Loading...
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10#include <linux/sched.h>
11#include <linux/slab.h>
12#include <linux/spinlock.h>
13#include <linux/completion.h>
14#include <linux/buffer_head.h>
15#include <linux/gfs2_ondisk.h>
16#include <linux/bio.h>
17#include <linux/fs.h>
18
19#include "gfs2.h"
20#include "incore.h"
21#include "inode.h"
22#include "glock.h"
23#include "log.h"
24#include "lops.h"
25#include "meta_io.h"
26#include "recovery.h"
27#include "rgrp.h"
28#include "trans.h"
29#include "util.h"
30#include "trace_gfs2.h"
31
32/**
33 * gfs2_pin - Pin a buffer in memory
34 * @sdp: The superblock
35 * @bh: The buffer to be pinned
36 *
37 * The log lock must be held when calling this function
38 */
39static void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
40{
41 struct gfs2_bufdata *bd;
42
43 BUG_ON(!current->journal_info);
44
45 clear_buffer_dirty(bh);
46 if (test_set_buffer_pinned(bh))
47 gfs2_assert_withdraw(sdp, 0);
48 if (!buffer_uptodate(bh))
49 gfs2_io_error_bh(sdp, bh);
50 bd = bh->b_private;
51 /* If this buffer is in the AIL and it has already been written
52 * to in-place disk block, remove it from the AIL.
53 */
54 spin_lock(&sdp->sd_ail_lock);
55 if (bd->bd_ail)
56 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
57 spin_unlock(&sdp->sd_ail_lock);
58 get_bh(bh);
59 atomic_inc(&sdp->sd_log_pinned);
60 trace_gfs2_pin(bd, 1);
61}
62
63/**
64 * gfs2_unpin - Unpin a buffer
65 * @sdp: the filesystem the buffer belongs to
66 * @bh: The buffer to unpin
67 * @ai:
68 * @flags: The inode dirty flags
69 *
70 */
71
72static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
73 struct gfs2_ail *ai)
74{
75 struct gfs2_bufdata *bd = bh->b_private;
76
77 BUG_ON(!buffer_uptodate(bh));
78 BUG_ON(!buffer_pinned(bh));
79
80 lock_buffer(bh);
81 mark_buffer_dirty(bh);
82 clear_buffer_pinned(bh);
83
84 spin_lock(&sdp->sd_ail_lock);
85 if (bd->bd_ail) {
86 list_del(&bd->bd_ail_st_list);
87 brelse(bh);
88 } else {
89 struct gfs2_glock *gl = bd->bd_gl;
90 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
91 atomic_inc(&gl->gl_ail_count);
92 }
93 bd->bd_ail = ai;
94 list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
95 spin_unlock(&sdp->sd_ail_lock);
96
97 clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
98 trace_gfs2_pin(bd, 0);
99 unlock_buffer(bh);
100 atomic_dec(&sdp->sd_log_pinned);
101}
102
103
104static inline struct gfs2_log_descriptor *bh_log_desc(struct buffer_head *bh)
105{
106 return (struct gfs2_log_descriptor *)bh->b_data;
107}
108
109static inline __be64 *bh_log_ptr(struct buffer_head *bh)
110{
111 struct gfs2_log_descriptor *ld = bh_log_desc(bh);
112 return (__force __be64 *)(ld + 1);
113}
114
115static inline __be64 *bh_ptr_end(struct buffer_head *bh)
116{
117 return (__force __be64 *)(bh->b_data + bh->b_size);
118}
119
120
121static struct buffer_head *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type)
122{
123 struct buffer_head *bh = gfs2_log_get_buf(sdp);
124 struct gfs2_log_descriptor *ld = bh_log_desc(bh);
125 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
126 ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
127 ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
128 ld->ld_type = cpu_to_be32(ld_type);
129 ld->ld_length = 0;
130 ld->ld_data1 = 0;
131 ld->ld_data2 = 0;
132 memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
133 return bh;
134}
135
136static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
137{
138 struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
139 struct gfs2_meta_header *mh;
140 struct gfs2_trans *tr;
141
142 lock_buffer(bd->bd_bh);
143 gfs2_log_lock(sdp);
144 if (!list_empty(&bd->bd_list_tr))
145 goto out;
146 tr = current->journal_info;
147 tr->tr_touched = 1;
148 tr->tr_num_buf++;
149 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
150 if (!list_empty(&le->le_list))
151 goto out;
152 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
153 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
154 gfs2_meta_check(sdp, bd->bd_bh);
155 gfs2_pin(sdp, bd->bd_bh);
156 mh = (struct gfs2_meta_header *)bd->bd_bh->b_data;
157 mh->__pad0 = cpu_to_be64(0);
158 mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
159 sdp->sd_log_num_buf++;
160 list_add(&le->le_list, &sdp->sd_log_le_buf);
161 tr->tr_num_buf_new++;
162out:
163 gfs2_log_unlock(sdp);
164 unlock_buffer(bd->bd_bh);
165}
166
167static void buf_lo_before_commit(struct gfs2_sbd *sdp)
168{
169 struct buffer_head *bh;
170 struct gfs2_log_descriptor *ld;
171 struct gfs2_bufdata *bd1 = NULL, *bd2;
172 unsigned int total;
173 unsigned int limit;
174 unsigned int num;
175 unsigned n;
176 __be64 *ptr;
177
178 limit = buf_limit(sdp);
179 /* for 4k blocks, limit = 503 */
180
181 gfs2_log_lock(sdp);
182 total = sdp->sd_log_num_buf;
183 bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
184 while(total) {
185 num = total;
186 if (total > limit)
187 num = limit;
188 gfs2_log_unlock(sdp);
189 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA);
190 gfs2_log_lock(sdp);
191 ld = bh_log_desc(bh);
192 ptr = bh_log_ptr(bh);
193 ld->ld_length = cpu_to_be32(num + 1);
194 ld->ld_data1 = cpu_to_be32(num);
195
196 n = 0;
197 list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
198 bd_le.le_list) {
199 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
200 if (++n >= num)
201 break;
202 }
203
204 gfs2_log_unlock(sdp);
205 submit_bh(WRITE_SYNC, bh);
206 gfs2_log_lock(sdp);
207
208 n = 0;
209 list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
210 bd_le.le_list) {
211 get_bh(bd2->bd_bh);
212 gfs2_log_unlock(sdp);
213 lock_buffer(bd2->bd_bh);
214 bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
215 submit_bh(WRITE_SYNC, bh);
216 gfs2_log_lock(sdp);
217 if (++n >= num)
218 break;
219 }
220
221 BUG_ON(total < num);
222 total -= num;
223 }
224 gfs2_log_unlock(sdp);
225}
226
227static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
228{
229 struct list_head *head = &sdp->sd_log_le_buf;
230 struct gfs2_bufdata *bd;
231
232 while (!list_empty(head)) {
233 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
234 list_del_init(&bd->bd_le.le_list);
235 sdp->sd_log_num_buf--;
236
237 gfs2_unpin(sdp, bd->bd_bh, ai);
238 }
239 gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
240}
241
242static void buf_lo_before_scan(struct gfs2_jdesc *jd,
243 struct gfs2_log_header_host *head, int pass)
244{
245 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
246
247 if (pass != 0)
248 return;
249
250 sdp->sd_found_blocks = 0;
251 sdp->sd_replayed_blocks = 0;
252}
253
254static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
