2 ** Write ahead logging implementation copyright Chris Mason 2000
4 ** The background commits make this code very interelated, and
5 ** overly complex. I need to rethink things a bit....The major players:
7 ** journal_begin -- call with the number of blocks you expect to log.
8 ** If the current transaction is too
9 ** old, it will block until the current transaction is
10 ** finished, and then start a new one.
11 ** Usually, your transaction will get joined in with
12 ** previous ones for speed.
14 ** journal_join -- same as journal_begin, but won't block on the current
15 ** transaction regardless of age. Don't ever call
16 ** this. Ever. There are only two places it should be
17 ** called from, and they are both inside this file.
19 ** journal_mark_dirty -- adds blocks into this transaction. clears any flags
20 ** that might make them get sent to disk
21 ** and then marks them BH_JDirty. Puts the buffer head
22 ** into the current transaction hash.
24 ** journal_end -- if the current transaction is batchable, it does nothing
25 ** otherwise, it could do an async/synchronous commit, or
26 ** a full flush of all log and real blocks in the
29 ** flush_old_commits -- if the current transaction is too old, it is ended and
30 ** commit blocks are sent to disk. Forces commit blocks
31 ** to disk for all backgrounded commits that have been
33 ** -- Note, if you call this as an immediate flush from
34 ** from within kupdate, it will ignore the immediate flag
37 #include <linux/time.h>
38 #include <linux/semaphore.h>
39 #include <linux/vmalloc.h>
40 #include <linux/reiserfs_fs.h>
41 #include <linux/kernel.h>
42 #include <linux/errno.h>
43 #include <linux/fcntl.h>
44 #include <linux/stat.h>
45 #include <linux/string.h>
46 #include <linux/smp_lock.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>
54 #include <asm/system.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 /* the number of mounted filesystems. This is used to decide when to
63 ** start and kill the commit workqueue
65 static int reiserfs_mounted_fs_count
;
67 static struct workqueue_struct
*commit_wq
;
69 #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
71 #define BUFNR 64 /*read ahead */
73 /* cnode stat bits. Move these into reiserfs_fs.h */
75 #define BLOCK_FREED 2 /* this block was freed, and can't be written. */
76 #define BLOCK_FREED_HOLDER 3 /* this block was freed during this transaction, and can't be written */
78 #define BLOCK_NEEDS_FLUSH 4 /* used in flush_journal_list */
79 #define BLOCK_DIRTIED 5
81 /* journal list state bits */
82 #define LIST_TOUCHED 1
84 #define LIST_COMMIT_PENDING 4 /* someone will commit this list */
86 /* flags for do_journal_end */
87 #define FLUSH_ALL 1 /* flush commit and real blocks */
88 #define COMMIT_NOW 2 /* end and commit this transaction */
89 #define WAIT 4 /* wait for the log blocks to hit the disk */
91 static int do_journal_end(struct reiserfs_transaction_handle
*,
92 struct super_block
*, unsigned long nblocks
,
94 static int flush_journal_list(struct super_block
*s
,
95 struct reiserfs_journal_list
*jl
, int flushall
);
96 static int flush_commit_list(struct super_block
*s
,
97 struct reiserfs_journal_list
*jl
, int flushall
);
98 static int can_dirty(struct reiserfs_journal_cnode
*cn
);
99 static int journal_join(struct reiserfs_transaction_handle
*th
,
100 struct super_block
*sb
, unsigned long nblocks
);
101 static int release_journal_dev(struct super_block
*super
,
102 struct reiserfs_journal
*journal
);
103 static int dirty_one_transaction(struct super_block
*s
,
104 struct reiserfs_journal_list
*jl
);
105 static void flush_async_commits(struct work_struct
*work
);
106 static void queue_log_writer(struct super_block
*s
);
108 /* values for join in do_journal_begin_r */
110 JBEGIN_REG
= 0, /* regular journal begin */
111 JBEGIN_JOIN
= 1, /* join the running transaction if at all possible */
112 JBEGIN_ABORT
= 2, /* called from cleanup code, ignores aborted flag */
115 static int do_journal_begin_r(struct reiserfs_transaction_handle
*th
,
116 struct super_block
*sb
,
117 unsigned long nblocks
, int join
);
119 static void init_journal_hash(struct super_block
*sb
)
121 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
122 memset(journal
->j_hash_table
, 0,
123 JOURNAL_HASH_SIZE
* sizeof(struct reiserfs_journal_cnode
*));
127 ** clears BH_Dirty and sticks the buffer on the clean list. Called because I can't allow refile_buffer to
128 ** make schedule happen after I've freed a block. Look at remove_from_transaction and journal_mark_freed for
131 static int reiserfs_clean_and_file_buffer(struct buffer_head
*bh
)
134 clear_buffer_dirty(bh
);
135 clear_buffer_journal_test(bh
);
140 static void disable_barrier(struct super_block
*s
)
142 REISERFS_SB(s
)->s_mount_opt
&= ~(1 << REISERFS_BARRIER_FLUSH
);
143 printk("reiserfs: disabling flush barriers on %s\n",
144 reiserfs_bdevname(s
));
147 static struct reiserfs_bitmap_node
*allocate_bitmap_node(struct super_block
150 struct reiserfs_bitmap_node
*bn
;
153 bn
= kmalloc(sizeof(struct reiserfs_bitmap_node
), GFP_NOFS
);
157 bn
->data
= kzalloc(sb
->s_blocksize
, GFP_NOFS
);
163 INIT_LIST_HEAD(&bn
->list
);
167 static struct reiserfs_bitmap_node
*get_bitmap_node(struct super_block
*sb
)
169 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
170 struct reiserfs_bitmap_node
*bn
= NULL
;
171 struct list_head
*entry
= journal
->j_bitmap_nodes
.next
;
173 journal
->j_used_bitmap_nodes
++;
176 if (entry
!= &journal
->j_bitmap_nodes
) {
177 bn
= list_entry(entry
, struct reiserfs_bitmap_node
, list
);
179 memset(bn
->data
, 0, sb
->s_blocksize
);
180 journal
->j_free_bitmap_nodes
--;
183 bn
= allocate_bitmap_node(sb
);
190 static inline void free_bitmap_node(struct super_block
*sb
,
191 struct reiserfs_bitmap_node
*bn
)
193 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
194 journal
->j_used_bitmap_nodes
--;
195 if (journal
->j_free_bitmap_nodes
> REISERFS_MAX_BITMAP_NODES
) {
199 list_add(&bn
->list
, &journal
->j_bitmap_nodes
);
200 journal
->j_free_bitmap_nodes
++;
204 static void allocate_bitmap_nodes(struct super_block
*sb
)
207 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
208 struct reiserfs_bitmap_node
*bn
= NULL
;
209 for (i
= 0; i
< REISERFS_MIN_BITMAP_NODES
; i
++) {
210 bn
= allocate_bitmap_node(sb
);
212 list_add(&bn
->list
, &journal
->j_bitmap_nodes
);
213 journal
->j_free_bitmap_nodes
++;
215 break; /* this is ok, we'll try again when more are needed */
220 static int set_bit_in_list_bitmap(struct super_block
*sb
,
222 struct reiserfs_list_bitmap
*jb
)
224 unsigned int bmap_nr
= block
/ (sb
->s_blocksize
<< 3);
225 unsigned int bit_nr
= block
% (sb
->s_blocksize
<< 3);
227 if (!jb
->bitmaps
[bmap_nr
]) {
228 jb
->bitmaps
[bmap_nr
] = get_bitmap_node(sb
);
230 set_bit(bit_nr
, (unsigned long *)jb
->bitmaps
[bmap_nr
]->data
);
234 static void cleanup_bitmap_list(struct super_block
*sb
,
235 struct reiserfs_list_bitmap
*jb
)
238 if (jb
->bitmaps
== NULL
)
241 for (i
= 0; i
< reiserfs_bmap_count(sb
); i
++) {
242 if (jb
->bitmaps
[i
]) {
243 free_bitmap_node(sb
, jb
->bitmaps
[i
]);
244 jb
->bitmaps
[i
] = NULL
;
250 ** only call this on FS unmount.
252 static int free_list_bitmaps(struct super_block
*sb
,
253 struct reiserfs_list_bitmap
*jb_array
)
256 struct reiserfs_list_bitmap
*jb
;
257 for (i
= 0; i
< JOURNAL_NUM_BITMAPS
; i
++) {
259 jb
->journal_list
= NULL
;
260 cleanup_bitmap_list(sb
, jb
);
267 static int free_bitmap_nodes(struct super_block
*sb
)
269 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
270 struct list_head
*next
= journal
->j_bitmap_nodes
.next
;
271 struct reiserfs_bitmap_node
*bn
;
273 while (next
!= &journal
->j_bitmap_nodes
) {
274 bn
= list_entry(next
, struct reiserfs_bitmap_node
, list
);
278 next
= journal
->j_bitmap_nodes
.next
;
279 journal
->j_free_bitmap_nodes
--;
286 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
287 ** jb_array is the array to be filled in.
289 int reiserfs_allocate_list_bitmaps(struct super_block
*sb
,
290 struct reiserfs_list_bitmap
*jb_array
,
291 unsigned int bmap_nr
)
295 struct reiserfs_list_bitmap
*jb
;
296 int mem
= bmap_nr
* sizeof(struct reiserfs_bitmap_node
*);
298 for (i
= 0; i
< JOURNAL_NUM_BITMAPS
; i
++) {
300 jb
->journal_list
= NULL
;
301 jb
->bitmaps
= vmalloc(mem
);
303 reiserfs_warning(sb
, "clm-2000", "unable to "
304 "allocate bitmaps for journal lists");
308 memset(jb
->bitmaps
, 0, mem
);
311 free_list_bitmaps(sb
, jb_array
);
318 ** find an available list bitmap. If you can't find one, flush a commit list
321 static struct reiserfs_list_bitmap
*get_list_bitmap(struct super_block
*sb
,
322 struct reiserfs_journal_list
326 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
327 struct reiserfs_list_bitmap
*jb
= NULL
;
329 for (j
= 0; j
< (JOURNAL_NUM_BITMAPS
* 3); j
++) {
330 i
= journal
->j_list_bitmap_index
;
331 journal
->j_list_bitmap_index
= (i
+ 1) % JOURNAL_NUM_BITMAPS
;
332 jb
= journal
->j_list_bitmap
+ i
;
333 if (journal
->j_list_bitmap
[i
].journal_list
) {
334 flush_commit_list(sb
,
335 journal
->j_list_bitmap
[i
].
337 if (!journal
->j_list_bitmap
[i
].journal_list
) {
344 if (jb
->journal_list
) { /* double check to make sure if flushed correctly */
347 jb
->journal_list
= jl
;
352 ** allocates a new chunk of X nodes, and links them all together as a list.
353 ** Uses the cnode->next and cnode->prev pointers
354 ** returns NULL on failure
356 static struct reiserfs_journal_cnode
*allocate_cnodes(int num_cnodes
)
358 struct reiserfs_journal_cnode
*head
;
360 if (num_cnodes
<= 0) {
363 head
= vmalloc(num_cnodes
* sizeof(struct reiserfs_journal_cnode
));
367 memset(head
, 0, num_cnodes
* sizeof(struct reiserfs_journal_cnode
));
369 head
[0].next
= head
+ 1;
370 for (i
= 1; i
< num_cnodes
; i
++) {
371 head
[i
].prev
= head
+ (i
- 1);
372 head
[i
].next
= head
+ (i
+ 1); /* if last one, overwrite it after the if */
374 head
[num_cnodes
- 1].next
= NULL
;
379 ** pulls a cnode off the free list, or returns NULL on failure
381 static struct reiserfs_journal_cnode
*get_cnode(struct super_block
*sb
)
383 struct reiserfs_journal_cnode
*cn
;
384 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
386 reiserfs_check_lock_depth(sb
, "get_cnode");
388 if (journal
->j_cnode_free
<= 0) {
391 journal
->j_cnode_used
++;
392 journal
->j_cnode_free
--;
393 cn
= journal
->j_cnode_free_list
;
398 cn
->next
->prev
= NULL
;
400 journal
->j_cnode_free_list
= cn
->next
;
401 memset(cn
, 0, sizeof(struct reiserfs_journal_cnode
));
406 ** returns a cnode to the free list
408 static void free_cnode(struct super_block
*sb
,
409 struct reiserfs_journal_cnode
*cn
)
411 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
413 reiserfs_check_lock_depth(sb
, "free_cnode");
415 journal
->j_cnode_used
--;
416 journal
->j_cnode_free
++;
417 /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
418 cn
->next
= journal
->j_cnode_free_list
;
419 if (journal
->j_cnode_free_list
) {
420 journal
->j_cnode_free_list
->prev
= cn
;
422 cn
->prev
= NULL
; /* not needed with the memset, but I might kill the memset, and forget to do this */
423 journal
->j_cnode_free_list
= cn
;
426 static void clear_prepared_bits(struct buffer_head
*bh
)
428 clear_buffer_journal_prepared(bh
);
429 clear_buffer_journal_restore_dirty(bh
);
432 /* return a cnode with same dev, block number and size in table, or null if not found */
433 static inline struct reiserfs_journal_cnode
*get_journal_hash_dev(struct
437 reiserfs_journal_cnode
441 struct reiserfs_journal_cnode
*cn
;
442 cn
= journal_hash(table
, sb
, bl
);
444 if (cn
->blocknr
== bl
&& cn
->sb
== sb
)
448 return (struct reiserfs_journal_cnode
*)0;
452 ** this actually means 'can this block be reallocated yet?'. If you set search_all, a block can only be allocated
453 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
454 ** being overwritten by a replay after crashing.
456 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction. Since deleting
457 ** a block removes it from the current transaction, this case should never happen. If you don't set search_all, make
458 ** sure you never write the block without logging it.
460 ** next_zero_bit is a suggestion about the next block to try for find_forward.
461 ** when bl is rejected because it is set in a journal list bitmap, we search
462 ** for the next zero bit in the bitmap that rejected bl. Then, we return that
463 ** through next_zero_bit for find_forward to try.
465 ** Just because we return something in next_zero_bit does not mean we won't
466 ** reject it on the next call to reiserfs_in_journal
469 int reiserfs_in_journal(struct super_block
*sb
,
470 unsigned int bmap_nr
, int bit_nr
, int search_all
,
471 b_blocknr_t
* next_zero_bit
)
473 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
474 struct reiserfs_journal_cnode
*cn
;
475 struct reiserfs_list_bitmap
*jb
;
479 *next_zero_bit
= 0; /* always start this at zero. */
481 PROC_INFO_INC(sb
, journal
.in_journal
);
482 /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
483 ** if we crash before the transaction that freed it commits, this transaction won't
484 ** have committed either, and the block will never be written
487 for (i
= 0; i
< JOURNAL_NUM_BITMAPS
; i
++) {
488 PROC_INFO_INC(sb
, journal
.in_journal_bitmap
);
489 jb
= journal
->j_list_bitmap
+ i
;
490 if (jb
->journal_list
&& jb
->bitmaps
[bmap_nr
] &&
492 (unsigned long *)jb
->bitmaps
[bmap_nr
]->
495 find_next_zero_bit((unsigned long *)
496 (jb
->bitmaps
[bmap_nr
]->
498 sb
->s_blocksize
<< 3,
505 bl
= bmap_nr
* (sb
->s_blocksize
<< 3) + bit_nr
;
506 /* is it in any old transactions? */
509 get_journal_hash_dev(sb
, journal
->j_list_hash_table
, bl
))) {
513 /* is it in the current transaction. This should never happen */
514 if ((cn
= get_journal_hash_dev(sb
, journal
->j_hash_table
, bl
))) {
519 PROC_INFO_INC(sb
, journal
.in_journal_reusable
);
524 /* insert cn into table
526 static inline void insert_journal_hash(struct reiserfs_journal_cnode
**table
,
527 struct reiserfs_journal_cnode
*cn
)
529 struct reiserfs_journal_cnode
*cn_orig
;
531 cn_orig
= journal_hash(table
, cn
->sb
, cn
->blocknr
);
537 journal_hash(table
, cn
->sb
, cn
->blocknr
) = cn
;
540 /* lock the current transaction */
541 static inline void lock_journal(struct super_block
*sb
)
543 PROC_INFO_INC(sb
, journal
.lock_journal
);
545 reiserfs_mutex_lock_safe(&SB_JOURNAL(sb
)->j_mutex
, sb
);
548 /* unlock the current transaction */
549 static inline void unlock_journal(struct super_block
*sb
)
551 mutex_unlock(&SB_JOURNAL(sb
)->j_mutex
);
554 static inline void get_journal_list(struct reiserfs_journal_list
*jl
)
559 static inline void put_journal_list(struct super_block
*s
,
560 struct reiserfs_journal_list
*jl
)
562 if (jl
->j_refcount
< 1) {
563 reiserfs_panic(s
, "journal-2", "trans id %u, refcount at %d",
564 jl
->j_trans_id
, jl
->j_refcount
);
566 if (--jl
->j_refcount
== 0)
571 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
572 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
575 static void cleanup_freed_for_journal_list(struct super_block
*sb
,
576 struct reiserfs_journal_list
*jl
)
579 struct reiserfs_list_bitmap
*jb
= jl
->j_list_bitmap
;
581 cleanup_bitmap_list(sb
, jb
);
583 jl
->j_list_bitmap
->journal_list
= NULL
;
584 jl
->j_list_bitmap
= NULL
;
587 static int journal_list_still_alive(struct super_block
*s
,
588 unsigned int trans_id
)
590 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
591 struct list_head
*entry
= &journal
->j_journal_list
;
592 struct reiserfs_journal_list
*jl
;
594 if (!list_empty(entry
)) {
595 jl
= JOURNAL_LIST_ENTRY(entry
->next
);
596 if (jl
->j_trans_id
<= trans_id
) {
604 * If page->mapping was null, we failed to truncate this page for
605 * some reason. Most likely because it was truncated after being
606 * logged via data=journal.
