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 = jd->jd_inode->i_gl;
36 error = gfs2_block_map(jd->jd_inode, blk, &new, &dblock, &extlen);
40 gfs2_consist_inode(jd->jd_inode);
44 gfs2_meta_ra(gl, dblock, extlen);
45 error = gfs2_meta_read(gl, dblock, DIO_START | DIO_WAIT, bh);
50 int gfs2_revoke_add(struct gfs2_sbd *sdp, uint64_t blkno, unsigned int where)
52 struct list_head *head = &sdp->sd_revoke_list;
53 struct gfs2_revoke_replay *rr;
56 list_for_each_entry(rr, head, rr_list) {
57 if (rr->rr_blkno == blkno) {
68 rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_KERNEL);
74 list_add(&rr->rr_list, head);
79 int gfs2_revoke_check(struct gfs2_sbd *sdp, uint64_t blkno, unsigned int where)
81 struct gfs2_revoke_replay *rr;
82 int wrap, a, b, revoke;
85 list_for_each_entry(rr, &sdp->sd_revoke_list, rr_list) {
86 if (rr->rr_blkno == blkno) {
95 wrap = (rr->rr_where < sdp->sd_replay_tail);
96 a = (sdp->sd_replay_tail < where);
97 b = (where < rr->rr_where);
98 revoke = (wrap) ? (a || b) : (a && b);
103 void gfs2_revoke_clean(struct gfs2_sbd *sdp)
105 struct list_head *head = &sdp->sd_revoke_list;
106 struct gfs2_revoke_replay *rr;
108 while (!list_empty(head)) {
109 rr = list_entry(head->next, struct gfs2_revoke_replay, rr_list);
110 list_del(&rr->rr_list);
116 * get_log_header - read the log header for a given segment
118 * @blk: the block to look at
119 * @lh: the log header to return
121 * Read the log header for a given segement in a given journal. Do a few
122 * sanity checks on it.
124 * Returns: 0 on success,
125 * 1 if the header was invalid or incomplete,
129 static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
130 struct gfs2_log_header *head)
132 struct buffer_head *bh;
133 struct gfs2_log_header lh;
137 error = gfs2_replay_read_block(jd, blk, &bh);
141 memcpy(&lh, bh->b_data, sizeof(struct gfs2_log_header));
143 hash = gfs2_disk_hash((char *)&lh, sizeof(struct gfs2_log_header));
144 gfs2_log_header_in(&lh, bh->b_data);
148 if (lh.lh_header.mh_magic != GFS2_MAGIC ||
149 lh.lh_header.mh_type != GFS2_METATYPE_LH ||
150 lh.lh_blkno != blk ||
160 * find_good_lh - find a good log header
162 * @blk: the segment to start searching from
163 * @lh: the log header to fill in
164 * @forward: if true search forward in the log, else search backward
166 * Call get_log_header() to get a log header for a segment, but if the
167 * segment is bad, either scan forward or backward until we find a good one.
172 static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk,
173 struct gfs2_log_header *head)
175 unsigned int orig_blk = *blk;
179 error = get_log_header(jd, *blk, head);
183 if (++*blk == jd->jd_blocks)
186 if (*blk == orig_blk) {
187 gfs2_consist_inode(jd->jd_inode);
194 * jhead_scan - make sure we've found the head of the log
196 * @head: this is filled in with the log descriptor of the head
198 * At this point, seg and lh should be either the head of the log or just
199 * before. Scan forward until we find the head.
204 static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
206 unsigned int blk = head->lh_blkno;
207 struct gfs2_log_header lh;
211 if (++blk == jd->jd_blocks)
214 error = get_log_header(jd, blk, &lh);
220 if (lh.lh_sequence == head->lh_sequence) {
221 gfs2_consist_inode(jd->jd_inode);
224 if (lh.lh_sequence < head->lh_sequence)
234 * gfs2_find_jhead - find the head of a log
236 * @head: the log descriptor for the head of the log is returned here
238 * Do a binary search of a journal and find the valid log entry with the
239 * highest sequence number. (i.e. the log head)
244 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
246 struct gfs2_log_header lh_1, lh_m;
247 uint32_t blk_1, blk_2, blk_m;
251 blk_2 = jd->jd_blocks - 1;
254 blk_m = (blk_1 + blk_2) / 2;
256 error = find_good_lh(jd, &blk_1, &lh_1);
260 error = find_good_lh(jd, &blk_m, &lh_m);
264 if (blk_1 == blk_m || blk_m == blk_2)
267 if (lh_1.lh_sequence <= lh_m.lh_sequence)
273 error = jhead_scan(jd, &lh_1);
283 * foreach_descriptor - go through the active part of the log
285 * @start: the first log header in the active region
286 * @end: the last log header (don't process the contents of this entry))
288 * Call a given function once for every log descriptor in the active
289 * portion of the log.
