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);
87 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
91 if (!buffer_busy(bh)) {
92 if (!buffer_uptodate(bh)) {
94 gfs2_io_error_bh(sdp, bh);
97 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
101 if (!buffer_dirty(bh))
104 list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
106 gfs2_log_unlock(sdp);
108 ll_rw_block(WRITE, 1, &bh);
118 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
119 * @sdp: the filesystem
124 static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags)
126 struct gfs2_bufdata *bd, *s;
127 struct buffer_head *bh;
129 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
134 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
138 gfs2_assert(sdp, bd->bd_ail == ai);
140 if (buffer_busy(bh)) {
147 if (!buffer_uptodate(bh))
148 gfs2_io_error_bh(sdp, bh);
150 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
153 return list_empty(&ai->ai_ail1_list);
156 static void gfs2_ail1_start(struct gfs2_sbd *sdp, int flags)
158 struct list_head *head = &sdp->sd_ail1_list;
160 struct list_head *first;
161 struct gfs2_ail *first_ai, *ai, *tmp;
165 if (list_empty(head)) {
166 gfs2_log_unlock(sdp);
169 sync_gen = sdp->sd_ail_sync_gen++;
172 first_ai = list_entry(first, struct gfs2_ail, ai_list);
173 first_ai->ai_sync_gen = sync_gen;
174 gfs2_ail1_start_one(sdp, first_ai); /* This may drop log lock */
180 if (first && (head->prev != first ||
181 gfs2_ail1_empty_one(sdp, first_ai, 0)))
185 list_for_each_entry_safe_reverse(ai, tmp, head, ai_list) {
186 if (ai->ai_sync_gen >= sync_gen)
188 ai->ai_sync_gen = sync_gen;
189 gfs2_ail1_start_one(sdp, ai); /* This may drop log lock */
195 gfs2_log_unlock(sdp);
198 int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
200 struct gfs2_ail *ai, *s;
205 list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) {
206 if (gfs2_ail1_empty_one(sdp, ai, flags))
207 list_move(&ai->ai_list, &sdp->sd_ail2_list);
208 else if (!(flags & DIO_ALL))
212 ret = list_empty(&sdp->sd_ail1_list);
214 gfs2_log_unlock(sdp);
221 * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
222 * @sdp: the filesystem
227 static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
229 struct list_head *head = &ai->ai_ail2_list;
230 struct gfs2_bufdata *bd;
232 while (!list_empty(head)) {
233 bd = list_entry(head->prev, struct gfs2_bufdata,
235 gfs2_assert(sdp, bd->bd_ail == ai);
237 list_del(&bd->bd_ail_st_list);
238 list_del(&bd->bd_ail_gl_list);
239 atomic_dec(&bd->bd_gl->gl_ail_count);
244 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
246 struct gfs2_ail *ai, *safe;
247 unsigned int old_tail = sdp->sd_log_tail;
248 int wrap = (new_tail < old_tail);
253 list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) {
254 a = (old_tail <= ai->ai_first);
255 b = (ai->ai_first < new_tail);
256 rm = (wrap) ? (a || b) : (a && b);
260 gfs2_ail2_empty_one(sdp, ai);
261 list_del(&ai->ai_list);
262 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail1_list));
263 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail2_list));
267 gfs2_log_unlock(sdp);
271 * gfs2_log_reserve - Make a log reservation
272 * @sdp: The GFS2 superblock
273 * @blks: The number of blocks to reserve
275 * Note that we never give out the last few blocks of the journal. Thats
276 * due to the fact that there is a small number of header blocks
277 * associated with each log flush. The exact number can't be known until
278 * flush time, so we ensure that we have just enough free blocks at all
279 * times to avoid running out during a log flush.
