[NETFILTER]: nf_conntrack: properly use RCU for nf_conntrack_destroyed callback
[linux-2.6.22.y-op.git] / fs / gfs2 / log.c
blob291415ddfe51cb291ccd66b23703e63a58b089d7
1 /*
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.
8 */
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>
20 #include "gfs2.h"
21 #include "incore.h"
22 #include "bmap.h"
23 #include "glock.h"
24 #include "log.h"
25 #include "lops.h"
26 #include "meta_io.h"
27 #include "util.h"
28 #include "dir.h"
30 #define PULL 1
32 /**
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,
45 unsigned int ssize)
47 unsigned int blks;
48 unsigned int first, second;
50 blks = 1;
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);
59 return blks;
62 /**
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;
73 int retry;
75 BUG_ON(!spin_is_locked(&sdp->sd_log_lock));
77 do {
78 retry = 0;
80 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
81 bd_ail_st_list) {
82 bh = bd->bd_bh;
84 gfs2_assert(sdp, bd->bd_ail == ai);
86 if (!buffer_busy(bh)) {
87 if (!buffer_uptodate(bh)) {
88 gfs2_log_unlock(sdp);
89 gfs2_io_error_bh(sdp, bh);
90 gfs2_log_lock(sdp);
92 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
93 continue;
96 if (!buffer_dirty(bh))
97 continue;
99 list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
101 gfs2_log_unlock(sdp);
102 wait_on_buffer(bh);
103 ll_rw_block(WRITE, 1, &bh);
104 gfs2_log_lock(sdp);
106 retry = 1;
107 break;
109 } while (retry);
113 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
114 * @sdp: the filesystem
115 * @ai: the AIL entry
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,
125 bd_ail_st_list) {
126 bh = bd->bd_bh;
128 gfs2_assert(sdp, bd->bd_ail == ai);
130 if (buffer_busy(bh)) {
131 if (flags & DIO_ALL)
132 continue;
133 else
134 break;
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;
149 u64 sync_gen;
150 struct list_head *first;
151 struct gfs2_ail *first_ai, *ai, *tmp;
152 int done = 0;
154 gfs2_log_lock(sdp);
155 if (list_empty(head)) {
156 gfs2_log_unlock(sdp);
157 return;
159 sync_gen = sdp->sd_ail_sync_gen++;
161 first = head->prev;
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 */
166 if (flags & DIO_ALL)
167 first = NULL;
169 while(!done) {
170 if (first && (head->prev != first ||
171 gfs2_ail1_empty_one(sdp, first_ai, 0)))
172 break;
174 done = 1;
175 list_for_each_entry_safe_reverse(ai, tmp, head, ai_list) {
176 if (ai->ai_sync_gen >= sync_gen)
177 continue;
178 ai->ai_sync_gen = sync_gen;
179 gfs2_ail1_start_one(sdp, ai); /* This may drop log lock */
180 done = 0;
181 break;
185 gfs2_log_unlock(sdp);
188 int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
190 struct gfs2_ail *ai, *s;
191 int ret;
193 gfs2_log_lock(sdp);
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))
199 break;
202 ret = list_empty(&sdp->sd_ail1_list);
204 gfs2_log_unlock(sdp);
206 return ret;
211 * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
212 * @sdp: the filesystem
213 * @ai: the AIL entry
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,
224 bd_ail_st_list);
225 gfs2_assert(sdp, bd->bd_ail == ai);
226 bd->bd_ail = NULL;
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);
230 brelse(bd->bd_bh);
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);
239 int a, b, rm;
241 gfs2_log_lock(sdp);
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);
247 if (!rm)
248 continue;
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));
254 kfree(ai);
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 6 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.
