Merge branch 'master'
[linux-2.6] / fs / gfs2 / log.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2005 Red Hat, Inc.  All rights reserved.
4  *
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.
8  */
9
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 <asm/semaphore.h>
16
17 #include "gfs2.h"
18 #include "bmap.h"
19 #include "glock.h"
20 #include "log.h"
21 #include "lops.h"
22 #include "meta_io.h"
23
24 #define PULL 1
25
26 static void do_lock_wait(struct gfs2_sbd *sdp, wait_queue_head_t *wq,
27                          atomic_t *a)
28 {
29         wait_event(*wq, atomic_read(a) ? 0 : 1);
30 }
31
32 static void lock_for_trans(struct gfs2_sbd *sdp)
33 {
34         do_lock_wait(sdp, &sdp->sd_log_trans_wq, &sdp->sd_log_flush_count);
35         atomic_inc(&sdp->sd_log_trans_count);
36 }
37
38 static void unlock_from_trans(struct gfs2_sbd *sdp)
39 {
40         gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_trans_count));
41         if (atomic_dec_and_test(&sdp->sd_log_trans_count))
42                 wake_up(&sdp->sd_log_flush_wq);
43 }
44
45 static void gfs2_lock_for_flush(struct gfs2_sbd *sdp)
46 {
47         atomic_inc(&sdp->sd_log_flush_count);
48         do_lock_wait(sdp, &sdp->sd_log_flush_wq, &sdp->sd_log_trans_count);
49 }
50
51 static void gfs2_unlock_from_flush(struct gfs2_sbd *sdp)
52 {
53         gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_flush_count));
54         if (atomic_dec_and_test(&sdp->sd_log_flush_count))
55                 wake_up(&sdp->sd_log_trans_wq);
56 }
57
58 /**
59  * gfs2_struct2blk - compute stuff
60  * @sdp: the filesystem
61  * @nstruct: the number of structures
62  * @ssize: the size of the structures
63  *
64  * Compute the number of log descriptor blocks needed to hold a certain number
65  * of structures of a certain size.
66  *
67  * Returns: the number of blocks needed (minimum is always 1)
68  */
69
70 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
71                              unsigned int ssize)
72 {
73         unsigned int blks;
74         unsigned int first, second;
75
76         blks = 1;
77         first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) /
78                 ssize;
79
80         if (nstruct > first) {
81                 second = (sdp->sd_sb.sb_bsize -
82                           sizeof(struct gfs2_meta_header)) / ssize;
83                 blks += DIV_RU(nstruct - first, second);
84         }
85
86         return blks;
87 }
88
89 void gfs2_ail1_start(struct gfs2_sbd *sdp, int flags)
90 {
91         struct list_head *head = &sdp->sd_ail1_list;
92         uint64_t sync_gen;
93         struct list_head *first, *tmp;
94         struct gfs2_ail *first_ai, *ai;
95
96         gfs2_log_lock(sdp);
97         if (list_empty(head)) {
98                 gfs2_log_unlock(sdp);
99                 return;
100         }
101         sync_gen = sdp->sd_ail_sync_gen++;
102
103         first = head->prev;
104         first_ai = list_entry(first, struct gfs2_ail, ai_list);
105         first_ai->ai_sync_gen = sync_gen;
106         gfs2_ail1_start_one(sdp, first_ai);
107
108         if (flags & DIO_ALL)
109                 first = NULL;
110
111         for (;;) {
112                 if (first &&
113                     (head->prev != first ||
114                      gfs2_ail1_empty_one(sdp, first_ai, 0)))
115                         break;
116
117                 for (tmp = head->prev; tmp != head; tmp = tmp->prev) {
118                         ai = list_entry(tmp, struct gfs2_ail, ai_list);
119                         if (ai->ai_sync_gen >= sync_gen)
120                                 continue;
121                         ai->ai_sync_gen = sync_gen;
122                         gfs2_ail1_start_one(sdp, ai);
123                         break;
124                 }
125
126                 if (tmp == head)
127                         break;
128         }
129
130         gfs2_log_unlock(sdp);
131 }
132
133 int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
134 {
135         struct gfs2_ail *ai, *s;
136         int ret;
137
138         gfs2_log_lock(sdp);
139
140         list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) {
141                 if (gfs2_ail1_empty_one(sdp, ai, flags))
142                         list_move(&ai->ai_list, &sdp->sd_ail2_list);
