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
= kmalloc(sizeof(struct reiserfs_bitmap_node
), GFP_NOFS
);
159 bn
->data
= kzalloc(p_s_sb
->s_blocksize
, GFP_NOFS
);
165 INIT_LIST_HEAD(&bn
->list
);
169 static struct reiserfs_bitmap_node
*get_bitmap_node(struct super_block
*p_s_sb
)
171 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
172 struct reiserfs_bitmap_node
*bn
= NULL
;
173 struct list_head
*entry
= journal
->j_bitmap_nodes
.next
;
175 journal
->j_used_bitmap_nodes
++;
178 if (entry
!= &journal
->j_bitmap_nodes
) {
179 bn
= list_entry(entry
, struct reiserfs_bitmap_node
, list
);
181 memset(bn
->data
, 0, p_s_sb
->s_blocksize
);
182 journal
->j_free_bitmap_nodes
--;
185 bn
= allocate_bitmap_node(p_s_sb
);
192 static inline void free_bitmap_node(struct super_block
*p_s_sb
,
193 struct reiserfs_bitmap_node
*bn
)
195 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
196 journal
->j_used_bitmap_nodes
--;
197 if (journal
->j_free_bitmap_nodes
> REISERFS_MAX_BITMAP_NODES
) {
201 list_add(&bn
->list
, &journal
->j_bitmap_nodes
);
202 journal
->j_free_bitmap_nodes
++;
206 static void allocate_bitmap_nodes(struct super_block
*p_s_sb
)
209 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
210 struct reiserfs_bitmap_node
*bn
= NULL
;
211 for (i
= 0; i
< REISERFS_MIN_BITMAP_NODES
; i
++) {
212 bn
= allocate_bitmap_node(p_s_sb
);
214 list_add(&bn
->list
, &journal
->j_bitmap_nodes
);
215 journal
->j_free_bitmap_nodes
++;
217 break; // this is ok, we'll try again when more are needed
222 static int set_bit_in_list_bitmap(struct super_block
*p_s_sb
, int block
,
223 struct reiserfs_list_bitmap
*jb
)
225 int bmap_nr
= block
/ (p_s_sb
->s_blocksize
<< 3);
226 int bit_nr
= block
% (p_s_sb
->s_blocksize
<< 3);
228 if (!jb
->bitmaps
[bmap_nr
]) {
229 jb
->bitmaps
[bmap_nr
] = get_bitmap_node(p_s_sb
);
231 set_bit(bit_nr
, (unsigned long *)jb
->bitmaps
[bmap_nr
]->data
);
235 static void cleanup_bitmap_list(struct super_block
*p_s_sb
,
236 struct reiserfs_list_bitmap
*jb
)
239 if (jb
->bitmaps
== NULL
)
242 for (i
= 0; i
< SB_BMAP_NR(p_s_sb
); i
++) {
243 if (jb
->bitmaps
[i
]) {
244 free_bitmap_node(p_s_sb
, jb
->bitmaps
[i
]);
245 jb
->bitmaps
[i
] = NULL
;
251 ** only call this on FS unmount.
253 static int free_list_bitmaps(struct super_block
*p_s_sb
,
254 struct reiserfs_list_bitmap
*jb_array
)
257 struct reiserfs_list_bitmap
*jb
;
258 for (i
= 0; i
< JOURNAL_NUM_BITMAPS
; i
++) {
260 jb
->journal_list
= NULL
;
261 cleanup_bitmap_list(p_s_sb
, jb
);
268 static int free_bitmap_nodes(struct super_block
*p_s_sb
)
270 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
271 struct list_head
*next
= journal
->j_bitmap_nodes
.next
;
272 struct reiserfs_bitmap_node
*bn
;
274 while (next
!= &journal
->j_bitmap_nodes
) {
275 bn
= list_entry(next
, struct reiserfs_bitmap_node
, list
);
279 next
= journal
->j_bitmap_nodes
.next
;
280 journal
->j_free_bitmap_nodes
--;
287 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
288 ** jb_array is the array to be filled in.
290 int reiserfs_allocate_list_bitmaps(struct super_block
*p_s_sb
,
291 struct reiserfs_list_bitmap
*jb_array
,
296 struct reiserfs_list_bitmap
*jb
;
297 int mem
= bmap_nr
* sizeof(struct reiserfs_bitmap_node
*);
299 for (i
= 0; i
< JOURNAL_NUM_BITMAPS
; i
++) {
301 jb
->journal_list
= NULL
;
302 jb
->bitmaps
= vmalloc(mem
);
304 reiserfs_warning(p_s_sb
,
305 "clm-2000, unable to allocate bitmaps for journal lists");
309 memset(jb
->bitmaps
, 0, mem
);
312 free_list_bitmaps(p_s_sb
, jb_array
);
319 ** find an available list bitmap. If you can't find one, flush a commit list
322 static struct reiserfs_list_bitmap
*get_list_bitmap(struct super_block
*p_s_sb
,
323 struct reiserfs_journal_list
327 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
328 struct reiserfs_list_bitmap
*jb
= NULL
;
330 for (j
= 0; j
< (JOURNAL_NUM_BITMAPS
* 3); j
++) {
331 i
= journal
->j_list_bitmap_index
;
332 journal
->j_list_bitmap_index
= (i
+ 1) % JOURNAL_NUM_BITMAPS
;
333 jb
= journal
->j_list_bitmap
+ i
;
334 if (journal
->j_list_bitmap
[i
].journal_list
) {
335 flush_commit_list(p_s_sb
,
336 journal
->j_list_bitmap
[i
].
338 if (!journal
->j_list_bitmap
[i
].journal_list
) {
345 if (jb
->journal_list
) { /* double check to make sure if flushed correctly */
348 jb
->journal_list
= jl
;
353 ** allocates a new chunk of X nodes, and links them all together as a list.
354 ** Uses the cnode->next and cnode->prev pointers
355 ** returns NULL on failure
357 static struct reiserfs_journal_cnode
*allocate_cnodes(int num_cnodes
)
359 struct reiserfs_journal_cnode
*head
;
361 if (num_cnodes
<= 0) {
364 head
= vmalloc(num_cnodes
* sizeof(struct reiserfs_journal_cnode
));
368 memset(head
, 0, num_cnodes
* sizeof(struct reiserfs_journal_cnode
));
370 head
[0].next
= head
+ 1;
371 for (i
= 1; i
< num_cnodes
; i
++) {
372 head
[i
].prev
= head
+ (i
- 1);
373 head
[i
].next
= head
+ (i
+ 1); /* if last one, overwrite it after the if */
375 head
[num_cnodes
- 1].next
= NULL
;
380 ** pulls a cnode off the free list, or returns NULL on failure
382 static struct reiserfs_journal_cnode
*get_cnode(struct super_block
*p_s_sb
)
384 struct reiserfs_journal_cnode
*cn
;
385 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
387 reiserfs_check_lock_depth(p_s_sb
, "get_cnode");
389 if (journal
->j_cnode_free
<= 0) {
392 journal
->j_cnode_used
++;
393 journal
->j_cnode_free
--;
394 cn
= journal
->j_cnode_free_list
;
399 cn
->next
->prev
= NULL
;
401 journal
->j_cnode_free_list
= cn
->next
;
402 memset(cn
, 0, sizeof(struct reiserfs_journal_cnode
));
407 ** returns a cnode to the free list
409 static void free_cnode(struct super_block
*p_s_sb
,
410 struct reiserfs_journal_cnode
*cn
)
412 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
414 reiserfs_check_lock_depth(p_s_sb
, "free_cnode");
416 journal
->j_cnode_used
--;
417 journal
->j_cnode_free
++;
418 /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
419 cn
->next
= journal
->j_cnode_free_list
;
420 if (journal
->j_cnode_free_list
) {
421 journal
->j_cnode_free_list
->prev
= cn
;
423 cn
->prev
= NULL
; /* not needed with the memset, but I might kill the memset, and forget to do this */
424 journal
->j_cnode_free_list
= cn
;
427 static void clear_prepared_bits(struct buffer_head
*bh
)
429 clear_buffer_journal_prepared(bh
);
430 clear_buffer_journal_restore_dirty(bh
);
433 /* utility function to force a BUG if it is called without the big
434 ** kernel lock held. caller is the string printed just before calling BUG()
436 void reiserfs_check_lock_depth(struct super_block
*sb
, char *caller
)
439 if (current
->lock_depth
< 0) {
440 reiserfs_panic(sb
, "%s called without kernel lock held",
448 /* return a cnode with same dev, block number and size in table, or null if not found */
449 static inline struct reiserfs_journal_cnode
*get_journal_hash_dev(struct
453 reiserfs_journal_cnode
457 struct reiserfs_journal_cnode
*cn
;
458 cn
= journal_hash(table
, sb
, bl
);
460 if (cn
->blocknr
== bl
&& cn
->sb
== sb
)
464 return (struct reiserfs_journal_cnode
*)0;
468 ** this actually means 'can this block be reallocated yet?'. If you set search_all, a block can only be allocated
469 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
470 ** being overwritten by a replay after crashing.
472 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction. Since deleting
473 ** a block removes it from the current transaction, this case should never happen. If you don't set search_all, make
474 ** sure you never write the block without logging it.
476 ** next_zero_bit is a suggestion about the next block to try for find_forward.
477 ** when bl is rejected because it is set in a journal list bitmap, we search
478 ** for the next zero bit in the bitmap that rejected bl. Then, we return that
479 ** through next_zero_bit for find_forward to try.
481 ** Just because we return something in next_zero_bit does not mean we won't
482 ** reject it on the next call to reiserfs_in_journal
485 int reiserfs_in_journal(struct super_block
*p_s_sb
,
486 int bmap_nr
, int bit_nr
, int search_all
,
487 b_blocknr_t
* next_zero_bit
)
489 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
490 struct reiserfs_journal_cnode
*cn
;
491 struct reiserfs_list_bitmap
*jb
;
495 *next_zero_bit
= 0; /* always start this at zero. */
497 PROC_INFO_INC(p_s_sb
, journal
.in_journal
);
498 /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
499 ** if we crash before the transaction that freed it commits, this transaction won't
500 ** have committed either, and the block will never be written
503 for (i
= 0; i
< JOURNAL_NUM_BITMAPS
; i
++) {
504 PROC_INFO_INC(p_s_sb
, journal
.in_journal_bitmap
);
505 jb
= journal
->j_list_bitmap
+ i
;
506 if (jb
->journal_list
&& jb
->bitmaps
[bmap_nr
] &&
508 (unsigned long *)jb
->bitmaps
[bmap_nr
]->
511 find_next_zero_bit((unsigned long *)
512 (jb
->bitmaps
[bmap_nr
]->
514 p_s_sb
->s_blocksize
<< 3,
521 bl
= bmap_nr
* (p_s_sb
->s_blocksize
<< 3) + bit_nr
;
522 /* is it in any old transactions? */
525 get_journal_hash_dev(p_s_sb
, journal
->j_list_hash_table
, bl
))) {
529 /* is it in the current transaction. This should never happen */
530 if ((cn
= get_journal_hash_dev(p_s_sb
, journal
->j_hash_table
, bl
))) {
535 PROC_INFO_INC(p_s_sb
, journal
.in_journal_reusable
);
540 /* insert cn into table
542 static inline void insert_journal_hash(struct reiserfs_journal_cnode
**table
,
543 struct reiserfs_journal_cnode
*cn
)
545 struct reiserfs_journal_cnode
*cn_orig
;
547 cn_orig
= journal_hash(table
, cn
->sb
, cn
->blocknr
);
553 journal_hash(table
, cn
->sb
, cn
->blocknr
) = cn
;
556 /* lock the current transaction */
557 static inline void lock_journal(struct super_block
*p_s_sb
)
559 PROC_INFO_INC(p_s_sb
, journal
.lock_journal
);
560 down(&SB_JOURNAL(p_s_sb
)->j_lock
);
563 /* unlock the current transaction */
564 static inline void unlock_journal(struct super_block
*p_s_sb
)
566 up(&SB_JOURNAL(p_s_sb
)->j_lock
);
569 static inline void get_journal_list(struct reiserfs_journal_list
*jl
)
574 static inline void put_journal_list(struct super_block
*s
,
575 struct reiserfs_journal_list
*jl
)
577 if (jl
->j_refcount
< 1) {
578 reiserfs_panic(s
, "trans id %lu, refcount at %d",
579 jl
->j_trans_id
, jl
->j_refcount
);
581 if (--jl
->j_refcount
== 0)
586 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
587 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
590 static void cleanup_freed_for_journal_list(struct super_block
*p_s_sb
,
591 struct reiserfs_journal_list
*jl
)
594 struct reiserfs_list_bitmap
*jb
= jl
->j_list_bitmap
;
596 cleanup_bitmap_list(p_s_sb
, jb
);
598 jl
->j_list_bitmap
->journal_list
= NULL
;
599 jl
->j_list_bitmap
= NULL
;
602 static int journal_list_still_alive(struct super_block
*s
,
603 unsigned long trans_id
)
605 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
606 struct list_head
*entry
= &journal
->j_journal_list
;
607 struct reiserfs_journal_list
*jl
;
609 if (!list_empty(entry
)) {
610 jl
= JOURNAL_LIST_ENTRY(entry
->next
);
611 if (jl
->j_trans_id
<= trans_id
) {
618 static void reiserfs_end_buffer_io_sync(struct buffer_head
*bh
, int uptodate
)
620 char b
[BDEVNAME_SIZE
];
622 if (buffer_journaled(bh
)) {
623 reiserfs_warning(NULL
,
624 "clm-2084: pinned buffer %lu:%s sent to disk",
625 bh
->b_blocknr
, bdevname(bh
->b_bdev
, b
));
628 set_buffer_uptodate(bh
);
630 clear_buffer_uptodate(bh
);
635 static void reiserfs_end_ordered_io(struct buffer_head
*bh
, int uptodate
)
638 set_buffer_uptodate(bh
);
640 clear_buffer_uptodate(bh
);
645 static void submit_logged_buffer(struct buffer_head
*bh
)
648 bh
->b_end_io
= reiserfs_end_buffer_io_sync
;
649 clear_buffer_journal_new(bh
);
650 clear_buffer_dirty(bh
);
651 if (!test_clear_buffer_journal_test(bh
))
653 if (!buffer_uptodate(bh
))
655 submit_bh(WRITE
, bh
);
658 static void submit_ordered_buffer(struct buffer_head
*bh
)
661 bh
->b_end_io
= reiserfs_end_ordered_io
;
662 clear_buffer_dirty(bh
);
663 if (!buffer_uptodate(bh
))
665 submit_bh(WRITE
, bh
);
668 static int submit_barrier_buffer(struct buffer_head
*bh
)
671 bh
->b_end_io
= reiserfs_end_ordered_io
;
672 clear_buffer_dirty(bh
);
673 if (!buffer_uptodate(bh
))
675 return submit_bh(WRITE_BARRIER
, bh
);
678 static void check_barrier_completion(struct super_block
*s
,
679 struct buffer_head
*bh
)
681 if (buffer_eopnotsupp(bh
)) {
682 clear_buffer_eopnotsupp(bh
);
684 set_buffer_uptodate(bh
);
685 set_buffer_dirty(bh
);
686 sync_dirty_buffer(bh
);
690 #define CHUNK_SIZE 32
691 struct buffer_chunk
{
692 struct buffer_head
*bh
[CHUNK_SIZE
];
696 static void write_chunk(struct buffer_chunk
*chunk
)
700 for (i
= 0; i
< chunk
->nr
; i
++) {
701 submit_logged_buffer(chunk
->bh
[i
]);
707 static void write_ordered_chunk(struct buffer_chunk
*chunk
)
711 for (i
= 0; i
< chunk
->nr
; i
++) {
712 submit_ordered_buffer(chunk
->bh
[i
]);
718 static int add_to_chunk(struct buffer_chunk
*chunk
, struct buffer_head
*bh
,
719 spinlock_t
* lock
, void (fn
) (struct buffer_chunk
*))
722 if (chunk
->nr
>= CHUNK_SIZE
)
724 chunk
->bh
[chunk
->nr
++] = bh
;
725 if (chunk
->nr
>= CHUNK_SIZE
) {
736 static atomic_t nr_reiserfs_jh
= ATOMIC_INIT(0);
737 static struct reiserfs_jh
*alloc_jh(void)
739 struct reiserfs_jh
*jh
;
741 jh
= kmalloc(sizeof(*jh
), GFP_NOFS
);
743 atomic_inc(&nr_reiserfs_jh
);
751 * we want to free the jh when the buffer has been written
754 void reiserfs_free_jh(struct buffer_head
*bh
)
756 struct reiserfs_jh
*jh
;
760 bh
->b_private
= NULL
;
762 list_del_init(&jh
->list
);
764 if (atomic_read(&nr_reiserfs_jh
) <= 0)
766 atomic_dec(&nr_reiserfs_jh
);
771 static inline int __add_jh(struct reiserfs_journal
*j
, struct buffer_head
*bh
,
774 struct reiserfs_jh
*jh
;
777 spin_lock(&j
->j_dirty_buffers_lock
);
778 if (!bh
->b_private
) {
779 spin_unlock(&j
->j_dirty_buffers_lock
);
783 list_del_init(&jh
->list
);
788 spin_lock(&j
->j_dirty_buffers_lock
);
789 /* buffer must be locked for __add_jh, should be able to have
790 * two adds at the same time
797 jh
->jl
= j
->j_current_jl
;
799 list_add_tail(&jh
->list
, &jh
->jl
->j_tail_bh_list
);
801 list_add_tail(&jh
->list
, &jh
->jl
->j_bh_list
);
803 spin_unlock(&j
->j_dirty_buffers_lock
);
807 int reiserfs_add_tail_list(struct inode
*inode
, struct buffer_head
*bh
)
809 return __add_jh(SB_JOURNAL(inode
->i_sb
), bh
, 1);
811 int reiserfs_add_ordered_list(struct inode
*inode
, struct buffer_head
*bh
)
813 return __add_jh(SB_JOURNAL(inode
->i_sb
), bh
, 0);
816 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
817 static int write_ordered_buffers(spinlock_t
* lock
,
818 struct reiserfs_journal
*j
,
819 struct reiserfs_journal_list
*jl
,
820 struct list_head
*list
)
822 struct buffer_head
*bh
;
823 struct reiserfs_jh
*jh
;
824 int ret
= j
->j_errno
;
825 struct buffer_chunk chunk
;
826 struct list_head tmp
;
827 INIT_LIST_HEAD(&tmp
);
831 while (!list_empty(list
)) {
832 jh
= JH_ENTRY(list
->next
);
835 if (test_set_buffer_locked(bh
)) {
836 if (!buffer_dirty(bh
)) {
837 list_move(&jh
->list
, &tmp
);
842 write_ordered_chunk(&chunk
);
848 /* in theory, dirty non-uptodate buffers should never get here,
849 * but the upper layer io error paths still have a few quirks.
