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/config.h>
38 #include <asm/uaccess.h>
39 #include <asm/system.h>
41 #include <linux/time.h>
42 #include <asm/semaphore.h>
44 #include <linux/vmalloc.h>
45 #include <linux/reiserfs_fs.h>
47 #include <linux/kernel.h>
48 #include <linux/errno.h>
49 #include <linux/fcntl.h>
50 #include <linux/stat.h>
51 #include <linux/string.h>
52 #include <linux/smp_lock.h>
53 #include <linux/buffer_head.h>
54 #include <linux/workqueue.h>
55 #include <linux/writeback.h>
56 #include <linux/blkdev.h>
58 /* gets a struct reiserfs_journal_list * from a list head */
59 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
61 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
64 /* the number of mounted filesystems. This is used to decide when to
65 ** start and kill the commit workqueue
67 static int reiserfs_mounted_fs_count
;
69 static struct workqueue_struct
*commit_wq
;
71 #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
73 #define BUFNR 64 /*read ahead */
75 /* cnode stat bits. Move these into reiserfs_fs.h */
77 #define BLOCK_FREED 2 /* this block was freed, and can't be written. */
78 #define BLOCK_FREED_HOLDER 3 /* this block was freed during this transaction, and can't be written */
80 #define BLOCK_NEEDS_FLUSH 4 /* used in flush_journal_list */
81 #define BLOCK_DIRTIED 5
83 /* journal list state bits */
84 #define LIST_TOUCHED 1
86 #define LIST_COMMIT_PENDING 4 /* someone will commit this list */
88 /* flags for do_journal_end */
89 #define FLUSH_ALL 1 /* flush commit and real blocks */
90 #define COMMIT_NOW 2 /* end and commit this transaction */
91 #define WAIT 4 /* wait for the log blocks to hit the disk */
93 static int do_journal_end(struct reiserfs_transaction_handle
*,
94 struct super_block
*, unsigned long nblocks
,
96 static int flush_journal_list(struct super_block
*s
,
97 struct reiserfs_journal_list
*jl
, int flushall
);
98 static int flush_commit_list(struct super_block
*s
,
99 struct reiserfs_journal_list
*jl
, int flushall
);
100 static int can_dirty(struct reiserfs_journal_cnode
*cn
);
101 static int journal_join(struct reiserfs_transaction_handle
*th
,
102 struct super_block
*p_s_sb
, unsigned long nblocks
);
103 static int release_journal_dev(struct super_block
*super
,
104 struct reiserfs_journal
*journal
);
105 static int dirty_one_transaction(struct super_block
*s
,
106 struct reiserfs_journal_list
*jl
);
107 static void flush_async_commits(void *p
);
108 static void queue_log_writer(struct super_block
*s
);
110 /* values for join in do_journal_begin_r */
112 JBEGIN_REG
= 0, /* regular journal begin */
113 JBEGIN_JOIN
= 1, /* join the running transaction if at all possible */
114 JBEGIN_ABORT
= 2, /* called from cleanup code, ignores aborted flag */
117 static int do_journal_begin_r(struct reiserfs_transaction_handle
*th
,
118 struct super_block
*p_s_sb
,
119 unsigned long nblocks
, int join
);
121 static void init_journal_hash(struct super_block
*p_s_sb
)
123 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
124 memset(journal
->j_hash_table
, 0,
125 JOURNAL_HASH_SIZE
* sizeof(struct reiserfs_journal_cnode
*));
129 ** clears BH_Dirty and sticks the buffer on the clean list. Called because I can't allow refile_buffer to
130 ** make schedule happen after I've freed a block. Look at remove_from_transaction and journal_mark_freed for
133 static int reiserfs_clean_and_file_buffer(struct buffer_head
*bh
)
136 clear_buffer_dirty(bh
);
137 clear_buffer_journal_test(bh
);
142 static void disable_barrier(struct super_block
*s
)
144 REISERFS_SB(s
)->s_mount_opt
&= ~(1 << REISERFS_BARRIER_FLUSH
);
145 printk("reiserfs: disabling flush barriers on %s\n",
146 reiserfs_bdevname(s
));
149 static struct reiserfs_bitmap_node
*allocate_bitmap_node(struct super_block
152 struct reiserfs_bitmap_node
*bn
;
155 bn
= reiserfs_kmalloc(sizeof(struct reiserfs_bitmap_node
), GFP_NOFS
,
160 bn
->data
= reiserfs_kmalloc(p_s_sb
->s_blocksize
, GFP_NOFS
, p_s_sb
);
162 reiserfs_kfree(bn
, sizeof(struct reiserfs_bitmap_node
), p_s_sb
);
166 memset(bn
->data
, 0, p_s_sb
->s_blocksize
);
167 INIT_LIST_HEAD(&bn
->list
);
171 static struct reiserfs_bitmap_node
*get_bitmap_node(struct super_block
*p_s_sb
)
173 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
174 struct reiserfs_bitmap_node
*bn
= NULL
;
175 struct list_head
*entry
= journal
->j_bitmap_nodes
.next
;
177 journal
->j_used_bitmap_nodes
++;
180 if (entry
!= &journal
->j_bitmap_nodes
) {
181 bn
= list_entry(entry
, struct reiserfs_bitmap_node
, list
);
183 memset(bn
->data
, 0, p_s_sb
->s_blocksize
);
184 journal
->j_free_bitmap_nodes
--;
187 bn
= allocate_bitmap_node(p_s_sb
);
194 static inline void free_bitmap_node(struct super_block
*p_s_sb
,
195 struct reiserfs_bitmap_node
*bn
)
197 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
198 journal
->j_used_bitmap_nodes
--;
199 if (journal
->j_free_bitmap_nodes
> REISERFS_MAX_BITMAP_NODES
) {
200 reiserfs_kfree(bn
->data
, p_s_sb
->s_blocksize
, p_s_sb
);
201 reiserfs_kfree(bn
, sizeof(struct reiserfs_bitmap_node
), p_s_sb
);
203 list_add(&bn
->list
, &journal
->j_bitmap_nodes
);
204 journal
->j_free_bitmap_nodes
++;
208 static void allocate_bitmap_nodes(struct super_block
*p_s_sb
)
211 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
212 struct reiserfs_bitmap_node
*bn
= NULL
;
213 for (i
= 0; i
< REISERFS_MIN_BITMAP_NODES
; i
++) {
214 bn
= allocate_bitmap_node(p_s_sb
);
216 list_add(&bn
->list
, &journal
->j_bitmap_nodes
);
217 journal
->j_free_bitmap_nodes
++;
219 break; // this is ok, we'll try again when more are needed
224 static int set_bit_in_list_bitmap(struct super_block
*p_s_sb
, int block
,
225 struct reiserfs_list_bitmap
*jb
)
227 int bmap_nr
= block
/ (p_s_sb
->s_blocksize
<< 3);
228 int bit_nr
= block
% (p_s_sb
->s_blocksize
<< 3);
230 if (!jb
->bitmaps
[bmap_nr
]) {
231 jb
->bitmaps
[bmap_nr
] = get_bitmap_node(p_s_sb
);
233 set_bit(bit_nr
, (unsigned long *)jb
->bitmaps
[bmap_nr
]->data
);
237 static void cleanup_bitmap_list(struct super_block
*p_s_sb
,
238 struct reiserfs_list_bitmap
*jb
)
241 if (jb
->bitmaps
== NULL
)
244 for (i
= 0; i
< SB_BMAP_NR(p_s_sb
); i
++) {
245 if (jb
->bitmaps
[i
]) {
246 free_bitmap_node(p_s_sb
, jb
->bitmaps
[i
]);
247 jb
->bitmaps
[i
] = NULL
;
253 ** only call this on FS unmount.
255 static int free_list_bitmaps(struct super_block
*p_s_sb
,
256 struct reiserfs_list_bitmap
*jb_array
)
259 struct reiserfs_list_bitmap
*jb
;
260 for (i
= 0; i
< JOURNAL_NUM_BITMAPS
; i
++) {
262 jb
->journal_list
= NULL
;
263 cleanup_bitmap_list(p_s_sb
, jb
);
270 static int free_bitmap_nodes(struct super_block
*p_s_sb
)
272 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
273 struct list_head
*next
= journal
->j_bitmap_nodes
.next
;
274 struct reiserfs_bitmap_node
*bn
;
276 while (next
!= &journal
->j_bitmap_nodes
) {
277 bn
= list_entry(next
, struct reiserfs_bitmap_node
, list
);
279 reiserfs_kfree(bn
->data
, p_s_sb
->s_blocksize
, p_s_sb
);
280 reiserfs_kfree(bn
, sizeof(struct reiserfs_bitmap_node
), p_s_sb
);
281 next
= journal
->j_bitmap_nodes
.next
;
282 journal
->j_free_bitmap_nodes
--;
289 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
290 ** jb_array is the array to be filled in.
292 int reiserfs_allocate_list_bitmaps(struct super_block
*p_s_sb
,
293 struct reiserfs_list_bitmap
*jb_array
,
298 struct reiserfs_list_bitmap
*jb
;
299 int mem
= bmap_nr
* sizeof(struct reiserfs_bitmap_node
*);
301 for (i
= 0; i
< JOURNAL_NUM_BITMAPS
; i
++) {
303 jb
->journal_list
= NULL
;
304 jb
->bitmaps
= vmalloc(mem
);
306 reiserfs_warning(p_s_sb
,
307 "clm-2000, unable to allocate bitmaps for journal lists");
311 memset(jb
->bitmaps
, 0, mem
);
314 free_list_bitmaps(p_s_sb
, jb_array
);
321 ** find an available list bitmap. If you can't find one, flush a commit list
324 static struct reiserfs_list_bitmap
*get_list_bitmap(struct super_block
*p_s_sb
,
325 struct reiserfs_journal_list
329 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
330 struct reiserfs_list_bitmap
*jb
= NULL
;
332 for (j
= 0; j
< (JOURNAL_NUM_BITMAPS
* 3); j
++) {
333 i
= journal
->j_list_bitmap_index
;
334 journal
->j_list_bitmap_index
= (i
+ 1) % JOURNAL_NUM_BITMAPS
;
335 jb
= journal
->j_list_bitmap
+ i
;
336 if (journal
->j_list_bitmap
[i
].journal_list
) {
337 flush_commit_list(p_s_sb
,
338 journal
->j_list_bitmap
[i
].
340 if (!journal
->j_list_bitmap
[i
].journal_list
) {
347 if (jb
->journal_list
) { /* double check to make sure if flushed correctly */
350 jb
->journal_list
= jl
;
355 ** allocates a new chunk of X nodes, and links them all together as a list.
356 ** Uses the cnode->next and cnode->prev pointers
357 ** returns NULL on failure
359 static struct reiserfs_journal_cnode
*allocate_cnodes(int num_cnodes
)
361 struct reiserfs_journal_cnode
*head
;
363 if (num_cnodes
<= 0) {
366 head
= vmalloc(num_cnodes
* sizeof(struct reiserfs_journal_cnode
));
370 memset(head
, 0, num_cnodes
* sizeof(struct reiserfs_journal_cnode
));
372 head
[0].next
= head
+ 1;
373 for (i
= 1; i
< num_cnodes
; i
++) {
374 head
[i
].prev
= head
+ (i
- 1);
375 head
[i
].next
= head
+ (i
+ 1); /* if last one, overwrite it after the if */
377 head
[num_cnodes
- 1].next
= NULL
;
382 ** pulls a cnode off the free list, or returns NULL on failure
384 static struct reiserfs_journal_cnode
*get_cnode(struct super_block
*p_s_sb
)
386 struct reiserfs_journal_cnode
*cn
;
387 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
389 reiserfs_check_lock_depth(p_s_sb
, "get_cnode");
391 if (journal
->j_cnode_free
<= 0) {
394 journal
->j_cnode_used
++;
395 journal
->j_cnode_free
--;
396 cn
= journal
->j_cnode_free_list
;
401 cn
->next
->prev
= NULL
;
403 journal
->j_cnode_free_list
= cn
->next
;
404 memset(cn
, 0, sizeof(struct reiserfs_journal_cnode
));
409 ** returns a cnode to the free list
411 static void free_cnode(struct super_block
*p_s_sb
,
412 struct reiserfs_journal_cnode
*cn
)
414 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
416 reiserfs_check_lock_depth(p_s_sb
, "free_cnode");
418 journal
->j_cnode_used
--;
419 journal
->j_cnode_free
++;
420 /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
421 cn
->next
= journal
->j_cnode_free_list
;
422 if (journal
->j_cnode_free_list
) {
423 journal
->j_cnode_free_list
->prev
= cn
;
425 cn
->prev
= NULL
; /* not needed with the memset, but I might kill the memset, and forget to do this */
426 journal
->j_cnode_free_list
= cn
;
429 static void clear_prepared_bits(struct buffer_head
*bh
)
431 clear_buffer_journal_prepared(bh
);
432 clear_buffer_journal_restore_dirty(bh
);
435 /* utility function to force a BUG if it is called without the big
436 ** kernel lock held. caller is the string printed just before calling BUG()
438 void reiserfs_check_lock_depth(struct super_block
*sb
, char *caller
)
441 if (current
->lock_depth
< 0) {
442 reiserfs_panic(sb
, "%s called without kernel lock held",
450 /* return a cnode with same dev, block number and size in table, or null if not found */
451 static inline struct reiserfs_journal_cnode
*get_journal_hash_dev(struct
455 reiserfs_journal_cnode
459 struct reiserfs_journal_cnode
*cn
;
460 cn
= journal_hash(table
, sb
, bl
);
462 if (cn
->blocknr
== bl
&& cn
->sb
== sb
)
466 return (struct reiserfs_journal_cnode
*)0;
470 ** this actually means 'can this block be reallocated yet?'. If you set search_all, a block can only be allocated
471 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
472 ** being overwritten by a replay after crashing.
474 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction. Since deleting
475 ** a block removes it from the current transaction, this case should never happen. If you don't set search_all, make
476 ** sure you never write the block without logging it.
478 ** next_zero_bit is a suggestion about the next block to try for find_forward.
479 ** when bl is rejected because it is set in a journal list bitmap, we search
480 ** for the next zero bit in the bitmap that rejected bl. Then, we return that
481 ** through next_zero_bit for find_forward to try.
