]> Pileus Git - ~andy/linux/blob - fs/gfs2/lops.c
Merge branch 'for-3.14/core' of git://git.kernel.dk/linux-block
[~andy/linux] / fs / gfs2 / lops.c
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  */
41 void 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
65 static bool buffer_is_rgrp(const struct gfs2_bufdata *bd)
66 {
67         return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP;
68 }
69
70 static void maybe_release_space(struct gfs2_bufdata *bd)
71 {
72         struct gfs2_glock *gl = bd->bd_gl;
73         struct gfs2_sbd *sdp = gl->gl_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 == 0)
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
98 static 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
132 static 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
143 static 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, extent_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
172 static 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
205 static void gfs2_end_log_write(struct bio *bio, int error)
206 {
207         struct gfs2_sbd *sdp = bio->bi_private;
208         struct bio_vec *bvec;
209         struct page *page;
210         int i;
211
212         if (error) {
213                 sdp->sd_log_error = error;
214                 fs_err(sdp, "Error %d writing to log\n", 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, 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  * @rw: The rw flags
234  *
235  * Submit any pending part-built or full bio to the block device. If
236  * there is no pending bio, then this is a no-op.
237  */
238
239 void gfs2_log_flush_bio(struct gfs2_sbd *sdp, int rw)
240 {
241         if (sdp->sd_log_bio) {
242                 atomic_inc(&sdp->sd_log_in_flight);
243                 submit_bio(rw, sdp->sd_log_bio);
244                 sdp->sd_log_bio = NULL;
245         }
246 }
247
248 /**
249  * gfs2_log_alloc_bio - Allocate a new bio for log writing
250  * @sdp: The superblock
251  * @blkno: The next device block number we want to write to
252  *
253  * This should never be called when there is a cached bio in the
254  * super block. When it returns, there will be a cached bio in the
255  * super block which will have as many bio_vecs as the device is
256  * happy to handle.
257  *
258  * Returns: Newly allocated bio
259  */
260
261 static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno)
262 {
263         struct super_block *sb = sdp->sd_vfs;
264         unsigned nrvecs = bio_get_nr_vecs(sb->s_bdev);
265         struct bio *bio;
266
267         BUG_ON(sdp->sd_log_bio);
268
269         while (1) {
270                 bio = bio_alloc(GFP_NOIO, nrvecs);
271                 if (likely(bio))
272                         break;
273                 nrvecs = max(nrvecs/2, 1U);
274         }
275
276         bio->bi_iter.bi_sector = blkno * (sb->s_blocksize >> 9);
277         bio->bi_bdev = sb->s_bdev;
278         bio->bi_end_io = gfs2_end_log_write;
279         bio->bi_private = sdp;
280
281         sdp->sd_log_bio = bio;
282
283         return bio;
284 }
285
286 /**
287  * gfs2_log_get_bio - Get cached log bio, or allocate a new one
288  * @sdp: The superblock
289  * @blkno: The device block number we want to write to
290  *
291  * If there is a cached bio, then if the next block number is sequential
292  * with the previous one, return it, otherwise flush the bio to the
293  * device. If there is not a cached bio, or we just flushed it, then
294  * allocate a new one.
295  *
296  * Returns: The bio to use for log writes
297  */
298
299 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno)
300 {
301         struct bio *bio = sdp->sd_log_bio;
302         u64 nblk;
303
304         if (bio) {
305                 nblk = bio_end_sector(bio);
306                 nblk >>= sdp->sd_fsb2bb_shift;
307                 if (blkno == nblk)
308                         return bio;
309                 gfs2_log_flush_bio(sdp, WRITE);
310         }
311
312         return gfs2_log_alloc_bio(sdp, blkno);
313 }
314
315
316 /**
317  * gfs2_log_write - write to log
318  * @sdp: the filesystem
319  * @page: the page to write
320  * @size: the size of the data to write
321  * @offset: the offset within the page 
322  *
323  * Try and add the page segment to the current bio. If that fails,
324  * submit the current bio to the device and create a new one, and
325  * then add the page segment to that.
