Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/teigland/dlm
[linux-2.6] / fs / gfs2 / log.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2007 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 version 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 <linux/lm_interface.h>
18 #include <linux/delay.h>
19 #include <linux/kthread.h>
20 #include <linux/freezer.h>
21 #include <linux/bio.h>
22
23 #include "gfs2.h"
24 #include "incore.h"
25 #include "bmap.h"
26 #include "glock.h"
27 #include "log.h"
28 #include "lops.h"
29 #include "meta_io.h"
30 #include "util.h"
31 #include "dir.h"
32
33 #define PULL 1
34
35 /**
36  * gfs2_struct2blk - compute stuff
37  * @sdp: the filesystem
38  * @nstruct: the number of structures
39  * @ssize: the size of the structures
40  *
41  * Compute the number of log descriptor blocks needed to hold a certain number
42  * of structures of a certain size.
43  *
44  * Returns: the number of blocks needed (minimum is always 1)
45  */
46
47 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
48                              unsigned int ssize)
49 {
50         unsigned int blks;
51         unsigned int first, second;
52
53         blks = 1;
54         first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
55
56         if (nstruct > first) {
57                 second = (sdp->sd_sb.sb_bsize -
58                           sizeof(struct gfs2_meta_header)) / ssize;
59                 blks += DIV_ROUND_UP(nstruct - first, second);
60         }
61
62         return blks;
63 }
64
65 /**
66  * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
67  * @mapping: The associated mapping (maybe NULL)
68  * @bd: The gfs2_bufdata to remove
69  *
70  * The log lock _must_ be held when calling this function
71  *
72  */
73
74 void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
75 {
76         bd->bd_ail = NULL;
77         list_del_init(&bd->bd_ail_st_list);
78         list_del_init(&bd->bd_ail_gl_list);
79         atomic_dec(&bd->bd_gl->gl_ail_count);
80         brelse(bd->bd_bh);
81 }
82
83 /**
84  * gfs2_ail1_start_one - Start I/O on a part of the AIL
85  * @sdp: the filesystem
86  * @tr: the part of the AIL
87  *
88  */
89
90 static void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
91 __releases(&sdp->sd_log_lock)
92 __acquires(&sdp->sd_log_lock)
93 {
94         struct gfs2_bufdata *bd, *s;
95         struct buffer_head *bh;
96         int retry;
97
98         do {
99                 retry = 0;
100
101                 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
102                                                  bd_ail_st_list) {
103                         bh = bd->bd_bh;
104
105                         gfs2_assert(sdp, bd->bd_ail == ai);
106
107                         if (!buffer_busy(bh)) {
108                                 if (!buffer_uptodate(bh))
109                                         gfs2_io_error_bh(sdp, bh);
110                                 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
111                                 continue;
112                         }
113
114                         if (!buffer_dirty(bh))
115                                 continue;
116
117                         list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
118
119                         get_bh(bh);
120                         gfs2_log_unlock(sdp);
121                         lock_buffer(bh);
122                         if (test_clear_buffer_dirty(bh)) {
123                                 bh->b_end_io = end_buffer_write_sync;
124                                 submit_bh(WRITE, bh);
125                         } else {
126                                 unlock_buffer(bh);
127                                 brelse(bh);
128                         }
129                         gfs2_log_lock(sdp);
130
131                         retry = 1;
132                         break;
133                 }
134         } while (retry);
135 }
136
137 /**
138  * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
139  * @sdp: the filesystem
140  * @ai: the AIL entry
141  *
142  */
143
144 static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags)
145 {
146         struct gfs2_bufdata *bd, *s;
147         struct buffer_head *bh;
148
149         list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
150                                          bd_ail_st_list) {
151                 bh = bd->bd_bh;
152
153                 gfs2_assert(sdp, bd->bd_ail == ai);
154
155                 if (buffer_busy(bh)) {
