xref: /openbmc/linux/fs/gfs2/lops.c (revision 54525552)
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  */
39 static 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 
72 static 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 
104 static inline struct gfs2_log_descriptor *bh_log_desc(struct buffer_head *bh)
105 {
106 	return (struct gfs2_log_descriptor *)bh->b_data;
107 }
108 
109 static 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 
115 static inline __be64 *bh_ptr_end(struct buffer_head *bh)
116 {
117 	return (__force __be64 *)(bh->b_data + bh->b_size);
118 }
119 
120 
121 static 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 
136 static 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++;
162 out:
163 	gfs2_log_unlock(sdp);
164 	unlock_buffer(bd->bd_bh);
165 }
166 
167 static 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 
227 static 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 
242 static 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 
254 static 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 
303 static 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 
321 static 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 
336 static 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 
377 static 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 
393 static 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 
405 static 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 
455 static 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 
472 static 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 
492 static 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  */
524 static 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 	}
555 out:
556 	gfs2_log_unlock(sdp);
557 	unlock_buffer(bd->bd_bh);
558 }
559 
560 static 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 
573 static 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 
631 static 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 
666 static 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 
719 static 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 
738 static 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 
753 const 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 
763 const 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 
773 const 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 
779 const 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 
788 const 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 
796