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>
32 struct workqueue_struct *gfs_recovery_wq;
34 int gfs2_replay_read_block(struct gfs2_jdesc *jd, unsigned int blk,
35 struct buffer_head **bh)
37 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
38 struct gfs2_glock *gl = ip->i_gl;
44 error = gfs2_extent_map(&ip->i_inode, blk, &new, &dblock, &extlen);
48 gfs2_consist_inode(ip);
52 *bh = gfs2_meta_ra(gl, dblock, extlen);
57 int gfs2_revoke_add(struct gfs2_jdesc *jd, u64 blkno, unsigned int where)
59 struct list_head *head = &jd->jd_revoke_list;
60 struct gfs2_revoke_replay *rr;
63 list_for_each_entry(rr, head, rr_list) {
64 if (rr->rr_blkno == blkno) {
75 rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_NOFS);
81 list_add(&rr->rr_list, head);
86 int gfs2_revoke_check(struct gfs2_jdesc *jd, u64 blkno, unsigned int where)
88 struct gfs2_revoke_replay *rr;
89 int wrap, a, b, revoke;
92 list_for_each_entry(rr, &jd->jd_revoke_list, rr_list) {
93 if (rr->rr_blkno == blkno) {
102 wrap = (rr->rr_where < jd->jd_replay_tail);
103 a = (jd->jd_replay_tail < where);
104 b = (where < rr->rr_where);
105 revoke = (wrap) ? (a || b) : (a && b);
110 void gfs2_revoke_clean(struct gfs2_jdesc *jd)
112 struct list_head *head = &jd->jd_revoke_list;
113 struct gfs2_revoke_replay *rr;
115 while (!list_empty(head)) {
116 rr = list_entry(head->next, struct gfs2_revoke_replay, rr_list);
117 list_del(&rr->rr_list);
123 * get_log_header - read the log header for a given segment
125 * @blk: the block to look at
126 * @lh: the log header to return
128 * Read the log header for a given segement in a given journal. Do a few
129 * sanity checks on it.
131 * Returns: 0 on success,
132 * 1 if the header was invalid or incomplete,
136 static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
137 struct gfs2_log_header_host *head)
139 struct gfs2_log_header *lh;
140 struct buffer_head *bh;
144 error = gfs2_replay_read_block(jd, blk, &bh);
147 lh = (void *)bh->b_data;
149 hash = crc32(~0, lh, LH_V1_SIZE - 4);
150 hash = ~crc32_le_shift(hash, 4); /* assume lh_hash is zero */
152 crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
153 bh->b_size - LH_V1_SIZE - 4);
155 error = lh->lh_header.mh_magic != cpu_to_be32(GFS2_MAGIC) ||
156 lh->lh_header.mh_type != cpu_to_be32(GFS2_METATYPE_LH) ||
157 be32_to_cpu(lh->lh_blkno) != blk ||
158 be32_to_cpu(lh->lh_hash) != hash ||
159 (lh->lh_crc != 0 && be32_to_cpu(lh->lh_crc) != crc);
164 head->lh_sequence = be64_to_cpu(lh->lh_sequence);
165 head->lh_flags = be32_to_cpu(lh->lh_flags);
166 head->lh_tail = be32_to_cpu(lh->lh_tail);
167 head->lh_blkno = be32_to_cpu(lh->lh_blkno);
173 * find_good_lh - find a good log header
175 * @blk: the segment to start searching from
176 * @lh: the log header to fill in
177 * @forward: if true search forward in the log, else search backward
179 * Call get_log_header() to get a log header for a segment, but if the
180 * segment is bad, either scan forward or backward until we find a good one.
185 static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk,
186 struct gfs2_log_header_host *head)
188 unsigned int orig_blk = *blk;
192 error = get_log_header(jd, *blk, head);
196 if (++*blk == jd->jd_blocks)
199 if (*blk == orig_blk) {
200 gfs2_consist_inode(GFS2_I(jd->jd_inode));
207 * jhead_scan - make sure we've found the head of the log
209 * @head: this is filled in with the log descriptor of the head
211 * At this point, seg and lh should be either the head of the log or just
212 * before. Scan forward until we find the head.
217 static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
219 unsigned int blk = head->lh_blkno;
220 struct gfs2_log_header_host lh;
224 if (++blk == jd->jd_blocks)
227 error = get_log_header(jd, blk, &lh);
233 if (lh.lh_sequence == head->lh_sequence) {
234 gfs2_consist_inode(GFS2_I(jd->jd_inode));
237 if (lh.lh_sequence < head->lh_sequence)
247 * gfs2_find_jhead - find the head of a log
249 * @head: the log descriptor for the head of the log is returned here
251 * Do a binary search of a journal and find the valid log entry with the
252 * highest sequence number. (i.e. the log head)
257 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
259 struct gfs2_log_header_host lh_1, lh_m;
260 u32 blk_1, blk_2, blk_m;
264 blk_2 = jd->jd_blocks - 1;
267 blk_m = (blk_1 + blk_2) / 2;
269 error = find_good_lh(jd, &blk_1, &lh_1);
273 error = find_good_lh(jd, &blk_m, &lh_m);
277 if (blk_1 == blk_m || blk_m == blk_2)
280 if (lh_1.lh_sequence <= lh_m.lh_sequence)
286 error = jhead_scan(jd, &lh_1);
296 * foreach_descriptor - go through the active part of the log
298 * @start: the first log header in the active region
299 * @end: the last log header (don't process the contents of this entry))
301 * Call a given function once for every log descriptor in the active
302 * portion of the log.
