]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blob - fs/reiserfs/journal.c
projects / mirror_ubuntu-hirsute-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
powerpc/cacheinfo: Remove double free
[mirror_ubuntu-hirsute-kernel.git] / fs / reiserfs / journal.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Write ahead logging implementation copyright Chris Mason 2000
4 *
5 * The background commits make this code very interrelated, and
6 * overly complex. I need to rethink things a bit....The major players:
7 *
8 * journal_begin -- call with the number of blocks you expect to log.
9 * If the current transaction is too
10 * old, it will block until the current transaction is
11 * finished, and then start a new one.
12 * Usually, your transaction will get joined in with
13 * previous ones for speed.
14 *
15 * journal_join -- same as journal_begin, but won't block on the current
16 * transaction regardless of age. Don't ever call
17 * this. Ever. There are only two places it should be
18 * called from, and they are both inside this file.
19 *
20 * journal_mark_dirty -- adds blocks into this transaction. clears any flags
21 * that might make them get sent to disk
22 * and then marks them BH_JDirty. Puts the buffer head
23 * into the current transaction hash.
24 *
25 * journal_end -- if the current transaction is batchable, it does nothing
26 * otherwise, it could do an async/synchronous commit, or
27 * a full flush of all log and real blocks in the
28 * transaction.
29 *
30 * flush_old_commits -- if the current transaction is too old, it is ended and
31 * commit blocks are sent to disk. Forces commit blocks
32 * to disk for all backgrounded commits that have been
33 * around too long.
34 * -- Note, if you call this as an immediate flush from
35 * from within kupdate, it will ignore the immediate flag
36 */
37
38 #include <linux/time.h>
39 #include <linux/semaphore.h>
40 #include <linux/vmalloc.h>
41 #include "reiserfs.h"
42 #include <linux/kernel.h>
43 #include <linux/errno.h>
44 #include <linux/fcntl.h>
45 #include <linux/stat.h>
46 #include <linux/string.h>
47 #include <linux/buffer_head.h>
48 #include <linux/workqueue.h>
49 #include <linux/writeback.h>
50 #include <linux/blkdev.h>
51 #include <linux/backing-dev.h>
52 #include <linux/uaccess.h>
53 #include <linux/slab.h>
54
55
56 /* gets a struct reiserfs_journal_list * from a list head */
57 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
58 j_list))
59 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
60 j_working_list))
61
62 /* must be correct to keep the desc and commit structs at 4k */
63 #define JOURNAL_TRANS_HALF 1018
64 #define BUFNR 64 /*read ahead */
65
66 /* cnode stat bits. Move these into reiserfs_fs.h */
67
68 /* this block was freed, and can't be written. */
69 #define BLOCK_FREED 2
70 /* this block was freed during this transaction, and can't be written */
71 #define BLOCK_FREED_HOLDER 3
72
73 /* used in flush_journal_list */
74 #define BLOCK_NEEDS_FLUSH 4
75 #define BLOCK_DIRTIED 5
76
77 /* journal list state bits */
78 #define LIST_TOUCHED 1
79 #define LIST_DIRTY 2
80 #define LIST_COMMIT_PENDING 4 /* someone will commit this list */
81
82 /* flags for do_journal_end */
83 #define FLUSH_ALL 1 /* flush commit and real blocks */
84 #define COMMIT_NOW 2 /* end and commit this transaction */
85 #define WAIT 4 /* wait for the log blocks to hit the disk */
86
87 static int do_journal_end(struct reiserfs_transaction_handle *, int flags);
88 static int flush_journal_list(struct super_block *s,
89 struct reiserfs_journal_list *jl, int flushall);
90 static int flush_commit_list(struct super_block *s,
91 struct reiserfs_journal_list *jl, int flushall);
92 static int can_dirty(struct reiserfs_journal_cnode *cn);
93 static int journal_join(struct reiserfs_transaction_handle *th,
94 struct super_block *sb);
95 static void release_journal_dev(struct super_block *super,
96 struct reiserfs_journal *journal);
97 static int dirty_one_transaction(struct super_block *s,
98 struct reiserfs_journal_list *jl);
99 static void flush_async_commits(struct work_struct *work);
100 static void queue_log_writer(struct super_block *s);
101
102 /* values for join in do_journal_begin_r */
103 enum {
104 JBEGIN_REG = 0, /* regular journal begin */
105 /* join the running transaction if at all possible */
106 JBEGIN_JOIN = 1,
107 /* called from cleanup code, ignores aborted flag */
108 JBEGIN_ABORT = 2,
109 };
110
111 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
112 struct super_block *sb,
113 unsigned long nblocks, int join);
114
115 static void init_journal_hash(struct super_block *sb)
116 {
117 struct reiserfs_journal *journal = SB_JOURNAL(sb);
118 memset(journal->j_hash_table, 0,
119 JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
120 }
121
122 /*
123 * clears BH_Dirty and sticks the buffer on the clean list. Called because
124 * I can't allow refile_buffer to make schedule happen after I've freed a
125 * block. Look at remove_from_transaction and journal_mark_freed for
126 * more details.
127 */
128 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
129 {
130 if (bh) {
131 clear_buffer_dirty(bh);
132 clear_buffer_journal_test(bh);
133 }
134 return 0;
135 }
136
137 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
138 *sb)
139 {
140 struct reiserfs_bitmap_node *bn;
141 static int id;
142
143 bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
144 if (!bn) {
145 return NULL;
146 }
147 bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
148 if (!bn->data) {
149 kfree(bn);
150 return NULL;
151 }
152 bn->id = id++;
153 INIT_LIST_HEAD(&bn->list);
154 return bn;
155 }
156
157 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
158 {
159 struct reiserfs_journal *journal = SB_JOURNAL(sb);
160 struct reiserfs_bitmap_node *bn = NULL;
161 struct list_head *entry = journal->j_bitmap_nodes.next;
162
163 journal->j_used_bitmap_nodes++;
164 repeat:
165
166 if (entry != &journal->j_bitmap_nodes) {
167 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
168 list_del(entry);
169 memset(bn->data, 0, sb->s_blocksize);
170 journal->j_free_bitmap_nodes--;
171 return bn;
172 }
173 bn = allocate_bitmap_node(sb);
174 if (!bn) {
175 yield();
176 goto repeat;
177 }
178 return bn;
179 }
180 static inline void free_bitmap_node(struct super_block *sb,
181 struct reiserfs_bitmap_node *bn)
182 {
183 struct reiserfs_journal *journal = SB_JOURNAL(sb);
184 journal->j_used_bitmap_nodes--;
185 if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
186 kfree(bn->data);
187 kfree(bn);
188 } else {
189 list_add(&bn->list, &journal->j_bitmap_nodes);
190 journal->j_free_bitmap_nodes++;
191 }
192 }
193
194 static void allocate_bitmap_nodes(struct super_block *sb)
195 {
196 int i;
197 struct reiserfs_journal *journal = SB_JOURNAL(sb);
198 struct reiserfs_bitmap_node *bn = NULL;
199 for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
200 bn = allocate_bitmap_node(sb);
201 if (bn) {
202 list_add(&bn->list, &journal->j_bitmap_nodes);
203 journal->j_free_bitmap_nodes++;
204 } else {
205 /* this is ok, we'll try again when more are needed */
206 break;
207 }
208 }
209 }
210
211 static int set_bit_in_list_bitmap(struct super_block *sb,
212 b_blocknr_t block,
213 struct reiserfs_list_bitmap *jb)
214 {
215 unsigned int bmap_nr = block / (sb->s_blocksize << 3);
216 unsigned int bit_nr = block % (sb->s_blocksize << 3);
217
218 if (!jb->bitmaps[bmap_nr]) {
219 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
220 }
221 set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
222 return 0;
223 }
224
225 static void cleanup_bitmap_list(struct super_block *sb,
226 struct reiserfs_list_bitmap *jb)
227 {
228 int i;
229 if (jb->bitmaps == NULL)
230 return;
231
232 for (i = 0; i < reiserfs_bmap_count(sb); i++) {
233 if (jb->bitmaps[i]) {
234 free_bitmap_node(sb, jb->bitmaps[i]);
235 jb->bitmaps[i] = NULL;
236 }
237 }
238 }
239
240 /*
241 * only call this on FS unmount.
242 */
243 static int free_list_bitmaps(struct super_block *sb,
244 struct reiserfs_list_bitmap *jb_array)
245 {
246 int i;
247 struct reiserfs_list_bitmap *jb;
248 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
249 jb = jb_array + i;
250 jb->journal_list = NULL;
251 cleanup_bitmap_list(sb, jb);
252 vfree(jb->bitmaps);
253 jb->bitmaps = NULL;
254 }
255 return 0;
256 }
257
258 static int free_bitmap_nodes(struct super_block *sb)
259 {
260 struct reiserfs_journal *journal = SB_JOURNAL(sb);
261 struct list_head *next = journal->j_bitmap_nodes.next;
262 struct reiserfs_bitmap_node *bn;
263
264 while (next != &journal->j_bitmap_nodes) {
265 bn = list_entry(next, struct reiserfs_bitmap_node, list);
266 list_del(next);
267 kfree(bn->data);
268 kfree(bn);
269 next = journal->j_bitmap_nodes.next;
270 journal->j_free_bitmap_nodes--;
271 }
272
273 return 0;
274 }
275
276 /*
277 * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
278 * jb_array is the array to be filled in.
279 */
280 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
281 struct reiserfs_list_bitmap *jb_array,
282 unsigned int bmap_nr)
283 {
284 int i;
285 int failed = 0;
286 struct reiserfs_list_bitmap *jb;
287 int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
288
289 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
290 jb = jb_array + i;
291 jb->journal_list = NULL;
292 jb->bitmaps = vzalloc(mem);
293 if (!jb->bitmaps) {
294 reiserfs_warning(sb, "clm-2000", "unable to "
295 "allocate bitmaps for journal lists");
296 failed = 1;
297 break;
298 }
299 }
300 if (failed) {
301 free_list_bitmaps(sb, jb_array);
302 return -1;
303 }
304 return 0;
305 }
306
307 /*
308 * find an available list bitmap. If you can't find one, flush a commit list
309 * and try again
310 */
311 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
312 struct reiserfs_journal_list
313 *jl)
314 {
315 int i, j;
316 struct reiserfs_journal *journal = SB_JOURNAL(sb);
317 struct reiserfs_list_bitmap *jb = NULL;
318
319 for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
320 i = journal->j_list_bitmap_index;
321 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
322 jb = journal->j_list_bitmap + i;
323 if (journal->j_list_bitmap[i].journal_list) {
324 flush_commit_list(sb,
325 journal->j_list_bitmap[i].
326 journal_list, 1);
327 if (!journal->j_list_bitmap[i].journal_list) {
328 break;
329 }
330 } else {
331 break;
332 }
333 }
334 /* double check to make sure if flushed correctly */
335 if (jb->journal_list)
336 return NULL;
337 jb->journal_list = jl;
338 return jb;
339 }
340
341 /*
342 * allocates a new chunk of X nodes, and links them all together as a list.
343 * Uses the cnode->next and cnode->prev pointers
344 * returns NULL on failure
345 */
346 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
347 {
348 struct reiserfs_journal_cnode *head;
349 int i;
350 if (num_cnodes <= 0) {
351 return NULL;
352 }
353 head = vzalloc(array_size(num_cnodes,
354 sizeof(struct reiserfs_journal_cnode)));
355 if (!head) {
356 return NULL;
357 }
358 head[0].prev = NULL;
359 head[0].next = head + 1;
360 for (i = 1; i < num_cnodes; i++) {
361 head[i].prev = head + (i - 1);
362 head[i].next = head + (i + 1); /* if last one, overwrite it after the if */
363 }
364 head[num_cnodes - 1].next = NULL;
365 return head;
366 }
367
368 /* pulls a cnode off the free list, or returns NULL on failure */
369 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
370 {
371 struct reiserfs_journal_cnode *cn;
372 struct reiserfs_journal *journal = SB_JOURNAL(sb);
373
374 reiserfs_check_lock_depth(sb, "get_cnode");
375
376 if (journal->j_cnode_free <= 0) {
377 return NULL;
378 }
379 journal->j_cnode_used++;
380 journal->j_cnode_free--;
381 cn = journal->j_cnode_free_list;
382 if (!cn) {
383 return cn;
384 }
385 if (cn->next) {
386 cn->next->prev = NULL;
387 }
388 journal->j_cnode_free_list = cn->next;
389 memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
390 return cn;
391 }
392
393 /*
394 * returns a cnode to the free list
395 */
396 static void free_cnode(struct super_block *sb,
397 struct reiserfs_journal_cnode *cn)
398 {
399 struct reiserfs_journal *journal = SB_JOURNAL(sb);
400
401 reiserfs_check_lock_depth(sb, "free_cnode");
402
403 journal->j_cnode_used--;
404 journal->j_cnode_free++;
405 /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
406 cn->next = journal->j_cnode_free_list;
407 if (journal->j_cnode_free_list) {
408 journal->j_cnode_free_list->prev = cn;
409 }
410 cn->prev = NULL; /* not needed with the memset, but I might kill the memset, and forget to do this */
411 journal->j_cnode_free_list = cn;
412 }
413
414 static void clear_prepared_bits(struct buffer_head *bh)
415 {
416 clear_buffer_journal_prepared(bh);
417 clear_buffer_journal_restore_dirty(bh);
418 }
419
420 /*
421 * return a cnode with same dev, block number and size in table,
422 * or null if not found
423 */
424 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
425 super_block
426 *sb,
427 struct
428 reiserfs_journal_cnode
429 **table,
430 long bl)
431 {
432 struct reiserfs_journal_cnode *cn;
433 cn = journal_hash(table, sb, bl);
434 while (cn) {
435 if (cn->blocknr == bl && cn->sb == sb)
436 return cn;
437 cn = cn->hnext;
438 }
439 return (struct reiserfs_journal_cnode *)0;
440 }
441
442 /*
443 * this actually means 'can this block be reallocated yet?'. If you set
444 * search_all, a block can only be allocated if it is not in the current
445 * transaction, was not freed by the current transaction, and has no chance
446 * of ever being overwritten by a replay after crashing.
447 *
448 * If you don't set search_all, a block can only be allocated if it is not
449 * in the current transaction. Since deleting a block removes it from the
450 * current transaction, this case should never happen. If you don't set
451 * search_all, make sure you never write the block without logging it.
452 *
453 * next_zero_bit is a suggestion about the next block to try for find_forward.
454 * when bl is rejected because it is set in a journal list bitmap, we search
455 * for the next zero bit in the bitmap that rejected bl. Then, we return
456 * that through next_zero_bit for find_forward to try.
457 *
458 * Just because we return something in next_zero_bit does not mean we won't
459 * reject it on the next call to reiserfs_in_journal
460 */
461 int reiserfs_in_journal(struct super_block *sb,
462 unsigned int bmap_nr, int bit_nr, int search_all,
463 b_blocknr_t * next_zero_bit)
464 {
465 struct reiserfs_journal *journal = SB_JOURNAL(sb);
466 struct reiserfs_journal_cnode *cn;
467 struct reiserfs_list_bitmap *jb;
468 int i;
469 unsigned long bl;
470
471 *next_zero_bit = 0; /* always start this at zero. */
472
473 PROC_INFO_INC(sb, journal.in_journal);
474 /*
475 * If we aren't doing a search_all, this is a metablock, and it
476 * will be logged before use. if we crash before the transaction
477 * that freed it commits, this transaction won't have committed
478 * either, and the block will never be written
479 */
480 if (search_all) {
481 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
482 PROC_INFO_INC(sb, journal.in_journal_bitmap);
483 jb = journal->j_list_bitmap + i;
484 if (jb->journal_list && jb->bitmaps[bmap_nr] &&
485 test_bit(bit_nr,
486 (unsigned long *)jb->bitmaps[bmap_nr]->
487 data)) {
488 *next_zero_bit =
489 find_next_zero_bit((unsigned long *)
490 (jb->bitmaps[bmap_nr]->
491 data),
492 sb->s_blocksize << 3,
493 bit_nr + 1);
494 return 1;
495 }
496 }
497 }
498
499 bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
500 /* is it in any old transactions? */
501 if (search_all
502 && (cn =
503 get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
504 return 1;
505 }
506
507 /* is it in the current transaction. This should never happen */
508 if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
509 BUG();
510 return 1;
511 }
512
513 PROC_INFO_INC(sb, journal.in_journal_reusable);
514 /* safe for reuse */
515 return 0;
516 }
517
518 /* insert cn into table */
519 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
520 struct reiserfs_journal_cnode *cn)
521 {
522 struct reiserfs_journal_cnode *cn_orig;
523
524 cn_orig = journal_hash(table, cn->sb, cn->blocknr);
525 cn->hnext = cn_orig;
526 cn->hprev = NULL;
527 if (cn_orig) {
528 cn_orig->hprev = cn;
529 }
530 journal_hash(table, cn->sb, cn->blocknr) = cn;
531 }
532
533 /* lock the current transaction */
534 static inline void lock_journal(struct super_block *sb)
535 {
536 PROC_INFO_INC(sb, journal.lock_journal);
537
538 reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
539 }
540
541 /* unlock the current transaction */
542 static inline void unlock_journal(struct super_block *sb)
543 {
544 mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
545 }
546
547 static inline void get_journal_list(struct reiserfs_journal_list *jl)
548 {
549 jl->j_refcount++;
550 }
551
552 static inline void put_journal_list(struct super_block *s,
553 struct reiserfs_journal_list *jl)
554 {
555 if (jl->j_refcount < 1) {
556 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
557 jl->j_trans_id, jl->j_refcount);
558 }
559 if (--jl->j_refcount == 0)
560 kfree(jl);
561 }
562
563 /*
564 * this used to be much more involved, and I'm keeping it just in case
565 * things get ugly again. it gets called by flush_commit_list, and
566 * cleans up any data stored about blocks freed during a transaction.