255 struct gfs2_log_descriptor *ld, __be64 *ptr,
256 int pass)
257{
258 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
259 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
260 struct gfs2_glock *gl = ip->i_gl;
261 unsigned int blks = be32_to_cpu(ld->ld_data1);
262 struct buffer_head *bh_log, *bh_ip;
263 u64 blkno;
264 int error = 0;
265
266 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
267 return 0;
268
269 gfs2_replay_incr_blk(sdp, &start);
270
271 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
272 blkno = be64_to_cpu(*ptr++);
273
274 sdp->sd_found_blocks++;
275
276 if (gfs2_revoke_check(sdp, blkno, start))
277 continue;
278
279 error = gfs2_replay_read_block(jd, start, &bh_log);
280 if (error)
281 return error;
282
283 bh_ip = gfs2_meta_new(gl, blkno);
284 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
285
286 if (gfs2_meta_check(sdp, bh_ip))
287 error = -EIO;
288 else
289 mark_buffer_dirty(bh_ip);
290
291 brelse(bh_log);
292 brelse(bh_ip);
293
294 if (error)
295 break;
296
297 sdp->sd_replayed_blocks++;
298 }
299
300 return error;
301}
302
303static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
304{
305 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
306 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
307
308 if (error) {
309 gfs2_meta_sync(ip->i_gl);
310 return;
311 }
312 if (pass != 1)
313 return;
314
315 gfs2_meta_sync(ip->i_gl);
316
317 fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
318 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
319}
320
321static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
322{
323 struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
324 struct gfs2_glock *gl = bd->bd_gl;
325 struct gfs2_trans *tr;
326
327 tr = current->journal_info;
328 tr->tr_touched = 1;
329 tr->tr_num_revoke++;
330 sdp->sd_log_num_revoke++;
331 atomic_inc(&gl->gl_revokes);
332 set_bit(GLF_LFLUSH, &gl->gl_flags);
333 list_add(&le->le_list, &sdp->sd_log_le_revoke);
334}
335
336static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
337{
338 struct gfs2_log_descriptor *ld;
339 struct gfs2_meta_header *mh;
340 struct buffer_head *bh;
341 unsigned int offset;
342 struct list_head *head = &sdp->sd_log_le_revoke;
343 struct gfs2_bufdata *bd;
344
345 if (!sdp->sd_log_num_revoke)
346 return;
347
348 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE);
349 ld = bh_log_desc(bh);
350 ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
351 sizeof(u64)));
352 ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
353 offset = sizeof(struct gfs2_log_descriptor);
354
355 list_for_each_entry(bd, head, bd_le.le_list) {
356 sdp->sd_log_num_revoke--;
357
358 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
359 submit_bh(WRITE_SYNC, bh);
360
361 bh = gfs2_log_get_buf(sdp);
362 mh = (struct gfs2_meta_header *)bh->b_data;
363 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
364 mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
365 mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
366 offset = sizeof(struct gfs2_meta_header);
367 }
368
369 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(bd->bd_blkno);
370 offset += sizeof(u64);
371 }
372 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
373
374 submit_bh(WRITE_SYNC, bh);
375}
376
377static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
378{
379 struct list_head *head = &sdp->sd_log_le_revoke;
380 struct gfs2_bufdata *bd;
381 struct gfs2_glock *gl;
382
383 while (!list_empty(head)) {
384 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
385 list_del_init(&bd->bd_le.le_list);
386 gl = bd->bd_gl;
387 atomic_dec(&gl->gl_revokes);
388 clear_bit(GLF_LFLUSH, &gl->gl_flags);
389 kmem_cache_free(gfs2_bufdata_cachep, bd);
390 }
391}
392
393static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
394 struct gfs2_log_header_host *head, int pass)
395{
396 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
397
398 if (pass != 0)
399 return;
400
401 sdp->sd_found_revokes = 0;
402 sdp->sd_replay_tail = head->lh_tail;
403}
404
405static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
406 struct gfs2_log_descriptor *ld, __be64 *ptr,
407 int pass)
408{
409 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
410 unsigned int blks = be32_to_cpu(ld->ld_length);
411 unsigned int revokes = be32_to_cpu(ld->ld_data1);
412 struct buffer_head *bh;
413 unsigned int offset;
414 u64 blkno;
415 int first = 1;
416 int error;
417
418 if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
419 return 0;
420
421 offset = sizeof(struct gfs2_log_descriptor);
422
423 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
424 error = gfs2_replay_read_block(jd, start, &bh);
425 if (error)
426 return error;
427
428 if (!first)
429 gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
430
431 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
432 blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
433
434 error = gfs2_revoke_add(sdp, blkno, start);
435 if (error < 0) {
436 brelse(bh);
437 return error;
438 }
439 else if (error)
440 sdp->sd_found_revokes++;
441
442 if (!--revokes)
443 break;
444 offset += sizeof(u64);
445 }
446
447 brelse(bh);
448 offset = sizeof(struct gfs2_meta_header);
449 first = 0;
450 }
451
452 return 0;
453}
454
455static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
456{
457 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
458
459 if (error) {
460 gfs2_revoke_clean(sdp);
461 return;
462 }
463 if (pass != 1)
464 return;
465
466 fs_info(sdp, "jid=%u: Found %u revoke tags\n",
467 jd->jd_jid, sdp->sd_found_revokes);
468
469 gfs2_revoke_clean(sdp);
470}
471
472static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
473{
474 struct gfs2_rgrpd *rgd;
475 struct gfs2_trans *tr = current->journal_info;
476
477 tr->tr_touched = 1;
478
479 rgd = container_of(le, struct gfs2_rgrpd, rd_le);
480
481 gfs2_log_lock(sdp);
482 if (!list_empty(&le->le_list)){
483 gfs2_log_unlock(sdp);
484 return;
485 }
486 gfs2_rgrp_bh_hold(rgd);
487 sdp->sd_log_num_rg++;
488 list_add(&le->le_list, &sdp->sd_log_le_rg);
489 gfs2_log_unlock(sdp);
490}
491
492static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
493{
494 struct list_head *head = &sdp->sd_log_le_rg;
495 struct gfs2_rgrpd *rgd;
496
497 while (!list_empty(head)) {
498 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
499 list_del_init(&rgd->rd_le.le_list);
500 sdp->sd_log_num_rg--;
501
502 gfs2_rgrp_repolish_clones(rgd);
503 gfs2_rgrp_bh_put(rgd);
504 }
505 gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
506}
507
508/**
509 * databuf_lo_add - Add a databuf to the transaction.
