]>
Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3 | * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved. | |
4 | * | |
5 | * This copyrighted material is made available to anyone wishing to use, | |
6 | * modify, copy, or redistribute it subject to the terms and conditions | |
7 | * of the GNU General Public License v.2. | |
8 | */ | |
9 | ||
10 | #include <linux/sched.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/spinlock.h> | |
13 | #include <linux/completion.h> | |
14 | #include <linux/buffer_head.h> | |
15 | #include <asm/semaphore.h> | |
16 | ||
17 | #include "gfs2.h" | |
18 | #include "bmap.h" | |
19 | #include "glock.h" | |
20 | #include "glops.h" | |
21 | #include "lm.h" | |
22 | #include "lops.h" | |
23 | #include "meta_io.h" | |
24 | #include "recovery.h" | |
25 | #include "super.h" | |
26 | ||
27 | int gfs2_replay_read_block(struct gfs2_jdesc *jd, unsigned int blk, | |
28 | struct buffer_head **bh) | |
29 | { | |
7359a19c | 30 | struct gfs2_glock *gl = get_v2ip(jd->jd_inode)->i_gl; |
b3b94faa DT |
31 | int new = 0; |
32 | uint64_t dblock; | |
33 | uint32_t extlen; | |
34 | int error; | |
35 | ||
568f4c96 SW |
36 | error = gfs2_block_map(get_v2ip(jd->jd_inode), blk, &new, &dblock, |
37 | &extlen); | |
b3b94faa DT |
38 | if (error) |
39 | return error; | |
40 | if (!dblock) { | |
7359a19c | 41 | gfs2_consist_inode(get_v2ip(jd->jd_inode)); |
b3b94faa DT |
42 | return -EIO; |
43 | } | |
44 | ||
45 | gfs2_meta_ra(gl, dblock, extlen); | |
46 | error = gfs2_meta_read(gl, dblock, DIO_START | DIO_WAIT, bh); | |
47 | ||
48 | return error; | |
49 | } | |
50 | ||
51 | int gfs2_revoke_add(struct gfs2_sbd *sdp, uint64_t blkno, unsigned int where) | |
52 | { | |
53 | struct list_head *head = &sdp->sd_revoke_list; | |
54 | struct gfs2_revoke_replay *rr; | |
55 | int found = 0; | |
56 | ||
57 | list_for_each_entry(rr, head, rr_list) { | |
58 | if (rr->rr_blkno == blkno) { | |
59 | found = 1; | |
60 | break; | |
61 | } | |
62 | } | |
63 | ||
64 | if (found) { | |
65 | rr->rr_where = where; | |
66 | return 0; | |
67 | } | |
68 | ||
69 | rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_KERNEL); | |
70 | if (!rr) | |
71 | return -ENOMEM; | |
72 | ||
73 | rr->rr_blkno = blkno; | |
74 | rr->rr_where = where; | |
75 | list_add(&rr->rr_list, head); | |
76 | ||
77 | return 1; | |
78 | } | |
79 | ||
80 | int gfs2_revoke_check(struct gfs2_sbd *sdp, uint64_t blkno, unsigned int where) | |
81 | { | |
82 | struct gfs2_revoke_replay *rr; | |
83 | int wrap, a, b, revoke; | |
84 | int found = 0; | |
85 | ||
86 | list_for_each_entry(rr, &sdp->sd_revoke_list, rr_list) { | |
87 | if (rr->rr_blkno == blkno) { | |
88 | found = 1; | |
89 | break; | |
90 | } | |
91 | } | |
92 | ||
93 | if (!found) | |
94 | return 0; | |
95 | ||
96 | wrap = (rr->rr_where < sdp->sd_replay_tail); | |
97 | a = (sdp->sd_replay_tail < where); | |
98 | b = (where < rr->rr_where); | |
99 | revoke = (wrap) ? (a || b) : (a && b); | |
100 | ||
101 | return revoke; | |
102 | } | |
103 | ||
104 | void gfs2_revoke_clean(struct gfs2_sbd *sdp) | |
105 | { | |
106 | struct list_head *head = &sdp->sd_revoke_list; | |
107 | struct gfs2_revoke_replay *rr; | |
108 | ||
109 | while (!list_empty(head)) { | |
110 | rr = list_entry(head->next, struct gfs2_revoke_replay, rr_list); | |
111 | list_del(&rr->rr_list); | |
112 | kfree(rr); | |
113 | } | |
114 | } | |
115 | ||
116 | /** | |
117 | * get_log_header - read the log header for a given segment | |
