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#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
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 */
6
7#include <linux/sched.h>
8#include <linux/slab.h>
9#include <linux/spinlock.h>
10#include <linux/completion.h>
11#include <linux/buffer_head.h>
12#include <linux/mempool.h>
13#include <linux/gfs2_ondisk.h>
14#include <linux/bio.h>
15#include <linux/fs.h>
16#include <linux/list_sort.h>
17#include <linux/blkdev.h>
18
19#include "bmap.h"
20#include "dir.h"
21#include "gfs2.h"
22#include "incore.h"
23#include "inode.h"
24#include "glock.h"
25#include "glops.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_wd(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 rgrp_lock_local(rgd);
80 if (bi->bi_clone == NULL)
81 goto out;
82 if (sdp->sd_args.ar_discard)
83 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
84 memcpy(bi->bi_clone + bi->bi_offset,
85 bd->bd_bh->b_data + bi->bi_offset, bi->bi_bytes);
86 clear_bit(GBF_FULL, &bi->bi_flags);
87 rgd->rd_free_clone = rgd->rd_free;
88 BUG_ON(rgd->rd_free_clone < rgd->rd_reserved);
89 rgd->rd_extfail_pt = rgd->rd_free;
90
91out:
92 rgrp_unlock_local(rgd);
93}
94
95/**
96 * gfs2_unpin - Unpin a buffer
97 * @sdp: the filesystem the buffer belongs to
98 * @bh: The buffer to unpin
99 * @tr: The system transaction being flushed
100 */
101
102static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
103 struct gfs2_trans *tr)
104{
105 struct gfs2_bufdata *bd = bh->b_private;
106
107 BUG_ON(!buffer_uptodate(bh));
108 BUG_ON(!buffer_pinned(bh));
109
110 lock_buffer(bh);
111 mark_buffer_dirty(bh);
112 clear_buffer_pinned(bh);
113
114 if (buffer_is_rgrp(bd))
115 maybe_release_space(bd);
116
117 spin_lock(&sdp->sd_ail_lock);
118 if (bd->bd_tr) {
119 list_del(&bd->bd_ail_st_list);
120 brelse(bh);
121 } else {
122 struct gfs2_glock *gl = bd->bd_gl;
123 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
124 atomic_inc(&gl->gl_ail_count);
125 }
126 bd->bd_tr = tr;
127 list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list);
128 spin_unlock(&sdp->sd_ail_lock);
129
130 clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
131 trace_gfs2_pin(bd, 0);
132 unlock_buffer(bh);
133 atomic_dec(&sdp->sd_log_pinned);
134}
135
136void gfs2_log_incr_head(struct gfs2_sbd *sdp)
137{
138 BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
139 (sdp->sd_log_flush_head != sdp->sd_log_head));
140
141 if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks)
142 sdp->sd_log_flush_head = 0;
143}
144
145u64 gfs2_log_bmap(struct gfs2_jdesc *jd, unsigned int lblock)
146{
147 struct gfs2_journal_extent *je;
148
149 list_for_each_entry(je, &jd->extent_list, list) {
150 if (lblock >= je->lblock && lblock < je->lblock + je->blocks)
151 return je->dblock + lblock - je->lblock;
152 }
153
154 return -1;
155}
156
157/**
158 * gfs2_end_log_write_bh - end log write of pagecache data with buffers
159 * @sdp: The superblock
160 * @bvec: The bio_vec
161 * @error: The i/o status
162 *
163 * This finds the relevant buffers and unlocks them and sets the
164 * error flag according to the status of the i/o request. This is
165 * used when the log is writing data which has an in-place version
166 * that is pinned in the pagecache.
167 */
168
169static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp,
170 struct bio_vec *bvec,
171 blk_status_t 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 mark_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 *
196 * Each bio_vec contains either data from the pagecache or data
197 * relating to the log itself. Here we iterate over the bio_vec
198 * array, processing both kinds of data.