567 */
568 static void cleanup_freed_for_journal_list(struct super_block *sb,
569 struct reiserfs_journal_list *jl)
570 {
571
572 struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
573 if (jb) {
574 cleanup_bitmap_list(sb, jb);
575 }
576 jl->j_list_bitmap->journal_list = NULL;
577 jl->j_list_bitmap = NULL;
578 }
579
580 static int journal_list_still_alive(struct super_block *s,
581 unsigned int trans_id)
582 {
583 struct reiserfs_journal *journal = SB_JOURNAL(s);
584 struct list_head *entry = &journal->j_journal_list;
585 struct reiserfs_journal_list *jl;
586
587 if (!list_empty(entry)) {
588 jl = JOURNAL_LIST_ENTRY(entry->next);
589 if (jl->j_trans_id <= trans_id) {
590 return 1;
591 }
592 }
593 return 0;
594 }
595
596 /*
597 * If page->mapping was null, we failed to truncate this page for
598 * some reason. Most likely because it was truncated after being
599 * logged via data=journal.
600 *
601 * This does a check to see if the buffer belongs to one of these
602 * lost pages before doing the final put_bh. If page->mapping was
603 * null, it tries to free buffers on the page, which should make the
604 * final put_page drop the page from the lru.
605 */
606 static void release_buffer_page(struct buffer_head *bh)
607 {
608 struct page *page = bh->b_page;
609 if (!page->mapping && trylock_page(page)) {
610 get_page(page);
611 put_bh(bh);
612 if (!page->mapping)
613 try_to_free_buffers(page);
614 unlock_page(page);
615 put_page(page);
616 } else {
617 put_bh(bh);
618 }
619 }
620
621 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
622 {
623 if (buffer_journaled(bh)) {
624 reiserfs_warning(NULL, "clm-2084",
625 "pinned buffer %lu:%pg sent to disk",
626 bh->b_blocknr, bh->b_bdev);
627 }
628 if (uptodate)
629 set_buffer_uptodate(bh);
630 else
631 clear_buffer_uptodate(bh);
632
633 unlock_buffer(bh);
634 release_buffer_page(bh);
635 }
636
637 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
638 {
639 if (uptodate)
640 set_buffer_uptodate(bh);
641 else
642 clear_buffer_uptodate(bh);
643 unlock_buffer(bh);
644 put_bh(bh);
645 }
646
647 static void submit_logged_buffer(struct buffer_head *bh)
648 {
649 get_bh(bh);
650 bh->b_end_io = reiserfs_end_buffer_io_sync;
651 clear_buffer_journal_new(bh);
652 clear_buffer_dirty(bh);
653 if (!test_clear_buffer_journal_test(bh))
654 BUG();
655 if (!buffer_uptodate(bh))
656 BUG();
657 submit_bh(REQ_OP_WRITE, 0, bh);
658 }
659
660 static void submit_ordered_buffer(struct buffer_head *bh)
661 {
662 get_bh(bh);
663 bh->b_end_io = reiserfs_end_ordered_io;
664 clear_buffer_dirty(bh);
665 if (!buffer_uptodate(bh))
666 BUG();
667 submit_bh(REQ_OP_WRITE, 0, bh);
668 }
669
670 #define CHUNK_SIZE 32
671 struct buffer_chunk {
672 struct buffer_head *bh[CHUNK_SIZE];
673 int nr;
674 };
675
676 static void write_chunk(struct buffer_chunk *chunk)
677 {
678 int i;
679 for (i = 0; i < chunk->nr; i++) {
680 submit_logged_buffer(chunk->bh[i]);
681 }
682 chunk->nr = 0;
683 }
684
685 static void write_ordered_chunk(struct buffer_chunk *chunk)
686 {
687 int i;
688 for (i = 0; i < chunk->nr; i++) {
689 submit_ordered_buffer(chunk->bh[i]);
690 }
691 chunk->nr = 0;
692 }
693
694 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
695 spinlock_t * lock, void (fn) (struct buffer_chunk *))
696 {
697 int ret = 0;
698 BUG_ON(chunk->nr >= CHUNK_SIZE);
699 chunk->bh[chunk->nr++] = bh;
700 if (chunk->nr >= CHUNK_SIZE) {
701 ret = 1;
702 if (lock) {
703 spin_unlock(lock);
704 fn(chunk);
705 spin_lock(lock);
706 } else {
707 fn(chunk);
708 }
709 }
710 return ret;
711 }
712
713 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
714 static struct reiserfs_jh *alloc_jh(void)
715 {
716 struct reiserfs_jh *jh;
717 while (1) {
718 jh = kmalloc(sizeof(*jh), GFP_NOFS);
719 if (jh) {
720 atomic_inc(&nr_reiserfs_jh);
721 return jh;
722 }
723 yield();
724 }
725 }
726
727 /*
728 * we want to free the jh when the buffer has been written
729 * and waited on
730 */
731 void reiserfs_free_jh(struct buffer_head *bh)
732 {
733 struct reiserfs_jh *jh;
734
735 jh = bh->b_private;
736 if (jh) {
737 bh->b_private = NULL;
738 jh->bh = NULL;
739 list_del_init(&jh->list);
740 kfree(jh);
741 if (atomic_read(&nr_reiserfs_jh) <= 0)
742 BUG();
743 atomic_dec(&nr_reiserfs_jh);
744 put_bh(bh);
745 }
746 }
747
748 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
749 int tail)
750 {
751 struct reiserfs_jh *jh;
752
753 if (bh->b_private) {
754 spin_lock(&j->j_dirty_buffers_lock);
755 if (!bh->b_private) {
756 spin_unlock(&j->j_dirty_buffers_lock);
757 goto no_jh;
758 }
759 jh = bh->b_private;
760 list_del_init(&jh->list);
761 } else {
762 no_jh:
763 get_bh(bh);
764 jh = alloc_jh();
765 spin_lock(&j->j_dirty_buffers_lock);
766 /*
767 * buffer must be locked for __add_jh, should be able to have
768 * two adds at the same time
769 */
770 BUG_ON(bh->b_private);
771 jh->bh = bh;
772 bh->b_private = jh;
773 }
774 jh->jl = j->j_current_jl;
775 if (tail)
776 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
777 else {
778 list_add_tail(&jh->list, &jh->jl->j_bh_list);
779 }
780 spin_unlock(&j->j_dirty_buffers_lock);
781 return 0;
782 }
783
784 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
785 {
786 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
787 }
788 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
789 {
790 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
791 }
792
793 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
794 static int write_ordered_buffers(spinlock_t * lock,
795 struct reiserfs_journal *j,
796 struct reiserfs_journal_list *jl,
797 struct list_head *list)
798 {
799 struct buffer_head *bh;
800 struct reiserfs_jh *jh;
801 int ret = j->j_errno;
802 struct buffer_chunk chunk;
803 struct list_head tmp;
804 INIT_LIST_HEAD(&tmp);
805
806 chunk.nr = 0;
807 spin_lock(lock);
808 while (!list_empty(list)) {
809 jh = JH_ENTRY(list->next);
810 bh = jh->bh;
811 get_bh(bh);
812 if (!trylock_buffer(bh)) {
813 if (!buffer_dirty(bh)) {
814 list_move(&jh->list, &tmp);
815 goto loop_next;
816 }
817 spin_unlock(lock);
818 if (chunk.nr)
819 write_ordered_chunk(&chunk);
820 wait_on_buffer(bh);
821 cond_resched();
822 spin_lock(lock);
823 goto loop_next;
824 }
825 /*
826 * in theory, dirty non-uptodate buffers should never get here,
827 * but the upper layer io error paths still have a few quirks.
828 * Handle them here as gracefully as we can
829 */
830 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
831 clear_buffer_dirty(bh);
832 ret = -EIO;
833 }
834 if (buffer_dirty(bh)) {
835 list_move(&jh->list, &tmp);
836 add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
837 } else {
838 reiserfs_free_jh(bh);
839 unlock_buffer(bh);
840 }
841 loop_next:
842 put_bh(bh);
843 cond_resched_lock(lock);
844 }
845 if (chunk.nr) {
846 spin_unlock(lock);
847 write_ordered_chunk(&chunk);
848 spin_lock(lock);
849 }
850 while (!list_empty(&tmp)) {
851 jh = JH_ENTRY(tmp.prev);
852 bh = jh->bh;
853 get_bh(bh);
854 reiserfs_free_jh(bh);
855
856 if (buffer_locked(bh)) {
857 spin_unlock(lock);
858 wait_on_buffer(bh);
859 spin_lock(lock);
860 }
861 if (!buffer_uptodate(bh)) {
862 ret = -EIO;
863 }
864 /*
865 * ugly interaction with invalidatepage here.
866 * reiserfs_invalidate_page will pin any buffer that has a
867 * valid journal head from an older transaction. If someone
868 * else sets our buffer dirty after we write it in the first
869 * loop, and then someone truncates the page away, nobody
870 * will ever write the buffer. We're safe if we write the
871 * page one last time after freeing the journal header.
872 */
873 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
874 spin_unlock(lock);
875 ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
876 spin_lock(lock);
877 }
878 put_bh(bh);
879 cond_resched_lock(lock);
880 }
881 spin_unlock(lock);
882 return ret;
883 }
884
885 static int flush_older_commits(struct super_block *s,
886 struct reiserfs_journal_list *jl)
887 {
888 struct reiserfs_journal *journal = SB_JOURNAL(s);
889 struct reiserfs_journal_list *other_jl;
890 struct reiserfs_journal_list *first_jl;
891 struct list_head *entry;
892 unsigned int trans_id = jl->j_trans_id;
893 unsigned int other_trans_id;
894 unsigned int first_trans_id;
895
896 find_first:
897 /*
898 * first we walk backwards to find the oldest uncommitted transation
899 */
900 first_jl = jl;
901 entry = jl->j_list.prev;
902 while (1) {
903 other_jl = JOURNAL_LIST_ENTRY(entry);
904 if (entry == &journal->j_journal_list ||
905 atomic_read(&other_jl->j_older_commits_done))
906 break;
907
908 first_jl = other_jl;
909 entry = other_jl->j_list.prev;
910 }
911
912 /* if we didn't find any older uncommitted transactions, return now */
913 if (first_jl == jl) {
914 return 0;
915 }
916
917 first_trans_id = first_jl->j_trans_id;
918
919 entry = &first_jl->j_list;
920 while (1) {
921 other_jl = JOURNAL_LIST_ENTRY(entry);
922 other_trans_id = other_jl->j_trans_id;
923
924 if (other_trans_id < trans_id) {
925 if (atomic_read(&other_jl->j_commit_left) != 0) {
926 flush_commit_list(s, other_jl, 0);
927
928 /* list we were called with is gone, return */
929 if (!journal_list_still_alive(s, trans_id))
930 return 1;
931
932 /*
933 * the one we just flushed is gone, this means
934 * all older lists are also gone, so first_jl
935 * is no longer valid either. Go back to the
936 * beginning.
937 */
938 if (!journal_list_still_alive
939 (s, other_trans_id)) {
940 goto find_first;
941 }
942 }
943 entry = entry->next;
944 if (entry == &journal->j_journal_list)
945 return 0;
946 } else {
947 return 0;
948 }
949 }
950 return 0;
951 }
952
953 static int reiserfs_async_progress_wait(struct super_block *s)
954 {
955 struct reiserfs_journal *j = SB_JOURNAL(s);
956
957 if (atomic_read(&j->j_async_throttle)) {
958 int depth;
959
960 depth = reiserfs_write_unlock_nested(s);
961 congestion_wait(BLK_RW_ASYNC, HZ / 10);
962 reiserfs_write_lock_nested(s, depth);
963 }
964
965 return 0;
966 }
967
968 /*
969 * if this journal list still has commit blocks unflushed, send them to disk.
970 *
971 * log areas must be flushed in order (transaction 2 can't commit before
972 * transaction 1) Before the commit block can by written, every other log
973 * block must be safely on disk
974 */
975 static int flush_commit_list(struct super_block *s,
976 struct reiserfs_journal_list *jl, int flushall)
977 {
978 int i;
979 b_blocknr_t bn;
980 struct buffer_head *tbh = NULL;
981 unsigned int trans_id = jl->j_trans_id;
982 struct reiserfs_journal *journal = SB_JOURNAL(s);
983 int retval = 0;
984 int write_len;
985 int depth;
986
987 reiserfs_check_lock_depth(s, "flush_commit_list");
988
989 if (atomic_read(&jl->j_older_commits_done)) {
990 return 0;
991 }
992
993 /*
994 * before we can put our commit blocks on disk, we have to make
995 * sure everyone older than us is on disk too
996 */
997 BUG_ON(jl->j_len <= 0);
998 BUG_ON(trans_id == journal->j_trans_id);
999
1000 get_journal_list(jl);
1001 if (flushall) {
1002 if (flush_older_commits(s, jl) == 1) {
1003 /*
1004 * list disappeared during flush_older_commits.
1005 * return
1006 */
1007 goto put_jl;
1008 }
1009 }
1010
1011 /* make sure nobody is trying to flush this one at the same time */
1012 reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1013
1014 if (!journal_list_still_alive(s, trans_id)) {
1015 mutex_unlock(&jl->j_commit_mutex);
1016 goto put_jl;
1017 }
1018 BUG_ON(jl->j_trans_id == 0);
1019
1020 /* this commit is done, exit */
1021 if (atomic_read(&jl->j_commit_left) <= 0) {
1022 if (flushall) {
1023 atomic_set(&jl->j_older_commits_done, 1);
1024 }
1025 mutex_unlock(&jl->j_commit_mutex);
1026 goto put_jl;
1027 }
1028
1029 if (!list_empty(&jl->j_bh_list)) {
1030 int ret;
1031
1032 /*
1033 * We might sleep in numerous places inside
1034 * write_ordered_buffers. Relax the write lock.
1035 */
1036 depth = reiserfs_write_unlock_nested(s);
1037 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1038 journal, jl, &jl->j_bh_list);
1039 if (ret < 0 && retval == 0)
1040 retval = ret;
1041 reiserfs_write_lock_nested(s, depth);
1042 }
1043 BUG_ON(!list_empty(&jl->j_bh_list));
1044 /*
1045 * for the description block and all the log blocks, submit any buffers
1046 * that haven't already reached the disk. Try to write at least 256
1047 * log blocks. later on, we will only wait on blocks that correspond
1048 * to this transaction, but while we're unplugging we might as well
1049 * get a chunk of data on there.
1050 */
1051 atomic_inc(&journal->j_async_throttle);
1052 write_len = jl->j_len + 1;
1053 if (write_len < 256)
1054 write_len = 256;
1055 for (i = 0 ; i < write_len ; i++) {
1056 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1057 SB_ONDISK_JOURNAL_SIZE(s);
1058 tbh = journal_find_get_block(s, bn);
1059 if (tbh) {
1060 if (buffer_dirty(tbh)) {
1061 depth = reiserfs_write_unlock_nested(s);
1062 ll_rw_block(REQ_OP_WRITE, 0, 1, &tbh);
1063 reiserfs_write_lock_nested(s, depth);
1064 }
1065 put_bh(tbh) ;
1066 }
1067 }
1068 atomic_dec(&journal->j_async_throttle);
1069
1070 for (i = 0; i < (jl->j_len + 1); i++) {
1071 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1072 (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1073 tbh = journal_find_get_block(s, bn);
1074
1075 depth = reiserfs_write_unlock_nested(s);
1076 __wait_on_buffer(tbh);
1077 reiserfs_write_lock_nested(s, depth);
1078 /*
1079 * since we're using ll_rw_blk above, it might have skipped
1080 * over a locked buffer. Double check here
1081 */
1082 /* redundant, sync_dirty_buffer() checks */
1083 if (buffer_dirty(tbh)) {
1084 depth = reiserfs_write_unlock_nested(s);
1085 sync_dirty_buffer(tbh);
1086 reiserfs_write_lock_nested(s, depth);
1087 }
1088 if (unlikely(!buffer_uptodate(tbh))) {
1089 #ifdef CONFIG_REISERFS_CHECK
1090 reiserfs_warning(s, "journal-601",
1091 "buffer write failed");
1092 #endif
1093 retval = -EIO;
1094 }
1095 /* once for journal_find_get_block */
1096 put_bh(tbh);
1097 /* once due to original getblk in do_journal_end */
1098 put_bh(tbh);
1099 atomic_dec(&jl->j_commit_left);
1100 }
1101
1102 BUG_ON(atomic_read(&jl->j_commit_left) != 1);
1103
1104 /*
1105 * If there was a write error in the journal - we can't commit
1106 * this transaction - it will be invalid and, if successful,
1107 * will just end up propagating the write error out to
1108 * the file system.
1109 */
1110 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1111 if (buffer_dirty(jl->j_commit_bh))
1112 BUG();
1113 mark_buffer_dirty(jl->j_commit_bh) ;
1114 depth = reiserfs_write_unlock_nested(s);
1115 if (reiserfs_barrier_flush(s))
1116 __sync_dirty_buffer(jl->j_commit_bh,
1117 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1118 else
1119 sync_dirty_buffer(jl->j_commit_bh);
1120 reiserfs_write_lock_nested(s, depth);
1121 }
1122
1123 /*
1124 * If there was a write error in the journal - we can't commit this
1125 * transaction - it will be invalid and, if successful, will just end
1126 * up propagating the write error out to the filesystem.