284 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
286 unsigned int try = 0;
287 unsigned reserved_blks = 6 * (4096 / sdp->sd_vfs->s_blocksize);
289 if (gfs2_assert_warn(sdp, blks) ||
290 gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
293 mutex_lock(&sdp->sd_log_reserve_mutex);
295 while(sdp->sd_log_blks_free <= (blks + reserved_blks)) {
296 gfs2_log_unlock(sdp);
297 gfs2_ail1_empty(sdp, 0);
298 gfs2_log_flush(sdp, NULL);
301 gfs2_ail1_start(sdp, 0);
304 sdp->sd_log_blks_free -= blks;
305 gfs2_log_unlock(sdp);
306 mutex_unlock(&sdp->sd_log_reserve_mutex);
308 down_read(&sdp->sd_log_flush_lock);
314 * gfs2_log_release - Release a given number of log blocks
315 * @sdp: The GFS2 superblock
316 * @blks: The number of blocks
320 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
324 sdp->sd_log_blks_free += blks;
325 gfs2_assert_withdraw(sdp,
326 sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
327 gfs2_log_unlock(sdp);
328 up_read(&sdp->sd_log_flush_lock);
331 static u64 log_bmap(struct gfs2_sbd *sdp, unsigned int lbn)
333 struct inode *inode = sdp->sd_jdesc->jd_inode;
335 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
337 bh_map.b_size = 1 << inode->i_blkbits;
338 error = gfs2_block_map(inode, lbn, 0, &bh_map);
339 if (error || !bh_map.b_blocknr)
340 printk(KERN_INFO "error=%d, dbn=%llu lbn=%u", error,
341 (unsigned long long)bh_map.b_blocknr, lbn);
342 gfs2_assert_withdraw(sdp, !error && bh_map.b_blocknr);
344 return bh_map.b_blocknr;
348 * log_distance - Compute distance between two journal blocks
349 * @sdp: The GFS2 superblock
350 * @newer: The most recent journal block of the pair
351 * @older: The older journal block of the pair
353 * Compute the distance (in the journal direction) between two
354 * blocks in the journal
356 * Returns: the distance in blocks
359 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
364 dist = newer - older;
366 dist += sdp->sd_jdesc->jd_blocks;
372 * calc_reserved - Calculate the number of blocks to reserve when
373 * refunding a transaction's unused buffers.
374 * @sdp: The GFS2 superblock
376 * This is complex. We need to reserve room for all our currently used
377 * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
378 * all our journaled data buffers for journaled files (e.g. files in the
379 * meta_fs like rindex, or files for which chattr +j was done.)
380 * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
381 * will count it as free space (sd_log_blks_free) and corruption will follow.
383 * We can have metadata bufs and jdata bufs in the same journal. So each
384 * type gets its own log header, for which we need to reserve a block.
385 * In fact, each type has the potential for needing more than one header
386 * in cases where we have more buffers than will fit on a journal page.
387 * Metadata journal entries take up half the space of journaled buffer entries.
388 * Thus, metadata entries have buf_limit (502) and journaled buffers have
389 * databuf_limit (251) before they cause a wrap around.
391 * Also, we need to reserve blocks for revoke journal entries and one for an
392 * overall header for the lot.
394 * Returns: the number of blocks reserved
396 static unsigned int calc_reserved(struct gfs2_sbd *sdp)
398 unsigned int reserved = 0;
399 unsigned int mbuf_limit, metabufhdrs_needed;
400 unsigned int dbuf_limit, databufhdrs_needed;
401 unsigned int revokes = 0;
403 mbuf_limit = buf_limit(sdp);
404 metabufhdrs_needed = (sdp->sd_log_commited_buf +
405 (mbuf_limit - 1)) / mbuf_limit;
406 dbuf_limit = databuf_limit(sdp);
407 databufhdrs_needed = (sdp->sd_log_commited_databuf +
408 (dbuf_limit - 1)) / dbuf_limit;
410 if (sdp->sd_log_commited_revoke)
411 revokes = gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
414 reserved = sdp->sd_log_commited_buf + metabufhdrs_needed +
415 sdp->sd_log_commited_databuf + databufhdrs_needed +
417 /* One for the overall header */
423 static unsigned int current_tail(struct gfs2_sbd *sdp)
430 if (list_empty(&sdp->sd_ail1_list)) {
431 tail = sdp->sd_log_head;
433 ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail, ai_list);
437 gfs2_log_unlock(sdp);
