1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 */
6
7 #include <linux/module.h>
8 #include <linux/slab.h>
9 #include <linux/spinlock.h>
10 #include <linux/completion.h>
11 #include <linux/buffer_head.h>
12 #include <linux/gfs2_ondisk.h>
13 #include <linux/crc32.h>
14 #include <linux/crc32c.h>
15 #include <linux/ktime.h>
16
17 #include "gfs2.h"
18 #include "incore.h"
19 #include "bmap.h"
20 #include "glock.h"
21 #include "glops.h"
22 #include "log.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
gfs2_replay_read_block(struct gfs2_jdesc * jd,unsigned int blk,struct buffer_head ** bh)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
gfs2_revoke_add(struct gfs2_jdesc * jd,u64 blkno,unsigned int where)55 int gfs2_revoke_add(struct gfs2_jdesc *jd, u64 blkno, unsigned int where)
56 {
57 struct list_head *head = &jd->jd_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
gfs2_revoke_check(struct gfs2_jdesc * jd,u64 blkno,unsigned int where)84 int gfs2_revoke_check(struct gfs2_jdesc *jd, 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, &jd->jd_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 < jd->jd_replay_tail);
101 a = (jd->jd_replay_tail < where);
102 b = (where < rr->rr_where);
103 revoke = (wrap) ? (a || b) : (a && b);
104
105 return revoke;
106 }
107
gfs2_revoke_clean(struct gfs2_jdesc * jd)108 void gfs2_revoke_clean(struct gfs2_jdesc *jd)
109 {
110 struct list_head *head = &jd->jd_revoke_list;
111 struct gfs2_revoke_replay *rr;
112
113 while (!list_empty(head)) {
114 rr = list_first_entry(head, struct gfs2_revoke_replay, rr_list);
115 list_del(&rr->rr_list);
116 kfree(rr);
117 }
118 }
119
__get_log_header(struct gfs2_sbd * sdp,const struct gfs2_log_header * lh,unsigned int blkno,struct gfs2_log_header_host * head)120 int __get_log_header(struct gfs2_sbd *sdp, const struct gfs2_log_header *lh,
121 unsigned int blkno, struct gfs2_log_header_host *head)
122 {
123 u32 hash, crc;
124
125 if (lh->lh_header.mh_magic != cpu_to_be32(GFS2_MAGIC) ||
126 lh->lh_header.mh_type != cpu_to_be32(GFS2_METATYPE_LH) ||
127 (blkno && be32_to_cpu(lh->lh_blkno) != blkno))
128 return 1;
129
130 hash = crc32(~0, lh, LH_V1_SIZE - 4);
131 hash = ~crc32_le_shift(hash, 4); /* assume lh_hash is zero */
132
133 if (be32_to_cpu(lh->lh_hash) != hash)
134 return 1;
135
136 crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
137 sdp->sd_sb.sb_bsize - LH_V1_SIZE - 4);
138
139 if ((lh->lh_crc != 0 && be32_to_cpu(lh->lh_crc) != crc))
140 return 1;
141
142 head->lh_sequence = be64_to_cpu(lh->lh_sequence);
143 head->lh_flags = be32_to_cpu(lh->lh_flags);
144 head->lh_tail = be32_to_cpu(lh->lh_tail);
145 head->lh_blkno = be32_to_cpu(lh->lh_blkno);
146
147 head->lh_local_total = be64_to_cpu(lh->lh_local_total);
148 head->lh_local_free = be64_to_cpu(lh->lh_local_free);
149 head->lh_local_dinodes = be64_to_cpu(lh->lh_local_dinodes);
150
151 return 0;
152 }
153 /**
154 * get_log_header - read the log header for a given segment
155 * @jd: the journal
156 * @blk: the block to look at
157 * @lh: the log header to return
158 *
159 * Read the log header for a given segement in a given journal. Do a few
160 * sanity checks on it.
161 *
162 * Returns: 0 on success,
163 * 1 if the header was invalid or incomplete,
164 * errno on error
165 */
166
get_log_header(struct gfs2_jdesc * jd,unsigned int blk,struct gfs2_log_header_host * head)167 static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
168 struct gfs2_log_header_host *head)
169 {
170 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
171 struct buffer_head *bh;
172 int error;
173
174 error = gfs2_replay_read_block(jd, blk, &bh);
175 if (error)
176 return error;
177
178 error = __get_log_header(sdp, (const struct gfs2_log_header *)bh->b_data,
179 blk, head);
180 brelse(bh);
181
182 return error;
183 }
184
185 /**
186 * foreach_descriptor - go through the active part of the log
187 * @jd: the journal
188 * @start: the first log header in the active region
189 * @end: the last log header (don't process the contents of this entry))
190 *
191 * Call a given function once for every log descriptor in the active
192 * portion of the log.