1127 */
1128 if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1129 #ifdef CONFIG_REISERFS_CHECK
1130 reiserfs_warning(s, "journal-615", "buffer write failed");
1131 #endif
1132 retval = -EIO;
1133 }
1134 bforget(jl->j_commit_bh);
1135 if (journal->j_last_commit_id != 0 &&
1136 (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1137 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1138 journal->j_last_commit_id, jl->j_trans_id);
1139 }
1140 journal->j_last_commit_id = jl->j_trans_id;
1141
1142 /*
1143 * now, every commit block is on the disk. It is safe to allow
1144 * blocks freed during this transaction to be reallocated
1145 */
1146 cleanup_freed_for_journal_list(s, jl);
1147
1148 retval = retval ? retval : journal->j_errno;
1149
1150 /* mark the metadata dirty */
1151 if (!retval)
1152 dirty_one_transaction(s, jl);
1153 atomic_dec(&jl->j_commit_left);
1154
1155 if (flushall) {
1156 atomic_set(&jl->j_older_commits_done, 1);
1157 }
1158 mutex_unlock(&jl->j_commit_mutex);
1159 put_jl:
1160 put_journal_list(s, jl);
1161
1162 if (retval)
1163 reiserfs_abort(s, retval, "Journal write error in %s",
1164 __func__);
1165 return retval;
1166 }
1167
1168 /*
1169 * flush_journal_list frequently needs to find a newer transaction for a
1170 * given block. This does that, or returns NULL if it can't find anything
1171 */
1172 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1173 reiserfs_journal_cnode
1174 *cn)
1175 {
1176 struct super_block *sb = cn->sb;
1177 b_blocknr_t blocknr = cn->blocknr;
1178
1179 cn = cn->hprev;
1180 while (cn) {
1181 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1182 return cn->jlist;
1183 }
1184 cn = cn->hprev;
1185 }
1186 return NULL;
1187 }
1188
1189 static void remove_journal_hash(struct super_block *,
1190 struct reiserfs_journal_cnode **,
1191 struct reiserfs_journal_list *, unsigned long,
1192 int);
1193
1194 /*
1195 * once all the real blocks have been flushed, it is safe to remove them
1196 * from the journal list for this transaction. Aside from freeing the
1197 * cnode, this also allows the block to be reallocated for data blocks
1198 * if it had been deleted.
1199 */
1200 static void remove_all_from_journal_list(struct super_block *sb,
1201 struct reiserfs_journal_list *jl,
1202 int debug)
1203 {
1204 struct reiserfs_journal *journal = SB_JOURNAL(sb);
1205 struct reiserfs_journal_cnode *cn, *last;
1206 cn = jl->j_realblock;
1207
1208 /*
1209 * which is better, to lock once around the whole loop, or
1210 * to lock for each call to remove_journal_hash?
1211 */
1212 while (cn) {
1213 if (cn->blocknr != 0) {
1214 if (debug) {
1215 reiserfs_warning(sb, "reiserfs-2201",
1216 "block %u, bh is %d, state %ld",
1217 cn->blocknr, cn->bh ? 1 : 0,
1218 cn->state);
1219 }
1220 cn->state = 0;
1221 remove_journal_hash(sb, journal->j_list_hash_table,
1222 jl, cn->blocknr, 1);
1223 }
1224 last = cn;
1225 cn = cn->next;
1226 free_cnode(sb, last);
1227 }
1228 jl->j_realblock = NULL;
1229 }
1230
1231 /*
1232 * if this timestamp is greater than the timestamp we wrote last to the
1233 * header block, write it to the header block. once this is done, I can
1234 * safely say the log area for this transaction won't ever be replayed,
1235 * and I can start releasing blocks in this transaction for reuse as data
1236 * blocks. called by flush_journal_list, before it calls
1237 * remove_all_from_journal_list
1238 */
1239 static int _update_journal_header_block(struct super_block *sb,
1240 unsigned long offset,
1241 unsigned int trans_id)
1242 {
1243 struct reiserfs_journal_header *jh;
1244 struct reiserfs_journal *journal = SB_JOURNAL(sb);
1245 int depth;
1246
1247 if (reiserfs_is_journal_aborted(journal))
1248 return -EIO;
1249
1250 if (trans_id >= journal->j_last_flush_trans_id) {
1251 if (buffer_locked((journal->j_header_bh))) {
1252 depth = reiserfs_write_unlock_nested(sb);
1253 __wait_on_buffer(journal->j_header_bh);
1254 reiserfs_write_lock_nested(sb, depth);
1255 if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1256 #ifdef CONFIG_REISERFS_CHECK
1257 reiserfs_warning(sb, "journal-699",
1258 "buffer write failed");
1259 #endif
1260 return -EIO;
1261 }
1262 }
1263 journal->j_last_flush_trans_id = trans_id;
1264 journal->j_first_unflushed_offset = offset;
1265 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1266 b_data);
1267 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1268 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1269 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1270
1271 set_buffer_dirty(journal->j_header_bh);
1272 depth = reiserfs_write_unlock_nested(sb);
1273
1274 if (reiserfs_barrier_flush(sb))
1275 __sync_dirty_buffer(journal->j_header_bh,
1276 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1277 else
1278 sync_dirty_buffer(journal->j_header_bh);
1279
1280 reiserfs_write_lock_nested(sb, depth);
1281 if (!buffer_uptodate(journal->j_header_bh)) {
1282 reiserfs_warning(sb, "journal-837",
1283 "IO error during journal replay");
1284 return -EIO;
1285 }
1286 }
1287 return 0;
1288 }
1289
1290 static int update_journal_header_block(struct super_block *sb,
1291 unsigned long offset,
1292 unsigned int trans_id)
1293 {
1294 return _update_journal_header_block(sb, offset, trans_id);
1295 }
1296
1297 /*
1298 ** flush any and all journal lists older than you are
1299 ** can only be called from flush_journal_list
1300 */
1301 static int flush_older_journal_lists(struct super_block *sb,
1302 struct reiserfs_journal_list *jl)
1303 {
1304 struct list_head *entry;
1305 struct reiserfs_journal_list *other_jl;
1306 struct reiserfs_journal *journal = SB_JOURNAL(sb);
1307 unsigned int trans_id = jl->j_trans_id;
1308
1309 /*
1310 * we know we are the only ones flushing things, no extra race
1311 * protection is required.
1312 */
1313 restart:
1314 entry = journal->j_journal_list.next;
1315 /* Did we wrap? */
1316 if (entry == &journal->j_journal_list)
1317 return 0;
1318 other_jl = JOURNAL_LIST_ENTRY(entry);
1319 if (other_jl->j_trans_id < trans_id) {
1320 BUG_ON(other_jl->j_refcount <= 0);
1321 /* do not flush all */
1322 flush_journal_list(sb, other_jl, 0);
1323
1324 /* other_jl is now deleted from the list */
1325 goto restart;
1326 }
1327 return 0;
1328 }
1329
1330 static void del_from_work_list(struct super_block *s,
1331 struct reiserfs_journal_list *jl)
1332 {
1333 struct reiserfs_journal *journal = SB_JOURNAL(s);
1334 if (!list_empty(&jl->j_working_list)) {
1335 list_del_init(&jl->j_working_list);
1336 journal->j_num_work_lists--;
1337 }
1338 }
1339
1340 /*
1341 * flush a journal list, both commit and real blocks
1342 *
1343 * always set flushall to 1, unless you are calling from inside
1344 * flush_journal_list
1345 *
1346 * IMPORTANT. This can only be called while there are no journal writers,
1347 * and the journal is locked. That means it can only be called from
1348 * do_journal_end, or by journal_release
1349 */
1350 static int flush_journal_list(struct super_block *s,
1351 struct reiserfs_journal_list *jl, int flushall)
1352 {
1353 struct reiserfs_journal_list *pjl;
1354 struct reiserfs_journal_cnode *cn, *last;
1355 int count;
1356 int was_jwait = 0;
1357 int was_dirty = 0;
1358 struct buffer_head *saved_bh;
1359 unsigned long j_len_saved = jl->j_len;
1360 struct reiserfs_journal *journal = SB_JOURNAL(s);
1361 int err = 0;
1362 int depth;
1363
1364 BUG_ON(j_len_saved <= 0);
1365
1366 if (atomic_read(&journal->j_wcount) != 0) {
1367 reiserfs_warning(s, "clm-2048", "called with wcount %d",
1368 atomic_read(&journal->j_wcount));
1369 }
1370
1371 /* if flushall == 0, the lock is already held */
1372 if (flushall) {
1373 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1374 } else if (mutex_trylock(&journal->j_flush_mutex)) {
1375 BUG();
1376 }
1377
1378 count = 0;
1379 if (j_len_saved > journal->j_trans_max) {
1380 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1381 j_len_saved, jl->j_trans_id);
1382 return 0;
1383 }
1384
1385 /* if all the work is already done, get out of here */
1386 if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1387 atomic_read(&jl->j_commit_left) <= 0) {
1388 goto flush_older_and_return;
1389 }
1390
1391 /*
1392 * start by putting the commit list on disk. This will also flush
1393 * the commit lists of any olders transactions
1394 */
1395 flush_commit_list(s, jl, 1);
1396
1397 if (!(jl->j_state & LIST_DIRTY)
1398 && !reiserfs_is_journal_aborted(journal))
1399 BUG();
1400
1401 /* are we done now? */
1402 if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1403 atomic_read(&jl->j_commit_left) <= 0) {
1404 goto flush_older_and_return;
1405 }
1406
1407 /*
1408 * loop through each cnode, see if we need to write it,
1409 * or wait on a more recent transaction, or just ignore it
1410 */
1411 if (atomic_read(&journal->j_wcount) != 0) {
1412 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1413 "wcount is not 0");
1414 }
1415 cn = jl->j_realblock;
1416 while (cn) {
1417 was_jwait = 0;
1418 was_dirty = 0;
1419 saved_bh = NULL;
1420 /* blocknr of 0 is no longer in the hash, ignore it */
1421 if (cn->blocknr == 0) {
1422 goto free_cnode;
1423 }
1424
1425 /*
1426 * This transaction failed commit.
1427 * Don't write out to the disk
1428 */
1429 if (!(jl->j_state & LIST_DIRTY))
1430 goto free_cnode;
1431
1432 pjl = find_newer_jl_for_cn(cn);
1433 /*
1434 * the order is important here. We check pjl to make sure we
1435 * don't clear BH_JDirty_wait if we aren't the one writing this
1436 * block to disk
1437 */
1438 if (!pjl && cn->bh) {
1439 saved_bh = cn->bh;
1440
1441 /*
1442 * we do this to make sure nobody releases the
1443 * buffer while we are working with it
1444 */
1445 get_bh(saved_bh);
1446
1447 if (buffer_journal_dirty(saved_bh)) {
1448 BUG_ON(!can_dirty(cn));
1449 was_jwait = 1;
1450 was_dirty = 1;
1451 } else if (can_dirty(cn)) {
1452 /*
1453 * everything with !pjl && jwait
1454 * should be writable
1455 */
1456 BUG();
1457 }
1458 }
1459
1460 /*
1461 * if someone has this block in a newer transaction, just make
1462 * sure they are committed, and don't try writing it to disk
1463 */
1464 if (pjl) {
1465 if (atomic_read(&pjl->j_commit_left))
1466 flush_commit_list(s, pjl, 1);
1467 goto free_cnode;
1468 }
1469
1470 /*
1471 * bh == NULL when the block got to disk on its own, OR,
1472 * the block got freed in a future transaction
1473 */
1474 if (saved_bh == NULL) {
1475 goto free_cnode;
1476 }
1477
1478 /*
1479 * this should never happen. kupdate_one_transaction has
1480 * this list locked while it works, so we should never see a
1481 * buffer here that is not marked JDirty_wait
1482 */
1483 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1484 reiserfs_warning(s, "journal-813",
1485 "BAD! buffer %llu %cdirty %cjwait, "
1486 "not in a newer transaction",
1487 (unsigned long long)saved_bh->
1488 b_blocknr, was_dirty ? ' ' : '!',
1489 was_jwait ? ' ' : '!');
1490 }
1491 if (was_dirty) {
1492 /*
1493 * we inc again because saved_bh gets decremented
1494 * at free_cnode
1495 */
1496 get_bh(saved_bh);
1497 set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1498 lock_buffer(saved_bh);
1499 BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1500 if (buffer_dirty(saved_bh))
1501 submit_logged_buffer(saved_bh);
1502 else
1503 unlock_buffer(saved_bh);
1504 count++;
1505 } else {
1506 reiserfs_warning(s, "clm-2082",
1507 "Unable to flush buffer %llu in %s",
1508 (unsigned long long)saved_bh->
1509 b_blocknr, __func__);
1510 }
1511 free_cnode:
1512 last = cn;
1513 cn = cn->next;
1514 if (saved_bh) {
1515 /*
1516 * we incremented this to keep others from
1517 * taking the buffer head away
1518 */
1519 put_bh(saved_bh);
1520 if (atomic_read(&saved_bh->b_count) < 0) {
1521 reiserfs_warning(s, "journal-945",
1522 "saved_bh->b_count < 0");
1523 }
1524 }
1525 }
1526 if (count > 0) {
1527 cn = jl->j_realblock;
1528 while (cn) {
1529 if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1530 if (!cn->bh) {
1531 reiserfs_panic(s, "journal-1011",
1532 "cn->bh is NULL");
1533 }
1534
1535 depth = reiserfs_write_unlock_nested(s);
1536 __wait_on_buffer(cn->bh);
1537 reiserfs_write_lock_nested(s, depth);
1538
1539 if (!cn->bh) {
1540 reiserfs_panic(s, "journal-1012",
1541 "cn->bh is NULL");
1542 }
1543 if (unlikely(!buffer_uptodate(cn->bh))) {
1544 #ifdef CONFIG_REISERFS_CHECK
1545 reiserfs_warning(s, "journal-949",
1546 "buffer write failed");
1547 #endif
1548 err = -EIO;
1549 }
1550 /*
1551 * note, we must clear the JDirty_wait bit
1552 * after the up to date check, otherwise we
1553 * race against our flushpage routine
1554 */
1555 BUG_ON(!test_clear_buffer_journal_dirty
1556 (cn->bh));
1557
1558 /* drop one ref for us */
1559 put_bh(cn->bh);
1560 /* drop one ref for journal_mark_dirty */
1561 release_buffer_page(cn->bh);
1562 }
1563 cn = cn->next;
1564 }
1565 }
1566
1567 if (err)
1568 reiserfs_abort(s, -EIO,
1569 "Write error while pushing transaction to disk in %s",
1570 __func__);
1571 flush_older_and_return:
1572
1573 /*
1574 * before we can update the journal header block, we _must_ flush all
1575 * real blocks from all older transactions to disk. This is because
1576 * once the header block is updated, this transaction will not be
1577 * replayed after a crash
1578 */
1579 if (flushall) {
1580 flush_older_journal_lists(s, jl);
1581 }
1582
1583 err = journal->j_errno;
1584 /*
1585 * before we can remove everything from the hash tables for this
1586 * transaction, we must make sure it can never be replayed
1587 *
1588 * since we are only called from do_journal_end, we know for sure there
1589 * are no allocations going on while we are flushing journal lists. So,
1590 * we only need to update the journal header block for the last list
1591 * being flushed
1592 */
1593 if (!err && flushall) {
1594 err =
1595 update_journal_header_block(s,
1596 (jl->j_start + jl->j_len +
1597 2) % SB_ONDISK_JOURNAL_SIZE(s),
1598 jl->j_trans_id);
1599 if (err)
1600 reiserfs_abort(s, -EIO,
1601 "Write error while updating journal header in %s",
1602 __func__);
1603 }
1604 remove_all_from_journal_list(s, jl, 0);
1605 list_del_init(&jl->j_list);
1606 journal->j_num_lists--;
1607 del_from_work_list(s, jl);
1608
1609 if (journal->j_last_flush_id != 0 &&
1610 (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1611 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1612 journal->j_last_flush_id, jl->j_trans_id);
1613 }
1614 journal->j_last_flush_id = jl->j_trans_id;
1615
1616 /*
1617 * not strictly required since we are freeing the list, but it should
1618 * help find code using dead lists later on
1619 */
1620 jl->j_len = 0;
1621 atomic_set(&jl->j_nonzerolen, 0);
1622 jl->j_start = 0;
1623 jl->j_realblock = NULL;
1624 jl->j_commit_bh = NULL;
1625 jl->j_trans_id = 0;
1626 jl->j_state = 0;
1627 put_journal_list(s, jl);
1628 if (flushall)
1629 mutex_unlock(&journal->j_flush_mutex);
1630 return err;
1631 }
1632
1633 static int write_one_transaction(struct super_block *s,
1634 struct reiserfs_journal_list *jl,
1635 struct buffer_chunk *chunk)
1636 {
1637 struct reiserfs_journal_cnode *cn;
1638 int ret = 0;
1639
1640 jl->j_state |= LIST_TOUCHED;
1641 del_from_work_list(s, jl);
1642 if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1643 return 0;
1644 }
1645
1646 cn = jl->j_realblock;
1647 while (cn) {
1648 /*
1649 * if the blocknr == 0, this has been cleared from the hash,
1650 * skip it
1651 */
1652 if (cn->blocknr == 0) {
1653 goto next;
1654 }
1655 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1656 struct buffer_head *tmp_bh;
1657 /*
1658 * we can race against journal_mark_freed when we try
1659 * to lock_buffer(cn->bh), so we have to inc the buffer
1660 * count, and recheck things after locking
1661 */
1662 tmp_bh = cn->bh;
1663 get_bh(tmp_bh);
1664 lock_buffer(tmp_bh);
1665 if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1666 if (!buffer_journal_dirty(tmp_bh) ||
1667 buffer_journal_prepared(tmp_bh))
1668 BUG();
1669 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1670 ret++;
1671 } else {
1672 /* note, cn->bh might be null now */
1673 unlock_buffer(tmp_bh);
1674 }
1675 put_bh(tmp_bh);
1676 }
1677 next:
1678 cn = cn->next;
1679 cond_resched();
1680 }
1681 return ret;
1682 }
1683
1684 /* used by flush_commit_list */
1685 static int dirty_one_transaction(struct super_block *s,
1686 struct reiserfs_journal_list *jl)
1687 {
1688 struct reiserfs_journal_cnode *cn;
1689 struct reiserfs_journal_list *pjl;
1690 int ret = 0;
1691
1692 jl->j_state |= LIST_DIRTY;
1693 cn = jl->j_realblock;
1694 while (cn) {
1695 /*
1696 * look for a more recent transaction that logged this
1697 * buffer. Only the most recent transaction with a buffer in
1698 * it is allowed to send that buffer to disk
1699 */
1700 pjl = find_newer_jl_for_cn(cn);
1701 if (!pjl && cn->blocknr && cn->bh
1702 && buffer_journal_dirty(cn->bh)) {
1703 BUG_ON(!can_dirty(cn));
1704 /*
1705 * if the buffer is prepared, it will either be logged
1706 * or restored. If restored, we need to make sure
1707 * it actually gets marked dirty
1708 */
1709 clear_buffer_journal_new(cn->bh);
1710 if (buffer_journal_prepared(cn->bh)) {
1711 set_buffer_journal_restore_dirty(cn->bh);
1712 } else {
1713 set_buffer_journal_test(cn->bh);
1714 mark_buffer_dirty(cn->bh);
1715 }
1716 }
1717 cn = cn->next;
1718 }
1719 return ret;
1720 }
1721
1722 static int kupdate_transactions(struct super_block *s,
1723 struct reiserfs_journal_list *jl,
1724 struct reiserfs_journal_list **next_jl,
1725 unsigned int *next_trans_id,
1726 int num_blocks, int num_trans)
1727 {
1728 int ret = 0;
1729 int written = 0;
1730 int transactions_flushed = 0;
1731 unsigned int orig_trans_id = jl->j_trans_id;
1732 struct buffer_chunk chunk;
1733 struct list_head *entry;
1734 struct reiserfs_journal *journal = SB_JOURNAL(s);
1735 chunk.nr = 0;
1736
1737 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1738 if (!journal_list_still_alive(s, orig_trans_id)) {
1739 goto done;
1740 }
1741
1742 /*
1743 * we've got j_flush_mutex held, nobody is going to delete any
1744 * of these lists out from underneath us
1745 */
1746 while ((num_trans && transactions_flushed < num_trans) ||
1747 (!num_trans && written < num_blocks)) {
1748
1749 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1750 atomic_read(&jl->j_commit_left)
1751 || !(jl->j_state & LIST_DIRTY)) {
1752 del_from_work_list(s, jl);
1753 break;
1754 }
1755 ret = write_one_transaction(s, jl, &chunk);
1756
1757 if (ret < 0)
1758 goto done;
1759 transactions_flushed++;
1760 written += ret;
1761 entry = jl->j_list.next;
1762
1763 /* did we wrap? */
1764 if (entry == &journal->j_journal_list) {
1765 break;
1766 }
1767 jl = JOURNAL_LIST_ENTRY(entry);
1768
1769 /* don't bother with older transactions */
1770 if (jl->j_trans_id <= orig_trans_id)
1771 break;
1772 }
1773 if (chunk.nr) {
1774 write_chunk(&chunk);
1775 }
1776
1777 done:
1778 mutex_unlock(&journal->j_flush_mutex);
1779 return ret;
1780 }
1781
1782 /*
1783 * for o_sync and fsync heavy applications, they tend to use
1784 * all the journa list slots with tiny transactions. These
1785 * trigger lots and lots of calls to update the header block, which
1786 * adds seeks and slows things down.
