[GFS2] Finally get ref counting correct
[linux-2.6.git] / 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 <linux/gfs2_ondisk.h>
16 #include <linux/crc32.h>
17 #include <asm/semaphore.h>
18
19 #include "gfs2.h"
20 #include "lm_interface.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"
29
30 #define PULL 1
31
32 /**
33  * gfs2_struct2blk - compute stuff
34  * @sdp: the filesystem
35  * @nstruct: the number of structures
36  * @ssize: the size of the structures
37  *
38  * Compute the number of log descriptor blocks needed to hold a certain number
39  * of structures of a certain size.
40  *
41  * Returns: the number of blocks needed (minimum is always 1)
42  */
43
44 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
45                              unsigned int ssize)
46 {
47         unsigned int blks;
48         unsigned int first, second;
49
50         blks = 1;
51         first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) /
52                 ssize;
53
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);
58         }
59
60         return blks;
61 }
62
63 void gfs2_ail1_start(struct gfs2_sbd *sdp, int flags)
64 {
65         struct list_head *head = &sdp->sd_ail1_list;
66         uint64_t sync_gen;
67         struct list_head *first, *tmp;
68         struct gfs2_ail *first_ai, *ai;
69
70         gfs2_log_lock(sdp);
71         if (list_empty(head)) {
72                 gfs2_log_unlock(sdp);
73                 return;
74         }
75         sync_gen = sdp->sd_ail_sync_gen++;
76
77         first = head->prev;
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);
81
82         if (flags & DIO_ALL)
83                 first = NULL;
84
85         for (;;) {
86                 if (first && (head->prev != first ||
87                               gfs2_ail1_empty_one(sdp, first_ai, 0)))
88                         break;
89
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)
93                                 continue;
94                         ai->ai_sync_gen = sync_gen;
95                         gfs2_ail1_start_one(sdp, ai);
96                         break;
97                 }
98
99                 if (tmp == head)
100                         break;
101         }
102
103         gfs2_log_unlock(sdp);
104 }
105
106 int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
107 {
108         struct gfs2_ail *ai, *s;
109         int ret;
110
111         gfs2_log_lock(sdp);
112
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))
117                         break;
118         }
119
120         ret = list_empty(&sdp->sd_ail1_list);
121
122         gfs2_log_unlock(sdp);
123
124         return ret;
125 }
126
127 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
128 {
129         struct gfs2_ail *ai, *safe;
130         unsigned int old_tail = sdp->sd_log_tail;
131         int wrap = (new_tail < old_tail);
132         int a, b, rm;
133
134         gfs2_log_lock(sdp);
135
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);
140                 if (!rm)
141                         continue;
142
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));
147                 kfree(ai);
148         }
149
150         gfs2_log_unlock(sdp);
151 }
152
153 /**
154  * gfs2_log_reserve - Make a log reservation
155  * @sdp: The GFS2 superblock
156  * @blks: The number of blocks to reserve
157  *
158  * Returns: errno
159  */
160
161 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
162 {
163         unsigned int try = 0;
164
165         if (gfs2_assert_warn(sdp, blks) ||
166             gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
167                 return -EINVAL;
168
169         mutex_lock(&sdp->sd_log_reserve_mutex);
170         gfs2_log_lock(sdp);
171         while(sdp->sd_log_blks_free <= blks) {
172                 gfs2_log_unlock(sdp);
173                 gfs2_ail1_empty(sdp, 0);
174                 gfs2_log_flush(sdp, NULL);
175
176                 if (try++)
177                         gfs2_ail1_start(sdp, 0);
178                 gfs2_log_lock(sdp);
179         }
180         sdp->sd_log_blks_free -= blks;
181         /* printk(KERN_INFO "reserved %u blocks (%u left)\n", blks, sdp->sd_log_blks_free); */
182         gfs2_log_unlock(sdp);
183         mutex_unlock(&sdp->sd_log_reserve_mutex);
184
185         down_read(&sdp->sd_log_flush_lock);
186
187         return 0;
188 }
189
190 /**
191  * gfs2_log_release - Release a given number of log blocks
192  * @sdp: The GFS2 superblock
193  * @blks: The number of blocks
194  *
195  */
196
197 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
198 {