850 * Handle them here as gracefully as we can
852 if (!buffer_uptodate(bh
) && buffer_dirty(bh
)) {
853 clear_buffer_dirty(bh
);
856 if (buffer_dirty(bh
)) {
857 list_move(&jh
->list
, &tmp
);
858 add_to_chunk(&chunk
, bh
, lock
, write_ordered_chunk
);
860 reiserfs_free_jh(bh
);
865 cond_resched_lock(lock
);
869 write_ordered_chunk(&chunk
);
872 while (!list_empty(&tmp
)) {
873 jh
= JH_ENTRY(tmp
.prev
);
876 reiserfs_free_jh(bh
);
878 if (buffer_locked(bh
)) {
883 if (!buffer_uptodate(bh
)) {
886 /* ugly interaction with invalidatepage here.
887 * reiserfs_invalidate_page will pin any buffer that has a valid
888 * journal head from an older transaction. If someone else sets
889 * our buffer dirty after we write it in the first loop, and
890 * then someone truncates the page away, nobody will ever write
891 * the buffer. We're safe if we write the page one last time
892 * after freeing the journal header.
894 if (buffer_dirty(bh
) && unlikely(bh
->b_page
->mapping
== NULL
)) {
896 ll_rw_block(WRITE
, 1, &bh
);
900 cond_resched_lock(lock
);
906 static int flush_older_commits(struct super_block
*s
,
907 struct reiserfs_journal_list
*jl
)
909 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
910 struct reiserfs_journal_list
*other_jl
;
911 struct reiserfs_journal_list
*first_jl
;
912 struct list_head
*entry
;
913 unsigned long trans_id
= jl
->j_trans_id
;
914 unsigned long other_trans_id
;
915 unsigned long first_trans_id
;
919 * first we walk backwards to find the oldest uncommitted transation
922 entry
= jl
->j_list
.prev
;
924 other_jl
= JOURNAL_LIST_ENTRY(entry
);
925 if (entry
== &journal
->j_journal_list
||
926 atomic_read(&other_jl
->j_older_commits_done
))
930 entry
= other_jl
->j_list
.prev
;
933 /* if we didn't find any older uncommitted transactions, return now */
934 if (first_jl
== jl
) {
938 first_trans_id
= first_jl
->j_trans_id
;
940 entry
= &first_jl
->j_list
;
942 other_jl
= JOURNAL_LIST_ENTRY(entry
);
943 other_trans_id
= other_jl
->j_trans_id
;
945 if (other_trans_id
< trans_id
) {
946 if (atomic_read(&other_jl
->j_commit_left
) != 0) {
947 flush_commit_list(s
, other_jl
, 0);
949 /* list we were called with is gone, return */
950 if (!journal_list_still_alive(s
, trans_id
))
953 /* the one we just flushed is gone, this means all
954 * older lists are also gone, so first_jl is no longer
955 * valid either. Go back to the beginning.
957 if (!journal_list_still_alive
958 (s
, other_trans_id
)) {
963 if (entry
== &journal
->j_journal_list
)
971 int reiserfs_async_progress_wait(struct super_block
*s
)
974 struct reiserfs_journal
*j
= SB_JOURNAL(s
);
975 if (atomic_read(&j
->j_async_throttle
))
976 blk_congestion_wait(WRITE
, HZ
/ 10);
981 ** if this journal list still has commit blocks unflushed, send them to disk.
983 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
984 ** Before the commit block can by written, every other log block must be safely on disk
987 static int flush_commit_list(struct super_block
*s
,
988 struct reiserfs_journal_list
*jl
, int flushall
)
992 struct buffer_head
*tbh
= NULL
;
993 unsigned long trans_id
= jl
->j_trans_id
;
994 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
999 reiserfs_check_lock_depth(s
, "flush_commit_list");
1001 if (atomic_read(&jl
->j_older_commits_done
)) {
1007 /* before we can put our commit blocks on disk, we have to make sure everyone older than
1008 ** us is on disk too
1010 BUG_ON(jl
->j_len
<= 0);
1011 BUG_ON(trans_id
== journal
->j_trans_id
);
1013 get_journal_list(jl
);
1015 if (flush_older_commits(s
, jl
) == 1) {
1016 /* list disappeared during flush_older_commits. return */
1021 /* make sure nobody is trying to flush this one at the same time */
1022 down(&jl
->j_commit_lock
);
1023 if (!journal_list_still_alive(s
, trans_id
)) {
1024 up(&jl
->j_commit_lock
);
1027 BUG_ON(jl
->j_trans_id
== 0);
1029 /* this commit is done, exit */
1030 if (atomic_read(&(jl
->j_commit_left
)) <= 0) {
1032 atomic_set(&(jl
->j_older_commits_done
), 1);
1034 up(&jl
->j_commit_lock
);
1038 if (!list_empty(&jl
->j_bh_list
)) {
1041 ret
= write_ordered_buffers(&journal
->j_dirty_buffers_lock
,
1042 journal
, jl
, &jl
->j_bh_list
);
1043 if (ret
< 0 && retval
== 0)
1047 BUG_ON(!list_empty(&jl
->j_bh_list
));
1049 * for the description block and all the log blocks, submit any buffers
1050 * that haven't already reached the disk. Try to write at least 256
1051 * log blocks. later on, we will only wait on blocks that correspond
1052 * to this transaction, but while we're unplugging we might as well
1053 * get a chunk of data on there.
1055 atomic_inc(&journal
->j_async_throttle
);
1056 write_len
= jl
->j_len
+ 1;
1057 if (write_len
< 256)
1059 for (i
= 0 ; i
< write_len
; i
++) {
1060 bn
= SB_ONDISK_JOURNAL_1st_BLOCK(s
) + (jl
->j_start
+ i
) %
1061 SB_ONDISK_JOURNAL_SIZE(s
);
1062 tbh
= journal_find_get_block(s
, bn
);
1064 if (buffer_dirty(tbh
))
1065 ll_rw_block(WRITE
, 1, &tbh
) ;
1069 atomic_dec(&journal
->j_async_throttle
);
1071 /* We're skipping the commit if there's an error */
1072 if (retval
|| reiserfs_is_journal_aborted(journal
))
1075 /* wait on everything written so far before writing the commit
1076 * if we are in barrier mode, send the commit down now
1078 barrier
= reiserfs_barrier_flush(s
);
1081 lock_buffer(jl
->j_commit_bh
);
1082 ret
= submit_barrier_buffer(jl
->j_commit_bh
);
1083 if (ret
== -EOPNOTSUPP
) {
1084 set_buffer_uptodate(jl
->j_commit_bh
);
1089 for (i
= 0; i
< (jl
->j_len
+ 1); i
++) {
1090 bn
= SB_ONDISK_JOURNAL_1st_BLOCK(s
) +
1091 (jl
->j_start
+ i
) % SB_ONDISK_JOURNAL_SIZE(s
);
1092 tbh
= journal_find_get_block(s
, bn
);
1093 wait_on_buffer(tbh
);
1094 // since we're using ll_rw_blk above, it might have skipped over
1095 // a locked buffer. Double check here
1097 if (buffer_dirty(tbh
)) /* redundant, sync_dirty_buffer() checks */
1098 sync_dirty_buffer(tbh
);
1099 if (unlikely(!buffer_uptodate(tbh
))) {
1100 #ifdef CONFIG_REISERFS_CHECK
1101 reiserfs_warning(s
, "journal-601, buffer write failed");
1105 put_bh(tbh
); /* once for journal_find_get_block */
1106 put_bh(tbh
); /* once due to original getblk in do_journal_end */
1107 atomic_dec(&(jl
->j_commit_left
));
1110 BUG_ON(atomic_read(&(jl
->j_commit_left
)) != 1);
1113 /* If there was a write error in the journal - we can't commit
1114 * this transaction - it will be invalid and, if successful,
1115 * will just end up propogating the write error out to
1116 * the file system. */
1117 if (likely(!retval
&& !reiserfs_is_journal_aborted (journal
))) {
1118 if (buffer_dirty(jl
->j_commit_bh
))
1120 mark_buffer_dirty(jl
->j_commit_bh
) ;
1121 sync_dirty_buffer(jl
->j_commit_bh
) ;
1124 wait_on_buffer(jl
->j_commit_bh
);
1126 check_barrier_completion(s
, jl
->j_commit_bh
);
1128 /* If there was a write error in the journal - we can't commit this
1129 * transaction - it will be invalid and, if successful, will just end
1130 * up propogating the write error out to the filesystem. */
1131 if (unlikely(!buffer_uptodate(jl
->j_commit_bh
))) {
1132 #ifdef CONFIG_REISERFS_CHECK
1133 reiserfs_warning(s
, "journal-615: buffer write failed");
1137 bforget(jl
->j_commit_bh
);
1138 if (journal
->j_last_commit_id
!= 0 &&
1139 (jl
->j_trans_id
- journal
->j_last_commit_id
) != 1) {
1140 reiserfs_warning(s
, "clm-2200: last commit %lu, current %lu",
1141 journal
->j_last_commit_id
, jl
->j_trans_id
);
1143 journal
->j_last_commit_id
= jl
->j_trans_id
;
1145 /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */
1146 cleanup_freed_for_journal_list(s
, jl
);
1148 retval
= retval
? retval
: journal
->j_errno
;
1150 /* mark the metadata dirty */
1152 dirty_one_transaction(s
, jl
);
1153 atomic_dec(&(jl
->j_commit_left
));
1156 atomic_set(&(jl
->j_older_commits_done
), 1);
1158 up(&jl
->j_commit_lock
);
1160 put_journal_list(s
, jl
);
1163 reiserfs_abort(s
, retval
, "Journal write error in %s",
1170 ** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or
1171 ** returns NULL if it can't find anything
1173 static struct reiserfs_journal_list
*find_newer_jl_for_cn(struct
1174 reiserfs_journal_cnode
1177 struct super_block
*sb
= cn
->sb
;
1178 b_blocknr_t blocknr
= cn
->blocknr
;
1182 if (cn
->sb
== sb
&& cn
->blocknr
== blocknr
&& cn
->jlist
) {
1190 static void remove_journal_hash(struct super_block
*,
1191 struct reiserfs_journal_cnode
**,
1192 struct reiserfs_journal_list
*, unsigned long,
1196 ** once all the real blocks have been flushed, it is safe to remove them from the
1197 ** journal list for this transaction. Aside from freeing the cnode, this also allows the
1198 ** block to be reallocated for data blocks if it had been deleted.
1200 static void remove_all_from_journal_list(struct super_block
*p_s_sb
,
1201 struct reiserfs_journal_list
*jl
,
1204 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
1205 struct reiserfs_journal_cnode
*cn
, *last
;
1206 cn
= jl
->j_realblock
;
1208 /* which is better, to lock once around the whole loop, or
1209 ** to lock for each call to remove_journal_hash?
1212 if (cn
->blocknr
!= 0) {
1214 reiserfs_warning(p_s_sb
,
1215 "block %u, bh is %d, state %ld",
1216 cn
->blocknr
, cn
->bh
? 1 : 0,
1220 remove_journal_hash(p_s_sb
, journal
->j_list_hash_table
,
1221 jl
, cn
->blocknr
, 1);
1225 free_cnode(p_s_sb
, last
);
1227 jl
->j_realblock
= NULL
;
1231 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1232 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1233 ** releasing blocks in this transaction for reuse as data blocks.
1234 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1237 static int _update_journal_header_block(struct super_block
*p_s_sb
,
1238 unsigned long offset
,
1239 unsigned long trans_id
)
1241 struct reiserfs_journal_header
*jh
;
1242 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
1244 if (reiserfs_is_journal_aborted(journal
))
1247 if (trans_id
>= journal
->j_last_flush_trans_id
) {
1248 if (buffer_locked((journal
->j_header_bh
))) {
1249 wait_on_buffer((journal
->j_header_bh
));
1250 if (unlikely(!buffer_uptodate(journal
->j_header_bh
))) {
1251 #ifdef CONFIG_REISERFS_CHECK
1252 reiserfs_warning(p_s_sb
,
1253 "journal-699: buffer write failed");
1258 journal
->j_last_flush_trans_id
= trans_id
;
1259 journal
->j_first_unflushed_offset
= offset
;
1260 jh
= (struct reiserfs_journal_header
*)(journal
->j_header_bh
->
1262 jh
->j_last_flush_trans_id
= cpu_to_le32(trans_id
);
1263 jh
->j_first_unflushed_offset
= cpu_to_le32(offset
);
1264 jh
->j_mount_id
= cpu_to_le32(journal
->j_mount_id
);
1266 if (reiserfs_barrier_flush(p_s_sb
)) {
1268 lock_buffer(journal
->j_header_bh
);
1269 ret
= submit_barrier_buffer(journal
->j_header_bh
);
1270 if (ret
== -EOPNOTSUPP
) {
1271 set_buffer_uptodate(journal
->j_header_bh
);
1272 disable_barrier(p_s_sb
);
1275 wait_on_buffer(journal
->j_header_bh
);
1276 check_barrier_completion(p_s_sb
, journal
->j_header_bh
);
1279 set_buffer_dirty(journal
->j_header_bh
);
1280 sync_dirty_buffer(journal
->j_header_bh
);
1282 if (!buffer_uptodate(journal
->j_header_bh
)) {
1283 reiserfs_warning(p_s_sb
,
1284 "journal-837: IO error during journal replay");
1291 static int update_journal_header_block(struct super_block
*p_s_sb
,
1292 unsigned long offset
,
1293 unsigned long trans_id
)
1295 return _update_journal_header_block(p_s_sb
, offset
, trans_id
);
1299 ** flush any and all journal lists older than you are
1300 ** can only be called from flush_journal_list
1302 static int flush_older_journal_lists(struct super_block
*p_s_sb
,
1303 struct reiserfs_journal_list
*jl
)
1305 struct list_head
*entry
;
1306 struct reiserfs_journal_list
*other_jl
;
1307 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
1308 unsigned long trans_id
= jl
->j_trans_id
;
1310 /* we know we are the only ones flushing things, no extra race
1311 * protection is required.