1787 *
1788 * This function tries to clear out a large chunk of the journal lists
1789 * at once, which makes everything faster since only the newest journal
1790 * list updates the header block
1791 */
1792 static int flush_used_journal_lists(struct super_block *s,
1793 struct reiserfs_journal_list *jl)
1794 {
1795 unsigned long len = 0;
1796 unsigned long cur_len;
1797 int ret;
1798 int i;
1799 int limit = 256;
1800 struct reiserfs_journal_list *tjl;
1801 struct reiserfs_journal_list *flush_jl;
1802 unsigned int trans_id;
1803 struct reiserfs_journal *journal = SB_JOURNAL(s);
1804
1805 flush_jl = tjl = jl;
1806
1807 /* in data logging mode, try harder to flush a lot of blocks */
1808 if (reiserfs_data_log(s))
1809 limit = 1024;
1810 /* flush for 256 transactions or limit blocks, whichever comes first */
1811 for (i = 0; i < 256 && len < limit; i++) {
1812 if (atomic_read(&tjl->j_commit_left) ||
1813 tjl->j_trans_id < jl->j_trans_id) {
1814 break;
1815 }
1816 cur_len = atomic_read(&tjl->j_nonzerolen);
1817 if (cur_len > 0) {
1818 tjl->j_state &= ~LIST_TOUCHED;
1819 }
1820 len += cur_len;
1821 flush_jl = tjl;
1822 if (tjl->j_list.next == &journal->j_journal_list)
1823 break;
1824 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1825 }
1826 get_journal_list(jl);
1827 get_journal_list(flush_jl);
1828
1829 /*
1830 * try to find a group of blocks we can flush across all the
1831 * transactions, but only bother if we've actually spanned
1832 * across multiple lists
1833 */
1834 if (flush_jl != jl) {
1835 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1836 }
1837 flush_journal_list(s, flush_jl, 1);
1838 put_journal_list(s, flush_jl);
1839 put_journal_list(s, jl);
1840 return 0;
1841 }
1842
1843 /*
1844 * removes any nodes in table with name block and dev as bh.
1845 * only touchs the hnext and hprev pointers.
1846 */
1847 void remove_journal_hash(struct super_block *sb,
1848 struct reiserfs_journal_cnode **table,
1849 struct reiserfs_journal_list *jl,
1850 unsigned long block, int remove_freed)
1851 {
1852 struct reiserfs_journal_cnode *cur;
1853 struct reiserfs_journal_cnode **head;
1854
1855 head = &(journal_hash(table, sb, block));
1856 if (!head) {
1857 return;
1858 }
1859 cur = *head;
1860 while (cur) {
1861 if (cur->blocknr == block && cur->sb == sb
1862 && (jl == NULL || jl == cur->jlist)
1863 && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1864 if (cur->hnext) {
1865 cur->hnext->hprev = cur->hprev;
1866 }
1867 if (cur->hprev) {
1868 cur->hprev->hnext = cur->hnext;
1869 } else {
1870 *head = cur->hnext;
1871 }
1872 cur->blocknr = 0;
1873 cur->sb = NULL;
1874 cur->state = 0;
1875 /*
1876 * anybody who clears the cur->bh will also
1877 * dec the nonzerolen
1878 */
1879 if (cur->bh && cur->jlist)
1880 atomic_dec(&cur->jlist->j_nonzerolen);
1881 cur->bh = NULL;
1882 cur->jlist = NULL;
1883 }
1884 cur = cur->hnext;
1885 }
1886 }
1887
1888 static void free_journal_ram(struct super_block *sb)
1889 {
1890 struct reiserfs_journal *journal = SB_JOURNAL(sb);
1891 kfree(journal->j_current_jl);
1892 journal->j_num_lists--;
1893
1894 vfree(journal->j_cnode_free_orig);
1895 free_list_bitmaps(sb, journal->j_list_bitmap);
1896 free_bitmap_nodes(sb); /* must be after free_list_bitmaps */
1897 if (journal->j_header_bh) {
1898 brelse(journal->j_header_bh);
1899 }
1900 /*
1901 * j_header_bh is on the journal dev, make sure
1902 * not to release the journal dev until we brelse j_header_bh
1903 */
1904 release_journal_dev(sb, journal);
1905 vfree(journal);
1906 }
1907
1908 /*
1909 * call on unmount. Only set error to 1 if you haven't made your way out
1910 * of read_super() yet. Any other caller must keep error at 0.
1911 */
1912 static int do_journal_release(struct reiserfs_transaction_handle *th,
1913 struct super_block *sb, int error)
1914 {
1915 struct reiserfs_transaction_handle myth;
1916 int flushed = 0;
1917 struct reiserfs_journal *journal = SB_JOURNAL(sb);
1918
1919 /*
1920 * we only want to flush out transactions if we were
1921 * called with error == 0
1922 */
1923 if (!error && !sb_rdonly(sb)) {
1924 /* end the current trans */
1925 BUG_ON(!th->t_trans_id);
1926 do_journal_end(th, FLUSH_ALL);
1927
1928 /*
1929 * make sure something gets logged to force
1930 * our way into the flush code
1931 */
1932 if (!journal_join(&myth, sb)) {
1933 reiserfs_prepare_for_journal(sb,
1934 SB_BUFFER_WITH_SB(sb),
1935 1);
1936 journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1937 do_journal_end(&myth, FLUSH_ALL);
1938 flushed = 1;
1939 }
1940 }
1941
1942 /* this also catches errors during the do_journal_end above */
1943 if (!error && reiserfs_is_journal_aborted(journal)) {
1944 memset(&myth, 0, sizeof(myth));
1945 if (!journal_join_abort(&myth, sb)) {
1946 reiserfs_prepare_for_journal(sb,
1947 SB_BUFFER_WITH_SB(sb),
1948 1);
1949 journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1950 do_journal_end(&myth, FLUSH_ALL);
1951 }
1952 }
1953
1954
1955 /*
1956 * We must release the write lock here because
1957 * the workqueue job (flush_async_commit) needs this lock
1958 */
1959 reiserfs_write_unlock(sb);
1960
1961 /*
1962 * Cancel flushing of old commits. Note that neither of these works
1963 * will be requeued because superblock is being shutdown and doesn't
1964 * have SB_ACTIVE set.
1965 */
1966 reiserfs_cancel_old_flush(sb);
1967 /* wait for all commits to finish */
1968 cancel_delayed_work_sync(&SB_JOURNAL(sb)->j_work);
1969
1970 free_journal_ram(sb);
1971
1972 reiserfs_write_lock(sb);
1973
1974 return 0;
1975 }
1976
1977 /* * call on unmount. flush all journal trans, release all alloc'd ram */
1978 int journal_release(struct reiserfs_transaction_handle *th,
1979 struct super_block *sb)
1980 {
1981 return do_journal_release(th, sb, 0);
1982 }
1983
1984 /* only call from an error condition inside reiserfs_read_super! */
1985 int journal_release_error(struct reiserfs_transaction_handle *th,
1986 struct super_block *sb)
1987 {
1988 return do_journal_release(th, sb, 1);
1989 }
1990
1991 /*
1992 * compares description block with commit block.
1993 * returns 1 if they differ, 0 if they are the same
1994 */
1995 static int journal_compare_desc_commit(struct super_block *sb,
1996 struct reiserfs_journal_desc *desc,
1997 struct reiserfs_journal_commit *commit)
1998 {
1999 if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
2000 get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
2001 get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
2002 get_commit_trans_len(commit) <= 0) {
2003 return 1;
2004 }
2005 return 0;
2006 }
2007
2008 /*
2009 * returns 0 if it did not find a description block
2010 * returns -1 if it found a corrupt commit block
2011 * returns 1 if both desc and commit were valid
2012 * NOTE: only called during fs mount
2013 */
2014 static int journal_transaction_is_valid(struct super_block *sb,
2015 struct buffer_head *d_bh,
2016 unsigned int *oldest_invalid_trans_id,
2017 unsigned long *newest_mount_id)
2018 {
2019 struct reiserfs_journal_desc *desc;
2020 struct reiserfs_journal_commit *commit;
2021 struct buffer_head *c_bh;
2022 unsigned long offset;
2023
2024 if (!d_bh)
2025 return 0;
2026
2027 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2028 if (get_desc_trans_len(desc) > 0
2029 && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2030 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2031 && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2032 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2033 "journal-986: transaction "
2034 "is valid returning because trans_id %d is greater than "
2035 "oldest_invalid %lu",
2036 get_desc_trans_id(desc),
2037 *oldest_invalid_trans_id);
2038 return 0;
2039 }
2040 if (newest_mount_id
2041 && *newest_mount_id > get_desc_mount_id(desc)) {
2042 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2043 "journal-1087: transaction "
2044 "is valid returning because mount_id %d is less than "
2045 "newest_mount_id %lu",
2046 get_desc_mount_id(desc),
2047 *newest_mount_id);
2048 return -1;
2049 }
2050 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2051 reiserfs_warning(sb, "journal-2018",
2052 "Bad transaction length %d "
2053 "encountered, ignoring transaction",
2054 get_desc_trans_len(desc));
2055 return -1;
2056 }
2057 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2058
2059 /*
2060 * ok, we have a journal description block,
2061 * let's see if the transaction was valid
2062 */
2063 c_bh =
2064 journal_bread(sb,
2065 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2066 ((offset + get_desc_trans_len(desc) +
2067 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2068 if (!c_bh)
2069 return 0;
2070 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2071 if (journal_compare_desc_commit(sb, desc, commit)) {
2072 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2073 "journal_transaction_is_valid, commit offset %ld had bad "
2074 "time %d or length %d",
2075 c_bh->b_blocknr -
2076 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2077 get_commit_trans_id(commit),
2078 get_commit_trans_len(commit));
2079 brelse(c_bh);
2080 if (oldest_invalid_trans_id) {
2081 *oldest_invalid_trans_id =
2082 get_desc_trans_id(desc);
2083 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2084 "journal-1004: "
2085 "transaction_is_valid setting oldest invalid trans_id "
2086 "to %d",
2087 get_desc_trans_id(desc));
2088 }
2089 return -1;
2090 }
2091 brelse(c_bh);
2092 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2093 "journal-1006: found valid "
2094 "transaction start offset %llu, len %d id %d",
2095 d_bh->b_blocknr -
2096 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2097 get_desc_trans_len(desc),
2098 get_desc_trans_id(desc));
2099 return 1;
2100 } else {
2101 return 0;
2102 }
2103 }
2104
2105 static void brelse_array(struct buffer_head **heads, int num)
2106 {
2107 int i;
2108 for (i = 0; i < num; i++) {
2109 brelse(heads[i]);
2110 }
2111 }
2112
2113 /*
2114 * given the start, and values for the oldest acceptable transactions,
2115 * this either reads in a replays a transaction, or returns because the
2116 * transaction is invalid, or too old.
2117 * NOTE: only called during fs mount
2118 */
2119 static int journal_read_transaction(struct super_block *sb,
2120 unsigned long cur_dblock,
2121 unsigned long oldest_start,
2122 unsigned int oldest_trans_id,
2123 unsigned long newest_mount_id)
2124 {
2125 struct reiserfs_journal *journal = SB_JOURNAL(sb);
2126 struct reiserfs_journal_desc *desc;
2127 struct reiserfs_journal_commit *commit;
2128 unsigned int trans_id = 0;
2129 struct buffer_head *c_bh;
2130 struct buffer_head *d_bh;
2131 struct buffer_head **log_blocks = NULL;
2132 struct buffer_head **real_blocks = NULL;
2133 unsigned int trans_offset;
2134 int i;
2135 int trans_half;
2136
2137 d_bh = journal_bread(sb, cur_dblock);
2138 if (!d_bh)
2139 return 1;
2140 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2141 trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2142 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2143 "journal_read_transaction, offset %llu, len %d mount_id %d",
2144 d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2145 get_desc_trans_len(desc), get_desc_mount_id(desc));
2146 if (get_desc_trans_id(desc) < oldest_trans_id) {
2147 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2148 "journal_read_trans skipping because %lu is too old",
2149 cur_dblock -
2150 SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2151 brelse(d_bh);
2152 return 1;
2153 }
2154 if (get_desc_mount_id(desc) != newest_mount_id) {
2155 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2156 "journal_read_trans skipping because %d is != "
2157 "newest_mount_id %lu", get_desc_mount_id(desc),
2158 newest_mount_id);
2159 brelse(d_bh);
2160 return 1;
2161 }
2162 c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2163 ((trans_offset + get_desc_trans_len(desc) + 1) %
2164 SB_ONDISK_JOURNAL_SIZE(sb)));
2165 if (!c_bh) {
2166 brelse(d_bh);
2167 return 1;
2168 }
2169 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2170 if (journal_compare_desc_commit(sb, desc, commit)) {
2171 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2172 "journal_read_transaction, "
2173 "commit offset %llu had bad time %d or length %d",
2174 c_bh->b_blocknr -
2175 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2176 get_commit_trans_id(commit),
2177 get_commit_trans_len(commit));
2178 brelse(c_bh);
2179 brelse(d_bh);
2180 return 1;
2181 }
2182
2183 if (bdev_read_only(sb->s_bdev)) {
2184 reiserfs_warning(sb, "clm-2076",
2185 "device is readonly, unable to replay log");
2186 brelse(c_bh);
2187 brelse(d_bh);
2188 return -EROFS;
2189 }
2190
2191 trans_id = get_desc_trans_id(desc);
2192 /*
2193 * now we know we've got a good transaction, and it was
2194 * inside the valid time ranges
2195 */
2196 log_blocks = kmalloc_array(get_desc_trans_len(desc),
2197 sizeof(struct buffer_head *),
2198 GFP_NOFS);
2199 real_blocks = kmalloc_array(get_desc_trans_len(desc),
2200 sizeof(struct buffer_head *),
2201 GFP_NOFS);
2202 if (!log_blocks || !real_blocks) {
2203 brelse(c_bh);
2204 brelse(d_bh);
2205 kfree(log_blocks);
2206 kfree(real_blocks);
2207 reiserfs_warning(sb, "journal-1169",
2208 "kmalloc failed, unable to mount FS");
2209 return -1;
2210 }
2211 /* get all the buffer heads */
2212 trans_half = journal_trans_half(sb->s_blocksize);
2213 for (i = 0; i < get_desc_trans_len(desc); i++) {
2214 log_blocks[i] =
2215 journal_getblk(sb,
2216 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2217 (trans_offset + 1 +
2218 i) % SB_ONDISK_JOURNAL_SIZE(sb));
2219 if (i < trans_half) {
2220 real_blocks[i] =
2221 sb_getblk(sb,
2222 le32_to_cpu(desc->j_realblock[i]));
2223 } else {
2224 real_blocks[i] =
2225 sb_getblk(sb,
2226 le32_to_cpu(commit->
2227 j_realblock[i - trans_half]));
2228 }
2229 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2230 reiserfs_warning(sb, "journal-1207",
2231 "REPLAY FAILURE fsck required! "
2232 "Block to replay is outside of "
2233 "filesystem");
2234 goto abort_replay;
2235 }
2236 /* make sure we don't try to replay onto log or reserved area */
2237 if (is_block_in_log_or_reserved_area
2238 (sb, real_blocks[i]->b_blocknr)) {
2239 reiserfs_warning(sb, "journal-1204",
2240 "REPLAY FAILURE fsck required! "
2241 "Trying to replay onto a log block");
2242 abort_replay:
2243 brelse_array(log_blocks, i);
2244 brelse_array(real_blocks, i);
2245 brelse(c_bh);
2246 brelse(d_bh);
2247 kfree(log_blocks);
2248 kfree(real_blocks);
2249 return -1;
2250 }
2251 }
2252 /* read in the log blocks, memcpy to the corresponding real block */
2253 ll_rw_block(REQ_OP_READ, 0, get_desc_trans_len(desc), log_blocks);
2254 for (i = 0; i < get_desc_trans_len(desc); i++) {
2255
2256 wait_on_buffer(log_blocks[i]);
2257 if (!buffer_uptodate(log_blocks[i])) {
2258 reiserfs_warning(sb, "journal-1212",
2259 "REPLAY FAILURE fsck required! "
2260 "buffer write failed");
2261 brelse_array(log_blocks + i,
2262 get_desc_trans_len(desc) - i);
2263 brelse_array(real_blocks, get_desc_trans_len(desc));
2264 brelse(c_bh);
2265 brelse(d_bh);
2266 kfree(log_blocks);
2267 kfree(real_blocks);
2268 return -1;
2269 }
2270 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2271 real_blocks[i]->b_size);
2272 set_buffer_uptodate(real_blocks[i]);
2273 brelse(log_blocks[i]);
2274 }
2275 /* flush out the real blocks */
2276 for (i = 0; i < get_desc_trans_len(desc); i++) {
2277 set_buffer_dirty(real_blocks[i]);
2278 write_dirty_buffer(real_blocks[i], 0);
2279 }
2280 for (i = 0; i < get_desc_trans_len(desc); i++) {
2281 wait_on_buffer(real_blocks[i]);
2282 if (!buffer_uptodate(real_blocks[i])) {
2283 reiserfs_warning(sb, "journal-1226",
2284 "REPLAY FAILURE, fsck required! "
2285 "buffer write failed");
2286 brelse_array(real_blocks + i,
2287 get_desc_trans_len(desc) - i);
2288 brelse(c_bh);
2289 brelse(d_bh);
2290 kfree(log_blocks);
2291 kfree(real_blocks);
2292 return -1;
2293 }
2294 brelse(real_blocks[i]);
2295 }
2296 cur_dblock =
2297 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2298 ((trans_offset + get_desc_trans_len(desc) +
2299 2) % SB_ONDISK_JOURNAL_SIZE(sb));
2300 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2301 "journal-1095: setting journal " "start to offset %ld",
2302 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2303
2304 /*
2305 * init starting values for the first transaction, in case
2306 * this is the last transaction to be replayed.