326  */
327
328 static void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page,
329                            unsigned size, unsigned offset)
330 {
331         u64 blkno = gfs2_log_bmap(sdp);
332         struct bio *bio;
333         int ret;
334
335         bio = gfs2_log_get_bio(sdp, blkno);
336         ret = bio_add_page(bio, page, size, offset);
337         if (ret == 0) {
338                 gfs2_log_flush_bio(sdp, WRITE);
339                 bio = gfs2_log_alloc_bio(sdp, blkno);
340                 ret = bio_add_page(bio, page, size, offset);
341                 WARN_ON(ret == 0);
342         }
343 }
344
345 /**
346  * gfs2_log_write_bh - write a buffer's content to the log
347  * @sdp: The super block
348  * @bh: The buffer pointing to the in-place location
349  * 
350  * This writes the content of the buffer to the next available location
351  * in the log. The buffer will be unlocked once the i/o to the log has
352  * completed.
353  */
354
355 static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
356 {
357         gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh));
358 }
359
360 /**
361  * gfs2_log_write_page - write one block stored in a page, into the log
362  * @sdp: The superblock
363  * @page: The struct page
364  *
365  * This writes the first block-sized part of the page into the log. Note
366  * that the page must have been allocated from the gfs2_page_pool mempool
367  * and that after this has been called, ownership has been transferred and
368  * the page may be freed at any time.
369  */
370
371 void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
372 {
373         struct super_block *sb = sdp->sd_vfs;
374         gfs2_log_write(sdp, page, sb->s_blocksize, 0);
375 }
376
377 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
378                                       u32 ld_length, u32 ld_data1)
379 {
380         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
381         struct gfs2_log_descriptor *ld = page_address(page);
382         clear_page(ld);
383         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
384         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
385         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
386         ld->ld_type = cpu_to_be32(ld_type);
387         ld->ld_length = cpu_to_be32(ld_length);
388         ld->ld_data1 = cpu_to_be32(ld_data1);
389         ld->ld_data2 = 0;
390         return page;
391 }
392
393 static void gfs2_check_magic(struct buffer_head *bh)
394 {
395         void *kaddr;
396         __be32 *ptr;
397
398         clear_buffer_escaped(bh);
399         kaddr = kmap_atomic(bh->b_page);
400         ptr = kaddr + bh_offset(bh);
401         if (*ptr == cpu_to_be32(GFS2_MAGIC))
402                 set_buffer_escaped(bh);
403         kunmap_atomic(kaddr);
404 }
405
406 static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b)
407 {
408         struct gfs2_bufdata *bda, *bdb;
409
410         bda = list_entry(a, struct gfs2_bufdata, bd_list);
411         bdb = list_entry(b, struct gfs2_bufdata, bd_list);
412
413         if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
414                 return -1;
415         if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
416                 return 1;
417         return 0;
418 }
419
420 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
421                                 unsigned int total, struct list_head *blist,
422                                 bool is_databuf)
423 {
424         struct gfs2_log_descriptor *ld;
425         struct gfs2_bufdata *bd1 = NULL, *bd2;
426         struct page *page;
427         unsigned int num;
428         unsigned n;
429         __be64 *ptr;
430
431         gfs2_log_lock(sdp);
432         list_sort(NULL, blist, blocknr_cmp);
433         bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
434         while(total) {
435                 num = total;
436                 if (total > limit)
437                         num = limit;
438                 gfs2_log_unlock(sdp);
439                 page = gfs2_get_log_desc(sdp,
440                                          is_databuf ? GFS2_LOG_DESC_JDATA :
441                                          GFS2_LOG_DESC_METADATA, num + 1, num);
442                 ld = page_address(page);
443                 gfs2_log_lock(sdp);
444                 ptr = (__be64 *)(ld + 1);
445
446                 n = 0;
447                 list_for_each_entry_continue(bd1, blist, bd_list) {
448                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
449                         if (is_databuf) {
450                                 gfs2_check_magic(bd1->bd_bh);
451                                 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
452                         }
453                         if (++n >= num)
454                                 break;
455                 }
456
457                 gfs2_log_unlock(sdp);
458                 gfs2_log_write_page(sdp, page);
459                 gfs2_log_lock(sdp);
460
