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/buffer_head.h>
47 #include <linux/workqueue.h>
48 #include <linux/writeback.h>
49 #include <linux/blkdev.h>
50 #include <linux/backing-dev.h>
51 #include <linux/uaccess.h>
52 #include <linux/slab.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 struct reiserfs_bitmap_node
*allocate_bitmap_node(struct super_block
143 struct reiserfs_bitmap_node
*bn
;
146 bn
= kmalloc(sizeof(struct reiserfs_bitmap_node
), GFP_NOFS
);
150 bn
->data
= kzalloc(sb
->s_blocksize
, GFP_NOFS
);
156 INIT_LIST_HEAD(&bn
->list
);
160 static struct reiserfs_bitmap_node
*get_bitmap_node(struct super_block
*sb
)
162 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
163 struct reiserfs_bitmap_node
*bn
= NULL
;
164 struct list_head
*entry
= journal
->j_bitmap_nodes
.next
;
166 journal
->j_used_bitmap_nodes
++;
169 if (entry
!= &journal
->j_bitmap_nodes
) {
170 bn
= list_entry(entry
, struct reiserfs_bitmap_node
, list
);
172 memset(bn
->data
, 0, sb
->s_blocksize
);
173 journal
->j_free_bitmap_nodes
--;
176 bn
= allocate_bitmap_node(sb
);
183 static inline void free_bitmap_node(struct super_block
*sb
,
184 struct reiserfs_bitmap_node
*bn
)
186 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
187 journal
->j_used_bitmap_nodes
--;
188 if (journal
->j_free_bitmap_nodes
> REISERFS_MAX_BITMAP_NODES
) {
192 list_add(&bn
->list
, &journal
->j_bitmap_nodes
);
193 journal
->j_free_bitmap_nodes
++;
197 static void allocate_bitmap_nodes(struct super_block
*sb
)
200 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
201 struct reiserfs_bitmap_node
*bn
= NULL
;
202 for (i
= 0; i
< REISERFS_MIN_BITMAP_NODES
; i
++) {
203 bn
= allocate_bitmap_node(sb
);
205 list_add(&bn
->list
, &journal
->j_bitmap_nodes
);
206 journal
->j_free_bitmap_nodes
++;
208 break; /* this is ok, we'll try again when more are needed */
213 static int set_bit_in_list_bitmap(struct super_block
*sb
,
215 struct reiserfs_list_bitmap
*jb
)
217 unsigned int bmap_nr
= block
/ (sb
->s_blocksize
<< 3);
218 unsigned int bit_nr
= block
% (sb
->s_blocksize
<< 3);
220 if (!jb
->bitmaps
[bmap_nr
]) {
221 jb
->bitmaps
[bmap_nr
] = get_bitmap_node(sb
);
223 set_bit(bit_nr
, (unsigned long *)jb
->bitmaps
[bmap_nr
]->data
);
227 static void cleanup_bitmap_list(struct super_block
*sb
,
228 struct reiserfs_list_bitmap
*jb
)
231 if (jb
->bitmaps
== NULL
)
234 for (i
= 0; i
< reiserfs_bmap_count(sb
); i
++) {
235 if (jb
->bitmaps
[i
]) {
236 free_bitmap_node(sb
, jb
->bitmaps
[i
]);
237 jb
->bitmaps
[i
] = NULL
;
243 ** only call this on FS unmount.
245 static int free_list_bitmaps(struct super_block
*sb
,
246 struct reiserfs_list_bitmap
*jb_array
)
249 struct reiserfs_list_bitmap
*jb
;
250 for (i
= 0; i
< JOURNAL_NUM_BITMAPS
; i
++) {
252 jb
->journal_list
= NULL
;
253 cleanup_bitmap_list(sb
, jb
);
260 static int free_bitmap_nodes(struct super_block
*sb
)
262 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
263 struct list_head
*next
= journal
->j_bitmap_nodes
.next
;
264 struct reiserfs_bitmap_node
*bn
;
266 while (next
!= &journal
->j_bitmap_nodes
) {
267 bn
= list_entry(next
, struct reiserfs_bitmap_node
, list
);
271 next
= journal
->j_bitmap_nodes
.next
;
272 journal
->j_free_bitmap_nodes
--;
279 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
280 ** jb_array is the array to be filled in.
282 int reiserfs_allocate_list_bitmaps(struct super_block
*sb
,
283 struct reiserfs_list_bitmap
*jb_array
,
284 unsigned int bmap_nr
)
288 struct reiserfs_list_bitmap
*jb
;
289 int mem
= bmap_nr
* sizeof(struct reiserfs_bitmap_node
*);
291 for (i
= 0; i
< JOURNAL_NUM_BITMAPS
; i
++) {
293 jb
->journal_list
= NULL
;
294 jb
->bitmaps
= vmalloc(mem
);
296 reiserfs_warning(sb
, "clm-2000", "unable to "
297 "allocate bitmaps for journal lists");
301 memset(jb
->bitmaps
, 0, mem
);
304 free_list_bitmaps(sb
, jb_array
);
311 ** find an available list bitmap. If you can't find one, flush a commit list
314 static struct reiserfs_list_bitmap
*get_list_bitmap(struct super_block
*sb
,
315 struct reiserfs_journal_list
319 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
320 struct reiserfs_list_bitmap
*jb
= NULL
;
322 for (j
= 0; j
< (JOURNAL_NUM_BITMAPS
* 3); j
++) {
323 i
= journal
->j_list_bitmap_index
;
324 journal
->j_list_bitmap_index
= (i
+ 1) % JOURNAL_NUM_BITMAPS
;
325 jb
= journal
->j_list_bitmap
+ i
;
326 if (journal
->j_list_bitmap
[i
].journal_list
) {
327 flush_commit_list(sb
,
328 journal
->j_list_bitmap
[i
].
330 if (!journal
->j_list_bitmap
[i
].journal_list
) {
337 if (jb
->journal_list
) { /* double check to make sure if flushed correctly */
340 jb
->journal_list
= jl
;
345 ** allocates a new chunk of X nodes, and links them all together as a list.
346 ** Uses the cnode->next and cnode->prev pointers
347 ** returns NULL on failure
349 static struct reiserfs_journal_cnode
*allocate_cnodes(int num_cnodes
)
351 struct reiserfs_journal_cnode
*head
;
353 if (num_cnodes
<= 0) {
356 head
= vmalloc(num_cnodes
* sizeof(struct reiserfs_journal_cnode
));
360 memset(head
, 0, num_cnodes
* sizeof(struct reiserfs_journal_cnode
));
362 head
[0].next
= head
+ 1;
363 for (i
= 1; i
< num_cnodes
; i
++) {
364 head
[i
].prev
= head
+ (i
- 1);
365 head
[i
].next
= head
+ (i
+ 1); /* if last one, overwrite it after the if */
367 head
[num_cnodes
- 1].next
= NULL
;
372 ** pulls a cnode off the free list, or returns NULL on failure
374 static struct reiserfs_journal_cnode
*get_cnode(struct super_block
*sb
)
376 struct reiserfs_journal_cnode
*cn
;
377 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
379 reiserfs_check_lock_depth(sb
, "get_cnode");
381 if (journal
->j_cnode_free
<= 0) {
384 journal
->j_cnode_used
++;
385 journal
->j_cnode_free
--;
386 cn
= journal
->j_cnode_free_list
;
391 cn
->next
->prev
= NULL
;
393 journal
->j_cnode_free_list
= cn
->next
;
394 memset(cn
, 0, sizeof(struct reiserfs_journal_cnode
));
399 ** returns a cnode to the free list
401 static void free_cnode(struct super_block
*sb
,
402 struct reiserfs_journal_cnode
*cn
)
404 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
406 reiserfs_check_lock_depth(sb
, "free_cnode");
408 journal
->j_cnode_used
--;
409 journal
->j_cnode_free
++;
410 /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
411 cn
->next
= journal
->j_cnode_free_list
;
412 if (journal
->j_cnode_free_list
) {
413 journal
->j_cnode_free_list
->prev
= cn
;
415 cn
->prev
= NULL
; /* not needed with the memset, but I might kill the memset, and forget to do this */
416 journal
->j_cnode_free_list
= cn
;
419 static void clear_prepared_bits(struct buffer_head
*bh
)
421 clear_buffer_journal_prepared(bh
);
422 clear_buffer_journal_restore_dirty(bh
);
425 /* return a cnode with same dev, block number and size in table, or null if not found */
426 static inline struct reiserfs_journal_cnode
*get_journal_hash_dev(struct
430 reiserfs_journal_cnode
434 struct reiserfs_journal_cnode
*cn
;
435 cn
= journal_hash(table
, sb
, bl
);
437 if (cn
->blocknr
== bl
&& cn
->sb
== sb
)
441 return (struct reiserfs_journal_cnode
*)0;
445 ** this actually means 'can this block be reallocated yet?'. If you set search_all, a block can only be allocated
446 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
447 ** being overwritten by a replay after crashing.
449 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction. Since deleting
450 ** a block removes it from the current transaction, this case should never happen. If you don't set search_all, make
451 ** sure you never write the block without logging it.
453 ** next_zero_bit is a suggestion about the next block to try for find_forward.
454 ** when bl is rejected because it is set in a journal list bitmap, we search
455 ** for the next zero bit in the bitmap that rejected bl. Then, we return that
456 ** through next_zero_bit for find_forward to try.
458 ** Just because we return something in next_zero_bit does not mean we won't
459 ** reject it on the next call to reiserfs_in_journal
462 int reiserfs_in_journal(struct super_block
*sb
,
463 unsigned int bmap_nr
, int bit_nr
, int search_all
,
464 b_blocknr_t
* next_zero_bit
)
466 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
467 struct reiserfs_journal_cnode
*cn
;
468 struct reiserfs_list_bitmap
*jb
;
472 *next_zero_bit
= 0; /* always start this at zero. */
474 PROC_INFO_INC(sb
, journal
.in_journal
);
475 /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
476 ** if we crash before the transaction that freed it commits, this transaction won't
477 ** have committed either, and the block will never be written
480 for (i
= 0; i
< JOURNAL_NUM_BITMAPS
; i
++) {
481 PROC_INFO_INC(sb
, journal
.in_journal_bitmap
);
482 jb
= journal
->j_list_bitmap
+ i
;
483 if (jb
->journal_list
&& jb
->bitmaps
[bmap_nr
] &&
485 (unsigned long *)jb
->bitmaps
[bmap_nr
]->
488 find_next_zero_bit((unsigned long *)
489 (jb
->bitmaps
[bmap_nr
]->
491 sb
->s_blocksize
<< 3,
498 bl
= bmap_nr
* (sb
->s_blocksize
<< 3) + bit_nr
;
499 /* is it in any old transactions? */
502 get_journal_hash_dev(sb
, journal
->j_list_hash_table
, bl
))) {
506 /* is it in the current transaction. This should never happen */
507 if ((cn
= get_journal_hash_dev(sb
, journal
->j_hash_table
, bl
))) {
512 PROC_INFO_INC(sb
, journal
.in_journal_reusable
);
517 /* insert cn into table
519 static inline void insert_journal_hash(struct reiserfs_journal_cnode
**table
,
520 struct reiserfs_journal_cnode
*cn
)
522 struct reiserfs_journal_cnode
*cn_orig
;
524 cn_orig
= journal_hash(table
, cn
->sb
, cn
->blocknr
);
530 journal_hash(table
, cn
->sb
, cn
->blocknr
) = cn
;
533 /* lock the current transaction */
534 static inline void lock_journal(struct super_block
*sb
)
536 PROC_INFO_INC(sb
, journal
.lock_journal
);
538 reiserfs_mutex_lock_safe(&SB_JOURNAL(sb
)->j_mutex
, sb
);
541 /* unlock the current transaction */
542 static inline void unlock_journal(struct super_block
*sb
)
544 mutex_unlock(&SB_JOURNAL(sb
)->j_mutex
);
547 static inline void get_journal_list(struct reiserfs_journal_list
*jl
)
552 static inline void put_journal_list(struct super_block
*s
,
553 struct reiserfs_journal_list
*jl
)
555 if (jl
->j_refcount
< 1) {
556 reiserfs_panic(s
, "journal-2", "trans id %u, refcount at %d",
557 jl
->j_trans_id
, jl
->j_refcount
);
559 if (--jl
->j_refcount
== 0)
564 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
565 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
568 static void cleanup_freed_for_journal_list(struct super_block
*sb
,
569 struct reiserfs_journal_list
*jl
)
572 struct reiserfs_list_bitmap
*jb
= jl
->j_list_bitmap
;
574 cleanup_bitmap_list(sb
, jb
);
576 jl
->j_list_bitmap
->journal_list
= NULL
;
577 jl
->j_list_bitmap
= NULL
;
580 static int journal_list_still_alive(struct super_block
*s
,
581 unsigned int trans_id
)
583 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
584 struct list_head
*entry
= &journal
->j_journal_list
;
585 struct reiserfs_journal_list
*jl
;
587 if (!list_empty(entry
)) {
588 jl
= JOURNAL_LIST_ENTRY(entry
->next
);
589 if (jl
->j_trans_id
<= trans_id
) {
597 * If page->mapping was null, we failed to truncate this page for
598 * some reason. Most likely because it was truncated after being
599 * logged via data=journal.
601 * This does a check to see if the buffer belongs to one of these
602 * lost pages before doing the final put_bh. If page->mapping was
603 * null, it tries to free buffers on the page, which should make the
604 * final page_cache_release drop the page from the lru.
606 static void release_buffer_page(struct buffer_head
*bh
)
608 struct page
*page
= bh
->b_page
;
609 if (!page
->mapping
&& trylock_page(page
)) {
610 page_cache_get(page
);
613 try_to_free_buffers(page
);
615 page_cache_release(page
);
621 static void reiserfs_end_buffer_io_sync(struct buffer_head
*bh
, int uptodate
)
623 char b
[BDEVNAME_SIZE
];
625 if (buffer_journaled(bh
)) {
626 reiserfs_warning(NULL
, "clm-2084",
627 "pinned buffer %lu:%s sent to disk",
628 bh
->b_blocknr
, bdevname(bh
->b_bdev
, b
));
631 set_buffer_uptodate(bh
);
633 clear_buffer_uptodate(bh
);
636 release_buffer_page(bh
);
639 static void reiserfs_end_ordered_io(struct buffer_head
*bh
, int uptodate
)
642 set_buffer_uptodate(bh
);
644 clear_buffer_uptodate(bh
);
649 static void submit_logged_buffer(struct buffer_head
*bh
)
652 bh
->b_end_io
= reiserfs_end_buffer_io_sync
;
653 clear_buffer_journal_new(bh
);
654 clear_buffer_dirty(bh
);
655 if (!test_clear_buffer_journal_test(bh
))
657 if (!buffer_uptodate(bh
))
659 submit_bh(WRITE
, bh
);
662 static void submit_ordered_buffer(struct buffer_head
*bh
)
665 bh
->b_end_io
= reiserfs_end_ordered_io
;
666 clear_buffer_dirty(bh
);
667 if (!buffer_uptodate(bh
))
669 submit_bh(WRITE
, bh
);
672 #define CHUNK_SIZE 32
673 struct buffer_chunk
{
674 struct buffer_head
*bh
[CHUNK_SIZE
];
678 static void write_chunk(struct buffer_chunk
*chunk
)
682 for (i
= 0; i
< chunk
->nr
; i
++) {
683 submit_logged_buffer(chunk
->bh
[i
]);
689 static void write_ordered_chunk(struct buffer_chunk
*chunk
)
693 for (i
= 0; i
< chunk
->nr
; i
++) {
694 submit_ordered_buffer(chunk
->bh
[i
]);
700 static int add_to_chunk(struct buffer_chunk
*chunk
, struct buffer_head
*bh
,
701 spinlock_t
* lock
, void (fn
) (struct buffer_chunk
*))
704 BUG_ON(chunk
->nr
>= CHUNK_SIZE
);
705 chunk
->bh
[chunk
->nr
++] = bh
;
706 if (chunk
->nr
>= CHUNK_SIZE
) {
717 static atomic_t nr_reiserfs_jh
= ATOMIC_INIT(0);
718 static struct reiserfs_jh
*alloc_jh(void)
720 struct reiserfs_jh
*jh
;
722 jh
= kmalloc(sizeof(*jh
), GFP_NOFS
);
724 atomic_inc(&nr_reiserfs_jh
);
732 * we want to free the jh when the buffer has been written
735 void reiserfs_free_jh(struct buffer_head
*bh
)
737 struct reiserfs_jh
*jh
;
741 bh
->b_private
= NULL
;
743 list_del_init(&jh
->list
);
745 if (atomic_read(&nr_reiserfs_jh
) <= 0)
747 atomic_dec(&nr_reiserfs_jh
);
752 static inline int __add_jh(struct reiserfs_journal
*j
, struct buffer_head
*bh
,
755 struct reiserfs_jh
*jh
;
758 spin_lock(&j
->j_dirty_buffers_lock
);
759 if (!bh
->b_private
) {
760 spin_unlock(&j
->j_dirty_buffers_lock
);
764 list_del_init(&jh
->list
);
769 spin_lock(&j
->j_dirty_buffers_lock
);
770 /* buffer must be locked for __add_jh, should be able to have
771 * two adds at the same time
773 BUG_ON(bh
->b_private
);
777 jh
->jl
= j
->j_current_jl
;
779 list_add_tail(&jh
->list
, &jh
->jl
->j_tail_bh_list
);
781 list_add_tail(&jh
->list
, &jh
->jl
->j_bh_list
);
783 spin_unlock(&j
->j_dirty_buffers_lock
);
787 int reiserfs_add_tail_list(struct inode
*inode
, struct buffer_head
*bh
)
789 return __add_jh(SB_JOURNAL(inode
->i_sb
), bh
, 1);
791 int reiserfs_add_ordered_list(struct inode
*inode
, struct buffer_head
*bh
)
793 return __add_jh(SB_JOURNAL(inode
->i_sb
), bh
, 0);
796 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
797 static int write_ordered_buffers(spinlock_t
* lock
,
798 struct reiserfs_journal
*j
,
799 struct reiserfs_journal_list
*jl
,
800 struct list_head
*list
)
802 struct buffer_head
*bh
;
803 struct reiserfs_jh
*jh
;
804 int ret
= j
->j_errno
;
805 struct buffer_chunk chunk
;
806 struct list_head tmp
;
807 INIT_LIST_HEAD(&tmp
);
811 while (!list_empty(list
)) {
812 jh
= JH_ENTRY(list
->next
);
815 if (!trylock_buffer(bh
)) {
816 if (!buffer_dirty(bh
)) {
817 list_move(&jh
->list
, &tmp
);
822 write_ordered_chunk(&chunk
);
828 /* in theory, dirty non-uptodate buffers should never get here,
829 * but the upper layer io error paths still have a few quirks.
