]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/gfs2/lops.c
GFS2: Use filemap_fdatawrite() to write back the AIL
[mirror_ubuntu-artful-kernel.git] / fs / gfs2 / lops.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>
c969f58c
SW
16#include <linux/bio.h>
17#include <linux/fs.h>
b3b94faa
DT
18
19#include "gfs2.h"
5c676f6d 20#include "incore.h"
2332c443 21#include "inode.h"
b3b94faa
DT
22#include "glock.h"
23#include "log.h"
24#include "lops.h"
25#include "meta_io.h"
26#include "recovery.h"
27#include "rgrp.h"
28#include "trans.h"
5c676f6d 29#include "util.h"
63997775 30#include "trace_gfs2.h"
b3b94faa 31
9b9107a5
SW
32/**
33 * gfs2_pin - Pin a buffer in memory
34 * @sdp: The superblock
35 * @bh: The buffer to be pinned
36 *
37 * The log lock must be held when calling this function
38 */
39static void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
40{
41 struct gfs2_bufdata *bd;
42
43 gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
44
45 clear_buffer_dirty(bh);
46 if (test_set_buffer_pinned(bh))
47 gfs2_assert_withdraw(sdp, 0);
48 if (!buffer_uptodate(bh))
49 gfs2_io_error_bh(sdp, bh);
50 bd = bh->b_private;
51 /* If this buffer is in the AIL and it has already been written
52 * to in-place disk block, remove it from the AIL.
53 */
c618e87a 54 spin_lock(&sdp->sd_ail_lock);
9b9107a5
SW
55 if (bd->bd_ail)
56 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
c618e87a 57 spin_unlock(&sdp->sd_ail_lock);
9b9107a5 58 get_bh(bh);
5e687eac 59 atomic_inc(&sdp->sd_log_pinned);
63997775 60 trace_gfs2_pin(bd, 1);
9b9107a5
SW
61}
62
63/**
64 * gfs2_unpin - Unpin a buffer
65 * @sdp: the filesystem the buffer belongs to
66 * @bh: The buffer to unpin
67 * @ai:
68 *
69 */
70
71static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
72 struct gfs2_ail *ai)
73{
74 struct gfs2_bufdata *bd = bh->b_private;
75
76 gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
77
78 if (!buffer_pinned(bh))
79 gfs2_assert_withdraw(sdp, 0);
80
81 lock_buffer(bh);
82 mark_buffer_dirty(bh);
83 clear_buffer_pinned(bh);
84
d6a079e8 85 spin_lock(&sdp->sd_ail_lock);
9b9107a5
SW
86 if (bd->bd_ail) {
87 list_del(&bd->bd_ail_st_list);
88 brelse(bh);
89 } else {
90 struct gfs2_glock *gl = bd->bd_gl;
91 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
92 atomic_inc(&gl->gl_ail_count);
93 }
94 bd->bd_ail = ai;
95 list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
d6a079e8
DC
96 spin_unlock(&sdp->sd_ail_lock);
97
bc015cb8
SW
98 if (test_and_clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags))
99 gfs2_glock_schedule_for_reclaim(bd->bd_gl);
63997775 100 trace_gfs2_pin(bd, 0);
9b9107a5 101 unlock_buffer(bh);
5e687eac 102 atomic_dec(&sdp->sd_log_pinned);
9b9107a5
SW
103}
104
16615be1
SW
105
106static inline struct gfs2_log_descriptor *bh_log_desc(struct buffer_head *bh)
107{
108 return (struct gfs2_log_descriptor *)bh->b_data;
109}
110
111static inline __be64 *bh_log_ptr(struct buffer_head *bh)
112{
113 struct gfs2_log_descriptor *ld = bh_log_desc(bh);
114 return (__force __be64 *)(ld + 1);
115}
116
117static inline __be64 *bh_ptr_end(struct buffer_head *bh)
118{
119 return (__force __be64 *)(bh->b_data + bh->b_size);
120}
121
122
123static struct buffer_head *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type)
124{
125 struct buffer_head *bh = gfs2_log_get_buf(sdp);
126 struct gfs2_log_descriptor *ld = bh_log_desc(bh);
127 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
128 ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
129 ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
130 ld->ld_type = cpu_to_be32(ld_type);
