xref: /openbmc/linux/fs/gfs2/lops.c (revision a1e58bbd)
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 "inode.h"
21 #include "glock.h"
22 #include "log.h"
23 #include "lops.h"
24 #include "meta_io.h"
25 #include "recovery.h"
26 #include "rgrp.h"
27 #include "trans.h"
28 #include "util.h"
29 
30 /**
31  * gfs2_pin - Pin a buffer in memory
32  * @sdp: The superblock
33  * @bh: The buffer to be pinned
34  *
35  * The log lock must be held when calling this function
36  */
37 static void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
38 {
39 	struct gfs2_bufdata *bd;
40 
41 	gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
42 
43 	clear_buffer_dirty(bh);
44 	if (test_set_buffer_pinned(bh))
45 		gfs2_assert_withdraw(sdp, 0);
46 	if (!buffer_uptodate(bh))
47 		gfs2_io_error_bh(sdp, bh);
48 	bd = bh->b_private;
49 	/* If this buffer is in the AIL and it has already been written
50 	 * to in-place disk block, remove it from the AIL.
51 	 */
52 	if (bd->bd_ail)
53 		list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
54 	get_bh(bh);
55 }
56 
57 /**
58  * gfs2_unpin - Unpin a buffer
59  * @sdp: the filesystem the buffer belongs to
60  * @bh: The buffer to unpin
61  * @ai:
62  *
63  */
64 
65 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
66 		       struct gfs2_ail *ai)
67 {
68 	struct gfs2_bufdata *bd = bh->b_private;
69 
70 	gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
71 
72 	if (!buffer_pinned(bh))
73 		gfs2_assert_withdraw(sdp, 0);
74 
75 	lock_buffer(bh);
76 	mark_buffer_dirty(bh);
77 	clear_buffer_pinned(bh);
78 
79 	gfs2_log_lock(sdp);
80 	if (bd->bd_ail) {
81 		list_del(&bd->bd_ail_st_list);
82 		brelse(bh);
83 	} else {
84 		struct gfs2_glock *gl = bd->bd_gl;
85 		list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
86 		atomic_inc(&gl->gl_ail_count);
87 	}
88 	bd->bd_ail = ai;
89 	list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
90 	clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
91 	gfs2_log_unlock(sdp);
92 	unlock_buffer(bh);
93 }
94 
95 
96 static inline struct gfs2_log_descriptor *bh_log_desc(struct buffer_head *bh)
97 {
98 	return (struct gfs2_log_descriptor *)bh->b_data;
99 }
100 
101 static inline __be64 *bh_log_ptr(struct buffer_head *bh)
102 {
103 	struct gfs2_log_descriptor *ld = bh_log_desc(bh);
104 	return (__force __be64 *)(ld + 1);
105 }
106 
107 static inline __be64 *bh_ptr_end(struct buffer_head *bh)
108 {
109 	return (__force __be64 *)(bh->b_data + bh->b_size);
110 }
111 
112 
113 static struct buffer_head *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type)
114 {
115 	struct buffer_head *bh = gfs2_log_get_buf(sdp);
116 	struct gfs2_log_descriptor *ld = bh_log_desc(bh);
117 	ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
118 	ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
119 	ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
120 	ld->ld_type = cpu_to_be32(ld_type);
121 	ld->ld_length = 0;
122 	ld->ld_data1 = 0;
123 	ld->ld_data2 = 0;
124 	memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
125 	return bh;
126 }
127 
128 static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
129 {
130 	struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
131 	struct gfs2_trans *tr;
132 
133 	lock_buffer(bd->bd_bh);
134 	gfs2_log_lock(sdp);
135 	if (!list_empty(&bd->bd_list_tr))
136 		goto out;
137 	tr = current->journal_info;
138 	tr->tr_touched = 1;
139 	tr->tr_num_buf++;
140 	list_add(&bd->bd_list_tr, &tr->tr_list_buf);
141 	if (!list_empty(&le->le_list))
142 		goto out;
143 	set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
144 	set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