143                 else if (!(flags & DIO_ALL))
144                         break;
145         }
146
147         ret = list_empty(&sdp->sd_ail1_list);
148
149         gfs2_log_unlock(sdp);
150
151         return ret;
152 }
153
154 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
155 {
156         struct gfs2_ail *ai, *safe;
157         unsigned int old_tail = sdp->sd_log_tail;
158         int wrap = (new_tail < old_tail);
159         int a, b, rm;
160
161         gfs2_log_lock(sdp);
162
163         list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) {
164                 a = (old_tail <= ai->ai_first);
165                 b = (ai->ai_first < new_tail);
166                 rm = (wrap) ? (a || b) : (a && b);
167                 if (!rm)
168                         continue;
169
170                 gfs2_ail2_empty_one(sdp, ai);
171                 list_del(&ai->ai_list);
172                 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail1_list));
173                 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail2_list));
174                 kfree(ai);
175         }
176
177         gfs2_log_unlock(sdp);
178 }
179
180 /**
181  * gfs2_log_reserve - Make a log reservation
182  * @sdp: The GFS2 superblock
183  * @blks: The number of blocks to reserve
184  *
185  * Returns: errno
186  */
187
188 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
189 {
190         LIST_HEAD(list);
191         unsigned int try = 0;
192
193         if (gfs2_assert_warn(sdp, blks) ||
194             gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
195                 return -EINVAL;
196
197         for (;;) {
198                 gfs2_log_lock(sdp);
199                 if (list_empty(&list)) {
200                         list_add_tail(&list, &sdp->sd_log_blks_list);
201                         while (sdp->sd_log_blks_list.next != &list) {
202                                 DECLARE_WAITQUEUE(__wait_chan, current);
203                                 set_current_state(TASK_UNINTERRUPTIBLE);
204                                 add_wait_queue(&sdp->sd_log_blks_wait,
205                                                &__wait_chan);
206                                 gfs2_log_unlock(sdp);
207                                 schedule();
208                                 gfs2_log_lock(sdp);
209                                 remove_wait_queue(&sdp->sd_log_blks_wait,
210                                                   &__wait_chan);
211                                 set_current_state(TASK_RUNNING);
212                         }
213                 }
214                 /* Never give away the last block so we can
215                    always pull the tail if we need to. */
216                 if (sdp->sd_log_blks_free > blks) {
217                         sdp->sd_log_blks_free -= blks;
218                         list_del(&list);
219                         gfs2_log_unlock(sdp);
220                         wake_up(&sdp->sd_log_blks_wait);
221                         break;
222                 }
223
224                 gfs2_log_unlock(sdp);
225                 gfs2_ail1_empty(sdp, 0);
226                 gfs2_log_flush(sdp);
227
228                 if (try++)
229                         gfs2_ail1_start(sdp, 0);
230         }
231         lock_for_trans(sdp);
232
233         return 0;
234 }
235
236 /**
237  * gfs2_log_release - Release a given number of log blocks
238  * @sdp: The GFS2 superblock
239  * @blks: The number of blocks
240  *
241  */
242
243 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
244 {
245         unlock_from_trans(sdp);
246
247         gfs2_log_lock(sdp);
248         sdp->sd_log_blks_free += blks;
249         gfs2_assert_withdraw(sdp,
250                              sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
251         gfs2_log_unlock(sdp);
252 }
253
254 static uint64_t log_bmap(struct gfs2_sbd *sdp, unsigned int lbn)
255 {
256         int new = 0;
257         uint64_t dbn;
258         int error;
259
260         error = gfs2_block_map(get_v2ip(sdp->sd_jdesc->jd_inode),
261                                lbn, &new, &dbn, NULL);
262         gfs2_assert_withdraw(sdp, !error && dbn);
263
264         return dbn;
265 }
266
267 /**
268  * log_distance - Compute distance between two journal blocks
