2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 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 v.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 <asm/semaphore.h>
20 #include "lm_interface.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)) /
54 if (nstruct > first) {
55 second = (sdp->sd_sb.sb_bsize -
56 sizeof(struct gfs2_meta_header)) / ssize;
57 blks += DIV_ROUND_UP(nstruct - first, second);
63 void gfs2_ail1_start(struct gfs2_sbd *sdp, int flags)
65 struct list_head *head = &sdp->sd_ail1_list;
67 struct list_head *first, *tmp;
68 struct gfs2_ail *first_ai, *ai;
71 if (list_empty(head)) {
75 sync_gen = sdp->sd_ail_sync_gen++;
78 first_ai = list_entry(first, struct gfs2_ail, ai_list);
79 first_ai->ai_sync_gen = sync_gen;
80 gfs2_ail1_start_one(sdp, first_ai);
86 if (first && (head->prev != first ||
87 gfs2_ail1_empty_one(sdp, first_ai, 0)))
90 for (tmp = head->prev; tmp != head; tmp = tmp->prev) {
91 ai = list_entry(tmp, struct gfs2_ail, ai_list);
92 if (ai->ai_sync_gen >= sync_gen)
94 ai->ai_sync_gen = sync_gen;
95 gfs2_ail1_start_one(sdp, ai);
103 gfs2_log_unlock(sdp);
106 int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
108 struct gfs2_ail *ai, *s;
113 list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) {
114 if (gfs2_ail1_empty_one(sdp, ai, flags))
115 list_move(&ai->ai_list, &sdp->sd_ail2_list);
116 else if (!(flags & DIO_ALL))
120 ret = list_empty(&sdp->sd_ail1_list);
122 gfs2_log_unlock(sdp);
127 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
129 struct gfs2_ail *ai, *safe;
130 unsigned int old_tail = sdp->sd_log_tail;
131 int wrap = (new_tail < old_tail);
136 list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) {
137 a = (old_tail <= ai->ai_first);
138 b = (ai->ai_first < new_tail);
139 rm = (wrap) ? (a || b) : (a && b);
143 gfs2_ail2_empty_one(sdp, ai);
144 list_del(&ai->ai_list);
145 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail1_list));
146 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail2_list));
150 gfs2_log_unlock(sdp);
154 * gfs2_log_reserve - Make a log reservation
155 * @sdp: The GFS2 superblock
156 * @blks: The number of blocks to reserve
161 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
163 unsigned int try = 0;
165 if (gfs2_assert_warn(sdp, blks) ||
166 gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
169 mutex_lock(&sdp->sd_log_reserve_mutex);
171 while(sdp->sd_log_blks_free <= blks) {
172 gfs2_log_unlock(sdp);
173 gfs2_ail1_empty(sdp, 0);
177 gfs2_ail1_start(sdp, 0);
180 sdp->sd_log_blks_free -= blks;
181 gfs2_log_unlock(sdp);
182 mutex_unlock(&sdp->sd_log_reserve_mutex);
184 down_read(&sdp->sd_log_flush_lock);
190 * gfs2_log_release - Release a given number of log blocks
191 * @sdp: The GFS2 superblock
192 * @blks: The number of blocks
196 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
198 up_read(&sdp->sd_log_flush_lock);
201 sdp->sd_log_blks_free += blks;
202 gfs2_assert_withdraw(sdp,
203 sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
204 gfs2_log_unlock(sdp);
207 static uint64_t log_bmap(struct gfs2_sbd *sdp, unsigned int lbn)
213 error = gfs2_block_map(sdp->sd_jdesc->jd_inode->u.generic_ip,
214 lbn, &new, &dbn, NULL);
215 gfs2_assert_withdraw(sdp, !error && dbn);
221 * log_distance - Compute distance between two journal blocks
222 * @sdp: The GFS2 superblock
223 * @newer: The most recent journal block of the pair
224 * @older: The older journal block of the pair
226 * Compute the distance (in the journal direction) between two
227 * blocks in the journal
229 * Returns: the distance in blocks
232 static inline unsigned int log_distance(struct gfs2_sbd *sdp,
238 dist = newer - older;
240 dist += sdp->sd_jdesc->jd_blocks;
245 static unsigned int current_tail(struct gfs2_sbd *sdp)
252 if (list_empty(&sdp->sd_ail1_list))
253 tail = sdp->sd_log_head;
255 ai = list_entry(sdp->sd_ail1_list.prev,
256 struct gfs2_ail, ai_list);
260 gfs2_log_unlock(sdp);
265 static inline void log_incr_head(struct gfs2_sbd *sdp)
