xref: /linux/fs/gfs2/lops.c (revision c537b994505099b7197e7d3125b942ecbcc51eb6)
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 "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 static void glock_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
30 {
31 	struct gfs2_glock *gl;
32 	struct gfs2_trans *tr = current->journal_info;
33 
34 	tr->tr_touched = 1;
35 
36 	if (!list_empty(&le->le_list))
37 		return;
38 
39 	gl = container_of(le, struct gfs2_glock, gl_le);
40 	if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(gl)))
41 		return;
42 	gfs2_glock_hold(gl);
43 	set_bit(GLF_DIRTY, &gl->gl_flags);
44 
45 	gfs2_log_lock(sdp);
46 	sdp->sd_log_num_gl++;
47 	list_add(&le->le_list, &sdp->sd_log_le_gl);
48 	gfs2_log_unlock(sdp);
49 }
50 
51 static void glock_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
52 {
53 	struct list_head *head = &sdp->sd_log_le_gl;
54 	struct gfs2_glock *gl;
55 
56 	while (!list_empty(head)) {
57 		gl = list_entry(head->next, struct gfs2_glock, gl_le.le_list);
58 		list_del_init(&gl->gl_le.le_list);
59 		sdp->sd_log_num_gl--;
60 
61 		gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(gl));
62 		gfs2_glock_put(gl);
63 	}
64 	gfs2_assert_warn(sdp, !sdp->sd_log_num_gl);
65 }
66 
67 static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
68 {
69 	struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
70 	struct gfs2_trans *tr;
71 
72 	gfs2_log_lock(sdp);
73 	if (!list_empty(&bd->bd_list_tr)) {
74 		gfs2_log_unlock(sdp);
75 		return;
76 	}
77 	tr = current->journal_info;
78 	tr->tr_touched = 1;
79 	tr->tr_num_buf++;
80 	list_add(&bd->bd_list_tr, &tr->tr_list_buf);
81 	gfs2_log_unlock(sdp);
82 
83 	if (!list_empty(&le->le_list))
84 		return;
85 
86 	gfs2_trans_add_gl(bd->bd_gl);
87 
88 	gfs2_meta_check(sdp, bd->bd_bh);
89 	gfs2_pin(sdp, bd->bd_bh);
90 	gfs2_log_lock(sdp);
91 	sdp->sd_log_num_buf++;
92 	list_add(&le->le_list, &sdp->sd_log_le_buf);
93 	gfs2_log_unlock(sdp);
94 
95 	tr->tr_num_buf_new++;
96 }
97 
98 static void buf_lo_incore_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
99 {
100 	struct list_head *head = &tr->tr_list_buf;
101 	struct gfs2_bufdata *bd;
102 
103 	gfs2_log_lock(sdp);
104 	while (!list_empty(head)) {
105 		bd = list_entry(head->next, struct gfs2_bufdata, bd_list_tr);
106 		list_del_init(&bd->bd_list_tr);
107 		tr->tr_num_buf--;
108 	}
109 	gfs2_log_unlock(sdp);
110 	gfs2_assert_warn(sdp, !tr->tr_num_buf);
111 }
112 
113 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
114 {
115 	struct buffer_head *bh;
116 	struct gfs2_log_descriptor *ld;
117 	struct gfs2_bufdata *bd1 = NULL, *bd2;
118 	unsigned int total = sdp->sd_log_num_buf;
119 	unsigned int offset = sizeof(struct gfs2_log_descriptor);
120 	unsigned int limit;
121 	unsigned int num;
122 	unsigned n;
123 	__be64 *ptr;
124 
125 	offset += sizeof(__be64) - 1;
126 	offset &= ~(sizeof(__be64) - 1);
127 	limit = (sdp->sd_sb.sb_bsize - offset)/sizeof(__be64);
128 	/* for 4k blocks, limit = 503 */
129 
130 	bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
131 	while(total) {
132 		num = total;
133 		if (total > limit)
134 			num = limit;
135 		bh = gfs2_log_get_buf(sdp);
136 		sdp->sd_log_num_hdrs++;
137 		ld = (struct gfs2_log_descriptor *)bh->b_data;
138 		ptr = (__be64 *)(bh->b_data + offset);
139 		ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
140 		ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
141 		ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
142 		ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_METADATA);
143 		ld->ld_length = cpu_to_be32(num + 1);
144 		ld->ld_data1 = cpu_to_be32(num);
145 		ld->ld_data2 = cpu_to_be32(0);
146 		memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
147 
148 		n = 0;
149 		list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
150 					     bd_le.le_list) {
151 			*ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
152 			if (++n >= num)
153 				break;
154 		}
155 
156 		set_buffer_dirty(bh);