830 * Handle them here as gracefully as we can
832 if (!buffer_uptodate(bh
) && buffer_dirty(bh
)) {
833 clear_buffer_dirty(bh
);
836 if (buffer_dirty(bh
)) {
837 list_move(&jh
->list
, &tmp
);
838 add_to_chunk(&chunk
, bh
, lock
, write_ordered_chunk
);
840 reiserfs_free_jh(bh
);
845 cond_resched_lock(lock
);
849 write_ordered_chunk(&chunk
);
852 while (!list_empty(&tmp
)) {
853 jh
= JH_ENTRY(tmp
.prev
);
856 reiserfs_free_jh(bh
);
858 if (buffer_locked(bh
)) {
863 if (!buffer_uptodate(bh
)) {
866 /* ugly interaction with invalidatepage here.
867 * reiserfs_invalidate_page will pin any buffer that has a valid
868 * journal head from an older transaction. If someone else sets
869 * our buffer dirty after we write it in the first loop, and
870 * then someone truncates the page away, nobody will ever write
871 * the buffer. We're safe if we write the page one last time
872 * after freeing the journal header.
874 if (buffer_dirty(bh
) && unlikely(bh
->b_page
->mapping
== NULL
)) {
876 ll_rw_block(WRITE
, 1, &bh
);
880 cond_resched_lock(lock
);
886 static int flush_older_commits(struct super_block
*s
,
887 struct reiserfs_journal_list
*jl
)
889 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
890 struct reiserfs_journal_list
*other_jl
;
891 struct reiserfs_journal_list
*first_jl
;
892 struct list_head
*entry
;
893 unsigned int trans_id
= jl
->j_trans_id
;
894 unsigned int other_trans_id
;
895 unsigned int first_trans_id
;
899 * first we walk backwards to find the oldest uncommitted transation
902 entry
= jl
->j_list
.prev
;
904 other_jl
= JOURNAL_LIST_ENTRY(entry
);
905 if (entry
== &journal
->j_journal_list
||
906 atomic_read(&other_jl
->j_older_commits_done
))
910 entry
= other_jl
->j_list
.prev
;
913 /* if we didn't find any older uncommitted transactions, return now */
914 if (first_jl
== jl
) {
918 first_trans_id
= first_jl
->j_trans_id
;
920 entry
= &first_jl
->j_list
;
922 other_jl
= JOURNAL_LIST_ENTRY(entry
);
923 other_trans_id
= other_jl
->j_trans_id
;
925 if (other_trans_id
< trans_id
) {
926 if (atomic_read(&other_jl
->j_commit_left
) != 0) {
927 flush_commit_list(s
, other_jl
, 0);
929 /* list we were called with is gone, return */
930 if (!journal_list_still_alive(s
, trans_id
))
933 /* the one we just flushed is gone, this means all
934 * older lists are also gone, so first_jl is no longer
935 * valid either. Go back to the beginning.
937 if (!journal_list_still_alive
938 (s
, other_trans_id
)) {
943 if (entry
== &journal
->j_journal_list
)
952 static int reiserfs_async_progress_wait(struct super_block
*s
)
954 struct reiserfs_journal
*j
= SB_JOURNAL(s
);
956 if (atomic_read(&j
->j_async_throttle
)) {
957 reiserfs_write_unlock(s
);
958 congestion_wait(BLK_RW_ASYNC
, HZ
/ 10);
959 reiserfs_write_lock(s
);
966 ** if this journal list still has commit blocks unflushed, send them to disk.
968 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
969 ** Before the commit block can by written, every other log block must be safely on disk
972 static int flush_commit_list(struct super_block
*s
,
973 struct reiserfs_journal_list
*jl
, int flushall
)
977 struct buffer_head
*tbh
= NULL
;
978 unsigned int trans_id
= jl
->j_trans_id
;
979 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
983 reiserfs_check_lock_depth(s
, "flush_commit_list");
985 if (atomic_read(&jl
->j_older_commits_done
)) {
991 /* before we can put our commit blocks on disk, we have to make sure everyone older than
994 BUG_ON(jl
->j_len
<= 0);
995 BUG_ON(trans_id
== journal
->j_trans_id
);
997 get_journal_list(jl
);
999 if (flush_older_commits(s
, jl
) == 1) {
1000 /* list disappeared during flush_older_commits. return */
1005 /* make sure nobody is trying to flush this one at the same time */
1006 reiserfs_mutex_lock_safe(&jl
->j_commit_mutex
, s
);
1008 if (!journal_list_still_alive(s
, trans_id
)) {
1009 mutex_unlock(&jl
->j_commit_mutex
);
1012 BUG_ON(jl
->j_trans_id
== 0);
1014 /* this commit is done, exit */
1015 if (atomic_read(&(jl
->j_commit_left
)) <= 0) {
1017 atomic_set(&(jl
->j_older_commits_done
), 1);
1019 mutex_unlock(&jl
->j_commit_mutex
);
1023 if (!list_empty(&jl
->j_bh_list
)) {
1027 * We might sleep in numerous places inside
1028 * write_ordered_buffers. Relax the write lock.
1030 reiserfs_write_unlock(s
);
1031 ret
= write_ordered_buffers(&journal
->j_dirty_buffers_lock
,
1032 journal
, jl
, &jl
->j_bh_list
);
1033 if (ret
< 0 && retval
== 0)
1035 reiserfs_write_lock(s
);
1037 BUG_ON(!list_empty(&jl
->j_bh_list
));
1039 * for the description block and all the log blocks, submit any buffers
1040 * that haven't already reached the disk. Try to write at least 256
1041 * log blocks. later on, we will only wait on blocks that correspond
1042 * to this transaction, but while we're unplugging we might as well
1043 * get a chunk of data on there.
1045 atomic_inc(&journal
->j_async_throttle
);
1046 write_len
= jl
->j_len
+ 1;
1047 if (write_len
< 256)
1049 for (i
= 0 ; i
< write_len
; i
++) {
1050 bn
= SB_ONDISK_JOURNAL_1st_BLOCK(s
) + (jl
->j_start
+ i
) %
1051 SB_ONDISK_JOURNAL_SIZE(s
);
1052 tbh
= journal_find_get_block(s
, bn
);
1054 if (buffer_dirty(tbh
)) {
1055 reiserfs_write_unlock(s
);
1056 ll_rw_block(WRITE
, 1, &tbh
);
1057 reiserfs_write_lock(s
);
1062 atomic_dec(&journal
->j_async_throttle
);
1064 for (i
= 0; i
< (jl
->j_len
+ 1); i
++) {
1065 bn
= SB_ONDISK_JOURNAL_1st_BLOCK(s
) +
1066 (jl
->j_start
+ i
) % SB_ONDISK_JOURNAL_SIZE(s
);
1067 tbh
= journal_find_get_block(s
, bn
);
1069 reiserfs_write_unlock(s
);
1070 wait_on_buffer(tbh
);
1071 reiserfs_write_lock(s
);
1072 // since we're using ll_rw_blk above, it might have skipped over
1073 // a locked buffer. Double check here
1075 /* redundant, sync_dirty_buffer() checks */
1076 if (buffer_dirty(tbh
)) {
1077 reiserfs_write_unlock(s
);
1078 sync_dirty_buffer(tbh
);
1079 reiserfs_write_lock(s
);
1081 if (unlikely(!buffer_uptodate(tbh
))) {
1082 #ifdef CONFIG_REISERFS_CHECK
1083 reiserfs_warning(s
, "journal-601",
1084 "buffer write failed");
1088 put_bh(tbh
); /* once for journal_find_get_block */
1089 put_bh(tbh
); /* once due to original getblk in do_journal_end */
1090 atomic_dec(&(jl
->j_commit_left
));
1093 BUG_ON(atomic_read(&(jl
->j_commit_left
)) != 1);
1095 /* If there was a write error in the journal - we can't commit
1096 * this transaction - it will be invalid and, if successful,
1097 * will just end up propagating the write error out to
1098 * the file system. */
1099 if (likely(!retval
&& !reiserfs_is_journal_aborted (journal
))) {
1100 if (buffer_dirty(jl
->j_commit_bh
))
1102 mark_buffer_dirty(jl
->j_commit_bh
) ;
1103 reiserfs_write_unlock(s
);
1104 if (reiserfs_barrier_flush(s
))
1105 __sync_dirty_buffer(jl
->j_commit_bh
, WRITE_FLUSH_FUA
);
1107 sync_dirty_buffer(jl
->j_commit_bh
);
1108 reiserfs_write_lock(s
);
1111 /* If there was a write error in the journal - we can't commit this
1112 * transaction - it will be invalid and, if successful, will just end
1113 * up propagating the write error out to the filesystem. */
1114 if (unlikely(!buffer_uptodate(jl
->j_commit_bh
))) {
1115 #ifdef CONFIG_REISERFS_CHECK
1116 reiserfs_warning(s
, "journal-615", "buffer write failed");
1120 bforget(jl
->j_commit_bh
);
1121 if (journal
->j_last_commit_id
!= 0 &&
1122 (jl
->j_trans_id
- journal
->j_last_commit_id
) != 1) {
1123 reiserfs_warning(s
, "clm-2200", "last commit %lu, current %lu",
1124 journal
->j_last_commit_id
, jl
->j_trans_id
);
1126 journal
->j_last_commit_id
= jl
->j_trans_id
;
1128 /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */
1129 cleanup_freed_for_journal_list(s
, jl
);
1131 retval
= retval
? retval
: journal
->j_errno
;
1133 /* mark the metadata dirty */
1135 dirty_one_transaction(s
, jl
);
1136 atomic_dec(&(jl
->j_commit_left
));
1139 atomic_set(&(jl
->j_older_commits_done
), 1);
1141 mutex_unlock(&jl
->j_commit_mutex
);
1143 put_journal_list(s
, jl
);
1146 reiserfs_abort(s
, retval
, "Journal write error in %s",
1153 ** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or
1154 ** returns NULL if it can't find anything
1156 static struct reiserfs_journal_list
*find_newer_jl_for_cn(struct
1157 reiserfs_journal_cnode
1160 struct super_block
*sb
= cn
->sb
;
1161 b_blocknr_t blocknr
= cn
->blocknr
;
1165 if (cn
->sb
== sb
&& cn
->blocknr
== blocknr
&& cn
->jlist
) {
1173 static int newer_jl_done(struct reiserfs_journal_cnode
*cn
)
1175 struct super_block
*sb
= cn
->sb
;
1176 b_blocknr_t blocknr
= cn
->blocknr
;
1180 if (cn
->sb
== sb
&& cn
->blocknr
== blocknr
&& cn
->jlist
&&
1181 atomic_read(&cn
->jlist
->j_commit_left
) != 0)
1188 static void remove_journal_hash(struct super_block
*,
1189 struct reiserfs_journal_cnode
**,
1190 struct reiserfs_journal_list
*, unsigned long,
1194 ** once all the real blocks have been flushed, it is safe to remove them from the
1195 ** journal list for this transaction. Aside from freeing the cnode, this also allows the
1196 ** block to be reallocated for data blocks if it had been deleted.
1198 static void remove_all_from_journal_list(struct super_block
*sb
,
1199 struct reiserfs_journal_list
*jl
,
1202 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1203 struct reiserfs_journal_cnode
*cn
, *last
;
1204 cn
= jl
->j_realblock
;
1206 /* which is better, to lock once around the whole loop, or
1207 ** to lock for each call to remove_journal_hash?
1210 if (cn
->blocknr
!= 0) {
1212 reiserfs_warning(sb
, "reiserfs-2201",
1213 "block %u, bh is %d, state %ld",
1214 cn
->blocknr
, cn
->bh
? 1 : 0,
1218 remove_journal_hash(sb
, journal
->j_list_hash_table
,
1219 jl
, cn
->blocknr
, 1);
1223 free_cnode(sb
, last
);
1225 jl
->j_realblock
= NULL
;
1229 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1230 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1231 ** releasing blocks in this transaction for reuse as data blocks.
1232 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1235 static int _update_journal_header_block(struct super_block
*sb
,
1236 unsigned long offset
,
1237 unsigned int trans_id
)
1239 struct reiserfs_journal_header
*jh
;
1240 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1242 if (reiserfs_is_journal_aborted(journal
))
1245 if (trans_id
>= journal
->j_last_flush_trans_id
) {
1246 if (buffer_locked((journal
->j_header_bh
))) {
1247 reiserfs_write_unlock(sb
);
1248 wait_on_buffer((journal
->j_header_bh
));
1249 reiserfs_write_lock(sb
);
1250 if (unlikely(!buffer_uptodate(journal
->j_header_bh
))) {
1251 #ifdef CONFIG_REISERFS_CHECK
1252 reiserfs_warning(sb
, "journal-699",
1253 "buffer write failed");
1258 journal
->j_last_flush_trans_id
= trans_id
;
1259 journal
->j_first_unflushed_offset
= offset
;
1260 jh
= (struct reiserfs_journal_header
*)(journal
->j_header_bh
->
1262 jh
->j_last_flush_trans_id
= cpu_to_le32(trans_id
);
1263 jh
->j_first_unflushed_offset
= cpu_to_le32(offset
);
1264 jh
->j_mount_id
= cpu_to_le32(journal
->j_mount_id
);
1266 set_buffer_dirty(journal
->j_header_bh
);
1267 reiserfs_write_unlock(sb
);
1269 if (reiserfs_barrier_flush(sb
))
1270 __sync_dirty_buffer(journal
->j_header_bh
, WRITE_FLUSH_FUA
);
1272 sync_dirty_buffer(journal
->j_header_bh
);
1274 reiserfs_write_lock(sb
);
1275 if (!buffer_uptodate(journal
->j_header_bh
)) {
1276 reiserfs_warning(sb
, "journal-837",
1277 "IO error during journal replay");
1284 static int update_journal_header_block(struct super_block
*sb
,
1285 unsigned long offset
,
1286 unsigned int trans_id
)
1288 return _update_journal_header_block(sb
, offset
, trans_id
);
1292 ** flush any and all journal lists older than you are
1293 ** can only be called from flush_journal_list
1295 static int flush_older_journal_lists(struct super_block
*sb
,
1296 struct reiserfs_journal_list
*jl
)
1298 struct list_head
*entry
;
1299 struct reiserfs_journal_list
*other_jl
;
1300 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1301 unsigned int trans_id
= jl
->j_trans_id
;
1303 /* we know we are the only ones flushing things, no extra race
1304 * protection is required.
