1 // SPDX-License-Identifier: GPL-2.0
3 * Write ahead logging implementation copyright Chris Mason 2000
5 * The background commits make this code very interrelated, and
6 * overly complex. I need to rethink things a bit....The major players:
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.
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.
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.
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
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
34 * -- Note, if you call this as an immediate flush from
35 * from within kupdate, it will ignore the immediate flag
38 #include <linux/time.h>
39 #include <linux/semaphore.h>
40 #include <linux/vmalloc.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>
56 /* gets a struct reiserfs_journal_list * from a list head */
57 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
59 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
62 /* must be correct to keep the desc and commit structs at 4k */
63 #define JOURNAL_TRANS_HALF 1018
64 #define BUFNR 64 /*read ahead */
66 /* cnode stat bits. Move these into reiserfs_fs.h */
68 /* this block was freed, and can't be written. */
70 /* this block was freed during this transaction, and can't be written */
71 #define BLOCK_FREED_HOLDER 3
73 /* used in flush_journal_list */
74 #define BLOCK_NEEDS_FLUSH 4
75 #define BLOCK_DIRTIED 5
77 /* journal list state bits */
78 #define LIST_TOUCHED 1
80 #define LIST_COMMIT_PENDING 4 /* someone will commit this list */
82 /* flags for do_journal_end */
83 #define FLUSH_ALL 1 /* flush commit and real blocks */
84 #define COMMIT_NOW 2 /* end and commit this transaction */
85 #define WAIT 4 /* wait for the log blocks to hit the disk */
87 static int do_journal_end(struct reiserfs_transaction_handle
*, int flags
);
88 static int flush_journal_list(struct super_block
*s
,
89 struct reiserfs_journal_list
*jl
, int flushall
);
90 static int flush_commit_list(struct super_block
*s
,
91 struct reiserfs_journal_list
*jl
, int flushall
);
92 static int can_dirty(struct reiserfs_journal_cnode
*cn
);
93 static int journal_join(struct reiserfs_transaction_handle
*th
,
94 struct super_block
*sb
);
95 static void release_journal_dev(struct super_block
*super
,
96 struct reiserfs_journal
*journal
);
97 static int dirty_one_transaction(struct super_block
*s
,
98 struct reiserfs_journal_list
*jl
);
99 static void flush_async_commits(struct work_struct
*work
);
100 static void queue_log_writer(struct super_block
*s
);
102 /* values for join in do_journal_begin_r */
104 JBEGIN_REG
= 0, /* regular journal begin */
105 /* join the running transaction if at all possible */
107 /* called from cleanup code, ignores aborted flag */
111 static int do_journal_begin_r(struct reiserfs_transaction_handle
*th
,
112 struct super_block
*sb
,
113 unsigned long nblocks
, int join
);
115 static void init_journal_hash(struct super_block
*sb
)
117 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
118 memset(journal
->j_hash_table
, 0,
119 JOURNAL_HASH_SIZE
* sizeof(struct reiserfs_journal_cnode
*));
123 * clears BH_Dirty and sticks the buffer on the clean list. Called because
124 * I can't allow refile_buffer to make schedule happen after I've freed a
125 * block. Look at remove_from_transaction and journal_mark_freed for
128 static int reiserfs_clean_and_file_buffer(struct buffer_head
*bh
)
131 clear_buffer_dirty(bh
);
132 clear_buffer_journal_test(bh
);
137 static struct reiserfs_bitmap_node
*allocate_bitmap_node(struct super_block
140 struct reiserfs_bitmap_node
*bn
;
143 bn
= kmalloc(sizeof(struct reiserfs_bitmap_node
), GFP_NOFS
);
147 bn
->data
= kzalloc(sb
->s_blocksize
, GFP_NOFS
);
153 INIT_LIST_HEAD(&bn
->list
);
157 static struct reiserfs_bitmap_node
*get_bitmap_node(struct super_block
*sb
)
159 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
160 struct reiserfs_bitmap_node
*bn
= NULL
;
161 struct list_head
*entry
= journal
->j_bitmap_nodes
.next
;
163 journal
->j_used_bitmap_nodes
++;
166 if (entry
!= &journal
->j_bitmap_nodes
) {
167 bn
= list_entry(entry
, struct reiserfs_bitmap_node
, list
);
169 memset(bn
->data
, 0, sb
->s_blocksize
);
170 journal
->j_free_bitmap_nodes
--;
173 bn
= allocate_bitmap_node(sb
);
180 static inline void free_bitmap_node(struct super_block
*sb
,
181 struct reiserfs_bitmap_node
*bn
)
183 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
184 journal
->j_used_bitmap_nodes
--;
185 if (journal
->j_free_bitmap_nodes
> REISERFS_MAX_BITMAP_NODES
) {
189 list_add(&bn
->list
, &journal
->j_bitmap_nodes
);
190 journal
->j_free_bitmap_nodes
++;
194 static void allocate_bitmap_nodes(struct super_block
*sb
)
197 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
198 struct reiserfs_bitmap_node
*bn
= NULL
;
199 for (i
= 0; i
< REISERFS_MIN_BITMAP_NODES
; i
++) {
200 bn
= allocate_bitmap_node(sb
);
202 list_add(&bn
->list
, &journal
->j_bitmap_nodes
);
203 journal
->j_free_bitmap_nodes
++;
205 /* this is ok, we'll try again when more are needed */
211 static int set_bit_in_list_bitmap(struct super_block
*sb
,
213 struct reiserfs_list_bitmap
*jb
)
215 unsigned int bmap_nr
= block
/ (sb
->s_blocksize
<< 3);
216 unsigned int bit_nr
= block
% (sb
->s_blocksize
<< 3);
218 if (!jb
->bitmaps
[bmap_nr
]) {
219 jb
->bitmaps
[bmap_nr
] = get_bitmap_node(sb
);
221 set_bit(bit_nr
, (unsigned long *)jb
->bitmaps
[bmap_nr
]->data
);
225 static void cleanup_bitmap_list(struct super_block
*sb
,
226 struct reiserfs_list_bitmap
*jb
)
229 if (jb
->bitmaps
== NULL
)
232 for (i
= 0; i
< reiserfs_bmap_count(sb
); i
++) {
233 if (jb
->bitmaps
[i
]) {
234 free_bitmap_node(sb
, jb
->bitmaps
[i
]);
235 jb
->bitmaps
[i
] = NULL
;
241 * only call this on FS unmount.
243 static int free_list_bitmaps(struct super_block
*sb
,
244 struct reiserfs_list_bitmap
*jb_array
)
247 struct reiserfs_list_bitmap
*jb
;
248 for (i
= 0; i
< JOURNAL_NUM_BITMAPS
; i
++) {
250 jb
->journal_list
= NULL
;
251 cleanup_bitmap_list(sb
, jb
);
258 static int free_bitmap_nodes(struct super_block
*sb
)
260 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
261 struct list_head
*next
= journal
->j_bitmap_nodes
.next
;
262 struct reiserfs_bitmap_node
*bn
;
264 while (next
!= &journal
->j_bitmap_nodes
) {
265 bn
= list_entry(next
, struct reiserfs_bitmap_node
, list
);
269 next
= journal
->j_bitmap_nodes
.next
;
270 journal
->j_free_bitmap_nodes
--;
277 * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
278 * jb_array is the array to be filled in.
280 int reiserfs_allocate_list_bitmaps(struct super_block
*sb
,
281 struct reiserfs_list_bitmap
*jb_array
,
282 unsigned int bmap_nr
)
286 struct reiserfs_list_bitmap
*jb
;
287 int mem
= bmap_nr
* sizeof(struct reiserfs_bitmap_node
*);
289 for (i
= 0; i
< JOURNAL_NUM_BITMAPS
; i
++) {
291 jb
->journal_list
= NULL
;
292 jb
->bitmaps
= vzalloc(mem
);
294 reiserfs_warning(sb
, "clm-2000", "unable to "
295 "allocate bitmaps for journal lists");
301 free_list_bitmaps(sb
, jb_array
);
308 * find an available list bitmap. If you can't find one, flush a commit list
311 static struct reiserfs_list_bitmap
*get_list_bitmap(struct super_block
*sb
,
312 struct reiserfs_journal_list
316 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
317 struct reiserfs_list_bitmap
*jb
= NULL
;
319 for (j
= 0; j
< (JOURNAL_NUM_BITMAPS
* 3); j
++) {
320 i
= journal
->j_list_bitmap_index
;
321 journal
->j_list_bitmap_index
= (i
+ 1) % JOURNAL_NUM_BITMAPS
;
322 jb
= journal
->j_list_bitmap
+ i
;
323 if (journal
->j_list_bitmap
[i
].journal_list
) {
324 flush_commit_list(sb
,
325 journal
->j_list_bitmap
[i
].
327 if (!journal
->j_list_bitmap
[i
].journal_list
) {
334 /* double check to make sure if flushed correctly */
335 if (jb
->journal_list
)
337 jb
->journal_list
= jl
;
342 * allocates a new chunk of X nodes, and links them all together as a list.
343 * Uses the cnode->next and cnode->prev pointers
344 * returns NULL on failure
346 static struct reiserfs_journal_cnode
*allocate_cnodes(int num_cnodes
)
348 struct reiserfs_journal_cnode
*head
;
350 if (num_cnodes
<= 0) {
353 head
= vzalloc(num_cnodes
* sizeof(struct reiserfs_journal_cnode
));
358 head
[0].next
= head
+ 1;
359 for (i
= 1; i
< num_cnodes
; i
++) {
360 head
[i
].prev
= head
+ (i
- 1);
361 head
[i
].next
= head
+ (i
+ 1); /* if last one, overwrite it after the if */
363 head
[num_cnodes
- 1].next
= NULL
;
367 /* pulls a cnode off the free list, or returns NULL on failure */
368 static struct reiserfs_journal_cnode
*get_cnode(struct super_block
*sb
)
370 struct reiserfs_journal_cnode
*cn
;
371 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
373 reiserfs_check_lock_depth(sb
, "get_cnode");
375 if (journal
->j_cnode_free
<= 0) {
378 journal
->j_cnode_used
++;
379 journal
->j_cnode_free
--;
380 cn
= journal
->j_cnode_free_list
;
385 cn
->next
->prev
= NULL
;
387 journal
->j_cnode_free_list
= cn
->next
;
388 memset(cn
, 0, sizeof(struct reiserfs_journal_cnode
));
393 * returns a cnode to the free list
395 static void free_cnode(struct super_block
*sb
,
396 struct reiserfs_journal_cnode
*cn
)
398 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
400 reiserfs_check_lock_depth(sb
, "free_cnode");
402 journal
->j_cnode_used
--;
403 journal
->j_cnode_free
++;
404 /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
405 cn
->next
= journal
->j_cnode_free_list
;
406 if (journal
->j_cnode_free_list
) {
407 journal
->j_cnode_free_list
->prev
= cn
;
409 cn
->prev
= NULL
; /* not needed with the memset, but I might kill the memset, and forget to do this */
410 journal
->j_cnode_free_list
= cn
;
413 static void clear_prepared_bits(struct buffer_head
*bh
)
415 clear_buffer_journal_prepared(bh
);
416 clear_buffer_journal_restore_dirty(bh
);
420 * return a cnode with same dev, block number and size in table,
421 * or null if not found
423 static inline struct reiserfs_journal_cnode
*get_journal_hash_dev(struct
427 reiserfs_journal_cnode
431 struct reiserfs_journal_cnode
*cn
;
432 cn
= journal_hash(table
, sb
, bl
);
434 if (cn
->blocknr
== bl
&& cn
->sb
== sb
)
438 return (struct reiserfs_journal_cnode
*)0;
442 * this actually means 'can this block be reallocated yet?'. If you set
443 * search_all, a block can only be allocated if it is not in the current
444 * transaction, was not freed by the current transaction, and has no chance
445 * of ever being overwritten by a replay after crashing.
447 * If you don't set search_all, a block can only be allocated if it is not
448 * in the current transaction. Since deleting a block removes it from the
449 * current transaction, this case should never happen. If you don't set
450 * search_all, make sure you never write the block without logging it.
452 * next_zero_bit is a suggestion about the next block to try for find_forward.
453 * when bl is rejected because it is set in a journal list bitmap, we search
454 * for the next zero bit in the bitmap that rejected bl. Then, we return
455 * that through next_zero_bit for find_forward to try.
457 * Just because we return something in next_zero_bit does not mean we won't
458 * reject it on the next call to reiserfs_in_journal
460 int reiserfs_in_journal(struct super_block
*sb
,
461 unsigned int bmap_nr
, int bit_nr
, int search_all
,
462 b_blocknr_t
* next_zero_bit
)
464 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
465 struct reiserfs_journal_cnode
*cn
;
466 struct reiserfs_list_bitmap
*jb
;
470 *next_zero_bit
= 0; /* always start this at zero. */
472 PROC_INFO_INC(sb
, journal
.in_journal
);
474 * If we aren't doing a search_all, this is a metablock, and it
475 * will be logged before use. if we crash before the transaction
476 * that freed it commits, this transaction won't have committed
477 * either, and the block will never be written
480 for (i
= 0; i
< JOURNAL_NUM_BITMAPS
; i
++) {
481 PROC_INFO_INC(sb
, journal
.in_journal_bitmap
);
482 jb
= journal
->j_list_bitmap
+ i
;
483 if (jb
->journal_list
&& jb
->bitmaps
[bmap_nr
] &&
485 (unsigned long *)jb
->bitmaps
[bmap_nr
]->
488 find_next_zero_bit((unsigned long *)
489 (jb
->bitmaps
[bmap_nr
]->
491 sb
->s_blocksize
<< 3,
498 bl
= bmap_nr
* (sb
->s_blocksize
<< 3) + bit_nr
;
499 /* is it in any old transactions? */
502 get_journal_hash_dev(sb
, journal
->j_list_hash_table
, bl
))) {
506 /* is it in the current transaction. This should never happen */
507 if ((cn
= get_journal_hash_dev(sb
, journal
->j_hash_table
, bl
))) {
512 PROC_INFO_INC(sb
, journal
.in_journal_reusable
);
517 /* insert cn into table */
518 static inline void insert_journal_hash(struct reiserfs_journal_cnode
**table
,
519 struct reiserfs_journal_cnode
*cn
)
521 struct reiserfs_journal_cnode
*cn_orig
;
523 cn_orig
= journal_hash(table
, cn
->sb
, cn
->blocknr
);
529 journal_hash(table
, cn
->sb
, cn
->blocknr
) = cn
;
532 /* lock the current transaction */
533 static inline void lock_journal(struct super_block
*sb
)
535 PROC_INFO_INC(sb
, journal
.lock_journal
);
537 reiserfs_mutex_lock_safe(&SB_JOURNAL(sb
)->j_mutex
, sb
);
540 /* unlock the current transaction */
541 static inline void unlock_journal(struct super_block
*sb
)
543 mutex_unlock(&SB_JOURNAL(sb
)->j_mutex
);
546 static inline void get_journal_list(struct reiserfs_journal_list
*jl
)
551 static inline void put_journal_list(struct super_block
*s
,
552 struct reiserfs_journal_list
*jl
)
554 if (jl
->j_refcount
< 1) {
555 reiserfs_panic(s
, "journal-2", "trans id %u, refcount at %d",
556 jl
->j_trans_id
, jl
->j_refcount
);
558 if (--jl
->j_refcount
== 0)
563 * this used to be much more involved, and I'm keeping it just in case
564 * things get ugly again. it gets called by flush_commit_list, and
565 * cleans up any data stored about blocks freed during a transaction.
567 static void cleanup_freed_for_journal_list(struct super_block
*sb
,
568 struct reiserfs_journal_list
*jl
)
571 struct reiserfs_list_bitmap
*jb
= jl
->j_list_bitmap
;
573 cleanup_bitmap_list(sb
, jb
);
575 jl
->j_list_bitmap
->journal_list
= NULL
;
576 jl
->j_list_bitmap
= NULL
;
579 static int journal_list_still_alive(struct super_block
*s
,
580 unsigned int trans_id
)
582 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
583 struct list_head
*entry
= &journal
->j_journal_list
;
584 struct reiserfs_journal_list
*jl
;
586 if (!list_empty(entry
)) {
587 jl
= JOURNAL_LIST_ENTRY(entry
->next
);
588 if (jl
->j_trans_id
<= trans_id
) {
596 * If page->mapping was null, we failed to truncate this page for
597 * some reason. Most likely because it was truncated after being
598 * logged via data=journal.
600 * This does a check to see if the buffer belongs to one of these
601 * lost pages before doing the final put_bh. If page->mapping was
602 * null, it tries to free buffers on the page, which should make the
603 * final put_page drop the page from the lru.
