1
2
3
4
5
6
7
8
9
10#include <linux/module.h>
11#include <linux/slab.h>
12#include <linux/spinlock.h>
13#include <linux/completion.h>
14#include <linux/buffer_head.h>
15#include <linux/gfs2_ondisk.h>
16#include <linux/crc32.h>
17#include <linux/slow-work.h>
18
19#include "gfs2.h"
20#include "incore.h"
21#include "bmap.h"
22#include "glock.h"
23#include "glops.h"
24#include "lops.h"
25#include "meta_io.h"
26#include "recovery.h"
27#include "super.h"
28#include "util.h"
29#include "dir.h"
30
31int gfs2_replay_read_block(struct gfs2_jdesc *jd, unsigned int blk,
32 struct buffer_head **bh)
33{
34 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
35 struct gfs2_glock *gl = ip->i_gl;
36 int new = 0;
37 u64 dblock;
38 u32 extlen;
39 int error;
40
41 error = gfs2_extent_map(&ip->i_inode, blk, &new, &dblock, &extlen);
42 if (error)
43 return error;
44 if (!dblock) {
45 gfs2_consist_inode(ip);
46 return -EIO;
47 }
48
49 *bh = gfs2_meta_ra(gl, dblock, extlen);
50
51 return error;
52}
53
54int gfs2_revoke_add(struct gfs2_sbd *sdp, u64 blkno, unsigned int where)
55{
56 struct list_head *head = &sdp->sd_revoke_list;
57 struct gfs2_revoke_replay *rr;
58 int found = 0;
59
60 list_for_each_entry(rr, head, rr_list) {
61 if (rr->rr_blkno == blkno) {
62 found = 1;
63 break;
64 }
65 }
66
67 if (found) {
68 rr->rr_where = where;
69 return 0;
70 }
71
72 rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_NOFS);
73 if (!rr)
74 return -ENOMEM;
75
76 rr->rr_blkno = blkno;
77 rr->rr_where = where;
78 list_add(&rr->rr_list, head);
79
80 return 1;
81}
82
83int gfs2_revoke_check(struct gfs2_sbd *sdp, u64 blkno, unsigned int where)
84{
85 struct gfs2_revoke_replay *rr;
86 int wrap, a, b, revoke;
87 int found = 0;
88
89 list_for_each_entry(rr, &sdp->sd_revoke_list, rr_list) {
90 if (rr->rr_blkno == blkno) {
91 found = 1;
92 break;
93 }
94 }
95
96 if (!found)
97 return 0;
98
99 wrap = (rr->rr_where < sdp->sd_replay_tail);
100 a = (sdp->sd_replay_tail < where);
101 b = (where < rr->rr_where);
102 revoke = (wrap) ? (a || b) : (a && b);
103
104 return revoke;
105}
106
107void gfs2_revoke_clean(struct gfs2_sbd *sdp)
108{
109 struct list_head *head = &sdp->sd_revoke_list;
110 struct gfs2_revoke_replay *rr;
111
112 while (!list_empty(head)) {
113 rr = list_entry(head->next, struct gfs2_revoke_replay, rr_list);
114 list_del(&rr->rr_list);
115 kfree(rr);
116 }
117}
118
119static int gfs2_log_header_in(struct gfs2_log_header_host *lh, const void *buf)
120{
121 const struct gfs2_log_header *str = buf;
122
123 if (str->lh_header.mh_magic != cpu_to_be32(GFS2_MAGIC) ||
124 str->lh_header.mh_type != cpu_to_be32(GFS2_METATYPE_LH))
125 return 1;
126
127 lh->lh_sequence = be64_to_cpu(str->lh_sequence);
128 lh->lh_flags = be32_to_cpu(str->lh_flags);
129 lh->lh_tail = be32_to_cpu(str->lh_tail);
130 lh->lh_blkno = be32_to_cpu(str->lh_blkno);
131 lh->lh_hash = be32_to_cpu(str->lh_hash);
132 return 0;
133}
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
150 struct gfs2_log_header_host *head)
151{
152 struct buffer_head *bh;
153 struct gfs2_log_header_host uninitialized_var(lh);
154 const u32 nothing = 0;
155 u32 hash;
156 int error;
157
158 error = gfs2_replay_read_block(jd, blk, &bh);
159 if (error)
160 return error;
161
162 hash = crc32_le((u32)~0, bh->b_data, sizeof(struct gfs2_log_header) -
163 sizeof(u32));
164 hash = crc32_le(hash, (unsigned char const *)¬hing, sizeof(nothing));
