4553 | __le32 crc; | |
4554 | ||
4555 | crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len)); | |
4556 | ||
4557 | /* | |
4558 | * Nothing else to do if this is a CRC verification pass. Just return | |
4559 | * if this a record with a non-zero crc. Unfortunately, mkfs always | |
4560 | * sets h_crc to 0 so we must consider this valid even on v5 supers. | |
4561 | * Otherwise, return EFSBADCRC on failure so the callers up the stack | |
4562 | * know precisely what failed. | |
4563 | */ | |
4564 | if (pass == XLOG_RECOVER_CRCPASS) { | |
4565 | if (rhead->h_crc && crc != rhead->h_crc) | |
4566 | return -EFSBADCRC; | |
4567 | return 0; | |
4568 | } | |
4569 | ||
4570 | /* | |
4571 | * We're in the normal recovery path. Issue a warning if and only if the | |
4572 | * CRC in the header is non-zero. This is an advisory warning and the | |
4573 | * zero CRC check prevents warnings from being emitted when upgrading | |
4574 | * the kernel from one that does not add CRCs by default. | |
4575 | */ | |
4576 | if (crc != rhead->h_crc) { | |
4577 | if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) { | |
4578 | xfs_alert(log->l_mp, | |
4579 | "log record CRC mismatch: found 0x%x, expected 0x%x.", | |
4580 | le32_to_cpu(rhead->h_crc), | |
4581 | le32_to_cpu(crc)); | |
4582 | xfs_hex_dump(dp, 32); | |
4583 | } | |
4584 | ||
4585 | /* | |
4586 | * If the filesystem is CRC enabled, this mismatch becomes a | |
4587 | * fatal log corruption failure. | |
4588 | */ | |
4589 | if (xfs_sb_version_hascrc(&log->l_mp->m_sb)) | |
4590 | return -EFSCORRUPTED; | |
4591 | } | |
4592 | ||
4593 | error = xlog_unpack_data(rhead, dp, log); | |
4594 | if (error) | |
4595 | return error; | |
4596 | ||
4597 | return xlog_recover_process_data(log, rhash, rhead, dp, pass); | |
4598 | } | |
4599 | ||
4600 | STATIC int | |
4601 | xlog_valid_rec_header( | |
4602 | struct xlog *log, | |
4603 | struct xlog_rec_header *rhead, | |
4604 | xfs_daddr_t blkno) | |
4605 | { | |
4606 | int hlen; | |
4607 | ||
4608 | if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) { | |
4609 | XFS_ERROR_REPORT("xlog_valid_rec_header(1)", | |
4610 | XFS_ERRLEVEL_LOW, log->l_mp); | |
4611 | return -EFSCORRUPTED; | |
4612 | } | |
4613 | if (unlikely( | |
4614 | (!rhead->h_version || | |
4615 | (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) { | |
4616 | xfs_warn(log->l_mp, "%s: unrecognised log version (%d).", | |
4617 | __func__, be32_to_cpu(rhead->h_version)); | |
4618 | return -EIO; | |
4619 | } | |
4620 | ||
4621 | /* LR body must have data or it wouldn't have been written */ | |
4622 | hlen = be32_to_cpu(rhead->h_len); | |
4623 | if (unlikely( hlen <= 0 || hlen > INT_MAX )) { | |
4624 | XFS_ERROR_REPORT("xlog_valid_rec_header(2)", | |
4625 | XFS_ERRLEVEL_LOW, log->l_mp); | |
4626 | return -EFSCORRUPTED; | |
4627 | } | |
4628 | if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) { | |
4629 | XFS_ERROR_REPORT("xlog_valid_rec_header(3)", | |
4630 | XFS_ERRLEVEL_LOW, log->l_mp); | |
4631 | return -EFSCORRUPTED; | |
4632 | } | |
4633 | return 0; | |
4634 | } | |
4635 | ||
4636 | /* | |
4637 | * Read the log from tail to head and process the log records found. | |
4638 | * Handle the two cases where the tail and head are in the same cycle | |
4639 | * and where the active portion of the log wraps around the end of | |
4640 | * the physical log separately. The pass parameter is passed through | |
4641 | * to the routines called to process the data and is not looked at | |
4642 | * here. | |
4643 | */ | |