157 		ll_rw_block(WRITE, 1, &bh);
158 
159 		n = 0;
160 		list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
161 					     bd_le.le_list) {
162 			bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
163 			set_buffer_dirty(bh);
164 			ll_rw_block(WRITE, 1, &bh);
165 			if (++n >= num)
166 				break;
167 		}
168 
169 		total -= num;
170 	}
171 }
172 
173 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
174 {
175 	struct list_head *head = &sdp->sd_log_le_buf;
176 	struct gfs2_bufdata *bd;
177 
178 	while (!list_empty(head)) {
179 		bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
180 		list_del_init(&bd->bd_le.le_list);
181 		sdp->sd_log_num_buf--;
182 
183 		gfs2_unpin(sdp, bd->bd_bh, ai);
184 	}
185 	gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
186 }
187 
188 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
189 			       struct gfs2_log_header_host *head, int pass)
190 {
191 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
192 
193 	if (pass != 0)
194 		return;
195 
196 	sdp->sd_found_blocks = 0;
197 	sdp->sd_replayed_blocks = 0;
198 }
199 
200 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
201 				struct gfs2_log_descriptor *ld, __be64 *ptr,
202 				int pass)
203 {
204 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
205 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
206 	struct gfs2_glock *gl = ip->i_gl;
207 	unsigned int blks = be32_to_cpu(ld->ld_data1);
208 	struct buffer_head *bh_log, *bh_ip;
209 	u64 blkno;
210 	int error = 0;
211 
212 	if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
213 		return 0;
214 
215 	gfs2_replay_incr_blk(sdp, &start);
216 
217 	for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
218 		blkno = be64_to_cpu(*ptr++);
219 
220 		sdp->sd_found_blocks++;
221 
222 		if (gfs2_revoke_check(sdp, blkno, start))
223 			continue;
224 
225 		error = gfs2_replay_read_block(jd, start, &bh_log);
226 		if (error)
227 			return error;
228 
229 		bh_ip = gfs2_meta_new(gl, blkno);
230 		memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
231 
232 		if (gfs2_meta_check(sdp, bh_ip))
233 			error = -EIO;
234 		else
235 			mark_buffer_dirty(bh_ip);
236 
237 		brelse(bh_log);
238 		brelse(bh_ip);
239 
240 		if (error)
241 			break;
242 
243 		sdp->sd_replayed_blocks++;
244 	}
245 
246 	return error;
247 }
248 
249 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
250 {
251 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
252 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
253 
254 	if (error) {
255 		gfs2_meta_sync(ip->i_gl);
256 		return;
257 	}
258 	if (pass != 1)
259 		return;
260 
261 	gfs2_meta_sync(ip->i_gl);
262 
263 	fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
264 	        jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
265 }
266 
267 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
268 {
269 	struct gfs2_trans *tr;
270 
271 	tr = current->journal_info;
272 	tr->tr_touched = 1;
273 	tr->tr_num_revoke++;
274 
275 	gfs2_log_lock(sdp);
276 	sdp->sd_log_num_revoke++;
277 	list_add(&le->le_list, &sdp->sd_log_le_revoke);
278 	gfs2_log_unlock(sdp);
279 }
280 
281 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
282 {
283 	struct gfs2_log_descriptor *ld;
284 	struct gfs2_meta_header *mh;
285 	struct buffer_head *bh;
286 	unsigned int offset;
287 	struct list_head *head = &sdp->sd_log_le_revoke;
288 	struct gfs2_revoke *rv;
289 
290 	if (!sdp->sd_log_num_revoke)
291 		return;
292 
293 	bh = gfs2_log_get_buf(sdp);
294 	ld = (struct gfs2_log_descriptor *)bh->b_data;
295 	ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
296 	ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
297 	ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
298 	ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_REVOKE);
299 	ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
300 						    sizeof(u64)));
301 	ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
302 	ld->ld_data2 = cpu_to_be32(0);