608 * This does a check to see if the buffer belongs to one of these
609 * lost pages before doing the final put_bh. If page->mapping was
610 * null, it tries to free buffers on the page, which should make the
611 * final page_cache_release drop the page from the lru.
613 static void release_buffer_page(struct buffer_head
*bh
)
615 struct page
*page
= bh
->b_page
;
616 if (!page
->mapping
&& trylock_page(page
)) {
617 page_cache_get(page
);
620 try_to_free_buffers(page
);
622 page_cache_release(page
);
628 static void reiserfs_end_buffer_io_sync(struct buffer_head
*bh
, int uptodate
)
630 char b
[BDEVNAME_SIZE
];
632 if (buffer_journaled(bh
)) {
633 reiserfs_warning(NULL
, "clm-2084",
634 "pinned buffer %lu:%s sent to disk",
635 bh
->b_blocknr
, bdevname(bh
->b_bdev
, b
));
638 set_buffer_uptodate(bh
);
640 clear_buffer_uptodate(bh
);
643 release_buffer_page(bh
);
646 static void reiserfs_end_ordered_io(struct buffer_head
*bh
, int uptodate
)
649 set_buffer_uptodate(bh
);
651 clear_buffer_uptodate(bh
);
656 static void submit_logged_buffer(struct buffer_head
*bh
)
659 bh
->b_end_io
= reiserfs_end_buffer_io_sync
;
660 clear_buffer_journal_new(bh
);
661 clear_buffer_dirty(bh
);
662 if (!test_clear_buffer_journal_test(bh
))
664 if (!buffer_uptodate(bh
))
666 submit_bh(WRITE
, bh
);
669 static void submit_ordered_buffer(struct buffer_head
*bh
)
672 bh
->b_end_io
= reiserfs_end_ordered_io
;
673 clear_buffer_dirty(bh
);
674 if (!buffer_uptodate(bh
))
676 submit_bh(WRITE
, bh
);
679 static int submit_barrier_buffer(struct buffer_head
*bh
)
682 bh
->b_end_io
= reiserfs_end_ordered_io
;
683 clear_buffer_dirty(bh
);
684 if (!buffer_uptodate(bh
))
686 return submit_bh(WRITE_BARRIER
, bh
);
689 static void check_barrier_completion(struct super_block
*s
,
690 struct buffer_head
*bh
)
692 if (buffer_eopnotsupp(bh
)) {
693 clear_buffer_eopnotsupp(bh
);
695 set_buffer_uptodate(bh
);
696 set_buffer_dirty(bh
);
697 reiserfs_write_unlock(s
);
698 sync_dirty_buffer(bh
);
699 reiserfs_write_lock(s
);
703 #define CHUNK_SIZE 32
704 struct buffer_chunk
{
705 struct buffer_head
*bh
[CHUNK_SIZE
];
709 static void write_chunk(struct buffer_chunk
*chunk
)
713 for (i
= 0; i
< chunk
->nr
; i
++) {
714 submit_logged_buffer(chunk
->bh
[i
]);
720 static void write_ordered_chunk(struct buffer_chunk
*chunk
)
724 for (i
= 0; i
< chunk
->nr
; i
++) {
725 submit_ordered_buffer(chunk
->bh
[i
]);
731 static int add_to_chunk(struct buffer_chunk
*chunk
, struct buffer_head
*bh
,
732 spinlock_t
* lock
, void (fn
) (struct buffer_chunk
*))
735 BUG_ON(chunk
->nr
>= CHUNK_SIZE
);
736 chunk
->bh
[chunk
->nr
++] = bh
;
737 if (chunk
->nr
>= CHUNK_SIZE
) {
748 static atomic_t nr_reiserfs_jh
= ATOMIC_INIT(0);
749 static struct reiserfs_jh
*alloc_jh(void)
751 struct reiserfs_jh
*jh
;
753 jh
= kmalloc(sizeof(*jh
), GFP_NOFS
);
755 atomic_inc(&nr_reiserfs_jh
);
763 * we want to free the jh when the buffer has been written
766 void reiserfs_free_jh(struct buffer_head
*bh
)
768 struct reiserfs_jh
*jh
;
772 bh
->b_private
= NULL
;
774 list_del_init(&jh
->list
);
776 if (atomic_read(&nr_reiserfs_jh
) <= 0)
778 atomic_dec(&nr_reiserfs_jh
);
783 static inline int __add_jh(struct reiserfs_journal
*j
, struct buffer_head
*bh
,
786 struct reiserfs_jh
*jh
;
789 spin_lock(&j
->j_dirty_buffers_lock
);
790 if (!bh
->b_private
) {
791 spin_unlock(&j
->j_dirty_buffers_lock
);
795 list_del_init(&jh
->list
);
800 spin_lock(&j
->j_dirty_buffers_lock
);
801 /* buffer must be locked for __add_jh, should be able to have
802 * two adds at the same time
804 BUG_ON(bh
->b_private
);
808 jh
->jl
= j
->j_current_jl
;
810 list_add_tail(&jh
->list
, &jh
->jl
->j_tail_bh_list
);
812 list_add_tail(&jh
->list
, &jh
->jl
->j_bh_list
);
814 spin_unlock(&j
->j_dirty_buffers_lock
);
818 int reiserfs_add_tail_list(struct inode
*inode
, struct buffer_head
*bh
)
820 return __add_jh(SB_JOURNAL(inode
->i_sb
), bh
, 1);
822 int reiserfs_add_ordered_list(struct inode
*inode
, struct buffer_head
*bh
)
824 return __add_jh(SB_JOURNAL(inode
->i_sb
), bh
, 0);
827 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
828 static int write_ordered_buffers(spinlock_t
* lock
,
829 struct reiserfs_journal
*j
,
830 struct reiserfs_journal_list
*jl
,
831 struct list_head
*list
)
833 struct buffer_head
*bh
;
834 struct reiserfs_jh
*jh
;
835 int ret
= j
->j_errno
;
836 struct buffer_chunk chunk
;
837 struct list_head tmp
;
838 INIT_LIST_HEAD(&tmp
);
842 while (!list_empty(list
)) {
843 jh
= JH_ENTRY(list
->next
);
846 if (!trylock_buffer(bh
)) {
847 if (!buffer_dirty(bh
)) {
848 list_move(&jh
->list
, &tmp
);
853 write_ordered_chunk(&chunk
);
859 /* in theory, dirty non-uptodate buffers should never get here,
860 * but the upper layer io error paths still have a few quirks.
861 * Handle them here as gracefully as we can
863 if (!buffer_uptodate(bh
) && buffer_dirty(bh
)) {
864 clear_buffer_dirty(bh
);
867 if (buffer_dirty(bh
)) {
868 list_move(&jh
->list
, &tmp
);
869 add_to_chunk(&chunk
, bh
, lock
, write_ordered_chunk
);
871 reiserfs_free_jh(bh
);
876 cond_resched_lock(lock
);
880 write_ordered_chunk(&chunk
);
883 while (!list_empty(&tmp
)) {
884 jh
= JH_ENTRY(tmp
.prev
);
887 reiserfs_free_jh(bh
);
889 if (buffer_locked(bh
)) {
894 if (!buffer_uptodate(bh
)) {
897 /* ugly interaction with invalidatepage here.
898 * reiserfs_invalidate_page will pin any buffer that has a valid
899 * journal head from an older transaction. If someone else sets
900 * our buffer dirty after we write it in the first loop, and
901 * then someone truncates the page away, nobody will ever write
902 * the buffer. We're safe if we write the page one last time
903 * after freeing the journal header.
905 if (buffer_dirty(bh
) && unlikely(bh
->b_page
->mapping
== NULL
)) {
907 ll_rw_block(WRITE
, 1, &bh
);
911 cond_resched_lock(lock
);
917 static int flush_older_commits(struct super_block
*s
,
918 struct reiserfs_journal_list
*jl
)
920 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
921 struct reiserfs_journal_list
*other_jl
;
922 struct reiserfs_journal_list
*first_jl
;
923 struct list_head
*entry
;
924 unsigned int trans_id
= jl
->j_trans_id
;
925 unsigned int other_trans_id
;
926 unsigned int first_trans_id
;
930 * first we walk backwards to find the oldest uncommitted transation
933 entry
= jl
->j_list
.prev
;
935 other_jl
= JOURNAL_LIST_ENTRY(entry
);
936 if (entry
== &journal
->j_journal_list
||
937 atomic_read(&other_jl
->j_older_commits_done
))
941 entry
= other_jl
->j_list
.prev
;
944 /* if we didn't find any older uncommitted transactions, return now */
945 if (first_jl
== jl
) {
949 first_trans_id
= first_jl
->j_trans_id
;
951 entry
= &first_jl
->j_list
;
953 other_jl
= JOURNAL_LIST_ENTRY(entry
);
954 other_trans_id
= other_jl
->j_trans_id
;
956 if (other_trans_id
< trans_id
) {
957 if (atomic_read(&other_jl
->j_commit_left
) != 0) {
958 flush_commit_list(s
, other_jl
, 0);
960 /* list we were called with is gone, return */
961 if (!journal_list_still_alive(s
, trans_id
))
964 /* the one we just flushed is gone, this means all
965 * older lists are also gone, so first_jl is no longer
966 * valid either. Go back to the beginning.
968 if (!journal_list_still_alive
969 (s
, other_trans_id
)) {
974 if (entry
== &journal
->j_journal_list
)
983 static int reiserfs_async_progress_wait(struct super_block
*s
)
986 struct reiserfs_journal
*j
= SB_JOURNAL(s
);
988 if (atomic_read(&j
->j_async_throttle
)) {
989 reiserfs_write_unlock(s
);
990 congestion_wait(BLK_RW_ASYNC
, HZ
/ 10);
991 reiserfs_write_lock(s
);
998 ** if this journal list still has commit blocks unflushed, send them to disk.
1000 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
1001 ** Before the commit block can by written, every other log block must be safely on disk
1004 static int flush_commit_list(struct super_block
*s
,
1005 struct reiserfs_journal_list
*jl
, int flushall
)
1009 struct buffer_head
*tbh
= NULL
;
1010 unsigned int trans_id
= jl
->j_trans_id
;
1011 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1016 reiserfs_check_lock_depth(s
, "flush_commit_list");
1018 if (atomic_read(&jl
->j_older_commits_done
)) {
1024 /* before we can put our commit blocks on disk, we have to make sure everyone older than
1025 ** us is on disk too
1027 BUG_ON(jl
->j_len
<= 0);
1028 BUG_ON(trans_id
== journal
->j_trans_id
);
1030 get_journal_list(jl
);
1032 if (flush_older_commits(s
, jl
) == 1) {
1033 /* list disappeared during flush_older_commits. return */
1038 /* make sure nobody is trying to flush this one at the same time */
1039 reiserfs_mutex_lock_safe(&jl
->j_commit_mutex
, s
);
1041 if (!journal_list_still_alive(s
, trans_id
)) {
1042 mutex_unlock(&jl
->j_commit_mutex
);
1045 BUG_ON(jl
->j_trans_id
== 0);
1047 /* this commit is done, exit */
1048 if (atomic_read(&(jl
->j_commit_left
)) <= 0) {
1050 atomic_set(&(jl
->j_older_commits_done
), 1);
1052 mutex_unlock(&jl
->j_commit_mutex
);
1056 if (!list_empty(&jl
->j_bh_list
)) {
1060 * We might sleep in numerous places inside
1061 * write_ordered_buffers. Relax the write lock.
1063 reiserfs_write_unlock(s
);
1064 ret
= write_ordered_buffers(&journal
->j_dirty_buffers_lock
,
1065 journal
, jl
, &jl
->j_bh_list
);
1066 if (ret
< 0 && retval
== 0)
1068 reiserfs_write_lock(s
);
1070 BUG_ON(!list_empty(&jl
->j_bh_list
));
1072 * for the description block and all the log blocks, submit any buffers
1073 * that haven't already reached the disk. Try to write at least 256
1074 * log blocks. later on, we will only wait on blocks that correspond
1075 * to this transaction, but while we're unplugging we might as well
1076 * get a chunk of data on there.
1078 atomic_inc(&journal
->j_async_throttle
);
1079 write_len
= jl
->j_len
+ 1;
1080 if (write_len
< 256)
1082 for (i
= 0 ; i
< write_len
; i
++) {
1083 bn
= SB_ONDISK_JOURNAL_1st_BLOCK(s
) + (jl
->j_start
+ i
) %
1084 SB_ONDISK_JOURNAL_SIZE(s
);
1085 tbh
= journal_find_get_block(s
, bn
);
1087 if (buffer_dirty(tbh
)) {
1088 reiserfs_write_unlock(s
);
1089 ll_rw_block(WRITE
, 1, &tbh
);
1090 reiserfs_write_lock(s
);
1095 atomic_dec(&journal
->j_async_throttle
);
1097 /* We're skipping the commit if there's an error */
1098 if (retval
|| reiserfs_is_journal_aborted(journal
))
1101 /* wait on everything written so far before writing the commit
1102 * if we are in barrier mode, send the commit down now
1104 barrier
= reiserfs_barrier_flush(s
);
1107 lock_buffer(jl
->j_commit_bh
);
1108 ret
= submit_barrier_buffer(jl
->j_commit_bh
);
1109 if (ret
== -EOPNOTSUPP
) {
1110 set_buffer_uptodate(jl
->j_commit_bh
);
1115 for (i
= 0; i
< (jl
->j_len
+ 1); i
++) {
1116 bn
= SB_ONDISK_JOURNAL_1st_BLOCK(s
) +
1117 (jl
->j_start
+ i
) % SB_ONDISK_JOURNAL_SIZE(s
);
1118 tbh
= journal_find_get_block(s
, bn
);
1120 reiserfs_write_unlock(s
);
1121 wait_on_buffer(tbh
);
1122 reiserfs_write_lock(s
);
1123 // since we're using ll_rw_blk above, it might have skipped over
1124 // a locked buffer. Double check here
1126 /* redundant, sync_dirty_buffer() checks */
1127 if (buffer_dirty(tbh
)) {
1128 reiserfs_write_unlock(s
);
1129 sync_dirty_buffer(tbh
);
1130 reiserfs_write_lock(s
);
1132 if (unlikely(!buffer_uptodate(tbh
))) {
1133 #ifdef CONFIG_REISERFS_CHECK
1134 reiserfs_warning(s
, "journal-601",
1135 "buffer write failed");
1139 put_bh(tbh
); /* once for journal_find_get_block */
1140 put_bh(tbh
); /* once due to original getblk in do_journal_end */
1141 atomic_dec(&(jl
->j_commit_left
));
1144 BUG_ON(atomic_read(&(jl
->j_commit_left
)) != 1);
1147 /* If there was a write error in the journal - we can't commit
1148 * this transaction - it will be invalid and, if successful,
1149 * will just end up propagating the write error out to
1150 * the file system. */
1151 if (likely(!retval
&& !reiserfs_is_journal_aborted (journal
))) {
1152 if (buffer_dirty(jl
->j_commit_bh
))
1154 mark_buffer_dirty(jl
->j_commit_bh
) ;
1155 reiserfs_write_unlock(s
);
1156 sync_dirty_buffer(jl
->j_commit_bh
) ;
1157 reiserfs_write_lock(s
);
1160 reiserfs_write_unlock(s
);
1161 wait_on_buffer(jl
->j_commit_bh
);
1162 reiserfs_write_lock(s
);
1165 check_barrier_completion(s
, jl
->j_commit_bh
);
1167 /* If there was a write error in the journal - we can't commit this
1168 * transaction - it will be invalid and, if successful, will just end
1169 * up propagating the write error out to the filesystem. */
1170 if (unlikely(!buffer_uptodate(jl
->j_commit_bh
))) {
1171 #ifdef CONFIG_REISERFS_CHECK
1172 reiserfs_warning(s
, "journal-615", "buffer write failed");
1176 bforget(jl
->j_commit_bh
);
1177 if (journal
->j_last_commit_id
!= 0 &&
1178 (jl
->j_trans_id
- journal
->j_last_commit_id
) != 1) {
1179 reiserfs_warning(s
, "clm-2200", "last commit %lu, current %lu",
1180 journal
->j_last_commit_id
, jl
->j_trans_id
);
1182 journal
->j_last_commit_id
= jl
->j_trans_id
;
1184 /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */
1185 cleanup_freed_for_journal_list(s
, jl
);
1187 retval
= retval
? retval
: journal
->j_errno
;
1189 /* mark the metadata dirty */
1191 dirty_one_transaction(s
, jl
);
1192 atomic_dec(&(jl
->j_commit_left
));
1195 atomic_set(&(jl
->j_older_commits_done
), 1);
1197 mutex_unlock(&jl
->j_commit_mutex
);
1199 put_journal_list(s
, jl
);
1202 reiserfs_abort(s
, retval
, "Journal write error in %s",
1209 ** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or
1210 ** returns NULL if it can't find anything
1212 static struct reiserfs_journal_list
*find_newer_jl_for_cn(struct
1213 reiserfs_journal_cnode
1216 struct super_block
*sb
= cn
->sb
;
1217 b_blocknr_t blocknr
= cn
->blocknr
;
1221 if (cn
->sb
== sb
&& cn
->blocknr
== blocknr
&& cn
->jlist
) {
1229 static int newer_jl_done(struct reiserfs_journal_cnode
*cn
)
1231 struct super_block
*sb
= cn
->sb
;
1232 b_blocknr_t blocknr
= cn
->blocknr
;
1236 if (cn
->sb
== sb
&& cn
->blocknr
== blocknr
&& cn
->jlist
&&
1237 atomic_read(&cn
->jlist
->j_commit_left
) != 0)
1244 static void remove_journal_hash(struct super_block
*,
1245 struct reiserfs_journal_cnode
**,
1246 struct reiserfs_journal_list
*, unsigned long,
1250 ** once all the real blocks have been flushed, it is safe to remove them from the
1251 ** journal list for this transaction. Aside from freeing the cnode, this also allows the
1252 ** block to be reallocated for data blocks if it had been deleted.