510 *
511 * This is used in two distinct cases:
512 * i) In ordered write mode
513 * We put the data buffer on a list so that we can ensure that its
514 * synced to disk at the right time
515 * ii) In journaled data mode
516 * We need to journal the data block in the same way as metadata in
517 * the functions above. The difference is that here we have a tag
518 * which is two __be64's being the block number (as per meta data)
519 * and a flag which says whether the data block needs escaping or
520 * not. This means we need a new log entry for each 251 or so data
521 * blocks, which isn't an enormous overhead but twice as much as
522 * for normal metadata blocks.
523 */
524static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
525{
526 struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
527 struct gfs2_trans *tr = current->journal_info;
528 struct address_space *mapping = bd->bd_bh->b_page->mapping;
529 struct gfs2_inode *ip = GFS2_I(mapping->host);
530
531 lock_buffer(bd->bd_bh);
532 gfs2_log_lock(sdp);
533 if (tr) {
534 if (!list_empty(&bd->bd_list_tr))
535 goto out;
536 tr->tr_touched = 1;
537 if (gfs2_is_jdata(ip)) {
538 tr->tr_num_buf++;
539 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
540 }
541 }
542 if (!list_empty(&le->le_list))
543 goto out;
544
545 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
546 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
547 if (gfs2_is_jdata(ip)) {
548 gfs2_pin(sdp, bd->bd_bh);
549 tr->tr_num_databuf_new++;
550 sdp->sd_log_num_databuf++;
551 list_add_tail(&le->le_list, &sdp->sd_log_le_databuf);
552 } else {
553 list_add_tail(&le->le_list, &sdp->sd_log_le_ordered);
554 }
555out:
556 gfs2_log_unlock(sdp);
557 unlock_buffer(bd->bd_bh);
558}
559
560static void gfs2_check_magic(struct buffer_head *bh)
561{
562 void *kaddr;
563 __be32 *ptr;
564
565 clear_buffer_escaped(bh);
566 kaddr = kmap_atomic(bh->b_page, KM_USER0);
567 ptr = kaddr + bh_offset(bh);
568 if (*ptr == cpu_to_be32(GFS2_MAGIC))
569 set_buffer_escaped(bh);
570 kunmap_atomic(kaddr, KM_USER0);
571}
572
573static void gfs2_write_blocks(struct gfs2_sbd *sdp, struct buffer_head *bh,
574 struct list_head *list, struct list_head *done,
575 unsigned int n)
576{
577 struct buffer_head *bh1;
578 struct gfs2_log_descriptor *ld;
579 struct gfs2_bufdata *bd;
580 __be64 *ptr;
581
582 if (!bh)
583 return;
584
585 ld = bh_log_desc(bh);
586 ld->ld_length = cpu_to_be32(n + 1);
587 ld->ld_data1 = cpu_to_be32(n);
588
589 ptr = bh_log_ptr(bh);
590
591 get_bh(bh);
592 submit_bh(WRITE_SYNC, bh);
593 gfs2_log_lock(sdp);
594 while(!list_empty(list)) {
595 bd = list_entry(list->next, struct gfs2_bufdata, bd_le.le_list);
596 list_move_tail(&bd->bd_le.le_list, done);
597 get_bh(bd->bd_bh);
598 while (be64_to_cpu(*ptr) != bd->bd_bh->b_blocknr) {
599 gfs2_log_incr_head(sdp);
600 ptr += 2;
601 }
602 gfs2_log_unlock(sdp);
603 lock_buffer(bd->bd_bh);
604 if (buffer_escaped(bd->bd_bh)) {
605 void *kaddr;
606 bh1 = gfs2_log_get_buf(sdp);
607 kaddr = kmap_atomic(bd->bd_bh->b_page, KM_USER0);
608 memcpy(bh1->b_data, kaddr + bh_offset(bd->bd_bh),
609 bh1->b_size);
610 kunmap_atomic(kaddr, KM_USER0);
611 *(__be32 *)bh1->b_data = 0;
612 clear_buffer_escaped(bd->bd_bh);
613 unlock_buffer(bd->bd_bh);
614 brelse(bd->bd_bh);
615 } else {
616 bh1 = gfs2_log_fake_buf(sdp, bd->bd_bh);
617 }
618 submit_bh(WRITE_SYNC, bh1);
619 gfs2_log_lock(sdp);
620 ptr += 2;
621 }
622 gfs2_log_unlock(sdp);
623 brelse(bh);
624}
625
626/**
627 * databuf_lo_before_commit - Scan the data buffers, writing as we go
628 *
629 */
630
631static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
632{
633 struct gfs2_bufdata *bd = NULL;
634 struct buffer_head *bh = NULL;
635 unsigned int n = 0;
636 __be64 *ptr = NULL, *end = NULL;
637 LIST_HEAD(processed);
638 LIST_HEAD(in_progress);
639
640 gfs2_log_lock(sdp);
641 while (!list_empty(&sdp->sd_log_le_databuf)) {
642 if (ptr == end) {
643 gfs2_log_unlock(sdp);
644 gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
645 n = 0;
646 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_JDATA);
647 ptr = bh_log_ptr(bh);
648 end = bh_ptr_end(bh) - 1;
649 gfs2_log_lock(sdp);
650 continue;
651 }
652 bd = list_entry(sdp->sd_log_le_databuf.next, struct gfs2_bufdata, bd_le.le_list);
653 list_move_tail(&bd->bd_le.le_list, &in_progress);
654 gfs2_check_magic(bd->bd_bh);
655 *ptr++ = cpu_to_be64(bd->bd_bh->b_blocknr);
656 *ptr++ = cpu_to_be64(buffer_escaped(bh) ? 1 : 0);
657 n++;
658 }
659 gfs2_log_unlock(sdp);
660 gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
661 gfs2_log_lock(sdp);
662 list_splice(&processed, &sdp->sd_log_le_databuf);