483 ** Just because we return something in next_zero_bit does not mean we won't
484 ** reject it on the next call to reiserfs_in_journal
487 int reiserfs_in_journal(struct super_block
*p_s_sb
,
488 int bmap_nr
, int bit_nr
, int search_all
,
489 b_blocknr_t
* next_zero_bit
)
491 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
492 struct reiserfs_journal_cnode
*cn
;
493 struct reiserfs_list_bitmap
*jb
;
497 *next_zero_bit
= 0; /* always start this at zero. */
499 PROC_INFO_INC(p_s_sb
, journal
.in_journal
);
500 /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
501 ** if we crash before the transaction that freed it commits, this transaction won't
502 ** have committed either, and the block will never be written
505 for (i
= 0; i
< JOURNAL_NUM_BITMAPS
; i
++) {
506 PROC_INFO_INC(p_s_sb
, journal
.in_journal_bitmap
);
507 jb
= journal
->j_list_bitmap
+ i
;
508 if (jb
->journal_list
&& jb
->bitmaps
[bmap_nr
] &&
510 (unsigned long *)jb
->bitmaps
[bmap_nr
]->
513 find_next_zero_bit((unsigned long *)
514 (jb
->bitmaps
[bmap_nr
]->
516 p_s_sb
->s_blocksize
<< 3,
523 bl
= bmap_nr
* (p_s_sb
->s_blocksize
<< 3) + bit_nr
;
524 /* is it in any old transactions? */
527 get_journal_hash_dev(p_s_sb
, journal
->j_list_hash_table
, bl
))) {
531 /* is it in the current transaction. This should never happen */
532 if ((cn
= get_journal_hash_dev(p_s_sb
, journal
->j_hash_table
, bl
))) {
537 PROC_INFO_INC(p_s_sb
, journal
.in_journal_reusable
);
542 /* insert cn into table
544 static inline void insert_journal_hash(struct reiserfs_journal_cnode
**table
,
545 struct reiserfs_journal_cnode
*cn
)
547 struct reiserfs_journal_cnode
*cn_orig
;
549 cn_orig
= journal_hash(table
, cn
->sb
, cn
->blocknr
);
555 journal_hash(table
, cn
->sb
, cn
->blocknr
) = cn
;
558 /* lock the current transaction */
559 static inline void lock_journal(struct super_block
*p_s_sb
)
561 PROC_INFO_INC(p_s_sb
, journal
.lock_journal
);
562 down(&SB_JOURNAL(p_s_sb
)->j_lock
);
565 /* unlock the current transaction */
566 static inline void unlock_journal(struct super_block
*p_s_sb
)
568 up(&SB_JOURNAL(p_s_sb
)->j_lock
);
571 static inline void get_journal_list(struct reiserfs_journal_list
*jl
)
576 static inline void put_journal_list(struct super_block
*s
,
577 struct reiserfs_journal_list
*jl
)
579 if (jl
->j_refcount
< 1) {
580 reiserfs_panic(s
, "trans id %lu, refcount at %d",
581 jl
->j_trans_id
, jl
->j_refcount
);
583 if (--jl
->j_refcount
== 0)
584 reiserfs_kfree(jl
, sizeof(struct reiserfs_journal_list
), s
);
588 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
589 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
592 static void cleanup_freed_for_journal_list(struct super_block
*p_s_sb
,
593 struct reiserfs_journal_list
*jl
)
596 struct reiserfs_list_bitmap
*jb
= jl
->j_list_bitmap
;
598 cleanup_bitmap_list(p_s_sb
, jb
);
600 jl
->j_list_bitmap
->journal_list
= NULL
;
601 jl
->j_list_bitmap
= NULL
;
604 static int journal_list_still_alive(struct super_block
*s
,
605 unsigned long trans_id
)
607 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
608 struct list_head
*entry
= &journal
->j_journal_list
;
609 struct reiserfs_journal_list
*jl
;
611 if (!list_empty(entry
)) {
612 jl
= JOURNAL_LIST_ENTRY(entry
->next
);
613 if (jl
->j_trans_id
<= trans_id
) {
620 static void reiserfs_end_buffer_io_sync(struct buffer_head
*bh
, int uptodate
)
622 char b
[BDEVNAME_SIZE
];
624 if (buffer_journaled(bh
)) {
625 reiserfs_warning(NULL
,
626 "clm-2084: pinned buffer %lu:%s sent to disk",
627 bh
->b_blocknr
, bdevname(bh
->b_bdev
, b
));
630 set_buffer_uptodate(bh
);
632 clear_buffer_uptodate(bh
);
637 static void reiserfs_end_ordered_io(struct buffer_head
*bh
, int uptodate
)
640 set_buffer_uptodate(bh
);
642 clear_buffer_uptodate(bh
);
647 static void submit_logged_buffer(struct buffer_head
*bh
)
650 bh
->b_end_io
= reiserfs_end_buffer_io_sync
;
651 clear_buffer_journal_new(bh
);
652 clear_buffer_dirty(bh
);
653 if (!test_clear_buffer_journal_test(bh
))
655 if (!buffer_uptodate(bh
))
657 submit_bh(WRITE
, bh
);
660 static void submit_ordered_buffer(struct buffer_head
*bh
)
663 bh
->b_end_io
= reiserfs_end_ordered_io
;
664 clear_buffer_dirty(bh
);
665 if (!buffer_uptodate(bh
))
667 submit_bh(WRITE
, bh
);
670 static int submit_barrier_buffer(struct buffer_head
*bh
)
673 bh
->b_end_io
= reiserfs_end_ordered_io
;
674 clear_buffer_dirty(bh
);
675 if (!buffer_uptodate(bh
))
677 return submit_bh(WRITE_BARRIER
, bh
);
680 static void check_barrier_completion(struct super_block
*s
,
681 struct buffer_head
*bh
)
683 if (buffer_eopnotsupp(bh
)) {
684 clear_buffer_eopnotsupp(bh
);
686 set_buffer_uptodate(bh
);
687 set_buffer_dirty(bh
);
688 sync_dirty_buffer(bh
);
692 #define CHUNK_SIZE 32
693 struct buffer_chunk
{
694 struct buffer_head
*bh
[CHUNK_SIZE
];
698 static void write_chunk(struct buffer_chunk
*chunk
)
702 for (i
= 0; i
< chunk
->nr
; i
++) {
703 submit_logged_buffer(chunk
->bh
[i
]);
709 static void write_ordered_chunk(struct buffer_chunk
*chunk
)
713 for (i
= 0; i
< chunk
->nr
; i
++) {
714 submit_ordered_buffer(chunk
->bh
[i
]);
720 static int add_to_chunk(struct buffer_chunk
*chunk
, struct buffer_head
*bh
,
721 spinlock_t
* lock
, void (fn
) (struct buffer_chunk
*))
724 if (chunk
->nr
>= CHUNK_SIZE
)
726 chunk
->bh
[chunk
->nr
++] = bh
;
727 if (chunk
->nr
>= CHUNK_SIZE
) {
738 static atomic_t nr_reiserfs_jh
= ATOMIC_INIT(0);
739 static struct reiserfs_jh
*alloc_jh(void)
741 struct reiserfs_jh
*jh
;
743 jh
= kmalloc(sizeof(*jh
), GFP_NOFS
);
745 atomic_inc(&nr_reiserfs_jh
);
753 * we want to free the jh when the buffer has been written
756 void reiserfs_free_jh(struct buffer_head
*bh
)
758 struct reiserfs_jh
*jh
;
762 bh
->b_private
= NULL
;
764 list_del_init(&jh
->list
);
766 if (atomic_read(&nr_reiserfs_jh
) <= 0)
768 atomic_dec(&nr_reiserfs_jh
);
773 static inline int __add_jh(struct reiserfs_journal
*j
, struct buffer_head
*bh
,
776 struct reiserfs_jh
*jh
;
779 spin_lock(&j
->j_dirty_buffers_lock
);
780 if (!bh
->b_private
) {
781 spin_unlock(&j
->j_dirty_buffers_lock
);
785 list_del_init(&jh
->list
);
790 spin_lock(&j
->j_dirty_buffers_lock
);
791 /* buffer must be locked for __add_jh, should be able to have
792 * two adds at the same time
799 jh
->jl
= j
->j_current_jl
;
801 list_add_tail(&jh
->list
, &jh
->jl
->j_tail_bh_list
);
803 list_add_tail(&jh
->list
, &jh
->jl
->j_bh_list
);
805 spin_unlock(&j
->j_dirty_buffers_lock
);
809 int reiserfs_add_tail_list(struct inode
*inode
, struct buffer_head
*bh
)
811 return __add_jh(SB_JOURNAL(inode
->i_sb
), bh
, 1);
813 int reiserfs_add_ordered_list(struct inode
*inode
, struct buffer_head
*bh
)
815 return __add_jh(SB_JOURNAL(inode
->i_sb
), bh
, 0);
818 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
819 static int write_ordered_buffers(spinlock_t
* lock
,
820 struct reiserfs_journal
*j
,
821 struct reiserfs_journal_list
*jl
,
822 struct list_head
*list
)
824 struct buffer_head
*bh
;
825 struct reiserfs_jh
*jh
;
826 int ret
= j
->j_errno
;
827 struct buffer_chunk chunk
;
828 struct list_head tmp
;
829 INIT_LIST_HEAD(&tmp
);
833 while (!list_empty(list
)) {
834 jh
= JH_ENTRY(list
->next
);
837 if (test_set_buffer_locked(bh
)) {
838 if (!buffer_dirty(bh
)) {
839 list_del_init(&jh
->list
);
840 list_add(&jh
->list
, &tmp
);
845 write_ordered_chunk(&chunk
);
851 if (buffer_dirty(bh
)) {
852 list_del_init(&jh
->list
);
853 list_add(&jh
->list
, &tmp
);
854 add_to_chunk(&chunk
, bh
, lock
, write_ordered_chunk
);
856 reiserfs_free_jh(bh
);
861 cond_resched_lock(lock
);
865 write_ordered_chunk(&chunk
);
868 while (!list_empty(&tmp
)) {
869 jh
= JH_ENTRY(tmp
.prev
);
872 reiserfs_free_jh(bh
);
874 if (buffer_locked(bh
)) {
879 if (!buffer_uptodate(bh
)) {
883 cond_resched_lock(lock
);
889 static int flush_older_commits(struct super_block
*s
,
890 struct reiserfs_journal_list
*jl
)
892 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
893 struct reiserfs_journal_list
*other_jl
;
894 struct reiserfs_journal_list
*first_jl
;
895 struct list_head
*entry
;
896 unsigned long trans_id
= jl
->j_trans_id
;
897 unsigned long other_trans_id
;
898 unsigned long first_trans_id
;
902 * first we walk backwards to find the oldest uncommitted transation
905 entry
= jl
->j_list
.prev
;
907 other_jl
= JOURNAL_LIST_ENTRY(entry
);
908 if (entry
== &journal
->j_journal_list
||
909 atomic_read(&other_jl
->j_older_commits_done
))
913 entry
= other_jl
->j_list
.prev
;
916 /* if we didn't find any older uncommitted transactions, return now */
917 if (first_jl
== jl
) {
921 first_trans_id
= first_jl
->j_trans_id
;
923 entry
= &first_jl
->j_list
;
925 other_jl
= JOURNAL_LIST_ENTRY(entry
);
926 other_trans_id
= other_jl
->j_trans_id
;
928 if (other_trans_id
< trans_id
) {
929 if (atomic_read(&other_jl
->j_commit_left
) != 0) {
930 flush_commit_list(s
, other_jl
, 0);
932 /* list we were called with is gone, return */
933 if (!journal_list_still_alive(s
, trans_id
))
936 /* the one we just flushed is gone, this means all
937 * older lists are also gone, so first_jl is no longer
938 * valid either. Go back to the beginning.
940 if (!journal_list_still_alive
941 (s
, other_trans_id
)) {
946 if (entry
== &journal
->j_journal_list
)
954 int reiserfs_async_progress_wait(struct super_block
*s
)
957 struct reiserfs_journal
*j
= SB_JOURNAL(s
);
958 if (atomic_read(&j
->j_async_throttle
))
959 blk_congestion_wait(WRITE
, HZ
/ 10);
964 ** if this journal list still has commit blocks unflushed, send them to disk.
966 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
967 ** Before the commit block can by written, every other log block must be safely on disk
970 static int flush_commit_list(struct super_block
*s
,
971 struct reiserfs_journal_list
*jl
, int flushall
)
975 struct buffer_head
*tbh
= NULL
;
976 unsigned long trans_id
= jl
->j_trans_id
;
977 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
981 reiserfs_check_lock_depth(s
, "flush_commit_list");
983 if (atomic_read(&jl
->j_older_commits_done
)) {
989 /* before we can put our commit blocks on disk, we have to make sure everyone older than
992 BUG_ON(jl
->j_len
<= 0);
993 BUG_ON(trans_id
== journal
->j_trans_id
);
995 get_journal_list(jl
);
997 if (flush_older_commits(s
, jl
) == 1) {
998 /* list disappeared during flush_older_commits. return */
1003 /* make sure nobody is trying to flush this one at the same time */
1004 down(&jl
->j_commit_lock
);
1005 if (!journal_list_still_alive(s
, trans_id
)) {
1006 up(&jl
->j_commit_lock
);
1009 BUG_ON(jl
->j_trans_id
== 0);
1011 /* this commit is done, exit */
1012 if (atomic_read(&(jl
->j_commit_left
)) <= 0) {
1014 atomic_set(&(jl
->j_older_commits_done
), 1);
1016 up(&jl
->j_commit_lock
);
1020 if (!list_empty(&jl
->j_bh_list
)) {
1022 write_ordered_buffers(&journal
->j_dirty_buffers_lock
,
1023 journal
, jl
, &jl
->j_bh_list
);
1026 BUG_ON(!list_empty(&jl
->j_bh_list
));
1028 * for the description block and all the log blocks, submit any buffers
1029 * that haven't already reached the disk
1031 atomic_inc(&journal
->j_async_throttle
);
1032 for (i
= 0; i
< (jl
->j_len
+ 1); i
++) {
1033 bn
= SB_ONDISK_JOURNAL_1st_BLOCK(s
) + (jl
->j_start
+ i
) %
1034 SB_ONDISK_JOURNAL_SIZE(s
);
1035 tbh
= journal_find_get_block(s
, bn
);
1036 if (buffer_dirty(tbh
)) /* redundant, ll_rw_block() checks */
1037 ll_rw_block(SWRITE
, 1, &tbh
);
1040 atomic_dec(&journal
->j_async_throttle
);
1042 /* wait on everything written so far before writing the commit
1043 * if we are in barrier mode, send the commit down now
1045 barrier
= reiserfs_barrier_flush(s
);
1048 lock_buffer(jl
->j_commit_bh
);
1049 ret
= submit_barrier_buffer(jl
->j_commit_bh
);
1050 if (ret
== -EOPNOTSUPP
) {
1051 set_buffer_uptodate(jl
->j_commit_bh
);
1056 for (i
= 0; i
< (jl
->j_len
+ 1); i
++) {
1057 bn
= SB_ONDISK_JOURNAL_1st_BLOCK(s
) +
1058 (jl
->j_start
+ i
) % SB_ONDISK_JOURNAL_SIZE(s
);
1059 tbh
= journal_find_get_block(s
, bn
);
1060 wait_on_buffer(tbh
);
1061 // since we're using ll_rw_blk above, it might have skipped over
1062 // a locked buffer. Double check here
1064 if (buffer_dirty(tbh
)) /* redundant, sync_dirty_buffer() checks */
1065 sync_dirty_buffer(tbh
);
1066 if (unlikely(!buffer_uptodate(tbh
))) {
1067 #ifdef CONFIG_REISERFS_CHECK
1068 reiserfs_warning(s
, "journal-601, buffer write failed");
1072 put_bh(tbh
); /* once for journal_find_get_block */
1073 put_bh(tbh
); /* once due to original getblk in do_journal_end */
1074 atomic_dec(&(jl
->j_commit_left
));
1077 BUG_ON(atomic_read(&(jl
->j_commit_left
)) != 1);
1080 if (buffer_dirty(jl
->j_commit_bh
))
1082 mark_buffer_dirty(jl
->j_commit_bh
);
1083 sync_dirty_buffer(jl
->j_commit_bh
);
1085 wait_on_buffer(jl
->j_commit_bh
);
1087 check_barrier_completion(s
, jl
->j_commit_bh
);
1089 /* If there was a write error in the journal - we can't commit this
1090 * transaction - it will be invalid and, if successful, will just end
1091 * up propogating the write error out to the filesystem. */
1092 if (unlikely(!buffer_uptodate(jl
->j_commit_bh
))) {
1093 #ifdef CONFIG_REISERFS_CHECK
1094 reiserfs_warning(s
, "journal-615: buffer write failed");
1098 bforget(jl
->j_commit_bh
);
1099 if (journal
->j_last_commit_id
!= 0 &&
1100 (jl
->j_trans_id
- journal
->j_last_commit_id
) != 1) {
1101 reiserfs_warning(s
, "clm-2200: last commit %lu, current %lu",
1102 journal
->j_last_commit_id
, jl
->j_trans_id
);
1104 journal
->j_last_commit_id
= jl
->j_trans_id
;
1106 /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */
1107 cleanup_freed_for_journal_list(s
, jl
);
1109 retval
= retval
? retval
: journal
->j_errno
;
1111 /* mark the metadata dirty */
1113 dirty_one_transaction(s
, jl
);
1114 atomic_dec(&(jl
->j_commit_left
));
1117 atomic_set(&(jl
->j_older_commits_done
), 1);
1119 up(&jl
->j_commit_lock
);
1121 put_journal_list(s
, jl
);
1124 reiserfs_abort(s
, retval
, "Journal write error in %s",
1131 ** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or
1132 ** returns NULL if it can't find anything
1134 static struct reiserfs_journal_list
*find_newer_jl_for_cn(struct
1135 reiserfs_journal_cnode
1138 struct super_block
*sb
= cn
->sb
;
1139 b_blocknr_t blocknr
= cn
->blocknr
;
1143 if (cn
->sb
== sb
&& cn
->blocknr
== blocknr
&& cn
->jlist
) {
1151 static void remove_journal_hash(struct super_block
*,
1152 struct reiserfs_journal_cnode
**,
1153 struct reiserfs_journal_list
*, unsigned long,
1157 ** once all the real blocks have been flushed, it is safe to remove them from the
1158 ** journal list for this transaction. Aside from freeing the cnode, this also allows the
1159 ** block to be reallocated for data blocks if it had been deleted.
1161 static void remove_all_from_journal_list(struct super_block
*p_s_sb
,
1162 struct reiserfs_journal_list
*jl
,
1165 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
1166 struct reiserfs_journal_cnode
*cn
, *last
;
1167 cn
= jl
->j_realblock
;
1169 /* which is better, to lock once around the whole loop, or
1170 ** to lock for each call to remove_journal_hash?
1173 if (cn
->blocknr
!= 0) {
1175 reiserfs_warning(p_s_sb
,
1176 "block %u, bh is %d, state %ld",
1177 cn
->blocknr
, cn
->bh
? 1 : 0,
1181 remove_journal_hash(p_s_sb
, journal
->j_list_hash_table
,
1182 jl
, cn
->blocknr
, 1);
1186 free_cnode(p_s_sb
, last
);
1188 jl
->j_realblock
= NULL
;
1192 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1193 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1194 ** releasing blocks in this transaction for reuse as data blocks.
