Linux Audio

Check our new training course

Yocto distribution development and maintenance

Need a Yocto distribution for your embedded project?
Loading...
v4.17
  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/mempool.h>
 16#include <linux/gfs2_ondisk.h>
 17#include <linux/bio.h>
 18#include <linux/fs.h>
 19#include <linux/list_sort.h>
 20
 21#include "dir.h"
 22#include "gfs2.h"
 23#include "incore.h"
 24#include "inode.h"
 25#include "glock.h"
 26#include "log.h"
 27#include "lops.h"
 28#include "meta_io.h"
 29#include "recovery.h"
 30#include "rgrp.h"
 31#include "trans.h"
 32#include "util.h"
 33#include "trace_gfs2.h"
 34
 35/**
 36 * gfs2_pin - Pin a buffer in memory
 37 * @sdp: The superblock
 38 * @bh: The buffer to be pinned
 39 *
 40 * The log lock must be held when calling this function
 41 */
 42void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
 43{
 44	struct gfs2_bufdata *bd;
 45
 46	BUG_ON(!current->journal_info);
 47
 48	clear_buffer_dirty(bh);
 49	if (test_set_buffer_pinned(bh))
 50		gfs2_assert_withdraw(sdp, 0);
 51	if (!buffer_uptodate(bh))
 52		gfs2_io_error_bh(sdp, bh);
 53	bd = bh->b_private;
 54	/* If this buffer is in the AIL and it has already been written
 55	 * to in-place disk block, remove it from the AIL.
 56	 */
 57	spin_lock(&sdp->sd_ail_lock);
 58	if (bd->bd_tr)
 59		list_move(&bd->bd_ail_st_list, &bd->bd_tr->tr_ail2_list);
 60	spin_unlock(&sdp->sd_ail_lock);
 61	get_bh(bh);
 62	atomic_inc(&sdp->sd_log_pinned);
 63	trace_gfs2_pin(bd, 1);
 64}
 65
 66static bool buffer_is_rgrp(const struct gfs2_bufdata *bd)
 67{
 68	return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP;
 69}
 70
 71static void maybe_release_space(struct gfs2_bufdata *bd)
 72{
 73	struct gfs2_glock *gl = bd->bd_gl;
 74	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 75	struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
 76	unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number;
 77	struct gfs2_bitmap *bi = rgd->rd_bits + index;
 78
 79	if (bi->bi_clone == NULL)
 80		return;
 81	if (sdp->sd_args.ar_discard)
 82		gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
 83	memcpy(bi->bi_clone + bi->bi_offset,
 84	       bd->bd_bh->b_data + bi->bi_offset, bi->bi_len);
 85	clear_bit(GBF_FULL, &bi->bi_flags);
 86	rgd->rd_free_clone = rgd->rd_free;
 87	rgd->rd_extfail_pt = rgd->rd_free;
 88}
 89
 90/**
 91 * gfs2_unpin - Unpin a buffer
 92 * @sdp: the filesystem the buffer belongs to
 93 * @bh: The buffer to unpin
 94 * @ai:
 95 * @flags: The inode dirty flags
 96 *
 97 */
 98
 99static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
100		       struct gfs2_trans *tr)
101{
102	struct gfs2_bufdata *bd = bh->b_private;
103
104	BUG_ON(!buffer_uptodate(bh));
105	BUG_ON(!buffer_pinned(bh));
106
107	lock_buffer(bh);
108	mark_buffer_dirty(bh);
109	clear_buffer_pinned(bh);
110
111	if (buffer_is_rgrp(bd))
112		maybe_release_space(bd);
113
114	spin_lock(&sdp->sd_ail_lock);
115	if (bd->bd_tr) {
116		list_del(&bd->bd_ail_st_list);
117		brelse(bh);
118	} else {
119		struct gfs2_glock *gl = bd->bd_gl;
120		list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
121		atomic_inc(&gl->gl_ail_count);
122	}
123	bd->bd_tr = tr;
124	list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list);
125	spin_unlock(&sdp->sd_ail_lock);
126
127	clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
128	trace_gfs2_pin(bd, 0);
129	unlock_buffer(bh);
130	atomic_dec(&sdp->sd_log_pinned);
131}
132
133static void gfs2_log_incr_head(struct gfs2_sbd *sdp)
134{
135	BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
136	       (sdp->sd_log_flush_head != sdp->sd_log_head));
137
138	if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks)
139		sdp->sd_log_flush_head = 0;
 
 
140}
141
142u64 gfs2_log_bmap(struct gfs2_sbd *sdp)
143{
144	unsigned int lbn = sdp->sd_log_flush_head;
145	struct gfs2_journal_extent *je;
146	u64 block;
147
148	list_for_each_entry(je, &sdp->sd_jdesc->extent_list, list) {
149		if ((lbn >= je->lblock) && (lbn < (je->lblock + je->blocks))) {
150			block = je->dblock + lbn - je->lblock;
151			gfs2_log_incr_head(sdp);
152			return block;
153		}
154	}
155
156	return -1;
157}
158
159/**
160 * gfs2_end_log_write_bh - end log write of pagecache data with buffers
161 * @sdp: The superblock
162 * @bvec: The bio_vec
163 * @error: The i/o status
164 *
165 * This finds the relevant buffers and unlocks them and sets the
166 * error flag according to the status of the i/o request. This is
167 * used when the log is writing data which has an in-place version
168 * that is pinned in the pagecache.
169 */
170
171static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp, struct bio_vec *bvec,
172				  blk_status_t error)
173{
174	struct buffer_head *bh, *next;
175	struct page *page = bvec->bv_page;
176	unsigned size;
177
178	bh = page_buffers(page);
179	size = bvec->bv_len;
180	while (bh_offset(bh) < bvec->bv_offset)
181		bh = bh->b_this_page;
182	do {
183		if (error)
184			mark_buffer_write_io_error(bh);
185		unlock_buffer(bh);
186		next = bh->b_this_page;
187		size -= bh->b_size;
188		brelse(bh);
189		bh = next;
190	} while(bh && size);
191}
192
193/**
194 * gfs2_end_log_write - end of i/o to the log
195 * @bio: The bio
196 * @error: Status of i/o request
197 *
198 * Each bio_vec contains either data from the pagecache or data
199 * relating to the log itself. Here we iterate over the bio_vec
200 * array, processing both kinds of data.
