xref: /openbmc/linux/fs/gfs2/lops.c (revision 4800cd83)
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 	gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
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 	if (bd->bd_ail)
55 		list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
56 	get_bh(bh);
57 	atomic_inc(&sdp->sd_log_pinned);
58 	trace_gfs2_pin(bd, 1);
59 }
60 
61 /**
62  * gfs2_unpin - Unpin a buffer
63  * @sdp: the filesystem the buffer belongs to
64  * @bh: The buffer to unpin
65  * @ai:
66  *
67  */
68 
69 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
70 		       struct gfs2_ail *ai)
71 {
72 	struct gfs2_bufdata *bd = bh->b_private;
73 
74 	gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
75 
76 	if (!buffer_pinned(bh))
77 		gfs2_assert_withdraw(sdp, 0);
78 
79 	lock_buffer(bh);
80 	mark_buffer_dirty(bh);
81 	clear_buffer_pinned(bh);
82 
83 	gfs2_log_lock(sdp);
84 	if (bd->bd_ail) {
85 		list_del(&bd->bd_ail_st_list);
86 		brelse(bh);
87 	} else {
88 		struct gfs2_glock *gl = bd->bd_gl;
89 		list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
90 		atomic_inc(&gl->gl_ail_count);
91 	}
92 	bd->bd_ail = ai;
93 	list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
94 	clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
95 	trace_gfs2_pin(bd, 0);
96 	gfs2_log_unlock(sdp);
97 	unlock_buffer(bh);
98 	atomic_dec(&sdp->sd_log_pinned);
99 }
100 
101 
102 static inline struct gfs2_log_descriptor *bh_log_desc(struct buffer_head *bh)
103 {
104 	return (struct gfs2_log_descriptor *)bh->b_data;
105 }
106 
107 static inline __be64 *bh_log_ptr(struct buffer_head *bh)
108 {
109 	struct gfs2_log_descriptor *ld = bh_log_desc(bh);
110 	return (__force __be64 *)(ld + 1);
111 }
112 
113 static inline __be64 *bh_ptr_end(struct buffer_head *bh)
114 {
115 	return (__force __be64 *)(bh->b_data + bh->b_size);
116 }
117 
118 
119 static struct buffer_head *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type)
120 {
121 	struct buffer_head *bh = gfs2_log_get_buf(sdp);
122 	struct gfs2_log_descriptor *ld = bh_log_desc(bh);
123 	ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
124 	ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
125 	ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
126 	ld->ld_type = cpu_to_be32(ld_type);
127 	ld->ld_length = 0;
128 	ld->ld_data1 = 0;
129 	ld->ld_data2 = 0;
130 	memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
131 	return bh;
132 }
133 
134 static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
135 {
136 	struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
137 	struct gfs2_meta_header *mh;
138 	struct gfs2_trans *tr;
139 
140 	lock_buffer(bd->bd_bh);
141 	gfs2_log_lock(sdp);
142 	if (!list_empty(&bd->bd_list_tr))
143 		goto out;
144 	tr = current->journal_info;
145 	tr->tr_touched = 1;
146 	tr->tr_num_buf++;
147 	list_add(&bd->bd_list_tr, &tr->tr_list_buf);
148 	if (!list_empty(&le->le_list))
149 		goto out;
150 	set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
151 	set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
152 	gfs2_meta_check(sdp, bd->bd_bh);
153 	gfs2_pin(sdp, bd->bd_bh);
154 	mh = (struct gfs2_meta_header *)bd->bd_bh->b_data;
155 	mh->__pad0 = cpu_to_be64(0);
156 	mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
157 	sdp->sd_log_num_buf++;
158 	list_add(&le->le_list, &sdp->sd_log_le_buf);
159 	tr->tr_num_buf_new++;
160 out:
161 	gfs2_log_unlock(sdp);
162 	unlock_buffer(bd->bd_bh);
163 }
164 
