]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/gfs2/log.c
[GFS2] Correct lock ordering in unlink
[mirror_ubuntu-bionic-kernel.git] / fs / gfs2 / log.c
CommitLineData
b3b94faa
DT
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3a8a9a10 3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
b3b94faa
DT
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
e9fc2aa0 7 * of the GNU General Public License version 2.
b3b94faa
DT
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>
5c676f6d 15#include <linux/gfs2_ondisk.h>
71b86f56 16#include <linux/crc32.h>
7d308590 17#include <linux/lm_interface.h>
a25311c8 18#include <linux/delay.h>
b3b94faa
DT
19
20#include "gfs2.h"
5c676f6d 21#include "incore.h"
b3b94faa
DT
22#include "bmap.h"
23#include "glock.h"
24#include "log.h"
25#include "lops.h"
26#include "meta_io.h"
5c676f6d 27#include "util.h"
71b86f56 28#include "dir.h"
b3b94faa
DT
29
30#define PULL 1
31
b3b94faa
DT
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
44unsigned 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;
faa31ce8 51 first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
b3b94faa
DT
52
53 if (nstruct > first) {
568f4c96
SW
54 second = (sdp->sd_sb.sb_bsize -
55 sizeof(struct gfs2_meta_header)) / ssize;
5c676f6d 56 blks += DIV_ROUND_UP(nstruct - first, second);
b3b94faa
DT
57 }
58
59 return blks;
60}
61
ddacfaf7
SW
62/**
63 * gfs2_ail1_start_one - Start I/O on a part of the AIL
64 * @sdp: the filesystem
65 * @tr: the part of the AIL
66 *
67 */
68
69static void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
70{
71 struct gfs2_bufdata *bd, *s;
72 struct buffer_head *bh;
73 int retry;
74
75 BUG_ON(!spin_is_locked(&sdp->sd_log_lock));
76
77 do {
78 retry = 0;
79
80 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
81 bd_ail_st_list) {
82 bh = bd->bd_bh;
83
84 gfs2_assert(sdp, bd->bd_ail == ai);
85
86 if (!buffer_busy(bh)) {
87 if (!buffer_uptodate(bh)) {
88 gfs2_log_unlock(sdp);
89 gfs2_io_error_bh(sdp, bh);
90 gfs2_log_lock(sdp);
91 }
92 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
93 continue;
94 }
95
96 if (!buffer_dirty(bh))
97 continue;
98
99 list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
100
101 gfs2_log_unlock(sdp);
102 wait_on_buffer(bh);
103 ll_rw_block(WRITE, 1, &bh);
104 gfs2_log_lock(sdp);
105
106 retry = 1;
107 break;
108 }
109 } while (retry);
110}
111
112/**
113 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
114 * @sdp: the filesystem
115 * @ai: the AIL entry
116 *
117 */
118
119static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags)
120{
121 struct gfs2_bufdata *bd, *s;
122 struct buffer_head *bh;
123
124 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
125 bd_ail_st_list) {
126 bh = bd->bd_bh;
127
128 gfs2_assert(sdp, bd->bd_ail == ai);
129
130 if (buffer_busy(bh)) {
131 if (flags & DIO_ALL)
132 continue;
133 else
134 break;
135 }
136
137 if (!buffer_uptodate(bh))
138 gfs2_io_error_bh(sdp, bh);
139
140 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
141 }
142
143 return list_empty(&ai->ai_ail1_list);
144}
145
a25311c8 146static void gfs2_ail1_start(struct gfs2_sbd *sdp, int flags)
b3b94faa 147{
693ddeab 148 struct list_head *head;
cd915493 149 u64 sync_gen;
74669416
SW
150 struct list_head *first;
151 struct gfs2_ail *first_ai, *ai, *tmp;
152 int done = 0;
b3b94faa
DT
153
154 gfs2_log_lock(sdp);
693ddeab 155 head = &sdp->sd_ail1_list;
b3b94faa