4644 | STATIC int | |
4645 | xlog_do_recovery_pass( | |
4646 | struct xlog *log, | |
4647 | xfs_daddr_t head_blk, | |
4648 | xfs_daddr_t tail_blk, | |
4649 | int pass, | |
4650 | xfs_daddr_t *first_bad) /* out: first bad log rec */ | |
4651 | { | |
4652 | xlog_rec_header_t *rhead; | |
4653 | xfs_daddr_t blk_no; | |
4654 | xfs_daddr_t rhead_blk; | |
4655 | char *offset; | |
4656 | xfs_buf_t *hbp, *dbp; | |
4657 | int error = 0, h_size, h_len; | |
4658 | int bblks, split_bblks; | |
4659 | int hblks, split_hblks, wrapped_hblks; | |
4660 | struct hlist_head rhash[XLOG_RHASH_SIZE]; | |
4661 | ||
4662 | ASSERT(head_blk != tail_blk); | |
4663 | rhead_blk = 0; | |
4664 | ||
4665 | /* | |
4666 | * Read the header of the tail block and get the iclog buffer size from | |
4667 | * h_size. Use this to tell how many sectors make up the log header. | |
4668 | */ | |
4669 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
4670 | /* | |
4671 | * When using variable length iclogs, read first sector of | |
4672 | * iclog header and extract the header size from it. Get a | |
4673 | * new hbp that is the correct size. | |
4674 | */ | |
4675 | hbp = xlog_get_bp(log, 1); | |
4676 | if (!hbp) | |
4677 | return -ENOMEM; | |
4678 | ||
4679 | error = xlog_bread(log, tail_blk, 1, hbp, &offset); | |
4680 | if (error) | |
4681 | goto bread_err1; | |
4682 | ||
4683 | rhead = (xlog_rec_header_t *)offset; | |
4684 | error = xlog_valid_rec_header(log, rhead, tail_blk); | |
4685 | if (error) | |
4686 | goto bread_err1; | |
4687 | ||
4688 | /* | |
4689 | * xfsprogs has a bug where record length is based on lsunit but | |
4690 | * h_size (iclog size) is hardcoded to 32k. Now that we | |
4691 | * unconditionally CRC verify the unmount record, this means the | |
4692 | * log buffer can be too small for the record and cause an | |
4693 | * overrun. | |
4694 | * | |
4695 | * Detect this condition here. Use lsunit for the buffer size as | |
4696 | * long as this looks like the mkfs case. Otherwise, return an | |
4697 | * error to avoid a buffer overrun. | |
4698 | */ | |
4699 | h_size = be32_to_cpu(rhead->h_size); | |
4700 | h_len = be32_to_cpu(rhead->h_len); | |
4701 | if (h_len > h_size) { | |
4702 | if (h_len <= log->l_mp->m_logbsize && | |
4703 | be32_to_cpu(rhead->h_num_logops) == 1) { | |
4704 | xfs_warn(log->l_mp, | |
4705 | "invalid iclog size (%d bytes), using lsunit (%d bytes)", | |
4706 | h_size, log->l_mp->m_logbsize); | |
4707 | h_size = log->l_mp->m_logbsize; | |
4708 | } else | |
4709 | return -EFSCORRUPTED; | |
4710 | } | |
4711 | ||
4712 | if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) && | |
4713 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { | |
4714 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; | |
4715 | if (h_size % XLOG_HEADER_CYCLE_SIZE) | |
4716 | hblks++; | |
4717 | xlog_put_bp(hbp); | |
4718 | hbp = xlog_get_bp(log, hblks); | |
4719 | } else { | |
4720 | hblks = 1; | |
4721 | } | |
4722 | } else { | |
4723 | ASSERT(log->l_sectBBsize == 1); | |
4724 | hblks = 1; | |
4725 | hbp = xlog_get_bp(log, 1); | |
4726 | h_size = XLOG_BIG_RECORD_BSIZE; | |
4727 | } | |
4728 | ||
4729 | if (!hbp) | |
4730 | return -ENOMEM; | |
4731 | dbp = xlog_get_bp(log, BTOBB(h_size)); | |
4732 | if (!dbp) { | |
4733 | xlog_put_bp(hbp); | |
4734 | return -ENOMEM; | |
4735 | } | |
4736 | ||
4737 | memset(rhash, 0, sizeof(rhash)); | |
4738 | blk_no = rhead_blk = tail_blk; | |
4739 | if (tail_blk > head_blk) { | |
4740 | /* | |
4741 | * Perform recovery around the end of the physical log. | |
