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 <linux/gfs2_ondisk.h>
16 #include <linux/crc32.h>
17 #include <asm/semaphore.h>
20 #include "lm_interface.h"
33 int gfs2_replay_read_block(struct gfs2_jdesc *jd, unsigned int blk,
34 struct buffer_head **bh)
36 struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
37 struct gfs2_glock *gl = ip->i_gl;
43 error = gfs2_extent_map(ip->i_vnode, blk, &new, &dblock, &extlen);
47 gfs2_consist_inode(ip);
51 gfs2_meta_ra(gl, dblock, extlen);
52 error = gfs2_meta_read(gl, dblock, DIO_START | DIO_WAIT, bh);
57 int gfs2_revoke_add(struct gfs2_sbd *sdp, uint64_t blkno, unsigned int where)
59 struct list_head *head = &sdp->sd_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_KERNEL);
81 list_add(&rr->rr_list, head);
86 int gfs2_revoke_check(struct gfs2_sbd *sdp, uint64_t blkno, unsigned int where)
88 struct gfs2_revoke_replay *rr;
89 int wrap, a, b, revoke;
92 list_for_each_entry(rr, &sdp->sd_revoke_list, rr_list) {
93 if (rr->rr_blkno == blkno) {
102 wrap = (rr->rr_where < sdp->sd_replay_tail);
103 a = (sdp->sd_replay_tail < where);
104 b = (where < rr->rr_where);
105 revoke = (wrap) ? (a || b) : (a && b);
110 void gfs2_revoke_clean(struct gfs2_sbd *sdp)
112 struct list_head *head = &sdp->sd_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 *head)
139 struct buffer_head *bh;
140 struct gfs2_log_header lh;
144 error = gfs2_replay_read_block(jd, blk, &bh);
148 memcpy(&lh, bh->b_data, sizeof(struct gfs2_log_header));
150 hash = gfs2_disk_hash((char *)&lh, sizeof(struct gfs2_log_header));
151 gfs2_log_header_in(&lh, bh->b_data);
155 if (lh.lh_header.mh_magic != GFS2_MAGIC ||
156 lh.lh_header.mh_type != GFS2_METATYPE_LH ||
157 lh.lh_blkno != blk ||
167 * find_good_lh - find a good log header
169 * @blk: the segment to start searching from
170 * @lh: the log header to fill in
171 * @forward: if true search forward in the log, else search backward
173 * Call get_log_header() to get a log header for a segment, but if the
174 * segment is bad, either scan forward or backward until we find a good one.
179 static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk,
180 struct gfs2_log_header *head)
182 unsigned int orig_blk = *blk;
186 error = get_log_header(jd, *blk, head);
190 if (++*blk == jd->jd_blocks)
193 if (*blk == orig_blk) {
194 gfs2_consist_inode(jd->jd_inode->u.generic_ip);
201 * jhead_scan - make sure we've found the head of the log
203 * @head: this is filled in with the log descriptor of the head
205 * At this point, seg and lh should be either the head of the log or just
206 * before. Scan forward until we find the head.
211 static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
213 unsigned int blk = head->lh_blkno;
214 struct gfs2_log_header lh;
218 if (++blk == jd->jd_blocks)
221 error = get_log_header(jd, blk, &lh);
227 if (lh.lh_sequence == head->lh_sequence) {
228 gfs2_consist_inode(jd->jd_inode->u.generic_ip);
231 if (lh.lh_sequence < head->lh_sequence)
241 * gfs2_find_jhead - find the head of a log
243 * @head: the log descriptor for the head of the log is returned here
245 * Do a binary search of a journal and find the valid log entry with the
246 * highest sequence number. (i.e. the log head)
251 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
253 struct gfs2_log_header lh_1, lh_m;
254 uint32_t blk_1, blk_2, blk_m;
258 blk_2 = jd->jd_blocks - 1;
261 blk_m = (blk_1 + blk_2) / 2;
263 error = find_good_lh(jd, &blk_1, &lh_1);
267 error = find_good_lh(jd, &blk_m, &lh_m);
271 if (blk_1 == blk_m || blk_m == blk_2)
274 if (lh_1.lh_sequence <= lh_m.lh_sequence)
280 error = jhead_scan(jd, &lh_1);
290 * foreach_descriptor - go through the active part of the log
292 * @start: the first log header in the active region
293 * @end: the last log header (don't process the contents of this entry))
295 * Call a given function once for every log descriptor in the active
296 * portion of the log.