193 *
194 * Returns: errno
195 */
196
foreach_descriptor(struct gfs2_jdesc * jd,u32 start,unsigned int end,int pass)197 static int foreach_descriptor(struct gfs2_jdesc *jd, u32 start,
198 unsigned int end, int pass)
199 {
200 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
201 struct buffer_head *bh;
202 struct gfs2_log_descriptor *ld;
203 int error = 0;
204 u32 length;
205 __be64 *ptr;
206 unsigned int offset = sizeof(struct gfs2_log_descriptor);
207 offset += sizeof(__be64) - 1;
208 offset &= ~(sizeof(__be64) - 1);
209
210 while (start != end) {
211 error = gfs2_replay_read_block(jd, start, &bh);
212 if (error)
213 return error;
214 if (gfs2_meta_check(sdp, bh)) {
215 brelse(bh);
216 return -EIO;
217 }
218 ld = (struct gfs2_log_descriptor *)bh->b_data;
219 length = be32_to_cpu(ld->ld_length);
220
221 if (be32_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
222 struct gfs2_log_header_host lh;
223 error = get_log_header(jd, start, &lh);
224 if (!error) {
225 gfs2_replay_incr_blk(jd, &start);
226 brelse(bh);
227 continue;
228 }
229 if (error == 1) {
230 gfs2_consist_inode(GFS2_I(jd->jd_inode));
231 error = -EIO;
232 }
233 brelse(bh);
234 return error;
235 } else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
236 brelse(bh);
237 return -EIO;
238 }
239 ptr = (__be64 *)(bh->b_data + offset);
240 error = lops_scan_elements(jd, start, ld, ptr, pass);
241 if (error) {
242 brelse(bh);
243 return error;
244 }
245
246 while (length--)
247 gfs2_replay_incr_blk(jd, &start);
248
249 brelse(bh);
250 }
251
252 return 0;
253 }
254
255 /**
256 * clean_journal - mark a dirty journal as being clean
257 * @jd: the journal
258 * @head: the head journal to start from
259 *
260 * Returns: errno
261 */
262
clean_journal(struct gfs2_jdesc * jd,struct gfs2_log_header_host * head)263 static void clean_journal(struct gfs2_jdesc *jd,
264 struct gfs2_log_header_host *head)
265 {
266 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
267 u32 lblock = head->lh_blkno;
268
269 gfs2_replay_incr_blk(jd, &lblock);
270 gfs2_write_log_header(sdp, jd, head->lh_sequence + 1, 0, lblock,
271 GFS2_LOG_HEAD_UNMOUNT | GFS2_LOG_HEAD_RECOVERY,
272 REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC);
273 if (jd->jd_jid == sdp->sd_lockstruct.ls_jid) {
274 sdp->sd_log_flush_head = lblock;
275 gfs2_log_incr_head(sdp);
276 }
277 }
278
279
gfs2_recovery_done(struct gfs2_sbd * sdp,unsigned int jid,unsigned int message)280 static void gfs2_recovery_done(struct gfs2_sbd *sdp, unsigned int jid,
281 unsigned int message)
282 {
283 char env_jid[20];
284 char env_status[20];
285 char *envp[] = { env_jid, env_status, NULL };
286 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
287
288 ls->ls_recover_jid_done = jid;
289 ls->ls_recover_jid_status = message;
290 sprintf(env_jid, "JID=%u", jid);
291 sprintf(env_status, "RECOVERY=%s",
292 message == LM_RD_SUCCESS ? "Done" : "Failed");
293 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
294
295 if (sdp->sd_lockstruct.ls_ops->lm_recovery_result)
296 sdp->sd_lockstruct.ls_ops->lm_recovery_result(sdp, jid, message);
297 }
298
299 /**
300 * update_statfs_inode - Update the master statfs inode or zero out the local
301 * statfs inode for a given journal.
302 * @jd: The journal
303 * @head: If NULL, @inode is the local statfs inode and we need to zero it out.
304 * Otherwise, it @head contains the statfs change info that needs to be
305 * synced to the master statfs inode (pointed to by @inode).
306 * @inode: statfs inode to update.