1314 entry
= journal
->j_journal_list
.next
;
1316 if (entry
== &journal
->j_journal_list
)
1318 other_jl
= JOURNAL_LIST_ENTRY(entry
);
1319 if (other_jl
->j_trans_id
< trans_id
) {
1320 BUG_ON(other_jl
->j_refcount
<= 0);
1321 /* do not flush all */
1322 flush_journal_list(p_s_sb
, other_jl
, 0);
1324 /* other_jl is now deleted from the list */
1330 static void del_from_work_list(struct super_block
*s
,
1331 struct reiserfs_journal_list
*jl
)
1333 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1334 if (!list_empty(&jl
->j_working_list
)) {
1335 list_del_init(&jl
->j_working_list
);
1336 journal
->j_num_work_lists
--;
1340 /* flush a journal list, both commit and real blocks
1342 ** always set flushall to 1, unless you are calling from inside
1343 ** flush_journal_list
1345 ** IMPORTANT. This can only be called while there are no journal writers,
1346 ** and the journal is locked. That means it can only be called from
1347 ** do_journal_end, or by journal_release
1349 static int flush_journal_list(struct super_block
*s
,
1350 struct reiserfs_journal_list
*jl
, int flushall
)
1352 struct reiserfs_journal_list
*pjl
;
1353 struct reiserfs_journal_cnode
*cn
, *last
;
1357 struct buffer_head
*saved_bh
;
1358 unsigned long j_len_saved
= jl
->j_len
;
1359 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1362 BUG_ON(j_len_saved
<= 0);
1364 if (atomic_read(&journal
->j_wcount
) != 0) {
1366 "clm-2048: flush_journal_list called with wcount %d",
1367 atomic_read(&journal
->j_wcount
));
1369 BUG_ON(jl
->j_trans_id
== 0);
1371 /* if flushall == 0, the lock is already held */
1373 down(&journal
->j_flush_sem
);
1374 } else if (!down_trylock(&journal
->j_flush_sem
)) {
1379 if (j_len_saved
> journal
->j_trans_max
) {
1381 "journal-715: flush_journal_list, length is %lu, trans id %lu\n",
1382 j_len_saved
, jl
->j_trans_id
);
1388 /* if all the work is already done, get out of here */
1389 if (atomic_read(&(jl
->j_nonzerolen
)) <= 0 &&
1390 atomic_read(&(jl
->j_commit_left
)) <= 0) {
1391 goto flush_older_and_return
;
1394 /* start by putting the commit list on disk. This will also flush
1395 ** the commit lists of any olders transactions
1397 flush_commit_list(s
, jl
, 1);
1399 if (!(jl
->j_state
& LIST_DIRTY
)
1400 && !reiserfs_is_journal_aborted(journal
))
1403 /* are we done now? */
1404 if (atomic_read(&(jl
->j_nonzerolen
)) <= 0 &&
1405 atomic_read(&(jl
->j_commit_left
)) <= 0) {
1406 goto flush_older_and_return
;
1409 /* loop through each cnode, see if we need to write it,
1410 ** or wait on a more recent transaction, or just ignore it
1412 if (atomic_read(&(journal
->j_wcount
)) != 0) {
1414 "journal-844: panic journal list is flushing, wcount is not 0\n");
1416 cn
= jl
->j_realblock
;
1421 /* blocknr of 0 is no longer in the hash, ignore it */
1422 if (cn
->blocknr
== 0) {
1426 /* This transaction failed commit. Don't write out to the disk */
1427 if (!(jl
->j_state
& LIST_DIRTY
))
1430 pjl
= find_newer_jl_for_cn(cn
);
1431 /* the order is important here. We check pjl to make sure we
1432 ** don't clear BH_JDirty_wait if we aren't the one writing this
1435 if (!pjl
&& cn
->bh
) {
1438 /* we do this to make sure nobody releases the buffer while
1439 ** we are working with it
1443 if (buffer_journal_dirty(saved_bh
)) {
1444 BUG_ON(!can_dirty(cn
));
1447 } else if (can_dirty(cn
)) {
1448 /* everything with !pjl && jwait should be writable */
1453 /* if someone has this block in a newer transaction, just make
1454 ** sure they are commited, and don't try writing it to disk
1457 if (atomic_read(&pjl
->j_commit_left
))
1458 flush_commit_list(s
, pjl
, 1);
1462 /* bh == NULL when the block got to disk on its own, OR,
1463 ** the block got freed in a future transaction
1465 if (saved_bh
== NULL
) {
1469 /* this should never happen. kupdate_one_transaction has this list
1470 ** locked while it works, so we should never see a buffer here that
1471 ** is not marked JDirty_wait
1473 if ((!was_jwait
) && !buffer_locked(saved_bh
)) {
1475 "journal-813: BAD! buffer %llu %cdirty %cjwait, "
1476 "not in a newer tranasction",
1477 (unsigned long long)saved_bh
->
1478 b_blocknr
, was_dirty
? ' ' : '!',
1479 was_jwait
? ' ' : '!');
1482 /* we inc again because saved_bh gets decremented at free_cnode */
1484 set_bit(BLOCK_NEEDS_FLUSH
, &cn
->state
);
1485 lock_buffer(saved_bh
);
1486 BUG_ON(cn
->blocknr
!= saved_bh
->b_blocknr
);
1487 if (buffer_dirty(saved_bh
))
1488 submit_logged_buffer(saved_bh
);
1490 unlock_buffer(saved_bh
);
1494 "clm-2082: Unable to flush buffer %llu in %s",
1495 (unsigned long long)saved_bh
->
1496 b_blocknr
, __FUNCTION__
);
1502 /* we incremented this to keep others from taking the buffer head away */
1504 if (atomic_read(&(saved_bh
->b_count
)) < 0) {
1506 "journal-945: saved_bh->b_count < 0");
1511 cn
= jl
->j_realblock
;
1513 if (test_bit(BLOCK_NEEDS_FLUSH
, &cn
->state
)) {
1516 "journal-1011: cn->bh is NULL\n");
1518 wait_on_buffer(cn
->bh
);
1521 "journal-1012: cn->bh is NULL\n");
1523 if (unlikely(!buffer_uptodate(cn
->bh
))) {
1524 #ifdef CONFIG_REISERFS_CHECK
1526 "journal-949: buffer write failed\n");
1530 /* note, we must clear the JDirty_wait bit after the up to date
1531 ** check, otherwise we race against our flushpage routine
1533 BUG_ON(!test_clear_buffer_journal_dirty
1536 /* undo the inc from journal_mark_dirty */
1545 reiserfs_abort(s
, -EIO
,
1546 "Write error while pushing transaction to disk in %s",
1548 flush_older_and_return
:
1550 /* before we can update the journal header block, we _must_ flush all
1551 ** real blocks from all older transactions to disk. This is because
1552 ** once the header block is updated, this transaction will not be
1553 ** replayed after a crash
1556 flush_older_journal_lists(s
, jl
);
1559 err
= journal
->j_errno
;
1560 /* before we can remove everything from the hash tables for this
1561 ** transaction, we must make sure it can never be replayed
1563 ** since we are only called from do_journal_end, we know for sure there
1564 ** are no allocations going on while we are flushing journal lists. So,
1565 ** we only need to update the journal header block for the last list
1568 if (!err
&& flushall
) {
1570 update_journal_header_block(s
,
1571 (jl
->j_start
+ jl
->j_len
+
1572 2) % SB_ONDISK_JOURNAL_SIZE(s
),
1575 reiserfs_abort(s
, -EIO
,
1576 "Write error while updating journal header in %s",
1579 remove_all_from_journal_list(s
, jl
, 0);
1580 list_del_init(&jl
->j_list
);
1581 journal
->j_num_lists
--;
1582 del_from_work_list(s
, jl
);
1584 if (journal
->j_last_flush_id
!= 0 &&
1585 (jl
->j_trans_id
- journal
->j_last_flush_id
) != 1) {
1586 reiserfs_warning(s
, "clm-2201: last flush %lu, current %lu",
1587 journal
->j_last_flush_id
, jl
->j_trans_id
);
1589 journal
->j_last_flush_id
= jl
->j_trans_id
;
1591 /* not strictly required since we are freeing the list, but it should
1592 * help find code using dead lists later on
1595 atomic_set(&(jl
->j_nonzerolen
), 0);
1597 jl
->j_realblock
= NULL
;
1598 jl
->j_commit_bh
= NULL
;
1601 put_journal_list(s
, jl
);
1603 up(&journal
->j_flush_sem
);
1608 static int write_one_transaction(struct super_block
*s
,
1609 struct reiserfs_journal_list
*jl
,
1610 struct buffer_chunk
*chunk
)
1612 struct reiserfs_journal_cnode
*cn
;
1615 jl
->j_state
|= LIST_TOUCHED
;
1616 del_from_work_list(s
, jl
);
1617 if (jl
->j_len
== 0 || atomic_read(&jl
->j_nonzerolen
) == 0) {
1621 cn
= jl
->j_realblock
;
1623 /* if the blocknr == 0, this has been cleared from the hash,
1626 if (cn
->blocknr
== 0) {
1629 if (cn
->bh
&& can_dirty(cn
) && buffer_dirty(cn
->bh
)) {
1630 struct buffer_head
*tmp_bh
;
1631 /* we can race against journal_mark_freed when we try
1632 * to lock_buffer(cn->bh), so we have to inc the buffer
1633 * count, and recheck things after locking
1637 lock_buffer(tmp_bh
);
1638 if (cn
->bh
&& can_dirty(cn
) && buffer_dirty(tmp_bh
)) {
1639 if (!buffer_journal_dirty(tmp_bh
) ||
1640 buffer_journal_prepared(tmp_bh
))
1642 add_to_chunk(chunk
, tmp_bh
, NULL
, write_chunk
);
1645 /* note, cn->bh might be null now */
1646 unlock_buffer(tmp_bh
);
1657 /* used by flush_commit_list */
1658 static int dirty_one_transaction(struct super_block
*s
,
1659 struct reiserfs_journal_list
*jl
)
1661 struct reiserfs_journal_cnode
*cn
;
1662 struct reiserfs_journal_list
*pjl
;
1665 jl
->j_state
|= LIST_DIRTY
;
1666 cn
= jl
->j_realblock
;
1668 /* look for a more recent transaction that logged this
1669 ** buffer. Only the most recent transaction with a buffer in
1670 ** it is allowed to send that buffer to disk
1672 pjl
= find_newer_jl_for_cn(cn
);
1673 if (!pjl
&& cn
->blocknr
&& cn
->bh
1674 && buffer_journal_dirty(cn
->bh
)) {
1675 BUG_ON(!can_dirty(cn
));
1676 /* if the buffer is prepared, it will either be logged
1677 * or restored. If restored, we need to make sure
1678 * it actually gets marked dirty
1680 clear_buffer_journal_new(cn
->bh
);
1681 if (buffer_journal_prepared(cn
->bh
)) {
1682 set_buffer_journal_restore_dirty(cn
->bh
);
1684 set_buffer_journal_test(cn
->bh
);
1685 mark_buffer_dirty(cn
->bh
);
1693 static int kupdate_transactions(struct super_block
*s
,
1694 struct reiserfs_journal_list
*jl
,
1695 struct reiserfs_journal_list
**next_jl
,
1696 unsigned long *next_trans_id
,
1697 int num_blocks
, int num_trans
)
1701 int transactions_flushed
= 0;
1702 unsigned long orig_trans_id
= jl
->j_trans_id
;
1703 struct buffer_chunk chunk
;
1704 struct list_head
*entry
;
1705 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1708 down(&journal
->j_flush_sem
);
1709 if (!journal_list_still_alive(s
, orig_trans_id
)) {
1713 /* we've got j_flush_sem held, nobody is going to delete any
1714 * of these lists out from underneath us
1716 while ((num_trans
&& transactions_flushed
< num_trans
) ||
1717 (!num_trans
&& written
< num_blocks
)) {
1719 if (jl
->j_len
== 0 || (jl
->j_state
& LIST_TOUCHED
) ||
1720 atomic_read(&jl
->j_commit_left
)
1721 || !(jl
->j_state
& LIST_DIRTY
)) {
1722 del_from_work_list(s
, jl
);
1725 ret
= write_one_transaction(s
, jl
, &chunk
);
1729 transactions_flushed
++;
1731 entry
= jl
->j_list
.next
;
1734 if (entry
== &journal
->j_journal_list
) {
1737 jl
= JOURNAL_LIST_ENTRY(entry
);
1739 /* don't bother with older transactions */
1740 if (jl
->j_trans_id
<= orig_trans_id
)
1744 write_chunk(&chunk
);
1748 up(&journal
->j_flush_sem
);
1752 /* for o_sync and fsync heavy applications, they tend to use
1753 ** all the journa list slots with tiny transactions. These
1754 ** trigger lots and lots of calls to update the header block, which
1755 ** adds seeks and slows things down.
1757 ** This function tries to clear out a large chunk of the journal lists
1758 ** at once, which makes everything faster since only the newest journal
1759 ** list updates the header block
1761 static int flush_used_journal_lists(struct super_block
*s
,
1762 struct reiserfs_journal_list
*jl
)
1764 unsigned long len
= 0;
1765 unsigned long cur_len
;
1769 struct reiserfs_journal_list
*tjl
;
1770 struct reiserfs_journal_list
*flush_jl
;
1771 unsigned long trans_id
;
1772 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1774 flush_jl
= tjl
= jl
;
1776 /* in data logging mode, try harder to flush a lot of blocks */
1777 if (reiserfs_data_log(s
))
1779 /* flush for 256 transactions or limit blocks, whichever comes first */
1780 for (i
= 0; i
< 256 && len
< limit
; i
++) {
1781 if (atomic_read(&tjl
->j_commit_left
) ||
1782 tjl
->j_trans_id
< jl
->j_trans_id
) {
1785 cur_len
= atomic_read(&tjl
->j_nonzerolen
);
1787 tjl
->j_state
&= ~LIST_TOUCHED
;
1791 if (tjl
->j_list
.next
== &journal
->j_journal_list
)
1793 tjl
= JOURNAL_LIST_ENTRY(tjl
->j_list
.next
);
1795 /* try to find a group of blocks we can flush across all the
1796 ** transactions, but only bother if we've actually spanned
1797 ** across multiple lists
1799 if (flush_jl
!= jl
) {
1800 ret
= kupdate_transactions(s
, jl
, &tjl
, &trans_id
, len
, i
);
1802 flush_journal_list(s
, flush_jl
, 1);
1807 ** removes any nodes in table with name block and dev as bh.
1808 ** only touchs the hnext and hprev pointers.
1810 void remove_journal_hash(struct super_block
*sb
,
1811 struct reiserfs_journal_cnode
**table
,
1812 struct reiserfs_journal_list
*jl
,
1813 unsigned long block
, int remove_freed
)
1815 struct reiserfs_journal_cnode
*cur
;
1816 struct reiserfs_journal_cnode
**head
;
1818 head
= &(journal_hash(table
, sb
, block
));
1824 if (cur
->blocknr
== block
&& cur
->sb
== sb
1825 && (jl
== NULL
|| jl
== cur
->jlist
)
1826 && (!test_bit(BLOCK_FREED
, &cur
->state
) || remove_freed
)) {
1828 cur
->hnext
->hprev
= cur
->hprev
;
1831 cur
->hprev
->hnext
= cur
->hnext
;
1838 if (cur
->bh
&& cur
->jlist
) /* anybody who clears the cur->bh will also dec the nonzerolen */
1839 atomic_dec(&(cur
->jlist
->j_nonzerolen
));
1847 static void free_journal_ram(struct super_block
*p_s_sb
)
1849 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
1850 kfree(journal
->j_current_jl
);
1851 journal
->j_num_lists
--;
1853 vfree(journal
->j_cnode_free_orig
);
1854 free_list_bitmaps(p_s_sb
, journal
->j_list_bitmap
);
1855 free_bitmap_nodes(p_s_sb
); /* must be after free_list_bitmaps */
1856 if (journal
->j_header_bh
) {
1857 brelse(journal
->j_header_bh
);
1859 /* j_header_bh is on the journal dev, make sure not to release the journal
1860 * dev until we brelse j_header_bh
1862 release_journal_dev(p_s_sb
, journal
);
1867 ** call on unmount. Only set error to 1 if you haven't made your way out
1868 ** of read_super() yet. Any other caller must keep error at 0.