663 gfs2_log_unlock(sdp);
664}
665
666static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
667 struct gfs2_log_descriptor *ld,
668 __be64 *ptr, int pass)
669{
670 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
671 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
672 struct gfs2_glock *gl = ip->i_gl;
673 unsigned int blks = be32_to_cpu(ld->ld_data1);
674 struct buffer_head *bh_log, *bh_ip;
675 u64 blkno;
676 u64 esc;
677 int error = 0;
678
679 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
680 return 0;
681
682 gfs2_replay_incr_blk(sdp, &start);
683 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
684 blkno = be64_to_cpu(*ptr++);
685 esc = be64_to_cpu(*ptr++);
686
687 sdp->sd_found_blocks++;
688
689 if (gfs2_revoke_check(sdp, blkno, start))
690 continue;
691
692 error = gfs2_replay_read_block(jd, start, &bh_log);
693 if (error)
694 return error;
695
696 bh_ip = gfs2_meta_new(gl, blkno);
697 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
698
699 /* Unescape */
700 if (esc) {
701 __be32 *eptr = (__be32 *)bh_ip->b_data;
702 *eptr = cpu_to_be32(GFS2_MAGIC);
703 }
704 mark_buffer_dirty(bh_ip);
705
706 brelse(bh_log);
707 brelse(bh_ip);
708 if (error)
709 break;
710
711 sdp->sd_replayed_blocks++;
712 }
713
714 return error;
715}
716
717/* FIXME: sort out accounting for log blocks etc. */
718
719static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
720{
721 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
722 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
723
724 if (error) {
725 gfs2_meta_sync(ip->i_gl);
726 return;
727 }
728 if (pass != 1)
729 return;
730
731 /* data sync? */
732 gfs2_meta_sync(ip->i_gl);
733
734 fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
735 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
736}
737
738static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
739{
740 struct list_head *head = &sdp->sd_log_le_databuf;
741 struct gfs2_bufdata *bd;
742
743 while (!list_empty(head)) {
744 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
745 list_del_init(&bd->bd_le.le_list);
746 sdp->sd_log_num_databuf--;
747 gfs2_unpin(sdp, bd->bd_bh, ai);
748 }
749 gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
750}
751
752
753const struct gfs2_log_operations gfs2_buf_lops = {
754 .lo_add = buf_lo_add,
755 .lo_before_commit = buf_lo_before_commit,
756 .lo_after_commit = buf_lo_after_commit,
757 .lo_before_scan = buf_lo_before_scan,
758 .lo_scan_elements = buf_lo_scan_elements,
759 .lo_after_scan = buf_lo_after_scan,
760 .lo_name = "buf",
761};
762
763const struct gfs2_log_operations gfs2_revoke_lops = {
764 .lo_add = revoke_lo_add,
765 .lo_before_commit = revoke_lo_before_commit,
766 .lo_after_commit = revoke_lo_after_commit,
767 .lo_before_scan = revoke_lo_before_scan,
768 .lo_scan_elements = revoke_lo_scan_elements,
769 .lo_after_scan = revoke_lo_after_scan,
770 .lo_name = "revoke",
771};
772
773const struct gfs2_log_operations gfs2_rg_lops = {
774 .lo_add = rg_lo_add,
775 .lo_after_commit = rg_lo_after_commit,
776 .lo_name = "rg",
777};
778
779const struct gfs2_log_operations gfs2_databuf_lops = {
780 .lo_add = databuf_lo_add,
781 .lo_before_commit = databuf_lo_before_commit,
782 .lo_after_commit = databuf_lo_after_commit,
783 .lo_scan_elements = databuf_lo_scan_elements,
784 .lo_after_scan = databuf_lo_after_scan,
785 .lo_name = "databuf",
786};
787
788const struct gfs2_log_operations *gfs2_log_ops[] = {
789 &gfs2_databuf_lops,
790 &gfs2_buf_lops,
791 &gfs2_rg_lops,
792 &gfs2_revoke_lops,
793 NULL,
794};
795
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/mempool.h>
16#include <linux/gfs2_ondisk.h>
17#include <linux/bio.h>
18#include <linux/fs.h>
19#include <linux/list_sort.h>
20
21#include "gfs2.h"
22#include "incore.h"
23#include "inode.h"
24#include "glock.h"
25#include "log.h"
26#include "lops.h"
27#include "meta_io.h"
28#include "recovery.h"
29#include "rgrp.h"
30#include "trans.h"
31#include "util.h"
32#include "trace_gfs2.h"
33
34/**
35 * gfs2_pin - Pin a buffer in memory
36 * @sdp: The superblock
37 * @bh: The buffer to be pinned
38 *
39 * The log lock must be held when calling this function
40 */
41void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
42{
43 struct gfs2_bufdata *bd;
44
45 BUG_ON(!current->journal_info);
46
47 clear_buffer_dirty(bh);
48 if (test_set_buffer_pinned(bh))
49 gfs2_assert_withdraw(sdp, 0);
50 if (!buffer_uptodate(bh))
51 gfs2_io_error_bh(sdp, bh);
52 bd = bh->b_private;
53 /* If this buffer is in the AIL and it has already been written
54 * to in-place disk block, remove it from the AIL.
