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/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 <linux/lm_interface.h>
18 #include <linux/delay.h>
33 * gfs2_struct2blk - compute stuff
34 * @sdp: the filesystem
35 * @nstruct: the number of structures
36 * @ssize: the size of the structures
38 * Compute the number of log descriptor blocks needed to hold a certain number
39 * of structures of a certain size.
41 * Returns: the number of blocks needed (minimum is always 1)
44 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
48 unsigned int first, second;
51 first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
53 if (nstruct > first) {
54 second = (sdp->sd_sb.sb_bsize -
55 sizeof(struct gfs2_meta_header)) / ssize;
56 blks += DIV_ROUND_UP(nstruct - first, second);
63 * gfs2_ail1_start_one - Start I/O on a part of the AIL
64 * @sdp: the filesystem
65 * @tr: the part of the AIL
69 static void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
71 struct gfs2_bufdata *bd, *s;
72 struct buffer_head *bh;
75 BUG_ON(!spin_is_locked(&sdp->sd_log_lock));
80 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
84 gfs2_assert(sdp, bd->bd_ail == ai);
86 if (!buffer_busy(bh)) {
87 if (!buffer_uptodate(bh)) {
89 gfs2_io_error_bh(sdp, bh);
92 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
96 if (!buffer_dirty(bh))
99 list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
101 gfs2_log_unlock(sdp);
103 ll_rw_block(WRITE, 1, &bh);
113 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
114 * @sdp: the filesystem
119 static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags)
121 struct gfs2_bufdata *bd, *s;
122 struct buffer_head *bh;
124 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
128 gfs2_assert(sdp, bd->bd_ail == ai);
130 if (buffer_busy(bh)) {
137 if (!buffer_uptodate(bh))
138 gfs2_io_error_bh(sdp, bh);
140 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
143 return list_empty(&ai->ai_ail1_list);
146 static void gfs2_ail1_start(struct gfs2_sbd *sdp, int flags)
148 struct list_head *head = &sdp->sd_ail1_list;
150 struct list_head *first;
151 struct gfs2_ail *first_ai, *ai, *tmp;
155 if (list_empty(head)) {
156 gfs2_log_unlock(sdp);
159 sync_gen = sdp->sd_ail_sync_gen++;
162 first_ai = list_entry(first, struct gfs2_ail, ai_list);
163 first_ai->ai_sync_gen = sync_gen;
164 gfs2_ail1_start_one(sdp, first_ai); /* This may drop log lock */
170 if (first && (head->prev != first ||
171 gfs2_ail1_empty_one(sdp, first_ai, 0)))
175 list_for_each_entry_safe_reverse(ai, tmp, head, ai_list) {
176 if (ai->ai_sync_gen >= sync_gen)
178 ai->ai_sync_gen = sync_gen;
179 gfs2_ail1_start_one(sdp, ai); /* This may drop log lock */
185 gfs2_log_unlock(sdp);
188 int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
190 struct gfs2_ail *ai, *s;
195 list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) {
196 if (gfs2_ail1_empty_one(sdp, ai, flags))
197 list_move(&ai->ai_list, &sdp->sd_ail2_list);
198 else if (!(flags & DIO_ALL))
202 ret = list_empty(&sdp->sd_ail1_list);
204 gfs2_log_unlock(sdp);
211 * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
212 * @sdp: the filesystem
217 static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
219 struct list_head *head = &ai->ai_ail2_list;
220 struct gfs2_bufdata *bd;
222 while (!list_empty(head)) {
223 bd = list_entry(head->prev, struct gfs2_bufdata,
225 gfs2_assert(sdp, bd->bd_ail == ai);
227 list_del(&bd->bd_ail_st_list);
228 list_del(&bd->bd_ail_gl_list);
229 atomic_dec(&bd->bd_gl->gl_ail_count);
234 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
236 struct gfs2_ail *ai, *safe;
237 unsigned int old_tail = sdp->sd_log_tail;
238 int wrap = (new_tail < old_tail);
243 list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) {
244 a = (old_tail <= ai->ai_first);
245 b = (ai->ai_first < new_tail);
246 rm = (wrap) ? (a || b) : (a && b);
250 gfs2_ail2_empty_one(sdp, ai);
251 list_del(&ai->ai_list);
252 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail1_list));
253 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail2_list));
257 gfs2_log_unlock(sdp);
261 * gfs2_log_reserve - Make a log reservation
262 * @sdp: The GFS2 superblock
263 * @blks: The number of blocks to reserve
265 * Note that we never give out the last few blocks of the journal. Thats
266 * due to the fact that there is are a small number of header blocks
267 * associated with each log flush. The exact number can't be known until
268 * flush time, so we ensure that we have just enough free blocks at all
269 * times to avoid running out during a log flush.
