2 ** Write ahead logging implementation copyright Chris Mason 2000
4 ** The background commits make this code very interrelated, and
5 ** overly complex. I need to rethink things a bit....The major players:
7 ** journal_begin -- call with the number of blocks you expect to log.
8 ** If the current transaction is too
9 ** old, it will block until the current transaction is
10 ** finished, and then start a new one.
11 ** Usually, your transaction will get joined in with
12 ** previous ones for speed.
14 ** journal_join -- same as journal_begin, but won't block on the current
15 ** transaction regardless of age. Don't ever call
16 ** this. Ever. There are only two places it should be
17 ** called from, and they are both inside this file.
19 ** journal_mark_dirty -- adds blocks into this transaction. clears any flags
20 ** that might make them get sent to disk
21 ** and then marks them BH_JDirty. Puts the buffer head
22 ** into the current transaction hash.
24 ** journal_end -- if the current transaction is batchable, it does nothing
25 ** otherwise, it could do an async/synchronous commit, or
26 ** a full flush of all log and real blocks in the
29 ** flush_old_commits -- if the current transaction is too old, it is ended and
30 ** commit blocks are sent to disk. Forces commit blocks
31 ** to disk for all backgrounded commits that have been
33 ** -- Note, if you call this as an immediate flush from
34 ** from within kupdate, it will ignore the immediate flag
37 #include <linux/time.h>
38 #include <linux/semaphore.h>
39 #include <linux/vmalloc.h>
41 #include <linux/kernel.h>
42 #include <linux/errno.h>
43 #include <linux/fcntl.h>
44 #include <linux/stat.h>
45 #include <linux/string.h>
46 #include <linux/buffer_head.h>
47 #include <linux/workqueue.h>
48 #include <linux/writeback.h>
49 #include <linux/blkdev.h>
50 #include <linux/backing-dev.h>
51 #include <linux/uaccess.h>
52 #include <linux/slab.h>
55 /* gets a struct reiserfs_journal_list * from a list head */
56 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
58 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
61 /* the number of mounted filesystems. This is used to decide when to
62 ** start and kill the commit workqueue
64 static int reiserfs_mounted_fs_count
;
66 static struct workqueue_struct
*commit_wq
;
68 #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
70 #define BUFNR 64 /*read ahead */
72 /* cnode stat bits. Move these into reiserfs_fs.h */
74 #define BLOCK_FREED 2 /* this block was freed, and can't be written. */
75 #define BLOCK_FREED_HOLDER 3 /* this block was freed during this transaction, and can't be written */
77 #define BLOCK_NEEDS_FLUSH 4 /* used in flush_journal_list */
78 #define BLOCK_DIRTIED 5
80 /* journal list state bits */
81 #define LIST_TOUCHED 1
83 #define LIST_COMMIT_PENDING 4 /* someone will commit this list */
85 /* flags for do_journal_end */
86 #define FLUSH_ALL 1 /* flush commit and real blocks */
87 #define COMMIT_NOW 2 /* end and commit this transaction */
88 #define WAIT 4 /* wait for the log blocks to hit the disk */
90 static int do_journal_end(struct reiserfs_transaction_handle
*,
91 struct super_block
*, unsigned long nblocks
,
93 static int flush_journal_list(struct super_block
*s
,
94 struct reiserfs_journal_list
*jl
, int flushall
);
95 static int flush_commit_list(struct super_block
*s
,
96 struct reiserfs_journal_list
*jl
, int flushall
);
97 static int can_dirty(struct reiserfs_journal_cnode
*cn
);
98 static int journal_join(struct reiserfs_transaction_handle
*th
,
99 struct super_block
*sb
, unsigned long nblocks
);
100 static int release_journal_dev(struct super_block
*super
,
101 struct reiserfs_journal
*journal
);
102 static int dirty_one_transaction(struct super_block
*s
,
103 struct reiserfs_journal_list
*jl
);
104 static void flush_async_commits(struct work_struct
*work
);
105 static void queue_log_writer(struct super_block
*s
);
107 /* values for join in do_journal_begin_r */