267 if (sdp->sd_log_flush_head == sdp->sd_log_tail)
268 gfs2_assert_withdraw(sdp,
269 sdp->sd_log_flush_head == sdp->sd_log_head);
271 if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
272 sdp->sd_log_flush_head = 0;
273 sdp->sd_log_flush_wrapped = 1;
278 * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
279 * @sdp: The GFS2 superblock
281 * Returns: the buffer_head
284 struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
286 uint64_t blkno = log_bmap(sdp, sdp->sd_log_flush_head);
287 struct gfs2_log_buf *lb;
288 struct buffer_head *bh;
290 lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
291 list_add(&lb->lb_list, &sdp->sd_log_flush_list);
293 bh = lb->lb_bh = sb_getblk(sdp->sd_vfs, blkno);
295 memset(bh->b_data, 0, bh->b_size);
296 set_buffer_uptodate(bh);
297 clear_buffer_dirty(bh);
306 * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
307 * @sdp: the filesystem
308 * @data: the data the buffer_head should point to
310 * Returns: the log buffer descriptor
313 struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
314 struct buffer_head *real)
316 uint64_t blkno = log_bmap(sdp, sdp->sd_log_flush_head);
317 struct gfs2_log_buf *lb;
318 struct buffer_head *bh;
320 lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
321 list_add(&lb->lb_list, &sdp->sd_log_flush_list);
324 bh = lb->lb_bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL);
325 atomic_set(&bh->b_count, 1);
326 bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate);
327 set_bh_page(bh, real->b_page, bh_offset(real));
328 bh->b_blocknr = blkno;
329 bh->b_size = sdp->sd_sb.sb_bsize;
330 bh->b_bdev = sdp->sd_vfs->s_bdev;
337 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail, int pull)
339 unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
341 ail2_empty(sdp, new_tail);
344 sdp->sd_log_blks_free += dist - ((pull) ? 1 : 0);
345 gfs2_assert_withdraw(sdp,
346 sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
347 gfs2_log_unlock(sdp);
349 sdp->sd_log_tail = new_tail;
353 * log_write_header - Get and initialize a journal header buffer
354 * @sdp: The GFS2 superblock
356 * Returns: the initialized log buffer descriptor
359 static void log_write_header(struct gfs2_sbd *sdp, uint32_t flags, int pull)
361 uint64_t blkno = log_bmap(sdp, sdp->sd_log_flush_head);
362 struct buffer_head *bh;
363 struct gfs2_log_header *lh;
367 bh = sb_getblk(sdp->sd_vfs, blkno);
369 memset(bh->b_data, 0, bh->b_size);
370 set_buffer_uptodate(bh);
371 clear_buffer_dirty(bh);
374 gfs2_ail1_empty(sdp, 0);
375 tail = current_tail(sdp);
377 lh = (struct gfs2_log_header *)bh->b_data;
378 memset(lh, 0, sizeof(struct gfs2_log_header));
379 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
380 lh->lh_header.mh_type = cpu_to_be16(GFS2_METATYPE_LH);
381 lh->lh_header.mh_format = cpu_to_be16(GFS2_FORMAT_LH);
382 lh->lh_sequence = be64_to_cpu(sdp->sd_log_sequence++);
383 lh->lh_flags = be32_to_cpu(flags);
384 lh->lh_tail = be32_to_cpu(tail);
385 lh->lh_blkno = be32_to_cpu(sdp->sd_log_flush_head);
386 hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
387 lh->lh_hash = cpu_to_be32(hash);
389 set_buffer_dirty(bh);
390 if (sync_dirty_buffer(bh))
391 gfs2_io_error_bh(sdp, bh);
394 if (sdp->sd_log_tail != tail)
395 log_pull_tail(sdp, tail, pull);
397 gfs2_assert_withdraw(sdp, !pull);
399 sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
403 static void log_flush_commit(struct gfs2_sbd *sdp)
405 struct list_head *head = &sdp->sd_log_flush_list;
406 struct gfs2_log_buf *lb;
407 struct buffer_head *bh;
410 d = log_distance(sdp, sdp->sd_log_flush_head, sdp->sd_log_head);
412 gfs2_assert_withdraw(sdp, d + 1 == sdp->sd_log_blks_reserved);
414 while (!list_empty(head)) {
415 lb = list_entry(head->next, struct gfs2_log_buf, lb_list);
416 list_del(&lb->lb_list);
420 if (!buffer_uptodate(bh))
421 gfs2_io_error_bh(sdp, bh);
423 while (atomic_read(&bh->b_count) != 1) /* Grrrr... */
425 free_buffer_head(bh);