307 static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start,
308 unsigned int end, int pass)
310 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
311 struct buffer_head *bh;
312 struct gfs2_log_descriptor *ld;
316 unsigned int offset = sizeof(struct gfs2_log_descriptor);
317 offset += sizeof(__be64) - 1;
318 offset &= ~(sizeof(__be64) - 1);
320 while (start != end) {
321 error = gfs2_replay_read_block(jd, start, &bh);
324 if (gfs2_meta_check(sdp, bh)) {
328 ld = (struct gfs2_log_descriptor *)bh->b_data;
329 length = be32_to_cpu(ld->ld_length);
331 if (be32_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
332 struct gfs2_log_header_host lh;
333 error = get_log_header(jd, start, &lh);
335 gfs2_replay_incr_blk(jd, &start);
340 gfs2_consist_inode(GFS2_I(jd->jd_inode));
345 } else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
349 ptr = (__be64 *)(bh->b_data + offset);
350 error = lops_scan_elements(jd, start, ld, ptr, pass);
357 gfs2_replay_incr_blk(jd, &start);
366 * clean_journal - mark a dirty journal as being clean
368 * @head: the head journal to start from
373 static void clean_journal(struct gfs2_jdesc *jd,
374 struct gfs2_log_header_host *head)
376 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
378 sdp->sd_log_flush_head = head->lh_blkno;
379 gfs2_replay_incr_blk(jd, &sdp->sd_log_flush_head);
380 gfs2_write_log_header(sdp, jd, head->lh_sequence + 1, 0,
381 GFS2_LOG_HEAD_UNMOUNT | GFS2_LOG_HEAD_RECOVERY,
382 REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC);
386 static void gfs2_recovery_done(struct gfs2_sbd *sdp, unsigned int jid,
387 unsigned int message)
391 char *envp[] = { env_jid, env_status, NULL };
392 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
394 ls->ls_recover_jid_done = jid;
395 ls->ls_recover_jid_status = message;
396 sprintf(env_jid, "JID=%u", jid);
397 sprintf(env_status, "RECOVERY=%s",
398 message == LM_RD_SUCCESS ? "Done" : "Failed");
399 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
401 if (sdp->sd_lockstruct.ls_ops->lm_recovery_result)
402 sdp->sd_lockstruct.ls_ops->lm_recovery_result(sdp, jid, message);
405 void gfs2_recover_func(struct work_struct *work)
407 struct gfs2_jdesc *jd = container_of(work, struct gfs2_jdesc, jd_work);
408 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
409 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
410 struct gfs2_log_header_host head;
411 struct gfs2_holder j_gh, ji_gh, thaw_gh;
418 if (sdp->sd_args.ar_spectator ||
419 (jd->jd_jid != sdp->sd_lockstruct.ls_jid)) {
420 fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
423 /* Acquire the journal lock so we can do recovery */
425 error = gfs2_glock_nq_num(sdp, jd->jd_jid, &gfs2_journal_glops,
427 LM_FLAG_NOEXP | LM_FLAG_TRY | GL_NOCACHE,
434 fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
441 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
442 LM_FLAG_NOEXP | GL_NOCACHE, &ji_gh);
446 fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
449 fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
451 error = gfs2_jdesc_check(jd);
453 goto fail_gunlock_ji;
455 error = gfs2_find_jhead(jd, &head);
457 goto fail_gunlock_ji;
459 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
460 fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
465 /* Acquire a shared hold on the freeze lock */
467 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
468 LM_FLAG_NOEXP | LM_FLAG_PRIORITY,
471 goto fail_gunlock_ji;
473 if (test_bit(SDF_RORECOVERY, &sdp->sd_flags)) {
475 } else if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
476 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
479 if (sb_rdonly(sdp->sd_vfs)) {
480 /* check if device itself is read-only */
481 ro = bdev_read_only(sdp->sd_vfs->s_bdev);
483 fs_info(sdp, "recovery required on "
484 "read-only filesystem.\n");
485 fs_info(sdp, "write access will be "
486 "enabled during recovery.\n");
492 fs_warn(sdp, "jid=%u: Can't replay: read-only block "
493 "device\n", jd->jd_jid);
495 goto fail_gunlock_thaw;
498 fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
500 for (pass = 0; pass < 2; pass++) {
501 lops_before_scan(jd, &head, pass);
502 error = foreach_descriptor(jd, head.lh_tail,
503 head.lh_blkno, pass);
504 lops_after_scan(jd, error, pass);
506 goto fail_gunlock_thaw;
509 clean_journal(jd, &head);
511 gfs2_glock_dq_uninit(&thaw_gh);
512 t = DIV_ROUND_UP(jiffies - t, HZ);
513 fs_info(sdp, "jid=%u: Journal replayed in %lus\n",
517 gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
520 gfs2_glock_dq_uninit(&ji_gh);
521 gfs2_glock_dq_uninit(&j_gh);
524 fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
528 gfs2_glock_dq_uninit(&thaw_gh);
531 gfs2_glock_dq_uninit(&ji_gh);
533 gfs2_glock_dq_uninit(&j_gh);
536 fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
538 jd->jd_recover_error = error;
539 gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
541 clear_bit(JDF_RECOVERY, &jd->jd_flags);
542 smp_mb__after_atomic();
543 wake_up_bit(&jd->jd_flags, JDF_RECOVERY);
546 int gfs2_recover_journal(struct gfs2_jdesc *jd, bool wait)
550 if (test_and_set_bit(JDF_RECOVERY, &jd->jd_flags))
553 /* we have JDF_RECOVERY, queue should always succeed */
554 rv = queue_work(gfs_recovery_wq, &jd->jd_work);
558 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
559 TASK_UNINTERRUPTIBLE);
561 return wait ? jd->jd_recover_error : 0;