269  * @sdp: The GFS2 superblock
270  * @newer: The most recent journal block of the pair
271  * @older: The older journal block of the pair
272  *
273  *   Compute the distance (in the journal direction) between two
274  *   blocks in the journal
275  *
276  * Returns: the distance in blocks
277  */
278
279 static inline unsigned int log_distance(struct gfs2_sbd *sdp,
280                                         unsigned int newer,
281                                         unsigned int older)
282 {
283         int dist;
284
285         dist = newer - older;
286         if (dist < 0)
287                 dist += sdp->sd_jdesc->jd_blocks;
288
289         return dist;
290 }
291
292 static unsigned int current_tail(struct gfs2_sbd *sdp)
293 {
294         struct gfs2_ail *ai;
295         unsigned int tail;
296
297         gfs2_log_lock(sdp);
298
299         if (list_empty(&sdp->sd_ail1_list))
300                 tail = sdp->sd_log_head;
301         else {
302                 ai = list_entry(sdp->sd_ail1_list.prev,
303                                 struct gfs2_ail, ai_list);
304                 tail = ai->ai_first;
305         }
306
307         gfs2_log_unlock(sdp);
308
309         return tail;
310 }
311
312 static inline void log_incr_head(struct gfs2_sbd *sdp)
313 {
314         if (sdp->sd_log_flush_head == sdp->sd_log_tail)
315                 gfs2_assert_withdraw(sdp,
316                                 sdp->sd_log_flush_head == sdp->sd_log_head);
317
318         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
319                 sdp->sd_log_flush_head = 0;
320                 sdp->sd_log_flush_wrapped = 1;
321         }
322 }
323
324 /**
325  * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
326  * @sdp: The GFS2 superblock
327  *
328  * Returns: the buffer_head
329  */
330
331 struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
332 {
333         uint64_t blkno = log_bmap(sdp, sdp->sd_log_flush_head);
334         struct gfs2_log_buf *lb;
335         struct buffer_head *bh;
336
337         lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
338         list_add(&lb->lb_list, &sdp->sd_log_flush_list);
339
340         bh = lb->lb_bh = sb_getblk(sdp->sd_vfs, blkno);
341         lock_buffer(bh);
342         memset(bh->b_data, 0, bh->b_size);
343         set_buffer_uptodate(bh);
344         clear_buffer_dirty(bh);
345         unlock_buffer(bh);
346
347         log_incr_head(sdp);
348
349         return bh;
350 }
351
352 /**
353  * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
354  * @sdp: the filesystem
355  * @data: the data the buffer_head should point to
356  *
357  * Returns: the log buffer descriptor
358  */
359
360 struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
361                                       struct buffer_head *real)
362 {
363         uint64_t blkno = log_bmap(sdp, sdp->sd_log_flush_head);
364         struct gfs2_log_buf *lb;
365         struct buffer_head *bh;
366
367         lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
368         list_add(&lb->lb_list, &sdp->sd_log_flush_list);
369         lb->lb_real = real;
370
371         bh = lb->lb_bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL);
372         atomic_set(&bh->b_count, 1);
373         bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate);
374         set_bh_page(bh, real->b_page, bh_offset(real));
375         bh->b_blocknr = blkno;
376         bh->b_size = sdp->sd_sb.sb_bsize;
377         bh->b_bdev = sdp->sd_vfs->s_bdev;
378
379         log_incr_head(sdp);
380
381         return bh;
382 }
383
384 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail, int pull)
385 {
386         unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
387
388         ail2_empty(sdp, new_tail);
389
390         gfs2_log_lock(sdp);
391         sdp->sd_log_blks_free += dist - ((pull) ? 1 : 0);
392         gfs2_assert_withdraw(sdp,
393                              sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
394         gfs2_log_unlock(sdp);
395
396         sdp->sd_log_tail = new_tail;
397 }
398
399 /**
400  * log_write_header - Get and initialize a journal header buffer
401  * @sdp: The GFS2 superblock
402  *
403  * Returns: the initialized log buffer descriptor
404  */
405
406 static void log_write_header(struct gfs2_sbd *sdp, uint32_t flags, int pull)