605 static void release_buffer_page(struct buffer_head
*bh
)
607 struct page
*page
= bh
->b_page
;
608 if (!page
->mapping
&& trylock_page(page
)) {
612 try_to_free_buffers(page
);
620 static void reiserfs_end_buffer_io_sync(struct buffer_head
*bh
, int uptodate
)
622 if (buffer_journaled(bh
)) {
623 reiserfs_warning(NULL
, "clm-2084",
624 "pinned buffer %lu:%pg sent to disk",
625 bh
->b_blocknr
, bh
->b_bdev
);
628 set_buffer_uptodate(bh
);
630 clear_buffer_uptodate(bh
);
633 release_buffer_page(bh
);
636 static void reiserfs_end_ordered_io(struct buffer_head
*bh
, int uptodate
)
639 set_buffer_uptodate(bh
);
641 clear_buffer_uptodate(bh
);
646 static void submit_logged_buffer(struct buffer_head
*bh
)
649 bh
->b_end_io
= reiserfs_end_buffer_io_sync
;
650 clear_buffer_journal_new(bh
);
651 clear_buffer_dirty(bh
);
652 if (!test_clear_buffer_journal_test(bh
))
654 if (!buffer_uptodate(bh
))
656 submit_bh(REQ_OP_WRITE
, 0, bh
);
659 static void submit_ordered_buffer(struct buffer_head
*bh
)
662 bh
->b_end_io
= reiserfs_end_ordered_io
;
663 clear_buffer_dirty(bh
);
664 if (!buffer_uptodate(bh
))
666 submit_bh(REQ_OP_WRITE
, 0, bh
);
669 #define CHUNK_SIZE 32
670 struct buffer_chunk
{
671 struct buffer_head
*bh
[CHUNK_SIZE
];
675 static void write_chunk(struct buffer_chunk
*chunk
)
678 for (i
= 0; i
< chunk
->nr
; i
++) {
679 submit_logged_buffer(chunk
->bh
[i
]);
684 static void write_ordered_chunk(struct buffer_chunk
*chunk
)
687 for (i
= 0; i
< chunk
->nr
; i
++) {
688 submit_ordered_buffer(chunk
->bh
[i
]);
693 static int add_to_chunk(struct buffer_chunk
*chunk
, struct buffer_head
*bh
,
694 spinlock_t
* lock
, void (fn
) (struct buffer_chunk
*))
697 BUG_ON(chunk
->nr
>= CHUNK_SIZE
);
698 chunk
->bh
[chunk
->nr
++] = bh
;
699 if (chunk
->nr
>= CHUNK_SIZE
) {
712 static atomic_t nr_reiserfs_jh
= ATOMIC_INIT(0);
713 static struct reiserfs_jh
*alloc_jh(void)
715 struct reiserfs_jh
*jh
;
717 jh
= kmalloc(sizeof(*jh
), GFP_NOFS
);
719 atomic_inc(&nr_reiserfs_jh
);
727 * we want to free the jh when the buffer has been written
730 void reiserfs_free_jh(struct buffer_head
*bh
)
732 struct reiserfs_jh
*jh
;
736 bh
->b_private
= NULL
;
738 list_del_init(&jh
->list
);
740 if (atomic_read(&nr_reiserfs_jh
) <= 0)
742 atomic_dec(&nr_reiserfs_jh
);
747 static inline int __add_jh(struct reiserfs_journal
*j
, struct buffer_head
*bh
,
750 struct reiserfs_jh
*jh
;
753 spin_lock(&j
->j_dirty_buffers_lock
);
754 if (!bh
->b_private
) {
755 spin_unlock(&j
->j_dirty_buffers_lock
);
759 list_del_init(&jh
->list
);
764 spin_lock(&j
->j_dirty_buffers_lock
);
766 * buffer must be locked for __add_jh, should be able to have
767 * two adds at the same time
769 BUG_ON(bh
->b_private
);
773 jh
->jl
= j
->j_current_jl
;
775 list_add_tail(&jh
->list
, &jh
->jl
->j_tail_bh_list
);
777 list_add_tail(&jh
->list
, &jh
->jl
->j_bh_list
);
779 spin_unlock(&j
->j_dirty_buffers_lock
);
783 int reiserfs_add_tail_list(struct inode
*inode
, struct buffer_head
*bh
)
785 return __add_jh(SB_JOURNAL(inode
->i_sb
), bh
, 1);
787 int reiserfs_add_ordered_list(struct inode
*inode
, struct buffer_head
*bh
)
789 return __add_jh(SB_JOURNAL(inode
->i_sb
), bh
, 0);
792 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
793 static int write_ordered_buffers(spinlock_t
* lock
,
794 struct reiserfs_journal
*j
,
795 struct reiserfs_journal_list
*jl
,
796 struct list_head
*list
)
798 struct buffer_head
*bh
;
799 struct reiserfs_jh
*jh
;
800 int ret
= j
->j_errno
;
801 struct buffer_chunk chunk
;
802 struct list_head tmp
;
803 INIT_LIST_HEAD(&tmp
);
807 while (!list_empty(list
)) {
808 jh
= JH_ENTRY(list
->next
);
811 if (!trylock_buffer(bh
)) {
812 if (!buffer_dirty(bh
)) {
813 list_move(&jh
->list
, &tmp
);
818 write_ordered_chunk(&chunk
);
825 * in theory, dirty non-uptodate buffers should never get here,
826 * but the upper layer io error paths still have a few quirks.
827 * Handle them here as gracefully as we can
829 if (!buffer_uptodate(bh
) && buffer_dirty(bh
)) {
830 clear_buffer_dirty(bh
);
833 if (buffer_dirty(bh
)) {
834 list_move(&jh
->list
, &tmp
);
835 add_to_chunk(&chunk
, bh
, lock
, write_ordered_chunk
);
837 reiserfs_free_jh(bh
);
842 cond_resched_lock(lock
);
846 write_ordered_chunk(&chunk
);
849 while (!list_empty(&tmp
)) {
850 jh
= JH_ENTRY(tmp
.prev
);
853 reiserfs_free_jh(bh
);
855 if (buffer_locked(bh
)) {
860 if (!buffer_uptodate(bh
)) {
864 * ugly interaction with invalidatepage here.
865 * reiserfs_invalidate_page will pin any buffer that has a
866 * valid journal head from an older transaction. If someone
867 * else sets our buffer dirty after we write it in the first
868 * loop, and then someone truncates the page away, nobody
869 * will ever write the buffer. We're safe if we write the
870 * page one last time after freeing the journal header.
872 if (buffer_dirty(bh
) && unlikely(bh
->b_page
->mapping
== NULL
)) {
874 ll_rw_block(REQ_OP_WRITE
, 0, 1, &bh
);
878 cond_resched_lock(lock
);
884 static int flush_older_commits(struct super_block
*s
,
885 struct reiserfs_journal_list
*jl
)
887 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
888 struct reiserfs_journal_list
*other_jl
;
889 struct reiserfs_journal_list
*first_jl
;
890 struct list_head
*entry
;
891 unsigned int trans_id
= jl
->j_trans_id
;
892 unsigned int other_trans_id
;
893 unsigned int first_trans_id
;
897 * first we walk backwards to find the oldest uncommitted transation
900 entry
= jl
->j_list
.prev
;
902 other_jl
= JOURNAL_LIST_ENTRY(entry
);
903 if (entry
== &journal
->j_journal_list
||
904 atomic_read(&other_jl
->j_older_commits_done
))
908 entry
= other_jl
->j_list
.prev
;
911 /* if we didn't find any older uncommitted transactions, return now */
912 if (first_jl
== jl
) {
916 first_trans_id
= first_jl
->j_trans_id
;
918 entry
= &first_jl
->j_list
;
920 other_jl
= JOURNAL_LIST_ENTRY(entry
);
921 other_trans_id
= other_jl
->j_trans_id
;
923 if (other_trans_id
< trans_id
) {
924 if (atomic_read(&other_jl
->j_commit_left
) != 0) {
925 flush_commit_list(s
, other_jl
, 0);
927 /* list we were called with is gone, return */
928 if (!journal_list_still_alive(s
, trans_id
))
932 * the one we just flushed is gone, this means
933 * all older lists are also gone, so first_jl
934 * is no longer valid either. Go back to the
937 if (!journal_list_still_alive
938 (s
, other_trans_id
)) {
943 if (entry
== &journal
->j_journal_list
)
952 static int reiserfs_async_progress_wait(struct super_block
*s
)
954 struct reiserfs_journal
*j
= SB_JOURNAL(s
);
956 if (atomic_read(&j
->j_async_throttle
)) {
959 depth
= reiserfs_write_unlock_nested(s
);
960 congestion_wait(BLK_RW_ASYNC
, HZ
/ 10);
961 reiserfs_write_lock_nested(s
, depth
);
968 * if this journal list still has commit blocks unflushed, send them to disk.
970 * log areas must be flushed in order (transaction 2 can't commit before
971 * transaction 1) Before the commit block can by written, every other log
972 * block must be safely on disk
974 static int flush_commit_list(struct super_block
*s
,
975 struct reiserfs_journal_list
*jl
, int flushall
)
979 struct buffer_head
*tbh
= NULL
;
980 unsigned int trans_id
= jl
->j_trans_id
;
981 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
986 reiserfs_check_lock_depth(s
, "flush_commit_list");
988 if (atomic_read(&jl
->j_older_commits_done
)) {
993 * before we can put our commit blocks on disk, we have to make
994 * sure everyone older than us is on disk too
996 BUG_ON(jl
->j_len
<= 0);
997 BUG_ON(trans_id
== journal
->j_trans_id
);
999 get_journal_list(jl
);
1001 if (flush_older_commits(s
, jl
) == 1) {
1003 * list disappeared during flush_older_commits.
1010 /* make sure nobody is trying to flush this one at the same time */
1011 reiserfs_mutex_lock_safe(&jl
->j_commit_mutex
, s
);
1013 if (!journal_list_still_alive(s
, trans_id
)) {
1014 mutex_unlock(&jl
->j_commit_mutex
);
1017 BUG_ON(jl
->j_trans_id
== 0);
1019 /* this commit is done, exit */
1020 if (atomic_read(&jl
->j_commit_left
) <= 0) {
1022 atomic_set(&jl
->j_older_commits_done
, 1);
1024 mutex_unlock(&jl
->j_commit_mutex
);
1028 if (!list_empty(&jl
->j_bh_list
)) {
1032 * We might sleep in numerous places inside
1033 * write_ordered_buffers. Relax the write lock.
1035 depth
= reiserfs_write_unlock_nested(s
);
1036 ret
= write_ordered_buffers(&journal
->j_dirty_buffers_lock
,
1037 journal
, jl
, &jl
->j_bh_list
);
1038 if (ret
< 0 && retval
== 0)
1040 reiserfs_write_lock_nested(s
, depth
);
1042 BUG_ON(!list_empty(&jl
->j_bh_list
));
1044 * for the description block and all the log blocks, submit any buffers
1045 * that haven't already reached the disk. Try to write at least 256
1046 * log blocks. later on, we will only wait on blocks that correspond
1047 * to this transaction, but while we're unplugging we might as well
1048 * get a chunk of data on there.
1050 atomic_inc(&journal
->j_async_throttle
);
1051 write_len
= jl
->j_len
+ 1;
1052 if (write_len
< 256)
1054 for (i
= 0 ; i
< write_len
; i
++) {
1055 bn
= SB_ONDISK_JOURNAL_1st_BLOCK(s
) + (jl
->j_start
+ i
) %
1056 SB_ONDISK_JOURNAL_SIZE(s
);
1057 tbh
= journal_find_get_block(s
, bn
);
1059 if (buffer_dirty(tbh
)) {
1060 depth
= reiserfs_write_unlock_nested(s
);
1061 ll_rw_block(REQ_OP_WRITE
, 0, 1, &tbh
);
1062 reiserfs_write_lock_nested(s
, depth
);
1067 atomic_dec(&journal
->j_async_throttle
);
1069 for (i
= 0; i
< (jl
->j_len
+ 1); i
++) {
1070 bn
= SB_ONDISK_JOURNAL_1st_BLOCK(s
) +
1071 (jl
->j_start
+ i
) % SB_ONDISK_JOURNAL_SIZE(s
);
1072 tbh
= journal_find_get_block(s
, bn
);
1074 depth
= reiserfs_write_unlock_nested(s
);
1075 __wait_on_buffer(tbh
);
1076 reiserfs_write_lock_nested(s
, depth
);
1078 * since we're using ll_rw_blk above, it might have skipped
1079 * over a locked buffer. Double check here
1081 /* redundant, sync_dirty_buffer() checks */
1082 if (buffer_dirty(tbh
)) {
1083 depth
= reiserfs_write_unlock_nested(s
);
1084 sync_dirty_buffer(tbh
);
1085 reiserfs_write_lock_nested(s
, depth
);
1087 if (unlikely(!buffer_uptodate(tbh
))) {
1088 #ifdef CONFIG_REISERFS_CHECK
1089 reiserfs_warning(s
, "journal-601",
1090 "buffer write failed");
1094 /* once for journal_find_get_block */
1096 /* once due to original getblk in do_journal_end */
1098 atomic_dec(&jl
->j_commit_left
);
1101 BUG_ON(atomic_read(&jl
->j_commit_left
) != 1);
1104 * If there was a write error in the journal - we can't commit
1105 * this transaction - it will be invalid and, if successful,
1106 * will just end up propagating the write error out to
1109 if (likely(!retval
&& !reiserfs_is_journal_aborted (journal
))) {
1110 if (buffer_dirty(jl
->j_commit_bh
))
1112 mark_buffer_dirty(jl
->j_commit_bh
) ;
1113 depth
= reiserfs_write_unlock_nested(s
);
1114 if (reiserfs_barrier_flush(s
))
1115 __sync_dirty_buffer(jl
->j_commit_bh
,
1116 REQ_SYNC
| REQ_PREFLUSH
| REQ_FUA
);
1118 sync_dirty_buffer(jl
->j_commit_bh
);
1119 reiserfs_write_lock_nested(s
, depth
);
1123 * If there was a write error in the journal - we can't commit this
1124 * transaction - it will be invalid and, if successful, will just end
1125 * up propagating the write error out to the filesystem.
1127 if (unlikely(!buffer_uptodate(jl
->j_commit_bh
))) {
1128 #ifdef CONFIG_REISERFS_CHECK
1129 reiserfs_warning(s
, "journal-615", "buffer write failed");
1133 bforget(jl
->j_commit_bh
);
1134 if (journal
->j_last_commit_id
!= 0 &&
1135 (jl
->j_trans_id
- journal
->j_last_commit_id
) != 1) {
1136 reiserfs_warning(s
, "clm-2200", "last commit %lu, current %lu",
1137 journal
->j_last_commit_id
, jl
->j_trans_id
);
1139 journal
->j_last_commit_id
= jl
->j_trans_id
;
1142 * now, every commit block is on the disk. It is safe to allow
1143 * blocks freed during this transaction to be reallocated
1145 cleanup_freed_for_journal_list(s
, jl
);
1147 retval
= retval
? retval
: journal
->j_errno
;
1149 /* mark the metadata dirty */
1151 dirty_one_transaction(s
, jl
);
1152 atomic_dec(&jl
->j_commit_left
);
1155 atomic_set(&jl
->j_older_commits_done
, 1);
1157 mutex_unlock(&jl
->j_commit_mutex
);
1159 put_journal_list(s
, jl
);
1162 reiserfs_abort(s
, retval
, "Journal write error in %s",
1168 * flush_journal_list frequently needs to find a newer transaction for a
1169 * given block. This does that, or returns NULL if it can't find anything
1171 static struct reiserfs_journal_list
*find_newer_jl_for_cn(struct
1172 reiserfs_journal_cnode
1175 struct super_block
*sb
= cn
->sb
;
1176 b_blocknr_t blocknr
= cn
->blocknr
;
1180 if (cn
->sb
== sb
&& cn
->blocknr
== blocknr
&& cn
->jlist
) {
1188 static void remove_journal_hash(struct super_block
*,
1189 struct reiserfs_journal_cnode
**,
1190 struct reiserfs_journal_list
*, unsigned long,
1194 * once all the real blocks have been flushed, it is safe to remove them
1195 * from the journal list for this transaction. Aside from freeing the
1196 * cnode, this also allows the block to be reallocated for data blocks
1197 * if it had been deleted.
1199 static void remove_all_from_journal_list(struct super_block
*sb
,
1200 struct reiserfs_journal_list
*jl
,
1203 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1204 struct reiserfs_journal_cnode
*cn
, *last
;
1205 cn
= jl
->j_realblock
;
1208 * which is better, to lock once around the whole loop, or
1209 * to lock for each call to remove_journal_hash?