DT
156 if (list_empty(head)) {
157 gfs2_log_unlock(sdp);
158 return;
159 }
160 sync_gen = sdp->sd_ail_sync_gen++;
161
162 first = head->prev;
163 first_ai = list_entry(first, struct gfs2_ail, ai_list);
164 first_ai->ai_sync_gen = sync_gen;
74669416 165 gfs2_ail1_start_one(sdp, first_ai); /* This may drop log lock */
b3b94faa
DT
166
167 if (flags & DIO_ALL)
168 first = NULL;
169
74669416 170 while(!done) {
484adff8
SW
171 if (first && (head->prev != first ||
172 gfs2_ail1_empty_one(sdp, first_ai, 0)))
b3b94faa
DT
173 break;
174
74669416
SW
175 done = 1;
176 list_for_each_entry_safe_reverse(ai, tmp, head, ai_list) {
b3b94faa
DT
177 if (ai->ai_sync_gen >= sync_gen)
178 continue;
179 ai->ai_sync_gen = sync_gen;
74669416
SW
180 gfs2_ail1_start_one(sdp, ai); /* This may drop log lock */
181 done = 0;
b3b94faa
DT
182 break;
183 }
b3b94faa
DT
184 }
185
186 gfs2_log_unlock(sdp);
187}
188
189int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
190{
191 struct gfs2_ail *ai, *s;
192 int ret;
193
194 gfs2_log_lock(sdp);
195
196 list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) {
197 if (gfs2_ail1_empty_one(sdp, ai, flags))
198 list_move(&ai->ai_list, &sdp->sd_ail2_list);
199 else if (!(flags & DIO_ALL))
200 break;
201 }
202
203 ret = list_empty(&sdp->sd_ail1_list);
204
205 gfs2_log_unlock(sdp);
206
207 return ret;
208}
209
ddacfaf7
SW
210
211/**
212 * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
213 * @sdp: the filesystem
214 * @ai: the AIL entry
215 *
216 */
217
218static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
219{
220 struct list_head *head = &ai->ai_ail2_list;
221 struct gfs2_bufdata *bd;
0f8468c8 222 struct gfs2_inode *bh_ip;
ddacfaf7
SW
223
224 while (!list_empty(head)) {
225 bd = list_entry(head->prev, struct gfs2_bufdata,
226 bd_ail_st_list);
227 gfs2_assert(sdp, bd->bd_ail == ai);
228 bd->bd_ail = NULL;
229 list_del(&bd->bd_ail_st_list);
230 list_del(&bd->bd_ail_gl_list);
231 atomic_dec(&bd->bd_gl->gl_ail_count);
bb3b0e3d
SW
232 if (bd->bd_bh->b_page->mapping) {
233 bh_ip = GFS2_I(bd->bd_bh->b_page->mapping->host);
234 gfs2_meta_cache_flush(bh_ip);
235 }
a0a24741 236 brelse(bd->bd_bh);
ddacfaf7
SW
237 }
238}
239
b3b94faa
DT
240static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
241{
242 struct gfs2_ail *ai, *safe;
243 unsigned int old_tail = sdp->sd_log_tail;
244 int wrap = (new_tail < old_tail);
245 int a, b, rm;
246
247 gfs2_log_lock(sdp);
248
249 list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) {
250 a = (old_tail <= ai->ai_first);
251 b = (ai->ai_first < new_tail);
252 rm = (wrap) ? (a || b) : (a && b);
253 if (!rm)
254 continue;
255
256 gfs2_ail2_empty_one(sdp, ai);
257 list_del(&ai->ai_list);
258 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail1_list));
259 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail2_list));
260 kfree(ai);
261 }
262
263 gfs2_log_unlock(sdp);
264}
265
266/**
267 * gfs2_log_reserve - Make a log reservation
268 * @sdp: The GFS2 superblock
269 * @blks: The number of blocks to reserve
270 *
89918647 271 * Note that we never give out the last few blocks of the journal. Thats
2332c443 272 * due to the fact that there is a small number of header blocks
b004157a
SW
273 * associated with each log flush. The exact number can't be known until
274 * flush time, so we ensure that we have just enough free blocks at all
275 * times to avoid running out during a log flush.