1307 entry
= journal
->j_journal_list
.next
;
1309 if (entry
== &journal
->j_journal_list
)
1311 other_jl
= JOURNAL_LIST_ENTRY(entry
);
1312 if (other_jl
->j_trans_id
< trans_id
) {
1313 BUG_ON(other_jl
->j_refcount
<= 0);
1314 /* do not flush all */
1315 flush_journal_list(sb
, other_jl
, 0);
1317 /* other_jl is now deleted from the list */
1323 static void del_from_work_list(struct super_block
*s
,
1324 struct reiserfs_journal_list
*jl
)
1326 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1327 if (!list_empty(&jl
->j_working_list
)) {
1328 list_del_init(&jl
->j_working_list
);
1329 journal
->j_num_work_lists
--;
1333 /* flush a journal list, both commit and real blocks
1335 ** always set flushall to 1, unless you are calling from inside
1336 ** flush_journal_list
1338 ** IMPORTANT. This can only be called while there are no journal writers,
1339 ** and the journal is locked. That means it can only be called from
1340 ** do_journal_end, or by journal_release
1342 static int flush_journal_list(struct super_block
*s
,
1343 struct reiserfs_journal_list
*jl
, int flushall
)
1345 struct reiserfs_journal_list
*pjl
;
1346 struct reiserfs_journal_cnode
*cn
, *last
;
1350 struct buffer_head
*saved_bh
;
1351 unsigned long j_len_saved
= jl
->j_len
;
1352 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1355 BUG_ON(j_len_saved
<= 0);
1357 if (atomic_read(&journal
->j_wcount
) != 0) {
1358 reiserfs_warning(s
, "clm-2048", "called with wcount %d",
1359 atomic_read(&journal
->j_wcount
));
1361 BUG_ON(jl
->j_trans_id
== 0);
1363 /* if flushall == 0, the lock is already held */
1365 reiserfs_mutex_lock_safe(&journal
->j_flush_mutex
, s
);
1366 } else if (mutex_trylock(&journal
->j_flush_mutex
)) {
1371 if (j_len_saved
> journal
->j_trans_max
) {
1372 reiserfs_panic(s
, "journal-715", "length is %lu, trans id %lu",
1373 j_len_saved
, jl
->j_trans_id
);
1379 /* if all the work is already done, get out of here */
1380 if (atomic_read(&(jl
->j_nonzerolen
)) <= 0 &&
1381 atomic_read(&(jl
->j_commit_left
)) <= 0) {
1382 goto flush_older_and_return
;
1385 /* start by putting the commit list on disk. This will also flush
1386 ** the commit lists of any olders transactions
1388 flush_commit_list(s
, jl
, 1);
1390 if (!(jl
->j_state
& LIST_DIRTY
)
1391 && !reiserfs_is_journal_aborted(journal
))
1394 /* are we done now? */
1395 if (atomic_read(&(jl
->j_nonzerolen
)) <= 0 &&
1396 atomic_read(&(jl
->j_commit_left
)) <= 0) {
1397 goto flush_older_and_return
;
1400 /* loop through each cnode, see if we need to write it,
1401 ** or wait on a more recent transaction, or just ignore it
1403 if (atomic_read(&(journal
->j_wcount
)) != 0) {
1404 reiserfs_panic(s
, "journal-844", "journal list is flushing, "
1407 cn
= jl
->j_realblock
;
1412 /* blocknr of 0 is no longer in the hash, ignore it */
1413 if (cn
->blocknr
== 0) {
1417 /* This transaction failed commit. Don't write out to the disk */
1418 if (!(jl
->j_state
& LIST_DIRTY
))
1421 pjl
= find_newer_jl_for_cn(cn
);
1422 /* the order is important here. We check pjl to make sure we
1423 ** don't clear BH_JDirty_wait if we aren't the one writing this
1426 if (!pjl
&& cn
->bh
) {
1429 /* we do this to make sure nobody releases the buffer while
1430 ** we are working with it
1434 if (buffer_journal_dirty(saved_bh
)) {
1435 BUG_ON(!can_dirty(cn
));
1438 } else if (can_dirty(cn
)) {
1439 /* everything with !pjl && jwait should be writable */
1444 /* if someone has this block in a newer transaction, just make
1445 ** sure they are committed, and don't try writing it to disk
1448 if (atomic_read(&pjl
->j_commit_left
))
1449 flush_commit_list(s
, pjl
, 1);
1453 /* bh == NULL when the block got to disk on its own, OR,
1454 ** the block got freed in a future transaction
1456 if (saved_bh
== NULL
) {
1460 /* this should never happen. kupdate_one_transaction has this list
1461 ** locked while it works, so we should never see a buffer here that
1462 ** is not marked JDirty_wait
1464 if ((!was_jwait
) && !buffer_locked(saved_bh
)) {
1465 reiserfs_warning(s
, "journal-813",
1466 "BAD! buffer %llu %cdirty %cjwait, "
1467 "not in a newer tranasction",
1468 (unsigned long long)saved_bh
->
1469 b_blocknr
, was_dirty
? ' ' : '!',
1470 was_jwait
? ' ' : '!');
1473 /* we inc again because saved_bh gets decremented at free_cnode */
1475 set_bit(BLOCK_NEEDS_FLUSH
, &cn
->state
);
1476 lock_buffer(saved_bh
);
1477 BUG_ON(cn
->blocknr
!= saved_bh
->b_blocknr
);
1478 if (buffer_dirty(saved_bh
))
1479 submit_logged_buffer(saved_bh
);
1481 unlock_buffer(saved_bh
);
1484 reiserfs_warning(s
, "clm-2082",
1485 "Unable to flush buffer %llu in %s",
1486 (unsigned long long)saved_bh
->
1487 b_blocknr
, __func__
);
1493 /* we incremented this to keep others from taking the buffer head away */
1495 if (atomic_read(&(saved_bh
->b_count
)) < 0) {
1496 reiserfs_warning(s
, "journal-945",
1497 "saved_bh->b_count < 0");
1502 cn
= jl
->j_realblock
;
1504 if (test_bit(BLOCK_NEEDS_FLUSH
, &cn
->state
)) {
1506 reiserfs_panic(s
, "journal-1011",
1510 reiserfs_write_unlock(s
);
1511 wait_on_buffer(cn
->bh
);
1512 reiserfs_write_lock(s
);
1515 reiserfs_panic(s
, "journal-1012",
1518 if (unlikely(!buffer_uptodate(cn
->bh
))) {
1519 #ifdef CONFIG_REISERFS_CHECK
1520 reiserfs_warning(s
, "journal-949",
1521 "buffer write failed");
1525 /* note, we must clear the JDirty_wait bit after the up to date
1526 ** check, otherwise we race against our flushpage routine
1528 BUG_ON(!test_clear_buffer_journal_dirty
1531 /* drop one ref for us */
1533 /* drop one ref for journal_mark_dirty */
1534 release_buffer_page(cn
->bh
);
1541 reiserfs_abort(s
, -EIO
,
1542 "Write error while pushing transaction to disk in %s",
1544 flush_older_and_return
:
1546 /* before we can update the journal header block, we _must_ flush all
1547 ** real blocks from all older transactions to disk. This is because
1548 ** once the header block is updated, this transaction will not be
1549 ** replayed after a crash
1552 flush_older_journal_lists(s
, jl
);
1555 err
= journal
->j_errno
;
1556 /* before we can remove everything from the hash tables for this
1557 ** transaction, we must make sure it can never be replayed
1559 ** since we are only called from do_journal_end, we know for sure there
1560 ** are no allocations going on while we are flushing journal lists. So,
1561 ** we only need to update the journal header block for the last list
1564 if (!err
&& flushall
) {
1566 update_journal_header_block(s
,
1567 (jl
->j_start
+ jl
->j_len
+
1568 2) % SB_ONDISK_JOURNAL_SIZE(s
),
1571 reiserfs_abort(s
, -EIO
,
1572 "Write error while updating journal header in %s",
1575 remove_all_from_journal_list(s
, jl
, 0);
1576 list_del_init(&jl
->j_list
);
1577 journal
->j_num_lists
--;
1578 del_from_work_list(s
, jl
);
1580 if (journal
->j_last_flush_id
!= 0 &&
1581 (jl
->j_trans_id
- journal
->j_last_flush_id
) != 1) {
1582 reiserfs_warning(s
, "clm-2201", "last flush %lu, current %lu",
1583 journal
->j_last_flush_id
, jl
->j_trans_id
);
1585 journal
->j_last_flush_id
= jl
->j_trans_id
;
1587 /* not strictly required since we are freeing the list, but it should
1588 * help find code using dead lists later on
1591 atomic_set(&(jl
->j_nonzerolen
), 0);
1593 jl
->j_realblock
= NULL
;
1594 jl
->j_commit_bh
= NULL
;
1597 put_journal_list(s
, jl
);
1599 mutex_unlock(&journal
->j_flush_mutex
);
1604 static int test_transaction(struct super_block
*s
,
1605 struct reiserfs_journal_list
*jl
)
1607 struct reiserfs_journal_cnode
*cn
;
1609 if (jl
->j_len
== 0 || atomic_read(&jl
->j_nonzerolen
) == 0)
1612 cn
= jl
->j_realblock
;
1614 /* if the blocknr == 0, this has been cleared from the hash,
1617 if (cn
->blocknr
== 0) {
1620 if (cn
->bh
&& !newer_jl_done(cn
))
1629 static int write_one_transaction(struct super_block
*s
,
1630 struct reiserfs_journal_list
*jl
,
1631 struct buffer_chunk
*chunk
)
1633 struct reiserfs_journal_cnode
*cn
;
1636 jl
->j_state
|= LIST_TOUCHED
;
1637 del_from_work_list(s
, jl
);
1638 if (jl
->j_len
== 0 || atomic_read(&jl
->j_nonzerolen
) == 0) {
1642 cn
= jl
->j_realblock
;
1644 /* if the blocknr == 0, this has been cleared from the hash,
1647 if (cn
->blocknr
== 0) {
1650 if (cn
->bh
&& can_dirty(cn
) && buffer_dirty(cn
->bh
)) {
1651 struct buffer_head
*tmp_bh
;
1652 /* we can race against journal_mark_freed when we try
1653 * to lock_buffer(cn->bh), so we have to inc the buffer
1654 * count, and recheck things after locking
1658 lock_buffer(tmp_bh
);
1659 if (cn
->bh
&& can_dirty(cn
) && buffer_dirty(tmp_bh
)) {
1660 if (!buffer_journal_dirty(tmp_bh
) ||
1661 buffer_journal_prepared(tmp_bh
))
1663 add_to_chunk(chunk
, tmp_bh
, NULL
, write_chunk
);
1666 /* note, cn->bh might be null now */
1667 unlock_buffer(tmp_bh
);
1678 /* used by flush_commit_list */
1679 static int dirty_one_transaction(struct super_block
*s
,
1680 struct reiserfs_journal_list
*jl
)
1682 struct reiserfs_journal_cnode
*cn
;
1683 struct reiserfs_journal_list
*pjl
;
1686 jl
->j_state
|= LIST_DIRTY
;
1687 cn
= jl
->j_realblock
;
1689 /* look for a more recent transaction that logged this
1690 ** buffer. Only the most recent transaction with a buffer in
1691 ** it is allowed to send that buffer to disk
1693 pjl
= find_newer_jl_for_cn(cn
);
1694 if (!pjl
&& cn
->blocknr
&& cn
->bh
1695 && buffer_journal_dirty(cn
->bh
)) {
1696 BUG_ON(!can_dirty(cn
));
1697 /* if the buffer is prepared, it will either be logged
1698 * or restored. If restored, we need to make sure
1699 * it actually gets marked dirty
1701 clear_buffer_journal_new(cn
->bh
);
1702 if (buffer_journal_prepared(cn
->bh
)) {
1703 set_buffer_journal_restore_dirty(cn
->bh
);
1705 set_buffer_journal_test(cn
->bh
);
1706 mark_buffer_dirty(cn
->bh
);
1714 static int kupdate_transactions(struct super_block
*s
,
1715 struct reiserfs_journal_list
*jl
,
1716 struct reiserfs_journal_list
**next_jl
,
1717 unsigned int *next_trans_id
,
1718 int num_blocks
, int num_trans
)
1722 int transactions_flushed
= 0;
1723 unsigned int orig_trans_id
= jl
->j_trans_id
;
1724 struct buffer_chunk chunk
;
1725 struct list_head
*entry
;
1726 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1729 reiserfs_mutex_lock_safe(&journal
->j_flush_mutex
, s
);
1730 if (!journal_list_still_alive(s
, orig_trans_id
)) {
1734 /* we've got j_flush_mutex held, nobody is going to delete any
1735 * of these lists out from underneath us
1737 while ((num_trans
&& transactions_flushed
< num_trans
) ||
1738 (!num_trans
&& written
< num_blocks
)) {
1740 if (jl
->j_len
== 0 || (jl
->j_state
& LIST_TOUCHED
) ||
1741 atomic_read(&jl
->j_commit_left
)
1742 || !(jl
->j_state
& LIST_DIRTY
)) {
1743 del_from_work_list(s
, jl
);
1746 ret
= write_one_transaction(s
, jl
, &chunk
);
1750 transactions_flushed
++;
1752 entry
= jl
->j_list
.next
;
1755 if (entry
== &journal
->j_journal_list
) {
1758 jl
= JOURNAL_LIST_ENTRY(entry
);
1760 /* don't bother with older transactions */
1761 if (jl
->j_trans_id
<= orig_trans_id
)
1765 write_chunk(&chunk
);
1769 mutex_unlock(&journal
->j_flush_mutex
);
1773 /* for o_sync and fsync heavy applications, they tend to use
1774 ** all the journa list slots with tiny transactions. These
1775 ** trigger lots and lots of calls to update the header block, which
1776 ** adds seeks and slows things down.
1778 ** This function tries to clear out a large chunk of the journal lists
1779 ** at once, which makes everything faster since only the newest journal
1780 ** list updates the header block
1782 static int flush_used_journal_lists(struct super_block
*s
,
1783 struct reiserfs_journal_list
*jl
)
1785 unsigned long len
= 0;
1786 unsigned long cur_len
;
1790 struct reiserfs_journal_list
*tjl
;
1791 struct reiserfs_journal_list
*flush_jl
;
1792 unsigned int trans_id
;
1793 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1795 flush_jl
= tjl
= jl
;
1797 /* in data logging mode, try harder to flush a lot of blocks */
1798 if (reiserfs_data_log(s
))
1800 /* flush for 256 transactions or limit blocks, whichever comes first */
1801 for (i
= 0; i
< 256 && len
< limit
; i
++) {
1802 if (atomic_read(&tjl
->j_commit_left
) ||
1803 tjl
->j_trans_id
< jl
->j_trans_id
) {
1806 cur_len
= atomic_read(&tjl
->j_nonzerolen
);
1808 tjl
->j_state
&= ~LIST_TOUCHED
;
1812 if (tjl
->j_list
.next
== &journal
->j_journal_list
)
1814 tjl
= JOURNAL_LIST_ENTRY(tjl
->j_list
.next
);
1816 /* try to find a group of blocks we can flush across all the
1817 ** transactions, but only bother if we've actually spanned
1818 ** across multiple lists
1820 if (flush_jl
!= jl
) {
1821 ret
= kupdate_transactions(s
, jl
, &tjl
, &trans_id
, len
, i
);
1823 flush_journal_list(s
, flush_jl
, 1);
1828 ** removes any nodes in table with name block and dev as bh.
1829 ** only touchs the hnext and hprev pointers.
1831 void remove_journal_hash(struct super_block
*sb
,
1832 struct reiserfs_journal_cnode
**table
,
1833 struct reiserfs_journal_list
*jl
,
1834 unsigned long block
, int remove_freed
)
1836 struct reiserfs_journal_cnode
*cur
;
1837 struct reiserfs_journal_cnode
**head
;
1839 head
= &(journal_hash(table
, sb
, block
));
1845 if (cur
->blocknr
== block
&& cur
->sb
== sb
1846 && (jl
== NULL
|| jl
== cur
->jlist
)
1847 && (!test_bit(BLOCK_FREED
, &cur
->state
) || remove_freed
)) {
1849 cur
->hnext
->hprev
= cur
->hprev
;
1852 cur
->hprev
->hnext
= cur
->hnext
;
1859 if (cur
->bh
&& cur
->jlist
) /* anybody who clears the cur->bh will also dec the nonzerolen */
1860 atomic_dec(&(cur
->jlist
->j_nonzerolen
));
1868 static void free_journal_ram(struct super_block
*sb
)
1870 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1871 kfree(journal
->j_current_jl
);
1872 journal
->j_num_lists
--;
1874 vfree(journal
->j_cnode_free_orig
);
1875 free_list_bitmaps(sb
, journal
->j_list_bitmap
);
1876 free_bitmap_nodes(sb
); /* must be after free_list_bitmaps */
1877 if (journal
->j_header_bh
) {
1878 brelse(journal
->j_header_bh
);
1880 /* j_header_bh is on the journal dev, make sure not to release the journal
1881 * dev until we brelse j_header_bh
1883 release_journal_dev(sb
, journal
);
1888 ** call on unmount. Only set error to 1 if you haven't made your way out
1889 ** of read_super() yet. Any other caller must keep error at 0.