55 */
56 spin_lock(&sdp->sd_ail_lock);
57 if (bd->bd_tr)
58 list_move(&bd->bd_ail_st_list, &bd->bd_tr->tr_ail2_list);
59 spin_unlock(&sdp->sd_ail_lock);
60 get_bh(bh);
61 atomic_inc(&sdp->sd_log_pinned);
62 trace_gfs2_pin(bd, 1);
63}
64
65static bool buffer_is_rgrp(const struct gfs2_bufdata *bd)
66{
67 return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP;
68}
69
70static void maybe_release_space(struct gfs2_bufdata *bd)
71{
72 struct gfs2_glock *gl = bd->bd_gl;
73 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
74 struct gfs2_rgrpd *rgd = gl->gl_object;
75 unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number;
76 struct gfs2_bitmap *bi = rgd->rd_bits + index;
77
78 if (bi->bi_clone == NULL)
79 return;
80 if (sdp->sd_args.ar_discard)
81 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
82 memcpy(bi->bi_clone + bi->bi_offset,
83 bd->bd_bh->b_data + bi->bi_offset, bi->bi_len);
84 clear_bit(GBF_FULL, &bi->bi_flags);
85 rgd->rd_free_clone = rgd->rd_free;
86 rgd->rd_extfail_pt = rgd->rd_free;
87}
88
89/**
90 * gfs2_unpin - Unpin a buffer
91 * @sdp: the filesystem the buffer belongs to
92 * @bh: The buffer to unpin
93 * @ai:
94 * @flags: The inode dirty flags
95 *
96 */
97
98static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
99 struct gfs2_trans *tr)
100{
101 struct gfs2_bufdata *bd = bh->b_private;
102
103 BUG_ON(!buffer_uptodate(bh));
104 BUG_ON(!buffer_pinned(bh));
105
106 lock_buffer(bh);
107 mark_buffer_dirty(bh);
108 clear_buffer_pinned(bh);
109
110 if (buffer_is_rgrp(bd))
111 maybe_release_space(bd);
112
113 spin_lock(&sdp->sd_ail_lock);
114 if (bd->bd_tr) {
115 list_del(&bd->bd_ail_st_list);
116 brelse(bh);
117 } else {
118 struct gfs2_glock *gl = bd->bd_gl;
119 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
120 atomic_inc(&gl->gl_ail_count);
121 }
122 bd->bd_tr = tr;
123 list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list);
124 spin_unlock(&sdp->sd_ail_lock);
125
126 clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
127 trace_gfs2_pin(bd, 0);
128 unlock_buffer(bh);
129 atomic_dec(&sdp->sd_log_pinned);
130}
131
132static void gfs2_log_incr_head(struct gfs2_sbd *sdp)
133{
134 BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
135 (sdp->sd_log_flush_head != sdp->sd_log_head));
136
137 if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
138 sdp->sd_log_flush_head = 0;
139 sdp->sd_log_flush_wrapped = 1;
140 }
141}
142
143static u64 gfs2_log_bmap(struct gfs2_sbd *sdp)
144{
145 unsigned int lbn = sdp->sd_log_flush_head;
146 struct gfs2_journal_extent *je;
147 u64 block;
148
149 list_for_each_entry(je, &sdp->sd_jdesc->extent_list, list) {
150 if ((lbn >= je->lblock) && (lbn < (je->lblock + je->blocks))) {
151 block = je->dblock + lbn - je->lblock;
152 gfs2_log_incr_head(sdp);
153 return block;
154 }
155 }
156
157 return -1;
158}
159
160/**
161 * gfs2_end_log_write_bh - end log write of pagecache data with buffers
162 * @sdp: The superblock
163 * @bvec: The bio_vec
164 * @error: The i/o status
165 *
166 * This finds the relavent buffers and unlocks then and sets the
167 * error flag according to the status of the i/o request. This is
168 * used when the log is writing data which has an in-place version
169 * that is pinned in the pagecache.
170 */
171
172static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp, struct bio_vec *bvec,
173 int error)
174{
175 struct buffer_head *bh, *next;
176 struct page *page = bvec->bv_page;
177 unsigned size;
178
179 bh = page_buffers(page);
180 size = bvec->bv_len;
181 while (bh_offset(bh) < bvec->bv_offset)
182 bh = bh->b_this_page;
183 do {
184 if (error)
185 set_buffer_write_io_error(bh);
186 unlock_buffer(bh);
187 next = bh->b_this_page;
188 size -= bh->b_size;
189 brelse(bh);
190 bh = next;
191 } while(bh && size);
192}
193
194/**
195 * gfs2_end_log_write - end of i/o to the log
196 * @bio: The bio
197 * @error: Status of i/o request
198 *
199 * Each bio_vec contains either data from the pagecache or data
200 * relating to the log itself. Here we iterate over the bio_vec
201 * array, processing both kinds of data.
