22a4f038e3b9eb497fa600566d7d58952bac2829
[linux-2.6.git] / fs / gfs2 / lops.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2005 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 v.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_inode *ip = jd->jd_inode->u.generic_ip;
188         struct gfs2_sbd *sdp = ip->i_sbd;
189
190         if (pass != 0)
191                 return;
192
193         sdp->sd_found_blocks = 0;
194         sdp->sd_replayed_blocks = 0;
195 }
196
197 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
198                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
199                                 int pass)
200 {
201         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
202         struct gfs2_sbd *sdp = ip->i_sbd;
203         struct gfs2_glock *gl = ip->i_gl;
204         unsigned int blks = be32_to_cpu(ld->ld_data1);
205         struct buffer_head *bh_log, *bh_ip;
206         uint64_t blkno;
207         int error = 0;
208
209         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
210                 return 0;
211
212         gfs2_replay_incr_blk(sdp, &start);
213
214         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
215                 blkno = be64_to_cpu(*ptr++);
216
217                 sdp->sd_found_blocks++;
218
219                 if (gfs2_revoke_check(sdp, blkno, start))
220                         continue;
221
222                 error = gfs2_replay_read_block(jd, start, &bh_log);
223                 if (error)
224                         return error;
225
226                 bh_ip = gfs2_meta_new(gl, blkno);
227                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
228
229                 if (gfs2_meta_check(sdp, bh_ip))
230                         error = -EIO;
231                 else
232                         mark_buffer_dirty(bh_ip);
233
234                 brelse(bh_log);
235                 brelse(bh_ip);
236
237                 if (error)
238                         break;
239
240                 sdp->sd_replayed_blocks++;
241         }
242
243         return error;
244 }
245
246 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
247 {
248         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
249         struct gfs2_sbd *sdp = ip->i_sbd;
250
251         if (error) {
252                 gfs2_meta_sync(ip->i_gl,
253                                DIO_START | DIO_WAIT);
254                 return;
255         }
256         if (pass != 1)
257                 return;
258
259         gfs2_meta_sync(ip->i_gl, DIO_START | DIO_WAIT);
260
261         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
262                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
263 }
264
265 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
266 {
267         struct gfs2_trans *tr;
268
269         tr = current->journal_info;
270         tr->tr_touched = 1;
271         tr->tr_num_revoke++;
272
273         gfs2_log_lock(sdp);
274         sdp->sd_log_num_revoke++;
275         list_add(&le->le_list, &sdp->sd_log_le_revoke);
276         gfs2_log_unlock(sdp);
277 }
278
279 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
280 {
281         struct gfs2_log_descriptor *ld;
282         struct gfs2_meta_header *mh;
283         struct buffer_head *bh;
284         unsigned int offset;
285         struct list_head *head = &sdp->sd_log_le_revoke;
286         struct gfs2_revoke *rv;
287
288         if (!sdp->sd_log_num_revoke)
289                 return;
290
291         bh = gfs2_log_get_buf(sdp);
292         ld = (struct gfs2_log_descriptor *)bh->b_data;
293         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
294         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
295         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
296         ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_REVOKE);
297         ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
298                                                     sizeof(uint64_t)));
299         ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
300         ld->ld_data2 = cpu_to_be32(0);
301         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
302         offset = sizeof(struct gfs2_log_descriptor);
303
304         while (!list_empty(head)) {
305                 rv = list_entry(head->next, struct gfs2_revoke, rv_le.le_list);
306                 list_del_init(&rv->rv_le.le_list);
307                 sdp->sd_log_num_revoke--;
308
309                 if (offset + sizeof(uint64_t) > sdp->sd_sb.sb_bsize) {
310                         set_buffer_dirty(bh);
311                         ll_rw_block(WRITE, 1, &bh);
312
313                         bh = gfs2_log_get_buf(sdp);
314                         mh = (struct gfs2_meta_header *)bh->b_data;
315                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
316                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
317                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
