GFS2: dlm based recovery coordination
[linux-2.6.git] / fs / gfs2 / recovery.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2006 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 version 2.
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
18 #include "gfs2.h"
19 #include "incore.h"
20 #include "bmap.h"
21 #include "glock.h"
22 #include "glops.h"
23 #include "lops.h"
24 #include "meta_io.h"
25 #include "recovery.h"
26 #include "super.h"
27 #include "util.h"
28 #include "dir.h"
29
30 struct workqueue_struct *gfs_recovery_wq;
31
32 int gfs2_replay_read_block(struct gfs2_jdesc *jd, unsigned int blk,
33                            struct buffer_head **bh)
34 {
35         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
36         struct gfs2_glock *gl = ip->i_gl;
37         int new = 0;
38         u64 dblock;
39         u32 extlen;
40         int error;
41
42         error = gfs2_extent_map(&ip->i_inode, blk, &new, &dblock, &extlen);
43         if (error)
44                 return error;
45         if (!dblock) {
46                 gfs2_consist_inode(ip);
47                 return -EIO;
48         }
49
50         *bh = gfs2_meta_ra(gl, dblock, extlen);
51
52         return error;
53 }
54
55 int gfs2_revoke_add(struct gfs2_sbd *sdp, u64 blkno, unsigned int where)
56 {
57         struct list_head *head = &sdp->sd_revoke_list;
58         struct gfs2_revoke_replay *rr;
59         int found = 0;
60
61         list_for_each_entry(rr, head, rr_list) {
62                 if (rr->rr_blkno == blkno) {
63                         found = 1;
64                         break;
65                 }
66         }
67
68         if (found) {
69                 rr->rr_where = where;
70                 return 0;
71         }
72
73         rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_NOFS);
74         if (!rr)
75                 return -ENOMEM;
76
77         rr->rr_blkno = blkno;
78         rr->rr_where = where;
79         list_add(&rr->rr_list, head);
80
81         return 1;
82 }
83
84 int gfs2_revoke_check(struct gfs2_sbd *sdp, u64 blkno, unsigned int where)
85 {
86         struct gfs2_revoke_replay *rr;
87         int wrap, a, b, revoke;
88         int found = 0;
89
90         list_for_each_entry(rr, &sdp->sd_revoke_list, rr_list) {
91                 if (rr->rr_blkno == blkno) {
92                         found = 1;
93                         break;
94                 }
95         }
96
97         if (!found)
98                 return 0;
99
100         wrap = (rr->rr_where < sdp->sd_replay_tail);
101         a = (sdp->sd_replay_tail < where);
102         b = (where < rr->rr_where);
103         revoke = (wrap) ? (a || b) : (a && b);
104
105         return revoke;
106 }
107
108 void gfs2_revoke_clean(struct gfs2_sbd *sdp)
109 {
110         struct list_head *head = &sdp->sd_revoke_list;
111         struct gfs2_revoke_replay *rr;
112
113         while (!list_empty(head)) {
114                 rr = list_entry(head->next, struct gfs2_revoke_replay, rr_list);
115                 list_del(&rr->rr_list);
116                 kfree(rr);
117         }
118 }
119
120 static int gfs2_log_header_in(struct gfs2_log_header_host *lh, const void *buf)
121 {
122         const struct gfs2_log_header *str = buf;
123
124         if (str->lh_header.mh_magic != cpu_to_be32(GFS2_MAGIC) ||
125             str->lh_header.mh_type != cpu_to_be32(GFS2_METATYPE_LH))
126                 return 1;
127
128         lh->lh_sequence = be64_to_cpu(str->lh_sequence);
129         lh->lh_flags = be32_to_cpu(str->lh_flags);
130         lh->lh_tail = be32_to_cpu(str->lh_tail);
131         lh->lh_blkno = be32_to_cpu(str->lh_blkno);
132         lh->lh_hash = be32_to_cpu(str->lh_hash);
133         return 0;
134 }
135
136 /**
137  * get_log_header - read the log header for a given segment
138  * @jd: the journal
139  * @blk: the block to look at
140  * @lh: the log header to return
141  *
142  * Read the log header for a given segement in a given journal.  Do a few
143  * sanity checks on it.
