[POWERPC] ps3: add htab routines
[linux-2.6/linux-loongson.git] / fs / gfs2 / log.c
blob0cace3da9dbb41795b6cacdbe97de52786d3678e
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>
19 #include "gfs2.h"
20 #include "incore.h"
21 #include "bmap.h"
22 #include "glock.h"
23 #include "log.h"
24 #include "lops.h"
25 #include "meta_io.h"
26 #include "util.h"
27 #include "dir.h"
29 #define PULL 1
31 /**
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,
44 unsigned int ssize)
46 unsigned int blks;
47 unsigned int first, second;
49 blks = 1;
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);
58 return blks;
61 /**
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;
72 int retry;
74 BUG_ON(!spin_is_locked(&sdp->sd_log_lock));
76 do {
77 retry = 0;
79 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
80 bd_ail_st_list) {
81 bh = bd->bd_bh;
83 gfs2_assert(sdp, bd->bd_ail == ai);
85 if (!buffer_busy(bh)) {
86 if (!buffer_uptodate(bh)) {
87 gfs2_log_unlock(sdp);
88 gfs2_io_error_bh(sdp, bh);
89 gfs2_log_lock(sdp);
91 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
92 continue;
95 if (!buffer_dirty(bh))
96 continue;
98 list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
100 gfs2_log_unlock(sdp);
101 wait_on_buffer(bh);
102 ll_rw_block(WRITE, 1, &bh);
103 gfs2_log_lock(sdp);
105 retry = 1;
106 break;
108 } while (retry);
112 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
113 * @sdp: the filesystem
114 * @ai: the AIL entry
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,
124 bd_ail_st_list) {
125 bh = bd->bd_bh;
127 gfs2_assert(sdp, bd->bd_ail == ai);
129 if (buffer_busy(bh)) {
130 if (flags & DIO_ALL)
131 continue;
132 else
133 break;
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;
148 u64 sync_gen;
149 struct list_head *first;
150 struct gfs2_ail *first_ai, *ai, *tmp;
151 int done = 0;
153 gfs2_log_lock(sdp);
154 if (list_empty(head)) {
155 gfs2_log_unlock(sdp);
156 return;
158 sync_gen = sdp->sd_ail_sync_gen++;
160 first = head->prev;
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 */
165 if (flags & DIO_ALL)
166 first = NULL;
168 while(!done) {
169 if (first && (head->prev != first ||
170 gfs2_ail1_empty_one(sdp, first_ai, 0)))
171 break;
173 done = 1;
174 list_for_each_entry_safe_reverse(ai, tmp, head, ai_list) {
175 if (ai->ai_sync_gen >= sync_gen)
176 continue;
177 ai->ai_sync_gen = sync_gen;
178 gfs2_ail1_start_one(sdp, ai); /* This may drop log lock */
179 done = 0;
180 break;
184 gfs2_log_unlock(sdp);
187 int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
189 struct gfs2_ail *ai, *s;
190 int ret;
192 gfs2_log_lock(sdp);
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))
198 break;
201 ret = list_empty(&sdp->sd_ail1_list);
203 gfs2_log_unlock(sdp);
205 return ret;
210 * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
211 * @sdp: the filesystem
212 * @ai: the AIL entry
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,
223 bd_ail_st_list);
224 gfs2_assert(sdp, bd->bd_ail == ai);
225 bd->bd_ail = NULL;
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);
229 brelse(bd->bd_bh);
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);
238 int a, b, rm;
240 gfs2_log_lock(sdp);
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);
246 if (!rm)
247 continue;
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));
253 kfree(ai);
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
264 * Returns: errno
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))
273 return -EINVAL;
275 mutex_lock(&sdp->sd_log_reserve_mutex);
276 gfs2_log_lock(sdp);
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);
282 if (try++)
283 gfs2_ail1_start(sdp, 0);
284 gfs2_log_lock(sdp);
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);
292 return 0;
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)
305 gfs2_log_lock(sdp);
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)
315 struct inode *inode = sdp->sd_jdesc->jd_inode;
316 int error;
317 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
319 bh_map.b_size = 1 << inode->i_blkbits;
320 error = gfs2_block_map(inode, lbn, 0, &bh_map);
321 if (error || !bh_map.b_blocknr)
322 printk(KERN_INFO "error=%d, dbn=%llu lbn=%u", error, bh_map.b_blocknr, lbn);
323 gfs2_assert_withdraw(sdp, !error && bh_map.b_blocknr);
325 return bh_map.b_blocknr;
329 * log_distance - Compute distance between two journal blocks
330 * @sdp: The GFS2 superblock
331 * @newer: The most recent journal block of the pair
332 * @older: The older journal block of the pair
334 * Compute the distance (in the journal direction) between two