199
200         gfs2_log_lock(sdp);
201         sdp->sd_log_blks_free += blks;
202         /* printk(KERN_INFO "released %u blocks (%u left)\n", blks, sdp->sd_log_blks_free); */
203         gfs2_assert_withdraw(sdp,
204                              sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
205         gfs2_log_unlock(sdp);
206         up_read(&sdp->sd_log_flush_lock);
207 }
208
209 static uint64_t log_bmap(struct gfs2_sbd *sdp, unsigned int lbn)
210 {
211         int new = 0;
212         uint64_t dbn;
213         int error;
214
215         error = gfs2_block_map(sdp->sd_jdesc->jd_inode->u.generic_ip,
216                                lbn, &new, &dbn, NULL);
217         gfs2_assert_withdraw(sdp, !error && dbn);
218
219         return dbn;
220 }
221
222 /**
223  * log_distance - Compute distance between two journal blocks
224  * @sdp: The GFS2 superblock
225  * @newer: The most recent journal block of the pair
226  * @older: The older journal block of the pair
227  *
228  *   Compute the distance (in the journal direction) between two
229  *   blocks in the journal
230  *
231  * Returns: the distance in blocks
232  */
233
234 static inline unsigned int log_distance(struct gfs2_sbd *sdp,
235                                         unsigned int newer,
236                                         unsigned int older)
237 {
238         int dist;
239
240         dist = newer - older;
241         if (dist < 0)
242                 dist += sdp->sd_jdesc->jd_blocks;
243
244         return dist;
245 }
246
247 static unsigned int current_tail(struct gfs2_sbd *sdp)
248 {
249         struct gfs2_ail *ai;
250         unsigned int tail;
251
252         gfs2_log_lock(sdp);
253
254         if (list_empty(&sdp->sd_ail1_list))
255                 tail = sdp->sd_log_head;
256         else {
257                 ai = list_entry(sdp->sd_ail1_list.prev,
258                                 struct gfs2_ail, ai_list);
259                 tail = ai->ai_first;
260         }
261
262         gfs2_log_unlock(sdp);
263
264         return tail;
265 }
266
267 static inline void log_incr_head(struct gfs2_sbd *sdp)
268 {
269         if (sdp->sd_log_flush_head == sdp->sd_log_tail)
270                 gfs2_assert_withdraw(sdp,
271                                 sdp->sd_log_flush_head == sdp->sd_log_head);
272
273         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
274                 sdp->sd_log_flush_head = 0;
275                 sdp->sd_log_flush_wrapped = 1;
276         }
277 }
278
279 /**
280  * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
281  * @sdp: The GFS2 superblock
282  *
283  * Returns: the buffer_head
284  */
285
286 struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
287 {
288         uint64_t blkno = log_bmap(sdp, sdp->sd_log_flush_head);
289         struct gfs2_log_buf *lb;
290         struct buffer_head *bh;
291
292         lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
293         list_add(&lb->lb_list, &sdp->sd_log_flush_list);
294
295         bh = lb->lb_bh = sb_getblk(sdp->sd_vfs, blkno);
296         lock_buffer(bh);
297         memset(bh->b_data, 0, bh->b_size);
298         set_buffer_uptodate(bh);
299         clear_buffer_dirty(bh);
300         unlock_buffer(bh);
301
302         log_incr_head(sdp);
303
304         return bh;
305 }
306
307 /**
308  * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
309  * @sdp: the filesystem
310  * @data: the data the buffer_head should point to
311  *
312  * Returns: the log buffer descriptor
313  */
314
315 struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
316                                       struct buffer_head *real)
317 {
318         uint64_t blkno = log_bmap(sdp, sdp->sd_log_flush_head);
319         struct gfs2_log_buf *lb;
320         struct buffer_head *bh;
321
322         lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
323         list_add(&lb->lb_list, &sdp->sd_log_flush_list);
324         lb->lb_real = real;
325
326         bh = lb->lb_bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL);
327         atomic_set(&bh->b_count, 1);
328         bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate);
329         set_bh_page(bh, real->b_page, bh_offset(real));
330         bh->b_blocknr = blkno;
331         bh->b_size = sdp->sd_sb.sb_bsize;
332         bh->b_bdev = sdp->sd_vfs->s_bdev;
333
334         log_incr_head(sdp);
335
336         return bh;
337 }
338
339 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail, int pull)
340 {
341         unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
342
343         ail2_empty(sdp, new_tail);