202 *
203 */
204
205static void gfs2_end_log_write(struct bio *bio)
206{
207 struct gfs2_sbd *sdp = bio->bi_private;
208 struct bio_vec *bvec;
209 struct page *page;
210 int i;
211
212 if (bio->bi_error) {
213 sdp->sd_log_error = bio->bi_error;
214 fs_err(sdp, "Error %d writing to log\n", bio->bi_error);
215 }
216
217 bio_for_each_segment_all(bvec, bio, i) {
218 page = bvec->bv_page;
219 if (page_has_buffers(page))
220 gfs2_end_log_write_bh(sdp, bvec, bio->bi_error);
221 else
222 mempool_free(page, gfs2_page_pool);
223 }
224
225 bio_put(bio);
226 if (atomic_dec_and_test(&sdp->sd_log_in_flight))
227 wake_up(&sdp->sd_log_flush_wait);
228}
229
230/**
231 * gfs2_log_flush_bio - Submit any pending log bio
232 * @sdp: The superblock
233 * @op: REQ_OP
234 * @op_flags: req_flag_bits
235 *
236 * Submit any pending part-built or full bio to the block device. If
237 * there is no pending bio, then this is a no-op.
238 */
239
240void gfs2_log_flush_bio(struct gfs2_sbd *sdp, int op, int op_flags)
241{
242 if (sdp->sd_log_bio) {
243 atomic_inc(&sdp->sd_log_in_flight);
244 bio_set_op_attrs(sdp->sd_log_bio, op, op_flags);
245 submit_bio(sdp->sd_log_bio);
246 sdp->sd_log_bio = NULL;
247 }
248}
249
250/**
251 * gfs2_log_alloc_bio - Allocate a new bio for log writing
252 * @sdp: The superblock
253 * @blkno: The next device block number we want to write to
254 *
255 * This should never be called when there is a cached bio in the
256 * super block. When it returns, there will be a cached bio in the
257 * super block which will have as many bio_vecs as the device is
258 * happy to handle.
259 *
260 * Returns: Newly allocated bio
261 */
262
263static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno)
264{
265 struct super_block *sb = sdp->sd_vfs;
266 struct bio *bio;
267
268 BUG_ON(sdp->sd_log_bio);
269
270 bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
271 bio->bi_iter.bi_sector = blkno * (sb->s_blocksize >> 9);
272 bio->bi_bdev = sb->s_bdev;
273 bio->bi_end_io = gfs2_end_log_write;
274 bio->bi_private = sdp;
275
276 sdp->sd_log_bio = bio;
277
278 return bio;
279}
280
281/**
282 * gfs2_log_get_bio - Get cached log bio, or allocate a new one
283 * @sdp: The superblock
284 * @blkno: The device block number we want to write to
285 *
286 * If there is a cached bio, then if the next block number is sequential
287 * with the previous one, return it, otherwise flush the bio to the
288 * device. If there is not a cached bio, or we just flushed it, then
289 * allocate a new one.
290 *
291 * Returns: The bio to use for log writes
292 */
293
294static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno)
295{
296 struct bio *bio = sdp->sd_log_bio;
297 u64 nblk;
298
299 if (bio) {
300 nblk = bio_end_sector(bio);
301 nblk >>= sdp->sd_fsb2bb_shift;
302 if (blkno == nblk)
303 return bio;
304 gfs2_log_flush_bio(sdp, REQ_OP_WRITE, 0);
305 }
306
307 return gfs2_log_alloc_bio(sdp, blkno);
308}
309
310
311/**
312 * gfs2_log_write - write to log
313 * @sdp: the filesystem
314 * @page: the page to write
315 * @size: the size of the data to write
316 * @offset: the offset within the page
317 *
318 * Try and add the page segment to the current bio. If that fails,
319 * submit the current bio to the device and create a new one, and
320 * then add the page segment to that.
321 */
322
323static void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page,
324 unsigned size, unsigned offset)
325{
326 u64 blkno = gfs2_log_bmap(sdp);
327 struct bio *bio;
328 int ret;
329
330 bio = gfs2_log_get_bio(sdp, blkno);
331 ret = bio_add_page(bio, page, size, offset);
332 if (ret == 0) {
333 gfs2_log_flush_bio(sdp, REQ_OP_WRITE, 0);
334 bio = gfs2_log_alloc_bio(sdp, blkno);
335 ret = bio_add_page(bio, page, size, offset);
336 WARN_ON(ret == 0);
337 }
338}
339
340/**
341 * gfs2_log_write_bh - write a buffer's content to the log
342 * @sdp: The super block
343 * @bh: The buffer pointing to the in-place location
344 *
345 * This writes the content of the buffer to the next available location
346 * in the log. The buffer will be unlocked once the i/o to the log has
347 * completed.
348 */
349
350static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
351{
352 gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh));
353}
354
355/**
356 * gfs2_log_write_page - write one block stored in a page, into the log
357 * @sdp: The superblock
358 * @page: The struct page
359 *
360 * This writes the first block-sized part of the page into the log. Note
361 * that the page must have been allocated from the gfs2_page_pool mempool
362 * and that after this has been called, ownership has been transferred and
363 * the page may be freed at any time.