156                         if (flags & DIO_ALL)
157                                 continue;
158                         else
159                                 break;
160                 }
161
162                 if (!buffer_uptodate(bh))
163                         gfs2_io_error_bh(sdp, bh);
164
165                 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
166         }
167
168         return list_empty(&ai->ai_ail1_list);
169 }
170
171 static void gfs2_ail1_start(struct gfs2_sbd *sdp, int flags)
172 {
173         struct list_head *head;
174         u64 sync_gen;
175         struct list_head *first;
176         struct gfs2_ail *first_ai, *ai, *tmp;
177         int done = 0;
178
179         gfs2_log_lock(sdp);
180         head = &sdp->sd_ail1_list;
181         if (list_empty(head)) {
182                 gfs2_log_unlock(sdp);
183                 return;
184         }
185         sync_gen = sdp->sd_ail_sync_gen++;
186
187         first = head->prev;
188         first_ai = list_entry(first, struct gfs2_ail, ai_list);
189         first_ai->ai_sync_gen = sync_gen;
190         gfs2_ail1_start_one(sdp, first_ai); /* This may drop log lock */
191
192         if (flags & DIO_ALL)
193                 first = NULL;
194
195         while(!done) {
196                 if (first && (head->prev != first ||
197                               gfs2_ail1_empty_one(sdp, first_ai, 0)))
198                         break;
199
200                 done = 1;
201                 list_for_each_entry_safe_reverse(ai, tmp, head, ai_list) {
202                         if (ai->ai_sync_gen >= sync_gen)
203                                 continue;
204                         ai->ai_sync_gen = sync_gen;
205                         gfs2_ail1_start_one(sdp, ai); /* This may drop log lock */
206                         done = 0;
207                         break;
208                 }
209         }
210
211         gfs2_log_unlock(sdp);
212 }
213
214 static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
215 {
216         struct gfs2_ail *ai, *s;
217         int ret;
218
219         gfs2_log_lock(sdp);
220
221         list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) {
222                 if (gfs2_ail1_empty_one(sdp, ai, flags))
223                         list_move(&ai->ai_list, &sdp->sd_ail2_list);
224                 else if (!(flags & DIO_ALL))
225                         break;
226         }
227
228         ret = list_empty(&sdp->sd_ail1_list);
229
230         gfs2_log_unlock(sdp);
231
232         return ret;
233 }
234
235
236 /**
237  * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
238  * @sdp: the filesystem
239  * @ai: the AIL entry
240  *
241  */
242
243 static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
244 {
245         struct list_head *head = &ai->ai_ail2_list;
246         struct gfs2_bufdata *bd;
247
248         while (!list_empty(head)) {
249                 bd = list_entry(head->prev, struct gfs2_bufdata,
250                                 bd_ail_st_list);
251                 gfs2_assert(sdp, bd->bd_ail == ai);
252                 gfs2_remove_from_ail(bd);
253         }
254 }
255
256 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
257 {
258         struct gfs2_ail *ai, *safe;
259         unsigned int old_tail = sdp->sd_log_tail;
260         int wrap = (new_tail < old_tail);
261         int a, b, rm;
262
263         gfs2_log_lock(sdp);
264
265         list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) {
266                 a = (old_tail <= ai->ai_first);
267                 b = (ai->ai_first < new_tail);
268                 rm = (wrap) ? (a || b) : (a && b);
269                 if (!rm)
270                         continue;
271
272                 gfs2_ail2_empty_one(sdp, ai);
273                 list_del(&ai->ai_list);
274                 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail1_list));
275                 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail2_list));
276                 kfree(ai);
277         }
278
279         gfs2_log_unlock(sdp);
280 }
281
282 /**
283  * gfs2_log_reserve - Make a log reservation
284  * @sdp: The GFS2 superblock
285  * @blks: The number of blocks to reserve
286  *
287  * Note that we never give out the last few blocks of the journal. Thats
288  * due to the fact that there is a small number of header blocks
289  * associated with each log flush. The exact number can't be known until
290  * flush time, so we ensure that we have just enough free blocks at all
291  * times to avoid running out during a log flush.