301 static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start,
302 unsigned int end, int pass)
304 struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
305 struct gfs2_sbd *sdp = ip->i_sbd;
306 struct buffer_head *bh;
307 struct gfs2_log_descriptor *ld;
311 unsigned int offset = sizeof(struct gfs2_log_descriptor);
312 offset += (sizeof(__be64)-1);
313 offset &= ~(sizeof(__be64)-1);
315 while (start != end) {
316 error = gfs2_replay_read_block(jd, start, &bh);
319 if (gfs2_meta_check(sdp, bh)) {
323 ld = (struct gfs2_log_descriptor *)bh->b_data;
324 length = be32_to_cpu(ld->ld_length);
326 if (be32_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
327 struct gfs2_log_header lh;
328 error = get_log_header(jd, start, &lh);
330 gfs2_replay_incr_blk(sdp, &start);
334 gfs2_consist_inode(jd->jd_inode->u.generic_ip);
339 } else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
343 ptr = (__be64 *)(bh->b_data + offset);
344 error = lops_scan_elements(jd, start, ld, ptr, pass);
351 gfs2_replay_incr_blk(sdp, &start);
360 * clean_journal - mark a dirty journal as being clean
361 * @sdp: the filesystem
363 * @gl: the journal's glock
364 * @head: the head journal to start from
369 static int clean_journal(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
371 struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
372 struct gfs2_sbd *sdp = ip->i_sbd;
376 struct gfs2_log_header *lh;
378 struct buffer_head *bh;
382 lblock = head->lh_blkno;
383 gfs2_replay_incr_blk(sdp, &lblock);
384 error = gfs2_block_map(ip->i_vnode, lblock, &new, &dblock, &boundary);
388 gfs2_consist_inode(ip);
392 bh = sb_getblk(sdp->sd_vfs, dblock);
394 memset(bh->b_data, 0, bh->b_size);
395 set_buffer_uptodate(bh);
396 clear_buffer_dirty(bh);
399 lh = (struct gfs2_log_header *)bh->b_data;
400 memset(lh, 0, sizeof(struct gfs2_log_header));
401 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
402 lh->lh_header.mh_type = cpu_to_be16(GFS2_METATYPE_LH);
403 lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
404 lh->lh_sequence = cpu_to_be64(head->lh_sequence + 1);
405 lh->lh_flags = cpu_to_be32(GFS2_LOG_HEAD_UNMOUNT);
406 lh->lh_blkno = cpu_to_be32(lblock);
407 hash = gfs2_disk_hash((const char *)lh, sizeof(struct gfs2_log_header));
408 lh->lh_hash = cpu_to_be32(hash);
410 set_buffer_dirty(bh);
411 if (sync_dirty_buffer(bh))
412 gfs2_io_error_bh(sdp, bh);
419 * gfs2_recover_journal - recovery a given journal
420 * @jd: the struct gfs2_jdesc describing the journal
422 * Acquire the journal's lock, check to see if the journal is clean, and
423 * do recovery if necessary.
428 int gfs2_recover_journal(struct gfs2_jdesc *jd)
430 struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
431 struct gfs2_sbd *sdp = ip->i_sbd;
432 struct gfs2_log_header head;
433 struct gfs2_holder j_gh, ji_gh, t_gh;
439 if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
440 fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
443 /* Aquire the journal lock so we can do recovery */
445 error = gfs2_glock_nq_num(sdp, jd->jd_jid, &gfs2_journal_glops,
447 LM_FLAG_NOEXP | LM_FLAG_TRY | GL_NOCACHE,
454 fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
461 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
462 LM_FLAG_NOEXP, &ji_gh);
466 fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
469 fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
471 error = gfs2_jdesc_check(jd);
473 goto fail_gunlock_ji;
475 error = gfs2_find_jhead(jd, &head);
477 goto fail_gunlock_ji;
479 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
480 fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
485 /* Acquire a shared hold on the transaction lock */
487 error = gfs2_glock_nq_init(sdp->sd_trans_gl,
489 LM_FLAG_NOEXP | LM_FLAG_PRIORITY |
490 GL_NOCANCEL | GL_NOCACHE,
493 goto fail_gunlock_ji;
495 if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
496 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
499 if (sdp->sd_vfs->s_flags & MS_RDONLY)
504 fs_warn(sdp, "jid=%u: Can't replay: read-only FS\n",
507 goto fail_gunlock_tr;
510 fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
512 for (pass = 0; pass < 2; pass++) {
513 lops_before_scan(jd, &head, pass);
514 error = foreach_descriptor(jd, head.lh_tail,
515 head.lh_blkno, pass);
516 lops_after_scan(jd, error, pass);
518 goto fail_gunlock_tr;
521 error = clean_journal(jd, &head);
523 goto fail_gunlock_tr;
525 gfs2_glock_dq_uninit(&t_gh);
526 t = DIV_ROUND_UP(jiffies - t, HZ);
527 fs_info(sdp, "jid=%u: Journal replayed in %lus\n",
531 if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
532 gfs2_glock_dq_uninit(&ji_gh);
534 gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
536 if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
537 gfs2_glock_dq_uninit(&j_gh);
539 fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
543 gfs2_glock_dq_uninit(&t_gh);
545 if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
546 gfs2_glock_dq_uninit(&ji_gh);
548 gfs2_glock_dq_uninit(&j_gh);
551 fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
554 gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
559 * gfs2_check_journals - Recover any dirty journals
560 * @sdp: the filesystem
564 void gfs2_check_journals(struct gfs2_sbd *sdp)
566 struct gfs2_jdesc *jd;
569 jd = gfs2_jdesc_find_dirty(sdp);
573 if (jd != sdp->sd_jdesc)
574 gfs2_recover_journal(jd);