144  *
145  * Returns: 0 on success,
146  *          1 if the header was invalid or incomplete,
147  *          errno on error
148  */
149
150 static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
151                           struct gfs2_log_header_host *head)
152 {
153         struct buffer_head *bh;
154         struct gfs2_log_header_host uninitialized_var(lh);
155         const u32 nothing = 0;
156         u32 hash;
157         int error;
158
159         error = gfs2_replay_read_block(jd, blk, &bh);
160         if (error)
161                 return error;
162
163         hash = crc32_le((u32)~0, bh->b_data, sizeof(struct gfs2_log_header) -
164                                              sizeof(u32));
165         hash = crc32_le(hash, (unsigned char const *)&nothing, sizeof(nothing));
166         hash ^= (u32)~0;
167         error = gfs2_log_header_in(&lh, bh->b_data);
168         brelse(bh);
169
170         if (error || lh.lh_blkno != blk || lh.lh_hash != hash)
171                 return 1;
172
173         *head = lh;
174
175         return 0;
176 }
177
178 /**
179  * find_good_lh - find a good log header
180  * @jd: the journal
181  * @blk: the segment to start searching from
182  * @lh: the log header to fill in
183  * @forward: if true search forward in the log, else search backward
184  *
185  * Call get_log_header() to get a log header for a segment, but if the
186  * segment is bad, either scan forward or backward until we find a good one.
187  *
188  * Returns: errno
189  */
190
191 static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk,
192                         struct gfs2_log_header_host *head)
193 {
194         unsigned int orig_blk = *blk;
195         int error;
196
197         for (;;) {
198                 error = get_log_header(jd, *blk, head);
199                 if (error <= 0)
200                         return error;
201
202                 if (++*blk == jd->jd_blocks)
203                         *blk = 0;
204
205                 if (*blk == orig_blk) {
206                         gfs2_consist_inode(GFS2_I(jd->jd_inode));
207                         return -EIO;
208                 }
209         }
210 }
211
212 /**
213  * jhead_scan - make sure we've found the head of the log
214  * @jd: the journal
215  * @head: this is filled in with the log descriptor of the head
216  *
217  * At this point, seg and lh should be either the head of the log or just
218  * before.  Scan forward until we find the head.
219  *
220  * Returns: errno
221  */
222
223 static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
224 {
225         unsigned int blk = head->lh_blkno;
226         struct gfs2_log_header_host lh;
227         int error;
228
229         for (;;) {
230                 if (++blk == jd->jd_blocks)
231                         blk = 0;
232
233                 error = get_log_header(jd, blk, &lh);
234                 if (error < 0)
235                         return error;
236                 if (error == 1)
237                         continue;
238
239                 if (lh.lh_sequence == head->lh_sequence) {
240                         gfs2_consist_inode(GFS2_I(jd->jd_inode));
241                         return -EIO;
242                 }
243                 if (lh.lh_sequence < head->lh_sequence)
244                         break;
245
246                 *head = lh;
247         }
248
249         return 0;
250 }
251
252 /**
253  * gfs2_find_jhead - find the head of a log
254  * @jd: the journal
255  * @head: the log descriptor for the head of the log is returned here
256  *
257  * Do a binary search of a journal and find the valid log entry with the
258  * highest sequence number.  (i.e. the log head)
259  *
260  * Returns: errno
261  */
262
263 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
264 {
265         struct gfs2_log_header_host lh_1, lh_m;
266         u32 blk_1, blk_2, blk_m;
267         int error;
268
269         blk_1 = 0;
270         blk_2 = jd->jd_blocks - 1;
271
272         for (;;) {
273                 blk_m = (blk_1 + blk_2) / 2;
274
275                 error = find_good_lh(jd, &blk_1, &lh_1);
276                 if (error)
277                         return error;
278
279                 error = find_good_lh(jd, &blk_m, &lh_m);
280                 if (error)
281                         return error;
282
283                 if (blk_1 == blk_m || blk_m == blk_2)
284                         break;
285
286                 if (lh_1.lh_sequence <= lh_m.lh_sequence)
287                         blk_1 = blk_m;
288                 else
289                         blk_2 = blk_m;
290         }
291
292         error = jhead_scan(jd, &lh_1);
293         if (error)
294                 return error;
295
296         *head = lh_1;
297
298         return error;
299 }
300
301 /**
302  * foreach_descriptor - go through the active part of the log
303  * @jd: the journal
304  * @start: the first log header in the active region
305  * @end: the last log header (don't process the contents of this entry))
306  *
307  * Call a given function once for every log descriptor in the active
308  * portion of the log.
