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>
32 * gfs2_struct2blk - compute stuff
33 * @sdp: the filesystem
34 * @nstruct: the number of structures
35 * @ssize: the size of the structures
37 * Compute the number of log descriptor blocks needed to hold a certain number
38 * of structures of a certain size.
40 * Returns: the number of blocks needed (minimum is always 1)
43 unsigned int gfs2_struct2blk(struct gfs2_sbd
*sdp
, unsigned int nstruct
,
47 unsigned int first
, second
;
50 first
= (sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_log_descriptor
)) / ssize
;
52 if (nstruct
> first
) {
53 second
= (sdp
->sd_sb
.sb_bsize
-
54 sizeof(struct gfs2_meta_header
)) / ssize
;
55 blks
+= DIV_ROUND_UP(nstruct
- first
, second
);
62 * gfs2_ail1_start_one - Start I/O on a part of the AIL
63 * @sdp: the filesystem
64 * @tr: the part of the AIL
68 static void gfs2_ail1_start_one(struct gfs2_sbd
*sdp
, struct gfs2_ail
*ai
)
70 struct gfs2_bufdata
*bd
, *s
;
71 struct buffer_head
*bh
;
74 BUG_ON(!spin_is_locked(&sdp
->sd_log_lock
));
79 list_for_each_entry_safe_reverse(bd
, s
, &ai
->ai_ail1_list
,
83 gfs2_assert(sdp
, bd
->bd_ail
== ai
);
85 if (!buffer_busy(bh
)) {
86 if (!buffer_uptodate(bh
)) {
88 gfs2_io_error_bh(sdp
, bh
);
91 list_move(&bd
->bd_ail_st_list
, &ai
->ai_ail2_list
);
95 if (!buffer_dirty(bh
))
98 list_move(&bd
->bd_ail_st_list
, &ai
->ai_ail1_list
);
100 gfs2_log_unlock(sdp
);
102 ll_rw_block(WRITE
, 1, &bh
);
112 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
113 * @sdp: the filesystem
118 static int gfs2_ail1_empty_one(struct gfs2_sbd
*sdp
, struct gfs2_ail
*ai
, int flags
)
120 struct gfs2_bufdata
*bd
, *s
;
121 struct buffer_head
*bh
;
123 list_for_each_entry_safe_reverse(bd
, s
, &ai
->ai_ail1_list
,
127 gfs2_assert(sdp
, bd
->bd_ail
== ai
);
129 if (buffer_busy(bh
)) {
136 if (!buffer_uptodate(bh
))
137 gfs2_io_error_bh(sdp
, bh
);
139 list_move(&bd
->bd_ail_st_list
, &ai
->ai_ail2_list
);
142 return list_empty(&ai
->ai_ail1_list
);
145 void gfs2_ail1_start(struct gfs2_sbd
*sdp
, int flags
)
147 struct list_head
*head
= &sdp
->sd_ail1_list
;
149 struct list_head
*first
;
150 struct gfs2_ail
*first_ai
, *ai
, *tmp
;
154 if (list_empty(head
)) {
155 gfs2_log_unlock(sdp
);
158 sync_gen
= sdp
->sd_ail_sync_gen
++;
161 first_ai
= list_entry(first
, struct gfs2_ail
, ai_list
);
162 first_ai
->ai_sync_gen
= sync_gen
;
163 gfs2_ail1_start_one(sdp
, first_ai
); /* This may drop log lock */
169 if (first
&& (head
->prev
!= first
||
170 gfs2_ail1_empty_one(sdp
, first_ai
, 0)))
174 list_for_each_entry_safe_reverse(ai
, tmp
, head
, ai_list
) {
175 if (ai
->ai_sync_gen
>= sync_gen
)
177 ai
->ai_sync_gen
= sync_gen
;
178 gfs2_ail1_start_one(sdp
, ai
); /* This may drop log lock */
184 gfs2_log_unlock(sdp
);
187 int gfs2_ail1_empty(struct gfs2_sbd
*sdp
, int flags
)
189 struct gfs2_ail
*ai
, *s
;
194 list_for_each_entry_safe_reverse(ai
, s
, &sdp
->sd_ail1_list