2307 */
2308 journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2309 journal->j_last_flush_trans_id = trans_id;
2310 journal->j_trans_id = trans_id + 1;
2311 /* check for trans_id overflow */
2312 if (journal->j_trans_id == 0)
2313 journal->j_trans_id = 10;
2314 brelse(c_bh);
2315 brelse(d_bh);
2316 kfree(log_blocks);
2317 kfree(real_blocks);
2318 return 0;
2319 }
2320
2321 /*
2322 * This function reads blocks starting from block and to max_block of bufsize
2323 * size (but no more than BUFNR blocks at a time). This proved to improve
2324 * mounting speed on self-rebuilding raid5 arrays at least.
2325 * Right now it is only used from journal code. But later we might use it
2326 * from other places.
2327 * Note: Do not use journal_getblk/sb_getblk functions here!
2328 */
2329 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2330 b_blocknr_t block, int bufsize,
2331 b_blocknr_t max_block)
2332 {
2333 struct buffer_head *bhlist[BUFNR];
2334 unsigned int blocks = BUFNR;
2335 struct buffer_head *bh;
2336 int i, j;
2337
2338 bh = __getblk(dev, block, bufsize);
2339 if (buffer_uptodate(bh))
2340 return (bh);
2341
2342 if (block + BUFNR > max_block) {
2343 blocks = max_block - block;
2344 }
2345 bhlist[0] = bh;
2346 j = 1;
2347 for (i = 1; i < blocks; i++) {
2348 bh = __getblk(dev, block + i, bufsize);
2349 if (buffer_uptodate(bh)) {
2350 brelse(bh);
2351 break;
2352 } else
2353 bhlist[j++] = bh;
2354 }
2355 ll_rw_block(REQ_OP_READ, 0, j, bhlist);
2356 for (i = 1; i < j; i++)
2357 brelse(bhlist[i]);
2358 bh = bhlist[0];
2359 wait_on_buffer(bh);
2360 if (buffer_uptodate(bh))
2361 return bh;
2362 brelse(bh);
2363 return NULL;
2364 }
2365
2366 /*
2367 * read and replay the log
2368 * on a clean unmount, the journal header's next unflushed pointer will be
2369 * to an invalid transaction. This tests that before finding all the
2370 * transactions in the log, which makes normal mount times fast.
2371 *
2372 * After a crash, this starts with the next unflushed transaction, and
2373 * replays until it finds one too old, or invalid.
2374 *
2375 * On exit, it sets things up so the first transaction will work correctly.
2376 * NOTE: only called during fs mount
2377 */
2378 static int journal_read(struct super_block *sb)
2379 {
2380 struct reiserfs_journal *journal = SB_JOURNAL(sb);
2381 struct reiserfs_journal_desc *desc;
2382 unsigned int oldest_trans_id = 0;
2383 unsigned int oldest_invalid_trans_id = 0;
2384 time64_t start;
2385 unsigned long oldest_start = 0;
2386 unsigned long cur_dblock = 0;
2387 unsigned long newest_mount_id = 9;
2388 struct buffer_head *d_bh;
2389 struct reiserfs_journal_header *jh;
2390 int valid_journal_header = 0;
2391 int replay_count = 0;
2392 int continue_replay = 1;
2393 int ret;
2394
2395 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2396 reiserfs_info(sb, "checking transaction log (%pg)\n",
2397 journal->j_dev_bd);
2398 start = ktime_get_seconds();
2399
2400 /*
2401 * step 1, read in the journal header block. Check the transaction
2402 * it says is the first unflushed, and if that transaction is not
2403 * valid, replay is done
2404 */
2405 journal->j_header_bh = journal_bread(sb,
2406 SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2407 + SB_ONDISK_JOURNAL_SIZE(sb));
2408 if (!journal->j_header_bh) {
2409 return 1;
2410 }
2411 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2412 if (le32_to_cpu(jh->j_first_unflushed_offset) <
2413 SB_ONDISK_JOURNAL_SIZE(sb)
2414 && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2415 oldest_start =
2416 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2417 le32_to_cpu(jh->j_first_unflushed_offset);
2418 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2419 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2420 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2421 "journal-1153: found in "
2422 "header: first_unflushed_offset %d, last_flushed_trans_id "
2423 "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2424 le32_to_cpu(jh->j_last_flush_trans_id));
2425 valid_journal_header = 1;
2426
2427 /*
2428 * now, we try to read the first unflushed offset. If it
2429 * is not valid, there is nothing more we can do, and it
2430 * makes no sense to read through the whole log.
2431 */
2432 d_bh =
2433 journal_bread(sb,
2434 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2435 le32_to_cpu(jh->j_first_unflushed_offset));
2436 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2437 if (!ret) {
2438 continue_replay = 0;
2439 }
2440 brelse(d_bh);
2441 goto start_log_replay;
2442 }
2443
2444 /*
2445 * ok, there are transactions that need to be replayed. start
2446 * with the first log block, find all the valid transactions, and
2447 * pick out the oldest.
2448 */
2449 while (continue_replay
2450 && cur_dblock <
2451 (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2452 SB_ONDISK_JOURNAL_SIZE(sb))) {
2453 /*
2454 * Note that it is required for blocksize of primary fs
2455 * device and journal device to be the same
2456 */
2457 d_bh =
2458 reiserfs_breada(journal->j_dev_bd, cur_dblock,
2459 sb->s_blocksize,
2460 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2461 SB_ONDISK_JOURNAL_SIZE(sb));
2462 ret =
2463 journal_transaction_is_valid(sb, d_bh,
2464 &oldest_invalid_trans_id,
2465 &newest_mount_id);
2466 if (ret == 1) {
2467 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2468 if (oldest_start == 0) { /* init all oldest_ values */
2469 oldest_trans_id = get_desc_trans_id(desc);
2470 oldest_start = d_bh->b_blocknr;
2471 newest_mount_id = get_desc_mount_id(desc);
2472 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2473 "journal-1179: Setting "
2474 "oldest_start to offset %llu, trans_id %lu",
2475 oldest_start -
2476 SB_ONDISK_JOURNAL_1st_BLOCK
2477 (sb), oldest_trans_id);
2478 } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2479 /* one we just read was older */
2480 oldest_trans_id = get_desc_trans_id(desc);
2481 oldest_start = d_bh->b_blocknr;
2482 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2483 "journal-1180: Resetting "
2484 "oldest_start to offset %lu, trans_id %lu",
2485 oldest_start -
2486 SB_ONDISK_JOURNAL_1st_BLOCK
2487 (sb), oldest_trans_id);
2488 }
2489 if (newest_mount_id < get_desc_mount_id(desc)) {
2490 newest_mount_id = get_desc_mount_id(desc);
2491 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2492 "journal-1299: Setting "
2493 "newest_mount_id to %d",
2494 get_desc_mount_id(desc));
2495 }
2496 cur_dblock += get_desc_trans_len(desc) + 2;
2497 } else {
2498 cur_dblock++;
2499 }
2500 brelse(d_bh);
2501 }
2502
2503 start_log_replay:
2504 cur_dblock = oldest_start;
2505 if (oldest_trans_id) {
2506 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2507 "journal-1206: Starting replay "
2508 "from offset %llu, trans_id %lu",
2509 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2510 oldest_trans_id);
2511
2512 }
2513 replay_count = 0;
2514 while (continue_replay && oldest_trans_id > 0) {
2515 ret =
2516 journal_read_transaction(sb, cur_dblock, oldest_start,
2517 oldest_trans_id, newest_mount_id);
2518 if (ret < 0) {
2519 return ret;
2520 } else if (ret != 0) {
2521 break;
2522 }
2523 cur_dblock =
2524 SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2525 replay_count++;
2526 if (cur_dblock == oldest_start)
2527 break;
2528 }
2529
2530 if (oldest_trans_id == 0) {
2531 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2532 "journal-1225: No valid " "transactions found");
2533 }
2534 /*
2535 * j_start does not get set correctly if we don't replay any
2536 * transactions. if we had a valid journal_header, set j_start
2537 * to the first unflushed transaction value, copy the trans_id
2538 * from the header
2539 */
2540 if (valid_journal_header && replay_count == 0) {
2541 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2542 journal->j_trans_id =
2543 le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2544 /* check for trans_id overflow */
2545 if (journal->j_trans_id == 0)
2546 journal->j_trans_id = 10;
2547 journal->j_last_flush_trans_id =
2548 le32_to_cpu(jh->j_last_flush_trans_id);
2549 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2550 } else {
2551 journal->j_mount_id = newest_mount_id + 1;
2552 }
2553 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2554 "newest_mount_id to %lu", journal->j_mount_id);
2555 journal->j_first_unflushed_offset = journal->j_start;
2556 if (replay_count > 0) {
2557 reiserfs_info(sb,
2558 "replayed %d transactions in %lu seconds\n",
2559 replay_count, ktime_get_seconds() - start);
2560 }
2561 /* needed to satisfy the locking in _update_journal_header_block */
2562 reiserfs_write_lock(sb);
2563 if (!bdev_read_only(sb->s_bdev) &&
2564 _update_journal_header_block(sb, journal->j_start,
2565 journal->j_last_flush_trans_id)) {
2566 reiserfs_write_unlock(sb);
2567 /*
2568 * replay failed, caller must call free_journal_ram and abort
2569 * the mount
2570 */
2571 return -1;
2572 }
2573 reiserfs_write_unlock(sb);
2574 return 0;
2575 }
2576
2577 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2578 {
2579 struct reiserfs_journal_list *jl;
2580 jl = kzalloc(sizeof(struct reiserfs_journal_list),
2581 GFP_NOFS | __GFP_NOFAIL);
2582 INIT_LIST_HEAD(&jl->j_list);
2583 INIT_LIST_HEAD(&jl->j_working_list);
2584 INIT_LIST_HEAD(&jl->j_tail_bh_list);
2585 INIT_LIST_HEAD(&jl->j_bh_list);
2586 mutex_init(&jl->j_commit_mutex);
2587 SB_JOURNAL(s)->j_num_lists++;
2588 get_journal_list(jl);
2589 return jl;
2590 }
2591
2592 static void journal_list_init(struct super_block *sb)
2593 {
2594 SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2595 }
2596
2597 static void release_journal_dev(struct super_block *super,
2598 struct reiserfs_journal *journal)
2599 {
2600 if (journal->j_dev_bd != NULL) {
2601 blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2602 journal->j_dev_bd = NULL;
2603 }
2604 }
2605
2606 static int journal_init_dev(struct super_block *super,
2607 struct reiserfs_journal *journal,
2608 const char *jdev_name)
2609 {
2610 int result;
2611 dev_t jdev;
2612 fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
2613 char b[BDEVNAME_SIZE];
2614
2615 result = 0;
2616
2617 journal->j_dev_bd = NULL;
2618 jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2619 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2620
2621 if (bdev_read_only(super->s_bdev))
2622 blkdev_mode = FMODE_READ;
2623
2624 /* there is no "jdev" option and journal is on separate device */
2625 if ((!jdev_name || !jdev_name[0])) {
2626 if (jdev == super->s_dev)
2627 blkdev_mode &= ~FMODE_EXCL;
2628 journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2629 journal);
2630 journal->j_dev_mode = blkdev_mode;
2631 if (IS_ERR(journal->j_dev_bd)) {
2632 result = PTR_ERR(journal->j_dev_bd);
2633 journal->j_dev_bd = NULL;
2634 reiserfs_warning(super, "sh-458",
2635 "cannot init journal device '%s': %i",
2636 __bdevname(jdev, b), result);
2637 return result;
2638 } else if (jdev != super->s_dev)
2639 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2640
2641 return 0;
2642 }
2643
2644 journal->j_dev_mode = blkdev_mode;
2645 journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
2646 if (IS_ERR(journal->j_dev_bd)) {
2647 result = PTR_ERR(journal->j_dev_bd);
2648 journal->j_dev_bd = NULL;
2649 reiserfs_warning(super, "sh-457",
2650 "journal_init_dev: Cannot open '%s': %i",
2651 jdev_name, result);
2652 return result;
2653 }
2654
2655 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2656 reiserfs_info(super,
2657 "journal_init_dev: journal device: %pg\n",
2658 journal->j_dev_bd);
2659 return 0;
2660 }
2661
2662 /*
2663 * When creating/tuning a file system user can assign some
2664 * journal params within boundaries which depend on the ratio
2665 * blocksize/standard_blocksize.
2666 *
2667 * For blocks >= standard_blocksize transaction size should
2668 * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2669 * then JOURNAL_TRANS_MAX_DEFAULT.
2670 *
2671 * For blocks < standard_blocksize these boundaries should be
2672 * decreased proportionally.
2673 */
2674 #define REISERFS_STANDARD_BLKSIZE (4096)
2675
2676 static int check_advise_trans_params(struct super_block *sb,
2677 struct reiserfs_journal *journal)
2678 {
2679 if (journal->j_trans_max) {
2680 /* Non-default journal params. Do sanity check for them. */
2681 int ratio = 1;
2682 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2683 ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2684
2685 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2686 journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2687 SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2688 JOURNAL_MIN_RATIO) {
2689 reiserfs_warning(sb, "sh-462",
2690 "bad transaction max size (%u). "
2691 "FSCK?", journal->j_trans_max);
2692 return 1;
2693 }
2694 if (journal->j_max_batch != (journal->j_trans_max) *
2695 JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2696 reiserfs_warning(sb, "sh-463",
2697 "bad transaction max batch (%u). "
2698 "FSCK?", journal->j_max_batch);
2699 return 1;
2700 }
2701 } else {
2702 /*
2703 * Default journal params.