201 *
202 */
203
204static void gfs2_end_log_write(struct bio *bio)
205{
206	struct gfs2_sbd *sdp = bio->bi_private;
207	struct bio_vec *bvec;
208	struct page *page;
209	int i;
210
211	if (bio->bi_status) {
212		fs_err(sdp, "Error %d writing to journal, jid=%u\n",
213		       bio->bi_status, sdp->sd_jdesc->jd_jid);
214		wake_up(&sdp->sd_logd_waitq);
215	}
216
217	bio_for_each_segment_all(bvec, bio, i) {
218		page = bvec->bv_page;
219		if (page_has_buffers(page))
220			gfs2_end_log_write_bh(sdp, bvec, bio->bi_status);
221		else
222			mempool_free(page, gfs2_page_pool);
223	}
224
225	bio_put(bio);
226	if (atomic_dec_and_test(&sdp->sd_log_in_flight))
227		wake_up(&sdp->sd_log_flush_wait);
228}
229
230/**
231 * gfs2_log_flush_bio - Submit any pending log bio
232 * @sdp: The superblock
233 * @op: REQ_OP
234 * @op_flags: req_flag_bits
235 *
236 * Submit any pending part-built or full bio to the block device. If
237 * there is no pending bio, then this is a no-op.
238 */
239
240void gfs2_log_flush_bio(struct gfs2_sbd *sdp, int op, int op_flags)
241{
242	if (sdp->sd_log_bio) {
243		atomic_inc(&sdp->sd_log_in_flight);
244		bio_set_op_attrs(sdp->sd_log_bio, op, op_flags);
245		submit_bio(sdp->sd_log_bio);
246		sdp->sd_log_bio = NULL;
247	}
248}
249
250/**
251 * gfs2_log_alloc_bio - Allocate a new bio for log writing
252 * @sdp: The superblock
253 * @blkno: The next device block number we want to write to
254 *
255 * This should never be called when there is a cached bio in the
256 * super block. When it returns, there will be a cached bio in the
257 * super block which will have as many bio_vecs as the device is
258 * happy to handle.
259 *
260 * Returns: Newly allocated bio
261 */
262
263static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno)
264{
265	struct super_block *sb = sdp->sd_vfs;
266	struct bio *bio;
267
268	BUG_ON(sdp->sd_log_bio);
269
270	bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
271	bio->bi_iter.bi_sector = blkno * (sb->s_blocksize >> 9);
272	bio_set_dev(bio, sb->s_bdev);
273	bio->bi_end_io = gfs2_end_log_write;
274	bio->bi_private = sdp;
275
276	sdp->sd_log_bio = bio;
277
278	return bio;
279}
280
281/**
282 * gfs2_log_get_bio - Get cached log bio, or allocate a new one
283 * @sdp: The superblock
284 * @blkno: The device block number we want to write to
285 *
286 * If there is a cached bio, then if the next block number is sequential
287 * with the previous one, return it, otherwise flush the bio to the
288 * device. If there is not a cached bio, or we just flushed it, then
289 * allocate a new one.
290 *
291 * Returns: The bio to use for log writes
292 */
293
294static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno)
295{
296	struct bio *bio = sdp->sd_log_bio;
297	u64 nblk;
298
299	if (bio) {
300		nblk = bio_end_sector(bio);
301		nblk >>= sdp->sd_fsb2bb_shift;
302		if (blkno == nblk)
303			return bio;
304		gfs2_log_flush_bio(sdp, REQ_OP_WRITE, 0);
305	}
306
307	return gfs2_log_alloc_bio(sdp, blkno);
308}
309
 
310/**
311 * gfs2_log_write - write to log
312 * @sdp: the filesystem
313 * @page: the page to write
314 * @size: the size of the data to write
315 * @offset: the offset within the page 
316 * @blkno: block number of the log entry
317 *
318 * Try and add the page segment to the current bio. If that fails,
319 * submit the current bio to the device and create a new one, and
320 * then add the page segment to that.
321 */
322
323void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page,
324		    unsigned size, unsigned offset, u64 blkno)
325{
 
326	struct bio *bio;
327	int ret;
328
329	bio = gfs2_log_get_bio(sdp, blkno);
330	ret = bio_add_page(bio, page, size, offset);
331	if (ret == 0) {
332		gfs2_log_flush_bio(sdp, REQ_OP_WRITE, 0);
333		bio = gfs2_log_alloc_bio(sdp, blkno);
334		ret = bio_add_page(bio, page, size, offset);
335		WARN_ON(ret == 0);
336	}
337}
338
339/**
340 * gfs2_log_write_bh - write a buffer's content to the log
341 * @sdp: The super block
342 * @bh: The buffer pointing to the in-place location
343 * 
344 * This writes the content of the buffer to the next available location
345 * in the log. The buffer will be unlocked once the i/o to the log has
346 * completed.
347 */
348
349static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
350{
351	gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh),
352		       gfs2_log_bmap(sdp));
353}
354
355/**
356 * gfs2_log_write_page - write one block stored in a page, into the log
357 * @sdp: The superblock
358 * @page: The struct page
359 *
360 * This writes the first block-sized part of the page into the log. Note
361 * that the page must have been allocated from the gfs2_page_pool mempool
362 * and that after this has been called, ownership has been transferred and
363 * the page may be freed at any time.
