Loading...
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
20#include "gfs2.h"
21#include "incore.h"
22#include "inode.h"
23#include "glock.h"
24#include "log.h"
25#include "lops.h"
26#include "meta_io.h"
27#include "recovery.h"
28#include "rgrp.h"
29#include "trans.h"
30#include "util.h"
31#include "trace_gfs2.h"
32
33/**
34 * gfs2_pin - Pin a buffer in memory
35 * @sdp: The superblock
36 * @bh: The buffer to be pinned
37 *
38 * The log lock must be held when calling this function
39 */
40static void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
41{
42 struct gfs2_bufdata *bd;
43
44 BUG_ON(!current->journal_info);
45
46 clear_buffer_dirty(bh);
47 if (test_set_buffer_pinned(bh))
48 gfs2_assert_withdraw(sdp, 0);
49 if (!buffer_uptodate(bh))
50 gfs2_io_error_bh(sdp, bh);
51 bd = bh->b_private;
52 /* If this buffer is in the AIL and it has already been written
53 * to in-place disk block, remove it from the AIL.
54 */
55 spin_lock(&sdp->sd_ail_lock);
56 if (bd->bd_ail)
57 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
58 spin_unlock(&sdp->sd_ail_lock);
59 get_bh(bh);
60 atomic_inc(&sdp->sd_log_pinned);
61 trace_gfs2_pin(bd, 1);
62}
63
64static bool buffer_is_rgrp(const struct gfs2_bufdata *bd)
65{
66 return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP;
67}
68
69static void maybe_release_space(struct gfs2_bufdata *bd)
70{
71 struct gfs2_glock *gl = bd->bd_gl;
72 struct gfs2_sbd *sdp = gl->gl_sbd;
73 struct gfs2_rgrpd *rgd = gl->gl_object;
74 unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number;
75 struct gfs2_bitmap *bi = rgd->rd_bits + index;
76
77 if (bi->bi_clone == 0)
78 return;
79 if (sdp->sd_args.ar_discard)
80 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
81 memcpy(bi->bi_clone + bi->bi_offset,
82 bd->bd_bh->b_data + bi->bi_offset, bi->bi_len);
83 clear_bit(GBF_FULL, &bi->bi_flags);
84 rgd->rd_free_clone = rgd->rd_free;
85}
86
87/**
88 * gfs2_unpin - Unpin a buffer
89 * @sdp: the filesystem the buffer belongs to
90 * @bh: The buffer to unpin
91 * @ai:
92 * @flags: The inode dirty flags
93 *
94 */
95
96static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
97 struct gfs2_ail *ai)
98{
99 struct gfs2_bufdata *bd = bh->b_private;
100
101 BUG_ON(!buffer_uptodate(bh));
102 BUG_ON(!buffer_pinned(bh));
103
104 lock_buffer(bh);
105 mark_buffer_dirty(bh);
106 clear_buffer_pinned(bh);
107
108 if (buffer_is_rgrp(bd))
109 maybe_release_space(bd);
110
111 spin_lock(&sdp->sd_ail_lock);
112 if (bd->bd_ail) {
113 list_del(&bd->bd_ail_st_list);
114 brelse(bh);
115 } else {
116 struct gfs2_glock *gl = bd->bd_gl;
117 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
118 atomic_inc(&gl->gl_ail_count);
119 }
120 bd->bd_ail = ai;
121 list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
122 spin_unlock(&sdp->sd_ail_lock);
123
124 clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
125 trace_gfs2_pin(bd, 0);
126 unlock_buffer(bh);
127 atomic_dec(&sdp->sd_log_pinned);
128}
129
130static void gfs2_log_incr_head(struct gfs2_sbd *sdp)
131{
132 BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
133 (sdp->sd_log_flush_head != sdp->sd_log_head));
134
135 if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
136 sdp->sd_log_flush_head = 0;
137 sdp->sd_log_flush_wrapped = 1;
138 }
139}
140
141static u64 gfs2_log_bmap(struct gfs2_sbd *sdp)
142{
143 unsigned int lbn = sdp->sd_log_flush_head;
144 struct gfs2_journal_extent *je;
145 u64 block;
146
147 list_for_each_entry(je, &sdp->sd_jdesc->extent_list, extent_list) {
148 if (lbn >= je->lblock && lbn < je->lblock + je->blocks) {
149 block = je->dblock + lbn - je->lblock;
150 gfs2_log_incr_head(sdp);
151 return block;
152 }
153 }
154
155 return -1;
156}
157
158/**
159 * gfs2_end_log_write_bh - end log write of pagecache data with buffers
160 * @sdp: The superblock
161 * @bvec: The bio_vec
162 * @error: The i/o status
163 *
164 * This finds the relavent buffers and unlocks then and sets the
165 * error flag according to the status of the i/o request. This is
166 * used when the log is writing data which has an in-place version
167 * that is pinned in the pagecache.