145 	gfs2_meta_check(sdp, bd->bd_bh);
146 	gfs2_pin(sdp, bd->bd_bh);
147 	sdp->sd_log_num_buf++;
148 	list_add(&le->le_list, &sdp->sd_log_le_buf);
149 	tr->tr_num_buf_new++;
150 out:
151 	gfs2_log_unlock(sdp);
152 	unlock_buffer(bd->bd_bh);
153 }
154 
155 static void buf_lo_incore_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
156 {
157 	struct list_head *head = &tr->tr_list_buf;
158 	struct gfs2_bufdata *bd;
159 
160 	gfs2_log_lock(sdp);
161 	while (!list_empty(head)) {
162 		bd = list_entry(head->next, struct gfs2_bufdata, bd_list_tr);
163 		list_del_init(&bd->bd_list_tr);
164 		tr->tr_num_buf--;
165 	}
166 	gfs2_log_unlock(sdp);
167 	gfs2_assert_warn(sdp, !tr->tr_num_buf);
168 }
169 
170 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
171 {
172 	struct buffer_head *bh;
173 	struct gfs2_log_descriptor *ld;
174 	struct gfs2_bufdata *bd1 = NULL, *bd2;
175 	unsigned int total;
176 	unsigned int limit;
177 	unsigned int num;
178 	unsigned n;
179 	__be64 *ptr;
180 
181 	limit = buf_limit(sdp);
182 	/* for 4k blocks, limit = 503 */
183 
184 	gfs2_log_lock(sdp);
185 	total = sdp->sd_log_num_buf;
186 	bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
187 	while(total) {
188 		num = total;
189 		if (total > limit)
190 			num = limit;
191 		gfs2_log_unlock(sdp);
192 		bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA);
193 		gfs2_log_lock(sdp);
194 		ld = bh_log_desc(bh);
195 		ptr = bh_log_ptr(bh);
196 		ld->ld_length = cpu_to_be32(num + 1);
197 		ld->ld_data1 = cpu_to_be32(num);
198 
199 		n = 0;
200 		list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
201 					     bd_le.le_list) {
202 			*ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
203 			if (++n >= num)
204 				break;
205 		}
206 
207 		gfs2_log_unlock(sdp);
208 		submit_bh(WRITE, bh);
209 		gfs2_log_lock(sdp);
210 
211 		n = 0;
212 		list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
213 					     bd_le.le_list) {
214 			get_bh(bd2->bd_bh);
215 			gfs2_log_unlock(sdp);
216 			lock_buffer(bd2->bd_bh);
217 			bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
218 			submit_bh(WRITE, bh);
219 			gfs2_log_lock(sdp);
220 			if (++n >= num)
221 				break;
222 		}
223 
224 		BUG_ON(total < num);
225 		total -= num;
226 	}
227 	gfs2_log_unlock(sdp);
228 }
229 
230 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
231 {
232 	struct list_head *head = &sdp->sd_log_le_buf;
233 	struct gfs2_bufdata *bd;
234 
235 	while (!list_empty(head)) {
236 		bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
237 		list_del_init(&bd->bd_le.le_list);
238 		sdp->sd_log_num_buf--;
239 
240 		gfs2_unpin(sdp, bd->bd_bh, ai);
241 	}
242 	gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
243 }
244 
245 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
246 			       struct gfs2_log_header_host *head, int pass)
247 {
248 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
249 
250 	if (pass != 0)
251 		return;
252 
253 	sdp->sd_found_blocks = 0;
254 	sdp->sd_replayed_blocks = 0;
255 }
256 
257 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
258 				struct gfs2_log_descriptor *ld, __be64 *ptr,
259 				int pass)
260 {
261 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
262 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
263 	struct gfs2_glock *gl = ip->i_gl;
264 	unsigned int blks = be32_to_cpu(ld->ld_data1);
265 	struct buffer_head *bh_log, *bh_ip;
266 	u64 blkno;
267 	int error = 0;
268 
269 	if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
270 		return 0;
271 
272 	gfs2_replay_incr_blk(sdp, &start);
273 