318                         offset = sizeof(struct gfs2_meta_header);
319                 }
320
321                 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(rv->rv_blkno);
322                 kfree(rv);
323
324                 offset += sizeof(uint64_t);
325         }
326         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
327
328         set_buffer_dirty(bh);
329         ll_rw_block(WRITE, 1, &bh);
330 }
331
332 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
333                                   struct gfs2_log_header *head, int pass)
334 {
335         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
336         struct gfs2_sbd *sdp = ip->i_sbd;
337
338         if (pass != 0)
339                 return;
340
341         sdp->sd_found_revokes = 0;
342         sdp->sd_replay_tail = head->lh_tail;
343 }
344
345 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
346                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
347                                    int pass)
348 {
349         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
350         struct gfs2_sbd *sdp = ip->i_sbd;
351         unsigned int blks = be32_to_cpu(ld->ld_length);
352         unsigned int revokes = be32_to_cpu(ld->ld_data1);
353         struct buffer_head *bh;
354         unsigned int offset;
355         uint64_t blkno;
356         int first = 1;
357         int error;
358
359         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
360                 return 0;
361
362         offset = sizeof(struct gfs2_log_descriptor);
363
364         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
365                 error = gfs2_replay_read_block(jd, start, &bh);
366                 if (error)
367                         return error;
368
369                 if (!first)
370                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
371
372                 while (offset + sizeof(uint64_t) <= sdp->sd_sb.sb_bsize) {
373                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
374
375                         error = gfs2_revoke_add(sdp, blkno, start);
376                         if (error < 0)
377                                 return error;
378                         else if (error)
379                                 sdp->sd_found_revokes++;
380
381                         if (!--revokes)
382                                 break;
383                         offset += sizeof(uint64_t);
384                 }
385
386                 brelse(bh);
387                 offset = sizeof(struct gfs2_meta_header);
388                 first = 0;
389         }
390
391         return 0;
392 }
393
394 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
395 {
396         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
397         struct gfs2_sbd *sdp = ip->i_sbd;
398
399         if (error) {
400                 gfs2_revoke_clean(sdp);
401                 return;
402         }
403         if (pass != 1)
404                 return;
405
406         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
407                 jd->jd_jid, sdp->sd_found_revokes);
408
409         gfs2_revoke_clean(sdp);
410 }
411
412 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
413 {
414         struct gfs2_rgrpd *rgd;
415         struct gfs2_trans *tr = current->journal_info;
416
417         tr->tr_touched = 1;
418
419         if (!list_empty(&le->le_list))
420                 return;
421
422         rgd = container_of(le, struct gfs2_rgrpd, rd_le);
423         gfs2_rgrp_bh_hold(rgd);
424
425         gfs2_log_lock(sdp);
426         sdp->sd_log_num_rg++;
427         list_add(&le->le_list, &sdp->sd_log_le_rg);
428         gfs2_log_unlock(sdp);   
429 }
430
431 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
432 {
433         struct list_head *head = &sdp->sd_log_le_rg;
434         struct gfs2_rgrpd *rgd;
435
436         while (!list_empty(head)) {
437                 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
438                 list_del_init(&rgd->rd_le.le_list);
439                 sdp->sd_log_num_rg--;
440
441                 gfs2_rgrp_repolish_clones(rgd);
442                 gfs2_rgrp_bh_put(rgd);
443         }
444         gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
445 }
446
447 /**
448  * databuf_lo_add - Add a databuf to the transaction.
449  *
450  * This is used in two distinct cases:
451  * i) In ordered write mode
452  *    We put the data buffer on a list so that we can ensure that its
453  *    synced to disk at the right time
454  * ii) In journaled data mode
455  *    We need to journal the data block in the same way as metadata in
456  *    the functions above. The difference is that here we have a tag
457  *    which is two __be64's being the block number (as per meta data)
458  *    and a flag which says whether the data block needs escaping or
459  *    not. This means we need a new log entry for each 251 or so data
460  *    blocks, which isn't an enormous overhead but twice as much as
461  *    for normal metadata blocks.