199 *
200 */
201
202static void gfs2_end_log_write(struct bio *bio)
203{
204 struct gfs2_sbd *sdp = bio->bi_private;
205 struct bio_vec *bvec;
206 struct page *page;
207 struct bvec_iter_all iter_all;
208
209 if (bio->bi_status) {
210 if (!cmpxchg(&sdp->sd_log_error, 0, (int)bio->bi_status))
211 fs_err(sdp, "Error %d writing to journal, jid=%u\n",
212 bio->bi_status, sdp->sd_jdesc->jd_jid);
213 gfs2_withdraw_delayed(sdp);
214 /* prevent more writes to the journal */
215 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
216 wake_up(&sdp->sd_logd_waitq);
217 }
218
219 bio_for_each_segment_all(bvec, bio, iter_all) {
220 page = bvec->bv_page;
221 if (page_has_buffers(page))
222 gfs2_end_log_write_bh(sdp, bvec, bio->bi_status);
223 else
224 mempool_free(page, gfs2_page_pool);
225 }
226
227 bio_put(bio);
228 if (atomic_dec_and_test(&sdp->sd_log_in_flight))
229 wake_up(&sdp->sd_log_flush_wait);
230}
231
232/**
233 * gfs2_log_submit_bio - Submit any pending log bio
234 * @biop: Address of the bio pointer
235 * @opf: REQ_OP | op_flags
236 *
237 * Submit any pending part-built or full bio to the block device. If
238 * there is no pending bio, then this is a no-op.
239 */
240
241void gfs2_log_submit_bio(struct bio **biop, blk_opf_t opf)
242{
243 struct bio *bio = *biop;
244 if (bio) {
245 struct gfs2_sbd *sdp = bio->bi_private;
246 atomic_inc(&sdp->sd_log_in_flight);
247 bio->bi_opf = opf;
248 submit_bio(bio);
249 *biop = NULL;
250 }
251}
252
253/**
254 * gfs2_log_alloc_bio - Allocate a bio
255 * @sdp: The super block
256 * @blkno: The device block number we want to write to
257 * @end_io: The bi_end_io callback
258 *
259 * Allocate a new bio, initialize it with the given parameters and return it.
260 *
261 * Returns: The newly allocated bio
262 */
263
264static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno,
265 bio_end_io_t *end_io)
266{
267 struct super_block *sb = sdp->sd_vfs;
268 struct bio *bio = bio_alloc(sb->s_bdev, BIO_MAX_VECS, 0, GFP_NOIO);
269
270 bio->bi_iter.bi_sector = blkno << sdp->sd_fsb2bb_shift;
271 bio->bi_end_io = end_io;
272 bio->bi_private = sdp;
273
274 return bio;
275}
276
277/**
278 * gfs2_log_get_bio - Get cached log bio, or allocate a new one
279 * @sdp: The super block
280 * @blkno: The device block number we want to write to
281 * @biop: The bio to get or allocate
282 * @op: REQ_OP
283 * @end_io: The bi_end_io callback
284 * @flush: Always flush the current bio and allocate a new one?
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 no 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 struct bio **biop, enum req_op op,
296 bio_end_io_t *end_io, bool flush)
297{
298 struct bio *bio = *biop;
299
300 if (bio) {
301 u64 nblk;
302
303 nblk = bio_end_sector(bio);
304 nblk >>= sdp->sd_fsb2bb_shift;
305 if (blkno == nblk && !flush)
306 return bio;
307 gfs2_log_submit_bio(biop, op);
308 }
309
310 *biop = gfs2_log_alloc_bio(sdp, blkno, end_io);
311 return *biop;
312}
313
314/**
315 * gfs2_log_write - write to log
316 * @sdp: the filesystem
317 * @jd: The journal descriptor
318 * @page: the page to write
319 * @size: the size of the data to write
320 * @offset: the offset within the page
321 * @blkno: block number of the log entry
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
328void gfs2_log_write(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
329 struct page *page, unsigned size, unsigned offset,
330 u64 blkno)
331{
332 struct bio *bio;
333 int ret;
334
335 bio = gfs2_log_get_bio(sdp, blkno, &jd->jd_log_bio, REQ_OP_WRITE,
336 gfs2_end_log_write, false);
337 ret = bio_add_page(bio, page, size, offset);
338 if (ret == 0) {
339 bio = gfs2_log_get_bio(sdp, blkno, &jd->jd_log_bio,
340 REQ_OP_WRITE, gfs2_end_log_write, true);
341 ret = bio_add_page(bio, page, size, offset);
342 WARN_ON(ret == 0);
343 }
344}
345
346/**
347 * gfs2_log_write_bh - write a buffer's content to the log
348 * @sdp: The super block
349 * @bh: The buffer pointing to the in-place location
350 *
351 * This writes the content of the buffer to the next available location
352 * in the log. The buffer will be unlocked once the i/o to the log has
353 * completed.