309  *
310  * Returns: errno
311  */
312
313 static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start,
314                               unsigned int end, int pass)
315 {
316         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
317         struct buffer_head *bh;
318         struct gfs2_log_descriptor *ld;
319         int error = 0;
320         u32 length;
321         __be64 *ptr;
322         unsigned int offset = sizeof(struct gfs2_log_descriptor);
323         offset += sizeof(__be64) - 1;
324         offset &= ~(sizeof(__be64) - 1);
325
326         while (start != end) {
327                 error = gfs2_replay_read_block(jd, start, &bh);
328                 if (error)
329                         return error;
330                 if (gfs2_meta_check(sdp, bh)) {
331                         brelse(bh);
332                         return -EIO;
333                 }
334                 ld = (struct gfs2_log_descriptor *)bh->b_data;
335                 length = be32_to_cpu(ld->ld_length);
336
337                 if (be32_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
338                         struct gfs2_log_header_host lh;
339                         error = get_log_header(jd, start, &lh);
340                         if (!error) {
341                                 gfs2_replay_incr_blk(sdp, &start);
342                                 brelse(bh);
343                                 continue;
344                         }
345                         if (error == 1) {
346                                 gfs2_consist_inode(GFS2_I(jd->jd_inode));
347                                 error = -EIO;
348                         }
349                         brelse(bh);
350                         return error;
351                 } else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
352                         brelse(bh);
353                         return -EIO;
354                 }
355                 ptr = (__be64 *)(bh->b_data + offset);
356                 error = lops_scan_elements(jd, start, ld, ptr, pass);
357                 if (error) {
358                         brelse(bh);
359                         return error;
360                 }
361
362                 while (length--)
363                         gfs2_replay_incr_blk(sdp, &start);
364
365                 brelse(bh);
366         }
367
368         return 0;
369 }
370
371 /**
372  * clean_journal - mark a dirty journal as being clean
373  * @sdp: the filesystem
374  * @jd: the journal
375  * @gl: the journal's glock
376  * @head: the head journal to start from
377  *
378  * Returns: errno
379  */
380
381 static int clean_journal(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
382 {
383         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
384         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
385         unsigned int lblock;
386         struct gfs2_log_header *lh;
387         u32 hash;
388         struct buffer_head *bh;
389         int error;
390         struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
391
392         lblock = head->lh_blkno;
393         gfs2_replay_incr_blk(sdp, &lblock);
394         bh_map.b_size = 1 << ip->i_inode.i_blkbits;
395         error = gfs2_block_map(&ip->i_inode, lblock, &bh_map, 0);
396         if (error)
397                 return error;
398         if (!bh_map.b_blocknr) {
399                 gfs2_consist_inode(ip);
400                 return -EIO;
401         }
402
403         bh = sb_getblk(sdp->sd_vfs, bh_map.b_blocknr);
404         lock_buffer(bh);
405         memset(bh->b_data, 0, bh->b_size);
406         set_buffer_uptodate(bh);
407         clear_buffer_dirty(bh);
408         unlock_buffer(bh);
409
410         lh = (struct gfs2_log_header *)bh->b_data;
411         memset(lh, 0, sizeof(struct gfs2_log_header));
412         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
413         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
414         lh->lh_header.__pad0 = cpu_to_be64(0);
415         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
416         lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
417         lh->lh_sequence = cpu_to_be64(head->lh_sequence + 1);
418         lh->lh_flags = cpu_to_be32(GFS2_LOG_HEAD_UNMOUNT);
419         lh->lh_blkno = cpu_to_be32(lblock);
420         hash = gfs2_disk_hash((const char *)lh, sizeof(struct gfs2_log_header));
421         lh->lh_hash = cpu_to_be32(hash);
422
423         set_buffer_dirty(bh);
424         if (sync_dirty_buffer(bh))
425                 gfs2_io_error_bh(sdp, bh);
426         brelse(bh);
427
428         return error;
429 }
430
431
432 static void gfs2_recovery_done(struct gfs2_sbd *sdp, unsigned int jid,
433                                unsigned int message)
434 {
435         char env_jid[20];
436         char env_status[20];
437         char *envp[] = { env_jid, env_status, NULL };
438         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
439
440         ls->ls_recover_jid_done = jid;
441         ls->ls_recover_jid_status = message;
442         sprintf(env_jid, "JID=%d", jid);
443         sprintf(env_status, "RECOVERY=%s",
444                 message == LM_RD_SUCCESS ? "Done" : "Failed");
445         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
446
447         if (sdp->sd_lockstruct.ls_ops->lm_recovery_result)
448                 sdp->sd_lockstruct.ls_ops->lm_recovery_result(sdp, jid, message);
449 }
450
451 void gfs2_recover_func(struct work_struct *work)
452 {
453         struct gfs2_jdesc *jd = container_of(work, struct gfs2_jdesc, jd_work);
454         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
455         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
456         struct gfs2_log_header_host head;
457         struct gfs2_holder j_gh, ji_gh, t_gh;
458         unsigned long t;
459         int ro = 0;
460         unsigned int pass;
461         int error;
