Merge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable
[linux-2.6] / fs / gfs2 / lops.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2006 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/bio.h>
17 #include <linux/fs.h>
18
19 #include "gfs2.h"
20 #include "incore.h"
21 #include "inode.h"
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"
29 #include "util.h"
30
31 /**
32  * gfs2_pin - Pin a buffer in memory
33  * @sdp: The superblock
34  * @bh: The buffer to be pinned
35  *
36  * The log lock must be held when calling this function
37  */
38 static void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
39 {
40         struct gfs2_bufdata *bd;
41
42         gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
43
44         clear_buffer_dirty(bh);
45         if (test_set_buffer_pinned(bh))
46                 gfs2_assert_withdraw(sdp, 0);
47         if (!buffer_uptodate(bh))
48                 gfs2_io_error_bh(sdp, bh);
49         bd = bh->b_private;
50         /* If this buffer is in the AIL and it has already been written
51          * to in-place disk block, remove it from the AIL.
52          */
53         if (bd->bd_ail)
54                 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
55         get_bh(bh);
56 }
57
58 /**
59  * gfs2_unpin - Unpin a buffer
60  * @sdp: the filesystem the buffer belongs to
61  * @bh: The buffer to unpin
62  * @ai:
63  *
64  */
65
66 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
67                        struct gfs2_ail *ai)
68 {
69         struct gfs2_bufdata *bd = bh->b_private;
70
71         gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
72
73         if (!buffer_pinned(bh))
74                 gfs2_assert_withdraw(sdp, 0);
75
76         lock_buffer(bh);
77         mark_buffer_dirty(bh);
78         clear_buffer_pinned(bh);
79
80         gfs2_log_lock(sdp);
81         if (bd->bd_ail) {
82                 list_del(&bd->bd_ail_st_list);
83                 brelse(bh);
84         } else {
85                 struct gfs2_glock *gl = bd->bd_gl;
86                 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
87                 atomic_inc(&gl->gl_ail_count);
88         }
89         bd->bd_ail = ai;
90         list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
91         clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
92         gfs2_log_unlock(sdp);
93         unlock_buffer(bh);
94 }
95
96
97 static inline struct gfs2_log_descriptor *bh_log_desc(struct buffer_head *bh)
98 {
99         return (struct gfs2_log_descriptor *)bh->b_data;
100 }
101
102 static inline __be64 *bh_log_ptr(struct buffer_head *bh)
103 {
104         struct gfs2_log_descriptor *ld = bh_log_desc(bh);
105         return (__force __be64 *)(ld + 1);
106 }
107
108 static inline __be64 *bh_ptr_end(struct buffer_head *bh)
109 {
110         return (__force __be64 *)(bh->b_data + bh->b_size);
111 }
112
113
114 static struct buffer_head *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type)
115 {
116         struct buffer_head *bh = gfs2_log_get_buf(sdp);
117         struct gfs2_log_descriptor *ld = bh_log_desc(bh);
118         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
119         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
120         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
121         ld->ld_type = cpu_to_be32(ld_type);
122         ld->ld_length = 0;
123         ld->ld_data1 = 0;
124         ld->ld_data2 = 0;
125         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
126         return bh;
127 }
128
129 static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
130 {
131         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
132         struct gfs2_trans *tr;
133
134         lock_buffer(bd->bd_bh);
135         gfs2_log_lock(sdp);
136         if (!list_empty(&bd->bd_list_tr))
137                 goto out;
138         tr = current->journal_info;
139         tr->tr_touched = 1;
140         tr->tr_num_buf++;
141         list_add(&bd->bd_list_tr, &tr->tr_list_buf);
142         if (!list_empty(&le->le_list))
143                 goto out;
144         set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
145         set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
146         gfs2_meta_check(sdp, bd->bd_bh);
147         gfs2_pin(sdp, bd->bd_bh);
148         sdp->sd_log_num_buf++;
149         list_add(&le->le_list, &sdp->sd_log_le_buf);
150         tr->tr_num_buf_new++;
151 out:
152         gfs2_log_unlock(sdp);
153         unlock_buffer(bd->bd_bh);
154 }
155
156 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
157 {
158         struct buffer_head *bh;
159         struct gfs2_log_descriptor *ld;
160         struct gfs2_bufdata *bd1 = NULL, *bd2;
161         unsigned int total;
162         unsigned int limit;
163         unsigned int num;
164         unsigned n;
165         __be64 *ptr;
166
167         limit = buf_limit(sdp);
168         /* for 4k blocks, limit = 503 */
169
170         gfs2_log_lock(sdp);
171         total = sdp->sd_log_num_buf;
172         bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
173         while(total) {
174                 num = total;
175                 if (total > limit)
176                         num = limit;
177                 gfs2_log_unlock(sdp);
178                 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA);
179                 gfs2_log_lock(sdp);
180                 ld = bh_log_desc(bh);
181                 ptr = bh_log_ptr(bh);
182                 ld->ld_length = cpu_to_be32(num + 1);
183                 ld->ld_data1 = cpu_to_be32(num);