335 * blocks in the journal
337 * Returns: the distance in blocks
340 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
341 unsigned int older)
343 int dist;
345 dist = newer - older;
346 if (dist < 0)
347 dist += sdp->sd_jdesc->jd_blocks;
349 return dist;
352 static unsigned int current_tail(struct gfs2_sbd *sdp)
354 struct gfs2_ail *ai;
355 unsigned int tail;
357 gfs2_log_lock(sdp);
359 if (list_empty(&sdp->sd_ail1_list)) {
360 tail = sdp->sd_log_head;
361 } else {
362 ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail, ai_list);
363 tail = ai->ai_first;
366 gfs2_log_unlock(sdp);
368 return tail;
371 static inline void log_incr_head(struct gfs2_sbd *sdp)
373 if (sdp->sd_log_flush_head == sdp->sd_log_tail)
374 gfs2_assert_withdraw(sdp, sdp->sd_log_flush_head == sdp->sd_log_head);
376 if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
377 sdp->sd_log_flush_head = 0;
378 sdp->sd_log_flush_wrapped = 1;
383 * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
384 * @sdp: The GFS2 superblock
386 * Returns: the buffer_head
389 struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
391 u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
392 struct gfs2_log_buf *lb;
393 struct buffer_head *bh;
395 lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
396 list_add(&lb->lb_list, &sdp->sd_log_flush_list);
398 bh = lb->lb_bh = sb_getblk(sdp->sd_vfs, blkno);
399 lock_buffer(bh);
400 memset(bh->b_data, 0, bh->b_size);
401 set_buffer_uptodate(bh);
402 clear_buffer_dirty(bh);
403 unlock_buffer(bh);
405 log_incr_head(sdp);
407 return bh;
411 * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
412 * @sdp: the filesystem
413 * @data: the data the buffer_head should point to
415 * Returns: the log buffer descriptor
418 struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
419 struct buffer_head *real)
421 u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
422 struct gfs2_log_buf *lb;
423 struct buffer_head *bh;
425 lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
426 list_add(&lb->lb_list, &sdp->sd_log_flush_list);
427 lb->lb_real = real;
429 bh = lb->lb_bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL);
430 atomic_set(&bh->b_count, 1);
431 bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate);
432 set_bh_page(bh, real->b_page, bh_offset(real));
433 bh->b_blocknr = blkno;
434 bh->b_size = sdp->sd_sb.sb_bsize;
435 bh->b_bdev = sdp->sd_vfs->s_bdev;
437 log_incr_head(sdp);
439 return bh;
442 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail, int pull)
444 unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
446 ail2_empty(sdp, new_tail);
448 gfs2_log_lock(sdp);
449 sdp->sd_log_blks_free += dist - (pull ? 1 : 0);
450 gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
451 gfs2_log_unlock(sdp);
453 sdp->sd_log_tail = new_tail;
457 * log_write_header - Get and initialize a journal header buffer
458 * @sdp: The GFS2 superblock
460 * Returns: the initialized log buffer descriptor
463 static void log_write_header(struct gfs2_sbd *sdp, u32 flags, int pull)
465 u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
466 struct buffer_head *bh;
467 struct gfs2_log_header *lh;
468 unsigned int tail;
469 u32 hash;
471 bh = sb_getblk(sdp->sd_vfs, blkno);
472 lock_buffer(bh);
473 memset(bh->b_data, 0, bh->b_size);
474 set_buffer_uptodate(bh);
475 clear_buffer_dirty(bh);
476 unlock_buffer(bh);
478 gfs2_ail1_empty(sdp, 0);
479 tail = current_tail(sdp);
481 lh = (struct gfs2_log_header *)bh->b_data;
482 memset(lh, 0, sizeof(struct gfs2_log_header));
483 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
484 lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
485 lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
486 lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++);
487 lh->lh_flags = cpu_to_be32(flags);
488 lh->lh_tail = cpu_to_be32(tail);
489 lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
490 hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
491 lh->lh_hash = cpu_to_be32(hash);
493 set_buffer_dirty(bh);
494 if (sync_dirty_buffer(bh))
495 gfs2_io_error_bh(sdp, bh);
496 brelse(bh);
498 if (sdp->sd_log_tail != tail)
499 log_pull_tail(sdp, tail, pull);
500 else
501 gfs2_assert_withdraw(sdp, !pull);
503 sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
504 log_incr_head(sdp);
507 static void log_flush_commit(struct gfs2_sbd *sdp)
509 struct list_head *head = &sdp->sd_log_flush_list;
510 struct gfs2_log_buf *lb;
511 struct buffer_head *bh;
513 while (!list_empty(head)) {
514 lb = list_entry(head->next, struct gfs2_log_buf, lb_list);
515 list_del(&lb->lb_list);