168 */
169
170static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp, struct bio_vec *bvec,
171 int error)
172{
173 struct buffer_head *bh, *next;
174 struct page *page = bvec->bv_page;
175 unsigned size;
176
177 bh = page_buffers(page);
178 size = bvec->bv_len;
179 while (bh_offset(bh) < bvec->bv_offset)
180 bh = bh->b_this_page;
181 do {
182 if (error)
183 set_buffer_write_io_error(bh);
184 unlock_buffer(bh);
185 next = bh->b_this_page;
186 size -= bh->b_size;
187 brelse(bh);
188 bh = next;
189 } while(bh && size);
190}
191
192/**
193 * gfs2_end_log_write - end of i/o to the log
194 * @bio: The bio
195 * @error: Status of i/o request
196 *
197 * Each bio_vec contains either data from the pagecache or data
198 * relating to the log itself. Here we iterate over the bio_vec
199 * array, processing both kinds of data.
200 *
201 */
202
203static void gfs2_end_log_write(struct bio *bio, int error)
204{
205 struct gfs2_sbd *sdp = bio->bi_private;
206 struct bio_vec *bvec;
207 struct page *page;
208 int i;
209
210 if (error) {
211 sdp->sd_log_error = error;
212 fs_err(sdp, "Error %d writing to log\n", error);
213 }
214
215 bio_for_each_segment(bvec, bio, i) {
216 page = bvec->bv_page;
217 if (page_has_buffers(page))
218 gfs2_end_log_write_bh(sdp, bvec, error);
219 else
220 mempool_free(page, gfs2_page_pool);
221 }
222
223 bio_put(bio);
224 if (atomic_dec_and_test(&sdp->sd_log_in_flight))
225 wake_up(&sdp->sd_log_flush_wait);
226}
227
228/**
229 * gfs2_log_flush_bio - Submit any pending log bio
230 * @sdp: The superblock
231 * @rw: The rw flags
232 *
233 * Submit any pending part-built or full bio to the block device. If
234 * there is no pending bio, then this is a no-op.
235 */
236
237void gfs2_log_flush_bio(struct gfs2_sbd *sdp, int rw)
238{
239 if (sdp->sd_log_bio) {
240 atomic_inc(&sdp->sd_log_in_flight);
241 submit_bio(rw, sdp->sd_log_bio);
242 sdp->sd_log_bio = NULL;
243 }
244}
245
246/**
247 * gfs2_log_alloc_bio - Allocate a new bio for log writing
248 * @sdp: The superblock
249 * @blkno: The next device block number we want to write to
250 *
251 * This should never be called when there is a cached bio in the
252 * super block. When it returns, there will be a cached bio in the
253 * super block which will have as many bio_vecs as the device is
254 * happy to handle.
255 *
256 * Returns: Newly allocated bio
257 */
258
259static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno)
260{
261 struct super_block *sb = sdp->sd_vfs;
262 unsigned nrvecs = bio_get_nr_vecs(sb->s_bdev);
263 struct bio *bio;
264
265 BUG_ON(sdp->sd_log_bio);
266
267 while (1) {
268 bio = bio_alloc(GFP_NOIO, nrvecs);
269 if (likely(bio))
270 break;
271 nrvecs = max(nrvecs/2, 1U);
272 }
273
274 bio->bi_sector = blkno * (sb->s_blocksize >> 9);
275 bio->bi_bdev = sb->s_bdev;
276 bio->bi_end_io = gfs2_end_log_write;
277 bio->bi_private = sdp;
278
279 sdp->sd_log_bio = bio;
280
281 return bio;
282}
283
284/**
285 * gfs2_log_get_bio - Get cached log bio, or allocate a new one
286 * @sdp: The superblock
287 * @blkno: The device block number we want to write to
288 *
289 * If there is a cached bio, then if the next block number is sequential
290 * with the previous one, return it, otherwise flush the bio to the
291 * device. If there is not a cached bio, or we just flushed it, then
292 * allocate a new one.
293 *
294 * Returns: The bio to use for log writes
295 */
296
297static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno)
298{
299 struct bio *bio = sdp->sd_log_bio;
300 u64 nblk;
301
302 if (bio) {
303 nblk = bio->bi_sector + bio_sectors(bio);
304 nblk >>= sdp->sd_fsb2bb_shift;
305 if (blkno == nblk)
306 return bio;
307 gfs2_log_flush_bio(sdp, WRITE);
308 }
309
310 return gfs2_log_alloc_bio(sdp, blkno);
311}
312
313
314/**
315 * gfs2_log_write - write to log
316 * @sdp: the filesystem
317 * @page: the page to write
318 * @size: the size of the data to write
319 * @offset: the offset within the page
320 *
321 * Try and add the page segment to the current bio. If that fails,
322 * submit the current bio to the device and create a new one, and
323 * then add the page segment to that.