274 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
276 unsigned int try = 0;
277 unsigned reserved_blks = 6 * (4096 / sdp->sd_vfs->s_blocksize);
279 if (gfs2_assert_warn(sdp, blks) ||
280 gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
283 mutex_lock(&sdp->sd_log_reserve_mutex);
285 while(sdp->sd_log_blks_free <= (blks + reserved_blks)) {
286 gfs2_log_unlock(sdp);
287 gfs2_ail1_empty(sdp, 0);
288 gfs2_log_flush(sdp, NULL);
291 gfs2_ail1_start(sdp, 0);
294 sdp->sd_log_blks_free -= blks;
295 gfs2_log_unlock(sdp);
296 mutex_unlock(&sdp->sd_log_reserve_mutex);
298 down_read(&sdp->sd_log_flush_lock);
304 * gfs2_log_release - Release a given number of log blocks
305 * @sdp: The GFS2 superblock
306 * @blks: The number of blocks
310 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
314 sdp->sd_log_blks_free += blks;
315 gfs2_assert_withdraw(sdp,
316 sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
317 gfs2_log_unlock(sdp);
318 up_read(&sdp->sd_log_flush_lock);
321 static u64 log_bmap(struct gfs2_sbd *sdp, unsigned int lbn)
323 struct inode *inode = sdp->sd_jdesc->jd_inode;
325 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
327 bh_map.b_size = 1 << inode->i_blkbits;
328 error = gfs2_block_map(inode, lbn, 0, &bh_map);
329 if (error || !bh_map.b_blocknr)
330 printk(KERN_INFO "error=%d, dbn=%llu lbn=%u", error,
331 (unsigned long long)bh_map.b_blocknr, lbn);
332 gfs2_assert_withdraw(sdp, !error && bh_map.b_blocknr);
334 return bh_map.b_blocknr;
338 * log_distance - Compute distance between two journal blocks
339 * @sdp: The GFS2 superblock
340 * @newer: The most recent journal block of the pair
341 * @older: The older journal block of the pair
343 * Compute the distance (in the journal direction) between two
344 * blocks in the journal
346 * Returns: the distance in blocks
349 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
354 dist = newer - older;
356 dist += sdp->sd_jdesc->jd_blocks;
361 static unsigned int current_tail(struct gfs2_sbd *sdp)
368 if (list_empty(&sdp->sd_ail1_list)) {
369 tail = sdp->sd_log_head;
371 ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail, ai_list);
375 gfs2_log_unlock(sdp);
380 static inline void log_incr_head(struct gfs2_sbd *sdp)
382 if (sdp->sd_log_flush_head == sdp->sd_log_tail)
383 gfs2_assert_withdraw(sdp, sdp->sd_log_flush_head == sdp->sd_log_head);
385 if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
386 sdp->sd_log_flush_head = 0;
387 sdp->sd_log_flush_wrapped = 1;
392 * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
393 * @sdp: The GFS2 superblock
395 * Returns: the buffer_head
398 struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
400 u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
401 struct gfs2_log_buf *lb;
402 struct buffer_head *bh;
404 lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
405 list_add(&lb->lb_list, &sdp->sd_log_flush_list);
407 bh = lb->lb_bh = sb_getblk(sdp->sd_vfs, blkno);
409 memset(bh->b_data, 0, bh->b_size);
410 set_buffer_uptodate(bh);
411 clear_buffer_dirty(bh);
420 * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
421 * @sdp: the filesystem
422 * @data: the data the buffer_head should point to
424 * Returns: the log buffer descriptor
427 struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
428 struct buffer_head *real)
430 u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
431 struct gfs2_log_buf *lb;
432 struct buffer_head *bh;
434 lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
435 list_add(&lb->lb_list, &sdp->sd_log_flush_list);
438 bh = lb->lb_bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL);
439 atomic_set(&bh->b_count, 1);
440 bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate);
441 set_bh_page(bh, real->b_page, bh_offset(real));
442 bh->b_blocknr = blkno;
443 bh->b_size = sdp->sd_sb.sb_bsize;
444 bh->b_bdev = sdp->sd_vfs->s_bdev;
451 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail, int pull)
453 unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
455 ail2_empty(sdp, new_tail);
458 sdp->sd_log_blks_free += dist - (pull ? 1 : 0);
459 gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
460 gfs2_log_unlock(sdp);
462 sdp->sd_log_tail = new_tail;
466 * log_write_header - Get and initialize a journal header buffer
467 * @sdp: The GFS2 superblock
469 * Returns: the initialized log buffer descriptor
472 static void log_write_header(struct gfs2_sbd *sdp, u32 flags, int pull)
474 u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
475 struct buffer_head *bh;
476 struct gfs2_log_header *lh;
480 bh = sb_getblk(sdp->sd_vfs, blkno);
482 memset(bh->b_data, 0, bh->b_size);
483 set_buffer_uptodate(bh);
484 clear_buffer_dirty(bh);
487 gfs2_ail1_empty(sdp, 0);
488 tail = current_tail(sdp);
490 lh = (struct gfs2_log_header *)bh->b_data;
491 memset(lh, 0, sizeof(struct gfs2_log_header));
492 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
493 lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
494 lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
495 lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++);
496 lh->lh_flags = cpu_to_be32(flags);
497 lh->lh_tail = cpu_to_be32(tail);
