xref: /linux/fs/gfs2/lops.c (revision b8bb76713ec50df2f11efee386e16f93d51e1076)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
4  *
5  * This copyrighted material is made available to anyone wishing to use,
6  * modify, copy, or redistribute it subject to the terms and conditions
7  * of the GNU General Public License version 2.
8  */
9 
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/gfs2_ondisk.h>
16 
17 #include "gfs2.h"
18 #include "incore.h"
19 #include "inode.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 /**
30  * gfs2_pin - Pin a buffer in memory
31  * @sdp: The superblock
32  * @bh: The buffer to be pinned
33  *
34  * The log lock must be held when calling this function
35  */
36 static void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
37 {
38 	struct gfs2_bufdata *bd;
39 
40 	gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
41 
42 	clear_buffer_dirty(bh);
43 	if (test_set_buffer_pinned(bh))
44 		gfs2_assert_withdraw(sdp, 0);
45 	if (!buffer_uptodate(bh))
46 		gfs2_io_error_bh(sdp, bh);
47 	bd = bh->b_private;
48 	/* If this buffer is in the AIL and it has already been written
49 	 * to in-place disk block, remove it from the AIL.
50 	 */
51 	if (bd->bd_ail)
52 		list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
53 	get_bh(bh);
54 }
55 
56 /**
57  * gfs2_unpin - Unpin a buffer
58  * @sdp: the filesystem the buffer belongs to
59  * @bh: The buffer to unpin
60  * @ai:
61  *
62  */
63 
64 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
65 		       struct gfs2_ail *ai)
66 {
67 	struct gfs2_bufdata *bd = bh->b_private;
68 
69 	gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
70 
71 	if (!buffer_pinned(bh))
72 		gfs2_assert_withdraw(sdp, 0);
73 
74 	lock_buffer(bh);
75 	mark_buffer_dirty(bh);
76 	clear_buffer_pinned(bh);
77 
78 	gfs2_log_lock(sdp);
79 	if (bd->bd_ail) {
80 		list_del(&bd->bd_ail_st_list);
81 		brelse(bh);
82 	} else {
83 		struct gfs2_glock *gl = bd->bd_gl;
84 		list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
85 		atomic_inc(&gl->gl_ail_count);
86 	}
87 	bd->bd_ail = ai;
88 	list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
89 	clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
90 	gfs2_log_unlock(sdp);
91 	unlock_buffer(bh);
92 }
93 
94 
95 static inline struct gfs2_log_descriptor *bh_log_desc(struct buffer_head *bh)
96 {
97 	return (struct gfs2_log_descriptor *)bh->b_data;
98 }
99 
100 static inline __be64 *bh_log_ptr(struct buffer_head *bh)
101 {
102 	struct gfs2_log_descriptor *ld = bh_log_desc(bh);
103 	return (__force __be64 *)(ld + 1);
104 }
105 
106 static inline __be64 *bh_ptr_end(struct buffer_head *bh)
107 {
108 	return (__force __be64 *)(bh->b_data + bh->b_size);
109 }
110 
111 
112 static struct buffer_head *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type)
113 {
114 	struct buffer_head *bh = gfs2_log_get_buf(sdp);
115 	struct gfs2_log_descriptor *ld = bh_log_desc(bh);
116 	ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
117 	ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
118 	ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
119 	ld->ld_type = cpu_to_be32(ld_type);
120 	ld->ld_length = 0;
121 	ld->ld_data1 = 0;
122 	ld->ld_data2 = 0;
123 	memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
124 	return bh;
125 }
126 
127 static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
128 {
129 	struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
130 	struct gfs2_trans *tr;
131 
132 	lock_buffer(bd->bd_bh);
133 	gfs2_log_lock(sdp);
134 	if (!list_empty(&bd->bd_list_tr))
135 		goto out;