292  *
293  * Returns: errno
294  */
295
296 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
297 {
298         unsigned int try = 0;
299         unsigned reserved_blks = 6 * (4096 / sdp->sd_vfs->s_blocksize);
300
301         if (gfs2_assert_warn(sdp, blks) ||
302             gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
303                 return -EINVAL;
304
305         mutex_lock(&sdp->sd_log_reserve_mutex);
306         gfs2_log_lock(sdp);
307         while(atomic_read(&sdp->sd_log_blks_free) <= (blks + reserved_blks)) {
308                 gfs2_log_unlock(sdp);
309                 gfs2_ail1_empty(sdp, 0);
310                 gfs2_log_flush(sdp, NULL);
311
312                 if (try++)
313                         gfs2_ail1_start(sdp, 0);
314                 gfs2_log_lock(sdp);
315         }
316         atomic_sub(blks, &sdp->sd_log_blks_free);
317         gfs2_log_unlock(sdp);
318         mutex_unlock(&sdp->sd_log_reserve_mutex);
319
320         down_read(&sdp->sd_log_flush_lock);
321
322         return 0;
323 }
324
325 /**
326  * gfs2_log_release - Release a given number of log blocks
327  * @sdp: The GFS2 superblock
328  * @blks: The number of blocks
329  *
330  */
331
332 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
333 {
334
335         gfs2_log_lock(sdp);
336         atomic_add(blks, &sdp->sd_log_blks_free);
337         gfs2_assert_withdraw(sdp,
338                              atomic_read(&sdp->sd_log_blks_free) <= sdp->sd_jdesc->jd_blocks);
339         gfs2_log_unlock(sdp);
340         up_read(&sdp->sd_log_flush_lock);
341 }
342
343 static u64 log_bmap(struct gfs2_sbd *sdp, unsigned int lbn)
344 {
345         struct gfs2_journal_extent *je;
346
347         list_for_each_entry(je, &sdp->sd_jdesc->extent_list, extent_list) {
348                 if (lbn >= je->lblock && lbn < je->lblock + je->blocks)
349                         return je->dblock + lbn - je->lblock;
350         }
351
352         return -1;
353 }
354
355 /**
356  * log_distance - Compute distance between two journal blocks
357  * @sdp: The GFS2 superblock
358  * @newer: The most recent journal block of the pair
359  * @older: The older journal block of the pair
360  *
361  *   Compute the distance (in the journal direction) between two
362  *   blocks in the journal
363  *
364  * Returns: the distance in blocks
365  */
366
367 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
368                                         unsigned int older)
369 {
370         int dist;
371
372         dist = newer - older;
373         if (dist < 0)
374                 dist += sdp->sd_jdesc->jd_blocks;
375
376         return dist;
377 }
378
379 /**
380  * calc_reserved - Calculate the number of blocks to reserve when
381  *                 refunding a transaction's unused buffers.
382  * @sdp: The GFS2 superblock
383  *
384  * This is complex.  We need to reserve room for all our currently used
385  * metadata buffers (e.g. normal file I/O rewriting file time stamps) and 
386  * all our journaled data buffers for journaled files (e.g. files in the 
387  * meta_fs like rindex, or files for which chattr +j was done.)
388  * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
389  * will count it as free space (sd_log_blks_free) and corruption will follow.
390  *
391  * We can have metadata bufs and jdata bufs in the same journal.  So each
392  * type gets its own log header, for which we need to reserve a block.
393  * In fact, each type has the potential for needing more than one header 
394  * in cases where we have more buffers than will fit on a journal page.
395  * Metadata journal entries take up half the space of journaled buffer entries.
396  * Thus, metadata entries have buf_limit (502) and journaled buffers have
397  * databuf_limit (251) before they cause a wrap around.
398  *
399  * Also, we need to reserve blocks for revoke journal entries and one for an
400  * overall header for the lot.