1212 if (cn
->blocknr
!= 0) {
1214 reiserfs_warning(sb
, "reiserfs-2201",
1215 "block %u, bh is %d, state %ld",
1216 cn
->blocknr
, cn
->bh
? 1 : 0,
1220 remove_journal_hash(sb
, journal
->j_list_hash_table
,
1221 jl
, cn
->blocknr
, 1);
1225 free_cnode(sb
, last
);
1227 jl
->j_realblock
= NULL
;
1231 * if this timestamp is greater than the timestamp we wrote last to the
1232 * header block, write it to the header block. once this is done, I can
1233 * safely say the log area for this transaction won't ever be replayed,
1234 * and I can start releasing blocks in this transaction for reuse as data
1235 * blocks. called by flush_journal_list, before it calls
1236 * remove_all_from_journal_list
1238 static int _update_journal_header_block(struct super_block
*sb
,
1239 unsigned long offset
,
1240 unsigned int trans_id
)
1242 struct reiserfs_journal_header
*jh
;
1243 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1246 if (reiserfs_is_journal_aborted(journal
))
1249 if (trans_id
>= journal
->j_last_flush_trans_id
) {
1250 if (buffer_locked((journal
->j_header_bh
))) {
1251 depth
= reiserfs_write_unlock_nested(sb
);
1252 __wait_on_buffer(journal
->j_header_bh
);
1253 reiserfs_write_lock_nested(sb
, depth
);
1254 if (unlikely(!buffer_uptodate(journal
->j_header_bh
))) {
1255 #ifdef CONFIG_REISERFS_CHECK
1256 reiserfs_warning(sb
, "journal-699",
1257 "buffer write failed");
1262 journal
->j_last_flush_trans_id
= trans_id
;
1263 journal
->j_first_unflushed_offset
= offset
;
1264 jh
= (struct reiserfs_journal_header
*)(journal
->j_header_bh
->
1266 jh
->j_last_flush_trans_id
= cpu_to_le32(trans_id
);
1267 jh
->j_first_unflushed_offset
= cpu_to_le32(offset
);
1268 jh
->j_mount_id
= cpu_to_le32(journal
->j_mount_id
);
1270 set_buffer_dirty(journal
->j_header_bh
);
1271 depth
= reiserfs_write_unlock_nested(sb
);
1273 if (reiserfs_barrier_flush(sb
))
1274 __sync_dirty_buffer(journal
->j_header_bh
,
1275 REQ_SYNC
| REQ_PREFLUSH
| REQ_FUA
);
1277 sync_dirty_buffer(journal
->j_header_bh
);
1279 reiserfs_write_lock_nested(sb
, depth
);
1280 if (!buffer_uptodate(journal
->j_header_bh
)) {
1281 reiserfs_warning(sb
, "journal-837",
1282 "IO error during journal replay");
1289 static int update_journal_header_block(struct super_block
*sb
,
1290 unsigned long offset
,
1291 unsigned int trans_id
)
1293 return _update_journal_header_block(sb
, offset
, trans_id
);
1297 ** flush any and all journal lists older than you are
1298 ** can only be called from flush_journal_list
1300 static int flush_older_journal_lists(struct super_block
*sb
,
1301 struct reiserfs_journal_list
*jl
)
1303 struct list_head
*entry
;
1304 struct reiserfs_journal_list
*other_jl
;
1305 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1306 unsigned int trans_id
= jl
->j_trans_id
;
1309 * we know we are the only ones flushing things, no extra race
1310 * protection is required.
1313 entry
= journal
->j_journal_list
.next
;
1315 if (entry
== &journal
->j_journal_list
)
1317 other_jl
= JOURNAL_LIST_ENTRY(entry
);
1318 if (other_jl
->j_trans_id
< trans_id
) {
1319 BUG_ON(other_jl
->j_refcount
<= 0);
1320 /* do not flush all */
1321 flush_journal_list(sb
, other_jl
, 0);
1323 /* other_jl is now deleted from the list */
1329 static void del_from_work_list(struct super_block
*s
,
1330 struct reiserfs_journal_list
*jl
)
1332 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1333 if (!list_empty(&jl
->j_working_list
)) {
1334 list_del_init(&jl
->j_working_list
);
1335 journal
->j_num_work_lists
--;
1340 * flush a journal list, both commit and real blocks
1342 * always set flushall to 1, unless you are calling from inside
1343 * flush_journal_list
1345 * IMPORTANT. This can only be called while there are no journal writers,
1346 * and the journal is locked. That means it can only be called from
1347 * do_journal_end, or by journal_release
1349 static int flush_journal_list(struct super_block
*s
,
1350 struct reiserfs_journal_list
*jl
, int flushall
)
1352 struct reiserfs_journal_list
*pjl
;
1353 struct reiserfs_journal_cnode
*cn
, *last
;
1357 struct buffer_head
*saved_bh
;
1358 unsigned long j_len_saved
= jl
->j_len
;
1359 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1363 BUG_ON(j_len_saved
<= 0);
1365 if (atomic_read(&journal
->j_wcount
) != 0) {
1366 reiserfs_warning(s
, "clm-2048", "called with wcount %d",
1367 atomic_read(&journal
->j_wcount
));
1370 /* if flushall == 0, the lock is already held */
1372 reiserfs_mutex_lock_safe(&journal
->j_flush_mutex
, s
);
1373 } else if (mutex_trylock(&journal
->j_flush_mutex
)) {
1378 if (j_len_saved
> journal
->j_trans_max
) {
1379 reiserfs_panic(s
, "journal-715", "length is %lu, trans id %lu",
1380 j_len_saved
, jl
->j_trans_id
);
1384 /* if all the work is already done, get out of here */
1385 if (atomic_read(&jl
->j_nonzerolen
) <= 0 &&
1386 atomic_read(&jl
->j_commit_left
) <= 0) {
1387 goto flush_older_and_return
;
1391 * start by putting the commit list on disk. This will also flush
1392 * the commit lists of any olders transactions
1394 flush_commit_list(s
, jl
, 1);
1396 if (!(jl
->j_state
& LIST_DIRTY
)
1397 && !reiserfs_is_journal_aborted(journal
))
1400 /* are we done now? */
1401 if (atomic_read(&jl
->j_nonzerolen
) <= 0 &&
1402 atomic_read(&jl
->j_commit_left
) <= 0) {
1403 goto flush_older_and_return
;
1407 * loop through each cnode, see if we need to write it,
1408 * or wait on a more recent transaction, or just ignore it
1410 if (atomic_read(&journal
->j_wcount
) != 0) {
1411 reiserfs_panic(s
, "journal-844", "journal list is flushing, "
1414 cn
= jl
->j_realblock
;
1419 /* blocknr of 0 is no longer in the hash, ignore it */
1420 if (cn
->blocknr
== 0) {
1425 * This transaction failed commit.
1426 * Don't write out to the disk
1428 if (!(jl
->j_state
& LIST_DIRTY
))
1431 pjl
= find_newer_jl_for_cn(cn
);
1433 * the order is important here. We check pjl to make sure we
1434 * don't clear BH_JDirty_wait if we aren't the one writing this
1437 if (!pjl
&& cn
->bh
) {
1441 * we do this to make sure nobody releases the
1442 * buffer while we are working with it
1446 if (buffer_journal_dirty(saved_bh
)) {
1447 BUG_ON(!can_dirty(cn
));
1450 } else if (can_dirty(cn
)) {
1452 * everything with !pjl && jwait
1453 * should be writable
1460 * if someone has this block in a newer transaction, just make
1461 * sure they are committed, and don't try writing it to disk
1464 if (atomic_read(&pjl
->j_commit_left
))
1465 flush_commit_list(s
, pjl
, 1);
1470 * bh == NULL when the block got to disk on its own, OR,
1471 * the block got freed in a future transaction
1473 if (saved_bh
== NULL
) {
1478 * this should never happen. kupdate_one_transaction has
1479 * this list locked while it works, so we should never see a
1480 * buffer here that is not marked JDirty_wait
1482 if ((!was_jwait
) && !buffer_locked(saved_bh
)) {
1483 reiserfs_warning(s
, "journal-813",
1484 "BAD! buffer %llu %cdirty %cjwait, "
1485 "not in a newer transaction",
1486 (unsigned long long)saved_bh
->
1487 b_blocknr
, was_dirty
? ' ' : '!',
1488 was_jwait
? ' ' : '!');
1492 * we inc again because saved_bh gets decremented
1496 set_bit(BLOCK_NEEDS_FLUSH
, &cn
->state
);
1497 lock_buffer(saved_bh
);
1498 BUG_ON(cn
->blocknr
!= saved_bh
->b_blocknr
);
1499 if (buffer_dirty(saved_bh
))
1500 submit_logged_buffer(saved_bh
);
1502 unlock_buffer(saved_bh
);
1505 reiserfs_warning(s
, "clm-2082",
1506 "Unable to flush buffer %llu in %s",
1507 (unsigned long long)saved_bh
->
1508 b_blocknr
, __func__
);
1515 * we incremented this to keep others from
1516 * taking the buffer head away
1519 if (atomic_read(&saved_bh
->b_count
) < 0) {
1520 reiserfs_warning(s
, "journal-945",
1521 "saved_bh->b_count < 0");
1526 cn
= jl
->j_realblock
;
1528 if (test_bit(BLOCK_NEEDS_FLUSH
, &cn
->state
)) {
1530 reiserfs_panic(s
, "journal-1011",
1534 depth
= reiserfs_write_unlock_nested(s
);
1535 __wait_on_buffer(cn
->bh
);
1536 reiserfs_write_lock_nested(s
, depth
);
1539 reiserfs_panic(s
, "journal-1012",
1542 if (unlikely(!buffer_uptodate(cn
->bh
))) {
1543 #ifdef CONFIG_REISERFS_CHECK
1544 reiserfs_warning(s
, "journal-949",
1545 "buffer write failed");
1550 * note, we must clear the JDirty_wait bit
1551 * after the up to date check, otherwise we
1552 * race against our flushpage routine
1554 BUG_ON(!test_clear_buffer_journal_dirty
1557 /* drop one ref for us */
1559 /* drop one ref for journal_mark_dirty */
1560 release_buffer_page(cn
->bh
);
1567 reiserfs_abort(s
, -EIO
,
1568 "Write error while pushing transaction to disk in %s",
1570 flush_older_and_return
:
1573 * before we can update the journal header block, we _must_ flush all
1574 * real blocks from all older transactions to disk. This is because
1575 * once the header block is updated, this transaction will not be
1576 * replayed after a crash
1579 flush_older_journal_lists(s
, jl
);
1582 err
= journal
->j_errno
;
1584 * before we can remove everything from the hash tables for this
1585 * transaction, we must make sure it can never be replayed
1587 * since we are only called from do_journal_end, we know for sure there
1588 * are no allocations going on while we are flushing journal lists. So,
1589 * we only need to update the journal header block for the last list
1592 if (!err
&& flushall
) {
1594 update_journal_header_block(s
,
1595 (jl
->j_start
+ jl
->j_len
+
1596 2) % SB_ONDISK_JOURNAL_SIZE(s
),
1599 reiserfs_abort(s
, -EIO
,
1600 "Write error while updating journal header in %s",
1603 remove_all_from_journal_list(s
, jl
, 0);
1604 list_del_init(&jl
->j_list
);
1605 journal
->j_num_lists
--;
1606 del_from_work_list(s
, jl
);
1608 if (journal
->j_last_flush_id
!= 0 &&
1609 (jl
->j_trans_id
- journal
->j_last_flush_id
) != 1) {
1610 reiserfs_warning(s
, "clm-2201", "last flush %lu, current %lu",
1611 journal
->j_last_flush_id
, jl
->j_trans_id
);
1613 journal
->j_last_flush_id
= jl
->j_trans_id
;
1616 * not strictly required since we are freeing the list, but it should
1617 * help find code using dead lists later on
1620 atomic_set(&jl
->j_nonzerolen
, 0);
1622 jl
->j_realblock
= NULL
;
1623 jl
->j_commit_bh
= NULL
;
1626 put_journal_list(s
, jl
);
1628 mutex_unlock(&journal
->j_flush_mutex
);
1632 static int write_one_transaction(struct super_block
*s
,
1633 struct reiserfs_journal_list
*jl
,
1634 struct buffer_chunk
*chunk
)
1636 struct reiserfs_journal_cnode
*cn
;
1639 jl
->j_state
|= LIST_TOUCHED
;
1640 del_from_work_list(s
, jl
);
1641 if (jl
->j_len
== 0 || atomic_read(&jl
->j_nonzerolen
) == 0) {
1645 cn
= jl
->j_realblock
;
1648 * if the blocknr == 0, this has been cleared from the hash,
1651 if (cn
->blocknr
== 0) {
1654 if (cn
->bh
&& can_dirty(cn
) && buffer_dirty(cn
->bh
)) {
1655 struct buffer_head
*tmp_bh
;
1657 * we can race against journal_mark_freed when we try
1658 * to lock_buffer(cn->bh), so we have to inc the buffer
1659 * count, and recheck things after locking
1663 lock_buffer(tmp_bh
);
1664 if (cn
->bh
&& can_dirty(cn
) && buffer_dirty(tmp_bh
)) {
1665 if (!buffer_journal_dirty(tmp_bh
) ||
1666 buffer_journal_prepared(tmp_bh
))
1668 add_to_chunk(chunk
, tmp_bh
, NULL
, write_chunk
);
1671 /* note, cn->bh might be null now */
1672 unlock_buffer(tmp_bh
);
1683 /* used by flush_commit_list */
1684 static int dirty_one_transaction(struct super_block
*s
,
1685 struct reiserfs_journal_list
*jl
)
1687 struct reiserfs_journal_cnode
*cn
;
1688 struct reiserfs_journal_list
*pjl
;
1691 jl
->j_state
|= LIST_DIRTY
;
1692 cn
= jl
->j_realblock
;
1695 * look for a more recent transaction that logged this
1696 * buffer. Only the most recent transaction with a buffer in
1697 * it is allowed to send that buffer to disk
1699 pjl
= find_newer_jl_for_cn(cn
);
1700 if (!pjl
&& cn
->blocknr
&& cn
->bh
1701 && buffer_journal_dirty(cn
->bh
)) {
1702 BUG_ON(!can_dirty(cn
));
1704 * if the buffer is prepared, it will either be logged
1705 * or restored. If restored, we need to make sure
1706 * it actually gets marked dirty
1708 clear_buffer_journal_new(cn
->bh
);
1709 if (buffer_journal_prepared(cn
->bh
)) {
1710 set_buffer_journal_restore_dirty(cn
->bh
);
1712 set_buffer_journal_test(cn
->bh
);
1713 mark_buffer_dirty(cn
->bh
);
1721 static int kupdate_transactions(struct super_block
*s
,
1722 struct reiserfs_journal_list
*jl
,
1723 struct reiserfs_journal_list
**next_jl
,
1724 unsigned int *next_trans_id
,
1725 int num_blocks
, int num_trans
)
1729 int transactions_flushed
= 0;
1730 unsigned int orig_trans_id
= jl
->j_trans_id
;
1731 struct buffer_chunk chunk
;
1732 struct list_head
*entry
;
1733 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1736 reiserfs_mutex_lock_safe(&journal
->j_flush_mutex
, s
);
1737 if (!journal_list_still_alive(s
, orig_trans_id
)) {
1742 * we've got j_flush_mutex held, nobody is going to delete any
1743 * of these lists out from underneath us
1745 while ((num_trans
&& transactions_flushed
< num_trans
) ||
1746 (!num_trans
&& written
< num_blocks
)) {
1748 if (jl
->j_len
== 0 || (jl
->j_state
& LIST_TOUCHED
) ||
1749 atomic_read(&jl
->j_commit_left
)
1750 || !(jl
->j_state
& LIST_DIRTY
)) {
1751 del_from_work_list(s
, jl
);
1754 ret
= write_one_transaction(s
, jl
, &chunk
);
1758 transactions_flushed
++;
1760 entry
= jl
->j_list
.next
;
1763 if (entry
== &journal
->j_journal_list
) {
1766 jl
= JOURNAL_LIST_ENTRY(entry
);
1768 /* don't bother with older transactions */
1769 if (jl
->j_trans_id
<= orig_trans_id
)
1773 write_chunk(&chunk
);
1777 mutex_unlock(&journal
->j_flush_mutex
);
1782 * for o_sync and fsync heavy applications, they tend to use
1783 * all the journa list slots with tiny transactions. These
1784 * trigger lots and lots of calls to update the header block, which
1785 * adds seeks and slows things down.
