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 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 <linux/gfs2_ondisk.h>
16 #include <linux/crc32.h>
19 #include "lm_interface.h"
32 int gfs2_replay_read_block(struct gfs2_jdesc *jd, unsigned int blk,
33 struct buffer_head **bh)
35 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
36 struct gfs2_glock *gl = ip->i_gl;
42 error = gfs2_extent_map(&ip->i_inode, blk, &new, &dblock, &extlen);
46 gfs2_consist_inode(ip);
50 gfs2_meta_ra(gl, dblock, extlen);
51 error = gfs2_meta_read(gl, dblock, DIO_START | DIO_WAIT, bh);
56 int gfs2_revoke_add(struct gfs2_sbd *sdp, uint64_t blkno, unsigned int where)
58 struct list_head *head = &sdp->sd_revoke_list;
59 struct gfs2_revoke_replay *rr;
62 list_for_each_entry(rr, head, rr_list) {
63 if (rr->rr_blkno == blkno) {
74 rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_KERNEL);
80 list_add(&rr->rr_list, head);
85 int gfs2_revoke_check(struct gfs2_sbd *sdp, uint64_t blkno, unsigned int where)
87 struct gfs2_revoke_replay *rr;
88 int wrap, a, b, revoke;
91 list_for_each_entry(rr, &sdp->sd_revoke_list, rr_list) {
92 if (rr->rr_blkno == blkno) {
101 wrap = (rr->rr_where < sdp->sd_replay_tail);
102 a = (sdp->sd_replay_tail < where);
103 b = (where < rr->rr_where);
104 revoke = (wrap) ? (a || b) : (a && b);
109 void gfs2_revoke_clean(struct gfs2_sbd *sdp)
111 struct list_head *head = &sdp->sd_revoke_list;
112 struct gfs2_revoke_replay *rr;
114 while (!list_empty(head)) {
115 rr = list_entry(head->next, struct gfs2_revoke_replay, rr_list);
116 list_del(&rr->rr_list);
122 * get_log_header - read the log header for a given segment
124 * @blk: the block to look at
125 * @lh: the log header to return
127 * Read the log header for a given segement in a given journal. Do a few
128 * sanity checks on it.
130 * Returns: 0 on success,
131 * 1 if the header was invalid or incomplete,
135 static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
136 struct gfs2_log_header *head)
138 struct buffer_head *bh;
139 struct gfs2_log_header lh;
143 error = gfs2_replay_read_block(jd, blk, &bh);
147 memcpy(&lh, bh->b_data, sizeof(struct gfs2_log_header));
149 hash = gfs2_disk_hash((char *)&lh, sizeof(struct gfs2_log_header));
150 gfs2_log_header_in(&lh, bh->b_data);
154 if (lh.lh_header.mh_magic != GFS2_MAGIC ||
155 lh.lh_header.mh_type != GFS2_METATYPE_LH ||
156 lh.lh_blkno != blk ||
166 * find_good_lh - find a good log header
168 * @blk: the segment to start searching from
169 * @lh: the log header to fill in
170 * @forward: if true search forward in the log, else search backward
172 * Call get_log_header() to get a log header for a segment, but if the
173 * segment is bad, either scan forward or backward until we find a good one.
178 static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk,
179 struct gfs2_log_header *head)
181 unsigned int orig_blk = *blk;
185 error = get_log_header(jd, *blk, head);
189 if (++*blk == jd->jd_blocks)
192 if (*blk == orig_blk) {
193 gfs2_consist_inode(GFS2_I(jd->jd_inode));
200 * jhead_scan - make sure we've found the head of the log
202 * @head: this is filled in with the log descriptor of the head
204 * At this point, seg and lh should be either the head of the log or just
205 * before. Scan forward until we find the head.
210 static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
212 unsigned int blk = head->lh_blkno;
213 struct gfs2_log_header lh;
217 if (++blk == jd->jd_blocks)
220 error = get_log_header(jd, blk, &lh);
226 if (lh.lh_sequence == head->lh_sequence) {
227 gfs2_consist_inode(GFS2_I(jd->jd_inode));
230 if (lh.lh_sequence < head->lh_sequence)
240 * gfs2_find_jhead - find the head of a log
242 * @head: the log descriptor for the head of the log is returned here
244 * Do a binary search of a journal and find the valid log entry with the
245 * highest sequence number. (i.e. the log head)
250 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
252 struct gfs2_log_header lh_1, lh_m;
253 uint32_t blk_1, blk_2, blk_m;
257 blk_2 = jd->jd_blocks - 1;
260 blk_m = (blk_1 + blk_2) / 2;
262 error = find_good_lh(jd, &blk_1, &lh_1);
266 error = find_good_lh(jd, &blk_m, &lh_m);
270 if (blk_1 == blk_m || blk_m == blk_2)
273 if (lh_1.lh_sequence <= lh_m.lh_sequence)
279 error = jhead_scan(jd, &lh_1);
289 * foreach_descriptor - go through the active part of the log
291 * @start: the first log header in the active region
292 * @end: the last log header (don't process the contents of this entry))
294 * Call a given function once for every log descriptor in the active
295 * portion of the log.