364 */
365
366void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
367{
368	struct super_block *sb = sdp->sd_vfs;
369	gfs2_log_write(sdp, page, sb->s_blocksize, 0,
370		       gfs2_log_bmap(sdp));
371}
372
373static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
374				      u32 ld_length, u32 ld_data1)
375{
376	struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
377	struct gfs2_log_descriptor *ld = page_address(page);
378	clear_page(ld);
379	ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
380	ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
381	ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
382	ld->ld_type = cpu_to_be32(ld_type);
383	ld->ld_length = cpu_to_be32(ld_length);
384	ld->ld_data1 = cpu_to_be32(ld_data1);
385	ld->ld_data2 = 0;
386	return page;
387}
388
389static void gfs2_check_magic(struct buffer_head *bh)
390{
391	void *kaddr;
392	__be32 *ptr;
393
394	clear_buffer_escaped(bh);
395	kaddr = kmap_atomic(bh->b_page);
396	ptr = kaddr + bh_offset(bh);
397	if (*ptr == cpu_to_be32(GFS2_MAGIC))
398		set_buffer_escaped(bh);
399	kunmap_atomic(kaddr);
400}
401
402static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b)
403{
404	struct gfs2_bufdata *bda, *bdb;
405
406	bda = list_entry(a, struct gfs2_bufdata, bd_list);
407	bdb = list_entry(b, struct gfs2_bufdata, bd_list);
408
409	if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
410		return -1;
411	if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
412		return 1;
413	return 0;
414}
415
416static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
417				unsigned int total, struct list_head *blist,
418				bool is_databuf)
419{
420	struct gfs2_log_descriptor *ld;
421	struct gfs2_bufdata *bd1 = NULL, *bd2;
422	struct page *page;
423	unsigned int num;
424	unsigned n;
425	__be64 *ptr;
426
427	gfs2_log_lock(sdp);
428	list_sort(NULL, blist, blocknr_cmp);
429	bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
430	while(total) {
431		num = total;
432		if (total > limit)
433			num = limit;
434		gfs2_log_unlock(sdp);
435		page = gfs2_get_log_desc(sdp,
436					 is_databuf ? GFS2_LOG_DESC_JDATA :
437					 GFS2_LOG_DESC_METADATA, num + 1, num);
438		ld = page_address(page);
439		gfs2_log_lock(sdp);
440		ptr = (__be64 *)(ld + 1);
441
442		n = 0;
443		list_for_each_entry_continue(bd1, blist, bd_list) {
444			*ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
445			if (is_databuf) {
446				gfs2_check_magic(bd1->bd_bh);
447				*ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
448			}
449			if (++n >= num)
450				break;
451		}
452
453		gfs2_log_unlock(sdp);
454		gfs2_log_write_page(sdp, page);
455		gfs2_log_lock(sdp);
456
457		n = 0;
458		list_for_each_entry_continue(bd2, blist, bd_list) {
459			get_bh(bd2->bd_bh);
460			gfs2_log_unlock(sdp);
461			lock_buffer(bd2->bd_bh);
462
463			if (buffer_escaped(bd2->bd_bh)) {
464				void *kaddr;
465				page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
466				ptr = page_address(page);
467				kaddr = kmap_atomic(bd2->bd_bh->b_page);
468				memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
469				       bd2->bd_bh->b_size);
470				kunmap_atomic(kaddr);
471				*(__be32 *)ptr = 0;
472				clear_buffer_escaped(bd2->bd_bh);
473				unlock_buffer(bd2->bd_bh);
474				brelse(bd2->bd_bh);
475				gfs2_log_write_page(sdp, page);
476			} else {
477				gfs2_log_write_bh(sdp, bd2->bd_bh);
478			}
479			gfs2_log_lock(sdp);
480			if (++n >= num)
481				break;
482		}
483
484		BUG_ON(total < num);
485		total -= num;
486	}
487	gfs2_log_unlock(sdp);
488}
489
490static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
491{
492	unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
493	unsigned int nbuf;
494	if (tr == NULL)
495		return;
496	nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
497	gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
498}
499
500static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
501{
502	struct list_head *head;
503	struct gfs2_bufdata *bd;
504
505	if (tr == NULL)
506		return;
507
508	head = &tr->tr_buf;
509	while (!list_empty(head)) {
510		bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
511		list_del_init(&bd->bd_list);
512		gfs2_unpin(sdp, bd->bd_bh, tr);
513	}
514}
515
516static void buf_lo_before_scan(struct gfs2_jdesc *jd,
517			       struct gfs2_log_header_host *head, int pass)
518{
519	if (pass != 0)
520		return;
521
522	jd->jd_found_blocks = 0;
523	jd->jd_replayed_blocks = 0;
524}
525
526static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
527				struct gfs2_log_descriptor *ld, __be64 *ptr,
528				int pass)
529{
530	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
531	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
532	struct gfs2_glock *gl = ip->i_gl;
533	unsigned int blks = be32_to_cpu(ld->ld_data1);
534	struct buffer_head *bh_log, *bh_ip;
535	u64 blkno;
536	int error = 0;
537
538	if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
539		return 0;
540
541	gfs2_replay_incr_blk(jd, &start);
542
543	for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
544		blkno = be64_to_cpu(*ptr++);
545
546		jd->jd_found_blocks++;
547
548		if (gfs2_revoke_check(jd, blkno, start))
549			continue;
550
551		error = gfs2_replay_read_block(jd, start, &bh_log);
552		if (error)
553			return error;
554
555		bh_ip = gfs2_meta_new(gl, blkno);
556		memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
557
558		if (gfs2_meta_check(sdp, bh_ip))
559			error = -EIO;
560		else
561			mark_buffer_dirty(bh_ip);
562
563		brelse(bh_log);
564		brelse(bh_ip);
565
566		if (error)
567			break;
568
569		jd->jd_replayed_blocks++;
570	}
571
572	return error;
573}
574
575/**
576 * gfs2_meta_sync - Sync all buffers associated with a glock
577 * @gl: The glock
578 *
579 */
580
581static void gfs2_meta_sync(struct gfs2_glock *gl)
582{
583	struct address_space *mapping = gfs2_glock2aspace(gl);
584	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
585	int error;
586
587	if (mapping == NULL)
588		mapping = &sdp->sd_aspace;
589
590	filemap_fdatawrite(mapping);
591	error = filemap_fdatawait(mapping);
592
593	if (error)
594		gfs2_io_error(gl->gl_name.ln_sbd);
595}
596
597static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
598{
599	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
600	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
601
602	if (error) {
603		gfs2_meta_sync(ip->i_gl);
604		return;
605	}
606	if (pass != 1)
607		return;
608
609	gfs2_meta_sync(ip->i_gl);
610
611	fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
612	        jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
613}
614
615static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
616{
617	struct gfs2_meta_header *mh;
618	unsigned int offset;
619	struct list_head *head = &sdp->sd_log_le_revoke;
620	struct gfs2_bufdata *bd;
621	struct page *page;
622	unsigned int length;
623
624	gfs2_write_revokes(sdp);
625	if (!sdp->sd_log_num_revoke)
626		return;
627
628	length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
629	page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
630	offset = sizeof(struct gfs2_log_descriptor);
631
632	list_for_each_entry(bd, head, bd_list) {
633		sdp->sd_log_num_revoke--;
634
635		if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
636
637			gfs2_log_write_page(sdp, page);
638			page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
639			mh = page_address(page);
640			clear_page(mh);