401  *
402  * Returns: the number of blocks reserved
403  */
404 static unsigned int calc_reserved(struct gfs2_sbd *sdp)
405 {
406         unsigned int reserved = 0;
407         unsigned int mbuf_limit, metabufhdrs_needed;
408         unsigned int dbuf_limit, databufhdrs_needed;
409         unsigned int revokes = 0;
410
411         mbuf_limit = buf_limit(sdp);
412         metabufhdrs_needed = (sdp->sd_log_commited_buf +
413                               (mbuf_limit - 1)) / mbuf_limit;
414         dbuf_limit = databuf_limit(sdp);
415         databufhdrs_needed = (sdp->sd_log_commited_databuf +
416                               (dbuf_limit - 1)) / dbuf_limit;
417
418         if (sdp->sd_log_commited_revoke)
419                 revokes = gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
420                                           sizeof(u64));
421
422         reserved = sdp->sd_log_commited_buf + metabufhdrs_needed +
423                 sdp->sd_log_commited_databuf + databufhdrs_needed +
424                 revokes;
425         /* One for the overall header */
426         if (reserved)
427                 reserved++;
428         return reserved;
429 }
430
431 static unsigned int current_tail(struct gfs2_sbd *sdp)
432 {
433         struct gfs2_ail *ai;
434         unsigned int tail;
435
436         gfs2_log_lock(sdp);
437
438         if (list_empty(&sdp->sd_ail1_list)) {
439                 tail = sdp->sd_log_head;
440         } else {
441                 ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail, ai_list);
442                 tail = ai->ai_first;
443         }
444
445         gfs2_log_unlock(sdp);
446
447         return tail;
448 }
449
450 void gfs2_log_incr_head(struct gfs2_sbd *sdp)
451 {
452         if (sdp->sd_log_flush_head == sdp->sd_log_tail)
453                 BUG_ON(sdp->sd_log_flush_head != sdp->sd_log_head);
454
455         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
456                 sdp->sd_log_flush_head = 0;
457                 sdp->sd_log_flush_wrapped = 1;
458         }
459 }
460
461 /**
462  * gfs2_log_write_endio - End of I/O for a log buffer
463  * @bh: The buffer head
464  * @uptodate: I/O Status
465  *
466  */
467
468 static void gfs2_log_write_endio(struct buffer_head *bh, int uptodate)
469 {
470         struct gfs2_sbd *sdp = bh->b_private;
471         bh->b_private = NULL;
472
473         end_buffer_write_sync(bh, uptodate);
474         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
475                 wake_up(&sdp->sd_log_flush_wait);
476 }
477
478 /**
479  * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
480  * @sdp: The GFS2 superblock
481  *
482  * Returns: the buffer_head
483  */
484
485 struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
486 {
487         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
488         struct buffer_head *bh;
489
490         bh = sb_getblk(sdp->sd_vfs, blkno);
491         lock_buffer(bh);
492         memset(bh->b_data, 0, bh->b_size);
493         set_buffer_uptodate(bh);
494         clear_buffer_dirty(bh);
495         gfs2_log_incr_head(sdp);
496         atomic_inc(&sdp->sd_log_in_flight);
497         bh->b_private = sdp;
498         bh->b_end_io = gfs2_log_write_endio;
499
500         return bh;
501 }
502
503 /**
504  * gfs2_fake_write_endio - 
505  * @bh: The buffer head
506  * @uptodate: The I/O Status
507  *
508  */
509
510 static void gfs2_fake_write_endio(struct buffer_head *bh, int uptodate)
511 {
512         struct buffer_head *real_bh = bh->b_private;
513         struct gfs2_bufdata *bd = real_bh->b_private;
514         struct gfs2_sbd *sdp = bd->bd_gl->gl_sbd;
515
516         end_buffer_write_sync(bh, uptodate);
517         free_buffer_head(bh);
518         unlock_buffer(real_bh);
519         brelse(real_bh);
520         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
521                 wake_up(&sdp->sd_log_flush_wait);
522 }
523
524 /**
525  * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
526  * @sdp: the filesystem
527  * @data: the data the buffer_head should point to
528  *
529  * Returns: the log buffer descriptor
530  */
531
532 struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
533                                       struct buffer_head *real)
534 {
535         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
536         struct buffer_head *bh;
537
538         bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL);
539         atomic_set(&bh->b_count, 1);
540         bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate) | (1 << BH_Lock);
541         set_bh_page(bh, real->b_page, bh_offset(real));