1195 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1198 static int _update_journal_header_block(struct super_block
*p_s_sb
,
1199 unsigned long offset
,
1200 unsigned long trans_id
)
1202 struct reiserfs_journal_header
*jh
;
1203 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
1205 if (reiserfs_is_journal_aborted(journal
))
1208 if (trans_id
>= journal
->j_last_flush_trans_id
) {
1209 if (buffer_locked((journal
->j_header_bh
))) {
1210 wait_on_buffer((journal
->j_header_bh
));
1211 if (unlikely(!buffer_uptodate(journal
->j_header_bh
))) {
1212 #ifdef CONFIG_REISERFS_CHECK
1213 reiserfs_warning(p_s_sb
,
1214 "journal-699: buffer write failed");
1219 journal
->j_last_flush_trans_id
= trans_id
;
1220 journal
->j_first_unflushed_offset
= offset
;
1221 jh
= (struct reiserfs_journal_header
*)(journal
->j_header_bh
->
1223 jh
->j_last_flush_trans_id
= cpu_to_le32(trans_id
);
1224 jh
->j_first_unflushed_offset
= cpu_to_le32(offset
);
1225 jh
->j_mount_id
= cpu_to_le32(journal
->j_mount_id
);
1227 if (reiserfs_barrier_flush(p_s_sb
)) {
1229 lock_buffer(journal
->j_header_bh
);
1230 ret
= submit_barrier_buffer(journal
->j_header_bh
);
1231 if (ret
== -EOPNOTSUPP
) {
1232 set_buffer_uptodate(journal
->j_header_bh
);
1233 disable_barrier(p_s_sb
);
1236 wait_on_buffer(journal
->j_header_bh
);
1237 check_barrier_completion(p_s_sb
, journal
->j_header_bh
);
1240 set_buffer_dirty(journal
->j_header_bh
);
1241 sync_dirty_buffer(journal
->j_header_bh
);
1243 if (!buffer_uptodate(journal
->j_header_bh
)) {
1244 reiserfs_warning(p_s_sb
,
1245 "journal-837: IO error during journal replay");
1252 static int update_journal_header_block(struct super_block
*p_s_sb
,
1253 unsigned long offset
,
1254 unsigned long trans_id
)
1256 return _update_journal_header_block(p_s_sb
, offset
, trans_id
);
1260 ** flush any and all journal lists older than you are
1261 ** can only be called from flush_journal_list
1263 static int flush_older_journal_lists(struct super_block
*p_s_sb
,
1264 struct reiserfs_journal_list
*jl
)
1266 struct list_head
*entry
;
1267 struct reiserfs_journal_list
*other_jl
;
1268 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
1269 unsigned long trans_id
= jl
->j_trans_id
;
1271 /* we know we are the only ones flushing things, no extra race
1272 * protection is required.
1275 entry
= journal
->j_journal_list
.next
;
1277 if (entry
== &journal
->j_journal_list
)
1279 other_jl
= JOURNAL_LIST_ENTRY(entry
);
1280 if (other_jl
->j_trans_id
< trans_id
) {
1281 BUG_ON(other_jl
->j_refcount
<= 0);
1282 /* do not flush all */
1283 flush_journal_list(p_s_sb
, other_jl
, 0);
1285 /* other_jl is now deleted from the list */
1291 static void del_from_work_list(struct super_block
*s
,
1292 struct reiserfs_journal_list
*jl
)
1294 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1295 if (!list_empty(&jl
->j_working_list
)) {
1296 list_del_init(&jl
->j_working_list
);
1297 journal
->j_num_work_lists
--;
1301 /* flush a journal list, both commit and real blocks
1303 ** always set flushall to 1, unless you are calling from inside
1304 ** flush_journal_list
1306 ** IMPORTANT. This can only be called while there are no journal writers,
1307 ** and the journal is locked. That means it can only be called from
1308 ** do_journal_end, or by journal_release
1310 static int flush_journal_list(struct super_block
*s
,
1311 struct reiserfs_journal_list
*jl
, int flushall
)
1313 struct reiserfs_journal_list
*pjl
;
1314 struct reiserfs_journal_cnode
*cn
, *last
;
1318 struct buffer_head
*saved_bh
;
1319 unsigned long j_len_saved
= jl
->j_len
;
1320 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1323 BUG_ON(j_len_saved
<= 0);
1325 if (atomic_read(&journal
->j_wcount
) != 0) {
1327 "clm-2048: flush_journal_list called with wcount %d",
1328 atomic_read(&journal
->j_wcount
));
1330 BUG_ON(jl
->j_trans_id
== 0);
1332 /* if flushall == 0, the lock is already held */
1334 down(&journal
->j_flush_sem
);
1335 } else if (!down_trylock(&journal
->j_flush_sem
)) {
1340 if (j_len_saved
> journal
->j_trans_max
) {
1342 "journal-715: flush_journal_list, length is %lu, trans id %lu\n",
1343 j_len_saved
, jl
->j_trans_id
);
1349 /* if all the work is already done, get out of here */
1350 if (atomic_read(&(jl
->j_nonzerolen
)) <= 0 &&
1351 atomic_read(&(jl
->j_commit_left
)) <= 0) {
1352 goto flush_older_and_return
;
1355 /* start by putting the commit list on disk. This will also flush
1356 ** the commit lists of any olders transactions
1358 flush_commit_list(s
, jl
, 1);
1360 if (!(jl
->j_state
& LIST_DIRTY
)
1361 && !reiserfs_is_journal_aborted(journal
))
1364 /* are we done now? */
1365 if (atomic_read(&(jl
->j_nonzerolen
)) <= 0 &&
1366 atomic_read(&(jl
->j_commit_left
)) <= 0) {
1367 goto flush_older_and_return
;
1370 /* loop through each cnode, see if we need to write it,
1371 ** or wait on a more recent transaction, or just ignore it
1373 if (atomic_read(&(journal
->j_wcount
)) != 0) {
1375 "journal-844: panic journal list is flushing, wcount is not 0\n");
1377 cn
= jl
->j_realblock
;
1382 /* blocknr of 0 is no longer in the hash, ignore it */
1383 if (cn
->blocknr
== 0) {
1387 /* This transaction failed commit. Don't write out to the disk */
1388 if (!(jl
->j_state
& LIST_DIRTY
))
1391 pjl
= find_newer_jl_for_cn(cn
);
1392 /* the order is important here. We check pjl to make sure we
1393 ** don't clear BH_JDirty_wait if we aren't the one writing this
1396 if (!pjl
&& cn
->bh
) {
1399 /* we do this to make sure nobody releases the buffer while
1400 ** we are working with it
1404 if (buffer_journal_dirty(saved_bh
)) {
1405 BUG_ON(!can_dirty(cn
));
1408 } else if (can_dirty(cn
)) {
1409 /* everything with !pjl && jwait should be writable */
1414 /* if someone has this block in a newer transaction, just make
1415 ** sure they are commited, and don't try writing it to disk
1418 if (atomic_read(&pjl
->j_commit_left
))
1419 flush_commit_list(s
, pjl
, 1);
1423 /* bh == NULL when the block got to disk on its own, OR,
1424 ** the block got freed in a future transaction
1426 if (saved_bh
== NULL
) {
1430 /* this should never happen. kupdate_one_transaction has this list
1431 ** locked while it works, so we should never see a buffer here that
1432 ** is not marked JDirty_wait
1434 if ((!was_jwait
) && !buffer_locked(saved_bh
)) {
1436 "journal-813: BAD! buffer %llu %cdirty %cjwait, "
1437 "not in a newer tranasction",
1438 (unsigned long long)saved_bh
->
1439 b_blocknr
, was_dirty
? ' ' : '!',
1440 was_jwait
? ' ' : '!');
1443 /* we inc again because saved_bh gets decremented at free_cnode */
1445 set_bit(BLOCK_NEEDS_FLUSH
, &cn
->state
);
1446 lock_buffer(saved_bh
);
1447 BUG_ON(cn
->blocknr
!= saved_bh
->b_blocknr
);
1448 if (buffer_dirty(saved_bh
))
1449 submit_logged_buffer(saved_bh
);
1451 unlock_buffer(saved_bh
);
1455 "clm-2082: Unable to flush buffer %llu in %s",
1456 (unsigned long long)saved_bh
->
1457 b_blocknr
, __FUNCTION__
);
1463 /* we incremented this to keep others from taking the buffer head away */
1465 if (atomic_read(&(saved_bh
->b_count
)) < 0) {
1467 "journal-945: saved_bh->b_count < 0");
1472 cn
= jl
->j_realblock
;
1474 if (test_bit(BLOCK_NEEDS_FLUSH
, &cn
->state
)) {
1477 "journal-1011: cn->bh is NULL\n");
1479 wait_on_buffer(cn
->bh
);
1482 "journal-1012: cn->bh is NULL\n");
1484 if (unlikely(!buffer_uptodate(cn
->bh
))) {
1485 #ifdef CONFIG_REISERFS_CHECK
1487 "journal-949: buffer write failed\n");
1491 /* note, we must clear the JDirty_wait bit after the up to date
1492 ** check, otherwise we race against our flushpage routine
1494 BUG_ON(!test_clear_buffer_journal_dirty
1497 /* undo the inc from journal_mark_dirty */
1506 reiserfs_abort(s
, -EIO
,
1507 "Write error while pushing transaction to disk in %s",
1509 flush_older_and_return
:
1511 /* before we can update the journal header block, we _must_ flush all
1512 ** real blocks from all older transactions to disk. This is because
1513 ** once the header block is updated, this transaction will not be
1514 ** replayed after a crash
1517 flush_older_journal_lists(s
, jl
);
1520 err
= journal
->j_errno
;
1521 /* before we can remove everything from the hash tables for this
1522 ** transaction, we must make sure it can never be replayed
1524 ** since we are only called from do_journal_end, we know for sure there
1525 ** are no allocations going on while we are flushing journal lists. So,
1526 ** we only need to update the journal header block for the last list
1529 if (!err
&& flushall
) {
1531 update_journal_header_block(s
,
1532 (jl
->j_start
+ jl
->j_len
+
1533 2) % SB_ONDISK_JOURNAL_SIZE(s
),
1536 reiserfs_abort(s
, -EIO
,
1537 "Write error while updating journal header in %s",
1540 remove_all_from_journal_list(s
, jl
, 0);
1541 list_del_init(&jl
->j_list
);
1542 journal
->j_num_lists
--;
1543 del_from_work_list(s
, jl
);
1545 if (journal
->j_last_flush_id
!= 0 &&
1546 (jl
->j_trans_id
- journal
->j_last_flush_id
) != 1) {
1547 reiserfs_warning(s
, "clm-2201: last flush %lu, current %lu",
1548 journal
->j_last_flush_id
, jl
->j_trans_id
);
1550 journal
->j_last_flush_id
= jl
->j_trans_id
;
1552 /* not strictly required since we are freeing the list, but it should
1553 * help find code using dead lists later on
1556 atomic_set(&(jl
->j_nonzerolen
), 0);
1558 jl
->j_realblock
= NULL
;
1559 jl
->j_commit_bh
= NULL
;
1562 put_journal_list(s
, jl
);
1564 up(&journal
->j_flush_sem
);
1569 static int write_one_transaction(struct super_block
*s
,
1570 struct reiserfs_journal_list
*jl
,
1571 struct buffer_chunk
*chunk
)
1573 struct reiserfs_journal_cnode
*cn
;
1576 jl
->j_state
|= LIST_TOUCHED
;
1577 del_from_work_list(s
, jl
);
1578 if (jl
->j_len
== 0 || atomic_read(&jl
->j_nonzerolen
) == 0) {
1582 cn
= jl
->j_realblock
;
1584 /* if the blocknr == 0, this has been cleared from the hash,
1587 if (cn
->blocknr
== 0) {
1590 if (cn
->bh
&& can_dirty(cn
) && buffer_dirty(cn
->bh
)) {
1591 struct buffer_head
*tmp_bh
;
1592 /* we can race against journal_mark_freed when we try
1593 * to lock_buffer(cn->bh), so we have to inc the buffer
1594 * count, and recheck things after locking
1598 lock_buffer(tmp_bh
);
1599 if (cn
->bh
&& can_dirty(cn
) && buffer_dirty(tmp_bh
)) {
1600 if (!buffer_journal_dirty(tmp_bh
) ||
1601 buffer_journal_prepared(tmp_bh
))
1603 add_to_chunk(chunk
, tmp_bh
, NULL
, write_chunk
);
1606 /* note, cn->bh might be null now */
1607 unlock_buffer(tmp_bh
);
1618 /* used by flush_commit_list */
1619 static int dirty_one_transaction(struct super_block
*s
,
1620 struct reiserfs_journal_list
*jl
)
1622 struct reiserfs_journal_cnode
*cn
;
1623 struct reiserfs_journal_list
*pjl
;
1626 jl
->j_state
|= LIST_DIRTY
;
1627 cn
= jl
->j_realblock
;
1629 /* look for a more recent transaction that logged this
1630 ** buffer. Only the most recent transaction with a buffer in
1631 ** it is allowed to send that buffer to disk
1633 pjl
= find_newer_jl_for_cn(cn
);
1634 if (!pjl
&& cn
->blocknr
&& cn
->bh
1635 && buffer_journal_dirty(cn
->bh
)) {
1636 BUG_ON(!can_dirty(cn
));
1637 /* if the buffer is prepared, it will either be logged
1638 * or restored. If restored, we need to make sure
1639 * it actually gets marked dirty
1641 clear_buffer_journal_new(cn
->bh
);
1642 if (buffer_journal_prepared(cn
->bh
)) {
1643 set_buffer_journal_restore_dirty(cn
->bh
);
1645 set_buffer_journal_test(cn
->bh
);
1646 mark_buffer_dirty(cn
->bh
);
1654 static int kupdate_transactions(struct super_block
*s
,
1655 struct reiserfs_journal_list
*jl
,
1656 struct reiserfs_journal_list
**next_jl
,
1657 unsigned long *next_trans_id
,
1658 int num_blocks
, int num_trans
)
1662 int transactions_flushed
= 0;
1663 unsigned long orig_trans_id
= jl
->j_trans_id
;
1664 struct buffer_chunk chunk
;
1665 struct list_head
*entry
;
1666 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1669 down(&journal
->j_flush_sem
);
1670 if (!journal_list_still_alive(s
, orig_trans_id
)) {
1674 /* we've got j_flush_sem held, nobody is going to delete any
1675 * of these lists out from underneath us
1677 while ((num_trans
&& transactions_flushed
< num_trans
) ||
1678 (!num_trans
&& written
< num_blocks
)) {
1680 if (jl
->j_len
== 0 || (jl
->j_state
& LIST_TOUCHED
) ||
1681 atomic_read(&jl
->j_commit_left
)
1682 || !(jl
->j_state
& LIST_DIRTY
)) {
1683 del_from_work_list(s
, jl
);
1686 ret
= write_one_transaction(s
, jl
, &chunk
);
1690 transactions_flushed
++;
1692 entry
= jl
->j_list
.next
;
1695 if (entry
== &journal
->j_journal_list
) {
1698 jl
= JOURNAL_LIST_ENTRY(entry
);
1700 /* don't bother with older transactions */
1701 if (jl
->j_trans_id
<= orig_trans_id
)
1705 write_chunk(&chunk
);
1709 up(&journal
->j_flush_sem
);
1713 /* for o_sync and fsync heavy applications, they tend to use
1714 ** all the journa list slots with tiny transactions. These
1715 ** trigger lots and lots of calls to update the header block, which
1716 ** adds seeks and slows things down.
1718 ** This function tries to clear out a large chunk of the journal lists
1719 ** at once, which makes everything faster since only the newest journal
1720 ** list updates the header block
1722 static int flush_used_journal_lists(struct super_block
*s
,
1723 struct reiserfs_journal_list
*jl
)
1725 unsigned long len
= 0;
1726 unsigned long cur_len
;
1730 struct reiserfs_journal_list
*tjl
;
1731 struct reiserfs_journal_list
*flush_jl
;
1732 unsigned long trans_id
;
1733 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1735 flush_jl
= tjl
= jl
;
1737 /* in data logging mode, try harder to flush a lot of blocks */
1738 if (reiserfs_data_log(s
))
1740 /* flush for 256 transactions or limit blocks, whichever comes first */
1741 for (i
= 0; i
< 256 && len
< limit
; i
++) {
1742 if (atomic_read(&tjl
->j_commit_left
) ||
1743 tjl
->j_trans_id
< jl
->j_trans_id
) {
1746 cur_len
= atomic_read(&tjl
->j_nonzerolen
);
1748 tjl
->j_state
&= ~LIST_TOUCHED
;
1752 if (tjl
->j_list
.next
== &journal
->j_journal_list
)
1754 tjl
= JOURNAL_LIST_ENTRY(tjl
->j_list
.next
);
1756 /* try to find a group of blocks we can flush across all the
1757 ** transactions, but only bother if we've actually spanned
1758 ** across multiple lists
1760 if (flush_jl
!= jl
) {
1761 ret
= kupdate_transactions(s
, jl
, &tjl
, &trans_id
, len
, i
);
1763 flush_journal_list(s
, flush_jl
, 1);
1768 ** removes any nodes in table with name block and dev as bh.
1769 ** only touchs the hnext and hprev pointers.