641			mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
642			mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
643			mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
644			offset = sizeof(struct gfs2_meta_header);
645		}
646
647		*(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
648		offset += sizeof(u64);
649	}
650	gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
651
652	gfs2_log_write_page(sdp, page);
653}
654
655static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
656{
657	struct list_head *head = &sdp->sd_log_le_revoke;
658	struct gfs2_bufdata *bd;
659	struct gfs2_glock *gl;
660
661	while (!list_empty(head)) {
662		bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
663		list_del_init(&bd->bd_list);
664		gl = bd->bd_gl;
665		atomic_dec(&gl->gl_revokes);
666		clear_bit(GLF_LFLUSH, &gl->gl_flags);
667		kmem_cache_free(gfs2_bufdata_cachep, bd);
668	}
669}
670
671static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
672				  struct gfs2_log_header_host *head, int pass)
673{
674	if (pass != 0)
675		return;
676
677	jd->jd_found_revokes = 0;
678	jd->jd_replay_tail = head->lh_tail;
679}
680
681static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
682				   struct gfs2_log_descriptor *ld, __be64 *ptr,
683				   int pass)
684{
685	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
686	unsigned int blks = be32_to_cpu(ld->ld_length);
687	unsigned int revokes = be32_to_cpu(ld->ld_data1);
688	struct buffer_head *bh;
689	unsigned int offset;
690	u64 blkno;
691	int first = 1;
692	int error;
693
694	if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
695		return 0;
696
697	offset = sizeof(struct gfs2_log_descriptor);
698
699	for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
700		error = gfs2_replay_read_block(jd, start, &bh);
701		if (error)
702			return error;
703
704		if (!first)
705			gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
706
707		while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
708			blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
709
710			error = gfs2_revoke_add(jd, blkno, start);
711			if (error < 0) {
712				brelse(bh);
713				return error;
714			}
715			else if (error)
716				jd->jd_found_revokes++;
717
718			if (!--revokes)
719				break;
720			offset += sizeof(u64);
721		}
722
723		brelse(bh);
724		offset = sizeof(struct gfs2_meta_header);
725		first = 0;
726	}
727
728	return 0;
729}
730
731static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
732{
733	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
734
735	if (error) {
736		gfs2_revoke_clean(jd);
737		return;
738	}
739	if (pass != 1)
740		return;
741
742	fs_info(sdp, "jid=%u: Found %u revoke tags\n",
743	        jd->jd_jid, jd->jd_found_revokes);
744
745	gfs2_revoke_clean(jd);
746}
747
748/**
749 * databuf_lo_before_commit - Scan the data buffers, writing as we go
750 *
751 */
752
753static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
754{
755	unsigned int limit = databuf_limit(sdp);
756	unsigned int nbuf;
757	if (tr == NULL)
758		return;
759	nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
760	gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
761}
762
763static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
764				    struct gfs2_log_descriptor *ld,
765				    __be64 *ptr, int pass)
766{
767	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
768	struct gfs2_glock *gl = ip->i_gl;
769	unsigned int blks = be32_to_cpu(ld->ld_data1);
770	struct buffer_head *bh_log, *bh_ip;
771	u64 blkno;
772	u64 esc;
773	int error = 0;
774
775	if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
776		return 0;
777
778	gfs2_replay_incr_blk(jd, &start);
779	for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
780		blkno = be64_to_cpu(*ptr++);
781		esc = be64_to_cpu(*ptr++);
782
783		jd->jd_found_blocks++;
784
785		if (gfs2_revoke_check(jd, blkno, start))
786			continue;
787
788		error = gfs2_replay_read_block(jd, start, &bh_log);
789		if (error)
790			return error;
791
792		bh_ip = gfs2_meta_new(gl, blkno);
793		memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
794
795		/* Unescape */
796		if (esc) {
797			__be32 *eptr = (__be32 *)bh_ip->b_data;
798			*eptr = cpu_to_be32(GFS2_MAGIC);
799		}
800		mark_buffer_dirty(bh_ip);
801
802		brelse(bh_log);
803		brelse(bh_ip);
804
805		jd->jd_replayed_blocks++;
806	}
807
808	return error;
809}
810
811/* FIXME: sort out accounting for log blocks etc. */
812
813static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
814{
815	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
816	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
817
818	if (error) {
819		gfs2_meta_sync(ip->i_gl);
820		return;
821	}
822	if (pass != 1)
823		return;
824
825	/* data sync? */
826	gfs2_meta_sync(ip->i_gl);
827
828	fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
829		jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
830}
831
832static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
833{
834	struct list_head *head;
835	struct gfs2_bufdata *bd;
836
837	if (tr == NULL)
838		return;
839
840	head = &tr->tr_databuf;
841	while (!list_empty(head)) {
842		bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
843		list_del_init(&bd->bd_list);
844		gfs2_unpin(sdp, bd->bd_bh, tr);
845	}
846}
847
848
849const struct gfs2_log_operations gfs2_buf_lops = {
850	.lo_before_commit = buf_lo_before_commit,
851	.lo_after_commit = buf_lo_after_commit,
852	.lo_before_scan = buf_lo_before_scan,
853	.lo_scan_elements = buf_lo_scan_elements,
854	.lo_after_scan = buf_lo_after_scan,
855	.lo_name = "buf",
856};
857
858const struct gfs2_log_operations gfs2_revoke_lops = {
859	.lo_before_commit = revoke_lo_before_commit,
860	.lo_after_commit = revoke_lo_after_commit,
861	.lo_before_scan = revoke_lo_before_scan,
862	.lo_scan_elements = revoke_lo_scan_elements,
863	.lo_after_scan = revoke_lo_after_scan,
864	.lo_name = "revoke",
865};
866
867const struct gfs2_log_operations gfs2_databuf_lops = {
868	.lo_before_commit = databuf_lo_before_commit,
869	.lo_after_commit = databuf_lo_after_commit,
870	.lo_scan_elements = databuf_lo_scan_elements,
871	.lo_after_scan = databuf_lo_after_scan,
872	.lo_name = "databuf",
873};
874
875const struct gfs2_log_operations *gfs2_log_ops[] = {
876	&gfs2_databuf_lops,
877	&gfs2_buf_lops,
878	&gfs2_revoke_lops,
879	NULL,
880};
881
v4.10.11
  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/mempool.h>
 16#include <linux/gfs2_ondisk.h>
 17#include <linux/bio.h>
 18#include <linux/fs.h>
 19#include <linux/list_sort.h>
 20
 
 21#include "gfs2.h"
 22#include "incore.h"
 23#include "inode.h"
 24#include "glock.h"
 25#include "log.h"
 26#include "lops.h"
 27#include "meta_io.h"
 28#include "recovery.h"
 29#include "rgrp.h"
 30#include "trans.h"
 31#include "util.h"
 32#include "trace_gfs2.h"
 33
 34/**
 35 * gfs2_pin - Pin a buffer in memory
 36 * @sdp: The superblock
 37 * @bh: The buffer to be pinned
 38 *
 39 * The log lock must be held when calling this function
 40 */
 41void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
 42{
 43	struct gfs2_bufdata *bd;
 44
 45	BUG_ON(!current->journal_info);
 46
 47	clear_buffer_dirty(bh);
 48	if (test_set_buffer_pinned(bh))
 49		gfs2_assert_withdraw(sdp, 0);
 50	if (!buffer_uptodate(bh))
 51		gfs2_io_error_bh(sdp, bh);
 52	bd = bh->b_private;
 53	/* If this buffer is in the AIL and it has already been written
 54	 * to in-place disk block, remove it from the AIL.