131 ld->ld_length = 0;
132 ld->ld_data1 = 0;
133 ld->ld_data2 = 0;
134 memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
135 return bh;
136}
137
b3b94faa
DT
138static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
139{
140 struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
0ab7d13f 141 struct gfs2_meta_header *mh;
b3b94faa
DT
142 struct gfs2_trans *tr;
143
9b9107a5 144 lock_buffer(bd->bd_bh);
8bd95727 145 gfs2_log_lock(sdp);
9b9107a5
SW
146 if (!list_empty(&bd->bd_list_tr))
147 goto out;
5c676f6d 148 tr = current->journal_info;
b3b94faa
DT
149 tr->tr_touched = 1;
150 tr->tr_num_buf++;
151 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
b3b94faa 152 if (!list_empty(&le->le_list))
9b9107a5 153 goto out;
2bcd610d
SW
154 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
155 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
b3b94faa 156 gfs2_meta_check(sdp, bd->bd_bh);
a98ab220 157 gfs2_pin(sdp, bd->bd_bh);
0ab7d13f
SW
158 mh = (struct gfs2_meta_header *)bd->bd_bh->b_data;
159 mh->__pad0 = cpu_to_be64(0);
160 mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
b3b94faa
DT
161 sdp->sd_log_num_buf++;
162 list_add(&le->le_list, &sdp->sd_log_le_buf);
b3b94faa 163 tr->tr_num_buf_new++;
9b9107a5
SW
164out:
165 gfs2_log_unlock(sdp);
166 unlock_buffer(bd->bd_bh);
b3b94faa
DT
167}
168
b3b94faa
DT
169static void buf_lo_before_commit(struct gfs2_sbd *sdp)
170{
171 struct buffer_head *bh;
172 struct gfs2_log_descriptor *ld;
173 struct gfs2_bufdata *bd1 = NULL, *bd2;
905d2aef 174 unsigned int total;
b3b94faa
DT
175 unsigned int limit;
176 unsigned int num;
177 unsigned n;
178 __be64 *ptr;
179
2332c443 180 limit = buf_limit(sdp);
b3b94faa
DT
181 /* for 4k blocks, limit = 503 */
182
905d2aef
BP
183 gfs2_log_lock(sdp);
184 total = sdp->sd_log_num_buf;
b3b94faa
DT
185 bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
186 while(total) {
187 num = total;
188 if (total > limit)
189 num = limit;
905d2aef 190 gfs2_log_unlock(sdp);
16615be1 191 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA);
905d2aef 192 gfs2_log_lock(sdp);
16615be1
SW
193 ld = bh_log_desc(bh);
194 ptr = bh_log_ptr(bh);
b3b94faa
DT
195 ld->ld_length = cpu_to_be32(num + 1);
196 ld->ld_data1 = cpu_to_be32(num);
b3b94faa
DT
197
198 n = 0;
568f4c96
SW
199 list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
200 bd_le.le_list) {
b3b94faa
DT
201 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
202 if (++n >= num)
203 break;
204 }
205
905d2aef 206 gfs2_log_unlock(sdp);
721a9602 207 submit_bh(WRITE_SYNC, bh);
905d2aef 208 gfs2_log_lock(sdp);
b3b94faa
DT
209
210 n = 0;
568f4c96
SW
211 list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
212 bd_le.le_list) {
16615be1 213 get_bh(bd2->bd_bh);
905d2aef 214 gfs2_log_unlock(sdp);
16615be1 215 lock_buffer(bd2->bd_bh);
b3b94faa 216 bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
721a9602 217 submit_bh(WRITE_SYNC, bh);
905d2aef 218 gfs2_log_lock(sdp);
b3b94faa
DT
219 if (++n >= num)
220 break;
221 }
222
905d2aef 223 BUG_ON(total < num);
b3b94faa
DT
224 total -= num;
225 }
905d2aef 226 gfs2_log_unlock(sdp);
b3b94faa
DT
227}
228
229static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
230{
231 struct list_head *head = &sdp->sd_log_le_buf;
232 struct gfs2_bufdata *bd;
233
234 while (!list_empty(head)) {
235 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
236 list_del_init(&bd->bd_le.le_list);
237 sdp->sd_log_num_buf--;
238
a98ab220 239 gfs2_unpin(sdp, bd->bd_bh, ai);
b3b94faa
DT
240 }
241 gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
242}
243