271 * Returns: errno
274 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
276 unsigned int try = 0;
278 if (gfs2_assert_warn(sdp, blks) ||
279 gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
280 return -EINVAL;
282 mutex_lock(&sdp->sd_log_reserve_mutex);
283 gfs2_log_lock(sdp);
284 while(sdp->sd_log_blks_free <= (blks + 6)) {
285 gfs2_log_unlock(sdp);
286 gfs2_ail1_empty(sdp, 0);
287 gfs2_log_flush(sdp, NULL);
289 if (try++)
290 gfs2_ail1_start(sdp, 0);
291 gfs2_log_lock(sdp);
293 sdp->sd_log_blks_free -= blks;
294 gfs2_log_unlock(sdp);
295 mutex_unlock(&sdp->sd_log_reserve_mutex);
297 down_read(&sdp->sd_log_flush_lock);
299 return 0;
303 * gfs2_log_release - Release a given number of log blocks
304 * @sdp: The GFS2 superblock
305 * @blks: The number of blocks
309 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
312 gfs2_log_lock(sdp);
313 sdp->sd_log_blks_free += blks;
314 gfs2_assert_withdraw(sdp,
315 sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
316 gfs2_log_unlock(sdp);
317 up_read(&sdp->sd_log_flush_lock);
320 static u64 log_bmap(struct gfs2_sbd *sdp, unsigned int lbn)
322 struct inode *inode = sdp->sd_jdesc->jd_inode;
323 int error;
324 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
326 bh_map.b_size = 1 << inode->i_blkbits;
327 error = gfs2_block_map(inode, lbn, 0, &bh_map);
328 if (error || !bh_map.b_blocknr)
329 printk(KERN_INFO "error=%d, dbn=%llu lbn=%u", error,
330 (unsigned long long)bh_map.b_blocknr, lbn);
331 gfs2_assert_withdraw(sdp, !error && bh_map.b_blocknr);
333 return bh_map.b_blocknr;
337 * log_distance - Compute distance between two journal blocks
338 * @sdp: The GFS2 superblock
339 * @newer: The most recent journal block of the pair
340 * @older: The older journal block of the pair
342 * Compute the distance (in the journal direction) between two
343 * blocks in the journal
345 * Returns: the distance in blocks
348 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
349 unsigned int older)
351 int dist;
353 dist = newer - older;
354 if (dist < 0)
355 dist += sdp->sd_jdesc->jd_blocks;
357 return dist;
360 static unsigned int current_tail(struct gfs2_sbd *sdp)
362 struct gfs2_ail *ai;
363 unsigned int tail;
365 gfs2_log_lock(sdp);
367 if (list_empty(&sdp->sd_ail1_list)) {
368 tail = sdp->sd_log_head;
369 } else {
370 ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail, ai_list);
371 tail = ai->ai_first;
374 gfs2_log_unlock(sdp);
376 return tail;
379 static inline void log_incr_head(struct gfs2_sbd *sdp)
381 if (sdp->sd_log_flush_head == sdp->sd_log_tail)
382 gfs2_assert_withdraw(sdp, sdp->sd_log_flush_head == sdp->sd_log_head);
384 if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
385 sdp->sd_log_flush_head = 0;
386 sdp->sd_log_flush_wrapped = 1;
391 * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
392 * @sdp: The GFS2 superblock
394 * Returns: the buffer_head
397 struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
399 u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
400 struct gfs2_log_buf *lb;
401 struct buffer_head *bh;
403 lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
404 list_add(&lb->lb_list, &sdp->sd_log_flush_list);
406 bh = lb->lb_bh = sb_getblk(sdp->sd_vfs, blkno);
407 lock_buffer(bh);
408 memset(bh->b_data, 0, bh->b_size);
409 set_buffer_uptodate(bh);
410 clear_buffer_dirty(bh);
411 unlock_buffer(bh);
413 log_incr_head(sdp);
415 return bh;
419 * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
420 * @sdp: the filesystem
421 * @data: the data the buffer_head should point to
423 * Returns: the log buffer descriptor
426 struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
427 struct buffer_head *real)
429 u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
430 struct gfs2_log_buf *lb;