461                 n = 0;
462                 list_for_each_entry_continue(bd2, blist, bd_list) {
463                         get_bh(bd2->bd_bh);
464                         gfs2_log_unlock(sdp);
465                         lock_buffer(bd2->bd_bh);
466
467                         if (buffer_escaped(bd2->bd_bh)) {
468                                 void *kaddr;
469                                 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
470                                 ptr = page_address(page);
471                                 kaddr = kmap_atomic(bd2->bd_bh->b_page);
472                                 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
473                                        bd2->bd_bh->b_size);
474                                 kunmap_atomic(kaddr);
475                                 *(__be32 *)ptr = 0;
476                                 clear_buffer_escaped(bd2->bd_bh);
477                                 unlock_buffer(bd2->bd_bh);
478                                 brelse(bd2->bd_bh);
479                                 gfs2_log_write_page(sdp, page);
480                         } else {
481                                 gfs2_log_write_bh(sdp, bd2->bd_bh);
482                         }
483                         gfs2_log_lock(sdp);
484                         if (++n >= num)
485                                 break;
486                 }
487
488                 BUG_ON(total < num);
489                 total -= num;
490         }
491         gfs2_log_unlock(sdp);
492 }
493
494 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
495 {
496         unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
497
498         gfs2_before_commit(sdp, limit, sdp->sd_log_num_buf,
499                            &sdp->sd_log_le_buf, 0);
500 }
501
502 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
503 {
504         struct list_head *head = &sdp->sd_log_le_buf;
505         struct gfs2_bufdata *bd;
506
507         if (tr == NULL) {
508                 gfs2_assert(sdp, list_empty(head));
509                 return;
510         }
511
512         while (!list_empty(head)) {
513                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
514                 list_del_init(&bd->bd_list);
515                 sdp->sd_log_num_buf--;
516
517                 gfs2_unpin(sdp, bd->bd_bh, tr);
518         }
519         gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
520 }
521
522 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
523                                struct gfs2_log_header_host *head, int pass)
524 {
525         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
526
527         if (pass != 0)
528                 return;
529
530         sdp->sd_found_blocks = 0;
531         sdp->sd_replayed_blocks = 0;
532 }
533
534 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
535                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
536                                 int pass)
537 {
538         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
539         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
540         struct gfs2_glock *gl = ip->i_gl;
541         unsigned int blks = be32_to_cpu(ld->ld_data1);
542         struct buffer_head *bh_log, *bh_ip;
543         u64 blkno;
544         int error = 0;
545
546         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
547                 return 0;
548
549         gfs2_replay_incr_blk(sdp, &start);
550
551         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
552                 blkno = be64_to_cpu(*ptr++);
553
554                 sdp->sd_found_blocks++;
555
556                 if (gfs2_revoke_check(sdp, blkno, start))
557                         continue;
558
559                 error = gfs2_replay_read_block(jd, start, &bh_log);
560                 if (error)
561                         return error;
562
563                 bh_ip = gfs2_meta_new(gl, blkno);
564                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
565
566                 if (gfs2_meta_check(sdp, bh_ip))
567                         error = -EIO;
568                 else
569                         mark_buffer_dirty(bh_ip);
570
571                 brelse(bh_log);
572                 brelse(bh_ip);
573
574                 if (error)
575                         break;
576
577                 sdp->sd_replayed_blocks++;
578         }
579
580         return error;
581 }
582
583 /**
584  * gfs2_meta_sync - Sync all buffers associated with a glock
585  * @gl: The glock
586  *
587  */
588
589 static void gfs2_meta_sync(struct gfs2_glock *gl)
590 {
591         struct address_space *mapping = gfs2_glock2aspace(gl);
592         struct gfs2_sbd *sdp = gl->gl_sbd;
593         int error;
594
595         if (mapping == NULL)
596                 mapping = &sdp->sd_aspace;
597
598         filemap_fdatawrite(mapping);
599         error = filemap_fdatawait(mapping);
600
601         if (error)
602                 gfs2_io_error(gl->gl_sbd);
603 }
604