136 	tr = current->journal_info;
137 	tr->tr_touched = 1;
138 	tr->tr_num_buf++;
139 	list_add(&bd->bd_list_tr, &tr->tr_list_buf);
140 	if (!list_empty(&le->le_list))
141 		goto out;
142 	set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
143 	set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
144 	gfs2_meta_check(sdp, bd->bd_bh);
145 	gfs2_pin(sdp, bd->bd_bh);
146 	sdp->sd_log_num_buf++;
147 	list_add(&le->le_list, &sdp->sd_log_le_buf);
148 	tr->tr_num_buf_new++;
149 out:
150 	gfs2_log_unlock(sdp);
151 	unlock_buffer(bd->bd_bh);
152 }
153 
154 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
155 {
156 	struct buffer_head *bh;
157 	struct gfs2_log_descriptor *ld;
158 	struct gfs2_bufdata *bd1 = NULL, *bd2;
159 	unsigned int total;
160 	unsigned int limit;
161 	unsigned int num;
162 	unsigned n;
163 	__be64 *ptr;
164 
165 	limit = buf_limit(sdp);
166 	/* for 4k blocks, limit = 503 */
167 
168 	gfs2_log_lock(sdp);
169 	total = sdp->sd_log_num_buf;
170 	bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
171 	while(total) {
172 		num = total;
173 		if (total > limit)
174 			num = limit;
175 		gfs2_log_unlock(sdp);
176 		bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA);
177 		gfs2_log_lock(sdp);
178 		ld = bh_log_desc(bh);
179 		ptr = bh_log_ptr(bh);
180 		ld->ld_length = cpu_to_be32(num + 1);
181 		ld->ld_data1 = cpu_to_be32(num);
182 
183 		n = 0;
184 		list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
185 					     bd_le.le_list) {
186 			*ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
187 			if (++n >= num)
188 				break;
189 		}
190 
191 		gfs2_log_unlock(sdp);
192 		submit_bh(WRITE, bh);
193 		gfs2_log_lock(sdp);
194 
195 		n = 0;
196 		list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
197 					     bd_le.le_list) {
198 			get_bh(bd2->bd_bh);
199 			gfs2_log_unlock(sdp);
200 			lock_buffer(bd2->bd_bh);
201 			bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
202 			submit_bh(WRITE, bh);
203 			gfs2_log_lock(sdp);
204 			if (++n >= num)
205 				break;
206 		}
207 
208 		BUG_ON(total < num);
209 		total -= num;
210 	}
211 	gfs2_log_unlock(sdp);
212 }
213 
214 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
215 {
216 	struct list_head *head = &sdp->sd_log_le_buf;
217 	struct gfs2_bufdata *bd;
218 
219 	while (!list_empty(head)) {
220 		bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
221 		list_del_init(&bd->bd_le.le_list);
222 		sdp->sd_log_num_buf--;
223 
224 		gfs2_unpin(sdp, bd->bd_bh, ai);
225 	}
226 	gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
227 }
228 
229 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
230 			       struct gfs2_log_header_host *head, int pass)
231 {
232 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
233 
234 	if (pass != 0)
235 		return;
236 
237 	sdp->sd_found_blocks = 0;
238 	sdp->sd_replayed_blocks = 0;
239 }
240 
241 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
242 				struct gfs2_log_descriptor *ld, __be64 *ptr,
243 				int pass)
244 {
245 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
246 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
247 	struct gfs2_glock *gl = ip->i_gl;
248 	unsigned int blks = be32_to_cpu(ld->ld_data1);
249 	struct buffer_head *bh_log, *bh_ip;
250 	u64 blkno;
251 	int error = 0;
252 
253 	if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
254 		return 0;
255 
256 	gfs2_replay_incr_blk(sdp, &start);
257 
258 	for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
259 		blkno = be64_to_cpu(*ptr++);
260 
261 		sdp->sd_found_blocks++;
262 