1891 static int do_journal_release(struct reiserfs_transaction_handle
*th
,
1892 struct super_block
*sb
, int error
)
1894 struct reiserfs_transaction_handle myth
;
1896 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1898 /* we only want to flush out transactions if we were called with error == 0
1900 if (!error
&& !(sb
->s_flags
& MS_RDONLY
)) {
1901 /* end the current trans */
1902 BUG_ON(!th
->t_trans_id
);
1903 do_journal_end(th
, sb
, 10, FLUSH_ALL
);
1905 /* make sure something gets logged to force our way into the flush code */
1906 if (!journal_join(&myth
, sb
, 1)) {
1907 reiserfs_prepare_for_journal(sb
,
1908 SB_BUFFER_WITH_SB(sb
),
1910 journal_mark_dirty(&myth
, sb
,
1911 SB_BUFFER_WITH_SB(sb
));
1912 do_journal_end(&myth
, sb
, 1, FLUSH_ALL
);
1917 /* this also catches errors during the do_journal_end above */
1918 if (!error
&& reiserfs_is_journal_aborted(journal
)) {
1919 memset(&myth
, 0, sizeof(myth
));
1920 if (!journal_join_abort(&myth
, sb
, 1)) {
1921 reiserfs_prepare_for_journal(sb
,
1922 SB_BUFFER_WITH_SB(sb
),
1924 journal_mark_dirty(&myth
, sb
,
1925 SB_BUFFER_WITH_SB(sb
));
1926 do_journal_end(&myth
, sb
, 1, FLUSH_ALL
);
1930 reiserfs_mounted_fs_count
--;
1931 /* wait for all commits to finish */
1932 cancel_delayed_work(&SB_JOURNAL(sb
)->j_work
);
1935 * We must release the write lock here because
1936 * the workqueue job (flush_async_commit) needs this lock
1938 reiserfs_write_unlock(sb
);
1939 flush_workqueue(commit_wq
);
1941 if (!reiserfs_mounted_fs_count
) {
1942 destroy_workqueue(commit_wq
);
1946 free_journal_ram(sb
);
1948 reiserfs_write_lock(sb
);
1954 ** call on unmount. flush all journal trans, release all alloc'd ram
1956 int journal_release(struct reiserfs_transaction_handle
*th
,
1957 struct super_block
*sb
)
1959 return do_journal_release(th
, sb
, 0);
1963 ** only call from an error condition inside reiserfs_read_super!
1965 int journal_release_error(struct reiserfs_transaction_handle
*th
,
1966 struct super_block
*sb
)
1968 return do_journal_release(th
, sb
, 1);
1971 /* compares description block with commit block. returns 1 if they differ, 0 if they are the same */
1972 static int journal_compare_desc_commit(struct super_block
*sb
,
1973 struct reiserfs_journal_desc
*desc
,
1974 struct reiserfs_journal_commit
*commit
)
1976 if (get_commit_trans_id(commit
) != get_desc_trans_id(desc
) ||
1977 get_commit_trans_len(commit
) != get_desc_trans_len(desc
) ||
1978 get_commit_trans_len(commit
) > SB_JOURNAL(sb
)->j_trans_max
||
1979 get_commit_trans_len(commit
) <= 0) {
1985 /* returns 0 if it did not find a description block
1986 ** returns -1 if it found a corrupt commit block
1987 ** returns 1 if both desc and commit were valid
1989 static int journal_transaction_is_valid(struct super_block
*sb
,
1990 struct buffer_head
*d_bh
,
1991 unsigned int *oldest_invalid_trans_id
,
1992 unsigned long *newest_mount_id
)
1994 struct reiserfs_journal_desc
*desc
;
1995 struct reiserfs_journal_commit
*commit
;
1996 struct buffer_head
*c_bh
;
1997 unsigned long offset
;
2002 desc
= (struct reiserfs_journal_desc
*)d_bh
->b_data
;
2003 if (get_desc_trans_len(desc
) > 0
2004 && !memcmp(get_journal_desc_magic(d_bh
), JOURNAL_DESC_MAGIC
, 8)) {
2005 if (oldest_invalid_trans_id
&& *oldest_invalid_trans_id
2006 && get_desc_trans_id(desc
) > *oldest_invalid_trans_id
) {
2007 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2008 "journal-986: transaction "
2009 "is valid returning because trans_id %d is greater than "
2010 "oldest_invalid %lu",
2011 get_desc_trans_id(desc
),
2012 *oldest_invalid_trans_id
);
2016 && *newest_mount_id
> get_desc_mount_id(desc
)) {
2017 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2018 "journal-1087: transaction "
2019 "is valid returning because mount_id %d is less than "
2020 "newest_mount_id %lu",
2021 get_desc_mount_id(desc
),
2025 if (get_desc_trans_len(desc
) > SB_JOURNAL(sb
)->j_trans_max
) {
2026 reiserfs_warning(sb
, "journal-2018",
2027 "Bad transaction length %d "
2028 "encountered, ignoring transaction",
2029 get_desc_trans_len(desc
));
2032 offset
= d_bh
->b_blocknr
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
);
2034 /* ok, we have a journal description block, lets see if the transaction was valid */
2037 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2038 ((offset
+ get_desc_trans_len(desc
) +
2039 1) % SB_ONDISK_JOURNAL_SIZE(sb
)));
2042 commit
= (struct reiserfs_journal_commit
*)c_bh
->b_data
;
2043 if (journal_compare_desc_commit(sb
, desc
, commit
)) {
2044 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2045 "journal_transaction_is_valid, commit offset %ld had bad "
2046 "time %d or length %d",
2048 SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2049 get_commit_trans_id(commit
),
2050 get_commit_trans_len(commit
));
2052 if (oldest_invalid_trans_id
) {
2053 *oldest_invalid_trans_id
=
2054 get_desc_trans_id(desc
);
2055 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2057 "transaction_is_valid setting oldest invalid trans_id "
2059 get_desc_trans_id(desc
));
2064 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2065 "journal-1006: found valid "
2066 "transaction start offset %llu, len %d id %d",
2068 SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2069 get_desc_trans_len(desc
),
2070 get_desc_trans_id(desc
));
2077 static void brelse_array(struct buffer_head
**heads
, int num
)
2080 for (i
= 0; i
< num
; i
++) {
2086 ** given the start, and values for the oldest acceptable transactions,
2087 ** this either reads in a replays a transaction, or returns because the transaction
2088 ** is invalid, or too old.
2090 static int journal_read_transaction(struct super_block
*sb
,
2091 unsigned long cur_dblock
,
2092 unsigned long oldest_start
,
2093 unsigned int oldest_trans_id
,
2094 unsigned long newest_mount_id
)
2096 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
2097 struct reiserfs_journal_desc
*desc
;
2098 struct reiserfs_journal_commit
*commit
;
2099 unsigned int trans_id
= 0;
2100 struct buffer_head
*c_bh
;
2101 struct buffer_head
*d_bh
;
2102 struct buffer_head
**log_blocks
= NULL
;
2103 struct buffer_head
**real_blocks
= NULL
;
2104 unsigned int trans_offset
;
2108 d_bh
= journal_bread(sb
, cur_dblock
);
2111 desc
= (struct reiserfs_journal_desc
*)d_bh
->b_data
;
2112 trans_offset
= d_bh
->b_blocknr
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
);
2113 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
, "journal-1037: "
2114 "journal_read_transaction, offset %llu, len %d mount_id %d",
2115 d_bh
->b_blocknr
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2116 get_desc_trans_len(desc
), get_desc_mount_id(desc
));
2117 if (get_desc_trans_id(desc
) < oldest_trans_id
) {
2118 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
, "journal-1039: "
2119 "journal_read_trans skipping because %lu is too old",
2121 SB_ONDISK_JOURNAL_1st_BLOCK(sb
));
2125 if (get_desc_mount_id(desc
) != newest_mount_id
) {
2126 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
, "journal-1146: "
2127 "journal_read_trans skipping because %d is != "
2128 "newest_mount_id %lu", get_desc_mount_id(desc
),
2133 c_bh
= journal_bread(sb
, SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2134 ((trans_offset
+ get_desc_trans_len(desc
) + 1) %
2135 SB_ONDISK_JOURNAL_SIZE(sb
)));
2140 commit
= (struct reiserfs_journal_commit
*)c_bh
->b_data
;
2141 if (journal_compare_desc_commit(sb
, desc
, commit
)) {
2142 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2143 "journal_read_transaction, "
2144 "commit offset %llu had bad time %d or length %d",
2146 SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2147 get_commit_trans_id(commit
),
2148 get_commit_trans_len(commit
));
2154 if (bdev_read_only(sb
->s_bdev
)) {
2155 reiserfs_warning(sb
, "clm-2076",
2156 "device is readonly, unable to replay log");
2162 trans_id
= get_desc_trans_id(desc
);
2163 /* now we know we've got a good transaction, and it was inside the valid time ranges */
2164 log_blocks
= kmalloc(get_desc_trans_len(desc
) *
2165 sizeof(struct buffer_head
*), GFP_NOFS
);
2166 real_blocks
= kmalloc(get_desc_trans_len(desc
) *
2167 sizeof(struct buffer_head
*), GFP_NOFS
);
2168 if (!log_blocks
|| !real_blocks
) {
2173 reiserfs_warning(sb
, "journal-1169",
2174 "kmalloc failed, unable to mount FS");
2177 /* get all the buffer heads */
2178 trans_half
= journal_trans_half(sb
->s_blocksize
);
2179 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2182 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2184 i
) % SB_ONDISK_JOURNAL_SIZE(sb
));
2185 if (i
< trans_half
) {
2188 le32_to_cpu(desc
->j_realblock
[i
]));
2192 le32_to_cpu(commit
->
2193 j_realblock
[i
- trans_half
]));
2195 if (real_blocks
[i
]->b_blocknr
> SB_BLOCK_COUNT(sb
)) {
2196 reiserfs_warning(sb
, "journal-1207",
2197 "REPLAY FAILURE fsck required! "
2198 "Block to replay is outside of "
2202 /* make sure we don't try to replay onto log or reserved area */
2203 if (is_block_in_log_or_reserved_area
2204 (sb
, real_blocks
[i
]->b_blocknr
)) {
2205 reiserfs_warning(sb
, "journal-1204",
2206 "REPLAY FAILURE fsck required! "
2207 "Trying to replay onto a log block");
2209 brelse_array(log_blocks
, i
);
2210 brelse_array(real_blocks
, i
);
2218 /* read in the log blocks, memcpy to the corresponding real block */
2219 ll_rw_block(READ
, get_desc_trans_len(desc
), log_blocks
);
2220 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2222 reiserfs_write_unlock(sb
);
2223 wait_on_buffer(log_blocks
[i
]);
2224 reiserfs_write_lock(sb
);
2226 if (!buffer_uptodate(log_blocks
[i
])) {
2227 reiserfs_warning(sb
, "journal-1212",
2228 "REPLAY FAILURE fsck required! "
2229 "buffer write failed");
2230 brelse_array(log_blocks
+ i
,
2231 get_desc_trans_len(desc
) - i
);
2232 brelse_array(real_blocks
, get_desc_trans_len(desc
));
2239 memcpy(real_blocks
[i
]->b_data
, log_blocks
[i
]->b_data
,
2240 real_blocks
[i
]->b_size
);
2241 set_buffer_uptodate(real_blocks
[i
]);
2242 brelse(log_blocks
[i
]);
2244 /* flush out the real blocks */
2245 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2246 set_buffer_dirty(real_blocks
[i
]);
2247 write_dirty_buffer(real_blocks
[i
], WRITE
);
2249 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2250 wait_on_buffer(real_blocks
[i
]);
2251 if (!buffer_uptodate(real_blocks
[i
])) {
2252 reiserfs_warning(sb
, "journal-1226",
2253 "REPLAY FAILURE, fsck required! "
2254 "buffer write failed");
2255 brelse_array(real_blocks
+ i
,
2256 get_desc_trans_len(desc
) - i
);
2263 brelse(real_blocks
[i
]);
2266 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2267 ((trans_offset
+ get_desc_trans_len(desc
) +
2268 2) % SB_ONDISK_JOURNAL_SIZE(sb
));
2269 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2270 "journal-1095: setting journal " "start to offset %ld",
2271 cur_dblock
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
));
2273 /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2274 journal
->j_start
= cur_dblock
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
);
2275 journal
->j_last_flush_trans_id
= trans_id
;
2276 journal
->j_trans_id
= trans_id
+ 1;
2277 /* check for trans_id overflow */
2278 if (journal
->j_trans_id
== 0)
2279 journal
->j_trans_id
= 10;
2287 /* This function reads blocks starting from block and to max_block of bufsize
2288 size (but no more than BUFNR blocks at a time). This proved to improve
2289 mounting speed on self-rebuilding raid5 arrays at least.
2290 Right now it is only used from journal code. But later we might use it
2292 Note: Do not use journal_getblk/sb_getblk functions here! */
2293 static struct buffer_head
*reiserfs_breada(struct block_device
*dev
,
2294 b_blocknr_t block
, int bufsize
,
2295 b_blocknr_t max_block
)
2297 struct buffer_head
*bhlist
[BUFNR
];
2298 unsigned int blocks
= BUFNR
;
2299 struct buffer_head
*bh
;
2302 bh
= __getblk(dev
, block
, bufsize
);
2303 if (buffer_uptodate(bh
))
2306 if (block
+ BUFNR
> max_block
) {
2307 blocks
= max_block
- block
;
2311 for (i
= 1; i
< blocks
; i
++) {
2312 bh
= __getblk(dev
, block
+ i
, bufsize
);
2313 if (buffer_uptodate(bh
)) {
2319 ll_rw_block(READ
, j
, bhlist
);
2320 for (i
= 1; i
< j
; i
++)
2324 if (buffer_uptodate(bh
))
2331 ** read and replay the log
2332 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2333 ** transaction. This tests that before finding all the transactions in the log, which makes normal mount times fast.
2335 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2337 ** On exit, it sets things up so the first transaction will work correctly.
2339 static int journal_read(struct super_block
*sb
)
2341 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
2342 struct reiserfs_journal_desc
*desc
;
2343 unsigned int oldest_trans_id
= 0;
2344 unsigned int oldest_invalid_trans_id
= 0;
2346 unsigned long oldest_start
= 0;
2347 unsigned long cur_dblock
= 0;
2348 unsigned long newest_mount_id
= 9;
2349 struct buffer_head
*d_bh
;
2350 struct reiserfs_journal_header
*jh
;
2351 int valid_journal_header
= 0;
2352 int replay_count
= 0;
2353 int continue_replay
= 1;
2355 char b
[BDEVNAME_SIZE
];
2357 cur_dblock
= SB_ONDISK_JOURNAL_1st_BLOCK(sb
);
2358 reiserfs_info(sb
, "checking transaction log (%s)\n",
2359 bdevname(journal
->j_dev_bd
, b
));
2360 start
= get_seconds();
2362 /* step 1, read in the journal header block. Check the transaction it says
2363 ** is the first unflushed, and if that transaction is not valid,
2366 journal
->j_header_bh
= journal_bread(sb
,
2367 SB_ONDISK_JOURNAL_1st_BLOCK(sb
)
2368 + SB_ONDISK_JOURNAL_SIZE(sb
));
2369 if (!journal
->j_header_bh
) {
2372 jh
= (struct reiserfs_journal_header
*)(journal
->j_header_bh
->b_data
);
2373 if (le32_to_cpu(jh
->j_first_unflushed_offset
) <
2374 SB_ONDISK_JOURNAL_SIZE(sb
)
2375 && le32_to_cpu(jh
->j_last_flush_trans_id
) > 0) {
2377 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2378 le32_to_cpu(jh
->j_first_unflushed_offset
);
2379 oldest_trans_id
= le32_to_cpu(jh
->j_last_flush_trans_id
) + 1;
2380 newest_mount_id
= le32_to_cpu(jh
->j_mount_id
);
2381 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2382 "journal-1153: found in "
2383 "header: first_unflushed_offset %d, last_flushed_trans_id "
2384 "%lu", le32_to_cpu(jh
->j_first_unflushed_offset
),
2385 le32_to_cpu(jh
->j_last_flush_trans_id
));
2386 valid_journal_header
= 1;
2388 /* now, we try to read the first unflushed offset. If it is not valid,
2389 ** there is nothing more we can do, and it makes no sense to read
2390 ** through the whole log.
