Merge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee13...
[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/lm_interface.h>
17
18 #include "gfs2.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                 return;
253         }
254         if (pass != 1)
255                 return;
256
257         gfs2_meta_sync(ip->i_gl);
258
259         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
260                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
261 }
262
263 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
264 {
265         struct gfs2_trans *tr;
266
267         tr = current->journal_info;
268         tr->tr_touched = 1;
269         tr->tr_num_revoke++;
270
271         gfs2_log_lock(sdp);
272         sdp->sd_log_num_revoke++;
273         list_add(&le->le_list, &sdp->sd_log_le_revoke);
274         gfs2_log_unlock(sdp);
275 }
276
277 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
278 {
279         struct gfs2_log_descriptor *ld;
280         struct gfs2_meta_header *mh;
281         struct buffer_head *bh;
282         unsigned int offset;
283         struct list_head *head = &sdp->sd_log_le_revoke;
284         struct gfs2_revoke *rv;
285
286         if (!sdp->sd_log_num_revoke)
287                 return;
288
289         bh = gfs2_log_get_buf(sdp);
290         ld = (struct gfs2_log_descriptor *)bh->b_data;
291         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
292         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
293         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
294         ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_REVOKE);
295         ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
296                                                     sizeof(u64)));
297         ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
298         ld->ld_data2 = cpu_to_be32(0);
299         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
300         offset = sizeof(struct gfs2_log_descriptor);
301
302         while (!list_empty(head)) {
303                 rv = list_entry(head->next, struct gfs2_revoke, rv_le.le_list);
304                 list_del_init(&rv->rv_le.le_list);
305                 sdp->sd_log_num_revoke--;
306
307                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
308                         set_buffer_dirty(bh);
309                         ll_rw_block(WRITE, 1, &bh);
310
311                         bh = gfs2_log_get_buf(sdp);
312                         mh = (struct gfs2_meta_header *)bh->b_data;
313                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
314                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
315                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
316                         offset = sizeof(struct gfs2_meta_header);
317                 }
318
319                 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(rv->rv_blkno);
320                 kfree(rv);
321
322                 offset += sizeof(u64);
323         }
324         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
325
326         set_buffer_dirty(bh);
327         ll_rw_block(WRITE, 1, &bh);
328 }
329
330 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
331                                   struct gfs2_log_header *head, int pass)
332 {
333         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
334
335         if (pass != 0)
336                 return;
337
338         sdp->sd_found_revokes = 0;
339         sdp->sd_replay_tail = head->lh_tail;
340 }
341
342 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
343                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
344                                    int pass)
345 {
346         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
347         unsigned int blks = be32_to_cpu(ld->ld_length);
348         unsigned int revokes = be32_to_cpu(ld->ld_data1);
349         struct buffer_head *bh;
350         unsigned int offset;
351         u64 blkno;
352         int first = 1;
353         int error;
354
355         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
356                 return 0;
357
358         offset = sizeof(struct gfs2_log_descriptor);
359
360         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
361                 error = gfs2_replay_read_block(jd, start, &bh);
362                 if (error)
363                         return error;
364
365                 if (!first)
366                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
367
368                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
369                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
370
371                         error = gfs2_revoke_add(sdp, blkno, start);
372                         if (error < 0)
373                                 return error;
374                         else if (error)
375                                 sdp->sd_found_revokes++;
376
377                         if (!--revokes)
378                                 break;
379                         offset += sizeof(u64);
380                 }
381
382                 brelse(bh);
383                 offset = sizeof(struct gfs2_meta_header);
384                 first = 0;
385         }
386
387         return 0;
388 }
389
390 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
391 {
392         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
393
394         if (error) {
395                 gfs2_revoke_clean(sdp);
396                 return;
397         }
398         if (pass != 1)
399                 return;
400
401         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
402                 jd->jd_jid, sdp->sd_found_revokes);
403
404         gfs2_revoke_clean(sdp);
405 }
406
407 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
408 {
409         struct gfs2_rgrpd *rgd;
410         struct gfs2_trans *tr = current->journal_info;
411
412         tr->tr_touched = 1;
413
414         if (!list_empty(&le->le_list))
415                 return;
416
417         rgd = container_of(le, struct gfs2_rgrpd, rd_le);
418         gfs2_rgrp_bh_hold(rgd);
419
420         gfs2_log_lock(sdp);
421         sdp->sd_log_num_rg++;
422         list_add(&le->le_list, &sdp->sd_log_le_rg);
423         gfs2_log_unlock(sdp);
424 }
425
426 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
427 {
428         struct list_head *head = &sdp->sd_log_le_rg;
429         struct gfs2_rgrpd *rgd;
430
431         while (!list_empty(head)) {
432                 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
433                 list_del_init(&rgd->rd_le.le_list);
434                 sdp->sd_log_num_rg--;
435
436                 gfs2_rgrp_repolish_clones(rgd);
437                 gfs2_rgrp_bh_put(rgd);
438         }
439         gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
440 }
441
442 /**
443  * databuf_lo_add - Add a databuf to the transaction.