542         bh->b_blocknr = blkno;
543         bh->b_size = sdp->sd_sb.sb_bsize;
544         bh->b_bdev = sdp->sd_vfs->s_bdev;
545         bh->b_private = real;
546         bh->b_end_io = gfs2_fake_write_endio;
547
548         gfs2_log_incr_head(sdp);
549         atomic_inc(&sdp->sd_log_in_flight);
550
551         return bh;
552 }
553
554 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
555 {
556         unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
557
558         ail2_empty(sdp, new_tail);
559
560         gfs2_log_lock(sdp);
561         atomic_add(dist, &sdp->sd_log_blks_free);
562         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= sdp->sd_jdesc->jd_blocks);
563         gfs2_log_unlock(sdp);
564
565         sdp->sd_log_tail = new_tail;
566 }
567
568 /**
569  * log_write_header - Get and initialize a journal header buffer
570  * @sdp: The GFS2 superblock
571  *
572  * Returns: the initialized log buffer descriptor
573  */
574
575 static void log_write_header(struct gfs2_sbd *sdp, u32 flags, int pull)
576 {
577         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
578         struct buffer_head *bh;
579         struct gfs2_log_header *lh;
580         unsigned int tail;
581         u32 hash;
582
583         bh = sb_getblk(sdp->sd_vfs, blkno);
584         lock_buffer(bh);
585         memset(bh->b_data, 0, bh->b_size);
586         set_buffer_uptodate(bh);
587         clear_buffer_dirty(bh);
588
589         gfs2_ail1_empty(sdp, 0);
590         tail = current_tail(sdp);
591
592         lh = (struct gfs2_log_header *)bh->b_data;
593         memset(lh, 0, sizeof(struct gfs2_log_header));
594         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
595         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
596         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
597         lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++);
598         lh->lh_flags = cpu_to_be32(flags);
599         lh->lh_tail = cpu_to_be32(tail);
600         lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
601         hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
602         lh->lh_hash = cpu_to_be32(hash);
603
604         bh->b_end_io = end_buffer_write_sync;
605         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
606                 goto skip_barrier;
607         get_bh(bh);
608         submit_bh(WRITE_BARRIER | (1 << BIO_RW_META), bh);
609         wait_on_buffer(bh);
610         if (buffer_eopnotsupp(bh)) {
611                 clear_buffer_eopnotsupp(bh);
612                 set_buffer_uptodate(bh);
613                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
614                 lock_buffer(bh);
615 skip_barrier:
616                 get_bh(bh);
617                 submit_bh(WRITE_SYNC | (1 << BIO_RW_META), bh);
618                 wait_on_buffer(bh);
619         }
620         if (!buffer_uptodate(bh))
621                 gfs2_io_error_bh(sdp, bh);
622         brelse(bh);
623
624         if (sdp->sd_log_tail != tail)
625                 log_pull_tail(sdp, tail);
626         else
627                 gfs2_assert_withdraw(sdp, !pull);
628
629         sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
630         gfs2_log_incr_head(sdp);
631 }
632
633 static void log_flush_commit(struct gfs2_sbd *sdp)
634 {
635         DEFINE_WAIT(wait);
636
637         if (atomic_read(&sdp->sd_log_in_flight)) {
638                 do {
639                         prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
640                                         TASK_UNINTERRUPTIBLE);
641                         if (atomic_read(&sdp->sd_log_in_flight))
642                                 io_schedule();
643                 } while(atomic_read(&sdp->sd_log_in_flight));
644                 finish_wait(&sdp->sd_log_flush_wait, &wait);
645         }
646
647         log_write_header(sdp, 0, 0);
648 }
649
650 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
651 {
652         struct gfs2_bufdata *bd;
653         struct buffer_head *bh;
654         LIST_HEAD(written);
655
656         gfs2_log_lock(sdp);
657         while (!list_empty(&sdp->sd_log_le_ordered)) {
658                 bd = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_bufdata, bd_le.le_list);
659                 list_move(&bd->bd_le.le_list, &written);
660                 bh = bd->bd_bh;
661                 if (!buffer_dirty(bh))
662                         continue;
663                 get_bh(bh);
664                 gfs2_log_unlock(sdp);
665                 lock_buffer(bh);
666                 if (buffer_mapped(bh) && test_clear_buffer_dirty(bh)) {