354 */
355
356static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
357{
358 u64 dblock;
359
360 dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head);
361 gfs2_log_incr_head(sdp);
362 gfs2_log_write(sdp, sdp->sd_jdesc, bh->b_page, bh->b_size,
363 bh_offset(bh), dblock);
364}
365
366/**
367 * gfs2_log_write_page - write one block stored in a page, into the log
368 * @sdp: The superblock
369 * @page: The struct page
370 *
371 * This writes the first block-sized part of the page into the log. Note
372 * that the page must have been allocated from the gfs2_page_pool mempool
373 * and that after this has been called, ownership has been transferred and
374 * the page may be freed at any time.
375 */
376
377static void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
378{
379 struct super_block *sb = sdp->sd_vfs;
380 u64 dblock;
381
382 dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head);
383 gfs2_log_incr_head(sdp);
384 gfs2_log_write(sdp, sdp->sd_jdesc, page, sb->s_blocksize, 0, dblock);
385}
386
387/**
388 * gfs2_end_log_read - end I/O callback for reads from the log
389 * @bio: The bio
390 *
391 * Simply unlock the pages in the bio. The main thread will wait on them and
392 * process them in order as necessary.
393 */
394static void gfs2_end_log_read(struct bio *bio)
395{
396 int error = blk_status_to_errno(bio->bi_status);
397 struct folio_iter fi;
398
399 bio_for_each_folio_all(fi, bio) {
400 /* We're abusing wb_err to get the error to gfs2_find_jhead */
401 filemap_set_wb_err(fi.folio->mapping, error);
402 folio_end_read(fi.folio, !error);
403 }
404
405 bio_put(bio);
406}
407
408/**
409 * gfs2_jhead_pg_srch - Look for the journal head in a given page.
410 * @jd: The journal descriptor
411 * @head: The journal head to start from
412 * @page: The page to look in
413 *
414 * Returns: 1 if found, 0 otherwise.
415 */
416
417static bool gfs2_jhead_pg_srch(struct gfs2_jdesc *jd,
418 struct gfs2_log_header_host *head,
419 struct page *page)
420{
421 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
422 struct gfs2_log_header_host lh;
423 void *kaddr;
424 unsigned int offset;
425 bool ret = false;
426
427 kaddr = kmap_local_page(page);
428 for (offset = 0; offset < PAGE_SIZE; offset += sdp->sd_sb.sb_bsize) {
429 if (!__get_log_header(sdp, kaddr + offset, 0, &lh)) {
430 if (lh.lh_sequence >= head->lh_sequence)
431 *head = lh;
432 else {
433 ret = true;
434 break;
435 }
436 }
437 }
438 kunmap_local(kaddr);
439 return ret;
440}
441
442/**
443 * gfs2_jhead_process_page - Search/cleanup a page
444 * @jd: The journal descriptor
445 * @index: Index of the page to look into
446 * @head: The journal head to start from
447 * @done: If set, perform only cleanup, else search and set if found.
448 *
449 * Find the folio with 'index' in the journal's mapping. Search the folio for
450 * the journal head if requested (cleanup == false). Release refs on the
451 * folio so the page cache can reclaim it. We grabbed a
452 * reference on this folio twice, first when we did a grab_cache_page()
453 * to obtain the folio to add it to the bio and second when we do a
454 * filemap_get_folio() here to get the folio to wait on while I/O on it is being
455 * completed.