324 */
325
326static void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page,
327 unsigned size, unsigned offset)
328{
329 u64 blkno = gfs2_log_bmap(sdp);
330 struct bio *bio;
331 int ret;
332
333 bio = gfs2_log_get_bio(sdp, blkno);
334 ret = bio_add_page(bio, page, size, offset);
335 if (ret == 0) {
336 gfs2_log_flush_bio(sdp, WRITE);
337 bio = gfs2_log_alloc_bio(sdp, blkno);
338 ret = bio_add_page(bio, page, size, offset);
339 WARN_ON(ret == 0);
340 }
341}
342
343/**
344 * gfs2_log_write_bh - write a buffer's content to the log
345 * @sdp: The super block
346 * @bh: The buffer pointing to the in-place location
347 *
348 * This writes the content of the buffer to the next available location
349 * in the log. The buffer will be unlocked once the i/o to the log has
350 * completed.
351 */
352
353static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
354{
355 gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh));
356}
357
358/**
359 * gfs2_log_write_page - write one block stored in a page, into the log
360 * @sdp: The superblock
361 * @page: The struct page
362 *
363 * This writes the first block-sized part of the page into the log. Note
364 * that the page must have been allocated from the gfs2_page_pool mempool
365 * and that after this has been called, ownership has been transferred and
366 * the page may be freed at any time.
367 */
368
369void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
370{
371 struct super_block *sb = sdp->sd_vfs;
372 gfs2_log_write(sdp, page, sb->s_blocksize, 0);
373}
374
375static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
376 u32 ld_length, u32 ld_data1)
377{
378 struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
379 struct gfs2_log_descriptor *ld = page_address(page);
380 clear_page(ld);
381 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
382 ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
383 ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
384 ld->ld_type = cpu_to_be32(ld_type);
385 ld->ld_length = cpu_to_be32(ld_length);
386 ld->ld_data1 = cpu_to_be32(ld_data1);
387 ld->ld_data2 = 0;
388 return page;
389}
390
391static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
392{
393 struct gfs2_meta_header *mh;
394 struct gfs2_trans *tr;
395
396 lock_buffer(bd->bd_bh);
397 gfs2_log_lock(sdp);
398 tr = current->journal_info;
399 tr->tr_touched = 1;
400 if (!list_empty(&bd->bd_list))
401 goto out;
402 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
403 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
404 gfs2_meta_check(sdp, bd->bd_bh);
405 gfs2_pin(sdp, bd->bd_bh);
406 mh = (struct gfs2_meta_header *)bd->bd_bh->b_data;
407 mh->__pad0 = cpu_to_be64(0);
408 mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
409 sdp->sd_log_num_buf++;
410 list_add(&bd->bd_list, &sdp->sd_log_le_buf);
411 tr->tr_num_buf_new++;
412out:
413 gfs2_log_unlock(sdp);
414 unlock_buffer(bd->bd_bh);
415}
416
417static void gfs2_check_magic(struct buffer_head *bh)
418{
419 void *kaddr;
420 __be32 *ptr;
421
422 clear_buffer_escaped(bh);
423 kaddr = kmap_atomic(bh->b_page);
424 ptr = kaddr + bh_offset(bh);
425 if (*ptr == cpu_to_be32(GFS2_MAGIC))
426 set_buffer_escaped(bh);
427 kunmap_atomic(kaddr);
428}
429
430static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
431 unsigned int total, struct list_head *blist,
432 bool is_databuf)
433{
434 struct gfs2_log_descriptor *ld;
435 struct gfs2_bufdata *bd1 = NULL, *bd2;
436 struct page *page;
437 unsigned int num;
438 unsigned n;
439 __be64 *ptr;
440
441 gfs2_log_lock(sdp);
442 bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
443 while(total) {
444 num = total;
445 if (total > limit)
446 num = limit;
447 gfs2_log_unlock(sdp);
448 page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA, num + 1, num);
449 ld = page_address(page);
450 gfs2_log_lock(sdp);
451 ptr = (__be64 *)(ld + 1);
452
453 n = 0;
454 list_for_each_entry_continue(bd1, blist, bd_list) {
455 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
456 if (is_databuf) {
457 gfs2_check_magic(bd1->bd_bh);
458 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