1771 void remove_journal_hash(struct super_block
*sb
,
1772 struct reiserfs_journal_cnode
**table
,
1773 struct reiserfs_journal_list
*jl
,
1774 unsigned long block
, int remove_freed
)
1776 struct reiserfs_journal_cnode
*cur
;
1777 struct reiserfs_journal_cnode
**head
;
1779 head
= &(journal_hash(table
, sb
, block
));
1785 if (cur
->blocknr
== block
&& cur
->sb
== sb
1786 && (jl
== NULL
|| jl
== cur
->jlist
)
1787 && (!test_bit(BLOCK_FREED
, &cur
->state
) || remove_freed
)) {
1789 cur
->hnext
->hprev
= cur
->hprev
;
1792 cur
->hprev
->hnext
= cur
->hnext
;
1799 if (cur
->bh
&& cur
->jlist
) /* anybody who clears the cur->bh will also dec the nonzerolen */
1800 atomic_dec(&(cur
->jlist
->j_nonzerolen
));
1808 static void free_journal_ram(struct super_block
*p_s_sb
)
1810 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
1811 reiserfs_kfree(journal
->j_current_jl
,
1812 sizeof(struct reiserfs_journal_list
), p_s_sb
);
1813 journal
->j_num_lists
--;
1815 vfree(journal
->j_cnode_free_orig
);
1816 free_list_bitmaps(p_s_sb
, journal
->j_list_bitmap
);
1817 free_bitmap_nodes(p_s_sb
); /* must be after free_list_bitmaps */
1818 if (journal
->j_header_bh
) {
1819 brelse(journal
->j_header_bh
);
1821 /* j_header_bh is on the journal dev, make sure not to release the journal
1822 * dev until we brelse j_header_bh
1824 release_journal_dev(p_s_sb
, journal
);
1829 ** call on unmount. Only set error to 1 if you haven't made your way out
1830 ** of read_super() yet. Any other caller must keep error at 0.
1832 static int do_journal_release(struct reiserfs_transaction_handle
*th
,
1833 struct super_block
*p_s_sb
, int error
)
1835 struct reiserfs_transaction_handle myth
;
1837 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
1839 /* we only want to flush out transactions if we were called with error == 0
1841 if (!error
&& !(p_s_sb
->s_flags
& MS_RDONLY
)) {
1842 /* end the current trans */
1843 BUG_ON(!th
->t_trans_id
);
1844 do_journal_end(th
, p_s_sb
, 10, FLUSH_ALL
);
1846 /* make sure something gets logged to force our way into the flush code */
1847 if (!journal_join(&myth
, p_s_sb
, 1)) {
1848 reiserfs_prepare_for_journal(p_s_sb
,
1849 SB_BUFFER_WITH_SB(p_s_sb
),
1851 journal_mark_dirty(&myth
, p_s_sb
,
1852 SB_BUFFER_WITH_SB(p_s_sb
));
1853 do_journal_end(&myth
, p_s_sb
, 1, FLUSH_ALL
);
1858 /* this also catches errors during the do_journal_end above */
1859 if (!error
&& reiserfs_is_journal_aborted(journal
)) {
1860 memset(&myth
, 0, sizeof(myth
));
1861 if (!journal_join_abort(&myth
, p_s_sb
, 1)) {
1862 reiserfs_prepare_for_journal(p_s_sb
,
1863 SB_BUFFER_WITH_SB(p_s_sb
),
1865 journal_mark_dirty(&myth
, p_s_sb
,
1866 SB_BUFFER_WITH_SB(p_s_sb
));
1867 do_journal_end(&myth
, p_s_sb
, 1, FLUSH_ALL
);
1871 reiserfs_mounted_fs_count
--;
1872 /* wait for all commits to finish */
1873 cancel_delayed_work(&SB_JOURNAL(p_s_sb
)->j_work
);
1874 flush_workqueue(commit_wq
);
1875 if (!reiserfs_mounted_fs_count
) {
1876 destroy_workqueue(commit_wq
);
1880 free_journal_ram(p_s_sb
);
1886 ** call on unmount. flush all journal trans, release all alloc'd ram
1888 int journal_release(struct reiserfs_transaction_handle
*th
,
1889 struct super_block
*p_s_sb
)
1891 return do_journal_release(th
, p_s_sb
, 0);
1895 ** only call from an error condition inside reiserfs_read_super!
1897 int journal_release_error(struct reiserfs_transaction_handle
*th
,
1898 struct super_block
*p_s_sb
)
1900 return do_journal_release(th
, p_s_sb
, 1);
1903 /* compares description block with commit block. returns 1 if they differ, 0 if they are the same */
1904 static int journal_compare_desc_commit(struct super_block
*p_s_sb
,
1905 struct reiserfs_journal_desc
*desc
,
1906 struct reiserfs_journal_commit
*commit
)
1908 if (get_commit_trans_id(commit
) != get_desc_trans_id(desc
) ||
1909 get_commit_trans_len(commit
) != get_desc_trans_len(desc
) ||
1910 get_commit_trans_len(commit
) > SB_JOURNAL(p_s_sb
)->j_trans_max
||
1911 get_commit_trans_len(commit
) <= 0) {
1917 /* returns 0 if it did not find a description block
1918 ** returns -1 if it found a corrupt commit block
1919 ** returns 1 if both desc and commit were valid
1921 static int journal_transaction_is_valid(struct super_block
*p_s_sb
,
1922 struct buffer_head
*d_bh
,
1923 unsigned long *oldest_invalid_trans_id
,
1924 unsigned long *newest_mount_id
)
1926 struct reiserfs_journal_desc
*desc
;
1927 struct reiserfs_journal_commit
*commit
;
1928 struct buffer_head
*c_bh
;
1929 unsigned long offset
;
1934 desc
= (struct reiserfs_journal_desc
*)d_bh
->b_data
;
1935 if (get_desc_trans_len(desc
) > 0
1936 && !memcmp(get_journal_desc_magic(d_bh
), JOURNAL_DESC_MAGIC
, 8)) {
1937 if (oldest_invalid_trans_id
&& *oldest_invalid_trans_id
1938 && get_desc_trans_id(desc
) > *oldest_invalid_trans_id
) {
1939 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
1940 "journal-986: transaction "
1941 "is valid returning because trans_id %d is greater than "
1942 "oldest_invalid %lu",
1943 get_desc_trans_id(desc
),
1944 *oldest_invalid_trans_id
);
1948 && *newest_mount_id
> get_desc_mount_id(desc
)) {
1949 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
1950 "journal-1087: transaction "
1951 "is valid returning because mount_id %d is less than "
1952 "newest_mount_id %lu",
1953 get_desc_mount_id(desc
),
1957 if (get_desc_trans_len(desc
) > SB_JOURNAL(p_s_sb
)->j_trans_max
) {
1958 reiserfs_warning(p_s_sb
,
1959 "journal-2018: Bad transaction length %d encountered, ignoring transaction",
1960 get_desc_trans_len(desc
));
1963 offset
= d_bh
->b_blocknr
- SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
);
1965 /* ok, we have a journal description block, lets see if the transaction was valid */
1967 journal_bread(p_s_sb
,
1968 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
1969 ((offset
+ get_desc_trans_len(desc
) +
1970 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb
)));
1973 commit
= (struct reiserfs_journal_commit
*)c_bh
->b_data
;
1974 if (journal_compare_desc_commit(p_s_sb
, desc
, commit
)) {
1975 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
1976 "journal_transaction_is_valid, commit offset %ld had bad "
1977 "time %d or length %d",
1979 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
),
1980 get_commit_trans_id(commit
),
1981 get_commit_trans_len(commit
));
1983 if (oldest_invalid_trans_id
) {
1984 *oldest_invalid_trans_id
=
1985 get_desc_trans_id(desc
);
1986 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
1988 "transaction_is_valid setting oldest invalid trans_id "
1990 get_desc_trans_id(desc
));
1995 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
1996 "journal-1006: found valid "
1997 "transaction start offset %llu, len %d id %d",
1999 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
),
2000 get_desc_trans_len(desc
),
2001 get_desc_trans_id(desc
));
2008 static void brelse_array(struct buffer_head
**heads
, int num
)
2011 for (i
= 0; i
< num
; i
++) {
2017 ** given the start, and values for the oldest acceptable transactions,
2018 ** this either reads in a replays a transaction, or returns because the transaction
2019 ** is invalid, or too old.
2021 static int journal_read_transaction(struct super_block
*p_s_sb
,
2022 unsigned long cur_dblock
,
2023 unsigned long oldest_start
,
2024 unsigned long oldest_trans_id
,
2025 unsigned long newest_mount_id
)
2027 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
2028 struct reiserfs_journal_desc
*desc
;
2029 struct reiserfs_journal_commit
*commit
;
2030 unsigned long trans_id
= 0;
2031 struct buffer_head
*c_bh
;
2032 struct buffer_head
*d_bh
;
2033 struct buffer_head
**log_blocks
= NULL
;
2034 struct buffer_head
**real_blocks
= NULL
;
2035 unsigned long trans_offset
;
2039 d_bh
= journal_bread(p_s_sb
, cur_dblock
);
2042 desc
= (struct reiserfs_journal_desc
*)d_bh
->b_data
;
2043 trans_offset
= d_bh
->b_blocknr
- SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
);
2044 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
, "journal-1037: "
2045 "journal_read_transaction, offset %llu, len %d mount_id %d",
2046 d_bh
->b_blocknr
- SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
),
2047 get_desc_trans_len(desc
), get_desc_mount_id(desc
));
2048 if (get_desc_trans_id(desc
) < oldest_trans_id
) {
2049 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
, "journal-1039: "
2050 "journal_read_trans skipping because %lu is too old",
2052 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
));
2056 if (get_desc_mount_id(desc
) != newest_mount_id
) {
2057 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
, "journal-1146: "
2058 "journal_read_trans skipping because %d is != "
2059 "newest_mount_id %lu", get_desc_mount_id(desc
),
2064 c_bh
= journal_bread(p_s_sb
, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
2065 ((trans_offset
+ get_desc_trans_len(desc
) + 1) %
2066 SB_ONDISK_JOURNAL_SIZE(p_s_sb
)));
2071 commit
= (struct reiserfs_journal_commit
*)c_bh
->b_data
;
2072 if (journal_compare_desc_commit(p_s_sb
, desc
, commit
)) {
2073 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2074 "journal_read_transaction, "
2075 "commit offset %llu had bad time %d or length %d",
2077 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
),
2078 get_commit_trans_id(commit
),
2079 get_commit_trans_len(commit
));
2084 trans_id
= get_desc_trans_id(desc
);
2085 /* now we know we've got a good transaction, and it was inside the valid time ranges */
2087 reiserfs_kmalloc(get_desc_trans_len(desc
) *
2088 sizeof(struct buffer_head
*), GFP_NOFS
, p_s_sb
);
2090 reiserfs_kmalloc(get_desc_trans_len(desc
) *
2091 sizeof(struct buffer_head
*), GFP_NOFS
, p_s_sb
);
2092 if (!log_blocks
|| !real_blocks
) {
2095 reiserfs_kfree(log_blocks
,
2096 get_desc_trans_len(desc
) *
2097 sizeof(struct buffer_head
*), p_s_sb
);
2098 reiserfs_kfree(real_blocks
,
2099 get_desc_trans_len(desc
) *
2100 sizeof(struct buffer_head
*), p_s_sb
);
2101 reiserfs_warning(p_s_sb
,
2102 "journal-1169: kmalloc failed, unable to mount FS");
2105 /* get all the buffer heads */
2106 trans_half
= journal_trans_half(p_s_sb
->s_blocksize
);
2107 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2109 journal_getblk(p_s_sb
,
2110 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
2112 i
) % SB_ONDISK_JOURNAL_SIZE(p_s_sb
));
2113 if (i
< trans_half
) {
2116 le32_to_cpu(desc
->j_realblock
[i
]));
2120 le32_to_cpu(commit
->
2121 j_realblock
[i
- trans_half
]));
2123 if (real_blocks
[i
]->b_blocknr
> SB_BLOCK_COUNT(p_s_sb
)) {
2124 reiserfs_warning(p_s_sb
,
2125 "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem");
2128 /* make sure we don't try to replay onto log or reserved area */
2129 if (is_block_in_log_or_reserved_area
2130 (p_s_sb
, real_blocks
[i
]->b_blocknr
)) {
2131 reiserfs_warning(p_s_sb
,
2132 "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block");
2134 brelse_array(log_blocks
, i
);
2135 brelse_array(real_blocks
, i
);
2138 reiserfs_kfree(log_blocks
,
2139 get_desc_trans_len(desc
) *
2140 sizeof(struct buffer_head
*), p_s_sb
);
2141 reiserfs_kfree(real_blocks
,
2142 get_desc_trans_len(desc
) *
2143 sizeof(struct buffer_head
*), p_s_sb
);
2147 /* read in the log blocks, memcpy to the corresponding real block */
2148 ll_rw_block(READ
, get_desc_trans_len(desc
), log_blocks
);
2149 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2150 wait_on_buffer(log_blocks
[i
]);
2151 if (!buffer_uptodate(log_blocks
[i
])) {
2152 reiserfs_warning(p_s_sb
,
2153 "journal-1212: REPLAY FAILURE fsck required! buffer write failed");
2154 brelse_array(log_blocks
+ i
,
2155 get_desc_trans_len(desc
) - i
);
2156 brelse_array(real_blocks
, get_desc_trans_len(desc
));
2159 reiserfs_kfree(log_blocks
,
2160 get_desc_trans_len(desc
) *
2161 sizeof(struct buffer_head
*), p_s_sb
);
2162 reiserfs_kfree(real_blocks
,
2163 get_desc_trans_len(desc
) *
2164 sizeof(struct buffer_head
*), p_s_sb
);
2167 memcpy(real_blocks
[i
]->b_data
, log_blocks
[i
]->b_data
,
2168 real_blocks
[i
]->b_size
);
2169 set_buffer_uptodate(real_blocks
[i
]);
2170 brelse(log_blocks
[i
]);
2172 /* flush out the real blocks */
2173 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2174 set_buffer_dirty(real_blocks
[i
]);
2175 ll_rw_block(SWRITE
, 1, real_blocks
+ i
);
2177 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2178 wait_on_buffer(real_blocks
[i
]);
2179 if (!buffer_uptodate(real_blocks
[i
])) {
2180 reiserfs_warning(p_s_sb
,
2181 "journal-1226: REPLAY FAILURE, fsck required! buffer write failed");
2182 brelse_array(real_blocks
+ i
,
2183 get_desc_trans_len(desc
) - i
);
2186 reiserfs_kfree(log_blocks
,
2187 get_desc_trans_len(desc
) *
2188 sizeof(struct buffer_head
*), p_s_sb
);
2189 reiserfs_kfree(real_blocks
,
2190 get_desc_trans_len(desc
) *
2191 sizeof(struct buffer_head
*), p_s_sb
);
2194 brelse(real_blocks
[i
]);
2197 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
2198 ((trans_offset
+ get_desc_trans_len(desc
) +
2199 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb
));
2200 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2201 "journal-1095: setting journal " "start to offset %ld",
2202 cur_dblock
- SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
));
2204 /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2205 journal
->j_start
= cur_dblock
- SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
);
2206 journal
->j_last_flush_trans_id
= trans_id
;
2207 journal
->j_trans_id
= trans_id
+ 1;
2210 reiserfs_kfree(log_blocks
,
2211 le32_to_cpu(desc
->j_len
) * sizeof(struct buffer_head
*),
2213 reiserfs_kfree(real_blocks
,
2214 le32_to_cpu(desc
->j_len
) * sizeof(struct buffer_head
*),
2219 /* This function reads blocks starting from block and to max_block of bufsize
2220 size (but no more than BUFNR blocks at a time). This proved to improve
2221 mounting speed on self-rebuilding raid5 arrays at least.
2222 Right now it is only used from journal code. But later we might use it
2224 Note: Do not use journal_getblk/sb_getblk functions here! */
2225 static struct buffer_head
*reiserfs_breada(struct block_device
*dev
, int block
,
2226 int bufsize
, unsigned int max_block
)
2228 struct buffer_head
*bhlist
[BUFNR
];
2229 unsigned int blocks
= BUFNR
;
2230 struct buffer_head
*bh
;
2233 bh
= __getblk(dev
, block
, bufsize
);
2234 if (buffer_uptodate(bh
))
2237 if (block
+ BUFNR
> max_block
) {
2238 blocks
= max_block
- block
;
2242 for (i
= 1; i
< blocks
; i
++) {
2243 bh
= __getblk(dev
, block
+ i
, bufsize
);
2244 if (buffer_uptodate(bh
)) {
2250 ll_rw_block(READ
, j
, bhlist
);
2251 for (i
= 1; i
< j
; i
++)
2255 if (buffer_uptodate(bh
))
2262 ** read and replay the log
2263 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2264 ** transaction. This tests that before finding all the transactions in the log, which makes normal mount times fast.
2266 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2268 ** On exit, it sets things up so the first transaction will work correctly.