462  */
463 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
464 {
465         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
466         struct gfs2_trans *tr = current->journal_info;
467         struct address_space *mapping = bd->bd_bh->b_page->mapping;
468         struct gfs2_inode *ip = mapping->host->u.generic_ip;
469
470         tr->tr_touched = 1;
471         if (!list_empty(&bd->bd_list_tr) &&
472             (ip->i_di.di_flags & GFS2_DIF_JDATA)) {
473                 tr->tr_num_buf++;
474                 gfs2_trans_add_gl(bd->bd_gl);
475                 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
476                 gfs2_pin(sdp, bd->bd_bh);
477                 tr->tr_num_buf_new++;
478         }
479         gfs2_log_lock(sdp);
480         if (!list_empty(&le->le_list)) {
481                 if (ip->i_di.di_flags & GFS2_DIF_JDATA)
482                         sdp->sd_log_num_jdata++;
483                 sdp->sd_log_num_databuf++;
484                 list_add(&le->le_list, &sdp->sd_log_le_databuf);
485         }
486         gfs2_log_unlock(sdp);
487 }
488
489 static int gfs2_check_magic(struct buffer_head *bh)
490 {
491         struct page *page = bh->b_page;
492         void *kaddr;
493         __be32 *ptr;
494         int rv = 0;
495
496         kaddr = kmap_atomic(page, KM_USER0);
497         ptr = kaddr + bh_offset(bh);
498         if (*ptr == cpu_to_be32(GFS2_MAGIC))
499                 rv = 1;
500         kunmap_atomic(page, KM_USER0);
501
502         return rv;
503 }
504
505 /**
506  * databuf_lo_before_commit - Scan the data buffers, writing as we go
507  *
508  * Here we scan through the lists of buffers and make the assumption
509  * that any buffer thats been pinned is being journaled, and that
510  * any unpinned buffer is an ordered write data buffer and therefore
511  * will be written back rather than journaled.
512  */
513 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
514 {
515         LIST_HEAD(started);
516         struct gfs2_bufdata *bd1 = NULL, *bd2, *bdt;
517         struct buffer_head *bh = NULL;
518         unsigned int offset = sizeof(struct gfs2_log_descriptor);
519         struct gfs2_log_descriptor *ld;
520         unsigned int limit;
521         unsigned int total_dbuf = sdp->sd_log_num_databuf;
522         unsigned int total_jdata = sdp->sd_log_num_jdata;
523         unsigned int num, n;
524         __be64 *ptr = NULL;
525
526         offset += (2*sizeof(__be64) - 1);
527         offset &= ~(2*sizeof(__be64) - 1);
528         limit = (sdp->sd_sb.sb_bsize - offset)/sizeof(__be64);
529
530         /*
531          * Start writing ordered buffers, write journaled buffers
532          * into the log along with a header
533          */
534         gfs2_log_lock(sdp);
535         bd2 = bd1 = list_prepare_entry(bd1, &sdp->sd_log_le_databuf,
536                                        bd_le.le_list);
537         while(total_dbuf) {
538                 num = total_jdata;
539                 if (num > limit)
540                         num = limit;
541                 n = 0;
542                 list_for_each_entry_safe_continue(bd1, bdt,
543                                                   &sdp->sd_log_le_databuf,
544                                                   bd_le.le_list) {
545                         /* An ordered write buffer */
546                         if (bd1->bd_bh && !buffer_pinned(bd1->bd_bh)) {
547                                 list_move(&bd1->bd_le.le_list, &started);
548                                 if (bd1 == bd2) {
549                                         bd2 = NULL;
550                                         bd2 = list_prepare_entry(bd2,
551                                                         &sdp->sd_log_le_databuf,
552                                                         bd_le.le_list);
553                                 }
554                                 total_dbuf--;
555                                 if (bd1->bd_bh) {
556                                         get_bh(bd1->bd_bh);
557                                         if (buffer_dirty(bd1->bd_bh)) {
558                                                 gfs2_log_unlock(sdp);
559                                                 wait_on_buffer(bd1->bd_bh);
560                                                 ll_rw_block(WRITE, 1,
561                                                             &bd1->bd_bh);
562                                                 gfs2_log_lock(sdp);
563                                         }
564                                         brelse(bd1->bd_bh);
565                                         continue;
566                                 }
567                                 continue;
568                         } else if (bd1->bd_bh) { /* A journaled buffer */
569                                 int magic;