364 */
365
366void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
367{
368 struct super_block *sb = sdp->sd_vfs;
369 gfs2_log_write(sdp, page, sb->s_blocksize, 0);
370}
371
372static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
373 u32 ld_length, u32 ld_data1)
374{
375 struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
376 struct gfs2_log_descriptor *ld = page_address(page);
377 clear_page(ld);
378 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
379 ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
380 ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
381 ld->ld_type = cpu_to_be32(ld_type);
382 ld->ld_length = cpu_to_be32(ld_length);
383 ld->ld_data1 = cpu_to_be32(ld_data1);
384 ld->ld_data2 = 0;
385 return page;
386}
387
388static void gfs2_check_magic(struct buffer_head *bh)
389{
390 void *kaddr;
391 __be32 *ptr;
392
393 clear_buffer_escaped(bh);
394 kaddr = kmap_atomic(bh->b_page);
395 ptr = kaddr + bh_offset(bh);
396 if (*ptr == cpu_to_be32(GFS2_MAGIC))
397 set_buffer_escaped(bh);
398 kunmap_atomic(kaddr);
399}
400
401static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b)
402{
403 struct gfs2_bufdata *bda, *bdb;
404
405 bda = list_entry(a, struct gfs2_bufdata, bd_list);
406 bdb = list_entry(b, struct gfs2_bufdata, bd_list);
407
408 if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
409 return -1;
410 if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
411 return 1;
412 return 0;
413}
414
415static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
416 unsigned int total, struct list_head *blist,
417 bool is_databuf)
418{
419 struct gfs2_log_descriptor *ld;
420 struct gfs2_bufdata *bd1 = NULL, *bd2;
421 struct page *page;
422 unsigned int num;
423 unsigned n;
424 __be64 *ptr;
425
426 gfs2_log_lock(sdp);
427 list_sort(NULL, blist, blocknr_cmp);
428 bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
429 while(total) {
430 num = total;
431 if (total > limit)
432 num = limit;
433 gfs2_log_unlock(sdp);
434 page = gfs2_get_log_desc(sdp,
435 is_databuf ? GFS2_LOG_DESC_JDATA :
436 GFS2_LOG_DESC_METADATA, num + 1, num);
437 ld = page_address(page);
438 gfs2_log_lock(sdp);
439 ptr = (__be64 *)(ld + 1);
440
441 n = 0;
442 list_for_each_entry_continue(bd1, blist, bd_list) {
443 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
444 if (is_databuf) {
445 gfs2_check_magic(bd1->bd_bh);
446 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
447 }
448 if (++n >= num)
449 break;
450 }
451
452 gfs2_log_unlock(sdp);
453 gfs2_log_write_page(sdp, page);
454 gfs2_log_lock(sdp);
455
456 n = 0;
457 list_for_each_entry_continue(bd2, blist, bd_list) {
458 get_bh(bd2->bd_bh);
459 gfs2_log_unlock(sdp);
460 lock_buffer(bd2->bd_bh);
461
462 if (buffer_escaped(bd2->bd_bh)) {
463 void *kaddr;
464 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
465 ptr = page_address(page);
466 kaddr = kmap_atomic(bd2->bd_bh->b_page);
467 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
468 bd2->bd_bh->b_size);
469 kunmap_atomic(kaddr);
470 *(__be32 *)ptr = 0;
471 clear_buffer_escaped(bd2->bd_bh);
472 unlock_buffer(bd2->bd_bh);
473 brelse(bd2->bd_bh);
474 gfs2_log_write_page(sdp, page);
475 } else {
476 gfs2_log_write_bh(sdp, bd2->bd_bh);
477 }
478 gfs2_log_lock(sdp);
479 if (++n >= num)
480 break;
481 }
482
483 BUG_ON(total < num);
484 total -= num;
485 }
486 gfs2_log_unlock(sdp);
487}
488
489static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
490{
491 unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
492 unsigned int nbuf;
493 if (tr == NULL)
494 return;
495 nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
496 gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
497}
498
499static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
500{
501 struct list_head *head;
502 struct gfs2_bufdata *bd;
503
504 if (tr == NULL)
505 return;
506
507 head = &tr->tr_buf;
508 while (!list_empty(head)) {
509 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
510 list_del_init(&bd->bd_list);
511 gfs2_unpin(sdp, bd->bd_bh, tr);
512 }
513}
514
515static void buf_lo_before_scan(struct gfs2_jdesc *jd,
516 struct gfs2_log_header_host *head, int pass)
517{
518 if (pass != 0)
519 return;
520
521 jd->jd_found_blocks = 0;
522 jd->jd_replayed_blocks = 0;
523}
524
525static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
526 struct gfs2_log_descriptor *ld, __be64 *ptr,
527 int pass)
528{
529 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
530 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
531 struct gfs2_glock *gl = ip->i_gl;
532 unsigned int blks = be32_to_cpu(ld->ld_data1);
533 struct buffer_head *bh_log, *bh_ip;
534 u64 blkno;
535 int error = 0;
536
537 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
538 return 0;
539
540 gfs2_replay_incr_blk(jd, &start);
541
542 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
543 blkno = be64_to_cpu(*ptr++);
544
545 jd->jd_found_blocks++;
546
547 if (gfs2_revoke_check(jd, blkno, start))
548 continue;
549
550 error = gfs2_replay_read_block(jd, start, &bh_log);
551 if (error)
552 return error;
553
554 bh_ip = gfs2_meta_new(gl, blkno);
555 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
556
557 if (gfs2_meta_check(sdp, bh_ip))
558 error = -EIO;
559 else
560 mark_buffer_dirty(bh_ip);
561
562 brelse(bh_log);
563 brelse(bh_ip);
564
565 if (error)
566 break;
567
568 jd->jd_replayed_blocks++;
569 }
570
571 return error;
572}
573
574/**
575 * gfs2_meta_sync - Sync all buffers associated with a glock
576 * @gl: The glock
577 *
578 */
579
580static void gfs2_meta_sync(struct gfs2_glock *gl)
581{
582 struct address_space *mapping = gfs2_glock2aspace(gl);
583 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
584 int error;
585
586 if (mapping == NULL)
587 mapping = &sdp->sd_aspace;
588
589 filemap_fdatawrite(mapping);
590 error = filemap_fdatawait(mapping);
591
592 if (error)
593 gfs2_io_error(gl->gl_name.ln_sbd);
594}
595
596static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
597{
598 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
599 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
600
601 if (error) {
602 gfs2_meta_sync(ip->i_gl);
603 return;
604 }
605 if (pass != 1)
606 return;
607
608 gfs2_meta_sync(ip->i_gl);
609
610 fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
611 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
612}
613
614static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
615{
616 struct gfs2_meta_header *mh;
617 unsigned int offset;
618 struct list_head *head = &sdp->sd_log_le_revoke;
619 struct gfs2_bufdata *bd;
620 struct page *page;
621 unsigned int length;
622
623 gfs2_write_revokes(sdp);
624 if (!sdp->sd_log_num_revoke)