165 hash ^= (u32)~0;
166 error = gfs2_log_header_in(&lh, bh->b_data);
167 brelse(bh);
168
169 if (error || lh.lh_blkno != blk || lh.lh_hash != hash)
170 return 1;
171
172 *head = lh;
173
174 return 0;
175}
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk,
191 struct gfs2_log_header_host *head)
192{
193 unsigned int orig_blk = *blk;
194 int error;
195
196 for (;;) {
197 error = get_log_header(jd, *blk, head);
198 if (error <= 0)
199 return error;
200
201 if (++*blk == jd->jd_blocks)
202 *blk = 0;
203
204 if (*blk == orig_blk) {
205 gfs2_consist_inode(GFS2_I(jd->jd_inode));
206 return -EIO;
207 }
208 }
209}
210
211
212
213
214
215
216
217
218
219
220
221
222static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
223{
224 unsigned int blk = head->lh_blkno;
225 struct gfs2_log_header_host lh;
226 int error;
227
228 for (;;) {
229 if (++blk == jd->jd_blocks)
230 blk = 0;
231
232 error = get_log_header(jd, blk, &lh);
233 if (error < 0)
234 return error;
235 if (error == 1)
236 continue;
237
238 if (lh.lh_sequence == head->lh_sequence) {
239 gfs2_consist_inode(GFS2_I(jd->jd_inode));
240 return -EIO;
241 }
242 if (lh.lh_sequence < head->lh_sequence)
243 break;
244
245 *head = lh;
246 }
247
248 return 0;
249}
250
251
252
253
254
255
256
257
258
259
260
261
262int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
263{
264 struct gfs2_log_header_host lh_1, lh_m;
265 u32 blk_1, blk_2, blk_m;
266 int error;
267
268 blk_1 = 0;
269 blk_2 = jd->jd_blocks - 1;
270
271 for (;;) {
272 blk_m = (blk_1 + blk_2) / 2;
273
274 error = find_good_lh(jd, &blk_1, &lh_1);
275 if (error)
276 return error;
277
278 error = find_good_lh(jd, &blk_m, &lh_m);
279 if (error)
280 return error;
281
282 if (blk_1 == blk_m || blk_m == blk_2)
283 break;
284
285 if (lh_1.lh_sequence <= lh_m.lh_sequence)
286 blk_1 = blk_m;
287 else
288 blk_2 = blk_m;
289 }
290
291 error = jhead_scan(jd, &lh_1);
292 if (error)
293 return error;
294
295 *head = lh_1;
296
297 return error;
298}
299
300
301
302
303
304
305
306
307
308
309
310
311
312static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start,
313 unsigned int end, int pass)
314{
315 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
316 struct buffer_head *bh;
317 struct gfs2_log_descriptor *ld;
318 int error = 0;
319 u32 length;
320 __be64 *ptr;
321 unsigned int offset = sizeof(struct gfs2_log_descriptor);
322 offset += sizeof(__be64) - 1;
323 offset &= ~(sizeof(__be64) - 1);
324
325 while (start != end) {
326 error = gfs2_replay_read_block(jd, start, &bh);
327 if (error)
328 return error;
329 if (gfs2_meta_check(sdp, bh)) {
330 brelse(bh);
331 return -EIO;
332 }
333 ld = (struct gfs2_log_descriptor *)bh->b_data;
334 length = be32_to_cpu(ld->ld_length);
335
336 if (be32_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
337 struct gfs2_log_header_host lh;
338 error = get_log_header(jd, start, &lh);
339 if (!error) {
340 gfs2_replay_incr_blk(sdp, &start);
341 brelse(bh);
342 continue;
343 }
344 if (error == 1) {
345 gfs2_consist_inode(GFS2_I(jd->jd_inode));
346 error = -EIO;
347 }
348 brelse(bh);
349 return error;
350 } else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
351 brelse(bh);
352 return -EIO;
353 }
354 ptr = (__be64 *)(bh->b_data + offset);
355 error = lops_scan_elements(jd, start, ld, ptr, pass);
356 if (error) {
357 brelse(bh);
358 return error;
359 }
360