4742 | * When the head is not on the same cycle number as the tail, | |
4743 | * we can't do a sequential recovery. | |
4744 | */ | |
4745 | while (blk_no < log->l_logBBsize) { | |
4746 | /* | |
4747 | * Check for header wrapping around physical end-of-log | |
4748 | */ | |
4749 | offset = hbp->b_addr; | |
4750 | split_hblks = 0; | |
4751 | wrapped_hblks = 0; | |
4752 | if (blk_no + hblks <= log->l_logBBsize) { | |
4753 | /* Read header in one read */ | |
4754 | error = xlog_bread(log, blk_no, hblks, hbp, | |
4755 | &offset); | |
4756 | if (error) | |
4757 | goto bread_err2; | |
4758 | } else { | |
4759 | /* This LR is split across physical log end */ | |
4760 | if (blk_no != log->l_logBBsize) { | |
4761 | /* some data before physical log end */ | |
4762 | ASSERT(blk_no <= INT_MAX); | |
4763 | split_hblks = log->l_logBBsize - (int)blk_no; | |
4764 | ASSERT(split_hblks > 0); | |
4765 | error = xlog_bread(log, blk_no, | |
4766 | split_hblks, hbp, | |
4767 | &offset); | |
4768 | if (error) | |
4769 | goto bread_err2; | |
4770 | } | |
4771 | ||
4772 | /* | |
4773 | * Note: this black magic still works with | |
4774 | * large sector sizes (non-512) only because: | |
4775 | * - we increased the buffer size originally | |
4776 | * by 1 sector giving us enough extra space | |
4777 | * for the second read; | |
4778 | * - the log start is guaranteed to be sector | |
4779 | * aligned; | |
4780 | * - we read the log end (LR header start) | |
4781 | * _first_, then the log start (LR header end) | |
4782 | * - order is important. | |
4783 | */ | |
4784 | wrapped_hblks = hblks - split_hblks; | |
4785 | error = xlog_bread_offset(log, 0, | |
4786 | wrapped_hblks, hbp, | |
4787 | offset + BBTOB(split_hblks)); | |
4788 | if (error) | |
4789 | goto bread_err2; | |
4790 | } | |
4791 | rhead = (xlog_rec_header_t *)offset; | |
4792 | error = xlog_valid_rec_header(log, rhead, | |
4793 | split_hblks ? blk_no : 0); | |
4794 | if (error) | |
4795 | goto bread_err2; | |
4796 | ||
4797 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); | |
4798 | blk_no += hblks; | |
4799 | ||
4800 | /* Read in data for log record */ | |
4801 | if (blk_no + bblks <= log->l_logBBsize) { | |
4802 | error = xlog_bread(log, blk_no, bblks, dbp, | |
4803 | &offset); | |
4804 | if (error) | |
4805 | goto bread_err2; | |
4806 | } else { | |
4807 | /* This log record is split across the | |
4808 | * physical end of log */ | |
4809 | offset = dbp->b_addr; | |
4810 | split_bblks = 0; | |
4811 | if (blk_no != log->l_logBBsize) { | |
4812 | /* some data is before the physical | |
4813 | * end of log */ | |
4814 | ASSERT(!wrapped_hblks); | |
4815 | ASSERT(blk_no <= INT_MAX); | |
4816 | split_bblks = | |
4817 | log->l_logBBsize - (int)blk_no; | |
4818 | ASSERT(split_bblks > 0); | |
4819 | error = xlog_bread(log, blk_no, | |
4820 | split_bblks, dbp, | |
4821 | &offset); | |
4822 | if (error) | |
4823 | goto bread_err2; | |
4824 | } | |
4825 | ||
4826 | /* | |
4827 | * Note: this black magic still works with | |
4828 | * large sector sizes (non-512) only because: | |
4829 | * - we increased the buffer size originally | |
4830 | * by 1 sector giving us enough extra space | |
4831 | * for the second read; | |
4832 | * - the log start is guaranteed to be sector | |
4833 | * aligned; | |
4834 | * - we read the log end (LR header start) | |
4835 | * _first_, then the log start (LR header end) | |
4836 | * - order is important. | |
4837 | */ | |
4838 | error = xlog_bread_offset(log, 0, | |
4839 | bblks - split_bblks, dbp, | |
