2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
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.
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/crc32c.h>
18 #include <linux/ktime.h>
33 struct workqueue_struct *gfs_recovery_wq;
35 int gfs2_replay_read_block(struct gfs2_jdesc *jd, unsigned int blk,
36 struct buffer_head **bh)
38 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
39 struct gfs2_glock *gl = ip->i_gl;
45 error = gfs2_extent_map(&ip->i_inode, blk, &new, &dblock, &extlen);
49 gfs2_consist_inode(ip);
53 *bh = gfs2_meta_ra(gl, dblock, extlen);
58 int gfs2_revoke_add(struct gfs2_jdesc *jd, u64 blkno, unsigned int where)
60 struct list_head *head = &jd->jd_revoke_list;
61 struct gfs2_revoke_replay *rr;
64 list_for_each_entry(rr, head, rr_list) {
65 if (rr->rr_blkno == blkno) {
76 rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_NOFS);
82 list_add(&rr->rr_list, head);
87 int gfs2_revoke_check(struct gfs2_jdesc *jd, u64 blkno, unsigned int where)
89 struct gfs2_revoke_replay *rr;
90 int wrap, a, b, revoke;
93 list_for_each_entry(rr, &jd->jd_revoke_list, rr_list) {
94 if (rr->rr_blkno == blkno) {
103 wrap = (rr->rr_where < jd->jd_replay_tail);
104 a = (jd->jd_replay_tail < where);
105 b = (where < rr->rr_where);
106 revoke = (wrap) ? (a || b) : (a && b);
111 void gfs2_revoke_clean(struct gfs2_jdesc *jd)
113 struct list_head *head = &jd->jd_revoke_list;
114 struct gfs2_revoke_replay *rr;
116 while (!list_empty(head)) {
117 rr = list_entry(head->next, struct gfs2_revoke_replay, rr_list);
118 list_del(&rr->rr_list);
123 int __get_log_header(struct gfs2_sbd *sdp, const struct gfs2_log_header *lh,
124 unsigned int blkno, struct gfs2_log_header_host *head)
128 if (lh->lh_header.mh_magic != cpu_to_be32(GFS2_MAGIC) ||
129 lh->lh_header.mh_type != cpu_to_be32(GFS2_METATYPE_LH) ||
130 (blkno && be32_to_cpu(lh->lh_blkno) != blkno))
133 hash = crc32(~0, lh, LH_V1_SIZE - 4);
134 hash = ~crc32_le_shift(hash, 4); /* assume lh_hash is zero */
136 if (be32_to_cpu(lh->lh_hash) != hash)
139 crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
140 sdp->sd_sb.sb_bsize - LH_V1_SIZE - 4);
142 if ((lh->lh_crc != 0 && be32_to_cpu(lh->lh_crc) != crc))
145 head->lh_sequence = be64_to_cpu(lh->lh_sequence);
146 head->lh_flags = be32_to_cpu(lh->lh_flags);
147 head->lh_tail = be32_to_cpu(lh->lh_tail);
148 head->lh_blkno = be32_to_cpu(lh->lh_blkno);
153 * get_log_header - read the log header for a given segment
155 * @blk: the block to look at
156 * @lh: the log header to return
158 * Read the log header for a given segement in a given journal. Do a few
159 * sanity checks on it.
161 * Returns: 0 on success,
162 * 1 if the header was invalid or incomplete,
166 static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
167 struct gfs2_log_header_host *head)
169 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
170 struct buffer_head *bh;
173 error = gfs2_replay_read_block(jd, blk, &bh);
177 error = __get_log_header(sdp, (const struct gfs2_log_header *)bh->b_data,
185 * find_good_lh - find a good log header
187 * @blk: the segment to start searching from
188 * @lh: the log header to fill in
189 * @forward: if true search forward in the log, else search backward
191 * Call get_log_header() to get a log header for a segment, but if the
192 * segment is bad, either scan forward or backward until we find a good one.