2704 * The file system was created by old version
2705 * of mkreiserfs, so some fields contain zeros,
2706 * and we need to advise proper values for them
2707 */
2708 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2709 reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2710 sb->s_blocksize);
2711 return 1;
2712 }
2713 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2714 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2715 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2716 }
2717 return 0;
2718 }
2719
2720 /* must be called once on fs mount. calls journal_read for you */
2721 int journal_init(struct super_block *sb, const char *j_dev_name,
2722 int old_format, unsigned int commit_max_age)
2723 {
2724 int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2725 struct buffer_head *bhjh;
2726 struct reiserfs_super_block *rs;
2727 struct reiserfs_journal_header *jh;
2728 struct reiserfs_journal *journal;
2729 struct reiserfs_journal_list *jl;
2730 int ret;
2731
2732 journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal));
2733 if (!journal) {
2734 reiserfs_warning(sb, "journal-1256",
2735 "unable to get memory for journal structure");
2736 return 1;
2737 }
2738 INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2739 INIT_LIST_HEAD(&journal->j_prealloc_list);
2740 INIT_LIST_HEAD(&journal->j_working_list);
2741 INIT_LIST_HEAD(&journal->j_journal_list);
2742 journal->j_persistent_trans = 0;
2743 if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2744 reiserfs_bmap_count(sb)))
2745 goto free_and_return;
2746
2747 allocate_bitmap_nodes(sb);
2748
2749 /* reserved for journal area support */
2750 SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2751 REISERFS_OLD_DISK_OFFSET_IN_BYTES
2752 / sb->s_blocksize +
2753 reiserfs_bmap_count(sb) +
2754 1 :
2755 REISERFS_DISK_OFFSET_IN_BYTES /
2756 sb->s_blocksize + 2);
2757
2758 /*
2759 * Sanity check to see is the standard journal fitting
2760 * within first bitmap (actual for small blocksizes)
2761 */
2762 if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2763 (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2764 SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2765 reiserfs_warning(sb, "journal-1393",
2766 "journal does not fit for area addressed "
2767 "by first of bitmap blocks. It starts at "
2768 "%u and its size is %u. Block size %ld",
2769 SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2770 SB_ONDISK_JOURNAL_SIZE(sb),
2771 sb->s_blocksize);
2772 goto free_and_return;
2773 }
2774
2775 if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2776 reiserfs_warning(sb, "sh-462",
2777 "unable to initialize journal device");
2778 goto free_and_return;
2779 }
2780
2781 rs = SB_DISK_SUPER_BLOCK(sb);
2782
2783 /* read journal header */
2784 bhjh = journal_bread(sb,
2785 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2786 SB_ONDISK_JOURNAL_SIZE(sb));
2787 if (!bhjh) {
2788 reiserfs_warning(sb, "sh-459",
2789 "unable to read journal header");
2790 goto free_and_return;
2791 }
2792 jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2793
2794 /* make sure that journal matches to the super block */
2795 if (is_reiserfs_jr(rs)
2796 && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2797 sb_jp_journal_magic(rs))) {
2798 reiserfs_warning(sb, "sh-460",
2799 "journal header magic %x (device %pg) does "
2800 "not match to magic found in super block %x",
2801 jh->jh_journal.jp_journal_magic,
2802 journal->j_dev_bd,
2803 sb_jp_journal_magic(rs));
2804 brelse(bhjh);
2805 goto free_and_return;
2806 }
2807
2808 journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2809 journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2810 journal->j_max_commit_age =
2811 le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2812 journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2813
2814 if (check_advise_trans_params(sb, journal) != 0)
2815 goto free_and_return;
2816 journal->j_default_max_commit_age = journal->j_max_commit_age;
2817
2818 if (commit_max_age != 0) {
2819 journal->j_max_commit_age = commit_max_age;
2820 journal->j_max_trans_age = commit_max_age;
2821 }
2822
2823 reiserfs_info(sb, "journal params: device %pg, size %u, "
2824 "journal first block %u, max trans len %u, max batch %u, "
2825 "max commit age %u, max trans age %u\n",
2826 journal->j_dev_bd,
2827 SB_ONDISK_JOURNAL_SIZE(sb),
2828 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2829 journal->j_trans_max,
2830 journal->j_max_batch,
2831 journal->j_max_commit_age, journal->j_max_trans_age);
2832
2833 brelse(bhjh);
2834
2835 journal->j_list_bitmap_index = 0;
2836 journal_list_init(sb);
2837
2838 memset(journal->j_list_hash_table, 0,
2839 JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2840
2841 INIT_LIST_HEAD(&journal->j_dirty_buffers);
2842 spin_lock_init(&journal->j_dirty_buffers_lock);
2843
2844 journal->j_start = 0;
2845 journal->j_len = 0;
2846 journal->j_len_alloc = 0;
2847 atomic_set(&journal->j_wcount, 0);
2848 atomic_set(&journal->j_async_throttle, 0);
2849 journal->j_bcount = 0;
2850 journal->j_trans_start_time = 0;
2851 journal->j_last = NULL;
2852 journal->j_first = NULL;
2853 init_waitqueue_head(&journal->j_join_wait);
2854 mutex_init(&journal->j_mutex);
2855 mutex_init(&journal->j_flush_mutex);
2856
2857 journal->j_trans_id = 10;
2858 journal->j_mount_id = 10;
2859 journal->j_state = 0;
2860 atomic_set(&journal->j_jlock, 0);
2861 journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2862 journal->j_cnode_free_orig = journal->j_cnode_free_list;
2863 journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2864 journal->j_cnode_used = 0;
2865 journal->j_must_wait = 0;
2866
2867 if (journal->j_cnode_free == 0) {
2868 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2869 "allocation failed (%ld bytes). Journal is "
2870 "too large for available memory. Usually "
2871 "this is due to a journal that is too large.",
2872 sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2873 goto free_and_return;
2874 }
2875
2876 init_journal_hash(sb);
2877 jl = journal->j_current_jl;
2878
2879 /*
2880 * get_list_bitmap() may call flush_commit_list() which
2881 * requires the lock. Calling flush_commit_list() shouldn't happen
2882 * this early but I like to be paranoid.
2883 */
2884 reiserfs_write_lock(sb);
2885 jl->j_list_bitmap = get_list_bitmap(sb, jl);
2886 reiserfs_write_unlock(sb);
2887 if (!jl->j_list_bitmap) {
2888 reiserfs_warning(sb, "journal-2005",
2889 "get_list_bitmap failed for journal list 0");
2890 goto free_and_return;
2891 }
2892
2893 ret = journal_read(sb);
2894 if (ret < 0) {
2895 reiserfs_warning(sb, "reiserfs-2006",
2896 "Replay Failure, unable to mount");
2897 goto free_and_return;
2898 }
2899
2900 INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2901 journal->j_work_sb = sb;
2902 return 0;
2903 free_and_return:
2904 free_journal_ram(sb);
2905 return 1;
2906 }
2907
2908 /*
2909 * test for a polite end of the current transaction. Used by file_write,
2910 * and should be used by delete to make sure they don't write more than
2911 * can fit inside a single transaction
2912 */
2913 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2914 int new_alloc)
2915 {
2916 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2917 time64_t now = ktime_get_seconds();
2918 /* cannot restart while nested */
2919 BUG_ON(!th->t_trans_id);
2920 if (th->t_refcount > 1)
2921 return 0;
2922 if (journal->j_must_wait > 0 ||
2923 (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2924 atomic_read(&journal->j_jlock) ||
2925 (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2926 journal->j_cnode_free < (journal->j_trans_max * 3)) {
2927 return 1;
2928 }
2929
2930 journal->j_len_alloc += new_alloc;
2931 th->t_blocks_allocated += new_alloc ;
2932 return 0;
2933 }
2934
2935 /* this must be called inside a transaction */
2936 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2937 {
2938 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2939 BUG_ON(!th->t_trans_id);
2940 journal->j_must_wait = 1;
2941 set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2942 return;
2943 }
2944
2945 /* this must be called without a transaction started */
2946 void reiserfs_allow_writes(struct super_block *s)
2947 {
2948 struct reiserfs_journal *journal = SB_JOURNAL(s);
2949 clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2950 wake_up(&journal->j_join_wait);
2951 }
2952
2953 /* this must be called without a transaction started */
2954 void reiserfs_wait_on_write_block(struct super_block *s)
2955 {
2956 struct reiserfs_journal *journal = SB_JOURNAL(s);
2957 wait_event(journal->j_join_wait,
2958 !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2959 }
2960
2961 static void queue_log_writer(struct super_block *s)
2962 {
2963 wait_queue_entry_t wait;
2964 struct reiserfs_journal *journal = SB_JOURNAL(s);
2965 set_bit(J_WRITERS_QUEUED, &journal->j_state);
2966
2967 /*
2968 * we don't want to use wait_event here because
2969 * we only want to wait once.
2970 */
2971 init_waitqueue_entry(&wait, current);
2972 add_wait_queue(&journal->j_join_wait, &wait);
2973 set_current_state(TASK_UNINTERRUPTIBLE);
2974 if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
2975 int depth = reiserfs_write_unlock_nested(s);
2976 schedule();
2977 reiserfs_write_lock_nested(s, depth);
2978 }
2979 __set_current_state(TASK_RUNNING);
2980 remove_wait_queue(&journal->j_join_wait, &wait);
2981 }
2982
2983 static void wake_queued_writers(struct super_block *s)
2984 {
2985 struct reiserfs_journal *journal = SB_JOURNAL(s);
2986 if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2987 wake_up(&journal->j_join_wait);
2988 }
2989
2990 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
2991 {
2992 struct reiserfs_journal *journal = SB_JOURNAL(sb);
2993 unsigned long bcount = journal->j_bcount;
2994 while (1) {
2995 int depth;
2996
2997 depth = reiserfs_write_unlock_nested(sb);
2998 schedule_timeout_uninterruptible(1);
2999 reiserfs_write_lock_nested(sb, depth);
3000
3001 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
3002 while ((atomic_read(&journal->j_wcount) > 0 ||
3003 atomic_read(&journal->j_jlock)) &&
3004 journal->j_trans_id == trans_id) {
3005 queue_log_writer(sb);
3006 }
3007 if (journal->j_trans_id != trans_id)
3008 break;
3009 if (bcount == journal->j_bcount)
3010 break;
3011 bcount = journal->j_bcount;
3012 }
3013 }
3014
3015 /*
3016 * join == true if you must join an existing transaction.
3017 * join == false if you can deal with waiting for others to finish
3018 *
3019 * this will block until the transaction is joinable. send the number of
3020 * blocks you expect to use in nblocks.
3021 */
3022 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3023 struct super_block *sb, unsigned long nblocks,
3024 int join)
3025 {
3026 time64_t now = ktime_get_seconds();
3027 unsigned int old_trans_id;
3028 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3029 struct reiserfs_transaction_handle myth;
3030 int sched_count = 0;
3031 int retval;
3032 int depth;
3033
3034 reiserfs_check_lock_depth(sb, "journal_begin");
3035 BUG_ON(nblocks > journal->j_trans_max);
3036
3037 PROC_INFO_INC(sb, journal.journal_being);
3038 /* set here for journal_join */
3039 th->t_refcount = 1;
3040 th->t_super = sb;
3041
3042 relock:
3043 lock_journal(sb);
3044 if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3045 unlock_journal(sb);
3046 retval = journal->j_errno;
3047 goto out_fail;
3048 }
3049 journal->j_bcount++;
3050
3051 if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3052 unlock_journal(sb);
3053 depth = reiserfs_write_unlock_nested(sb);
3054 reiserfs_wait_on_write_block(sb);
3055 reiserfs_write_lock_nested(sb, depth);
3056 PROC_INFO_INC(sb, journal.journal_relock_writers);
3057 goto relock;
3058 }
3059 now = ktime_get_seconds();
3060
3061 /*
3062 * if there is no room in the journal OR
3063 * if this transaction is too old, and we weren't called joinable,
3064 * wait for it to finish before beginning we don't sleep if there
3065 * aren't other writers
3066 */
3067
3068 if ((!join && journal->j_must_wait > 0) ||
3069 (!join
3070 && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3071 || (!join && atomic_read(&journal->j_wcount) > 0
3072 && journal->j_trans_start_time > 0
3073 && (now - journal->j_trans_start_time) >
3074 journal->j_max_trans_age) || (!join
3075 && atomic_read(&journal->j_jlock))
3076 || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3077
3078 old_trans_id = journal->j_trans_id;
3079 /* allow others to finish this transaction */
3080 unlock_journal(sb);
3081
3082 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3083 journal->j_max_batch &&
3084 ((journal->j_len + nblocks + 2) * 100) <
3085 (journal->j_len_alloc * 75)) {
3086 if (atomic_read(&journal->j_wcount) > 10) {
3087 sched_count++;
3088 queue_log_writer(sb);
3089 goto relock;
3090 }
3091 }
3092 /*
3093 * don't mess with joining the transaction if all we
3094 * have to do is wait for someone else to do a commit
3095 */
3096 if (atomic_read(&journal->j_jlock)) {
3097 while (journal->j_trans_id == old_trans_id &&
3098 atomic_read(&journal->j_jlock)) {
3099 queue_log_writer(sb);
3100 }
3101 goto relock;
3102 }
3103 retval = journal_join(&myth, sb);
3104 if (retval)
3105 goto out_fail;
3106
3107 /* someone might have ended the transaction while we joined */
3108 if (old_trans_id != journal->j_trans_id) {
3109 retval = do_journal_end(&myth, 0);
3110 } else {
3111 retval = do_journal_end(&myth, COMMIT_NOW);
3112 }
3113
3114 if (retval)
3115 goto out_fail;
3116
3117 PROC_INFO_INC(sb, journal.journal_relock_wcount);
3118 goto relock;
3119 }
3120 /* we are the first writer, set trans_id */
3121 if (journal->j_trans_start_time == 0) {
3122 journal->j_trans_start_time = ktime_get_seconds();
3123 }
3124 atomic_inc(&journal->j_wcount);
3125 journal->j_len_alloc += nblocks;
3126 th->t_blocks_logged = 0;
3127 th->t_blocks_allocated = nblocks;
3128 th->t_trans_id = journal->j_trans_id;
3129 unlock_journal(sb);
3130 INIT_LIST_HEAD(&th->t_list);
3131 return 0;
3132
3133 out_fail:
3134 memset(th, 0, sizeof(*th));
3135 /*
3136 * Re-set th->t_super, so we can properly keep track of how many
3137 * persistent transactions there are. We need to do this so if this
3138 * call is part of a failed restart_transaction, we can free it later
3139 */
3140 th->t_super = sb;
3141 return retval;
3142 }
3143
3144 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3145 super_block
3146 *s,
3147 int nblocks)
3148 {
3149 int ret;
3150 struct reiserfs_transaction_handle *th;
3151
3152 /*
3153 * if we're nesting into an existing transaction. It will be
3154 * persistent on its own
3155 */
3156 if (reiserfs_transaction_running(s)) {
3157 th = current->journal_info;
3158 th->t_refcount++;
3159 BUG_ON(th->t_refcount < 2);
3160
3161 return th;
3162 }
3163 th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3164 if (!th)
3165 return NULL;
3166 ret = journal_begin(th, s, nblocks);
3167 if (ret) {
3168 kfree(th);
3169 return NULL;
3170 }
3171
3172 SB_JOURNAL(s)->j_persistent_trans++;
3173 return th;
3174 }
3175
3176 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3177 {
3178 struct super_block *s = th->t_super;
3179 int ret = 0;
3180 if (th->t_trans_id)
3181 ret = journal_end(th);
3182 else
3183 ret = -EIO;
3184 if (th->t_refcount == 0) {
3185 SB_JOURNAL(s)->j_persistent_trans--;
3186 kfree(th);
3187 }
3188 return ret;
3189 }
3190
3191 static int journal_join(struct reiserfs_transaction_handle *th,
3192 struct super_block *sb)
3193 {
3194 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3195
3196 /*
3197 * this keeps do_journal_end from NULLing out the
3198 * current->journal_info pointer
3199 */
3200 th->t_handle_save = cur_th;
3201 BUG_ON(cur_th && cur_th->t_refcount > 1);
3202 return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN);
3203 }
3204
3205 int journal_join_abort(struct reiserfs_transaction_handle *th,
3206 struct super_block *sb)
3207 {
3208 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3209
3210 /*
3211 * this keeps do_journal_end from NULLing out the
3212 * current->journal_info pointer
3213 */
3214 th->t_handle_save = cur_th;
3215 BUG_ON(cur_th && cur_th->t_refcount > 1);
3216 return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT);
3217 }
3218
3219 int journal_begin(struct reiserfs_transaction_handle *th,
3220 struct super_block *sb, unsigned long nblocks)
3221 {
3222 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3223 int ret;
3224
3225 th->t_handle_save = NULL;
3226 if (cur_th) {
3227 /* we are nesting into the current transaction */
3228 if (cur_th->t_super == sb) {
3229 BUG_ON(!cur_th->t_refcount);
3230 cur_th->t_refcount++;
3231 memcpy(th, cur_th, sizeof(*th));
3232 if (th->t_refcount <= 1)
3233 reiserfs_warning(sb, "reiserfs-2005",
3234 "BAD: refcount <= 1, but "
3235 "journal_info != 0");
3236 return 0;
3237 } else {
3238 /*
3239 * we've ended up with a handle from a different
3240 * filesystem. save it and restore on journal_end.
3241 * This should never really happen...
3242 */
3243 reiserfs_warning(sb, "clm-2100",
3244 "nesting info a different FS");
3245 th->t_handle_save = current->journal_info;
3246 current->journal_info = th;
3247 }
3248 } else {
3249 current->journal_info = th;
3250 }
3251 ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3252 BUG_ON(current->journal_info != th);
3253
3254 /*
3255 * I guess this boils down to being the reciprocal of clm-2100 above.
3256 * If do_journal_begin_r fails, we need to put it back, since
3257 * journal_end won't be called to do it. */
3258 if (ret)
3259 current->journal_info = th->t_handle_save;
3260 else
3261 BUG_ON(!th->t_refcount);
3262
3263 return ret;
3264 }
3265
3266 /*
3267 * puts bh into the current transaction. If it was already there, reorders
3268 * removes the old pointers from the hash, and puts new ones in (to make
3269 * sure replay happen in the right order).
3270 *
3271 * if it was dirty, cleans and files onto the clean list. I can't let it
3272 * be dirty again until the transaction is committed.
3273 *
3274 * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3275 */
3276 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3277 struct buffer_head *bh)
3278 {
3279 struct super_block *sb = th->t_super;
3280 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3281 struct reiserfs_journal_cnode *cn = NULL;
3282 int count_already_incd = 0;
3283 int prepared = 0;
3284 BUG_ON(!th->t_trans_id);
3285
3286 PROC_INFO_INC(sb, journal.mark_dirty);
3287 if (th->t_trans_id != journal->j_trans_id) {
3288 reiserfs_panic(th->t_super, "journal-1577",
3289 "handle trans id %ld != current trans id %ld",
3290 th->t_trans_id, journal->j_trans_id);
3291 }
3292
3293 prepared = test_clear_buffer_journal_prepared(bh);
3294 clear_buffer_journal_restore_dirty(bh);
3295 /* already in this transaction, we are done */
3296 if (buffer_journaled(bh)) {
3297 PROC_INFO_INC(sb, journal.mark_dirty_already);
3298 return 0;
3299 }
3300
3301 /*
3302 * this must be turned into a panic instead of a warning. We can't
3303 * allow a dirty or journal_dirty or locked buffer to be logged, as
3304 * some changes could get to disk too early. NOT GOOD.
3305 */
3306 if (!prepared || buffer_dirty(bh)) {
3307 reiserfs_warning(sb, "journal-1777",
3308 "buffer %llu bad state "
3309 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3310 (unsigned long long)bh->b_blocknr,
3311 prepared ? ' ' : '!',
3312 buffer_locked(bh) ? ' ' : '!',
3313 buffer_dirty(bh) ? ' ' : '!',
3314 buffer_journal_dirty(bh) ? ' ' : '!');
3315 }
3316
3317 if (atomic_read(&journal->j_wcount) <= 0) {
3318 reiserfs_warning(sb, "journal-1409",
3319 "returning because j_wcount was %d",
3320 atomic_read(&journal->j_wcount));
3321 return 1;
3322 }
3323 /*
3324 * this error means I've screwed up, and we've overflowed
3325 * the transaction. Nothing can be done here, except make the
3326 * FS readonly or panic.
