[GFS2] Change all types to uX style
[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
17 #include "gfs2.h"
18 #include "lm_interface.h"
19 #include "incore.h"
20 #include "glock.h"
21 #include "log.h"
22 #include "lops.h"
23 #include "meta_io.h"
24 #include "recovery.h"
25 #include "rgrp.h"
26 #include "trans.h"
27 #include "util.h"
28
29 static void glock_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
30 {
31         struct gfs2_glock *gl;
32         struct gfs2_trans *tr = current->journal_info;
33
34         tr->tr_touched = 1;
35
36         if (!list_empty(&le->le_list))
37                 return;
38
39         gl = container_of(le, struct gfs2_glock, gl_le);
40         if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(gl)))
41                 return;
42         gfs2_glock_hold(gl);
43         set_bit(GLF_DIRTY, &gl->gl_flags);
44
45         gfs2_log_lock(sdp);
46         sdp->sd_log_num_gl++;
47         list_add(&le->le_list, &sdp->sd_log_le_gl);
48         gfs2_log_unlock(sdp);
49 }
50
51 static void glock_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
52 {
53         struct list_head *head = &sdp->sd_log_le_gl;
54         struct gfs2_glock *gl;
55
56         while (!list_empty(head)) {
57                 gl = list_entry(head->next, struct gfs2_glock, gl_le.le_list);
58                 list_del_init(&gl->gl_le.le_list);
59                 sdp->sd_log_num_gl--;
60
61                 gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(gl));
62                 gfs2_glock_put(gl);
63         }
64         gfs2_assert_warn(sdp, !sdp->sd_log_num_gl);
65 }
66
67 static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
68 {
69         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
70         struct gfs2_trans *tr;
71
72         if (!list_empty(&bd->bd_list_tr))
73                 return;
74
75         tr = current->journal_info;
76         tr->tr_touched = 1;
77         tr->tr_num_buf++;
78         list_add(&bd->bd_list_tr, &tr->tr_list_buf);
79
80         if (!list_empty(&le->le_list))
81                 return;
82
83         gfs2_trans_add_gl(bd->bd_gl);
84
85         gfs2_meta_check(sdp, bd->bd_bh);
86         gfs2_pin(sdp, bd->bd_bh);
87
88         gfs2_log_lock(sdp);
89         sdp->sd_log_num_buf++;
90         list_add(&le->le_list, &sdp->sd_log_le_buf);
91         gfs2_log_unlock(sdp);
92
93         tr->tr_num_buf_new++;
94 }
95
96 static void buf_lo_incore_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
97 {
98         struct list_head *head = &tr->tr_list_buf;
99         struct gfs2_bufdata *bd;
100
101         while (!list_empty(head)) {
102                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list_tr);
103                 list_del_init(&bd->bd_list_tr);
104                 tr->tr_num_buf--;
105         }
106         gfs2_assert_warn(sdp, !tr->tr_num_buf);
107 }
108
109 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
110 {
111         struct buffer_head *bh;
112         struct gfs2_log_descriptor *ld;
113         struct gfs2_bufdata *bd1 = NULL, *bd2;
114         unsigned int total = sdp->sd_log_num_buf;
115         unsigned int offset = sizeof(struct gfs2_log_descriptor);
116         unsigned int limit;
117         unsigned int num;
118         unsigned n;
119         __be64 *ptr;
120
121         offset += (sizeof(__be64) - 1);
122         offset &= ~(sizeof(__be64) - 1);
123         limit = (sdp->sd_sb.sb_bsize - offset)/sizeof(__be64);
124         /* for 4k blocks, limit = 503 */
125
126         bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
127         while(total) {
128                 num = total;
129                 if (total > limit)
130                         num = limit;
131                 bh = gfs2_log_get_buf(sdp);
132                 sdp->sd_log_num_hdrs++;
133                 ld = (struct gfs2_log_descriptor *)bh->b_data;
134                 ptr = (__be64 *)(bh->b_data + offset);
135                 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
136                 ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
137                 ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
138                 ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_METADATA);
139                 ld->ld_length = cpu_to_be32(num + 1);
140                 ld->ld_data1 = cpu_to_be32(num);
141                 ld->ld_data2 = cpu_to_be32(0);
142                 memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
143
144                 n = 0;
145                 list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
146                                              bd_le.le_list) {
147                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
148                         if (++n >= num)
149                                 break;
150                 }
151
152                 set_buffer_dirty(bh);
153                 ll_rw_block(WRITE, 1, &bh);
154
155                 n = 0;
156                 list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
157                                              bd_le.le_list) {
158                         bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
159                         set_buffer_dirty(bh);
160                         ll_rw_block(WRITE, 1, &bh);