361 while (length--)
362 gfs2_replay_incr_blk(sdp, &start);
363
364 brelse(bh);
365 }
366
367 return 0;
368}
369
370
371
372
373
374
375
376
377
378
379
380static int clean_journal(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
381{
382 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
383 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
384 unsigned int lblock;
385 struct gfs2_log_header *lh;
386 u32 hash;
387 struct buffer_head *bh;
388 int error;
389 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
390
391 lblock = head->lh_blkno;
392 gfs2_replay_incr_blk(sdp, &lblock);
393 bh_map.b_size = 1 << ip->i_inode.i_blkbits;
394 error = gfs2_block_map(&ip->i_inode, lblock, &bh_map, 0);
395 if (error)
396 return error;
397 if (!bh_map.b_blocknr) {
398 gfs2_consist_inode(ip);
399 return -EIO;
400 }
401
402 bh = sb_getblk(sdp->sd_vfs, bh_map.b_blocknr);
403 lock_buffer(bh);
404 memset(bh->b_data, 0, bh->b_size);
405 set_buffer_uptodate(bh);
406 clear_buffer_dirty(bh);
407 unlock_buffer(bh);
408
409 lh = (struct gfs2_log_header *)bh->b_data;
410 memset(lh, 0, sizeof(struct gfs2_log_header));
411 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
412 lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
413 lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
414 lh->lh_sequence = cpu_to_be64(head->lh_sequence + 1);
415 lh->lh_flags = cpu_to_be32(GFS2_LOG_HEAD_UNMOUNT);
416 lh->lh_blkno = cpu_to_be32(lblock);
417 hash = gfs2_disk_hash((const char *)lh, sizeof(struct gfs2_log_header));
418 lh->lh_hash = cpu_to_be32(hash);
419
420 set_buffer_dirty(bh);
421 if (sync_dirty_buffer(bh))
422 gfs2_io_error_bh(sdp, bh);
423 brelse(bh);
424
425 return error;
426}
427
428
429static void gfs2_recovery_done(struct gfs2_sbd *sdp, unsigned int jid,
430 unsigned int message)
431{
432 char env_jid[20];
433 char env_status[20];
434 char *envp[] = { env_jid, env_status, NULL };
435 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
436 ls->ls_recover_jid_done = jid;
437 ls->ls_recover_jid_status = message;
438 sprintf(env_jid, "JID=%d", jid);
439 sprintf(env_status, "RECOVERY=%s",
440 message == LM_RD_SUCCESS ? "Done" : "Failed");
441 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
442}
443
444static int gfs2_recover_get_ref(struct slow_work *work)
445{
446 struct gfs2_jdesc *jd = container_of(work, struct gfs2_jdesc, jd_work);
447 if (test_and_set_bit(JDF_RECOVERY, &jd->jd_flags))
448 return -EBUSY;
449 return 0;
450}
451
452static void gfs2_recover_put_ref(struct slow_work *work)
453{
454 struct gfs2_jdesc *jd = container_of(work, struct gfs2_jdesc, jd_work);
455 clear_bit(JDF_RECOVERY, &jd->jd_flags);
456 smp_mb__after_clear_bit();
457 wake_up_bit(&jd->jd_flags, JDF_RECOVERY);
458}
459
460static void gfs2_recover_work(struct slow_work *work)
461{
462 struct gfs2_jdesc *jd = container_of(work, struct gfs2_jdesc, jd_work);
463 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
464 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
465 struct gfs2_log_header_host head;
466 struct gfs2_holder j_gh, ji_gh, t_gh;
467 unsigned long t;
468 int ro = 0;
469 unsigned int pass;
470 int error;
471
472 if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
473 fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
474 jd->jd_jid);
475
476
477
478 error = gfs2_glock_nq_num(sdp, jd->jd_jid, &gfs2_journal_glops,
479 LM_ST_EXCLUSIVE,