570                                 gfs2_log_unlock(sdp);
571                                 if (!bh) {
572                                         bh = gfs2_log_get_buf(sdp);
573                                         sdp->sd_log_num_hdrs++;
574                                         ld = (struct gfs2_log_descriptor *)
575                                              bh->b_data;
576                                         ptr = (__be64 *)(bh->b_data + offset);
577                                         ld->ld_header.mh_magic =
578                                                 cpu_to_be32(GFS2_MAGIC);
579                                         ld->ld_header.mh_type =
580                                                 cpu_to_be32(GFS2_METATYPE_LD);
581                                         ld->ld_header.mh_format =
582                                                 cpu_to_be32(GFS2_FORMAT_LD);
583                                         ld->ld_type =
584                                                 cpu_to_be32(GFS2_LOG_DESC_JDATA);
585                                         ld->ld_length = cpu_to_be32(num + 1);
586                                         ld->ld_data1 = cpu_to_be32(num);
587                                         ld->ld_data2 = cpu_to_be32(0);
588                                         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
589                                 }
590                                 magic = gfs2_check_magic(bd1->bd_bh);
591                                 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
592                                 *ptr++ = cpu_to_be64((__u64)magic);
593                                 clear_buffer_escaped(bd1->bd_bh);
594                                 if (unlikely(magic != 0))
595                                         set_buffer_escaped(bd1->bd_bh);
596                                 gfs2_log_lock(sdp);
597                                 if (n++ > num)
598                                         break;
599                         }
600                 }
601                 gfs2_log_unlock(sdp);
602                 if (bh) {
603                         set_buffer_dirty(bh);
604                         ll_rw_block(WRITE, 1, &bh);
605                         bh = NULL;
606                 }
607                 n = 0;
608                 gfs2_log_lock(sdp);
609                 list_for_each_entry_continue(bd2, &sdp->sd_log_le_databuf,
610                                              bd_le.le_list) {
611                         if (!bd2->bd_bh)
612                                 continue;
613                         /* copy buffer if it needs escaping */
614                         gfs2_log_unlock(sdp);
615                         if (unlikely(buffer_escaped(bd2->bd_bh))) {
616                                 void *kaddr;
617                                 struct page *page = bd2->bd_bh->b_page;
618                                 bh = gfs2_log_get_buf(sdp);
619                                 kaddr = kmap_atomic(page, KM_USER0);
620                                 memcpy(bh->b_data,
621                                        kaddr + bh_offset(bd2->bd_bh),
622                                        sdp->sd_sb.sb_bsize);
623                                 kunmap_atomic(page, KM_USER0);
624                                 *(__be32 *)bh->b_data = 0;
625                         } else {
626                                 bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
627                         }
628                         set_buffer_dirty(bh);
629                         ll_rw_block(WRITE, 1, &bh);
630                         gfs2_log_lock(sdp);
631                         if (++n >= num)
632                                 break;
633                 }
634                 bh = NULL;
635                 total_dbuf -= num;
636                 total_jdata -= num;
637         }
638         gfs2_log_unlock(sdp);
639
640         /* Wait on all ordered buffers */
641         while (!list_empty(&started)) {
642                 gfs2_log_lock(sdp);
643                 bd1 = list_entry(started.next, struct gfs2_bufdata,
644                                  bd_le.le_list);
645                 list_del(&bd1->bd_le.le_list);
646                 sdp->sd_log_num_databuf--;
647
648                 bh = bd1->bd_bh;
649                 if (bh) {
650                         bh->b_private = NULL;
651                         gfs2_log_unlock(sdp);
652                         wait_on_buffer(bh);
653                         brelse(bh);
654                 } else
655                         gfs2_log_unlock(sdp);
656
657                 kfree(bd1);
658         }
659
660         /* We've removed all the ordered write bufs here, so only jdata left */
661         gfs2_assert_warn(sdp, sdp->sd_log_num_databuf == sdp->sd_log_num_jdata);
662 }
663
664 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