109 JBEGIN_REG
= 0, /* regular journal begin */
110 JBEGIN_JOIN
= 1, /* join the running transaction if at all possible */
111 JBEGIN_ABORT
= 2, /* called from cleanup code, ignores aborted flag */
114 static int do_journal_begin_r(struct reiserfs_transaction_handle
*th
,
115 struct super_block
*sb
,
116 unsigned long nblocks
, int join
);
118 static void init_journal_hash(struct super_block
*sb
)
120 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
121 memset(journal
->j_hash_table
, 0,
122 JOURNAL_HASH_SIZE
* sizeof(struct reiserfs_journal_cnode
*));
126 ** clears BH_Dirty and sticks the buffer on the clean list. Called because I can't allow refile_buffer to
127 ** make schedule happen after I've freed a block. Look at remove_from_transaction and journal_mark_freed for
130 static int reiserfs_clean_and_file_buffer(struct buffer_head
*bh
)
133 clear_buffer_dirty(bh
);
134 clear_buffer_journal_test(bh
);
139 static struct reiserfs_bitmap_node
*allocate_bitmap_node(struct super_block
142 struct reiserfs_bitmap_node
*bn
;
145 bn
= kmalloc(sizeof(struct reiserfs_bitmap_node
), GFP_NOFS
);
149 bn
->data
= kzalloc(sb
->s_blocksize
, GFP_NOFS
);
155 INIT_LIST_HEAD(&bn
->list
);
159 static struct reiserfs_bitmap_node
*get_bitmap_node(struct super_block
*sb
)
161 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
162 struct reiserfs_bitmap_node
*bn
= NULL
;
163 struct list_head
*entry
= journal
->j_bitmap_nodes
.next
;
165 journal
->j_used_bitmap_nodes
++;
168 if (entry
!= &journal
->j_bitmap_nodes
) {
169 bn
= list_entry(entry
, struct reiserfs_bitmap_node
, list
);
171 memset(bn
->data
, 0, sb
->s_blocksize
);
172 journal
->j_free_bitmap_nodes
--;
175 bn
= allocate_bitmap_node(sb
);
182 static inline void free_bitmap_node(struct super_block
*sb
,
183 struct reiserfs_bitmap_node
*bn
)
185 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
186 journal
->j_used_bitmap_nodes
--;
187 if (journal
->j_free_bitmap_nodes
> REISERFS_MAX_BITMAP_NODES
) {
191 list_add(&bn
->list
, &journal
->j_bitmap_nodes
);
192 journal
->j_free_bitmap_nodes
++;
196 static void allocate_bitmap_nodes(struct super_block
*sb
)
199 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
200 struct reiserfs_bitmap_node
*bn
= NULL
;
201 for (i
= 0; i
< REISERFS_MIN_BITMAP_NODES
; i
++) {
202 bn
= allocate_bitmap_node(sb
);
204 list_add(&bn
->list
, &journal
->j_bitmap_nodes
);
205 journal
->j_free_bitmap_nodes
++;
207 break; /* this is ok, we'll try again when more are needed */
212 static int set_bit_in_list_bitmap(struct super_block
*sb
,
214 struct reiserfs_list_bitmap
*jb
)
216 unsigned int bmap_nr
= block
/ (sb
->s_blocksize
<< 3);
217 unsigned int bit_nr
= block
% (sb
->s_blocksize
<< 3);
219 if (!jb
->bitmaps
[bmap_nr
]) {
220 jb
->bitmaps
[bmap_nr
] = get_bitmap_node(sb
);
222 set_bit(bit_nr
, (unsigned long *)jb
->bitmaps
[bmap_nr
]->data
);
226 static void cleanup_bitmap_list(struct super_block
*sb
,
227 struct reiserfs_list_bitmap
*jb
)
230 if (jb
->bitmaps
== NULL
)
233 for (i
= 0; i
< reiserfs_bmap_count(sb
); i
++) {
234 if (jb
->bitmaps
[i
]) {
235 free_bitmap_node(sb
, jb
->bitmaps
[i
]);
236 jb
->bitmaps
[i
] = NULL
;
242 ** only call this on FS unmount.
244 static int free_list_bitmaps(struct super_block
*sb
,
245 struct reiserfs_list_bitmap
*jb_array
)
248 struct reiserfs_list_bitmap
*jb
;
249 for (i
= 0; i
< JOURNAL_NUM_BITMAPS
; i
++) {
251 jb
->journal_list
= NULL
;
252 cleanup_bitmap_list(sb
, jb
);
259 static int free_bitmap_nodes(struct super_block
*sb
)
261 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
262 struct list_head
*next
= journal
->j_bitmap_nodes
.next
;
263 struct reiserfs_bitmap_node
*bn
;
265 while (next
!= &journal
->j_bitmap_nodes
) {
266 bn
= list_entry(next
, struct reiserfs_bitmap_node
, list
);
270 next
= journal
->j_bitmap_nodes
.next
;
271 journal
->j_free_bitmap_nodes
--;
278 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
279 ** jb_array is the array to be filled in.