244static void buf_lo_before_scan(struct gfs2_jdesc *jd,
55167622 245 struct gfs2_log_header_host *head, int pass)
b3b94faa 246{
feaa7bba 247 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
b3b94faa
DT
248
249 if (pass != 0)
250 return;
251
252 sdp->sd_found_blocks = 0;
253 sdp->sd_replayed_blocks = 0;
254}
255
256static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
257 struct gfs2_log_descriptor *ld, __be64 *ptr,
258 int pass)
259{
feaa7bba
SW
260 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
261 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
5c676f6d 262 struct gfs2_glock *gl = ip->i_gl;
b3b94faa
DT
263 unsigned int blks = be32_to_cpu(ld->ld_data1);
264 struct buffer_head *bh_log, *bh_ip;
cd915493 265 u64 blkno;
b3b94faa
DT
266 int error = 0;
267
268 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
269 return 0;
270
271 gfs2_replay_incr_blk(sdp, &start);
272
273 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
274 blkno = be64_to_cpu(*ptr++);
275
276 sdp->sd_found_blocks++;
277
278 if (gfs2_revoke_check(sdp, blkno, start))
279 continue;
280
281 error = gfs2_replay_read_block(jd, start, &bh_log);
82ffa516
SW
282 if (error)
283 return error;
b3b94faa
DT
284
285 bh_ip = gfs2_meta_new(gl, blkno);
286 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
287
288 if (gfs2_meta_check(sdp, bh_ip))
289 error = -EIO;
290 else
291 mark_buffer_dirty(bh_ip);
292
293 brelse(bh_log);
294 brelse(bh_ip);
295
296 if (error)
297 break;
298
299 sdp->sd_replayed_blocks++;
300 }
301
302 return error;
303}
304
305static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
306{
feaa7bba
SW
307 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
308 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
b3b94faa
DT
309
310 if (error) {
7276b3b0 311 gfs2_meta_sync(ip->i_gl);
b3b94faa
DT
312 return;
313 }
314 if (pass != 1)
315 return;
316
7276b3b0 317 gfs2_meta_sync(ip->i_gl);
b3b94faa
DT
318
319 fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
320 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
321}
322
323static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
324{
325 struct gfs2_trans *tr;
326
5c676f6d 327 tr = current->journal_info;
b3b94faa
DT
328 tr->tr_touched = 1;
329 tr->tr_num_revoke++;
b3b94faa
DT
330 sdp->sd_log_num_revoke++;
331 list_add(&le->le_list, &sdp->sd_log_le_revoke);
b3b94faa
DT
332}
333
334static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
335{
336 struct gfs2_log_descriptor *ld;
337 struct gfs2_meta_header *mh;
338 struct buffer_head *bh;
339 unsigned int offset;
340 struct list_head *head = &sdp->sd_log_le_revoke;
82e86087 341 struct gfs2_bufdata *bd;
b3b94faa
DT
342
343 if (!sdp->sd_log_num_revoke)
344 return;
345
16615be1
SW
346 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE);
347 ld = bh_log_desc(bh);
568f4c96 348 ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
cd915493 349 sizeof(u64)));
b3b94faa 350 ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
b3b94faa
DT
351 offset = sizeof(struct gfs2_log_descriptor);
352
353 while (!list_empty(head)) {
82e86087
SW
354 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
355 list_del_init(&bd->bd_le.le_list);
b3b94faa
DT
356 sdp->sd_log_num_revoke--;
357
cd915493 358 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
721a9602 359 submit_bh(WRITE_SYNC, bh);
b3b94faa
DT
360
361 bh = gfs2_log_get_buf(sdp);
362 mh = (struct gfs2_meta_header *)bh->b_data;
363 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
e3167ded
SW
364 mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