431 struct buffer_head *bh;
433 lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
434 list_add(&lb->lb_list, &sdp->sd_log_flush_list);
435 lb->lb_real = real;
437 bh = lb->lb_bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL);
438 atomic_set(&bh->b_count, 1);
439 bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate);
440 set_bh_page(bh, real->b_page, bh_offset(real));
441 bh->b_blocknr = blkno;
442 bh->b_size = sdp->sd_sb.sb_bsize;
443 bh->b_bdev = sdp->sd_vfs->s_bdev;
445 log_incr_head(sdp);
447 return bh;
450 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail, int pull)
452 unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
454 ail2_empty(sdp, new_tail);
456 gfs2_log_lock(sdp);
457 sdp->sd_log_blks_free += dist - (pull ? 1 : 0);
458 gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
459 gfs2_log_unlock(sdp);
461 sdp->sd_log_tail = new_tail;
465 * log_write_header - Get and initialize a journal header buffer
466 * @sdp: The GFS2 superblock
468 * Returns: the initialized log buffer descriptor
471 static void log_write_header(struct gfs2_sbd *sdp, u32 flags, int pull)
473 u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
474 struct buffer_head *bh;
475 struct gfs2_log_header *lh;
476 unsigned int tail;
477 u32 hash;
479 bh = sb_getblk(sdp->sd_vfs, blkno);
480 lock_buffer(bh);
481 memset(bh->b_data, 0, bh->b_size);
482 set_buffer_uptodate(bh);
483 clear_buffer_dirty(bh);
484 unlock_buffer(bh);
486 gfs2_ail1_empty(sdp, 0);
487 tail = current_tail(sdp);
489 lh = (struct gfs2_log_header *)bh->b_data;
490 memset(lh, 0, sizeof(struct gfs2_log_header));
491 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
492 lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
493 lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
494 lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++);
495 lh->lh_flags = cpu_to_be32(flags);
496 lh->lh_tail = cpu_to_be32(tail);
497 lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
498 hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
499 lh->lh_hash = cpu_to_be32(hash);
501 set_buffer_dirty(bh);
502 if (sync_dirty_buffer(bh))
503 gfs2_io_error_bh(sdp, bh);
504 brelse(bh);
506 if (sdp->sd_log_tail != tail)
507 log_pull_tail(sdp, tail, pull);
508 else
509 gfs2_assert_withdraw(sdp, !pull);
511 sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
512 log_incr_head(sdp);
515 static void log_flush_commit(struct gfs2_sbd *sdp)
517 struct list_head *head = &sdp->sd_log_flush_list;
518 struct gfs2_log_buf *lb;
519 struct buffer_head *bh;
521 while (!list_empty(head)) {
522 lb = list_entry(head->next, struct gfs2_log_buf, lb_list);
523 list_del(&lb->lb_list);
524 bh = lb->lb_bh;
526 wait_on_buffer(bh);
527 if (!buffer_uptodate(bh))
528 gfs2_io_error_bh(sdp, bh);
529 if (lb->lb_real) {
530 while (atomic_read(&bh->b_count) != 1) /* Grrrr... */
531 schedule();
532 free_buffer_head(bh);
533 } else
534 brelse(bh);
535 kfree(lb);
538 log_write_header(sdp, 0, 0);
542 * gfs2_log_flush - flush incore transaction(s)
543 * @sdp: the filesystem
544 * @gl: The glock structure to flush. If NULL, flush the whole incore log
548 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
550 struct gfs2_ail *ai;
552 down_write(&sdp->sd_log_flush_lock);
554 if (gl) {
555 gfs2_log_lock(sdp);
556 if (list_empty(&gl->gl_le.le_list)) {
557 gfs2_log_unlock(sdp);
558 up_write(&sdp->sd_log_flush_lock);
559 return;
561 gfs2_log_unlock(sdp);
564 ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
565 INIT_LIST_HEAD(&ai->ai_ail1_list);
566 INIT_LIST_HEAD(&ai->ai_ail2_list);
568 gfs2_assert_withdraw(sdp, sdp->sd_log_num_buf == sdp->sd_log_commited_buf);
569 gfs2_assert_withdraw(sdp,
570 sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
572 sdp->sd_log_flush_head = sdp->sd_log_head;