456 * This function is also used to free up a folio we might've grabbed but not
457 * used. Maybe we added it to a bio, but not submitted it for I/O. Or we
458 * submitted the I/O, but we already found the jhead so we only need to drop
459 * our references to the folio.
460 */
461
462static void gfs2_jhead_process_page(struct gfs2_jdesc *jd, unsigned long index,
463 struct gfs2_log_header_host *head,
464 bool *done)
465{
466 struct folio *folio;
467
468 folio = filemap_get_folio(jd->jd_inode->i_mapping, index);
469
470 folio_wait_locked(folio);
471 if (!folio_test_uptodate(folio))
472 *done = true;
473
474 if (!*done)
475 *done = gfs2_jhead_pg_srch(jd, head, &folio->page);
476
477 /* filemap_get_folio() and the earlier grab_cache_page() */
478 folio_put_refs(folio, 2);
479}
480
481static struct bio *gfs2_chain_bio(struct bio *prev, unsigned int nr_iovecs)
482{
483 struct bio *new;
484
485 new = bio_alloc(prev->bi_bdev, nr_iovecs, prev->bi_opf, GFP_NOIO);
486 bio_clone_blkg_association(new, prev);
487 new->bi_iter.bi_sector = bio_end_sector(prev);
488 bio_chain(new, prev);
489 submit_bio(prev);
490 return new;
491}
492
493/**
494 * gfs2_find_jhead - find the head of a log
495 * @jd: The journal descriptor
496 * @head: The log descriptor for the head of the log is returned here
497 * @keep_cache: If set inode pages will not be truncated
498 *
499 * Do a search of a journal by reading it in large chunks using bios and find
500 * the valid log entry with the highest sequence number. (i.e. the log head)
501 *
502 * Returns: 0 on success, errno otherwise
503 */
504int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head,
505 bool keep_cache)
506{
507 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
508 struct address_space *mapping = jd->jd_inode->i_mapping;
509 unsigned int block = 0, blocks_submitted = 0, blocks_read = 0;
510 unsigned int bsize = sdp->sd_sb.sb_bsize, off;
511 unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift;
512 unsigned int shift = PAGE_SHIFT - bsize_shift;
513 unsigned int max_blocks = 2 * 1024 * 1024 >> bsize_shift;
514 struct gfs2_journal_extent *je;
515 int sz, ret = 0;
516 struct bio *bio = NULL;
517 struct page *page = NULL;
518 bool done = false;
519 errseq_t since;
520
521 memset(head, 0, sizeof(*head));
522 if (list_empty(&jd->extent_list))
523 gfs2_map_journal_extents(sdp, jd);
524
525 since = filemap_sample_wb_err(mapping);
526 list_for_each_entry(je, &jd->extent_list, list) {
527 u64 dblock = je->dblock;
528
529 for (; block < je->lblock + je->blocks; block++, dblock++) {
530 if (!page) {
531 page = grab_cache_page(mapping, block >> shift);
532 if (!page) {
533 ret = -ENOMEM;
534 done = true;
535 goto out;
536 }
537 off = 0;
538 }
539
540 if (bio && (off || block < blocks_submitted + max_blocks)) {
541 sector_t sector = dblock << sdp->sd_fsb2bb_shift;
542
543 if (bio_end_sector(bio) == sector) {
544 sz = bio_add_page(bio, page, bsize, off);
545 if (sz == bsize)
546 goto block_added;
547 }
548 if (off) {
549 unsigned int blocks =
550 (PAGE_SIZE - off) >> bsize_shift;
551
552 bio = gfs2_chain_bio(bio, blocks);
553 goto add_block_to_new_bio;
554 }
555 }
556
557 if (bio) {
558 blocks_submitted = block;
559 submit_bio(bio);
560 }
561
562 bio = gfs2_log_alloc_bio(sdp, dblock, gfs2_end_log_read);
563 bio->bi_opf = REQ_OP_READ;
564add_block_to_new_bio:
565 sz = bio_add_page(bio, page, bsize, off);
566 BUG_ON(sz != bsize);
567block_added:
568 off += bsize;
569 if (off == PAGE_SIZE)
570 page = NULL;
571 if (blocks_submitted <= blocks_read + max_blocks) {
572 /* Keep at least one bio in flight */
573 continue;
574 }
575