, ai_list
) {
195 if (gfs2_ail1_empty_one(sdp
, ai
, flags
))
196 list_move(&ai
->ai_list
, &sdp
->sd_ail2_list
);
197 else if (!(flags
& DIO_ALL
))
201 ret
= list_empty(&sdp
->sd_ail1_list
);
203 gfs2_log_unlock(sdp
);
210 * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
211 * @sdp: the filesystem
216 static void gfs2_ail2_empty_one(struct gfs2_sbd
*sdp
, struct gfs2_ail
*ai
)
218 struct list_head
*head
= &ai
->ai_ail2_list
;
219 struct gfs2_bufdata
*bd
;
221 while (!list_empty(head
)) {
222 bd
= list_entry(head
->prev
, struct gfs2_bufdata
,
224 gfs2_assert(sdp
, bd
->bd_ail
== ai
);
226 list_del(&bd
->bd_ail_st_list
);
227 list_del(&bd
->bd_ail_gl_list
);
228 atomic_dec(&bd
->bd_gl
->gl_ail_count
);
233 static void ail2_empty(struct gfs2_sbd
*sdp
, unsigned int new_tail
)
235 struct gfs2_ail
*ai
, *safe
;
236 unsigned int old_tail
= sdp
->sd_log_tail
;
237 int wrap
= (new_tail
< old_tail
);
242 list_for_each_entry_safe(ai
, safe
, &sdp
->sd_ail2_list
, ai_list
) {
243 a
= (old_tail
<= ai
->ai_first
);
244 b
= (ai
->ai_first
< new_tail
);
245 rm
= (wrap
) ? (a
|| b
) : (a
&& b
);
249 gfs2_ail2_empty_one(sdp
, ai
);
250 list_del(&ai
->ai_list
);
251 gfs2_assert_warn(sdp
, list_empty(&ai
->ai_ail1_list
));
252 gfs2_assert_warn(sdp
, list_empty(&ai
->ai_ail2_list
));
256 gfs2_log_unlock(sdp
);
260 * gfs2_log_reserve - Make a log reservation
261 * @sdp: The GFS2 superblock
262 * @blks: The number of blocks to reserve
267 int gfs2_log_reserve(struct gfs2_sbd
*sdp
, unsigned int blks
)
269 unsigned int try = 0;
271 if (gfs2_assert_warn(sdp
, blks
) ||
272 gfs2_assert_warn(sdp
, blks
<= sdp
->sd_jdesc
->jd_blocks
))
275 mutex_lock(&sdp
->sd_log_reserve_mutex
);
277 while(sdp
->sd_log_blks_free
<= blks
) {
278 gfs2_log_unlock(sdp
);
279 gfs2_ail1_empty(sdp
, 0);
280 gfs2_log_flush(sdp
, NULL
);
283 gfs2_ail1_start(sdp
, 0);
286 sdp
->sd_log_blks_free
-= blks
;
287 gfs2_log_unlock(sdp
);
288 mutex_unlock(&sdp
->sd_log_reserve_mutex
);
290 down_read(&sdp
->sd_log_flush_lock
);
296 * gfs2_log_release - Release a given number of log blocks
297 * @sdp: The GFS2 superblock
298 * @blks: The number of blocks
302 void gfs2_log_release(struct gfs2_sbd
*sdp
, unsigned int blks
)
306 sdp
->sd_log_blks_free
+= blks
;
307 gfs2_assert_withdraw(sdp
,
308 sdp
->sd_log_blks_free
<= sdp
->sd_jdesc
->jd_blocks
);
309 gfs2_log_unlock(sdp
);
310 up_read(&sdp
->sd_log_flush_lock
);
313 static u64
log_bmap(struct gfs2_sbd
*sdp
, unsigned int lbn
)
316 struct buffer_head bh_map
;
318 error
= gfs2_block_map(sdp
->sd_jdesc
->jd_inode
, lbn
, 0, &bh_map
, 1);
319 if (error
|| !bh_map
.b_blocknr
)
320 printk(KERN_INFO
"error=%d, dbn=%llu lbn=%u", error
, bh_map
.b_blocknr
, lbn
);
321 gfs2_assert_withdraw(sdp
, !error
&& bh_map
.b_blocknr
);
323 return bh_map
.b_blocknr
;
327 * log_distance - Compute distance between two journal blocks
328 * @sdp: The GFS2 superblock
329 * @newer: The most recent journal block of the pair