459 }
460 if (++n >= num)
461 break;
462 }
463
464 gfs2_log_unlock(sdp);
465 gfs2_log_write_page(sdp, page);
466 gfs2_log_lock(sdp);
467
468 n = 0;
469 list_for_each_entry_continue(bd2, blist, bd_list) {
470 get_bh(bd2->bd_bh);
471 gfs2_log_unlock(sdp);
472 lock_buffer(bd2->bd_bh);
473
474 if (buffer_escaped(bd2->bd_bh)) {
475 void *kaddr;
476 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
477 ptr = page_address(page);
478 kaddr = kmap_atomic(bd2->bd_bh->b_page);
479 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
480 bd2->bd_bh->b_size);
481 kunmap_atomic(kaddr);
482 *(__be32 *)ptr = 0;
483 clear_buffer_escaped(bd2->bd_bh);
484 unlock_buffer(bd2->bd_bh);
485 brelse(bd2->bd_bh);
486 gfs2_log_write_page(sdp, page);
487 } else {
488 gfs2_log_write_bh(sdp, bd2->bd_bh);
489 }
490 gfs2_log_lock(sdp);
491 if (++n >= num)
492 break;
493 }
494
495 BUG_ON(total < num);
496 total -= num;
497 }
498 gfs2_log_unlock(sdp);
499}
500
501static void buf_lo_before_commit(struct gfs2_sbd *sdp)
502{
503 unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
504
505 gfs2_before_commit(sdp, limit, sdp->sd_log_num_buf,
506 &sdp->sd_log_le_buf, 0);
507}
508
509static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
510{
511 struct list_head *head = &sdp->sd_log_le_buf;
512 struct gfs2_bufdata *bd;
513
514 while (!list_empty(head)) {
515 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
516 list_del_init(&bd->bd_list);
517 sdp->sd_log_num_buf--;
518
519 gfs2_unpin(sdp, bd->bd_bh, ai);
520 }
521 gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
522}
523
524static void buf_lo_before_scan(struct gfs2_jdesc *jd,
525 struct gfs2_log_header_host *head, int pass)
526{
527 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
528
529 if (pass != 0)
530 return;
531
532 sdp->sd_found_blocks = 0;
533 sdp->sd_replayed_blocks = 0;
534}
535
536static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
537 struct gfs2_log_descriptor *ld, __be64 *ptr,
538 int pass)
539{
540 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
541 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
542 struct gfs2_glock *gl = ip->i_gl;
543 unsigned int blks = be32_to_cpu(ld->ld_data1);
544 struct buffer_head *bh_log, *bh_ip;
545 u64 blkno;
546 int error = 0;
547
548 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
549 return 0;
550
551 gfs2_replay_incr_blk(sdp, &start);
552
553 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
554 blkno = be64_to_cpu(*ptr++);
555
556 sdp->sd_found_blocks++;
557
558 if (gfs2_revoke_check(sdp, blkno, start))
559 continue;
560
561 error = gfs2_replay_read_block(jd, start, &bh_log);
562 if (error)
563 return error;
564
565 bh_ip = gfs2_meta_new(gl, blkno);
566 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
567
568 if (gfs2_meta_check(sdp, bh_ip))
569 error = -EIO;
570 else
571 mark_buffer_dirty(bh_ip);
572
573 brelse(bh_log);
574 brelse(bh_ip);
575
576 if (error)
577 break;
578
579 sdp->sd_replayed_blocks++;
580 }
581
582 return error;
583}
584
585static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
586{
587 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
588 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
589
590 if (error) {
591 gfs2_meta_sync(ip->i_gl);
592 return;
593 }
594 if (pass != 1)
595 return;
596
597 gfs2_meta_sync(ip->i_gl);
598
599 fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
600 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
601}
602
603static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
604{
605 struct gfs2_glock *gl = bd->bd_gl;
606 struct gfs2_trans *tr;
607
608 tr = current->journal_info;
609 tr->tr_touched = 1;
610 tr->tr_num_revoke++;
611 sdp->sd_log_num_revoke++;
612 atomic_inc(&gl->gl_revokes);
613 set_bit(GLF_LFLUSH, &gl->gl_flags);
614 list_add(&bd->bd_list, &sdp->sd_log_le_revoke);
615}
616
617static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
618{
619 struct gfs2_log_descriptor *ld;
620 struct gfs2_meta_header *mh;
621 unsigned int offset;
622 struct list_head *head = &sdp->sd_log_le_revoke;
623 struct gfs2_bufdata *bd;