184
185                 n = 0;
186                 list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
187                                              bd_le.le_list) {
188                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
189                         if (++n >= num)
190                                 break;
191                 }
192
193                 gfs2_log_unlock(sdp);
194                 submit_bh(WRITE_SYNC_PLUG, bh);
195                 gfs2_log_lock(sdp);
196
197                 n = 0;
198                 list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
199                                              bd_le.le_list) {
200                         get_bh(bd2->bd_bh);
201                         gfs2_log_unlock(sdp);
202                         lock_buffer(bd2->bd_bh);
203                         bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
204                         submit_bh(WRITE_SYNC_PLUG, bh);
205                         gfs2_log_lock(sdp);
206                         if (++n >= num)
207                                 break;
208                 }
209
210                 BUG_ON(total < num);
211                 total -= num;
212         }
213         gfs2_log_unlock(sdp);
214 }
215
216 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
217 {
218         struct list_head *head = &sdp->sd_log_le_buf;
219         struct gfs2_bufdata *bd;
220
221         while (!list_empty(head)) {
222                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
223                 list_del_init(&bd->bd_le.le_list);
224                 sdp->sd_log_num_buf--;
225
226                 gfs2_unpin(sdp, bd->bd_bh, ai);
227         }
228         gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
229 }
230
231 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
232                                struct gfs2_log_header_host *head, int pass)
233 {
234         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
235
236         if (pass != 0)
237                 return;
238
239         sdp->sd_found_blocks = 0;
240         sdp->sd_replayed_blocks = 0;
241 }
242
243 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
244                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
245                                 int pass)
246 {
247         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
248         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
249         struct gfs2_glock *gl = ip->i_gl;
250         unsigned int blks = be32_to_cpu(ld->ld_data1);
251         struct buffer_head *bh_log, *bh_ip;
252         u64 blkno;
253         int error = 0;
254
255         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
256                 return 0;
257
258         gfs2_replay_incr_blk(sdp, &start);
259
260         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
261                 blkno = be64_to_cpu(*ptr++);
262
263                 sdp->sd_found_blocks++;
264
265                 if (gfs2_revoke_check(sdp, blkno, start))
266                         continue;
267
268                 error = gfs2_replay_read_block(jd, start, &bh_log);
269                 if (error)
270                         return error;
271
272                 bh_ip = gfs2_meta_new(gl, blkno);
273                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
274
275                 if (gfs2_meta_check(sdp, bh_ip))
276                         error = -EIO;
277                 else
278                         mark_buffer_dirty(bh_ip);
279
280                 brelse(bh_log);
281                 brelse(bh_ip);
282
283                 if (error)
284                         break;
285
286                 sdp->sd_replayed_blocks++;
287         }
288
289         return error;
290 }
291
292 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
293 {
294         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
295         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
296
297         if (error) {
298                 gfs2_meta_sync(ip->i_gl);
299                 return;
300         }
301         if (pass != 1)
302                 return;
303
304         gfs2_meta_sync(ip->i_gl);
305
306         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
307                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
308 }
309
310 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
311 {
312         struct gfs2_trans *tr;
313
314         tr = current->journal_info;
315         tr->tr_touched = 1;
316         tr->tr_num_revoke++;
317         sdp->sd_log_num_revoke++;
318         list_add(&le->le_list, &sdp->sd_log_le_revoke);
319 }
320
321 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
322 {
323         struct gfs2_log_descriptor *ld;
324         struct gfs2_meta_header *mh;
325         struct buffer_head *bh;
326         unsigned int offset;
327         struct list_head *head = &sdp->sd_log_le_revoke;
328         struct gfs2_bufdata *bd;
329
330         if (!sdp->sd_log_num_revoke)
331                 return;
332
333         bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE);
334         ld = bh_log_desc(bh);
335         ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
336                                                     sizeof(u64)));
337         ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
338         offset = sizeof(struct gfs2_log_descriptor);
339
340         while (!list_empty(head)) {
341                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
342                 list_del_init(&bd->bd_le.le_list);