118 | * @jd: the journal | |
119 | * @blk: the block to look at | |
120 | * @lh: the log header to return | |
121 | * | |
122 | * Read the log header for a given segement in a given journal. Do a few | |
123 | * sanity checks on it. | |
124 | * | |
125 | * Returns: 0 on success, | |
126 | * 1 if the header was invalid or incomplete, | |
127 | * errno on error | |
128 | */ | |
129 | ||
130 | static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk, | |
131 | struct gfs2_log_header *head) | |
132 | { | |
133 | struct buffer_head *bh; | |
134 | struct gfs2_log_header lh; | |
135 | uint32_t hash; | |
136 | int error; | |
137 | ||
138 | error = gfs2_replay_read_block(jd, blk, &bh); | |
139 | if (error) | |
140 | return error; | |
141 | ||
142 | memcpy(&lh, bh->b_data, sizeof(struct gfs2_log_header)); | |
143 | lh.lh_hash = 0; | |
144 | hash = gfs2_disk_hash((char *)&lh, sizeof(struct gfs2_log_header)); | |
145 | gfs2_log_header_in(&lh, bh->b_data); | |
146 | ||
147 | brelse(bh); | |
148 | ||
149 | if (lh.lh_header.mh_magic != GFS2_MAGIC || | |
150 | lh.lh_header.mh_type != GFS2_METATYPE_LH || | |
151 | lh.lh_blkno != blk || | |
152 | lh.lh_hash != hash) | |
153 | return 1; | |
154 | ||
155 | *head = lh; | |
156 | ||
157 | return 0; | |
158 | } | |
159 | ||
160 | /** | |
161 | * find_good_lh - find a good log header | |
162 | * @jd: the journal | |
163 | * @blk: the segment to start searching from | |
164 | * @lh: the log header to fill in | |
165 | * @forward: if true search forward in the log, else search backward | |
166 | * | |
167 | * Call get_log_header() to get a log header for a segment, but if the | |
168 | * segment is bad, either scan forward or backward until we find a good one. | |
169 | * | |
170 | * Returns: errno | |
171 | */ | |
172 | ||
173 | static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk, | |
174 | struct gfs2_log_header *head) | |
175 | { | |
176 | unsigned int orig_blk = *blk; | |
177 | int error; | |
178 | ||
179 | for (;;) { | |
180 | error = get_log_header(jd, *blk, head); | |
181 | if (error <= 0) | |
182 | return error; | |
183 | ||
184 | if (++*blk == jd->jd_blocks) | |
185 | *blk = 0; | |
186 | ||
187 | if (*blk == orig_blk) { | |
7359a19c | 188 | gfs2_consist_inode(get_v2ip(jd->jd_inode)); |
b3b94faa DT |
189 | return -EIO; |
190 | } | |
191 | } | |
192 | } | |
193 | ||
194 | /** | |
195 | * jhead_scan - make sure we've found the head of the log | |
196 | * @jd: the journal | |
197 | * @head: this is filled in with the log descriptor of the head | |
198 | * | |
199 | * At this point, seg and lh should be either the head of the log or just | |
200 | * before. Scan forward until we find the head. | |
201 | * | |
202 | * Returns: errno | |
203 | */ | |
204 | ||
205 | static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header *head) | |
206 | { | |
207 | unsigned int blk = head->lh_blkno; | |
208 | struct gfs2_log_header lh; | |
209 | int error; | |
210 | ||
211 | for (;;) { | |
212 | if (++blk == jd->jd_blocks) | |
213 | blk = 0; | |
214 | ||
215 | error = get_log_header(jd, blk, &lh); | |
216 | if (error < 0) | |
217 | return error; | |
218 | if (error == 1) | |
219 | continue; | |
220 | ||
221 | if (lh.lh_sequence == head->lh_sequence) { | |
7359a19c | 222 | gfs2_consist_inode(get_v2ip(jd->jd_inode)); |
b3b94faa DT |
223 | return -EIO; |
224 | } | |
225 | if (lh.lh_sequence < head->lh_sequence) | |
226 | break; | |
227 | ||
228 | *head = lh; | |
229 | } | |
230 | ||