625 return;
626
627 length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
628 page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
629 offset = sizeof(struct gfs2_log_descriptor);
630
631 list_for_each_entry(bd, head, bd_list) {
632 sdp->sd_log_num_revoke--;
633
634 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
635
636 gfs2_log_write_page(sdp, page);
637 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
638 mh = page_address(page);
639 clear_page(mh);
640 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
641 mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
642 mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
643 offset = sizeof(struct gfs2_meta_header);
644 }
645
646 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
647 offset += sizeof(u64);
648 }
649 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
650
651 gfs2_log_write_page(sdp, page);
652}
653
654static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
655{
656 struct list_head *head = &sdp->sd_log_le_revoke;
657 struct gfs2_bufdata *bd;
658 struct gfs2_glock *gl;
659
660 while (!list_empty(head)) {
661 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
662 list_del_init(&bd->bd_list);
663 gl = bd->bd_gl;
664 atomic_dec(&gl->gl_revokes);
665 clear_bit(GLF_LFLUSH, &gl->gl_flags);
666 kmem_cache_free(gfs2_bufdata_cachep, bd);
667 }
668}
669
670static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
671 struct gfs2_log_header_host *head, int pass)
672{
673 if (pass != 0)
674 return;
675
676 jd->jd_found_revokes = 0;
677 jd->jd_replay_tail = head->lh_tail;
678}
679
680static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
681 struct gfs2_log_descriptor *ld, __be64 *ptr,
682 int pass)
683{
684 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
685 unsigned int blks = be32_to_cpu(ld->ld_length);
686 unsigned int revokes = be32_to_cpu(ld->ld_data1);
687 struct buffer_head *bh;
688 unsigned int offset;
689 u64 blkno;
690 int first = 1;
691 int error;
692
693 if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
694 return 0;
695
696 offset = sizeof(struct gfs2_log_descriptor);
697
698 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
699 error = gfs2_replay_read_block(jd, start, &bh);
700 if (error)
701 return error;
702
703 if (!first)
704 gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
705
706 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
707 blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
708
709 error = gfs2_revoke_add(jd, blkno, start);
710 if (error < 0) {
711 brelse(bh);
712 return error;
713 }
714 else if (error)
715 jd->jd_found_revokes++;
716
717 if (!--revokes)
718 break;
719 offset += sizeof(u64);
720 }
721
722 brelse(bh);
723 offset = sizeof(struct gfs2_meta_header);
724 first = 0;
725 }
726
727 return 0;
728}
729
730static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
731{
732 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
733
734 if (error) {
735 gfs2_revoke_clean(jd);
736 return;
737 }
738 if (pass != 1)
739 return;
740
741 fs_info(sdp, "jid=%u: Found %u revoke tags\n",
742 jd->jd_jid, jd->jd_found_revokes);
743
744 gfs2_revoke_clean(jd);
745}
746
747/**
748 * databuf_lo_before_commit - Scan the data buffers, writing as we go
749 *
750 */
751
752static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
753{
754 unsigned int limit = databuf_limit(sdp);
755 unsigned int nbuf;
756 if (tr == NULL)
757 return;
758 nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
759 gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
760}
761
762static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
763 struct gfs2_log_descriptor *ld,
764 __be64 *ptr, int pass)
765{
766 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
767 struct gfs2_glock *gl = ip->i_gl;
768 unsigned int blks = be32_to_cpu(ld->ld_data1);
769 struct buffer_head *bh_log, *bh_ip;
770 u64 blkno;
771 u64 esc;
772 int error = 0;
773
774 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
775 return 0;
776
777 gfs2_replay_incr_blk(jd, &start);
778 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
779 blkno = be64_to_cpu(*ptr++);
780 esc = be64_to_cpu(*ptr++);
781
782 jd->jd_found_blocks++;
783
784 if (gfs2_revoke_check(jd, blkno, start))
785 continue;
786
787 error = gfs2_replay_read_block(jd, start, &bh_log);
788 if (error)
789 return error;
790
791 bh_ip = gfs2_meta_new(gl, blkno);
792 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
793
794 /* Unescape */
795 if (esc) {
796 __be32 *eptr = (__be32 *)bh_ip->b_data;
797 *eptr = cpu_to_be32(GFS2_MAGIC);
798 }
799 mark_buffer_dirty(bh_ip);
800
801 brelse(bh_log);
802 brelse(bh_ip);
803
804 jd->jd_replayed_blocks++;
805 }
806
807 return error;
808}
809
810/* FIXME: sort out accounting for log blocks etc. */
811
812static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
813{
814 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
815 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
816
817 if (error) {
818 gfs2_meta_sync(ip->i_gl);
819 return;
820 }
821 if (pass != 1)
822 return;
823
824 /* data sync? */
825 gfs2_meta_sync(ip->i_gl);
826
827 fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
828 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
829}
830
831static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
832{
833 struct list_head *head;
834 struct gfs2_bufdata *bd;
835
836 if (tr == NULL)
837 return;
838
839 head = &tr->tr_databuf;
840 while (!list_empty(head)) {
841 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
842 list_del_init(&bd->bd_list);
843 gfs2_unpin(sdp, bd->bd_bh, tr);
844 }
845}
846
847
848const struct gfs2_log_operations gfs2_buf_lops = {
849 .lo_before_commit = buf_lo_before_commit,
850 .lo_after_commit = buf_lo_after_commit,
851 .lo_before_scan = buf_lo_before_scan,
852 .lo_scan_elements = buf_lo_scan_elements,
853 .lo_after_scan = buf_lo_after_scan,
854 .lo_name = "buf",
855};
856
857const struct gfs2_log_operations gfs2_revoke_lops = {
858 .lo_before_commit = revoke_lo_before_commit,
859 .lo_after_commit = revoke_lo_after_commit,
860 .lo_before_scan = revoke_lo_before_scan,
861 .lo_scan_elements = revoke_lo_scan_elements,
862 .lo_after_scan = revoke_lo_after_scan,
863 .lo_name = "revoke",
864};
865
866const struct gfs2_log_operations gfs2_databuf_lops = {
867 .lo_before_commit = databuf_lo_before_commit,
868 .lo_after_commit = databuf_lo_after_commit,
869 .lo_scan_elements = databuf_lo_scan_elements,
870 .lo_after_scan = databuf_lo_after_scan,
871 .lo_name = "databuf",
872};
873
874const struct gfs2_log_operations *gfs2_log_ops[] = {
875 &gfs2_databuf_lops,
876 &gfs2_buf_lops,
877 &gfs2_revoke_lops,
878 NULL,
879};
880