3327 */
3328 if (journal->j_len >= journal->j_trans_max) {
3329 reiserfs_panic(th->t_super, "journal-1413",
3330 "j_len (%lu) is too big",
3331 journal->j_len);
3332 }
3333
3334 if (buffer_journal_dirty(bh)) {
3335 count_already_incd = 1;
3336 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3337 clear_buffer_journal_dirty(bh);
3338 }
3339
3340 if (journal->j_len > journal->j_len_alloc) {
3341 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3342 }
3343
3344 set_buffer_journaled(bh);
3345
3346 /* now put this guy on the end */
3347 if (!cn) {
3348 cn = get_cnode(sb);
3349 if (!cn) {
3350 reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3351 }
3352
3353 if (th->t_blocks_logged == th->t_blocks_allocated) {
3354 th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3355 journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3356 }
3357 th->t_blocks_logged++;
3358 journal->j_len++;
3359
3360 cn->bh = bh;
3361 cn->blocknr = bh->b_blocknr;
3362 cn->sb = sb;
3363 cn->jlist = NULL;
3364 insert_journal_hash(journal->j_hash_table, cn);
3365 if (!count_already_incd) {
3366 get_bh(bh);
3367 }
3368 }
3369 cn->next = NULL;
3370 cn->prev = journal->j_last;
3371 cn->bh = bh;
3372 if (journal->j_last) {
3373 journal->j_last->next = cn;
3374 journal->j_last = cn;
3375 } else {
3376 journal->j_first = cn;
3377 journal->j_last = cn;
3378 }
3379 reiserfs_schedule_old_flush(sb);
3380 return 0;
3381 }
3382
3383 int journal_end(struct reiserfs_transaction_handle *th)
3384 {
3385 struct super_block *sb = th->t_super;
3386 if (!current->journal_info && th->t_refcount > 1)
3387 reiserfs_warning(sb, "REISER-NESTING",
3388 "th NULL, refcount %d", th->t_refcount);
3389
3390 if (!th->t_trans_id) {
3391 WARN_ON(1);
3392 return -EIO;
3393 }
3394
3395 th->t_refcount--;
3396 if (th->t_refcount > 0) {
3397 struct reiserfs_transaction_handle *cur_th =
3398 current->journal_info;
3399
3400 /*
3401 * we aren't allowed to close a nested transaction on a
3402 * different filesystem from the one in the task struct
3403 */
3404 BUG_ON(cur_th->t_super != th->t_super);
3405
3406 if (th != cur_th) {
3407 memcpy(current->journal_info, th, sizeof(*th));
3408 th->t_trans_id = 0;
3409 }
3410 return 0;
3411 } else {
3412 return do_journal_end(th, 0);
3413 }
3414 }
3415
3416 /*
3417 * removes from the current transaction, relsing and descrementing any counters.
3418 * also files the removed buffer directly onto the clean list
3419 *
3420 * called by journal_mark_freed when a block has been deleted
3421 *
3422 * returns 1 if it cleaned and relsed the buffer. 0 otherwise
3423 */
3424 static int remove_from_transaction(struct super_block *sb,
3425 b_blocknr_t blocknr, int already_cleaned)
3426 {
3427 struct buffer_head *bh;
3428 struct reiserfs_journal_cnode *cn;
3429 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3430 int ret = 0;
3431
3432 cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3433 if (!cn || !cn->bh) {
3434 return ret;
3435 }
3436 bh = cn->bh;
3437 if (cn->prev) {
3438 cn->prev->next = cn->next;
3439 }
3440 if (cn->next) {
3441 cn->next->prev = cn->prev;
3442 }
3443 if (cn == journal->j_first) {
3444 journal->j_first = cn->next;
3445 }
3446 if (cn == journal->j_last) {
3447 journal->j_last = cn->prev;
3448 }
3449 if (bh)
3450 remove_journal_hash(sb, journal->j_hash_table, NULL,
3451 bh->b_blocknr, 0);
3452 clear_buffer_journaled(bh); /* don't log this one */
3453
3454 if (!already_cleaned) {
3455 clear_buffer_journal_dirty(bh);
3456 clear_buffer_dirty(bh);
3457 clear_buffer_journal_test(bh);
3458 put_bh(bh);
3459 if (atomic_read(&bh->b_count) < 0) {
3460 reiserfs_warning(sb, "journal-1752",
3461 "b_count < 0");
3462 }
3463 ret = 1;
3464 }
3465 journal->j_len--;
3466 journal->j_len_alloc--;
3467 free_cnode(sb, cn);
3468 return ret;
3469 }
3470
3471 /*
3472 * for any cnode in a journal list, it can only be dirtied of all the
3473 * transactions that include it are committed to disk.
3474 * this checks through each transaction, and returns 1 if you are allowed
3475 * to dirty, and 0 if you aren't
3476 *
3477 * it is called by dirty_journal_list, which is called after
3478 * flush_commit_list has gotten all the log blocks for a given
3479 * transaction on disk
3480 *
3481 */
3482 static int can_dirty(struct reiserfs_journal_cnode *cn)
3483 {
3484 struct super_block *sb = cn->sb;
3485 b_blocknr_t blocknr = cn->blocknr;
3486 struct reiserfs_journal_cnode *cur = cn->hprev;
3487 int can_dirty = 1;
3488
3489 /*
3490 * first test hprev. These are all newer than cn, so any node here
3491 * with the same block number and dev means this node can't be sent
3492 * to disk right now.
3493 */
3494 while (cur && can_dirty) {
3495 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3496 cur->blocknr == blocknr) {
3497 can_dirty = 0;
3498 }
3499 cur = cur->hprev;
3500 }
3501 /*
3502 * then test hnext. These are all older than cn. As long as they
3503 * are committed to the log, it is safe to write cn to disk
3504 */
3505 cur = cn->hnext;
3506 while (cur && can_dirty) {
3507 if (cur->jlist && cur->jlist->j_len > 0 &&
3508 atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh &&
3509 cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3510 can_dirty = 0;
3511 }
3512 cur = cur->hnext;
3513 }
3514 return can_dirty;
3515 }
3516
3517 /*
3518 * syncs the commit blocks, but does not force the real buffers to disk
3519 * will wait until the current transaction is done/committed before returning
3520 */
3521 int journal_end_sync(struct reiserfs_transaction_handle *th)
3522 {
3523 struct super_block *sb = th->t_super;
3524 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3525
3526 BUG_ON(!th->t_trans_id);
3527 /* you can sync while nested, very, very bad */
3528 BUG_ON(th->t_refcount > 1);
3529 if (journal->j_len == 0) {
3530 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3531 1);
3532 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
3533 }
3534 return do_journal_end(th, COMMIT_NOW | WAIT);
3535 }
3536
3537 /* writeback the pending async commits to disk */
3538 static void flush_async_commits(struct work_struct *work)
3539 {
3540 struct reiserfs_journal *journal =
3541 container_of(work, struct reiserfs_journal, j_work.work);
3542 struct super_block *sb = journal->j_work_sb;
3543 struct reiserfs_journal_list *jl;
3544 struct list_head *entry;
3545
3546 reiserfs_write_lock(sb);
3547 if (!list_empty(&journal->j_journal_list)) {
3548 /* last entry is the youngest, commit it and you get everything */
3549 entry = journal->j_journal_list.prev;
3550 jl = JOURNAL_LIST_ENTRY(entry);
3551 flush_commit_list(sb, jl, 1);
3552 }
3553 reiserfs_write_unlock(sb);
3554 }
3555
3556 /*
3557 * flushes any old transactions to disk
3558 * ends the current transaction if it is too old
3559 */
3560 void reiserfs_flush_old_commits(struct super_block *sb)
3561 {
3562 time64_t now;
3563 struct reiserfs_transaction_handle th;
3564 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3565
3566 now = ktime_get_seconds();
3567 /*
3568 * safety check so we don't flush while we are replaying the log during
3569 * mount
3570 */
3571 if (list_empty(&journal->j_journal_list))
3572 return;
3573
3574 /*
3575 * check the current transaction. If there are no writers, and it is
3576 * too old, finish it, and force the commit blocks to disk
3577 */
3578 if (atomic_read(&journal->j_wcount) <= 0 &&
3579 journal->j_trans_start_time > 0 &&
3580 journal->j_len > 0 &&
3581 (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3582 if (!journal_join(&th, sb)) {
3583 reiserfs_prepare_for_journal(sb,
3584 SB_BUFFER_WITH_SB(sb),
3585 1);
3586 journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3587
3588 /*
3589 * we're only being called from kreiserfsd, it makes
3590 * no sense to do an async commit so that kreiserfsd
3591 * can do it later
3592 */
3593 do_journal_end(&th, COMMIT_NOW | WAIT);
3594 }
3595 }
3596 }
3597
3598 /*
3599 * returns 0 if do_journal_end should return right away, returns 1 if
3600 * do_journal_end should finish the commit
3601 *
3602 * if the current transaction is too old, but still has writers, this will
3603 * wait on j_join_wait until all the writers are done. By the time it
3604 * wakes up, the transaction it was called has already ended, so it just
3605 * flushes the commit list and returns 0.
3606 *
3607 * Won't batch when flush or commit_now is set. Also won't batch when
3608 * others are waiting on j_join_wait.
3609 *
3610 * Note, we can't allow the journal_end to proceed while there are still
3611 * writers in the log.
3612 */
3613 static int check_journal_end(struct reiserfs_transaction_handle *th, int flags)
3614 {
3615
3616 time64_t now;
3617 int flush = flags & FLUSH_ALL;
3618 int commit_now = flags & COMMIT_NOW;
3619 int wait_on_commit = flags & WAIT;
3620 struct reiserfs_journal_list *jl;
3621 struct super_block *sb = th->t_super;
3622 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3623
3624 BUG_ON(!th->t_trans_id);
3625
3626 if (th->t_trans_id != journal->j_trans_id) {
3627 reiserfs_panic(th->t_super, "journal-1577",
3628 "handle trans id %ld != current trans id %ld",
3629 th->t_trans_id, journal->j_trans_id);
3630 }
3631
3632 journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3633 /* <= 0 is allowed. unmounting might not call begin */
3634 if (atomic_read(&journal->j_wcount) > 0)
3635 atomic_dec(&journal->j_wcount);
3636
3637 /*
3638 * BUG, deal with case where j_len is 0, but people previously
3639 * freed blocks need to be released will be dealt with by next
3640 * transaction that actually writes something, but should be taken
3641 * care of in this trans
3642 */
3643 BUG_ON(journal->j_len == 0);
3644
3645 /*
3646 * if wcount > 0, and we are called to with flush or commit_now,
3647 * we wait on j_join_wait. We will wake up when the last writer has
3648 * finished the transaction, and started it on its way to the disk.
3649 * Then, we flush the commit or journal list, and just return 0
3650 * because the rest of journal end was already done for this
3651 * transaction.
3652 */
3653 if (atomic_read(&journal->j_wcount) > 0) {
3654 if (flush || commit_now) {
3655 unsigned trans_id;
3656
3657 jl = journal->j_current_jl;
3658 trans_id = jl->j_trans_id;
3659 if (wait_on_commit)
3660 jl->j_state |= LIST_COMMIT_PENDING;
3661 atomic_set(&journal->j_jlock, 1);
3662 if (flush) {
3663 journal->j_next_full_flush = 1;
3664 }
3665 unlock_journal(sb);
3666
3667 /*
3668 * sleep while the current transaction is
3669 * still j_jlocked
3670 */
3671 while (journal->j_trans_id == trans_id) {
3672 if (atomic_read(&journal->j_jlock)) {
3673 queue_log_writer(sb);
3674 } else {
3675 lock_journal(sb);
3676 if (journal->j_trans_id == trans_id) {
3677 atomic_set(&journal->j_jlock,
3678 1);
3679 }
3680 unlock_journal(sb);
3681 }
3682 }
3683 BUG_ON(journal->j_trans_id == trans_id);
3684
3685 if (commit_now
3686 && journal_list_still_alive(sb, trans_id)
3687 && wait_on_commit) {
3688 flush_commit_list(sb, jl, 1);
3689 }
3690 return 0;
3691 }
3692 unlock_journal(sb);
3693 return 0;
3694 }
3695
3696 /* deal with old transactions where we are the last writers */
3697 now = ktime_get_seconds();
3698 if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3699 commit_now = 1;
3700 journal->j_next_async_flush = 1;
3701 }
3702 /* don't batch when someone is waiting on j_join_wait */
3703 /* don't batch when syncing the commit or flushing the whole trans */
3704 if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock))
3705 && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3706 && journal->j_len_alloc < journal->j_max_batch
3707 && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3708 journal->j_bcount++;
3709 unlock_journal(sb);
3710 return 0;
3711 }
3712
3713 if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3714 reiserfs_panic(sb, "journal-003",
3715 "j_start (%ld) is too high",
3716 journal->j_start);
3717 }
3718 return 1;
3719 }
3720
3721 /*
3722 * Does all the work that makes deleting blocks safe.
3723 * when deleting a block mark BH_JNew, just remove it from the current
3724 * transaction, clean it's buffer_head and move on.
3725 *
3726 * otherwise:
3727 * set a bit for the block in the journal bitmap. That will prevent it from
3728 * being allocated for unformatted nodes before this transaction has finished.
3729 *
3730 * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.
3731 * That will prevent any old transactions with this block from trying to flush
3732 * to the real location. Since we aren't removing the cnode from the
3733 * journal_list_hash, *the block can't be reallocated yet.
3734 *
3735 * Then remove it from the current transaction, decrementing any counters and
3736 * filing it on the clean list.
3737 */
3738 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3739 struct super_block *sb, b_blocknr_t blocknr)
3740 {
3741 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3742 struct reiserfs_journal_cnode *cn = NULL;
3743 struct buffer_head *bh = NULL;
3744 struct reiserfs_list_bitmap *jb = NULL;
3745 int cleaned = 0;
3746 BUG_ON(!th->t_trans_id);
3747
3748 cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3749 if (cn && cn->bh) {
3750 bh = cn->bh;
3751 get_bh(bh);
3752 }
3753 /* if it is journal new, we just remove it from this transaction */
3754 if (bh && buffer_journal_new(bh)) {
3755 clear_buffer_journal_new(bh);
3756 clear_prepared_bits(bh);
3757 reiserfs_clean_and_file_buffer(bh);
3758 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3759 } else {
3760 /*
3761 * set the bit for this block in the journal bitmap
3762 * for this transaction
3763 */
3764 jb = journal->j_current_jl->j_list_bitmap;
3765 if (!jb) {
3766 reiserfs_panic(sb, "journal-1702",
3767 "journal_list_bitmap is NULL");
3768 }
3769 set_bit_in_list_bitmap(sb, blocknr, jb);
3770
3771 /* Note, the entire while loop is not allowed to schedule. */
3772
3773 if (bh) {
3774 clear_prepared_bits(bh);
3775 reiserfs_clean_and_file_buffer(bh);
3776 }
3777 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3778
3779 /*
3780 * find all older transactions with this block,
3781 * make sure they don't try to write it out
3782 */
3783 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3784 blocknr);
3785 while (cn) {
3786 if (sb == cn->sb && blocknr == cn->blocknr) {
3787 set_bit(BLOCK_FREED, &cn->state);
3788 if (cn->bh) {
3789 /*
3790 * remove_from_transaction will brelse
3791 * the buffer if it was in the current
3792 * trans
3793 */
3794 if (!cleaned) {
3795 clear_buffer_journal_dirty(cn->
3796 bh);
3797 clear_buffer_dirty(cn->bh);
3798 clear_buffer_journal_test(cn->
3799 bh);
3800 cleaned = 1;
3801 put_bh(cn->bh);
3802 if (atomic_read
3803 (&cn->bh->b_count) < 0) {
3804 reiserfs_warning(sb,
3805 "journal-2138",
3806 "cn->bh->b_count < 0");
3807 }
3808 }
3809 /*
3810 * since we are clearing the bh,
3811 * we MUST dec nonzerolen
3812 */
3813 if (cn->jlist) {
3814 atomic_dec(&cn->jlist->
3815 j_nonzerolen);
3816 }
3817 cn->bh = NULL;
3818 }
3819 }
3820 cn = cn->hnext;
3821 }
3822 }
3823
3824 if (bh)
3825 release_buffer_page(bh); /* get_hash grabs the buffer */
3826 return 0;
3827 }
3828
3829 void reiserfs_update_inode_transaction(struct inode *inode)
3830 {
3831 struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3832 REISERFS_I(inode)->i_jl = journal->j_current_jl;
3833 REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3834 }
3835
3836 /*
3837 * returns -1 on error, 0 if no commits/barriers were done and 1
3838 * if a transaction was actually committed and the barrier was done
3839 */
3840 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3841 struct reiserfs_journal_list *jl)
3842 {
3843 struct reiserfs_transaction_handle th;
3844 struct super_block *sb = inode->i_sb;
3845 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3846 int ret = 0;
3847
3848 /*
3849 * is it from the current transaction,
3850 * or from an unknown transaction?
3851 */
3852 if (id == journal->j_trans_id) {
3853 jl = journal->j_current_jl;
3854 /*
3855 * try to let other writers come in and
3856 * grow this transaction
3857 */
3858 let_transaction_grow(sb, id);
3859 if (journal->j_trans_id != id) {
3860 goto flush_commit_only;
3861 }
3862
3863 ret = journal_begin(&th, sb, 1);
3864 if (ret)
3865 return ret;
3866
3867 /* someone might have ended this transaction while we joined */
3868 if (journal->j_trans_id != id) {
3869 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3870 1);
3871 journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3872 ret = journal_end(&th);
3873 goto flush_commit_only;
3874 }
3875
3876 ret = journal_end_sync(&th);
3877 if (!ret)
3878 ret = 1;
3879
3880 } else {
3881 /*
3882 * this gets tricky, we have to make sure the journal list in
3883 * the inode still exists. We know the list is still around
3884 * if we've got a larger transaction id than the oldest list
3885 */
3886 flush_commit_only:
3887 if (journal_list_still_alive(inode->i_sb, id)) {
3888 /*
3889 * we only set ret to 1 when we know for sure
3890 * the barrier hasn't been started yet on the commit
3891 * block.