576 gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
577 blocks_read += PAGE_SIZE >> bsize_shift;
578 if (done)
579 goto out; /* found */
580 }
581 }
582
583out:
584 if (bio)
585 submit_bio(bio);
586 while (blocks_read < block) {
587 gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
588 blocks_read += PAGE_SIZE >> bsize_shift;
589 }
590
591 if (!ret)
592 ret = filemap_check_wb_err(mapping, since);
593
594 if (!keep_cache)
595 truncate_inode_pages(mapping, 0);
596
597 return ret;
598}
599
600static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
601 u32 ld_length, u32 ld_data1)
602{
603 struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
604 struct gfs2_log_descriptor *ld = page_address(page);
605 clear_page(ld);
606 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
607 ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
608 ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
609 ld->ld_type = cpu_to_be32(ld_type);
610 ld->ld_length = cpu_to_be32(ld_length);
611 ld->ld_data1 = cpu_to_be32(ld_data1);
612 ld->ld_data2 = 0;
613 return page;
614}
615
616static void gfs2_check_magic(struct buffer_head *bh)
617{
618 void *kaddr;
619 __be32 *ptr;
620
621 clear_buffer_escaped(bh);
622 kaddr = kmap_local_page(bh->b_page);
623 ptr = kaddr + bh_offset(bh);
624 if (*ptr == cpu_to_be32(GFS2_MAGIC))
625 set_buffer_escaped(bh);
626 kunmap_local(kaddr);
627}
628
629static int blocknr_cmp(void *priv, const struct list_head *a,
630 const struct list_head *b)
631{
632 struct gfs2_bufdata *bda, *bdb;
633
634 bda = list_entry(a, struct gfs2_bufdata, bd_list);
635 bdb = list_entry(b, struct gfs2_bufdata, bd_list);
636
637 if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
638 return -1;
639 if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
640 return 1;
641 return 0;
642}
643
644static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
645 unsigned int total, struct list_head *blist,
646 bool is_databuf)
647{
648 struct gfs2_log_descriptor *ld;
649 struct gfs2_bufdata *bd1 = NULL, *bd2;
650 struct page *page;
651 unsigned int num;
652 unsigned n;
653 __be64 *ptr;
654
655 gfs2_log_lock(sdp);
656 list_sort(NULL, blist, blocknr_cmp);
657 bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
658 while(total) {
659 num = total;
660 if (total > limit)
661 num = limit;
662 gfs2_log_unlock(sdp);
663 page = gfs2_get_log_desc(sdp,
664 is_databuf ? GFS2_LOG_DESC_JDATA :
665 GFS2_LOG_DESC_METADATA, num + 1, num);
666 ld = page_address(page);
667 gfs2_log_lock(sdp);
668 ptr = (__be64 *)(ld + 1);
669
670 n = 0;
671 list_for_each_entry_continue(bd1, blist, bd_list) {
672 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
673 if (is_databuf) {
674 gfs2_check_magic(bd1->bd_bh);
675 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
676 }
677 if (++n >= num)
678 break;
679 }
680
681 gfs2_log_unlock(sdp);
682 gfs2_log_write_page(sdp, page);
683 gfs2_log_lock(sdp);
684
685 n = 0;
686 list_for_each_entry_continue(bd2, blist, bd_list) {
687 get_bh(bd2->bd_bh);
688 gfs2_log_unlock(sdp);
689 lock_buffer(bd2->bd_bh);
690
691 if (buffer_escaped(bd2->bd_bh)) {
692 void *p;
693
694 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
695 p = page_address(page);
696 memcpy_from_page(p, page, bh_offset(bd2->bd_bh), bd2->bd_bh->b_size);
697 *(__be32 *)p = 0;
698 clear_buffer_escaped(bd2->bd_bh);
699 unlock_buffer(bd2->bd_bh);
700 brelse(bd2->bd_bh);
701 gfs2_log_write_page(sdp, page);
702 } else {
703 gfs2_log_write_bh(sdp, bd2->bd_bh);
704 }
705 gfs2_log_lock(sdp);
706 if (++n >= num)
707 break;
708 }
709
710 BUG_ON(total < num);
711 total -= num;
712 }
713 gfs2_log_unlock(sdp);
714}
