2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 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 v.2.
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>
27 int gfs2_replay_read_block(struct gfs2_jdesc *jd, unsigned int blk,
28 struct buffer_head **bh)
30 struct gfs2_glock *gl = get_v2ip(jd->jd_inode)->i_gl;
36 error = gfs2_block_map(get_v2ip(jd->jd_inode), blk, &new, &dblock,
41 gfs2_consist_inode(get_v2ip(jd->jd_inode));
45 gfs2_meta_ra(gl, dblock, extlen);
46 error = gfs2_meta_read(gl, dblock, DIO_START | DIO_WAIT, bh);
51 int gfs2_revoke_add(struct gfs2_sbd *sdp, uint64_t blkno, unsigned int where)
53 struct list_head *head = &sdp->sd_revoke_list;
54 struct gfs2_revoke_replay *rr;
57 list_for_each_entry(rr, head, rr_list) {
58 if (rr->rr_blkno == blkno) {
69 rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_KERNEL);
75 list_add(&rr->rr_list, head);
80 int gfs2_revoke_check(struct gfs2_sbd *sdp, uint64_t blkno, unsigned int where)
82 struct gfs2_revoke_replay *rr;
83 int wrap, a, b, revoke;
86 list_for_each_entry(rr, &sdp->sd_revoke_list, rr_list) {
87 if (rr->rr_blkno == blkno) {
96 wrap = (rr->rr_where < sdp->sd_replay_tail);
97 a = (sdp->sd_replay_tail < where);
98 b = (where < rr->rr_where);
99 revoke = (wrap) ? (a || b) : (a && b);
104 void gfs2_revoke_clean(struct gfs2_sbd *sdp)
106 struct list_head *head = &sdp->sd_revoke_list;
107 struct gfs2_revoke_replay *rr;
109 while (!list_empty(head)) {
110 rr = list_entry(head->next, struct gfs2_revoke_replay, rr_list);
111 list_del(&rr->rr_list);
117 * get_log_header - read the log header for a given segment
119 * @blk: the block to look at
120 * @lh: the log header to return
122 * Read the log header for a given segement in a given journal. Do a few
123 * sanity checks on it.
125 * Returns: 0 on success,
126 * 1 if the header was invalid or incomplete,
130 static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
131 struct gfs2_log_header *head)
133 struct buffer_head *bh;
134 struct gfs2_log_header lh;
138 error = gfs2_replay_read_block(jd, blk, &bh);
142 memcpy(&lh, bh->b_data, sizeof(struct gfs2_log_header));
144 hash = gfs2_disk_hash((char *)&lh, sizeof(struct gfs2_log_header));
145 gfs2_log_header_in(&lh, bh->b_data);
149 if (lh.lh_header.mh_magic != GFS2_MAGIC ||
150 lh.lh_header.mh_type != GFS2_METATYPE_LH ||
151 lh.lh_blkno != blk ||
161 * find_good_lh - find a good log header
163 * @blk: the segment to start searching from
164 * @lh: the log header to fill in
165 * @forward: if true search forward in the log, else search backward
167 * Call get_log_header() to get a log header for a segment, but if the
168 * segment is bad, either scan forward or backward until we find a good one.
173 static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk,
174 struct gfs2_log_header *head)
176 unsigned int orig_blk = *blk;
180 error = get_log_header(jd, *blk, head);
184 if (++*blk == jd->jd_blocks)
187 if (*blk == orig_blk) {
188 gfs2_consist_inode(get_v2ip(jd->jd_inode));
195 * jhead_scan - make sure we've found the head of the log
197 * @head: this is filled in with the log descriptor of the head
199 * At this point, seg and lh should be either the head of the log or just
200 * before. Scan forward until we find the head.
205 static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
207 unsigned int blk = head->lh_blkno;
208 struct gfs2_log_header lh;
212 if (++blk == jd->jd_blocks)
215 error = get_log_header(jd, blk, &lh);
221 if (lh.lh_sequence == head->lh_sequence) {
222 gfs2_consist_inode(get_v2ip(jd->jd_inode));
225 if (lh.lh_sequence < head->lh_sequence)
235 * gfs2_find_jhead - find the head of a log
237 * @head: the log descriptor for the head of the log is returned here
239 * Do a binary search of a journal and find the valid log entry with the
240 * highest sequence number. (i.e. the log head)
245 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
247 struct gfs2_log_header lh_1, lh_m;
248 uint32_t blk_1, blk_2, blk_m;
252 blk_2 = jd->jd_blocks - 1;
255 blk_m = (blk_1 + blk_2) / 2;
257 error = find_good_lh(jd, &blk_1, &lh_1);
261 error = find_good_lh(jd, &blk_m, &lh_m);
265 if (blk_1 == blk_m || blk_m == blk_2)
268 if (lh_1.lh_sequence <= lh_m.lh_sequence)
274 error = jhead_scan(jd, &lh_1);
284 * foreach_descriptor - go through the active part of the log
286 * @start: the first log header in the active region
287 * @end: the last log header (don't process the contents of this entry))
289 * Call a given function once for every log descriptor in the active
290 * portion of the log.