274 	for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
275 		blkno = be64_to_cpu(*ptr++);
276 
277 		sdp->sd_found_blocks++;
278 
279 		if (gfs2_revoke_check(sdp, blkno, start))
280 			continue;
281 
282 		error = gfs2_replay_read_block(jd, start, &bh_log);
283 		if (error)
284 			return error;
285 
286 		bh_ip = gfs2_meta_new(gl, blkno);
287 		memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
288 
289 		if (gfs2_meta_check(sdp, bh_ip))
290 			error = -EIO;
291 		else
292 			mark_buffer_dirty(bh_ip);
293 
294 		brelse(bh_log);
295 		brelse(bh_ip);
296 
297 		if (error)
298 			break;
299 
300 		sdp->sd_replayed_blocks++;
301 	}
302 
303 	return error;
304 }
305 
306 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
307 {
308 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
309 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
310 
311 	if (error) {
312 		gfs2_meta_sync(ip->i_gl);
313 		return;
314 	}
315 	if (pass != 1)
316 		return;
317 
318 	gfs2_meta_sync(ip->i_gl);
319 
320 	fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
321 	        jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
322 }
323 
324 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
325 {
326 	struct gfs2_trans *tr;
327 
328 	tr = current->journal_info;
329 	tr->tr_touched = 1;
330 	tr->tr_num_revoke++;
331 	sdp->sd_log_num_revoke++;
332 	list_add(&le->le_list, &sdp->sd_log_le_revoke);
333 }
334 
335 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
336 {
337 	struct gfs2_log_descriptor *ld;
338 	struct gfs2_meta_header *mh;
339 	struct buffer_head *bh;
340 	unsigned int offset;
341 	struct list_head *head = &sdp->sd_log_le_revoke;
342 	struct gfs2_bufdata *bd;
343 
344 	if (!sdp->sd_log_num_revoke)
345 		return;
346 
347 	bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE);
348 	ld = bh_log_desc(bh);
349 	ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
350 						    sizeof(u64)));
351 	ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
352 	offset = sizeof(struct gfs2_log_descriptor);
353 
354 	while (!list_empty(head)) {
355 		bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
356 		list_del_init(&bd->bd_le.le_list);
357 		sdp->sd_log_num_revoke--;
358 
359 		if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
360 			submit_bh(WRITE, bh);
361 
362 			bh = gfs2_log_get_buf(sdp);
363 			mh = (struct gfs2_meta_header *)bh->b_data;
364 			mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
365 			mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
366 			mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
367 			offset = sizeof(struct gfs2_meta_header);
368 		}
369 
370 		*(__be64 *)(bh->b_data + offset) = cpu_to_be64(bd->bd_blkno);
371 		kmem_cache_free(gfs2_bufdata_cachep, bd);
372 
373 		offset += sizeof(u64);
374 	}
375 	gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
376 
377 	submit_bh(WRITE, bh);
378 }
379 
380 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
381 				  struct gfs2_log_header_host *head, int pass)
382 {
383 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
384 
385 	if (pass != 0)
386 		return;
387 
388 	sdp->sd_found_revokes = 0;
389 	sdp->sd_replay_tail = head->lh_tail;
390 }
391 
392 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
393 				   struct gfs2_log_descriptor *ld, __be64 *ptr,
394 				   int pass)
395 {
396 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
397 	unsigned int blks = be32_to_cpu(ld->ld_length);
398 	unsigned int revokes = be32_to_cpu(ld->ld_data1);
399 	struct buffer_head *bh;
400 	unsigned int offset;
401 	u64 blkno;
402 	int first = 1;
403 	int error;
404 