365 mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
b3b94faa
DT
366 offset = sizeof(struct gfs2_meta_header);
367 }
368
82e86087 369 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(bd->bd_blkno);
0820ab51 370 kmem_cache_free(gfs2_bufdata_cachep, bd);
b3b94faa 371
cd915493 372 offset += sizeof(u64);
b3b94faa
DT
373 }
374 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
375
721a9602 376 submit_bh(WRITE_SYNC, bh);
b3b94faa
DT
377}
378
379static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
55167622 380 struct gfs2_log_header_host *head, int pass)
b3b94faa 381{
feaa7bba 382 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
b3b94faa
DT
383
384 if (pass != 0)
385 return;
386
387 sdp->sd_found_revokes = 0;
388 sdp->sd_replay_tail = head->lh_tail;
389}
390
391static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
392 struct gfs2_log_descriptor *ld, __be64 *ptr,
393 int pass)
394{
feaa7bba 395 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
b3b94faa
DT
396 unsigned int blks = be32_to_cpu(ld->ld_length);
397 unsigned int revokes = be32_to_cpu(ld->ld_data1);
398 struct buffer_head *bh;
399 unsigned int offset;
cd915493 400 u64 blkno;
b3b94faa
DT
401 int first = 1;
402 int error;
403
404 if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
405 return 0;
406
407 offset = sizeof(struct gfs2_log_descriptor);
408
409 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
410 error = gfs2_replay_read_block(jd, start, &bh);
411 if (error)
412 return error;
413
414 if (!first)
415 gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
416
cd915493 417 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
b3b94faa
DT
418 blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
419
420 error = gfs2_revoke_add(sdp, blkno, start);
3ad62e87
BP
421 if (error < 0) {
422 brelse(bh);
b3b94faa 423 return error;
3ad62e87 424 }
b3b94faa
DT
425 else if (error)
426 sdp->sd_found_revokes++;
427
428 if (!--revokes)
429 break;
cd915493 430 offset += sizeof(u64);
b3b94faa
DT
431 }
432
433 brelse(bh);
434 offset = sizeof(struct gfs2_meta_header);
435 first = 0;
436 }
437
438 return 0;
439}
440
441static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
442{
feaa7bba 443 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
b3b94faa
DT
444
445 if (error) {
446 gfs2_revoke_clean(sdp);
447 return;
448 }
449 if (pass != 1)
450 return;
451
452 fs_info(sdp, "jid=%u: Found %u revoke tags\n",
453 jd->jd_jid, sdp->sd_found_revokes);
454
455 gfs2_revoke_clean(sdp);
456}
457
458static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
459{
460 struct gfs2_rgrpd *rgd;
5c676f6d 461 struct gfs2_trans *tr = current->journal_info;
b3b94faa 462
5c676f6d 463 tr->tr_touched = 1;
b3b94faa 464
b3b94faa 465 rgd = container_of(le, struct gfs2_rgrpd, rd_le);
b3b94faa
DT
466
467 gfs2_log_lock(sdp);
68835625
BM
468 if (!list_empty(&le->le_list)){
469 gfs2_log_unlock(sdp);
470 return;
471 }
472 gfs2_rgrp_bh_hold(rgd);
b3b94faa
DT
473 sdp->sd_log_num_rg++;
474 list_add(&le->le_list, &sdp->sd_log_le_rg);
907b9bce 475 gfs2_log_unlock(sdp);
b3b94faa
DT
476}
477
478static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
479{
480 struct list_head *head = &sdp->sd_log_le_rg;
481 struct gfs2_rgrpd *rgd;
482
483 while (!list_empty(head)) {
484 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
485 list_del_init(&rgd->rd_le.le_list);
486 sdp->sd_log_num_rg--;
487
488 gfs2_rgrp_repolish_clones(rgd);
489 gfs2_rgrp_bh_put(rgd);
490 }
491 gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
492}
493
18ec7d5c
SW
494/**
495 * databuf_lo_add - Add a databuf to the transaction.