573 sdp->sd_log_flush_wrapped = 0;
574 ai->ai_first = sdp->sd_log_flush_head;
576 lops_before_commit(sdp);
577 if (!list_empty(&sdp->sd_log_flush_list))
578 log_flush_commit(sdp);
579 else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle)
580 log_write_header(sdp, 0, PULL);
581 lops_after_commit(sdp, ai);
583 gfs2_log_lock(sdp);
584 sdp->sd_log_head = sdp->sd_log_flush_head;
585 sdp->sd_log_blks_free -= sdp->sd_log_num_hdrs;
586 sdp->sd_log_blks_reserved = 0;
587 sdp->sd_log_commited_buf = 0;
588 sdp->sd_log_num_hdrs = 0;
589 sdp->sd_log_commited_revoke = 0;
591 if (!list_empty(&ai->ai_ail1_list)) {
592 list_add(&ai->ai_list, &sdp->sd_ail1_list);
593 ai = NULL;
595 gfs2_log_unlock(sdp);
597 sdp->sd_vfs->s_dirt = 0;
598 up_write(&sdp->sd_log_flush_lock);
600 kfree(ai);
603 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
605 unsigned int reserved = 0;
606 unsigned int old;
608 gfs2_log_lock(sdp);
610 sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
611 gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_buf) >= 0);
612 sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
613 gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0);
615 if (sdp->sd_log_commited_buf)
616 reserved += sdp->sd_log_commited_buf;
617 if (sdp->sd_log_commited_revoke)
618 reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
619 sizeof(u64));
620 if (reserved)
621 reserved++;
623 old = sdp->sd_log_blks_free;
624 sdp->sd_log_blks_free += tr->tr_reserved -
625 (reserved - sdp->sd_log_blks_reserved);
627 gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free >= old);
628 gfs2_assert_withdraw(sdp,
629 sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks +
630 sdp->sd_log_num_hdrs);
632 sdp->sd_log_blks_reserved = reserved;
634 gfs2_log_unlock(sdp);
638 * gfs2_log_commit - Commit a transaction to the log
639 * @sdp: the filesystem
640 * @tr: the transaction
642 * Returns: errno
645 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
647 log_refund(sdp, tr);
648 lops_incore_commit(sdp, tr);
650 sdp->sd_vfs->s_dirt = 1;
651 up_read(&sdp->sd_log_flush_lock);
653 gfs2_log_lock(sdp);
654 if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks))
655 wake_up_process(sdp->sd_logd_process);
656 gfs2_log_unlock(sdp);
660 * gfs2_log_shutdown - write a shutdown header into a journal
661 * @sdp: the filesystem
665 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
667 down_write(&sdp->sd_log_flush_lock);
669 gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
670 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_gl);
671 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
672 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_jdata);
673 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
674 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
675 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
676 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_hdrs);
677 gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
679 sdp->sd_log_flush_head = sdp->sd_log_head;
680 sdp->sd_log_flush_wrapped = 0;
682 log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT, 0);
684 gfs2_assert_warn(sdp, sdp->sd_log_blks_free == sdp->sd_jdesc->jd_blocks);
685 gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
686 gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
688 sdp->sd_log_head = sdp->sd_log_flush_head;
689 sdp->sd_log_tail = sdp->sd_log_head;
691 up_write(&sdp->sd_log_flush_lock);
696 * gfs2_meta_syncfs - sync all the buffers in a filesystem
697 * @sdp: the filesystem
701 void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
703 gfs2_log_flush(sdp, NULL);
704 for (;;) {
705 gfs2_ail1_start(sdp, DIO_ALL);
706 if (gfs2_ail1_empty(sdp, DIO_ALL))
707 break;
708 msleep(10);