444  *
445  * This is used in two distinct cases:
446  * i) In ordered write mode
447  *    We put the data buffer on a list so that we can ensure that its
448  *    synced to disk at the right time
449  * ii) In journaled data mode
450  *    We need to journal the data block in the same way as metadata in
451  *    the functions above. The difference is that here we have a tag
452  *    which is two __be64's being the block number (as per meta data)
453  *    and a flag which says whether the data block needs escaping or
454  *    not. This means we need a new log entry for each 251 or so data
455  *    blocks, which isn't an enormous overhead but twice as much as
456  *    for normal metadata blocks.
457  */
458 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
459 {
460         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
461         struct gfs2_trans *tr = current->journal_info;
462         struct address_space *mapping = bd->bd_bh->b_page->mapping;
463         struct gfs2_inode *ip = GFS2_I(mapping->host);
464
465         tr->tr_touched = 1;
466         if (list_empty(&bd->bd_list_tr) &&
467             (ip->i_di.di_flags & GFS2_DIF_JDATA)) {
468                 tr->tr_num_buf++;
469                 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
470                 gfs2_pin(sdp, bd->bd_bh);
471                 tr->tr_num_buf_new++;
472         }
473         gfs2_trans_add_gl(bd->bd_gl);
474         gfs2_log_lock(sdp);
475         if (list_empty(&le->le_list)) {
476                 if (ip->i_di.di_flags & GFS2_DIF_JDATA)
477                         sdp->sd_log_num_jdata++;
478                 sdp->sd_log_num_databuf++;
479                 list_add(&le->le_list, &sdp->sd_log_le_databuf);
480         }
481         gfs2_log_unlock(sdp);
482 }
483
484 static int gfs2_check_magic(struct buffer_head *bh)
485 {
486         struct page *page = bh->b_page;
487         void *kaddr;
488         __be32 *ptr;
489         int rv = 0;
490
491         kaddr = kmap_atomic(page, KM_USER0);
492         ptr = kaddr + bh_offset(bh);
493         if (*ptr == cpu_to_be32(GFS2_MAGIC))
494                 rv = 1;
495         kunmap_atomic(page, KM_USER0);
496
497         return rv;
498 }
499
500 /**
501  * databuf_lo_before_commit - Scan the data buffers, writing as we go
502  *
503  * Here we scan through the lists of buffers and make the assumption
504  * that any buffer thats been pinned is being journaled, and that
505  * any unpinned buffer is an ordered write data buffer and therefore
506  * will be written back rather than journaled.
507  */
508 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
509 {
510         LIST_HEAD(started);
511         struct gfs2_bufdata *bd1 = NULL, *bd2, *bdt;
512         struct buffer_head *bh = NULL;
513         unsigned int offset = sizeof(struct gfs2_log_descriptor);
514         struct gfs2_log_descriptor *ld;
515         unsigned int limit;
516         unsigned int total_dbuf = sdp->sd_log_num_databuf;
517         unsigned int total_jdata = sdp->sd_log_num_jdata;
518         unsigned int num, n;
519         __be64 *ptr = NULL;
520
521         offset += 2*sizeof(__be64) - 1;
522         offset &= ~(2*sizeof(__be64) - 1);
523         limit = (sdp->sd_sb.sb_bsize - offset)/sizeof(__be64);
524
525         /*
526          * Start writing ordered buffers, write journaled buffers
527          * into the log along with a header
528          */
529         gfs2_log_lock(sdp);
530         bd2 = bd1 = list_prepare_entry(bd1, &sdp->sd_log_le_databuf,
531                                        bd_le.le_list);
532         while(total_dbuf) {
533                 num = total_jdata;
534                 if (num > limit)
535                         num = limit;
536                 n = 0;
537                 list_for_each_entry_safe_continue(bd1, bdt,
538                                                   &sdp->sd_log_le_databuf,
539                                                   bd_le.le_list) {
540                         /* An ordered write buffer */
541                         if (bd1->bd_bh && !buffer_pinned(bd1->bd_bh)) {
542                                 list_move(&bd1->bd_le.le_list, &started);
543                                 if (bd1 == bd2) {
544                                         bd2 = NULL;
545                                         bd2 = list_prepare_entry(bd2,
546                                                         &sdp->sd_log_le_databuf,
547                                                         bd_le.le_list);
548                                 }
549                                 total_dbuf--;
550                                 if (bd1->bd_bh) {
551                                         get_bh(bd1->bd_bh);
552                                         if (buffer_dirty(bd1->bd_bh)) {
553                                                 gfs2_log_unlock(sdp);
554                                                 wait_on_buffer(bd1->bd_bh);
555                                                 ll_rw_block(WRITE, 1,
556                                                             &bd1->bd_bh);
557                                                 gfs2_log_lock(sdp);
558                                         }
559                                         brelse(bd1->bd_bh);
560                                         continue;