303 	memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
304 	offset = sizeof(struct gfs2_log_descriptor);
305 
306 	while (!list_empty(head)) {
307 		rv = list_entry(head->next, struct gfs2_revoke, rv_le.le_list);
308 		list_del_init(&rv->rv_le.le_list);
309 		sdp->sd_log_num_revoke--;
310 
311 		if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
312 			set_buffer_dirty(bh);
313 			ll_rw_block(WRITE, 1, &bh);
314 
315 			bh = gfs2_log_get_buf(sdp);
316 			mh = (struct gfs2_meta_header *)bh->b_data;
317 			mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
318 			mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
319 			mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
320 			offset = sizeof(struct gfs2_meta_header);
321 		}
322 
323 		*(__be64 *)(bh->b_data + offset) = cpu_to_be64(rv->rv_blkno);
324 		kfree(rv);
325 
326 		offset += sizeof(u64);
327 	}
328 	gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
329 
330 	set_buffer_dirty(bh);
331 	ll_rw_block(WRITE, 1, &bh);
332 }
333 
334 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
335 				  struct gfs2_log_header_host *head, int pass)
336 {
337 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
338 
339 	if (pass != 0)
340 		return;
341 
342 	sdp->sd_found_revokes = 0;
343 	sdp->sd_replay_tail = head->lh_tail;
344 }
345 
346 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
347 				   struct gfs2_log_descriptor *ld, __be64 *ptr,
348 				   int pass)
349 {
350 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
351 	unsigned int blks = be32_to_cpu(ld->ld_length);
352 	unsigned int revokes = be32_to_cpu(ld->ld_data1);
353 	struct buffer_head *bh;
354 	unsigned int offset;
355 	u64 blkno;
356 	int first = 1;
357 	int error;
358 
359 	if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
360 		return 0;
361 
362 	offset = sizeof(struct gfs2_log_descriptor);
363 
364 	for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
365 		error = gfs2_replay_read_block(jd, start, &bh);
366 		if (error)
367 			return error;
368 
369 		if (!first)
370 			gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
371 
372 		while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
373 			blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
374 
375 			error = gfs2_revoke_add(sdp, blkno, start);
376 			if (error < 0)
377 				return error;
378 			else if (error)
379 				sdp->sd_found_revokes++;
380 
381 			if (!--revokes)
382 				break;
383 			offset += sizeof(u64);
384 		}
385 
386 		brelse(bh);
387 		offset = sizeof(struct gfs2_meta_header);
388 		first = 0;
389 	}
390 
391 	return 0;
392 }
393 
394 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
395 {
396 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
397 
398 	if (error) {
399 		gfs2_revoke_clean(sdp);
400 		return;
401 	}
402 	if (pass != 1)
403 		return;
404 
405 	fs_info(sdp, "jid=%u: Found %u revoke tags\n",
406 	        jd->jd_jid, sdp->sd_found_revokes);
407 
408 	gfs2_revoke_clean(sdp);
409 }
410 
411 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
412 {
413 	struct gfs2_rgrpd *rgd;
414 	struct gfs2_trans *tr = current->journal_info;
415 
416 	tr->tr_touched = 1;
417 
418 	if (!list_empty(&le->le_list))
419 		return;
420 
421 	rgd = container_of(le, struct gfs2_rgrpd, rd_le);
422 	gfs2_rgrp_bh_hold(rgd);
423 
424 	gfs2_log_lock(sdp);
425 	sdp->sd_log_num_rg++;
426 	list_add(&le->le_list, &sdp->sd_log_le_rg);
427 	gfs2_log_unlock(sdp);
428 }
429 
430 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
431 {
432 	struct list_head *head = &sdp->sd_log_le_rg;
433 	struct gfs2_rgrpd *rgd;
434 
435 	while (!list_empty(head)) {
436 		rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
437 		list_del_init(&rgd->rd_le.le_list);
438 		sdp->sd_log_num_rg--;
439 
440 		gfs2_rgrp_repolish_clones(rgd);
441 		gfs2_rgrp_bh_put(rgd);
442 	}
443 	gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
444 }
445 
446 /**
447  * databuf_lo_add - Add a databuf to the transaction.