1254 static void remove_all_from_journal_list(struct super_block
*sb
,
1255 struct reiserfs_journal_list
*jl
,
1258 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1259 struct reiserfs_journal_cnode
*cn
, *last
;
1260 cn
= jl
->j_realblock
;
1262 /* which is better, to lock once around the whole loop, or
1263 ** to lock for each call to remove_journal_hash?
1266 if (cn
->blocknr
!= 0) {
1268 reiserfs_warning(sb
, "reiserfs-2201",
1269 "block %u, bh is %d, state %ld",
1270 cn
->blocknr
, cn
->bh
? 1 : 0,
1274 remove_journal_hash(sb
, journal
->j_list_hash_table
,
1275 jl
, cn
->blocknr
, 1);
1279 free_cnode(sb
, last
);
1281 jl
->j_realblock
= NULL
;
1285 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1286 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1287 ** releasing blocks in this transaction for reuse as data blocks.
1288 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1291 static int _update_journal_header_block(struct super_block
*sb
,
1292 unsigned long offset
,
1293 unsigned int trans_id
)
1295 struct reiserfs_journal_header
*jh
;
1296 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1298 if (reiserfs_is_journal_aborted(journal
))
1301 if (trans_id
>= journal
->j_last_flush_trans_id
) {
1302 if (buffer_locked((journal
->j_header_bh
))) {
1303 reiserfs_write_unlock(sb
);
1304 wait_on_buffer((journal
->j_header_bh
));
1305 reiserfs_write_lock(sb
);
1306 if (unlikely(!buffer_uptodate(journal
->j_header_bh
))) {
1307 #ifdef CONFIG_REISERFS_CHECK
1308 reiserfs_warning(sb
, "journal-699",
1309 "buffer write failed");
1314 journal
->j_last_flush_trans_id
= trans_id
;
1315 journal
->j_first_unflushed_offset
= offset
;
1316 jh
= (struct reiserfs_journal_header
*)(journal
->j_header_bh
->
1318 jh
->j_last_flush_trans_id
= cpu_to_le32(trans_id
);
1319 jh
->j_first_unflushed_offset
= cpu_to_le32(offset
);
1320 jh
->j_mount_id
= cpu_to_le32(journal
->j_mount_id
);
1322 if (reiserfs_barrier_flush(sb
)) {
1324 lock_buffer(journal
->j_header_bh
);
1325 ret
= submit_barrier_buffer(journal
->j_header_bh
);
1326 if (ret
== -EOPNOTSUPP
) {
1327 set_buffer_uptodate(journal
->j_header_bh
);
1328 disable_barrier(sb
);
1331 reiserfs_write_unlock(sb
);
1332 wait_on_buffer(journal
->j_header_bh
);
1333 reiserfs_write_lock(sb
);
1334 check_barrier_completion(sb
, journal
->j_header_bh
);
1337 set_buffer_dirty(journal
->j_header_bh
);
1338 reiserfs_write_unlock(sb
);
1339 sync_dirty_buffer(journal
->j_header_bh
);
1340 reiserfs_write_lock(sb
);
1342 if (!buffer_uptodate(journal
->j_header_bh
)) {
1343 reiserfs_warning(sb
, "journal-837",
1344 "IO error during journal replay");
1351 static int update_journal_header_block(struct super_block
*sb
,
1352 unsigned long offset
,
1353 unsigned int trans_id
)
1355 return _update_journal_header_block(sb
, offset
, trans_id
);
1359 ** flush any and all journal lists older than you are
1360 ** can only be called from flush_journal_list
1362 static int flush_older_journal_lists(struct super_block
*sb
,
1363 struct reiserfs_journal_list
*jl
)
1365 struct list_head
*entry
;
1366 struct reiserfs_journal_list
*other_jl
;
1367 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1368 unsigned int trans_id
= jl
->j_trans_id
;
1370 /* we know we are the only ones flushing things, no extra race
1371 * protection is required.
1374 entry
= journal
->j_journal_list
.next
;
1376 if (entry
== &journal
->j_journal_list
)
1378 other_jl
= JOURNAL_LIST_ENTRY(entry
);
1379 if (other_jl
->j_trans_id
< trans_id
) {
1380 BUG_ON(other_jl
->j_refcount
<= 0);
1381 /* do not flush all */
1382 flush_journal_list(sb
, other_jl
, 0);
1384 /* other_jl is now deleted from the list */
1390 static void del_from_work_list(struct super_block
*s
,
1391 struct reiserfs_journal_list
*jl
)
1393 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1394 if (!list_empty(&jl
->j_working_list
)) {
1395 list_del_init(&jl
->j_working_list
);
1396 journal
->j_num_work_lists
--;
1400 /* flush a journal list, both commit and real blocks
1402 ** always set flushall to 1, unless you are calling from inside
1403 ** flush_journal_list
1405 ** IMPORTANT. This can only be called while there are no journal writers,
1406 ** and the journal is locked. That means it can only be called from
1407 ** do_journal_end, or by journal_release
1409 static int flush_journal_list(struct super_block
*s
,
1410 struct reiserfs_journal_list
*jl
, int flushall
)
1412 struct reiserfs_journal_list
*pjl
;
1413 struct reiserfs_journal_cnode
*cn
, *last
;
1417 struct buffer_head
*saved_bh
;
1418 unsigned long j_len_saved
= jl
->j_len
;
1419 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1422 BUG_ON(j_len_saved
<= 0);
1424 if (atomic_read(&journal
->j_wcount
) != 0) {
1425 reiserfs_warning(s
, "clm-2048", "called with wcount %d",
1426 atomic_read(&journal
->j_wcount
));
1428 BUG_ON(jl
->j_trans_id
== 0);
1430 /* if flushall == 0, the lock is already held */
1432 reiserfs_mutex_lock_safe(&journal
->j_flush_mutex
, s
);
1433 } else if (mutex_trylock(&journal
->j_flush_mutex
)) {
1438 if (j_len_saved
> journal
->j_trans_max
) {
1439 reiserfs_panic(s
, "journal-715", "length is %lu, trans id %lu",
1440 j_len_saved
, jl
->j_trans_id
);
1446 /* if all the work is already done, get out of here */
1447 if (atomic_read(&(jl
->j_nonzerolen
)) <= 0 &&
1448 atomic_read(&(jl
->j_commit_left
)) <= 0) {
1449 goto flush_older_and_return
;
1452 /* start by putting the commit list on disk. This will also flush
1453 ** the commit lists of any olders transactions
1455 flush_commit_list(s
, jl
, 1);
1457 if (!(jl
->j_state
& LIST_DIRTY
)
1458 && !reiserfs_is_journal_aborted(journal
))
1461 /* are we done now? */
1462 if (atomic_read(&(jl
->j_nonzerolen
)) <= 0 &&
1463 atomic_read(&(jl
->j_commit_left
)) <= 0) {
1464 goto flush_older_and_return
;
1467 /* loop through each cnode, see if we need to write it,
1468 ** or wait on a more recent transaction, or just ignore it
1470 if (atomic_read(&(journal
->j_wcount
)) != 0) {
1471 reiserfs_panic(s
, "journal-844", "journal list is flushing, "
1474 cn
= jl
->j_realblock
;
1479 /* blocknr of 0 is no longer in the hash, ignore it */
1480 if (cn
->blocknr
== 0) {
1484 /* This transaction failed commit. Don't write out to the disk */
1485 if (!(jl
->j_state
& LIST_DIRTY
))
1488 pjl
= find_newer_jl_for_cn(cn
);
1489 /* the order is important here. We check pjl to make sure we
1490 ** don't clear BH_JDirty_wait if we aren't the one writing this
1493 if (!pjl
&& cn
->bh
) {
1496 /* we do this to make sure nobody releases the buffer while
1497 ** we are working with it
1501 if (buffer_journal_dirty(saved_bh
)) {
1502 BUG_ON(!can_dirty(cn
));
1505 } else if (can_dirty(cn
)) {
1506 /* everything with !pjl && jwait should be writable */
1511 /* if someone has this block in a newer transaction, just make
1512 ** sure they are committed, and don't try writing it to disk
1515 if (atomic_read(&pjl
->j_commit_left
))
1516 flush_commit_list(s
, pjl
, 1);
1520 /* bh == NULL when the block got to disk on its own, OR,
1521 ** the block got freed in a future transaction
1523 if (saved_bh
== NULL
) {
1527 /* this should never happen. kupdate_one_transaction has this list
1528 ** locked while it works, so we should never see a buffer here that
1529 ** is not marked JDirty_wait
1531 if ((!was_jwait
) && !buffer_locked(saved_bh
)) {
1532 reiserfs_warning(s
, "journal-813",
1533 "BAD! buffer %llu %cdirty %cjwait, "
1534 "not in a newer tranasction",
1535 (unsigned long long)saved_bh
->
1536 b_blocknr
, was_dirty
? ' ' : '!',
1537 was_jwait
? ' ' : '!');
1540 /* we inc again because saved_bh gets decremented at free_cnode */
1542 set_bit(BLOCK_NEEDS_FLUSH
, &cn
->state
);
1543 lock_buffer(saved_bh
);
1544 BUG_ON(cn
->blocknr
!= saved_bh
->b_blocknr
);
1545 if (buffer_dirty(saved_bh
))
1546 submit_logged_buffer(saved_bh
);
1548 unlock_buffer(saved_bh
);
1551 reiserfs_warning(s
, "clm-2082",
1552 "Unable to flush buffer %llu in %s",
1553 (unsigned long long)saved_bh
->
1554 b_blocknr
, __func__
);
1560 /* we incremented this to keep others from taking the buffer head away */
1562 if (atomic_read(&(saved_bh
->b_count
)) < 0) {
1563 reiserfs_warning(s
, "journal-945",
1564 "saved_bh->b_count < 0");
1569 cn
= jl
->j_realblock
;
1571 if (test_bit(BLOCK_NEEDS_FLUSH
, &cn
->state
)) {
1573 reiserfs_panic(s
, "journal-1011",
1577 reiserfs_write_unlock(s
);
1578 wait_on_buffer(cn
->bh
);
1579 reiserfs_write_lock(s
);
1582 reiserfs_panic(s
, "journal-1012",
1585 if (unlikely(!buffer_uptodate(cn
->bh
))) {
1586 #ifdef CONFIG_REISERFS_CHECK
1587 reiserfs_warning(s
, "journal-949",
1588 "buffer write failed");
1592 /* note, we must clear the JDirty_wait bit after the up to date
1593 ** check, otherwise we race against our flushpage routine
1595 BUG_ON(!test_clear_buffer_journal_dirty
1598 /* drop one ref for us */
1600 /* drop one ref for journal_mark_dirty */
1601 release_buffer_page(cn
->bh
);
1608 reiserfs_abort(s
, -EIO
,
1609 "Write error while pushing transaction to disk in %s",
1611 flush_older_and_return
:
1613 /* before we can update the journal header block, we _must_ flush all
1614 ** real blocks from all older transactions to disk. This is because
1615 ** once the header block is updated, this transaction will not be
1616 ** replayed after a crash
1619 flush_older_journal_lists(s
, jl
);
1622 err
= journal
->j_errno
;
1623 /* before we can remove everything from the hash tables for this
1624 ** transaction, we must make sure it can never be replayed
1626 ** since we are only called from do_journal_end, we know for sure there
1627 ** are no allocations going on while we are flushing journal lists. So,
1628 ** we only need to update the journal header block for the last list
1631 if (!err
&& flushall
) {
1633 update_journal_header_block(s
,
1634 (jl
->j_start
+ jl
->j_len
+
1635 2) % SB_ONDISK_JOURNAL_SIZE(s
),
1638 reiserfs_abort(s
, -EIO
,
1639 "Write error while updating journal header in %s",
1642 remove_all_from_journal_list(s
, jl
, 0);
1643 list_del_init(&jl
->j_list
);
1644 journal
->j_num_lists
--;
1645 del_from_work_list(s
, jl
);
1647 if (journal
->j_last_flush_id
!= 0 &&
1648 (jl
->j_trans_id
- journal
->j_last_flush_id
) != 1) {
1649 reiserfs_warning(s
, "clm-2201", "last flush %lu, current %lu",
1650 journal
->j_last_flush_id
, jl
->j_trans_id
);
1652 journal
->j_last_flush_id
= jl
->j_trans_id
;
1654 /* not strictly required since we are freeing the list, but it should
1655 * help find code using dead lists later on
1658 atomic_set(&(jl
->j_nonzerolen
), 0);
1660 jl
->j_realblock
= NULL
;
1661 jl
->j_commit_bh
= NULL
;
1664 put_journal_list(s
, jl
);
1666 mutex_unlock(&journal
->j_flush_mutex
);
1671 static int test_transaction(struct super_block
*s
,
1672 struct reiserfs_journal_list
*jl
)
1674 struct reiserfs_journal_cnode
*cn
;
1676 if (jl
->j_len
== 0 || atomic_read(&jl
->j_nonzerolen
) == 0)
1679 cn
= jl
->j_realblock
;
1681 /* if the blocknr == 0, this has been cleared from the hash,
1684 if (cn
->blocknr
== 0) {
1687 if (cn
->bh
&& !newer_jl_done(cn
))
1696 static int write_one_transaction(struct super_block
*s
,
1697 struct reiserfs_journal_list
*jl
,
1698 struct buffer_chunk
*chunk
)
1700 struct reiserfs_journal_cnode
*cn
;
1703 jl
->j_state
|= LIST_TOUCHED
;
1704 del_from_work_list(s
, jl
);
1705 if (jl
->j_len
== 0 || atomic_read(&jl
->j_nonzerolen
) == 0) {
1709 cn
= jl
->j_realblock
;
1711 /* if the blocknr == 0, this has been cleared from the hash,
1714 if (cn
->blocknr
== 0) {
1717 if (cn
->bh
&& can_dirty(cn
) && buffer_dirty(cn
->bh
)) {
1718 struct buffer_head
*tmp_bh
;
1719 /* we can race against journal_mark_freed when we try
1720 * to lock_buffer(cn->bh), so we have to inc the buffer
1721 * count, and recheck things after locking
1725 lock_buffer(tmp_bh
);
1726 if (cn
->bh
&& can_dirty(cn
) && buffer_dirty(tmp_bh
)) {
1727 if (!buffer_journal_dirty(tmp_bh
) ||
1728 buffer_journal_prepared(tmp_bh
))
1730 add_to_chunk(chunk
, tmp_bh
, NULL
, write_chunk
);
1733 /* note, cn->bh might be null now */
1734 unlock_buffer(tmp_bh
);
1745 /* used by flush_commit_list */
1746 static int dirty_one_transaction(struct super_block
*s
,
1747 struct reiserfs_journal_list
*jl
)
1749 struct reiserfs_journal_cnode
*cn
;
1750 struct reiserfs_journal_list
*pjl
;
1753 jl
->j_state
|= LIST_DIRTY
;
1754 cn
= jl
->j_realblock
;
1756 /* look for a more recent transaction that logged this
1757 ** buffer. Only the most recent transaction with a buffer in
1758 ** it is allowed to send that buffer to disk
1760 pjl
= find_newer_jl_for_cn(cn
);
1761 if (!pjl
&& cn
->blocknr
&& cn
->bh
1762 && buffer_journal_dirty(cn
->bh
)) {
1763 BUG_ON(!can_dirty(cn
));
1764 /* if the buffer is prepared, it will either be logged
1765 * or restored. If restored, we need to make sure
1766 * it actually gets marked dirty
1768 clear_buffer_journal_new(cn
->bh
);
1769 if (buffer_journal_prepared(cn
->bh
)) {
1770 set_buffer_journal_restore_dirty(cn
->bh
);
1772 set_buffer_journal_test(cn
->bh
);
1773 mark_buffer_dirty(cn
->bh
);
1781 static int kupdate_transactions(struct super_block
*s
,
1782 struct reiserfs_journal_list
*jl
,
1783 struct reiserfs_journal_list
**next_jl
,
1784 unsigned int *next_trans_id
,
1785 int num_blocks
, int num_trans
)
1789 int transactions_flushed
= 0;
1790 unsigned int orig_trans_id
= jl
->j_trans_id
;
1791 struct buffer_chunk chunk
;
1792 struct list_head
*entry
;
1793 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1796 reiserfs_mutex_lock_safe(&journal
->j_flush_mutex
, s
);
1797 if (!journal_list_still_alive(s
, orig_trans_id
)) {
1801 /* we've got j_flush_mutex held, nobody is going to delete any
1802 * of these lists out from underneath us
1804 while ((num_trans
&& transactions_flushed
< num_trans
) ||
1805 (!num_trans
&& written
< num_blocks
)) {
1807 if (jl
->j_len
== 0 || (jl
->j_state
& LIST_TOUCHED
) ||
1808 atomic_read(&jl
->j_commit_left
)
1809 || !(jl
->j_state
& LIST_DIRTY
)) {
1810 del_from_work_list(s
, jl
);
1813 ret
= write_one_transaction(s
, jl
, &chunk
);
1817 transactions_flushed
++;
1819 entry
= jl
->j_list
.next
;
1822 if (entry
== &journal
->j_journal_list
) {
1825 jl
= JOURNAL_LIST_ENTRY(entry
);
1827 /* don't bother with older transactions */
1828 if (jl
->j_trans_id
<= orig_trans_id
)
1832 write_chunk(&chunk
);
1836 mutex_unlock(&journal
->j_flush_mutex
);
1840 /* for o_sync and fsync heavy applications, they tend to use
1841 ** all the journa list slots with tiny transactions. These
1842 ** trigger lots and lots of calls to update the header block, which
1843 ** adds seeks and slows things down.
