[GFS2] The core of GFS2
[linux-2.6.git] / fs / gfs2 / recovery.c
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 {
30         struct gfs2_glock *gl = jd->jd_inode->i_gl;
31         int new = 0;
32         uint64_t dblock;
33         uint32_t extlen;
34         int error;
35
36         error = gfs2_block_map(jd->jd_inode, blk, &new, &dblock, &extlen);
37         if (error)
38                 return error;
39         if (!dblock) {
40                 gfs2_consist_inode(jd->jd_inode);
41                 return -EIO;
42         }
43
44         gfs2_meta_ra(gl, dblock, extlen);
45         error = gfs2_meta_read(gl, dblock, DIO_START | DIO_WAIT, bh);
46
47         return error;
48 }
49
50 int gfs2_revoke_add(struct gfs2_sbd *sdp, uint64_t blkno, unsigned int where)
51 {
52         struct list_head *head = &sdp->sd_revoke_list;
53         struct gfs2_revoke_replay *rr;
54         int found = 0;
55
56         list_for_each_entry(rr, head, rr_list) {
57                 if (rr->rr_blkno == blkno) {
58                         found = 1;
59                         break;
60                 }
61         }
62
63         if (found) {
64                 rr->rr_where = where;
65                 return 0;
66         }
67
68         rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_KERNEL);
69         if (!rr)
70                 return -ENOMEM;
71
72         rr->rr_blkno = blkno;
73         rr->rr_where = where;
74         list_add(&rr->rr_list, head);
75
76         return 1;
77 }
78
79 int gfs2_revoke_check(struct gfs2_sbd *sdp, uint64_t blkno, unsigned int where)
80 {
81         struct gfs2_revoke_replay *rr;
82         int wrap, a, b, revoke;
83         int found = 0;
84
85         list_for_each_entry(rr, &sdp->sd_revoke_list, rr_list) {
86                 if (rr->rr_blkno == blkno) {
87                         found = 1;
88                         break;
89                 }
90         }
91
92         if (!found)
93                 return 0;
94
95         wrap = (rr->rr_where < sdp->sd_replay_tail);
96         a = (sdp->sd_replay_tail < where);
97         b = (where < rr->rr_where);
98         revoke = (wrap) ? (a || b) : (a && b);
99
100         return revoke;
101 }
102
103 void gfs2_revoke_clean(struct gfs2_sbd *sdp)
104 {
105         struct list_head *head = &sdp->sd_revoke_list;
106         struct gfs2_revoke_replay *rr;
107
108         while (!list_empty(head)) {
109                 rr = list_entry(head->next, struct gfs2_revoke_replay, rr_list);
110                 list_del(&rr->rr_list);
111                 kfree(rr);
112         }
113 }
114
115 /**
116  * get_log_header - read the log header for a given segment
117  * @jd: the journal
118  * @blk: the block to look at
119  * @lh: the log header to return
120  *
121  * Read the log header for a given segement in a given journal.  Do a few
122  * sanity checks on it.
123  *
124  * Returns: 0 on success,
125  *          1 if the header was invalid or incomplete,
126  *          errno on error
127  */
128
129 static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
130                           struct gfs2_log_header *head)
131 {
132         struct buffer_head *bh;
133         struct gfs2_log_header lh;
134         uint32_t hash;
135         int error;
136
137         error = gfs2_replay_read_block(jd, blk, &bh);
138         if (error)
139                 return error;
140
141         memcpy(&lh, bh->b_data, sizeof(struct gfs2_log_header));
142         lh.lh_hash = 0;
143         hash = gfs2_disk_hash((char *)&lh, sizeof(struct gfs2_log_header));
144         gfs2_log_header_in(&lh, bh->b_data);
145
146         brelse(bh);
147
148         if (lh.lh_header.mh_magic != GFS2_MAGIC ||
149             lh.lh_header.mh_type != GFS2_METATYPE_LH ||
150             lh.lh_blkno != blk ||
151             lh.lh_hash != hash)
152                 return 1;
153
154         *head = lh;
155
156         return 0;
157 }
158
159 /**
160  * find_good_lh - find a good log header
161  * @jd: the journal
162  * @blk: the segment to start searching from
163  * @lh: the log header to fill in
164  * @forward: if true search forward in the log, else search backward
165  *
166  * Call get_log_header() to get a log header for a segment, but if the
167  * segment is bad, either scan forward or backward until we find a good one.