405 	if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
406 		return 0;
407 
408 	offset = sizeof(struct gfs2_log_descriptor);
409 
410 	for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
411 		error = gfs2_replay_read_block(jd, start, &bh);
412 		if (error)
413 			return error;
414 
415 		if (!first)
416 			gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
417 
418 		while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
419 			blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
420 
421 			error = gfs2_revoke_add(sdp, blkno, start);
422 			if (error < 0)
423 				return error;
424 			else if (error)
425 				sdp->sd_found_revokes++;
426 
427 			if (!--revokes)
428 				break;
429 			offset += sizeof(u64);
430 		}
431 
432 		brelse(bh);
433 		offset = sizeof(struct gfs2_meta_header);
434 		first = 0;
435 	}
436 
437 	return 0;
438 }
439 
440 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
441 {
442 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
443 
444 	if (error) {
445 		gfs2_revoke_clean(sdp);
446 		return;
447 	}
448 	if (pass != 1)
449 		return;
450 
451 	fs_info(sdp, "jid=%u: Found %u revoke tags\n",
452 	        jd->jd_jid, sdp->sd_found_revokes);
453 
454 	gfs2_revoke_clean(sdp);
455 }
456 
457 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
458 {
459 	struct gfs2_rgrpd *rgd;
460 	struct gfs2_trans *tr = current->journal_info;
461 
462 	tr->tr_touched = 1;
463 
464 	rgd = container_of(le, struct gfs2_rgrpd, rd_le);
465 
466 	gfs2_log_lock(sdp);
467 	if (!list_empty(&le->le_list)){
468 		gfs2_log_unlock(sdp);
469 		return;
470 	}
471 	gfs2_rgrp_bh_hold(rgd);
472 	sdp->sd_log_num_rg++;
473 	list_add(&le->le_list, &sdp->sd_log_le_rg);
474 	gfs2_log_unlock(sdp);
475 }
476 
477 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
478 {
479 	struct list_head *head = &sdp->sd_log_le_rg;
480 	struct gfs2_rgrpd *rgd;
481 
482 	while (!list_empty(head)) {
483 		rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
484 		list_del_init(&rgd->rd_le.le_list);
485 		sdp->sd_log_num_rg--;
486 
487 		gfs2_rgrp_repolish_clones(rgd);
488 		gfs2_rgrp_bh_put(rgd);
489 	}
490 	gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
491 }
492 
493 /**
494  * databuf_lo_add - Add a databuf to the transaction.
495  *
496  * This is used in two distinct cases:
497  * i) In ordered write mode
498  *    We put the data buffer on a list so that we can ensure that its
499  *    synced to disk at the right time
500  * ii) In journaled data mode
501  *    We need to journal the data block in the same way as metadata in
502  *    the functions above. The difference is that here we have a tag
503  *    which is two __be64's being the block number (as per meta data)
504  *    and a flag which says whether the data block needs escaping or
505  *    not. This means we need a new log entry for each 251 or so data
506  *    blocks, which isn't an enormous overhead but twice as much as
507  *    for normal metadata blocks.
508  */
509 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
510 {
511 	struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
512 	struct gfs2_trans *tr = current->journal_info;
513 	struct address_space *mapping = bd->bd_bh->b_page->mapping;
514 	struct gfs2_inode *ip = GFS2_I(mapping->host);
515 
516 	lock_buffer(bd->bd_bh);
517 	gfs2_log_lock(sdp);
518 	if (tr) {
519 		if (!list_empty(&bd->bd_list_tr))
520 			goto out;
521 		tr->tr_touched = 1;
522 		if (gfs2_is_jdata(ip)) {
523 			tr->tr_num_buf++;
524 			list_add(&bd->bd_list_tr, &tr->tr_list_buf);
525 		}
526 	}
527 	if (!list_empty(&le->le_list))
528 		goto out;
529 