 55	 */
 56	spin_lock(&sdp->sd_ail_lock);
 57	if (bd->bd_tr)
 58		list_move(&bd->bd_ail_st_list, &bd->bd_tr->tr_ail2_list);
 59	spin_unlock(&sdp->sd_ail_lock);
 60	get_bh(bh);
 61	atomic_inc(&sdp->sd_log_pinned);
 62	trace_gfs2_pin(bd, 1);
 63}
 64
 65static bool buffer_is_rgrp(const struct gfs2_bufdata *bd)
 66{
 67	return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP;
 68}
 69
 70static void maybe_release_space(struct gfs2_bufdata *bd)
 71{
 72	struct gfs2_glock *gl = bd->bd_gl;
 73	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 74	struct gfs2_rgrpd *rgd = gl->gl_object;
 75	unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number;
 76	struct gfs2_bitmap *bi = rgd->rd_bits + index;
 77
 78	if (bi->bi_clone == NULL)
 79		return;
 80	if (sdp->sd_args.ar_discard)
 81		gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
 82	memcpy(bi->bi_clone + bi->bi_offset,
 83	       bd->bd_bh->b_data + bi->bi_offset, bi->bi_len);
 84	clear_bit(GBF_FULL, &bi->bi_flags);
 85	rgd->rd_free_clone = rgd->rd_free;
 86	rgd->rd_extfail_pt = rgd->rd_free;
 87}
 88
 89/**
 90 * gfs2_unpin - Unpin a buffer
 91 * @sdp: the filesystem the buffer belongs to
 92 * @bh: The buffer to unpin
 93 * @ai:
 94 * @flags: The inode dirty flags
 95 *
 96 */
 97
 98static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
 99		       struct gfs2_trans *tr)
100{
101	struct gfs2_bufdata *bd = bh->b_private;
102
103	BUG_ON(!buffer_uptodate(bh));
104	BUG_ON(!buffer_pinned(bh));
105
106	lock_buffer(bh);
107	mark_buffer_dirty(bh);
108	clear_buffer_pinned(bh);
109
110	if (buffer_is_rgrp(bd))
111		maybe_release_space(bd);
112
113	spin_lock(&sdp->sd_ail_lock);
114	if (bd->bd_tr) {
115		list_del(&bd->bd_ail_st_list);
116		brelse(bh);
117	} else {
118		struct gfs2_glock *gl = bd->bd_gl;
119		list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
120		atomic_inc(&gl->gl_ail_count);
121	}
122	bd->bd_tr = tr;
123	list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list);
124	spin_unlock(&sdp->sd_ail_lock);
125
126	clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
127	trace_gfs2_pin(bd, 0);
128	unlock_buffer(bh);
129	atomic_dec(&sdp->sd_log_pinned);
130}
131
132static void gfs2_log_incr_head(struct gfs2_sbd *sdp)
133{
134	BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
135	       (sdp->sd_log_flush_head != sdp->sd_log_head));
136
137	if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
138		sdp->sd_log_flush_head = 0;
139		sdp->sd_log_flush_wrapped = 1;
140	}
141}
142
143static u64 gfs2_log_bmap(struct gfs2_sbd *sdp)
144{
145	unsigned int lbn = sdp->sd_log_flush_head;
146	struct gfs2_journal_extent *je;
147	u64 block;
148
149	list_for_each_entry(je, &sdp->sd_jdesc->extent_list, list) {
150		if ((lbn >= je->lblock) && (lbn < (je->lblock + je->blocks))) {
151			block = je->dblock + lbn - je->lblock;
152			gfs2_log_incr_head(sdp);
153			return block;
154		}
155	}
156
157	return -1;
158}
159
160/**
161 * gfs2_end_log_write_bh - end log write of pagecache data with buffers
162 * @sdp: The superblock
163 * @bvec: The bio_vec
164 * @error: The i/o status
165 *
166 * This finds the relavent buffers and unlocks then and sets the
167 * error flag according to the status of the i/o request. This is
168 * used when the log is writing data which has an in-place version
169 * that is pinned in the pagecache.
170 */
171
172static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp, struct bio_vec *bvec,
173				  int error)
174{
175	struct buffer_head *bh, *next;
176	struct page *page = bvec->bv_page;
177	unsigned size;
178
179	bh = page_buffers(page);
180	size = bvec->bv_len;
181	while (bh_offset(bh) < bvec->bv_offset)
182		bh = bh->b_this_page;
183	do {
184		if (error)
185			set_buffer_write_io_error(bh);
186		unlock_buffer(bh);
187		next = bh->b_this_page;
188		size -= bh->b_size;
189		brelse(bh);
190		bh = next;
191	} while(bh && size);
192}
193
194/**
195 * gfs2_end_log_write - end of i/o to the log
196 * @bio: The bio
197 * @error: Status of i/o request
198 *
199 * Each bio_vec contains either data from the pagecache or data
200 * relating to the log itself. Here we iterate over the bio_vec
201 * array, processing both kinds of data.