498 lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
499 hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
500 lh->lh_hash = cpu_to_be32(hash);
502 set_buffer_dirty(bh);
503 if (sync_dirty_buffer(bh))
504 gfs2_io_error_bh(sdp, bh);
507 if (sdp->sd_log_tail != tail)
508 log_pull_tail(sdp, tail, pull);
510 gfs2_assert_withdraw(sdp, !pull);
512 sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
516 static void log_flush_commit(struct gfs2_sbd *sdp)
518 struct list_head *head = &sdp->sd_log_flush_list;
519 struct gfs2_log_buf *lb;
520 struct buffer_head *bh;
522 while (!list_empty(head)) {
523 lb = list_entry(head->next, struct gfs2_log_buf, lb_list);
524 list_del(&lb->lb_list);
528 if (!buffer_uptodate(bh))
529 gfs2_io_error_bh(sdp, bh);
531 while (atomic_read(&bh->b_count) != 1) /* Grrrr... */
533 free_buffer_head(bh);
539 log_write_header(sdp, 0, 0);
543 * gfs2_log_flush - flush incore transaction(s)
544 * @sdp: the filesystem
545 * @gl: The glock structure to flush. If NULL, flush the whole incore log
549 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
553 down_write(&sdp->sd_log_flush_lock);
557 if (list_empty(&gl->gl_le.le_list)) {
558 gfs2_log_unlock(sdp);
559 up_write(&sdp->sd_log_flush_lock);
562 gfs2_log_unlock(sdp);
565 ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
566 INIT_LIST_HEAD(&ai->ai_ail1_list);
567 INIT_LIST_HEAD(&ai->ai_ail2_list);
569 gfs2_assert_withdraw(sdp, sdp->sd_log_num_buf + sdp->sd_log_num_jdata == sdp->sd_log_commited_buf);
570 gfs2_assert_withdraw(sdp,
571 sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
573 sdp->sd_log_flush_head = sdp->sd_log_head;
574 sdp->sd_log_flush_wrapped = 0;
575 ai->ai_first = sdp->sd_log_flush_head;
577 lops_before_commit(sdp);
578 if (!list_empty(&sdp->sd_log_flush_list))
579 log_flush_commit(sdp);
580 else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle)
581 log_write_header(sdp, 0, PULL);
582 lops_after_commit(sdp, ai);
585 sdp->sd_log_head = sdp->sd_log_flush_head;
586 sdp->sd_log_blks_free -= sdp->sd_log_num_hdrs;
587 sdp->sd_log_blks_reserved = 0;
588 sdp->sd_log_commited_buf = 0;
589 sdp->sd_log_num_hdrs = 0;
590 sdp->sd_log_commited_revoke = 0;
592 if (!list_empty(&ai->ai_ail1_list)) {
593 list_add(&ai->ai_list, &sdp->sd_ail1_list);
596 gfs2_log_unlock(sdp);
598 sdp->sd_vfs->s_dirt = 0;
599 up_write(&sdp->sd_log_flush_lock);
604 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
606 unsigned int reserved = 0;
611 sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
612 gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_buf) >= 0);
613 sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
614 gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0);
616 if (sdp->sd_log_commited_buf)
617 reserved += sdp->sd_log_commited_buf;
618 if (sdp->sd_log_commited_revoke)
619 reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
624 old = sdp->sd_log_blks_free;
625 sdp->sd_log_blks_free += tr->tr_reserved -
626 (reserved - sdp->sd_log_blks_reserved);
628 gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free >= old);
629 gfs2_assert_withdraw(sdp,
630 sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks +
631 sdp->sd_log_num_hdrs);
633 sdp->sd_log_blks_reserved = reserved;
635 gfs2_log_unlock(sdp);
639 * gfs2_log_commit - Commit a transaction to the log
640 * @sdp: the filesystem
641 * @tr: the transaction
646 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
649 lops_incore_commit(sdp, tr);
651 sdp->sd_vfs->s_dirt = 1;
652 up_read(&sdp->sd_log_flush_lock);
655 if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks))
656 wake_up_process(sdp->sd_logd_process);
657 gfs2_log_unlock(sdp);
661 * gfs2_log_shutdown - write a shutdown header into a journal
662 * @sdp: the filesystem
666 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
668 down_write(&sdp->sd_log_flush_lock);
670 gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
671 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_gl);
672 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
673 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_jdata);
674 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
675 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
676 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
677 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_hdrs);
678 gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
680 sdp->sd_log_flush_head = sdp->sd_log_head;
681 sdp->sd_log_flush_wrapped = 0;
683 log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT, 0);
685 gfs2_assert_warn(sdp, sdp->sd_log_blks_free == sdp->sd_jdesc->jd_blocks);
686 gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
687 gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
689 sdp->sd_log_head = sdp->sd_log_flush_head;
690 sdp->sd_log_tail = sdp->sd_log_head;
692 up_write(&sdp->sd_log_flush_lock);
697 * gfs2_meta_syncfs - sync all the buffers in a filesystem
698 * @sdp: the filesystem
702 void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
704 gfs2_log_flush(sdp, NULL);
706 gfs2_ail1_start(sdp, DIO_ALL);
707 if (gfs2_ail1_empty(sdp, DIO_ALL))