448  *
449  * This is used in two distinct cases:
450  * i) In ordered write mode
451  *    We put the data buffer on a list so that we can ensure that its
452  *    synced to disk at the right time
453  * ii) In journaled data mode
454  *    We need to journal the data block in the same way as metadata in
455  *    the functions above. The difference is that here we have a tag
456  *    which is two __be64's being the block number (as per meta data)
457  *    and a flag which says whether the data block needs escaping or
458  *    not. This means we need a new log entry for each 251 or so data
459  *    blocks, which isn't an enormous overhead but twice as much as
460  *    for normal metadata blocks.
461  */
462 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
463 {
464 	struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
465 	struct gfs2_trans *tr = current->journal_info;
466 	struct address_space *mapping = bd->bd_bh->b_page->mapping;
467 	struct gfs2_inode *ip = GFS2_I(mapping->host);
468 
469 	gfs2_log_lock(sdp);
470 	tr->tr_touched = 1;
471 	if (list_empty(&bd->bd_list_tr) &&
472 	    (ip->i_di.di_flags & GFS2_DIF_JDATA)) {
473 		tr->tr_num_buf++;
474 		list_add(&bd->bd_list_tr, &tr->tr_list_buf);
475 		gfs2_log_unlock(sdp);
476 		gfs2_pin(sdp, bd->bd_bh);
477 		tr->tr_num_buf_new++;
478 	} else {
479 		gfs2_log_unlock(sdp);
480 	}
481 	gfs2_trans_add_gl(bd->bd_gl);
482 	gfs2_log_lock(sdp);
483 	if (list_empty(&le->le_list)) {
484 		if (ip->i_di.di_flags & GFS2_DIF_JDATA)
485 			sdp->sd_log_num_jdata++;
486 		sdp->sd_log_num_databuf++;
487 		list_add(&le->le_list, &sdp->sd_log_le_databuf);
488 	}
489 	gfs2_log_unlock(sdp);
490 }
491 
492 static int gfs2_check_magic(struct buffer_head *bh)
493 {
494 	struct page *page = bh->b_page;
495 	void *kaddr;
496 	__be32 *ptr;
497 	int rv = 0;
498 
499 	kaddr = kmap_atomic(page, KM_USER0);
500 	ptr = kaddr + bh_offset(bh);
501 	if (*ptr == cpu_to_be32(GFS2_MAGIC))
502 		rv = 1;
503 	kunmap_atomic(kaddr, KM_USER0);
504 
505 	return rv;
506 }
507 
508 /**
509  * databuf_lo_before_commit - Scan the data buffers, writing as we go
510  *
511  * Here we scan through the lists of buffers and make the assumption
512  * that any buffer thats been pinned is being journaled, and that
513  * any unpinned buffer is an ordered write data buffer and therefore
514  * will be written back rather than journaled.