294 static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start,
295 unsigned int end, int pass)
297 struct gfs2_sbd *sdp = jd->jd_inode->i_sbd;
298 struct buffer_head *bh;
299 struct gfs2_log_descriptor *ld;
303 unsigned int offset = sizeof(struct gfs2_log_descriptor);
304 offset += (sizeof(__be64)-1);
305 offset &= ~(sizeof(__be64)-1);
307 while (start != end) {
308 error = gfs2_replay_read_block(jd, start, &bh);
311 if (gfs2_meta_check(sdp, bh)) {
315 ld = (struct gfs2_log_descriptor *)bh->b_data;
316 length = be32_to_cpu(ld->ld_length);
318 if (be16_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
319 struct gfs2_log_header lh;
320 error = get_log_header(jd, start, &lh);
322 gfs2_replay_incr_blk(sdp, &start);
326 gfs2_consist_inode(jd->jd_inode);
331 } else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
335 ptr = (__be64 *)(bh->b_data + offset);
336 error = lops_scan_elements(jd, start, ld, ptr, pass);
343 gfs2_replay_incr_blk(sdp, &start);
352 * clean_journal - mark a dirty journal as being clean
353 * @sdp: the filesystem
355 * @gl: the journal's glock
356 * @head: the head journal to start from
361 static int clean_journal(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
363 struct gfs2_inode *ip = jd->jd_inode;
364 struct gfs2_sbd *sdp = ip->i_sbd;
368 struct gfs2_log_header *lh;
370 struct buffer_head *bh;
373 lblock = head->lh_blkno;
374 gfs2_replay_incr_blk(sdp, &lblock);
375 error = gfs2_block_map(ip, lblock, &new, &dblock, NULL);
379 gfs2_consist_inode(ip);
383 bh = sb_getblk(sdp->sd_vfs, dblock);
385 memset(bh->b_data, 0, bh->b_size);
386 set_buffer_uptodate(bh);
387 clear_buffer_dirty(bh);
390 lh = (struct gfs2_log_header *)bh->b_data;
391 memset(lh, 0, sizeof(struct gfs2_log_header));
392 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
393 lh->lh_header.mh_type = cpu_to_be16(GFS2_METATYPE_LH);
394 lh->lh_header.mh_format = cpu_to_be16(GFS2_FORMAT_LH);
395 lh->lh_sequence = cpu_to_be64(head->lh_sequence + 1);
396 lh->lh_flags = cpu_to_be32(GFS2_LOG_HEAD_UNMOUNT);
397 lh->lh_blkno = cpu_to_be32(lblock);
398 hash = gfs2_disk_hash((const char *)lh, sizeof(struct gfs2_log_header));
399 lh->lh_hash = cpu_to_be32(hash);
401 set_buffer_dirty(bh);
402 if (sync_dirty_buffer(bh))
403 gfs2_io_error_bh(sdp, bh);
410 * gfs2_recover_journal - recovery a given journal
411 * @jd: the struct gfs2_jdesc describing the journal
412 * @wait: Don't return until the journal is clean (or an error is encountered)
414 * Acquire the journal's lock, check to see if the journal is clean, and
415 * do recovery if necessary.
420 int gfs2_recover_journal(struct gfs2_jdesc *jd, int wait)
422 struct gfs2_sbd *sdp = jd->jd_inode->i_sbd;
423 struct gfs2_log_header head;
424 struct gfs2_holder j_gh, ji_gh, t_gh;
430 fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n", jd->jd_jid);
432 /* Aquire the journal lock so we can do recovery */
434 error = gfs2_glock_nq_num(sdp,
435 jd->jd_jid, &gfs2_journal_glops,
438 ((wait) ? 0 : LM_FLAG_TRY) |
445 fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
452 error = gfs2_glock_nq_init(jd->jd_inode->i_gl, LM_ST_SHARED,
453 LM_FLAG_NOEXP, &ji_gh);
457 fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
459 error = gfs2_jdesc_check(jd);
461 goto fail_gunlock_ji;
463 error = gfs2_find_jhead(jd, &head);
465 goto fail_gunlock_ji;
467 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
468 fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
473 /* Acquire a shared hold on the transaction lock */
475 error = gfs2_glock_nq_init(sdp->sd_trans_gl,
484 goto fail_gunlock_ji;
486 if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
487 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
490 if (sdp->sd_vfs->s_flags & MS_RDONLY)
495 fs_warn(sdp, "jid=%u: Can't replay: read-only FS\n",
498 goto fail_gunlock_tr;
501 fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
503 for (pass = 0; pass < 2; pass++) {
504 lops_before_scan(jd, &head, pass);
505 error = foreach_descriptor(jd, head.lh_tail,
506 head.lh_blkno, pass);
507 lops_after_scan(jd, error, pass);
509 goto fail_gunlock_tr;
512 error = clean_journal(jd, &head);
514 goto fail_gunlock_tr;
516 gfs2_glock_dq_uninit(&t_gh);
518 t = DIV_RU(jiffies - t, HZ);
520 fs_info(sdp, "jid=%u: Journal replayed in %lus\n",
524 gfs2_glock_dq_uninit(&ji_gh);
526 gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
528 gfs2_glock_dq_uninit(&j_gh);
530 fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
535 gfs2_glock_dq_uninit(&t_gh);
538 gfs2_glock_dq_uninit(&ji_gh);
541 gfs2_glock_dq_uninit(&j_gh);
543 fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
546 gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
552 * gfs2_check_journals - Recover any dirty journals
553 * @sdp: the filesystem
557 void gfs2_check_journals(struct gfs2_sbd *sdp)
559 struct gfs2_jdesc *jd;
562 jd = gfs2_jdesc_find_dirty(sdp);
566 if (jd != sdp->sd_jdesc)
567 gfs2_recover_journal(jd, NO_WAIT);