561                                 }
562                                 continue;
563                         } else if (bd1->bd_bh) { /* A journaled buffer */
564                                 int magic;
565                                 gfs2_log_unlock(sdp);
566                                 if (!bh) {
567                                         bh = gfs2_log_get_buf(sdp);
568                                         sdp->sd_log_num_hdrs++;
569                                         ld = (struct gfs2_log_descriptor *)
570                                              bh->b_data;
571                                         ptr = (__be64 *)(bh->b_data + offset);
572                                         ld->ld_header.mh_magic =
573                                                 cpu_to_be32(GFS2_MAGIC);
574                                         ld->ld_header.mh_type =
575                                                 cpu_to_be32(GFS2_METATYPE_LD);
576                                         ld->ld_header.mh_format =
577                                                 cpu_to_be32(GFS2_FORMAT_LD);
578                                         ld->ld_type =
579                                                 cpu_to_be32(GFS2_LOG_DESC_JDATA);
580                                         ld->ld_length = cpu_to_be32(num + 1);
581                                         ld->ld_data1 = cpu_to_be32(num);
582                                         ld->ld_data2 = cpu_to_be32(0);
583                                         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
584                                 }
585                                 magic = gfs2_check_magic(bd1->bd_bh);
586                                 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
587                                 *ptr++ = cpu_to_be64((__u64)magic);
588                                 clear_buffer_escaped(bd1->bd_bh);
589                                 if (unlikely(magic != 0))
590                                         set_buffer_escaped(bd1->bd_bh);
591                                 gfs2_log_lock(sdp);
592                                 if (n++ > num)
593                                         break;
594                         } else if (!bd1->bd_bh) {
595                                 total_dbuf--;
596                                 sdp->sd_log_num_databuf--;
597                                 list_del_init(&bd1->bd_le.le_list);
598                                 if (bd1 == bd2) {
599                                         bd2 = NULL;
600                                         bd2 = list_prepare_entry(bd2,
601                                                 &sdp->sd_log_le_databuf,
602                                                 bd_le.le_list);
603                                 }
604                                 kmem_cache_free(gfs2_bufdata_cachep, bd1);
605                         }
606                 }
607                 gfs2_log_unlock(sdp);
608                 if (bh) {
609                         set_buffer_dirty(bh);
610                         ll_rw_block(WRITE, 1, &bh);
611                         bh = NULL;
612                 }
613                 n = 0;
614                 gfs2_log_lock(sdp);
615                 list_for_each_entry_continue(bd2, &sdp->sd_log_le_databuf,
616                                              bd_le.le_list) {
617                         if (!bd2->bd_bh)
618                                 continue;
619                         /* copy buffer if it needs escaping */
620                         gfs2_log_unlock(sdp);
621                         if (unlikely(buffer_escaped(bd2->bd_bh))) {
622                                 void *kaddr;
623                                 struct page *page = bd2->bd_bh->b_page;
624                                 bh = gfs2_log_get_buf(sdp);
625                                 kaddr = kmap_atomic(page, KM_USER0);
626                                 memcpy(bh->b_data,
627                                        kaddr + bh_offset(bd2->bd_bh),
628                                        sdp->sd_sb.sb_bsize);
629                                 kunmap_atomic(page, KM_USER0);
630                                 *(__be32 *)bh->b_data = 0;
631                         } else {
632                                 bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
633                         }
634                         set_buffer_dirty(bh);
635                         ll_rw_block(WRITE, 1, &bh);
636                         gfs2_log_lock(sdp);
637                         if (++n >= num)
638                                 break;
639                 }
640                 bh = NULL;
641                 total_dbuf -= num;
642                 total_jdata -= num;
643         }
644         gfs2_log_unlock(sdp);
645
646         /* Wait on all ordered buffers */
647         while (!list_empty(&started)) {
648                 gfs2_log_lock(sdp);
649                 bd1 = list_entry(started.next, struct gfs2_bufdata,
650                                  bd_le.le_list);
651                 list_del_init(&bd1->bd_le.le_list);
652                 sdp->sd_log_num_databuf--;
653                 bh = bd1->bd_bh;
654                 if (bh) {
655                         bh->b_private = NULL;
656                         get_bh(bh);