431 log_write_header(sdp, 0, 0);
435 * gfs2_log_flush_i - flush incore transaction(s)
436 * @sdp: the filesystem
437 * @gl: The glock structure to flush. If NULL, flush the whole incore log
441 void gfs2_log_flush_i(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
445 ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
446 INIT_LIST_HEAD(&ai->ai_ail1_list);
447 INIT_LIST_HEAD(&ai->ai_ail2_list);
449 down_write(&sdp->sd_log_flush_lock);
453 if (list_empty(&gl->gl_le.le_list)) {
454 gfs2_log_unlock(sdp);
455 up_write(&sdp->sd_log_flush_lock);
459 gfs2_log_unlock(sdp);
463 gfs2_assert_withdraw(sdp,
464 sdp->sd_log_num_buf == sdp->sd_log_commited_buf);
465 gfs2_assert_withdraw(sdp,
466 sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
468 sdp->sd_log_flush_head = sdp->sd_log_head;
469 sdp->sd_log_flush_wrapped = 0;
470 ai->ai_first = sdp->sd_log_flush_head;
472 lops_before_commit(sdp);
473 if (!list_empty(&sdp->sd_log_flush_list))
474 log_flush_commit(sdp);
475 else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle)
476 log_write_header(sdp, 0, PULL);
477 lops_after_commit(sdp, ai);
478 sdp->sd_log_head = sdp->sd_log_flush_head;
479 if (sdp->sd_log_flush_wrapped)
482 sdp->sd_log_blks_reserved =
483 sdp->sd_log_commited_buf =
484 sdp->sd_log_commited_revoke = 0;
487 if (!list_empty(&ai->ai_ail1_list)) {
488 list_add(&ai->ai_list, &sdp->sd_ail1_list);
491 gfs2_log_unlock(sdp);
493 sdp->sd_vfs->s_dirt = 0;
494 up_write(&sdp->sd_log_flush_lock);
499 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
501 unsigned int reserved = 1;
506 sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
507 gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_buf) >= 0);
508 sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
509 gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0);
511 if (sdp->sd_log_commited_buf)
512 reserved += 1 + sdp->sd_log_commited_buf +
513 sdp->sd_log_commited_buf/503;
514 if (sdp->sd_log_commited_revoke)
515 reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
518 old = sdp->sd_log_blks_free;
519 sdp->sd_log_blks_free += tr->tr_reserved -
520 (reserved - sdp->sd_log_blks_reserved);
522 gfs2_assert_withdraw(sdp,
523 sdp->sd_log_blks_free >= old);
524 gfs2_assert_withdraw(sdp,
525 sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
527 sdp->sd_log_blks_reserved = reserved;
529 gfs2_log_unlock(sdp);
533 * gfs2_log_commit - Commit a transaction to the log
534 * @sdp: the filesystem
535 * @tr: the transaction
540 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
543 lops_incore_commit(sdp, tr);
545 sdp->sd_vfs->s_dirt = 1;
546 up_read(&sdp->sd_log_flush_lock);
549 if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks)) {
550 gfs2_log_unlock(sdp);
553 gfs2_log_unlock(sdp);
557 * gfs2_log_shutdown - write a shutdown header into a journal
558 * @sdp: the filesystem
562 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
564 down_write(&sdp->sd_log_flush_lock);
566 gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
567 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_gl);
568 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
569 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_jdata);
570 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
571 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
572 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
573 gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
575 sdp->sd_log_flush_head = sdp->sd_log_head;
576 sdp->sd_log_flush_wrapped = 0;
578 log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT, 0);
580 gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free ==
581 sdp->sd_jdesc->jd_blocks);
582 gfs2_assert_withdraw(sdp, sdp->sd_log_head == sdp->sd_log_tail);
583 gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail2_list));
585 sdp->sd_log_head = sdp->sd_log_flush_head;
586 if (sdp->sd_log_flush_wrapped)
588 sdp->sd_log_tail = sdp->sd_log_head;
590 up_write(&sdp->sd_log_flush_lock);