1870 static int do_journal_release(struct reiserfs_transaction_handle
*th
,
1871 struct super_block
*p_s_sb
, int error
)
1873 struct reiserfs_transaction_handle myth
;
1875 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
1877 /* we only want to flush out transactions if we were called with error == 0
1879 if (!error
&& !(p_s_sb
->s_flags
& MS_RDONLY
)) {
1880 /* end the current trans */
1881 BUG_ON(!th
->t_trans_id
);
1882 do_journal_end(th
, p_s_sb
, 10, FLUSH_ALL
);
1884 /* make sure something gets logged to force our way into the flush code */
1885 if (!journal_join(&myth
, p_s_sb
, 1)) {
1886 reiserfs_prepare_for_journal(p_s_sb
,
1887 SB_BUFFER_WITH_SB(p_s_sb
),
1889 journal_mark_dirty(&myth
, p_s_sb
,
1890 SB_BUFFER_WITH_SB(p_s_sb
));
1891 do_journal_end(&myth
, p_s_sb
, 1, FLUSH_ALL
);
1896 /* this also catches errors during the do_journal_end above */
1897 if (!error
&& reiserfs_is_journal_aborted(journal
)) {
1898 memset(&myth
, 0, sizeof(myth
));
1899 if (!journal_join_abort(&myth
, p_s_sb
, 1)) {
1900 reiserfs_prepare_for_journal(p_s_sb
,
1901 SB_BUFFER_WITH_SB(p_s_sb
),
1903 journal_mark_dirty(&myth
, p_s_sb
,
1904 SB_BUFFER_WITH_SB(p_s_sb
));
1905 do_journal_end(&myth
, p_s_sb
, 1, FLUSH_ALL
);
1909 reiserfs_mounted_fs_count
--;
1910 /* wait for all commits to finish */
1911 cancel_delayed_work(&SB_JOURNAL(p_s_sb
)->j_work
);
1912 flush_workqueue(commit_wq
);
1913 if (!reiserfs_mounted_fs_count
) {
1914 destroy_workqueue(commit_wq
);
1918 free_journal_ram(p_s_sb
);
1924 ** call on unmount. flush all journal trans, release all alloc'd ram
1926 int journal_release(struct reiserfs_transaction_handle
*th
,
1927 struct super_block
*p_s_sb
)
1929 return do_journal_release(th
, p_s_sb
, 0);
1933 ** only call from an error condition inside reiserfs_read_super!
1935 int journal_release_error(struct reiserfs_transaction_handle
*th
,
1936 struct super_block
*p_s_sb
)
1938 return do_journal_release(th
, p_s_sb
, 1);
1941 /* compares description block with commit block. returns 1 if they differ, 0 if they are the same */
1942 static int journal_compare_desc_commit(struct super_block
*p_s_sb
,
1943 struct reiserfs_journal_desc
*desc
,
1944 struct reiserfs_journal_commit
*commit
)
1946 if (get_commit_trans_id(commit
) != get_desc_trans_id(desc
) ||
1947 get_commit_trans_len(commit
) != get_desc_trans_len(desc
) ||
1948 get_commit_trans_len(commit
) > SB_JOURNAL(p_s_sb
)->j_trans_max
||
1949 get_commit_trans_len(commit
) <= 0) {
1955 /* returns 0 if it did not find a description block
1956 ** returns -1 if it found a corrupt commit block
1957 ** returns 1 if both desc and commit were valid
1959 static int journal_transaction_is_valid(struct super_block
*p_s_sb
,
1960 struct buffer_head
*d_bh
,
1961 unsigned long *oldest_invalid_trans_id
,
1962 unsigned long *newest_mount_id
)
1964 struct reiserfs_journal_desc
*desc
;
1965 struct reiserfs_journal_commit
*commit
;
1966 struct buffer_head
*c_bh
;
1967 unsigned long offset
;
1972 desc
= (struct reiserfs_journal_desc
*)d_bh
->b_data
;
1973 if (get_desc_trans_len(desc
) > 0
1974 && !memcmp(get_journal_desc_magic(d_bh
), JOURNAL_DESC_MAGIC
, 8)) {
1975 if (oldest_invalid_trans_id
&& *oldest_invalid_trans_id
1976 && get_desc_trans_id(desc
) > *oldest_invalid_trans_id
) {
1977 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
1978 "journal-986: transaction "
1979 "is valid returning because trans_id %d is greater than "
1980 "oldest_invalid %lu",
1981 get_desc_trans_id(desc
),
1982 *oldest_invalid_trans_id
);
1986 && *newest_mount_id
> get_desc_mount_id(desc
)) {
1987 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
1988 "journal-1087: transaction "
1989 "is valid returning because mount_id %d is less than "
1990 "newest_mount_id %lu",
1991 get_desc_mount_id(desc
),
1995 if (get_desc_trans_len(desc
) > SB_JOURNAL(p_s_sb
)->j_trans_max
) {
1996 reiserfs_warning(p_s_sb
,
1997 "journal-2018: Bad transaction length %d encountered, ignoring transaction",
1998 get_desc_trans_len(desc
));
2001 offset
= d_bh
->b_blocknr
- SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
);
2003 /* ok, we have a journal description block, lets see if the transaction was valid */
2005 journal_bread(p_s_sb
,
2006 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
2007 ((offset
+ get_desc_trans_len(desc
) +
2008 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb
)));
2011 commit
= (struct reiserfs_journal_commit
*)c_bh
->b_data
;
2012 if (journal_compare_desc_commit(p_s_sb
, desc
, commit
)) {
2013 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2014 "journal_transaction_is_valid, commit offset %ld had bad "
2015 "time %d or length %d",
2017 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
),
2018 get_commit_trans_id(commit
),
2019 get_commit_trans_len(commit
));
2021 if (oldest_invalid_trans_id
) {
2022 *oldest_invalid_trans_id
=
2023 get_desc_trans_id(desc
);
2024 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2026 "transaction_is_valid setting oldest invalid trans_id "
2028 get_desc_trans_id(desc
));
2033 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2034 "journal-1006: found valid "
2035 "transaction start offset %llu, len %d id %d",
2037 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
),
2038 get_desc_trans_len(desc
),
2039 get_desc_trans_id(desc
));
2046 static void brelse_array(struct buffer_head
**heads
, int num
)
2049 for (i
= 0; i
< num
; i
++) {
2055 ** given the start, and values for the oldest acceptable transactions,
2056 ** this either reads in a replays a transaction, or returns because the transaction
2057 ** is invalid, or too old.
2059 static int journal_read_transaction(struct super_block
*p_s_sb
,
2060 unsigned long cur_dblock
,
2061 unsigned long oldest_start
,
2062 unsigned long oldest_trans_id
,
2063 unsigned long newest_mount_id
)
2065 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
2066 struct reiserfs_journal_desc
*desc
;
2067 struct reiserfs_journal_commit
*commit
;
2068 unsigned long trans_id
= 0;
2069 struct buffer_head
*c_bh
;
2070 struct buffer_head
*d_bh
;
2071 struct buffer_head
**log_blocks
= NULL
;
2072 struct buffer_head
**real_blocks
= NULL
;
2073 unsigned long trans_offset
;
2077 d_bh
= journal_bread(p_s_sb
, cur_dblock
);
2080 desc
= (struct reiserfs_journal_desc
*)d_bh
->b_data
;
2081 trans_offset
= d_bh
->b_blocknr
- SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
);
2082 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
, "journal-1037: "
2083 "journal_read_transaction, offset %llu, len %d mount_id %d",
2084 d_bh
->b_blocknr
- SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
),
2085 get_desc_trans_len(desc
), get_desc_mount_id(desc
));
2086 if (get_desc_trans_id(desc
) < oldest_trans_id
) {
2087 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
, "journal-1039: "
2088 "journal_read_trans skipping because %lu is too old",
2090 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
));
2094 if (get_desc_mount_id(desc
) != newest_mount_id
) {
2095 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
, "journal-1146: "
2096 "journal_read_trans skipping because %d is != "
2097 "newest_mount_id %lu", get_desc_mount_id(desc
),
2102 c_bh
= journal_bread(p_s_sb
, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
2103 ((trans_offset
+ get_desc_trans_len(desc
) + 1) %
2104 SB_ONDISK_JOURNAL_SIZE(p_s_sb
)));
2109 commit
= (struct reiserfs_journal_commit
*)c_bh
->b_data
;
2110 if (journal_compare_desc_commit(p_s_sb
, desc
, commit
)) {
2111 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2112 "journal_read_transaction, "
2113 "commit offset %llu had bad time %d or length %d",
2115 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
),
2116 get_commit_trans_id(commit
),
2117 get_commit_trans_len(commit
));
2122 trans_id
= get_desc_trans_id(desc
);
2123 /* now we know we've got a good transaction, and it was inside the valid time ranges */
2124 log_blocks
= kmalloc(get_desc_trans_len(desc
) *
2125 sizeof(struct buffer_head
*), GFP_NOFS
);
2126 real_blocks
= kmalloc(get_desc_trans_len(desc
) *
2127 sizeof(struct buffer_head
*), GFP_NOFS
);
2128 if (!log_blocks
|| !real_blocks
) {
2133 reiserfs_warning(p_s_sb
,
2134 "journal-1169: kmalloc failed, unable to mount FS");
2137 /* get all the buffer heads */
2138 trans_half
= journal_trans_half(p_s_sb
->s_blocksize
);
2139 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2141 journal_getblk(p_s_sb
,
2142 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
2144 i
) % SB_ONDISK_JOURNAL_SIZE(p_s_sb
));
2145 if (i
< trans_half
) {
2148 le32_to_cpu(desc
->j_realblock
[i
]));
2152 le32_to_cpu(commit
->
2153 j_realblock
[i
- trans_half
]));
2155 if (real_blocks
[i
]->b_blocknr
> SB_BLOCK_COUNT(p_s_sb
)) {
2156 reiserfs_warning(p_s_sb
,
2157 "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem");
2160 /* make sure we don't try to replay onto log or reserved area */
2161 if (is_block_in_log_or_reserved_area
2162 (p_s_sb
, real_blocks
[i
]->b_blocknr
)) {
2163 reiserfs_warning(p_s_sb
,
2164 "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block");
2166 brelse_array(log_blocks
, i
);
2167 brelse_array(real_blocks
, i
);
2175 /* read in the log blocks, memcpy to the corresponding real block */
2176 ll_rw_block(READ
, get_desc_trans_len(desc
), log_blocks
);
2177 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2178 wait_on_buffer(log_blocks
[i
]);
2179 if (!buffer_uptodate(log_blocks
[i
])) {
2180 reiserfs_warning(p_s_sb
,
2181 "journal-1212: REPLAY FAILURE fsck required! buffer write failed");
2182 brelse_array(log_blocks
+ i
,
2183 get_desc_trans_len(desc
) - i
);
2184 brelse_array(real_blocks
, get_desc_trans_len(desc
));
2191 memcpy(real_blocks
[i
]->b_data
, log_blocks
[i
]->b_data
,
2192 real_blocks
[i
]->b_size
);
2193 set_buffer_uptodate(real_blocks
[i
]);
2194 brelse(log_blocks
[i
]);
2196 /* flush out the real blocks */
2197 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2198 set_buffer_dirty(real_blocks
[i
]);
2199 ll_rw_block(SWRITE
, 1, real_blocks
+ i
);
2201 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2202 wait_on_buffer(real_blocks
[i
]);
2203 if (!buffer_uptodate(real_blocks
[i
])) {
2204 reiserfs_warning(p_s_sb
,
2205 "journal-1226: REPLAY FAILURE, fsck required! buffer write failed");
2206 brelse_array(real_blocks
+ i
,
2207 get_desc_trans_len(desc
) - i
);
2214 brelse(real_blocks
[i
]);
2217 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
2218 ((trans_offset
+ get_desc_trans_len(desc
) +
2219 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb
));
2220 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2221 "journal-1095: setting journal " "start to offset %ld",
2222 cur_dblock
- SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
));
2224 /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2225 journal
->j_start
= cur_dblock
- SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
);
2226 journal
->j_last_flush_trans_id
= trans_id
;
2227 journal
->j_trans_id
= trans_id
+ 1;
2228 /* check for trans_id overflow */
2229 if (journal
->j_trans_id
== 0)
2230 journal
->j_trans_id
= 10;
2238 /* This function reads blocks starting from block and to max_block of bufsize
2239 size (but no more than BUFNR blocks at a time). This proved to improve
2240 mounting speed on self-rebuilding raid5 arrays at least.
2241 Right now it is only used from journal code. But later we might use it
2243 Note: Do not use journal_getblk/sb_getblk functions here! */
2244 static struct buffer_head
*reiserfs_breada(struct block_device
*dev
, int block
,
2245 int bufsize
, unsigned int max_block
)
2247 struct buffer_head
*bhlist
[BUFNR
];
2248 unsigned int blocks
= BUFNR
;
2249 struct buffer_head
*bh
;
2252 bh
= __getblk(dev
, block
, bufsize
);
2253 if (buffer_uptodate(bh
))
2256 if (block
+ BUFNR
> max_block
) {
2257 blocks
= max_block
- block
;
2261 for (i
= 1; i
< blocks
; i
++) {
2262 bh
= __getblk(dev
, block
+ i
, bufsize
);
2263 if (buffer_uptodate(bh
)) {
2269 ll_rw_block(READ
, j
, bhlist
);
2270 for (i
= 1; i
< j
; i
++)
2274 if (buffer_uptodate(bh
))
2281 ** read and replay the log
2282 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2283 ** transaction. This tests that before finding all the transactions in the log, which makes normal mount times fast.
2285 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2287 ** On exit, it sets things up so the first transaction will work correctly.
2289 static int journal_read(struct super_block
*p_s_sb
)
2291 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
2292 struct reiserfs_journal_desc
*desc
;
2293 unsigned long oldest_trans_id
= 0;
2294 unsigned long oldest_invalid_trans_id
= 0;
2296 unsigned long oldest_start
= 0;
2297 unsigned long cur_dblock
= 0;
2298 unsigned long newest_mount_id
= 9;
2299 struct buffer_head
*d_bh
;
2300 struct reiserfs_journal_header
*jh
;
2301 int valid_journal_header
= 0;
2302 int replay_count
= 0;
2303 int continue_replay
= 1;
2305 char b
[BDEVNAME_SIZE
];
2307 cur_dblock
= SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
);
2308 reiserfs_info(p_s_sb
, "checking transaction log (%s)\n",
2309 bdevname(journal
->j_dev_bd
, b
));
2310 start
= get_seconds();
2312 /* step 1, read in the journal header block. Check the transaction it says
2313 ** is the first unflushed, and if that transaction is not valid,
2316 journal
->j_header_bh
= journal_bread(p_s_sb
,
2317 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
)
2318 + SB_ONDISK_JOURNAL_SIZE(p_s_sb
));
2319 if (!journal
->j_header_bh
) {
2322 jh
= (struct reiserfs_journal_header
*)(journal
->j_header_bh
->b_data
);
2323 if (le32_to_cpu(jh
->j_first_unflushed_offset
) <
2324 SB_ONDISK_JOURNAL_SIZE(p_s_sb
)
2325 && le32_to_cpu(jh
->j_last_flush_trans_id
) > 0) {
2327 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
2328 le32_to_cpu(jh
->j_first_unflushed_offset
);
2329 oldest_trans_id
= le32_to_cpu(jh
->j_last_flush_trans_id
) + 1;
2330 newest_mount_id
= le32_to_cpu(jh
->j_mount_id
);
2331 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2332 "journal-1153: found in "
2333 "header: first_unflushed_offset %d, last_flushed_trans_id "
2334 "%lu", le32_to_cpu(jh
->j_first_unflushed_offset
),
2335 le32_to_cpu(jh
->j_last_flush_trans_id
));
2336 valid_journal_header
= 1;
2338 /* now, we try to read the first unflushed offset. If it is not valid,
2339 ** there is nothing more we can do, and it makes no sense to read
2340 ** through the whole log.
2343 journal_bread(p_s_sb
,
2344 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
2345 le32_to_cpu(jh
->j_first_unflushed_offset
));
2346 ret
= journal_transaction_is_valid(p_s_sb
, d_bh
, NULL
, NULL
);
2348 continue_replay
= 0;
2351 goto start_log_replay
;
2354 if (continue_replay
&& bdev_read_only(p_s_sb
->s_bdev
)) {
2355 reiserfs_warning(p_s_sb
,
2356 "clm-2076: device is readonly, unable to replay log");
2360 /* ok, there are transactions that need to be replayed. start with the first log block, find
2361 ** all the valid transactions, and pick out the oldest.