4840 | offset + BBTOB(split_bblks)); | |
4841 | if (error) | |
4842 | goto bread_err2; | |
4843 | } | |
4844 | ||
4845 | error = xlog_recover_process(log, rhash, rhead, offset, | |
4846 | pass); | |
4847 | if (error) | |
4848 | goto bread_err2; | |
4849 | ||
4850 | blk_no += bblks; | |
4851 | rhead_blk = blk_no; | |
4852 | } | |
4853 | ||
4854 | ASSERT(blk_no >= log->l_logBBsize); | |
4855 | blk_no -= log->l_logBBsize; | |
4856 | rhead_blk = blk_no; | |
4857 | } | |
4858 | ||
4859 | /* read first part of physical log */ | |
4860 | while (blk_no < head_blk) { | |
4861 | error = xlog_bread(log, blk_no, hblks, hbp, &offset); | |
4862 | if (error) | |
4863 | goto bread_err2; | |
4864 | ||
4865 | rhead = (xlog_rec_header_t *)offset; | |
4866 | error = xlog_valid_rec_header(log, rhead, blk_no); | |
4867 | if (error) | |
4868 | goto bread_err2; | |
4869 | ||
4870 | /* blocks in data section */ | |
4871 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); | |
4872 | error = xlog_bread(log, blk_no+hblks, bblks, dbp, | |
4873 | &offset); | |
4874 | if (error) | |
4875 | goto bread_err2; | |
4876 | ||
4877 | error = xlog_recover_process(log, rhash, rhead, offset, pass); | |
4878 | if (error) | |
4879 | goto bread_err2; | |
4880 | ||
4881 | blk_no += bblks + hblks; | |
4882 | rhead_blk = blk_no; | |
4883 | } | |
4884 | ||
4885 | bread_err2: | |
4886 | xlog_put_bp(dbp); | |
4887 | bread_err1: | |
4888 | xlog_put_bp(hbp); | |
4889 | ||
4890 | if (error && first_bad) | |
4891 | *first_bad = rhead_blk; | |
4892 | ||
4893 | return error; | |
4894 | } | |
4895 | ||
4896 | /* | |
4897 | * Do the recovery of the log. We actually do this in two phases. | |
4898 | * The two passes are necessary in order to implement the function | |
4899 | * of cancelling a record written into the log. The first pass | |
4900 | * determines those things which have been cancelled, and the | |
4901 | * second pass replays log items normally except for those which | |
4902 | * have been cancelled. The handling of the replay and cancellations | |
4903 | * takes place in the log item type specific routines. | |
4904 | * | |
4905 | * The table of items which have cancel records in the log is allocated | |
4906 | * and freed at this level, since only here do we know when all of | |
4907 | * the log recovery has been completed. | |
4908 | */ | |
4909 | STATIC int | |
4910 | xlog_do_log_recovery( | |
4911 | struct xlog *log, | |
4912 | xfs_daddr_t head_blk, | |
4913 | xfs_daddr_t tail_blk) | |
4914 | { | |
4915 | int error, i; | |
4916 | ||
4917 | ASSERT(head_blk != tail_blk); | |
4918 | ||
4919 | /* | |
4920 | * First do a pass to find all of the cancelled buf log items. | |
4921 | * Store them in the buf_cancel_table for use in the second pass. | |
4922 | */ | |
4923 | log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE * | |
4924 | sizeof(struct list_head), | |
4925 | KM_SLEEP); | |
4926 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) | |
4927 | INIT_LIST_HEAD(&log->l_buf_cancel_table[i]); | |
4928 | ||
4929 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, | |
4930 | XLOG_RECOVER_PASS1, NULL); | |
4931 | if (error != 0) { | |
4932 | kmem_free(log->l_buf_cancel_table); | |
4933 | log->l_buf_cancel_table = NULL; | |
4934 | return error; | |
4935 | } | |
4936 | /* | |
4937 | * Then do a second pass to actually recover the items in the log. | |
4938 | * When it is complete free the table of buf cancel items. | |
4939 | */ | |
4940 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, | |