462         int jlocked = 0;
463
464         if (sdp->sd_args.ar_spectator ||
465             (jd->jd_jid != sdp->sd_lockstruct.ls_jid)) {
466                 fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
467                         jd->jd_jid);
468                 jlocked = 1;
469                 /* Acquire the journal lock so we can do recovery */
470
471                 error = gfs2_glock_nq_num(sdp, jd->jd_jid, &gfs2_journal_glops,
472                                           LM_ST_EXCLUSIVE,
473                                           LM_FLAG_NOEXP | LM_FLAG_TRY | GL_NOCACHE,
474                                           &j_gh);
475                 switch (error) {
476                 case 0:
477                         break;
478
479                 case GLR_TRYFAILED:
480                         fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
481                         error = 0;
482
483                 default:
484                         goto fail;
485                 };
486
487                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
488                                            LM_FLAG_NOEXP | GL_NOCACHE, &ji_gh);
489                 if (error)
490                         goto fail_gunlock_j;
491         } else {
492                 fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
493         }
494
495         fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
496
497         error = gfs2_jdesc_check(jd);
498         if (error)
499                 goto fail_gunlock_ji;
500
501         error = gfs2_find_jhead(jd, &head);
502         if (error)
503                 goto fail_gunlock_ji;
504
505         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
506                 fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
507                         jd->jd_jid);
508
509                 t = jiffies;
510
511                 /* Acquire a shared hold on the transaction lock */
512
513                 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED,
514                                            LM_FLAG_NOEXP | LM_FLAG_PRIORITY |
515                                            GL_NOCACHE, &t_gh);
516                 if (error)
517                         goto fail_gunlock_ji;
518
519                 if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
520                         if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
521                                 ro = 1;
522                 } else {
523                         if (sdp->sd_vfs->s_flags & MS_RDONLY) {
524                                 /* check if device itself is read-only */
525                                 ro = bdev_read_only(sdp->sd_vfs->s_bdev);
526                                 if (!ro) {
527                                         fs_info(sdp, "recovery required on "
528                                                 "read-only filesystem.\n");
529                                         fs_info(sdp, "write access will be "
530                                                 "enabled during recovery.\n");
531                                 }
532                         }
533                 }
534
535                 if (ro) {
536                         fs_warn(sdp, "jid=%u: Can't replay: read-only block "
537                                 "device\n", jd->jd_jid);
538                         error = -EROFS;
539                         goto fail_gunlock_tr;
540                 }
541
542                 fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
543
544                 for (pass = 0; pass < 2; pass++) {
545                         lops_before_scan(jd, &head, pass);
546                         error = foreach_descriptor(jd, head.lh_tail,
547                                                    head.lh_blkno, pass);
548                         lops_after_scan(jd, error, pass);
549                         if (error)
550                                 goto fail_gunlock_tr;
551                 }
552
553                 error = clean_journal(jd, &head);
554                 if (error)
555                         goto fail_gunlock_tr;
556
557                 gfs2_glock_dq_uninit(&t_gh);
558                 t = DIV_ROUND_UP(jiffies - t, HZ);
559                 fs_info(sdp, "jid=%u: Journal replayed in %lus\n",
560                         jd->jd_jid, t);
561         }
562
563         gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
564
565         if (jlocked) {
566                 gfs2_glock_dq_uninit(&ji_gh);
567                 gfs2_glock_dq_uninit(&j_gh);
568         }
569
570         fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
571         goto done;
572
573 fail_gunlock_tr:
574         gfs2_glock_dq_uninit(&t_gh);
575 fail_gunlock_ji:
576         if (jlocked) {
577                 gfs2_glock_dq_uninit(&ji_gh);
578 fail_gunlock_j:
579                 gfs2_glock_dq_uninit(&j_gh);
580         }
581
582         fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
583 fail:
584         gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
585 done:
586         clear_bit(JDF_RECOVERY, &jd->jd_flags);
587         smp_mb__after_clear_bit();
588         wake_up_bit(&jd->jd_flags, JDF_RECOVERY);
589 }
590
591 static int gfs2_recovery_wait(void *word)
592 {
593         schedule();
594         return 0;
595 }
596
597 int gfs2_recover_journal(struct gfs2_jdesc *jd, bool wait)
598 {
599         int rv;
600
601         if (test_and_set_bit(JDF_RECOVERY, &jd->jd_flags))
602                 return -EBUSY;
603
604         /* we have JDF_RECOVERY, queue should always succeed */
605         rv = queue_work(gfs_recovery_wq, &jd->jd_work);
606         BUG_ON(!rv);
607
608         if (wait)
609                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY, gfs2_recovery_wait,
610                             TASK_UNINTERRUPTIBLE);
611
612         return 0;
613 }
614