276 *
b3b94faa
DT
277 * Returns: errno
278 */
279
280int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
281{
b3b94faa 282 unsigned int try = 0;
89918647 283 unsigned reserved_blks = 6 * (4096 / sdp->sd_vfs->s_blocksize);
b3b94faa
DT
284
285 if (gfs2_assert_warn(sdp, blks) ||
286 gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
287 return -EINVAL;
288
71b86f56 289 mutex_lock(&sdp->sd_log_reserve_mutex);
484adff8 290 gfs2_log_lock(sdp);
89918647 291 while(sdp->sd_log_blks_free <= (blks + reserved_blks)) {
b3b94faa 292 gfs2_log_unlock(sdp);
b3b94faa 293 gfs2_ail1_empty(sdp, 0);
b09e593d 294 gfs2_log_flush(sdp, NULL);
b3b94faa
DT
295
296 if (try++)
297 gfs2_ail1_start(sdp, 0);
484adff8 298 gfs2_log_lock(sdp);
b3b94faa 299 }
484adff8
SW
300 sdp->sd_log_blks_free -= blks;
301 gfs2_log_unlock(sdp);
302 mutex_unlock(&sdp->sd_log_reserve_mutex);
303
304 down_read(&sdp->sd_log_flush_lock);
b3b94faa
DT
305
306 return 0;
307}
308
309/**
310 * gfs2_log_release - Release a given number of log blocks
311 * @sdp: The GFS2 superblock
312 * @blks: The number of blocks
313 *
314 */
315
316void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
317{
b3b94faa
DT
318
319 gfs2_log_lock(sdp);
320 sdp->sd_log_blks_free += blks;
321 gfs2_assert_withdraw(sdp,
322 sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
323 gfs2_log_unlock(sdp);
ed386507 324 up_read(&sdp->sd_log_flush_lock);
b3b94faa
DT
325}
326
cd915493 327static u64 log_bmap(struct gfs2_sbd *sdp, unsigned int lbn)
b3b94faa 328{
23591256 329 struct inode *inode = sdp->sd_jdesc->jd_inode;
b3b94faa 330 int error;
23591256 331 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
b3b94faa 332
23591256
SW
333 bh_map.b_size = 1 << inode->i_blkbits;
334 error = gfs2_block_map(inode, lbn, 0, &bh_map);
7a6bbacb 335 if (error || !bh_map.b_blocknr)
aed3255f
RK
336 printk(KERN_INFO "error=%d, dbn=%llu lbn=%u", error,
337 (unsigned long long)bh_map.b_blocknr, lbn);
7a6bbacb 338 gfs2_assert_withdraw(sdp, !error && bh_map.b_blocknr);
b3b94faa 339
7a6bbacb 340 return bh_map.b_blocknr;
b3b94faa
DT
341}
342
343/**
344 * log_distance - Compute distance between two journal blocks
345 * @sdp: The GFS2 superblock
346 * @newer: The most recent journal block of the pair
347 * @older: The older journal block of the pair
348 *
349 * Compute the distance (in the journal direction) between two
350 * blocks in the journal
351 *
352 * Returns: the distance in blocks
353 */
354
faa31ce8 355static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
b3b94faa
DT
356 unsigned int older)
357{
358 int dist;
359
360 dist = newer - older;
361 if (dist < 0)
362 dist += sdp->sd_jdesc->jd_blocks;
363
364 return dist;
365}
366
2332c443
RP
367/**
368 * calc_reserved - Calculate the number of blocks to reserve when
369 * refunding a transaction's unused buffers.
370 * @sdp: The GFS2 superblock
371 *
372 * This is complex. We need to reserve room for all our currently used
373 * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
374 * all our journaled data buffers for journaled files (e.g. files in the
375 * meta_fs like rindex, or files for which chattr +j was done.)
376 * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
377 * will count it as free space (sd_log_blks_free) and corruption will follow.
378 *
379 * We can have metadata bufs and jdata bufs in the same journal. So each
380 * type gets its own log header, for which we need to reserve a block.
381 * In fact, each type has the potential for needing more than one header
382 * in cases where we have more buffers than will fit on a journal page.
383 * Metadata journal entries take up half the space of journaled buffer entries.
384 * Thus, metadata entries have buf_limit (502) and journaled buffers have
385 * databuf_limit (251) before they cause a wrap around.
386 *
387 * Also, we need to reserve blocks for revoke journal entries and one for an
388 * overall header for the lot.