1845 ** This function tries to clear out a large chunk of the journal lists
1846 ** at once, which makes everything faster since only the newest journal
1847 ** list updates the header block
1849 static int flush_used_journal_lists(struct super_block
*s
,
1850 struct reiserfs_journal_list
*jl
)
1852 unsigned long len
= 0;
1853 unsigned long cur_len
;
1857 struct reiserfs_journal_list
*tjl
;
1858 struct reiserfs_journal_list
*flush_jl
;
1859 unsigned int trans_id
;
1860 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1862 flush_jl
= tjl
= jl
;
1864 /* in data logging mode, try harder to flush a lot of blocks */
1865 if (reiserfs_data_log(s
))
1867 /* flush for 256 transactions or limit blocks, whichever comes first */
1868 for (i
= 0; i
< 256 && len
< limit
; i
++) {
1869 if (atomic_read(&tjl
->j_commit_left
) ||
1870 tjl
->j_trans_id
< jl
->j_trans_id
) {
1873 cur_len
= atomic_read(&tjl
->j_nonzerolen
);
1875 tjl
->j_state
&= ~LIST_TOUCHED
;
1879 if (tjl
->j_list
.next
== &journal
->j_journal_list
)
1881 tjl
= JOURNAL_LIST_ENTRY(tjl
->j_list
.next
);
1883 /* try to find a group of blocks we can flush across all the
1884 ** transactions, but only bother if we've actually spanned
1885 ** across multiple lists
1887 if (flush_jl
!= jl
) {
1888 ret
= kupdate_transactions(s
, jl
, &tjl
, &trans_id
, len
, i
);
1890 flush_journal_list(s
, flush_jl
, 1);
1895 ** removes any nodes in table with name block and dev as bh.
1896 ** only touchs the hnext and hprev pointers.
1898 void remove_journal_hash(struct super_block
*sb
,
1899 struct reiserfs_journal_cnode
**table
,
1900 struct reiserfs_journal_list
*jl
,
1901 unsigned long block
, int remove_freed
)
1903 struct reiserfs_journal_cnode
*cur
;
1904 struct reiserfs_journal_cnode
**head
;
1906 head
= &(journal_hash(table
, sb
, block
));
1912 if (cur
->blocknr
== block
&& cur
->sb
== sb
1913 && (jl
== NULL
|| jl
== cur
->jlist
)
1914 && (!test_bit(BLOCK_FREED
, &cur
->state
) || remove_freed
)) {
1916 cur
->hnext
->hprev
= cur
->hprev
;
1919 cur
->hprev
->hnext
= cur
->hnext
;
1926 if (cur
->bh
&& cur
->jlist
) /* anybody who clears the cur->bh will also dec the nonzerolen */
1927 atomic_dec(&(cur
->jlist
->j_nonzerolen
));
1935 static void free_journal_ram(struct super_block
*sb
)
1937 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1938 kfree(journal
->j_current_jl
);
1939 journal
->j_num_lists
--;
1941 vfree(journal
->j_cnode_free_orig
);
1942 free_list_bitmaps(sb
, journal
->j_list_bitmap
);
1943 free_bitmap_nodes(sb
); /* must be after free_list_bitmaps */
1944 if (journal
->j_header_bh
) {
1945 brelse(journal
->j_header_bh
);
1947 /* j_header_bh is on the journal dev, make sure not to release the journal
1948 * dev until we brelse j_header_bh
1950 release_journal_dev(sb
, journal
);
1955 ** call on unmount. Only set error to 1 if you haven't made your way out
1956 ** of read_super() yet. Any other caller must keep error at 0.
1958 static int do_journal_release(struct reiserfs_transaction_handle
*th
,
1959 struct super_block
*sb
, int error
)
1961 struct reiserfs_transaction_handle myth
;
1963 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1965 /* we only want to flush out transactions if we were called with error == 0
1967 if (!error
&& !(sb
->s_flags
& MS_RDONLY
)) {
1968 /* end the current trans */
1969 BUG_ON(!th
->t_trans_id
);
1970 do_journal_end(th
, sb
, 10, FLUSH_ALL
);
1972 /* make sure something gets logged to force our way into the flush code */
1973 if (!journal_join(&myth
, sb
, 1)) {
1974 reiserfs_prepare_for_journal(sb
,
1975 SB_BUFFER_WITH_SB(sb
),
1977 journal_mark_dirty(&myth
, sb
,
1978 SB_BUFFER_WITH_SB(sb
));
1979 do_journal_end(&myth
, sb
, 1, FLUSH_ALL
);
1984 /* this also catches errors during the do_journal_end above */
1985 if (!error
&& reiserfs_is_journal_aborted(journal
)) {
1986 memset(&myth
, 0, sizeof(myth
));
1987 if (!journal_join_abort(&myth
, sb
, 1)) {
1988 reiserfs_prepare_for_journal(sb
,
1989 SB_BUFFER_WITH_SB(sb
),
1991 journal_mark_dirty(&myth
, sb
,
1992 SB_BUFFER_WITH_SB(sb
));
1993 do_journal_end(&myth
, sb
, 1, FLUSH_ALL
);
1997 reiserfs_mounted_fs_count
--;
1998 /* wait for all commits to finish */
1999 cancel_delayed_work(&SB_JOURNAL(sb
)->j_work
);
2002 * We must release the write lock here because
2003 * the workqueue job (flush_async_commit) needs this lock
2005 reiserfs_write_unlock(sb
);
2006 flush_workqueue(commit_wq
);
2008 if (!reiserfs_mounted_fs_count
) {
2009 destroy_workqueue(commit_wq
);
2012 reiserfs_write_lock(sb
);
2014 free_journal_ram(sb
);
2020 ** call on unmount. flush all journal trans, release all alloc'd ram
2022 int journal_release(struct reiserfs_transaction_handle
*th
,
2023 struct super_block
*sb
)
2025 return do_journal_release(th
, sb
, 0);
2029 ** only call from an error condition inside reiserfs_read_super!
2031 int journal_release_error(struct reiserfs_transaction_handle
*th
,
2032 struct super_block
*sb
)
2034 return do_journal_release(th
, sb
, 1);
2037 /* compares description block with commit block. returns 1 if they differ, 0 if they are the same */
2038 static int journal_compare_desc_commit(struct super_block
*sb
,
2039 struct reiserfs_journal_desc
*desc
,
2040 struct reiserfs_journal_commit
*commit
)
2042 if (get_commit_trans_id(commit
) != get_desc_trans_id(desc
) ||
2043 get_commit_trans_len(commit
) != get_desc_trans_len(desc
) ||
2044 get_commit_trans_len(commit
) > SB_JOURNAL(sb
)->j_trans_max
||
2045 get_commit_trans_len(commit
) <= 0) {
2051 /* returns 0 if it did not find a description block
2052 ** returns -1 if it found a corrupt commit block
2053 ** returns 1 if both desc and commit were valid
2055 static int journal_transaction_is_valid(struct super_block
*sb
,
2056 struct buffer_head
*d_bh
,
2057 unsigned int *oldest_invalid_trans_id
,
2058 unsigned long *newest_mount_id
)
2060 struct reiserfs_journal_desc
*desc
;
2061 struct reiserfs_journal_commit
*commit
;
2062 struct buffer_head
*c_bh
;
2063 unsigned long offset
;
2068 desc
= (struct reiserfs_journal_desc
*)d_bh
->b_data
;
2069 if (get_desc_trans_len(desc
) > 0
2070 && !memcmp(get_journal_desc_magic(d_bh
), JOURNAL_DESC_MAGIC
, 8)) {
2071 if (oldest_invalid_trans_id
&& *oldest_invalid_trans_id
2072 && get_desc_trans_id(desc
) > *oldest_invalid_trans_id
) {
2073 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2074 "journal-986: transaction "
2075 "is valid returning because trans_id %d is greater than "
2076 "oldest_invalid %lu",
2077 get_desc_trans_id(desc
),
2078 *oldest_invalid_trans_id
);
2082 && *newest_mount_id
> get_desc_mount_id(desc
)) {
2083 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2084 "journal-1087: transaction "
2085 "is valid returning because mount_id %d is less than "
2086 "newest_mount_id %lu",
2087 get_desc_mount_id(desc
),
2091 if (get_desc_trans_len(desc
) > SB_JOURNAL(sb
)->j_trans_max
) {
2092 reiserfs_warning(sb
, "journal-2018",
2093 "Bad transaction length %d "
2094 "encountered, ignoring transaction",
2095 get_desc_trans_len(desc
));
2098 offset
= d_bh
->b_blocknr
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
);
2100 /* ok, we have a journal description block, lets see if the transaction was valid */
2103 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2104 ((offset
+ get_desc_trans_len(desc
) +
2105 1) % SB_ONDISK_JOURNAL_SIZE(sb
)));
2108 commit
= (struct reiserfs_journal_commit
*)c_bh
->b_data
;
2109 if (journal_compare_desc_commit(sb
, desc
, commit
)) {
2110 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2111 "journal_transaction_is_valid, commit offset %ld had bad "
2112 "time %d or length %d",
2114 SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2115 get_commit_trans_id(commit
),
2116 get_commit_trans_len(commit
));
2118 if (oldest_invalid_trans_id
) {
2119 *oldest_invalid_trans_id
=
2120 get_desc_trans_id(desc
);
2121 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2123 "transaction_is_valid setting oldest invalid trans_id "
2125 get_desc_trans_id(desc
));
2130 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2131 "journal-1006: found valid "
2132 "transaction start offset %llu, len %d id %d",
2134 SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2135 get_desc_trans_len(desc
),
2136 get_desc_trans_id(desc
));
2143 static void brelse_array(struct buffer_head
**heads
, int num
)
2146 for (i
= 0; i
< num
; i
++) {
2152 ** given the start, and values for the oldest acceptable transactions,
2153 ** this either reads in a replays a transaction, or returns because the transaction
2154 ** is invalid, or too old.
2156 static int journal_read_transaction(struct super_block
*sb
,
2157 unsigned long cur_dblock
,
2158 unsigned long oldest_start
,
2159 unsigned int oldest_trans_id
,
2160 unsigned long newest_mount_id
)
2162 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
2163 struct reiserfs_journal_desc
*desc
;
2164 struct reiserfs_journal_commit
*commit
;
2165 unsigned int trans_id
= 0;
2166 struct buffer_head
*c_bh
;
2167 struct buffer_head
*d_bh
;
2168 struct buffer_head
**log_blocks
= NULL
;
2169 struct buffer_head
**real_blocks
= NULL
;
2170 unsigned int trans_offset
;
2174 d_bh
= journal_bread(sb
, cur_dblock
);
2177 desc
= (struct reiserfs_journal_desc
*)d_bh
->b_data
;
2178 trans_offset
= d_bh
->b_blocknr
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
);
2179 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
, "journal-1037: "
2180 "journal_read_transaction, offset %llu, len %d mount_id %d",
2181 d_bh
->b_blocknr
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2182 get_desc_trans_len(desc
), get_desc_mount_id(desc
));
2183 if (get_desc_trans_id(desc
) < oldest_trans_id
) {
2184 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
, "journal-1039: "
2185 "journal_read_trans skipping because %lu is too old",
2187 SB_ONDISK_JOURNAL_1st_BLOCK(sb
));
2191 if (get_desc_mount_id(desc
) != newest_mount_id
) {
2192 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
, "journal-1146: "
2193 "journal_read_trans skipping because %d is != "
2194 "newest_mount_id %lu", get_desc_mount_id(desc
),
2199 c_bh
= journal_bread(sb
, SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2200 ((trans_offset
+ get_desc_trans_len(desc
) + 1) %
2201 SB_ONDISK_JOURNAL_SIZE(sb
)));
2206 commit
= (struct reiserfs_journal_commit
*)c_bh
->b_data
;
2207 if (journal_compare_desc_commit(sb
, desc
, commit
)) {
2208 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2209 "journal_read_transaction, "
2210 "commit offset %llu had bad time %d or length %d",
2212 SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2213 get_commit_trans_id(commit
),
2214 get_commit_trans_len(commit
));
2219 trans_id
= get_desc_trans_id(desc
);
2220 /* now we know we've got a good transaction, and it was inside the valid time ranges */
2221 log_blocks
= kmalloc(get_desc_trans_len(desc
) *
2222 sizeof(struct buffer_head
*), GFP_NOFS
);
2223 real_blocks
= kmalloc(get_desc_trans_len(desc
) *
2224 sizeof(struct buffer_head
*), GFP_NOFS
);
2225 if (!log_blocks
|| !real_blocks
) {
2230 reiserfs_warning(sb
, "journal-1169",
2231 "kmalloc failed, unable to mount FS");
2234 /* get all the buffer heads */
2235 trans_half
= journal_trans_half(sb
->s_blocksize
);
2236 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2239 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2241 i
) % SB_ONDISK_JOURNAL_SIZE(sb
));
2242 if (i
< trans_half
) {
2245 le32_to_cpu(desc
->j_realblock
[i
]));
2249 le32_to_cpu(commit
->
2250 j_realblock
[i
- trans_half
]));
2252 if (real_blocks
[i
]->b_blocknr
> SB_BLOCK_COUNT(sb
)) {
2253 reiserfs_warning(sb
, "journal-1207",
2254 "REPLAY FAILURE fsck required! "
2255 "Block to replay is outside of "
2259 /* make sure we don't try to replay onto log or reserved area */
2260 if (is_block_in_log_or_reserved_area
2261 (sb
, real_blocks
[i
]->b_blocknr
)) {
2262 reiserfs_warning(sb
, "journal-1204",
2263 "REPLAY FAILURE fsck required! "
2264 "Trying to replay onto a log block");
2266 brelse_array(log_blocks
, i
);
2267 brelse_array(real_blocks
, i
);
2275 /* read in the log blocks, memcpy to the corresponding real block */
2276 ll_rw_block(READ
, get_desc_trans_len(desc
), log_blocks
);
2277 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2279 reiserfs_write_unlock(sb
);
2280 wait_on_buffer(log_blocks
[i
]);
2281 reiserfs_write_lock(sb
);
2283 if (!buffer_uptodate(log_blocks
[i
])) {
2284 reiserfs_warning(sb
, "journal-1212",
2285 "REPLAY FAILURE fsck required! "
2286 "buffer write failed");
2287 brelse_array(log_blocks
+ i
,
2288 get_desc_trans_len(desc
) - i
);
2289 brelse_array(real_blocks
, get_desc_trans_len(desc
));
2296 memcpy(real_blocks
[i
]->b_data
, log_blocks
[i
]->b_data
,
2297 real_blocks
[i
]->b_size
);
2298 set_buffer_uptodate(real_blocks
[i
]);
2299 brelse(log_blocks
[i
]);
2301 /* flush out the real blocks */
2302 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2303 set_buffer_dirty(real_blocks
[i
]);
2304 ll_rw_block(SWRITE
, 1, real_blocks
+ i
);
2306 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2307 wait_on_buffer(real_blocks
[i
]);
2308 if (!buffer_uptodate(real_blocks
[i
])) {
2309 reiserfs_warning(sb
, "journal-1226",
2310 "REPLAY FAILURE, fsck required! "
2311 "buffer write failed");
2312 brelse_array(real_blocks
+ i
,
2313 get_desc_trans_len(desc
) - i
);
2320 brelse(real_blocks
[i
]);
2323 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2324 ((trans_offset
+ get_desc_trans_len(desc
) +
2325 2) % SB_ONDISK_JOURNAL_SIZE(sb
));
2326 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2327 "journal-1095: setting journal " "start to offset %ld",
2328 cur_dblock
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
));
2330 /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2331 journal
->j_start
= cur_dblock
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
);
2332 journal
->j_last_flush_trans_id
= trans_id
;
2333 journal
->j_trans_id
= trans_id
+ 1;
2334 /* check for trans_id overflow */
2335 if (journal
->j_trans_id
== 0)
2336 journal
->j_trans_id
= 10;
2344 /* This function reads blocks starting from block and to max_block of bufsize
2345 size (but no more than BUFNR blocks at a time). This proved to improve
2346 mounting speed on self-rebuilding raid5 arrays at least.