667                         bh->b_end_io = end_buffer_write_sync;
668                         submit_bh(WRITE, bh);
669                 } else {
670                         unlock_buffer(bh);
671                         brelse(bh);
672                 }
673                 gfs2_log_lock(sdp);
674         }
675         list_splice(&written, &sdp->sd_log_le_ordered);
676         gfs2_log_unlock(sdp);
677 }
678
679 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
680 {
681         struct gfs2_bufdata *bd;
682         struct buffer_head *bh;
683
684         gfs2_log_lock(sdp);
685         while (!list_empty(&sdp->sd_log_le_ordered)) {
686                 bd = list_entry(sdp->sd_log_le_ordered.prev, struct gfs2_bufdata, bd_le.le_list);
687                 bh = bd->bd_bh;
688                 if (buffer_locked(bh)) {
689                         get_bh(bh);
690                         gfs2_log_unlock(sdp);
691                         wait_on_buffer(bh);
692                         brelse(bh);
693                         gfs2_log_lock(sdp);
694                         continue;
695                 }
696                 list_del_init(&bd->bd_le.le_list);
697         }
698         gfs2_log_unlock(sdp);
699 }
700
701 /**
702  * gfs2_log_flush - flush incore transaction(s)
703  * @sdp: the filesystem
704  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
705  *
706  */
707
708 void __gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
709 {
710         struct gfs2_ail *ai;
711
712         down_write(&sdp->sd_log_flush_lock);
713
714         /* Log might have been flushed while we waited for the flush lock */
715         if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
716                 up_write(&sdp->sd_log_flush_lock);
717                 return;
718         }
719
720         ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
721         INIT_LIST_HEAD(&ai->ai_ail1_list);
722         INIT_LIST_HEAD(&ai->ai_ail2_list);
723
724         if (sdp->sd_log_num_buf != sdp->sd_log_commited_buf) {
725                 printk(KERN_INFO "GFS2: log buf %u %u\n", sdp->sd_log_num_buf,
726                        sdp->sd_log_commited_buf);
727                 gfs2_assert_withdraw(sdp, 0);
728         }
729         if (sdp->sd_log_num_databuf != sdp->sd_log_commited_databuf) {
730                 printk(KERN_INFO "GFS2: log databuf %u %u\n",
731                        sdp->sd_log_num_databuf, sdp->sd_log_commited_databuf);
732                 gfs2_assert_withdraw(sdp, 0);
733         }
734         gfs2_assert_withdraw(sdp,
735                         sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
736
737         sdp->sd_log_flush_head = sdp->sd_log_head;
738         sdp->sd_log_flush_wrapped = 0;
739         ai->ai_first = sdp->sd_log_flush_head;
740
741         gfs2_ordered_write(sdp);
742         lops_before_commit(sdp);
743         gfs2_ordered_wait(sdp);
744
745         if (sdp->sd_log_head != sdp->sd_log_flush_head)
746                 log_flush_commit(sdp);
747         else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
748                 gfs2_log_lock(sdp);
749                 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
750                 gfs2_log_unlock(sdp);
751                 log_write_header(sdp, 0, PULL);
752         }
753         lops_after_commit(sdp, ai);
754
755         gfs2_log_lock(sdp);
756         sdp->sd_log_head = sdp->sd_log_flush_head;
757         sdp->sd_log_blks_reserved = 0;
758         sdp->sd_log_commited_buf = 0;
759         sdp->sd_log_commited_databuf = 0;
760         sdp->sd_log_commited_revoke = 0;
761
762         if (!list_empty(&ai->ai_ail1_list)) {
763                 list_add(&ai->ai_list, &sdp->sd_ail1_list);
764                 ai = NULL;
765         }
766         gfs2_log_unlock(sdp);
767
768         sdp->sd_vfs->s_dirt = 0;
769         up_write(&sdp->sd_log_flush_lock);
770
771         kfree(ai);
772 }
773
774 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
775 {
776         unsigned int reserved;
777         unsigned int unused;
778
779         gfs2_log_lock(sdp);
780
781         sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
782         sdp->sd_log_commited_databuf += tr->tr_num_databuf_new -
783                 tr->tr_num_databuf_rm;
784         gfs2_assert_withdraw(sdp, (((int)sdp->sd_log_commited_buf) >= 0) ||
785                              (((int)sdp->sd_log_commited_databuf) >= 0));
786         sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
787         gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0);