624 struct page *page;
625 unsigned int length;
626
627 if (!sdp->sd_log_num_revoke)
628 return;
629
630 length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
631 page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
632 ld = page_address(page);
633 offset = sizeof(struct gfs2_log_descriptor);
634
635 list_for_each_entry(bd, head, bd_list) {
636 sdp->sd_log_num_revoke--;
637
638 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
639
640 gfs2_log_write_page(sdp, page);
641 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
642 mh = page_address(page);
643 clear_page(mh);
644 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
645 mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
646 mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
647 offset = sizeof(struct gfs2_meta_header);
648 }
649
650 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
651 offset += sizeof(u64);
652 }
653 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
654
655 gfs2_log_write_page(sdp, page);
656}
657
658static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
659{
660 struct list_head *head = &sdp->sd_log_le_revoke;
661 struct gfs2_bufdata *bd;
662 struct gfs2_glock *gl;
663
664 while (!list_empty(head)) {
665 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
666 list_del_init(&bd->bd_list);
667 gl = bd->bd_gl;
668 atomic_dec(&gl->gl_revokes);
669 clear_bit(GLF_LFLUSH, &gl->gl_flags);
670 kmem_cache_free(gfs2_bufdata_cachep, bd);
671 }
672}
673
674static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
675 struct gfs2_log_header_host *head, int pass)
676{
677 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
678
679 if (pass != 0)
680 return;
681
682 sdp->sd_found_revokes = 0;
683 sdp->sd_replay_tail = head->lh_tail;
684}
685
686static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
687 struct gfs2_log_descriptor *ld, __be64 *ptr,
688 int pass)
689{
690 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
691 unsigned int blks = be32_to_cpu(ld->ld_length);
692 unsigned int revokes = be32_to_cpu(ld->ld_data1);
693 struct buffer_head *bh;
694 unsigned int offset;
695 u64 blkno;
696 int first = 1;
697 int error;
698
699 if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
700 return 0;
701
702 offset = sizeof(struct gfs2_log_descriptor);
703
704 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
705 error = gfs2_replay_read_block(jd, start, &bh);
706 if (error)
707 return error;
708
709 if (!first)
710 gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
711
712 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
713 blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
714
715 error = gfs2_revoke_add(sdp, blkno, start);
716 if (error < 0) {
717 brelse(bh);
718 return error;
719 }
720 else if (error)
721 sdp->sd_found_revokes++;
722
723 if (!--revokes)
724 break;
725 offset += sizeof(u64);
726 }
727
728 brelse(bh);
729 offset = sizeof(struct gfs2_meta_header);
730 first = 0;
731 }
732
733 return 0;
734}
735
736static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
737{
738 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
739
740 if (error) {
741 gfs2_revoke_clean(sdp);
742 return;
743 }
744 if (pass != 1)
745 return;
746
747 fs_info(sdp, "jid=%u: Found %u revoke tags\n",
748 jd->jd_jid, sdp->sd_found_revokes);
749
750 gfs2_revoke_clean(sdp);
751}
752
753/**
754 * databuf_lo_add - Add a databuf to the transaction.
755 *
756 * This is used in two distinct cases:
757 * i) In ordered write mode
758 * We put the data buffer on a list so that we can ensure that its
759 * synced to disk at the right time
760 * ii) In journaled data mode
761 * We need to journal the data block in the same way as metadata in
762 * the functions above. The difference is that here we have a tag
763 * which is two __be64's being the block number (as per meta data)
764 * and a flag which says whether the data block needs escaping or
765 * not. This means we need a new log entry for each 251 or so data
766 * blocks, which isn't an enormous overhead but twice as much as
767 * for normal metadata blocks.