389 *
390 * Returns: the number of blocks reserved
391 */
392static unsigned int calc_reserved(struct gfs2_sbd *sdp)
393{
394 unsigned int reserved = 0;
395 unsigned int mbuf_limit, metabufhdrs_needed;
396 unsigned int dbuf_limit, databufhdrs_needed;
397 unsigned int revokes = 0;
398
399 mbuf_limit = buf_limit(sdp);
400 metabufhdrs_needed = (sdp->sd_log_commited_buf +
401 (mbuf_limit - 1)) / mbuf_limit;
402 dbuf_limit = databuf_limit(sdp);
403 databufhdrs_needed = (sdp->sd_log_commited_databuf +
404 (dbuf_limit - 1)) / dbuf_limit;
405
406 if (sdp->sd_log_commited_revoke)
407 revokes = gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
408 sizeof(u64));
409
410 reserved = sdp->sd_log_commited_buf + metabufhdrs_needed +
411 sdp->sd_log_commited_databuf + databufhdrs_needed +
412 revokes;
413 /* One for the overall header */
414 if (reserved)
415 reserved++;
416 return reserved;
417}
418
b3b94faa
DT
419static unsigned int current_tail(struct gfs2_sbd *sdp)
420{
421 struct gfs2_ail *ai;
422 unsigned int tail;
423
424 gfs2_log_lock(sdp);
425
faa31ce8 426 if (list_empty(&sdp->sd_ail1_list)) {
b3b94faa 427 tail = sdp->sd_log_head;
faa31ce8
SW
428 } else {
429 ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail, ai_list);
b3b94faa
DT
430 tail = ai->ai_first;
431 }
432
433 gfs2_log_unlock(sdp);
434
435 return tail;
436}
437
438static inline void log_incr_head(struct gfs2_sbd *sdp)
439{
440 if (sdp->sd_log_flush_head == sdp->sd_log_tail)
faa31ce8 441 gfs2_assert_withdraw(sdp, sdp->sd_log_flush_head == sdp->sd_log_head);
b3b94faa
DT
442
443 if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
444 sdp->sd_log_flush_head = 0;
445 sdp->sd_log_flush_wrapped = 1;
446 }
447}
448
449/**
450 * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
451 * @sdp: The GFS2 superblock
452 *
453 * Returns: the buffer_head
454 */
455
456struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
457{
cd915493 458 u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
b3b94faa
DT
459 struct gfs2_log_buf *lb;
460 struct buffer_head *bh;
461
4f3df041 462 lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
b3b94faa
DT
463 list_add(&lb->lb_list, &sdp->sd_log_flush_list);
464
465 bh = lb->lb_bh = sb_getblk(sdp->sd_vfs, blkno);
466 lock_buffer(bh);
467 memset(bh->b_data, 0, bh->b_size);
468 set_buffer_uptodate(bh);
469 clear_buffer_dirty(bh);
470 unlock_buffer(bh);
471
472 log_incr_head(sdp);
473
474 return bh;
475}
476
477/**
478 * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
479 * @sdp: the filesystem
480 * @data: the data the buffer_head should point to
481 *
482 * Returns: the log buffer descriptor
483 */
484
485struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
486 struct buffer_head *real)
487{
cd915493 488 u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
b3b94faa
DT
489 struct gfs2_log_buf *lb;
490 struct buffer_head *bh;
491
4f3df041 492 lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
b3b94faa
DT
493 list_add(&lb->lb_list, &sdp->sd_log_flush_list);
494 lb->lb_real = real;
495
496 bh = lb->lb_bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL);
497 atomic_set(&bh->b_count, 1);
498 bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate);
18ec7d5c 499 set_bh_page(bh, real->b_page, bh_offset(real));
b3b94faa
DT
500 bh->b_blocknr = blkno;
501 bh->b_size = sdp->sd_sb.sb_bsize;
502 bh->b_bdev = sdp->sd_vfs->s_bdev;
503
504 log_incr_head(sdp);
505
506 return bh;
507}
508
2332c443 509static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
b3b94faa
DT
510{
511 unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
512
513 ail2_empty(sdp, new_tail);
514
515 gfs2_log_lock(sdp);
2332c443 516 sdp->sd_log_blks_free += dist;
faa31ce8 517 gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
b3b94faa
DT
518 gfs2_log_unlock(sdp);
519
520 sdp->sd_log_tail = new_tail;
521}
522
523/**
524 * log_write_header - Get and initialize a journal header buffer
525 * @sdp: The GFS2 superblock
526 *
527 * Returns: the initialized log buffer descriptor