715
716static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
717{
718 unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
719 unsigned int nbuf;
720 if (tr == NULL)
721 return;
722 nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
723 gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
724}
725
726static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
727{
728 struct list_head *head;
729 struct gfs2_bufdata *bd;
730
731 if (tr == NULL)
732 return;
733
734 head = &tr->tr_buf;
735 while (!list_empty(head)) {
736 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
737 list_del_init(&bd->bd_list);
738 gfs2_unpin(sdp, bd->bd_bh, tr);
739 }
740}
741
742static void buf_lo_before_scan(struct gfs2_jdesc *jd,
743 struct gfs2_log_header_host *head, int pass)
744{
745 if (pass != 0)
746 return;
747
748 jd->jd_found_blocks = 0;
749 jd->jd_replayed_blocks = 0;
750}
751
752#define obsolete_rgrp_replay \
753"Replaying 0x%llx from jid=%d/0x%llx but we already have a bh!\n"
754#define obsolete_rgrp_replay2 \
755"busy:%d, pinned:%d rg_gen:0x%llx, j_gen:0x%llx\n"
756
757static void obsolete_rgrp(struct gfs2_jdesc *jd, struct buffer_head *bh_log,
758 u64 blkno)
759{
760 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
761 struct gfs2_rgrpd *rgd;
762 struct gfs2_rgrp *jrgd = (struct gfs2_rgrp *)bh_log->b_data;
763
764 rgd = gfs2_blk2rgrpd(sdp, blkno, false);
765 if (rgd && rgd->rd_addr == blkno &&
766 rgd->rd_bits && rgd->rd_bits->bi_bh) {
767 fs_info(sdp, obsolete_rgrp_replay, (unsigned long long)blkno,
768 jd->jd_jid, bh_log->b_blocknr);
769 fs_info(sdp, obsolete_rgrp_replay2,
770 buffer_busy(rgd->rd_bits->bi_bh) ? 1 : 0,
771 buffer_pinned(rgd->rd_bits->bi_bh),
772 rgd->rd_igeneration,
773 be64_to_cpu(jrgd->rg_igeneration));
774 gfs2_dump_glock(NULL, rgd->rd_gl, true);
775 }
776}
777
778static int buf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
779 struct gfs2_log_descriptor *ld, __be64 *ptr,
780 int pass)
781{
782 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
783 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
784 struct gfs2_glock *gl = ip->i_gl;
785 unsigned int blks = be32_to_cpu(ld->ld_data1);
786 struct buffer_head *bh_log, *bh_ip;
787 u64 blkno;
788 int error = 0;
789
790 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
791 return 0;
792
793 gfs2_replay_incr_blk(jd, &start);
794
795 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
796 blkno = be64_to_cpu(*ptr++);
797
798 jd->jd_found_blocks++;
799
800 if (gfs2_revoke_check(jd, blkno, start))
801 continue;
802
803 error = gfs2_replay_read_block(jd, start, &bh_log);
804 if (error)
805 return error;
806
807 bh_ip = gfs2_meta_new(gl, blkno);
808 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
809
810 if (gfs2_meta_check(sdp, bh_ip))
811 error = -EIO;
812 else {
813 struct gfs2_meta_header *mh =
814 (struct gfs2_meta_header *)bh_ip->b_data;
815
816 if (mh->mh_type == cpu_to_be32(GFS2_METATYPE_RG))
817 obsolete_rgrp(jd, bh_log, blkno);
818
819 mark_buffer_dirty(bh_ip);
820 }
821 brelse(bh_log);
822 brelse(bh_ip);
823
824 if (error)
825 break;
826
827 jd->jd_replayed_blocks++;
828 }
829
830 return error;
831}
832
833static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
834{
835 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
836 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
837
838 if (error) {
839 gfs2_inode_metasync(ip->i_gl);
840 return;
841 }
842 if (pass != 1)
843 return;
844
845 gfs2_inode_metasync(ip->i_gl);
846
847 fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
848 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
849}
850