1787 * This function tries to clear out a large chunk of the journal lists
1788 * at once, which makes everything faster since only the newest journal
1789 * list updates the header block
1791 static int flush_used_journal_lists(struct super_block
*s
,
1792 struct reiserfs_journal_list
*jl
)
1794 unsigned long len
= 0;
1795 unsigned long cur_len
;
1799 struct reiserfs_journal_list
*tjl
;
1800 struct reiserfs_journal_list
*flush_jl
;
1801 unsigned int trans_id
;
1802 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1804 flush_jl
= tjl
= jl
;
1806 /* in data logging mode, try harder to flush a lot of blocks */
1807 if (reiserfs_data_log(s
))
1809 /* flush for 256 transactions or limit blocks, whichever comes first */
1810 for (i
= 0; i
< 256 && len
< limit
; i
++) {
1811 if (atomic_read(&tjl
->j_commit_left
) ||
1812 tjl
->j_trans_id
< jl
->j_trans_id
) {
1815 cur_len
= atomic_read(&tjl
->j_nonzerolen
);
1817 tjl
->j_state
&= ~LIST_TOUCHED
;
1821 if (tjl
->j_list
.next
== &journal
->j_journal_list
)
1823 tjl
= JOURNAL_LIST_ENTRY(tjl
->j_list
.next
);
1825 get_journal_list(jl
);
1826 get_journal_list(flush_jl
);
1829 * try to find a group of blocks we can flush across all the
1830 * transactions, but only bother if we've actually spanned
1831 * across multiple lists
1833 if (flush_jl
!= jl
) {
1834 ret
= kupdate_transactions(s
, jl
, &tjl
, &trans_id
, len
, i
);
1836 flush_journal_list(s
, flush_jl
, 1);
1837 put_journal_list(s
, flush_jl
);
1838 put_journal_list(s
, jl
);
1843 * removes any nodes in table with name block and dev as bh.
1844 * only touchs the hnext and hprev pointers.
1846 void remove_journal_hash(struct super_block
*sb
,
1847 struct reiserfs_journal_cnode
**table
,
1848 struct reiserfs_journal_list
*jl
,
1849 unsigned long block
, int remove_freed
)
1851 struct reiserfs_journal_cnode
*cur
;
1852 struct reiserfs_journal_cnode
**head
;
1854 head
= &(journal_hash(table
, sb
, block
));
1860 if (cur
->blocknr
== block
&& cur
->sb
== sb
1861 && (jl
== NULL
|| jl
== cur
->jlist
)
1862 && (!test_bit(BLOCK_FREED
, &cur
->state
) || remove_freed
)) {
1864 cur
->hnext
->hprev
= cur
->hprev
;
1867 cur
->hprev
->hnext
= cur
->hnext
;
1875 * anybody who clears the cur->bh will also
1876 * dec the nonzerolen
1878 if (cur
->bh
&& cur
->jlist
)
1879 atomic_dec(&cur
->jlist
->j_nonzerolen
);
1887 static void free_journal_ram(struct super_block
*sb
)
1889 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1890 kfree(journal
->j_current_jl
);
1891 journal
->j_num_lists
--;
1893 vfree(journal
->j_cnode_free_orig
);
1894 free_list_bitmaps(sb
, journal
->j_list_bitmap
);
1895 free_bitmap_nodes(sb
); /* must be after free_list_bitmaps */
1896 if (journal
->j_header_bh
) {
1897 brelse(journal
->j_header_bh
);
1900 * j_header_bh is on the journal dev, make sure
1901 * not to release the journal dev until we brelse j_header_bh
1903 release_journal_dev(sb
, journal
);
1908 * call on unmount. Only set error to 1 if you haven't made your way out
1909 * of read_super() yet. Any other caller must keep error at 0.
1911 static int do_journal_release(struct reiserfs_transaction_handle
*th
,
1912 struct super_block
*sb
, int error
)
1914 struct reiserfs_transaction_handle myth
;
1916 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1919 * we only want to flush out transactions if we were
1920 * called with error == 0
1922 if (!error
&& !sb_rdonly(sb
)) {
1923 /* end the current trans */
1924 BUG_ON(!th
->t_trans_id
);
1925 do_journal_end(th
, FLUSH_ALL
);
1928 * make sure something gets logged to force
1929 * our way into the flush code
1931 if (!journal_join(&myth
, sb
)) {
1932 reiserfs_prepare_for_journal(sb
,
1933 SB_BUFFER_WITH_SB(sb
),
1935 journal_mark_dirty(&myth
, SB_BUFFER_WITH_SB(sb
));
1936 do_journal_end(&myth
, FLUSH_ALL
);
1941 /* this also catches errors during the do_journal_end above */
1942 if (!error
&& reiserfs_is_journal_aborted(journal
)) {
1943 memset(&myth
, 0, sizeof(myth
));
1944 if (!journal_join_abort(&myth
, sb
)) {
1945 reiserfs_prepare_for_journal(sb
,
1946 SB_BUFFER_WITH_SB(sb
),
1948 journal_mark_dirty(&myth
, SB_BUFFER_WITH_SB(sb
));
1949 do_journal_end(&myth
, FLUSH_ALL
);
1955 * We must release the write lock here because
1956 * the workqueue job (flush_async_commit) needs this lock
1958 reiserfs_write_unlock(sb
);
1961 * Cancel flushing of old commits. Note that neither of these works
1962 * will be requeued because superblock is being shutdown and doesn't
1963 * have SB_ACTIVE set.
1965 reiserfs_cancel_old_flush(sb
);
1966 /* wait for all commits to finish */
1967 cancel_delayed_work_sync(&SB_JOURNAL(sb
)->j_work
);
1969 free_journal_ram(sb
);
1971 reiserfs_write_lock(sb
);
1976 /* * call on unmount. flush all journal trans, release all alloc'd ram */
1977 int journal_release(struct reiserfs_transaction_handle
*th
,
1978 struct super_block
*sb
)
1980 return do_journal_release(th
, sb
, 0);
1983 /* only call from an error condition inside reiserfs_read_super! */
1984 int journal_release_error(struct reiserfs_transaction_handle
*th
,
1985 struct super_block
*sb
)
1987 return do_journal_release(th
, sb
, 1);
1991 * compares description block with commit block.
1992 * returns 1 if they differ, 0 if they are the same
1994 static int journal_compare_desc_commit(struct super_block
*sb
,
1995 struct reiserfs_journal_desc
*desc
,
1996 struct reiserfs_journal_commit
*commit
)
1998 if (get_commit_trans_id(commit
) != get_desc_trans_id(desc
) ||
1999 get_commit_trans_len(commit
) != get_desc_trans_len(desc
) ||
2000 get_commit_trans_len(commit
) > SB_JOURNAL(sb
)->j_trans_max
||
2001 get_commit_trans_len(commit
) <= 0) {
2008 * returns 0 if it did not find a description block
2009 * returns -1 if it found a corrupt commit block
2010 * returns 1 if both desc and commit were valid
2011 * NOTE: only called during fs mount
2013 static int journal_transaction_is_valid(struct super_block
*sb
,
2014 struct buffer_head
*d_bh
,
2015 unsigned int *oldest_invalid_trans_id
,
2016 unsigned long *newest_mount_id
)
2018 struct reiserfs_journal_desc
*desc
;
2019 struct reiserfs_journal_commit
*commit
;
2020 struct buffer_head
*c_bh
;
2021 unsigned long offset
;
2026 desc
= (struct reiserfs_journal_desc
*)d_bh
->b_data
;
2027 if (get_desc_trans_len(desc
) > 0
2028 && !memcmp(get_journal_desc_magic(d_bh
), JOURNAL_DESC_MAGIC
, 8)) {
2029 if (oldest_invalid_trans_id
&& *oldest_invalid_trans_id
2030 && get_desc_trans_id(desc
) > *oldest_invalid_trans_id
) {
2031 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2032 "journal-986: transaction "
2033 "is valid returning because trans_id %d is greater than "
2034 "oldest_invalid %lu",
2035 get_desc_trans_id(desc
),
2036 *oldest_invalid_trans_id
);
2040 && *newest_mount_id
> get_desc_mount_id(desc
)) {
2041 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2042 "journal-1087: transaction "
2043 "is valid returning because mount_id %d is less than "
2044 "newest_mount_id %lu",
2045 get_desc_mount_id(desc
),
2049 if (get_desc_trans_len(desc
) > SB_JOURNAL(sb
)->j_trans_max
) {
2050 reiserfs_warning(sb
, "journal-2018",
2051 "Bad transaction length %d "
2052 "encountered, ignoring transaction",
2053 get_desc_trans_len(desc
));
2056 offset
= d_bh
->b_blocknr
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
);
2059 * ok, we have a journal description block,
2060 * let's see if the transaction was valid
2064 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2065 ((offset
+ get_desc_trans_len(desc
) +
2066 1) % SB_ONDISK_JOURNAL_SIZE(sb
)));
2069 commit
= (struct reiserfs_journal_commit
*)c_bh
->b_data
;
2070 if (journal_compare_desc_commit(sb
, desc
, commit
)) {
2071 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2072 "journal_transaction_is_valid, commit offset %ld had bad "
2073 "time %d or length %d",
2075 SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2076 get_commit_trans_id(commit
),
2077 get_commit_trans_len(commit
));
2079 if (oldest_invalid_trans_id
) {
2080 *oldest_invalid_trans_id
=
2081 get_desc_trans_id(desc
);
2082 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2084 "transaction_is_valid setting oldest invalid trans_id "
2086 get_desc_trans_id(desc
));
2091 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2092 "journal-1006: found valid "
2093 "transaction start offset %llu, len %d id %d",
2095 SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2096 get_desc_trans_len(desc
),
2097 get_desc_trans_id(desc
));
2104 static void brelse_array(struct buffer_head
**heads
, int num
)
2107 for (i
= 0; i
< num
; i
++) {
2113 * given the start, and values for the oldest acceptable transactions,
2114 * this either reads in a replays a transaction, or returns because the
2115 * transaction is invalid, or too old.
2116 * NOTE: only called during fs mount
2118 static int journal_read_transaction(struct super_block
*sb
,
2119 unsigned long cur_dblock
,
2120 unsigned long oldest_start
,
2121 unsigned int oldest_trans_id
,
2122 unsigned long newest_mount_id
)
2124 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
2125 struct reiserfs_journal_desc
*desc
;
2126 struct reiserfs_journal_commit
*commit
;
2127 unsigned int trans_id
= 0;
2128 struct buffer_head
*c_bh
;
2129 struct buffer_head
*d_bh
;
2130 struct buffer_head
**log_blocks
= NULL
;
2131 struct buffer_head
**real_blocks
= NULL
;
2132 unsigned int trans_offset
;
2136 d_bh
= journal_bread(sb
, cur_dblock
);
2139 desc
= (struct reiserfs_journal_desc
*)d_bh
->b_data
;
2140 trans_offset
= d_bh
->b_blocknr
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
);
2141 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
, "journal-1037: "
2142 "journal_read_transaction, offset %llu, len %d mount_id %d",
2143 d_bh
->b_blocknr
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2144 get_desc_trans_len(desc
), get_desc_mount_id(desc
));
2145 if (get_desc_trans_id(desc
) < oldest_trans_id
) {
2146 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
, "journal-1039: "
2147 "journal_read_trans skipping because %lu is too old",
2149 SB_ONDISK_JOURNAL_1st_BLOCK(sb
));
2153 if (get_desc_mount_id(desc
) != newest_mount_id
) {
2154 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
, "journal-1146: "
2155 "journal_read_trans skipping because %d is != "
2156 "newest_mount_id %lu", get_desc_mount_id(desc
),
2161 c_bh
= journal_bread(sb
, SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2162 ((trans_offset
+ get_desc_trans_len(desc
) + 1) %
2163 SB_ONDISK_JOURNAL_SIZE(sb
)));
2168 commit
= (struct reiserfs_journal_commit
*)c_bh
->b_data
;
2169 if (journal_compare_desc_commit(sb
, desc
, commit
)) {
2170 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2171 "journal_read_transaction, "
2172 "commit offset %llu had bad time %d or length %d",
2174 SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2175 get_commit_trans_id(commit
),
2176 get_commit_trans_len(commit
));
2182 if (bdev_read_only(sb
->s_bdev
)) {
2183 reiserfs_warning(sb
, "clm-2076",
2184 "device is readonly, unable to replay log");
2190 trans_id
= get_desc_trans_id(desc
);
2192 * now we know we've got a good transaction, and it was
2193 * inside the valid time ranges
2195 log_blocks
= kmalloc_array(get_desc_trans_len(desc
),
2196 sizeof(struct buffer_head
*),
2198 real_blocks
= kmalloc_array(get_desc_trans_len(desc
),
2199 sizeof(struct buffer_head
*),
2201 if (!log_blocks
|| !real_blocks
) {
2206 reiserfs_warning(sb
, "journal-1169",
2207 "kmalloc failed, unable to mount FS");
2210 /* get all the buffer heads */
2211 trans_half
= journal_trans_half(sb
->s_blocksize
);
2212 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2215 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2217 i
) % SB_ONDISK_JOURNAL_SIZE(sb
));
2218 if (i
< trans_half
) {
2221 le32_to_cpu(desc
->j_realblock
[i
]));
2225 le32_to_cpu(commit
->
2226 j_realblock
[i
- trans_half
]));
2228 if (real_blocks
[i
]->b_blocknr
> SB_BLOCK_COUNT(sb
)) {
2229 reiserfs_warning(sb
, "journal-1207",
2230 "REPLAY FAILURE fsck required! "
2231 "Block to replay is outside of "
2235 /* make sure we don't try to replay onto log or reserved area */
2236 if (is_block_in_log_or_reserved_area
2237 (sb
, real_blocks
[i
]->b_blocknr
)) {
2238 reiserfs_warning(sb
, "journal-1204",
2239 "REPLAY FAILURE fsck required! "
2240 "Trying to replay onto a log block");
2242 brelse_array(log_blocks
, i
);
2243 brelse_array(real_blocks
, i
);
2251 /* read in the log blocks, memcpy to the corresponding real block */
2252 ll_rw_block(REQ_OP_READ
, 0, get_desc_trans_len(desc
), log_blocks
);
2253 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2255 wait_on_buffer(log_blocks
[i
]);
2256 if (!buffer_uptodate(log_blocks
[i
])) {
2257 reiserfs_warning(sb
, "journal-1212",
2258 "REPLAY FAILURE fsck required! "
2259 "buffer write failed");
2260 brelse_array(log_blocks
+ i
,
2261 get_desc_trans_len(desc
) - i
);
2262 brelse_array(real_blocks
, get_desc_trans_len(desc
));
2269 memcpy(real_blocks
[i
]->b_data
, log_blocks
[i
]->b_data
,
2270 real_blocks
[i
]->b_size
);
2271 set_buffer_uptodate(real_blocks
[i
]);
2272 brelse(log_blocks
[i
]);
2274 /* flush out the real blocks */
2275 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2276 set_buffer_dirty(real_blocks
[i
]);
2277 write_dirty_buffer(real_blocks
[i
], 0);
2279 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2280 wait_on_buffer(real_blocks
[i
]);
2281 if (!buffer_uptodate(real_blocks
[i
])) {
2282 reiserfs_warning(sb
, "journal-1226",
2283 "REPLAY FAILURE, fsck required! "
2284 "buffer write failed");
2285 brelse_array(real_blocks
+ i
,
2286 get_desc_trans_len(desc
) - i
);
2293 brelse(real_blocks
[i
]);
2296 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2297 ((trans_offset
+ get_desc_trans_len(desc
) +
2298 2) % SB_ONDISK_JOURNAL_SIZE(sb
));
2299 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2300 "journal-1095: setting journal " "start to offset %ld",
2301 cur_dblock
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
));
2304 * init starting values for the first transaction, in case
2305 * this is the last transaction to be replayed.
2307 journal
->j_start
= cur_dblock
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
);
2308 journal
->j_last_flush_trans_id
= trans_id
;
2309 journal
->j_trans_id
= trans_id
+ 1;
2310 /* check for trans_id overflow */
2311 if (journal
->j_trans_id
== 0)
2312 journal
->j_trans_id
= 10;
2321 * This function reads blocks starting from block and to max_block of bufsize
2322 * size (but no more than BUFNR blocks at a time). This proved to improve
2323 * mounting speed on self-rebuilding raid5 arrays at least.
2324 * Right now it is only used from journal code. But later we might use it
2325 * from other places.
2326 * Note: Do not use journal_getblk/sb_getblk functions here!