516 bh = lb->lb_bh;
518 wait_on_buffer(bh);
519 if (!buffer_uptodate(bh))
520 gfs2_io_error_bh(sdp, bh);
521 if (lb->lb_real) {
522 while (atomic_read(&bh->b_count) != 1) /* Grrrr... */
523 schedule();
524 free_buffer_head(bh);
525 } else
526 brelse(bh);
527 kfree(lb);
530 log_write_header(sdp, 0, 0);
534 * gfs2_log_flush - flush incore transaction(s)
535 * @sdp: the filesystem
536 * @gl: The glock structure to flush. If NULL, flush the whole incore log
540 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
542 struct gfs2_ail *ai;
544 down_write(&sdp->sd_log_flush_lock);
546 if (gl) {
547 gfs2_log_lock(sdp);
548 if (list_empty(&gl->gl_le.le_list)) {
549 gfs2_log_unlock(sdp);
550 up_write(&sdp->sd_log_flush_lock);
551 return;
553 gfs2_log_unlock(sdp);
556 ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
557 INIT_LIST_HEAD(&ai->ai_ail1_list);
558 INIT_LIST_HEAD(&ai->ai_ail2_list);
560 gfs2_assert_withdraw(sdp, sdp->sd_log_num_buf == sdp->sd_log_commited_buf);
561 gfs2_assert_withdraw(sdp,
562 sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
564 sdp->sd_log_flush_head = sdp->sd_log_head;
565 sdp->sd_log_flush_wrapped = 0;
566 ai->ai_first = sdp->sd_log_flush_head;
568 lops_before_commit(sdp);
569 if (!list_empty(&sdp->sd_log_flush_list))
570 log_flush_commit(sdp);
571 else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle)
572 log_write_header(sdp, 0, PULL);
573 lops_after_commit(sdp, ai);
575 gfs2_log_lock(sdp);
576 sdp->sd_log_head = sdp->sd_log_flush_head;
577 sdp->sd_log_blks_free -= sdp->sd_log_num_hdrs;
578 sdp->sd_log_blks_reserved = 0;
579 sdp->sd_log_commited_buf = 0;
580 sdp->sd_log_num_hdrs = 0;
581 sdp->sd_log_commited_revoke = 0;
583 if (!list_empty(&ai->ai_ail1_list)) {
584 list_add(&ai->ai_list, &sdp->sd_ail1_list);
585 ai = NULL;
587 gfs2_log_unlock(sdp);
589 sdp->sd_vfs->s_dirt = 0;
590 up_write(&sdp->sd_log_flush_lock);
592 kfree(ai);
595 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
597 unsigned int reserved = 0;
598 unsigned int old;
600 gfs2_log_lock(sdp);
602 sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
603 gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_buf) >= 0);
604 sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
605 gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0);
607 if (sdp->sd_log_commited_buf)
608 reserved += sdp->sd_log_commited_buf;
609 if (sdp->sd_log_commited_revoke)
610 reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
611 sizeof(u64));
612 if (reserved)
613 reserved++;
615 old = sdp->sd_log_blks_free;
616 sdp->sd_log_blks_free += tr->tr_reserved -
617 (reserved - sdp->sd_log_blks_reserved);
619 gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free >= old);
620 gfs2_assert_withdraw(sdp,
621 sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks +
622 sdp->sd_log_num_hdrs);
624 sdp->sd_log_blks_reserved = reserved;
626 gfs2_log_unlock(sdp);
630 * gfs2_log_commit - Commit a transaction to the log
631 * @sdp: the filesystem
632 * @tr: the transaction
634 * Returns: errno
637 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
639 log_refund(sdp, tr);
640 lops_incore_commit(sdp, tr);
642 sdp->sd_vfs->s_dirt = 1;
643 up_read(&sdp->sd_log_flush_lock);
645 gfs2_log_lock(sdp);
646 if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks)) {
647 gfs2_log_unlock(sdp);
648 gfs2_log_flush(sdp, NULL);
649 } else {
650 gfs2_log_unlock(sdp);
655 * gfs2_log_shutdown - write a shutdown header into a journal
656 * @sdp: the filesystem
660 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
662 down_write(&sdp->sd_log_flush_lock);
664 gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
665 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_gl);
666 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
667 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_jdata);
668 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
669 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
670 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
671 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_hdrs);
672 gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
674 sdp->sd_log_flush_head = sdp->sd_log_head;
675 sdp->sd_log_flush_wrapped = 0;
677 log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT, 0);
679 gfs2_assert_warn(sdp, sdp->sd_log_blks_free == sdp->sd_jdesc->jd_blocks);
680 gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
681 gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
683 sdp->sd_log_head = sdp->sd_log_flush_head;
684 sdp->sd_log_tail = sdp->sd_log_head;
686 up_write(&sdp->sd_log_flush_lock);