343                 sdp->sd_log_num_revoke--;
344
345                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
346                         submit_bh(WRITE_SYNC_PLUG, bh);
347
348                         bh = gfs2_log_get_buf(sdp);
349                         mh = (struct gfs2_meta_header *)bh->b_data;
350                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
351                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
352                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
353                         offset = sizeof(struct gfs2_meta_header);
354                 }
355
356                 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(bd->bd_blkno);
357                 kmem_cache_free(gfs2_bufdata_cachep, bd);
358
359                 offset += sizeof(u64);
360         }
361         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
362
363         submit_bh(WRITE_SYNC_PLUG, bh);
364 }
365
366 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
367                                   struct gfs2_log_header_host *head, int pass)
368 {
369         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
370
371         if (pass != 0)
372                 return;
373
374         sdp->sd_found_revokes = 0;
375         sdp->sd_replay_tail = head->lh_tail;
376 }
377
378 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
379                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
380                                    int pass)
381 {
382         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
383         unsigned int blks = be32_to_cpu(ld->ld_length);
384         unsigned int revokes = be32_to_cpu(ld->ld_data1);
385         struct buffer_head *bh;
386         unsigned int offset;
387         u64 blkno;
388         int first = 1;
389         int error;
390
391         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
392                 return 0;
393
394         offset = sizeof(struct gfs2_log_descriptor);
395
396         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
397                 error = gfs2_replay_read_block(jd, start, &bh);
398                 if (error)
399                         return error;
400
401                 if (!first)
402                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
403
404                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
405                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
406
407                         error = gfs2_revoke_add(sdp, blkno, start);
408                         if (error < 0) {
409                                 brelse(bh);
410                                 return error;
411                         }
412                         else if (error)
413                                 sdp->sd_found_revokes++;
414
415                         if (!--revokes)
416                                 break;
417                         offset += sizeof(u64);
418                 }
419
420                 brelse(bh);
421                 offset = sizeof(struct gfs2_meta_header);
422                 first = 0;
423         }
424
425         return 0;
426 }
427
428 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
429 {
430         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
431
432         if (error) {
433                 gfs2_revoke_clean(sdp);
434                 return;
435         }
436         if (pass != 1)
437                 return;
438
439         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
440                 jd->jd_jid, sdp->sd_found_revokes);
441
442         gfs2_revoke_clean(sdp);
443 }
444
445 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
446 {
447         struct gfs2_rgrpd *rgd;
448         struct gfs2_trans *tr = current->journal_info;
449
450         tr->tr_touched = 1;
451
452         rgd = container_of(le, struct gfs2_rgrpd, rd_le);
453
454         gfs2_log_lock(sdp);
455         if (!list_empty(&le->le_list)){
456                 gfs2_log_unlock(sdp);
457                 return;
458         }
459         gfs2_rgrp_bh_hold(rgd);
460         sdp->sd_log_num_rg++;
461         list_add(&le->le_list, &sdp->sd_log_le_rg);
462         gfs2_log_unlock(sdp);
463 }
464
465 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
466 {
467         struct list_head *head = &sdp->sd_log_le_rg;
468         struct gfs2_rgrpd *rgd;
469
470         while (!list_empty(head)) {
471                 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
472                 list_del_init(&rgd->rd_le.le_list);
473                 sdp->sd_log_num_rg--;
474
475                 gfs2_rgrp_repolish_clones(rgd);
476                 gfs2_rgrp_bh_put(rgd);
477         }
478         gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
479 }
480
481 /**
482  * databuf_lo_add - Add a databuf to the transaction.
483  *
484  * This is used in two distinct cases:
485  * i) In ordered write mode
486  *    We put the data buffer on a list so that we can ensure that its
487  *    synced to disk at the right time
488  * ii) In journaled data mode
489  *    We need to journal the data block in the same way as metadata in
490  *    the functions above. The difference is that here we have a tag
491  *    which is two __be64's being the block number (as per meta data)
492  *    and a flag which says whether the data block needs escaping or
493  *    not. This means we need a new log entry for each 251 or so data
494  *    blocks, which isn't an enormous overhead but twice as much as
495  *    for normal metadata blocks.