295 static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start,
296 unsigned int end, int pass)
298 struct gfs2_sbd *sdp = get_v2ip(jd->jd_inode)->i_sbd;
299 struct buffer_head *bh;
300 struct gfs2_log_descriptor *ld;
304 unsigned int offset = sizeof(struct gfs2_log_descriptor);
305 offset += (sizeof(__be64)-1);
306 offset &= ~(sizeof(__be64)-1);
308 while (start != end) {
309 error = gfs2_replay_read_block(jd, start, &bh);
312 if (gfs2_meta_check(sdp, bh)) {
316 ld = (struct gfs2_log_descriptor *)bh->b_data;
317 length = be32_to_cpu(ld->ld_length);
319 if (be16_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
320 struct gfs2_log_header lh;
321 error = get_log_header(jd, start, &lh);
323 gfs2_replay_incr_blk(sdp, &start);
327 gfs2_consist_inode(get_v2ip(jd->jd_inode));
332 } else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
336 ptr = (__be64 *)(bh->b_data + offset);
337 error = lops_scan_elements(jd, start, ld, ptr, pass);
344 gfs2_replay_incr_blk(sdp, &start);
353 * clean_journal - mark a dirty journal as being clean
354 * @sdp: the filesystem
356 * @gl: the journal's glock
357 * @head: the head journal to start from
362 static int clean_journal(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
364 struct gfs2_inode *ip = get_v2ip(jd->jd_inode);
365 struct gfs2_sbd *sdp = ip->i_sbd;
369 struct gfs2_log_header *lh;
371 struct buffer_head *bh;
374 lblock = head->lh_blkno;
375 gfs2_replay_incr_blk(sdp, &lblock);
376 error = gfs2_block_map(ip, lblock, &new, &dblock, NULL);
380 gfs2_consist_inode(ip);
384 bh = sb_getblk(sdp->sd_vfs, dblock);
386 memset(bh->b_data, 0, bh->b_size);
387 set_buffer_uptodate(bh);
388 clear_buffer_dirty(bh);
391 lh = (struct gfs2_log_header *)bh->b_data;
392 memset(lh, 0, sizeof(struct gfs2_log_header));
393 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
394 lh->lh_header.mh_type = cpu_to_be16(GFS2_METATYPE_LH);
395 lh->lh_header.mh_format = cpu_to_be16(GFS2_FORMAT_LH);
396 lh->lh_sequence = cpu_to_be64(head->lh_sequence + 1);
397 lh->lh_flags = cpu_to_be32(GFS2_LOG_HEAD_UNMOUNT);
398 lh->lh_blkno = cpu_to_be32(lblock);
399 hash = gfs2_disk_hash((const char *)lh, sizeof(struct gfs2_log_header));
400 lh->lh_hash = cpu_to_be32(hash);
402 set_buffer_dirty(bh);
403 if (sync_dirty_buffer(bh))
404 gfs2_io_error_bh(sdp, bh);
411 * gfs2_recover_journal - recovery a given journal
412 * @jd: the struct gfs2_jdesc describing the journal
413 * @wait: Don't return until the journal is clean (or an error is encountered)
415 * Acquire the journal's lock, check to see if the journal is clean, and
416 * do recovery if necessary.
421 int gfs2_recover_journal(struct gfs2_jdesc *jd, int wait)
423 struct gfs2_sbd *sdp = get_v2ip(jd->jd_inode)->i_sbd;
424 struct gfs2_log_header head;
425 struct gfs2_holder j_gh, ji_gh, t_gh;
431 fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n", jd->jd_jid);
433 /* Aquire the journal lock so we can do recovery */
435 error = gfs2_glock_nq_num(sdp,
436 jd->jd_jid, &gfs2_journal_glops,
439 ((wait) ? 0 : LM_FLAG_TRY) |
446 fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
453 error = gfs2_glock_nq_init(get_v2ip(jd->jd_inode)->i_gl, LM_ST_SHARED,
454 LM_FLAG_NOEXP, &ji_gh);
458 fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
460 error = gfs2_jdesc_check(jd);
462 goto fail_gunlock_ji;
464 error = gfs2_find_jhead(jd, &head);
466 goto fail_gunlock_ji;
468 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
469 fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
474 /* Acquire a shared hold on the transaction lock */
476 error = gfs2_glock_nq_init(sdp->sd_trans_gl,
485 goto fail_gunlock_ji;
487 if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
488 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
491 if (sdp->sd_vfs->s_flags & MS_RDONLY)
496 fs_warn(sdp, "jid=%u: Can't replay: read-only FS\n",
499 goto fail_gunlock_tr;
502 fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
504 for (pass = 0; pass < 2; pass++) {
505 lops_before_scan(jd, &head, pass);
506 error = foreach_descriptor(jd, head.lh_tail,
507 head.lh_blkno, pass);
508 lops_after_scan(jd, error, pass);
510 goto fail_gunlock_tr;
513 error = clean_journal(jd, &head);
515 goto fail_gunlock_tr;
517 gfs2_glock_dq_uninit(&t_gh);
519 t = DIV_RU(jiffies - t, HZ);
521 fs_info(sdp, "jid=%u: Journal replayed in %lus\n",
525 gfs2_glock_dq_uninit(&ji_gh);
527 gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
529 gfs2_glock_dq_uninit(&j_gh);
531 fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
536 gfs2_glock_dq_uninit(&t_gh);
539 gfs2_glock_dq_uninit(&ji_gh);
542 gfs2_glock_dq_uninit(&j_gh);
544 fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
547 gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
553 * gfs2_check_journals - Recover any dirty journals
554 * @sdp: the filesystem
558 void gfs2_check_journals(struct gfs2_sbd *sdp)
560 struct gfs2_jdesc *jd;
563 jd = gfs2_jdesc_find_dirty(sdp);
567 if (jd != sdp->sd_jdesc)
568 gfs2_recover_journal(jd, NO_WAIT);