2328 static struct buffer_head
*reiserfs_breada(struct block_device
*dev
,
2329 b_blocknr_t block
, int bufsize
,
2330 b_blocknr_t max_block
)
2332 struct buffer_head
*bhlist
[BUFNR
];
2333 unsigned int blocks
= BUFNR
;
2334 struct buffer_head
*bh
;
2337 bh
= __getblk(dev
, block
, bufsize
);
2338 if (buffer_uptodate(bh
))
2341 if (block
+ BUFNR
> max_block
) {
2342 blocks
= max_block
- block
;
2346 for (i
= 1; i
< blocks
; i
++) {
2347 bh
= __getblk(dev
, block
+ i
, bufsize
);
2348 if (buffer_uptodate(bh
)) {
2354 ll_rw_block(REQ_OP_READ
, 0, j
, bhlist
);
2355 for (i
= 1; i
< j
; i
++)
2359 if (buffer_uptodate(bh
))
2366 * read and replay the log
2367 * on a clean unmount, the journal header's next unflushed pointer will be
2368 * to an invalid transaction. This tests that before finding all the
2369 * transactions in the log, which makes normal mount times fast.
2371 * After a crash, this starts with the next unflushed transaction, and
2372 * replays until it finds one too old, or invalid.
2374 * On exit, it sets things up so the first transaction will work correctly.
2375 * NOTE: only called during fs mount
2377 static int journal_read(struct super_block
*sb
)
2379 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
2380 struct reiserfs_journal_desc
*desc
;
2381 unsigned int oldest_trans_id
= 0;
2382 unsigned int oldest_invalid_trans_id
= 0;
2384 unsigned long oldest_start
= 0;
2385 unsigned long cur_dblock
= 0;
2386 unsigned long newest_mount_id
= 9;
2387 struct buffer_head
*d_bh
;
2388 struct reiserfs_journal_header
*jh
;
2389 int valid_journal_header
= 0;
2390 int replay_count
= 0;
2391 int continue_replay
= 1;
2394 cur_dblock
= SB_ONDISK_JOURNAL_1st_BLOCK(sb
);
2395 reiserfs_info(sb
, "checking transaction log (%pg)\n",
2397 start
= get_seconds();
2400 * step 1, read in the journal header block. Check the transaction
2401 * it says is the first unflushed, and if that transaction is not
2402 * valid, replay is done
2404 journal
->j_header_bh
= journal_bread(sb
,
2405 SB_ONDISK_JOURNAL_1st_BLOCK(sb
)
2406 + SB_ONDISK_JOURNAL_SIZE(sb
));
2407 if (!journal
->j_header_bh
) {
2410 jh
= (struct reiserfs_journal_header
*)(journal
->j_header_bh
->b_data
);
2411 if (le32_to_cpu(jh
->j_first_unflushed_offset
) <
2412 SB_ONDISK_JOURNAL_SIZE(sb
)
2413 && le32_to_cpu(jh
->j_last_flush_trans_id
) > 0) {
2415 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2416 le32_to_cpu(jh
->j_first_unflushed_offset
);
2417 oldest_trans_id
= le32_to_cpu(jh
->j_last_flush_trans_id
) + 1;
2418 newest_mount_id
= le32_to_cpu(jh
->j_mount_id
);
2419 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2420 "journal-1153: found in "
2421 "header: first_unflushed_offset %d, last_flushed_trans_id "
2422 "%lu", le32_to_cpu(jh
->j_first_unflushed_offset
),
2423 le32_to_cpu(jh
->j_last_flush_trans_id
));
2424 valid_journal_header
= 1;
2427 * now, we try to read the first unflushed offset. If it
2428 * is not valid, there is nothing more we can do, and it
2429 * makes no sense to read through the whole log.
2433 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2434 le32_to_cpu(jh
->j_first_unflushed_offset
));
2435 ret
= journal_transaction_is_valid(sb
, d_bh
, NULL
, NULL
);
2437 continue_replay
= 0;
2440 goto start_log_replay
;
2444 * ok, there are transactions that need to be replayed. start
2445 * with the first log block, find all the valid transactions, and
2446 * pick out the oldest.
2448 while (continue_replay
2450 (SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2451 SB_ONDISK_JOURNAL_SIZE(sb
))) {
2453 * Note that it is required for blocksize of primary fs
2454 * device and journal device to be the same
2457 reiserfs_breada(journal
->j_dev_bd
, cur_dblock
,
2459 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2460 SB_ONDISK_JOURNAL_SIZE(sb
));
2462 journal_transaction_is_valid(sb
, d_bh
,
2463 &oldest_invalid_trans_id
,
2466 desc
= (struct reiserfs_journal_desc
*)d_bh
->b_data
;
2467 if (oldest_start
== 0) { /* init all oldest_ values */
2468 oldest_trans_id
= get_desc_trans_id(desc
);
2469 oldest_start
= d_bh
->b_blocknr
;
2470 newest_mount_id
= get_desc_mount_id(desc
);
2471 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2472 "journal-1179: Setting "
2473 "oldest_start to offset %llu, trans_id %lu",
2475 SB_ONDISK_JOURNAL_1st_BLOCK
2476 (sb
), oldest_trans_id
);
2477 } else if (oldest_trans_id
> get_desc_trans_id(desc
)) {
2478 /* one we just read was older */
2479 oldest_trans_id
= get_desc_trans_id(desc
);
2480 oldest_start
= d_bh
->b_blocknr
;
2481 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2482 "journal-1180: Resetting "
2483 "oldest_start to offset %lu, trans_id %lu",
2485 SB_ONDISK_JOURNAL_1st_BLOCK
2486 (sb
), oldest_trans_id
);
2488 if (newest_mount_id
< get_desc_mount_id(desc
)) {
2489 newest_mount_id
= get_desc_mount_id(desc
);
2490 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2491 "journal-1299: Setting "
2492 "newest_mount_id to %d",
2493 get_desc_mount_id(desc
));
2495 cur_dblock
+= get_desc_trans_len(desc
) + 2;
2503 cur_dblock
= oldest_start
;
2504 if (oldest_trans_id
) {
2505 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2506 "journal-1206: Starting replay "
2507 "from offset %llu, trans_id %lu",
2508 cur_dblock
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2513 while (continue_replay
&& oldest_trans_id
> 0) {
2515 journal_read_transaction(sb
, cur_dblock
, oldest_start
,
2516 oldest_trans_id
, newest_mount_id
);
2519 } else if (ret
!= 0) {
2523 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) + journal
->j_start
;
2525 if (cur_dblock
== oldest_start
)
2529 if (oldest_trans_id
== 0) {
2530 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2531 "journal-1225: No valid " "transactions found");
2534 * j_start does not get set correctly if we don't replay any
2535 * transactions. if we had a valid journal_header, set j_start
2536 * to the first unflushed transaction value, copy the trans_id
2539 if (valid_journal_header
&& replay_count
== 0) {
2540 journal
->j_start
= le32_to_cpu(jh
->j_first_unflushed_offset
);
2541 journal
->j_trans_id
=
2542 le32_to_cpu(jh
->j_last_flush_trans_id
) + 1;
2543 /* check for trans_id overflow */
2544 if (journal
->j_trans_id
== 0)
2545 journal
->j_trans_id
= 10;
2546 journal
->j_last_flush_trans_id
=
2547 le32_to_cpu(jh
->j_last_flush_trans_id
);
2548 journal
->j_mount_id
= le32_to_cpu(jh
->j_mount_id
) + 1;
2550 journal
->j_mount_id
= newest_mount_id
+ 1;
2552 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
, "journal-1299: Setting "
2553 "newest_mount_id to %lu", journal
->j_mount_id
);
2554 journal
->j_first_unflushed_offset
= journal
->j_start
;
2555 if (replay_count
> 0) {
2557 "replayed %d transactions in %lu seconds\n",
2558 replay_count
, get_seconds() - start
);
2560 /* needed to satisfy the locking in _update_journal_header_block */
2561 reiserfs_write_lock(sb
);
2562 if (!bdev_read_only(sb
->s_bdev
) &&
2563 _update_journal_header_block(sb
, journal
->j_start
,
2564 journal
->j_last_flush_trans_id
)) {
2565 reiserfs_write_unlock(sb
);
2567 * replay failed, caller must call free_journal_ram and abort
2572 reiserfs_write_unlock(sb
);
2576 static struct reiserfs_journal_list
*alloc_journal_list(struct super_block
*s
)
2578 struct reiserfs_journal_list
*jl
;
2579 jl
= kzalloc(sizeof(struct reiserfs_journal_list
),
2580 GFP_NOFS
| __GFP_NOFAIL
);
2581 INIT_LIST_HEAD(&jl
->j_list
);
2582 INIT_LIST_HEAD(&jl
->j_working_list
);
2583 INIT_LIST_HEAD(&jl
->j_tail_bh_list
);
2584 INIT_LIST_HEAD(&jl
->j_bh_list
);
2585 mutex_init(&jl
->j_commit_mutex
);
2586 SB_JOURNAL(s
)->j_num_lists
++;
2587 get_journal_list(jl
);
2591 static void journal_list_init(struct super_block
*sb
)
2593 SB_JOURNAL(sb
)->j_current_jl
= alloc_journal_list(sb
);
2596 static void release_journal_dev(struct super_block
*super
,
2597 struct reiserfs_journal
*journal
)
2599 if (journal
->j_dev_bd
!= NULL
) {
2600 blkdev_put(journal
->j_dev_bd
, journal
->j_dev_mode
);
2601 journal
->j_dev_bd
= NULL
;
2605 static int journal_init_dev(struct super_block
*super
,
2606 struct reiserfs_journal
*journal
,
2607 const char *jdev_name
)
2611 fmode_t blkdev_mode
= FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
;
2612 char b
[BDEVNAME_SIZE
];
2616 journal
->j_dev_bd
= NULL
;
2617 jdev
= SB_ONDISK_JOURNAL_DEVICE(super
) ?
2618 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super
)) : super
->s_dev
;
2620 if (bdev_read_only(super
->s_bdev
))
2621 blkdev_mode
= FMODE_READ
;
2623 /* there is no "jdev" option and journal is on separate device */
2624 if ((!jdev_name
|| !jdev_name
[0])) {
2625 if (jdev
== super
->s_dev
)
2626 blkdev_mode
&= ~FMODE_EXCL
;
2627 journal
->j_dev_bd
= blkdev_get_by_dev(jdev
, blkdev_mode
,
2629 journal
->j_dev_mode
= blkdev_mode
;
2630 if (IS_ERR(journal
->j_dev_bd
)) {
2631 result
= PTR_ERR(journal
->j_dev_bd
);
2632 journal
->j_dev_bd
= NULL
;
2633 reiserfs_warning(super
, "sh-458",
2634 "cannot init journal device '%s': %i",
2635 __bdevname(jdev
, b
), result
);
2637 } else if (jdev
!= super
->s_dev
)
2638 set_blocksize(journal
->j_dev_bd
, super
->s_blocksize
);
2643 journal
->j_dev_mode
= blkdev_mode
;
2644 journal
->j_dev_bd
= blkdev_get_by_path(jdev_name
, blkdev_mode
, journal
);
2645 if (IS_ERR(journal
->j_dev_bd
)) {
2646 result
= PTR_ERR(journal
->j_dev_bd
);
2647 journal
->j_dev_bd
= NULL
;
2648 reiserfs_warning(super
, "sh-457",
2649 "journal_init_dev: Cannot open '%s': %i",
2654 set_blocksize(journal
->j_dev_bd
, super
->s_blocksize
);
2655 reiserfs_info(super
,
2656 "journal_init_dev: journal device: %pg\n",
2662 * When creating/tuning a file system user can assign some
2663 * journal params within boundaries which depend on the ratio
2664 * blocksize/standard_blocksize.
2666 * For blocks >= standard_blocksize transaction size should
2667 * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2668 * then JOURNAL_TRANS_MAX_DEFAULT.
2670 * For blocks < standard_blocksize these boundaries should be
2671 * decreased proportionally.
2673 #define REISERFS_STANDARD_BLKSIZE (4096)
2675 static int check_advise_trans_params(struct super_block
*sb
,
2676 struct reiserfs_journal
*journal
)
2678 if (journal
->j_trans_max
) {
2679 /* Non-default journal params. Do sanity check for them. */
2681 if (sb
->s_blocksize
< REISERFS_STANDARD_BLKSIZE
)
2682 ratio
= REISERFS_STANDARD_BLKSIZE
/ sb
->s_blocksize
;
2684 if (journal
->j_trans_max
> JOURNAL_TRANS_MAX_DEFAULT
/ ratio
||
2685 journal
->j_trans_max
< JOURNAL_TRANS_MIN_DEFAULT
/ ratio
||
2686 SB_ONDISK_JOURNAL_SIZE(sb
) / journal
->j_trans_max
<
2687 JOURNAL_MIN_RATIO
) {
2688 reiserfs_warning(sb
, "sh-462",
2689 "bad transaction max size (%u). "
2690 "FSCK?", journal
->j_trans_max
);
2693 if (journal
->j_max_batch
!= (journal
->j_trans_max
) *
2694 JOURNAL_MAX_BATCH_DEFAULT
/JOURNAL_TRANS_MAX_DEFAULT
) {
2695 reiserfs_warning(sb
, "sh-463",
2696 "bad transaction max batch (%u). "
2697 "FSCK?", journal
->j_max_batch
);
2702 * Default journal params.
2703 * The file system was created by old version
2704 * of mkreiserfs, so some fields contain zeros,
2705 * and we need to advise proper values for them
2707 if (sb
->s_blocksize
!= REISERFS_STANDARD_BLKSIZE
) {
2708 reiserfs_warning(sb
, "sh-464", "bad blocksize (%u)",
2712 journal
->j_trans_max
= JOURNAL_TRANS_MAX_DEFAULT
;
2713 journal
->j_max_batch
= JOURNAL_MAX_BATCH_DEFAULT
;
2714 journal
->j_max_commit_age
= JOURNAL_MAX_COMMIT_AGE
;
2719 /* must be called once on fs mount. calls journal_read for you */
2720 int journal_init(struct super_block
*sb
, const char *j_dev_name
,
2721 int old_format
, unsigned int commit_max_age
)
2723 int num_cnodes
= SB_ONDISK_JOURNAL_SIZE(sb
) * 2;
2724 struct buffer_head
*bhjh
;
2725 struct reiserfs_super_block
*rs
;
2726 struct reiserfs_journal_header
*jh
;
2727 struct reiserfs_journal
*journal
;
2728 struct reiserfs_journal_list
*jl
;
2731 journal
= SB_JOURNAL(sb
) = vzalloc(sizeof(struct reiserfs_journal
));
2733 reiserfs_warning(sb
, "journal-1256",
2734 "unable to get memory for journal structure");
2737 INIT_LIST_HEAD(&journal
->j_bitmap_nodes
);
2738 INIT_LIST_HEAD(&journal
->j_prealloc_list
);
2739 INIT_LIST_HEAD(&journal
->j_working_list
);
2740 INIT_LIST_HEAD(&journal
->j_journal_list
);
2741 journal
->j_persistent_trans
= 0;
2742 if (reiserfs_allocate_list_bitmaps(sb
, journal
->j_list_bitmap
,
2743 reiserfs_bmap_count(sb
)))
2744 goto free_and_return
;
2746 allocate_bitmap_nodes(sb
);
2748 /* reserved for journal area support */
2749 SB_JOURNAL_1st_RESERVED_BLOCK(sb
) = (old_format
?