768 */
769static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
770{
771 struct gfs2_trans *tr = current->journal_info;
772 struct address_space *mapping = bd->bd_bh->b_page->mapping;
773 struct gfs2_inode *ip = GFS2_I(mapping->host);
774
775 lock_buffer(bd->bd_bh);
776 gfs2_log_lock(sdp);
777 if (tr)
778 tr->tr_touched = 1;
779 if (!list_empty(&bd->bd_list))
780 goto out;
781 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
782 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
783 if (gfs2_is_jdata(ip)) {
784 gfs2_pin(sdp, bd->bd_bh);
785 tr->tr_num_databuf_new++;
786 sdp->sd_log_num_databuf++;
787 list_add_tail(&bd->bd_list, &sdp->sd_log_le_databuf);
788 } else {
789 list_add_tail(&bd->bd_list, &sdp->sd_log_le_ordered);
790 }
791out:
792 gfs2_log_unlock(sdp);
793 unlock_buffer(bd->bd_bh);
794}
795
796/**
797 * databuf_lo_before_commit - Scan the data buffers, writing as we go
798 *
799 */
800
801static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
802{
803 unsigned int limit = buf_limit(sdp) / 2;
804
805 gfs2_before_commit(sdp, limit, sdp->sd_log_num_databuf,
806 &sdp->sd_log_le_databuf, 1);
807}
808
809static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
810 struct gfs2_log_descriptor *ld,
811 __be64 *ptr, int pass)
812{
813 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
814 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
815 struct gfs2_glock *gl = ip->i_gl;
816 unsigned int blks = be32_to_cpu(ld->ld_data1);
817 struct buffer_head *bh_log, *bh_ip;
818 u64 blkno;
819 u64 esc;
820 int error = 0;
821
822 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
823 return 0;
824
825 gfs2_replay_incr_blk(sdp, &start);
826 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
827 blkno = be64_to_cpu(*ptr++);
828 esc = be64_to_cpu(*ptr++);
829
830 sdp->sd_found_blocks++;
831
832 if (gfs2_revoke_check(sdp, blkno, start))
833 continue;
834
835 error = gfs2_replay_read_block(jd, start, &bh_log);
836 if (error)
837 return error;
838
839 bh_ip = gfs2_meta_new(gl, blkno);
840 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
841
842 /* Unescape */
843 if (esc) {
844 __be32 *eptr = (__be32 *)bh_ip->b_data;
845 *eptr = cpu_to_be32(GFS2_MAGIC);
846 }
847 mark_buffer_dirty(bh_ip);
848
849 brelse(bh_log);
850 brelse(bh_ip);
851
852 sdp->sd_replayed_blocks++;
853 }
854
855 return error;
856}
857
858/* FIXME: sort out accounting for log blocks etc. */
859
860static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
861{
862 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
863 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
864
865 if (error) {
866 gfs2_meta_sync(ip->i_gl);
867 return;
868 }
869 if (pass != 1)
870 return;
871
872 /* data sync? */
873 gfs2_meta_sync(ip->i_gl);
874
875 fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
876 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
877}
878
879static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
880{
881 struct list_head *head = &sdp->sd_log_le_databuf;
882 struct gfs2_bufdata *bd;
883
884 while (!list_empty(head)) {
885 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
886 list_del_init(&bd->bd_list);
887 sdp->sd_log_num_databuf--;
888 gfs2_unpin(sdp, bd->bd_bh, ai);
889 }
890 gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
891}
892
893
894const struct gfs2_log_operations gfs2_buf_lops = {
895 .lo_add = buf_lo_add,
896 .lo_before_commit = buf_lo_before_commit,
897 .lo_after_commit = buf_lo_after_commit,
898 .lo_before_scan = buf_lo_before_scan,
899 .lo_scan_elements = buf_lo_scan_elements,
900 .lo_after_scan = buf_lo_after_scan,
901 .lo_name = "buf",
902};
903
904const struct gfs2_log_operations gfs2_revoke_lops = {
905 .lo_add = revoke_lo_add,
906 .lo_before_commit = revoke_lo_before_commit,
907 .lo_after_commit = revoke_lo_after_commit,
908 .lo_before_scan = revoke_lo_before_scan,
909 .lo_scan_elements = revoke_lo_scan_elements,
910 .lo_after_scan = revoke_lo_after_scan,
911 .lo_name = "revoke",
912};
913
914const struct gfs2_log_operations gfs2_rg_lops = {
915 .lo_name = "rg",
916};
917
918const struct gfs2_log_operations gfs2_databuf_lops = {
919 .lo_add = databuf_lo_add,
920 .lo_before_commit = databuf_lo_before_commit,
921 .lo_after_commit = databuf_lo_after_commit,
922 .lo_scan_elements = databuf_lo_scan_elements,
923 .lo_after_scan = databuf_lo_after_scan,
924 .lo_name = "databuf",
925};
926
927const struct gfs2_log_operations *gfs2_log_ops[] = {
928 &gfs2_databuf_lops,
929 &gfs2_buf_lops,
930 &gfs2_rg_lops,
931 &gfs2_revoke_lops,
932 NULL,
933};
934
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_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
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, 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
239void 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
261static 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
299static 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
328static 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
355static 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
371void 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