330 * @older: The older journal block of the pair
332 * Compute the distance (in the journal direction) between two
333 * blocks in the journal
335 * Returns: the distance in blocks
338 static inline unsigned int log_distance(struct gfs2_sbd
*sdp
, unsigned int newer
,
343 dist
= newer
- older
;
345 dist
+= sdp
->sd_jdesc
->jd_blocks
;
350 static unsigned int current_tail(struct gfs2_sbd
*sdp
)
357 if (list_empty(&sdp
->sd_ail1_list
)) {
358 tail
= sdp
->sd_log_head
;
360 ai
= list_entry(sdp
->sd_ail1_list
.prev
, struct gfs2_ail
, ai_list
);
364 gfs2_log_unlock(sdp
);
369 static inline void log_incr_head(struct gfs2_sbd
*sdp
)
371 if (sdp
->sd_log_flush_head
== sdp
->sd_log_tail
)
372 gfs2_assert_withdraw(sdp
, sdp
->sd_log_flush_head
== sdp
->sd_log_head
);
374 if (++sdp
->sd_log_flush_head
== sdp
->sd_jdesc
->jd_blocks
) {
375 sdp
->sd_log_flush_head
= 0;
376 sdp
->sd_log_flush_wrapped
= 1;
381 * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
382 * @sdp: The GFS2 superblock
384 * Returns: the buffer_head
387 struct buffer_head
*gfs2_log_get_buf(struct gfs2_sbd
*sdp
)
389 u64 blkno
= log_bmap(sdp
, sdp
->sd_log_flush_head
);
390 struct gfs2_log_buf
*lb
;
391 struct buffer_head
*bh
;
393 lb
= kzalloc(sizeof(struct gfs2_log_buf
), GFP_NOFS
| __GFP_NOFAIL
);
394 list_add(&lb
->lb_list
, &sdp
->sd_log_flush_list
);
396 bh
= lb
->lb_bh
= sb_getblk(sdp
->sd_vfs
, blkno
);
398 memset(bh
->b_data
, 0, bh
->b_size
);
399 set_buffer_uptodate(bh
);
400 clear_buffer_dirty(bh
);
409 * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
410 * @sdp: the filesystem
411 * @data: the data the buffer_head should point to
413 * Returns: the log buffer descriptor
416 struct buffer_head
*gfs2_log_fake_buf(struct gfs2_sbd
*sdp
,
417 struct buffer_head
*real
)
419 u64 blkno
= log_bmap(sdp
, sdp
->sd_log_flush_head
);
420 struct gfs2_log_buf
*lb
;
421 struct buffer_head
*bh
;
423 lb
= kzalloc(sizeof(struct gfs2_log_buf
), GFP_NOFS
| __GFP_NOFAIL
);
424 list_add(&lb
->lb_list
, &sdp
->sd_log_flush_list
);
427 bh
= lb
->lb_bh
= alloc_buffer_head(GFP_NOFS
| __GFP_NOFAIL
);
428 atomic_set(&bh
->b_count
, 1);
429 bh
->b_state
= (1 << BH_Mapped
) | (1 << BH_Uptodate
);
430 set_bh_page(bh
, real
->b_page
, bh_offset(real
));
431 bh
->b_blocknr
= blkno
;
432 bh
->b_size
= sdp
->sd_sb
.sb_bsize
;
433 bh
->b_bdev
= sdp
->sd_vfs
->s_bdev
;
440 static void log_pull_tail(struct gfs2_sbd
*sdp
, unsigned int new_tail
, int pull
)
442 unsigned int dist
= log_distance(sdp
, new_tail
, sdp
->sd_log_tail
);
444 ail2_empty(sdp
, new_tail
);
447 sdp
->sd_log_blks_free
+= dist
- (pull
? 1 : 0);
448 gfs2_assert_withdraw(sdp
, sdp
->sd_log_blks_free
<= sdp
->sd_jdesc
->jd_blocks
);
449 gfs2_log_unlock(sdp
);
451 sdp
->sd_log_tail
= new_tail
;
455 * log_write_header - Get and initialize a journal header buffer
456 * @sdp: The GFS2 superblock
458 * Returns: the initialized log buffer descriptor