168  *
169  * Returns: errno
170  */
171
172 static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk,
173                         struct gfs2_log_header *head)
174 {
175         unsigned int orig_blk = *blk;
176         int error;
177
178         for (;;) {
179                 error = get_log_header(jd, *blk, head);
180                 if (error <= 0)
181                         return error;
182
183                 if (++*blk == jd->jd_blocks)
184                         *blk = 0;
185
186                 if (*blk == orig_blk) {
187                         gfs2_consist_inode(jd->jd_inode);
188                         return -EIO;
189                 }
190         }
191 }
192
193 /**
194  * jhead_scan - make sure we've found the head of the log
195  * @jd: the journal
196  * @head: this is filled in with the log descriptor of the head
197  *
198  * At this point, seg and lh should be either the head of the log or just
199  * before.  Scan forward until we find the head.
200  *
201  * Returns: errno
202  */
203
204 static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
205 {
206         unsigned int blk = head->lh_blkno;
207         struct gfs2_log_header lh;
208         int error;
209
210         for (;;) {
211                 if (++blk == jd->jd_blocks)
212                         blk = 0;
213
214                 error = get_log_header(jd, blk, &lh);
215                 if (error < 0)
216                         return error;
217                 if (error == 1)
218                         continue;
219
220                 if (lh.lh_sequence == head->lh_sequence) {
221                         gfs2_consist_inode(jd->jd_inode);
222                         return -EIO;
223                 }
224                 if (lh.lh_sequence < head->lh_sequence)
225                         break;
226
227                 *head = lh;
228         }
229
230         return 0;
231 }
232
233 /**
234  * gfs2_find_jhead - find the head of a log
235  * @jd: the journal
236  * @head: the log descriptor for the head of the log is returned here
237  *
238  * Do a binary search of a journal and find the valid log entry with the
239  * highest sequence number.  (i.e. the log head)
240  *
241  * Returns: errno
242  */
243
244 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
245 {
246         struct gfs2_log_header lh_1, lh_m;
247         uint32_t blk_1, blk_2, blk_m;
248         int error;
249
250         blk_1 = 0;
251         blk_2 = jd->jd_blocks - 1;
252
253         for (;;) {
254                 blk_m = (blk_1 + blk_2) / 2;
255
256                 error = find_good_lh(jd, &blk_1, &lh_1);
257                 if (error)
258                         return error;
259
260                 error = find_good_lh(jd, &blk_m, &lh_m);
261                 if (error)
262                         return error;
263
264                 if (blk_1 == blk_m || blk_m == blk_2)
265                         break;
266
267                 if (lh_1.lh_sequence <= lh_m.lh_sequence)
268                         blk_1 = blk_m;
269                 else
270                         blk_2 = blk_m;
271         }
272
273         error = jhead_scan(jd, &lh_1);
274         if (error)
275                 return error;
276
277         *head = lh_1;
278
279         return error;
280 }
281
282 /**
283  * foreach_descriptor - go through the active part of the log
284  * @jd: the journal
285  * @start: the first log header in the active region
286  * @end: the last log header (don't process the contents of this entry))
287  *
288  * Call a given function once for every log descriptor in the active
289  * portion of the log.