231 | return 0; | |
232 | } | |
233 | ||
234 | /** | |
235 | * gfs2_find_jhead - find the head of a log | |
236 | * @jd: the journal | |
237 | * @head: the log descriptor for the head of the log is returned here | |
238 | * | |
239 | * Do a binary search of a journal and find the valid log entry with the | |
240 | * highest sequence number. (i.e. the log head) | |
241 | * | |
242 | * Returns: errno | |
243 | */ | |
244 | ||
245 | int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header *head) | |
246 | { | |
247 | struct gfs2_log_header lh_1, lh_m; | |
248 | uint32_t blk_1, blk_2, blk_m; | |
249 | int error; | |
250 | ||
251 | blk_1 = 0; | |
252 | blk_2 = jd->jd_blocks - 1; | |
253 | ||
254 | for (;;) { | |
255 | blk_m = (blk_1 + blk_2) / 2; | |
256 | ||
257 | error = find_good_lh(jd, &blk_1, &lh_1); | |
258 | if (error) | |
259 | return error; | |
260 | ||
261 | error = find_good_lh(jd, &blk_m, &lh_m); | |
262 | if (error) | |
263 | return error; | |
264 | ||
265 | if (blk_1 == blk_m || blk_m == blk_2) | |
266 | break; | |
267 | ||
268 | if (lh_1.lh_sequence <= lh_m.lh_sequence) | |
269 | blk_1 = blk_m; | |
270 | else | |
271 | blk_2 = blk_m; | |
272 | } | |
273 | ||
274 | error = jhead_scan(jd, &lh_1); | |
275 | if (error) | |
276 | return error; | |
277 | ||
278 | *head = lh_1; | |
279 | ||
280 | return error; | |
281 | } | |
282 | ||
283 | /** | |
284 | * foreach_descriptor - go through the active part of the log | |
285 | * @jd: the journal | |
286 | * @start: the first log header in the active region | |
287 | * @end: the last log header (don't process the contents of this entry)) | |
288 | * | |
289 | * Call a given function once for every log descriptor in the active | |
290 | * portion of the log. | |
291 | * | |
292 | * Returns: errno | |
293 | */ | |
294 | ||
295 | static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start, | |
296 | unsigned int end, int pass) | |
297 | { | |
7359a19c | 298 | struct gfs2_sbd *sdp = get_v2ip(jd->jd_inode)->i_sbd; |
b3b94faa DT |
299 | struct buffer_head *bh; |
300 | struct gfs2_log_descriptor *ld; | |
301 | int error = 0; | |
302 | u32 length; | |
303 | __be64 *ptr; | |
304 | unsigned int offset = sizeof(struct gfs2_log_descriptor); | |
305 | offset += (sizeof(__be64)-1); | |
306 | offset &= ~(sizeof(__be64)-1); | |
307 | ||
308 | while (start != end) { | |
309 | error = gfs2_replay_read_block(jd, start, &bh); | |
310 | if (error) | |
311 | return error; | |
312 | if (gfs2_meta_check(sdp, bh)) { | |
313 | brelse(bh); | |
314 | return -EIO; | |
315 | } | |
316 | ld = (struct gfs2_log_descriptor *)bh->b_data; | |
317 | length = be32_to_cpu(ld->ld_length); | |
318 | ||
319 | if (be16_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) { | |
320 | struct gfs2_log_header lh; | |
321 | error = get_log_header(jd, start, &lh); | |
322 | if (!error) { | |
323 | gfs2_replay_incr_blk(sdp, &start); | |
324 | continue; | |
325 | } | |
326 | if (error == 1) { | |
7359a19c | 327 | gfs2_consist_inode(get_v2ip(jd->jd_inode)); |
b3b94faa DT |
328 | error = -EIO; |
329 | } | |
330 | brelse(bh); | |
331 | return error; | |
332 | } else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) { | |
333 | brelse(bh); | |
334 | return -EIO; | |
335 | } | |
336 | ptr = (__be64 *)(bh->b_data + offset); | |
337 | error = lops_scan_elements(jd, start, ld, ptr, pass); | |
338 | if (error) { | |
339 | brelse(bh); | |
340 | return error; | |
341 | } | |
342 | ||
343 | while (length--) | |