281 int reiserfs_allocate_list_bitmaps(struct super_block
*sb
,
282 struct reiserfs_list_bitmap
*jb_array
,
283 unsigned int bmap_nr
)
287 struct reiserfs_list_bitmap
*jb
;
288 int mem
= bmap_nr
* sizeof(struct reiserfs_bitmap_node
*);
290 for (i
= 0; i
< JOURNAL_NUM_BITMAPS
; i
++) {
292 jb
->journal_list
= NULL
;
293 jb
->bitmaps
= vzalloc(mem
);
295 reiserfs_warning(sb
, "clm-2000", "unable to "
296 "allocate bitmaps for journal lists");
302 free_list_bitmaps(sb
, jb_array
);
309 ** find an available list bitmap. If you can't find one, flush a commit list
312 static struct reiserfs_list_bitmap
*get_list_bitmap(struct super_block
*sb
,
313 struct reiserfs_journal_list
317 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
318 struct reiserfs_list_bitmap
*jb
= NULL
;
320 for (j
= 0; j
< (JOURNAL_NUM_BITMAPS
* 3); j
++) {
321 i
= journal
->j_list_bitmap_index
;
322 journal
->j_list_bitmap_index
= (i
+ 1) % JOURNAL_NUM_BITMAPS
;
323 jb
= journal
->j_list_bitmap
+ i
;
324 if (journal
->j_list_bitmap
[i
].journal_list
) {
325 flush_commit_list(sb
,
326 journal
->j_list_bitmap
[i
].
328 if (!journal
->j_list_bitmap
[i
].journal_list
) {
335 if (jb
->journal_list
) { /* double check to make sure if flushed correctly */
338 jb
->journal_list
= jl
;
343 ** allocates a new chunk of X nodes, and links them all together as a list.
344 ** Uses the cnode->next and cnode->prev pointers
345 ** returns NULL on failure
347 static struct reiserfs_journal_cnode
*allocate_cnodes(int num_cnodes
)
349 struct reiserfs_journal_cnode
*head
;
351 if (num_cnodes
<= 0) {
354 head
= vzalloc(num_cnodes
* sizeof(struct reiserfs_journal_cnode
));
359 head
[0].next
= head
+ 1;
360 for (i
= 1; i
< num_cnodes
; i
++) {
361 head
[i
].prev
= head
+ (i
- 1);
362 head
[i
].next
= head
+ (i
+ 1); /* if last one, overwrite it after the if */
364 head
[num_cnodes
- 1].next
= NULL
;
369 ** pulls a cnode off the free list, or returns NULL on failure
371 static struct reiserfs_journal_cnode
*get_cnode(struct super_block
*sb
)
373 struct reiserfs_journal_cnode
*cn
;
374 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
376 reiserfs_check_lock_depth(sb
, "get_cnode");
378 if (journal
->j_cnode_free
<= 0) {
381 journal
->j_cnode_used
++;
382 journal
->j_cnode_free
--;
383 cn
= journal
->j_cnode_free_list
;
388 cn
->next
->prev
= NULL
;
390 journal
->j_cnode_free_list
= cn
->next
;
391 memset(cn
, 0, sizeof(struct reiserfs_journal_cnode
));
396 ** returns a cnode to the free list
398 static void free_cnode(struct super_block
*sb
,
399 struct reiserfs_journal_cnode
*cn
)
401 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
403 reiserfs_check_lock_depth(sb
, "free_cnode");
405 journal
->j_cnode_used
--;
406 journal
->j_cnode_free
++;
407 /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
408 cn
->next
= journal
->j_cnode_free_list
;
409 if (journal
->j_cnode_free_list
) {
410 journal
->j_cnode_free_list
->prev
= cn
;
412 cn
->prev
= NULL
; /* not needed with the memset, but I might kill the memset, and forget to do this */
413 journal
->j_cnode_free_list
= cn
;
416 static void clear_prepared_bits(struct buffer_head
*bh
)
418 clear_buffer_journal_prepared(bh
);
419 clear_buffer_journal_restore_dirty(bh
);
422 /* return a cnode with same dev, block number and size in table, or null if not found */
423 static inline struct reiserfs_journal_cnode
*get_journal_hash_dev(struct
427 reiserfs_journal_cnode
431 struct reiserfs_journal_cnode
*cn
;
432 cn
= journal_hash(table
, sb
, bl
);
434 if (cn
->blocknr
== bl
&& cn
->sb
== sb
)
438 return (struct reiserfs_journal_cnode
*)0;
442 ** this actually means 'can this block be reallocated yet?'. If you set search_all, a block can only be allocated
443 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
444 ** being overwritten by a replay after crashing.
446 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction. Since deleting
447 ** a block removes it from the current transaction, this case should never happen. If you don't set search_all, make
448 ** sure you never write the block without logging it.
450 ** next_zero_bit is a suggestion about the next block to try for find_forward.
451 ** when bl is rejected because it is set in a journal list bitmap, we search
452 ** for the next zero bit in the bitmap that rejected bl. Then, we return that
453 ** through next_zero_bit for find_forward to try.