496  */
497 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
498 {
499         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
500         struct gfs2_trans *tr = current->journal_info;
501         struct address_space *mapping = bd->bd_bh->b_page->mapping;
502         struct gfs2_inode *ip = GFS2_I(mapping->host);
503
504         lock_buffer(bd->bd_bh);
505         gfs2_log_lock(sdp);
506         if (tr) {
507                 if (!list_empty(&bd->bd_list_tr))
508                         goto out;
509                 tr->tr_touched = 1;
510                 if (gfs2_is_jdata(ip)) {
511                         tr->tr_num_buf++;
512                         list_add(&bd->bd_list_tr, &tr->tr_list_buf);
513                 }
514         }
515         if (!list_empty(&le->le_list))
516                 goto out;
517
518         set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
519         set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
520         if (gfs2_is_jdata(ip)) {
521                 gfs2_pin(sdp, bd->bd_bh);
522                 tr->tr_num_databuf_new++;
523                 sdp->sd_log_num_databuf++;
524                 list_add(&le->le_list, &sdp->sd_log_le_databuf);
525         } else {
526                 list_add(&le->le_list, &sdp->sd_log_le_ordered);
527         }
528 out:
529         gfs2_log_unlock(sdp);
530         unlock_buffer(bd->bd_bh);
531 }
532
533 static void gfs2_check_magic(struct buffer_head *bh)
534 {
535         void *kaddr;
536         __be32 *ptr;
537
538         clear_buffer_escaped(bh);
539         kaddr = kmap_atomic(bh->b_page, KM_USER0);
540         ptr = kaddr + bh_offset(bh);
541         if (*ptr == cpu_to_be32(GFS2_MAGIC))
542                 set_buffer_escaped(bh);
543         kunmap_atomic(kaddr, KM_USER0);
544 }
545
546 static void gfs2_write_blocks(struct gfs2_sbd *sdp, struct buffer_head *bh,
547                               struct list_head *list, struct list_head *done,
548                               unsigned int n)
549 {
550         struct buffer_head *bh1;
551         struct gfs2_log_descriptor *ld;
552         struct gfs2_bufdata *bd;
553         __be64 *ptr;
554
555         if (!bh)
556                 return;
557
558         ld = bh_log_desc(bh);
559         ld->ld_length = cpu_to_be32(n + 1);
560         ld->ld_data1 = cpu_to_be32(n);
561
562         ptr = bh_log_ptr(bh);
563         
564         get_bh(bh);
565         submit_bh(WRITE_SYNC_PLUG, bh);
566         gfs2_log_lock(sdp);
567         while(!list_empty(list)) {
568                 bd = list_entry(list->next, struct gfs2_bufdata, bd_le.le_list);
569                 list_move_tail(&bd->bd_le.le_list, done);
570                 get_bh(bd->bd_bh);
571                 while (be64_to_cpu(*ptr) != bd->bd_bh->b_blocknr) {
572                         gfs2_log_incr_head(sdp);
573                         ptr += 2;
574                 }
575                 gfs2_log_unlock(sdp);
576                 lock_buffer(bd->bd_bh);
577                 if (buffer_escaped(bd->bd_bh)) {
578                         void *kaddr;
579                         bh1 = gfs2_log_get_buf(sdp);
580                         kaddr = kmap_atomic(bd->bd_bh->b_page, KM_USER0);
581                         memcpy(bh1->b_data, kaddr + bh_offset(bd->bd_bh),
582                                bh1->b_size);
583                         kunmap_atomic(kaddr, KM_USER0);
584                         *(__be32 *)bh1->b_data = 0;
585                         clear_buffer_escaped(bd->bd_bh);
586                         unlock_buffer(bd->bd_bh);
587                         brelse(bd->bd_bh);
588                 } else {
589                         bh1 = gfs2_log_fake_buf(sdp, bd->bd_bh);
590                 }
591                 submit_bh(WRITE_SYNC_PLUG, bh1);
592                 gfs2_log_lock(sdp);
593                 ptr += 2;
594         }
595         gfs2_log_unlock(sdp);
596         brelse(bh);
597 }
598
599 /**
600  * databuf_lo_before_commit - Scan the data buffers, writing as we go
601  *
602  */
603
604 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
605 {
606         struct gfs2_bufdata *bd = NULL;
607         struct buffer_head *bh = NULL;
608         unsigned int n = 0;
609         __be64 *ptr = NULL, *end = NULL;
610         LIST_HEAD(processed);
611         LIST_HEAD(in_progress);
612
613         gfs2_log_lock(sdp);
614         while (!list_empty(&sdp->sd_log_le_databuf)) {
615                 if (ptr == end) {
616                         gfs2_log_unlock(sdp);
617                         gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
618                         n = 0;
619                         bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_JDATA);
620                         ptr = bh_log_ptr(bh);
621                         end = bh_ptr_end(bh) - 1;
622                         gfs2_log_lock(sdp);
623                         continue;
624                 }
625                 bd = list_entry(sdp->sd_log_le_databuf.next, struct gfs2_bufdata, bd_le.le_list);