515  */
516 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
517 {
518 	LIST_HEAD(started);
519 	struct gfs2_bufdata *bd1 = NULL, *bd2, *bdt;
520 	struct buffer_head *bh = NULL,*bh1 = NULL;
521 	unsigned int offset = sizeof(struct gfs2_log_descriptor);
522 	struct gfs2_log_descriptor *ld;
523 	unsigned int limit;
524 	unsigned int total_dbuf = sdp->sd_log_num_databuf;
525 	unsigned int total_jdata = sdp->sd_log_num_jdata;
526 	unsigned int num, n;
527 	__be64 *ptr = NULL;
528 
529 	offset += 2*sizeof(__be64) - 1;
530 	offset &= ~(2*sizeof(__be64) - 1);
531 	limit = (sdp->sd_sb.sb_bsize - offset)/sizeof(__be64);
532 
533 	/*
534 	 * Start writing ordered buffers, write journaled buffers
535 	 * into the log along with a header
536 	 */
537 	gfs2_log_lock(sdp);
538 	bd2 = bd1 = list_prepare_entry(bd1, &sdp->sd_log_le_databuf,
539 				       bd_le.le_list);
540 	while(total_dbuf) {
541 		num = total_jdata;
542 		if (num > limit)
543 			num = limit;
544 		n = 0;
545 		list_for_each_entry_safe_continue(bd1, bdt,
546 						  &sdp->sd_log_le_databuf,
547 						  bd_le.le_list) {
548 			/* store off the buffer head in a local ptr since
549 			 * gfs2_bufdata might change when we drop the log lock
550 			 */
551 			bh1 = bd1->bd_bh;
552 
553 			/* An ordered write buffer */
554 			if (bh1 && !buffer_pinned(bh1)) {
555 				list_move(&bd1->bd_le.le_list, &started);
556 				if (bd1 == bd2) {
557 					bd2 = NULL;
558 					bd2 = list_prepare_entry(bd2,
559 							&sdp->sd_log_le_databuf,
560 							bd_le.le_list);
561 				}
562 				total_dbuf--;
563 				if (bh1) {
564 					if (buffer_dirty(bh1)) {
565 						get_bh(bh1);
566 
567 						gfs2_log_unlock(sdp);
568 
569 						ll_rw_block(SWRITE, 1, &bh1);
570 						brelse(bh1);
571 
572 						gfs2_log_lock(sdp);
573 					}
574 					continue;
575 				}
576 				continue;
577 			} else if (bh1) { /* A journaled buffer */
578 				int magic;
579 				gfs2_log_unlock(sdp);
580 				if (!bh) {
581 					bh = gfs2_log_get_buf(sdp);
582 					sdp->sd_log_num_hdrs++;
583 					ld = (struct gfs2_log_descriptor *)
584 					     bh->b_data;
585 					ptr = (__be64 *)(bh->b_data + offset);
586 					ld->ld_header.mh_magic =
587 						cpu_to_be32(GFS2_MAGIC);
588 					ld->ld_header.mh_type =
589 						cpu_to_be32(GFS2_METATYPE_LD);
590 					ld->ld_header.mh_format =
591 						cpu_to_be32(GFS2_FORMAT_LD);
592 					ld->ld_type =
593 						cpu_to_be32(GFS2_LOG_DESC_JDATA);
594 					ld->ld_length = cpu_to_be32(num + 1);
595 					ld->ld_data1 = cpu_to_be32(num);
596 					ld->ld_data2 = cpu_to_be32(0);
597 					memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
598 				}
599 				magic = gfs2_check_magic(bh1);
600 				*ptr++ = cpu_to_be64(bh1->b_blocknr);
601 				*ptr++ = cpu_to_be64((__u64)magic);
602 				clear_buffer_escaped(bh1);
603 				if (unlikely(magic != 0))
604 					set_buffer_escaped(bh1);
605 				gfs2_log_lock(sdp);
606 				if (n++ > num)
607 					break;
608 			} else if (!bh1) {
609 				total_dbuf--;
610 				sdp->sd_log_num_databuf--;
611 				list_del_init(&bd1->bd_le.le_list);