455 ** Just because we return something in next_zero_bit does not mean we won't
456 ** reject it on the next call to reiserfs_in_journal
459 int reiserfs_in_journal(struct super_block
*sb
,
460 unsigned int bmap_nr
, int bit_nr
, int search_all
,
461 b_blocknr_t
* next_zero_bit
)
463 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
464 struct reiserfs_journal_cnode
*cn
;
465 struct reiserfs_list_bitmap
*jb
;
469 *next_zero_bit
= 0; /* always start this at zero. */
471 PROC_INFO_INC(sb
, journal
.in_journal
);
472 /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
473 ** if we crash before the transaction that freed it commits, this transaction won't
474 ** have committed either, and the block will never be written
477 for (i
= 0; i
< JOURNAL_NUM_BITMAPS
; i
++) {
478 PROC_INFO_INC(sb
, journal
.in_journal_bitmap
);
479 jb
= journal
->j_list_bitmap
+ i
;
480 if (jb
->journal_list
&& jb
->bitmaps
[bmap_nr
] &&
482 (unsigned long *)jb
->bitmaps
[bmap_nr
]->
485 find_next_zero_bit((unsigned long *)
486 (jb
->bitmaps
[bmap_nr
]->
488 sb
->s_blocksize
<< 3,
495 bl
= bmap_nr
* (sb
->s_blocksize
<< 3) + bit_nr
;
496 /* is it in any old transactions? */
499 get_journal_hash_dev(sb
, journal
->j_list_hash_table
, bl
))) {
503 /* is it in the current transaction. This should never happen */
504 if ((cn
= get_journal_hash_dev(sb
, journal
->j_hash_table
, bl
))) {
509 PROC_INFO_INC(sb
, journal
.in_journal_reusable
);
514 /* insert cn into table
516 static inline void insert_journal_hash(struct reiserfs_journal_cnode
**table
,
517 struct reiserfs_journal_cnode
*cn
)
519 struct reiserfs_journal_cnode
*cn_orig
;
521 cn_orig
= journal_hash(table
, cn
->sb
, cn
->blocknr
);
527 journal_hash(table
, cn
->sb
, cn
->blocknr
) = cn
;
530 /* lock the current transaction */
531 static inline void lock_journal(struct super_block
*sb
)
533 PROC_INFO_INC(sb
, journal
.lock_journal
);
535 reiserfs_mutex_lock_safe(&SB_JOURNAL(sb
)->j_mutex
, sb
);
538 /* unlock the current transaction */
539 static inline void unlock_journal(struct super_block
*sb
)
541 mutex_unlock(&SB_JOURNAL(sb
)->j_mutex
);
544 static inline void get_journal_list(struct reiserfs_journal_list
*jl
)
549 static inline void put_journal_list(struct super_block
*s
,
550 struct reiserfs_journal_list
*jl
)
552 if (jl
->j_refcount
< 1) {
553 reiserfs_panic(s
, "journal-2", "trans id %u, refcount at %d",
554 jl
->j_trans_id
, jl
->j_refcount
);
556 if (--jl
->j_refcount
== 0)
561 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
562 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
565 static void cleanup_freed_for_journal_list(struct super_block
*sb
,
566 struct reiserfs_journal_list
*jl
)
569 struct reiserfs_list_bitmap
*jb
= jl
->j_list_bitmap
;
571 cleanup_bitmap_list(sb
, jb
);
573 jl
->j_list_bitmap
->journal_list
= NULL
;
574 jl
->j_list_bitmap
= NULL
;
577 static int journal_list_still_alive(struct super_block
*s
,
578 unsigned int trans_id
)
580 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
581 struct list_head
*entry
= &journal
->j_journal_list
;
582 struct reiserfs_journal_list
*jl
;
584 if (!list_empty(entry
)) {
585 jl
= JOURNAL_LIST_ENTRY(entry
->next
);
586 if (jl
->j_trans_id
<= trans_id
) {
594 * If page->mapping was null, we failed to truncate this page for
595 * some reason. Most likely because it was truncated after being
596 * logged via data=journal.
598 * This does a check to see if the buffer belongs to one of these
599 * lost pages before doing the final put_bh. If page->mapping was
600 * null, it tries to free buffers on the page, which should make the
601 * final page_cache_release drop the page from the lru.