496 *
497 * This is used in two distinct cases:
498 * i) In ordered write mode
499 * We put the data buffer on a list so that we can ensure that its
500 * synced to disk at the right time
501 * ii) In journaled data mode
502 * We need to journal the data block in the same way as metadata in
503 * the functions above. The difference is that here we have a tag
504 * which is two __be64's being the block number (as per meta data)
505 * and a flag which says whether the data block needs escaping or
506 * not. This means we need a new log entry for each 251 or so data
507 * blocks, which isn't an enormous overhead but twice as much as
508 * for normal metadata blocks.
509 */
b3b94faa
DT
510static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
511{
18ec7d5c 512 struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
5c676f6d 513 struct gfs2_trans *tr = current->journal_info;
18ec7d5c 514 struct address_space *mapping = bd->bd_bh->b_page->mapping;
feaa7bba 515 struct gfs2_inode *ip = GFS2_I(mapping->host);
b3b94faa 516
9b9107a5 517 lock_buffer(bd->bd_bh);
8bd95727 518 gfs2_log_lock(sdp);
9ff8ec32
SW
519 if (tr) {
520 if (!list_empty(&bd->bd_list_tr))
521 goto out;
522 tr->tr_touched = 1;
523 if (gfs2_is_jdata(ip)) {
524 tr->tr_num_buf++;
525 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
526 }
773ed1a0 527 }
2332c443 528 if (!list_empty(&le->le_list))
9b9107a5 529 goto out;
2332c443 530
2bcd610d
SW
531 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
532 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
2332c443 533 if (gfs2_is_jdata(ip)) {
2332c443
RP
534 gfs2_pin(sdp, bd->bd_bh);
535 tr->tr_num_databuf_new++;
d7b616e2 536 sdp->sd_log_num_databuf++;
e5884636 537 list_add_tail(&le->le_list, &sdp->sd_log_le_databuf);
d7b616e2 538 } else {
e5884636 539 list_add_tail(&le->le_list, &sdp->sd_log_le_ordered);
9b9107a5 540 }
9b9107a5 541out:
b3b94faa 542 gfs2_log_unlock(sdp);
9b9107a5 543 unlock_buffer(bd->bd_bh);
b3b94faa
DT
544}
545
16615be1 546static void gfs2_check_magic(struct buffer_head *bh)
18ec7d5c 547{
18ec7d5c
SW
548 void *kaddr;
549 __be32 *ptr;
18ec7d5c 550
16615be1
SW
551 clear_buffer_escaped(bh);
552 kaddr = kmap_atomic(bh->b_page, KM_USER0);
18ec7d5c
SW
553 ptr = kaddr + bh_offset(bh);
554 if (*ptr == cpu_to_be32(GFS2_MAGIC))
16615be1 555 set_buffer_escaped(bh);
c312c4fd 556 kunmap_atomic(kaddr, KM_USER0);
18ec7d5c
SW
557}
558
16615be1
SW
559static void gfs2_write_blocks(struct gfs2_sbd *sdp, struct buffer_head *bh,
560 struct list_head *list, struct list_head *done,
561 unsigned int n)
b3b94faa 562{
16615be1 563 struct buffer_head *bh1;
18ec7d5c 564 struct gfs2_log_descriptor *ld;
16615be1
SW
565 struct gfs2_bufdata *bd;
566 __be64 *ptr;
d7b616e2 567
16615be1
SW
568 if (!bh)
569 return;
b3b94faa 570
16615be1
SW
571 ld = bh_log_desc(bh);
572 ld->ld_length = cpu_to_be32(n + 1);
573 ld->ld_data1 = cpu_to_be32(n);
b3b94faa 574
16615be1
SW
575 ptr = bh_log_ptr(bh);
576
577 get_bh(bh);
721a9602 578 submit_bh(WRITE_SYNC, bh);
f55ab26a 579 gfs2_log_lock(sdp);
16615be1
SW
580 while(!list_empty(list)) {