2363 while (continue_replay
2365 (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
2366 SB_ONDISK_JOURNAL_SIZE(p_s_sb
))) {
2367 /* Note that it is required for blocksize of primary fs device and journal
2368 device to be the same */
2370 reiserfs_breada(journal
->j_dev_bd
, cur_dblock
,
2371 p_s_sb
->s_blocksize
,
2372 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
2373 SB_ONDISK_JOURNAL_SIZE(p_s_sb
));
2375 journal_transaction_is_valid(p_s_sb
, d_bh
,
2376 &oldest_invalid_trans_id
,
2379 desc
= (struct reiserfs_journal_desc
*)d_bh
->b_data
;
2380 if (oldest_start
== 0) { /* init all oldest_ values */
2381 oldest_trans_id
= get_desc_trans_id(desc
);
2382 oldest_start
= d_bh
->b_blocknr
;
2383 newest_mount_id
= get_desc_mount_id(desc
);
2384 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2385 "journal-1179: Setting "
2386 "oldest_start to offset %llu, trans_id %lu",
2388 SB_ONDISK_JOURNAL_1st_BLOCK
2389 (p_s_sb
), oldest_trans_id
);
2390 } else if (oldest_trans_id
> get_desc_trans_id(desc
)) {
2391 /* one we just read was older */
2392 oldest_trans_id
= get_desc_trans_id(desc
);
2393 oldest_start
= d_bh
->b_blocknr
;
2394 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2395 "journal-1180: Resetting "
2396 "oldest_start to offset %lu, trans_id %lu",
2398 SB_ONDISK_JOURNAL_1st_BLOCK
2399 (p_s_sb
), oldest_trans_id
);
2401 if (newest_mount_id
< get_desc_mount_id(desc
)) {
2402 newest_mount_id
= get_desc_mount_id(desc
);
2403 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2404 "journal-1299: Setting "
2405 "newest_mount_id to %d",
2406 get_desc_mount_id(desc
));
2408 cur_dblock
+= get_desc_trans_len(desc
) + 2;
2416 cur_dblock
= oldest_start
;
2417 if (oldest_trans_id
) {
2418 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2419 "journal-1206: Starting replay "
2420 "from offset %llu, trans_id %lu",
2421 cur_dblock
- SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
),
2426 while (continue_replay
&& oldest_trans_id
> 0) {
2428 journal_read_transaction(p_s_sb
, cur_dblock
, oldest_start
,
2429 oldest_trans_id
, newest_mount_id
);
2432 } else if (ret
!= 0) {
2436 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) + journal
->j_start
;
2438 if (cur_dblock
== oldest_start
)
2442 if (oldest_trans_id
== 0) {
2443 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
,
2444 "journal-1225: No valid " "transactions found");
2446 /* j_start does not get set correctly if we don't replay any transactions.
2447 ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2448 ** copy the trans_id from the header
2450 if (valid_journal_header
&& replay_count
== 0) {
2451 journal
->j_start
= le32_to_cpu(jh
->j_first_unflushed_offset
);
2452 journal
->j_trans_id
=
2453 le32_to_cpu(jh
->j_last_flush_trans_id
) + 1;
2454 /* check for trans_id overflow */
2455 if (journal
->j_trans_id
== 0)
2456 journal
->j_trans_id
= 10;
2457 journal
->j_last_flush_trans_id
=
2458 le32_to_cpu(jh
->j_last_flush_trans_id
);
2459 journal
->j_mount_id
= le32_to_cpu(jh
->j_mount_id
) + 1;
2461 journal
->j_mount_id
= newest_mount_id
+ 1;
2463 reiserfs_debug(p_s_sb
, REISERFS_DEBUG_CODE
, "journal-1299: Setting "
2464 "newest_mount_id to %lu", journal
->j_mount_id
);
2465 journal
->j_first_unflushed_offset
= journal
->j_start
;
2466 if (replay_count
> 0) {
2467 reiserfs_info(p_s_sb
,
2468 "replayed %d transactions in %lu seconds\n",
2469 replay_count
, get_seconds() - start
);
2471 if (!bdev_read_only(p_s_sb
->s_bdev
) &&
2472 _update_journal_header_block(p_s_sb
, journal
->j_start
,
2473 journal
->j_last_flush_trans_id
)) {
2474 /* replay failed, caller must call free_journal_ram and abort
2482 static struct reiserfs_journal_list
*alloc_journal_list(struct super_block
*s
)
2484 struct reiserfs_journal_list
*jl
;
2485 jl
= kzalloc(sizeof(struct reiserfs_journal_list
),
2486 GFP_NOFS
| __GFP_NOFAIL
);
2487 INIT_LIST_HEAD(&jl
->j_list
);
2488 INIT_LIST_HEAD(&jl
->j_working_list
);
2489 INIT_LIST_HEAD(&jl
->j_tail_bh_list
);
2490 INIT_LIST_HEAD(&jl
->j_bh_list
);
2491 sema_init(&jl
->j_commit_lock
, 1);
2492 SB_JOURNAL(s
)->j_num_lists
++;
2493 get_journal_list(jl
);
2497 static void journal_list_init(struct super_block
*p_s_sb
)
2499 SB_JOURNAL(p_s_sb
)->j_current_jl
= alloc_journal_list(p_s_sb
);
2502 static int release_journal_dev(struct super_block
*super
,
2503 struct reiserfs_journal
*journal
)
2509 if (journal
->j_dev_file
!= NULL
) {
2510 result
= filp_close(journal
->j_dev_file
, NULL
);
2511 journal
->j_dev_file
= NULL
;
2512 journal
->j_dev_bd
= NULL
;
2513 } else if (journal
->j_dev_bd
!= NULL
) {
2514 result
= blkdev_put(journal
->j_dev_bd
);
2515 journal
->j_dev_bd
= NULL
;
2519 reiserfs_warning(super
,
2520 "sh-457: release_journal_dev: Cannot release journal device: %i",
2526 static int journal_init_dev(struct super_block
*super
,
2527 struct reiserfs_journal
*journal
,
2528 const char *jdev_name
)
2532 int blkdev_mode
= FMODE_READ
| FMODE_WRITE
;
2533 char b
[BDEVNAME_SIZE
];
2537 journal
->j_dev_bd
= NULL
;
2538 journal
->j_dev_file
= NULL
;
2539 jdev
= SB_ONDISK_JOURNAL_DEVICE(super
) ?
2540 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super
)) : super
->s_dev
;
2542 if (bdev_read_only(super
->s_bdev
))
2543 blkdev_mode
= FMODE_READ
;
2545 /* there is no "jdev" option and journal is on separate device */
2546 if ((!jdev_name
|| !jdev_name
[0])) {
2547 journal
->j_dev_bd
= open_by_devnum(jdev
, blkdev_mode
);
2548 if (IS_ERR(journal
->j_dev_bd
)) {
2549 result
= PTR_ERR(journal
->j_dev_bd
);
2550 journal
->j_dev_bd
= NULL
;
2551 reiserfs_warning(super
, "sh-458: journal_init_dev: "
2552 "cannot init journal device '%s': %i",
2553 __bdevname(jdev
, b
), result
);
2555 } else if (jdev
!= super
->s_dev
)
2556 set_blocksize(journal
->j_dev_bd
, super
->s_blocksize
);
2560 journal
->j_dev_file
= filp_open(jdev_name
, 0, 0);
2561 if (!IS_ERR(journal
->j_dev_file
)) {
2562 struct inode
*jdev_inode
= journal
->j_dev_file
->f_mapping
->host
;
2563 if (!S_ISBLK(jdev_inode
->i_mode
)) {
2564 reiserfs_warning(super
, "journal_init_dev: '%s' is "
2565 "not a block device", jdev_name
);
2567 release_journal_dev(super
, journal
);
2570 journal
->j_dev_bd
= I_BDEV(jdev_inode
);
2571 set_blocksize(journal
->j_dev_bd
, super
->s_blocksize
);
2572 reiserfs_info(super
,
2573 "journal_init_dev: journal device: %s\n",
2574 bdevname(journal
->j_dev_bd
, b
));
2577 result
= PTR_ERR(journal
->j_dev_file
);
2578 journal
->j_dev_file
= NULL
;
2579 reiserfs_warning(super
,
2580 "journal_init_dev: Cannot open '%s': %i",
2587 ** must be called once on fs mount. calls journal_read for you
2589 int journal_init(struct super_block
*p_s_sb
, const char *j_dev_name
,
2590 int old_format
, unsigned int commit_max_age
)
2592 int num_cnodes
= SB_ONDISK_JOURNAL_SIZE(p_s_sb
) * 2;
2593 struct buffer_head
*bhjh
;
2594 struct reiserfs_super_block
*rs
;
2595 struct reiserfs_journal_header
*jh
;
2596 struct reiserfs_journal
*journal
;
2597 struct reiserfs_journal_list
*jl
;
2598 char b
[BDEVNAME_SIZE
];
2600 journal
= SB_JOURNAL(p_s_sb
) = vmalloc(sizeof(struct reiserfs_journal
));
2602 reiserfs_warning(p_s_sb
,
2603 "journal-1256: unable to get memory for journal structure");
2606 memset(journal
, 0, sizeof(struct reiserfs_journal
));
2607 INIT_LIST_HEAD(&journal
->j_bitmap_nodes
);
2608 INIT_LIST_HEAD(&journal
->j_prealloc_list
);
2609 INIT_LIST_HEAD(&journal
->j_working_list
);
2610 INIT_LIST_HEAD(&journal
->j_journal_list
);
2611 journal
->j_persistent_trans
= 0;
2612 if (reiserfs_allocate_list_bitmaps(p_s_sb
,
2613 journal
->j_list_bitmap
,
2614 SB_BMAP_NR(p_s_sb
)))
2615 goto free_and_return
;
2616 allocate_bitmap_nodes(p_s_sb
);
2618 /* reserved for journal area support */
2619 SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb
) = (old_format
?
2620 REISERFS_OLD_DISK_OFFSET_IN_BYTES
2621 / p_s_sb
->s_blocksize
+
2622 SB_BMAP_NR(p_s_sb
) +
2624 REISERFS_DISK_OFFSET_IN_BYTES
/
2625 p_s_sb
->s_blocksize
+ 2);
2627 /* Sanity check to see is the standard journal fitting withing first bitmap
2628 (actual for small blocksizes) */
2629 if (!SB_ONDISK_JOURNAL_DEVICE(p_s_sb
) &&
2630 (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb
) +
2631 SB_ONDISK_JOURNAL_SIZE(p_s_sb
) > p_s_sb
->s_blocksize
* 8)) {
2632 reiserfs_warning(p_s_sb
,
2633 "journal-1393: journal does not fit for area "
2634 "addressed by first of bitmap blocks. It starts at "
2635 "%u and its size is %u. Block size %ld",
2636 SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb
),
2637 SB_ONDISK_JOURNAL_SIZE(p_s_sb
),
2638 p_s_sb
->s_blocksize
);
2639 goto free_and_return
;
2642 if (journal_init_dev(p_s_sb
, journal
, j_dev_name
) != 0) {
2643 reiserfs_warning(p_s_sb
,
2644 "sh-462: unable to initialize jornal device");
2645 goto free_and_return
;
2648 rs
= SB_DISK_SUPER_BLOCK(p_s_sb
);
2650 /* read journal header */
2651 bhjh
= journal_bread(p_s_sb
,
2652 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
2653 SB_ONDISK_JOURNAL_SIZE(p_s_sb
));
2655 reiserfs_warning(p_s_sb
,
2656 "sh-459: unable to read journal header");
2657 goto free_and_return
;
2659 jh
= (struct reiserfs_journal_header
*)(bhjh
->b_data
);
2661 /* make sure that journal matches to the super block */
2662 if (is_reiserfs_jr(rs
)
2663 && (le32_to_cpu(jh
->jh_journal
.jp_journal_magic
) !=
2664 sb_jp_journal_magic(rs
))) {
2665 reiserfs_warning(p_s_sb
,
2666 "sh-460: journal header magic %x "
2667 "(device %s) does not match to magic found in super "
2668 "block %x", jh
->jh_journal
.jp_journal_magic
,
2669 bdevname(journal
->j_dev_bd
, b
),
2670 sb_jp_journal_magic(rs
));
2672 goto free_and_return
;
2675 journal
->j_trans_max
= le32_to_cpu(jh
->jh_journal
.jp_journal_trans_max
);
2676 journal
->j_max_batch
= le32_to_cpu(jh
->jh_journal
.jp_journal_max_batch
);
2677 journal
->j_max_commit_age
=
2678 le32_to_cpu(jh
->jh_journal
.jp_journal_max_commit_age
);
2679 journal
->j_max_trans_age
= JOURNAL_MAX_TRANS_AGE
;
2681 if (journal
->j_trans_max
) {
2682 /* make sure these parameters are available, assign it if they are not */
2683 __u32 initial
= journal
->j_trans_max
;
2686 if (p_s_sb
->s_blocksize
< 4096)
2687 ratio
= 4096 / p_s_sb
->s_blocksize
;
2689 if (SB_ONDISK_JOURNAL_SIZE(p_s_sb
) / journal
->j_trans_max
<
2691 journal
->j_trans_max
=
2692 SB_ONDISK_JOURNAL_SIZE(p_s_sb
) / JOURNAL_MIN_RATIO
;
2693 if (journal
->j_trans_max
> JOURNAL_TRANS_MAX_DEFAULT
/ ratio
)
2694 journal
->j_trans_max
=
2695 JOURNAL_TRANS_MAX_DEFAULT
/ ratio
;
2696 if (journal
->j_trans_max
< JOURNAL_TRANS_MIN_DEFAULT
/ ratio
)
2697 journal
->j_trans_max
=
2698 JOURNAL_TRANS_MIN_DEFAULT
/ ratio
;
2700 if (journal
->j_trans_max
!= initial
)
2701 reiserfs_warning(p_s_sb
,
2702 "sh-461: journal_init: wrong transaction max size (%u). Changed to %u",
2703 initial
, journal
->j_trans_max
);
2705 journal
->j_max_batch
= journal
->j_trans_max
*
2706 JOURNAL_MAX_BATCH_DEFAULT
/ JOURNAL_TRANS_MAX_DEFAULT
;
2709 if (!journal
->j_trans_max
) {
2710 /*we have the file system was created by old version of mkreiserfs
2711 so this field contains zero value */
2712 journal
->j_trans_max
= JOURNAL_TRANS_MAX_DEFAULT
;
2713 journal
->j_max_batch
= JOURNAL_MAX_BATCH_DEFAULT
;
2714 journal
->j_max_commit_age
= JOURNAL_MAX_COMMIT_AGE
;
2716 /* for blocksize >= 4096 - max transaction size is 1024. For block size < 4096
2717 trans max size is decreased proportionally */
2718 if (p_s_sb
->s_blocksize
< 4096) {
2719 journal
->j_trans_max
/= (4096 / p_s_sb
->s_blocksize
);
2720 journal
->j_max_batch
= (journal
->j_trans_max
) * 9 / 10;
2724 journal
->j_default_max_commit_age
= journal
->j_max_commit_age
;
2726 if (commit_max_age
!= 0) {
2727 journal
->j_max_commit_age
= commit_max_age
;
2728 journal
->j_max_trans_age
= commit_max_age
;
2731 reiserfs_info(p_s_sb
, "journal params: device %s, size %u, "
2732 "journal first block %u, max trans len %u, max batch %u, "
2733 "max commit age %u, max trans age %u\n",
2734 bdevname(journal
->j_dev_bd
, b
),
2735 SB_ONDISK_JOURNAL_SIZE(p_s_sb
),
2736 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
),
2737 journal
->j_trans_max
,
2738 journal
->j_max_batch
,
2739 journal
->j_max_commit_age
, journal
->j_max_trans_age
);
2743 journal
->j_list_bitmap_index
= 0;
2744 journal_list_init(p_s_sb
);
2746 memset(journal
->j_list_hash_table
, 0,
2747 JOURNAL_HASH_SIZE
* sizeof(struct reiserfs_journal_cnode
*));
2749 INIT_LIST_HEAD(&journal
->j_dirty_buffers
);
2750 spin_lock_init(&journal
->j_dirty_buffers_lock
);
2752 journal
->j_start
= 0;
2754 journal
->j_len_alloc
= 0;
2755 atomic_set(&(journal
->j_wcount
), 0);
2756 atomic_set(&(journal
->j_async_throttle
), 0);
2757 journal
->j_bcount
= 0;
2758 journal
->j_trans_start_time
= 0;
2759 journal
->j_last
= NULL
;
2760 journal
->j_first
= NULL
;
2761 init_waitqueue_head(&(journal
->j_join_wait
));
2762 sema_init(&journal
->j_lock
, 1);
2763 sema_init(&journal
->j_flush_sem
, 1);
2765 journal
->j_trans_id
= 10;
2766 journal
->j_mount_id
= 10;
2767 journal
->j_state
= 0;
2768 atomic_set(&(journal
->j_jlock
), 0);
2769 journal
->j_cnode_free_list
= allocate_cnodes(num_cnodes
);
2770 journal
->j_cnode_free_orig
= journal
->j_cnode_free_list
;
2771 journal
->j_cnode_free
= journal
->j_cnode_free_list
? num_cnodes
: 0;
2772 journal
->j_cnode_used
= 0;
2773 journal
->j_must_wait
= 0;
2775 if (journal
->j_cnode_free
== 0) {
2776 reiserfs_warning(p_s_sb
, "journal-2004: Journal cnode memory "
2777 "allocation failed (%ld bytes). Journal is "
2778 "too large for available memory. Usually "
2779 "this is due to a journal that is too large.",
2780 sizeof (struct reiserfs_journal_cnode
) * num_cnodes
);
2781 goto free_and_return
;
2784 init_journal_hash(p_s_sb
);
2785 jl
= journal
->j_current_jl
;
2786 jl
->j_list_bitmap
= get_list_bitmap(p_s_sb
, jl
);
2787 if (!jl
->j_list_bitmap
) {
2788 reiserfs_warning(p_s_sb
,
2789 "journal-2005, get_list_bitmap failed for journal list 0");
2790 goto free_and_return
;
2792 if (journal_read(p_s_sb
) < 0) {
2793 reiserfs_warning(p_s_sb
, "Replay Failure, unable to mount");
2794 goto free_and_return
;
2797 reiserfs_mounted_fs_count
++;
2798 if (reiserfs_mounted_fs_count
<= 1)
2799 commit_wq
= create_workqueue("reiserfs");
2801 INIT_WORK(&journal
->j_work
, flush_async_commits
, p_s_sb
);
2804 free_journal_ram(p_s_sb
);
2809 ** test for a polite end of the current transaction. Used by file_write, and should
2810 ** be used by delete to make sure they don't write more than can fit inside a single
2813 int journal_transaction_should_end(struct reiserfs_transaction_handle
*th
,
2816 struct reiserfs_journal
*journal
= SB_JOURNAL(th
->t_super
);
2817 time_t now
= get_seconds();
2818 /* cannot restart while nested */
2819 BUG_ON(!th
->t_trans_id
);
2820 if (th
->t_refcount
> 1)
2822 if (journal
->j_must_wait
> 0 ||
2823 (journal
->j_len_alloc
+ new_alloc
) >= journal
->j_max_batch
||
2824 atomic_read(&(journal
->j_jlock
)) ||
2825 (now
- journal
->j_trans_start_time
) > journal
->j_max_trans_age
||
2826 journal
->j_cnode_free
< (journal
->j_trans_max
* 3)) {
2829 /* protected by the BKL here */
2830 journal
->j_len_alloc
+= new_alloc
;
2831 th
->t_blocks_allocated
+= new_alloc
;
2835 /* this must be called inside a transaction, and requires the
2836 ** kernel_lock to be held
2838 void reiserfs_block_writes(struct reiserfs_transaction_handle
*th
)
2840 struct reiserfs_journal
*journal
= SB_JOURNAL(th
->t_super
);
2841 BUG_ON(!th
->t_trans_id
);
2842 journal
->j_must_wait
= 1;
2843 set_bit(J_WRITERS_BLOCKED
, &journal
->j_state
);
2847 /* this must be called without a transaction started, and does not
2850 void reiserfs_allow_writes(struct super_block
*s
)
2852 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2853 clear_bit(J_WRITERS_BLOCKED
, &journal
->j_state
);
2854 wake_up(&journal
->j_join_wait
);
2857 /* this must be called without a transaction started, and does not
2860 void reiserfs_wait_on_write_block(struct super_block
*s
)
2862 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2863 wait_event(journal
->j_join_wait
,
2864 !test_bit(J_WRITERS_BLOCKED
, &journal
->j_state
));
2867 static void queue_log_writer(struct super_block
*s
)
2870 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2871 set_bit(J_WRITERS_QUEUED
, &journal
->j_state
);
2874 * we don't want to use wait_event here because
2875 * we only want to wait once.