442 static inline void log_incr_head(struct gfs2_sbd *sdp)
444 if (sdp->sd_log_flush_head == sdp->sd_log_tail)
445 gfs2_assert_withdraw(sdp, sdp->sd_log_flush_head == sdp->sd_log_head);
447 if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
448 sdp->sd_log_flush_head = 0;
449 sdp->sd_log_flush_wrapped = 1;
454 * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
455 * @sdp: The GFS2 superblock
457 * Returns: the buffer_head
460 struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
462 u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
463 struct gfs2_log_buf *lb;
464 struct buffer_head *bh;
466 lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
467 list_add(&lb->lb_list, &sdp->sd_log_flush_list);
469 bh = lb->lb_bh = sb_getblk(sdp->sd_vfs, blkno);
471 memset(bh->b_data, 0, bh->b_size);
472 set_buffer_uptodate(bh);
473 clear_buffer_dirty(bh);
482 * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
483 * @sdp: the filesystem
484 * @data: the data the buffer_head should point to
486 * Returns: the log buffer descriptor
489 struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
490 struct buffer_head *real)
492 u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
493 struct gfs2_log_buf *lb;
494 struct buffer_head *bh;
496 lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
497 list_add(&lb->lb_list, &sdp->sd_log_flush_list);
500 bh = lb->lb_bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL);
501 atomic_set(&bh->b_count, 1);
502 bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate);
503 set_bh_page(bh, real->b_page, bh_offset(real));
504 bh->b_blocknr = blkno;
505 bh->b_size = sdp->sd_sb.sb_bsize;
506 bh->b_bdev = sdp->sd_vfs->s_bdev;
513 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
515 unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
517 ail2_empty(sdp, new_tail);
520 sdp->sd_log_blks_free += dist;
521 gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
522 gfs2_log_unlock(sdp);
524 sdp->sd_log_tail = new_tail;
528 * log_write_header - Get and initialize a journal header buffer
529 * @sdp: The GFS2 superblock
531 * Returns: the initialized log buffer descriptor
534 static void log_write_header(struct gfs2_sbd *sdp, u32 flags, int pull)
536 u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
537 struct buffer_head *bh;
538 struct gfs2_log_header *lh;
542 bh = sb_getblk(sdp->sd_vfs, blkno);
544 memset(bh->b_data, 0, bh->b_size);
545 set_buffer_uptodate(bh);
546 clear_buffer_dirty(bh);
549 gfs2_ail1_empty(sdp, 0);
550 tail = current_tail(sdp);
552 lh = (struct gfs2_log_header *)bh->b_data;
553 memset(lh, 0, sizeof(struct gfs2_log_header));
554 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
555 lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
556 lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
557 lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++);
558 lh->lh_flags = cpu_to_be32(flags);
559 lh->lh_tail = cpu_to_be32(tail);
560 lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
561 hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
562 lh->lh_hash = cpu_to_be32(hash);
564 set_buffer_dirty(bh);
565 if (sync_dirty_buffer(bh))
566 gfs2_io_error_bh(sdp, bh);
569 if (sdp->sd_log_tail != tail)
570 log_pull_tail(sdp, tail);
572 gfs2_assert_withdraw(sdp, !pull);
574 sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
578 static void log_flush_commit(struct gfs2_sbd *sdp)
580 struct list_head *head = &sdp->sd_log_flush_list;
581 struct gfs2_log_buf *lb;
582 struct buffer_head *bh;
585 while (!list_empty(head)) {
586 lb = list_entry(head->next, struct gfs2_log_buf, lb_list);
587 list_del(&lb->lb_list);
591 if (!buffer_uptodate(bh))
592 gfs2_io_error_bh(sdp, bh);
594 while (atomic_read(&bh->b_count) != 1) /* Grrrr... */
596 free_buffer_head(bh);
603 /* If nothing was journaled, the header is unplanned and unwanted. */
605 log_write_header(sdp, 0, 0);
608 tail = current_tail(sdp);
610 gfs2_ail1_empty(sdp, 0);
611 if (sdp->sd_log_tail != tail)
612 log_pull_tail(sdp, tail);
617 * gfs2_log_flush - flush incore transaction(s)
618 * @sdp: the filesystem
619 * @gl: The glock structure to flush. If NULL, flush the whole incore log