344
345         gfs2_log_lock(sdp);
346         sdp->sd_log_blks_free += dist - ((pull) ? 1 : 0);
347         /* printk(KERN_INFO "pull tail refunding %u blocks (%u left) pull=%d\n", dist - ((pull) ? 1 : 0), sdp->sd_log_blks_free, pull); */
348         gfs2_assert_withdraw(sdp,
349                              sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
350         gfs2_log_unlock(sdp);
351
352         sdp->sd_log_tail = new_tail;
353 }
354
355 /**
356  * log_write_header - Get and initialize a journal header buffer
357  * @sdp: The GFS2 superblock
358  *
359  * Returns: the initialized log buffer descriptor
360  */
361
362 static void log_write_header(struct gfs2_sbd *sdp, uint32_t flags, int pull)
363 {
364         uint64_t blkno = log_bmap(sdp, sdp->sd_log_flush_head);
365         struct buffer_head *bh;
366         struct gfs2_log_header *lh;
367         unsigned int tail;
368         uint32_t hash;
369
370         /* printk(KERN_INFO "log write header start (flags=%08x, pull=%d)\n", flags, pull); */
371
372         bh = sb_getblk(sdp->sd_vfs, blkno);
373         lock_buffer(bh);
374         memset(bh->b_data, 0, bh->b_size);
375         set_buffer_uptodate(bh);
376         clear_buffer_dirty(bh);
377         unlock_buffer(bh);
378
379         gfs2_ail1_empty(sdp, 0);
380         tail = current_tail(sdp);
381
382         lh = (struct gfs2_log_header *)bh->b_data;
383         memset(lh, 0, sizeof(struct gfs2_log_header));
384         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
385         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
386         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
387         lh->lh_sequence = be64_to_cpu(sdp->sd_log_sequence++);
388         lh->lh_flags = be32_to_cpu(flags);
389         lh->lh_tail = be32_to_cpu(tail);
390         lh->lh_blkno = be32_to_cpu(sdp->sd_log_flush_head);
391         hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
392         lh->lh_hash = cpu_to_be32(hash);
393
394         set_buffer_dirty(bh);
395         if (sync_dirty_buffer(bh))
396                 gfs2_io_error_bh(sdp, bh);
397         brelse(bh);
398
399         if (sdp->sd_log_tail != tail)
400                 log_pull_tail(sdp, tail, pull);
401         else
402                 gfs2_assert_withdraw(sdp, !pull);
403
404         sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
405         log_incr_head(sdp);
406
407         /* printk(KERN_INFO "log write header out\n"); */
408 }
409
410 static void log_flush_commit(struct gfs2_sbd *sdp)
411 {
412         struct list_head *head = &sdp->sd_log_flush_list;
413         struct gfs2_log_buf *lb;
414         struct buffer_head *bh;
415         unsigned int d;
416
417         d = log_distance(sdp, sdp->sd_log_flush_head, sdp->sd_log_head);
418
419         gfs2_assert_withdraw(sdp, d + 1 == sdp->sd_log_blks_reserved);
420
421         while (!list_empty(head)) {
422                 lb = list_entry(head->next, struct gfs2_log_buf, lb_list);
423                 list_del(&lb->lb_list);
424                 bh = lb->lb_bh;
425
426                 wait_on_buffer(bh);
427                 if (!buffer_uptodate(bh))
428                         gfs2_io_error_bh(sdp, bh);
429                 if (lb->lb_real) {
430                         while (atomic_read(&bh->b_count) != 1)  /* Grrrr... */
431                                 schedule();
432                         free_buffer_head(bh);
433                 } else
434                         brelse(bh);
435                 kfree(lb);
436         }
437
438         log_write_header(sdp, 0, 0);
439 }
440
441 /**
442  * gfs2_log_flush - flush incore transaction(s)
443  * @sdp: the filesystem
444  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
445  *
446  */
447
448 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
449 {
450         struct gfs2_ail *ai;
451
452         down_write(&sdp->sd_log_flush_lock);
453
454         if (gl) {
455                 gfs2_log_lock(sdp);
456                 if (list_empty(&gl->gl_le.le_list)) {
457                         gfs2_log_unlock(sdp);
458                         up_write(&sdp->sd_log_flush_lock);
459                         return;
460                 }
461                 gfs2_log_unlock(sdp);
462         }
463
464         ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
465         INIT_LIST_HEAD(&ai->ai_ail1_list);
466         INIT_LIST_HEAD(&ai->ai_ail2_list);
467
468         gfs2_assert_withdraw(sdp,
469                         sdp->sd_log_num_buf == sdp->sd_log_commited_buf);
470         gfs2_assert_withdraw(sdp,