2270 static int journal_read(struct super_block
*p_s_sb
)
2272 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
2273 struct reiserfs_journal_desc
*desc
;
2274 unsigned long oldest_trans_id
= 0;
2275 unsigned long oldest_invalid_trans_id
= 0;
2277 unsigned long oldest_start
= 0;
2278 unsigned long cur_dblock
= 0;
2279 unsigned long newest_mount_id
= 9;
2280 struct buffer_head
*d_bh
;
2281 struct reiserfs_journal_header
*jh
;
2282 int valid_journal_header
= 0;
2283 int replay_count
= 0;
2284 int continue_replay
= 1;
2286 char b
[BDEVNAME_SIZE
];
2288 cur_dblock
= SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
);
2289 reiserfs_info(p_s_sb
, "checking transaction log (%s)\n",
2290 bdevname(journal
->j_dev_bd
, b
));
2291 start
= get_seconds();
2293 /* step 1, read in the journal header block. Check the transaction it says
2294 ** is the first unflushed, and if that transaction is not valid,
2297 journal
->j_header_bh
= journal_bread(p_s_sb
,
2298 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
)
2299 + SB_ONDISK_JOURNAL_SIZE(p_s_sb
));
2300 if (!journal
->j_header_bh
) {
2303 jh
= (struct reiserfs_journal_header
*)(journal
->j_header_bh
->b_data
);
2304 if (le32_to_cpu(jh
->j_first_unflushed_offset
) >= 0 &&
2305 le32_to_cpu(jh
->j_first_unflushed_offset
) <
2306 SB_ONDISK_JOURNAL_SIZE(p_s_sb
)
2307 && le32_to_cpu(jh
->j_last_flush_trans_id
) > 0) {
2309 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
2310 le32_to_cpu(jh
->j_first_unflushed_offset
);
2311 oldest_trans_id
= le32_to_cpu(jh
->j_last_flush_trans_id
) + 1;
2312 newest_mount_id
= le32_to_cpu(jh
->j_mount_id
);
2313 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2314 "journal-1153: found in "
2315 "header: first_unflushed_offset %d, last_flushed_trans_id "
2316 "%lu", le32_to_cpu(jh
->j_first_unflushed_offset
),
2317 le32_to_cpu(jh
->j_last_flush_trans_id
));
2318 valid_journal_header
= 1;
2320 /* now, we try to read the first unflushed offset. If it is not valid,
2321 ** there is nothing more we can do, and it makes no sense to read
2322 ** through the whole log.
2325 journal_bread(p_s_sb
,
2326 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
2327 le32_to_cpu(jh
->j_first_unflushed_offset
));
2328 ret
= journal_transaction_is_valid(p_s_sb
, d_bh
, NULL
, NULL
);
2330 continue_replay
= 0;
2333 goto start_log_replay
;
2336 if (continue_replay
&& bdev_read_only(p_s_sb
->s_bdev
)) {
2337 reiserfs_warning(p_s_sb
,
2338 "clm-2076: device is readonly, unable to replay log");
2342 /* ok, there are transactions that need to be replayed. start with the first log block, find
2343 ** all the valid transactions, and pick out the oldest.
2345 while (continue_replay
2347 (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
2348 SB_ONDISK_JOURNAL_SIZE(p_s_sb
))) {
2349 /* Note that it is required for blocksize of primary fs device and journal
2350 device to be the same */
2352 reiserfs_breada(journal
->j_dev_bd
, cur_dblock
,
2353 p_s_sb
->s_blocksize
,
2354 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
2355 SB_ONDISK_JOURNAL_SIZE(p_s_sb
));
2357 journal_transaction_is_valid(p_s_sb
, d_bh
,
2358 &oldest_invalid_trans_id
,
2361 desc
= (struct reiserfs_journal_desc
*)d_bh
->b_data
;
2362 if (oldest_start
== 0) { /* init all oldest_ values */
2363 oldest_trans_id
= get_desc_trans_id(desc
);
2364 oldest_start
= d_bh
->b_blocknr
;
2365 newest_mount_id
= get_desc_mount_id(desc
);
2366 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2367 "journal-1179: Setting "
2368 "oldest_start to offset %llu, trans_id %lu",
2370 SB_ONDISK_JOURNAL_1st_BLOCK
2371 (p_s_sb
), oldest_trans_id
);
2372 } else if (oldest_trans_id
> get_desc_trans_id(desc
)) {
2373 /* one we just read was older */
2374 oldest_trans_id
= get_desc_trans_id(desc
);
2375 oldest_start
= d_bh
->b_blocknr
;
2376 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2377 "journal-1180: Resetting "
2378 "oldest_start to offset %lu, trans_id %lu",
2380 SB_ONDISK_JOURNAL_1st_BLOCK
2381 (p_s_sb
), oldest_trans_id
);
2383 if (newest_mount_id
< get_desc_mount_id(desc
)) {
2384 newest_mount_id
= get_desc_mount_id(desc
);
2385 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2386 "journal-1299: Setting "
2387 "newest_mount_id to %d",
2388 get_desc_mount_id(desc
));
2390 cur_dblock
+= get_desc_trans_len(desc
) + 2;
2398 cur_dblock
= oldest_start
;
2399 if (oldest_trans_id
) {
2400 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2401 "journal-1206: Starting replay "
2402 "from offset %llu, trans_id %lu",
2403 cur_dblock
- SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
),
2408 while (continue_replay
&& oldest_trans_id
> 0) {
2410 journal_read_transaction(p_s_sb
, cur_dblock
, oldest_start
,
2411 oldest_trans_id
, newest_mount_id
);
2414 } else if (ret
!= 0) {
2418 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) + journal
->j_start
;
2420 if (cur_dblock
== oldest_start
)
2424 if (oldest_trans_id
== 0) {
2425 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2426 "journal-1225: No valid " "transactions found");
2428 /* j_start does not get set correctly if we don't replay any transactions.
2429 ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2430 ** copy the trans_id from the header
2432 if (valid_journal_header
&& replay_count
== 0) {
2433 journal
->j_start
= le32_to_cpu(jh
->j_first_unflushed_offset
);
2434 journal
->j_trans_id
=
2435 le32_to_cpu(jh
->j_last_flush_trans_id
) + 1;
2436 journal
->j_last_flush_trans_id
=
2437 le32_to_cpu(jh
->j_last_flush_trans_id
);
2438 journal
->j_mount_id
= le32_to_cpu(jh
->j_mount_id
) + 1;
2440 journal
->j_mount_id
= newest_mount_id
+ 1;
2442 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
, "journal-1299: Setting "
2443 "newest_mount_id to %lu", journal
->j_mount_id
);
2444 journal
->j_first_unflushed_offset
= journal
->j_start
;
2445 if (replay_count
> 0) {
2446 reiserfs_info(p_s_sb
,
2447 "replayed %d transactions in %lu seconds\n",
2448 replay_count
, get_seconds() - start
);
2450 if (!bdev_read_only(p_s_sb
->s_bdev
) &&
2451 _update_journal_header_block(p_s_sb
, journal
->j_start
,
2452 journal
->j_last_flush_trans_id
)) {
2453 /* replay failed, caller must call free_journal_ram and abort
2461 static struct reiserfs_journal_list
*alloc_journal_list(struct super_block
*s
)
2463 struct reiserfs_journal_list
*jl
;
2465 jl
= reiserfs_kmalloc(sizeof(struct reiserfs_journal_list
), GFP_NOFS
,
2471 memset(jl
, 0, sizeof(*jl
));
2472 INIT_LIST_HEAD(&jl
->j_list
);
2473 INIT_LIST_HEAD(&jl
->j_working_list
);
2474 INIT_LIST_HEAD(&jl
->j_tail_bh_list
);
2475 INIT_LIST_HEAD(&jl
->j_bh_list
);
2476 sema_init(&jl
->j_commit_lock
, 1);
2477 SB_JOURNAL(s
)->j_num_lists
++;
2478 get_journal_list(jl
);
2482 static void journal_list_init(struct super_block
*p_s_sb
)
2484 SB_JOURNAL(p_s_sb
)->j_current_jl
= alloc_journal_list(p_s_sb
);
2487 static int release_journal_dev(struct super_block
*super
,
2488 struct reiserfs_journal
*journal
)
2494 if (journal
->j_dev_file
!= NULL
) {
2495 result
= filp_close(journal
->j_dev_file
, NULL
);
2496 journal
->j_dev_file
= NULL
;
2497 journal
->j_dev_bd
= NULL
;
2498 } else if (journal
->j_dev_bd
!= NULL
) {
2499 result
= blkdev_put(journal
->j_dev_bd
);
2500 journal
->j_dev_bd
= NULL
;
2504 reiserfs_warning(super
,
2505 "sh-457: release_journal_dev: Cannot release journal device: %i",
2511 static int journal_init_dev(struct super_block
*super
,
2512 struct reiserfs_journal
*journal
,
2513 const char *jdev_name
)
2517 int blkdev_mode
= FMODE_READ
| FMODE_WRITE
;
2518 char b
[BDEVNAME_SIZE
];
2522 journal
->j_dev_bd
= NULL
;
2523 journal
->j_dev_file
= NULL
;
2524 jdev
= SB_ONDISK_JOURNAL_DEVICE(super
) ?
2525 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super
)) : super
->s_dev
;
2527 if (bdev_read_only(super
->s_bdev
))
2528 blkdev_mode
= FMODE_READ
;
2530 /* there is no "jdev" option and journal is on separate device */
2531 if ((!jdev_name
|| !jdev_name
[0])) {
2532 journal
->j_dev_bd
= open_by_devnum(jdev
, blkdev_mode
);
2533 if (IS_ERR(journal
->j_dev_bd
)) {
2534 result
= PTR_ERR(journal
->j_dev_bd
);
2535 journal
->j_dev_bd
= NULL
;
2536 reiserfs_warning(super
, "sh-458: journal_init_dev: "
2537 "cannot init journal device '%s': %i",
2538 __bdevname(jdev
, b
), result
);
2540 } else if (jdev
!= super
->s_dev
)
2541 set_blocksize(journal
->j_dev_bd
, super
->s_blocksize
);
2545 journal
->j_dev_file
= filp_open(jdev_name
, 0, 0);
2546 if (!IS_ERR(journal
->j_dev_file
)) {
2547 struct inode
*jdev_inode
= journal
->j_dev_file
->f_mapping
->host
;
2548 if (!S_ISBLK(jdev_inode
->i_mode
)) {
2549 reiserfs_warning(super
, "journal_init_dev: '%s' is "
2550 "not a block device", jdev_name
);
2552 release_journal_dev(super
, journal
);
2555 journal
->j_dev_bd
= I_BDEV(jdev_inode
);
2556 set_blocksize(journal
->j_dev_bd
, super
->s_blocksize
);
2557 reiserfs_info(super
,
2558 "journal_init_dev: journal device: %s\n",
2559 bdevname(journal
->j_dev_bd
, b
));
2562 result
= PTR_ERR(journal
->j_dev_file
);
2563 journal
->j_dev_file
= NULL
;
2564 reiserfs_warning(super
,
2565 "journal_init_dev: Cannot open '%s': %i",
2572 ** must be called once on fs mount. calls journal_read for you
2574 int journal_init(struct super_block
*p_s_sb
, const char *j_dev_name
,
2575 int old_format
, unsigned int commit_max_age
)
2577 int num_cnodes
= SB_ONDISK_JOURNAL_SIZE(p_s_sb
) * 2;
2578 struct buffer_head
*bhjh
;
2579 struct reiserfs_super_block
*rs
;
2580 struct reiserfs_journal_header
*jh
;
2581 struct reiserfs_journal
*journal
;
2582 struct reiserfs_journal_list
*jl
;
2583 char b
[BDEVNAME_SIZE
];
2585 journal
= SB_JOURNAL(p_s_sb
) = vmalloc(sizeof(struct reiserfs_journal
));
2587 reiserfs_warning(p_s_sb
,
2588 "journal-1256: unable to get memory for journal structure");
2591 memset(journal
, 0, sizeof(struct reiserfs_journal
));
2592 INIT_LIST_HEAD(&journal
->j_bitmap_nodes
);
2593 INIT_LIST_HEAD(&journal
->j_prealloc_list
);
2594 INIT_LIST_HEAD(&journal
->j_working_list
);
2595 INIT_LIST_HEAD(&journal
->j_journal_list
);
2596 journal
->j_persistent_trans
= 0;
2597 if (reiserfs_allocate_list_bitmaps(p_s_sb
,
2598 journal
->j_list_bitmap
,
2599 SB_BMAP_NR(p_s_sb
)))
2600 goto free_and_return
;
2601 allocate_bitmap_nodes(p_s_sb
);
2603 /* reserved for journal area support */
2604 SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb
) = (old_format
?