165 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
166 {
167 	struct buffer_head *bh;
168 	struct gfs2_log_descriptor *ld;
169 	struct gfs2_bufdata *bd1 = NULL, *bd2;
170 	unsigned int total;
171 	unsigned int limit;
172 	unsigned int num;
173 	unsigned n;
174 	__be64 *ptr;
175 
176 	limit = buf_limit(sdp);
177 	/* for 4k blocks, limit = 503 */
178 
179 	gfs2_log_lock(sdp);
180 	total = sdp->sd_log_num_buf;
181 	bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
182 	while(total) {
183 		num = total;
184 		if (total > limit)
185 			num = limit;
186 		gfs2_log_unlock(sdp);
187 		bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA);
188 		gfs2_log_lock(sdp);
189 		ld = bh_log_desc(bh);
190 		ptr = bh_log_ptr(bh);
191 		ld->ld_length = cpu_to_be32(num + 1);
192 		ld->ld_data1 = cpu_to_be32(num);
193 
194 		n = 0;
195 		list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
196 					     bd_le.le_list) {
197 			*ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
198 			if (++n >= num)
199 				break;
200 		}
201 
202 		gfs2_log_unlock(sdp);
203 		submit_bh(WRITE_SYNC_PLUG, bh);
204 		gfs2_log_lock(sdp);
205 
206 		n = 0;
207 		list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
208 					     bd_le.le_list) {
209 			get_bh(bd2->bd_bh);
210 			gfs2_log_unlock(sdp);
211 			lock_buffer(bd2->bd_bh);
212 			bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
213 			submit_bh(WRITE_SYNC_PLUG, bh);
214 			gfs2_log_lock(sdp);
215 			if (++n >= num)
216 				break;
217 		}
218 
219 		BUG_ON(total < num);
220 		total -= num;
221 	}
222 	gfs2_log_unlock(sdp);
223 }
224 
225 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
226 {
227 	struct list_head *head = &sdp->sd_log_le_buf;
228 	struct gfs2_bufdata *bd;
229 
230 	while (!list_empty(head)) {
231 		bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
232 		list_del_init(&bd->bd_le.le_list);
233 		sdp->sd_log_num_buf--;
234 
235 		gfs2_unpin(sdp, bd->bd_bh, ai);
236 	}
237 	gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
238 }
239 
240 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
241 			       struct gfs2_log_header_host *head, int pass)
242 {
243 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
244 
245 	if (pass != 0)
246 		return;
247 
248 	sdp->sd_found_blocks = 0;
249 	sdp->sd_replayed_blocks = 0;
250 }
251 
252 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
253 				struct gfs2_log_descriptor *ld, __be64 *ptr,
254 				int pass)
255 {
256 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
257 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
258 	struct gfs2_glock *gl = ip->i_gl;
259 	unsigned int blks = be32_to_cpu(ld->ld_data1);
260 	struct buffer_head *bh_log, *bh_ip;
261 	u64 blkno;
262 	int error = 0;
263 
264 	if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
265 		return 0;
266 
267 	gfs2_replay_incr_blk(sdp, &start);
268 
269 	for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
270 		blkno = be64_to_cpu(*ptr++);
271 
272 		sdp->sd_found_blocks++;
273 
274 		if (gfs2_revoke_check(sdp, blkno, start))
275 			continue;
276 
277 		error = gfs2_replay_read_block(jd, start, &bh_log);
278 		if (error)
279 			return error;
280 
281 		bh_ip = gfs2_meta_new(gl, blkno);
282 		memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
283 
284 		if (gfs2_meta_check(sdp, bh_ip))
285 			error = -EIO;
286 		else
287 			mark_buffer_dirty(bh_ip);
288 
289 		brelse(bh_log);
290 		brelse(bh_ip);
291 
292 		if (error)