161                         if (++n >= num)
162                                 break;
163                 }
164
165                 total -= num;
166         }
167 }
168
169 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
170 {
171         struct list_head *head = &sdp->sd_log_le_buf;
172         struct gfs2_bufdata *bd;
173
174         while (!list_empty(head)) {
175                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
176                 list_del_init(&bd->bd_le.le_list);
177                 sdp->sd_log_num_buf--;
178
179                 gfs2_unpin(sdp, bd->bd_bh, ai);
180         }
181         gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
182 }
183
184 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
185                                struct gfs2_log_header *head, int pass)
186 {
187         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
188
189         if (pass != 0)
190                 return;
191
192         sdp->sd_found_blocks = 0;
193         sdp->sd_replayed_blocks = 0;
194 }
195
196 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
197                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
198                                 int pass)
199 {
200         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
201         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
202         struct gfs2_glock *gl = ip->i_gl;
203         unsigned int blks = be32_to_cpu(ld->ld_data1);
204         struct buffer_head *bh_log, *bh_ip;
205         u64 blkno;
206         int error = 0;
207
208         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
209                 return 0;
210
211         gfs2_replay_incr_blk(sdp, &start);
212
213         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
214                 blkno = be64_to_cpu(*ptr++);
215
216                 sdp->sd_found_blocks++;
217
218                 if (gfs2_revoke_check(sdp, blkno, start))
219                         continue;
220
221                 error = gfs2_replay_read_block(jd, start, &bh_log);
222                 if (error)
223                         return error;
224
225                 bh_ip = gfs2_meta_new(gl, blkno);
226                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
227
228                 if (gfs2_meta_check(sdp, bh_ip))
229                         error = -EIO;
230                 else
231                         mark_buffer_dirty(bh_ip);
232
233                 brelse(bh_log);
234                 brelse(bh_ip);
235
236                 if (error)
237                         break;
238
239                 sdp->sd_replayed_blocks++;
240         }
241
242         return error;
243 }
244
245 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
246 {
247         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
248         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
249
250         if (error) {
251                 gfs2_meta_sync(ip->i_gl,
252                                DIO_START | DIO_WAIT);
253                 return;
254         }
255         if (pass != 1)
256                 return;
257
258         gfs2_meta_sync(ip->i_gl, DIO_START | DIO_WAIT);
259
260         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
261                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
262 }
263
264 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
265 {
266         struct gfs2_trans *tr;
267
268         tr = current->journal_info;
269         tr->tr_touched = 1;
270         tr->tr_num_revoke++;
271
272         gfs2_log_lock(sdp);
273         sdp->sd_log_num_revoke++;
274         list_add(&le->le_list, &sdp->sd_log_le_revoke);
275         gfs2_log_unlock(sdp);
276 }
277
278 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
279 {
280         struct gfs2_log_descriptor *ld;
281         struct gfs2_meta_header *mh;
282         struct buffer_head *bh;
283         unsigned int offset;
284         struct list_head *head = &sdp->sd_log_le_revoke;
285         struct gfs2_revoke *rv;
286
287         if (!sdp->sd_log_num_revoke)
288                 return;
289
290         bh = gfs2_log_get_buf(sdp);
291         ld = (struct gfs2_log_descriptor *)bh->b_data;
292         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
293         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
294         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
295         ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_REVOKE);
296         ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
297                                                     sizeof(u64)));
298         ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
299         ld->ld_data2 = cpu_to_be32(0);
300         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
301         offset = sizeof(struct gfs2_log_descriptor);
302
303         while (!list_empty(head)) {
304                 rv = list_entry(head->next, struct gfs2_revoke, rv_le.le_list);
305                 list_del_init(&rv->rv_le.le_list);
306                 sdp->sd_log_num_revoke--;
307
308                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
309                         set_buffer_dirty(bh);
310                         ll_rw_block(WRITE, 1, &bh);