377static 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
393static 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
406static 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
420static 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
494static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
495{
496 unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
497 unsigned int nbuf;
498 if (tr == NULL)
499 return;
500 nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
501 gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
502}
503
504static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
505{
506 struct list_head *head;
507 struct gfs2_bufdata *bd;
508
509 if (tr == NULL)
510 return;
511
512 head = &tr->tr_buf;
513 while (!list_empty(head)) {
514 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
515 list_del_init(&bd->bd_list);
516 gfs2_unpin(sdp, bd->bd_bh, tr);
517 }
518}
519
520static void buf_lo_before_scan(struct gfs2_jdesc *jd,
521 struct gfs2_log_header_host *head, int pass)
522{
523 if (pass != 0)
524 return;
525
526 jd->jd_found_blocks = 0;
527 jd->jd_replayed_blocks = 0;
528}
529
530static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
531 struct gfs2_log_descriptor *ld, __be64 *ptr,
532 int pass)
533{
534 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
535 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
536 struct gfs2_glock *gl = ip->i_gl;
537 unsigned int blks = be32_to_cpu(ld->ld_data1);
538 struct buffer_head *bh_log, *bh_ip;
539 u64 blkno;
540 int error = 0;
541
542 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
543 return 0;
544
545 gfs2_replay_incr_blk(sdp, &start);
546
547 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
548 blkno = be64_to_cpu(*ptr++);
549
550 jd->jd_found_blocks++;
551
552 if (gfs2_revoke_check(jd, blkno, start))
553 continue;
554
555 error = gfs2_replay_read_block(jd, start, &bh_log);
556 if (error)
557 return error;
558
559 bh_ip = gfs2_meta_new(gl, blkno);
560 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
561
562 if (gfs2_meta_check(sdp, bh_ip))
563 error = -EIO;
564 else
565 mark_buffer_dirty(bh_ip);
566
567 brelse(bh_log);
568 brelse(bh_ip);
569
570 if (error)
571 break;
572
573 jd->jd_replayed_blocks++;
574 }
575
576 return error;
577}
578
579/**
580 * gfs2_meta_sync - Sync all buffers associated with a glock
581 * @gl: The glock
582 *
583 */
584
585static void gfs2_meta_sync(struct gfs2_glock *gl)
586{
587 struct address_space *mapping = gfs2_glock2aspace(gl);
588 struct gfs2_sbd *sdp = gl->gl_sbd;
589 int error;
590
591 if (mapping == NULL)
592 mapping = &sdp->sd_aspace;
593
594 filemap_fdatawrite(mapping);
595 error = filemap_fdatawait(mapping);
596
597 if (error)
598 gfs2_io_error(gl->gl_sbd);
599}
600
601static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
602{
603 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
604 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
605
606 if (error) {
607 gfs2_meta_sync(ip->i_gl);
608 return;
609 }
610 if (pass != 1)
611 return;
612
613 gfs2_meta_sync(ip->i_gl);
614
615 fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
616 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
617}
618
619static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
620{
621 struct gfs2_meta_header *mh;
622 unsigned int offset;
623 struct list_head *head = &sdp->sd_log_le_revoke;
624 struct gfs2_bufdata *bd;
625 struct page *page;
626 unsigned int length;
627
628 gfs2_write_revokes(sdp);
629 if (!sdp->sd_log_num_revoke)
630 return;
631
632 length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
633 page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
634 offset = sizeof(struct gfs2_log_descriptor);
635
636 list_for_each_entry(bd, head, bd_list) {
637 sdp->sd_log_num_revoke--;
638
639 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
640
641 gfs2_log_write_page(sdp, page);
642 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
643 mh = page_address(page);
644 clear_page(mh);
645 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
646 mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
647 mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
648 offset = sizeof(struct gfs2_meta_header);
649 }
650
651 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
652 offset += sizeof(u64);
653 }
654 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
655
656 gfs2_log_write_page(sdp, page);
657}
658
659static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
660{
661 struct list_head *head = &sdp->sd_log_le_revoke;
662 struct gfs2_bufdata *bd;
663 struct gfs2_glock *gl;
664
665 while (!list_empty(head)) {
666 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
667 list_del_init(&bd->bd_list);
668 gl = bd->bd_gl;
669 atomic_dec(&gl->gl_revokes);
670 clear_bit(GLF_LFLUSH, &gl->gl_flags);
671 kmem_cache_free(gfs2_bufdata_cachep, bd);
672 }
673}
674
675static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
676 struct gfs2_log_header_host *head, int pass)
677{
678 if (pass != 0)
679 return;
680
681 jd->jd_found_revokes = 0;
682 jd->jd_replay_tail = head->lh_tail;
683}