2605 REISERFS_OLD_DISK_OFFSET_IN_BYTES
2606 / p_s_sb
->s_blocksize
+
2607 SB_BMAP_NR(p_s_sb
) +
2609 REISERFS_DISK_OFFSET_IN_BYTES
/
2610 p_s_sb
->s_blocksize
+ 2);
2612 /* Sanity check to see is the standard journal fitting withing first bitmap
2613 (actual for small blocksizes) */
2614 if (!SB_ONDISK_JOURNAL_DEVICE(p_s_sb
) &&
2615 (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb
) +
2616 SB_ONDISK_JOURNAL_SIZE(p_s_sb
) > p_s_sb
->s_blocksize
* 8)) {
2617 reiserfs_warning(p_s_sb
,
2618 "journal-1393: journal does not fit for area "
2619 "addressed by first of bitmap blocks. It starts at "
2620 "%u and its size is %u. Block size %ld",
2621 SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb
),
2622 SB_ONDISK_JOURNAL_SIZE(p_s_sb
),
2623 p_s_sb
->s_blocksize
);
2624 goto free_and_return
;
2627 if (journal_init_dev(p_s_sb
, journal
, j_dev_name
) != 0) {
2628 reiserfs_warning(p_s_sb
,
2629 "sh-462: unable to initialize jornal device");
2630 goto free_and_return
;
2633 rs
= SB_DISK_SUPER_BLOCK(p_s_sb
);
2635 /* read journal header */
2636 bhjh
= journal_bread(p_s_sb
,
2637 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
2638 SB_ONDISK_JOURNAL_SIZE(p_s_sb
));
2640 reiserfs_warning(p_s_sb
,
2641 "sh-459: unable to read journal header");
2642 goto free_and_return
;
2644 jh
= (struct reiserfs_journal_header
*)(bhjh
->b_data
);
2646 /* make sure that journal matches to the super block */
2647 if (is_reiserfs_jr(rs
)
2648 && (le32_to_cpu(jh
->jh_journal
.jp_journal_magic
) !=
2649 sb_jp_journal_magic(rs
))) {
2650 reiserfs_warning(p_s_sb
,
2651 "sh-460: journal header magic %x "
2652 "(device %s) does not match to magic found in super "
2653 "block %x", jh
->jh_journal
.jp_journal_magic
,
2654 bdevname(journal
->j_dev_bd
, b
),
2655 sb_jp_journal_magic(rs
));
2657 goto free_and_return
;
2660 journal
->j_trans_max
= le32_to_cpu(jh
->jh_journal
.jp_journal_trans_max
);
2661 journal
->j_max_batch
= le32_to_cpu(jh
->jh_journal
.jp_journal_max_batch
);
2662 journal
->j_max_commit_age
=
2663 le32_to_cpu(jh
->jh_journal
.jp_journal_max_commit_age
);
2664 journal
->j_max_trans_age
= JOURNAL_MAX_TRANS_AGE
;
2666 if (journal
->j_trans_max
) {
2667 /* make sure these parameters are available, assign it if they are not */
2668 __u32 initial
= journal
->j_trans_max
;
2671 if (p_s_sb
->s_blocksize
< 4096)
2672 ratio
= 4096 / p_s_sb
->s_blocksize
;
2674 if (SB_ONDISK_JOURNAL_SIZE(p_s_sb
) / journal
->j_trans_max
<
2676 journal
->j_trans_max
=
2677 SB_ONDISK_JOURNAL_SIZE(p_s_sb
) / JOURNAL_MIN_RATIO
;
2678 if (journal
->j_trans_max
> JOURNAL_TRANS_MAX_DEFAULT
/ ratio
)
2679 journal
->j_trans_max
=
2680 JOURNAL_TRANS_MAX_DEFAULT
/ ratio
;
2681 if (journal
->j_trans_max
< JOURNAL_TRANS_MIN_DEFAULT
/ ratio
)
2682 journal
->j_trans_max
=
2683 JOURNAL_TRANS_MIN_DEFAULT
/ ratio
;
2685 if (journal
->j_trans_max
!= initial
)
2686 reiserfs_warning(p_s_sb
,
2687 "sh-461: journal_init: wrong transaction max size (%u). Changed to %u",
2688 initial
, journal
->j_trans_max
);
2690 journal
->j_max_batch
= journal
->j_trans_max
*
2691 JOURNAL_MAX_BATCH_DEFAULT
/ JOURNAL_TRANS_MAX_DEFAULT
;
2694 if (!journal
->j_trans_max
) {
2695 /*we have the file system was created by old version of mkreiserfs
2696 so this field contains zero value */
2697 journal
->j_trans_max
= JOURNAL_TRANS_MAX_DEFAULT
;
2698 journal
->j_max_batch
= JOURNAL_MAX_BATCH_DEFAULT
;
2699 journal
->j_max_commit_age
= JOURNAL_MAX_COMMIT_AGE
;
2701 /* for blocksize >= 4096 - max transaction size is 1024. For block size < 4096
2702 trans max size is decreased proportionally */
2703 if (p_s_sb
->s_blocksize
< 4096) {
2704 journal
->j_trans_max
/= (4096 / p_s_sb
->s_blocksize
);
2705 journal
->j_max_batch
= (journal
->j_trans_max
) * 9 / 10;
2709 journal
->j_default_max_commit_age
= journal
->j_max_commit_age
;
2711 if (commit_max_age
!= 0) {
2712 journal
->j_max_commit_age
= commit_max_age
;
2713 journal
->j_max_trans_age
= commit_max_age
;
2716 reiserfs_info(p_s_sb
, "journal params: device %s, size %u, "
2717 "journal first block %u, max trans len %u, max batch %u, "
2718 "max commit age %u, max trans age %u\n",
2719 bdevname(journal
->j_dev_bd
, b
),
2720 SB_ONDISK_JOURNAL_SIZE(p_s_sb
),
2721 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
),
2722 journal
->j_trans_max
,
2723 journal
->j_max_batch
,
2724 journal
->j_max_commit_age
, journal
->j_max_trans_age
);
2728 journal
->j_list_bitmap_index
= 0;
2729 journal_list_init(p_s_sb
);
2731 memset(journal
->j_list_hash_table
, 0,
2732 JOURNAL_HASH_SIZE
* sizeof(struct reiserfs_journal_cnode
*));
2734 INIT_LIST_HEAD(&journal
->j_dirty_buffers
);
2735 spin_lock_init(&journal
->j_dirty_buffers_lock
);
2737 journal
->j_start
= 0;
2739 journal
->j_len_alloc
= 0;
2740 atomic_set(&(journal
->j_wcount
), 0);
2741 atomic_set(&(journal
->j_async_throttle
), 0);
2742 journal
->j_bcount
= 0;
2743 journal
->j_trans_start_time
= 0;
2744 journal
->j_last
= NULL
;
2745 journal
->j_first
= NULL
;
2746 init_waitqueue_head(&(journal
->j_join_wait
));
2747 sema_init(&journal
->j_lock
, 1);
2748 sema_init(&journal
->j_flush_sem
, 1);
2750 journal
->j_trans_id
= 10;
2751 journal
->j_mount_id
= 10;
2752 journal
->j_state
= 0;
2753 atomic_set(&(journal
->j_jlock
), 0);
2754 journal
->j_cnode_free_list
= allocate_cnodes(num_cnodes
);
2755 journal
->j_cnode_free_orig
= journal
->j_cnode_free_list
;
2756 journal
->j_cnode_free
= journal
->j_cnode_free_list
? num_cnodes
: 0;
2757 journal
->j_cnode_used
= 0;
2758 journal
->j_must_wait
= 0;
2760 init_journal_hash(p_s_sb
);
2761 jl
= journal
->j_current_jl
;
2762 jl
->j_list_bitmap
= get_list_bitmap(p_s_sb
, jl
);
2763 if (!jl
->j_list_bitmap
) {
2764 reiserfs_warning(p_s_sb
,
2765 "journal-2005, get_list_bitmap failed for journal list 0");
2766 goto free_and_return
;
2768 if (journal_read(p_s_sb
) < 0) {
2769 reiserfs_warning(p_s_sb
, "Replay Failure, unable to mount");
2770 goto free_and_return
;
2773 reiserfs_mounted_fs_count
++;
2774 if (reiserfs_mounted_fs_count
<= 1)
2775 commit_wq
= create_workqueue("reiserfs");
2777 INIT_WORK(&journal
->j_work
, flush_async_commits
, p_s_sb
);
2780 free_journal_ram(p_s_sb
);
2785 ** test for a polite end of the current transaction. Used by file_write, and should
2786 ** be used by delete to make sure they don't write more than can fit inside a single
2789 int journal_transaction_should_end(struct reiserfs_transaction_handle
*th
,
2792 struct reiserfs_journal
*journal
= SB_JOURNAL(th
->t_super
);
2793 time_t now
= get_seconds();
2794 /* cannot restart while nested */
2795 BUG_ON(!th
->t_trans_id
);
2796 if (th
->t_refcount
> 1)
2798 if (journal
->j_must_wait
> 0 ||
2799 (journal
->j_len_alloc
+ new_alloc
) >= journal
->j_max_batch
||
2800 atomic_read(&(journal
->j_jlock
)) ||
2801 (now
- journal
->j_trans_start_time
) > journal
->j_max_trans_age
||
2802 journal
->j_cnode_free
< (journal
->j_trans_max
* 3)) {
2808 /* this must be called inside a transaction, and requires the
2809 ** kernel_lock to be held
2811 void reiserfs_block_writes(struct reiserfs_transaction_handle
*th
)
2813 struct reiserfs_journal
*journal
= SB_JOURNAL(th
->t_super
);
2814 BUG_ON(!th
->t_trans_id
);
2815 journal
->j_must_wait
= 1;
2816 set_bit(J_WRITERS_BLOCKED
, &journal
->j_state
);
2820 /* this must be called without a transaction started, and does not
2823 void reiserfs_allow_writes(struct super_block
*s
)
2825 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2826 clear_bit(J_WRITERS_BLOCKED
, &journal
->j_state
);
2827 wake_up(&journal
->j_join_wait
);
2830 /* this must be called without a transaction started, and does not
2833 void reiserfs_wait_on_write_block(struct super_block
*s
)
2835 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2836 wait_event(journal
->j_join_wait
,
2837 !test_bit(J_WRITERS_BLOCKED
, &journal
->j_state
));
2840 static void queue_log_writer(struct super_block
*s
)
2843 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2844 set_bit(J_WRITERS_QUEUED
, &journal
->j_state
);
2847 * we don't want to use wait_event here because
2848 * we only want to wait once.
2850 init_waitqueue_entry(&wait
, current
);
2851 add_wait_queue(&journal
->j_join_wait
, &wait
);
2852 set_current_state(TASK_UNINTERRUPTIBLE
);
2853 if (test_bit(J_WRITERS_QUEUED
, &journal
->j_state
))
2855 current
->state
= TASK_RUNNING
;
2856 remove_wait_queue(&journal
->j_join_wait
, &wait
);
2859 static void wake_queued_writers(struct super_block
*s
)
2861 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2862 if (test_and_clear_bit(J_WRITERS_QUEUED
, &journal
->j_state
))
2863 wake_up(&journal
->j_join_wait
);
2866 static void let_transaction_grow(struct super_block
*sb
, unsigned long trans_id
)
2868 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
2869 unsigned long bcount
= journal
->j_bcount
;
2871 schedule_timeout_uninterruptible(1);
2872 journal
->j_current_jl
->j_state
|= LIST_COMMIT_PENDING
;
2873 while ((atomic_read(&journal
->j_wcount
) > 0 ||
2874 atomic_read(&journal
->j_jlock
)) &&
2875 journal
->j_trans_id
== trans_id
) {
2876 queue_log_writer(sb
);
2878 if (journal
->j_trans_id
!= trans_id
)
2880 if (bcount
== journal
->j_bcount
)
2882 bcount
= journal
->j_bcount
;
2886 /* join == true if you must join an existing transaction.
2887 ** join == false if you can deal with waiting for others to finish
2889 ** this will block until the transaction is joinable. send the number of blocks you
2890 ** expect to use in nblocks.
2892 static int do_journal_begin_r(struct reiserfs_transaction_handle
*th
,
2893 struct super_block
*p_s_sb
, unsigned long nblocks
,
2896 time_t now
= get_seconds();
2898 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
2899 struct reiserfs_transaction_handle myth
;
2900 int sched_count
= 0;
2903 reiserfs_check_lock_depth(p_s_sb
, "journal_begin");
2904 if (nblocks
> journal
->j_trans_max
)
2907 PROC_INFO_INC(p_s_sb
, journal
.journal_being
);
2908 /* set here for journal_join */
2910 th
->t_super
= p_s_sb
;
2913 lock_journal(p_s_sb
);
2914 if (join
!= JBEGIN_ABORT
&& reiserfs_is_journal_aborted(journal
)) {
2915 unlock_journal(p_s_sb
);
2916 retval
= journal
->j_errno
;
2919 journal
->j_bcount
++;
2921 if (test_bit(J_WRITERS_BLOCKED
, &journal
->j_state
)) {
2922 unlock_journal(p_s_sb
);
2923 reiserfs_wait_on_write_block(p_s_sb
);
2924 PROC_INFO_INC(p_s_sb
, journal
.journal_relock_writers
);
2927 now
= get_seconds();
2929 /* if there is no room in the journal OR
2930 ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
2931 ** we don't sleep if there aren't other writers
2934 if ((!join
&& journal
->j_must_wait
> 0) ||
2936 && (journal
->j_len_alloc
+ nblocks
+ 2) >= journal
->j_max_batch
)
2937 || (!join
&& atomic_read(&journal
->j_wcount
) > 0
2938 && journal
->j_trans_start_time
> 0
2939 && (now
- journal
->j_trans_start_time
) >
2940 journal
->j_max_trans_age
) || (!join
2941 && atomic_read(&journal
->j_jlock
))
2942 || (!join
&& journal
->j_cnode_free
< (journal
->j_trans_max
* 3))) {
2944 old_trans_id
= journal
->j_trans_id
;
2945 unlock_journal(p_s_sb
); /* allow others to finish this transaction */
2947 if (!join
&& (journal
->j_len_alloc
+ nblocks
+ 2) >=
2948 journal
->j_max_batch
&&
2949 ((journal
->j_len
+ nblocks
+ 2) * 100) <
2950 (journal
->j_len_alloc
* 75)) {
2951 if (atomic_read(&journal
->j_wcount
) > 10) {
2953 queue_log_writer(p_s_sb
);
2957 /* don't mess with joining the transaction if all we have to do is
2958 * wait for someone else to do a commit
2960 if (atomic_read(&journal
->j_jlock
)) {
2961 while (journal
->j_trans_id
== old_trans_id
&&
2962 atomic_read(&journal
->j_jlock
)) {
2963 queue_log_writer(p_s_sb
);
2967 retval
= journal_join(&myth
, p_s_sb
, 1);
2971 /* someone might have ended the transaction while we joined */
2972 if (old_trans_id
!= journal
->j_trans_id
) {
2973 retval
= do_journal_end(&myth
, p_s_sb
, 1, 0);
2975 retval
= do_journal_end(&myth
, p_s_sb
, 1, COMMIT_NOW
);
2981 PROC_INFO_INC(p_s_sb
, journal
.journal_relock_wcount
);
2984 /* we are the first writer, set trans_id */
2985 if (journal
->j_trans_start_time
== 0) {
2986 journal
->j_trans_start_time
= get_seconds();
2988 atomic_inc(&(journal
->j_wcount
));
2989 journal
->j_len_alloc
+= nblocks
;
2990 th
->t_blocks_logged
= 0;
2991 th
->t_blocks_allocated
= nblocks
;
2992 th
->t_trans_id
= journal
->j_trans_id
;
2993 unlock_journal(p_s_sb
);
2994 INIT_LIST_HEAD(&th
->t_list
);
2999 memset(th
, 0, sizeof(*th
));
3000 /* Re-set th->t_super, so we can properly keep track of how many
3001 * persistent transactions there are. We need to do this so if this
3002 * call is part of a failed restart_transaction, we can free it later */
3003 th
->t_super
= p_s_sb
;
3007 struct reiserfs_transaction_handle
*reiserfs_persistent_transaction(struct
3013 struct reiserfs_transaction_handle
*th
;
3015 /* if we're nesting into an existing transaction. It will be
3016 ** persistent on its own
3018 if (reiserfs_transaction_running(s
)) {
3019 th
= current
->journal_info
;
3021 if (th
->t_refcount
< 2) {
3026 th
= reiserfs_kmalloc(sizeof(struct reiserfs_transaction_handle
),
3030 ret
= journal_begin(th
, s
, nblocks
);
3032 reiserfs_kfree(th
, sizeof(struct reiserfs_transaction_handle
),
3037 SB_JOURNAL(s
)->j_persistent_trans
++;
3041 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle
*th
)
3043 struct super_block
*s
= th
->t_super
;
3046 ret
= journal_end(th
, th
->t_super
, th
->t_blocks_allocated
);
3049 if (th
->t_refcount
== 0) {
3050 SB_JOURNAL(s
)->j_persistent_trans
--;
3051 reiserfs_kfree(th
, sizeof(struct reiserfs_transaction_handle
),
3057 static int journal_join(struct reiserfs_transaction_handle
*th
,
3058 struct super_block
*p_s_sb
, unsigned long nblocks
)
3060 struct reiserfs_transaction_handle
*cur_th
= current
->journal_info
;
3062 /* this keeps do_journal_end from NULLing out the current->journal_info
3065 th
->t_handle_save
= cur_th
;
3066 if (cur_th
&& cur_th
->t_refcount
> 1) {
3069 return do_journal_begin_r(th
, p_s_sb
, nblocks
, JBEGIN_JOIN
);
3072 int journal_join_abort(struct reiserfs_transaction_handle
*th
,
3073 struct super_block
*p_s_sb
, unsigned long nblocks
)
3075 struct reiserfs_transaction_handle
*cur_th
= current
->journal_info
;
3077 /* this keeps do_journal_end from NULLing out the current->journal_info
3080 th
->t_handle_save
= cur_th
;
3081 if (cur_th
&& cur_th
->t_refcount
> 1) {
3084 return do_journal_begin_r(th
, p_s_sb
, nblocks
, JBEGIN_ABORT
);
3087 int journal_begin(struct reiserfs_transaction_handle
*th
,
3088 struct super_block
*p_s_sb
, unsigned long nblocks
)
3090 struct reiserfs_transaction_handle
*cur_th
= current
->journal_info
;
3093 th
->t_handle_save
= NULL
;
3095 /* we are nesting into the current transaction */
3096 if (cur_th
->t_super
== p_s_sb
) {
3097 BUG_ON(!cur_th
->t_refcount
);
3098 cur_th
->t_refcount
++;
3099 memcpy(th
, cur_th
, sizeof(*th
));
3100 if (th
->t_refcount
<= 1)
3101 reiserfs_warning(p_s_sb
,
3102 "BAD: refcount <= 1, but journal_info != 0");
3105 /* we've ended up with a handle from a different filesystem.
3106 ** save it and restore on journal_end. This should never
3109 reiserfs_warning(p_s_sb
,
3110 "clm-2100: nesting info a different FS");
3111 th
->t_handle_save
= current
->journal_info
;
3112 current
->journal_info
= th
;
3115 current
->journal_info
= th
;
3117 ret
= do_journal_begin_r(th
, p_s_sb
, nblocks
, JBEGIN_REG
);
3118 if (current
->journal_info
!= th
)
3121 /* I guess this boils down to being the reciprocal of clm-2100 above.
3122 * If do_journal_begin_r fails, we need to put it back, since journal_end
3123 * won't be called to do it. */
3125 current
->journal_info
= th
->t_handle_save
;
3127 BUG_ON(!th
->t_refcount
);
3133 ** puts bh into the current transaction. If it was already there, reorders removes the
3134 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3136 ** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the
3137 ** transaction is committed.
3139 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3141 int journal_mark_dirty(struct reiserfs_transaction_handle
*th
,
3142 struct super_block
*p_s_sb
, struct buffer_head
*bh
)
3144 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
3145 struct reiserfs_journal_cnode
*cn
= NULL
;
3146 int count_already_incd
= 0;
3148 BUG_ON(!th
->t_trans_id
);
3150 PROC_INFO_INC(p_s_sb
, journal
.mark_dirty
);
3151 if (th
->t_trans_id
!= journal
->j_trans_id
) {
3152 reiserfs_panic(th
->t_super
,
3153 "journal-1577: handle trans id %ld != current trans id %ld\n",
3154 th
->t_trans_id
, journal
->j_trans_id
);
3159 prepared
= test_clear_buffer_journal_prepared(bh
);
3160 clear_buffer_journal_restore_dirty(bh
);
3161 /* already in this transaction, we are done */
3162 if (buffer_journaled(bh
)) {
3163 PROC_INFO_INC(p_s_sb
, journal
.mark_dirty_already
);
3167 /* this must be turned into a panic instead of a warning. We can't allow
3168 ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3169 ** could get to disk too early. NOT GOOD.
3171 if (!prepared
|| buffer_dirty(bh
)) {
3172 reiserfs_warning(p_s_sb
, "journal-1777: buffer %llu bad state "
3173 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3174 (unsigned long long)bh
->b_blocknr
,
3175 prepared
? ' ' : '!',
3176 buffer_locked(bh
) ? ' ' : '!',
3177 buffer_dirty(bh
) ? ' ' : '!',
3178 buffer_journal_dirty(bh
) ? ' ' : '!');
3181 if (atomic_read(&(journal
->j_wcount
)) <= 0) {
3182 reiserfs_warning(p_s_sb
,
3183 "journal-1409: journal_mark_dirty returning because j_wcount was %d",
3184 atomic_read(&(journal
->j_wcount
)));
3187 /* this error means I've screwed up, and we've overflowed the transaction.
3188 ** Nothing can be done here, except make the FS readonly or panic.