311
312                         bh = gfs2_log_get_buf(sdp);
313                         mh = (struct gfs2_meta_header *)bh->b_data;
314                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
315                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
316                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
317                         offset = sizeof(struct gfs2_meta_header);
318                 }
319
320                 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(rv->rv_blkno);
321                 kfree(rv);
322
323                 offset += sizeof(u64);
324         }
325         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
326
327         set_buffer_dirty(bh);
328         ll_rw_block(WRITE, 1, &bh);
329 }
330
331 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
332                                   struct gfs2_log_header *head, int pass)
333 {
334         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
335
336         if (pass != 0)
337                 return;
338
339         sdp->sd_found_revokes = 0;
340         sdp->sd_replay_tail = head->lh_tail;
341 }
342
343 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
344                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
345                                    int pass)
346 {
347         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
348         unsigned int blks = be32_to_cpu(ld->ld_length);
349         unsigned int revokes = be32_to_cpu(ld->ld_data1);
350         struct buffer_head *bh;
351         unsigned int offset;
352         u64 blkno;
353         int first = 1;
354         int error;
355
356         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
357                 return 0;
358
359         offset = sizeof(struct gfs2_log_descriptor);
360
361         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
362                 error = gfs2_replay_read_block(jd, start, &bh);
363                 if (error)
364                         return error;
365
366                 if (!first)
367                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
368
369                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
370                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
371
372                         error = gfs2_revoke_add(sdp, blkno, start);
373                         if (error < 0)
374                                 return error;
375                         else if (error)
376                                 sdp->sd_found_revokes++;
377
378                         if (!--revokes)
379                                 break;
380                         offset += sizeof(u64);
381                 }
382
383                 brelse(bh);
384                 offset = sizeof(struct gfs2_meta_header);
385                 first = 0;
386         }
387
388         return 0;
389 }
390
391 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
392 {
393         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
394
395         if (error) {
396                 gfs2_revoke_clean(sdp);
397                 return;
398         }
399         if (pass != 1)
400                 return;
401
402         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
403                 jd->jd_jid, sdp->sd_found_revokes);
404
405         gfs2_revoke_clean(sdp);
406 }
407
408 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
409 {
410         struct gfs2_rgrpd *rgd;
411         struct gfs2_trans *tr = current->journal_info;
412
413         tr->tr_touched = 1;
414
415         if (!list_empty(&le->le_list))
416                 return;
417
418         rgd = container_of(le, struct gfs2_rgrpd, rd_le);
419         gfs2_rgrp_bh_hold(rgd);
420
421         gfs2_log_lock(sdp);
422         sdp->sd_log_num_rg++;
423         list_add(&le->le_list, &sdp->sd_log_le_rg);
424         gfs2_log_unlock(sdp);   
425 }
426
427 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
428 {
429         struct list_head *head = &sdp->sd_log_le_rg;
430         struct gfs2_rgrpd *rgd;
431
432         while (!list_empty(head)) {
433                 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
434                 list_del_init(&rgd->rd_le.le_list);
435                 sdp->sd_log_num_rg--;
436
437                 gfs2_rgrp_repolish_clones(rgd);
438                 gfs2_rgrp_bh_put(rgd);
439         }
440         gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
441 }
442
443 /**
444  * databuf_lo_add - Add a databuf to the transaction.
445  *
446  * This is used in two distinct cases:
447  * i) In ordered write mode
448  *    We put the data buffer on a list so that we can ensure that its
449  *    synced to disk at the right time
450  * ii) In journaled data mode
451  *    We need to journal the data block in the same way as metadata in
452  *    the functions above. The difference is that here we have a tag
453  *    which is two __be64's being the block number (as per meta data)
454  *    and a flag which says whether the data block needs escaping or
455  *    not. This means we need a new log entry for each 251 or so data
456  *    blocks, which isn't an enormous overhead but twice as much as
457  *    for normal metadata blocks.