2877 init_waitqueue_entry(&wait
, current
);
2878 add_wait_queue(&journal
->j_join_wait
, &wait
);
2879 set_current_state(TASK_UNINTERRUPTIBLE
);
2880 if (test_bit(J_WRITERS_QUEUED
, &journal
->j_state
))
2882 current
->state
= TASK_RUNNING
;
2883 remove_wait_queue(&journal
->j_join_wait
, &wait
);
2886 static void wake_queued_writers(struct super_block
*s
)
2888 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2889 if (test_and_clear_bit(J_WRITERS_QUEUED
, &journal
->j_state
))
2890 wake_up(&journal
->j_join_wait
);
2893 static void let_transaction_grow(struct super_block
*sb
, unsigned long trans_id
)
2895 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
2896 unsigned long bcount
= journal
->j_bcount
;
2898 schedule_timeout_uninterruptible(1);
2899 journal
->j_current_jl
->j_state
|= LIST_COMMIT_PENDING
;
2900 while ((atomic_read(&journal
->j_wcount
) > 0 ||
2901 atomic_read(&journal
->j_jlock
)) &&
2902 journal
->j_trans_id
== trans_id
) {
2903 queue_log_writer(sb
);
2905 if (journal
->j_trans_id
!= trans_id
)
2907 if (bcount
== journal
->j_bcount
)
2909 bcount
= journal
->j_bcount
;
2913 /* join == true if you must join an existing transaction.
2914 ** join == false if you can deal with waiting for others to finish
2916 ** this will block until the transaction is joinable. send the number of blocks you
2917 ** expect to use in nblocks.
2919 static int do_journal_begin_r(struct reiserfs_transaction_handle
*th
,
2920 struct super_block
*p_s_sb
, unsigned long nblocks
,
2923 time_t now
= get_seconds();
2925 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
2926 struct reiserfs_transaction_handle myth
;
2927 int sched_count
= 0;
2930 reiserfs_check_lock_depth(p_s_sb
, "journal_begin");
2931 if (nblocks
> journal
->j_trans_max
)
2934 PROC_INFO_INC(p_s_sb
, journal
.journal_being
);
2935 /* set here for journal_join */
2937 th
->t_super
= p_s_sb
;
2940 lock_journal(p_s_sb
);
2941 if (join
!= JBEGIN_ABORT
&& reiserfs_is_journal_aborted(journal
)) {
2942 unlock_journal(p_s_sb
);
2943 retval
= journal
->j_errno
;
2946 journal
->j_bcount
++;
2948 if (test_bit(J_WRITERS_BLOCKED
, &journal
->j_state
)) {
2949 unlock_journal(p_s_sb
);
2950 reiserfs_wait_on_write_block(p_s_sb
);
2951 PROC_INFO_INC(p_s_sb
, journal
.journal_relock_writers
);
2954 now
= get_seconds();
2956 /* if there is no room in the journal OR
2957 ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
2958 ** we don't sleep if there aren't other writers
2961 if ((!join
&& journal
->j_must_wait
> 0) ||
2963 && (journal
->j_len_alloc
+ nblocks
+ 2) >= journal
->j_max_batch
)
2964 || (!join
&& atomic_read(&journal
->j_wcount
) > 0
2965 && journal
->j_trans_start_time
> 0
2966 && (now
- journal
->j_trans_start_time
) >
2967 journal
->j_max_trans_age
) || (!join
2968 && atomic_read(&journal
->j_jlock
))
2969 || (!join
&& journal
->j_cnode_free
< (journal
->j_trans_max
* 3))) {
2971 old_trans_id
= journal
->j_trans_id
;
2972 unlock_journal(p_s_sb
); /* allow others to finish this transaction */
2974 if (!join
&& (journal
->j_len_alloc
+ nblocks
+ 2) >=
2975 journal
->j_max_batch
&&
2976 ((journal
->j_len
+ nblocks
+ 2) * 100) <
2977 (journal
->j_len_alloc
* 75)) {
2978 if (atomic_read(&journal
->j_wcount
) > 10) {
2980 queue_log_writer(p_s_sb
);
2984 /* don't mess with joining the transaction if all we have to do is
2985 * wait for someone else to do a commit
2987 if (atomic_read(&journal
->j_jlock
)) {
2988 while (journal
->j_trans_id
== old_trans_id
&&
2989 atomic_read(&journal
->j_jlock
)) {
2990 queue_log_writer(p_s_sb
);
2994 retval
= journal_join(&myth
, p_s_sb
, 1);
2998 /* someone might have ended the transaction while we joined */
2999 if (old_trans_id
!= journal
->j_trans_id
) {
3000 retval
= do_journal_end(&myth
, p_s_sb
, 1, 0);
3002 retval
= do_journal_end(&myth
, p_s_sb
, 1, COMMIT_NOW
);
3008 PROC_INFO_INC(p_s_sb
, journal
.journal_relock_wcount
);
3011 /* we are the first writer, set trans_id */
3012 if (journal
->j_trans_start_time
== 0) {
3013 journal
->j_trans_start_time
= get_seconds();
3015 atomic_inc(&(journal
->j_wcount
));
3016 journal
->j_len_alloc
+= nblocks
;
3017 th
->t_blocks_logged
= 0;
3018 th
->t_blocks_allocated
= nblocks
;
3019 th
->t_trans_id
= journal
->j_trans_id
;
3020 unlock_journal(p_s_sb
);
3021 INIT_LIST_HEAD(&th
->t_list
);
3026 memset(th
, 0, sizeof(*th
));
3027 /* Re-set th->t_super, so we can properly keep track of how many
3028 * persistent transactions there are. We need to do this so if this
3029 * call is part of a failed restart_transaction, we can free it later */
3030 th
->t_super
= p_s_sb
;
3034 struct reiserfs_transaction_handle
*reiserfs_persistent_transaction(struct
3040 struct reiserfs_transaction_handle
*th
;
3042 /* if we're nesting into an existing transaction. It will be
3043 ** persistent on its own
3045 if (reiserfs_transaction_running(s
)) {
3046 th
= current
->journal_info
;
3048 if (th
->t_refcount
< 2) {
3053 th
= kmalloc(sizeof(struct reiserfs_transaction_handle
), GFP_NOFS
);
3056 ret
= journal_begin(th
, s
, nblocks
);
3062 SB_JOURNAL(s
)->j_persistent_trans
++;
3066 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle
*th
)
3068 struct super_block
*s
= th
->t_super
;
3071 ret
= journal_end(th
, th
->t_super
, th
->t_blocks_allocated
);
3074 if (th
->t_refcount
== 0) {
3075 SB_JOURNAL(s
)->j_persistent_trans
--;
3081 static int journal_join(struct reiserfs_transaction_handle
*th
,
3082 struct super_block
*p_s_sb
, unsigned long nblocks
)
3084 struct reiserfs_transaction_handle
*cur_th
= current
->journal_info
;
3086 /* this keeps do_journal_end from NULLing out the current->journal_info
3089 th
->t_handle_save
= cur_th
;
3090 if (cur_th
&& cur_th
->t_refcount
> 1) {
3093 return do_journal_begin_r(th
, p_s_sb
, nblocks
, JBEGIN_JOIN
);
3096 int journal_join_abort(struct reiserfs_transaction_handle
*th
,
3097 struct super_block
*p_s_sb
, unsigned long nblocks
)
3099 struct reiserfs_transaction_handle
*cur_th
= current
->journal_info
;
3101 /* this keeps do_journal_end from NULLing out the current->journal_info
3104 th
->t_handle_save
= cur_th
;
3105 if (cur_th
&& cur_th
->t_refcount
> 1) {
3108 return do_journal_begin_r(th
, p_s_sb
, nblocks
, JBEGIN_ABORT
);
3111 int journal_begin(struct reiserfs_transaction_handle
*th
,
3112 struct super_block
*p_s_sb
, unsigned long nblocks
)
3114 struct reiserfs_transaction_handle
*cur_th
= current
->journal_info
;
3117 th
->t_handle_save
= NULL
;
3119 /* we are nesting into the current transaction */
3120 if (cur_th
->t_super
== p_s_sb
) {
3121 BUG_ON(!cur_th
->t_refcount
);
3122 cur_th
->t_refcount
++;
3123 memcpy(th
, cur_th
, sizeof(*th
));
3124 if (th
->t_refcount
<= 1)
3125 reiserfs_warning(p_s_sb
,
3126 "BAD: refcount <= 1, but journal_info != 0");
3129 /* we've ended up with a handle from a different filesystem.
3130 ** save it and restore on journal_end. This should never
3133 reiserfs_warning(p_s_sb
,
3134 "clm-2100: nesting info a different FS");
3135 th
->t_handle_save
= current
->journal_info
;
3136 current
->journal_info
= th
;
3139 current
->journal_info
= th
;
3141 ret
= do_journal_begin_r(th
, p_s_sb
, nblocks
, JBEGIN_REG
);
3142 if (current
->journal_info
!= th
)
3145 /* I guess this boils down to being the reciprocal of clm-2100 above.
3146 * If do_journal_begin_r fails, we need to put it back, since journal_end
3147 * won't be called to do it. */
3149 current
->journal_info
= th
->t_handle_save
;
3151 BUG_ON(!th
->t_refcount
);
3157 ** puts bh into the current transaction. If it was already there, reorders removes the
3158 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3160 ** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the
3161 ** transaction is committed.
3163 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3165 int journal_mark_dirty(struct reiserfs_transaction_handle
*th
,
3166 struct super_block
*p_s_sb
, struct buffer_head
*bh
)
3168 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
3169 struct reiserfs_journal_cnode
*cn
= NULL
;
3170 int count_already_incd
= 0;
3172 BUG_ON(!th
->t_trans_id
);
3174 PROC_INFO_INC(p_s_sb
, journal
.mark_dirty
);
3175 if (th
->t_trans_id
!= journal
->j_trans_id
) {
3176 reiserfs_panic(th
->t_super
,
3177 "journal-1577: handle trans id %ld != current trans id %ld\n",
3178 th
->t_trans_id
, journal
->j_trans_id
);
3183 prepared
= test_clear_buffer_journal_prepared(bh
);
3184 clear_buffer_journal_restore_dirty(bh
);
3185 /* already in this transaction, we are done */
3186 if (buffer_journaled(bh
)) {
3187 PROC_INFO_INC(p_s_sb
, journal
.mark_dirty_already
);
3191 /* this must be turned into a panic instead of a warning. We can't allow
3192 ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3193 ** could get to disk too early. NOT GOOD.
3195 if (!prepared
|| buffer_dirty(bh
)) {
3196 reiserfs_warning(p_s_sb
, "journal-1777: buffer %llu bad state "
3197 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3198 (unsigned long long)bh
->b_blocknr
,
3199 prepared
? ' ' : '!',
3200 buffer_locked(bh
) ? ' ' : '!',
3201 buffer_dirty(bh
) ? ' ' : '!',
3202 buffer_journal_dirty(bh
) ? ' ' : '!');
3205 if (atomic_read(&(journal
->j_wcount
)) <= 0) {
3206 reiserfs_warning(p_s_sb
,
3207 "journal-1409: journal_mark_dirty returning because j_wcount was %d",
3208 atomic_read(&(journal
->j_wcount
)));
3211 /* this error means I've screwed up, and we've overflowed the transaction.
3212 ** Nothing can be done here, except make the FS readonly or panic.
3214 if (journal
->j_len
>= journal
->j_trans_max
) {
3215 reiserfs_panic(th
->t_super
,
3216 "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n",
3220 if (buffer_journal_dirty(bh
)) {
3221 count_already_incd
= 1;
3222 PROC_INFO_INC(p_s_sb
, journal
.mark_dirty_notjournal
);
3223 clear_buffer_journal_dirty(bh
);
3226 if (journal
->j_len
> journal
->j_len_alloc
) {
3227 journal
->j_len_alloc
= journal
->j_len
+ JOURNAL_PER_BALANCE_CNT
;
3230 set_buffer_journaled(bh
);
3232 /* now put this guy on the end */
3234 cn
= get_cnode(p_s_sb
);
3236 reiserfs_panic(p_s_sb
, "get_cnode failed!\n");
3239 if (th
->t_blocks_logged
== th
->t_blocks_allocated
) {
3240 th
->t_blocks_allocated
+= JOURNAL_PER_BALANCE_CNT
;
3241 journal
->j_len_alloc
+= JOURNAL_PER_BALANCE_CNT
;
3243 th
->t_blocks_logged
++;
3247 cn
->blocknr
= bh
->b_blocknr
;
3250 insert_journal_hash(journal
->j_hash_table
, cn
);
3251 if (!count_already_incd
) {
3256 cn
->prev
= journal
->j_last
;
3258 if (journal
->j_last
) {
3259 journal
->j_last
->next
= cn
;
3260 journal
->j_last
= cn
;
3262 journal
->j_first
= cn
;
3263 journal
->j_last
= cn
;
3268 int journal_end(struct reiserfs_transaction_handle
*th
,
3269 struct super_block
*p_s_sb
, unsigned long nblocks
)
3271 if (!current
->journal_info
&& th
->t_refcount
> 1)
3272 reiserfs_warning(p_s_sb
, "REISER-NESTING: th NULL, refcount %d",
3275 if (!th
->t_trans_id
) {
3281 if (th
->t_refcount
> 0) {
3282 struct reiserfs_transaction_handle
*cur_th
=
3283 current
->journal_info
;
3285 /* we aren't allowed to close a nested transaction on a different
3286 ** filesystem from the one in the task struct
3288 if (cur_th
->t_super
!= th
->t_super
)
3292 memcpy(current
->journal_info
, th
, sizeof(*th
));
3297 return do_journal_end(th
, p_s_sb
, nblocks
, 0);
3301 /* removes from the current transaction, relsing and descrementing any counters.