530 	set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
531 	set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
532 	if (gfs2_is_jdata(ip)) {
533 		gfs2_pin(sdp, bd->bd_bh);
534 		tr->tr_num_databuf_new++;
535 		sdp->sd_log_num_databuf++;
536 		list_add(&le->le_list, &sdp->sd_log_le_databuf);
537 	} else {
538 		list_add(&le->le_list, &sdp->sd_log_le_ordered);
539 	}
540 out:
541 	gfs2_log_unlock(sdp);
542 	unlock_buffer(bd->bd_bh);
543 }
544 
545 static void gfs2_check_magic(struct buffer_head *bh)
546 {
547 	void *kaddr;
548 	__be32 *ptr;
549 
550 	clear_buffer_escaped(bh);
551 	kaddr = kmap_atomic(bh->b_page, KM_USER0);
552 	ptr = kaddr + bh_offset(bh);
553 	if (*ptr == cpu_to_be32(GFS2_MAGIC))
554 		set_buffer_escaped(bh);
555 	kunmap_atomic(kaddr, KM_USER0);
556 }
557 
558 static void gfs2_write_blocks(struct gfs2_sbd *sdp, struct buffer_head *bh,
559 			      struct list_head *list, struct list_head *done,
560 			      unsigned int n)
561 {
562 	struct buffer_head *bh1;
563 	struct gfs2_log_descriptor *ld;
564 	struct gfs2_bufdata *bd;
565 	__be64 *ptr;
566 
567 	if (!bh)
568 		return;
569 
570 	ld = bh_log_desc(bh);
571 	ld->ld_length = cpu_to_be32(n + 1);
572 	ld->ld_data1 = cpu_to_be32(n);
573 
574 	ptr = bh_log_ptr(bh);
575 
576 	get_bh(bh);
577 	submit_bh(WRITE, bh);
578 	gfs2_log_lock(sdp);
579 	while(!list_empty(list)) {
580 		bd = list_entry(list->next, struct gfs2_bufdata, bd_le.le_list);
581 		list_move_tail(&bd->bd_le.le_list, done);
582 		get_bh(bd->bd_bh);
583 		while (be64_to_cpu(*ptr) != bd->bd_bh->b_blocknr) {
584 			gfs2_log_incr_head(sdp);
585 			ptr += 2;
586 		}
587 		gfs2_log_unlock(sdp);
588 		lock_buffer(bd->bd_bh);
589 		if (buffer_escaped(bd->bd_bh)) {
590 			void *kaddr;
591 			bh1 = gfs2_log_get_buf(sdp);
592 			kaddr = kmap_atomic(bd->bd_bh->b_page, KM_USER0);
593 			memcpy(bh1->b_data, kaddr + bh_offset(bd->bd_bh),
594 			       bh1->b_size);
595 			kunmap_atomic(kaddr, KM_USER0);
596 			*(__be32 *)bh1->b_data = 0;
597 			clear_buffer_escaped(bd->bd_bh);
598 			unlock_buffer(bd->bd_bh);
599 			brelse(bd->bd_bh);
600 		} else {
601 			bh1 = gfs2_log_fake_buf(sdp, bd->bd_bh);
602 		}
603 		submit_bh(WRITE, bh1);
604 		gfs2_log_lock(sdp);
605 		ptr += 2;
606 	}
607 	gfs2_log_unlock(sdp);
608 	brelse(bh);
609 }
610 
611 /**
612  * databuf_lo_before_commit - Scan the data buffers, writing as we go
613  *
614  */
615 
616 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
617 {
618 	struct gfs2_bufdata *bd = NULL;
619 	struct buffer_head *bh = NULL;
620 	unsigned int n = 0;
621 	__be64 *ptr = NULL, *end = NULL;
622 	LIST_HEAD(processed);
623 	LIST_HEAD(in_progress);
624 
625 	gfs2_log_lock(sdp);
626 	while (!list_empty(&sdp->sd_log_le_databuf)) {
627 		if (ptr == end) {
628 			gfs2_log_unlock(sdp);
629 			gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
630 			n = 0;
631 			bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_JDATA);
632 			ptr = bh_log_ptr(bh);
633 			end = bh_ptr_end(bh) - 1;
634 			gfs2_log_lock(sdp);
635 			continue;
636 		}
637 		bd = list_entry(sdp->sd_log_le_databuf.next, struct gfs2_bufdata, bd_le.le_list);
638 		list_move_tail(&bd->bd_le.le_list, &in_progress);
639 		gfs2_check_magic(bd->bd_bh);
640 		*ptr++ = cpu_to_be64(bd->bd_bh->b_blocknr);
641 		*ptr++ = cpu_to_be64(buffer_escaped(bh) ? 1 : 0);
642 		n++;
643 	}
644 	gfs2_log_unlock(sdp);
645 	gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
646 	gfs2_log_lock(sdp);
647 	list_splice(&processed, &sdp->sd_log_le_databuf);
648 	gfs2_log_unlock(sdp);
649 }
650 
651 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
652 				    struct gfs2_log_descriptor *ld,