293 			break;
294 
295 		sdp->sd_replayed_blocks++;
296 	}
297 
298 	return error;
299 }
300 
301 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
302 {
303 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
304 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
305 
306 	if (error) {
307 		gfs2_meta_sync(ip->i_gl);
308 		return;
309 	}
310 	if (pass != 1)
311 		return;
312 
313 	gfs2_meta_sync(ip->i_gl);
314 
315 	fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
316 	        jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
317 }
318 
319 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
320 {
321 	struct gfs2_trans *tr;
322 
323 	tr = current->journal_info;
324 	tr->tr_touched = 1;
325 	tr->tr_num_revoke++;
326 	sdp->sd_log_num_revoke++;
327 	list_add(&le->le_list, &sdp->sd_log_le_revoke);
328 }
329 
330 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
331 {
332 	struct gfs2_log_descriptor *ld;
333 	struct gfs2_meta_header *mh;
334 	struct buffer_head *bh;
335 	unsigned int offset;
336 	struct list_head *head = &sdp->sd_log_le_revoke;
337 	struct gfs2_bufdata *bd;
338 
339 	if (!sdp->sd_log_num_revoke)
340 		return;
341 
342 	bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE);
343 	ld = bh_log_desc(bh);
344 	ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
345 						    sizeof(u64)));
346 	ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
347 	offset = sizeof(struct gfs2_log_descriptor);
348 
349 	while (!list_empty(head)) {
350 		bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
351 		list_del_init(&bd->bd_le.le_list);
352 		sdp->sd_log_num_revoke--;
353 
354 		if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
355 			submit_bh(WRITE_SYNC_PLUG, bh);
356 
357 			bh = gfs2_log_get_buf(sdp);
358 			mh = (struct gfs2_meta_header *)bh->b_data;
359 			mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
360 			mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
361 			mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
362 			offset = sizeof(struct gfs2_meta_header);
363 		}
364 
365 		*(__be64 *)(bh->b_data + offset) = cpu_to_be64(bd->bd_blkno);
366 		kmem_cache_free(gfs2_bufdata_cachep, bd);
367 
368 		offset += sizeof(u64);
369 	}
370 	gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
371 
372 	submit_bh(WRITE_SYNC_PLUG, bh);
373 }
374 
375 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
376 				  struct gfs2_log_header_host *head, int pass)
377 {
378 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
379 
380 	if (pass != 0)
381 		return;
382 
383 	sdp->sd_found_revokes = 0;
384 	sdp->sd_replay_tail = head->lh_tail;
385 }
386 
387 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
388 				   struct gfs2_log_descriptor *ld, __be64 *ptr,
389 				   int pass)
390 {
391 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
392 	unsigned int blks = be32_to_cpu(ld->ld_length);
393 	unsigned int revokes = be32_to_cpu(ld->ld_data1);
394 	struct buffer_head *bh;
395 	unsigned int offset;
396 	u64 blkno;
397 	int first = 1;
398 	int error;
399 
400 	if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
401 		return 0;
402 
403 	offset = sizeof(struct gfs2_log_descriptor);
404 
405 	for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
406 		error = gfs2_replay_read_block(jd, start, &bh);
407 		if (error)
408 			return error;
409 
410 		if (!first)
411 			gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
412 
413 		while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