202 *
203 */
204
205static void gfs2_end_log_write(struct bio *bio)
206{
207	struct gfs2_sbd *sdp = bio->bi_private;
208	struct bio_vec *bvec;
209	struct page *page;
210	int i;
211
212	if (bio->bi_error) {
213		sdp->sd_log_error = bio->bi_error;
214		fs_err(sdp, "Error %d writing to log\n", bio->bi_error);
 
215	}
216
217	bio_for_each_segment_all(bvec, bio, i) {
218		page = bvec->bv_page;
219		if (page_has_buffers(page))
220			gfs2_end_log_write_bh(sdp, bvec, bio->bi_error);
221		else
222			mempool_free(page, gfs2_page_pool);
223	}
224
225	bio_put(bio);
226	if (atomic_dec_and_test(&sdp->sd_log_in_flight))
227		wake_up(&sdp->sd_log_flush_wait);
228}
229
230/**
231 * gfs2_log_flush_bio - Submit any pending log bio
232 * @sdp: The superblock
233 * @op: REQ_OP
234 * @op_flags: req_flag_bits
235 *
236 * Submit any pending part-built or full bio to the block device. If
237 * there is no pending bio, then this is a no-op.
238 */
239
240void gfs2_log_flush_bio(struct gfs2_sbd *sdp, int op, int op_flags)
241{
242	if (sdp->sd_log_bio) {
243		atomic_inc(&sdp->sd_log_in_flight);
244		bio_set_op_attrs(sdp->sd_log_bio, op, op_flags);
245		submit_bio(sdp->sd_log_bio);
246		sdp->sd_log_bio = NULL;
247	}
248}
249
250/**
251 * gfs2_log_alloc_bio - Allocate a new bio for log writing
252 * @sdp: The superblock
253 * @blkno: The next device block number we want to write to
254 *
255 * This should never be called when there is a cached bio in the
256 * super block. When it returns, there will be a cached bio in the
257 * super block which will have as many bio_vecs as the device is
258 * happy to handle.
259 *
260 * Returns: Newly allocated bio
261 */
262
263static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno)
264{
265	struct super_block *sb = sdp->sd_vfs;
266	struct bio *bio;
267
268	BUG_ON(sdp->sd_log_bio);
269
270	bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
271	bio->bi_iter.bi_sector = blkno * (sb->s_blocksize >> 9);
272	bio->bi_bdev = sb->s_bdev;
273	bio->bi_end_io = gfs2_end_log_write;
274	bio->bi_private = sdp;
275
276	sdp->sd_log_bio = bio;
277
278	return bio;
279}
280
281/**
282 * gfs2_log_get_bio - Get cached log bio, or allocate a new one
283 * @sdp: The superblock
284 * @blkno: The device block number we want to write to
285 *
286 * If there is a cached bio, then if the next block number is sequential
287 * with the previous one, return it, otherwise flush the bio to the
288 * device. If there is not a cached bio, or we just flushed it, then
289 * allocate a new one.
290 *
291 * Returns: The bio to use for log writes
292 */
293
294static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno)
295{
296	struct bio *bio = sdp->sd_log_bio;
297	u64 nblk;
298
299	if (bio) {
300		nblk = bio_end_sector(bio);
301		nblk >>= sdp->sd_fsb2bb_shift;
302		if (blkno == nblk)
303			return bio;
304		gfs2_log_flush_bio(sdp, REQ_OP_WRITE, 0);
305	}
306
307	return gfs2_log_alloc_bio(sdp, blkno);
308}
309
310
311/**
312 * gfs2_log_write - write to log
313 * @sdp: the filesystem
314 * @page: the page to write
315 * @size: the size of the data to write
316 * @offset: the offset within the page 
 
317 *
318 * Try and add the page segment to the current bio. If that fails,
319 * submit the current bio to the device and create a new one, and
320 * then add the page segment to that.
321 */
322
323static void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page,
324			   unsigned size, unsigned offset)
325{
326	u64 blkno = gfs2_log_bmap(sdp);
327	struct bio *bio;
328	int ret;
329
330	bio = gfs2_log_get_bio(sdp, blkno);
331	ret = bio_add_page(bio, page, size, offset);
332	if (ret == 0) {
333		gfs2_log_flush_bio(sdp, REQ_OP_WRITE, 0);
334		bio = gfs2_log_alloc_bio(sdp, blkno);
335		ret = bio_add_page(bio, page, size, offset);
336		WARN_ON(ret == 0);
337	}
338}
339
340/**
341 * gfs2_log_write_bh - write a buffer's content to the log
342 * @sdp: The super block
343 * @bh: The buffer pointing to the in-place location
344 * 
345 * This writes the content of the buffer to the next available location
346 * in the log. The buffer will be unlocked once the i/o to the log has
347 * completed.
348 */
349
350static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
351{
352	gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh));
 
353}
354
355/**
356 * gfs2_log_write_page - write one block stored in a page, into the log
357 * @sdp: The superblock
358 * @page: The struct page
359 *
360 * This writes the first block-sized part of the page into the log. Note
361 * that the page must have been allocated from the gfs2_page_pool mempool
362 * and that after this has been called, ownership has been transferred and
363 * the page may be freed at any time.