300 static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start,
301 unsigned int end, int pass)
303 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
304 struct buffer_head *bh;
305 struct gfs2_log_descriptor *ld;
309 unsigned int offset = sizeof(struct gfs2_log_descriptor);
310 offset += (sizeof(__be64)-1);
311 offset &= ~(sizeof(__be64)-1);
313 while (start != end) {
314 error = gfs2_replay_read_block(jd, start, &bh);
317 if (gfs2_meta_check(sdp, bh)) {
321 ld = (struct gfs2_log_descriptor *)bh->b_data;
322 length = be32_to_cpu(ld->ld_length);
324 if (be32_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
325 struct gfs2_log_header lh;
326 error = get_log_header(jd, start, &lh);
328 gfs2_replay_incr_blk(sdp, &start);
332 gfs2_consist_inode(GFS2_I(jd->jd_inode));
337 } else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
341 ptr = (__be64 *)(bh->b_data + offset);
342 error = lops_scan_elements(jd, start, ld, ptr, pass);
349 gfs2_replay_incr_blk(sdp, &start);
358 * clean_journal - mark a dirty journal as being clean
359 * @sdp: the filesystem
361 * @gl: the journal's glock
362 * @head: the head journal to start from
367 static int clean_journal(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
369 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
370 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
374 struct gfs2_log_header *lh;
376 struct buffer_head *bh;
380 lblock = head->lh_blkno;
381 gfs2_replay_incr_blk(sdp, &lblock);
382 error = gfs2_block_map(&ip->i_inode, lblock, &new, &dblock, &boundary);
386 gfs2_consist_inode(ip);
390 bh = sb_getblk(sdp->sd_vfs, dblock);
392 memset(bh->b_data, 0, bh->b_size);
393 set_buffer_uptodate(bh);
394 clear_buffer_dirty(bh);
397 lh = (struct gfs2_log_header *)bh->b_data;
398 memset(lh, 0, sizeof(struct gfs2_log_header));
399 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
400 lh->lh_header.mh_type = cpu_to_be16(GFS2_METATYPE_LH);
401 lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
402 lh->lh_sequence = cpu_to_be64(head->lh_sequence + 1);
403 lh->lh_flags = cpu_to_be32(GFS2_LOG_HEAD_UNMOUNT);
404 lh->lh_blkno = cpu_to_be32(lblock);
405 hash = gfs2_disk_hash((const char *)lh, sizeof(struct gfs2_log_header));
406 lh->lh_hash = cpu_to_be32(hash);
408 set_buffer_dirty(bh);
409 if (sync_dirty_buffer(bh))
410 gfs2_io_error_bh(sdp, bh);
417 * gfs2_recover_journal - recovery a given journal
418 * @jd: the struct gfs2_jdesc describing the journal
420 * Acquire the journal's lock, check to see if the journal is clean, and
421 * do recovery if necessary.
426 int gfs2_recover_journal(struct gfs2_jdesc *jd)
428 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
429 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
430 struct gfs2_log_header head;
431 struct gfs2_holder j_gh, ji_gh, t_gh;
437 if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
438 fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
441 /* Aquire the journal lock so we can do recovery */
443 error = gfs2_glock_nq_num(sdp, jd->jd_jid, &gfs2_journal_glops,
445 LM_FLAG_NOEXP | LM_FLAG_TRY | GL_NOCACHE,
452 fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
459 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
460 LM_FLAG_NOEXP, &ji_gh);
464 fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
467 fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
469 error = gfs2_jdesc_check(jd);
471 goto fail_gunlock_ji;
473 error = gfs2_find_jhead(jd, &head);
475 goto fail_gunlock_ji;
477 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
478 fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
483 /* Acquire a shared hold on the transaction lock */
485 error = gfs2_glock_nq_init(sdp->sd_trans_gl,
487 LM_FLAG_NOEXP | LM_FLAG_PRIORITY |
488 GL_NOCANCEL | GL_NOCACHE,
491 goto fail_gunlock_ji;
493 if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
494 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
497 if (sdp->sd_vfs->s_flags & MS_RDONLY)
502 fs_warn(sdp, "jid=%u: Can't replay: read-only FS\n",
505 goto fail_gunlock_tr;
508 fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
510 for (pass = 0; pass < 2; pass++) {
511 lops_before_scan(jd, &head, pass);
512 error = foreach_descriptor(jd, head.lh_tail,
513 head.lh_blkno, pass);
514 lops_after_scan(jd, error, pass);
516 goto fail_gunlock_tr;
519 error = clean_journal(jd, &head);
521 goto fail_gunlock_tr;
523 gfs2_glock_dq_uninit(&t_gh);
524 t = DIV_ROUND_UP(jiffies - t, HZ);
525 fs_info(sdp, "jid=%u: Journal replayed in %lus\n",
529 if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
530 gfs2_glock_dq_uninit(&ji_gh);
532 gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
534 if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
535 gfs2_glock_dq_uninit(&j_gh);
537 fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
541 gfs2_glock_dq_uninit(&t_gh);
543 if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
544 gfs2_glock_dq_uninit(&ji_gh);
546 gfs2_glock_dq_uninit(&j_gh);
549 fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
552 gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
557 * gfs2_check_journals - Recover any dirty journals
558 * @sdp: the filesystem
562 void gfs2_check_journals(struct gfs2_sbd *sdp)
564 struct gfs2_jdesc *jd;
567 jd = gfs2_jdesc_find_dirty(sdp);
571 if (jd != sdp->sd_jdesc)
572 gfs2_recover_journal(jd);