603 static void release_buffer_page(struct buffer_head
*bh
)
605 struct page
*page
= bh
->b_page
;
606 if (!page
->mapping
&& trylock_page(page
)) {
607 page_cache_get(page
);
610 try_to_free_buffers(page
);
612 page_cache_release(page
);
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
, "clm-2084",
624 "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
);
633 release_buffer_page(bh
);
636 static void reiserfs_end_ordered_io(struct buffer_head
*bh
, int uptodate
)
639 set_buffer_uptodate(bh
);
641 clear_buffer_uptodate(bh
);
646 static void submit_logged_buffer(struct buffer_head
*bh
)
649 bh
->b_end_io
= reiserfs_end_buffer_io_sync
;
650 clear_buffer_journal_new(bh
);
651 clear_buffer_dirty(bh
);
652 if (!test_clear_buffer_journal_test(bh
))
654 if (!buffer_uptodate(bh
))
656 submit_bh(WRITE
, bh
);
659 static void submit_ordered_buffer(struct buffer_head
*bh
)
662 bh
->b_end_io
= reiserfs_end_ordered_io
;
663 clear_buffer_dirty(bh
);
664 if (!buffer_uptodate(bh
))
666 submit_bh(WRITE
, bh
);
669 #define CHUNK_SIZE 32
670 struct buffer_chunk
{
671 struct buffer_head
*bh
[CHUNK_SIZE
];
675 static void write_chunk(struct buffer_chunk
*chunk
)
678 for (i
= 0; i
< chunk
->nr
; i
++) {
679 submit_logged_buffer(chunk
->bh
[i
]);
684 static void write_ordered_chunk(struct buffer_chunk
*chunk
)
687 for (i
= 0; i
< chunk
->nr
; i
++) {
688 submit_ordered_buffer(chunk
->bh
[i
]);
693 static int add_to_chunk(struct buffer_chunk
*chunk
, struct buffer_head
*bh
,
694 spinlock_t
* lock
, void (fn
) (struct buffer_chunk
*))
697 BUG_ON(chunk
->nr
>= CHUNK_SIZE
);
698 chunk
->bh
[chunk
->nr
++] = bh
;
699 if (chunk
->nr
>= CHUNK_SIZE
) {
710 static atomic_t nr_reiserfs_jh
= ATOMIC_INIT(0);
711 static struct reiserfs_jh
*alloc_jh(void)
713 struct reiserfs_jh
*jh
;
715 jh
= kmalloc(sizeof(*jh
), GFP_NOFS
);
717 atomic_inc(&nr_reiserfs_jh
);
725 * we want to free the jh when the buffer has been written
728 void reiserfs_free_jh(struct buffer_head
*bh
)
730 struct reiserfs_jh
*jh
;
734 bh
->b_private
= NULL
;
736 list_del_init(&jh
->list
);
738 if (atomic_read(&nr_reiserfs_jh
) <= 0)
740 atomic_dec(&nr_reiserfs_jh
);
745 static inline int __add_jh(struct reiserfs_journal
*j
, struct buffer_head
*bh
,
748 struct reiserfs_jh
*jh
;
751 spin_lock(&j
->j_dirty_buffers_lock
);
752 if (!bh
->b_private
) {
753 spin_unlock(&j
->j_dirty_buffers_lock
);
757 list_del_init(&jh
->list
);
762 spin_lock(&j
->j_dirty_buffers_lock
);
763 /* buffer must be locked for __add_jh, should be able to have
764 * two adds at the same time
766 BUG_ON(bh
->b_private
);
770 jh
->jl
= j
->j_current_jl
;
772 list_add_tail(&jh
->list
, &jh
->jl
->j_tail_bh_list
);
774 list_add_tail(&jh
->list
, &jh
->jl
->j_bh_list
);
776 spin_unlock(&j
->j_dirty_buffers_lock
);
780 int reiserfs_add_tail_list(struct inode
*inode
, struct buffer_head
*bh
)
782 return __add_jh(SB_JOURNAL(inode
->i_sb
), bh
, 1);
784 int reiserfs_add_ordered_list(struct inode
*inode
, struct buffer_head
*bh
)
786 return __add_jh(SB_JOURNAL(inode
->i_sb
), bh
, 0);
789 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
790 static int write_ordered_buffers(spinlock_t
* lock
,
791 struct reiserfs_journal
*j
,
792 struct reiserfs_journal_list
*jl
,
793 struct list_head
*list
)
795 struct buffer_head
*bh
;
796 struct reiserfs_jh
*jh
;
797 int ret
= j
->j_errno
;
798 struct buffer_chunk chunk
;
799 struct list_head tmp
;
800 INIT_LIST_HEAD(&tmp
);
804 while (!list_empty(list
)) {
805 jh
= JH_ENTRY(list
->next
);
808 if (!trylock_buffer(bh
)) {
809 if (!buffer_dirty(bh
)) {
810 list_move(&jh
->list
, &tmp
);
815 write_ordered_chunk(&chunk
);
821 /* in theory, dirty non-uptodate buffers should never get here,
822 * but the upper layer io error paths still have a few quirks.
823 * Handle them here as gracefully as we can
825 if (!buffer_uptodate(bh
) && buffer_dirty(bh
)) {
826 clear_buffer_dirty(bh
);
829 if (buffer_dirty(bh
)) {
830 list_move(&jh
->list
, &tmp
);
831 add_to_chunk(&chunk
, bh
, lock
, write_ordered_chunk
);
833 reiserfs_free_jh(bh
);
838 cond_resched_lock(lock
);
842 write_ordered_chunk(&chunk
);
845 while (!list_empty(&tmp
)) {
846 jh
= JH_ENTRY(tmp
.prev
);
849 reiserfs_free_jh(bh
);
851 if (buffer_locked(bh
)) {
856 if (!buffer_uptodate(bh
)) {
859 /* ugly interaction with invalidatepage here.