665                                     struct gfs2_log_descriptor *ld,
666                                     __be64 *ptr, int pass)
667 {
668         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
669         struct gfs2_sbd *sdp = ip->i_sbd;
670         struct gfs2_glock *gl = ip->i_gl;
671         unsigned int blks = be32_to_cpu(ld->ld_data1);
672         struct buffer_head *bh_log, *bh_ip;
673         uint64_t blkno;
674         uint64_t esc;
675         int error = 0;
676
677         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
678                 return 0;
679
680         gfs2_replay_incr_blk(sdp, &start);
681         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
682                 blkno = be64_to_cpu(*ptr++);
683                 esc = be64_to_cpu(*ptr++);
684
685                 sdp->sd_found_blocks++;
686
687                 if (gfs2_revoke_check(sdp, blkno, start))
688                         continue;
689
690                 error = gfs2_replay_read_block(jd, start, &bh_log);
691                 if (error)
692                         return error;
693
694                 bh_ip = gfs2_meta_new(gl, blkno);
695                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
696
697                 /* Unescape */
698                 if (esc) {
699                         __be32 *eptr = (__be32 *)bh_ip->b_data;
700                         *eptr = cpu_to_be32(GFS2_MAGIC);
701                 }
702                 mark_buffer_dirty(bh_ip);
703
704                 brelse(bh_log);
705                 brelse(bh_ip);
706                 if (error)
707                         break;
708
709                 sdp->sd_replayed_blocks++;
710         }
711
712         return error;
713 }
714
715 /* FIXME: sort out accounting for log blocks etc. */
716
717 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
718 {
719         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
720         struct gfs2_sbd *sdp = ip->i_sbd;
721
722         if (error) {
723                 gfs2_meta_sync(ip->i_gl,
724                                DIO_START | DIO_WAIT);
725                 return;
726         }
727         if (pass != 1)
728                 return;
729
730         /* data sync? */
731         gfs2_meta_sync(ip->i_gl, DIO_START | DIO_WAIT);
732
733         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
734                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
735 }
736
737 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
738 {
739         struct list_head *head = &sdp->sd_log_le_databuf;
740         struct gfs2_bufdata *bd;
741
742         while (!list_empty(head)) {
743                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
744                 list_del(&bd->bd_le.le_list);
745                 sdp->sd_log_num_databuf--;
746                 sdp->sd_log_num_jdata--;
747                 gfs2_unpin(sdp, bd->bd_bh, ai);
748         }
749         gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
750         gfs2_assert_warn(sdp, !sdp->sd_log_num_jdata);
751 }
752
753
754 const struct gfs2_log_operations gfs2_glock_lops = {
755         .lo_add = glock_lo_add,
756         .lo_after_commit = glock_lo_after_commit,
757         .lo_name = "glock"
758 };
759
760 const struct gfs2_log_operations gfs2_buf_lops = {
761         .lo_add = buf_lo_add,
762         .lo_incore_commit = buf_lo_incore_commit,
763         .lo_before_commit = buf_lo_before_commit,
764         .lo_after_commit = buf_lo_after_commit,
765         .lo_before_scan = buf_lo_before_scan,
766         .lo_scan_elements = buf_lo_scan_elements,
767         .lo_after_scan = buf_lo_after_scan,
768         .lo_name = "buf"
769 };
770
771 const struct gfs2_log_operations gfs2_revoke_lops = {
772         .lo_add = revoke_lo_add,
773         .lo_before_commit = revoke_lo_before_commit,
774         .lo_before_scan = revoke_lo_before_scan,
775         .lo_scan_elements = revoke_lo_scan_elements,
776         .lo_after_scan = revoke_lo_after_scan,
777         .lo_name = "revoke"
778 };
779
780 const struct gfs2_log_operations gfs2_rg_lops = {
781         .lo_add = rg_lo_add,
782         .lo_after_commit = rg_lo_after_commit,
783         .lo_name = "rg"
784 };
785
786 const struct gfs2_log_operations gfs2_databuf_lops = {
787         .lo_add = databuf_lo_add,
788         .lo_incore_commit = buf_lo_incore_commit,
789         .lo_before_commit = databuf_lo_before_commit,
790         .lo_after_commit = databuf_lo_after_commit,
791         .lo_scan_elements = databuf_lo_scan_elements,
792         .lo_after_scan = databuf_lo_after_scan,
793         .lo_name = "databuf"
794 };
795
796 const struct gfs2_log_operations *gfs2_log_ops[] = {
797         &gfs2_glock_lops,
798         &gfs2_buf_lops,
799         &gfs2_revoke_lops,
800         &gfs2_rg_lops,
801         &gfs2_databuf_lops,
802         NULL
803 };
804