307 */
update_statfs_inode(struct gfs2_jdesc * jd,struct gfs2_log_header_host * head,struct inode * inode)308 static int update_statfs_inode(struct gfs2_jdesc *jd,
309 struct gfs2_log_header_host *head,
310 struct inode *inode)
311 {
312 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
313 struct gfs2_inode *ip;
314 struct buffer_head *bh;
315 struct gfs2_statfs_change_host sc;
316 int error = 0;
317
318 BUG_ON(!inode);
319 ip = GFS2_I(inode);
320
321 error = gfs2_meta_inode_buffer(ip, &bh);
322 if (error)
323 goto out;
324
325 spin_lock(&sdp->sd_statfs_spin);
326
327 if (head) { /* Update the master statfs inode */
328 gfs2_statfs_change_in(&sc, bh->b_data + sizeof(struct gfs2_dinode));
329 sc.sc_total += head->lh_local_total;
330 sc.sc_free += head->lh_local_free;
331 sc.sc_dinodes += head->lh_local_dinodes;
332 gfs2_statfs_change_out(&sc, bh->b_data + sizeof(struct gfs2_dinode));
333
334 fs_info(sdp, "jid=%u: Updated master statfs Total:%lld, "
335 "Free:%lld, Dinodes:%lld after change "
336 "[%+lld,%+lld,%+lld]\n", jd->jd_jid, sc.sc_total,
337 sc.sc_free, sc.sc_dinodes, head->lh_local_total,
338 head->lh_local_free, head->lh_local_dinodes);
339 } else { /* Zero out the local statfs inode */
340 memset(bh->b_data + sizeof(struct gfs2_dinode), 0,
341 sizeof(struct gfs2_statfs_change));
342 /* If it's our own journal, reset any in-memory changes too */
343 if (jd->jd_jid == sdp->sd_lockstruct.ls_jid) {
344 memset(&sdp->sd_statfs_local, 0,
345 sizeof(struct gfs2_statfs_change_host));
346 }
347 }
348 spin_unlock(&sdp->sd_statfs_spin);
349
350 mark_buffer_dirty(bh);
351 brelse(bh);
352 gfs2_inode_metasync(ip->i_gl);
353
354 out:
355 return error;
356 }
357
358 /**
359 * recover_local_statfs - Update the master and local statfs changes for this
360 * journal.
361 *
362 * Previously, statfs updates would be read in from the local statfs inode and
363 * synced to the master statfs inode during recovery.
364 *
365 * We now use the statfs updates in the journal head to update the master statfs
366 * inode instead of reading in from the local statfs inode. To preserve backward
367 * compatibility with kernels that can't do this, we still need to keep the
368 * local statfs inode up to date by writing changes to it. At some point in the
369 * future, we can do away with the local statfs inodes altogether and keep the
370 * statfs changes solely in the journal.
371 *
372 * @jd: the journal
373 * @head: the journal head
374 *
375 * Returns: errno
376 */
recover_local_statfs(struct gfs2_jdesc * jd,struct gfs2_log_header_host * head)377 static void recover_local_statfs(struct gfs2_jdesc *jd,
378 struct gfs2_log_header_host *head)
379 {
380 int error;
381 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
382
383 if (!head->lh_local_total && !head->lh_local_free
384 && !head->lh_local_dinodes) /* No change */
385 goto zero_local;
386
387 /* First update the master statfs inode with the changes we
388 * found in the journal. */
389 error = update_statfs_inode(jd, head, sdp->sd_statfs_inode);
390 if (error)
391 goto out;
392
393 zero_local:
394 /* Zero out the local statfs inode so any changes in there
395 * are not re-recovered. */
396 error = update_statfs_inode(jd, NULL,
397 find_local_statfs_inode(sdp, jd->jd_jid));
398 out:
399 return;
400 }
401
gfs2_recover_func(struct work_struct * work)402 void gfs2_recover_func(struct work_struct *work)
403 {
404 struct gfs2_jdesc *jd = container_of(work, struct gfs2_jdesc, jd_work);
405 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
406 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
407 struct gfs2_log_header_host head;
408 struct gfs2_holder j_gh, ji_gh, thaw_gh;
409 ktime_t t_start, t_jlck, t_jhd, t_tlck, t_rep;
410 int ro = 0;
411 unsigned int pass;
412 int error = 0;
413 int jlocked = 0;
414
415 if (gfs2_withdrawn(sdp)) {
416 fs_err(sdp, "jid=%u: Recovery not attempted due to withdraw.\n",
417 jd->jd_jid);
418 goto fail;
419 }
420 t_start = ktime_get();
421 if (sdp->sd_args.ar_spectator)
422 goto fail;
423 if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
424 fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
425 jd->jd_jid);
426 jlocked = 1;
427 /* Acquire the journal lock so we can do recovery */