581 bd = list_entry(list->next, struct gfs2_bufdata, bd_le.le_list);
582 list_move_tail(&bd->bd_le.le_list, done);
583 get_bh(bd->bd_bh);
584 while (be64_to_cpu(*ptr) != bd->bd_bh->b_blocknr) {
585 gfs2_log_incr_head(sdp);
586 ptr += 2;
18ec7d5c 587 }
f55ab26a 588 gfs2_log_unlock(sdp);
16615be1
SW
589 lock_buffer(bd->bd_bh);
590 if (buffer_escaped(bd->bd_bh)) {
591 void *kaddr;
592 bh1 = gfs2_log_get_buf(sdp);
593 kaddr = kmap_atomic(bd->bd_bh->b_page, KM_USER0);
594 memcpy(bh1->b_data, kaddr + bh_offset(bd->bd_bh),
595 bh1->b_size);
596 kunmap_atomic(kaddr, KM_USER0);
597 *(__be32 *)bh1->b_data = 0;
598 clear_buffer_escaped(bd->bd_bh);
599 unlock_buffer(bd->bd_bh);
600 brelse(bd->bd_bh);
601 } else {
602 bh1 = gfs2_log_fake_buf(sdp, bd->bd_bh);
18ec7d5c 603 }
721a9602 604 submit_bh(WRITE_SYNC, bh1);
f55ab26a 605 gfs2_log_lock(sdp);
16615be1
SW
606 ptr += 2;
607 }
608 gfs2_log_unlock(sdp);
609 brelse(bh);
610}
611
612/**
613 * databuf_lo_before_commit - Scan the data buffers, writing as we go
614 *
615 */
616
617static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
618{
619 struct gfs2_bufdata *bd = NULL;
620 struct buffer_head *bh = NULL;
621 unsigned int n = 0;
622 __be64 *ptr = NULL, *end = NULL;
623 LIST_HEAD(processed);
624 LIST_HEAD(in_progress);
625
626 gfs2_log_lock(sdp);
627 while (!list_empty(&sdp->sd_log_le_databuf)) {
628 if (ptr == end) {
f55ab26a 629 gfs2_log_unlock(sdp);
16615be1
SW
630 gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
631 n = 0;
632 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_JDATA);
633 ptr = bh_log_ptr(bh);
634 end = bh_ptr_end(bh) - 1;
f55ab26a 635 gfs2_log_lock(sdp);
16615be1 636 continue;
18ec7d5c 637 }
16615be1
SW
638 bd = list_entry(sdp->sd_log_le_databuf.next, struct gfs2_bufdata, bd_le.le_list);
639 list_move_tail(&bd->bd_le.le_list, &in_progress);
640 gfs2_check_magic(bd->bd_bh);
641 *ptr++ = cpu_to_be64(bd->bd_bh->b_blocknr);
642 *ptr++ = cpu_to_be64(buffer_escaped(bh) ? 1 : 0);
643 n++;
b3b94faa 644 }
f55ab26a 645 gfs2_log_unlock(sdp);
16615be1
SW
646 gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
647 gfs2_log_lock(sdp);
648 list_splice(&processed, &sdp->sd_log_le_databuf);
649 gfs2_log_unlock(sdp);
18ec7d5c
SW
650}
651
652static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
653 struct gfs2_log_descriptor *ld,
654 __be64 *ptr, int pass)
655{
feaa7bba
SW
656 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
657 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
5c676f6d 658 struct gfs2_glock *gl = ip->i_gl;
18ec7d5c
SW
659 unsigned int blks = be32_to_cpu(ld->ld_data1);
660 struct buffer_head *bh_log, *bh_ip;
cd915493
SW
661 u64 blkno;
662 u64 esc;
18ec7d5c
SW
663 int error = 0;
664
665 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
666 return 0;
667
668 gfs2_replay_incr_blk(sdp, &start);
669 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
670 blkno = be64_to_cpu(*ptr++);
671 esc = be64_to_cpu(*ptr++);
672
673 sdp->sd_found_blocks++;
674
675 if (gfs2_revoke_check(sdp, blkno, start))
676 continue;
677