263 		if (gfs2_revoke_check(sdp, blkno, start))
264 			continue;
265 
266 		error = gfs2_replay_read_block(jd, start, &bh_log);
267 		if (error)
268 			return error;
269 
270 		bh_ip = gfs2_meta_new(gl, blkno);
271 		memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
272 
273 		if (gfs2_meta_check(sdp, bh_ip))
274 			error = -EIO;
275 		else
276 			mark_buffer_dirty(bh_ip);
277 
278 		brelse(bh_log);
279 		brelse(bh_ip);
280 
281 		if (error)
282 			break;
283 
284 		sdp->sd_replayed_blocks++;
285 	}
286 
287 	return error;
288 }
289 
290 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
291 {
292 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
293 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
294 
295 	if (error) {
296 		gfs2_meta_sync(ip->i_gl);
297 		return;
298 	}
299 	if (pass != 1)
300 		return;
301 
302 	gfs2_meta_sync(ip->i_gl);
303 
304 	fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
305 	        jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
306 }
307 
308 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
309 {
310 	struct gfs2_trans *tr;
311 
312 	tr = current->journal_info;
313 	tr->tr_touched = 1;
314 	tr->tr_num_revoke++;
315 	sdp->sd_log_num_revoke++;
316 	list_add(&le->le_list, &sdp->sd_log_le_revoke);
317 }
318 
319 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
320 {
321 	struct gfs2_log_descriptor *ld;
322 	struct gfs2_meta_header *mh;
323 	struct buffer_head *bh;
324 	unsigned int offset;
325 	struct list_head *head = &sdp->sd_log_le_revoke;
326 	struct gfs2_bufdata *bd;
327 
328 	if (!sdp->sd_log_num_revoke)
329 		return;
330 
331 	bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE);
332 	ld = bh_log_desc(bh);
333 	ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
334 						    sizeof(u64)));
335 	ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
336 	offset = sizeof(struct gfs2_log_descriptor);
337 
338 	while (!list_empty(head)) {
339 		bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
340 		list_del_init(&bd->bd_le.le_list);
341 		sdp->sd_log_num_revoke--;
342 
343 		if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
344 			submit_bh(WRITE, bh);
345 
346 			bh = gfs2_log_get_buf(sdp);
347 			mh = (struct gfs2_meta_header *)bh->b_data;
348 			mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
349 			mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
350 			mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
351 			offset = sizeof(struct gfs2_meta_header);
352 		}
353 
354 		*(__be64 *)(bh->b_data + offset) = cpu_to_be64(bd->bd_blkno);
355 		kmem_cache_free(gfs2_bufdata_cachep, bd);
356 
357 		offset += sizeof(u64);
358 	}
359 	gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
360 
361 	submit_bh(WRITE, bh);
362 }
363 
364 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
365 				  struct gfs2_log_header_host *head, int pass)
366 {
367 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
368 
369 	if (pass != 0)
370 		return;
371 
372 	sdp->sd_found_revokes = 0;
373 	sdp->sd_replay_tail = head->lh_tail;
374 }
375 
376 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
377 				   struct gfs2_log_descriptor *ld, __be64 *ptr,
378 				   int pass)
379 {
380 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
381 	unsigned int blks = be32_to_cpu(ld->ld_length);
382 	unsigned int revokes = be32_to_cpu(ld->ld_data1);
383 	struct buffer_head *bh;
384 	unsigned int offset;
385 	u64 blkno;
386 	int first = 1;
387 	int error;
388 
389 	if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
390 		return 0;
391 