458  */
459 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
460 {
461         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
462         struct gfs2_trans *tr = current->journal_info;
463         struct address_space *mapping = bd->bd_bh->b_page->mapping;
464         struct gfs2_inode *ip = GFS2_I(mapping->host);
465
466         tr->tr_touched = 1;
467         if (list_empty(&bd->bd_list_tr) &&
468             (ip->i_di.di_flags & GFS2_DIF_JDATA)) {
469                 tr->tr_num_buf++;
470                 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
471                 gfs2_pin(sdp, bd->bd_bh);
472                 tr->tr_num_buf_new++;
473         }
474         gfs2_trans_add_gl(bd->bd_gl);
475         gfs2_log_lock(sdp);
476         if (list_empty(&le->le_list)) {
477                 if (ip->i_di.di_flags & GFS2_DIF_JDATA)
478                         sdp->sd_log_num_jdata++;
479                 sdp->sd_log_num_databuf++;
480                 list_add(&le->le_list, &sdp->sd_log_le_databuf);
481         }
482         gfs2_log_unlock(sdp);
483 }
484
485 static int gfs2_check_magic(struct buffer_head *bh)
486 {
487         struct page *page = bh->b_page;
488         void *kaddr;
489         __be32 *ptr;
490         int rv = 0;
491
492         kaddr = kmap_atomic(page, KM_USER0);
493         ptr = kaddr + bh_offset(bh);
494         if (*ptr == cpu_to_be32(GFS2_MAGIC))
495                 rv = 1;
496         kunmap_atomic(page, KM_USER0);
497
498         return rv;
499 }
500
501 /**
502  * databuf_lo_before_commit - Scan the data buffers, writing as we go
503  *
504  * Here we scan through the lists of buffers and make the assumption
505  * that any buffer thats been pinned is being journaled, and that
506  * any unpinned buffer is an ordered write data buffer and therefore
507  * will be written back rather than journaled.
508  */
509 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
510 {
511         LIST_HEAD(started);
512         struct gfs2_bufdata *bd1 = NULL, *bd2, *bdt;
513         struct buffer_head *bh = NULL;
514         unsigned int offset = sizeof(struct gfs2_log_descriptor);
515         struct gfs2_log_descriptor *ld;
516         unsigned int limit;
517         unsigned int total_dbuf = sdp->sd_log_num_databuf;
518         unsigned int total_jdata = sdp->sd_log_num_jdata;
519         unsigned int num, n;
520         __be64 *ptr = NULL;
521
522         offset += (2*sizeof(__be64) - 1);
523         offset &= ~(2*sizeof(__be64) - 1);
524         limit = (sdp->sd_sb.sb_bsize - offset)/sizeof(__be64);
525
526         /*
527          * Start writing ordered buffers, write journaled buffers
528          * into the log along with a header
529          */
530         gfs2_log_lock(sdp);
531         bd2 = bd1 = list_prepare_entry(bd1, &sdp->sd_log_le_databuf,
532                                        bd_le.le_list);
533         while(total_dbuf) {
534                 num = total_jdata;
535                 if (num > limit)
536                         num = limit;
537                 n = 0;
538                 list_for_each_entry_safe_continue(bd1, bdt,
539                                                   &sdp->sd_log_le_databuf,
540                                                   bd_le.le_list) {
541                         /* An ordered write buffer */
542                         if (bd1->bd_bh && !buffer_pinned(bd1->bd_bh)) {
543                                 list_move(&bd1->bd_le.le_list, &started);
544                                 if (bd1 == bd2) {
545                                         bd2 = NULL;
546                                         bd2 = list_prepare_entry(bd2,
547                                                         &sdp->sd_log_le_databuf,
548                                                         bd_le.le_list);
549                                 }
550                                 total_dbuf--;
551                                 if (bd1->bd_bh) {
552                                         get_bh(bd1->bd_bh);
553                                         if (buffer_dirty(bd1->bd_bh)) {
554                                                 gfs2_log_unlock(sdp);
555                                                 wait_on_buffer(bd1->bd_bh);
556                                                 ll_rw_block(WRITE, 1,
557                                                             &bd1->bd_bh);
558                                                 gfs2_log_lock(sdp);
559                                         }
560                                         brelse(bd1->bd_bh);
561                                         continue;
562                                 }
563                                 continue;
564                         } else if (bd1->bd_bh) { /* A journaled buffer */
565                                 int magic;
566                                 gfs2_log_unlock(sdp);
567                                 if (!bh) {
568                                         bh = gfs2_log_get_buf(sdp);
569                                         sdp->sd_log_num_hdrs++;
570                                         ld = (struct gfs2_log_descriptor *)