414 			blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
415 
416 			error = gfs2_revoke_add(sdp, blkno, start);
417 			if (error < 0) {
418 				brelse(bh);
419 				return error;
420 			}
421 			else if (error)
422 				sdp->sd_found_revokes++;
423 
424 			if (!--revokes)
425 				break;
426 			offset += sizeof(u64);
427 		}
428 
429 		brelse(bh);
430 		offset = sizeof(struct gfs2_meta_header);
431 		first = 0;
432 	}
433 
434 	return 0;
435 }
436 
437 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
438 {
439 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
440 
441 	if (error) {
442 		gfs2_revoke_clean(sdp);
443 		return;
444 	}
445 	if (pass != 1)
446 		return;
447 
448 	fs_info(sdp, "jid=%u: Found %u revoke tags\n",
449 	        jd->jd_jid, sdp->sd_found_revokes);
450 
451 	gfs2_revoke_clean(sdp);
452 }
453 
454 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
455 {
456 	struct gfs2_rgrpd *rgd;
457 	struct gfs2_trans *tr = current->journal_info;
458 
459 	tr->tr_touched = 1;
460 
461 	rgd = container_of(le, struct gfs2_rgrpd, rd_le);
462 
463 	gfs2_log_lock(sdp);
464 	if (!list_empty(&le->le_list)){
465 		gfs2_log_unlock(sdp);
466 		return;
467 	}
468 	gfs2_rgrp_bh_hold(rgd);
469 	sdp->sd_log_num_rg++;
470 	list_add(&le->le_list, &sdp->sd_log_le_rg);
471 	gfs2_log_unlock(sdp);
472 }
473 
474 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
475 {
476 	struct list_head *head = &sdp->sd_log_le_rg;
477 	struct gfs2_rgrpd *rgd;
478 
479 	while (!list_empty(head)) {
480 		rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
481 		list_del_init(&rgd->rd_le.le_list);
482 		sdp->sd_log_num_rg--;
483 
484 		gfs2_rgrp_repolish_clones(rgd);
485 		gfs2_rgrp_bh_put(rgd);
486 	}
487 	gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
488 }
489 
490 /**
491  * databuf_lo_add - Add a databuf to the transaction.
492  *
493  * This is used in two distinct cases:
494  * i) In ordered write mode
495  *    We put the data buffer on a list so that we can ensure that its
496  *    synced to disk at the right time
497  * ii) In journaled data mode
498  *    We need to journal the data block in the same way as metadata in
499  *    the functions above. The difference is that here we have a tag
500  *    which is two __be64's being the block number (as per meta data)
501  *    and a flag which says whether the data block needs escaping or
502  *    not. This means we need a new log entry for each 251 or so data
503  *    blocks, which isn't an enormous overhead but twice as much as
504  *    for normal metadata blocks.
505  */
506 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
507 {
508 	struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
509 	struct gfs2_trans *tr = current->journal_info;
510 	struct address_space *mapping = bd->bd_bh->b_page->mapping;
511 	struct gfs2_inode *ip = GFS2_I(mapping->host);
512 
513 	lock_buffer(bd->bd_bh);
514 	gfs2_log_lock(sdp);
515 	if (tr) {
516 		if (!list_empty(&bd->bd_list_tr))
517 			goto out;
518 		tr->tr_touched = 1;
519 		if (gfs2_is_jdata(ip)) {
520 			tr->tr_num_buf++;
521 			list_add(&bd->bd_list_tr, &tr->tr_list_buf);
522 		}
523 	}
524 	if (!list_empty(&le->le_list))
525 		goto out;
526 
527 	set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
528 	set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
529 	if (gfs2_is_jdata(ip)) {
530 		gfs2_pin(sdp, bd->bd_bh);
531 		tr->tr_num_databuf_new++;
532 		sdp->sd_log_num_databuf++;