684
685static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
686 struct gfs2_log_descriptor *ld, __be64 *ptr,
687 int pass)
688{
689 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
690 unsigned int blks = be32_to_cpu(ld->ld_length);
691 unsigned int revokes = be32_to_cpu(ld->ld_data1);
692 struct buffer_head *bh;
693 unsigned int offset;
694 u64 blkno;
695 int first = 1;
696 int error;
697
698 if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
699 return 0;
700
701 offset = sizeof(struct gfs2_log_descriptor);
702
703 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
704 error = gfs2_replay_read_block(jd, start, &bh);
705 if (error)
706 return error;
707
708 if (!first)
709 gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
710
711 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
712 blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
713
714 error = gfs2_revoke_add(jd, blkno, start);
715 if (error < 0) {
716 brelse(bh);
717 return error;
718 }
719 else if (error)
720 jd->jd_found_revokes++;
721
722 if (!--revokes)
723 break;
724 offset += sizeof(u64);
725 }
726
727 brelse(bh);
728 offset = sizeof(struct gfs2_meta_header);
729 first = 0;
730 }
731
732 return 0;
733}
734
735static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
736{
737 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
738
739 if (error) {
740 gfs2_revoke_clean(jd);
741 return;
742 }
743 if (pass != 1)
744 return;
745
746 fs_info(sdp, "jid=%u: Found %u revoke tags\n",
747 jd->jd_jid, jd->jd_found_revokes);
748
749 gfs2_revoke_clean(jd);
750}
751
752/**
753 * databuf_lo_before_commit - Scan the data buffers, writing as we go
754 *
755 */
756
757static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
758{
759 unsigned int limit = databuf_limit(sdp);
760 unsigned int nbuf;
761 if (tr == NULL)
762 return;
763 nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
764 gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
765}
766
767static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
768 struct gfs2_log_descriptor *ld,
769 __be64 *ptr, int pass)
770{
771 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
772 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
773 struct gfs2_glock *gl = ip->i_gl;
774 unsigned int blks = be32_to_cpu(ld->ld_data1);
775 struct buffer_head *bh_log, *bh_ip;
776 u64 blkno;
777 u64 esc;
778 int error = 0;
779
780 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
781 return 0;
782
783 gfs2_replay_incr_blk(sdp, &start);
784 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
785 blkno = be64_to_cpu(*ptr++);
786 esc = be64_to_cpu(*ptr++);
787
788 jd->jd_found_blocks++;
789
790 if (gfs2_revoke_check(jd, blkno, start))
791 continue;
792
793 error = gfs2_replay_read_block(jd, start, &bh_log);
794 if (error)
795 return error;
796
797 bh_ip = gfs2_meta_new(gl, blkno);
798 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
799
800 /* Unescape */
801 if (esc) {
802 __be32 *eptr = (__be32 *)bh_ip->b_data;
803 *eptr = cpu_to_be32(GFS2_MAGIC);
804 }
805 mark_buffer_dirty(bh_ip);
806
807 brelse(bh_log);
808 brelse(bh_ip);
809
810 jd->jd_replayed_blocks++;
811 }
812
813 return error;
814}
815
816/* FIXME: sort out accounting for log blocks etc. */
817
818static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
819{
820 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
821 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
822
823 if (error) {
824 gfs2_meta_sync(ip->i_gl);
825 return;
826 }
827 if (pass != 1)
828 return;
829
830 /* data sync? */
831 gfs2_meta_sync(ip->i_gl);
832
833 fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
834 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
835}
836
837static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
838{
839 struct list_head *head;
840 struct gfs2_bufdata *bd;
841
842 if (tr == NULL)
843 return;
844
845 head = &tr->tr_databuf;
846 while (!list_empty(head)) {
847 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
848 list_del_init(&bd->bd_list);
849 gfs2_unpin(sdp, bd->bd_bh, tr);
850 }
851}
852
853
854const struct gfs2_log_operations gfs2_buf_lops = {
855 .lo_before_commit = buf_lo_before_commit,
856 .lo_after_commit = buf_lo_after_commit,
857 .lo_before_scan = buf_lo_before_scan,
858 .lo_scan_elements = buf_lo_scan_elements,
859 .lo_after_scan = buf_lo_after_scan,
860 .lo_name = "buf",
861};
862
863const struct gfs2_log_operations gfs2_revoke_lops = {
864 .lo_before_commit = revoke_lo_before_commit,
865 .lo_after_commit = revoke_lo_after_commit,
866 .lo_before_scan = revoke_lo_before_scan,
867 .lo_scan_elements = revoke_lo_scan_elements,
868 .lo_after_scan = revoke_lo_after_scan,
869 .lo_name = "revoke",
870};
871
872const struct gfs2_log_operations gfs2_databuf_lops = {
873 .lo_before_commit = databuf_lo_before_commit,
874 .lo_after_commit = databuf_lo_after_commit,
875 .lo_scan_elements = databuf_lo_scan_elements,
876 .lo_after_scan = databuf_lo_after_scan,
877 .lo_name = "databuf",
878};
879
880const struct gfs2_log_operations *gfs2_log_ops[] = {
881 &gfs2_databuf_lops,
882 &gfs2_buf_lops,
883 &gfs2_revoke_lops,
884 NULL,
885};
886