860 * reiserfs_invalidate_page will pin any buffer that has a valid
861 * journal head from an older transaction. If someone else sets
862 * our buffer dirty after we write it in the first loop, and
863 * then someone truncates the page away, nobody will ever write
864 * the buffer. We're safe if we write the page one last time
865 * after freeing the journal header.
867 if (buffer_dirty(bh
) && unlikely(bh
->b_page
->mapping
== NULL
)) {
869 ll_rw_block(WRITE
, 1, &bh
);
873 cond_resched_lock(lock
);
879 static int flush_older_commits(struct super_block
*s
,
880 struct reiserfs_journal_list
*jl
)
882 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
883 struct reiserfs_journal_list
*other_jl
;
884 struct reiserfs_journal_list
*first_jl
;
885 struct list_head
*entry
;
886 unsigned int trans_id
= jl
->j_trans_id
;
887 unsigned int other_trans_id
;
888 unsigned int first_trans_id
;
892 * first we walk backwards to find the oldest uncommitted transation
895 entry
= jl
->j_list
.prev
;
897 other_jl
= JOURNAL_LIST_ENTRY(entry
);
898 if (entry
== &journal
->j_journal_list
||
899 atomic_read(&other_jl
->j_older_commits_done
))
903 entry
= other_jl
->j_list
.prev
;
906 /* if we didn't find any older uncommitted transactions, return now */
907 if (first_jl
== jl
) {
911 first_trans_id
= first_jl
->j_trans_id
;
913 entry
= &first_jl
->j_list
;
915 other_jl
= JOURNAL_LIST_ENTRY(entry
);
916 other_trans_id
= other_jl
->j_trans_id
;
918 if (other_trans_id
< trans_id
) {
919 if (atomic_read(&other_jl
->j_commit_left
) != 0) {
920 flush_commit_list(s
, other_jl
, 0);
922 /* list we were called with is gone, return */
923 if (!journal_list_still_alive(s
, trans_id
))
926 /* the one we just flushed is gone, this means all
927 * older lists are also gone, so first_jl is no longer
928 * valid either. Go back to the beginning.
930 if (!journal_list_still_alive
931 (s
, other_trans_id
)) {
936 if (entry
== &journal
->j_journal_list
)
945 static int reiserfs_async_progress_wait(struct super_block
*s
)
947 struct reiserfs_journal
*j
= SB_JOURNAL(s
);
949 if (atomic_read(&j
->j_async_throttle
)) {
950 reiserfs_write_unlock(s
);
951 congestion_wait(BLK_RW_ASYNC
, HZ
/ 10);
952 reiserfs_write_lock(s
);
959 ** if this journal list still has commit blocks unflushed, send them to disk.
961 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
962 ** Before the commit block can by written, every other log block must be safely on disk
965 static int flush_commit_list(struct super_block
*s
,
966 struct reiserfs_journal_list
*jl
, int flushall
)
970 struct buffer_head
*tbh
= NULL
;
971 unsigned int trans_id
= jl
->j_trans_id
;
972 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
976 reiserfs_check_lock_depth(s
, "flush_commit_list");
978 if (atomic_read(&jl
->j_older_commits_done
)) {
982 /* before we can put our commit blocks on disk, we have to make sure everyone older than
985 BUG_ON(jl
->j_len
<= 0);
986 BUG_ON(trans_id
== journal
->j_trans_id
);
988 get_journal_list(jl
);
990 if (flush_older_commits(s
, jl
) == 1) {
991 /* list disappeared during flush_older_commits. return */
996 /* make sure nobody is trying to flush this one at the same time */
997 reiserfs_mutex_lock_safe(&jl
->j_commit_mutex
, s
);
999 if (!journal_list_still_alive(s
, trans_id
)) {
1000 mutex_unlock(&jl
->j_commit_mutex
);
1003 BUG_ON(jl
->j_trans_id
== 0);
1005 /* this commit is done, exit */
1006 if (atomic_read(&(jl
->j_commit_left
)) <= 0) {
1008 atomic_set(&(jl
->j_older_commits_done
), 1);
1010 mutex_unlock(&jl
->j_commit_mutex
);
1014 if (!list_empty(&jl
->j_bh_list
)) {
1018 * We might sleep in numerous places inside
1019 * write_ordered_buffers. Relax the write lock.
1021 reiserfs_write_unlock(s
);
1022 ret
= write_ordered_buffers(&journal
->j_dirty_buffers_lock
,
1023 journal
, jl
, &jl
->j_bh_list
);
1024 if (ret
< 0 && retval
== 0)
1026 reiserfs_write_lock(s
);
1028 BUG_ON(!list_empty(&jl
->j_bh_list
));
1030 * for the description block and all the log blocks, submit any buffers
1031 * that haven't already reached the disk. Try to write at least 256
1032 * log blocks. later on, we will only wait on blocks that correspond
1033 * to this transaction, but while we're unplugging we might as well
1034 * get a chunk of data on there.