626                 list_move_tail(&bd->bd_le.le_list, &in_progress);
627                 gfs2_check_magic(bd->bd_bh);
628                 *ptr++ = cpu_to_be64(bd->bd_bh->b_blocknr);
629                 *ptr++ = cpu_to_be64(buffer_escaped(bh) ? 1 : 0);
630                 n++;
631         }
632         gfs2_log_unlock(sdp);
633         gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
634         gfs2_log_lock(sdp);
635         list_splice(&processed, &sdp->sd_log_le_databuf);
636         gfs2_log_unlock(sdp);
637 }
638
639 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
640                                     struct gfs2_log_descriptor *ld,
641                                     __be64 *ptr, int pass)
642 {
643         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
644         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
645         struct gfs2_glock *gl = ip->i_gl;
646         unsigned int blks = be32_to_cpu(ld->ld_data1);
647         struct buffer_head *bh_log, *bh_ip;
648         u64 blkno;
649         u64 esc;
650         int error = 0;
651
652         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
653                 return 0;
654
655         gfs2_replay_incr_blk(sdp, &start);
656         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
657                 blkno = be64_to_cpu(*ptr++);
658                 esc = be64_to_cpu(*ptr++);
659
660                 sdp->sd_found_blocks++;
661
662                 if (gfs2_revoke_check(sdp, blkno, start))
663                         continue;
664
665                 error = gfs2_replay_read_block(jd, start, &bh_log);
666                 if (error)
667                         return error;
668
669                 bh_ip = gfs2_meta_new(gl, blkno);
670                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
671
672                 /* Unescape */
673                 if (esc) {
674                         __be32 *eptr = (__be32 *)bh_ip->b_data;
675                         *eptr = cpu_to_be32(GFS2_MAGIC);
676                 }
677                 mark_buffer_dirty(bh_ip);
678
679                 brelse(bh_log);
680                 brelse(bh_ip);
681                 if (error)
682                         break;
683
684                 sdp->sd_replayed_blocks++;
685         }
686
687         return error;
688 }
689
690 /* FIXME: sort out accounting for log blocks etc. */
691
692 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
693 {
694         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
695         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
696
697         if (error) {
698                 gfs2_meta_sync(ip->i_gl);
699                 return;
700         }
701         if (pass != 1)
702                 return;
703
704         /* data sync? */
705         gfs2_meta_sync(ip->i_gl);
706
707         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
708                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
709 }
710
711 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
712 {
713         struct list_head *head = &sdp->sd_log_le_databuf;
714         struct gfs2_bufdata *bd;
715
716         while (!list_empty(head)) {
717                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
718                 list_del_init(&bd->bd_le.le_list);
719                 sdp->sd_log_num_databuf--;
720                 gfs2_unpin(sdp, bd->bd_bh, ai);
721         }
722         gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
723 }
724
725
726 const struct gfs2_log_operations gfs2_buf_lops = {
727         .lo_add = buf_lo_add,
728         .lo_before_commit = buf_lo_before_commit,
729         .lo_after_commit = buf_lo_after_commit,
730         .lo_before_scan = buf_lo_before_scan,
731         .lo_scan_elements = buf_lo_scan_elements,
732         .lo_after_scan = buf_lo_after_scan,
733         .lo_name = "buf",
734 };
735
736 const struct gfs2_log_operations gfs2_revoke_lops = {
737         .lo_add = revoke_lo_add,
738         .lo_before_commit = revoke_lo_before_commit,
739         .lo_before_scan = revoke_lo_before_scan,
740         .lo_scan_elements = revoke_lo_scan_elements,
741         .lo_after_scan = revoke_lo_after_scan,
742         .lo_name = "revoke",
743 };
744
745 const struct gfs2_log_operations gfs2_rg_lops = {
746         .lo_add = rg_lo_add,
747         .lo_after_commit = rg_lo_after_commit,
748         .lo_name = "rg",
749 };
750
751 const struct gfs2_log_operations gfs2_databuf_lops = {
752         .lo_add = databuf_lo_add,
753         .lo_before_commit = databuf_lo_before_commit,
754         .lo_after_commit = databuf_lo_after_commit,
755         .lo_scan_elements = databuf_lo_scan_elements,
756         .lo_after_scan = databuf_lo_after_scan,
757         .lo_name = "databuf",
758 };
759
760 const struct gfs2_log_operations *gfs2_log_ops[] = {
761         &gfs2_databuf_lops,
762         &gfs2_buf_lops,
763         &gfs2_rg_lops,
764         &gfs2_revoke_lops,
765         NULL,
766 };
767