392 	offset = sizeof(struct gfs2_log_descriptor);
393 
394 	for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
395 		error = gfs2_replay_read_block(jd, start, &bh);
396 		if (error)
397 			return error;
398 
399 		if (!first)
400 			gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
401 
402 		while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
403 			blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
404 
405 			error = gfs2_revoke_add(sdp, blkno, start);
406 			if (error < 0) {
407 				brelse(bh);
408 				return error;
409 			}
410 			else if (error)
411 				sdp->sd_found_revokes++;
412 
413 			if (!--revokes)
414 				break;
415 			offset += sizeof(u64);
416 		}
417 
418 		brelse(bh);
419 		offset = sizeof(struct gfs2_meta_header);
420 		first = 0;
421 	}
422 
423 	return 0;
424 }
425 
426 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
427 {
428 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
429 
430 	if (error) {
431 		gfs2_revoke_clean(sdp);
432 		return;
433 	}
434 	if (pass != 1)
435 		return;
436 
437 	fs_info(sdp, "jid=%u: Found %u revoke tags\n",
438 	        jd->jd_jid, sdp->sd_found_revokes);
439 
440 	gfs2_revoke_clean(sdp);
441 }
442 
443 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
444 {
445 	struct gfs2_rgrpd *rgd;
446 	struct gfs2_trans *tr = current->journal_info;
447 
448 	tr->tr_touched = 1;
449 
450 	rgd = container_of(le, struct gfs2_rgrpd, rd_le);
451 
452 	gfs2_log_lock(sdp);
453 	if (!list_empty(&le->le_list)){
454 		gfs2_log_unlock(sdp);
455 		return;
456 	}
457 	gfs2_rgrp_bh_hold(rgd);
458 	sdp->sd_log_num_rg++;
459 	list_add(&le->le_list, &sdp->sd_log_le_rg);
460 	gfs2_log_unlock(sdp);
461 }
462 
463 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
464 {
465 	struct list_head *head = &sdp->sd_log_le_rg;
466 	struct gfs2_rgrpd *rgd;
467 
468 	while (!list_empty(head)) {
469 		rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
470 		list_del_init(&rgd->rd_le.le_list);
471 		sdp->sd_log_num_rg--;
472 
473 		gfs2_rgrp_repolish_clones(rgd);
474 		gfs2_rgrp_bh_put(rgd);
475 	}
476 	gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
477 }
478 
479 /**
480  * databuf_lo_add - Add a databuf to the transaction.
481  *
482  * This is used in two distinct cases:
483  * i) In ordered write mode
484  *    We put the data buffer on a list so that we can ensure that its
485  *    synced to disk at the right time
486  * ii) In journaled data mode
487  *    We need to journal the data block in the same way as metadata in
488  *    the functions above. The difference is that here we have a tag
489  *    which is two __be64's being the block number (as per meta data)
490  *    and a flag which says whether the data block needs escaping or
491  *    not. This means we need a new log entry for each 251 or so data
492  *    blocks, which isn't an enormous overhead but twice as much as
493  *    for normal metadata blocks.
494  */
495 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
496 {
497 	struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
498 	struct gfs2_trans *tr = current->journal_info;
499 	struct address_space *mapping = bd->bd_bh->b_page->mapping;
500 	struct gfs2_inode *ip = GFS2_I(mapping->host);
501 
502 	lock_buffer(bd->bd_bh);
503 	gfs2_log_lock(sdp);
504 	if (tr) {
505 		if (!list_empty(&bd->bd_list_tr))
506 			goto out;
507 		tr->tr_touched = 1;
508 		if (gfs2_is_jdata(ip)) {
509 			tr->tr_num_buf++;
510 			list_add(&bd->bd_list_tr, &tr->tr_list_buf);
511 		}
512 	}
513 	if (!list_empty(&le->le_list))
514 		goto out;
515 