2347 Right now it is only used from journal code. But later we might use it
2349 Note: Do not use journal_getblk/sb_getblk functions here! */
2350 static struct buffer_head
*reiserfs_breada(struct block_device
*dev
,
2351 b_blocknr_t block
, int bufsize
,
2352 b_blocknr_t max_block
)
2354 struct buffer_head
*bhlist
[BUFNR
];
2355 unsigned int blocks
= BUFNR
;
2356 struct buffer_head
*bh
;
2359 bh
= __getblk(dev
, block
, bufsize
);
2360 if (buffer_uptodate(bh
))
2363 if (block
+ BUFNR
> max_block
) {
2364 blocks
= max_block
- block
;
2368 for (i
= 1; i
< blocks
; i
++) {
2369 bh
= __getblk(dev
, block
+ i
, bufsize
);
2370 if (buffer_uptodate(bh
)) {
2376 ll_rw_block(READ
, j
, bhlist
);
2377 for (i
= 1; i
< j
; i
++)
2381 if (buffer_uptodate(bh
))
2388 ** read and replay the log
2389 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2390 ** transaction. This tests that before finding all the transactions in the log, which makes normal mount times fast.
2392 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2394 ** On exit, it sets things up so the first transaction will work correctly.
2396 static int journal_read(struct super_block
*sb
)
2398 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
2399 struct reiserfs_journal_desc
*desc
;
2400 unsigned int oldest_trans_id
= 0;
2401 unsigned int oldest_invalid_trans_id
= 0;
2403 unsigned long oldest_start
= 0;
2404 unsigned long cur_dblock
= 0;
2405 unsigned long newest_mount_id
= 9;
2406 struct buffer_head
*d_bh
;
2407 struct reiserfs_journal_header
*jh
;
2408 int valid_journal_header
= 0;
2409 int replay_count
= 0;
2410 int continue_replay
= 1;
2412 char b
[BDEVNAME_SIZE
];
2414 cur_dblock
= SB_ONDISK_JOURNAL_1st_BLOCK(sb
);
2415 reiserfs_info(sb
, "checking transaction log (%s)\n",
2416 bdevname(journal
->j_dev_bd
, b
));
2417 start
= get_seconds();
2419 /* step 1, read in the journal header block. Check the transaction it says
2420 ** is the first unflushed, and if that transaction is not valid,
2423 journal
->j_header_bh
= journal_bread(sb
,
2424 SB_ONDISK_JOURNAL_1st_BLOCK(sb
)
2425 + SB_ONDISK_JOURNAL_SIZE(sb
));
2426 if (!journal
->j_header_bh
) {
2429 jh
= (struct reiserfs_journal_header
*)(journal
->j_header_bh
->b_data
);
2430 if (le32_to_cpu(jh
->j_first_unflushed_offset
) <
2431 SB_ONDISK_JOURNAL_SIZE(sb
)
2432 && le32_to_cpu(jh
->j_last_flush_trans_id
) > 0) {
2434 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2435 le32_to_cpu(jh
->j_first_unflushed_offset
);
2436 oldest_trans_id
= le32_to_cpu(jh
->j_last_flush_trans_id
) + 1;
2437 newest_mount_id
= le32_to_cpu(jh
->j_mount_id
);
2438 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2439 "journal-1153: found in "
2440 "header: first_unflushed_offset %d, last_flushed_trans_id "
2441 "%lu", le32_to_cpu(jh
->j_first_unflushed_offset
),
2442 le32_to_cpu(jh
->j_last_flush_trans_id
));
2443 valid_journal_header
= 1;
2445 /* now, we try to read the first unflushed offset. If it is not valid,
2446 ** there is nothing more we can do, and it makes no sense to read
2447 ** through the whole log.
2451 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2452 le32_to_cpu(jh
->j_first_unflushed_offset
));
2453 ret
= journal_transaction_is_valid(sb
, d_bh
, NULL
, NULL
);
2455 continue_replay
= 0;
2458 goto start_log_replay
;
2461 if (continue_replay
&& bdev_read_only(sb
->s_bdev
)) {
2462 reiserfs_warning(sb
, "clm-2076",
2463 "device is readonly, unable to replay log");
2467 /* ok, there are transactions that need to be replayed. start with the first log block, find
2468 ** all the valid transactions, and pick out the oldest.
2470 while (continue_replay
2472 (SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2473 SB_ONDISK_JOURNAL_SIZE(sb
))) {
2474 /* Note that it is required for blocksize of primary fs device and journal
2475 device to be the same */
2477 reiserfs_breada(journal
->j_dev_bd
, cur_dblock
,
2479 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2480 SB_ONDISK_JOURNAL_SIZE(sb
));
2482 journal_transaction_is_valid(sb
, d_bh
,
2483 &oldest_invalid_trans_id
,
2486 desc
= (struct reiserfs_journal_desc
*)d_bh
->b_data
;
2487 if (oldest_start
== 0) { /* init all oldest_ values */
2488 oldest_trans_id
= get_desc_trans_id(desc
);
2489 oldest_start
= d_bh
->b_blocknr
;
2490 newest_mount_id
= get_desc_mount_id(desc
);
2491 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2492 "journal-1179: Setting "
2493 "oldest_start to offset %llu, trans_id %lu",
2495 SB_ONDISK_JOURNAL_1st_BLOCK
2496 (sb
), oldest_trans_id
);
2497 } else if (oldest_trans_id
> get_desc_trans_id(desc
)) {
2498 /* one we just read was older */
2499 oldest_trans_id
= get_desc_trans_id(desc
);
2500 oldest_start
= d_bh
->b_blocknr
;
2501 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2502 "journal-1180: Resetting "
2503 "oldest_start to offset %lu, trans_id %lu",
2505 SB_ONDISK_JOURNAL_1st_BLOCK
2506 (sb
), oldest_trans_id
);
2508 if (newest_mount_id
< get_desc_mount_id(desc
)) {
2509 newest_mount_id
= get_desc_mount_id(desc
);
2510 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2511 "journal-1299: Setting "
2512 "newest_mount_id to %d",
2513 get_desc_mount_id(desc
));
2515 cur_dblock
+= get_desc_trans_len(desc
) + 2;
2523 cur_dblock
= oldest_start
;
2524 if (oldest_trans_id
) {
2525 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2526 "journal-1206: Starting replay "
2527 "from offset %llu, trans_id %lu",
2528 cur_dblock
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2533 while (continue_replay
&& oldest_trans_id
> 0) {
2535 journal_read_transaction(sb
, cur_dblock
, oldest_start
,
2536 oldest_trans_id
, newest_mount_id
);
2539 } else if (ret
!= 0) {
2543 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) + journal
->j_start
;
2545 if (cur_dblock
== oldest_start
)
2549 if (oldest_trans_id
== 0) {
2550 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2551 "journal-1225: No valid " "transactions found");
2553 /* j_start does not get set correctly if we don't replay any transactions.
2554 ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2555 ** copy the trans_id from the header
2557 if (valid_journal_header
&& replay_count
== 0) {
2558 journal
->j_start
= le32_to_cpu(jh
->j_first_unflushed_offset
);
2559 journal
->j_trans_id
=
2560 le32_to_cpu(jh
->j_last_flush_trans_id
) + 1;
2561 /* check for trans_id overflow */
2562 if (journal
->j_trans_id
== 0)
2563 journal
->j_trans_id
= 10;
2564 journal
->j_last_flush_trans_id
=
2565 le32_to_cpu(jh
->j_last_flush_trans_id
);
2566 journal
->j_mount_id
= le32_to_cpu(jh
->j_mount_id
) + 1;
2568 journal
->j_mount_id
= newest_mount_id
+ 1;
2570 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
, "journal-1299: Setting "
2571 "newest_mount_id to %lu", journal
->j_mount_id
);
2572 journal
->j_first_unflushed_offset
= journal
->j_start
;
2573 if (replay_count
> 0) {
2575 "replayed %d transactions in %lu seconds\n",
2576 replay_count
, get_seconds() - start
);
2578 if (!bdev_read_only(sb
->s_bdev
) &&
2579 _update_journal_header_block(sb
, journal
->j_start
,
2580 journal
->j_last_flush_trans_id
)) {
2581 /* replay failed, caller must call free_journal_ram and abort
2589 static struct reiserfs_journal_list
*alloc_journal_list(struct super_block
*s
)
2591 struct reiserfs_journal_list
*jl
;
2592 jl
= kzalloc(sizeof(struct reiserfs_journal_list
),
2593 GFP_NOFS
| __GFP_NOFAIL
);
2594 INIT_LIST_HEAD(&jl
->j_list
);
2595 INIT_LIST_HEAD(&jl
->j_working_list
);
2596 INIT_LIST_HEAD(&jl
->j_tail_bh_list
);
2597 INIT_LIST_HEAD(&jl
->j_bh_list
);
2598 mutex_init(&jl
->j_commit_mutex
);
2599 SB_JOURNAL(s
)->j_num_lists
++;
2600 get_journal_list(jl
);
2604 static void journal_list_init(struct super_block
*sb
)
2606 SB_JOURNAL(sb
)->j_current_jl
= alloc_journal_list(sb
);
2609 static int release_journal_dev(struct super_block
*super
,
2610 struct reiserfs_journal
*journal
)
2616 if (journal
->j_dev_bd
!= NULL
) {
2617 if (journal
->j_dev_bd
->bd_dev
!= super
->s_dev
)
2618 bd_release(journal
->j_dev_bd
);
2619 result
= blkdev_put(journal
->j_dev_bd
, journal
->j_dev_mode
);
2620 journal
->j_dev_bd
= NULL
;
2624 reiserfs_warning(super
, "sh-457",
2625 "Cannot release journal device: %i", result
);
2630 static int journal_init_dev(struct super_block
*super
,
2631 struct reiserfs_journal
*journal
,
2632 const char *jdev_name
)
2636 fmode_t blkdev_mode
= FMODE_READ
| FMODE_WRITE
;
2637 char b
[BDEVNAME_SIZE
];
2641 journal
->j_dev_bd
= NULL
;
2642 jdev
= SB_ONDISK_JOURNAL_DEVICE(super
) ?
2643 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super
)) : super
->s_dev
;
2645 if (bdev_read_only(super
->s_bdev
))
2646 blkdev_mode
= FMODE_READ
;
2648 /* there is no "jdev" option and journal is on separate device */
2649 if ((!jdev_name
|| !jdev_name
[0])) {
2650 journal
->j_dev_bd
= open_by_devnum(jdev
, blkdev_mode
);
2651 journal
->j_dev_mode
= blkdev_mode
;
2652 if (IS_ERR(journal
->j_dev_bd
)) {
2653 result
= PTR_ERR(journal
->j_dev_bd
);
2654 journal
->j_dev_bd
= NULL
;
2655 reiserfs_warning(super
, "sh-458",
2656 "cannot init journal device '%s': %i",
2657 __bdevname(jdev
, b
), result
);
2659 } else if (jdev
!= super
->s_dev
) {
2660 result
= bd_claim(journal
->j_dev_bd
, journal
);
2662 blkdev_put(journal
->j_dev_bd
, blkdev_mode
);
2666 set_blocksize(journal
->j_dev_bd
, super
->s_blocksize
);
2672 journal
->j_dev_mode
= blkdev_mode
;
2673 journal
->j_dev_bd
= open_bdev_exclusive(jdev_name
,
2674 blkdev_mode
, journal
);
2675 if (IS_ERR(journal
->j_dev_bd
)) {
2676 result
= PTR_ERR(journal
->j_dev_bd
);
2677 journal
->j_dev_bd
= NULL
;
2678 reiserfs_warning(super
,
2679 "journal_init_dev: Cannot open '%s': %i",
2684 set_blocksize(journal
->j_dev_bd
, super
->s_blocksize
);
2685 reiserfs_info(super
,
2686 "journal_init_dev: journal device: %s\n",
2687 bdevname(journal
->j_dev_bd
, b
));
2692 * When creating/tuning a file system user can assign some
2693 * journal params within boundaries which depend on the ratio
2694 * blocksize/standard_blocksize.
2696 * For blocks >= standard_blocksize transaction size should
2697 * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2698 * then JOURNAL_TRANS_MAX_DEFAULT.
2700 * For blocks < standard_blocksize these boundaries should be
2701 * decreased proportionally.
2703 #define REISERFS_STANDARD_BLKSIZE (4096)
2705 static int check_advise_trans_params(struct super_block
*sb
,
2706 struct reiserfs_journal
*journal
)
2708 if (journal
->j_trans_max
) {
2709 /* Non-default journal params.
2710 Do sanity check for them. */
2712 if (sb
->s_blocksize
< REISERFS_STANDARD_BLKSIZE
)
2713 ratio
= REISERFS_STANDARD_BLKSIZE
/ sb
->s_blocksize
;
2715 if (journal
->j_trans_max
> JOURNAL_TRANS_MAX_DEFAULT
/ ratio
||
2716 journal
->j_trans_max
< JOURNAL_TRANS_MIN_DEFAULT
/ ratio
||
2717 SB_ONDISK_JOURNAL_SIZE(sb
) / journal
->j_trans_max
<
2718 JOURNAL_MIN_RATIO
) {
2719 reiserfs_warning(sb
, "sh-462",
2720 "bad transaction max size (%u). "
2721 "FSCK?", journal
->j_trans_max
);
2724 if (journal
->j_max_batch
!= (journal
->j_trans_max
) *
2725 JOURNAL_MAX_BATCH_DEFAULT
/JOURNAL_TRANS_MAX_DEFAULT
) {
2726 reiserfs_warning(sb
, "sh-463",
2727 "bad transaction max batch (%u). "
2728 "FSCK?", journal
->j_max_batch
);
2732 /* Default journal params.
2733 The file system was created by old version
2734 of mkreiserfs, so some fields contain zeros,
2735 and we need to advise proper values for them */
2736 if (sb
->s_blocksize
!= REISERFS_STANDARD_BLKSIZE
) {
2737 reiserfs_warning(sb
, "sh-464", "bad blocksize (%u)",
2741 journal
->j_trans_max
= JOURNAL_TRANS_MAX_DEFAULT
;
2742 journal
->j_max_batch
= JOURNAL_MAX_BATCH_DEFAULT
;
2743 journal
->j_max_commit_age
= JOURNAL_MAX_COMMIT_AGE
;
2749 ** must be called once on fs mount. calls journal_read for you
2751 int journal_init(struct super_block
*sb
, const char *j_dev_name
,
2752 int old_format
, unsigned int commit_max_age
)
2754 int num_cnodes
= SB_ONDISK_JOURNAL_SIZE(sb
) * 2;
2755 struct buffer_head
*bhjh
;
2756 struct reiserfs_super_block
*rs
;
2757 struct reiserfs_journal_header
*jh
;
2758 struct reiserfs_journal
*journal
;
2759 struct reiserfs_journal_list
*jl
;
2760 char b
[BDEVNAME_SIZE
];
2762 journal
= SB_JOURNAL(sb
) = vmalloc(sizeof(struct reiserfs_journal
));
2764 reiserfs_warning(sb
, "journal-1256",
2765 "unable to get memory for journal structure");
2768 memset(journal
, 0, sizeof(struct reiserfs_journal
));
2769 INIT_LIST_HEAD(&journal
->j_bitmap_nodes
);
2770 INIT_LIST_HEAD(&journal
->j_prealloc_list
);
2771 INIT_LIST_HEAD(&journal
->j_working_list
);
2772 INIT_LIST_HEAD(&journal
->j_journal_list
);
2773 journal
->j_persistent_trans
= 0;
2774 if (reiserfs_allocate_list_bitmaps(sb
,
2775 journal
->j_list_bitmap
,
2776 reiserfs_bmap_count(sb
)))
2777 goto free_and_return
;
2778 allocate_bitmap_nodes(sb
);
2780 /* reserved for journal area support */
2781 SB_JOURNAL_1st_RESERVED_BLOCK(sb
) = (old_format
?