528 */
529
cd915493 530static void log_write_header(struct gfs2_sbd *sdp, u32 flags, int pull)
b3b94faa 531{
cd915493 532 u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
b3b94faa
DT
533 struct buffer_head *bh;
534 struct gfs2_log_header *lh;
535 unsigned int tail;
cd915493 536 u32 hash;
b3b94faa 537
b3b94faa
DT
538 bh = sb_getblk(sdp->sd_vfs, blkno);
539 lock_buffer(bh);
540 memset(bh->b_data, 0, bh->b_size);
541 set_buffer_uptodate(bh);
542 clear_buffer_dirty(bh);
543 unlock_buffer(bh);
544
545 gfs2_ail1_empty(sdp, 0);
546 tail = current_tail(sdp);
547
548 lh = (struct gfs2_log_header *)bh->b_data;
549 memset(lh, 0, sizeof(struct gfs2_log_header));
550 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
e3167ded
SW
551 lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
552 lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
e0f2bf78
SW
553 lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++);
554 lh->lh_flags = cpu_to_be32(flags);
555 lh->lh_tail = cpu_to_be32(tail);
556 lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
b3b94faa
DT
557 hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
558 lh->lh_hash = cpu_to_be32(hash);
559
560 set_buffer_dirty(bh);
561 if (sync_dirty_buffer(bh))
562 gfs2_io_error_bh(sdp, bh);
563 brelse(bh);
564
565 if (sdp->sd_log_tail != tail)
2332c443 566 log_pull_tail(sdp, tail);
b3b94faa
DT
567 else
568 gfs2_assert_withdraw(sdp, !pull);
569
570 sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
571 log_incr_head(sdp);
572}
573
574static void log_flush_commit(struct gfs2_sbd *sdp)
575{
576 struct list_head *head = &sdp->sd_log_flush_list;
577 struct gfs2_log_buf *lb;
578 struct buffer_head *bh;
a0a24741 579 int flushcount = 0;
b3b94faa
DT
580
581 while (!list_empty(head)) {
582 lb = list_entry(head->next, struct gfs2_log_buf, lb_list);
583 list_del(&lb->lb_list);
584 bh = lb->lb_bh;
585
586 wait_on_buffer(bh);
587 if (!buffer_uptodate(bh))
588 gfs2_io_error_bh(sdp, bh);
589 if (lb->lb_real) {
590 while (atomic_read(&bh->b_count) != 1) /* Grrrr... */
591 schedule();
592 free_buffer_head(bh);
593 } else
594 brelse(bh);
595 kfree(lb);
a0a24741 596 flushcount++;
b3b94faa
DT
597 }
598
a0a24741
SW
599 /* If nothing was journaled, the header is unplanned and unwanted. */
600 if (flushcount) {
601 log_write_header(sdp, 0, 0);
602 } else {
603 unsigned int tail;
604 tail = current_tail(sdp);
605
606 gfs2_ail1_empty(sdp, 0);
607 if (sdp->sd_log_tail != tail)
608 log_pull_tail(sdp, tail);
609 }
b3b94faa
DT
610}
611
612/**
b09e593d 613 * gfs2_log_flush - flush incore transaction(s)
b3b94faa
DT
614 * @sdp: the filesystem
615 * @gl: The glock structure to flush. If NULL, flush the whole incore log
616 *
617 */
618
b09e593d 619void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
b3b94faa
DT
620{
621 struct gfs2_ail *ai;
622
484adff8 623 down_write(&sdp->sd_log_flush_lock);
f55ab26a
SW
624
625 if (gl) {
626 gfs2_log_lock(sdp);
627 if (list_empty(&gl->gl_le.le_list)) {
628 gfs2_log_unlock(sdp);
484adff8 629 up_write(&sdp->sd_log_flush_lock);
f55ab26a
SW
630 return;
631 }
632 gfs2_log_unlock(sdp);
633 }
634
b09e593d
SW
635 ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
636 INIT_LIST_HEAD(&ai->ai_ail1_list);
637 INIT_LIST_HEAD(&ai->ai_ail2_list);
b3b94faa 638
2332c443
RP
639 gfs2_assert_withdraw(sdp,
640 sdp->sd_log_num_buf + sdp->sd_log_num_jdata ==
641 sdp->sd_log_commited_buf +
642 sdp->sd_log_commited_databuf);
b3b94faa
DT
643 gfs2_assert_withdraw(sdp,
644 sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
645
b3b94faa
DT
646 sdp->sd_log_flush_head = sdp->sd_log_head;
647 sdp->sd_log_flush_wrapped = 0;
648 ai->ai_first = sdp->sd_log_flush_head;
649
650 lops_before_commit(sdp);
651 if (!list_empty(&sdp->sd_log_flush_list))
652 log_flush_commit(sdp);
2332c443
RP
653 else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
654 gfs2_log_lock(sdp);
655 sdp->sd_log_blks_free--; /* Adjust for unreserved buffer */
656 gfs2_log_unlock(sdp);