653 				    __be64 *ptr, int pass)
654 {
655 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
656 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
657 	struct gfs2_glock *gl = ip->i_gl;
658 	unsigned int blks = be32_to_cpu(ld->ld_data1);
659 	struct buffer_head *bh_log, *bh_ip;
660 	u64 blkno;
661 	u64 esc;
662 	int error = 0;
663 
664 	if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
665 		return 0;
666 
667 	gfs2_replay_incr_blk(sdp, &start);
668 	for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
669 		blkno = be64_to_cpu(*ptr++);
670 		esc = be64_to_cpu(*ptr++);
671 
672 		sdp->sd_found_blocks++;
673 
674 		if (gfs2_revoke_check(sdp, blkno, start))
675 			continue;
676 
677 		error = gfs2_replay_read_block(jd, start, &bh_log);
678 		if (error)
679 			return error;
680 
681 		bh_ip = gfs2_meta_new(gl, blkno);
682 		memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
683 
684 		/* Unescape */
685 		if (esc) {
686 			__be32 *eptr = (__be32 *)bh_ip->b_data;
687 			*eptr = cpu_to_be32(GFS2_MAGIC);
688 		}
689 		mark_buffer_dirty(bh_ip);
690 
691 		brelse(bh_log);
692 		brelse(bh_ip);
693 		if (error)
694 			break;
695 
696 		sdp->sd_replayed_blocks++;
697 	}
698 
699 	return error;
700 }
701 
702 /* FIXME: sort out accounting for log blocks etc. */
703 
704 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
705 {
706 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
707 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
708 
709 	if (error) {
710 		gfs2_meta_sync(ip->i_gl);
711 		return;
712 	}
713 	if (pass != 1)
714 		return;
715 
716 	/* data sync? */
717 	gfs2_meta_sync(ip->i_gl);
718 
719 	fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
720 		jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
721 }
722 
723 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
724 {
725 	struct list_head *head = &sdp->sd_log_le_databuf;
726 	struct gfs2_bufdata *bd;
727 
728 	while (!list_empty(head)) {
729 		bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
730 		list_del_init(&bd->bd_le.le_list);
731 		sdp->sd_log_num_databuf--;
732 		gfs2_unpin(sdp, bd->bd_bh, ai);
733 	}
734 	gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
735 }
736 
737 
738 const struct gfs2_log_operations gfs2_buf_lops = {
739 	.lo_add = buf_lo_add,
740 	.lo_incore_commit = buf_lo_incore_commit,
741 	.lo_before_commit = buf_lo_before_commit,
742 	.lo_after_commit = buf_lo_after_commit,
743 	.lo_before_scan = buf_lo_before_scan,
744 	.lo_scan_elements = buf_lo_scan_elements,
745 	.lo_after_scan = buf_lo_after_scan,
746 	.lo_name = "buf",
747 };
748 
749 const struct gfs2_log_operations gfs2_revoke_lops = {
750 	.lo_add = revoke_lo_add,
751 	.lo_before_commit = revoke_lo_before_commit,
752 	.lo_before_scan = revoke_lo_before_scan,
753 	.lo_scan_elements = revoke_lo_scan_elements,
754 	.lo_after_scan = revoke_lo_after_scan,
755 	.lo_name = "revoke",
756 };
757 
758 const struct gfs2_log_operations gfs2_rg_lops = {
759 	.lo_add = rg_lo_add,
760 	.lo_after_commit = rg_lo_after_commit,
761 	.lo_name = "rg",
762 };
763 
764 const struct gfs2_log_operations gfs2_databuf_lops = {
765 	.lo_add = databuf_lo_add,
766 	.lo_incore_commit = buf_lo_incore_commit,
767 	.lo_before_commit = databuf_lo_before_commit,
768 	.lo_after_commit = databuf_lo_after_commit,
769 	.lo_scan_elements = databuf_lo_scan_elements,
770 	.lo_after_scan = databuf_lo_after_scan,
771 	.lo_name = "databuf",
772 };
773 
774 const struct gfs2_log_operations *gfs2_log_ops[] = {
775 	&gfs2_databuf_lops,
776 	&gfs2_buf_lops,
777 	&gfs2_rg_lops,
778 	&gfs2_revoke_lops,
779 	NULL,
780 };
781 
782