3190 if (journal
->j_len
>= journal
->j_trans_max
) {
3191 reiserfs_panic(th
->t_super
,
3192 "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n",
3196 if (buffer_journal_dirty(bh
)) {
3197 count_already_incd
= 1;
3198 PROC_INFO_INC(p_s_sb
, journal
.mark_dirty_notjournal
);
3199 clear_buffer_journal_dirty(bh
);
3202 if (journal
->j_len
> journal
->j_len_alloc
) {
3203 journal
->j_len_alloc
= journal
->j_len
+ JOURNAL_PER_BALANCE_CNT
;
3206 set_buffer_journaled(bh
);
3208 /* now put this guy on the end */
3210 cn
= get_cnode(p_s_sb
);
3212 reiserfs_panic(p_s_sb
, "get_cnode failed!\n");
3215 if (th
->t_blocks_logged
== th
->t_blocks_allocated
) {
3216 th
->t_blocks_allocated
+= JOURNAL_PER_BALANCE_CNT
;
3217 journal
->j_len_alloc
+= JOURNAL_PER_BALANCE_CNT
;
3219 th
->t_blocks_logged
++;
3223 cn
->blocknr
= bh
->b_blocknr
;
3226 insert_journal_hash(journal
->j_hash_table
, cn
);
3227 if (!count_already_incd
) {
3232 cn
->prev
= journal
->j_last
;
3234 if (journal
->j_last
) {
3235 journal
->j_last
->next
= cn
;
3236 journal
->j_last
= cn
;
3238 journal
->j_first
= cn
;
3239 journal
->j_last
= cn
;
3244 int journal_end(struct reiserfs_transaction_handle
*th
,
3245 struct super_block
*p_s_sb
, unsigned long nblocks
)
3247 if (!current
->journal_info
&& th
->t_refcount
> 1)
3248 reiserfs_warning(p_s_sb
, "REISER-NESTING: th NULL, refcount %d",
3251 if (!th
->t_trans_id
) {
3257 if (th
->t_refcount
> 0) {
3258 struct reiserfs_transaction_handle
*cur_th
=
3259 current
->journal_info
;
3261 /* we aren't allowed to close a nested transaction on a different
3262 ** filesystem from the one in the task struct
3264 if (cur_th
->t_super
!= th
->t_super
)
3268 memcpy(current
->journal_info
, th
, sizeof(*th
));
3273 return do_journal_end(th
, p_s_sb
, nblocks
, 0);
3277 /* removes from the current transaction, relsing and descrementing any counters.
3278 ** also files the removed buffer directly onto the clean list
3280 ** called by journal_mark_freed when a block has been deleted
3282 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3284 static int remove_from_transaction(struct super_block
*p_s_sb
,
3285 b_blocknr_t blocknr
, int already_cleaned
)
3287 struct buffer_head
*bh
;
3288 struct reiserfs_journal_cnode
*cn
;
3289 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
3292 cn
= get_journal_hash_dev(p_s_sb
, journal
->j_hash_table
, blocknr
);
3293 if (!cn
|| !cn
->bh
) {
3298 cn
->prev
->next
= cn
->next
;
3301 cn
->next
->prev
= cn
->prev
;
3303 if (cn
== journal
->j_first
) {
3304 journal
->j_first
= cn
->next
;
3306 if (cn
== journal
->j_last
) {
3307 journal
->j_last
= cn
->prev
;
3310 remove_journal_hash(p_s_sb
, journal
->j_hash_table
, NULL
,
3312 clear_buffer_journaled(bh
); /* don't log this one */
3314 if (!already_cleaned
) {
3315 clear_buffer_journal_dirty(bh
);
3316 clear_buffer_dirty(bh
);
3317 clear_buffer_journal_test(bh
);
3319 if (atomic_read(&(bh
->b_count
)) < 0) {
3320 reiserfs_warning(p_s_sb
,
3321 "journal-1752: remove from trans, b_count < 0");
3326 journal
->j_len_alloc
--;
3327 free_cnode(p_s_sb
, cn
);
3332 ** for any cnode in a journal list, it can only be dirtied of all the
3333 ** transactions that include it are commited to disk.
3334 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3335 ** and 0 if you aren't
3337 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3338 ** blocks for a given transaction on disk
3341 static int can_dirty(struct reiserfs_journal_cnode
*cn
)
3343 struct super_block
*sb
= cn
->sb
;
3344 b_blocknr_t blocknr
= cn
->blocknr
;
3345 struct reiserfs_journal_cnode
*cur
= cn
->hprev
;
3348 /* first test hprev. These are all newer than cn, so any node here
3349 ** with the same block number and dev means this node can't be sent
3350 ** to disk right now.
3352 while (cur
&& can_dirty
) {
3353 if (cur
->jlist
&& cur
->bh
&& cur
->blocknr
&& cur
->sb
== sb
&&
3354 cur
->blocknr
== blocknr
) {
3359 /* then test hnext. These are all older than cn. As long as they
3360 ** are committed to the log, it is safe to write cn to disk
3363 while (cur
&& can_dirty
) {
3364 if (cur
->jlist
&& cur
->jlist
->j_len
> 0 &&
3365 atomic_read(&(cur
->jlist
->j_commit_left
)) > 0 && cur
->bh
&&
3366 cur
->blocknr
&& cur
->sb
== sb
&& cur
->blocknr
== blocknr
) {
3374 /* syncs the commit blocks, but does not force the real buffers to disk
3375 ** will wait until the current transaction is done/commited before returning
3377 int journal_end_sync(struct reiserfs_transaction_handle
*th
,
3378 struct super_block
*p_s_sb
, unsigned long nblocks
)
3380 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
3382 BUG_ON(!th
->t_trans_id
);
3383 /* you can sync while nested, very, very bad */
3384 if (th
->t_refcount
> 1) {
3387 if (journal
->j_len
== 0) {
3388 reiserfs_prepare_for_journal(p_s_sb
, SB_BUFFER_WITH_SB(p_s_sb
),
3390 journal_mark_dirty(th
, p_s_sb
, SB_BUFFER_WITH_SB(p_s_sb
));
3392 return do_journal_end(th
, p_s_sb
, nblocks
, COMMIT_NOW
| WAIT
);
3396 ** writeback the pending async commits to disk
3398 static void flush_async_commits(void *p
)
3400 struct super_block
*p_s_sb
= p
;
3401 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
3402 struct reiserfs_journal_list
*jl
;
3403 struct list_head
*entry
;
3406 if (!list_empty(&journal
->j_journal_list
)) {
3407 /* last entry is the youngest, commit it and you get everything */
3408 entry
= journal
->j_journal_list
.prev
;
3409 jl
= JOURNAL_LIST_ENTRY(entry
);
3410 flush_commit_list(p_s_sb
, jl
, 1);
3414 * this is a little racey, but there's no harm in missing
3415 * the filemap_fdata_write
3417 if (!atomic_read(&journal
->j_async_throttle
)
3418 && !reiserfs_is_journal_aborted(journal
)) {
3419 atomic_inc(&journal
->j_async_throttle
);
3420 filemap_fdatawrite(p_s_sb
->s_bdev
->bd_inode
->i_mapping
);
3421 atomic_dec(&journal
->j_async_throttle
);
3426 ** flushes any old transactions to disk
3427 ** ends the current transaction if it is too old
3429 int reiserfs_flush_old_commits(struct super_block
*p_s_sb
)
3432 struct reiserfs_transaction_handle th
;
3433 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
3435 now
= get_seconds();
3436 /* safety check so we don't flush while we are replaying the log during
3439 if (list_empty(&journal
->j_journal_list
)) {
3443 /* check the current transaction. If there are no writers, and it is
3444 * too old, finish it, and force the commit blocks to disk
3446 if (atomic_read(&journal
->j_wcount
) <= 0 &&
3447 journal
->j_trans_start_time
> 0 &&
3448 journal
->j_len
> 0 &&
3449 (now
- journal
->j_trans_start_time
) > journal
->j_max_trans_age
) {
3450 if (!journal_join(&th
, p_s_sb
, 1)) {
3451 reiserfs_prepare_for_journal(p_s_sb
,
3452 SB_BUFFER_WITH_SB(p_s_sb
),
3454 journal_mark_dirty(&th
, p_s_sb
,
3455 SB_BUFFER_WITH_SB(p_s_sb
));
3457 /* we're only being called from kreiserfsd, it makes no sense to do
3458 ** an async commit so that kreiserfsd can do it later
3460 do_journal_end(&th
, p_s_sb
, 1, COMMIT_NOW
| WAIT
);
3463 return p_s_sb
->s_dirt
;
3467 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3469 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3470 ** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just
3471 ** flushes the commit list and returns 0.
3473 ** Won't batch when flush or commit_now is set. Also won't batch when others are waiting on j_join_wait.
3475 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3477 static int check_journal_end(struct reiserfs_transaction_handle
*th
,
3478 struct super_block
*p_s_sb
, unsigned long nblocks
,
3483 int flush
= flags
& FLUSH_ALL
;
3484 int commit_now
= flags
& COMMIT_NOW
;
3485 int wait_on_commit
= flags
& WAIT
;
3486 struct reiserfs_journal_list
*jl
;
3487 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
3489 BUG_ON(!th
->t_trans_id
);
3491 if (th
->t_trans_id
!= journal
->j_trans_id
) {
3492 reiserfs_panic(th
->t_super
,
3493 "journal-1577: handle trans id %ld != current trans id %ld\n",
3494 th
->t_trans_id
, journal
->j_trans_id
);
3497 journal
->j_len_alloc
-= (th
->t_blocks_allocated
- th
->t_blocks_logged
);
3498 if (atomic_read(&(journal
->j_wcount
)) > 0) { /* <= 0 is allowed. unmounting might not call begin */
3499 atomic_dec(&(journal
->j_wcount
));
3502 /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3503 ** will be dealt with by next transaction that actually writes something, but should be taken
3504 ** care of in this trans
3506 if (journal
->j_len
== 0) {
3509 /* if wcount > 0, and we are called to with flush or commit_now,
3510 ** we wait on j_join_wait. We will wake up when the last writer has
3511 ** finished the transaction, and started it on its way to the disk.
3512 ** Then, we flush the commit or journal list, and just return 0
3513 ** because the rest of journal end was already done for this transaction.
3515 if (atomic_read(&(journal
->j_wcount
)) > 0) {
3516 if (flush
|| commit_now
) {
3519 jl
= journal
->j_current_jl
;
3520 trans_id
= jl
->j_trans_id
;
3522 jl
->j_state
|= LIST_COMMIT_PENDING
;
3523 atomic_set(&(journal
->j_jlock
), 1);
3525 journal
->j_next_full_flush
= 1;
3527 unlock_journal(p_s_sb
);
3529 /* sleep while the current transaction is still j_jlocked */
3530 while (journal
->j_trans_id
== trans_id
) {
3531 if (atomic_read(&journal
->j_jlock
)) {
3532 queue_log_writer(p_s_sb
);
3534 lock_journal(p_s_sb
);
3535 if (journal
->j_trans_id
== trans_id
) {
3536 atomic_set(&(journal
->j_jlock
),
3539 unlock_journal(p_s_sb
);
3542 if (journal
->j_trans_id
== trans_id
) {
3546 && journal_list_still_alive(p_s_sb
, trans_id
)
3547 && wait_on_commit
) {
3548 flush_commit_list(p_s_sb
, jl
, 1);
3552 unlock_journal(p_s_sb
);
3556 /* deal with old transactions where we are the last writers */
3557 now
= get_seconds();
3558 if ((now
- journal
->j_trans_start_time
) > journal
->j_max_trans_age
) {
3560 journal
->j_next_async_flush
= 1;
3562 /* don't batch when someone is waiting on j_join_wait */
3563 /* don't batch when syncing the commit or flushing the whole trans */
3564 if (!(journal
->j_must_wait
> 0) && !(atomic_read(&(journal
->j_jlock
)))
3565 && !flush
&& !commit_now
&& (journal
->j_len
< journal
->j_max_batch
)
3566 && journal
->j_len_alloc
< journal
->j_max_batch
3567 && journal
->j_cnode_free
> (journal
->j_trans_max
* 3)) {
3568 journal
->j_bcount
++;
3569 unlock_journal(p_s_sb
);
3573 if (journal
->j_start
> SB_ONDISK_JOURNAL_SIZE(p_s_sb
)) {
3574 reiserfs_panic(p_s_sb
,
3575 "journal-003: journal_end: j_start (%ld) is too high\n",
3582 ** Does all the work that makes deleting blocks safe.
3583 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3586 ** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes
3587 ** before this transaction has finished.
3589 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with
3590 ** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash,
3591 ** the block can't be reallocated yet.
3593 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3595 int journal_mark_freed(struct reiserfs_transaction_handle
*th
,
3596 struct super_block
*p_s_sb
, b_blocknr_t blocknr
)
3598 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
3599 struct reiserfs_journal_cnode
*cn
= NULL
;
3600 struct buffer_head
*bh
= NULL
;
3601 struct reiserfs_list_bitmap
*jb
= NULL
;
3603 BUG_ON(!th
->t_trans_id
);
3605 cn
= get_journal_hash_dev(p_s_sb
, journal
->j_hash_table
, blocknr
);
3610 /* if it is journal new, we just remove it from this transaction */
3611 if (bh
&& buffer_journal_new(bh
)) {
3612 clear_buffer_journal_new(bh
);
3613 clear_prepared_bits(bh
);
3614 reiserfs_clean_and_file_buffer(bh
);
3615 cleaned
= remove_from_transaction(p_s_sb
, blocknr
, cleaned
);
3617 /* set the bit for this block in the journal bitmap for this transaction */
3618 jb
= journal
->j_current_jl
->j_list_bitmap
;
3620 reiserfs_panic(p_s_sb
,
3621 "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n");
3623 set_bit_in_list_bitmap(p_s_sb
, blocknr
, jb
);
3625 /* Note, the entire while loop is not allowed to schedule. */
3628 clear_prepared_bits(bh
);
3629 reiserfs_clean_and_file_buffer(bh
);
3631 cleaned
= remove_from_transaction(p_s_sb
, blocknr
, cleaned
);
3633 /* find all older transactions with this block, make sure they don't try to write it out */
3634 cn
= get_journal_hash_dev(p_s_sb
, journal
->j_list_hash_table
,
3637 if (p_s_sb
== cn
->sb
&& blocknr
== cn
->blocknr
) {
3638 set_bit(BLOCK_FREED
, &cn
->state
);
3641 /* remove_from_transaction will brelse the buffer if it was
3642 ** in the current trans
3644 clear_buffer_journal_dirty(cn
->
3646 clear_buffer_dirty(cn
->bh
);
3647 clear_buffer_journal_test(cn
->
3652 (&(cn
->bh
->b_count
)) < 0) {
3653 reiserfs_warning(p_s_sb
,
3654 "journal-2138: cn->bh->b_count < 0");
3657 if (cn
->jlist
) { /* since we are clearing the bh, we MUST dec nonzerolen */
3670 put_bh(bh
); /* get_hash grabs the buffer */
3671 if (atomic_read(&(bh
->b_count
)) < 0) {
3672 reiserfs_warning(p_s_sb
,
3673 "journal-2165: bh->b_count < 0");
3679 void reiserfs_update_inode_transaction(struct inode
*inode
)
3681 struct reiserfs_journal
*journal
= SB_JOURNAL(inode
->i_sb
);
3682 REISERFS_I(inode
)->i_jl
= journal
->j_current_jl
;
3683 REISERFS_I(inode
)->i_trans_id
= journal
->j_trans_id
;
3687 * returns -1 on error, 0 if no commits/barriers were done and 1
3688 * if a transaction was actually committed and the barrier was done
3690 static int __commit_trans_jl(struct inode
*inode
, unsigned long id
,
3691 struct reiserfs_journal_list
*jl
)
3693 struct reiserfs_transaction_handle th
;
3694 struct super_block
*sb
= inode
->i_sb
;
3695 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3698 /* is it from the current transaction, or from an unknown transaction? */
3699 if (id
== journal
->j_trans_id
) {
3700 jl
= journal
->j_current_jl
;
3701 /* try to let other writers come in and grow this transaction */
3702 let_transaction_grow(sb
, id
);
3703 if (journal
->j_trans_id
!= id
) {
3704 goto flush_commit_only
;
3707 ret
= journal_begin(&th
, sb
, 1);
3711 /* someone might have ended this transaction while we joined */
3712 if (journal
->j_trans_id
!= id
) {
3713 reiserfs_prepare_for_journal(sb
, SB_BUFFER_WITH_SB(sb
),
3715 journal_mark_dirty(&th
, sb
, SB_BUFFER_WITH_SB(sb
));
3716 ret
= journal_end(&th
, sb
, 1);
3717 goto flush_commit_only
;
3720 ret
= journal_end_sync(&th
, sb
, 1);
3725 /* this gets tricky, we have to make sure the journal list in
3726 * the inode still exists. We know the list is still around
3727 * if we've got a larger transaction id than the oldest list
3730 if (journal_list_still_alive(inode
->i_sb
, id
)) {
3732 * we only set ret to 1 when we know for sure
3733 * the barrier hasn't been started yet on the commit
3736 if (atomic_read(&jl
->j_commit_left
) > 1)
3738 flush_commit_list(sb
, jl
, 1);
3739 if (journal
->j_errno
)
3740 ret
= journal
->j_errno
;
3743 /* otherwise the list is gone, and long since committed */
3747 int reiserfs_commit_for_inode(struct inode
*inode
)
3749 unsigned long id
= REISERFS_I(inode
)->i_trans_id
;
3750 struct reiserfs_journal_list
*jl
= REISERFS_I(inode
)->i_jl
;
3752 /* for the whole inode, assume unset id means it was
3753 * changed in the current transaction. More conservative
3756 reiserfs_update_inode_transaction(inode
);
3757 id
= REISERFS_I(inode
)->i_trans_id
;
3758 /* jl will be updated in __commit_trans_jl */
3761 return __commit_trans_jl(inode
, id
, jl
);
3764 void reiserfs_restore_prepared_buffer(struct super_block
*p_s_sb
,
3765 struct buffer_head
*bh
)
3767 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
3768 PROC_INFO_INC(p_s_sb
, journal
.restore_prepared
);
3772 if (test_clear_buffer_journal_restore_dirty(bh
) &&
3773 buffer_journal_dirty(bh
)) {
3774 struct reiserfs_journal_cnode
*cn
;
3775 cn
= get_journal_hash_dev(p_s_sb
,
3776 journal
->j_list_hash_table
,
3778 if (cn
&& can_dirty(cn
)) {
3779 set_buffer_journal_test(bh
);
3780 mark_buffer_dirty(bh
);
3783 clear_buffer_journal_prepared(bh
);
3786 extern struct tree_balance
*cur_tb
;
3788 ** before we can change a metadata block, we have to make sure it won't
3789 ** be written to disk while we are altering it. So, we must:
3794 int reiserfs_prepare_for_journal(struct super_block
*p_s_sb
,
3795 struct buffer_head
*bh
, int wait
)
3797 PROC_INFO_INC(p_s_sb
, journal
.prepare
);
3799 if (test_set_buffer_locked(bh
)) {
3804 set_buffer_journal_prepared(bh
);
3805 if (test_clear_buffer_dirty(bh
) && buffer_journal_dirty(bh
)) {
3806 clear_buffer_journal_test(bh
);
3807 set_buffer_journal_restore_dirty(bh
);
3813 static void flush_old_journal_lists(struct super_block
*s
)
3815 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
3816 struct reiserfs_journal_list
*jl
;
3817 struct list_head
*entry
;
3818 time_t now
= get_seconds();
3820 while (!list_empty(&journal
->j_journal_list
)) {
3821 entry
= journal
->j_journal_list
.next
;
3822 jl
= JOURNAL_LIST_ENTRY(entry
);
3823 /* this check should always be run, to send old lists to disk */
3824 if (jl
->j_timestamp
< (now
- (JOURNAL_MAX_TRANS_AGE
* 4))) {
3825 flush_used_journal_lists(s
, jl
);
3833 ** long and ugly. If flush, will not return until all commit
3834 ** blocks and all real buffers in the trans are on disk.