851static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
852{
853 struct gfs2_meta_header *mh;
854 unsigned int offset;
855 struct list_head *head = &sdp->sd_log_revokes;
856 struct gfs2_bufdata *bd;
857 struct page *page;
858 unsigned int length;
859
860 gfs2_flush_revokes(sdp);
861 if (!sdp->sd_log_num_revoke)
862 return;
863
864 length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke);
865 page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
866 offset = sizeof(struct gfs2_log_descriptor);
867
868 list_for_each_entry(bd, head, bd_list) {
869 sdp->sd_log_num_revoke--;
870
871 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
872 gfs2_log_write_page(sdp, page);
873 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
874 mh = page_address(page);
875 clear_page(mh);
876 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
877 mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
878 mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
879 offset = sizeof(struct gfs2_meta_header);
880 }
881
882 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
883 offset += sizeof(u64);
884 }
885 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
886
887 gfs2_log_write_page(sdp, page);
888}
889
890void gfs2_drain_revokes(struct gfs2_sbd *sdp)
891{
892 struct list_head *head = &sdp->sd_log_revokes;
893 struct gfs2_bufdata *bd;
894 struct gfs2_glock *gl;
895
896 while (!list_empty(head)) {
897 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
898 list_del_init(&bd->bd_list);
899 gl = bd->bd_gl;
900 gfs2_glock_remove_revoke(gl);
901 kmem_cache_free(gfs2_bufdata_cachep, bd);
902 }
903}
904
905static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
906{
907 gfs2_drain_revokes(sdp);
908}
909
910static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
911 struct gfs2_log_header_host *head, int pass)
912{
913 if (pass != 0)
914 return;
915
916 jd->jd_found_revokes = 0;
917 jd->jd_replay_tail = head->lh_tail;
918}
919
920static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
921 struct gfs2_log_descriptor *ld, __be64 *ptr,
922 int pass)
923{
924 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
925 unsigned int blks = be32_to_cpu(ld->ld_length);
926 unsigned int revokes = be32_to_cpu(ld->ld_data1);
927 struct buffer_head *bh;
928 unsigned int offset;
929 u64 blkno;
930 int first = 1;
931 int error;
932
933 if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
934 return 0;
935
936 offset = sizeof(struct gfs2_log_descriptor);
937
938 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
939 error = gfs2_replay_read_block(jd, start, &bh);
940 if (error)
941 return error;
942
943 if (!first)
944 gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
945
946 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
947 blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
948
949 error = gfs2_revoke_add(jd, blkno, start);
950 if (error < 0) {
951 brelse(bh);
952 return error;
953 }
954 else if (error)
955 jd->jd_found_revokes++;
956
957 if (!--revokes)
958 break;
959 offset += sizeof(u64);
960 }
961
962 brelse(bh);
963 offset = sizeof(struct gfs2_meta_header);
964 first = 0;
965 }
966
967 return 0;
968}
969
970static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
971{
972 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
973
974 if (error) {
975 gfs2_revoke_clean(jd);
976 return;
977 }
978 if (pass != 1)
979 return;
980
981 fs_info(sdp, "jid=%u: Found %u revoke tags\n",
982 jd->jd_jid, jd->jd_found_revokes);
983
984 gfs2_revoke_clean(jd);
985}
986
987/**
988 * databuf_lo_before_commit - Scan the data buffers, writing as we go
989 * @sdp: The filesystem