364 */
365
366void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
367{
368	struct super_block *sb = sdp->sd_vfs;
369	gfs2_log_write(sdp, page, sb->s_blocksize, 0);
 
370}
371
372static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
373				      u32 ld_length, u32 ld_data1)
374{
375	struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
376	struct gfs2_log_descriptor *ld = page_address(page);
377	clear_page(ld);
378	ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
379	ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
380	ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
381	ld->ld_type = cpu_to_be32(ld_type);
382	ld->ld_length = cpu_to_be32(ld_length);
383	ld->ld_data1 = cpu_to_be32(ld_data1);
384	ld->ld_data2 = 0;
385	return page;
386}
387
388static void gfs2_check_magic(struct buffer_head *bh)
389{
390	void *kaddr;
391	__be32 *ptr;
392
393	clear_buffer_escaped(bh);
394	kaddr = kmap_atomic(bh->b_page);
395	ptr = kaddr + bh_offset(bh);
396	if (*ptr == cpu_to_be32(GFS2_MAGIC))
397		set_buffer_escaped(bh);
398	kunmap_atomic(kaddr);
399}
400
401static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b)
402{
403	struct gfs2_bufdata *bda, *bdb;
404
405	bda = list_entry(a, struct gfs2_bufdata, bd_list);
406	bdb = list_entry(b, struct gfs2_bufdata, bd_list);
407
408	if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
409		return -1;
410	if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
411		return 1;
412	return 0;
413}
414
415static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
416				unsigned int total, struct list_head *blist,
417				bool is_databuf)
418{
419	struct gfs2_log_descriptor *ld;
420	struct gfs2_bufdata *bd1 = NULL, *bd2;
421	struct page *page;
422	unsigned int num;
423	unsigned n;
424	__be64 *ptr;
425
426	gfs2_log_lock(sdp);
427	list_sort(NULL, blist, blocknr_cmp);
428	bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
429	while(total) {
430		num = total;
431		if (total > limit)
432			num = limit;
433		gfs2_log_unlock(sdp);
434		page = gfs2_get_log_desc(sdp,
435					 is_databuf ? GFS2_LOG_DESC_JDATA :
436					 GFS2_LOG_DESC_METADATA, num + 1, num);
437		ld = page_address(page);
438		gfs2_log_lock(sdp);
439		ptr = (__be64 *)(ld + 1);
440
441		n = 0;
442		list_for_each_entry_continue(bd1, blist, bd_list) {
443			*ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
444			if (is_databuf) {
445				gfs2_check_magic(bd1->bd_bh);
446				*ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
447			}
448			if (++n >= num)
449				break;
450		}
451
452		gfs2_log_unlock(sdp);
453		gfs2_log_write_page(sdp, page);
454		gfs2_log_lock(sdp);
455
456		n = 0;
457		list_for_each_entry_continue(bd2, blist, bd_list) {
458			get_bh(bd2->bd_bh);
459			gfs2_log_unlock(sdp);
460			lock_buffer(bd2->bd_bh);
461
462			if (buffer_escaped(bd2->bd_bh)) {
463				void *kaddr;
464				page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
465				ptr = page_address(page);
466				kaddr = kmap_atomic(bd2->bd_bh->b_page);
467				memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
468				       bd2->bd_bh->b_size);
469				kunmap_atomic(kaddr);
470				*(__be32 *)ptr = 0;
471				clear_buffer_escaped(bd2->bd_bh);
472				unlock_buffer(bd2->bd_bh);
473				brelse(bd2->bd_bh);
474				gfs2_log_write_page(sdp, page);
475			} else {
476				gfs2_log_write_bh(sdp, bd2->bd_bh);
477			}
478			gfs2_log_lock(sdp);
479			if (++n >= num)
480				break;
481		}
482
483		BUG_ON(total < num);
484		total -= num;
485	}
486	gfs2_log_unlock(sdp);
487}
488
489static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
490{
491	unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
492	unsigned int nbuf;
493	if (tr == NULL)
494		return;
495	nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
496	gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
497}
498
499static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
500{
501	struct list_head *head;
502	struct gfs2_bufdata *bd;
503
504	if (tr == NULL)
505		return;
506
507	head = &tr->tr_buf;
508	while (!list_empty(head)) {
509		bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
510		list_del_init(&bd->bd_list);
511		gfs2_unpin(sdp, bd->bd_bh, tr);
512	}
513}
514
515static void buf_lo_before_scan(struct gfs2_jdesc *jd,
516			       struct gfs2_log_header_host *head, int pass)
517{
518	if (pass != 0)
519		return;
520
521	jd->jd_found_blocks = 0;
522	jd->jd_replayed_blocks = 0;
523}
524
525static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
526				struct gfs2_log_descriptor *ld, __be64 *ptr,
527				int pass)
528{
529	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
530	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
531	struct gfs2_glock *gl = ip->i_gl;
532	unsigned int blks = be32_to_cpu(ld->ld_data1);
533	struct buffer_head *bh_log, *bh_ip;
534	u64 blkno;
535	int error = 0;
536
537	if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
538		return 0;
539
540	gfs2_replay_incr_blk(jd, &start);
541
542	for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
543		blkno = be64_to_cpu(*ptr++);
544
545		jd->jd_found_blocks++;
546
547		if (gfs2_revoke_check(jd, blkno, start))
548			continue;
549
550		error = gfs2_replay_read_block(jd, start, &bh_log);
551		if (error)
552			return error;
553
554		bh_ip = gfs2_meta_new(gl, blkno);
555		memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
556
557		if (gfs2_meta_check(sdp, bh_ip))
558			error = -EIO;
559		else
560			mark_buffer_dirty(bh_ip);
561
562		brelse(bh_log);
563		brelse(bh_ip);
564
565		if (error)
566			break;
567
568		jd->jd_replayed_blocks++;
569	}
570
571	return error;
572}
573
574/**
575 * gfs2_meta_sync - Sync all buffers associated with a glock
576 * @gl: The glock
577 *
578 */
579
580static void gfs2_meta_sync(struct gfs2_glock *gl)
581{
582	struct address_space *mapping = gfs2_glock2aspace(gl);
583	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
584	int error;
585
586	if (mapping == NULL)
587		mapping = &sdp->sd_aspace;
588
589	filemap_fdatawrite(mapping);
590	error = filemap_fdatawait(mapping);
591
592	if (error)
593		gfs2_io_error(gl->gl_name.ln_sbd);
594}
595
596static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
597{
598	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
599	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
600
601	if (error) {
602		gfs2_meta_sync(ip->i_gl);
603		return;
604	}
605	if (pass != 1)
606		return;
607
608	gfs2_meta_sync(ip->i_gl);
609
610	fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
611	        jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
612}
613
614static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
615{
616	struct gfs2_meta_header *mh;
617	unsigned int offset;
618	struct list_head *head = &sdp->sd_log_le_revoke;
619	struct gfs2_bufdata *bd;
620	struct page *page;
621	unsigned int length;
622
623	gfs2_write_revokes(sdp);