516 	set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
517 	set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
518 	if (gfs2_is_jdata(ip)) {
519 		gfs2_pin(sdp, bd->bd_bh);
520 		tr->tr_num_databuf_new++;
521 		sdp->sd_log_num_databuf++;
522 		list_add(&le->le_list, &sdp->sd_log_le_databuf);
523 	} else {
524 		list_add(&le->le_list, &sdp->sd_log_le_ordered);
525 	}
526 out:
527 	gfs2_log_unlock(sdp);
528 	unlock_buffer(bd->bd_bh);
529 }
530 
531 static void gfs2_check_magic(struct buffer_head *bh)
532 {
533 	void *kaddr;
534 	__be32 *ptr;
535 
536 	clear_buffer_escaped(bh);
537 	kaddr = kmap_atomic(bh->b_page, KM_USER0);
538 	ptr = kaddr + bh_offset(bh);
539 	if (*ptr == cpu_to_be32(GFS2_MAGIC))
540 		set_buffer_escaped(bh);
541 	kunmap_atomic(kaddr, KM_USER0);
542 }
543 
544 static void gfs2_write_blocks(struct gfs2_sbd *sdp, struct buffer_head *bh,
545 			      struct list_head *list, struct list_head *done,
546 			      unsigned int n)
547 {
548 	struct buffer_head *bh1;
549 	struct gfs2_log_descriptor *ld;
550 	struct gfs2_bufdata *bd;
551 	__be64 *ptr;
552 
553 	if (!bh)
554 		return;
555 
556 	ld = bh_log_desc(bh);
557 	ld->ld_length = cpu_to_be32(n + 1);
558 	ld->ld_data1 = cpu_to_be32(n);
559 
560 	ptr = bh_log_ptr(bh);
561 
562 	get_bh(bh);
563 	submit_bh(WRITE, bh);
564 	gfs2_log_lock(sdp);
565 	while(!list_empty(list)) {
566 		bd = list_entry(list->next, struct gfs2_bufdata, bd_le.le_list);
567 		list_move_tail(&bd->bd_le.le_list, done);
568 		get_bh(bd->bd_bh);
569 		while (be64_to_cpu(*ptr) != bd->bd_bh->b_blocknr) {
570 			gfs2_log_incr_head(sdp);
571 			ptr += 2;
572 		}
573 		gfs2_log_unlock(sdp);
574 		lock_buffer(bd->bd_bh);
575 		if (buffer_escaped(bd->bd_bh)) {
576 			void *kaddr;
577 			bh1 = gfs2_log_get_buf(sdp);
578 			kaddr = kmap_atomic(bd->bd_bh->b_page, KM_USER0);
579 			memcpy(bh1->b_data, kaddr + bh_offset(bd->bd_bh),
580 			       bh1->b_size);
581 			kunmap_atomic(kaddr, KM_USER0);
582 			*(__be32 *)bh1->b_data = 0;
583 			clear_buffer_escaped(bd->bd_bh);
584 			unlock_buffer(bd->bd_bh);
585 			brelse(bd->bd_bh);
586 		} else {
587 			bh1 = gfs2_log_fake_buf(sdp, bd->bd_bh);
588 		}
589 		submit_bh(WRITE, bh1);
590 		gfs2_log_lock(sdp);
591 		ptr += 2;
592 	}
593 	gfs2_log_unlock(sdp);
594 	brelse(bh);
595 }
596 
597 /**
598  * databuf_lo_before_commit - Scan the data buffers, writing as we go
599  *
600  */
601 
602 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
603 {
604 	struct gfs2_bufdata *bd = NULL;
605 	struct buffer_head *bh = NULL;
606 	unsigned int n = 0;
607 	__be64 *ptr = NULL, *end = NULL;
608 	LIST_HEAD(processed);
609 	LIST_HEAD(in_progress);
610 
611 	gfs2_log_lock(sdp);
612 	while (!list_empty(&sdp->sd_log_le_databuf)) {
613 		if (ptr == end) {
614 			gfs2_log_unlock(sdp);
615 			gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
616 			n = 0;
617 			bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_JDATA);
618 			ptr = bh_log_ptr(bh);
619 			end = bh_ptr_end(bh) - 1;
620 			gfs2_log_lock(sdp);
621 			continue;
622 		}
623 		bd = list_entry(sdp->sd_log_le_databuf.next, struct gfs2_bufdata, bd_le.le_list);
624 		list_move_tail(&bd->bd_le.le_list, &in_progress);
625 		gfs2_check_magic(bd->bd_bh);
626 		*ptr++ = cpu_to_be64(bd->bd_bh->b_blocknr);
627 		*ptr++ = cpu_to_be64(buffer_escaped(bh) ? 1 : 0);
628 		n++;
629 	}
630 	gfs2_log_unlock(sdp);
631 	gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
632 	gfs2_log_lock(sdp);
633 	list_splice(&processed, &sdp->sd_log_le_databuf);
634 	gfs2_log_unlock(sdp);
635 }
636 
637 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