2750 REISERFS_OLD_DISK_OFFSET_IN_BYTES
2752 reiserfs_bmap_count(sb
) +
2754 REISERFS_DISK_OFFSET_IN_BYTES
/
2755 sb
->s_blocksize
+ 2);
2758 * Sanity check to see is the standard journal fitting
2759 * within first bitmap (actual for small blocksizes)
2761 if (!SB_ONDISK_JOURNAL_DEVICE(sb
) &&
2762 (SB_JOURNAL_1st_RESERVED_BLOCK(sb
) +
2763 SB_ONDISK_JOURNAL_SIZE(sb
) > sb
->s_blocksize
* 8)) {
2764 reiserfs_warning(sb
, "journal-1393",
2765 "journal does not fit for area addressed "
2766 "by first of bitmap blocks. It starts at "
2767 "%u and its size is %u. Block size %ld",
2768 SB_JOURNAL_1st_RESERVED_BLOCK(sb
),
2769 SB_ONDISK_JOURNAL_SIZE(sb
),
2771 goto free_and_return
;
2774 if (journal_init_dev(sb
, journal
, j_dev_name
) != 0) {
2775 reiserfs_warning(sb
, "sh-462",
2776 "unable to initialize journal device");
2777 goto free_and_return
;
2780 rs
= SB_DISK_SUPER_BLOCK(sb
);
2782 /* read journal header */
2783 bhjh
= journal_bread(sb
,
2784 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2785 SB_ONDISK_JOURNAL_SIZE(sb
));
2787 reiserfs_warning(sb
, "sh-459",
2788 "unable to read journal header");
2789 goto free_and_return
;
2791 jh
= (struct reiserfs_journal_header
*)(bhjh
->b_data
);
2793 /* make sure that journal matches to the super block */
2794 if (is_reiserfs_jr(rs
)
2795 && (le32_to_cpu(jh
->jh_journal
.jp_journal_magic
) !=
2796 sb_jp_journal_magic(rs
))) {
2797 reiserfs_warning(sb
, "sh-460",
2798 "journal header magic %x (device %pg) does "
2799 "not match to magic found in super block %x",
2800 jh
->jh_journal
.jp_journal_magic
,
2802 sb_jp_journal_magic(rs
));
2804 goto free_and_return
;
2807 journal
->j_trans_max
= le32_to_cpu(jh
->jh_journal
.jp_journal_trans_max
);
2808 journal
->j_max_batch
= le32_to_cpu(jh
->jh_journal
.jp_journal_max_batch
);
2809 journal
->j_max_commit_age
=
2810 le32_to_cpu(jh
->jh_journal
.jp_journal_max_commit_age
);
2811 journal
->j_max_trans_age
= JOURNAL_MAX_TRANS_AGE
;
2813 if (check_advise_trans_params(sb
, journal
) != 0)
2814 goto free_and_return
;
2815 journal
->j_default_max_commit_age
= journal
->j_max_commit_age
;
2817 if (commit_max_age
!= 0) {
2818 journal
->j_max_commit_age
= commit_max_age
;
2819 journal
->j_max_trans_age
= commit_max_age
;
2822 reiserfs_info(sb
, "journal params: device %pg, size %u, "
2823 "journal first block %u, max trans len %u, max batch %u, "
2824 "max commit age %u, max trans age %u\n",
2826 SB_ONDISK_JOURNAL_SIZE(sb
),
2827 SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2828 journal
->j_trans_max
,
2829 journal
->j_max_batch
,
2830 journal
->j_max_commit_age
, journal
->j_max_trans_age
);
2834 journal
->j_list_bitmap_index
= 0;
2835 journal_list_init(sb
);
2837 memset(journal
->j_list_hash_table
, 0,
2838 JOURNAL_HASH_SIZE
* sizeof(struct reiserfs_journal_cnode
*));
2840 INIT_LIST_HEAD(&journal
->j_dirty_buffers
);
2841 spin_lock_init(&journal
->j_dirty_buffers_lock
);
2843 journal
->j_start
= 0;
2845 journal
->j_len_alloc
= 0;
2846 atomic_set(&journal
->j_wcount
, 0);
2847 atomic_set(&journal
->j_async_throttle
, 0);
2848 journal
->j_bcount
= 0;
2849 journal
->j_trans_start_time
= 0;
2850 journal
->j_last
= NULL
;
2851 journal
->j_first
= NULL
;
2852 init_waitqueue_head(&journal
->j_join_wait
);
2853 mutex_init(&journal
->j_mutex
);
2854 mutex_init(&journal
->j_flush_mutex
);
2856 journal
->j_trans_id
= 10;
2857 journal
->j_mount_id
= 10;
2858 journal
->j_state
= 0;
2859 atomic_set(&journal
->j_jlock
, 0);
2860 journal
->j_cnode_free_list
= allocate_cnodes(num_cnodes
);
2861 journal
->j_cnode_free_orig
= journal
->j_cnode_free_list
;
2862 journal
->j_cnode_free
= journal
->j_cnode_free_list
? num_cnodes
: 0;
2863 journal
->j_cnode_used
= 0;
2864 journal
->j_must_wait
= 0;
2866 if (journal
->j_cnode_free
== 0) {
2867 reiserfs_warning(sb
, "journal-2004", "Journal cnode memory "
2868 "allocation failed (%ld bytes). Journal is "
2869 "too large for available memory. Usually "
2870 "this is due to a journal that is too large.",
2871 sizeof (struct reiserfs_journal_cnode
) * num_cnodes
);
2872 goto free_and_return
;
2875 init_journal_hash(sb
);
2876 jl
= journal
->j_current_jl
;
2879 * get_list_bitmap() may call flush_commit_list() which
2880 * requires the lock. Calling flush_commit_list() shouldn't happen
2881 * this early but I like to be paranoid.
2883 reiserfs_write_lock(sb
);
2884 jl
->j_list_bitmap
= get_list_bitmap(sb
, jl
);
2885 reiserfs_write_unlock(sb
);
2886 if (!jl
->j_list_bitmap
) {
2887 reiserfs_warning(sb
, "journal-2005",
2888 "get_list_bitmap failed for journal list 0");
2889 goto free_and_return
;
2892 ret
= journal_read(sb
);
2894 reiserfs_warning(sb
, "reiserfs-2006",
2895 "Replay Failure, unable to mount");
2896 goto free_and_return
;
2899 INIT_DELAYED_WORK(&journal
->j_work
, flush_async_commits
);
2900 journal
->j_work_sb
= sb
;
2903 free_journal_ram(sb
);
2908 * test for a polite end of the current transaction. Used by file_write,
2909 * and should be used by delete to make sure they don't write more than
2910 * can fit inside a single transaction
2912 int journal_transaction_should_end(struct reiserfs_transaction_handle
*th
,
2915 struct reiserfs_journal
*journal
= SB_JOURNAL(th
->t_super
);
2916 time_t now
= get_seconds();
2917 /* cannot restart while nested */
2918 BUG_ON(!th
->t_trans_id
);
2919 if (th
->t_refcount
> 1)
2921 if (journal
->j_must_wait
> 0 ||
2922 (journal
->j_len_alloc
+ new_alloc
) >= journal
->j_max_batch
||
2923 atomic_read(&journal
->j_jlock
) ||
2924 (now
- journal
->j_trans_start_time
) > journal
->j_max_trans_age
||
2925 journal
->j_cnode_free
< (journal
->j_trans_max
* 3)) {
2929 journal
->j_len_alloc
+= new_alloc
;
2930 th
->t_blocks_allocated
+= new_alloc
;
2934 /* this must be called inside a transaction */
2935 void reiserfs_block_writes(struct reiserfs_transaction_handle
*th
)
2937 struct reiserfs_journal
*journal
= SB_JOURNAL(th
->t_super
);
2938 BUG_ON(!th
->t_trans_id
);
2939 journal
->j_must_wait
= 1;
2940 set_bit(J_WRITERS_BLOCKED
, &journal
->j_state
);
2944 /* this must be called without a transaction started */
2945 void reiserfs_allow_writes(struct super_block
*s
)
2947 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2948 clear_bit(J_WRITERS_BLOCKED
, &journal
->j_state
);
2949 wake_up(&journal
->j_join_wait
);
2952 /* this must be called without a transaction started */
2953 void reiserfs_wait_on_write_block(struct super_block
*s
)
2955 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2956 wait_event(journal
->j_join_wait
,
2957 !test_bit(J_WRITERS_BLOCKED
, &journal
->j_state
));
2960 static void queue_log_writer(struct super_block
*s
)
2962 wait_queue_entry_t wait
;
2963 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2964 set_bit(J_WRITERS_QUEUED
, &journal
->j_state
);
2967 * we don't want to use wait_event here because
2968 * we only want to wait once.
2970 init_waitqueue_entry(&wait
, current
);
2971 add_wait_queue(&journal
->j_join_wait
, &wait
);
2972 set_current_state(TASK_UNINTERRUPTIBLE
);
2973 if (test_bit(J_WRITERS_QUEUED
, &journal
->j_state
)) {
2974 int depth
= reiserfs_write_unlock_nested(s
);
2976 reiserfs_write_lock_nested(s
, depth
);
2978 __set_current_state(TASK_RUNNING
);
2979 remove_wait_queue(&journal
->j_join_wait
, &wait
);
2982 static void wake_queued_writers(struct super_block
*s
)
2984 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2985 if (test_and_clear_bit(J_WRITERS_QUEUED
, &journal
->j_state
))
2986 wake_up(&journal
->j_join_wait
);
2989 static void let_transaction_grow(struct super_block
*sb
, unsigned int trans_id
)
2991 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
2992 unsigned long bcount
= journal
->j_bcount
;
2996 depth
= reiserfs_write_unlock_nested(sb
);
2997 schedule_timeout_uninterruptible(1);
2998 reiserfs_write_lock_nested(sb
, depth
);
3000 journal
->j_current_jl
->j_state
|= LIST_COMMIT_PENDING
;
3001 while ((atomic_read(&journal
->j_wcount
) > 0 ||
3002 atomic_read(&journal
->j_jlock
)) &&
3003 journal
->j_trans_id
== trans_id
) {
3004 queue_log_writer(sb
);
3006 if (journal
->j_trans_id
!= trans_id
)
3008 if (bcount
== journal
->j_bcount
)
3010 bcount
= journal
->j_bcount
;
3015 * join == true if you must join an existing transaction.
3016 * join == false if you can deal with waiting for others to finish
3018 * this will block until the transaction is joinable. send the number of
3019 * blocks you expect to use in nblocks.
3021 static int do_journal_begin_r(struct reiserfs_transaction_handle
*th
,
3022 struct super_block
*sb
, unsigned long nblocks
,
3025 time_t now
= get_seconds();
3026 unsigned int old_trans_id
;
3027 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3028 struct reiserfs_transaction_handle myth
;
3029 int sched_count
= 0;
3033 reiserfs_check_lock_depth(sb
, "journal_begin");
3034 BUG_ON(nblocks
> journal
->j_trans_max
);
3036 PROC_INFO_INC(sb
, journal
.journal_being
);
3037 /* set here for journal_join */
3043 if (join
!= JBEGIN_ABORT
&& reiserfs_is_journal_aborted(journal
)) {
3045 retval
= journal
->j_errno
;
3048 journal
->j_bcount
++;
3050 if (test_bit(J_WRITERS_BLOCKED
, &journal
->j_state
)) {
3052 depth
= reiserfs_write_unlock_nested(sb
);
3053 reiserfs_wait_on_write_block(sb
);
3054 reiserfs_write_lock_nested(sb
, depth
);
3055 PROC_INFO_INC(sb
, journal
.journal_relock_writers
);
3058 now
= get_seconds();
3061 * if there is no room in the journal OR
3062 * if this transaction is too old, and we weren't called joinable,
3063 * wait for it to finish before beginning we don't sleep if there
3064 * aren't other writers
3067 if ((!join
&& journal
->j_must_wait
> 0) ||
3069 && (journal
->j_len_alloc
+ nblocks
+ 2) >= journal
->j_max_batch
)
3070 || (!join
&& atomic_read(&journal
->j_wcount
) > 0
3071 && journal
->j_trans_start_time
> 0
3072 && (now
- journal
->j_trans_start_time
) >
3073 journal
->j_max_trans_age
) || (!join
3074 && atomic_read(&journal
->j_jlock
))
3075 || (!join
&& journal
->j_cnode_free
< (journal
->j_trans_max
* 3))) {
3077 old_trans_id
= journal
->j_trans_id
;
3078 /* allow others to finish this transaction */
3081 if (!join
&& (journal
->j_len_alloc
+ nblocks
+ 2) >=
3082 journal
->j_max_batch
&&
3083 ((journal
->j_len
+ nblocks
+ 2) * 100) <
3084 (journal
->j_len_alloc
* 75)) {
3085 if (atomic_read(&journal
->j_wcount
) > 10) {
3087 queue_log_writer(sb
);
3092 * don't mess with joining the transaction if all we
3093 * have to do is wait for someone else to do a commit
3095 if (atomic_read(&journal
->j_jlock
)) {
3096 while (journal
->j_trans_id
== old_trans_id
&&
3097 atomic_read(&journal
->j_jlock
)) {
3098 queue_log_writer(sb
);
3102 retval
= journal_join(&myth
, sb
);
3106 /* someone might have ended the transaction while we joined */
3107 if (old_trans_id
!= journal
->j_trans_id
) {
3108 retval
= do_journal_end(&myth
, 0);
3110 retval
= do_journal_end(&myth
, COMMIT_NOW
);
3116 PROC_INFO_INC(sb
, journal
.journal_relock_wcount
);
3119 /* we are the first writer, set trans_id */
3120 if (journal
->j_trans_start_time
== 0) {
3121 journal
->j_trans_start_time
= get_seconds();
3123 atomic_inc(&journal
->j_wcount
);
3124 journal
->j_len_alloc
+= nblocks
;
3125 th
->t_blocks_logged
= 0;
3126 th
->t_blocks_allocated
= nblocks
;
3127 th
->t_trans_id
= journal
->j_trans_id
;
3129 INIT_LIST_HEAD(&th
->t_list
);
3133 memset(th
, 0, sizeof(*th
));
3135 * Re-set th->t_super, so we can properly keep track of how many
3136 * persistent transactions there are. We need to do this so if this
3137 * call is part of a failed restart_transaction, we can free it later
3143 struct reiserfs_transaction_handle
*reiserfs_persistent_transaction(struct
3149 struct reiserfs_transaction_handle
*th
;
3152 * if we're nesting into an existing transaction. It will be
3153 * persistent on its own
3155 if (reiserfs_transaction_running(s
)) {
3156 th
= current
->journal_info
;
3158 BUG_ON(th
->t_refcount
< 2);
3162 th
= kmalloc(sizeof(struct reiserfs_transaction_handle
), GFP_NOFS
);
3165 ret
= journal_begin(th
, s
, nblocks
);
3171 SB_JOURNAL(s
)->j_persistent_trans
++;
3175 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle
*th
)
3177 struct super_block
*s
= th
->t_super
;
3180 ret
= journal_end(th
);
3183 if (th
->t_refcount
== 0) {
3184 SB_JOURNAL(s
)->j_persistent_trans
--;
3190 static int journal_join(struct reiserfs_transaction_handle
*th
,
3191 struct super_block
*sb
)
3193 struct reiserfs_transaction_handle
*cur_th
= current
->journal_info
;
3196 * this keeps do_journal_end from NULLing out the
3197 * current->journal_info pointer
3199 th
->t_handle_save
= cur_th
;
3200 BUG_ON(cur_th
&& cur_th
->t_refcount
> 1);
3201 return do_journal_begin_r(th
, sb
, 1, JBEGIN_JOIN
);
3204 int journal_join_abort(struct reiserfs_transaction_handle
*th
,
3205 struct super_block
*sb
)
3207 struct reiserfs_transaction_handle
*cur_th
= current
->journal_info
;
3210 * this keeps do_journal_end from NULLing out the
3211 * current->journal_info pointer
3213 th
->t_handle_save
= cur_th
;
3214 BUG_ON(cur_th
&& cur_th
->t_refcount
> 1);
3215 return do_journal_begin_r(th
, sb
, 1, JBEGIN_ABORT
);
3218 int journal_begin(struct reiserfs_transaction_handle
*th
,
3219 struct super_block
*sb
, unsigned long nblocks
)
3221 struct reiserfs_transaction_handle
*cur_th
= current
->journal_info
;
3224 th
->t_handle_save
= NULL
;
3226 /* we are nesting into the current transaction */
3227 if (cur_th
->t_super
== sb
) {
3228 BUG_ON(!cur_th
->t_refcount
);
3229 cur_th
->t_refcount
++;
3230 memcpy(th
, cur_th
, sizeof(*th
));
3231 if (th
->t_refcount
<= 1)
3232 reiserfs_warning(sb
, "reiserfs-2005",
3233 "BAD: refcount <= 1, but "
3234 "journal_info != 0");
3238 * we've ended up with a handle from a different
3239 * filesystem. save it and restore on journal_end.
3240 * This should never really happen...
3242 reiserfs_warning(sb
, "clm-2100",
3243 "nesting info a different FS");
3244 th
->t_handle_save
= current
->journal_info
;
3245 current
->journal_info
= th
;
3248 current
->journal_info
= th
;
3250 ret
= do_journal_begin_r(th
, sb
, nblocks
, JBEGIN_REG
);
3251 BUG_ON(current
->journal_info
!= th
);
3254 * I guess this boils down to being the reciprocal of clm-2100 above.
3255 * If do_journal_begin_r fails, we need to put it back, since
3256 * journal_end won't be called to do it. */
3258 current
->journal_info
= th
->t_handle_save
;
3260 BUG_ON(!th
->t_refcount
);
3266 * puts bh into the current transaction. If it was already there, reorders
3267 * removes the old pointers from the hash, and puts new ones in (to make
3268 * sure replay happen in the right order).