407 {
408         uint64_t blkno = log_bmap(sdp, sdp->sd_log_flush_head);
409         struct buffer_head *bh;
410         struct gfs2_log_header *lh;
411         unsigned int tail;
412         uint32_t hash;
413
414         bh = sb_getblk(sdp->sd_vfs, blkno);
415         lock_buffer(bh);
416         memset(bh->b_data, 0, bh->b_size);
417         set_buffer_uptodate(bh);
418         clear_buffer_dirty(bh);
419         unlock_buffer(bh);
420
421         gfs2_ail1_empty(sdp, 0);
422         tail = current_tail(sdp);
423
424         lh = (struct gfs2_log_header *)bh->b_data;
425         memset(lh, 0, sizeof(struct gfs2_log_header));
426         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
427         lh->lh_header.mh_type = cpu_to_be16(GFS2_METATYPE_LH);
428         lh->lh_header.mh_format = cpu_to_be16(GFS2_FORMAT_LH);
429         lh->lh_sequence = be64_to_cpu(sdp->sd_log_sequence++);
430         lh->lh_flags = be32_to_cpu(flags);
431         lh->lh_tail = be32_to_cpu(tail);
432         lh->lh_blkno = be32_to_cpu(sdp->sd_log_flush_head);
433         hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
434         lh->lh_hash = cpu_to_be32(hash);
435
436         set_buffer_dirty(bh);
437         if (sync_dirty_buffer(bh))
438                 gfs2_io_error_bh(sdp, bh);
439         brelse(bh);
440
441         if (sdp->sd_log_tail != tail)
442                 log_pull_tail(sdp, tail, pull);
443         else
444                 gfs2_assert_withdraw(sdp, !pull);
445
446         sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
447         log_incr_head(sdp);
448 }
449
450 static void log_flush_commit(struct gfs2_sbd *sdp)
451 {
452         struct list_head *head = &sdp->sd_log_flush_list;
453         struct gfs2_log_buf *lb;
454         struct buffer_head *bh;
455         unsigned int d;
456
457         d = log_distance(sdp, sdp->sd_log_flush_head, sdp->sd_log_head);
458
459         gfs2_assert_withdraw(sdp, d + 1 == sdp->sd_log_blks_reserved);
460
461         while (!list_empty(head)) {
462                 lb = list_entry(head->next, struct gfs2_log_buf, lb_list);
463                 list_del(&lb->lb_list);
464                 bh = lb->lb_bh;
465
466                 wait_on_buffer(bh);
467                 if (!buffer_uptodate(bh))
468                         gfs2_io_error_bh(sdp, bh);
469                 if (lb->lb_real) {
470                         while (atomic_read(&bh->b_count) != 1)  /* Grrrr... */
471                                 schedule();
472                         free_buffer_head(bh);
473                 } else
474                         brelse(bh);
475                 kfree(lb);
476         }
477
478         log_write_header(sdp, 0, 0);
479 }
480
481 /**
482  * gfs2_log_flush_i - flush incore transaction(s)
483  * @sdp: the filesystem
484  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
485  *
486  */
487
488 void gfs2_log_flush_i(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
489 {
490         struct gfs2_ail *ai;
491
492         ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
493         INIT_LIST_HEAD(&ai->ai_ail1_list);
494         INIT_LIST_HEAD(&ai->ai_ail2_list);
495         gfs2_lock_for_flush(sdp);
496
497         if (gl) {
498                 gfs2_log_lock(sdp);
499                 if (list_empty(&gl->gl_le.le_list)) {
500                         gfs2_log_unlock(sdp);
501                         gfs2_unlock_from_flush(sdp);
502                         kfree(ai);
503                         return;
504                 }
505                 gfs2_log_unlock(sdp);
506         }
507
508         mutex_lock(&sdp->sd_log_flush_lock);
509
510         gfs2_assert_withdraw(sdp,
511                         sdp->sd_log_num_buf == sdp->sd_log_commited_buf);
512         gfs2_assert_withdraw(sdp,
513                         sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
514
515         sdp->sd_log_flush_head = sdp->sd_log_head;
516         sdp->sd_log_flush_wrapped = 0;
517         ai->ai_first = sdp->sd_log_flush_head;
518
519         lops_before_commit(sdp);
520         if (!list_empty(&sdp->sd_log_flush_list))
521                 log_flush_commit(sdp);
522         else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle)
523                 log_write_header(sdp, 0, PULL);