2394 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2395 le32_to_cpu(jh
->j_first_unflushed_offset
));
2396 ret
= journal_transaction_is_valid(sb
, d_bh
, NULL
, NULL
);
2398 continue_replay
= 0;
2401 goto start_log_replay
;
2404 /* ok, there are transactions that need to be replayed. start with the first log block, find
2405 ** all the valid transactions, and pick out the oldest.
2407 while (continue_replay
2409 (SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2410 SB_ONDISK_JOURNAL_SIZE(sb
))) {
2411 /* Note that it is required for blocksize of primary fs device and journal
2412 device to be the same */
2414 reiserfs_breada(journal
->j_dev_bd
, cur_dblock
,
2416 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2417 SB_ONDISK_JOURNAL_SIZE(sb
));
2419 journal_transaction_is_valid(sb
, d_bh
,
2420 &oldest_invalid_trans_id
,
2423 desc
= (struct reiserfs_journal_desc
*)d_bh
->b_data
;
2424 if (oldest_start
== 0) { /* init all oldest_ values */
2425 oldest_trans_id
= get_desc_trans_id(desc
);
2426 oldest_start
= d_bh
->b_blocknr
;
2427 newest_mount_id
= get_desc_mount_id(desc
);
2428 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2429 "journal-1179: Setting "
2430 "oldest_start to offset %llu, trans_id %lu",
2432 SB_ONDISK_JOURNAL_1st_BLOCK
2433 (sb
), oldest_trans_id
);
2434 } else if (oldest_trans_id
> get_desc_trans_id(desc
)) {
2435 /* one we just read was older */
2436 oldest_trans_id
= get_desc_trans_id(desc
);
2437 oldest_start
= d_bh
->b_blocknr
;
2438 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2439 "journal-1180: Resetting "
2440 "oldest_start to offset %lu, trans_id %lu",
2442 SB_ONDISK_JOURNAL_1st_BLOCK
2443 (sb
), oldest_trans_id
);
2445 if (newest_mount_id
< get_desc_mount_id(desc
)) {
2446 newest_mount_id
= get_desc_mount_id(desc
);
2447 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2448 "journal-1299: Setting "
2449 "newest_mount_id to %d",
2450 get_desc_mount_id(desc
));
2452 cur_dblock
+= get_desc_trans_len(desc
) + 2;
2460 cur_dblock
= oldest_start
;
2461 if (oldest_trans_id
) {
2462 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2463 "journal-1206: Starting replay "
2464 "from offset %llu, trans_id %lu",
2465 cur_dblock
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2470 while (continue_replay
&& oldest_trans_id
> 0) {
2472 journal_read_transaction(sb
, cur_dblock
, oldest_start
,
2473 oldest_trans_id
, newest_mount_id
);
2476 } else if (ret
!= 0) {
2480 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) + journal
->j_start
;
2482 if (cur_dblock
== oldest_start
)
2486 if (oldest_trans_id
== 0) {
2487 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2488 "journal-1225: No valid " "transactions found");
2490 /* j_start does not get set correctly if we don't replay any transactions.
2491 ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2492 ** copy the trans_id from the header
2494 if (valid_journal_header
&& replay_count
== 0) {
2495 journal
->j_start
= le32_to_cpu(jh
->j_first_unflushed_offset
);
2496 journal
->j_trans_id
=
2497 le32_to_cpu(jh
->j_last_flush_trans_id
) + 1;
2498 /* check for trans_id overflow */
2499 if (journal
->j_trans_id
== 0)
2500 journal
->j_trans_id
= 10;
2501 journal
->j_last_flush_trans_id
=
2502 le32_to_cpu(jh
->j_last_flush_trans_id
);
2503 journal
->j_mount_id
= le32_to_cpu(jh
->j_mount_id
) + 1;
2505 journal
->j_mount_id
= newest_mount_id
+ 1;
2507 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
, "journal-1299: Setting "
2508 "newest_mount_id to %lu", journal
->j_mount_id
);
2509 journal
->j_first_unflushed_offset
= journal
->j_start
;
2510 if (replay_count
> 0) {
2512 "replayed %d transactions in %lu seconds\n",
2513 replay_count
, get_seconds() - start
);
2515 if (!bdev_read_only(sb
->s_bdev
) &&
2516 _update_journal_header_block(sb
, journal
->j_start
,
2517 journal
->j_last_flush_trans_id
)) {
2518 /* replay failed, caller must call free_journal_ram and abort
2526 static struct reiserfs_journal_list
*alloc_journal_list(struct super_block
*s
)
2528 struct reiserfs_journal_list
*jl
;
2529 jl
= kzalloc(sizeof(struct reiserfs_journal_list
),
2530 GFP_NOFS
| __GFP_NOFAIL
);
2531 INIT_LIST_HEAD(&jl
->j_list
);
2532 INIT_LIST_HEAD(&jl
->j_working_list
);
2533 INIT_LIST_HEAD(&jl
->j_tail_bh_list
);
2534 INIT_LIST_HEAD(&jl
->j_bh_list
);
2535 mutex_init(&jl
->j_commit_mutex
);
2536 SB_JOURNAL(s
)->j_num_lists
++;
2537 get_journal_list(jl
);
2541 static void journal_list_init(struct super_block
*sb
)
2543 SB_JOURNAL(sb
)->j_current_jl
= alloc_journal_list(sb
);
2546 static int release_journal_dev(struct super_block
*super
,
2547 struct reiserfs_journal
*journal
)
2553 if (journal
->j_dev_bd
!= NULL
) {
2554 result
= blkdev_put(journal
->j_dev_bd
, journal
->j_dev_mode
);
2555 journal
->j_dev_bd
= NULL
;
2559 reiserfs_warning(super
, "sh-457",
2560 "Cannot release journal device: %i", result
);
2565 static int journal_init_dev(struct super_block
*super
,
2566 struct reiserfs_journal
*journal
,
2567 const char *jdev_name
)
2571 fmode_t blkdev_mode
= FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
;
2572 char b
[BDEVNAME_SIZE
];
2576 journal
->j_dev_bd
= NULL
;
2577 jdev
= SB_ONDISK_JOURNAL_DEVICE(super
) ?
2578 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super
)) : super
->s_dev
;
2580 if (bdev_read_only(super
->s_bdev
))
2581 blkdev_mode
= FMODE_READ
;
2583 /* there is no "jdev" option and journal is on separate device */
2584 if ((!jdev_name
|| !jdev_name
[0])) {
2585 if (jdev
== super
->s_dev
)
2586 blkdev_mode
&= ~FMODE_EXCL
;
2587 journal
->j_dev_bd
= blkdev_get_by_dev(jdev
, blkdev_mode
,
2589 journal
->j_dev_mode
= blkdev_mode
;
2590 if (IS_ERR(journal
->j_dev_bd
)) {
2591 result
= PTR_ERR(journal
->j_dev_bd
);
2592 journal
->j_dev_bd
= NULL
;
2593 reiserfs_warning(super
, "sh-458",
2594 "cannot init journal device '%s': %i",
2595 __bdevname(jdev
, b
), result
);
2597 } else if (jdev
!= super
->s_dev
)
2598 set_blocksize(journal
->j_dev_bd
, super
->s_blocksize
);
2603 journal
->j_dev_mode
= blkdev_mode
;
2604 journal
->j_dev_bd
= blkdev_get_by_path(jdev_name
, blkdev_mode
, journal
);
2605 if (IS_ERR(journal
->j_dev_bd
)) {
2606 result
= PTR_ERR(journal
->j_dev_bd
);
2607 journal
->j_dev_bd
= NULL
;
2608 reiserfs_warning(super
,
2609 "journal_init_dev: Cannot open '%s': %i",
2614 set_blocksize(journal
->j_dev_bd
, super
->s_blocksize
);
2615 reiserfs_info(super
,
2616 "journal_init_dev: journal device: %s\n",
2617 bdevname(journal
->j_dev_bd
, b
));
2622 * When creating/tuning a file system user can assign some
2623 * journal params within boundaries which depend on the ratio
2624 * blocksize/standard_blocksize.
2626 * For blocks >= standard_blocksize transaction size should
2627 * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2628 * then JOURNAL_TRANS_MAX_DEFAULT.
2630 * For blocks < standard_blocksize these boundaries should be
2631 * decreased proportionally.
2633 #define REISERFS_STANDARD_BLKSIZE (4096)
2635 static int check_advise_trans_params(struct super_block
*sb
,
2636 struct reiserfs_journal
*journal
)
2638 if (journal
->j_trans_max
) {
2639 /* Non-default journal params.
2640 Do sanity check for them. */
2642 if (sb
->s_blocksize
< REISERFS_STANDARD_BLKSIZE
)
2643 ratio
= REISERFS_STANDARD_BLKSIZE
/ sb
->s_blocksize
;
2645 if (journal
->j_trans_max
> JOURNAL_TRANS_MAX_DEFAULT
/ ratio
||
2646 journal
->j_trans_max
< JOURNAL_TRANS_MIN_DEFAULT
/ ratio
||
2647 SB_ONDISK_JOURNAL_SIZE(sb
) / journal
->j_trans_max
<
2648 JOURNAL_MIN_RATIO
) {
2649 reiserfs_warning(sb
, "sh-462",
2650 "bad transaction max size (%u). "
2651 "FSCK?", journal
->j_trans_max
);
2654 if (journal
->j_max_batch
!= (journal
->j_trans_max
) *
2655 JOURNAL_MAX_BATCH_DEFAULT
/JOURNAL_TRANS_MAX_DEFAULT
) {
2656 reiserfs_warning(sb
, "sh-463",
2657 "bad transaction max batch (%u). "
2658 "FSCK?", journal
->j_max_batch
);
2662 /* Default journal params.
2663 The file system was created by old version
2664 of mkreiserfs, so some fields contain zeros,
2665 and we need to advise proper values for them */
2666 if (sb
->s_blocksize
!= REISERFS_STANDARD_BLKSIZE
) {
2667 reiserfs_warning(sb
, "sh-464", "bad blocksize (%u)",
2671 journal
->j_trans_max
= JOURNAL_TRANS_MAX_DEFAULT
;
2672 journal
->j_max_batch
= JOURNAL_MAX_BATCH_DEFAULT
;
2673 journal
->j_max_commit_age
= JOURNAL_MAX_COMMIT_AGE
;
2679 ** must be called once on fs mount. calls journal_read for you
2681 int journal_init(struct super_block
*sb
, const char *j_dev_name
,
2682 int old_format
, unsigned int commit_max_age
)
2684 int num_cnodes
= SB_ONDISK_JOURNAL_SIZE(sb
) * 2;
2685 struct buffer_head
*bhjh
;
2686 struct reiserfs_super_block
*rs
;
2687 struct reiserfs_journal_header
*jh
;
2688 struct reiserfs_journal
*journal
;
2689 struct reiserfs_journal_list
*jl
;
2690 char b
[BDEVNAME_SIZE
];
2694 * Unlock here to avoid various RECLAIM-FS-ON <-> IN-RECLAIM-FS
2695 * dependency inversion warnings.
2697 reiserfs_write_unlock(sb
);
2698 journal
= SB_JOURNAL(sb
) = vmalloc(sizeof(struct reiserfs_journal
));
2700 reiserfs_warning(sb
, "journal-1256",
2701 "unable to get memory for journal structure");
2702 reiserfs_write_lock(sb
);
2705 memset(journal
, 0, sizeof(struct reiserfs_journal
));
2706 INIT_LIST_HEAD(&journal
->j_bitmap_nodes
);
2707 INIT_LIST_HEAD(&journal
->j_prealloc_list
);
2708 INIT_LIST_HEAD(&journal
->j_working_list
);
2709 INIT_LIST_HEAD(&journal
->j_journal_list
);
2710 journal
->j_persistent_trans
= 0;
2711 ret
= reiserfs_allocate_list_bitmaps(sb
, journal
->j_list_bitmap
,
2712 reiserfs_bmap_count(sb
));
2713 reiserfs_write_lock(sb
);
2715 goto free_and_return
;
2717 allocate_bitmap_nodes(sb
);
2719 /* reserved for journal area support */
2720 SB_JOURNAL_1st_RESERVED_BLOCK(sb
) = (old_format
?