b3b94faa 657 log_write_header(sdp, 0, PULL);
2332c443 658 }
b3b94faa 659 lops_after_commit(sdp, ai);
b09e593d 660
fe1a698f
SW
661 gfs2_log_lock(sdp);
662 sdp->sd_log_head = sdp->sd_log_flush_head;
faa31ce8
SW
663 sdp->sd_log_blks_reserved = 0;
664 sdp->sd_log_commited_buf = 0;
2332c443 665 sdp->sd_log_commited_databuf = 0;
faa31ce8 666 sdp->sd_log_commited_revoke = 0;
b3b94faa 667
b3b94faa
DT
668 if (!list_empty(&ai->ai_ail1_list)) {
669 list_add(&ai->ai_list, &sdp->sd_ail1_list);
670 ai = NULL;
671 }
672 gfs2_log_unlock(sdp);
673
b3b94faa 674 sdp->sd_vfs->s_dirt = 0;
484adff8 675 up_write(&sdp->sd_log_flush_lock);
b3b94faa
DT
676
677 kfree(ai);
678}
679
680static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
681{
2332c443 682 unsigned int reserved;
b3b94faa
DT
683 unsigned int old;
684
685 gfs2_log_lock(sdp);
686
687 sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
2332c443
RP
688 sdp->sd_log_commited_databuf += tr->tr_num_databuf_new -
689 tr->tr_num_databuf_rm;
690 gfs2_assert_withdraw(sdp, (((int)sdp->sd_log_commited_buf) >= 0) ||
691 (((int)sdp->sd_log_commited_databuf) >= 0));
b3b94faa
DT
692 sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
693 gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0);
2332c443 694 reserved = calc_reserved(sdp);
b3b94faa
DT
695 old = sdp->sd_log_blks_free;
696 sdp->sd_log_blks_free += tr->tr_reserved -
697 (reserved - sdp->sd_log_blks_reserved);
698
b09e593d 699 gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free >= old);
2332c443
RP
700 gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free <=
701 sdp->sd_jdesc->jd_blocks);
b3b94faa
DT
702
703 sdp->sd_log_blks_reserved = reserved;
704
705 gfs2_log_unlock(sdp);
706}
707
708/**
709 * gfs2_log_commit - Commit a transaction to the log
710 * @sdp: the filesystem
711 * @tr: the transaction
712 *
713 * Returns: errno
714 */
715
716void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
717{
718 log_refund(sdp, tr);
719 lops_incore_commit(sdp, tr);
720
721 sdp->sd_vfs->s_dirt = 1;
484adff8 722 up_read(&sdp->sd_log_flush_lock);
b3b94faa 723
b3b94faa 724 gfs2_log_lock(sdp);
b004157a
SW
725 if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks))
726 wake_up_process(sdp->sd_logd_process);
727 gfs2_log_unlock(sdp);
b3b94faa
DT
728}
729
730/**
731 * gfs2_log_shutdown - write a shutdown header into a journal
732 * @sdp: the filesystem
733 *
734 */
735
736void gfs2_log_shutdown(struct gfs2_sbd *sdp)
737{
484adff8 738 down_write(&sdp->sd_log_flush_lock);
b3b94faa 739
b3b94faa
DT
740 gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
741 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_gl);
742 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
18ec7d5c 743 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_jdata);
b3b94faa
DT
744 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
745 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
746 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
747 gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
748
749 sdp->sd_log_flush_head = sdp->sd_log_head;
750 sdp->sd_log_flush_wrapped = 0;
751
2332c443
RP
752 log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT,
753 (sdp->sd_log_tail == current_tail(sdp)) ? 0 : PULL);
b3b94faa 754
a74604be
SW
755 gfs2_assert_warn(sdp, sdp->sd_log_blks_free == sdp->sd_jdesc->jd_blocks);
756 gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
757 gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
b3b94faa
DT
758
759 sdp->sd_log_head = sdp->sd_log_flush_head;
b3b94faa
DT
760 sdp->sd_log_tail = sdp->sd_log_head;
761
484adff8 762 up_write(&sdp->sd_log_flush_lock);
b3b94faa
DT
763}
764
a25311c8
SW
765
766/**
767 * gfs2_meta_syncfs - sync all the buffers in a filesystem
768 * @sdp: the filesystem
769 *
770 */
771
772void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
773{
774 gfs2_log_flush(sdp, NULL);
775 for (;;) {
776 gfs2_ail1_start(sdp, DIO_ALL);
777 if (gfs2_ail1_empty(sdp, DIO_ALL))
778 break;
779 msleep(10);
780 }
781}
782