1036 atomic_inc(&journal
->j_async_throttle
);
1037 write_len
= jl
->j_len
+ 1;
1038 if (write_len
< 256)
1040 for (i
= 0 ; i
< write_len
; i
++) {
1041 bn
= SB_ONDISK_JOURNAL_1st_BLOCK(s
) + (jl
->j_start
+ i
) %
1042 SB_ONDISK_JOURNAL_SIZE(s
);
1043 tbh
= journal_find_get_block(s
, bn
);
1045 if (buffer_dirty(tbh
)) {
1046 reiserfs_write_unlock(s
);
1047 ll_rw_block(WRITE
, 1, &tbh
);
1048 reiserfs_write_lock(s
);
1053 atomic_dec(&journal
->j_async_throttle
);
1055 for (i
= 0; i
< (jl
->j_len
+ 1); i
++) {
1056 bn
= SB_ONDISK_JOURNAL_1st_BLOCK(s
) +
1057 (jl
->j_start
+ i
) % SB_ONDISK_JOURNAL_SIZE(s
);
1058 tbh
= journal_find_get_block(s
, bn
);
1060 reiserfs_write_unlock(s
);
1061 wait_on_buffer(tbh
);
1062 reiserfs_write_lock(s
);
1063 // since we're using ll_rw_blk above, it might have skipped over
1064 // a locked buffer. Double check here
1066 /* redundant, sync_dirty_buffer() checks */
1067 if (buffer_dirty(tbh
)) {
1068 reiserfs_write_unlock(s
);
1069 sync_dirty_buffer(tbh
);
1070 reiserfs_write_lock(s
);
1072 if (unlikely(!buffer_uptodate(tbh
))) {
1073 #ifdef CONFIG_REISERFS_CHECK
1074 reiserfs_warning(s
, "journal-601",
1075 "buffer write failed");
1079 put_bh(tbh
); /* once for journal_find_get_block */
1080 put_bh(tbh
); /* once due to original getblk in do_journal_end */
1081 atomic_dec(&(jl
->j_commit_left
));
1084 BUG_ON(atomic_read(&(jl
->j_commit_left
)) != 1);
1086 /* If there was a write error in the journal - we can't commit
1087 * this transaction - it will be invalid and, if successful,
1088 * will just end up propagating the write error out to
1089 * the file system. */
1090 if (likely(!retval
&& !reiserfs_is_journal_aborted (journal
))) {
1091 if (buffer_dirty(jl
->j_commit_bh
))
1093 mark_buffer_dirty(jl
->j_commit_bh
) ;
1094 reiserfs_write_unlock(s
);
1095 if (reiserfs_barrier_flush(s
))
1096 __sync_dirty_buffer(jl
->j_commit_bh
, WRITE_FLUSH_FUA
);
1098 sync_dirty_buffer(jl
->j_commit_bh
);
1099 reiserfs_write_lock(s
);
1102 /* If there was a write error in the journal - we can't commit this
1103 * transaction - it will be invalid and, if successful, will just end
1104 * up propagating the write error out to the filesystem. */
1105 if (unlikely(!buffer_uptodate(jl
->j_commit_bh
))) {
1106 #ifdef CONFIG_REISERFS_CHECK
1107 reiserfs_warning(s
, "journal-615", "buffer write failed");
1111 bforget(jl
->j_commit_bh
);
1112 if (journal
->j_last_commit_id
!= 0 &&
1113 (jl
->j_trans_id
- journal
->j_last_commit_id
) != 1) {
1114 reiserfs_warning(s
, "clm-2200", "last commit %lu, current %lu",
1115 journal
->j_last_commit_id
, jl
->j_trans_id
);
1117 journal
->j_last_commit_id
= jl
->j_trans_id
;
1119 /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */
1120 cleanup_freed_for_journal_list(s
, jl
);
1122 retval
= retval
? retval
: journal
->j_errno
;
1124 /* mark the metadata dirty */
1126 dirty_one_transaction(s
, jl
);
1127 atomic_dec(&(jl
->j_commit_left
));
1130 atomic_set(&(jl
->j_older_commits_done
), 1);
1132 mutex_unlock(&jl
->j_commit_mutex
);
1134 put_journal_list(s
, jl
);
1137 reiserfs_abort(s
, retval
, "Journal write error in %s",
1143 ** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or
1144 ** returns NULL if it can't find anything
1146 static struct reiserfs_journal_list
*find_newer_jl_for_cn(struct
1147 reiserfs_journal_cnode
1150 struct super_block
*sb
= cn
->sb
;
1151 b_blocknr_t blocknr
= cn
->blocknr
;
1155 if (cn
->sb
== sb
&& cn
->blocknr
== blocknr
&& cn
->jlist
) {
1163 static int newer_jl_done(struct reiserfs_journal_cnode
*cn
)
1165 struct super_block
*sb
= cn
->sb
;
1166 b_blocknr_t blocknr
= cn
->blocknr
;
1170 if (cn
->sb
== sb
&& cn
->blocknr
== blocknr
&& cn
->jlist
&&
1171 atomic_read(&cn
->jlist
->j_commit_left
) != 0)
1178 static void remove_journal_hash(struct super_block
*,
1179 struct reiserfs_journal_cnode
**,
1180 struct reiserfs_journal_list
*, unsigned long,
1184 ** once all the real blocks have been flushed, it is safe to remove them from the
1185 ** journal list for this transaction. Aside from freeing the cnode, this also allows the
1186 ** block to be reallocated for data blocks if it had been deleted.