2721 REISERFS_OLD_DISK_OFFSET_IN_BYTES
2723 reiserfs_bmap_count(sb
) +
2725 REISERFS_DISK_OFFSET_IN_BYTES
/
2726 sb
->s_blocksize
+ 2);
2728 /* Sanity check to see is the standard journal fitting withing first bitmap
2729 (actual for small blocksizes) */
2730 if (!SB_ONDISK_JOURNAL_DEVICE(sb
) &&
2731 (SB_JOURNAL_1st_RESERVED_BLOCK(sb
) +
2732 SB_ONDISK_JOURNAL_SIZE(sb
) > sb
->s_blocksize
* 8)) {
2733 reiserfs_warning(sb
, "journal-1393",
2734 "journal does not fit for area addressed "
2735 "by first of bitmap blocks. It starts at "
2736 "%u and its size is %u. Block size %ld",
2737 SB_JOURNAL_1st_RESERVED_BLOCK(sb
),
2738 SB_ONDISK_JOURNAL_SIZE(sb
),
2740 goto free_and_return
;
2744 * We need to unlock here to avoid creating the following
2746 * reiserfs_lock -> sysfs_mutex
2747 * Because the reiserfs mmap path creates the following dependency:
2748 * mm->mmap -> reiserfs_lock, hence we have
2749 * mm->mmap -> reiserfs_lock ->sysfs_mutex
2750 * This would ends up in a circular dependency with sysfs readdir path
2751 * which does sysfs_mutex -> mm->mmap_sem
2752 * This is fine because the reiserfs lock is useless in mount path,
2753 * at least until we call journal_begin. We keep it for paranoid
2756 reiserfs_write_unlock(sb
);
2757 if (journal_init_dev(sb
, journal
, j_dev_name
) != 0) {
2758 reiserfs_write_lock(sb
);
2759 reiserfs_warning(sb
, "sh-462",
2760 "unable to initialize jornal device");
2761 goto free_and_return
;
2763 reiserfs_write_lock(sb
);
2765 rs
= SB_DISK_SUPER_BLOCK(sb
);
2767 /* read journal header */
2768 bhjh
= journal_bread(sb
,
2769 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2770 SB_ONDISK_JOURNAL_SIZE(sb
));
2772 reiserfs_warning(sb
, "sh-459",
2773 "unable to read journal header");
2774 goto free_and_return
;
2776 jh
= (struct reiserfs_journal_header
*)(bhjh
->b_data
);
2778 /* make sure that journal matches to the super block */
2779 if (is_reiserfs_jr(rs
)
2780 && (le32_to_cpu(jh
->jh_journal
.jp_journal_magic
) !=
2781 sb_jp_journal_magic(rs
))) {
2782 reiserfs_warning(sb
, "sh-460",
2783 "journal header magic %x (device %s) does "
2784 "not match to magic found in super block %x",
2785 jh
->jh_journal
.jp_journal_magic
,
2786 bdevname(journal
->j_dev_bd
, b
),
2787 sb_jp_journal_magic(rs
));
2789 goto free_and_return
;
2792 journal
->j_trans_max
= le32_to_cpu(jh
->jh_journal
.jp_journal_trans_max
);
2793 journal
->j_max_batch
= le32_to_cpu(jh
->jh_journal
.jp_journal_max_batch
);
2794 journal
->j_max_commit_age
=
2795 le32_to_cpu(jh
->jh_journal
.jp_journal_max_commit_age
);
2796 journal
->j_max_trans_age
= JOURNAL_MAX_TRANS_AGE
;
2798 if (check_advise_trans_params(sb
, journal
) != 0)
2799 goto free_and_return
;
2800 journal
->j_default_max_commit_age
= journal
->j_max_commit_age
;
2802 if (commit_max_age
!= 0) {
2803 journal
->j_max_commit_age
= commit_max_age
;
2804 journal
->j_max_trans_age
= commit_max_age
;
2807 reiserfs_info(sb
, "journal params: device %s, size %u, "
2808 "journal first block %u, max trans len %u, max batch %u, "
2809 "max commit age %u, max trans age %u\n",
2810 bdevname(journal
->j_dev_bd
, b
),
2811 SB_ONDISK_JOURNAL_SIZE(sb
),
2812 SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2813 journal
->j_trans_max
,
2814 journal
->j_max_batch
,
2815 journal
->j_max_commit_age
, journal
->j_max_trans_age
);
2819 journal
->j_list_bitmap_index
= 0;
2820 journal_list_init(sb
);
2822 memset(journal
->j_list_hash_table
, 0,
2823 JOURNAL_HASH_SIZE
* sizeof(struct reiserfs_journal_cnode
*));
2825 INIT_LIST_HEAD(&journal
->j_dirty_buffers
);
2826 spin_lock_init(&journal
->j_dirty_buffers_lock
);
2828 journal
->j_start
= 0;
2830 journal
->j_len_alloc
= 0;
2831 atomic_set(&(journal
->j_wcount
), 0);
2832 atomic_set(&(journal
->j_async_throttle
), 0);
2833 journal
->j_bcount
= 0;
2834 journal
->j_trans_start_time
= 0;
2835 journal
->j_last
= NULL
;
2836 journal
->j_first
= NULL
;
2837 init_waitqueue_head(&(journal
->j_join_wait
));
2838 mutex_init(&journal
->j_mutex
);
2839 mutex_init(&journal
->j_flush_mutex
);
2841 journal
->j_trans_id
= 10;
2842 journal
->j_mount_id
= 10;
2843 journal
->j_state
= 0;
2844 atomic_set(&(journal
->j_jlock
), 0);
2845 reiserfs_write_unlock(sb
);
2846 journal
->j_cnode_free_list
= allocate_cnodes(num_cnodes
);
2847 reiserfs_write_lock(sb
);
2848 journal
->j_cnode_free_orig
= journal
->j_cnode_free_list
;
2849 journal
->j_cnode_free
= journal
->j_cnode_free_list
? num_cnodes
: 0;
2850 journal
->j_cnode_used
= 0;
2851 journal
->j_must_wait
= 0;
2853 if (journal
->j_cnode_free
== 0) {
2854 reiserfs_warning(sb
, "journal-2004", "Journal cnode memory "
2855 "allocation failed (%ld bytes). Journal is "
2856 "too large for available memory. Usually "
2857 "this is due to a journal that is too large.",
2858 sizeof (struct reiserfs_journal_cnode
) * num_cnodes
);
2859 goto free_and_return
;
2862 init_journal_hash(sb
);
2863 jl
= journal
->j_current_jl
;
2864 jl
->j_list_bitmap
= get_list_bitmap(sb
, jl
);
2865 if (!jl
->j_list_bitmap
) {
2866 reiserfs_warning(sb
, "journal-2005",
2867 "get_list_bitmap failed for journal list 0");
2868 goto free_and_return
;
2870 if (journal_read(sb
) < 0) {
2871 reiserfs_warning(sb
, "reiserfs-2006",
2872 "Replay Failure, unable to mount");
2873 goto free_and_return
;
2876 reiserfs_mounted_fs_count
++;
2877 if (reiserfs_mounted_fs_count
<= 1) {
2878 reiserfs_write_unlock(sb
);
2879 commit_wq
= alloc_workqueue("reiserfs", WQ_MEM_RECLAIM
, 0);
2880 reiserfs_write_lock(sb
);
2883 INIT_DELAYED_WORK(&journal
->j_work
, flush_async_commits
);
2884 journal
->j_work_sb
= sb
;
2887 free_journal_ram(sb
);
2892 ** test for a polite end of the current transaction. Used by file_write, and should
2893 ** be used by delete to make sure they don't write more than can fit inside a single
2896 int journal_transaction_should_end(struct reiserfs_transaction_handle
*th
,
2899 struct reiserfs_journal
*journal
= SB_JOURNAL(th
->t_super
);
2900 time_t now
= get_seconds();
2901 /* cannot restart while nested */
2902 BUG_ON(!th
->t_trans_id
);
2903 if (th
->t_refcount
> 1)
2905 if (journal
->j_must_wait
> 0 ||
2906 (journal
->j_len_alloc
+ new_alloc
) >= journal
->j_max_batch
||
2907 atomic_read(&(journal
->j_jlock
)) ||
2908 (now
- journal
->j_trans_start_time
) > journal
->j_max_trans_age
||
2909 journal
->j_cnode_free
< (journal
->j_trans_max
* 3)) {
2912 /* protected by the BKL here */
2913 journal
->j_len_alloc
+= new_alloc
;
2914 th
->t_blocks_allocated
+= new_alloc
;
2918 /* this must be called inside a transaction, and requires the
2919 ** kernel_lock to be held
2921 void reiserfs_block_writes(struct reiserfs_transaction_handle
*th
)
2923 struct reiserfs_journal
*journal
= SB_JOURNAL(th
->t_super
);
2924 BUG_ON(!th
->t_trans_id
);
2925 journal
->j_must_wait
= 1;
2926 set_bit(J_WRITERS_BLOCKED
, &journal
->j_state
);
2930 /* this must be called without a transaction started, and does not
2933 void reiserfs_allow_writes(struct super_block
*s
)
2935 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2936 clear_bit(J_WRITERS_BLOCKED
, &journal
->j_state
);
2937 wake_up(&journal
->j_join_wait
);
2940 /* this must be called without a transaction started, and does not
2943 void reiserfs_wait_on_write_block(struct super_block
*s
)
2945 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2946 wait_event(journal
->j_join_wait
,
2947 !test_bit(J_WRITERS_BLOCKED
, &journal
->j_state
));
2950 static void queue_log_writer(struct super_block
*s
)
2953 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2954 set_bit(J_WRITERS_QUEUED
, &journal
->j_state
);
2957 * we don't want to use wait_event here because
2958 * we only want to wait once.
2960 init_waitqueue_entry(&wait
, current
);
2961 add_wait_queue(&journal
->j_join_wait
, &wait
);
2962 set_current_state(TASK_UNINTERRUPTIBLE
);
2963 if (test_bit(J_WRITERS_QUEUED
, &journal
->j_state
)) {
2964 reiserfs_write_unlock(s
);
2966 reiserfs_write_lock(s
);
2968 __set_current_state(TASK_RUNNING
);
2969 remove_wait_queue(&journal
->j_join_wait
, &wait
);
2972 static void wake_queued_writers(struct super_block
*s
)
2974 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2975 if (test_and_clear_bit(J_WRITERS_QUEUED
, &journal
->j_state
))
2976 wake_up(&journal
->j_join_wait
);
2979 static void let_transaction_grow(struct super_block
*sb
, unsigned int trans_id
)
2981 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
2982 unsigned long bcount
= journal
->j_bcount
;
2984 reiserfs_write_unlock(sb
);
2985 schedule_timeout_uninterruptible(1);
2986 reiserfs_write_lock(sb
);
2987 journal
->j_current_jl
->j_state
|= LIST_COMMIT_PENDING
;
2988 while ((atomic_read(&journal
->j_wcount
) > 0 ||
2989 atomic_read(&journal
->j_jlock
)) &&
2990 journal
->j_trans_id
== trans_id
) {
2991 queue_log_writer(sb
);
2993 if (journal
->j_trans_id
!= trans_id
)
2995 if (bcount
== journal
->j_bcount
)
2997 bcount
= journal
->j_bcount
;
3001 /* join == true if you must join an existing transaction.
3002 ** join == false if you can deal with waiting for others to finish
3004 ** this will block until the transaction is joinable. send the number of blocks you
3005 ** expect to use in nblocks.
3007 static int do_journal_begin_r(struct reiserfs_transaction_handle
*th
,
3008 struct super_block
*sb
, unsigned long nblocks
,
3011 time_t now
= get_seconds();
3012 unsigned int old_trans_id
;
3013 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3014 struct reiserfs_transaction_handle myth
;
3015 int sched_count
= 0;
3018 reiserfs_check_lock_depth(sb
, "journal_begin");
3019 BUG_ON(nblocks
> journal
->j_trans_max
);
3021 PROC_INFO_INC(sb
, journal
.journal_being
);
3022 /* set here for journal_join */
3028 if (join
!= JBEGIN_ABORT
&& reiserfs_is_journal_aborted(journal
)) {
3030 retval
= journal
->j_errno
;
3033 journal
->j_bcount
++;
3035 if (test_bit(J_WRITERS_BLOCKED
, &journal
->j_state
)) {
3037 reiserfs_write_unlock(sb
);
3038 reiserfs_wait_on_write_block(sb
);
3039 reiserfs_write_lock(sb
);
3040 PROC_INFO_INC(sb
, journal
.journal_relock_writers
);
3043 now
= get_seconds();
3045 /* if there is no room in the journal OR
3046 ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
3047 ** we don't sleep if there aren't other writers
3050 if ((!join
&& journal
->j_must_wait
> 0) ||
3052 && (journal
->j_len_alloc
+ nblocks
+ 2) >= journal
->j_max_batch
)
3053 || (!join
&& atomic_read(&journal
->j_wcount
) > 0
3054 && journal
->j_trans_start_time
> 0
3055 && (now
- journal
->j_trans_start_time
) >
3056 journal
->j_max_trans_age
) || (!join
3057 && atomic_read(&journal
->j_jlock
))
3058 || (!join
&& journal
->j_cnode_free
< (journal
->j_trans_max
* 3))) {
3060 old_trans_id
= journal
->j_trans_id
;
3061 unlock_journal(sb
); /* allow others to finish this transaction */
3063 if (!join
&& (journal
->j_len_alloc
+ nblocks
+ 2) >=
3064 journal
->j_max_batch
&&
3065 ((journal
->j_len
+ nblocks
+ 2) * 100) <
3066 (journal
->j_len_alloc
* 75)) {
3067 if (atomic_read(&journal
->j_wcount
) > 10) {
3069 queue_log_writer(sb
);
3073 /* don't mess with joining the transaction if all we have to do is
3074 * wait for someone else to do a commit
3076 if (atomic_read(&journal
->j_jlock
)) {
3077 while (journal
->j_trans_id
== old_trans_id
&&
3078 atomic_read(&journal
->j_jlock
)) {
3079 queue_log_writer(sb
);
3083 retval
= journal_join(&myth
, sb
, 1);
3087 /* someone might have ended the transaction while we joined */
3088 if (old_trans_id
!= journal
->j_trans_id
) {
3089 retval
= do_journal_end(&myth
, sb
, 1, 0);
3091 retval
= do_journal_end(&myth
, sb
, 1, COMMIT_NOW
);
3097 PROC_INFO_INC(sb
, journal
.journal_relock_wcount
);
3100 /* we are the first writer, set trans_id */
3101 if (journal
->j_trans_start_time
== 0) {
3102 journal
->j_trans_start_time
= get_seconds();
3104 atomic_inc(&(journal
->j_wcount
));
3105 journal
->j_len_alloc
+= nblocks
;
3106 th
->t_blocks_logged
= 0;
3107 th
->t_blocks_allocated
= nblocks
;
3108 th
->t_trans_id
= journal
->j_trans_id
;
3110 INIT_LIST_HEAD(&th
->t_list
);
3115 memset(th
, 0, sizeof(*th
));
3116 /* Re-set th->t_super, so we can properly keep track of how many
3117 * persistent transactions there are. We need to do this so if this
3118 * call is part of a failed restart_transaction, we can free it later */
3123 struct reiserfs_transaction_handle
*reiserfs_persistent_transaction(struct
3129 struct reiserfs_transaction_handle
*th
;
3131 /* if we're nesting into an existing transaction. It will be
3132 ** persistent on its own
3134 if (reiserfs_transaction_running(s
)) {
3135 th
= current
->journal_info
;
3137 BUG_ON(th
->t_refcount
< 2);
3141 th
= kmalloc(sizeof(struct reiserfs_transaction_handle
), GFP_NOFS
);
3144 ret
= journal_begin(th
, s
, nblocks
);
3150 SB_JOURNAL(s
)->j_persistent_trans
++;
3154 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle
*th
)
3156 struct super_block
*s
= th
->t_super
;
3159 ret
= journal_end(th
, th
->t_super
, th
->t_blocks_allocated
);
3162 if (th
->t_refcount
== 0) {
3163 SB_JOURNAL(s
)->j_persistent_trans
--;
3169 static int journal_join(struct reiserfs_transaction_handle
*th
,
3170 struct super_block
*sb
, unsigned long nblocks
)
3172 struct reiserfs_transaction_handle
*cur_th
= current
->journal_info
;
3174 /* this keeps do_journal_end from NULLing out the current->journal_info
3177 th
->t_handle_save
= cur_th
;
3178 BUG_ON(cur_th
&& cur_th
->t_refcount
> 1);
3179 return do_journal_begin_r(th
, sb
, nblocks
, JBEGIN_JOIN
);
3182 int journal_join_abort(struct reiserfs_transaction_handle
*th
,
3183 struct super_block
*sb
, unsigned long nblocks
)
3185 struct reiserfs_transaction_handle
*cur_th
= current
->journal_info
;
3187 /* this keeps do_journal_end from NULLing out the current->journal_info
3190 th
->t_handle_save
= cur_th
;
3191 BUG_ON(cur_th
&& cur_th
->t_refcount
> 1);
3192 return do_journal_begin_r(th
, sb
, nblocks
, JBEGIN_ABORT
);
3195 int journal_begin(struct reiserfs_transaction_handle
*th
,
3196 struct super_block
*sb
, unsigned long nblocks
)
3198 struct reiserfs_transaction_handle
*cur_th
= current
->journal_info
;
3201 th
->t_handle_save
= NULL
;
3203 /* we are nesting into the current transaction */
3204 if (cur_th
->t_super
== sb
) {
3205 BUG_ON(!cur_th
->t_refcount
);
3206 cur_th
->t_refcount
++;
3207 memcpy(th
, cur_th
, sizeof(*th
));
3208 if (th
->t_refcount
<= 1)
3209 reiserfs_warning(sb
, "reiserfs-2005",
3210 "BAD: refcount <= 1, but "
3211 "journal_info != 0");
3214 /* we've ended up with a handle from a different filesystem.
3215 ** save it and restore on journal_end. This should never
3218 reiserfs_warning(sb
, "clm-2100",
3219 "nesting info a different FS");
3220 th
->t_handle_save
= current
->journal_info
;
3221 current
->journal_info
= th
;
3224 current
->journal_info
= th
;
3226 ret
= do_journal_begin_r(th
, sb
, nblocks
, JBEGIN_REG
);
3227 BUG_ON(current
->journal_info
!= th
);
3229 /* I guess this boils down to being the reciprocal of clm-2100 above.
3230 * If do_journal_begin_r fails, we need to put it back, since journal_end
3231 * won't be called to do it. */
3233 current
->journal_info
= th
->t_handle_save
;
3235 BUG_ON(!th
->t_refcount
);
3241 ** puts bh into the current transaction. If it was already there, reorders removes the
3242 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3244 ** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the
3245 ** transaction is committed.
3247 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3249 int journal_mark_dirty(struct reiserfs_transaction_handle
*th
,
3250 struct super_block
*sb
, struct buffer_head
*bh
)
3252 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3253 struct reiserfs_journal_cnode
*cn
= NULL
;
3254 int count_already_incd
= 0;
3256 BUG_ON(!th
->t_trans_id
);
3258 PROC_INFO_INC(sb
, journal
.mark_dirty
);
3259 if (th
->t_trans_id
!= journal
->j_trans_id
) {
3260 reiserfs_panic(th
->t_super
, "journal-1577",
3261 "handle trans id %ld != current trans id %ld",
3262 th
->t_trans_id
, journal
->j_trans_id
);
3267 prepared
= test_clear_buffer_journal_prepared(bh
);
3268 clear_buffer_journal_restore_dirty(bh
);
3269 /* already in this transaction, we are done */
3270 if (buffer_journaled(bh
)) {
3271 PROC_INFO_INC(sb
, journal
.mark_dirty_already
);
3275 /* this must be turned into a panic instead of a warning. We can't allow
3276 ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3277 ** could get to disk too early. NOT GOOD.
3279 if (!prepared
|| buffer_dirty(bh
)) {
3280 reiserfs_warning(sb
, "journal-1777",
3281 "buffer %llu bad state "
3282 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3283 (unsigned long long)bh
->b_blocknr
,
3284 prepared
? ' ' : '!',
3285 buffer_locked(bh
) ? ' ' : '!',
3286 buffer_dirty(bh
) ? ' ' : '!',
3287 buffer_journal_dirty(bh
) ? ' ' : '!');
3290 if (atomic_read(&(journal
->j_wcount
)) <= 0) {
3291 reiserfs_warning(sb
, "journal-1409",
3292 "returning because j_wcount was %d",
3293 atomic_read(&(journal
->j_wcount
)));
3296 /* this error means I've screwed up, and we've overflowed the transaction.
