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/slow-work.h>
31 int gfs2_replay_read_block(struct gfs2_jdesc *jd, unsigned int blk,
32 struct buffer_head **bh)
34 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
35 struct gfs2_glock *gl = ip->i_gl;
41 error = gfs2_extent_map(&ip->i_inode, blk, &new, &dblock, &extlen);
45 gfs2_consist_inode(ip);
49 *bh = gfs2_meta_ra(gl, dblock, extlen);
54 int gfs2_revoke_add(struct gfs2_sbd *sdp, u64 blkno, unsigned int where)
56 struct list_head *head = &sdp->sd_revoke_list;
57 struct gfs2_revoke_replay *rr;
60 list_for_each_entry(rr, head, rr_list) {
61 if (rr->rr_blkno == blkno) {
72 rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_NOFS);
78 list_add(&rr->rr_list, head);
83 int gfs2_revoke_check(struct gfs2_sbd *sdp, u64 blkno, unsigned int where)
85 struct gfs2_revoke_replay *rr;
86 int wrap, a, b, revoke;
89 list_for_each_entry(rr, &sdp->sd_revoke_list, rr_list) {
90 if (rr->rr_blkno == blkno) {
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);
107 void gfs2_revoke_clean(struct gfs2_sbd *sdp)
109 struct list_head *head = &sdp->sd_revoke_list;
110 struct gfs2_revoke_replay *rr;
112 while (!list_empty(head)) {
113 rr = list_entry(head->next, struct gfs2_revoke_replay, rr_list);
114 list_del(&rr->rr_list);
119 static int gfs2_log_header_in(struct gfs2_log_header_host *lh, const void *buf)
121 const struct gfs2_log_header *str = buf;
123 if (str->lh_header.mh_magic != cpu_to_be32(GFS2_MAGIC) ||
124 str->lh_header.mh_type != cpu_to_be32(GFS2_METATYPE_LH))
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);
136 * get_log_header - read the log header for a given segment
138 * @blk: the block to look at
139 * @lh: the log header to return
141 * Read the log header for a given segement in a given journal. Do a few
142 * sanity checks on it.
144 * Returns: 0 on success,
145 * 1 if the header was invalid or incomplete,
149 static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
150 struct gfs2_log_header_host *head)
152 struct buffer_head *bh;
153 struct gfs2_log_header_host uninitialized_var(lh);
154 const u32 nothing = 0;
158 error = gfs2_replay_read_block(jd, blk, &bh);
162 hash = crc32_le((u32)~0, bh->b_data, sizeof(struct gfs2_log_header) -
164 hash = crc32_le(hash, (unsigned char const *)¬hing, sizeof(nothing));
166 error = gfs2_log_header_in(&lh, bh->b_data);
169 if (error || lh.lh_blkno != blk || lh.lh_hash != hash)
178 * find_good_lh - find a good log header
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
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.
190 static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk,
191 struct gfs2_log_header_host *head)
193 unsigned int orig_blk = *blk;
197 error = get_log_header(jd, *blk, head);
201 if (++*blk == jd->jd_blocks)
204 if (*blk == orig_blk) {
205 gfs2_consist_inode(GFS2_I(jd->jd_inode));
212 * jhead_scan - make sure we've found the head of the log
214 * @head: this is filled in with the log descriptor of the head
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.
222 static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
224 unsigned int blk = head->lh_blkno;
225 struct gfs2_log_header_host lh;
229 if (++blk == jd->jd_blocks)
232 error = get_log_header(jd, blk, &lh);
238 if (lh.lh_sequence == head->lh_sequence) {
239 gfs2_consist_inode(GFS2_I(jd->jd_inode));
242 if (lh.lh_sequence < head->lh_sequence)
252 * gfs2_find_jhead - find the head of a log
254 * @head: the log descriptor for the head of the log is returned here
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)
262 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
264 struct gfs2_log_header_host lh_1, lh_m;
265 u32 blk_1, blk_2, blk_m;
269 blk_2 = jd->jd_blocks - 1;
272 blk_m = (blk_1 + blk_2) / 2;
274 error = find_good_lh(jd, &blk_1, &lh_1);
278 error = find_good_lh(jd, &blk_m, &lh_m);
282 if (blk_1 == blk_m || blk_m == blk_2)
285 if (lh_1.lh_sequence <= lh_m.lh_sequence)
291 error = jhead_scan(jd, &lh_1);
301 * foreach_descriptor - go through the active part of the log
303 * @start: the first log header in the active region
304 * @end: the last log header (don't process the contents of this entry))
306 * Call a given function once for every log descriptor in the active
307 * portion of the log.