571                                              bh->b_data;
572                                         ptr = (__be64 *)(bh->b_data + offset);
573                                         ld->ld_header.mh_magic =
574                                                 cpu_to_be32(GFS2_MAGIC);
575                                         ld->ld_header.mh_type =
576                                                 cpu_to_be32(GFS2_METATYPE_LD);
577                                         ld->ld_header.mh_format =
578                                                 cpu_to_be32(GFS2_FORMAT_LD);
579                                         ld->ld_type =
580                                                 cpu_to_be32(GFS2_LOG_DESC_JDATA);
581                                         ld->ld_length = cpu_to_be32(num + 1);
582                                         ld->ld_data1 = cpu_to_be32(num);
583                                         ld->ld_data2 = cpu_to_be32(0);
584                                         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
585                                 }
586                                 magic = gfs2_check_magic(bd1->bd_bh);
587                                 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
588                                 *ptr++ = cpu_to_be64((__u64)magic);
589                                 clear_buffer_escaped(bd1->bd_bh);
590                                 if (unlikely(magic != 0))
591                                         set_buffer_escaped(bd1->bd_bh);
592                                 gfs2_log_lock(sdp);
593                                 if (n++ > num)
594                                         break;
595                         } else if (!bd1->bd_bh) {
596                                 total_dbuf--;
597                                 sdp->sd_log_num_databuf--;
598                                 list_del_init(&bd1->bd_le.le_list);
599                                 if (bd1 == bd2) {
600                                         bd2 = NULL;
601                                         bd2 = list_prepare_entry(bd2,
602                                                 &sdp->sd_log_le_databuf,
603                                                 bd_le.le_list);
604                                 }
605                                 kmem_cache_free(gfs2_bufdata_cachep, bd1);
606                         }
607                 }
608                 gfs2_log_unlock(sdp);
609                 if (bh) {
610                         set_buffer_dirty(bh);
611                         ll_rw_block(WRITE, 1, &bh);
612                         bh = NULL;
613                 }
614                 n = 0;
615                 gfs2_log_lock(sdp);
616                 list_for_each_entry_continue(bd2, &sdp->sd_log_le_databuf,
617                                              bd_le.le_list) {
618                         if (!bd2->bd_bh)
619                                 continue;
620                         /* copy buffer if it needs escaping */
621                         gfs2_log_unlock(sdp);
622                         if (unlikely(buffer_escaped(bd2->bd_bh))) {
623                                 void *kaddr;
624                                 struct page *page = bd2->bd_bh->b_page;
625                                 bh = gfs2_log_get_buf(sdp);
626                                 kaddr = kmap_atomic(page, KM_USER0);
627                                 memcpy(bh->b_data,
628                                        kaddr + bh_offset(bd2->bd_bh),
629                                        sdp->sd_sb.sb_bsize);
630                                 kunmap_atomic(page, KM_USER0);
631                                 *(__be32 *)bh->b_data = 0;
632                         } else {
633                                 bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
634                         }
635                         set_buffer_dirty(bh);
636                         ll_rw_block(WRITE, 1, &bh);
637                         gfs2_log_lock(sdp);
638                         if (++n >= num)
639                                 break;
640                 }
641                 bh = NULL;
642                 total_dbuf -= num;
643                 total_jdata -= num;
644         }
645         gfs2_log_unlock(sdp);
646
647         /* Wait on all ordered buffers */
648         while (!list_empty(&started)) {
649                 gfs2_log_lock(sdp);
650                 bd1 = list_entry(started.next, struct gfs2_bufdata,
651                                  bd_le.le_list);
652                 list_del_init(&bd1->bd_le.le_list);
653                 sdp->sd_log_num_databuf--;
654                 bh = bd1->bd_bh;
655                 if (bh) {
656                         bh->b_private = NULL;
657                         get_bh(bh);
658                         gfs2_log_unlock(sdp);
659                         wait_on_buffer(bh);
660                         brelse(bh);
661                 } else
662                         gfs2_log_unlock(sdp);
663
664                 kmem_cache_free(gfs2_bufdata_cachep, bd1);
665         }
