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