533 		list_add_tail(&le->le_list, &sdp->sd_log_le_databuf);
534 	} else {
535 		list_add_tail(&le->le_list, &sdp->sd_log_le_ordered);
536 	}
537 out:
538 	gfs2_log_unlock(sdp);
539 	unlock_buffer(bd->bd_bh);
540 }
541 
542 static void gfs2_check_magic(struct buffer_head *bh)
543 {
544 	void *kaddr;
545 	__be32 *ptr;
546 
547 	clear_buffer_escaped(bh);
548 	kaddr = kmap_atomic(bh->b_page, KM_USER0);
549 	ptr = kaddr + bh_offset(bh);
550 	if (*ptr == cpu_to_be32(GFS2_MAGIC))
551 		set_buffer_escaped(bh);
552 	kunmap_atomic(kaddr, KM_USER0);
553 }
554 
555 static void gfs2_write_blocks(struct gfs2_sbd *sdp, struct buffer_head *bh,
556 			      struct list_head *list, struct list_head *done,
557 			      unsigned int n)
558 {
559 	struct buffer_head *bh1;
560 	struct gfs2_log_descriptor *ld;
561 	struct gfs2_bufdata *bd;
562 	__be64 *ptr;
563 
564 	if (!bh)
565 		return;
566 
567 	ld = bh_log_desc(bh);
568 	ld->ld_length = cpu_to_be32(n + 1);
569 	ld->ld_data1 = cpu_to_be32(n);
570 
571 	ptr = bh_log_ptr(bh);
572 
573 	get_bh(bh);
574 	submit_bh(WRITE_SYNC_PLUG, bh);
575 	gfs2_log_lock(sdp);
576 	while(!list_empty(list)) {
577 		bd = list_entry(list->next, struct gfs2_bufdata, bd_le.le_list);
578 		list_move_tail(&bd->bd_le.le_list, done);
579 		get_bh(bd->bd_bh);
580 		while (be64_to_cpu(*ptr) != bd->bd_bh->b_blocknr) {
581 			gfs2_log_incr_head(sdp);
582 			ptr += 2;
583 		}
584 		gfs2_log_unlock(sdp);
585 		lock_buffer(bd->bd_bh);
586 		if (buffer_escaped(bd->bd_bh)) {
587 			void *kaddr;
588 			bh1 = gfs2_log_get_buf(sdp);
589 			kaddr = kmap_atomic(bd->bd_bh->b_page, KM_USER0);
590 			memcpy(bh1->b_data, kaddr + bh_offset(bd->bd_bh),
591 			       bh1->b_size);
592 			kunmap_atomic(kaddr, KM_USER0);
593 			*(__be32 *)bh1->b_data = 0;
594 			clear_buffer_escaped(bd->bd_bh);
595 			unlock_buffer(bd->bd_bh);
596 			brelse(bd->bd_bh);
597 		} else {
598 			bh1 = gfs2_log_fake_buf(sdp, bd->bd_bh);
599 		}
600 		submit_bh(WRITE_SYNC_PLUG, bh1);
601 		gfs2_log_lock(sdp);
602 		ptr += 2;
603 	}
604 	gfs2_log_unlock(sdp);
605 	brelse(bh);
606 }
607 
608 /**
609  * databuf_lo_before_commit - Scan the data buffers, writing as we go
610  *
611  */
612 
613 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
614 {
615 	struct gfs2_bufdata *bd = NULL;
616 	struct buffer_head *bh = NULL;
617 	unsigned int n = 0;
618 	__be64 *ptr = NULL, *end = NULL;
619 	LIST_HEAD(processed);
620 	LIST_HEAD(in_progress);
621 
622 	gfs2_log_lock(sdp);
623 	while (!list_empty(&sdp->sd_log_le_databuf)) {
624 		if (ptr == end) {
625 			gfs2_log_unlock(sdp);
626 			gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
627 			n = 0;
628 			bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_JDATA);
629 			ptr = bh_log_ptr(bh);
630 			end = bh_ptr_end(bh) - 1;
631 			gfs2_log_lock(sdp);
632 			continue;
633 		}
634 		bd = list_entry(sdp->sd_log_le_databuf.next, struct gfs2_bufdata, bd_le.le_list);
635 		list_move_tail(&bd->bd_le.le_list, &in_progress);
636 		gfs2_check_magic(bd->bd_bh);
637 		*ptr++ = cpu_to_be64(bd->bd_bh->b_blocknr);
638 		*ptr++ = cpu_to_be64(buffer_escaped(bh) ? 1 : 0);
639 		n++;
640 	}
641 	gfs2_log_unlock(sdp);
642 	gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
643 	gfs2_log_lock(sdp);
644 	list_splice(&processed, &sdp->sd_log_le_databuf);
645 	gfs2_log_unlock(sdp);
646 }
647 
648 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
649 				    struct gfs2_log_descriptor *ld,
650 				    __be64 *ptr, int pass)
651 {