2782 REISERFS_OLD_DISK_OFFSET_IN_BYTES
2784 reiserfs_bmap_count(sb
) +
2786 REISERFS_DISK_OFFSET_IN_BYTES
/
2787 sb
->s_blocksize
+ 2);
2789 /* Sanity check to see is the standard journal fitting withing first bitmap
2790 (actual for small blocksizes) */
2791 if (!SB_ONDISK_JOURNAL_DEVICE(sb
) &&
2792 (SB_JOURNAL_1st_RESERVED_BLOCK(sb
) +
2793 SB_ONDISK_JOURNAL_SIZE(sb
) > sb
->s_blocksize
* 8)) {
2794 reiserfs_warning(sb
, "journal-1393",
2795 "journal does not fit for area addressed "
2796 "by first of bitmap blocks. It starts at "
2797 "%u and its size is %u. Block size %ld",
2798 SB_JOURNAL_1st_RESERVED_BLOCK(sb
),
2799 SB_ONDISK_JOURNAL_SIZE(sb
),
2801 goto free_and_return
;
2805 * We need to unlock here to avoid creating the following
2807 * reiserfs_lock -> sysfs_mutex
2808 * Because the reiserfs mmap path creates the following dependency:
2809 * mm->mmap -> reiserfs_lock, hence we have
2810 * mm->mmap -> reiserfs_lock ->sysfs_mutex
2811 * This would ends up in a circular dependency with sysfs readdir path
2812 * which does sysfs_mutex -> mm->mmap_sem
2813 * This is fine because the reiserfs lock is useless in mount path,
2814 * at least until we call journal_begin. We keep it for paranoid
2817 reiserfs_write_unlock(sb
);
2818 if (journal_init_dev(sb
, journal
, j_dev_name
) != 0) {
2819 reiserfs_write_lock(sb
);
2820 reiserfs_warning(sb
, "sh-462",
2821 "unable to initialize jornal device");
2822 goto free_and_return
;
2824 reiserfs_write_lock(sb
);
2826 rs
= SB_DISK_SUPER_BLOCK(sb
);
2828 /* read journal header */
2829 bhjh
= journal_bread(sb
,
2830 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2831 SB_ONDISK_JOURNAL_SIZE(sb
));
2833 reiserfs_warning(sb
, "sh-459",
2834 "unable to read journal header");
2835 goto free_and_return
;
2837 jh
= (struct reiserfs_journal_header
*)(bhjh
->b_data
);
2839 /* make sure that journal matches to the super block */
2840 if (is_reiserfs_jr(rs
)
2841 && (le32_to_cpu(jh
->jh_journal
.jp_journal_magic
) !=
2842 sb_jp_journal_magic(rs
))) {
2843 reiserfs_warning(sb
, "sh-460",
2844 "journal header magic %x (device %s) does "
2845 "not match to magic found in super block %x",
2846 jh
->jh_journal
.jp_journal_magic
,
2847 bdevname(journal
->j_dev_bd
, b
),
2848 sb_jp_journal_magic(rs
));
2850 goto free_and_return
;
2853 journal
->j_trans_max
= le32_to_cpu(jh
->jh_journal
.jp_journal_trans_max
);
2854 journal
->j_max_batch
= le32_to_cpu(jh
->jh_journal
.jp_journal_max_batch
);
2855 journal
->j_max_commit_age
=
2856 le32_to_cpu(jh
->jh_journal
.jp_journal_max_commit_age
);
2857 journal
->j_max_trans_age
= JOURNAL_MAX_TRANS_AGE
;
2859 if (check_advise_trans_params(sb
, journal
) != 0)
2860 goto free_and_return
;
2861 journal
->j_default_max_commit_age
= journal
->j_max_commit_age
;
2863 if (commit_max_age
!= 0) {
2864 journal
->j_max_commit_age
= commit_max_age
;
2865 journal
->j_max_trans_age
= commit_max_age
;
2868 reiserfs_info(sb
, "journal params: device %s, size %u, "
2869 "journal first block %u, max trans len %u, max batch %u, "
2870 "max commit age %u, max trans age %u\n",
2871 bdevname(journal
->j_dev_bd
, b
),
2872 SB_ONDISK_JOURNAL_SIZE(sb
),
2873 SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2874 journal
->j_trans_max
,
2875 journal
->j_max_batch
,
2876 journal
->j_max_commit_age
, journal
->j_max_trans_age
);
2880 journal
->j_list_bitmap_index
= 0;
2881 journal_list_init(sb
);
2883 memset(journal
->j_list_hash_table
, 0,
2884 JOURNAL_HASH_SIZE
* sizeof(struct reiserfs_journal_cnode
*));
2886 INIT_LIST_HEAD(&journal
->j_dirty_buffers
);
2887 spin_lock_init(&journal
->j_dirty_buffers_lock
);
2889 journal
->j_start
= 0;
2891 journal
->j_len_alloc
= 0;
2892 atomic_set(&(journal
->j_wcount
), 0);
2893 atomic_set(&(journal
->j_async_throttle
), 0);
2894 journal
->j_bcount
= 0;
2895 journal
->j_trans_start_time
= 0;
2896 journal
->j_last
= NULL
;
2897 journal
->j_first
= NULL
;
2898 init_waitqueue_head(&(journal
->j_join_wait
));
2899 mutex_init(&journal
->j_mutex
);
2900 mutex_init(&journal
->j_flush_mutex
);
2902 journal
->j_trans_id
= 10;
2903 journal
->j_mount_id
= 10;
2904 journal
->j_state
= 0;
2905 atomic_set(&(journal
->j_jlock
), 0);
2906 journal
->j_cnode_free_list
= allocate_cnodes(num_cnodes
);
2907 journal
->j_cnode_free_orig
= journal
->j_cnode_free_list
;
2908 journal
->j_cnode_free
= journal
->j_cnode_free_list
? num_cnodes
: 0;
2909 journal
->j_cnode_used
= 0;
2910 journal
->j_must_wait
= 0;
2912 if (journal
->j_cnode_free
== 0) {
2913 reiserfs_warning(sb
, "journal-2004", "Journal cnode memory "
2914 "allocation failed (%ld bytes). Journal is "
2915 "too large for available memory. Usually "
2916 "this is due to a journal that is too large.",
2917 sizeof (struct reiserfs_journal_cnode
) * num_cnodes
);
2918 goto free_and_return
;
2921 init_journal_hash(sb
);
2922 jl
= journal
->j_current_jl
;
2923 jl
->j_list_bitmap
= get_list_bitmap(sb
, jl
);
2924 if (!jl
->j_list_bitmap
) {
2925 reiserfs_warning(sb
, "journal-2005",
2926 "get_list_bitmap failed for journal list 0");
2927 goto free_and_return
;
2929 if (journal_read(sb
) < 0) {
2930 reiserfs_warning(sb
, "reiserfs-2006",
2931 "Replay Failure, unable to mount");
2932 goto free_and_return
;
2935 reiserfs_mounted_fs_count
++;
2936 if (reiserfs_mounted_fs_count
<= 1) {
2937 reiserfs_write_unlock(sb
);
2938 commit_wq
= create_workqueue("reiserfs");
2939 reiserfs_write_lock(sb
);
2942 INIT_DELAYED_WORK(&journal
->j_work
, flush_async_commits
);
2943 journal
->j_work_sb
= sb
;
2946 free_journal_ram(sb
);
2951 ** test for a polite end of the current transaction. Used by file_write, and should
2952 ** be used by delete to make sure they don't write more than can fit inside a single
2955 int journal_transaction_should_end(struct reiserfs_transaction_handle
*th
,
2958 struct reiserfs_journal
*journal
= SB_JOURNAL(th
->t_super
);
2959 time_t now
= get_seconds();
2960 /* cannot restart while nested */
2961 BUG_ON(!th
->t_trans_id
);
2962 if (th
->t_refcount
> 1)
2964 if (journal
->j_must_wait
> 0 ||
2965 (journal
->j_len_alloc
+ new_alloc
) >= journal
->j_max_batch
||
2966 atomic_read(&(journal
->j_jlock
)) ||
2967 (now
- journal
->j_trans_start_time
) > journal
->j_max_trans_age
||
2968 journal
->j_cnode_free
< (journal
->j_trans_max
* 3)) {
2971 /* protected by the BKL here */
2972 journal
->j_len_alloc
+= new_alloc
;
2973 th
->t_blocks_allocated
+= new_alloc
;
2977 /* this must be called inside a transaction, and requires the
2978 ** kernel_lock to be held
2980 void reiserfs_block_writes(struct reiserfs_transaction_handle
*th
)
2982 struct reiserfs_journal
*journal
= SB_JOURNAL(th
->t_super
);
2983 BUG_ON(!th
->t_trans_id
);
2984 journal
->j_must_wait
= 1;
2985 set_bit(J_WRITERS_BLOCKED
, &journal
->j_state
);
2989 /* this must be called without a transaction started, and does not
2992 void reiserfs_allow_writes(struct super_block
*s
)
2994 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2995 clear_bit(J_WRITERS_BLOCKED
, &journal
->j_state
);
2996 wake_up(&journal
->j_join_wait
);
2999 /* this must be called without a transaction started, and does not
3002 void reiserfs_wait_on_write_block(struct super_block
*s
)
3004 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
3005 wait_event(journal
->j_join_wait
,
3006 !test_bit(J_WRITERS_BLOCKED
, &journal
->j_state
));
3009 static void queue_log_writer(struct super_block
*s
)
3012 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
3013 set_bit(J_WRITERS_QUEUED
, &journal
->j_state
);
3016 * we don't want to use wait_event here because
3017 * we only want to wait once.
3019 init_waitqueue_entry(&wait
, current
);
3020 add_wait_queue(&journal
->j_join_wait
, &wait
);
3021 set_current_state(TASK_UNINTERRUPTIBLE
);
3022 if (test_bit(J_WRITERS_QUEUED
, &journal
->j_state
)) {
3023 reiserfs_write_unlock(s
);
3025 reiserfs_write_lock(s
);
3027 __set_current_state(TASK_RUNNING
);
3028 remove_wait_queue(&journal
->j_join_wait
, &wait
);
3031 static void wake_queued_writers(struct super_block
*s
)
3033 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
3034 if (test_and_clear_bit(J_WRITERS_QUEUED
, &journal
->j_state
))
3035 wake_up(&journal
->j_join_wait
);
3038 static void let_transaction_grow(struct super_block
*sb
, unsigned int trans_id
)
3040 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3041 unsigned long bcount
= journal
->j_bcount
;
3043 reiserfs_write_unlock(sb
);
3044 schedule_timeout_uninterruptible(1);
3045 reiserfs_write_lock(sb
);
3046 journal
->j_current_jl
->j_state
|= LIST_COMMIT_PENDING
;
3047 while ((atomic_read(&journal
->j_wcount
) > 0 ||
3048 atomic_read(&journal
->j_jlock
)) &&
3049 journal
->j_trans_id
== trans_id
) {
3050 queue_log_writer(sb
);
3052 if (journal
->j_trans_id
!= trans_id
)
3054 if (bcount
== journal
->j_bcount
)
3056 bcount
= journal
->j_bcount
;
3060 /* join == true if you must join an existing transaction.
3061 ** join == false if you can deal with waiting for others to finish
3063 ** this will block until the transaction is joinable. send the number of blocks you
3064 ** expect to use in nblocks.
3066 static int do_journal_begin_r(struct reiserfs_transaction_handle
*th
,
3067 struct super_block
*sb
, unsigned long nblocks
,
3070 time_t now
= get_seconds();
3071 unsigned int old_trans_id
;
3072 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3073 struct reiserfs_transaction_handle myth
;
3074 int sched_count
= 0;
3077 reiserfs_check_lock_depth(sb
, "journal_begin");
3078 BUG_ON(nblocks
> journal
->j_trans_max
);
3080 PROC_INFO_INC(sb
, journal
.journal_being
);
3081 /* set here for journal_join */
3087 if (join
!= JBEGIN_ABORT
&& reiserfs_is_journal_aborted(journal
)) {
3089 retval
= journal
->j_errno
;
3092 journal
->j_bcount
++;
3094 if (test_bit(J_WRITERS_BLOCKED
, &journal
->j_state
)) {
3096 reiserfs_write_unlock(sb
);
3097 reiserfs_wait_on_write_block(sb
);
3098 reiserfs_write_lock(sb
);
3099 PROC_INFO_INC(sb
, journal
.journal_relock_writers
);
3102 now
= get_seconds();
3104 /* if there is no room in the journal OR
3105 ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
3106 ** we don't sleep if there aren't other writers
3109 if ((!join
&& journal
->j_must_wait
> 0) ||
3111 && (journal
->j_len_alloc
+ nblocks
+ 2) >= journal
->j_max_batch
)
3112 || (!join
&& atomic_read(&journal
->j_wcount
) > 0
3113 && journal
->j_trans_start_time
> 0
3114 && (now
- journal
->j_trans_start_time
) >
3115 journal
->j_max_trans_age
) || (!join
3116 && atomic_read(&journal
->j_jlock
))
3117 || (!join
&& journal
->j_cnode_free
< (journal
->j_trans_max
* 3))) {
3119 old_trans_id
= journal
->j_trans_id
;
3120 unlock_journal(sb
); /* allow others to finish this transaction */
3122 if (!join
&& (journal
->j_len_alloc
+ nblocks
+ 2) >=
3123 journal
->j_max_batch
&&
3124 ((journal
->j_len
+ nblocks
+ 2) * 100) <
3125 (journal
->j_len_alloc
* 75)) {
3126 if (atomic_read(&journal
->j_wcount
) > 10) {
3128 queue_log_writer(sb
);
3132 /* don't mess with joining the transaction if all we have to do is
3133 * wait for someone else to do a commit
3135 if (atomic_read(&journal
->j_jlock
)) {
3136 while (journal
->j_trans_id
== old_trans_id
&&
3137 atomic_read(&journal
->j_jlock
)) {
3138 queue_log_writer(sb
);
3142 retval
= journal_join(&myth
, sb
, 1);
3146 /* someone might have ended the transaction while we joined */
3147 if (old_trans_id
!= journal
->j_trans_id
) {
3148 retval
= do_journal_end(&myth
, sb
, 1, 0);
3150 retval
= do_journal_end(&myth
, sb
, 1, COMMIT_NOW
);
3156 PROC_INFO_INC(sb
, journal
.journal_relock_wcount
);
3159 /* we are the first writer, set trans_id */
3160 if (journal
->j_trans_start_time
== 0) {
3161 journal
->j_trans_start_time
= get_seconds();
3163 atomic_inc(&(journal
->j_wcount
));
3164 journal
->j_len_alloc
+= nblocks
;
3165 th
->t_blocks_logged
= 0;
3166 th
->t_blocks_allocated
= nblocks
;
3167 th
->t_trans_id
= journal
->j_trans_id
;
3169 INIT_LIST_HEAD(&th
->t_list
);
3174 memset(th
, 0, sizeof(*th
));
3175 /* Re-set th->t_super, so we can properly keep track of how many
3176 * persistent transactions there are. We need to do this so if this
3177 * call is part of a failed restart_transaction, we can free it later */
3182 struct reiserfs_transaction_handle
*reiserfs_persistent_transaction(struct
3188 struct reiserfs_transaction_handle
*th
;
3190 /* if we're nesting into an existing transaction. It will be
3191 ** persistent on its own
3193 if (reiserfs_transaction_running(s
)) {
3194 th
= current
->journal_info
;
3196 BUG_ON(th
->t_refcount
< 2);
3200 th
= kmalloc(sizeof(struct reiserfs_transaction_handle
), GFP_NOFS
);
3203 ret
= journal_begin(th
, s
, nblocks
);
3209 SB_JOURNAL(s
)->j_persistent_trans
++;
3213 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle
*th
)
3215 struct super_block
*s
= th
->t_super
;
3218 ret
= journal_end(th
, th
->t_super
, th
->t_blocks_allocated
);
3221 if (th
->t_refcount
== 0) {
3222 SB_JOURNAL(s
)->j_persistent_trans
--;
3228 static int journal_join(struct reiserfs_transaction_handle
*th
,
3229 struct super_block
*sb
, unsigned long nblocks
)
3231 struct reiserfs_transaction_handle
*cur_th
= current
->journal_info
;
3233 /* this keeps do_journal_end from NULLing out the current->journal_info
3236 th
->t_handle_save
= cur_th
;
3237 BUG_ON(cur_th
&& cur_th
->t_refcount
> 1);
3238 return do_journal_begin_r(th
, sb
, nblocks
, JBEGIN_JOIN
);
3241 int journal_join_abort(struct reiserfs_transaction_handle
*th
,
3242 struct super_block
*sb
, unsigned long nblocks
)
3244 struct reiserfs_transaction_handle
*cur_th
= current
->journal_info
;
3246 /* this keeps do_journal_end from NULLing out the current->journal_info
3249 th
->t_handle_save
= cur_th
;
3250 BUG_ON(cur_th
&& cur_th
->t_refcount
> 1);
3251 return do_journal_begin_r(th
, sb
, nblocks
, JBEGIN_ABORT
);
3254 int journal_begin(struct reiserfs_transaction_handle
*th
,
3255 struct super_block
*sb
, unsigned long nblocks
)
3257 struct reiserfs_transaction_handle
*cur_th
= current
->journal_info
;
3260 th
->t_handle_save
= NULL
;
3262 /* we are nesting into the current transaction */
3263 if (cur_th
->t_super
== sb
) {
3264 BUG_ON(!cur_th
->t_refcount
);
3265 cur_th
->t_refcount
++;
3266 memcpy(th
, cur_th
, sizeof(*th
));
3267 if (th
->t_refcount
<= 1)
3268 reiserfs_warning(sb
, "reiserfs-2005",
3269 "BAD: refcount <= 1, but "
3270 "journal_info != 0");
3273 /* we've ended up with a handle from a different filesystem.
3274 ** save it and restore on journal_end. This should never
3277 reiserfs_warning(sb
, "clm-2100",
3278 "nesting info a different FS");
3279 th
->t_handle_save
= current
->journal_info
;
3280 current
->journal_info
= th
;
3283 current
->journal_info
= th
;
3285 ret
= do_journal_begin_r(th
, sb
, nblocks
, JBEGIN_REG
);
3286 BUG_ON(current
->journal_info
!= th
);
3288 /* I guess this boils down to being the reciprocal of clm-2100 above.