344 | gfs2_replay_incr_blk(sdp, &start); | |
345 | ||
346 | brelse(bh); | |
347 | } | |
348 | ||
349 | return 0; | |
350 | } | |
351 | ||
352 | /** | |
353 | * clean_journal - mark a dirty journal as being clean | |
354 | * @sdp: the filesystem | |
355 | * @jd: the journal | |
356 | * @gl: the journal's glock | |
357 | * @head: the head journal to start from | |
358 | * | |
359 | * Returns: errno | |
360 | */ | |
361 | ||
362 | static int clean_journal(struct gfs2_jdesc *jd, struct gfs2_log_header *head) | |
363 | { | |
7359a19c | 364 | struct gfs2_inode *ip = get_v2ip(jd->jd_inode); |
b3b94faa DT |
365 | struct gfs2_sbd *sdp = ip->i_sbd; |
366 | unsigned int lblock; | |
367 | int new = 0; | |
368 | uint64_t dblock; | |
369 | struct gfs2_log_header *lh; | |
370 | uint32_t hash; | |
371 | struct buffer_head *bh; | |
372 | int error; | |
373 | ||
374 | lblock = head->lh_blkno; | |
375 | gfs2_replay_incr_blk(sdp, &lblock); | |
376 | error = gfs2_block_map(ip, lblock, &new, &dblock, NULL); | |
377 | if (error) | |
378 | return error; | |
379 | if (!dblock) { | |
380 | gfs2_consist_inode(ip); | |
381 | return -EIO; | |
382 | } | |
383 | ||
384 | bh = sb_getblk(sdp->sd_vfs, dblock); | |
385 | lock_buffer(bh); | |
386 | memset(bh->b_data, 0, bh->b_size); | |
387 | set_buffer_uptodate(bh); | |
388 | clear_buffer_dirty(bh); | |
389 | unlock_buffer(bh); | |
390 | ||
391 | lh = (struct gfs2_log_header *)bh->b_data; | |
392 | memset(lh, 0, sizeof(struct gfs2_log_header)); | |
393 | lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC); | |
394 | lh->lh_header.mh_type = cpu_to_be16(GFS2_METATYPE_LH); | |
395 | lh->lh_header.mh_format = cpu_to_be16(GFS2_FORMAT_LH); | |
396 | lh->lh_sequence = cpu_to_be64(head->lh_sequence + 1); | |
397 | lh->lh_flags = cpu_to_be32(GFS2_LOG_HEAD_UNMOUNT); | |
398 | lh->lh_blkno = cpu_to_be32(lblock); | |
399 | hash = gfs2_disk_hash((const char *)lh, sizeof(struct gfs2_log_header)); | |
400 | lh->lh_hash = cpu_to_be32(hash); | |
401 | ||
402 | set_buffer_dirty(bh); | |
403 | if (sync_dirty_buffer(bh)) | |
404 | gfs2_io_error_bh(sdp, bh); | |
405 | brelse(bh); | |
406 | ||
407 | return error; | |
408 | } | |
409 | ||
410 | /** | |
411 | * gfs2_recover_journal - recovery a given journal | |
412 | * @jd: the struct gfs2_jdesc describing the journal | |
413 | * @wait: Don't return until the journal is clean (or an error is encountered) | |
414 | * | |
415 | * Acquire the journal's lock, check to see if the journal is clean, and | |
416 | * do recovery if necessary. | |
417 | * | |
418 | * Returns: errno | |
419 | */ | |
420 | ||
421 | int gfs2_recover_journal(struct gfs2_jdesc *jd, int wait) | |
422 | { | |
7359a19c | 423 | struct gfs2_sbd *sdp = get_v2ip(jd->jd_inode)->i_sbd; |
b3b94faa DT |
424 | struct gfs2_log_header head; |
425 | struct gfs2_holder j_gh, ji_gh, t_gh; | |
426 | unsigned long t; | |
427 | int ro = 0; | |
428 | unsigned int pass; | |
429 | int error; | |
430 | ||
431 | fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n", jd->jd_jid); | |
432 | ||
433 | /* Aquire the journal lock so we can do recovery */ | |
434 | ||
435 | error = gfs2_glock_nq_num(sdp, | |
436 | jd->jd_jid, &gfs2_journal_glops, | |
437 | LM_ST_EXCLUSIVE, | |
438 | LM_FLAG_NOEXP | | |
439 | ((wait) ? 0 : LM_FLAG_TRY) | | |
440 | GL_NOCACHE, &j_gh); | |
441 | switch (error) { | |
442 | case 0: | |
443 | break; | |
444 | ||
445 | case GLR_TRYFAILED: | |
446 | fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid); | |