471                         sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
472
473         sdp->sd_log_flush_head = sdp->sd_log_head;
474         sdp->sd_log_flush_wrapped = 0;
475         ai->ai_first = sdp->sd_log_flush_head;
476
477         lops_before_commit(sdp);
478         if (!list_empty(&sdp->sd_log_flush_list))
479                 log_flush_commit(sdp);
480         else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle)
481                 log_write_header(sdp, 0, PULL);
482         lops_after_commit(sdp, ai);
483         sdp->sd_log_head = sdp->sd_log_flush_head;
484
485         /* printk(KERN_INFO "sd_log_num_hdrs %u\n", sdp->sd_log_num_hdrs); */
486
487         sdp->sd_log_blks_reserved =
488                 sdp->sd_log_commited_buf =
489                 sdp->sd_log_num_hdrs = 
490                 sdp->sd_log_commited_revoke = 0;
491
492         gfs2_log_lock(sdp);
493         if (!list_empty(&ai->ai_ail1_list)) {
494                 list_add(&ai->ai_list, &sdp->sd_ail1_list);
495                 ai = NULL;
496         }
497         gfs2_log_unlock(sdp);
498
499         sdp->sd_vfs->s_dirt = 0;
500         up_write(&sdp->sd_log_flush_lock);
501
502         kfree(ai);
503 }
504
505 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
506 {
507         unsigned int reserved = 1;
508         unsigned int old;
509
510         gfs2_log_lock(sdp);
511
512         sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
513         gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_buf) >= 0);
514         sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
515         gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0);
516
517         if (sdp->sd_log_commited_buf)
518                 reserved += 1 + sdp->sd_log_commited_buf +
519                             sdp->sd_log_commited_buf/503;
520         if (sdp->sd_log_commited_revoke)
521                 reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
522                                             sizeof(uint64_t));
523
524         old = sdp->sd_log_blks_free;
525         sdp->sd_log_blks_free += tr->tr_reserved -
526                                  (reserved - sdp->sd_log_blks_reserved);
527
528         gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free >= old);
529         gfs2_assert_withdraw(sdp,
530                              sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
531
532         sdp->sd_log_blks_reserved = reserved;
533
534         gfs2_log_unlock(sdp);
535 }
536
537 /**
538  * gfs2_log_commit - Commit a transaction to the log
539  * @sdp: the filesystem
540  * @tr: the transaction
541  *
542  * Returns: errno
543  */
544
545 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
546 {
547         log_refund(sdp, tr);
548         lops_incore_commit(sdp, tr);
549
550         sdp->sd_vfs->s_dirt = 1;
551         up_read(&sdp->sd_log_flush_lock);
552
553         gfs2_log_lock(sdp);
554         if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks)) {
555                 gfs2_log_unlock(sdp);
556                 gfs2_log_flush(sdp, NULL);
557         } else
558                 gfs2_log_unlock(sdp);
559 }
560
561 /**
562  * gfs2_log_shutdown - write a shutdown header into a journal
563  * @sdp: the filesystem
564  *
565  */
566
567 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
568 {
569         down_write(&sdp->sd_log_flush_lock);
570
571         gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
572         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_gl);
573         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
574         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_jdata);
575         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
576         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
577         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
578         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
579
580         sdp->sd_log_flush_head = sdp->sd_log_head;
581         sdp->sd_log_flush_wrapped = 0;
582
583         log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT, 0);
584
585         gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free ==
586                              sdp->sd_jdesc->jd_blocks);
587         gfs2_assert_withdraw(sdp, sdp->sd_log_head == sdp->sd_log_tail);
588         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail2_list));
589
590         sdp->sd_log_head = sdp->sd_log_flush_head;
591         sdp->sd_log_tail = sdp->sd_log_head;
592
593         up_write(&sdp->sd_log_flush_lock);
594 }
595