678 error = gfs2_replay_read_block(jd, start, &bh_log);
679 if (error)
680 return error;
681
682 bh_ip = gfs2_meta_new(gl, blkno);
683 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
684
685 /* Unescape */
686 if (esc) {
687 __be32 *eptr = (__be32 *)bh_ip->b_data;
688 *eptr = cpu_to_be32(GFS2_MAGIC);
689 }
690 mark_buffer_dirty(bh_ip);
691
692 brelse(bh_log);
693 brelse(bh_ip);
694 if (error)
695 break;
696
697 sdp->sd_replayed_blocks++;
698 }
699
700 return error;
701}
702
703/* FIXME: sort out accounting for log blocks etc. */
704
705static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
706{
feaa7bba
SW
707 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
708 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
18ec7d5c
SW
709
710 if (error) {
7276b3b0 711 gfs2_meta_sync(ip->i_gl);
18ec7d5c
SW
712 return;
713 }
714 if (pass != 1)
715 return;
716
717 /* data sync? */
7276b3b0 718 gfs2_meta_sync(ip->i_gl);
18ec7d5c
SW
719
720 fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
721 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
722}
723
724static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
725{
726 struct list_head *head = &sdp->sd_log_le_databuf;
727 struct gfs2_bufdata *bd;
728
729 while (!list_empty(head)) {
730 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
b8e1aabf 731 list_del_init(&bd->bd_le.le_list);
18ec7d5c 732 sdp->sd_log_num_databuf--;
18ec7d5c 733 gfs2_unpin(sdp, bd->bd_bh, ai);
18ec7d5c 734 }
b3b94faa
DT
735 gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
736}
737
18ec7d5c 738
b09e593d 739const struct gfs2_log_operations gfs2_buf_lops = {
b3b94faa 740 .lo_add = buf_lo_add,
b3b94faa
DT
741 .lo_before_commit = buf_lo_before_commit,
742 .lo_after_commit = buf_lo_after_commit,
743 .lo_before_scan = buf_lo_before_scan,
744 .lo_scan_elements = buf_lo_scan_elements,
745 .lo_after_scan = buf_lo_after_scan,
ea67eedb 746 .lo_name = "buf",
b3b94faa
DT
747};
748
b09e593d 749const struct gfs2_log_operations gfs2_revoke_lops = {
b3b94faa
DT
750 .lo_add = revoke_lo_add,
751 .lo_before_commit = revoke_lo_before_commit,
752 .lo_before_scan = revoke_lo_before_scan,
753 .lo_scan_elements = revoke_lo_scan_elements,
754 .lo_after_scan = revoke_lo_after_scan,
ea67eedb 755 .lo_name = "revoke",
b3b94faa
DT
756};
757
b09e593d 758const struct gfs2_log_operations gfs2_rg_lops = {
b3b94faa
DT
759 .lo_add = rg_lo_add,
760 .lo_after_commit = rg_lo_after_commit,
ea67eedb 761 .lo_name = "rg",
b3b94faa
DT
762};
763
b09e593d 764const struct gfs2_log_operations gfs2_databuf_lops = {
b3b94faa
DT
765 .lo_add = databuf_lo_add,
766 .lo_before_commit = databuf_lo_before_commit,
18ec7d5c
SW
767 .lo_after_commit = databuf_lo_after_commit,
768 .lo_scan_elements = databuf_lo_scan_elements,
769 .lo_after_scan = databuf_lo_after_scan,
ea67eedb 770 .lo_name = "databuf",
b3b94faa
DT
771};
772
b09e593d 773const struct gfs2_log_operations *gfs2_log_ops[] = {
16615be1 774 &gfs2_databuf_lops,
b3b94faa 775 &gfs2_buf_lops,
b3b94faa 776 &gfs2_rg_lops,
16615be1 777 &gfs2_revoke_lops,
ea67eedb 778 NULL,
b3b94faa
DT
779};
780