3297 ** Nothing can be done here, except make the FS readonly or panic.
3299 if (journal
->j_len
>= journal
->j_trans_max
) {
3300 reiserfs_panic(th
->t_super
, "journal-1413",
3301 "j_len (%lu) is too big",
3305 if (buffer_journal_dirty(bh
)) {
3306 count_already_incd
= 1;
3307 PROC_INFO_INC(sb
, journal
.mark_dirty_notjournal
);
3308 clear_buffer_journal_dirty(bh
);
3311 if (journal
->j_len
> journal
->j_len_alloc
) {
3312 journal
->j_len_alloc
= journal
->j_len
+ JOURNAL_PER_BALANCE_CNT
;
3315 set_buffer_journaled(bh
);
3317 /* now put this guy on the end */
3321 reiserfs_panic(sb
, "journal-4", "get_cnode failed!");
3324 if (th
->t_blocks_logged
== th
->t_blocks_allocated
) {
3325 th
->t_blocks_allocated
+= JOURNAL_PER_BALANCE_CNT
;
3326 journal
->j_len_alloc
+= JOURNAL_PER_BALANCE_CNT
;
3328 th
->t_blocks_logged
++;
3332 cn
->blocknr
= bh
->b_blocknr
;
3335 insert_journal_hash(journal
->j_hash_table
, cn
);
3336 if (!count_already_incd
) {
3341 cn
->prev
= journal
->j_last
;
3343 if (journal
->j_last
) {
3344 journal
->j_last
->next
= cn
;
3345 journal
->j_last
= cn
;
3347 journal
->j_first
= cn
;
3348 journal
->j_last
= cn
;
3353 int journal_end(struct reiserfs_transaction_handle
*th
,
3354 struct super_block
*sb
, unsigned long nblocks
)
3356 if (!current
->journal_info
&& th
->t_refcount
> 1)
3357 reiserfs_warning(sb
, "REISER-NESTING",
3358 "th NULL, refcount %d", th
->t_refcount
);
3360 if (!th
->t_trans_id
) {
3366 if (th
->t_refcount
> 0) {
3367 struct reiserfs_transaction_handle
*cur_th
=
3368 current
->journal_info
;
3370 /* we aren't allowed to close a nested transaction on a different
3371 ** filesystem from the one in the task struct
3373 BUG_ON(cur_th
->t_super
!= th
->t_super
);
3376 memcpy(current
->journal_info
, th
, sizeof(*th
));
3381 return do_journal_end(th
, sb
, nblocks
, 0);
3385 /* removes from the current transaction, relsing and descrementing any counters.
3386 ** also files the removed buffer directly onto the clean list
3388 ** called by journal_mark_freed when a block has been deleted
3390 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3392 static int remove_from_transaction(struct super_block
*sb
,
3393 b_blocknr_t blocknr
, int already_cleaned
)
3395 struct buffer_head
*bh
;
3396 struct reiserfs_journal_cnode
*cn
;
3397 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3400 cn
= get_journal_hash_dev(sb
, journal
->j_hash_table
, blocknr
);
3401 if (!cn
|| !cn
->bh
) {
3406 cn
->prev
->next
= cn
->next
;
3409 cn
->next
->prev
= cn
->prev
;
3411 if (cn
== journal
->j_first
) {
3412 journal
->j_first
= cn
->next
;
3414 if (cn
== journal
->j_last
) {
3415 journal
->j_last
= cn
->prev
;
3418 remove_journal_hash(sb
, journal
->j_hash_table
, NULL
,
3420 clear_buffer_journaled(bh
); /* don't log this one */
3422 if (!already_cleaned
) {
3423 clear_buffer_journal_dirty(bh
);
3424 clear_buffer_dirty(bh
);
3425 clear_buffer_journal_test(bh
);
3427 if (atomic_read(&(bh
->b_count
)) < 0) {
3428 reiserfs_warning(sb
, "journal-1752",
3434 journal
->j_len_alloc
--;
3440 ** for any cnode in a journal list, it can only be dirtied of all the
3441 ** transactions that include it are committed to disk.
3442 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3443 ** and 0 if you aren't
3445 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3446 ** blocks for a given transaction on disk
3449 static int can_dirty(struct reiserfs_journal_cnode
*cn
)
3451 struct super_block
*sb
= cn
->sb
;
3452 b_blocknr_t blocknr
= cn
->blocknr
;
3453 struct reiserfs_journal_cnode
*cur
= cn
->hprev
;
3456 /* first test hprev. These are all newer than cn, so any node here
3457 ** with the same block number and dev means this node can't be sent
3458 ** to disk right now.
3460 while (cur
&& can_dirty
) {
3461 if (cur
->jlist
&& cur
->bh
&& cur
->blocknr
&& cur
->sb
== sb
&&
3462 cur
->blocknr
== blocknr
) {
3467 /* then test hnext. These are all older than cn. As long as they
3468 ** are committed to the log, it is safe to write cn to disk
3471 while (cur
&& can_dirty
) {
3472 if (cur
->jlist
&& cur
->jlist
->j_len
> 0 &&
3473 atomic_read(&(cur
->jlist
->j_commit_left
)) > 0 && cur
->bh
&&
3474 cur
->blocknr
&& cur
->sb
== sb
&& cur
->blocknr
== blocknr
) {
3482 /* syncs the commit blocks, but does not force the real buffers to disk
3483 ** will wait until the current transaction is done/committed before returning
3485 int journal_end_sync(struct reiserfs_transaction_handle
*th
,
3486 struct super_block
*sb
, unsigned long nblocks
)
3488 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3490 BUG_ON(!th
->t_trans_id
);
3491 /* you can sync while nested, very, very bad */
3492 BUG_ON(th
->t_refcount
> 1);
3493 if (journal
->j_len
== 0) {
3494 reiserfs_prepare_for_journal(sb
, SB_BUFFER_WITH_SB(sb
),
3496 journal_mark_dirty(th
, sb
, SB_BUFFER_WITH_SB(sb
));
3498 return do_journal_end(th
, sb
, nblocks
, COMMIT_NOW
| WAIT
);
3502 ** writeback the pending async commits to disk
3504 static void flush_async_commits(struct work_struct
*work
)
3506 struct reiserfs_journal
*journal
=
3507 container_of(work
, struct reiserfs_journal
, j_work
.work
);
3508 struct super_block
*sb
= journal
->j_work_sb
;
3509 struct reiserfs_journal_list
*jl
;
3510 struct list_head
*entry
;
3512 reiserfs_write_lock(sb
);
3513 if (!list_empty(&journal
->j_journal_list
)) {
3514 /* last entry is the youngest, commit it and you get everything */
3515 entry
= journal
->j_journal_list
.prev
;
3516 jl
= JOURNAL_LIST_ENTRY(entry
);
3517 flush_commit_list(sb
, jl
, 1);
3519 reiserfs_write_unlock(sb
);
3523 ** flushes any old transactions to disk
3524 ** ends the current transaction if it is too old
3526 int reiserfs_flush_old_commits(struct super_block
*sb
)
3529 struct reiserfs_transaction_handle th
;
3530 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3532 now
= get_seconds();
3533 /* safety check so we don't flush while we are replaying the log during
3536 if (list_empty(&journal
->j_journal_list
)) {
3540 /* check the current transaction. If there are no writers, and it is
3541 * too old, finish it, and force the commit blocks to disk
3543 if (atomic_read(&journal
->j_wcount
) <= 0 &&
3544 journal
->j_trans_start_time
> 0 &&
3545 journal
->j_len
> 0 &&
3546 (now
- journal
->j_trans_start_time
) > journal
->j_max_trans_age
) {
3547 if (!journal_join(&th
, sb
, 1)) {
3548 reiserfs_prepare_for_journal(sb
,
3549 SB_BUFFER_WITH_SB(sb
),
3551 journal_mark_dirty(&th
, sb
,
3552 SB_BUFFER_WITH_SB(sb
));
3554 /* we're only being called from kreiserfsd, it makes no sense to do
3555 ** an async commit so that kreiserfsd can do it later
3557 do_journal_end(&th
, sb
, 1, COMMIT_NOW
| WAIT
);
3564 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3566 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3567 ** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just
3568 ** flushes the commit list and returns 0.
3570 ** Won't batch when flush or commit_now is set. Also won't batch when others are waiting on j_join_wait.
3572 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3574 static int check_journal_end(struct reiserfs_transaction_handle
*th
,
3575 struct super_block
*sb
, unsigned long nblocks
,
3580 int flush
= flags
& FLUSH_ALL
;
3581 int commit_now
= flags
& COMMIT_NOW
;
3582 int wait_on_commit
= flags
& WAIT
;
3583 struct reiserfs_journal_list
*jl
;
3584 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3586 BUG_ON(!th
->t_trans_id
);
3588 if (th
->t_trans_id
!= journal
->j_trans_id
) {
3589 reiserfs_panic(th
->t_super
, "journal-1577",
3590 "handle trans id %ld != current trans id %ld",
3591 th
->t_trans_id
, journal
->j_trans_id
);
3594 journal
->j_len_alloc
-= (th
->t_blocks_allocated
- th
->t_blocks_logged
);
3595 if (atomic_read(&(journal
->j_wcount
)) > 0) { /* <= 0 is allowed. unmounting might not call begin */
3596 atomic_dec(&(journal
->j_wcount
));
3599 /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3600 ** will be dealt with by next transaction that actually writes something, but should be taken
3601 ** care of in this trans
3603 BUG_ON(journal
->j_len
== 0);
3605 /* if wcount > 0, and we are called to with flush or commit_now,
3606 ** we wait on j_join_wait. We will wake up when the last writer has
3607 ** finished the transaction, and started it on its way to the disk.
3608 ** Then, we flush the commit or journal list, and just return 0
3609 ** because the rest of journal end was already done for this transaction.
3611 if (atomic_read(&(journal
->j_wcount
)) > 0) {
3612 if (flush
|| commit_now
) {
3615 jl
= journal
->j_current_jl
;
3616 trans_id
= jl
->j_trans_id
;
3618 jl
->j_state
|= LIST_COMMIT_PENDING
;
3619 atomic_set(&(journal
->j_jlock
), 1);
3621 journal
->j_next_full_flush
= 1;
3625 /* sleep while the current transaction is still j_jlocked */
3626 while (journal
->j_trans_id
== trans_id
) {
3627 if (atomic_read(&journal
->j_jlock
)) {
3628 queue_log_writer(sb
);
3631 if (journal
->j_trans_id
== trans_id
) {
3632 atomic_set(&(journal
->j_jlock
),
3638 BUG_ON(journal
->j_trans_id
== trans_id
);
3641 && journal_list_still_alive(sb
, trans_id
)
3642 && wait_on_commit
) {
3643 flush_commit_list(sb
, jl
, 1);
3651 /* deal with old transactions where we are the last writers */
3652 now
= get_seconds();
3653 if ((now
- journal
->j_trans_start_time
) > journal
->j_max_trans_age
) {
3655 journal
->j_next_async_flush
= 1;
3657 /* don't batch when someone is waiting on j_join_wait */
3658 /* don't batch when syncing the commit or flushing the whole trans */
3659 if (!(journal
->j_must_wait
> 0) && !(atomic_read(&(journal
->j_jlock
)))
3660 && !flush
&& !commit_now
&& (journal
->j_len
< journal
->j_max_batch
)
3661 && journal
->j_len_alloc
< journal
->j_max_batch
3662 && journal
->j_cnode_free
> (journal
->j_trans_max
* 3)) {
3663 journal
->j_bcount
++;
3668 if (journal
->j_start
> SB_ONDISK_JOURNAL_SIZE(sb
)) {
3669 reiserfs_panic(sb
, "journal-003",
3670 "j_start (%ld) is too high",
3677 ** Does all the work that makes deleting blocks safe.
3678 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3681 ** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes
3682 ** before this transaction has finished.
3684 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with
3685 ** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash,
3686 ** the block can't be reallocated yet.
3688 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3690 int journal_mark_freed(struct reiserfs_transaction_handle
*th
,
3691 struct super_block
*sb
, b_blocknr_t blocknr
)
3693 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3694 struct reiserfs_journal_cnode
*cn
= NULL
;
3695 struct buffer_head
*bh
= NULL
;
3696 struct reiserfs_list_bitmap
*jb
= NULL
;
3698 BUG_ON(!th
->t_trans_id
);
3700 cn
= get_journal_hash_dev(sb
, journal
->j_hash_table
, blocknr
);
3705 /* if it is journal new, we just remove it from this transaction */
3706 if (bh
&& buffer_journal_new(bh
)) {
3707 clear_buffer_journal_new(bh
);
3708 clear_prepared_bits(bh
);
3709 reiserfs_clean_and_file_buffer(bh
);
3710 cleaned
= remove_from_transaction(sb
, blocknr
, cleaned
);
3712 /* set the bit for this block in the journal bitmap for this transaction */
3713 jb
= journal
->j_current_jl
->j_list_bitmap
;
3715 reiserfs_panic(sb
, "journal-1702",
3716 "journal_list_bitmap is NULL");
3718 set_bit_in_list_bitmap(sb
, blocknr
, jb
);
3720 /* Note, the entire while loop is not allowed to schedule. */
3723 clear_prepared_bits(bh
);
3724 reiserfs_clean_and_file_buffer(bh
);
3726 cleaned
= remove_from_transaction(sb
, blocknr
, cleaned
);
3728 /* find all older transactions with this block, make sure they don't try to write it out */
3729 cn
= get_journal_hash_dev(sb
, journal
->j_list_hash_table
,
3732 if (sb
== cn
->sb
&& blocknr
== cn
->blocknr
) {
3733 set_bit(BLOCK_FREED
, &cn
->state
);
3736 /* remove_from_transaction will brelse the buffer if it was
3737 ** in the current trans
3739 clear_buffer_journal_dirty(cn
->
3741 clear_buffer_dirty(cn
->bh
);
3742 clear_buffer_journal_test(cn
->
3747 (&(cn
->bh
->b_count
)) < 0) {
3748 reiserfs_warning(sb
,
3750 "cn->bh->b_count < 0");
3753 if (cn
->jlist
) { /* since we are clearing the bh, we MUST dec nonzerolen */
3766 release_buffer_page(bh
); /* get_hash grabs the buffer */
3770 void reiserfs_update_inode_transaction(struct inode
*inode
)
3772 struct reiserfs_journal
*journal
= SB_JOURNAL(inode
->i_sb
);
3773 REISERFS_I(inode
)->i_jl
= journal
->j_current_jl
;
3774 REISERFS_I(inode
)->i_trans_id
= journal
->j_trans_id
;
3778 * returns -1 on error, 0 if no commits/barriers were done and 1
3779 * if a transaction was actually committed and the barrier was done
3781 static int __commit_trans_jl(struct inode
*inode
, unsigned long id
,
3782 struct reiserfs_journal_list
*jl
)
3784 struct reiserfs_transaction_handle th
;
3785 struct super_block
*sb
= inode
->i_sb
;
3786 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3789 /* is it from the current transaction, or from an unknown transaction? */
3790 if (id
== journal
->j_trans_id
) {
3791 jl
= journal
->j_current_jl
;
3792 /* try to let other writers come in and grow this transaction */
3793 let_transaction_grow(sb
, id
);
3794 if (journal
->j_trans_id
!= id
) {
3795 goto flush_commit_only
;
3798 ret
= journal_begin(&th
, sb
, 1);
3802 /* someone might have ended this transaction while we joined */
3803 if (journal
->j_trans_id
!= id
) {
3804 reiserfs_prepare_for_journal(sb
, SB_BUFFER_WITH_SB(sb
),
3806 journal_mark_dirty(&th
, sb
, SB_BUFFER_WITH_SB(sb
));
3807 ret
= journal_end(&th
, sb
, 1);
3808 goto flush_commit_only
;
3811 ret
= journal_end_sync(&th
, sb
, 1);
3816 /* this gets tricky, we have to make sure the journal list in
3817 * the inode still exists. We know the list is still around
3818 * if we've got a larger transaction id than the oldest list
3821 if (journal_list_still_alive(inode
->i_sb
, id
)) {
3823 * we only set ret to 1 when we know for sure
3824 * the barrier hasn't been started yet on the commit
3827 if (atomic_read(&jl
->j_commit_left
) > 1)
3829 flush_commit_list(sb
, jl
, 1);
3830 if (journal
->j_errno
)
3831 ret
= journal
->j_errno
;
3834 /* otherwise the list is gone, and long since committed */
3838 int reiserfs_commit_for_inode(struct inode
*inode
)
3840 unsigned int id
= REISERFS_I(inode
)->i_trans_id
;
3841 struct reiserfs_journal_list
*jl
= REISERFS_I(inode
)->i_jl
;
3843 /* for the whole inode, assume unset id means it was
3844 * changed in the current transaction. More conservative
3847 reiserfs_update_inode_transaction(inode
);
3848 id
= REISERFS_I(inode
)->i_trans_id
;
3849 /* jl will be updated in __commit_trans_jl */
3852 return __commit_trans_jl(inode
, id
, jl
);
3855 void reiserfs_restore_prepared_buffer(struct super_block
*sb
,
3856 struct buffer_head
*bh
)
3858 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3859 PROC_INFO_INC(sb
, journal
.restore_prepared
);
3863 if (test_clear_buffer_journal_restore_dirty(bh
) &&
3864 buffer_journal_dirty(bh
)) {
3865 struct reiserfs_journal_cnode
*cn
;
3866 cn
= get_journal_hash_dev(sb
,
3867 journal
->j_list_hash_table
,
3869 if (cn
&& can_dirty(cn
)) {
3870 set_buffer_journal_test(bh
);
3871 mark_buffer_dirty(bh
);
3874 clear_buffer_journal_prepared(bh
);
3877 extern struct tree_balance
*cur_tb
;
3879 ** before we can change a metadata block, we have to make sure it won't
3880 ** be written to disk while we are altering it. So, we must:
3885 int reiserfs_prepare_for_journal(struct super_block
*sb
,
3886 struct buffer_head
*bh
, int wait
)
3888 PROC_INFO_INC(sb
, journal
.prepare
);
3890 if (!trylock_buffer(bh
)) {
3895 set_buffer_journal_prepared(bh
);
3896 if (test_clear_buffer_dirty(bh
) && buffer_journal_dirty(bh
)) {
3897 clear_buffer_journal_test(bh
);
3898 set_buffer_journal_restore_dirty(bh
);
3904 static void flush_old_journal_lists(struct super_block
*s
)
3906 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
3907 struct reiserfs_journal_list
*jl
;
3908 struct list_head
*entry
;
3909 time_t now
= get_seconds();
3911 while (!list_empty(&journal
->j_journal_list
)) {
3912 entry
= journal
->j_journal_list
.next
;
3913 jl
= JOURNAL_LIST_ENTRY(entry
);
3914 /* this check should always be run, to send old lists to disk */
3915 if (jl
->j_timestamp
< (now
- (JOURNAL_MAX_TRANS_AGE
* 4)) &&
3916 atomic_read(&jl
->j_commit_left
) == 0 &&
3917 test_transaction(s
, jl
)) {
3918 flush_used_journal_lists(s
, jl
);
3926 ** long and ugly. If flush, will not return until all commit
3927 ** blocks and all real buffers in the trans are on disk.