3270 * if it was dirty, cleans and files onto the clean list. I can't let it
3271 * be dirty again until the transaction is committed.
3273 * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3275 int journal_mark_dirty(struct reiserfs_transaction_handle
*th
,
3276 struct buffer_head
*bh
)
3278 struct super_block
*sb
= th
->t_super
;
3279 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3280 struct reiserfs_journal_cnode
*cn
= NULL
;
3281 int count_already_incd
= 0;
3283 BUG_ON(!th
->t_trans_id
);
3285 PROC_INFO_INC(sb
, journal
.mark_dirty
);
3286 if (th
->t_trans_id
!= journal
->j_trans_id
) {
3287 reiserfs_panic(th
->t_super
, "journal-1577",
3288 "handle trans id %ld != current trans id %ld",
3289 th
->t_trans_id
, journal
->j_trans_id
);
3292 prepared
= test_clear_buffer_journal_prepared(bh
);
3293 clear_buffer_journal_restore_dirty(bh
);
3294 /* already in this transaction, we are done */
3295 if (buffer_journaled(bh
)) {
3296 PROC_INFO_INC(sb
, journal
.mark_dirty_already
);
3301 * this must be turned into a panic instead of a warning. We can't
3302 * allow a dirty or journal_dirty or locked buffer to be logged, as
3303 * some changes could get to disk too early. NOT GOOD.
3305 if (!prepared
|| buffer_dirty(bh
)) {
3306 reiserfs_warning(sb
, "journal-1777",
3307 "buffer %llu bad state "
3308 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3309 (unsigned long long)bh
->b_blocknr
,
3310 prepared
? ' ' : '!',
3311 buffer_locked(bh
) ? ' ' : '!',
3312 buffer_dirty(bh
) ? ' ' : '!',
3313 buffer_journal_dirty(bh
) ? ' ' : '!');
3316 if (atomic_read(&journal
->j_wcount
) <= 0) {
3317 reiserfs_warning(sb
, "journal-1409",
3318 "returning because j_wcount was %d",
3319 atomic_read(&journal
->j_wcount
));
3323 * this error means I've screwed up, and we've overflowed
3324 * the transaction. Nothing can be done here, except make the
3325 * FS readonly or panic.
3327 if (journal
->j_len
>= journal
->j_trans_max
) {
3328 reiserfs_panic(th
->t_super
, "journal-1413",
3329 "j_len (%lu) is too big",
3333 if (buffer_journal_dirty(bh
)) {
3334 count_already_incd
= 1;
3335 PROC_INFO_INC(sb
, journal
.mark_dirty_notjournal
);
3336 clear_buffer_journal_dirty(bh
);
3339 if (journal
->j_len
> journal
->j_len_alloc
) {
3340 journal
->j_len_alloc
= journal
->j_len
+ JOURNAL_PER_BALANCE_CNT
;
3343 set_buffer_journaled(bh
);
3345 /* now put this guy on the end */
3349 reiserfs_panic(sb
, "journal-4", "get_cnode failed!");
3352 if (th
->t_blocks_logged
== th
->t_blocks_allocated
) {
3353 th
->t_blocks_allocated
+= JOURNAL_PER_BALANCE_CNT
;
3354 journal
->j_len_alloc
+= JOURNAL_PER_BALANCE_CNT
;
3356 th
->t_blocks_logged
++;
3360 cn
->blocknr
= bh
->b_blocknr
;
3363 insert_journal_hash(journal
->j_hash_table
, cn
);
3364 if (!count_already_incd
) {
3369 cn
->prev
= journal
->j_last
;
3371 if (journal
->j_last
) {
3372 journal
->j_last
->next
= cn
;
3373 journal
->j_last
= cn
;
3375 journal
->j_first
= cn
;
3376 journal
->j_last
= cn
;
3378 reiserfs_schedule_old_flush(sb
);
3382 int journal_end(struct reiserfs_transaction_handle
*th
)
3384 struct super_block
*sb
= th
->t_super
;
3385 if (!current
->journal_info
&& th
->t_refcount
> 1)
3386 reiserfs_warning(sb
, "REISER-NESTING",
3387 "th NULL, refcount %d", th
->t_refcount
);
3389 if (!th
->t_trans_id
) {
3395 if (th
->t_refcount
> 0) {
3396 struct reiserfs_transaction_handle
*cur_th
=
3397 current
->journal_info
;
3400 * we aren't allowed to close a nested transaction on a
3401 * different filesystem from the one in the task struct
3403 BUG_ON(cur_th
->t_super
!= th
->t_super
);
3406 memcpy(current
->journal_info
, th
, sizeof(*th
));
3411 return do_journal_end(th
, 0);
3416 * removes from the current transaction, relsing and descrementing any counters.
3417 * also files the removed buffer directly onto the clean list
3419 * called by journal_mark_freed when a block has been deleted
3421 * returns 1 if it cleaned and relsed the buffer. 0 otherwise
3423 static int remove_from_transaction(struct super_block
*sb
,
3424 b_blocknr_t blocknr
, int already_cleaned
)
3426 struct buffer_head
*bh
;
3427 struct reiserfs_journal_cnode
*cn
;
3428 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3431 cn
= get_journal_hash_dev(sb
, journal
->j_hash_table
, blocknr
);
3432 if (!cn
|| !cn
->bh
) {
3437 cn
->prev
->next
= cn
->next
;
3440 cn
->next
->prev
= cn
->prev
;
3442 if (cn
== journal
->j_first
) {
3443 journal
->j_first
= cn
->next
;
3445 if (cn
== journal
->j_last
) {
3446 journal
->j_last
= cn
->prev
;
3449 remove_journal_hash(sb
, journal
->j_hash_table
, NULL
,
3451 clear_buffer_journaled(bh
); /* don't log this one */
3453 if (!already_cleaned
) {
3454 clear_buffer_journal_dirty(bh
);
3455 clear_buffer_dirty(bh
);
3456 clear_buffer_journal_test(bh
);
3458 if (atomic_read(&bh
->b_count
) < 0) {
3459 reiserfs_warning(sb
, "journal-1752",
3465 journal
->j_len_alloc
--;
3471 * for any cnode in a journal list, it can only be dirtied of all the
3472 * transactions that include it are committed to disk.
3473 * this checks through each transaction, and returns 1 if you are allowed
3474 * to dirty, and 0 if you aren't
3476 * it is called by dirty_journal_list, which is called after
3477 * flush_commit_list has gotten all the log blocks for a given
3478 * transaction on disk
3481 static int can_dirty(struct reiserfs_journal_cnode
*cn
)
3483 struct super_block
*sb
= cn
->sb
;
3484 b_blocknr_t blocknr
= cn
->blocknr
;
3485 struct reiserfs_journal_cnode
*cur
= cn
->hprev
;
3489 * first test hprev. These are all newer than cn, so any node here
3490 * with the same block number and dev means this node can't be sent
3491 * to disk right now.
3493 while (cur
&& can_dirty
) {
3494 if (cur
->jlist
&& cur
->bh
&& cur
->blocknr
&& cur
->sb
== sb
&&
3495 cur
->blocknr
== blocknr
) {
3501 * then test hnext. These are all older than cn. As long as they
3502 * are committed to the log, it is safe to write cn to disk
3505 while (cur
&& can_dirty
) {
3506 if (cur
->jlist
&& cur
->jlist
->j_len
> 0 &&
3507 atomic_read(&cur
->jlist
->j_commit_left
) > 0 && cur
->bh
&&
3508 cur
->blocknr
&& cur
->sb
== sb
&& cur
->blocknr
== blocknr
) {
3517 * syncs the commit blocks, but does not force the real buffers to disk
3518 * will wait until the current transaction is done/committed before returning
3520 int journal_end_sync(struct reiserfs_transaction_handle
*th
)
3522 struct super_block
*sb
= th
->t_super
;
3523 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3525 BUG_ON(!th
->t_trans_id
);
3526 /* you can sync while nested, very, very bad */
3527 BUG_ON(th
->t_refcount
> 1);
3528 if (journal
->j_len
== 0) {
3529 reiserfs_prepare_for_journal(sb
, SB_BUFFER_WITH_SB(sb
),
3531 journal_mark_dirty(th
, SB_BUFFER_WITH_SB(sb
));
3533 return do_journal_end(th
, COMMIT_NOW
| WAIT
);
3536 /* writeback the pending async commits to disk */
3537 static void flush_async_commits(struct work_struct
*work
)
3539 struct reiserfs_journal
*journal
=
3540 container_of(work
, struct reiserfs_journal
, j_work
.work
);
3541 struct super_block
*sb
= journal
->j_work_sb
;
3542 struct reiserfs_journal_list
*jl
;
3543 struct list_head
*entry
;
3545 reiserfs_write_lock(sb
);
3546 if (!list_empty(&journal
->j_journal_list
)) {
3547 /* last entry is the youngest, commit it and you get everything */
3548 entry
= journal
->j_journal_list
.prev
;
3549 jl
= JOURNAL_LIST_ENTRY(entry
);
3550 flush_commit_list(sb
, jl
, 1);
3552 reiserfs_write_unlock(sb
);
3556 * flushes any old transactions to disk
3557 * ends the current transaction if it is too old
3559 void reiserfs_flush_old_commits(struct super_block
*sb
)
3562 struct reiserfs_transaction_handle th
;
3563 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3565 now
= get_seconds();
3567 * safety check so we don't flush while we are replaying the log during
3570 if (list_empty(&journal
->j_journal_list
))
3574 * check the current transaction. If there are no writers, and it is
3575 * too old, finish it, and force the commit blocks to disk
3577 if (atomic_read(&journal
->j_wcount
) <= 0 &&
3578 journal
->j_trans_start_time
> 0 &&
3579 journal
->j_len
> 0 &&
3580 (now
- journal
->j_trans_start_time
) > journal
->j_max_trans_age
) {
3581 if (!journal_join(&th
, sb
)) {
3582 reiserfs_prepare_for_journal(sb
,
3583 SB_BUFFER_WITH_SB(sb
),
3585 journal_mark_dirty(&th
, SB_BUFFER_WITH_SB(sb
));
3588 * we're only being called from kreiserfsd, it makes
3589 * no sense to do an async commit so that kreiserfsd
3592 do_journal_end(&th
, COMMIT_NOW
| WAIT
);
3598 * returns 0 if do_journal_end should return right away, returns 1 if
3599 * do_journal_end should finish the commit
3601 * if the current transaction is too old, but still has writers, this will
3602 * wait on j_join_wait until all the writers are done. By the time it
3603 * wakes up, the transaction it was called has already ended, so it just
3604 * flushes the commit list and returns 0.
3606 * Won't batch when flush or commit_now is set. Also won't batch when
3607 * others are waiting on j_join_wait.
3609 * Note, we can't allow the journal_end to proceed while there are still
3610 * writers in the log.
3612 static int check_journal_end(struct reiserfs_transaction_handle
*th
, int flags
)
3616 int flush
= flags
& FLUSH_ALL
;
3617 int commit_now
= flags
& COMMIT_NOW
;
3618 int wait_on_commit
= flags
& WAIT
;
3619 struct reiserfs_journal_list
*jl
;
3620 struct super_block
*sb
= th
->t_super
;
3621 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3623 BUG_ON(!th
->t_trans_id
);
3625 if (th
->t_trans_id
!= journal
->j_trans_id
) {
3626 reiserfs_panic(th
->t_super
, "journal-1577",
3627 "handle trans id %ld != current trans id %ld",
3628 th
->t_trans_id
, journal
->j_trans_id
);
3631 journal
->j_len_alloc
-= (th
->t_blocks_allocated
- th
->t_blocks_logged
);
3632 /* <= 0 is allowed. unmounting might not call begin */
3633 if (atomic_read(&journal
->j_wcount
) > 0)
3634 atomic_dec(&journal
->j_wcount
);
3637 * BUG, deal with case where j_len is 0, but people previously
3638 * freed blocks need to be released will be dealt with by next
3639 * transaction that actually writes something, but should be taken
3640 * care of in this trans
3642 BUG_ON(journal
->j_len
== 0);
3645 * if wcount > 0, and we are called to with flush or commit_now,
3646 * we wait on j_join_wait. We will wake up when the last writer has
3647 * finished the transaction, and started it on its way to the disk.
3648 * Then, we flush the commit or journal list, and just return 0
3649 * because the rest of journal end was already done for this
3652 if (atomic_read(&journal
->j_wcount
) > 0) {
3653 if (flush
|| commit_now
) {
3656 jl
= journal
->j_current_jl
;
3657 trans_id
= jl
->j_trans_id
;
3659 jl
->j_state
|= LIST_COMMIT_PENDING
;
3660 atomic_set(&journal
->j_jlock
, 1);
3662 journal
->j_next_full_flush
= 1;
3667 * sleep while the current transaction is
3670 while (journal
->j_trans_id
== trans_id
) {
3671 if (atomic_read(&journal
->j_jlock
)) {
3672 queue_log_writer(sb
);
3675 if (journal
->j_trans_id
== trans_id
) {
3676 atomic_set(&journal
->j_jlock
,
3682 BUG_ON(journal
->j_trans_id
== trans_id
);
3685 && journal_list_still_alive(sb
, trans_id
)
3686 && wait_on_commit
) {
3687 flush_commit_list(sb
, jl
, 1);
3695 /* deal with old transactions where we are the last writers */
3696 now
= get_seconds();
3697 if ((now
- journal
->j_trans_start_time
) > journal
->j_max_trans_age
) {
3699 journal
->j_next_async_flush
= 1;
3701 /* don't batch when someone is waiting on j_join_wait */
3702 /* don't batch when syncing the commit or flushing the whole trans */
3703 if (!(journal
->j_must_wait
> 0) && !(atomic_read(&journal
->j_jlock
))
3704 && !flush
&& !commit_now
&& (journal
->j_len
< journal
->j_max_batch
)
3705 && journal
->j_len_alloc
< journal
->j_max_batch
3706 && journal
->j_cnode_free
> (journal
->j_trans_max
* 3)) {
3707 journal
->j_bcount
++;
3712 if (journal
->j_start
> SB_ONDISK_JOURNAL_SIZE(sb
)) {
3713 reiserfs_panic(sb
, "journal-003",
3714 "j_start (%ld) is too high",
3721 * Does all the work that makes deleting blocks safe.
3722 * when deleting a block mark BH_JNew, just remove it from the current
3723 * transaction, clean it's buffer_head and move on.
3726 * set a bit for the block in the journal bitmap. That will prevent it from
3727 * being allocated for unformatted nodes before this transaction has finished.
3729 * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.
3730 * That will prevent any old transactions with this block from trying to flush
3731 * to the real location. Since we aren't removing the cnode from the
3732 * journal_list_hash, *the block can't be reallocated yet.
3734 * Then remove it from the current transaction, decrementing any counters and
3735 * filing it on the clean list.
3737 int journal_mark_freed(struct reiserfs_transaction_handle
*th
,
3738 struct super_block
*sb
, b_blocknr_t blocknr
)
3740 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3741 struct reiserfs_journal_cnode
*cn
= NULL
;
3742 struct buffer_head
*bh
= NULL
;
3743 struct reiserfs_list_bitmap
*jb
= NULL
;
3745 BUG_ON(!th
->t_trans_id
);
3747 cn
= get_journal_hash_dev(sb
, journal
->j_hash_table
, blocknr
);
3752 /* if it is journal new, we just remove it from this transaction */
3753 if (bh
&& buffer_journal_new(bh
)) {
3754 clear_buffer_journal_new(bh
);
3755 clear_prepared_bits(bh
);
3756 reiserfs_clean_and_file_buffer(bh
);
3757 cleaned
= remove_from_transaction(sb
, blocknr
, cleaned
);
3760 * set the bit for this block in the journal bitmap
3761 * for this transaction
3763 jb
= journal
->j_current_jl
->j_list_bitmap
;
3765 reiserfs_panic(sb
, "journal-1702",
3766 "journal_list_bitmap is NULL");
3768 set_bit_in_list_bitmap(sb
, blocknr
, jb
);
3770 /* Note, the entire while loop is not allowed to schedule. */
3773 clear_prepared_bits(bh
);
3774 reiserfs_clean_and_file_buffer(bh
);
3776 cleaned
= remove_from_transaction(sb
, blocknr
, cleaned
);
3779 * find all older transactions with this block,
3780 * make sure they don't try to write it out
3782 cn
= get_journal_hash_dev(sb
, journal
->j_list_hash_table
,
3785 if (sb
== cn
->sb
&& blocknr
== cn
->blocknr
) {
3786 set_bit(BLOCK_FREED
, &cn
->state
);
3789 * remove_from_transaction will brelse
3790 * the buffer if it was in the current
3794 clear_buffer_journal_dirty(cn
->
3796 clear_buffer_dirty(cn
->bh
);
3797 clear_buffer_journal_test(cn
->
3802 (&cn
->bh
->b_count
) < 0) {
3803 reiserfs_warning(sb
,
3805 "cn->bh->b_count < 0");
3809 * since we are clearing the bh,
3810 * we MUST dec nonzerolen
3813 atomic_dec(&cn
->jlist
->
3824 release_buffer_page(bh
); /* get_hash grabs the buffer */
3828 void reiserfs_update_inode_transaction(struct inode
*inode
)
3830 struct reiserfs_journal
*journal
= SB_JOURNAL(inode
->i_sb
);
3831 REISERFS_I(inode
)->i_jl
= journal
->j_current_jl
;
3832 REISERFS_I(inode
)->i_trans_id
= journal
->j_trans_id
;
3836 * returns -1 on error, 0 if no commits/barriers were done and 1
3837 * if a transaction was actually committed and the barrier was done
3839 static int __commit_trans_jl(struct inode
*inode
, unsigned long id
,
3840 struct reiserfs_journal_list
*jl
)
3842 struct reiserfs_transaction_handle th
;
3843 struct super_block
*sb
= inode
->i_sb
;
3844 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3848 * is it from the current transaction,
3849 * or from an unknown transaction?