290  *
291  * Returns: errno
292  */
293
294 static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start,
295                               unsigned int end, int pass)
296 {
297         struct gfs2_sbd *sdp = jd->jd_inode->i_sbd;
298         struct buffer_head *bh;
299         struct gfs2_log_descriptor *ld;
300         int error = 0;
301         u32 length;
302         __be64 *ptr;
303         unsigned int offset = sizeof(struct gfs2_log_descriptor);
304         offset += (sizeof(__be64)-1);
305         offset &= ~(sizeof(__be64)-1);
306
307         while (start != end) {
308                 error = gfs2_replay_read_block(jd, start, &bh);
309                 if (error)
310                         return error;
311                 if (gfs2_meta_check(sdp, bh)) {
312                         brelse(bh);
313                         return -EIO;
314                 }
315                 ld = (struct gfs2_log_descriptor *)bh->b_data;
316                 length = be32_to_cpu(ld->ld_length);
317
318                 if (be16_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
319                         struct gfs2_log_header lh;
320                         error = get_log_header(jd, start, &lh);
321                         if (!error) {
322                                 gfs2_replay_incr_blk(sdp, &start);
323                                 continue;
324                         }
325                         if (error == 1) {
326                                 gfs2_consist_inode(jd->jd_inode);
327                                 error = -EIO;
328                         }
329                         brelse(bh);
330                         return error;
331                 } else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
332                         brelse(bh);
333                         return -EIO;
334                 }
335                 ptr = (__be64 *)(bh->b_data + offset);
336                 error = lops_scan_elements(jd, start, ld, ptr, pass);
337                 if (error) {
338                         brelse(bh);
339                         return error;
340                 }
341
342                 while (length--)
343                         gfs2_replay_incr_blk(sdp, &start);
344
345                 brelse(bh);
346         }
347
348         return 0;
349 }
350
351 /**
352  * clean_journal - mark a dirty journal as being clean
353  * @sdp: the filesystem
354  * @jd: the journal
355  * @gl: the journal's glock
356  * @head: the head journal to start from
357  *
358  * Returns: errno
359  */
360
361 static int clean_journal(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
362 {
363         struct gfs2_inode *ip = jd->jd_inode;
364         struct gfs2_sbd *sdp = ip->i_sbd;
365         unsigned int lblock;
366         int new = 0;
367         uint64_t dblock;
368         struct gfs2_log_header *lh;
369         uint32_t hash;
370         struct buffer_head *bh;
371         int error;
372         
373         lblock = head->lh_blkno;
374         gfs2_replay_incr_blk(sdp, &lblock);
375         error = gfs2_block_map(ip, lblock, &new, &dblock, NULL);
376         if (error)
377                 return error;
378         if (!dblock) {
379                 gfs2_consist_inode(ip);
380                 return -EIO;
381         }
382
383         bh = sb_getblk(sdp->sd_vfs, dblock);
384         lock_buffer(bh);
385         memset(bh->b_data, 0, bh->b_size);
386         set_buffer_uptodate(bh);
387         clear_buffer_dirty(bh);
388         unlock_buffer(bh);
389
390         lh = (struct gfs2_log_header *)bh->b_data;
391         memset(lh, 0, sizeof(struct gfs2_log_header));
392         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
393         lh->lh_header.mh_type = cpu_to_be16(GFS2_METATYPE_LH);
394         lh->lh_header.mh_format = cpu_to_be16(GFS2_FORMAT_LH);
395         lh->lh_sequence = cpu_to_be64(head->lh_sequence + 1);
396         lh->lh_flags = cpu_to_be32(GFS2_LOG_HEAD_UNMOUNT);
397         lh->lh_blkno = cpu_to_be32(lblock);
398         hash = gfs2_disk_hash((const char *)lh, sizeof(struct gfs2_log_header));
399         lh->lh_hash = cpu_to_be32(hash);
400
401         set_buffer_dirty(bh);
402         if (sync_dirty_buffer(bh))
403                 gfs2_io_error_bh(sdp, bh);
404         brelse(bh);
405
406         return error;
407 }
408
409 /**
410  * gfs2_recover_journal - recovery a given journal
411  * @jd: the struct gfs2_jdesc describing the journal
412  * @wait: Don't return until the journal is clean (or an error is encountered)
413  *
414  * Acquire the journal's lock, check to see if the journal is clean, and
415  * do recovery if necessary.