638 				    struct gfs2_log_descriptor *ld,
639 				    __be64 *ptr, int pass)
640 {
641 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
642 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
643 	struct gfs2_glock *gl = ip->i_gl;
644 	unsigned int blks = be32_to_cpu(ld->ld_data1);
645 	struct buffer_head *bh_log, *bh_ip;
646 	u64 blkno;
647 	u64 esc;
648 	int error = 0;
649 
650 	if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
651 		return 0;
652 
653 	gfs2_replay_incr_blk(sdp, &start);
654 	for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
655 		blkno = be64_to_cpu(*ptr++);
656 		esc = be64_to_cpu(*ptr++);
657 
658 		sdp->sd_found_blocks++;
659 
660 		if (gfs2_revoke_check(sdp, blkno, start))
661 			continue;
662 
663 		error = gfs2_replay_read_block(jd, start, &bh_log);
664 		if (error)
665 			return error;
666 
667 		bh_ip = gfs2_meta_new(gl, blkno);
668 		memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
669 
670 		/* Unescape */
671 		if (esc) {
672 			__be32 *eptr = (__be32 *)bh_ip->b_data;
673 			*eptr = cpu_to_be32(GFS2_MAGIC);
674 		}
675 		mark_buffer_dirty(bh_ip);
676 
677 		brelse(bh_log);
678 		brelse(bh_ip);
679 		if (error)
680 			break;
681 
682 		sdp->sd_replayed_blocks++;
683 	}
684 
685 	return error;
686 }
687 
688 /* FIXME: sort out accounting for log blocks etc. */
689 
690 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
691 {
692 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
693 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
694 
695 	if (error) {
696 		gfs2_meta_sync(ip->i_gl);
697 		return;
698 	}
699 	if (pass != 1)
700 		return;
701 
702 	/* data sync? */
703 	gfs2_meta_sync(ip->i_gl);
704 
705 	fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
706 		jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
707 }
708 
709 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
710 {
711 	struct list_head *head = &sdp->sd_log_le_databuf;
712 	struct gfs2_bufdata *bd;
713 
714 	while (!list_empty(head)) {
715 		bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
716 		list_del_init(&bd->bd_le.le_list);
717 		sdp->sd_log_num_databuf--;
718 		gfs2_unpin(sdp, bd->bd_bh, ai);
719 	}
720 	gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
721 }
722 
723 
724 const struct gfs2_log_operations gfs2_buf_lops = {
725 	.lo_add = buf_lo_add,
726 	.lo_before_commit = buf_lo_before_commit,
727 	.lo_after_commit = buf_lo_after_commit,
728 	.lo_before_scan = buf_lo_before_scan,
729 	.lo_scan_elements = buf_lo_scan_elements,
730 	.lo_after_scan = buf_lo_after_scan,
731 	.lo_name = "buf",
732 };
733 
734 const struct gfs2_log_operations gfs2_revoke_lops = {
735 	.lo_add = revoke_lo_add,
736 	.lo_before_commit = revoke_lo_before_commit,
737 	.lo_before_scan = revoke_lo_before_scan,
738 	.lo_scan_elements = revoke_lo_scan_elements,
739 	.lo_after_scan = revoke_lo_after_scan,
740 	.lo_name = "revoke",
741 };
742 
743 const struct gfs2_log_operations gfs2_rg_lops = {
744 	.lo_add = rg_lo_add,
745 	.lo_after_commit = rg_lo_after_commit,
746 	.lo_name = "rg",
747 };
748 
749 const struct gfs2_log_operations gfs2_databuf_lops = {
750 	.lo_add = databuf_lo_add,
751 	.lo_before_commit = databuf_lo_before_commit,
752 	.lo_after_commit = databuf_lo_after_commit,
753 	.lo_scan_elements = databuf_lo_scan_elements,
754 	.lo_after_scan = databuf_lo_after_scan,
755 	.lo_name = "databuf",
756 };
757 
758 const struct gfs2_log_operations *gfs2_log_ops[] = {
759 	&gfs2_databuf_lops,
760 	&gfs2_buf_lops,
761 	&gfs2_rg_lops,
762 	&gfs2_revoke_lops,
763 	NULL,
764 };
765 
766