3851 if (id
== journal
->j_trans_id
) {
3852 jl
= journal
->j_current_jl
;
3854 * try to let other writers come in and
3855 * grow this transaction
3857 let_transaction_grow(sb
, id
);
3858 if (journal
->j_trans_id
!= id
) {
3859 goto flush_commit_only
;
3862 ret
= journal_begin(&th
, sb
, 1);
3866 /* someone might have ended this transaction while we joined */
3867 if (journal
->j_trans_id
!= id
) {
3868 reiserfs_prepare_for_journal(sb
, SB_BUFFER_WITH_SB(sb
),
3870 journal_mark_dirty(&th
, SB_BUFFER_WITH_SB(sb
));
3871 ret
= journal_end(&th
);
3872 goto flush_commit_only
;
3875 ret
= journal_end_sync(&th
);
3881 * this gets tricky, we have to make sure the journal list in
3882 * the inode still exists. We know the list is still around
3883 * if we've got a larger transaction id than the oldest list
3886 if (journal_list_still_alive(inode
->i_sb
, id
)) {
3888 * we only set ret to 1 when we know for sure
3889 * the barrier hasn't been started yet on the commit
3892 if (atomic_read(&jl
->j_commit_left
) > 1)
3894 flush_commit_list(sb
, jl
, 1);
3895 if (journal
->j_errno
)
3896 ret
= journal
->j_errno
;
3899 /* otherwise the list is gone, and long since committed */
3903 int reiserfs_commit_for_inode(struct inode
*inode
)
3905 unsigned int id
= REISERFS_I(inode
)->i_trans_id
;
3906 struct reiserfs_journal_list
*jl
= REISERFS_I(inode
)->i_jl
;
3909 * for the whole inode, assume unset id means it was
3910 * changed in the current transaction. More conservative
3913 reiserfs_update_inode_transaction(inode
);
3914 id
= REISERFS_I(inode
)->i_trans_id
;
3915 /* jl will be updated in __commit_trans_jl */
3918 return __commit_trans_jl(inode
, id
, jl
);
3921 void reiserfs_restore_prepared_buffer(struct super_block
*sb
,
3922 struct buffer_head
*bh
)
3924 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3925 PROC_INFO_INC(sb
, journal
.restore_prepared
);
3929 if (test_clear_buffer_journal_restore_dirty(bh
) &&
3930 buffer_journal_dirty(bh
)) {
3931 struct reiserfs_journal_cnode
*cn
;
3932 reiserfs_write_lock(sb
);
3933 cn
= get_journal_hash_dev(sb
,
3934 journal
->j_list_hash_table
,
3936 if (cn
&& can_dirty(cn
)) {
3937 set_buffer_journal_test(bh
);
3938 mark_buffer_dirty(bh
);
3940 reiserfs_write_unlock(sb
);
3942 clear_buffer_journal_prepared(bh
);
3945 extern struct tree_balance
*cur_tb
;
3947 * before we can change a metadata block, we have to make sure it won't
3948 * be written to disk while we are altering it. So, we must:
3952 int reiserfs_prepare_for_journal(struct super_block
*sb
,
3953 struct buffer_head
*bh
, int wait
)
3955 PROC_INFO_INC(sb
, journal
.prepare
);
3957 if (!trylock_buffer(bh
)) {
3962 set_buffer_journal_prepared(bh
);
3963 if (test_clear_buffer_dirty(bh
) && buffer_journal_dirty(bh
)) {
3964 clear_buffer_journal_test(bh
);
3965 set_buffer_journal_restore_dirty(bh
);
3972 * long and ugly. If flush, will not return until all commit
3973 * blocks and all real buffers in the trans are on disk.
3974 * If no_async, won't return until all commit blocks are on disk.
3976 * keep reading, there are comments as you go along
3978 * If the journal is aborted, we just clean up. Things like flushing
3979 * journal lists, etc just won't happen.
3981 static int do_journal_end(struct reiserfs_transaction_handle
*th
, int flags
)
3983 struct super_block
*sb
= th
->t_super
;
3984 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3985 struct reiserfs_journal_cnode
*cn
, *next
, *jl_cn
;
3986 struct reiserfs_journal_cnode
*last_cn
= NULL
;
3987 struct reiserfs_journal_desc
*desc
;
3988 struct reiserfs_journal_commit
*commit
;
3989 struct buffer_head
*c_bh
; /* commit bh */
3990 struct buffer_head
*d_bh
; /* desc bh */
3991 int cur_write_start
= 0; /* start index of current log write */
3996 struct reiserfs_journal_list
*jl
, *temp_jl
;
3997 struct list_head
*entry
, *safe
;
3998 unsigned long jindex
;
3999 unsigned int commit_trans_id
;
4003 BUG_ON(th
->t_refcount
> 1);
4004 BUG_ON(!th
->t_trans_id
);
4005 BUG_ON(!th
->t_super
);
4008 * protect flush_older_commits from doing mistakes if the
4009 * transaction ID counter gets overflowed.
4011 if (th
->t_trans_id
== ~0U)
4012 flags
|= FLUSH_ALL
| COMMIT_NOW
| WAIT
;
4013 flush
= flags
& FLUSH_ALL
;
4014 wait_on_commit
= flags
& WAIT
;
4016 current
->journal_info
= th
->t_handle_save
;
4017 reiserfs_check_lock_depth(sb
, "journal end");
4018 if (journal
->j_len
== 0) {
4019 reiserfs_prepare_for_journal(sb
, SB_BUFFER_WITH_SB(sb
),
4021 journal_mark_dirty(th
, SB_BUFFER_WITH_SB(sb
));
4025 if (journal
->j_next_full_flush
) {
4029 if (journal
->j_next_async_flush
) {
4030 flags
|= COMMIT_NOW
| WAIT
;
4035 * check_journal_end locks the journal, and unlocks if it does
4036 * not return 1 it tells us if we should continue with the
4037 * journal_end, or just return
4039 if (!check_journal_end(th
, flags
)) {
4040 reiserfs_schedule_old_flush(sb
);
4041 wake_queued_writers(sb
);
4042 reiserfs_async_progress_wait(sb
);
4046 /* check_journal_end might set these, check again */
4047 if (journal
->j_next_full_flush
) {
4052 * j must wait means we have to flush the log blocks, and the
4053 * real blocks for this transaction
4055 if (journal
->j_must_wait
> 0) {
4058 #ifdef REISERFS_PREALLOCATE
4060 * quota ops might need to nest, setup the journal_info pointer
4061 * for them and raise the refcount so that it is > 0.
4063 current
->journal_info
= th
;
4066 /* it should not involve new blocks into the transaction */
4067 reiserfs_discard_all_prealloc(th
);
4070 current
->journal_info
= th
->t_handle_save
;
4073 /* setup description block */
4076 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
4078 set_buffer_uptodate(d_bh
);
4079 desc
= (struct reiserfs_journal_desc
*)(d_bh
)->b_data
;
4080 memset(d_bh
->b_data
, 0, d_bh
->b_size
);
4081 memcpy(get_journal_desc_magic(d_bh
), JOURNAL_DESC_MAGIC
, 8);
4082 set_desc_trans_id(desc
, journal
->j_trans_id
);
4085 * setup commit block. Don't write (keep it clean too) this one
4086 * until after everyone else is written
4088 c_bh
= journal_getblk(sb
, SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
4089 ((journal
->j_start
+ journal
->j_len
+
4090 1) % SB_ONDISK_JOURNAL_SIZE(sb
)));
4091 commit
= (struct reiserfs_journal_commit
*)c_bh
->b_data
;
4092 memset(c_bh
->b_data
, 0, c_bh
->b_size
);
4093 set_commit_trans_id(commit
, journal
->j_trans_id
);
4094 set_buffer_uptodate(c_bh
);
4096 /* init this journal list */
4097 jl
= journal
->j_current_jl
;
4100 * we lock the commit before doing anything because
4101 * we want to make sure nobody tries to run flush_commit_list until
4102 * the new transaction is fully setup, and we've already flushed the
4105 reiserfs_mutex_lock_safe(&jl
->j_commit_mutex
, sb
);
4107 /* save the transaction id in case we need to commit it later */
4108 commit_trans_id
= jl
->j_trans_id
;
4110 atomic_set(&jl
->j_older_commits_done
, 0);
4111 jl
->j_trans_id
= journal
->j_trans_id
;
4112 jl
->j_timestamp
= journal
->j_trans_start_time
;
4113 jl
->j_commit_bh
= c_bh
;
4114 jl
->j_start
= journal
->j_start
;
4115 jl
->j_len
= journal
->j_len
;
4116 atomic_set(&jl
->j_nonzerolen
, journal
->j_len
);
4117 atomic_set(&jl
->j_commit_left
, journal
->j_len
+ 2);
4118 jl
->j_realblock
= NULL
;
4121 * The ENTIRE FOR LOOP MUST not cause schedule to occur.
4122 * for each real block, add it to the journal list hash,
4123 * copy into real block index array in the commit or desc block
4125 trans_half
= journal_trans_half(sb
->s_blocksize
);
4126 for (i
= 0, cn
= journal
->j_first
; cn
; cn
= cn
->next
, i
++) {
4127 if (buffer_journaled(cn
->bh
)) {
4128 jl_cn
= get_cnode(sb
);
4130 reiserfs_panic(sb
, "journal-1676",
4131 "get_cnode returned NULL");
4134 jl
->j_realblock
= jl_cn
;
4136 jl_cn
->prev
= last_cn
;
4139 last_cn
->next
= jl_cn
;
4143 * make sure the block we are trying to log
4144 * is not a block of journal or reserved area
4146 if (is_block_in_log_or_reserved_area
4147 (sb
, cn
->bh
->b_blocknr
)) {
4148 reiserfs_panic(sb
, "journal-2332",
4149 "Trying to log block %lu, "
4150 "which is a log block",
4153 jl_cn
->blocknr
= cn
->bh
->b_blocknr
;
4158 insert_journal_hash(journal
->j_list_hash_table
, jl_cn
);
4159 if (i
< trans_half
) {
4160 desc
->j_realblock
[i
] =
4161 cpu_to_le32(cn
->bh
->b_blocknr
);
4163 commit
->j_realblock
[i
- trans_half
] =
4164 cpu_to_le32(cn
->bh
->b_blocknr
);
4170 set_desc_trans_len(desc
, journal
->j_len
);
4171 set_desc_mount_id(desc
, journal
->j_mount_id
);
4172 set_desc_trans_id(desc
, journal
->j_trans_id
);
4173 set_commit_trans_len(commit
, journal
->j_len
);
4176 * special check in case all buffers in the journal
4177 * were marked for not logging
4179 BUG_ON(journal
->j_len
== 0);
4182 * we're about to dirty all the log blocks, mark the description block
4183 * dirty now too. Don't mark the commit block dirty until all the
4184 * others are on disk
4186 mark_buffer_dirty(d_bh
);
4189 * first data block is j_start + 1, so add one to
4190 * cur_write_start wherever you use it
4192 cur_write_start
= journal
->j_start
;
4193 cn
= journal
->j_first
;
4194 jindex
= 1; /* start at one so we don't get the desc again */
4196 clear_buffer_journal_new(cn
->bh
);
4197 /* copy all the real blocks into log area. dirty log blocks */
4198 if (buffer_journaled(cn
->bh
)) {
4199 struct buffer_head
*tmp_bh
;
4204 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
4207 SB_ONDISK_JOURNAL_SIZE(sb
)));
4208 set_buffer_uptodate(tmp_bh
);
4209 page
= cn
->bh
->b_page
;
4211 memcpy(tmp_bh
->b_data
,
4212 addr
+ offset_in_page(cn
->bh
->b_data
),
4215 mark_buffer_dirty(tmp_bh
);
4217 set_buffer_journal_dirty(cn
->bh
);
4218 clear_buffer_journaled(cn
->bh
);
4221 * JDirty cleared sometime during transaction.
4222 * don't log this one
4224 reiserfs_warning(sb
, "journal-2048",
4225 "BAD, buffer in journal hash, "
4232 reiserfs_cond_resched(sb
);
4236 * we are done with both the c_bh and d_bh, but
4237 * c_bh must be written after all other commit blocks,
4238 * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4241 journal
->j_current_jl
= alloc_journal_list(sb
);
4243 /* now it is safe to insert this transaction on the main list */
4244 list_add_tail(&jl
->j_list
, &journal
->j_journal_list
);
4245 list_add_tail(&jl
->j_working_list
, &journal
->j_working_list
);
4246 journal
->j_num_work_lists
++;
4248 /* reset journal values for the next transaction */
4249 old_start
= 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
;
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
);
4270 * make sure reiserfs_add_jh sees the new current_jl before we
4271 * write out the tails
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.
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
);
4288 BUG_ON(!list_empty(&jl
->j_tail_bh_list
));
4289 mutex_unlock(&jl
->j_commit_mutex
);
4292 * honor the flush wishes from the caller, simple commits can
4293 * be done outside the journal lock, they are done below
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.
4300 flush_commit_list(sb
, jl
, 1);
4301 flush_journal_list(sb
, jl
, 1);
4302 } else if (!(jl
->j_state
& LIST_COMMIT_PENDING
)) {
4304 * Avoid queueing work when sb is being shut down. Transaction
4305 * will be flushed on journal shutdown.
4307 if (sb
->s_flags
& SB_ACTIVE
)
4308 queue_delayed_work(REISERFS_SB(sb
)->commit_wq
,
4309 &journal
->j_work
, HZ
/ 10);
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.
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) >=
4323 flush_used_journal_lists(sb
, temp_jl
);
4325 } else if ((journal
->j_start
+
4326 journal
->j_trans_max
+ 1) <
4327 SB_ONDISK_JOURNAL_SIZE(sb
)) {
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
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
)) >=
4342 flush_used_journal_lists(sb
, temp_jl
);
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
4356 journal
->j_current_jl
->j_list_bitmap
=
4357 get_list_bitmap(sb
, journal
->j_current_jl
);
4359 if (!(journal
->j_current_jl
->j_list_bitmap
)) {
4360 reiserfs_panic(sb
, "journal-1996",
4361 "could not get a list bitmap");
4364 atomic_set(&journal
->j_jlock
, 0);
4366 /* wake up any body waiting to join. */
4367 clear_bit(J_WRITERS_QUEUED
, &journal
->j_state
);
4368 wake_up(&journal
->j_join_wait
);
4370 if (!flush
&& wait_on_commit
&&
4371 journal_list_still_alive(sb
, commit_trans_id
)) {
4372 flush_commit_list(sb
, jl
, 1);
4375 reiserfs_check_lock_depth(sb
, "journal end2");
4377 memset(th
, 0, sizeof(*th
));
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
4385 return journal
->j_errno
;
4388 /* Send the file system read only and refuse new transactions */
4389 void reiserfs_abort_journal(struct super_block
*sb
, int errno
)
4391 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
4392 if (test_bit(J_ABORTED
, &journal
->j_state
))
4395 if (!journal
->j_errno
)
4396 journal
->j_errno
= errno
;
4398 sb
->s_flags
|= SB_RDONLY
;
4399 set_bit(J_ABORTED
, &journal
->j_state
);
4401 #ifdef CONFIG_REISERFS_CHECK