416  *
417  * Returns: errno
418  */
419
420 int gfs2_recover_journal(struct gfs2_jdesc *jd, int wait)
421 {
422         struct gfs2_sbd *sdp = jd->jd_inode->i_sbd;
423         struct gfs2_log_header head;
424         struct gfs2_holder j_gh, ji_gh, t_gh;
425         unsigned long t;
426         int ro = 0;
427         unsigned int pass;
428         int error;
429
430         fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n", jd->jd_jid);
431
432         /* Aquire the journal lock so we can do recovery */
433
434         error = gfs2_glock_nq_num(sdp,
435                                   jd->jd_jid, &gfs2_journal_glops,
436                                   LM_ST_EXCLUSIVE,
437                                   LM_FLAG_NOEXP |
438                                   ((wait) ? 0 : LM_FLAG_TRY) |
439                                   GL_NOCACHE, &j_gh);
440         switch (error) {
441         case 0:
442                 break;
443
444         case GLR_TRYFAILED:
445                 fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
446                 error = 0;
447
448         default:
449                 goto fail;
450         };
451
452         error = gfs2_glock_nq_init(jd->jd_inode->i_gl, LM_ST_SHARED,
453                                    LM_FLAG_NOEXP, &ji_gh);
454         if (error)
455                 goto fail_gunlock_j;
456
457         fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
458
459         error = gfs2_jdesc_check(jd);
460         if (error)
461                 goto fail_gunlock_ji;
462
463         error = gfs2_find_jhead(jd, &head);
464         if (error)
465                 goto fail_gunlock_ji;
466
467         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
468                 fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
469                         jd->jd_jid);
470
471                 t = jiffies;
472
473                 /* Acquire a shared hold on the transaction lock */
474
475                 error = gfs2_glock_nq_init(sdp->sd_trans_gl,
476                                            LM_ST_SHARED,
477                                            LM_FLAG_NOEXP |
478                                            LM_FLAG_PRIORITY |
479                                            GL_NEVER_RECURSE |
480                                            GL_NOCANCEL |
481                                            GL_NOCACHE,
482                                            &t_gh);
483                 if (error)
484                         goto fail_gunlock_ji;
485
486                 if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
487                         if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
488                                 ro = 1;
489                 } else {
490                         if (sdp->sd_vfs->s_flags & MS_RDONLY)
491                                 ro = 1;
492                 }
493
494                 if (ro) {
495                         fs_warn(sdp, "jid=%u: Can't replay: read-only FS\n",
496                                 jd->jd_jid);
497                         error = -EROFS;
498                         goto fail_gunlock_tr;
499                 }
500
501                 fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
502
503                 for (pass = 0; pass < 2; pass++) {
504                         lops_before_scan(jd, &head, pass);
505                         error = foreach_descriptor(jd, head.lh_tail,
506                                                    head.lh_blkno, pass);
507                         lops_after_scan(jd, error, pass);
508                         if (error)
509                                 goto fail_gunlock_tr;
510                 }
511
512                 error = clean_journal(jd, &head);
513                 if (error)
514                         goto fail_gunlock_tr;
515
516                 gfs2_glock_dq_uninit(&t_gh);
517
518                 t = DIV_RU(jiffies - t, HZ);
519                 
520                 fs_info(sdp, "jid=%u: Journal replayed in %lus\n",
521                         jd->jd_jid, t);
522         }
523
524         gfs2_glock_dq_uninit(&ji_gh);
525
526         gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
527
528         gfs2_glock_dq_uninit(&j_gh);
529
530         fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
531
532         return 0;
533
534  fail_gunlock_tr:
535         gfs2_glock_dq_uninit(&t_gh);
536
537  fail_gunlock_ji:
538         gfs2_glock_dq_uninit(&ji_gh);
539
540  fail_gunlock_j:
541         gfs2_glock_dq_uninit(&j_gh);
542
543         fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
544
545  fail:
546         gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
547
548         return error;
549 }
550
551 /**
552  * gfs2_check_journals - Recover any dirty journals
553  * @sdp: the filesystem
554  *
555  */
556
557 void gfs2_check_journals(struct gfs2_sbd *sdp)
558 {
559         struct gfs2_jdesc *jd;
560
561         for (;;) {
562                 jd = gfs2_jdesc_find_dirty(sdp);
563                 if (!jd)
564                         break;
565
566                 if (jd != sdp->sd_jdesc)
567                         gfs2_recover_journal(jd, NO_WAIT);
568         }
569 }
570