3892 */
3893 if (atomic_read(&jl->j_commit_left) > 1)
3894 ret = 1;
3895 flush_commit_list(sb, jl, 1);
3896 if (journal->j_errno)
3897 ret = journal->j_errno;
3898 }
3899 }
3900 /* otherwise the list is gone, and long since committed */
3901 return ret;
3902 }
3903
3904 int reiserfs_commit_for_inode(struct inode *inode)
3905 {
3906 unsigned int id = REISERFS_I(inode)->i_trans_id;
3907 struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3908
3909 /*
3910 * for the whole inode, assume unset id means it was
3911 * changed in the current transaction. More conservative
3912 */
3913 if (!id || !jl) {
3914 reiserfs_update_inode_transaction(inode);
3915 id = REISERFS_I(inode)->i_trans_id;
3916 /* jl will be updated in __commit_trans_jl */
3917 }
3918
3919 return __commit_trans_jl(inode, id, jl);
3920 }
3921
3922 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3923 struct buffer_head *bh)
3924 {
3925 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3926 PROC_INFO_INC(sb, journal.restore_prepared);
3927 if (!bh) {
3928 return;
3929 }
3930 if (test_clear_buffer_journal_restore_dirty(bh) &&
3931 buffer_journal_dirty(bh)) {
3932 struct reiserfs_journal_cnode *cn;
3933 reiserfs_write_lock(sb);
3934 cn = get_journal_hash_dev(sb,
3935 journal->j_list_hash_table,
3936 bh->b_blocknr);
3937 if (cn && can_dirty(cn)) {
3938 set_buffer_journal_test(bh);
3939 mark_buffer_dirty(bh);
3940 }
3941 reiserfs_write_unlock(sb);
3942 }
3943 clear_buffer_journal_prepared(bh);
3944 }
3945
3946 extern struct tree_balance *cur_tb;
3947 /*
3948 * before we can change a metadata block, we have to make sure it won't
3949 * be written to disk while we are altering it. So, we must:
3950 * clean it
3951 * wait on it.
3952 */
3953 int reiserfs_prepare_for_journal(struct super_block *sb,
3954 struct buffer_head *bh, int wait)
3955 {
3956 PROC_INFO_INC(sb, journal.prepare);
3957
3958 if (!trylock_buffer(bh)) {
3959 if (!wait)
3960 return 0;
3961 lock_buffer(bh);
3962 }
3963 set_buffer_journal_prepared(bh);
3964 if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3965 clear_buffer_journal_test(bh);
3966 set_buffer_journal_restore_dirty(bh);
3967 }
3968 unlock_buffer(bh);
3969 return 1;
3970 }
3971
3972 /*
3973 * long and ugly. If flush, will not return until all commit
3974 * blocks and all real buffers in the trans are on disk.
3975 * If no_async, won't return until all commit blocks are on disk.
3976 *
3977 * keep reading, there are comments as you go along
3978 *
3979 * If the journal is aborted, we just clean up. Things like flushing
3980 * journal lists, etc just won't happen.
3981 */
3982 static int do_journal_end(struct reiserfs_transaction_handle *th, int flags)
3983 {
3984 struct super_block *sb = th->t_super;
3985 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3986 struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3987 struct reiserfs_journal_cnode *last_cn = NULL;
3988 struct reiserfs_journal_desc *desc;
3989 struct reiserfs_journal_commit *commit;
3990 struct buffer_head *c_bh; /* commit bh */
3991 struct buffer_head *d_bh; /* desc bh */
3992 int cur_write_start = 0; /* start index of current log write */
3993 int old_start;
3994 int i;
3995 int flush;
3996 int wait_on_commit;
3997 struct reiserfs_journal_list *jl, *temp_jl;
3998 struct list_head *entry, *safe;
3999 unsigned long jindex;
4000 unsigned int commit_trans_id;
4001 int trans_half;
4002 int depth;
4003
4004 BUG_ON(th->t_refcount > 1);
4005 BUG_ON(!th->t_trans_id);
4006 BUG_ON(!th->t_super);
4007
4008 /*
4009 * protect flush_older_commits from doing mistakes if the
4010 * transaction ID counter gets overflowed.
4011 */
4012 if (th->t_trans_id == ~0U)
4013 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
4014 flush = flags & FLUSH_ALL;
4015 wait_on_commit = flags & WAIT;
4016
4017 current->journal_info = th->t_handle_save;
4018 reiserfs_check_lock_depth(sb, "journal end");
4019 if (journal->j_len == 0) {
4020 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
4021 1);
4022 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
4023 }
4024
4025 lock_journal(sb);
4026 if (journal->j_next_full_flush) {
4027 flags |= FLUSH_ALL;
4028 flush = 1;
4029 }
4030 if (journal->j_next_async_flush) {
4031 flags |= COMMIT_NOW | WAIT;
4032 wait_on_commit = 1;
4033 }
4034
4035 /*
4036 * check_journal_end locks the journal, and unlocks if it does
4037 * not return 1 it tells us if we should continue with the
4038 * journal_end, or just return
4039 */
4040 if (!check_journal_end(th, flags)) {
4041 reiserfs_schedule_old_flush(sb);
4042 wake_queued_writers(sb);
4043 reiserfs_async_progress_wait(sb);
4044 goto out;
4045 }
4046
4047 /* check_journal_end might set these, check again */
4048 if (journal->j_next_full_flush) {
4049 flush = 1;
4050 }
4051
4052 /*
4053 * j must wait means we have to flush the log blocks, and the
4054 * real blocks for this transaction
4055 */
4056 if (journal->j_must_wait > 0) {
4057 flush = 1;
4058 }
4059 #ifdef REISERFS_PREALLOCATE
4060 /*
4061 * quota ops might need to nest, setup the journal_info pointer
4062 * for them and raise the refcount so that it is > 0.
4063 */
4064 current->journal_info = th;
4065 th->t_refcount++;
4066
4067 /* it should not involve new blocks into the transaction */
4068 reiserfs_discard_all_prealloc(th);
4069
4070 th->t_refcount--;
4071 current->journal_info = th->t_handle_save;
4072 #endif
4073
4074 /* setup description block */
4075 d_bh =
4076 journal_getblk(sb,
4077 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4078 journal->j_start);
4079 set_buffer_uptodate(d_bh);
4080 desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4081 memset(d_bh->b_data, 0, d_bh->b_size);
4082 memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4083 set_desc_trans_id(desc, journal->j_trans_id);
4084
4085 /*
4086 * setup commit block. Don't write (keep it clean too) this one
4087 * until after everyone else is written
4088 */
4089 c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4090 ((journal->j_start + journal->j_len +
4091 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4092 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4093 memset(c_bh->b_data, 0, c_bh->b_size);
4094 set_commit_trans_id(commit, journal->j_trans_id);
4095 set_buffer_uptodate(c_bh);
4096
4097 /* init this journal list */
4098 jl = journal->j_current_jl;
4099
4100 /*
4101 * we lock the commit before doing anything because
4102 * we want to make sure nobody tries to run flush_commit_list until
4103 * the new transaction is fully setup, and we've already flushed the
4104 * ordered bh list
4105 */
4106 reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4107
4108 /* save the transaction id in case we need to commit it later */
4109 commit_trans_id = jl->j_trans_id;
4110
4111 atomic_set(&jl->j_older_commits_done, 0);
4112 jl->j_trans_id = journal->j_trans_id;
4113 jl->j_timestamp = journal->j_trans_start_time;
4114 jl->j_commit_bh = c_bh;
4115 jl->j_start = journal->j_start;
4116 jl->j_len = journal->j_len;
4117 atomic_set(&jl->j_nonzerolen, journal->j_len);
4118 atomic_set(&jl->j_commit_left, journal->j_len + 2);
4119 jl->j_realblock = NULL;
4120
4121 /*
4122 * The ENTIRE FOR LOOP MUST not cause schedule to occur.
4123 * for each real block, add it to the journal list hash,
4124 * copy into real block index array in the commit or desc block
4125 */
4126 trans_half = journal_trans_half(sb->s_blocksize);
4127 for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4128 if (buffer_journaled(cn->bh)) {
4129 jl_cn = get_cnode(sb);
4130 if (!jl_cn) {
4131 reiserfs_panic(sb, "journal-1676",
4132 "get_cnode returned NULL");
4133 }
4134 if (i == 0) {
4135 jl->j_realblock = jl_cn;
4136 }
4137 jl_cn->prev = last_cn;
4138 jl_cn->next = NULL;
4139 if (last_cn) {
4140 last_cn->next = jl_cn;
4141 }
4142 last_cn = jl_cn;
4143 /*
4144 * make sure the block we are trying to log
4145 * is not a block of journal or reserved area
4146 */
4147 if (is_block_in_log_or_reserved_area
4148 (sb, cn->bh->b_blocknr)) {
4149 reiserfs_panic(sb, "journal-2332",
4150 "Trying to log block %lu, "
4151 "which is a log block",
4152 cn->bh->b_blocknr);
4153 }
4154 jl_cn->blocknr = cn->bh->b_blocknr;
4155 jl_cn->state = 0;
4156 jl_cn->sb = sb;
4157 jl_cn->bh = cn->bh;
4158 jl_cn->jlist = jl;
4159 insert_journal_hash(journal->j_list_hash_table, jl_cn);
4160 if (i < trans_half) {
4161 desc->j_realblock[i] =
4162 cpu_to_le32(cn->bh->b_blocknr);
4163 } else {
4164 commit->j_realblock[i - trans_half] =
4165 cpu_to_le32(cn->bh->b_blocknr);
4166 }
4167 } else {
4168 i--;
4169 }
4170 }
4171 set_desc_trans_len(desc, journal->j_len);
4172 set_desc_mount_id(desc, journal->j_mount_id);
4173 set_desc_trans_id(desc, journal->j_trans_id);
4174 set_commit_trans_len(commit, journal->j_len);
4175
4176 /*
4177 * special check in case all buffers in the journal
4178 * were marked for not logging
4179 */
4180 BUG_ON(journal->j_len == 0);
4181
4182 /*
4183 * we're about to dirty all the log blocks, mark the description block
4184 * dirty now too. Don't mark the commit block dirty until all the
4185 * others are on disk
4186 */
4187 mark_buffer_dirty(d_bh);
4188
4189 /*
4190 * first data block is j_start + 1, so add one to
4191 * cur_write_start wherever you use it
4192 */
4193 cur_write_start = journal->j_start;
4194 cn = journal->j_first;
4195 jindex = 1; /* start at one so we don't get the desc again */
4196 while (cn) {
4197 clear_buffer_journal_new(cn->bh);
4198 /* copy all the real blocks into log area. dirty log blocks */
4199 if (buffer_journaled(cn->bh)) {
4200 struct buffer_head *tmp_bh;
4201 char *addr;
4202 struct page *page;
4203 tmp_bh =
4204 journal_getblk(sb,
4205 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4206 ((cur_write_start +
4207 jindex) %
4208 SB_ONDISK_JOURNAL_SIZE(sb)));
4209 set_buffer_uptodate(tmp_bh);
4210 page = cn->bh->b_page;
4211 addr = kmap(page);
4212 memcpy(tmp_bh->b_data,
4213 addr + offset_in_page(cn->bh->b_data),
4214 cn->bh->b_size);
4215 kunmap(page);
4216 mark_buffer_dirty(tmp_bh);
4217 jindex++;
4218 set_buffer_journal_dirty(cn->bh);
4219 clear_buffer_journaled(cn->bh);
4220 } else {
4221 /*
4222 * JDirty cleared sometime during transaction.
4223 * don't log this one
4224 */
4225 reiserfs_warning(sb, "journal-2048",
4226 "BAD, buffer in journal hash, "
4227 "but not JDirty!");
4228 brelse(cn->bh);
4229 }
4230 next = cn->next;
4231 free_cnode(sb, cn);
4232 cn = next;
4233 reiserfs_cond_resched(sb);
4234 }
4235
4236 /*
4237 * we are done with both the c_bh and d_bh, but
4238 * c_bh must be written after all other commit blocks,
4239 * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4240 */
4241
4242 journal->j_current_jl = alloc_journal_list(sb);
4243
4244 /* now it is safe to insert this transaction on the main list */
4245 list_add_tail(&jl->j_list, &journal->j_journal_list);
4246 list_add_tail(&jl->j_working_list, &journal->j_working_list);
4247 journal->j_num_work_lists++;
4248
4249 /* reset journal values for the next transaction */
4250 old_start = journal->j_start;
4251 journal->j_start =
4252 (journal->j_start + journal->j_len +
4253 2) % SB_ONDISK_JOURNAL_SIZE(sb);
4254 atomic_set(&journal->j_wcount, 0);
4255 journal->j_bcount = 0;
4256 journal->j_last = NULL;
4257 journal->j_first = NULL;
4258 journal->j_len = 0;
4259 journal->j_trans_start_time = 0;
4260 /* check for trans_id overflow */
4261 if (++journal->j_trans_id == 0)
4262 journal->j_trans_id = 10;
4263 journal->j_current_jl->j_trans_id = journal->j_trans_id;
4264 journal->j_must_wait = 0;
4265 journal->j_len_alloc = 0;
4266 journal->j_next_full_flush = 0;
4267 journal->j_next_async_flush = 0;
4268 init_journal_hash(sb);
4269
4270 /*
4271 * make sure reiserfs_add_jh sees the new current_jl before we
4272 * write out the tails
4273 */
4274 smp_mb();
4275
4276 /*
4277 * tail conversion targets have to hit the disk before we end the
4278 * transaction. Otherwise a later transaction might repack the tail
4279 * before this transaction commits, leaving the data block unflushed
4280 * and clean, if we crash before the later transaction commits, the
4281 * data block is lost.
4282 */
4283 if (!list_empty(&jl->j_tail_bh_list)) {
4284 depth = reiserfs_write_unlock_nested(sb);
4285 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4286 journal, jl, &jl->j_tail_bh_list);
4287 reiserfs_write_lock_nested(sb, depth);
4288 }
4289 BUG_ON(!list_empty(&jl->j_tail_bh_list));
4290 mutex_unlock(&jl->j_commit_mutex);
4291
4292 /*
4293 * honor the flush wishes from the caller, simple commits can
4294 * be done outside the journal lock, they are done below
4295 *
4296 * if we don't flush the commit list right now, we put it into
4297 * the work queue so the people waiting on the async progress work
4298 * queue don't wait for this proc to flush journal lists and such.
4299 */
4300 if (flush) {
4301 flush_commit_list(sb, jl, 1);
4302 flush_journal_list(sb, jl, 1);
4303 } else if (!(jl->j_state & LIST_COMMIT_PENDING)) {
4304 /*
4305 * Avoid queueing work when sb is being shut down. Transaction
4306 * will be flushed on journal shutdown.
4307 */
4308 if (sb->s_flags & SB_ACTIVE)
4309 queue_delayed_work(REISERFS_SB(sb)->commit_wq,
4310 &journal->j_work, HZ / 10);
4311 }
4312
4313 /*
4314 * if the next transaction has any chance of wrapping, flush
4315 * transactions that might get overwritten. If any journal lists
4316 * are very old flush them as well.
4317 */
4318 first_jl:
4319 list_for_each_safe(entry, safe, &journal->j_journal_list) {
4320 temp_jl = JOURNAL_LIST_ENTRY(entry);
4321 if (journal->j_start <= temp_jl->j_start) {
4322 if ((journal->j_start + journal->j_trans_max + 1) >=
4323 temp_jl->j_start) {
4324 flush_used_journal_lists(sb, temp_jl);
4325 goto first_jl;
4326 } else if ((journal->j_start +
4327 journal->j_trans_max + 1) <
4328 SB_ONDISK_JOURNAL_SIZE(sb)) {
4329 /*
4330 * if we don't cross into the next
4331 * transaction and we don't wrap, there is
4332 * no way we can overlap any later transactions
4333 * break now
4334 */
4335 break;
4336 }
4337 } else if ((journal->j_start +
4338 journal->j_trans_max + 1) >
4339 SB_ONDISK_JOURNAL_SIZE(sb)) {
4340 if (((journal->j_start + journal->j_trans_max + 1) %
4341 SB_ONDISK_JOURNAL_SIZE(sb)) >=
4342 temp_jl->j_start) {
4343 flush_used_journal_lists(sb, temp_jl);
4344 goto first_jl;
4345 } else {
4346 /*
4347 * we don't overlap anything from out start
4348 * to the end of the log, and our wrapped
4349 * portion doesn't overlap anything at
4350 * the start of the log. We can break
4351 */
4352 break;
4353 }
4354 }
4355 }
4356
4357 journal->j_current_jl->j_list_bitmap =
4358 get_list_bitmap(sb, journal->j_current_jl);
4359
4360 if (!(journal->j_current_jl->j_list_bitmap)) {
4361 reiserfs_panic(sb, "journal-1996",
4362 "could not get a list bitmap");
4363 }
4364
4365 atomic_set(&journal->j_jlock, 0);
4366 unlock_journal(sb);
4367 /* wake up any body waiting to join. */
4368 clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4369 wake_up(&journal->j_join_wait);
4370
4371 if (!flush && wait_on_commit &&
4372 journal_list_still_alive(sb, commit_trans_id)) {
4373 flush_commit_list(sb, jl, 1);
4374 }
4375 out:
4376 reiserfs_check_lock_depth(sb, "journal end2");
4377
4378 memset(th, 0, sizeof(*th));
4379 /*
4380 * Re-set th->t_super, so we can properly keep track of how many
4381 * persistent transactions there are. We need to do this so if this
4382 * call is part of a failed restart_transaction, we can free it later
4383 */
4384 th->t_super = sb;
4385
4386 return journal->j_errno;
4387 }
4388
4389 /* Send the file system read only and refuse new transactions */
4390 void reiserfs_abort_journal(struct super_block *sb, int errno)
4391 {
4392 struct reiserfs_journal *journal = SB_JOURNAL(sb);
4393 if (test_bit(J_ABORTED, &journal->j_state))
4394 return;
4395
4396 if (!journal->j_errno)
4397 journal->j_errno = errno;
4398
4399 sb->s_flags |= SB_RDONLY;
4400 set_bit(J_ABORTED, &journal->j_state);
4401
4402 #ifdef CONFIG_REISERFS_CHECK
4403 dump_stack();
4404 #endif
4405 }
Ubuntu Hirsute Linux kernel mirror