3302 ** also files the removed buffer directly onto the clean list
3304 ** called by journal_mark_freed when a block has been deleted
3306 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3308 static int remove_from_transaction(struct super_block
*p_s_sb
,
3309 b_blocknr_t blocknr
, int already_cleaned
)
3311 struct buffer_head
*bh
;
3312 struct reiserfs_journal_cnode
*cn
;
3313 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
3316 cn
= get_journal_hash_dev(p_s_sb
, journal
->j_hash_table
, blocknr
);
3317 if (!cn
|| !cn
->bh
) {
3322 cn
->prev
->next
= cn
->next
;
3325 cn
->next
->prev
= cn
->prev
;
3327 if (cn
== journal
->j_first
) {
3328 journal
->j_first
= cn
->next
;
3330 if (cn
== journal
->j_last
) {
3331 journal
->j_last
= cn
->prev
;
3334 remove_journal_hash(p_s_sb
, journal
->j_hash_table
, NULL
,
3336 clear_buffer_journaled(bh
); /* don't log this one */
3338 if (!already_cleaned
) {
3339 clear_buffer_journal_dirty(bh
);
3340 clear_buffer_dirty(bh
);
3341 clear_buffer_journal_test(bh
);
3343 if (atomic_read(&(bh
->b_count
)) < 0) {
3344 reiserfs_warning(p_s_sb
,
3345 "journal-1752: remove from trans, b_count < 0");
3350 journal
->j_len_alloc
--;
3351 free_cnode(p_s_sb
, cn
);
3356 ** for any cnode in a journal list, it can only be dirtied of all the
3357 ** transactions that include it are commited to disk.
3358 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3359 ** and 0 if you aren't
3361 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3362 ** blocks for a given transaction on disk
3365 static int can_dirty(struct reiserfs_journal_cnode
*cn
)
3367 struct super_block
*sb
= cn
->sb
;
3368 b_blocknr_t blocknr
= cn
->blocknr
;
3369 struct reiserfs_journal_cnode
*cur
= cn
->hprev
;
3372 /* first test hprev. These are all newer than cn, so any node here
3373 ** with the same block number and dev means this node can't be sent
3374 ** to disk right now.
3376 while (cur
&& can_dirty
) {
3377 if (cur
->jlist
&& cur
->bh
&& cur
->blocknr
&& cur
->sb
== sb
&&
3378 cur
->blocknr
== blocknr
) {
3383 /* then test hnext. These are all older than cn. As long as they
3384 ** are committed to the log, it is safe to write cn to disk
3387 while (cur
&& can_dirty
) {
3388 if (cur
->jlist
&& cur
->jlist
->j_len
> 0 &&
3389 atomic_read(&(cur
->jlist
->j_commit_left
)) > 0 && cur
->bh
&&
3390 cur
->blocknr
&& cur
->sb
== sb
&& cur
->blocknr
== blocknr
) {
3398 /* syncs the commit blocks, but does not force the real buffers to disk
3399 ** will wait until the current transaction is done/commited before returning
3401 int journal_end_sync(struct reiserfs_transaction_handle
*th
,
3402 struct super_block
*p_s_sb
, unsigned long nblocks
)
3404 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
3406 BUG_ON(!th
->t_trans_id
);
3407 /* you can sync while nested, very, very bad */
3408 if (th
->t_refcount
> 1) {
3411 if (journal
->j_len
== 0) {
3412 reiserfs_prepare_for_journal(p_s_sb
, SB_BUFFER_WITH_SB(p_s_sb
),
3414 journal_mark_dirty(th
, p_s_sb
, SB_BUFFER_WITH_SB(p_s_sb
));
3416 return do_journal_end(th
, p_s_sb
, nblocks
, COMMIT_NOW
| WAIT
);
3420 ** writeback the pending async commits to disk
3422 static void flush_async_commits(void *p
)
3424 struct super_block
*p_s_sb
= p
;
3425 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
3426 struct reiserfs_journal_list
*jl
;
3427 struct list_head
*entry
;
3430 if (!list_empty(&journal
->j_journal_list
)) {
3431 /* last entry is the youngest, commit it and you get everything */
3432 entry
= journal
->j_journal_list
.prev
;
3433 jl
= JOURNAL_LIST_ENTRY(entry
);
3434 flush_commit_list(p_s_sb
, jl
, 1);
3438 * this is a little racey, but there's no harm in missing
3439 * the filemap_fdata_write
3441 if (!atomic_read(&journal
->j_async_throttle
)
3442 && !reiserfs_is_journal_aborted(journal
)) {
3443 atomic_inc(&journal
->j_async_throttle
);
3444 filemap_fdatawrite(p_s_sb
->s_bdev
->bd_inode
->i_mapping
);
3445 atomic_dec(&journal
->j_async_throttle
);
3450 ** flushes any old transactions to disk
3451 ** ends the current transaction if it is too old
3453 int reiserfs_flush_old_commits(struct super_block
*p_s_sb
)
3456 struct reiserfs_transaction_handle th
;
3457 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
3459 now
= get_seconds();
3460 /* safety check so we don't flush while we are replaying the log during
3463 if (list_empty(&journal
->j_journal_list
)) {
3467 /* check the current transaction. If there are no writers, and it is
3468 * too old, finish it, and force the commit blocks to disk
3470 if (atomic_read(&journal
->j_wcount
) <= 0 &&
3471 journal
->j_trans_start_time
> 0 &&
3472 journal
->j_len
> 0 &&
3473 (now
- journal
->j_trans_start_time
) > journal
->j_max_trans_age
) {
3474 if (!journal_join(&th
, p_s_sb
, 1)) {
3475 reiserfs_prepare_for_journal(p_s_sb
,
3476 SB_BUFFER_WITH_SB(p_s_sb
),
3478 journal_mark_dirty(&th
, p_s_sb
,
3479 SB_BUFFER_WITH_SB(p_s_sb
));
3481 /* we're only being called from kreiserfsd, it makes no sense to do
3482 ** an async commit so that kreiserfsd can do it later
3484 do_journal_end(&th
, p_s_sb
, 1, COMMIT_NOW
| WAIT
);
3487 return p_s_sb
->s_dirt
;
3491 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3493 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3494 ** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just
3495 ** flushes the commit list and returns 0.
3497 ** Won't batch when flush or commit_now is set. Also won't batch when others are waiting on j_join_wait.
3499 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3501 static int check_journal_end(struct reiserfs_transaction_handle
*th
,
3502 struct super_block
*p_s_sb
, unsigned long nblocks
,
3507 int flush
= flags
& FLUSH_ALL
;
3508 int commit_now
= flags
& COMMIT_NOW
;
3509 int wait_on_commit
= flags
& WAIT
;
3510 struct reiserfs_journal_list
*jl
;
3511 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
3513 BUG_ON(!th
->t_trans_id
);
3515 if (th
->t_trans_id
!= journal
->j_trans_id
) {
3516 reiserfs_panic(th
->t_super
,
3517 "journal-1577: handle trans id %ld != current trans id %ld\n",
3518 th
->t_trans_id
, journal
->j_trans_id
);
3521 journal
->j_len_alloc
-= (th
->t_blocks_allocated
- th
->t_blocks_logged
);
3522 if (atomic_read(&(journal
->j_wcount
)) > 0) { /* <= 0 is allowed. unmounting might not call begin */
3523 atomic_dec(&(journal
->j_wcount
));
3526 /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3527 ** will be dealt with by next transaction that actually writes something, but should be taken
3528 ** care of in this trans
3530 if (journal
->j_len
== 0) {
3533 /* if wcount > 0, and we are called to with flush or commit_now,
3534 ** we wait on j_join_wait. We will wake up when the last writer has
3535 ** finished the transaction, and started it on its way to the disk.
3536 ** Then, we flush the commit or journal list, and just return 0
3537 ** because the rest of journal end was already done for this transaction.
3539 if (atomic_read(&(journal
->j_wcount
)) > 0) {
3540 if (flush
|| commit_now
) {
3543 jl
= journal
->j_current_jl
;
3544 trans_id
= jl
->j_trans_id
;
3546 jl
->j_state
|= LIST_COMMIT_PENDING
;
3547 atomic_set(&(journal
->j_jlock
), 1);
3549 journal
->j_next_full_flush
= 1;
3551 unlock_journal(p_s_sb
);
3553 /* sleep while the current transaction is still j_jlocked */
3554 while (journal
->j_trans_id
== trans_id
) {
3555 if (atomic_read(&journal
->j_jlock
)) {
3556 queue_log_writer(p_s_sb
);
3558 lock_journal(p_s_sb
);
3559 if (journal
->j_trans_id
== trans_id
) {
3560 atomic_set(&(journal
->j_jlock
),
3563 unlock_journal(p_s_sb
);
3566 if (journal
->j_trans_id
== trans_id
) {
3570 && journal_list_still_alive(p_s_sb
, trans_id
)
3571 && wait_on_commit
) {
3572 flush_commit_list(p_s_sb
, jl
, 1);
3576 unlock_journal(p_s_sb
);
3580 /* deal with old transactions where we are the last writers */
3581 now
= get_seconds();
3582 if ((now
- journal
->j_trans_start_time
) > journal
->j_max_trans_age
) {
3584 journal
->j_next_async_flush
= 1;
3586 /* don't batch when someone is waiting on j_join_wait */
3587 /* don't batch when syncing the commit or flushing the whole trans */
3588 if (!(journal
->j_must_wait
> 0) && !(atomic_read(&(journal
->j_jlock
)))
3589 && !flush
&& !commit_now
&& (journal
->j_len
< journal
->j_max_batch
)
3590 && journal
->j_len_alloc
< journal
->j_max_batch
3591 && journal
->j_cnode_free
> (journal
->j_trans_max
* 3)) {
3592 journal
->j_bcount
++;
3593 unlock_journal(p_s_sb
);
3597 if (journal
->j_start
> SB_ONDISK_JOURNAL_SIZE(p_s_sb
)) {
3598 reiserfs_panic(p_s_sb
,
3599 "journal-003: journal_end: j_start (%ld) is too high\n",
3606 ** Does all the work that makes deleting blocks safe.
3607 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3610 ** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes
3611 ** before this transaction has finished.
3613 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with
3614 ** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash,
3615 ** the block can't be reallocated yet.
3617 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3619 int journal_mark_freed(struct reiserfs_transaction_handle
*th
,
3620 struct super_block
*p_s_sb
, b_blocknr_t blocknr
)
3622 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
3623 struct reiserfs_journal_cnode
*cn
= NULL
;
3624 struct buffer_head
*bh
= NULL
;
3625 struct reiserfs_list_bitmap
*jb
= NULL
;
3627 BUG_ON(!th
->t_trans_id
);
3629 cn
= get_journal_hash_dev(p_s_sb
, journal
->j_hash_table
, blocknr
);
3634 /* if it is journal new, we just remove it from this transaction */
3635 if (bh
&& buffer_journal_new(bh
)) {
3636 clear_buffer_journal_new(bh
);
3637 clear_prepared_bits(bh
);
3638 reiserfs_clean_and_file_buffer(bh
);
3639 cleaned
= remove_from_transaction(p_s_sb
, blocknr
, cleaned
);
3641 /* set the bit for this block in the journal bitmap for this transaction */
3642 jb
= journal
->j_current_jl
->j_list_bitmap
;
3644 reiserfs_panic(p_s_sb
,
3645 "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n");
3647 set_bit_in_list_bitmap(p_s_sb
, blocknr
, jb
);
3649 /* Note, the entire while loop is not allowed to schedule. */
3652 clear_prepared_bits(bh
);
3653 reiserfs_clean_and_file_buffer(bh
);
3655 cleaned
= remove_from_transaction(p_s_sb
, blocknr
, cleaned
);
3657 /* find all older transactions with this block, make sure they don't try to write it out */
3658 cn
= get_journal_hash_dev(p_s_sb
, journal
->j_list_hash_table
,
3661 if (p_s_sb
== cn
->sb
&& blocknr
== cn
->blocknr
) {
3662 set_bit(BLOCK_FREED
, &cn
->state
);
3665 /* remove_from_transaction will brelse the buffer if it was
3666 ** in the current trans
3668 clear_buffer_journal_dirty(cn
->
3670 clear_buffer_dirty(cn
->bh
);
3671 clear_buffer_journal_test(cn
->
3676 (&(cn
->bh
->b_count
)) < 0) {
3677 reiserfs_warning(p_s_sb
,
3678 "journal-2138: cn->bh->b_count < 0");
3681 if (cn
->jlist
) { /* since we are clearing the bh, we MUST dec nonzerolen */
3694 put_bh(bh
); /* get_hash grabs the buffer */
3695 if (atomic_read(&(bh
->b_count
)) < 0) {
3696 reiserfs_warning(p_s_sb
,
3697 "journal-2165: bh->b_count < 0");
3703 void reiserfs_update_inode_transaction(struct inode
*inode
)
3705 struct reiserfs_journal
*journal
= SB_JOURNAL(inode
->i_sb
);
3706 REISERFS_I(inode
)->i_jl
= journal
->j_current_jl
;
3707 REISERFS_I(inode
)->i_trans_id
= journal
->j_trans_id
;
3711 * returns -1 on error, 0 if no commits/barriers were done and 1
3712 * if a transaction was actually committed and the barrier was done
3714 static int __commit_trans_jl(struct inode
*inode
, unsigned long id
,
3715 struct reiserfs_journal_list
*jl
)
3717 struct reiserfs_transaction_handle th
;
3718 struct super_block
*sb
= inode
->i_sb
;
3719 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3722 /* is it from the current transaction, or from an unknown transaction? */
3723 if (id
== journal
->j_trans_id
) {
3724 jl
= journal
->j_current_jl
;
3725 /* try to let other writers come in and grow this transaction */
3726 let_transaction_grow(sb
, id
);
3727 if (journal
->j_trans_id
!= id
) {
3728 goto flush_commit_only
;
3731 ret
= journal_begin(&th
, sb
, 1);
3735 /* someone might have ended this transaction while we joined */
3736 if (journal
->j_trans_id
!= id
) {
3737 reiserfs_prepare_for_journal(sb
, SB_BUFFER_WITH_SB(sb
),
3739 journal_mark_dirty(&th
, sb
, SB_BUFFER_WITH_SB(sb
));
3740 ret
= journal_end(&th
, sb
, 1);
3741 goto flush_commit_only
;
3744 ret
= journal_end_sync(&th
, sb
, 1);
3749 /* this gets tricky, we have to make sure the journal list in
3750 * the inode still exists. We know the list is still around
3751 * if we've got a larger transaction id than the oldest list
3754 if (journal_list_still_alive(inode
->i_sb
, id
)) {
3756 * we only set ret to 1 when we know for sure
3757 * the barrier hasn't been started yet on the commit
3760 if (atomic_read(&jl
->j_commit_left
) > 1)
3762 flush_commit_list(sb
, jl
, 1);
3763 if (journal
->j_errno
)
3764 ret
= journal
->j_errno
;
3767 /* otherwise the list is gone, and long since committed */
3771 int reiserfs_commit_for_inode(struct inode
*inode
)
3773 unsigned long id
= REISERFS_I(inode
)->i_trans_id
;
3774 struct reiserfs_journal_list
*jl
= REISERFS_I(inode
)->i_jl
;
3776 /* for the whole inode, assume unset id means it was
3777 * changed in the current transaction. More conservative
3780 reiserfs_update_inode_transaction(inode
);
3781 id
= REISERFS_I(inode
)->i_trans_id
;
3782 /* jl will be updated in __commit_trans_jl */
3785 return __commit_trans_jl(inode
, id
, jl
);
3788 void reiserfs_restore_prepared_buffer(struct super_block
*p_s_sb
,
3789 struct buffer_head
*bh
)
3791 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
3792 PROC_INFO_INC(p_s_sb
, journal
.restore_prepared
);
3796 if (test_clear_buffer_journal_restore_dirty(bh
) &&
3797 buffer_journal_dirty(bh
)) {
3798 struct reiserfs_journal_cnode
*cn
;
3799 cn
= get_journal_hash_dev(p_s_sb
,
3800 journal
->j_list_hash_table
,
3802 if (cn
&& can_dirty(cn
)) {
3803 set_buffer_journal_test(bh
);
3804 mark_buffer_dirty(bh
);
3807 clear_buffer_journal_prepared(bh
);
3810 extern struct tree_balance
*cur_tb
;
3812 ** before we can change a metadata block, we have to make sure it won't
3813 ** be written to disk while we are altering it. So, we must:
3818 int reiserfs_prepare_for_journal(struct super_block
*p_s_sb
,
3819 struct buffer_head
*bh
, int wait
)
3821 PROC_INFO_INC(p_s_sb
, journal
.prepare
);
3823 if (test_set_buffer_locked(bh
)) {
3828 set_buffer_journal_prepared(bh
);
3829 if (test_clear_buffer_dirty(bh
) && buffer_journal_dirty(bh
)) {
3830 clear_buffer_journal_test(bh
);
3831 set_buffer_journal_restore_dirty(bh
);
3837 static void flush_old_journal_lists(struct super_block
*s
)
3839 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
3840 struct reiserfs_journal_list
*jl
;
3841 struct list_head
*entry
;
3842 time_t now
= get_seconds();
3844 while (!list_empty(&journal
->j_journal_list
)) {
3845 entry
= journal
->j_journal_list
.next
;
3846 jl
= JOURNAL_LIST_ENTRY(entry
);
3847 /* this check should always be run, to send old lists to disk */
3848 if (jl
->j_timestamp
< (now
- (JOURNAL_MAX_TRANS_AGE
* 4))) {
3849 flush_used_journal_lists(s
, jl
);
3857 ** long and ugly. If flush, will not return until all commit
3858 ** blocks and all real buffers in the trans are on disk.