4941 | XLOG_RECOVER_PASS2, NULL); | |
4942 | #ifdef DEBUG | |
4943 | if (!error) { | |
4944 | int i; | |
4945 | ||
4946 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) | |
4947 | ASSERT(list_empty(&log->l_buf_cancel_table[i])); | |
4948 | } | |
4949 | #endif /* DEBUG */ | |
4950 | ||
4951 | kmem_free(log->l_buf_cancel_table); | |
4952 | log->l_buf_cancel_table = NULL; | |
4953 | ||
4954 | return error; | |
4955 | } | |
4956 | ||
4957 | /* | |
4958 | * Do the actual recovery | |
4959 | */ | |
4960 | STATIC int | |
4961 | xlog_do_recover( | |
4962 | struct xlog *log, | |
4963 | xfs_daddr_t head_blk, | |
4964 | xfs_daddr_t tail_blk) | |
4965 | { | |
4966 | struct xfs_mount *mp = log->l_mp; | |
4967 | int error; | |
4968 | xfs_buf_t *bp; | |
4969 | xfs_sb_t *sbp; | |
4970 | ||
4971 | /* | |
4972 | * First replay the images in the log. | |
4973 | */ | |
4974 | error = xlog_do_log_recovery(log, head_blk, tail_blk); | |
4975 | if (error) | |
4976 | return error; | |
4977 | ||
4978 | /* | |
4979 | * If IO errors happened during recovery, bail out. | |
4980 | */ | |
4981 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
4982 | return -EIO; | |
4983 | } | |
4984 | ||
4985 | /* | |
4986 | * We now update the tail_lsn since much of the recovery has completed | |
4987 | * and there may be space available to use. If there were no extent | |
4988 | * or iunlinks, we can free up the entire log and set the tail_lsn to | |
4989 | * be the last_sync_lsn. This was set in xlog_find_tail to be the | |
4990 | * lsn of the last known good LR on disk. If there are extent frees | |
4991 | * or iunlinks they will have some entries in the AIL; so we look at | |
4992 | * the AIL to determine how to set the tail_lsn. | |
4993 | */ | |
4994 | xlog_assign_tail_lsn(mp); | |
4995 | ||
4996 | /* | |
4997 | * Now that we've finished replaying all buffer and inode | |
4998 | * updates, re-read in the superblock and reverify it. | |
4999 | */ | |
5000 | bp = xfs_getsb(mp, 0); | |
5001 | bp->b_flags &= ~(XBF_DONE | XBF_ASYNC); | |
5002 | ASSERT(!(bp->b_flags & XBF_WRITE)); | |
5003 | bp->b_flags |= XBF_READ; | |
5004 | bp->b_ops = &xfs_sb_buf_ops; | |
5005 | ||
5006 | error = xfs_buf_submit_wait(bp); | |
5007 | if (error) { | |
5008 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
5009 | xfs_buf_ioerror_alert(bp, __func__); | |
5010 | ASSERT(0); | |
5011 | } | |
5012 | xfs_buf_relse(bp); | |
5013 | return error; | |
5014 | } | |
5015 | ||
5016 | /* Convert superblock from on-disk format */ | |
5017 | sbp = &mp->m_sb; | |
5018 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); | |
5019 | xfs_buf_relse(bp); | |
5020 | ||
5021 | /* re-initialise in-core superblock and geometry structures */ | |
5022 | xfs_reinit_percpu_counters(mp); | |
5023 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); | |
5024 | if (error) { | |
5025 | xfs_warn(mp, "Failed post-recovery per-ag init: %d", error); | |
5026 | return error; | |
5027 | } | |
5028 | ||
5029 | xlog_recover_check_summary(log); | |
5030 | ||
5031 | /* Normal transactions can now occur */ | |
5032 | log->l_flags &= ~XLOG_ACTIVE_RECOVERY; | |
5033 | return 0; | |
5034 | } | |
5035 | ||
5036 | /* | |
5037 | * Perform recovery and re-initialize some log variables in xlog_find_tail. | |
5038 | * | |
5039 | * Return error or zero. | |
5040 | */ | |
5041 | int | |
5042 | xlog_recover( | |
5043 | struct xlog *log) | |
5044 | { | |
5045 | xfs_daddr_t head_blk, tail_blk; | |
5046 | int error; | |
5047 | ||
5048 | /* find the tail of the log */ | |
