Merge branch 'for-2.6.30' into for-2.6.31
[linux-2.6] / 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/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/gfs2_ondisk.h>
15 #include <linux/crc32.h>
16 #include <linux/kthread.h>
17 #include <linux/freezer.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
31 int 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
54 int 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
83 int 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
107 void 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
119 static 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  * get_log_header - read the log header for a given segment
137  * @jd: the journal
138  * @blk: the block to look at
139  * @lh: the log header to return
140  *
141  * Read the log header for a given segement in a given journal.  Do a few
142  * sanity checks on it.
143  *
144  * Returns: 0 on success,
145  *          1 if the header was invalid or incomplete,
146  *          errno on error
147  */
148
149 static 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 *)&nothing, 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  * find_good_lh - find a good log header
179  * @jd: the journal
180  * @blk: the segment to start searching from
181  * @lh: the log header to fill in
182  * @forward: if true search forward in the log, else search backward
183  *
184  * Call get_log_header() to get a log header for a segment, but if the
185  * segment is bad, either scan forward or backward until we find a good one.
186  *
187  * Returns: errno
188  */
189
190 static 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  * jhead_scan - make sure we've found the head of the log
213  * @jd: the journal
214  * @head: this is filled in with the log descriptor of the head
215  *
216  * At this point, seg and lh should be either the head of the log or just
217  * before.  Scan forward until we find the head.
218  *
219  * Returns: errno
220  */
221
222 static 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  * gfs2_find_jhead - find the head of a log
253  * @jd: the journal
254  * @head: the log descriptor for the head of the log is returned here
255  *
256  * Do a binary search of a journal and find the valid log entry with the
257  * highest sequence number.  (i.e. the log head)
258  *
259  * Returns: errno
260  */
261
262 int 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  * foreach_descriptor - go through the active part of the log
302  * @jd: the journal
303  * @start: the first log header in the active region
304  * @end: the last log header (don't process the contents of this entry))
305  *
306  * Call a given function once for every log descriptor in the active
307  * portion of the log.
308  *
309  * Returns: errno
310  */
311
312 static 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  * clean_journal - mark a dirty journal as being clean
372  * @sdp: the filesystem
373  * @jd: the journal
374  * @gl: the journal's glock
375  * @head: the head journal to start from
376  *
377  * Returns: errno
378  */
379
380 static 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
429 static 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
444 /**
445  * gfs2_recover_journal - recover a given journal
446  * @jd: the struct gfs2_jdesc describing the journal
447  *
448  * Acquire the journal's lock, check to see if the journal is clean, and
449  * do recovery if necessary.
450  *
451  * Returns: errno
452  */
453
454 int gfs2_recover_journal(struct gfs2_jdesc *jd)
455 {
456         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
457         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
458         struct gfs2_log_header_host head;
459         struct gfs2_holder j_gh, ji_gh, t_gh;
460         unsigned long t;
461         int ro = 0;
462         unsigned int pass;
463         int error;
464
465         if (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
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         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
564                 gfs2_glock_dq_uninit(&ji_gh);
565
566         gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
567
568         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
569                 gfs2_glock_dq_uninit(&j_gh);
570
571         fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
572         return 0;
573
574 fail_gunlock_tr:
575         gfs2_glock_dq_uninit(&t_gh);
576 fail_gunlock_ji:
577         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
578                 gfs2_glock_dq_uninit(&ji_gh);
579 fail_gunlock_j:
580                 gfs2_glock_dq_uninit(&j_gh);
581         }
582
583         fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
584
585 fail:
586         gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
587         return error;
588 }
589
590 static struct gfs2_jdesc *gfs2_jdesc_find_dirty(struct gfs2_sbd *sdp)
591 {
592         struct gfs2_jdesc *jd;
593         int found = 0;
594
595         spin_lock(&sdp->sd_jindex_spin);
596
597         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
598                 if (jd->jd_dirty) {
599                         jd->jd_dirty = 0;
600                         found = 1;
601                         break;
602                 }
603         }
604         spin_unlock(&sdp->sd_jindex_spin);
605
606         if (!found)
607                 jd = NULL;
608
609         return jd;
610 }
611
612 /**
613  * gfs2_check_journals - Recover any dirty journals
614  * @sdp: the filesystem
615  *
616  */
617
618 static void gfs2_check_journals(struct gfs2_sbd *sdp)
619 {
620         struct gfs2_jdesc *jd;
621
622         for (;;) {
623                 jd = gfs2_jdesc_find_dirty(sdp);
624                 if (!jd)
625                         break;
626
627                 if (jd != sdp->sd_jdesc)
628                         gfs2_recover_journal(jd);
629         }
630 }
631
632 /**
633  * gfs2_recoverd - Recover dead machine's journals
634  * @sdp: Pointer to GFS2 superblock
635  *
636  */
637
638 int gfs2_recoverd(void *data)
639 {
640         struct gfs2_sbd *sdp = data;
641         unsigned long t;
642
643         while (!kthread_should_stop()) {
644                 gfs2_check_journals(sdp);
645                 t = gfs2_tune_get(sdp,  gt_recoverd_secs) * HZ;
646                 if (freezing(current))
647                         refrigerator();
648                 schedule_timeout_interruptible(t);
649         }
650
651         return 0;
652 }
653