197 static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk,
198 struct gfs2_log_header_host *head)
200 unsigned int orig_blk = *blk;
204 error = get_log_header(jd, *blk, head);
208 if (++*blk == jd->jd_blocks)
211 if (*blk == orig_blk) {
212 gfs2_consist_inode(GFS2_I(jd->jd_inode));
219 * jhead_scan - make sure we've found the head of the log
221 * @head: this is filled in with the log descriptor of the head
223 * At this point, seg and lh should be either the head of the log or just
224 * before. Scan forward until we find the head.
229 static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
231 unsigned int blk = head->lh_blkno;
232 struct gfs2_log_header_host lh;
236 if (++blk == jd->jd_blocks)
239 error = get_log_header(jd, blk, &lh);
245 if (lh.lh_sequence == head->lh_sequence) {
246 gfs2_consist_inode(GFS2_I(jd->jd_inode));
249 if (lh.lh_sequence < head->lh_sequence)
259 * gfs2_find_jhead - find the head of a log
261 * @head: the log descriptor for the head of the log is returned here
263 * Do a binary search of a journal and find the valid log entry with the
264 * highest sequence number. (i.e. the log head)
269 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
271 struct gfs2_log_header_host lh_1, lh_m;
272 u32 blk_1, blk_2, blk_m;
276 blk_2 = jd->jd_blocks - 1;
279 blk_m = (blk_1 + blk_2) / 2;
281 error = find_good_lh(jd, &blk_1, &lh_1);
285 error = find_good_lh(jd, &blk_m, &lh_m);
289 if (blk_1 == blk_m || blk_m == blk_2)
292 if (lh_1.lh_sequence <= lh_m.lh_sequence)
298 error = jhead_scan(jd, &lh_1);
308 * foreach_descriptor - go through the active part of the log
310 * @start: the first log header in the active region
311 * @end: the last log header (don't process the contents of this entry))
313 * Call a given function once for every log descriptor in the active
314 * portion of the log.
319 static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start,
320 unsigned int end, int pass)
322 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
323 struct buffer_head *bh;
324 struct gfs2_log_descriptor *ld;
328 unsigned int offset = sizeof(struct gfs2_log_descriptor);
329 offset += sizeof(__be64) - 1;
330 offset &= ~(sizeof(__be64) - 1);
332 while (start != end) {
333 error = gfs2_replay_read_block(jd, start, &bh);
336 if (gfs2_meta_check(sdp, bh)) {
340 ld = (struct gfs2_log_descriptor *)bh->b_data;
341 length = be32_to_cpu(ld->ld_length);
343 if (be32_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
344 struct gfs2_log_header_host lh;
345 error = get_log_header(jd, start, &lh);
347 gfs2_replay_incr_blk(jd, &start);
352 gfs2_consist_inode(GFS2_I(jd->jd_inode));
357 } else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
361 ptr = (__be64 *)(bh->b_data + offset);
362 error = lops_scan_elements(jd, start, ld, ptr, pass);
369 gfs2_replay_incr_blk(jd, &start);
378 * clean_journal - mark a dirty journal as being clean
380 * @head: the head journal to start from
385 static void clean_journal(struct gfs2_jdesc *jd,
386 struct gfs2_log_header_host *head)
388 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
390 sdp->sd_log_flush_head = head->lh_blkno;
391 gfs2_replay_incr_blk(jd, &sdp->sd_log_flush_head);
392 gfs2_write_log_header(sdp, jd, head->lh_sequence + 1, 0,
393 GFS2_LOG_HEAD_UNMOUNT | GFS2_LOG_HEAD_RECOVERY,
394 REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC);
398 static void gfs2_recovery_done(struct gfs2_sbd *sdp, unsigned int jid,
399 unsigned int message)
403 char *envp[] = { env_jid, env_status, NULL };
404 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
406 ls->ls_recover_jid_done = jid;
407 ls->ls_recover_jid_status = message;
408 sprintf(env_jid, "JID=%u", jid);
409 sprintf(env_status, "RECOVERY=%s",
410 message == LM_RD_SUCCESS ? "Done" : "Failed");
411 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