623 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
627 down_write(&sdp->sd_log_flush_lock);
631 if (list_empty(&gl->gl_le.le_list)) {
632 gfs2_log_unlock(sdp);
633 up_write(&sdp->sd_log_flush_lock);
636 gfs2_log_unlock(sdp);
639 ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
640 INIT_LIST_HEAD(&ai->ai_ail1_list);
641 INIT_LIST_HEAD(&ai->ai_ail2_list);
643 gfs2_assert_withdraw(sdp,
644 sdp->sd_log_num_buf + sdp->sd_log_num_jdata ==
645 sdp->sd_log_commited_buf +
646 sdp->sd_log_commited_databuf);
647 gfs2_assert_withdraw(sdp,
648 sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
650 sdp->sd_log_flush_head = sdp->sd_log_head;
651 sdp->sd_log_flush_wrapped = 0;
652 ai->ai_first = sdp->sd_log_flush_head;
654 lops_before_commit(sdp);
655 if (!list_empty(&sdp->sd_log_flush_list))
656 log_flush_commit(sdp);
657 else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
659 sdp->sd_log_blks_free--; /* Adjust for unreserved buffer */
660 gfs2_log_unlock(sdp);
661 log_write_header(sdp, 0, PULL);
663 lops_after_commit(sdp, ai);
666 sdp->sd_log_head = sdp->sd_log_flush_head;
667 sdp->sd_log_blks_reserved = 0;
668 sdp->sd_log_commited_buf = 0;
669 sdp->sd_log_commited_databuf = 0;
670 sdp->sd_log_commited_revoke = 0;
672 if (!list_empty(&ai->ai_ail1_list)) {
673 list_add(&ai->ai_list, &sdp->sd_ail1_list);
676 gfs2_log_unlock(sdp);
678 sdp->sd_vfs->s_dirt = 0;
679 up_write(&sdp->sd_log_flush_lock);
684 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
686 unsigned int reserved;
691 sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
692 sdp->sd_log_commited_databuf += tr->tr_num_databuf_new -
693 tr->tr_num_databuf_rm;
694 gfs2_assert_withdraw(sdp, (((int)sdp->sd_log_commited_buf) >= 0) ||
695 (((int)sdp->sd_log_commited_databuf) >= 0));
696 sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
697 gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0);
698 reserved = calc_reserved(sdp);
699 old = sdp->sd_log_blks_free;
700 sdp->sd_log_blks_free += tr->tr_reserved -
701 (reserved - sdp->sd_log_blks_reserved);
703 gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free >= old);
704 gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free <=
705 sdp->sd_jdesc->jd_blocks);
707 sdp->sd_log_blks_reserved = reserved;
709 gfs2_log_unlock(sdp);
713 * gfs2_log_commit - Commit a transaction to the log
714 * @sdp: the filesystem
715 * @tr: the transaction
720 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
723 lops_incore_commit(sdp, tr);
725 sdp->sd_vfs->s_dirt = 1;
726 up_read(&sdp->sd_log_flush_lock);
729 if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks))
730 wake_up_process(sdp->sd_logd_process);
731 gfs2_log_unlock(sdp);
735 * gfs2_log_shutdown - write a shutdown header into a journal
736 * @sdp: the filesystem
740 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
742 down_write(&sdp->sd_log_flush_lock);
744 gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
745 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_gl);
746 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
747 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_jdata);
748 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
749 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
750 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
751 gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
753 sdp->sd_log_flush_head = sdp->sd_log_head;
754 sdp->sd_log_flush_wrapped = 0;
756 log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT,
757 (sdp->sd_log_tail == current_tail(sdp)) ? 0 : PULL);
759 gfs2_assert_warn(sdp, sdp->sd_log_blks_free == sdp->sd_jdesc->jd_blocks);
760 gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
761 gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
763 sdp->sd_log_head = sdp->sd_log_flush_head;
764 sdp->sd_log_tail = sdp->sd_log_head;
766 up_write(&sdp->sd_log_flush_lock);
771 * gfs2_meta_syncfs - sync all the buffers in a filesystem
772 * @sdp: the filesystem
776 void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
778 gfs2_log_flush(sdp, NULL);
780 gfs2_ail1_start(sdp, DIO_ALL);
781 if (gfs2_ail1_empty(sdp, DIO_ALL))