3859 ** If no_async, won't return until all commit blocks are on disk.
3861 ** keep reading, there are comments as you go along
3863 ** If the journal is aborted, we just clean up. Things like flushing
3864 ** journal lists, etc just won't happen.
3866 static int do_journal_end(struct reiserfs_transaction_handle
*th
,
3867 struct super_block
*p_s_sb
, unsigned long nblocks
,
3870 struct reiserfs_journal
*journal
= SB_JOURNAL(p_s_sb
);
3871 struct reiserfs_journal_cnode
*cn
, *next
, *jl_cn
;
3872 struct reiserfs_journal_cnode
*last_cn
= NULL
;
3873 struct reiserfs_journal_desc
*desc
;
3874 struct reiserfs_journal_commit
*commit
;
3875 struct buffer_head
*c_bh
; /* commit bh */
3876 struct buffer_head
*d_bh
; /* desc bh */
3877 int cur_write_start
= 0; /* start index of current log write */
3882 struct reiserfs_journal_list
*jl
, *temp_jl
;
3883 struct list_head
*entry
, *safe
;
3884 unsigned long jindex
;
3885 unsigned long commit_trans_id
;
3888 BUG_ON(th
->t_refcount
> 1);
3889 BUG_ON(!th
->t_trans_id
);
3891 /* protect flush_older_commits from doing mistakes if the
3892 transaction ID counter gets overflowed. */
3893 if (th
->t_trans_id
== ~0UL)
3894 flags
|= FLUSH_ALL
| COMMIT_NOW
| WAIT
;
3895 flush
= flags
& FLUSH_ALL
;
3896 wait_on_commit
= flags
& WAIT
;
3899 current
->journal_info
= th
->t_handle_save
;
3900 reiserfs_check_lock_depth(p_s_sb
, "journal end");
3901 if (journal
->j_len
== 0) {
3902 reiserfs_prepare_for_journal(p_s_sb
, SB_BUFFER_WITH_SB(p_s_sb
),
3904 journal_mark_dirty(th
, p_s_sb
, SB_BUFFER_WITH_SB(p_s_sb
));
3907 lock_journal(p_s_sb
);
3908 if (journal
->j_next_full_flush
) {
3912 if (journal
->j_next_async_flush
) {
3913 flags
|= COMMIT_NOW
| WAIT
;
3917 /* check_journal_end locks the journal, and unlocks if it does not return 1
3918 ** it tells us if we should continue with the journal_end, or just return
3920 if (!check_journal_end(th
, p_s_sb
, nblocks
, flags
)) {
3922 wake_queued_writers(p_s_sb
);
3923 reiserfs_async_progress_wait(p_s_sb
);
3927 /* check_journal_end might set these, check again */
3928 if (journal
->j_next_full_flush
) {
3933 ** j must wait means we have to flush the log blocks, and the real blocks for
3936 if (journal
->j_must_wait
> 0) {
3939 #ifdef REISERFS_PREALLOCATE
3940 /* quota ops might need to nest, setup the journal_info pointer for them
3941 * and raise the refcount so that it is > 0. */
3942 current
->journal_info
= th
;
3944 reiserfs_discard_all_prealloc(th
); /* it should not involve new blocks into
3945 * the transaction */
3947 current
->journal_info
= th
->t_handle_save
;
3950 /* setup description block */
3952 journal_getblk(p_s_sb
,
3953 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
3955 set_buffer_uptodate(d_bh
);
3956 desc
= (struct reiserfs_journal_desc
*)(d_bh
)->b_data
;
3957 memset(d_bh
->b_data
, 0, d_bh
->b_size
);
3958 memcpy(get_journal_desc_magic(d_bh
), JOURNAL_DESC_MAGIC
, 8);
3959 set_desc_trans_id(desc
, journal
->j_trans_id
);
3961 /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */
3962 c_bh
= journal_getblk(p_s_sb
, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
3963 ((journal
->j_start
+ journal
->j_len
+
3964 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb
)));
3965 commit
= (struct reiserfs_journal_commit
*)c_bh
->b_data
;
3966 memset(c_bh
->b_data
, 0, c_bh
->b_size
);
3967 set_commit_trans_id(commit
, journal
->j_trans_id
);
3968 set_buffer_uptodate(c_bh
);
3970 /* init this journal list */
3971 jl
= journal
->j_current_jl
;
3973 /* we lock the commit before doing anything because
3974 * we want to make sure nobody tries to run flush_commit_list until
3975 * the new transaction is fully setup, and we've already flushed the
3978 down(&jl
->j_commit_lock
);
3980 /* save the transaction id in case we need to commit it later */
3981 commit_trans_id
= jl
->j_trans_id
;
3983 atomic_set(&jl
->j_older_commits_done
, 0);
3984 jl
->j_trans_id
= journal
->j_trans_id
;
3985 jl
->j_timestamp
= journal
->j_trans_start_time
;
3986 jl
->j_commit_bh
= c_bh
;
3987 jl
->j_start
= journal
->j_start
;
3988 jl
->j_len
= journal
->j_len
;
3989 atomic_set(&jl
->j_nonzerolen
, journal
->j_len
);
3990 atomic_set(&jl
->j_commit_left
, journal
->j_len
+ 2);
3991 jl
->j_realblock
= NULL
;
3993 /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
3994 ** for each real block, add it to the journal list hash,
3995 ** copy into real block index array in the commit or desc block
3997 trans_half
= journal_trans_half(p_s_sb
->s_blocksize
);
3998 for (i
= 0, cn
= journal
->j_first
; cn
; cn
= cn
->next
, i
++) {
3999 if (buffer_journaled(cn
->bh
)) {
4000 jl_cn
= get_cnode(p_s_sb
);
4002 reiserfs_panic(p_s_sb
,
4003 "journal-1676, get_cnode returned NULL\n");
4006 jl
->j_realblock
= jl_cn
;
4008 jl_cn
->prev
= last_cn
;
4011 last_cn
->next
= jl_cn
;
4014 /* make sure the block we are trying to log is not a block
4015 of journal or reserved area */
4017 if (is_block_in_log_or_reserved_area
4018 (p_s_sb
, cn
->bh
->b_blocknr
)) {
4019 reiserfs_panic(p_s_sb
,
4020 "journal-2332: Trying to log block %lu, which is a log block\n",
4023 jl_cn
->blocknr
= cn
->bh
->b_blocknr
;
4028 insert_journal_hash(journal
->j_list_hash_table
, jl_cn
);
4029 if (i
< trans_half
) {
4030 desc
->j_realblock
[i
] =
4031 cpu_to_le32(cn
->bh
->b_blocknr
);
4033 commit
->j_realblock
[i
- trans_half
] =
4034 cpu_to_le32(cn
->bh
->b_blocknr
);
4040 set_desc_trans_len(desc
, journal
->j_len
);
4041 set_desc_mount_id(desc
, journal
->j_mount_id
);
4042 set_desc_trans_id(desc
, journal
->j_trans_id
);
4043 set_commit_trans_len(commit
, journal
->j_len
);
4045 /* special check in case all buffers in the journal were marked for not logging */
4046 if (journal
->j_len
== 0) {
4050 /* we're about to dirty all the log blocks, mark the description block
4051 * dirty now too. Don't mark the commit block dirty until all the
4052 * others are on disk
4054 mark_buffer_dirty(d_bh
);
4056 /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4057 cur_write_start
= journal
->j_start
;
4058 cn
= journal
->j_first
;
4059 jindex
= 1; /* start at one so we don't get the desc again */
4061 clear_buffer_journal_new(cn
->bh
);
4062 /* copy all the real blocks into log area. dirty log blocks */
4063 if (buffer_journaled(cn
->bh
)) {
4064 struct buffer_head
*tmp_bh
;
4068 journal_getblk(p_s_sb
,
4069 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb
) +
4072 SB_ONDISK_JOURNAL_SIZE(p_s_sb
)));
4073 set_buffer_uptodate(tmp_bh
);
4074 page
= cn
->bh
->b_page
;
4076 memcpy(tmp_bh
->b_data
,
4077 addr
+ offset_in_page(cn
->bh
->b_data
),
4080 mark_buffer_dirty(tmp_bh
);
4082 set_buffer_journal_dirty(cn
->bh
);
4083 clear_buffer_journaled(cn
->bh
);
4085 /* JDirty cleared sometime during transaction. don't log this one */
4086 reiserfs_warning(p_s_sb
,
4087 "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!");
4091 free_cnode(p_s_sb
, cn
);
4096 /* we are done with both the c_bh and d_bh, but
4097 ** c_bh must be written after all other commit blocks,
4098 ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4101 journal
->j_current_jl
= alloc_journal_list(p_s_sb
);
4103 /* now it is safe to insert this transaction on the main list */
4104 list_add_tail(&jl
->j_list
, &journal
->j_journal_list
);
4105 list_add_tail(&jl
->j_working_list
, &journal
->j_working_list
);
4106 journal
->j_num_work_lists
++;
4108 /* reset journal values for the next transaction */
4109 old_start
= journal
->j_start
;
4111 (journal
->j_start
+ journal
->j_len
+
4112 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb
);
4113 atomic_set(&(journal
->j_wcount
), 0);
4114 journal
->j_bcount
= 0;
4115 journal
->j_last
= NULL
;
4116 journal
->j_first
= NULL
;
4118 journal
->j_trans_start_time
= 0;
4119 /* check for trans_id overflow */
4120 if (++journal
->j_trans_id
== 0)
4121 journal
->j_trans_id
= 10;
4122 journal
->j_current_jl
->j_trans_id
= journal
->j_trans_id
;
4123 journal
->j_must_wait
= 0;
4124 journal
->j_len_alloc
= 0;
4125 journal
->j_next_full_flush
= 0;
4126 journal
->j_next_async_flush
= 0;
4127 init_journal_hash(p_s_sb
);
4129 // make sure reiserfs_add_jh sees the new current_jl before we
4130 // write out the tails
4133 /* tail conversion targets have to hit the disk before we end the
4134 * transaction. Otherwise a later transaction might repack the tail
4135 * before this transaction commits, leaving the data block unflushed and
4136 * clean, if we crash before the later transaction commits, the data block
4139 if (!list_empty(&jl
->j_tail_bh_list
)) {
4141 write_ordered_buffers(&journal
->j_dirty_buffers_lock
,
4142 journal
, jl
, &jl
->j_tail_bh_list
);
4145 if (!list_empty(&jl
->j_tail_bh_list
))
4147 up(&jl
->j_commit_lock
);
4149 /* honor the flush wishes from the caller, simple commits can
4150 ** be done outside the journal lock, they are done below
4152 ** if we don't flush the commit list right now, we put it into
4153 ** the work queue so the people waiting on the async progress work
4154 ** queue don't wait for this proc to flush journal lists and such.
4157 flush_commit_list(p_s_sb
, jl
, 1);
4158 flush_journal_list(p_s_sb
, jl
, 1);
4159 } else if (!(jl
->j_state
& LIST_COMMIT_PENDING
))
4160 queue_delayed_work(commit_wq
, &journal
->j_work
, HZ
/ 10);
4162 /* if the next transaction has any chance of wrapping, flush
4163 ** transactions that might get overwritten. If any journal lists are very
4164 ** old flush them as well.
4167 list_for_each_safe(entry
, safe
, &journal
->j_journal_list
) {
4168 temp_jl
= JOURNAL_LIST_ENTRY(entry
);
4169 if (journal
->j_start
<= temp_jl
->j_start
) {
4170 if ((journal
->j_start
+ journal
->j_trans_max
+ 1) >=
4172 flush_used_journal_lists(p_s_sb
, temp_jl
);
4174 } else if ((journal
->j_start
+
4175 journal
->j_trans_max
+ 1) <
4176 SB_ONDISK_JOURNAL_SIZE(p_s_sb
)) {
4177 /* if we don't cross into the next transaction and we don't
4178 * wrap, there is no way we can overlap any later transactions
4183 } else if ((journal
->j_start
+
4184 journal
->j_trans_max
+ 1) >
4185 SB_ONDISK_JOURNAL_SIZE(p_s_sb
)) {
4186 if (((journal
->j_start
+ journal
->j_trans_max
+ 1) %
4187 SB_ONDISK_JOURNAL_SIZE(p_s_sb
)) >=
4189 flush_used_journal_lists(p_s_sb
, temp_jl
);
4192 /* we don't overlap anything from out start to the end of the
4193 * log, and our wrapped portion doesn't overlap anything at
4194 * the start of the log. We can break
4200 flush_old_journal_lists(p_s_sb
);
4202 journal
->j_current_jl
->j_list_bitmap
=
4203 get_list_bitmap(p_s_sb
, journal
->j_current_jl
);
4205 if (!(journal
->j_current_jl
->j_list_bitmap
)) {
4206 reiserfs_panic(p_s_sb
,
4207 "journal-1996: do_journal_end, could not get a list bitmap\n");
4210 atomic_set(&(journal
->j_jlock
), 0);
4211 unlock_journal(p_s_sb
);
4212 /* wake up any body waiting to join. */
4213 clear_bit(J_WRITERS_QUEUED
, &journal
->j_state
);
4214 wake_up(&(journal
->j_join_wait
));
4216 if (!flush
&& wait_on_commit
&&
4217 journal_list_still_alive(p_s_sb
, commit_trans_id
)) {
4218 flush_commit_list(p_s_sb
, jl
, 1);
4221 reiserfs_check_lock_depth(p_s_sb
, "journal end2");
4223 memset(th
, 0, sizeof(*th
));
4224 /* Re-set th->t_super, so we can properly keep track of how many
4225 * persistent transactions there are. We need to do this so if this
4226 * call is part of a failed restart_transaction, we can free it later */
4227 th
->t_super
= p_s_sb
;
4229 return journal
->j_errno
;
4232 static void __reiserfs_journal_abort_hard(struct super_block
*sb
)
4234 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
4235 if (test_bit(J_ABORTED
, &journal
->j_state
))
4238 printk(KERN_CRIT
"REISERFS: Aborting journal for filesystem on %s\n",
4239 reiserfs_bdevname(sb
));
4241 sb
->s_flags
|= MS_RDONLY
;
4242 set_bit(J_ABORTED
, &journal
->j_state
);
4244 #ifdef CONFIG_REISERFS_CHECK
4249 static void __reiserfs_journal_abort_soft(struct super_block
*sb
, int errno
)
4251 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
4252 if (test_bit(J_ABORTED
, &journal
->j_state
))
4255 if (!journal
->j_errno
)
4256 journal
->j_errno
= errno
;
4258 __reiserfs_journal_abort_hard(sb
);
4261 void reiserfs_journal_abort(struct super_block
*sb
, int errno
)
4263 return __reiserfs_journal_abort_soft(sb
, errno
);