788         reserved = calc_reserved(sdp);
789         gfs2_assert_withdraw(sdp, sdp->sd_log_blks_reserved + tr->tr_reserved >= reserved);
790         unused = sdp->sd_log_blks_reserved - reserved + tr->tr_reserved;
791         atomic_add(unused, &sdp->sd_log_blks_free);
792         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
793                              sdp->sd_jdesc->jd_blocks);
794         sdp->sd_log_blks_reserved = reserved;
795
796         gfs2_log_unlock(sdp);
797 }
798
799 static void buf_lo_incore_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
800 {
801         struct list_head *head = &tr->tr_list_buf;
802         struct gfs2_bufdata *bd;
803
804         gfs2_log_lock(sdp);
805         while (!list_empty(head)) {
806                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list_tr);
807                 list_del_init(&bd->bd_list_tr);
808                 tr->tr_num_buf--;
809         }
810         gfs2_log_unlock(sdp);
811         gfs2_assert_warn(sdp, !tr->tr_num_buf);
812 }
813
814 /**
815  * gfs2_log_commit - Commit a transaction to the log
816  * @sdp: the filesystem
817  * @tr: the transaction
818  *
819  * Returns: errno
820  */
821
822 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
823 {
824         log_refund(sdp, tr);
825         buf_lo_incore_commit(sdp, tr);
826
827         sdp->sd_vfs->s_dirt = 1;
828         up_read(&sdp->sd_log_flush_lock);
829
830         gfs2_log_lock(sdp);
831         if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks))
832                 wake_up_process(sdp->sd_logd_process);
833         gfs2_log_unlock(sdp);
834 }
835
836 /**
837  * gfs2_log_shutdown - write a shutdown header into a journal
838  * @sdp: the filesystem
839  *
840  */
841
842 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
843 {
844         down_write(&sdp->sd_log_flush_lock);
845
846         gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
847         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
848         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
849         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
850         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
851         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
852
853         sdp->sd_log_flush_head = sdp->sd_log_head;
854         sdp->sd_log_flush_wrapped = 0;
855
856         log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT,
857                          (sdp->sd_log_tail == current_tail(sdp)) ? 0 : PULL);
858
859         gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
860         gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
861         gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
862
863         sdp->sd_log_head = sdp->sd_log_flush_head;
864         sdp->sd_log_tail = sdp->sd_log_head;
865
866         up_write(&sdp->sd_log_flush_lock);
867 }
868
869
870 /**
871  * gfs2_meta_syncfs - sync all the buffers in a filesystem
872  * @sdp: the filesystem
873  *
874  */
875
876 void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
877 {
878         gfs2_log_flush(sdp, NULL);
879         for (;;) {
880                 gfs2_ail1_start(sdp, DIO_ALL);
881                 if (gfs2_ail1_empty(sdp, DIO_ALL))
882                         break;
883                 msleep(10);
884         }
885 }
886
887
888 /**
889  * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
890  * @sdp: Pointer to GFS2 superblock
891  *
892  * Also, periodically check to make sure that we're using the most recent
893  * journal index.
894  */
895
896 int gfs2_logd(void *data)
897 {
898         struct gfs2_sbd *sdp = data;
899         unsigned long t;
900         int need_flush;
901
902         while (!kthread_should_stop()) {
903                 /* Advance the log tail */
904
905                 t = sdp->sd_log_flush_time +
906                     gfs2_tune_get(sdp, gt_log_flush_secs) * HZ;
907
908                 gfs2_ail1_empty(sdp, DIO_ALL);
909                 gfs2_log_lock(sdp);
910                 need_flush = sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks);
911                 gfs2_log_unlock(sdp);
912                 if (need_flush || time_after_eq(jiffies, t)) {
913                         gfs2_log_flush(sdp, NULL);
914                         sdp->sd_log_flush_time = jiffies;
915                 }
916
917                 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
918                 if (freezing(current))
919                         refrigerator();
920                 schedule_timeout_interruptible(t);
921         }
922
923         return 0;
924 }
925