524         lops_after_commit(sdp, ai);
525         sdp->sd_log_head = sdp->sd_log_flush_head;
526         if (sdp->sd_log_flush_wrapped)
527                 sdp->sd_log_wraps++;
528
529         sdp->sd_log_blks_reserved =
530                 sdp->sd_log_commited_buf =
531                 sdp->sd_log_commited_revoke = 0;
532
533         gfs2_log_lock(sdp);
534         if (!list_empty(&ai->ai_ail1_list)) {
535                 list_add(&ai->ai_list, &sdp->sd_ail1_list);
536                 ai = NULL;
537         }
538         gfs2_log_unlock(sdp);
539
540         mutex_unlock(&sdp->sd_log_flush_lock);
541         sdp->sd_vfs->s_dirt = 0;
542         gfs2_unlock_from_flush(sdp);
543
544         kfree(ai);
545 }
546
547 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
548 {
549         unsigned int reserved = 1;
550         unsigned int old;
551
552         gfs2_log_lock(sdp);
553
554         sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
555         gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_buf) >= 0);
556         sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
557         gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0);
558
559         if (sdp->sd_log_commited_buf)
560                 reserved += 1 + sdp->sd_log_commited_buf +
561                             sdp->sd_log_commited_buf/503;
562         if (sdp->sd_log_commited_revoke)
563                 reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
564                                             sizeof(uint64_t));
565
566         old = sdp->sd_log_blks_free;
567         sdp->sd_log_blks_free += tr->tr_reserved -
568                                  (reserved - sdp->sd_log_blks_reserved);
569
570         gfs2_assert_withdraw(sdp,
571                              sdp->sd_log_blks_free >= old);
572         gfs2_assert_withdraw(sdp,
573                              sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
574
575         sdp->sd_log_blks_reserved = reserved;
576
577         gfs2_log_unlock(sdp);
578 }
579
580 /**
581  * gfs2_log_commit - Commit a transaction to the log
582  * @sdp: the filesystem
583  * @tr: the transaction
584  *
585  * Returns: errno
586  */
587
588 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
589 {
590         log_refund(sdp, tr);
591         lops_incore_commit(sdp, tr);
592
593         sdp->sd_vfs->s_dirt = 1;
594         unlock_from_trans(sdp);
595
596         kfree(tr);
597
598         gfs2_log_lock(sdp);
599         if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks)) {
600                 gfs2_log_unlock(sdp);
601                 gfs2_log_flush(sdp);
602         } else
603                 gfs2_log_unlock(sdp);
604 }
605
606 /**
607  * gfs2_log_shutdown - write a shutdown header into a journal
608  * @sdp: the filesystem
609  *
610  */
611
612 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
613 {
614         mutex_lock(&sdp->sd_log_flush_lock);
615
616         gfs2_assert_withdraw(sdp, !atomic_read(&sdp->sd_log_trans_count));
617         gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
618         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_gl);
619         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
620         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_jdata);
621         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
622         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
623         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
624         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
625
626         sdp->sd_log_flush_head = sdp->sd_log_head;
627         sdp->sd_log_flush_wrapped = 0;
628
629         log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT, 0);
630
631         gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free ==
632                              sdp->sd_jdesc->jd_blocks);
633         gfs2_assert_withdraw(sdp, sdp->sd_log_head == sdp->sd_log_tail);
634         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail2_list));
635
636         sdp->sd_log_head = sdp->sd_log_flush_head;
637         if (sdp->sd_log_flush_wrapped)
638                 sdp->sd_log_wraps++;
639         sdp->sd_log_tail = sdp->sd_log_head;
640
641         mutex_unlock(&sdp->sd_log_flush_lock);
642 }
643