657                         gfs2_log_unlock(sdp);
658                         wait_on_buffer(bh);
659                         brelse(bh);
660                 } else
661                         gfs2_log_unlock(sdp);
662
663                 kmem_cache_free(gfs2_bufdata_cachep, bd1);
664         }
665
666         /* We've removed all the ordered write bufs here, so only jdata left */
667         gfs2_assert_warn(sdp, sdp->sd_log_num_databuf == sdp->sd_log_num_jdata);
668 }
669
670 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
671                                     struct gfs2_log_descriptor *ld,
672                                     __be64 *ptr, int pass)
673 {
674         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
675         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
676         struct gfs2_glock *gl = ip->i_gl;
677         unsigned int blks = be32_to_cpu(ld->ld_data1);
678         struct buffer_head *bh_log, *bh_ip;
679         u64 blkno;
680         u64 esc;
681         int error = 0;
682
683         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
684                 return 0;
685
686         gfs2_replay_incr_blk(sdp, &start);
687         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
688                 blkno = be64_to_cpu(*ptr++);
689                 esc = be64_to_cpu(*ptr++);
690
691                 sdp->sd_found_blocks++;
692
693                 if (gfs2_revoke_check(sdp, blkno, start))
694                         continue;
695
696                 error = gfs2_replay_read_block(jd, start, &bh_log);
697                 if (error)
698                         return error;
699
700                 bh_ip = gfs2_meta_new(gl, blkno);
701                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
702
703                 /* Unescape */
704                 if (esc) {
705                         __be32 *eptr = (__be32 *)bh_ip->b_data;
706                         *eptr = cpu_to_be32(GFS2_MAGIC);
707                 }
708                 mark_buffer_dirty(bh_ip);
709
710                 brelse(bh_log);
711                 brelse(bh_ip);
712                 if (error)
713                         break;
714
715                 sdp->sd_replayed_blocks++;
716         }
717
718         return error;
719 }
720
721 /* FIXME: sort out accounting for log blocks etc. */
722
723 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
724 {
725         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
726         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
727
728         if (error) {
729                 gfs2_meta_sync(ip->i_gl);
730                 return;
731         }
732         if (pass != 1)
733                 return;
734
735         /* data sync? */
736         gfs2_meta_sync(ip->i_gl);
737
738         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
739                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
740 }
741
742 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
743 {
744         struct list_head *head = &sdp->sd_log_le_databuf;
745         struct gfs2_bufdata *bd;
746
747         while (!list_empty(head)) {
748                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
749                 list_del_init(&bd->bd_le.le_list);
750                 sdp->sd_log_num_databuf--;
751                 sdp->sd_log_num_jdata--;
752                 gfs2_unpin(sdp, bd->bd_bh, ai);
753         }
754         gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
755         gfs2_assert_warn(sdp, !sdp->sd_log_num_jdata);
756 }
757
758
759 const struct gfs2_log_operations gfs2_glock_lops = {
760         .lo_add = glock_lo_add,
761         .lo_after_commit = glock_lo_after_commit,
762         .lo_name = "glock",
763 };
764
765 const struct gfs2_log_operations gfs2_buf_lops = {
766         .lo_add = buf_lo_add,
767         .lo_incore_commit = buf_lo_incore_commit,
768         .lo_before_commit = buf_lo_before_commit,
769         .lo_after_commit = buf_lo_after_commit,
770         .lo_before_scan = buf_lo_before_scan,
771         .lo_scan_elements = buf_lo_scan_elements,
772         .lo_after_scan = buf_lo_after_scan,
773         .lo_name = "buf",
774 };
775
776 const struct gfs2_log_operations gfs2_revoke_lops = {
777         .lo_add = revoke_lo_add,
778         .lo_before_commit = revoke_lo_before_commit,
779         .lo_before_scan = revoke_lo_before_scan,
780         .lo_scan_elements = revoke_lo_scan_elements,
781         .lo_after_scan = revoke_lo_after_scan,
782         .lo_name = "revoke",
783 };
784
785 const struct gfs2_log_operations gfs2_rg_lops = {
786         .lo_add = rg_lo_add,
787         .lo_after_commit = rg_lo_after_commit,
788         .lo_name = "rg",
789 };
790
791 const struct gfs2_log_operations gfs2_databuf_lops = {
792         .lo_add = databuf_lo_add,
793         .lo_incore_commit = buf_lo_incore_commit,
794         .lo_before_commit = databuf_lo_before_commit,
795         .lo_after_commit = databuf_lo_after_commit,
796         .lo_scan_elements = databuf_lo_scan_elements,
797         .lo_after_scan = databuf_lo_after_scan,
798         .lo_name = "databuf",
799 };
800
801 const struct gfs2_log_operations *gfs2_log_ops[] = {
802         &gfs2_glock_lops,
803         &gfs2_buf_lops,
804         &gfs2_revoke_lops,
805         &gfs2_rg_lops,
806         &gfs2_databuf_lops,
807         NULL,
808 };
809