652 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
653 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
654 	struct gfs2_glock *gl = ip->i_gl;
655 	unsigned int blks = be32_to_cpu(ld->ld_data1);
656 	struct buffer_head *bh_log, *bh_ip;
657 	u64 blkno;
658 	u64 esc;
659 	int error = 0;
660 
661 	if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
662 		return 0;
663 
664 	gfs2_replay_incr_blk(sdp, &start);
665 	for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
666 		blkno = be64_to_cpu(*ptr++);
667 		esc = be64_to_cpu(*ptr++);
668 
669 		sdp->sd_found_blocks++;
670 
671 		if (gfs2_revoke_check(sdp, blkno, start))
672 			continue;
673 
674 		error = gfs2_replay_read_block(jd, start, &bh_log);
675 		if (error)
676 			return error;
677 
678 		bh_ip = gfs2_meta_new(gl, blkno);
679 		memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
680 
681 		/* Unescape */
682 		if (esc) {
683 			__be32 *eptr = (__be32 *)bh_ip->b_data;
684 			*eptr = cpu_to_be32(GFS2_MAGIC);
685 		}
686 		mark_buffer_dirty(bh_ip);
687 
688 		brelse(bh_log);
689 		brelse(bh_ip);
690 		if (error)
691 			break;
692 
693 		sdp->sd_replayed_blocks++;
694 	}
695 
696 	return error;
697 }
698 
699 /* FIXME: sort out accounting for log blocks etc. */
700 
701 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
702 {
703 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
704 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
705 
706 	if (error) {
707 		gfs2_meta_sync(ip->i_gl);
708 		return;
709 	}
710 	if (pass != 1)
711 		return;
712 
713 	/* data sync? */
714 	gfs2_meta_sync(ip->i_gl);
715 
716 	fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
717 		jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
718 }
719 
720 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
721 {
722 	struct list_head *head = &sdp->sd_log_le_databuf;
723 	struct gfs2_bufdata *bd;
724 
725 	while (!list_empty(head)) {
726 		bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
727 		list_del_init(&bd->bd_le.le_list);
728 		sdp->sd_log_num_databuf--;
729 		gfs2_unpin(sdp, bd->bd_bh, ai);
730 	}
731 	gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
732 }
733 
734 
735 const struct gfs2_log_operations gfs2_buf_lops = {
736 	.lo_add = buf_lo_add,
737 	.lo_before_commit = buf_lo_before_commit,
738 	.lo_after_commit = buf_lo_after_commit,
739 	.lo_before_scan = buf_lo_before_scan,
740 	.lo_scan_elements = buf_lo_scan_elements,
741 	.lo_after_scan = buf_lo_after_scan,
742 	.lo_name = "buf",
743 };
744 
745 const struct gfs2_log_operations gfs2_revoke_lops = {
746 	.lo_add = revoke_lo_add,
747 	.lo_before_commit = revoke_lo_before_commit,
748 	.lo_before_scan = revoke_lo_before_scan,
749 	.lo_scan_elements = revoke_lo_scan_elements,
750 	.lo_after_scan = revoke_lo_after_scan,
751 	.lo_name = "revoke",
752 };
753 
754 const struct gfs2_log_operations gfs2_rg_lops = {
755 	.lo_add = rg_lo_add,
756 	.lo_after_commit = rg_lo_after_commit,
757 	.lo_name = "rg",
758 };
759 
760 const struct gfs2_log_operations gfs2_databuf_lops = {
761 	.lo_add = databuf_lo_add,
762 	.lo_before_commit = databuf_lo_before_commit,
763 	.lo_after_commit = databuf_lo_after_commit,
764 	.lo_scan_elements = databuf_lo_scan_elements,
765 	.lo_after_scan = databuf_lo_after_scan,
766 	.lo_name = "databuf",
767 };
768 
769 const struct gfs2_log_operations *gfs2_log_ops[] = {
770 	&gfs2_databuf_lops,
771 	&gfs2_buf_lops,
772 	&gfs2_rg_lops,
773 	&gfs2_revoke_lops,
774 	NULL,
775 };
776 
777