3289 * If do_journal_begin_r fails, we need to put it back, since journal_end
3290 * won't be called to do it. */
3292 current
->journal_info
= th
->t_handle_save
;
3294 BUG_ON(!th
->t_refcount
);
3300 ** puts bh into the current transaction. If it was already there, reorders removes the
3301 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3303 ** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the
3304 ** transaction is committed.
3306 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3308 int journal_mark_dirty(struct reiserfs_transaction_handle
*th
,
3309 struct super_block
*sb
, struct buffer_head
*bh
)
3311 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3312 struct reiserfs_journal_cnode
*cn
= NULL
;
3313 int count_already_incd
= 0;
3315 BUG_ON(!th
->t_trans_id
);
3317 PROC_INFO_INC(sb
, journal
.mark_dirty
);
3318 if (th
->t_trans_id
!= journal
->j_trans_id
) {
3319 reiserfs_panic(th
->t_super
, "journal-1577",
3320 "handle trans id %ld != current trans id %ld",
3321 th
->t_trans_id
, journal
->j_trans_id
);
3326 prepared
= test_clear_buffer_journal_prepared(bh
);
3327 clear_buffer_journal_restore_dirty(bh
);
3328 /* already in this transaction, we are done */
3329 if (buffer_journaled(bh
)) {
3330 PROC_INFO_INC(sb
, journal
.mark_dirty_already
);
3334 /* this must be turned into a panic instead of a warning. We can't allow
3335 ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3336 ** could get to disk too early. NOT GOOD.
3338 if (!prepared
|| buffer_dirty(bh
)) {
3339 reiserfs_warning(sb
, "journal-1777",
3340 "buffer %llu bad state "
3341 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3342 (unsigned long long)bh
->b_blocknr
,
3343 prepared
? ' ' : '!',
3344 buffer_locked(bh
) ? ' ' : '!',
3345 buffer_dirty(bh
) ? ' ' : '!',
3346 buffer_journal_dirty(bh
) ? ' ' : '!');
3349 if (atomic_read(&(journal
->j_wcount
)) <= 0) {
3350 reiserfs_warning(sb
, "journal-1409",
3351 "returning because j_wcount was %d",
3352 atomic_read(&(journal
->j_wcount
)));
3355 /* this error means I've screwed up, and we've overflowed the transaction.
3356 ** Nothing can be done here, except make the FS readonly or panic.
3358 if (journal
->j_len
>= journal
->j_trans_max
) {
3359 reiserfs_panic(th
->t_super
, "journal-1413",
3360 "j_len (%lu) is too big",
3364 if (buffer_journal_dirty(bh
)) {
3365 count_already_incd
= 1;
3366 PROC_INFO_INC(sb
, journal
.mark_dirty_notjournal
);
3367 clear_buffer_journal_dirty(bh
);
3370 if (journal
->j_len
> journal
->j_len_alloc
) {
3371 journal
->j_len_alloc
= journal
->j_len
+ JOURNAL_PER_BALANCE_CNT
;
3374 set_buffer_journaled(bh
);
3376 /* now put this guy on the end */
3380 reiserfs_panic(sb
, "journal-4", "get_cnode failed!");
3383 if (th
->t_blocks_logged
== th
->t_blocks_allocated
) {
3384 th
->t_blocks_allocated
+= JOURNAL_PER_BALANCE_CNT
;
3385 journal
->j_len_alloc
+= JOURNAL_PER_BALANCE_CNT
;
3387 th
->t_blocks_logged
++;
3391 cn
->blocknr
= bh
->b_blocknr
;
3394 insert_journal_hash(journal
->j_hash_table
, cn
);
3395 if (!count_already_incd
) {
3400 cn
->prev
= journal
->j_last
;
3402 if (journal
->j_last
) {
3403 journal
->j_last
->next
= cn
;
3404 journal
->j_last
= cn
;
3406 journal
->j_first
= cn
;
3407 journal
->j_last
= cn
;
3412 int journal_end(struct reiserfs_transaction_handle
*th
,
3413 struct super_block
*sb
, unsigned long nblocks
)
3415 if (!current
->journal_info
&& th
->t_refcount
> 1)
3416 reiserfs_warning(sb
, "REISER-NESTING",
3417 "th NULL, refcount %d", th
->t_refcount
);
3419 if (!th
->t_trans_id
) {
3425 if (th
->t_refcount
> 0) {
3426 struct reiserfs_transaction_handle
*cur_th
=
3427 current
->journal_info
;
3429 /* we aren't allowed to close a nested transaction on a different
3430 ** filesystem from the one in the task struct
3432 BUG_ON(cur_th
->t_super
!= th
->t_super
);
3435 memcpy(current
->journal_info
, th
, sizeof(*th
));
3440 return do_journal_end(th
, sb
, nblocks
, 0);
3444 /* removes from the current transaction, relsing and descrementing any counters.
3445 ** also files the removed buffer directly onto the clean list
3447 ** called by journal_mark_freed when a block has been deleted
3449 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3451 static int remove_from_transaction(struct super_block
*sb
,
3452 b_blocknr_t blocknr
, int already_cleaned
)
3454 struct buffer_head
*bh
;
3455 struct reiserfs_journal_cnode
*cn
;
3456 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3459 cn
= get_journal_hash_dev(sb
, journal
->j_hash_table
, blocknr
);
3460 if (!cn
|| !cn
->bh
) {
3465 cn
->prev
->next
= cn
->next
;
3468 cn
->next
->prev
= cn
->prev
;
3470 if (cn
== journal
->j_first
) {
3471 journal
->j_first
= cn
->next
;
3473 if (cn
== journal
->j_last
) {
3474 journal
->j_last
= cn
->prev
;
3477 remove_journal_hash(sb
, journal
->j_hash_table
, NULL
,
3479 clear_buffer_journaled(bh
); /* don't log this one */
3481 if (!already_cleaned
) {
3482 clear_buffer_journal_dirty(bh
);
3483 clear_buffer_dirty(bh
);
3484 clear_buffer_journal_test(bh
);
3486 if (atomic_read(&(bh
->b_count
)) < 0) {
3487 reiserfs_warning(sb
, "journal-1752",
3493 journal
->j_len_alloc
--;
3499 ** for any cnode in a journal list, it can only be dirtied of all the
3500 ** transactions that include it are committed to disk.
3501 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3502 ** and 0 if you aren't
3504 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3505 ** blocks for a given transaction on disk
3508 static int can_dirty(struct reiserfs_journal_cnode
*cn
)
3510 struct super_block
*sb
= cn
->sb
;
3511 b_blocknr_t blocknr
= cn
->blocknr
;
3512 struct reiserfs_journal_cnode
*cur
= cn
->hprev
;
3515 /* first test hprev. These are all newer than cn, so any node here
3516 ** with the same block number and dev means this node can't be sent
3517 ** to disk right now.
3519 while (cur
&& can_dirty
) {
3520 if (cur
->jlist
&& cur
->bh
&& cur
->blocknr
&& cur
->sb
== sb
&&
3521 cur
->blocknr
== blocknr
) {
3526 /* then test hnext. These are all older than cn. As long as they
3527 ** are committed to the log, it is safe to write cn to disk
3530 while (cur
&& can_dirty
) {
3531 if (cur
->jlist
&& cur
->jlist
->j_len
> 0 &&
3532 atomic_read(&(cur
->jlist
->j_commit_left
)) > 0 && cur
->bh
&&
3533 cur
->blocknr
&& cur
->sb
== sb
&& cur
->blocknr
== blocknr
) {
3541 /* syncs the commit blocks, but does not force the real buffers to disk
3542 ** will wait until the current transaction is done/committed before returning
3544 int journal_end_sync(struct reiserfs_transaction_handle
*th
,
3545 struct super_block
*sb
, unsigned long nblocks
)
3547 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3549 BUG_ON(!th
->t_trans_id
);
3550 /* you can sync while nested, very, very bad */
3551 BUG_ON(th
->t_refcount
> 1);
3552 if (journal
->j_len
== 0) {
3553 reiserfs_prepare_for_journal(sb
, SB_BUFFER_WITH_SB(sb
),
3555 journal_mark_dirty(th
, sb
, SB_BUFFER_WITH_SB(sb
));
3557 return do_journal_end(th
, sb
, nblocks
, COMMIT_NOW
| WAIT
);
3561 ** writeback the pending async commits to disk
3563 static void flush_async_commits(struct work_struct
*work
)
3565 struct reiserfs_journal
*journal
=
3566 container_of(work
, struct reiserfs_journal
, j_work
.work
);
3567 struct super_block
*sb
= journal
->j_work_sb
;
3568 struct reiserfs_journal_list
*jl
;
3569 struct list_head
*entry
;
3571 reiserfs_write_lock(sb
);
3572 if (!list_empty(&journal
->j_journal_list
)) {
3573 /* last entry is the youngest, commit it and you get everything */
3574 entry
= journal
->j_journal_list
.prev
;
3575 jl
= JOURNAL_LIST_ENTRY(entry
);
3576 flush_commit_list(sb
, jl
, 1);
3578 reiserfs_write_unlock(sb
);
3582 ** flushes any old transactions to disk
3583 ** ends the current transaction if it is too old
3585 int reiserfs_flush_old_commits(struct super_block
*sb
)
3588 struct reiserfs_transaction_handle th
;
3589 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3591 now
= get_seconds();
3592 /* safety check so we don't flush while we are replaying the log during
3595 if (list_empty(&journal
->j_journal_list
)) {
3599 /* check the current transaction. If there are no writers, and it is
3600 * too old, finish it, and force the commit blocks to disk
3602 if (atomic_read(&journal
->j_wcount
) <= 0 &&
3603 journal
->j_trans_start_time
> 0 &&
3604 journal
->j_len
> 0 &&
3605 (now
- journal
->j_trans_start_time
) > journal
->j_max_trans_age
) {
3606 if (!journal_join(&th
, sb
, 1)) {
3607 reiserfs_prepare_for_journal(sb
,
3608 SB_BUFFER_WITH_SB(sb
),
3610 journal_mark_dirty(&th
, sb
,
3611 SB_BUFFER_WITH_SB(sb
));
3613 /* we're only being called from kreiserfsd, it makes no sense to do
3614 ** an async commit so that kreiserfsd can do it later
3616 do_journal_end(&th
, sb
, 1, COMMIT_NOW
| WAIT
);
3623 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3625 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3626 ** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just
3627 ** flushes the commit list and returns 0.
3629 ** Won't batch when flush or commit_now is set. Also won't batch when others are waiting on j_join_wait.
3631 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3633 static int check_journal_end(struct reiserfs_transaction_handle
*th
,
3634 struct super_block
*sb
, unsigned long nblocks
,
3639 int flush
= flags
& FLUSH_ALL
;
3640 int commit_now
= flags
& COMMIT_NOW
;
3641 int wait_on_commit
= flags
& WAIT
;
3642 struct reiserfs_journal_list
*jl
;
3643 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3645 BUG_ON(!th
->t_trans_id
);
3647 if (th
->t_trans_id
!= journal
->j_trans_id
) {
3648 reiserfs_panic(th
->t_super
, "journal-1577",
3649 "handle trans id %ld != current trans id %ld",
3650 th
->t_trans_id
, journal
->j_trans_id
);
3653 journal
->j_len_alloc
-= (th
->t_blocks_allocated
- th
->t_blocks_logged
);
3654 if (atomic_read(&(journal
->j_wcount
)) > 0) { /* <= 0 is allowed. unmounting might not call begin */
3655 atomic_dec(&(journal
->j_wcount
));
3658 /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3659 ** will be dealt with by next transaction that actually writes something, but should be taken
3660 ** care of in this trans
3662 BUG_ON(journal
->j_len
== 0);
3664 /* if wcount > 0, and we are called to with flush or commit_now,
3665 ** we wait on j_join_wait. We will wake up when the last writer has
3666 ** finished the transaction, and started it on its way to the disk.
3667 ** Then, we flush the commit or journal list, and just return 0
3668 ** because the rest of journal end was already done for this transaction.
3670 if (atomic_read(&(journal
->j_wcount
)) > 0) {
3671 if (flush
|| commit_now
) {
3674 jl
= journal
->j_current_jl
;
3675 trans_id
= jl
->j_trans_id
;
3677 jl
->j_state
|= LIST_COMMIT_PENDING
;
3678 atomic_set(&(journal
->j_jlock
), 1);
3680 journal
->j_next_full_flush
= 1;
3684 /* sleep while the current transaction is still j_jlocked */
3685 while (journal
->j_trans_id
== trans_id
) {
3686 if (atomic_read(&journal
->j_jlock
)) {
3687 queue_log_writer(sb
);
3690 if (journal
->j_trans_id
== trans_id
) {
3691 atomic_set(&(journal
->j_jlock
),
3697 BUG_ON(journal
->j_trans_id
== trans_id
);
3700 && journal_list_still_alive(sb
, trans_id
)
3701 && wait_on_commit
) {
3702 flush_commit_list(sb
, jl
, 1);
3710 /* deal with old transactions where we are the last writers */
3711 now
= get_seconds();
3712 if ((now
- journal
->j_trans_start_time
) > journal
->j_max_trans_age
) {
3714 journal
->j_next_async_flush
= 1;
3716 /* don't batch when someone is waiting on j_join_wait */
3717 /* don't batch when syncing the commit or flushing the whole trans */
3718 if (!(journal
->j_must_wait
> 0) && !(atomic_read(&(journal
->j_jlock
)))
3719 && !flush
&& !commit_now
&& (journal
->j_len
< journal
->j_max_batch
)
3720 && journal
->j_len_alloc
< journal
->j_max_batch
3721 && journal
->j_cnode_free
> (journal
->j_trans_max
* 3)) {
3722 journal
->j_bcount
++;
3727 if (journal
->j_start
> SB_ONDISK_JOURNAL_SIZE(sb
)) {
3728 reiserfs_panic(sb
, "journal-003",
3729 "j_start (%ld) is too high",
3736 ** Does all the work that makes deleting blocks safe.
3737 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3740 ** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes
3741 ** before this transaction has finished.
3743 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with
3744 ** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash,
3745 ** the block can't be reallocated yet.