461 static void log_write_header(struct gfs2_sbd
*sdp
, u32 flags
, int pull
)
463 u64 blkno
= log_bmap(sdp
, sdp
->sd_log_flush_head
);
464 struct buffer_head
*bh
;
465 struct gfs2_log_header
*lh
;
469 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
);
476 gfs2_ail1_empty(sdp
, 0);
477 tail
= current_tail(sdp
);
479 lh
= (struct gfs2_log_header
*)bh
->b_data
;
480 memset(lh
, 0, sizeof(struct gfs2_log_header
));
481 lh
->lh_header
.mh_magic
= cpu_to_be32(GFS2_MAGIC
);
482 lh
->lh_header
.mh_type
= cpu_to_be32(GFS2_METATYPE_LH
);
483 lh
->lh_header
.mh_format
= cpu_to_be32(GFS2_FORMAT_LH
);
484 lh
->lh_sequence
= cpu_to_be64(sdp
->sd_log_sequence
++);
485 lh
->lh_flags
= cpu_to_be32(flags
);
486 lh
->lh_tail
= cpu_to_be32(tail
);
487 lh
->lh_blkno
= cpu_to_be32(sdp
->sd_log_flush_head
);
488 hash
= gfs2_disk_hash(bh
->b_data
, sizeof(struct gfs2_log_header
));
489 lh
->lh_hash
= cpu_to_be32(hash
);
491 set_buffer_dirty(bh
);
492 if (sync_dirty_buffer(bh
))
493 gfs2_io_error_bh(sdp
, bh
);
496 if (sdp
->sd_log_tail
!= tail
)
497 log_pull_tail(sdp
, tail
, pull
);
499 gfs2_assert_withdraw(sdp
, !pull
);
501 sdp
->sd_log_idle
= (tail
== sdp
->sd_log_flush_head
);
505 static void log_flush_commit(struct gfs2_sbd
*sdp
)
507 struct list_head
*head
= &sdp
->sd_log_flush_list
;
508 struct gfs2_log_buf
*lb
;
509 struct buffer_head
*bh
;
511 while (!list_empty(head
)) {
512 lb
= list_entry(head
->next
, struct gfs2_log_buf
, lb_list
);
513 list_del(&lb
->lb_list
);
517 if (!buffer_uptodate(bh
))
518 gfs2_io_error_bh(sdp
, bh
);
520 while (atomic_read(&bh
->b_count
) != 1) /* Grrrr... */
522 free_buffer_head(bh
);
528 log_write_header(sdp
, 0, 0);
532 * gfs2_log_flush - flush incore transaction(s)
533 * @sdp: the filesystem
534 * @gl: The glock structure to flush. If NULL, flush the whole incore log
538 void gfs2_log_flush(struct gfs2_sbd
*sdp
, struct gfs2_glock
*gl
)
542 down_write(&sdp
->sd_log_flush_lock
);
546 if (list_empty(&gl
->gl_le
.le_list
)) {
547 gfs2_log_unlock(sdp
);
548 up_write(&sdp
->sd_log_flush_lock
);
551 gfs2_log_unlock(sdp
);
554 ai
= kzalloc(sizeof(struct gfs2_ail
), GFP_NOFS
| __GFP_NOFAIL
);
555 INIT_LIST_HEAD(&ai
->ai_ail1_list
);
556 INIT_LIST_HEAD(&ai
->ai_ail2_list
);
558 gfs2_assert_withdraw(sdp
, sdp
->sd_log_num_buf
== sdp
->sd_log_commited_buf
);
559 gfs2_assert_withdraw(sdp
,
560 sdp
->sd_log_num_revoke
== sdp
->sd_log_commited_revoke
);
562 sdp
->sd_log_flush_head
= sdp
->sd_log_head
;
563 sdp
->sd_log_flush_wrapped
= 0;
564 ai
->ai_first
= sdp
->sd_log_flush_head
;
566 lops_before_commit(sdp
);
567 if (!list_empty(&sdp
->sd_log_flush_list
))
568 log_flush_commit(sdp
);
569 else if (sdp
->sd_log_tail
!= current_tail(sdp
) && !sdp
->sd_log_idle
)
570 log_write_header(sdp
, 0, PULL
);
571 lops_after_commit(sdp
, ai
);
572 sdp
->sd_log_head
= sdp
->sd_log_flush_head
;
574 sdp
->sd_log_blks_free
-= sdp
->sd_log_num_hdrs
;
576 sdp
->sd_log_blks_reserved
= 0;