5049 | error = xlog_find_tail(log, &head_blk, &tail_blk); | |
5050 | if (error) | |
5051 | return error; | |
5052 | ||
5053 | /* | |
5054 | * The superblock was read before the log was available and thus the LSN | |
5055 | * could not be verified. Check the superblock LSN against the current | |
5056 | * LSN now that it's known. | |
5057 | */ | |
5058 | if (xfs_sb_version_hascrc(&log->l_mp->m_sb) && | |
5059 | !xfs_log_check_lsn(log->l_mp, log->l_mp->m_sb.sb_lsn)) | |
5060 | return -EINVAL; | |
5061 | ||
5062 | if (tail_blk != head_blk) { | |
5063 | /* There used to be a comment here: | |
5064 | * | |
5065 | * disallow recovery on read-only mounts. note -- mount | |
5066 | * checks for ENOSPC and turns it into an intelligent | |
5067 | * error message. | |
5068 | * ...but this is no longer true. Now, unless you specify | |
5069 | * NORECOVERY (in which case this function would never be | |
5070 | * called), we just go ahead and recover. We do this all | |
5071 | * under the vfs layer, so we can get away with it unless | |
5072 | * the device itself is read-only, in which case we fail. | |
5073 | */ | |
5074 | if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) { | |
5075 | return error; | |
5076 | } | |
5077 | ||
5078 | /* | |
5079 | * Version 5 superblock log feature mask validation. We know the | |
5080 | * log is dirty so check if there are any unknown log features | |
5081 | * in what we need to recover. If there are unknown features | |
5082 | * (e.g. unsupported transactions, then simply reject the | |
5083 | * attempt at recovery before touching anything. | |
5084 | */ | |
5085 | if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 && | |
5086 | xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb, | |
5087 | XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) { | |
5088 | xfs_warn(log->l_mp, | |
5089 | "Superblock has unknown incompatible log features (0x%x) enabled.", | |
5090 | (log->l_mp->m_sb.sb_features_log_incompat & | |
5091 | XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)); | |
5092 | xfs_warn(log->l_mp, | |
5093 | "The log can not be fully and/or safely recovered by this kernel."); | |
5094 | xfs_warn(log->l_mp, | |
5095 | "Please recover the log on a kernel that supports the unknown features."); | |
5096 | return -EINVAL; | |
5097 | } | |
5098 | ||
5099 | /* | |
5100 | * Delay log recovery if the debug hook is set. This is debug | |
5101 | * instrumention to coordinate simulation of I/O failures with | |
5102 | * log recovery. | |
5103 | */ | |
5104 | if (xfs_globals.log_recovery_delay) { | |
5105 | xfs_notice(log->l_mp, | |
5106 | "Delaying log recovery for %d seconds.", | |
5107 | xfs_globals.log_recovery_delay); | |
5108 | msleep(xfs_globals.log_recovery_delay * 1000); | |
5109 | } | |
5110 | ||
5111 | xfs_notice(log->l_mp, "Starting recovery (logdev: %s)", | |
5112 | log->l_mp->m_logname ? log->l_mp->m_logname | |
5113 | : "internal"); | |
5114 | ||
5115 | error = xlog_do_recover(log, head_blk, tail_blk); | |
5116 | log->l_flags |= XLOG_RECOVERY_NEEDED; | |
5117 | } | |
5118 | return error; | |
5119 | } | |
5120 | ||
5121 | /* | |
5122 | * In the first part of recovery we replay inodes and buffers and build | |
5123 | * up the list of extent free items which need to be processed. Here | |
5124 | * we process the extent free items and clean up the on disk unlinked | |
5125 | * inode lists. This is separated from the first part of recovery so | |
5126 | * that the root and real-time bitmap inodes can be read in from disk in | |