3835 ** If no_async, won't return until all commit blocks are on disk.
3837 ** keep reading, there are comments as you go along
3839 ** If the journal is aborted, we just clean up. Things like flushing
3840 ** journal lists, etc just won't happen.
3842 static int do_journal_end(struct reiserfs_transaction_handle
*th
,
3843 struct super_block
*p_s_sb
, unsigned long nblocks
,
3846 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
3847 struct reiserfs_journal_cnode
*cn
, *next
, *jl_cn
;
3848 struct reiserfs_journal_cnode
*last_cn
= NULL
;
3849 struct reiserfs_journal_desc
*desc
;
3850 struct reiserfs_journal_commit
*commit
;
3851 struct buffer_head
*c_bh
; /* commit bh */
3852 struct buffer_head
*d_bh
; /* desc bh */
3853 int cur_write_start
= 0; /* start index of current log write */
3856 int flush
= flags
& FLUSH_ALL
;
3857 int wait_on_commit
= flags
& WAIT
;
3858 struct reiserfs_journal_list
*jl
, *temp_jl
;
3859 struct list_head
*entry
, *safe
;
3860 unsigned long jindex
;
3861 unsigned long commit_trans_id
;
3864 BUG_ON(th
->t_refcount
> 1);
3865 BUG_ON(!th
->t_trans_id
);
3868 current
->journal_info
= th
->t_handle_save
;
3869 reiserfs_check_lock_depth(p_s_sb
, "journal end");
3870 if (journal
->j_len
== 0) {
3871 reiserfs_prepare_for_journal(p_s_sb
, SB_BUFFER_WITH_SB(p_s_sb
),
3873 journal_mark_dirty(th
, p_s_sb
, SB_BUFFER_WITH_SB(p_s_sb
));
3876 lock_journal(p_s_sb
);
3877 if (journal
->j_next_full_flush
) {
3881 if (journal
->j_next_async_flush
) {
3882 flags
|= COMMIT_NOW
| WAIT
;
3886 /* check_journal_end locks the journal, and unlocks if it does not return 1
3887 ** it tells us if we should continue with the journal_end, or just return
3889 if (!check_journal_end(th
, p_s_sb
, nblocks
, flags
)) {
3891 wake_queued_writers(p_s_sb
);
3892 reiserfs_async_progress_wait(p_s_sb
);
3896 /* check_journal_end might set these, check again */
3897 if (journal
->j_next_full_flush
) {
3902 ** j must wait means we have to flush the log blocks, and the real blocks for
3905 if (journal
->j_must_wait
> 0) {
3908 #ifdef REISERFS_PREALLOCATE
3909 /* quota ops might need to nest, setup the journal_info pointer for them */
3910 current
->journal_info
= th
;
3911 reiserfs_discard_all_prealloc(th
); /* it should not involve new blocks into
3912 * the transaction */
3913 current
->journal_info
= th
->t_handle_save
;
3916 /* setup description block */
3918 journal_getblk(p_s_sb
,
3919 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
3921 set_buffer_uptodate(d_bh
);
3922 desc
= (struct reiserfs_journal_desc
*)(d_bh
)->b_data
;
3923 memset(d_bh
->b_data
, 0, d_bh
->b_size
);
3924 memcpy(get_journal_desc_magic(d_bh
), JOURNAL_DESC_MAGIC
, 8);
3925 set_desc_trans_id(desc
, journal
->j_trans_id
);
3927 /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */
3928 c_bh
= journal_getblk(p_s_sb
, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
3929 ((journal
->j_start
+ journal
->j_len
+
3930 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb
)));
3931 commit
= (struct reiserfs_journal_commit
*)c_bh
->b_data
;
3932 memset(c_bh
->b_data
, 0, c_bh
->b_size
);
3933 set_commit_trans_id(commit
, journal
->j_trans_id
);
3934 set_buffer_uptodate(c_bh
);
3936 /* init this journal list */
3937 jl
= journal
->j_current_jl
;
3939 /* we lock the commit before doing anything because
3940 * we want to make sure nobody tries to run flush_commit_list until
3941 * the new transaction is fully setup, and we've already flushed the
3944 down(&jl
->j_commit_lock
);
3946 /* save the transaction id in case we need to commit it later */
3947 commit_trans_id
= jl
->j_trans_id
;
3949 atomic_set(&jl
->j_older_commits_done
, 0);
3950 jl
->j_trans_id
= journal
->j_trans_id
;
3951 jl
->j_timestamp
= journal
->j_trans_start_time
;
3952 jl
->j_commit_bh
= c_bh
;
3953 jl
->j_start
= journal
->j_start
;
3954 jl
->j_len
= journal
->j_len
;
3955 atomic_set(&jl
->j_nonzerolen
, journal
->j_len
);
3956 atomic_set(&jl
->j_commit_left
, journal
->j_len
+ 2);
3957 jl
->j_realblock
= NULL
;
3959 /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
3960 ** for each real block, add it to the journal list hash,
3961 ** copy into real block index array in the commit or desc block
3963 trans_half
= journal_trans_half(p_s_sb
->s_blocksize
);
3964 for (i
= 0, cn
= journal
->j_first
; cn
; cn
= cn
->next
, i
++) {
3965 if (buffer_journaled(cn
->bh
)) {
3966 jl_cn
= get_cnode(p_s_sb
);
3968 reiserfs_panic(p_s_sb
,
3969 "journal-1676, get_cnode returned NULL\n");
3972 jl
->j_realblock
= jl_cn
;
3974 jl_cn
->prev
= last_cn
;
3977 last_cn
->next
= jl_cn
;
3980 /* make sure the block we are trying to log is not a block
3981 of journal or reserved area */
3983 if (is_block_in_log_or_reserved_area
3984 (p_s_sb
, cn
->bh
->b_blocknr
)) {
3985 reiserfs_panic(p_s_sb
,
3986 "journal-2332: Trying to log block %lu, which is a log block\n",
3989 jl_cn
->blocknr
= cn
->bh
->b_blocknr
;
3994 insert_journal_hash(journal
->j_list_hash_table
, jl_cn
);
3995 if (i
< trans_half
) {
3996 desc
->j_realblock
[i
] =
3997 cpu_to_le32(cn
->bh
->b_blocknr
);
3999 commit
->j_realblock
[i
- trans_half
] =
4000 cpu_to_le32(cn
->bh
->b_blocknr
);
4006 set_desc_trans_len(desc
, journal
->j_len
);
4007 set_desc_mount_id(desc
, journal
->j_mount_id
);
4008 set_desc_trans_id(desc
, journal
->j_trans_id
);
4009 set_commit_trans_len(commit
, journal
->j_len
);
4011 /* special check in case all buffers in the journal were marked for not logging */
4012 if (journal
->j_len
== 0) {
4016 /* we're about to dirty all the log blocks, mark the description block
4017 * dirty now too. Don't mark the commit block dirty until all the
4018 * others are on disk
4020 mark_buffer_dirty(d_bh
);
4022 /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4023 cur_write_start
= journal
->j_start
;
4024 cn
= journal
->j_first
;
4025 jindex
= 1; /* start at one so we don't get the desc again */
4027 clear_buffer_journal_new(cn
->bh
);
4028 /* copy all the real blocks into log area. dirty log blocks */
4029 if (buffer_journaled(cn
->bh
)) {
4030 struct buffer_head
*tmp_bh
;
4034 journal_getblk(p_s_sb
,
4035 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
4038 SB_ONDISK_JOURNAL_SIZE(p_s_sb
)));
4039 set_buffer_uptodate(tmp_bh
);
4040 page
= cn
->bh
->b_page
;
4042 memcpy(tmp_bh
->b_data
,
4043 addr
+ offset_in_page(cn
->bh
->b_data
),
4046 mark_buffer_dirty(tmp_bh
);
4048 set_buffer_journal_dirty(cn
->bh
);
4049 clear_buffer_journaled(cn
->bh
);
4051 /* JDirty cleared sometime during transaction. don't log this one */
4052 reiserfs_warning(p_s_sb
,
4053 "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!");
4057 free_cnode(p_s_sb
, cn
);
4062 /* we are done with both the c_bh and d_bh, but
4063 ** c_bh must be written after all other commit blocks,
4064 ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4067 journal
->j_current_jl
= alloc_journal_list(p_s_sb
);
4069 /* now it is safe to insert this transaction on the main list */
4070 list_add_tail(&jl
->j_list
, &journal
->j_journal_list
);
4071 list_add_tail(&jl
->j_working_list
, &journal
->j_working_list
);
4072 journal
->j_num_work_lists
++;
4074 /* reset journal values for the next transaction */
4075 old_start
= journal
->j_start
;
4077 (journal
->j_start
+ journal
->j_len
+
4078 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb
);
4079 atomic_set(&(journal
->j_wcount
), 0);
4080 journal
->j_bcount
= 0;
4081 journal
->j_last
= NULL
;
4082 journal
->j_first
= NULL
;
4084 journal
->j_trans_start_time
= 0;
4085 journal
->j_trans_id
++;
4086 journal
->j_current_jl
->j_trans_id
= journal
->j_trans_id
;
4087 journal
->j_must_wait
= 0;
4088 journal
->j_len_alloc
= 0;
4089 journal
->j_next_full_flush
= 0;
4090 journal
->j_next_async_flush
= 0;
4091 init_journal_hash(p_s_sb
);
4093 // make sure reiserfs_add_jh sees the new current_jl before we
4094 // write out the tails
4097 /* tail conversion targets have to hit the disk before we end the
4098 * transaction. Otherwise a later transaction might repack the tail
4099 * before this transaction commits, leaving the data block unflushed and
4100 * clean, if we crash before the later transaction commits, the data block
4103 if (!list_empty(&jl
->j_tail_bh_list
)) {
4105 write_ordered_buffers(&journal
->j_dirty_buffers_lock
,
4106 journal
, jl
, &jl
->j_tail_bh_list
);
4109 if (!list_empty(&jl
->j_tail_bh_list
))
4111 up(&jl
->j_commit_lock
);
4113 /* honor the flush wishes from the caller, simple commits can
4114 ** be done outside the journal lock, they are done below
4116 ** if we don't flush the commit list right now, we put it into
4117 ** the work queue so the people waiting on the async progress work
4118 ** queue don't wait for this proc to flush journal lists and such.
4121 flush_commit_list(p_s_sb
, jl
, 1);
4122 flush_journal_list(p_s_sb
, jl
, 1);
4123 } else if (!(jl
->j_state
& LIST_COMMIT_PENDING
))
4124 queue_delayed_work(commit_wq
, &journal
->j_work
, HZ
/ 10);
4126 /* if the next transaction has any chance of wrapping, flush
4127 ** transactions that might get overwritten. If any journal lists are very
4128 ** old flush them as well.
4131 list_for_each_safe(entry
, safe
, &journal
->j_journal_list
) {
4132 temp_jl
= JOURNAL_LIST_ENTRY(entry
);
4133 if (journal
->j_start
<= temp_jl
->j_start
) {
4134 if ((journal
->j_start
+ journal
->j_trans_max
+ 1) >=
4136 flush_used_journal_lists(p_s_sb
, temp_jl
);
4138 } else if ((journal
->j_start
+
4139 journal
->j_trans_max
+ 1) <
4140 SB_ONDISK_JOURNAL_SIZE(p_s_sb
)) {
4141 /* if we don't cross into the next transaction and we don't
4142 * wrap, there is no way we can overlap any later transactions
4147 } else if ((journal
->j_start
+
4148 journal
->j_trans_max
+ 1) >
4149 SB_ONDISK_JOURNAL_SIZE(p_s_sb
)) {
4150 if (((journal
->j_start
+ journal
->j_trans_max
+ 1) %
4151 SB_ONDISK_JOURNAL_SIZE(p_s_sb
)) >=
4153 flush_used_journal_lists(p_s_sb
, temp_jl
);
4156 /* we don't overlap anything from out start to the end of the
4157 * log, and our wrapped portion doesn't overlap anything at
4158 * the start of the log. We can break
4164 flush_old_journal_lists(p_s_sb
);
4166 journal
->j_current_jl
->j_list_bitmap
=
4167 get_list_bitmap(p_s_sb
, journal
->j_current_jl
);
4169 if (!(journal
->j_current_jl
->j_list_bitmap
)) {
4170 reiserfs_panic(p_s_sb
,
4171 "journal-1996: do_journal_end, could not get a list bitmap\n");
4174 atomic_set(&(journal
->j_jlock
), 0);
4175 unlock_journal(p_s_sb
);
4176 /* wake up any body waiting to join. */
4177 clear_bit(J_WRITERS_QUEUED
, &journal
->j_state
);
4178 wake_up(&(journal
->j_join_wait
));
4180 if (!flush
&& wait_on_commit
&&
4181 journal_list_still_alive(p_s_sb
, commit_trans_id
)) {
4182 flush_commit_list(p_s_sb
, jl
, 1);
4185 reiserfs_check_lock_depth(p_s_sb
, "journal end2");
4187 memset(th
, 0, sizeof(*th
));
4188 /* Re-set th->t_super, so we can properly keep track of how many
4189 * persistent transactions there are. We need to do this so if this
4190 * call is part of a failed restart_transaction, we can free it later */
4191 th
->t_super
= p_s_sb
;
4193 return journal
->j_errno
;
4196 static void __reiserfs_journal_abort_hard(struct super_block
*sb
)
4198 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
4199 if (test_bit(J_ABORTED
, &journal
->j_state
))
4202 printk(KERN_CRIT
"REISERFS: Aborting journal for filesystem on %s\n",
4203 reiserfs_bdevname(sb
));
4205 sb
->s_flags
|= MS_RDONLY
;
4206 set_bit(J_ABORTED
, &journal
->j_state
);
4208 #ifdef CONFIG_REISERFS_CHECK
4213 static void __reiserfs_journal_abort_soft(struct super_block
*sb
, int errno
)
4215 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
4216 if (test_bit(J_ABORTED
, &journal
->j_state
))
4219 if (!journal
->j_errno
)
4220 journal
->j_errno
= errno
;
4222 __reiserfs_journal_abort_hard(sb
);
4225 void reiserfs_journal_abort(struct super_block
*sb
, int errno
)
4227 return __reiserfs_journal_abort_soft(sb
, errno
);