413 if (sdp->sd_lockstruct.ls_ops->lm_recovery_result)
414 sdp->sd_lockstruct.ls_ops->lm_recovery_result(sdp, jid, message);
417 void gfs2_recover_func(struct work_struct *work)
419 struct gfs2_jdesc *jd = container_of(work, struct gfs2_jdesc, jd_work);
420 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
421 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
422 struct gfs2_log_header_host head;
423 struct gfs2_holder j_gh, ji_gh, thaw_gh;
424 ktime_t t_start, t_jlck, t_jhd, t_tlck, t_rep;
430 t_start = ktime_get();
431 if (sdp->sd_args.ar_spectator)
433 if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
434 fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
437 /* Acquire the journal lock so we can do recovery */
439 error = gfs2_glock_nq_num(sdp, jd->jd_jid, &gfs2_journal_glops,
441 LM_FLAG_NOEXP | LM_FLAG_TRY | GL_NOCACHE,
448 fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
455 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
456 LM_FLAG_NOEXP | GL_NOCACHE, &ji_gh);
460 fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
463 t_jlck = ktime_get();
464 fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
466 error = gfs2_jdesc_check(jd);
468 goto fail_gunlock_ji;
470 error = gfs2_find_jhead(jd, &head);
472 goto fail_gunlock_ji;
474 fs_info(sdp, "jid=%u: Journal head lookup took %lldms\n", jd->jd_jid,
475 ktime_ms_delta(t_jhd, t_jlck));
477 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
478 fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
481 /* Acquire a shared hold on the freeze lock */
483 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
484 LM_FLAG_NOEXP | LM_FLAG_PRIORITY,
487 goto fail_gunlock_ji;
489 if (test_bit(SDF_RORECOVERY, &sdp->sd_flags)) {
491 } else if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
492 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
495 if (sb_rdonly(sdp->sd_vfs)) {
496 /* check if device itself is read-only */
497 ro = bdev_read_only(sdp->sd_vfs->s_bdev);
499 fs_info(sdp, "recovery required on "
500 "read-only filesystem.\n");
501 fs_info(sdp, "write access will be "
502 "enabled during recovery.\n");
508 fs_warn(sdp, "jid=%u: Can't replay: read-only block "
509 "device\n", jd->jd_jid);
511 goto fail_gunlock_thaw;
514 t_tlck = ktime_get();
515 fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
517 for (pass = 0; pass < 2; pass++) {
518 lops_before_scan(jd, &head, pass);
519 error = foreach_descriptor(jd, head.lh_tail,
520 head.lh_blkno, pass);
521 lops_after_scan(jd, error, pass);
523 goto fail_gunlock_thaw;
526 clean_journal(jd, &head);
528 gfs2_glock_dq_uninit(&thaw_gh);
530 fs_info(sdp, "jid=%u: Journal replayed in %lldms [jlck:%lldms, "
531 "jhead:%lldms, tlck:%lldms, replay:%lldms]\n",
532 jd->jd_jid, ktime_ms_delta(t_rep, t_start),
533 ktime_ms_delta(t_jlck, t_start),
534 ktime_ms_delta(t_jhd, t_jlck),
535 ktime_ms_delta(t_tlck, t_jhd),
536 ktime_ms_delta(t_rep, t_tlck));
539 gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
542 gfs2_glock_dq_uninit(&ji_gh);
543 gfs2_glock_dq_uninit(&j_gh);
546 fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
550 gfs2_glock_dq_uninit(&thaw_gh);
553 gfs2_glock_dq_uninit(&ji_gh);
555 gfs2_glock_dq_uninit(&j_gh);
558 fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
560 jd->jd_recover_error = error;
561 gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
563 clear_bit(JDF_RECOVERY, &jd->jd_flags);
564 smp_mb__after_atomic();
565 wake_up_bit(&jd->jd_flags, JDF_RECOVERY);
568 int gfs2_recover_journal(struct gfs2_jdesc *jd, bool wait)
572 if (test_and_set_bit(JDF_RECOVERY, &jd->jd_flags))
575 /* we have JDF_RECOVERY, queue should always succeed */
576 rv = queue_work(gfs_recovery_wq, &jd->jd_work);
580 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
581 TASK_UNINTERRUPTIBLE);
583 return wait ? jd->jd_recover_error : 0;