612 				if (bd1 == bd2) {
613 					bd2 = NULL;
614 					bd2 = list_prepare_entry(bd2,
615 						&sdp->sd_log_le_databuf,
616 						bd_le.le_list);
617                                 }
618 				kmem_cache_free(gfs2_bufdata_cachep, bd1);
619 			}
620 		}
621 		gfs2_log_unlock(sdp);
622 		if (bh) {
623 			set_buffer_dirty(bh);
624 			ll_rw_block(WRITE, 1, &bh);
625 			bh = NULL;
626 		}
627 		n = 0;
628 		gfs2_log_lock(sdp);
629 		list_for_each_entry_continue(bd2, &sdp->sd_log_le_databuf,
630 					     bd_le.le_list) {
631 			if (!bd2->bd_bh)
632 				continue;
633 			/* copy buffer if it needs escaping */
634 			gfs2_log_unlock(sdp);
635 			if (unlikely(buffer_escaped(bd2->bd_bh))) {
636 				void *kaddr;
637 				struct page *page = bd2->bd_bh->b_page;
638 				bh = gfs2_log_get_buf(sdp);
639 				kaddr = kmap_atomic(page, KM_USER0);
640 				memcpy(bh->b_data,
641 				       kaddr + bh_offset(bd2->bd_bh),
642 				       sdp->sd_sb.sb_bsize);
643 				kunmap_atomic(kaddr, KM_USER0);
644 				*(__be32 *)bh->b_data = 0;
645 			} else {
646 				bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
647 			}
648 			set_buffer_dirty(bh);
649 			ll_rw_block(WRITE, 1, &bh);
650 			gfs2_log_lock(sdp);
651 			if (++n >= num)
652 				break;
653 		}
654 		bh = NULL;
655 		total_dbuf -= num;
656 		total_jdata -= num;
657 	}
658 	gfs2_log_unlock(sdp);
659 
660 	/* Wait on all ordered buffers */
661 	while (!list_empty(&started)) {
662 		gfs2_log_lock(sdp);
663 		bd1 = list_entry(started.next, struct gfs2_bufdata,
664 				 bd_le.le_list);
665 		list_del_init(&bd1->bd_le.le_list);
666 		sdp->sd_log_num_databuf--;
667 		bh = bd1->bd_bh;
668 		if (bh) {
669 			bh->b_private = NULL;
670 			get_bh(bh);
671 			gfs2_log_unlock(sdp);
672 			wait_on_buffer(bh);
673 			brelse(bh);
674 		} else
675 			gfs2_log_unlock(sdp);
676 
677 		kmem_cache_free(gfs2_bufdata_cachep, bd1);
678 	}
679 
680 	/* We've removed all the ordered write bufs here, so only jdata left */
681 	gfs2_assert_warn(sdp, sdp->sd_log_num_databuf == sdp->sd_log_num_jdata);
682 }
683 
684 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
685 				    struct gfs2_log_descriptor *ld,
686 				    __be64 *ptr, int pass)
687 {
688 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
689 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
690 	struct gfs2_glock *gl = ip->i_gl;
691 	unsigned int blks = be32_to_cpu(ld->ld_data1);
692 	struct buffer_head *bh_log, *bh_ip;
693 	u64 blkno;
694 	u64 esc;
695 	int error = 0;
696 
697 	if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
698 		return 0;
699 
700 	gfs2_replay_incr_blk(sdp, &start);
701 	for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
702 		blkno = be64_to_cpu(*ptr++);
703 		esc = be64_to_cpu(*ptr++);
704 
705 		sdp->sd_found_blocks++;
706 
707 		if (gfs2_revoke_check(sdp, blkno, start))
708 			continue;
709 
710 		error = gfs2_replay_read_block(jd, start, &bh_log);
711 		if (error)
712 			return error;
713 
714 		bh_ip = gfs2_meta_new(gl, blkno);