3747 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3749 int journal_mark_freed(struct reiserfs_transaction_handle
*th
,
3750 struct super_block
*sb
, b_blocknr_t blocknr
)
3752 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3753 struct reiserfs_journal_cnode
*cn
= NULL
;
3754 struct buffer_head
*bh
= NULL
;
3755 struct reiserfs_list_bitmap
*jb
= NULL
;
3757 BUG_ON(!th
->t_trans_id
);
3759 cn
= get_journal_hash_dev(sb
, journal
->j_hash_table
, blocknr
);
3764 /* if it is journal new, we just remove it from this transaction */
3765 if (bh
&& buffer_journal_new(bh
)) {
3766 clear_buffer_journal_new(bh
);
3767 clear_prepared_bits(bh
);
3768 reiserfs_clean_and_file_buffer(bh
);
3769 cleaned
= remove_from_transaction(sb
, blocknr
, cleaned
);
3771 /* set the bit for this block in the journal bitmap for this transaction */
3772 jb
= journal
->j_current_jl
->j_list_bitmap
;
3774 reiserfs_panic(sb
, "journal-1702",
3775 "journal_list_bitmap is NULL");
3777 set_bit_in_list_bitmap(sb
, blocknr
, jb
);
3779 /* Note, the entire while loop is not allowed to schedule. */
3782 clear_prepared_bits(bh
);
3783 reiserfs_clean_and_file_buffer(bh
);
3785 cleaned
= remove_from_transaction(sb
, blocknr
, cleaned
);
3787 /* find all older transactions with this block, make sure they don't try to write it out */
3788 cn
= get_journal_hash_dev(sb
, journal
->j_list_hash_table
,
3791 if (sb
== cn
->sb
&& blocknr
== cn
->blocknr
) {
3792 set_bit(BLOCK_FREED
, &cn
->state
);
3795 /* remove_from_transaction will brelse the buffer if it was
3796 ** in the current trans
3798 clear_buffer_journal_dirty(cn
->
3800 clear_buffer_dirty(cn
->bh
);
3801 clear_buffer_journal_test(cn
->
3806 (&(cn
->bh
->b_count
)) < 0) {
3807 reiserfs_warning(sb
,
3809 "cn->bh->b_count < 0");
3812 if (cn
->jlist
) { /* since we are clearing the bh, we MUST dec nonzerolen */
3825 release_buffer_page(bh
); /* get_hash grabs the buffer */
3829 void reiserfs_update_inode_transaction(struct inode
*inode
)
3831 struct reiserfs_journal
*journal
= SB_JOURNAL(inode
->i_sb
);
3832 REISERFS_I(inode
)->i_jl
= journal
->j_current_jl
;
3833 REISERFS_I(inode
)->i_trans_id
= journal
->j_trans_id
;
3837 * returns -1 on error, 0 if no commits/barriers were done and 1
3838 * if a transaction was actually committed and the barrier was done
3840 static int __commit_trans_jl(struct inode
*inode
, unsigned long id
,
3841 struct reiserfs_journal_list
*jl
)
3843 struct reiserfs_transaction_handle th
;
3844 struct super_block
*sb
= inode
->i_sb
;
3845 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3848 /* is it from the current transaction, or from an unknown transaction? */
3849 if (id
== journal
->j_trans_id
) {
3850 jl
= journal
->j_current_jl
;
3851 /* try to let other writers come in and grow this transaction */
3852 let_transaction_grow(sb
, id
);
3853 if (journal
->j_trans_id
!= id
) {
3854 goto flush_commit_only
;
3857 ret
= journal_begin(&th
, sb
, 1);
3861 /* someone might have ended this transaction while we joined */
3862 if (journal
->j_trans_id
!= id
) {
3863 reiserfs_prepare_for_journal(sb
, SB_BUFFER_WITH_SB(sb
),
3865 journal_mark_dirty(&th
, sb
, SB_BUFFER_WITH_SB(sb
));
3866 ret
= journal_end(&th
, sb
, 1);
3867 goto flush_commit_only
;
3870 ret
= journal_end_sync(&th
, sb
, 1);
3875 /* this gets tricky, we have to make sure the journal list in
3876 * the inode still exists. We know the list is still around
3877 * if we've got a larger transaction id than the oldest list
3880 if (journal_list_still_alive(inode
->i_sb
, id
)) {
3882 * we only set ret to 1 when we know for sure
3883 * the barrier hasn't been started yet on the commit
3886 if (atomic_read(&jl
->j_commit_left
) > 1)
3888 flush_commit_list(sb
, jl
, 1);
3889 if (journal
->j_errno
)
3890 ret
= journal
->j_errno
;
3893 /* otherwise the list is gone, and long since committed */
3897 int reiserfs_commit_for_inode(struct inode
*inode
)
3899 unsigned int id
= REISERFS_I(inode
)->i_trans_id
;
3900 struct reiserfs_journal_list
*jl
= REISERFS_I(inode
)->i_jl
;
3902 /* for the whole inode, assume unset id means it was
3903 * changed in the current transaction. More conservative
3906 reiserfs_update_inode_transaction(inode
);
3907 id
= REISERFS_I(inode
)->i_trans_id
;
3908 /* jl will be updated in __commit_trans_jl */
3911 return __commit_trans_jl(inode
, id
, jl
);
3914 void reiserfs_restore_prepared_buffer(struct super_block
*sb
,
3915 struct buffer_head
*bh
)
3917 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3918 PROC_INFO_INC(sb
, journal
.restore_prepared
);
3922 if (test_clear_buffer_journal_restore_dirty(bh
) &&
3923 buffer_journal_dirty(bh
)) {
3924 struct reiserfs_journal_cnode
*cn
;
3925 cn
= get_journal_hash_dev(sb
,
3926 journal
->j_list_hash_table
,
3928 if (cn
&& can_dirty(cn
)) {
3929 set_buffer_journal_test(bh
);
3930 mark_buffer_dirty(bh
);
3933 clear_buffer_journal_prepared(bh
);
3936 extern struct tree_balance
*cur_tb
;
3938 ** before we can change a metadata block, we have to make sure it won't
3939 ** be written to disk while we are altering it. So, we must:
3944 int reiserfs_prepare_for_journal(struct super_block
*sb
,
3945 struct buffer_head
*bh
, int wait
)
3947 PROC_INFO_INC(sb
, journal
.prepare
);
3949 if (!trylock_buffer(bh
)) {
3954 set_buffer_journal_prepared(bh
);
3955 if (test_clear_buffer_dirty(bh
) && buffer_journal_dirty(bh
)) {
3956 clear_buffer_journal_test(bh
);
3957 set_buffer_journal_restore_dirty(bh
);
3963 static void flush_old_journal_lists(struct super_block
*s
)
3965 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
3966 struct reiserfs_journal_list
*jl
;
3967 struct list_head
*entry
;
3968 time_t now
= get_seconds();
3970 while (!list_empty(&journal
->j_journal_list
)) {
3971 entry
= journal
->j_journal_list
.next
;
3972 jl
= JOURNAL_LIST_ENTRY(entry
);
3973 /* this check should always be run, to send old lists to disk */
3974 if (jl
->j_timestamp
< (now
- (JOURNAL_MAX_TRANS_AGE
* 4)) &&
3975 atomic_read(&jl
->j_commit_left
) == 0 &&
3976 test_transaction(s
, jl
)) {
3977 flush_used_journal_lists(s
, jl
);
3985 ** long and ugly. If flush, will not return until all commit
3986 ** blocks and all real buffers in the trans are on disk.
3987 ** If no_async, won't return until all commit blocks are on disk.
3989 ** keep reading, there are comments as you go along
3991 ** If the journal is aborted, we just clean up. Things like flushing
3992 ** journal lists, etc just won't happen.
3994 static int do_journal_end(struct reiserfs_transaction_handle
*th
,
3995 struct super_block
*sb
, unsigned long nblocks
,
3998 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3999 struct reiserfs_journal_cnode
*cn
, *next
, *jl_cn
;
4000 struct reiserfs_journal_cnode
*last_cn
= NULL
;
4001 struct reiserfs_journal_desc
*desc
;
4002 struct reiserfs_journal_commit
*commit
;
4003 struct buffer_head
*c_bh
; /* commit bh */
4004 struct buffer_head
*d_bh
; /* desc bh */
4005 int cur_write_start
= 0; /* start index of current log write */
4010 struct reiserfs_journal_list
*jl
, *temp_jl
;
4011 struct list_head
*entry
, *safe
;
4012 unsigned long jindex
;
4013 unsigned int commit_trans_id
;
4016 BUG_ON(th
->t_refcount
> 1);
4017 BUG_ON(!th
->t_trans_id
);
4019 /* protect flush_older_commits from doing mistakes if the
4020 transaction ID counter gets overflowed. */
4021 if (th
->t_trans_id
== ~0U)
4022 flags
|= FLUSH_ALL
| COMMIT_NOW
| WAIT
;
4023 flush
= flags
& FLUSH_ALL
;
4024 wait_on_commit
= flags
& WAIT
;
4027 current
->journal_info
= th
->t_handle_save
;
4028 reiserfs_check_lock_depth(sb
, "journal end");
4029 if (journal
->j_len
== 0) {
4030 reiserfs_prepare_for_journal(sb
, SB_BUFFER_WITH_SB(sb
),
4032 journal_mark_dirty(th
, sb
, SB_BUFFER_WITH_SB(sb
));
4036 if (journal
->j_next_full_flush
) {
4040 if (journal
->j_next_async_flush
) {
4041 flags
|= COMMIT_NOW
| WAIT
;
4045 /* check_journal_end locks the journal, and unlocks if it does not return 1
4046 ** it tells us if we should continue with the journal_end, or just return
4048 if (!check_journal_end(th
, sb
, nblocks
, flags
)) {
4050 wake_queued_writers(sb
);
4051 reiserfs_async_progress_wait(sb
);
4055 /* check_journal_end might set these, check again */
4056 if (journal
->j_next_full_flush
) {
4061 ** j must wait means we have to flush the log blocks, and the real blocks for
4064 if (journal
->j_must_wait
> 0) {
4067 #ifdef REISERFS_PREALLOCATE
4068 /* quota ops might need to nest, setup the journal_info pointer for them
4069 * and raise the refcount so that it is > 0. */
4070 current
->journal_info
= th
;
4072 reiserfs_discard_all_prealloc(th
); /* it should not involve new blocks into
4073 * the transaction */
4075 current
->journal_info
= th
->t_handle_save
;
4078 /* setup description block */
4081 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
4083 set_buffer_uptodate(d_bh
);
4084 desc
= (struct reiserfs_journal_desc
*)(d_bh
)->b_data
;
4085 memset(d_bh
->b_data
, 0, d_bh
->b_size
);
4086 memcpy(get_journal_desc_magic(d_bh
), JOURNAL_DESC_MAGIC
, 8);
4087 set_desc_trans_id(desc
, journal
->j_trans_id
);
4089 /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */
4090 c_bh
= journal_getblk(sb
, SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
4091 ((journal
->j_start
+ journal
->j_len
+
4092 1) % SB_ONDISK_JOURNAL_SIZE(sb
)));
4093 commit
= (struct reiserfs_journal_commit
*)c_bh
->b_data
;
4094 memset(c_bh
->b_data
, 0, c_bh
->b_size
);
4095 set_commit_trans_id(commit
, journal
->j_trans_id
);
4096 set_buffer_uptodate(c_bh
);
4098 /* init this journal list */
4099 jl
= journal
->j_current_jl
;
4101 /* we lock the commit before doing anything because
4102 * we want to make sure nobody tries to run flush_commit_list until
4103 * the new transaction is fully setup, and we've already flushed the
4106 reiserfs_mutex_lock_safe(&jl
->j_commit_mutex
, sb
);
4108 /* save the transaction id in case we need to commit it later */
4109 commit_trans_id
= jl
->j_trans_id
;
4111 atomic_set(&jl
->j_older_commits_done
, 0);
4112 jl
->j_trans_id
= journal
->j_trans_id
;
4113 jl
->j_timestamp
= journal
->j_trans_start_time
;
4114 jl
->j_commit_bh
= c_bh
;
4115 jl
->j_start
= journal
->j_start
;
4116 jl
->j_len
= journal
->j_len
;
4117 atomic_set(&jl
->j_nonzerolen
, journal
->j_len
);
4118 atomic_set(&jl
->j_commit_left
, journal
->j_len
+ 2);
4119 jl
->j_realblock
= NULL
;
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
;
4142 /* make sure the block we are trying to log is not a block
4143 of journal or reserved area */
4145 if (is_block_in_log_or_reserved_area
4146 (sb
, cn
->bh
->b_blocknr
)) {
4147 reiserfs_panic(sb
, "journal-2332",
4148 "Trying to log block %lu, "
4149 "which is a log block",
4152 jl_cn
->blocknr
= cn
->bh
->b_blocknr
;
4157 insert_journal_hash(journal
->j_list_hash_table
, jl_cn
);
4158 if (i
< trans_half
) {
4159 desc
->j_realblock
[i
] =
4160 cpu_to_le32(cn
->bh
->b_blocknr
);
4162 commit
->j_realblock
[i
- trans_half
] =
4163 cpu_to_le32(cn
->bh
->b_blocknr
);
4169 set_desc_trans_len(desc
, journal
->j_len
);
4170 set_desc_mount_id(desc
, journal
->j_mount_id
);
4171 set_desc_trans_id(desc
, journal
->j_trans_id
);
4172 set_commit_trans_len(commit
, journal
->j_len
);
4174 /* special check in case all buffers in the journal were marked for not logging */
4175 BUG_ON(journal
->j_len
== 0);
4177 /* we're about to dirty all the log blocks, mark the description block
4178 * dirty now too. Don't mark the commit block dirty until all the
4179 * others are on disk
4181 mark_buffer_dirty(d_bh
);
4183 /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4184 cur_write_start
= journal
->j_start
;
4185 cn
= journal
->j_first
;
4186 jindex
= 1; /* start at one so we don't get the desc again */
4188 clear_buffer_journal_new(cn
->bh
);
4189 /* copy all the real blocks into log area. dirty log blocks */
4190 if (buffer_journaled(cn
->bh
)) {
4191 struct buffer_head
*tmp_bh
;
4196 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
4199 SB_ONDISK_JOURNAL_SIZE(sb
)));
4200 set_buffer_uptodate(tmp_bh
);
4201 page
= cn
->bh
->b_page
;
4203 memcpy(tmp_bh
->b_data
,
4204 addr
+ offset_in_page(cn
->bh
->b_data
),
4207 mark_buffer_dirty(tmp_bh
);
4209 set_buffer_journal_dirty(cn
->bh
);
4210 clear_buffer_journaled(cn
->bh
);
4212 /* JDirty cleared sometime during transaction. don't log this one */
4213 reiserfs_warning(sb
, "journal-2048",
4214 "BAD, buffer in journal hash, "
4221 reiserfs_write_unlock(sb
);
4223 reiserfs_write_lock(sb
);
4226 /* we are done with both the c_bh and d_bh, but
4227 ** c_bh must be written after all other commit blocks,
4228 ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4231 journal
->j_current_jl
= alloc_journal_list(sb
);
4233 /* now it is safe to insert this transaction on the main list */
4234 list_add_tail(&jl
->j_list
, &journal
->j_journal_list
);
4235 list_add_tail(&jl
->j_working_list
, &journal
->j_working_list
);
4236 journal
->j_num_work_lists
++;
4238 /* reset journal values for the next transaction */
4239 old_start
= journal
->j_start
;
4241 (journal
->j_start
+ journal
->j_len
+
4242 2) % SB_ONDISK_JOURNAL_SIZE(sb
);
4243 atomic_set(&(journal
->j_wcount
), 0);
4244 journal
->j_bcount
= 0;
4245 journal
->j_last
= NULL
;
4246 journal
->j_first
= NULL
;
4248 journal
->j_trans_start_time
= 0;
4249 /* check for trans_id overflow */
4250 if (++journal
->j_trans_id
== 0)
4251 journal
->j_trans_id
= 10;
4252 journal
->j_current_jl
->j_trans_id
= journal
->j_trans_id
;
4253 journal
->j_must_wait
= 0;
4254 journal
->j_len_alloc
= 0;
4255 journal
->j_next_full_flush
= 0;
4256 journal
->j_next_async_flush
= 0;
4257 init_journal_hash(sb
);
4259 // make sure reiserfs_add_jh sees the new current_jl before we
4260 // write out the tails
4263 /* tail conversion targets have to hit the disk before we end the
4264 * transaction. Otherwise a later transaction might repack the tail
4265 * before this transaction commits, leaving the data block unflushed and
4266 * clean, if we crash before the later transaction commits, the data block
4269 if (!list_empty(&jl
->j_tail_bh_list
)) {
4270 reiserfs_write_unlock(sb
);
4271 write_ordered_buffers(&journal
->j_dirty_buffers_lock
,
4272 journal
, jl
, &jl
->j_tail_bh_list
);
4273 reiserfs_write_lock(sb
);
4275 BUG_ON(!list_empty(&jl
->j_tail_bh_list
));
4276 mutex_unlock(&jl
->j_commit_mutex
);
4278 /* honor the flush wishes from the caller, simple commits can
4279 ** be done outside the journal lock, they are done below
4281 ** if we don't flush the commit list right now, we put it into
4282 ** the work queue so the people waiting on the async progress work
4283 ** queue don't wait for this proc to flush journal lists and such.
4286 flush_commit_list(sb
, jl
, 1);
4287 flush_journal_list(sb
, jl
, 1);
4288 } else if (!(jl
->j_state
& LIST_COMMIT_PENDING
))
4289 queue_delayed_work(commit_wq
, &journal
->j_work
, HZ
/ 10);
4291 /* if the next transaction has any chance of wrapping, flush
4292 ** transactions that might get overwritten. If any journal lists are very
4293 ** old flush them as well.
4296 list_for_each_safe(entry
, safe
, &journal
->j_journal_list
) {
4297 temp_jl
= JOURNAL_LIST_ENTRY(entry
);
4298 if (journal
->j_start
<= temp_jl
->j_start
) {
4299 if ((journal
->j_start
+ journal
->j_trans_max
+ 1) >=
4301 flush_used_journal_lists(sb
, temp_jl
);
4303 } else if ((journal
->j_start
+
4304 journal
->j_trans_max
+ 1) <
4305 SB_ONDISK_JOURNAL_SIZE(sb
)) {
4306 /* if we don't cross into the next transaction and we don't
4307 * wrap, there is no way we can overlap any later transactions
4312 } else if ((journal
->j_start
+
4313 journal
->j_trans_max
+ 1) >
4314 SB_ONDISK_JOURNAL_SIZE(sb
)) {
4315 if (((journal
->j_start
+ journal
->j_trans_max
+ 1) %
4316 SB_ONDISK_JOURNAL_SIZE(sb
)) >=
4318 flush_used_journal_lists(sb
, temp_jl
);
4321 /* we don't overlap anything from out start to the end of the
4322 * log, and our wrapped portion doesn't overlap anything at
4323 * the start of the log. We can break
4329 flush_old_journal_lists(sb
);
4331 journal
->j_current_jl
->j_list_bitmap
=
4332 get_list_bitmap(sb
, journal
->j_current_jl
);
4334 if (!(journal
->j_current_jl
->j_list_bitmap
)) {
4335 reiserfs_panic(sb
, "journal-1996",
4336 "could not get a list bitmap");
4339 atomic_set(&(journal
->j_jlock
), 0);
4341 /* wake up any body waiting to join. */
4342 clear_bit(J_WRITERS_QUEUED
, &journal
->j_state
);
4343 wake_up(&(journal
->j_join_wait
));
4345 if (!flush
&& wait_on_commit
&&
4346 journal_list_still_alive(sb
, commit_trans_id
)) {
4347 flush_commit_list(sb
, jl
, 1);
4350 reiserfs_check_lock_depth(sb
, "journal end2");
4352 memset(th
, 0, sizeof(*th
));
4353 /* Re-set th->t_super, so we can properly keep track of how many
4354 * persistent transactions there are. We need to do this so if this
4355 * call is part of a failed restart_transaction, we can free it later */
4358 return journal
->j_errno
;
4361 /* Send the file system read only and refuse new transactions */
4362 void reiserfs_abort_journal(struct super_block
*sb
, int errno
)
4364 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
4365 if (test_bit(J_ABORTED
, &journal
->j_state
))
4368 if (!journal
->j_errno
)
4369 journal
->j_errno
= errno
;
4371 sb
->s_flags
|= MS_RDONLY
;
4372 set_bit(J_ABORTED
, &journal
->j_state
);
4374 #ifdef CONFIG_REISERFS_CHECK