447 | error = 0; | |
448 | ||
449 | default: | |
450 | goto fail; | |
451 | }; | |
452 | ||
7359a19c | 453 | error = gfs2_glock_nq_init(get_v2ip(jd->jd_inode)->i_gl, LM_ST_SHARED, |
b3b94faa DT |
454 | LM_FLAG_NOEXP, &ji_gh); |
455 | if (error) | |
456 | goto fail_gunlock_j; | |
457 | ||
458 | fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid); | |
459 | ||
460 | error = gfs2_jdesc_check(jd); | |
461 | if (error) | |
462 | goto fail_gunlock_ji; | |
463 | ||
464 | error = gfs2_find_jhead(jd, &head); | |
465 | if (error) | |
466 | goto fail_gunlock_ji; | |
467 | ||
468 | if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) { | |
469 | fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n", | |
470 | jd->jd_jid); | |
471 | ||
472 | t = jiffies; | |
473 | ||
474 | /* Acquire a shared hold on the transaction lock */ | |
475 | ||
476 | error = gfs2_glock_nq_init(sdp->sd_trans_gl, | |
477 | LM_ST_SHARED, | |
478 | LM_FLAG_NOEXP | | |
479 | LM_FLAG_PRIORITY | | |
480 | GL_NEVER_RECURSE | | |
481 | GL_NOCANCEL | | |
482 | GL_NOCACHE, | |
483 | &t_gh); | |
484 | if (error) | |
485 | goto fail_gunlock_ji; | |
486 | ||
487 | if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) { | |
488 | if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) | |
489 | ro = 1; | |
490 | } else { | |
491 | if (sdp->sd_vfs->s_flags & MS_RDONLY) | |
492 | ro = 1; | |
493 | } | |
494 | ||
495 | if (ro) { | |
496 | fs_warn(sdp, "jid=%u: Can't replay: read-only FS\n", | |
497 | jd->jd_jid); | |
498 | error = -EROFS; | |
499 | goto fail_gunlock_tr; | |
500 | } | |
501 | ||
502 | fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid); | |
503 | ||
504 | for (pass = 0; pass < 2; pass++) { | |
505 | lops_before_scan(jd, &head, pass); | |
506 | error = foreach_descriptor(jd, head.lh_tail, | |
507 | head.lh_blkno, pass); | |
508 | lops_after_scan(jd, error, pass); | |
509 | if (error) | |
510 | goto fail_gunlock_tr; | |
511 | } | |
512 | ||
513 | error = clean_journal(jd, &head); | |
514 | if (error) | |
515 | goto fail_gunlock_tr; | |
516 | ||
517 | gfs2_glock_dq_uninit(&t_gh); | |
518 | ||
519 | t = DIV_RU(jiffies - t, HZ); | |
520 | ||
521 | fs_info(sdp, "jid=%u: Journal replayed in %lus\n", | |
522 | jd->jd_jid, t); | |
523 | } | |
524 | ||
525 | gfs2_glock_dq_uninit(&ji_gh); | |
526 | ||
527 | gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS); | |
528 | ||
529 | gfs2_glock_dq_uninit(&j_gh); | |
530 | ||
531 | fs_info(sdp, "jid=%u: Done\n", jd->jd_jid); | |
532 | ||
533 | return 0; | |
534 | ||
535 | fail_gunlock_tr: | |
536 | gfs2_glock_dq_uninit(&t_gh); | |
537 | ||
538 | fail_gunlock_ji: | |
539 | gfs2_glock_dq_uninit(&ji_gh); | |
540 | ||
541 | fail_gunlock_j: | |
542 | gfs2_glock_dq_uninit(&j_gh); | |
543 | ||
544 | fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done"); | |
545 | ||
546 | fail: | |
547 | gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP); | |
548 | ||
549 | return error; | |
550 | } | |
551 | ||
552 | /** | |
553 | * gfs2_check_journals - Recover any dirty journals | |
554 | * @sdp: the filesystem | |
555 | * | |
556 | */ | |
557 | ||
558 | void gfs2_check_journals(struct gfs2_sbd *sdp) | |
559 | { | |
560 | struct gfs2_jdesc *jd; | |
561 | ||
562 | for (;;) { | |
563 | jd = gfs2_jdesc_find_dirty(sdp); | |
564 | if (!jd) | |
565 | break; | |
566 | ||
567 | if (jd != sdp->sd_jdesc) | |
568 | gfs2_recover_journal(jd, NO_WAIT); | |
569 | } | |
570 | } | |
571 |