5127 | * between the two stages. This is necessary so that we can free space | |
5128 | * in the real-time portion of the file system. | |
5129 | */ | |
5130 | int | |
5131 | xlog_recover_finish( | |
5132 | struct xlog *log) | |
5133 | { | |
5134 | /* | |
5135 | * Now we're ready to do the transactions needed for the | |
5136 | * rest of recovery. Start with completing all the extent | |
5137 | * free intent records and then process the unlinked inode | |
5138 | * lists. At this point, we essentially run in normal mode | |
5139 | * except that we're still performing recovery actions | |
5140 | * rather than accepting new requests. | |
5141 | */ | |
5142 | if (log->l_flags & XLOG_RECOVERY_NEEDED) { | |
5143 | int error; | |
5144 | error = xlog_recover_process_efis(log); | |
5145 | if (error) { | |
5146 | xfs_alert(log->l_mp, "Failed to recover EFIs"); | |
5147 | return error; | |
5148 | } | |
5149 | /* | |
5150 | * Sync the log to get all the EFIs out of the AIL. | |
5151 | * This isn't absolutely necessary, but it helps in | |
5152 | * case the unlink transactions would have problems | |
5153 | * pushing the EFIs out of the way. | |
5154 | */ | |
5155 | xfs_log_force(log->l_mp, XFS_LOG_SYNC); | |
5156 | ||
5157 | xlog_recover_process_iunlinks(log); | |
5158 | ||
5159 | xlog_recover_check_summary(log); | |
5160 | ||
5161 | xfs_notice(log->l_mp, "Ending recovery (logdev: %s)", | |
5162 | log->l_mp->m_logname ? log->l_mp->m_logname | |
5163 | : "internal"); | |
5164 | log->l_flags &= ~XLOG_RECOVERY_NEEDED; | |
5165 | } else { | |
5166 | xfs_info(log->l_mp, "Ending clean mount"); | |
5167 | } | |
5168 | return 0; | |
5169 | } | |
5170 | ||
5171 | int | |
5172 | xlog_recover_cancel( | |
5173 | struct xlog *log) | |
5174 | { | |
5175 | int error = 0; | |
5176 | ||
5177 | if (log->l_flags & XLOG_RECOVERY_NEEDED) | |
5178 | error = xlog_recover_cancel_efis(log); | |
5179 | ||
5180 | return error; | |
5181 | } | |
5182 | ||
5183 | #if defined(DEBUG) | |
5184 | /* | |
5185 | * Read all of the agf and agi counters and check that they | |
5186 | * are consistent with the superblock counters. | |
5187 | */ | |
5188 | void | |
5189 | xlog_recover_check_summary( | |
5190 | struct xlog *log) | |
5191 | { | |
5192 | xfs_mount_t *mp; | |
5193 | xfs_agf_t *agfp; | |
5194 | xfs_buf_t *agfbp; | |
5195 | xfs_buf_t *agibp; | |
5196 | xfs_agnumber_t agno; | |
5197 | __uint64_t freeblks; | |
5198 | __uint64_t itotal; | |
5199 | __uint64_t ifree; | |
5200 | int error; | |
5201 | ||
5202 | mp = log->l_mp; | |
5203 | ||
5204 | freeblks = 0LL; | |
5205 | itotal = 0LL; | |
5206 | ifree = 0LL; | |
5207 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
5208 | error = xfs_read_agf(mp, NULL, agno, 0, &agfbp); | |
5209 | if (error) { | |
5210 | xfs_alert(mp, "%s agf read failed agno %d error %d", | |
5211 | __func__, agno, error); | |
5212 | } else { | |
5213 | agfp = XFS_BUF_TO_AGF(agfbp); | |
5214 | freeblks += be32_to_cpu(agfp->agf_freeblks) + | |
5215 | be32_to_cpu(agfp->agf_flcount); | |
5216 | xfs_buf_relse(agfbp); | |
5217 | } | |
5218 | ||
5219 | error = xfs_read_agi(mp, NULL, agno, &agibp); | |
5220 | if (error) { | |
5221 | xfs_alert(mp, "%s agi read failed agno %d error %d", | |
5222 | __func__, agno, error); | |
5223 | } else { | |
5224 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp); | |
5225 | ||
5226 | itotal += be32_to_cpu(agi->agi_count); | |
5227 | ifree += be32_to_cpu(agi->agi_freecount); | |
5228 | xfs_buf_relse(agibp); | |
5229 | } | |
5230 | } | |
5231 | } | |
5232 | #endif /* DEBUG */ |