605 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
606 {
607         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
608         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
609
610         if (error) {
611                 gfs2_meta_sync(ip->i_gl);
612                 return;
613         }
614         if (pass != 1)
615                 return;
616
617         gfs2_meta_sync(ip->i_gl);
618
619         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
620                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
621 }
622
623 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
624 {
625         struct gfs2_meta_header *mh;
626         unsigned int offset;
627         struct list_head *head = &sdp->sd_log_le_revoke;
628         struct gfs2_bufdata *bd;
629         struct page *page;
630         unsigned int length;
631
632         gfs2_write_revokes(sdp);
633         if (!sdp->sd_log_num_revoke)
634                 return;
635
636         length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
637         page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
638         offset = sizeof(struct gfs2_log_descriptor);
639
640         list_for_each_entry(bd, head, bd_list) {
641                 sdp->sd_log_num_revoke--;
642
643                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
644
645                         gfs2_log_write_page(sdp, page);
646                         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
647                         mh = page_address(page);
648                         clear_page(mh);
649                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
650                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
651                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
652                         offset = sizeof(struct gfs2_meta_header);
653                 }
654
655                 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
656                 offset += sizeof(u64);
657         }
658         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
659
660         gfs2_log_write_page(sdp, page);
661 }
662
663 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
664 {
665         struct list_head *head = &sdp->sd_log_le_revoke;
666         struct gfs2_bufdata *bd;
667         struct gfs2_glock *gl;
668
669         while (!list_empty(head)) {
670                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
671                 list_del_init(&bd->bd_list);
672                 gl = bd->bd_gl;
673                 atomic_dec(&gl->gl_revokes);
674                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
675                 kmem_cache_free(gfs2_bufdata_cachep, bd);
676         }
677 }
678
679 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
680                                   struct gfs2_log_header_host *head, int pass)
681 {
682         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
683
684         if (pass != 0)
685                 return;
686
687         sdp->sd_found_revokes = 0;
688         sdp->sd_replay_tail = head->lh_tail;
689 }
690
691 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
692                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
693                                    int pass)
694 {
695         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
696         unsigned int blks = be32_to_cpu(ld->ld_length);
697         unsigned int revokes = be32_to_cpu(ld->ld_data1);
698         struct buffer_head *bh;
699         unsigned int offset;
700         u64 blkno;
701         int first = 1;
702         int error;
703
704         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
705                 return 0;
706
707         offset = sizeof(struct gfs2_log_descriptor);
708
709         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
710                 error = gfs2_replay_read_block(jd, start, &bh);
711                 if (error)
712                         return error;
713
714                 if (!first)
715                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
716
717                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
718                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
719
720                         error = gfs2_revoke_add(sdp, blkno, start);
721                         if (error < 0) {
722                                 brelse(bh);
723                                 return error;
724                         }
725                         else if (error)
726                                 sdp->sd_found_revokes++;
727
728                         if (!--revokes)
729                                 break;
730                         offset += sizeof(u64);
731                 }
732
733                 brelse(bh);
734                 offset = sizeof(struct gfs2_meta_header);
735                 first = 0;
736         }
737
738         return 0;
739 }
740
741 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
742 {