3928 ** If no_async, won't return until all commit blocks are on disk.
3930 ** keep reading, there are comments as you go along
3932 ** If the journal is aborted, we just clean up. Things like flushing
3933 ** journal lists, etc just won't happen.
3935 static int do_journal_end(struct reiserfs_transaction_handle
*th
,
3936 struct super_block
*sb
, unsigned long nblocks
,
3939 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3940 struct reiserfs_journal_cnode
*cn
, *next
, *jl_cn
;
3941 struct reiserfs_journal_cnode
*last_cn
= NULL
;
3942 struct reiserfs_journal_desc
*desc
;
3943 struct reiserfs_journal_commit
*commit
;
3944 struct buffer_head
*c_bh
; /* commit bh */
3945 struct buffer_head
*d_bh
; /* desc bh */
3946 int cur_write_start
= 0; /* start index of current log write */
3951 struct reiserfs_journal_list
*jl
, *temp_jl
;
3952 struct list_head
*entry
, *safe
;
3953 unsigned long jindex
;
3954 unsigned int commit_trans_id
;
3957 BUG_ON(th
->t_refcount
> 1);
3958 BUG_ON(!th
->t_trans_id
);
3960 /* protect flush_older_commits from doing mistakes if the
3961 transaction ID counter gets overflowed. */
3962 if (th
->t_trans_id
== ~0U)
3963 flags
|= FLUSH_ALL
| COMMIT_NOW
| WAIT
;
3964 flush
= flags
& FLUSH_ALL
;
3965 wait_on_commit
= flags
& WAIT
;
3968 current
->journal_info
= th
->t_handle_save
;
3969 reiserfs_check_lock_depth(sb
, "journal end");
3970 if (journal
->j_len
== 0) {
3971 reiserfs_prepare_for_journal(sb
, SB_BUFFER_WITH_SB(sb
),
3973 journal_mark_dirty(th
, sb
, SB_BUFFER_WITH_SB(sb
));
3977 if (journal
->j_next_full_flush
) {
3981 if (journal
->j_next_async_flush
) {
3982 flags
|= COMMIT_NOW
| WAIT
;
3986 /* check_journal_end locks the journal, and unlocks if it does not return 1
3987 ** it tells us if we should continue with the journal_end, or just return
3989 if (!check_journal_end(th
, sb
, nblocks
, flags
)) {
3991 wake_queued_writers(sb
);
3992 reiserfs_async_progress_wait(sb
);
3996 /* check_journal_end might set these, check again */
3997 if (journal
->j_next_full_flush
) {
4002 ** j must wait means we have to flush the log blocks, and the real blocks for
4005 if (journal
->j_must_wait
> 0) {
4008 #ifdef REISERFS_PREALLOCATE
4009 /* quota ops might need to nest, setup the journal_info pointer for them
4010 * and raise the refcount so that it is > 0. */
4011 current
->journal_info
= th
;
4013 reiserfs_discard_all_prealloc(th
); /* it should not involve new blocks into
4014 * the transaction */
4016 current
->journal_info
= th
->t_handle_save
;
4019 /* setup description block */
4022 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
4024 set_buffer_uptodate(d_bh
);
4025 desc
= (struct reiserfs_journal_desc
*)(d_bh
)->b_data
;
4026 memset(d_bh
->b_data
, 0, d_bh
->b_size
);
4027 memcpy(get_journal_desc_magic(d_bh
), JOURNAL_DESC_MAGIC
, 8);
4028 set_desc_trans_id(desc
, journal
->j_trans_id
);
4030 /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */
4031 c_bh
= journal_getblk(sb
, SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
4032 ((journal
->j_start
+ journal
->j_len
+
4033 1) % SB_ONDISK_JOURNAL_SIZE(sb
)));
4034 commit
= (struct reiserfs_journal_commit
*)c_bh
->b_data
;
4035 memset(c_bh
->b_data
, 0, c_bh
->b_size
);
4036 set_commit_trans_id(commit
, journal
->j_trans_id
);
4037 set_buffer_uptodate(c_bh
);
4039 /* init this journal list */
4040 jl
= journal
->j_current_jl
;
4042 /* we lock the commit before doing anything because
4043 * we want to make sure nobody tries to run flush_commit_list until
4044 * the new transaction is fully setup, and we've already flushed the
4047 reiserfs_mutex_lock_safe(&jl
->j_commit_mutex
, sb
);
4049 /* save the transaction id in case we need to commit it later */
4050 commit_trans_id
= jl
->j_trans_id
;
4052 atomic_set(&jl
->j_older_commits_done
, 0);
4053 jl
->j_trans_id
= journal
->j_trans_id
;
4054 jl
->j_timestamp
= journal
->j_trans_start_time
;
4055 jl
->j_commit_bh
= c_bh
;
4056 jl
->j_start
= journal
->j_start
;
4057 jl
->j_len
= journal
->j_len
;
4058 atomic_set(&jl
->j_nonzerolen
, journal
->j_len
);
4059 atomic_set(&jl
->j_commit_left
, journal
->j_len
+ 2);
4060 jl
->j_realblock
= NULL
;
4062 /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4063 ** for each real block, add it to the journal list hash,
4064 ** copy into real block index array in the commit or desc block
4066 trans_half
= journal_trans_half(sb
->s_blocksize
);
4067 for (i
= 0, cn
= journal
->j_first
; cn
; cn
= cn
->next
, i
++) {
4068 if (buffer_journaled(cn
->bh
)) {
4069 jl_cn
= get_cnode(sb
);
4071 reiserfs_panic(sb
, "journal-1676",
4072 "get_cnode returned NULL");
4075 jl
->j_realblock
= jl_cn
;
4077 jl_cn
->prev
= last_cn
;
4080 last_cn
->next
= jl_cn
;
4083 /* make sure the block we are trying to log is not a block
4084 of journal or reserved area */
4086 if (is_block_in_log_or_reserved_area
4087 (sb
, cn
->bh
->b_blocknr
)) {
4088 reiserfs_panic(sb
, "journal-2332",
4089 "Trying to log block %lu, "
4090 "which is a log block",
4093 jl_cn
->blocknr
= cn
->bh
->b_blocknr
;
4098 insert_journal_hash(journal
->j_list_hash_table
, jl_cn
);
4099 if (i
< trans_half
) {
4100 desc
->j_realblock
[i
] =
4101 cpu_to_le32(cn
->bh
->b_blocknr
);
4103 commit
->j_realblock
[i
- trans_half
] =
4104 cpu_to_le32(cn
->bh
->b_blocknr
);
4110 set_desc_trans_len(desc
, journal
->j_len
);
4111 set_desc_mount_id(desc
, journal
->j_mount_id
);
4112 set_desc_trans_id(desc
, journal
->j_trans_id
);
4113 set_commit_trans_len(commit
, journal
->j_len
);
4115 /* special check in case all buffers in the journal were marked for not logging */
4116 BUG_ON(journal
->j_len
== 0);
4118 /* we're about to dirty all the log blocks, mark the description block
4119 * dirty now too. Don't mark the commit block dirty until all the
4120 * others are on disk
4122 mark_buffer_dirty(d_bh
);
4124 /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4125 cur_write_start
= journal
->j_start
;
4126 cn
= journal
->j_first
;
4127 jindex
= 1; /* start at one so we don't get the desc again */
4129 clear_buffer_journal_new(cn
->bh
);
4130 /* copy all the real blocks into log area. dirty log blocks */
4131 if (buffer_journaled(cn
->bh
)) {
4132 struct buffer_head
*tmp_bh
;
4137 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
4140 SB_ONDISK_JOURNAL_SIZE(sb
)));
4141 set_buffer_uptodate(tmp_bh
);
4142 page
= cn
->bh
->b_page
;
4144 memcpy(tmp_bh
->b_data
,
4145 addr
+ offset_in_page(cn
->bh
->b_data
),
4148 mark_buffer_dirty(tmp_bh
);
4150 set_buffer_journal_dirty(cn
->bh
);
4151 clear_buffer_journaled(cn
->bh
);
4153 /* JDirty cleared sometime during transaction. don't log this one */
4154 reiserfs_warning(sb
, "journal-2048",
4155 "BAD, buffer in journal hash, "
4162 reiserfs_write_unlock(sb
);
4164 reiserfs_write_lock(sb
);
4167 /* we are done with both the c_bh and d_bh, but
4168 ** c_bh must be written after all other commit blocks,
4169 ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4172 journal
->j_current_jl
= alloc_journal_list(sb
);
4174 /* now it is safe to insert this transaction on the main list */
4175 list_add_tail(&jl
->j_list
, &journal
->j_journal_list
);
4176 list_add_tail(&jl
->j_working_list
, &journal
->j_working_list
);
4177 journal
->j_num_work_lists
++;
4179 /* reset journal values for the next transaction */
4180 old_start
= journal
->j_start
;
4182 (journal
->j_start
+ journal
->j_len
+
4183 2) % SB_ONDISK_JOURNAL_SIZE(sb
);
4184 atomic_set(&(journal
->j_wcount
), 0);
4185 journal
->j_bcount
= 0;
4186 journal
->j_last
= NULL
;
4187 journal
->j_first
= NULL
;
4189 journal
->j_trans_start_time
= 0;
4190 /* check for trans_id overflow */
4191 if (++journal
->j_trans_id
== 0)
4192 journal
->j_trans_id
= 10;
4193 journal
->j_current_jl
->j_trans_id
= journal
->j_trans_id
;
4194 journal
->j_must_wait
= 0;
4195 journal
->j_len_alloc
= 0;
4196 journal
->j_next_full_flush
= 0;
4197 journal
->j_next_async_flush
= 0;
4198 init_journal_hash(sb
);
4200 // make sure reiserfs_add_jh sees the new current_jl before we
4201 // write out the tails
4204 /* tail conversion targets have to hit the disk before we end the
4205 * transaction. Otherwise a later transaction might repack the tail
4206 * before this transaction commits, leaving the data block unflushed and
4207 * clean, if we crash before the later transaction commits, the data block
4210 if (!list_empty(&jl
->j_tail_bh_list
)) {
4211 reiserfs_write_unlock(sb
);
4212 write_ordered_buffers(&journal
->j_dirty_buffers_lock
,
4213 journal
, jl
, &jl
->j_tail_bh_list
);
4214 reiserfs_write_lock(sb
);
4216 BUG_ON(!list_empty(&jl
->j_tail_bh_list
));
4217 mutex_unlock(&jl
->j_commit_mutex
);
4219 /* honor the flush wishes from the caller, simple commits can
4220 ** be done outside the journal lock, they are done below
4222 ** if we don't flush the commit list right now, we put it into
4223 ** the work queue so the people waiting on the async progress work
4224 ** queue don't wait for this proc to flush journal lists and such.
4227 flush_commit_list(sb
, jl
, 1);
4228 flush_journal_list(sb
, jl
, 1);
4229 } else if (!(jl
->j_state
& LIST_COMMIT_PENDING
))
4230 queue_delayed_work(commit_wq
, &journal
->j_work
, HZ
/ 10);
4232 /* if the next transaction has any chance of wrapping, flush
4233 ** transactions that might get overwritten. If any journal lists are very
4234 ** old flush them as well.
4237 list_for_each_safe(entry
, safe
, &journal
->j_journal_list
) {
4238 temp_jl
= JOURNAL_LIST_ENTRY(entry
);
4239 if (journal
->j_start
<= temp_jl
->j_start
) {
4240 if ((journal
->j_start
+ journal
->j_trans_max
+ 1) >=
4242 flush_used_journal_lists(sb
, temp_jl
);
4244 } else if ((journal
->j_start
+
4245 journal
->j_trans_max
+ 1) <
4246 SB_ONDISK_JOURNAL_SIZE(sb
)) {
4247 /* if we don't cross into the next transaction and we don't
4248 * wrap, there is no way we can overlap any later transactions
4253 } else if ((journal
->j_start
+
4254 journal
->j_trans_max
+ 1) >
4255 SB_ONDISK_JOURNAL_SIZE(sb
)) {
4256 if (((journal
->j_start
+ journal
->j_trans_max
+ 1) %
4257 SB_ONDISK_JOURNAL_SIZE(sb
)) >=
4259 flush_used_journal_lists(sb
, temp_jl
);
4262 /* we don't overlap anything from out start to the end of the
4263 * log, and our wrapped portion doesn't overlap anything at
4264 * the start of the log. We can break
4270 flush_old_journal_lists(sb
);
4272 journal
->j_current_jl
->j_list_bitmap
=
4273 get_list_bitmap(sb
, journal
->j_current_jl
);
4275 if (!(journal
->j_current_jl
->j_list_bitmap
)) {
4276 reiserfs_panic(sb
, "journal-1996",
4277 "could not get a list bitmap");
4280 atomic_set(&(journal
->j_jlock
), 0);
4282 /* wake up any body waiting to join. */
4283 clear_bit(J_WRITERS_QUEUED
, &journal
->j_state
);
4284 wake_up(&(journal
->j_join_wait
));
4286 if (!flush
&& wait_on_commit
&&
4287 journal_list_still_alive(sb
, commit_trans_id
)) {
4288 flush_commit_list(sb
, jl
, 1);
4291 reiserfs_check_lock_depth(sb
, "journal end2");
4293 memset(th
, 0, sizeof(*th
));
4294 /* Re-set th->t_super, so we can properly keep track of how many
4295 * persistent transactions there are. We need to do this so if this
4296 * call is part of a failed restart_transaction, we can free it later */
4299 return journal
->j_errno
;
4302 /* Send the file system read only and refuse new transactions */
4303 void reiserfs_abort_journal(struct super_block
*sb
, int errno
)
4305 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
4306 if (test_bit(J_ABORTED
, &journal
->j_state
))
4309 if (!journal
->j_errno
)
4310 journal
->j_errno
= errno
;
4312 sb
->s_flags
|= MS_RDONLY
;
4313 set_bit(J_ABORTED
, &journal
->j_state
);
4315 #ifdef CONFIG_REISERFS_CHECK