990 * @tr: The system transaction being flushed
991 */
992
993static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
994{
995 unsigned int limit = databuf_limit(sdp);
996 unsigned int nbuf;
997 if (tr == NULL)
998 return;
999 nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
1000 gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
1001}
1002
1003static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
1004 struct gfs2_log_descriptor *ld,
1005 __be64 *ptr, int pass)
1006{
1007 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1008 struct gfs2_glock *gl = ip->i_gl;
1009 unsigned int blks = be32_to_cpu(ld->ld_data1);
1010 struct buffer_head *bh_log, *bh_ip;
1011 u64 blkno;
1012 u64 esc;
1013 int error = 0;
1014
1015 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
1016 return 0;
1017
1018 gfs2_replay_incr_blk(jd, &start);
1019 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
1020 blkno = be64_to_cpu(*ptr++);
1021 esc = be64_to_cpu(*ptr++);
1022
1023 jd->jd_found_blocks++;
1024
1025 if (gfs2_revoke_check(jd, blkno, start))
1026 continue;
1027
1028 error = gfs2_replay_read_block(jd, start, &bh_log);
1029 if (error)
1030 return error;
1031
1032 bh_ip = gfs2_meta_new(gl, blkno);
1033 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
1034
1035 /* Unescape */
1036 if (esc) {
1037 __be32 *eptr = (__be32 *)bh_ip->b_data;
1038 *eptr = cpu_to_be32(GFS2_MAGIC);
1039 }
1040 mark_buffer_dirty(bh_ip);
1041
1042 brelse(bh_log);
1043 brelse(bh_ip);
1044
1045 jd->jd_replayed_blocks++;
1046 }
1047
1048 return error;
1049}
1050
1051/* FIXME: sort out accounting for log blocks etc. */
1052
1053static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
1054{
1055 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1056 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
1057
1058 if (error) {
1059 gfs2_inode_metasync(ip->i_gl);
1060 return;
1061 }
1062 if (pass != 1)
1063 return;
1064
1065 /* data sync? */
1066 gfs2_inode_metasync(ip->i_gl);
1067
1068 fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
1069 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
1070}
1071
1072static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1073{
1074 struct list_head *head;
1075 struct gfs2_bufdata *bd;
1076
1077 if (tr == NULL)
1078 return;
1079
1080 head = &tr->tr_databuf;
1081 while (!list_empty(head)) {
1082 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
1083 list_del_init(&bd->bd_list);
1084 gfs2_unpin(sdp, bd->bd_bh, tr);
1085 }
1086}
1087
1088
1089static const struct gfs2_log_operations gfs2_buf_lops = {
1090 .lo_before_commit = buf_lo_before_commit,
1091 .lo_after_commit = buf_lo_after_commit,
1092 .lo_before_scan = buf_lo_before_scan,
1093 .lo_scan_elements = buf_lo_scan_elements,
1094 .lo_after_scan = buf_lo_after_scan,
1095 .lo_name = "buf",
1096};
1097
1098static const struct gfs2_log_operations gfs2_revoke_lops = {
1099 .lo_before_commit = revoke_lo_before_commit,
1100 .lo_after_commit = revoke_lo_after_commit,
1101 .lo_before_scan = revoke_lo_before_scan,
1102 .lo_scan_elements = revoke_lo_scan_elements,
1103 .lo_after_scan = revoke_lo_after_scan,
1104 .lo_name = "revoke",
1105};
1106
1107static const struct gfs2_log_operations gfs2_databuf_lops = {
1108 .lo_before_commit = databuf_lo_before_commit,
1109 .lo_after_commit = databuf_lo_after_commit,
1110 .lo_scan_elements = databuf_lo_scan_elements,
1111 .lo_after_scan = databuf_lo_after_scan,
1112 .lo_name = "databuf",
1113};
1114
1115const struct gfs2_log_operations *gfs2_log_ops[] = {
1116 &gfs2_databuf_lops,
1117 &gfs2_buf_lops,
1118 &gfs2_revoke_lops,
1119 NULL,
1120};
1121