312 static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start,
313 unsigned int end, int pass)
315 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
316 struct buffer_head *bh;
317 struct gfs2_log_descriptor *ld;
321 unsigned int offset = sizeof(struct gfs2_log_descriptor);
322 offset += sizeof(__be64) - 1;
323 offset &= ~(sizeof(__be64) - 1);
325 while (start != end) {
326 error = gfs2_replay_read_block(jd, start, &bh);
329 if (gfs2_meta_check(sdp, bh)) {
333 ld = (struct gfs2_log_descriptor *)bh->b_data;
334 length = be32_to_cpu(ld->ld_length);
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);
340 gfs2_replay_incr_blk(sdp, &start);
345 gfs2_consist_inode(GFS2_I(jd->jd_inode));
350 } else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
354 ptr = (__be64 *)(bh->b_data + offset);
355 error = lops_scan_elements(jd, start, ld, ptr, pass);
362 gfs2_replay_incr_blk(sdp, &start);
371 * clean_journal - mark a dirty journal as being clean
372 * @sdp: the filesystem
374 * @gl: the journal's glock
375 * @head: the head journal to start from
380 static int clean_journal(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
382 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
383 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
385 struct gfs2_log_header *lh;
387 struct buffer_head *bh;
389 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
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);
397 if (!bh_map.b_blocknr) {
398 gfs2_consist_inode(ip);
402 bh = sb_getblk(sdp->sd_vfs, bh_map.b_blocknr);
404 memset(bh->b_data, 0, bh->b_size);
405 set_buffer_uptodate(bh);
406 clear_buffer_dirty(bh);
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.__pad0 = cpu_to_be64(0);
414 lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
415 lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
416 lh->lh_sequence = cpu_to_be64(head->lh_sequence + 1);
417 lh->lh_flags = cpu_to_be32(GFS2_LOG_HEAD_UNMOUNT);
418 lh->lh_blkno = cpu_to_be32(lblock);
419 hash = gfs2_disk_hash((const char *)lh, sizeof(struct gfs2_log_header));
420 lh->lh_hash = cpu_to_be32(hash);
422 set_buffer_dirty(bh);
423 if (sync_dirty_buffer(bh))
424 gfs2_io_error_bh(sdp, bh);
431 static void gfs2_recovery_done(struct gfs2_sbd *sdp, unsigned int jid,
432 unsigned int message)
436 char *envp[] = { env_jid, env_status, NULL };
437 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
438 ls->ls_recover_jid_done = jid;
439 ls->ls_recover_jid_status = message;
440 sprintf(env_jid, "JID=%d", jid);
441 sprintf(env_status, "RECOVERY=%s",
442 message == LM_RD_SUCCESS ? "Done" : "Failed");
443 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
446 static int gfs2_recover_get_ref(struct slow_work *work)
448 struct gfs2_jdesc *jd = container_of(work, struct gfs2_jdesc, jd_work);
449 if (test_and_set_bit(JDF_RECOVERY, &jd->jd_flags))
454 static void gfs2_recover_put_ref(struct slow_work *work)
456 struct gfs2_jdesc *jd = container_of(work, struct gfs2_jdesc, jd_work);
457 clear_bit(JDF_RECOVERY, &jd->jd_flags);
458 smp_mb__after_clear_bit();
459 wake_up_bit(&jd->jd_flags, JDF_RECOVERY);
462 static void gfs2_recover_work(struct slow_work *work)
464 struct gfs2_jdesc *jd = container_of(work, struct gfs2_jdesc, jd_work);
465 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
466 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
467 struct gfs2_log_header_host head;
468 struct gfs2_holder j_gh, ji_gh, t_gh;
474 if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
475 fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
478 /* Acquire the journal lock so we can do recovery */
480 error = gfs2_glock_nq_num(sdp, jd->jd_jid, &gfs2_journal_glops,
482 LM_FLAG_NOEXP | LM_FLAG_TRY | GL_NOCACHE,
489 fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
496 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
497 LM_FLAG_NOEXP | GL_NOCACHE, &ji_gh);
501 fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
504 fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
506 error = gfs2_jdesc_check(jd);
508 goto fail_gunlock_ji;
510 error = gfs2_find_jhead(jd, &head);
512 goto fail_gunlock_ji;
514 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
515 fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
520 /* Acquire a shared hold on the transaction lock */
522 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED,
523 LM_FLAG_NOEXP | LM_FLAG_PRIORITY |
526 goto fail_gunlock_ji;
528 if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
529 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
532 if (sdp->sd_vfs->s_flags & MS_RDONLY) {
533 /* check if device itself is read-only */
534 ro = bdev_read_only(sdp->sd_vfs->s_bdev);
536 fs_info(sdp, "recovery required on "
537 "read-only filesystem.\n");
538 fs_info(sdp, "write access will be "
539 "enabled during recovery.\n");
545 fs_warn(sdp, "jid=%u: Can't replay: read-only block "
546 "device\n", jd->jd_jid);
548 goto fail_gunlock_tr;
551 fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
553 for (pass = 0; pass < 2; pass++) {
554 lops_before_scan(jd, &head, pass);
555 error = foreach_descriptor(jd, head.lh_tail,
556 head.lh_blkno, pass);
557 lops_after_scan(jd, error, pass);
559 goto fail_gunlock_tr;
562 error = clean_journal(jd, &head);
564 goto fail_gunlock_tr;
566 gfs2_glock_dq_uninit(&t_gh);
567 t = DIV_ROUND_UP(jiffies - t, HZ);
568 fs_info(sdp, "jid=%u: Journal replayed in %lus\n",
572 if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
573 gfs2_glock_dq_uninit(&ji_gh);
575 gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
577 if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
578 gfs2_glock_dq_uninit(&j_gh);
580 fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
584 gfs2_glock_dq_uninit(&t_gh);
586 if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
587 gfs2_glock_dq_uninit(&ji_gh);
589 gfs2_glock_dq_uninit(&j_gh);
592 fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
595 gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
598 struct slow_work_ops gfs2_recover_ops = {
599 .owner = THIS_MODULE,
600 .get_ref = gfs2_recover_get_ref,
601 .put_ref = gfs2_recover_put_ref,
602 .execute = gfs2_recover_work,
606 static int gfs2_recovery_wait(void *word)
612 int gfs2_recover_journal(struct gfs2_jdesc *jd)
615 rv = slow_work_enqueue(&jd->jd_work);
618 wait_on_bit(&jd->jd_flags, JDF_RECOVERY, gfs2_recovery_wait, TASK_UNINTERRUPTIBLE);