666
667         /* We've removed all the ordered write bufs here, so only jdata left */
668         gfs2_assert_warn(sdp, sdp->sd_log_num_databuf == sdp->sd_log_num_jdata);
669 }
670
671 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
672                                     struct gfs2_log_descriptor *ld,
673                                     __be64 *ptr, int pass)
674 {
675         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
676         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
677         struct gfs2_glock *gl = ip->i_gl;
678         unsigned int blks = be32_to_cpu(ld->ld_data1);
679         struct buffer_head *bh_log, *bh_ip;
680         u64 blkno;
681         u64 esc;
682         int error = 0;
683
684         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
685                 return 0;
686
687         gfs2_replay_incr_blk(sdp, &start);
688         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
689                 blkno = be64_to_cpu(*ptr++);
690                 esc = be64_to_cpu(*ptr++);
691
692                 sdp->sd_found_blocks++;
693
694                 if (gfs2_revoke_check(sdp, blkno, start))
695                         continue;
696
697                 error = gfs2_replay_read_block(jd, start, &bh_log);
698                 if (error)
699                         return error;
700
701                 bh_ip = gfs2_meta_new(gl, blkno);
702                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
703
704                 /* Unescape */
705                 if (esc) {
706                         __be32 *eptr = (__be32 *)bh_ip->b_data;
707                         *eptr = cpu_to_be32(GFS2_MAGIC);
708                 }
709                 mark_buffer_dirty(bh_ip);
710
711                 brelse(bh_log);
712                 brelse(bh_ip);
713                 if (error)
714                         break;
715
716                 sdp->sd_replayed_blocks++;
717         }
718
719         return error;
720 }
721
722 /* FIXME: sort out accounting for log blocks etc. */
723
724 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
725 {
726         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
727         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
728
729         if (error) {
730                 gfs2_meta_sync(ip->i_gl,
731                                DIO_START | DIO_WAIT);
732                 return;
733         }
734         if (pass != 1)
735                 return;
736
737         /* data sync? */
738         gfs2_meta_sync(ip->i_gl, DIO_START | DIO_WAIT);
739
740         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
741                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
742 }
743
744 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
745 {
746         struct list_head *head = &sdp->sd_log_le_databuf;
747         struct gfs2_bufdata *bd;
748
749         while (!list_empty(head)) {
750                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
751                 list_del_init(&bd->bd_le.le_list);
752                 sdp->sd_log_num_databuf--;
753                 sdp->sd_log_num_jdata--;
754                 gfs2_unpin(sdp, bd->bd_bh, ai);
755         }
756         gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
757         gfs2_assert_warn(sdp, !sdp->sd_log_num_jdata);
758 }
759
760
761 const struct gfs2_log_operations gfs2_glock_lops = {
762         .lo_add = glock_lo_add,
763         .lo_after_commit = glock_lo_after_commit,
764         .lo_name = "glock"
765 };
766
767 const struct gfs2_log_operations gfs2_buf_lops = {
768         .lo_add = buf_lo_add,
769         .lo_incore_commit = buf_lo_incore_commit,
770         .lo_before_commit = buf_lo_before_commit,
771         .lo_after_commit = buf_lo_after_commit,
772         .lo_before_scan = buf_lo_before_scan,
773         .lo_scan_elements = buf_lo_scan_elements,
774         .lo_after_scan = buf_lo_after_scan,
775         .lo_name = "buf"
776 };
777
778 const struct gfs2_log_operations gfs2_revoke_lops = {
779         .lo_add = revoke_lo_add,
780         .lo_before_commit = revoke_lo_before_commit,
781         .lo_before_scan = revoke_lo_before_scan,
782         .lo_scan_elements = revoke_lo_scan_elements,
783         .lo_after_scan = revoke_lo_after_scan,
784         .lo_name = "revoke"
785 };
786
787 const struct gfs2_log_operations gfs2_rg_lops = {
788         .lo_add = rg_lo_add,
789         .lo_after_commit = rg_lo_after_commit,
790         .lo_name = "rg"
791 };
792
793 const struct gfs2_log_operations gfs2_databuf_lops = {
794         .lo_add = databuf_lo_add,
795         .lo_incore_commit = buf_lo_incore_commit,
796         .lo_before_commit = databuf_lo_before_commit,
797         .lo_after_commit = databuf_lo_after_commit,
798         .lo_scan_elements = databuf_lo_scan_elements,
799         .lo_after_scan = databuf_lo_after_scan,
800         .lo_name = "databuf"
801 };
802
803 const struct gfs2_log_operations *gfs2_log_ops[] = {
804         &gfs2_glock_lops,
805         &gfs2_buf_lops,
806         &gfs2_revoke_lops,
807         &gfs2_rg_lops,
808         &gfs2_databuf_lops,
809         NULL
810 };
811