428
429 error = gfs2_glock_nq_num(sdp, jd->jd_jid, &gfs2_journal_glops,
430 LM_ST_EXCLUSIVE,
431 LM_FLAG_NOEXP | LM_FLAG_TRY | GL_NOCACHE,
432 &j_gh);
433 switch (error) {
434 case 0:
435 break;
436
437 case GLR_TRYFAILED:
438 fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
439 error = 0;
440
441 default:
442 goto fail;
443 }
444
445 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
446 LM_FLAG_NOEXP | GL_NOCACHE, &ji_gh);
447 if (error)
448 goto fail_gunlock_j;
449 } else {
450 fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
451 }
452
453 t_jlck = ktime_get();
454 fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
455
456 error = gfs2_jdesc_check(jd);
457 if (error)
458 goto fail_gunlock_ji;
459
460 error = gfs2_find_jhead(jd, &head, true);
461 if (error)
462 goto fail_gunlock_ji;
463 t_jhd = ktime_get();
464 fs_info(sdp, "jid=%u: Journal head lookup took %lldms\n", jd->jd_jid,
465 ktime_ms_delta(t_jhd, t_jlck));
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 /* Acquire a shared hold on the freeze lock */
472
473 error = gfs2_freeze_lock(sdp, &thaw_gh, LM_FLAG_PRIORITY);
474 if (error)
475 goto fail_gunlock_ji;
476
477 if (test_bit(SDF_RORECOVERY, &sdp->sd_flags)) {
478 ro = 1;
479 } else if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
480 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
481 ro = 1;
482 } else {
483 if (sb_rdonly(sdp->sd_vfs)) {
484 /* check if device itself is read-only */
485 ro = bdev_read_only(sdp->sd_vfs->s_bdev);
486 if (!ro) {
487 fs_info(sdp, "recovery required on "
488 "read-only filesystem.\n");
489 fs_info(sdp, "write access will be "
490 "enabled during recovery.\n");
491 }
492 }
493 }
494
495 if (ro) {
496 fs_warn(sdp, "jid=%u: Can't replay: read-only block "
497 "device\n", jd->jd_jid);
498 error = -EROFS;
499 goto fail_gunlock_thaw;
500 }
501
502 t_tlck = ktime_get();
503 fs_info(sdp, "jid=%u: Replaying journal...0x%x to 0x%x\n",
504 jd->jd_jid, head.lh_tail, head.lh_blkno);
505
506 /* We take the sd_log_flush_lock here primarily to prevent log
507 * flushes and simultaneous journal replays from stomping on
508 * each other wrt sd_log_bio. */
509 down_read(&sdp->sd_log_flush_lock);
510 for (pass = 0; pass < 2; pass++) {
511 lops_before_scan(jd, &head, pass);
512 error = foreach_descriptor(jd, head.lh_tail,
513 head.lh_blkno, pass);
514 lops_after_scan(jd, error, pass);
515 if (error) {
516 up_read(&sdp->sd_log_flush_lock);
517 goto fail_gunlock_thaw;
518 }
519 }
520
521 recover_local_statfs(jd, &head);
522 clean_journal(jd, &head);
523 up_read(&sdp->sd_log_flush_lock);
524
525 gfs2_freeze_unlock(&thaw_gh);
526 t_rep = ktime_get();
527 fs_info(sdp, "jid=%u: Journal replayed in %lldms [jlck:%lldms, "
528 "jhead:%lldms, tlck:%lldms, replay:%lldms]\n",
529 jd->jd_jid, ktime_ms_delta(t_rep, t_start),
530 ktime_ms_delta(t_jlck, t_start),
531 ktime_ms_delta(t_jhd, t_jlck),
532 ktime_ms_delta(t_tlck, t_jhd),
533 ktime_ms_delta(t_rep, t_tlck));
534 }
535
536 gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
537
538 if (jlocked) {
539 gfs2_glock_dq_uninit(&ji_gh);
540 gfs2_glock_dq_uninit(&j_gh);
541 }
542
543 fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
544 goto done;
545
546 fail_gunlock_thaw:
547 gfs2_freeze_unlock(&thaw_gh);
548 fail_gunlock_ji:
549 if (jlocked) {
550 gfs2_glock_dq_uninit(&ji_gh);
551 fail_gunlock_j:
552 gfs2_glock_dq_uninit(&j_gh);
553 }
554
555 fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
556 fail:
557 jd->jd_recover_error = error;
558 gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
559 done:
560 clear_bit(JDF_RECOVERY, &jd->jd_flags);
561 smp_mb__after_atomic();
562 wake_up_bit(&jd->jd_flags, JDF_RECOVERY);
563 }
564
gfs2_recover_journal(struct gfs2_jdesc * jd,bool wait)565 int gfs2_recover_journal(struct gfs2_jdesc *jd, bool wait)
566 {
567 int rv;
568
569 if (test_and_set_bit(JDF_RECOVERY, &jd->jd_flags))
570 return -EBUSY;
571
572 /* we have JDF_RECOVERY, queue should always succeed */
573 rv = queue_work(gfs_recovery_wq, &jd->jd_work);
574 BUG_ON(!rv);
575
576 if (wait)
577 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
578 TASK_UNINTERRUPTIBLE);
579
580 return wait ? jd->jd_recover_error : 0;
581 }
582
583