624	if (!sdp->sd_log_num_revoke)
625		return;
626
627	length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
628	page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
629	offset = sizeof(struct gfs2_log_descriptor);
630
631	list_for_each_entry(bd, head, bd_list) {
632		sdp->sd_log_num_revoke--;
633
634		if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
635
636			gfs2_log_write_page(sdp, page);
637			page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
638			mh = page_address(page);
639			clear_page(mh);
640			mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
641			mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
642			mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
643			offset = sizeof(struct gfs2_meta_header);
644		}
645
646		*(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
647		offset += sizeof(u64);
648	}
649	gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
650
651	gfs2_log_write_page(sdp, page);
652}
653
654static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
655{
656	struct list_head *head = &sdp->sd_log_le_revoke;
657	struct gfs2_bufdata *bd;
658	struct gfs2_glock *gl;
659
660	while (!list_empty(head)) {
661		bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
662		list_del_init(&bd->bd_list);
663		gl = bd->bd_gl;
664		atomic_dec(&gl->gl_revokes);
665		clear_bit(GLF_LFLUSH, &gl->gl_flags);
666		kmem_cache_free(gfs2_bufdata_cachep, bd);
667	}
668}
669
670static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
671				  struct gfs2_log_header_host *head, int pass)
672{
673	if (pass != 0)
674		return;
675
676	jd->jd_found_revokes = 0;
677	jd->jd_replay_tail = head->lh_tail;
678}
679
680static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
681				   struct gfs2_log_descriptor *ld, __be64 *ptr,
682				   int pass)
683{
684	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
685	unsigned int blks = be32_to_cpu(ld->ld_length);
686	unsigned int revokes = be32_to_cpu(ld->ld_data1);
687	struct buffer_head *bh;
688	unsigned int offset;
689	u64 blkno;
690	int first = 1;
691	int error;
692
693	if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
694		return 0;
695
696	offset = sizeof(struct gfs2_log_descriptor);
697
698	for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
699		error = gfs2_replay_read_block(jd, start, &bh);
700		if (error)
701			return error;
702
703		if (!first)
704			gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
705
706		while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
707			blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
708
709			error = gfs2_revoke_add(jd, blkno, start);
710			if (error < 0) {
711				brelse(bh);
712				return error;
713			}
714			else if (error)
715				jd->jd_found_revokes++;
716
717			if (!--revokes)
718				break;
719			offset += sizeof(u64);
720		}
721
722		brelse(bh);
723		offset = sizeof(struct gfs2_meta_header);
724		first = 0;
725	}
726
727	return 0;
728}
729
730static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
731{
732	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
733
734	if (error) {
735		gfs2_revoke_clean(jd);
736		return;
737	}
738	if (pass != 1)
739		return;
740
741	fs_info(sdp, "jid=%u: Found %u revoke tags\n",
742	        jd->jd_jid, jd->jd_found_revokes);
743
744	gfs2_revoke_clean(jd);
745}
746
747/**
748 * databuf_lo_before_commit - Scan the data buffers, writing as we go
749 *
750 */
751
752static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
753{
754	unsigned int limit = databuf_limit(sdp);
755	unsigned int nbuf;
756	if (tr == NULL)
757		return;
758	nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
759	gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
760}
761
762static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
763				    struct gfs2_log_descriptor *ld,
764				    __be64 *ptr, int pass)
765{
766	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
767	struct gfs2_glock *gl = ip->i_gl;
768	unsigned int blks = be32_to_cpu(ld->ld_data1);
769	struct buffer_head *bh_log, *bh_ip;
770	u64 blkno;
771	u64 esc;
772	int error = 0;
773
774	if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
775		return 0;
776
777	gfs2_replay_incr_blk(jd, &start);
778	for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
779		blkno = be64_to_cpu(*ptr++);
780		esc = be64_to_cpu(*ptr++);
781
782		jd->jd_found_blocks++;
783
784		if (gfs2_revoke_check(jd, blkno, start))
785			continue;
786
787		error = gfs2_replay_read_block(jd, start, &bh_log);
788		if (error)
789			return error;
790
791		bh_ip = gfs2_meta_new(gl, blkno);
792		memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
793
794		/* Unescape */
795		if (esc) {
796			__be32 *eptr = (__be32 *)bh_ip->b_data;
797			*eptr = cpu_to_be32(GFS2_MAGIC);
798		}
799		mark_buffer_dirty(bh_ip);
800
801		brelse(bh_log);
802		brelse(bh_ip);
803
804		jd->jd_replayed_blocks++;
805	}
806
807	return error;
808}
809
810/* FIXME: sort out accounting for log blocks etc. */
811
812static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
813{
814	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
815	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
816
817	if (error) {
818		gfs2_meta_sync(ip->i_gl);
819		return;
820	}
821	if (pass != 1)
822		return;
823
824	/* data sync? */
825	gfs2_meta_sync(ip->i_gl);
826
827	fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
828		jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
829}
830
831static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
832{
833	struct list_head *head;
834	struct gfs2_bufdata *bd;
835
836	if (tr == NULL)
837		return;
838
839	head = &tr->tr_databuf;
840	while (!list_empty(head)) {
841		bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
842		list_del_init(&bd->bd_list);
843		gfs2_unpin(sdp, bd->bd_bh, tr);
844	}
845}
846
847
848const struct gfs2_log_operations gfs2_buf_lops = {
849	.lo_before_commit = buf_lo_before_commit,
850	.lo_after_commit = buf_lo_after_commit,
851	.lo_before_scan = buf_lo_before_scan,
852	.lo_scan_elements = buf_lo_scan_elements,
853	.lo_after_scan = buf_lo_after_scan,
854	.lo_name = "buf",
855};
856
857const struct gfs2_log_operations gfs2_revoke_lops = {
858	.lo_before_commit = revoke_lo_before_commit,
859	.lo_after_commit = revoke_lo_after_commit,
860	.lo_before_scan = revoke_lo_before_scan,
861	.lo_scan_elements = revoke_lo_scan_elements,
862	.lo_after_scan = revoke_lo_after_scan,
863	.lo_name = "revoke",
864};
865
866const struct gfs2_log_operations gfs2_databuf_lops = {
867	.lo_before_commit = databuf_lo_before_commit,
868	.lo_after_commit = databuf_lo_after_commit,
869	.lo_scan_elements = databuf_lo_scan_elements,
870	.lo_after_scan = databuf_lo_after_scan,
871	.lo_name = "databuf",
872};
873
874const struct gfs2_log_operations *gfs2_log_ops[] = {
875	&gfs2_databuf_lops,
876	&gfs2_buf_lops,
877	&gfs2_revoke_lops,
878	NULL,
879};
880