577 sdp
->sd_log_commited_buf
= 0;
578 sdp
->sd_log_num_hdrs
= 0;
579 sdp
->sd_log_commited_revoke
= 0;
582 if (!list_empty(&ai
->ai_ail1_list
)) {
583 list_add(&ai
->ai_list
, &sdp
->sd_ail1_list
);
586 gfs2_log_unlock(sdp
);
588 sdp
->sd_vfs
->s_dirt
= 0;
589 up_write(&sdp
->sd_log_flush_lock
);
594 static void log_refund(struct gfs2_sbd
*sdp
, struct gfs2_trans
*tr
)
596 unsigned int reserved
= 0;
601 sdp
->sd_log_commited_buf
+= tr
->tr_num_buf_new
- tr
->tr_num_buf_rm
;
602 gfs2_assert_withdraw(sdp
, ((int)sdp
->sd_log_commited_buf
) >= 0);
603 sdp
->sd_log_commited_revoke
+= tr
->tr_num_revoke
- tr
->tr_num_revoke_rm
;
604 gfs2_assert_withdraw(sdp
, ((int)sdp
->sd_log_commited_revoke
) >= 0);
606 if (sdp
->sd_log_commited_buf
)
607 reserved
+= sdp
->sd_log_commited_buf
;
608 if (sdp
->sd_log_commited_revoke
)
609 reserved
+= gfs2_struct2blk(sdp
, sdp
->sd_log_commited_revoke
,
614 old
= sdp
->sd_log_blks_free
;
615 sdp
->sd_log_blks_free
+= tr
->tr_reserved
-
616 (reserved
- sdp
->sd_log_blks_reserved
);
618 gfs2_assert_withdraw(sdp
, sdp
->sd_log_blks_free
>= old
);
619 gfs2_assert_withdraw(sdp
,
620 sdp
->sd_log_blks_free
<= sdp
->sd_jdesc
->jd_blocks
+
621 sdp
->sd_log_num_hdrs
);
623 sdp
->sd_log_blks_reserved
= reserved
;
625 gfs2_log_unlock(sdp
);
629 * gfs2_log_commit - Commit a transaction to the log
630 * @sdp: the filesystem
631 * @tr: the transaction
636 void gfs2_log_commit(struct gfs2_sbd
*sdp
, struct gfs2_trans
*tr
)
639 lops_incore_commit(sdp
, tr
);
641 sdp
->sd_vfs
->s_dirt
= 1;
642 up_read(&sdp
->sd_log_flush_lock
);
645 if (sdp
->sd_log_num_buf
> gfs2_tune_get(sdp
, gt_incore_log_blocks
)) {
646 gfs2_log_unlock(sdp
);
647 gfs2_log_flush(sdp
, NULL
);
649 gfs2_log_unlock(sdp
);
654 * gfs2_log_shutdown - write a shutdown header into a journal
655 * @sdp: the filesystem
659 void gfs2_log_shutdown(struct gfs2_sbd
*sdp
)
661 down_write(&sdp
->sd_log_flush_lock
);
663 gfs2_assert_withdraw(sdp
, !sdp
->sd_log_blks_reserved
);
664 gfs2_assert_withdraw(sdp
, !sdp
->sd_log_num_gl
);
665 gfs2_assert_withdraw(sdp
, !sdp
->sd_log_num_buf
);
666 gfs2_assert_withdraw(sdp
, !sdp
->sd_log_num_jdata
);
667 gfs2_assert_withdraw(sdp
, !sdp
->sd_log_num_revoke
);
668 gfs2_assert_withdraw(sdp
, !sdp
->sd_log_num_rg
);
669 gfs2_assert_withdraw(sdp
, !sdp
->sd_log_num_databuf
);
670 gfs2_assert_withdraw(sdp
, !sdp
->sd_log_num_hdrs
);
671 gfs2_assert_withdraw(sdp
, list_empty(&sdp
->sd_ail1_list
));
673 sdp
->sd_log_flush_head
= sdp
->sd_log_head
;
674 sdp
->sd_log_flush_wrapped
= 0;
676 log_write_header(sdp
, GFS2_LOG_HEAD_UNMOUNT
, 0);
678 gfs2_assert_warn(sdp
, sdp
->sd_log_blks_free
== sdp
->sd_jdesc
->jd_blocks
);
679 gfs2_assert_warn(sdp
, sdp
->sd_log_head
== sdp
->sd_log_tail
);
680 gfs2_assert_warn(sdp
, list_empty(&sdp
->sd_ail2_list
));
682 sdp
->sd_log_head
= sdp
->sd_log_flush_head
;
683 sdp
->sd_log_tail
= sdp
->sd_log_head
;
685 up_write(&sdp
->sd_log_flush_lock
);