715 		memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
716 
717 		/* Unescape */
718 		if (esc) {
719 			__be32 *eptr = (__be32 *)bh_ip->b_data;
720 			*eptr = cpu_to_be32(GFS2_MAGIC);
721 		}
722 		mark_buffer_dirty(bh_ip);
723 
724 		brelse(bh_log);
725 		brelse(bh_ip);
726 		if (error)
727 			break;
728 
729 		sdp->sd_replayed_blocks++;
730 	}
731 
732 	return error;
733 }
734 
735 /* FIXME: sort out accounting for log blocks etc. */
736 
737 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
738 {
739 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
740 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
741 
742 	if (error) {
743 		gfs2_meta_sync(ip->i_gl);
744 		return;
745 	}
746 	if (pass != 1)
747 		return;
748 
749 	/* data sync? */
750 	gfs2_meta_sync(ip->i_gl);
751 
752 	fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
753 		jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
754 }
755 
756 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
757 {
758 	struct list_head *head = &sdp->sd_log_le_databuf;
759 	struct gfs2_bufdata *bd;
760 
761 	while (!list_empty(head)) {
762 		bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
763 		list_del_init(&bd->bd_le.le_list);
764 		sdp->sd_log_num_databuf--;
765 		sdp->sd_log_num_jdata--;
766 		gfs2_unpin(sdp, bd->bd_bh, ai);
767 	}
768 	gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
769 	gfs2_assert_warn(sdp, !sdp->sd_log_num_jdata);
770 }
771 
772 
773 const struct gfs2_log_operations gfs2_glock_lops = {
774 	.lo_add = glock_lo_add,
775 	.lo_after_commit = glock_lo_after_commit,
776 	.lo_name = "glock",
777 };
778 
779 const struct gfs2_log_operations gfs2_buf_lops = {
780 	.lo_add = buf_lo_add,
781 	.lo_incore_commit = buf_lo_incore_commit,
782 	.lo_before_commit = buf_lo_before_commit,
783 	.lo_after_commit = buf_lo_after_commit,
784 	.lo_before_scan = buf_lo_before_scan,
785 	.lo_scan_elements = buf_lo_scan_elements,
786 	.lo_after_scan = buf_lo_after_scan,
787 	.lo_name = "buf",
788 };
789 
790 const struct gfs2_log_operations gfs2_revoke_lops = {
791 	.lo_add = revoke_lo_add,
792 	.lo_before_commit = revoke_lo_before_commit,
793 	.lo_before_scan = revoke_lo_before_scan,
794 	.lo_scan_elements = revoke_lo_scan_elements,
795 	.lo_after_scan = revoke_lo_after_scan,
796 	.lo_name = "revoke",
797 };
798 
799 const struct gfs2_log_operations gfs2_rg_lops = {
800 	.lo_add = rg_lo_add,
801 	.lo_after_commit = rg_lo_after_commit,
802 	.lo_name = "rg",
803 };
804 
805 const struct gfs2_log_operations gfs2_databuf_lops = {
806 	.lo_add = databuf_lo_add,
807 	.lo_incore_commit = buf_lo_incore_commit,
808 	.lo_before_commit = databuf_lo_before_commit,
809 	.lo_after_commit = databuf_lo_after_commit,
810 	.lo_scan_elements = databuf_lo_scan_elements,
811 	.lo_after_scan = databuf_lo_after_scan,
812 	.lo_name = "databuf",
813 };
814 
815 const struct gfs2_log_operations *gfs2_log_ops[] = {
816 	&gfs2_glock_lops,
817 	&gfs2_buf_lops,
818 	&gfs2_revoke_lops,
819 	&gfs2_rg_lops,
820 	&gfs2_databuf_lops,
821 	NULL,
822 };
823 
824