743         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
744
745         if (error) {
746                 gfs2_revoke_clean(sdp);
747                 return;
748         }
749         if (pass != 1)
750                 return;
751
752         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
753                 jd->jd_jid, sdp->sd_found_revokes);
754
755         gfs2_revoke_clean(sdp);
756 }
757
758 /**
759  * databuf_lo_before_commit - Scan the data buffers, writing as we go
760  *
761  */
762
763 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
764 {
765         unsigned int limit = buf_limit(sdp) / 2;
766
767         gfs2_before_commit(sdp, limit, sdp->sd_log_num_databuf,
768                            &sdp->sd_log_le_databuf, 1);
769 }
770
771 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
772                                     struct gfs2_log_descriptor *ld,
773                                     __be64 *ptr, int pass)
774 {
775         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
776         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
777         struct gfs2_glock *gl = ip->i_gl;
778         unsigned int blks = be32_to_cpu(ld->ld_data1);
779         struct buffer_head *bh_log, *bh_ip;
780         u64 blkno;
781         u64 esc;
782         int error = 0;
783
784         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
785                 return 0;
786
787         gfs2_replay_incr_blk(sdp, &start);
788         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
789                 blkno = be64_to_cpu(*ptr++);
790                 esc = be64_to_cpu(*ptr++);
791
792                 sdp->sd_found_blocks++;
793
794                 if (gfs2_revoke_check(sdp, blkno, start))
795                         continue;
796
797                 error = gfs2_replay_read_block(jd, start, &bh_log);
798                 if (error)
799                         return error;
800
801                 bh_ip = gfs2_meta_new(gl, blkno);
802                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
803
804                 /* Unescape */
805                 if (esc) {
806                         __be32 *eptr = (__be32 *)bh_ip->b_data;
807                         *eptr = cpu_to_be32(GFS2_MAGIC);
808                 }
809                 mark_buffer_dirty(bh_ip);
810
811                 brelse(bh_log);
812                 brelse(bh_ip);
813
814                 sdp->sd_replayed_blocks++;
815         }
816
817         return error;
818 }
819
820 /* FIXME: sort out accounting for log blocks etc. */
821
822 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
823 {
824         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
825         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
826
827         if (error) {
828                 gfs2_meta_sync(ip->i_gl);
829                 return;
830         }
831         if (pass != 1)
832                 return;
833
834         /* data sync? */
835         gfs2_meta_sync(ip->i_gl);
836
837         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
838                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
839 }
840
841 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
842 {
843         struct list_head *head = &sdp->sd_log_le_databuf;
844         struct gfs2_bufdata *bd;
845
846         if (tr == NULL) {
847                 gfs2_assert(sdp, list_empty(head));
848                 return;
849         }
850
851         while (!list_empty(head)) {
852                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
853                 list_del_init(&bd->bd_list);
854                 sdp->sd_log_num_databuf--;
855                 gfs2_unpin(sdp, bd->bd_bh, tr);
856         }
857         gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
858 }
859
860
861 const struct gfs2_log_operations gfs2_buf_lops = {
862         .lo_before_commit = buf_lo_before_commit,
863         .lo_after_commit = buf_lo_after_commit,
864         .lo_before_scan = buf_lo_before_scan,
865         .lo_scan_elements = buf_lo_scan_elements,
866         .lo_after_scan = buf_lo_after_scan,
867         .lo_name = "buf",
868 };
869
870 const struct gfs2_log_operations gfs2_revoke_lops = {
871         .lo_before_commit = revoke_lo_before_commit,
872         .lo_after_commit = revoke_lo_after_commit,
873         .lo_before_scan = revoke_lo_before_scan,
874         .lo_scan_elements = revoke_lo_scan_elements,
875         .lo_after_scan = revoke_lo_after_scan,
876         .lo_name = "revoke",
877 };
878
879 const struct gfs2_log_operations gfs2_databuf_lops = {
880         .lo_before_commit = databuf_lo_before_commit,
881         .lo_after_commit = databuf_lo_after_commit,
882         .lo_scan_elements = databuf_lo_scan_elements,
883         .lo_after_scan = databuf_lo_after_scan,
884         .lo_name = "databuf",
885 };
886
887 const struct gfs2_log_operations *gfs2_log_ops[] = {
888         &gfs2_databuf_lops,
889         &gfs2_buf_lops,
890         &gfs2_revoke_lops,
891         NULL,
892 };
893