480 LM_FLAG_NOEXP | LM_FLAG_TRY | GL_NOCACHE,
481 &j_gh);
482 switch (error) {
483 case 0:
484 break;
485
486 case GLR_TRYFAILED:
487 fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
488 error = 0;
489
490 default:
491 goto fail;
492 };
493
494 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
495 LM_FLAG_NOEXP | GL_NOCACHE, &ji_gh);
496 if (error)
497 goto fail_gunlock_j;
498 } else {
499 fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
500 }
501
502 fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
503
504 error = gfs2_jdesc_check(jd);
505 if (error)
506 goto fail_gunlock_ji;
507
508 error = gfs2_find_jhead(jd, &head);
509 if (error)
510 goto fail_gunlock_ji;
511
512 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
513 fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
514 jd->jd_jid);
515
516 t = jiffies;
517
518
519
520 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED,
521 LM_FLAG_NOEXP | LM_FLAG_PRIORITY |
522 GL_NOCACHE, &t_gh);
523 if (error)
524 goto fail_gunlock_ji;
525
526 if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
527 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
528 ro = 1;
529 } else {
530 if (sdp->sd_vfs->s_flags & MS_RDONLY) {
531
532 ro = bdev_read_only(sdp->sd_vfs->s_bdev);
533 if (!ro) {
534 fs_info(sdp, "recovery required on "
535 "read-only filesystem.\n");
536 fs_info(sdp, "write access will be "
537 "enabled during recovery.\n");
538 }
539 }
540 }
541
542 if (ro) {
543 fs_warn(sdp, "jid=%u: Can't replay: read-only block "
544 "device\n", jd->jd_jid);
545 error = -EROFS;
546 goto fail_gunlock_tr;
547 }
548
549 fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
550
551 for (pass = 0; pass < 2; pass++) {
552 lops_before_scan(jd, &head, pass);
553 error = foreach_descriptor(jd, head.lh_tail,
554 head.lh_blkno, pass);
555 lops_after_scan(jd, error, pass);
556 if (error)
557 goto fail_gunlock_tr;
558 }
559
560 error = clean_journal(jd, &head);
561 if (error)
562 goto fail_gunlock_tr;
563
564 gfs2_glock_dq_uninit(&t_gh);
565 t = DIV_ROUND_UP(jiffies - t, HZ);
566 fs_info(sdp, "jid=%u: Journal replayed in %lus\n",
567 jd->jd_jid, t);
568 }
569
570 if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
571 gfs2_glock_dq_uninit(&ji_gh);
572
573 gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
574
575 if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
576 gfs2_glock_dq_uninit(&j_gh);
577
578 fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
579 return;
580
581fail_gunlock_tr:
582 gfs2_glock_dq_uninit(&t_gh);
583fail_gunlock_ji:
584 if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
585 gfs2_glock_dq_uninit(&ji_gh);
586fail_gunlock_j:
587 gfs2_glock_dq_uninit(&j_gh);
588 }
589
590 fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
591
592fail:
593 gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
594}
595
596struct slow_work_ops gfs2_recover_ops = {
597 .owner = THIS_MODULE,
598 .get_ref = gfs2_recover_get_ref,
599 .put_ref = gfs2_recover_put_ref,
600 .execute = gfs2_recover_work,
601};
602
603
604static int gfs2_recovery_wait(void *word)
605{
606 schedule();
607 return 0;
608}
609
610int gfs2_recover_journal(struct gfs2_jdesc *jd)
611{
612 int rv;
613 rv = slow_work_enqueue(&jd->jd_work);
614 if (rv)
615 return rv;
616 wait_on_bit(&jd->jd_flags, JDF_RECOVERY, gfs2_recovery_wait, TASK_UNINTERRUPTIBLE);
617 return 0;
618}
619
620