1188 static void remove_all_from_journal_list(struct super_block
*sb
,
1189 struct reiserfs_journal_list
*jl
,
1192 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1193 struct reiserfs_journal_cnode
*cn
, *last
;
1194 cn
= jl
->j_realblock
;
1196 /* which is better, to lock once around the whole loop, or
1197 ** to lock for each call to remove_journal_hash?
1200 if (cn
->blocknr
!= 0) {
1202 reiserfs_warning(sb
, "reiserfs-2201",
1203 "block %u, bh is %d, state %ld",
1204 cn
->blocknr
, cn
->bh
? 1 : 0,
1208 remove_journal_hash(sb
, journal
->j_list_hash_table
,
1209 jl
, cn
->blocknr
, 1);
1213 free_cnode(sb
, last
);
1215 jl
->j_realblock
= NULL
;
1219 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1220 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1221 ** releasing blocks in this transaction for reuse as data blocks.
1222 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1225 static int _update_journal_header_block(struct super_block
*sb
,
1226 unsigned long offset
,
1227 unsigned int trans_id
)
1229 struct reiserfs_journal_header
*jh
;
1230 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1232 if (reiserfs_is_journal_aborted(journal
))
1235 if (trans_id
>= journal
->j_last_flush_trans_id
) {
1236 if (buffer_locked((journal
->j_header_bh
))) {
1237 reiserfs_write_unlock(sb
);
1238 wait_on_buffer((journal
->j_header_bh
));
1239 reiserfs_write_lock(sb
);
1240 if (unlikely(!buffer_uptodate(journal
->j_header_bh
))) {
1241 #ifdef CONFIG_REISERFS_CHECK
1242 reiserfs_warning(sb
, "journal-699",
1243 "buffer write failed");
1248 journal
->j_last_flush_trans_id
= trans_id
;
1249 journal
->j_first_unflushed_offset
= offset
;
1250 jh
= (struct reiserfs_journal_header
*)(journal
->j_header_bh
->
1252 jh
->j_last_flush_trans_id
= cpu_to_le32(trans_id
);
1253 jh
->j_first_unflushed_offset
= cpu_to_le32(offset
);
1254 jh
->j_mount_id
= cpu_to_le32(journal
->j_mount_id
);
1256 set_buffer_dirty(journal
->j_header_bh
);
1257 reiserfs_write_unlock(sb
);
1259 if (reiserfs_barrier_flush(sb
))
1260 __sync_dirty_buffer(journal
->j_header_bh
, WRITE_FLUSH_FUA
);
1262 sync_dirty_buffer(journal
->j_header_bh
);
1264 reiserfs_write_lock(sb
);
1265 if (!buffer_uptodate(journal
->j_header_bh
)) {
1266 reiserfs_warning(sb
, "journal-837",
1267 "IO error during journal replay");
1274 static int update_journal_header_block(struct super_block
*sb
,
1275 unsigned long offset
,
1276 unsigned int trans_id
)
1278 return _update_journal_header_block(sb
, offset
, trans_id
);
1282 ** flush any and all journal lists older than you are
1283 ** can only be called from flush_journal_list
1285 static int flush_older_journal_lists(struct super_block
*sb
,
1286 struct reiserfs_journal_list
*jl
)
1288 struct list_head
*entry
;
1289 struct reiserfs_journal_list
*other_jl
;
1290 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1291 unsigned int trans_id
= jl
->j_trans_id
;
1293 /* we know we are the only ones flushing things, no extra race
1294 * protection is required.
1297 entry
= journal
->j_journal_list
.next
;
1299 if (entry
== &journal
->j_journal_list
)
1301 other_jl
= JOURNAL_LIST_ENTRY(entry
);
1302 if (other_jl
->j_trans_id
< trans_id
) {
1303 BUG_ON(other_jl
->j_refcount
<= 0);
1304 /* do not flush all */
1305 flush_journal_list(sb
, other_jl
, 0);
1307 /* other_jl is now deleted from the list */
1313 static void del_from_work_list(struct super_block
*s
,
1314 struct reiserfs_journal_list
*jl
)
1316 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1317 if (!list_empty(&jl
->j_working_list
)) {
1318 list_del_init(&jl
->j_working_list
);
1319 journal
->j_num_work_lists
--;
1323 /* flush a journal list, both commit and real blocks
1325 ** always set flushall to 1, unless you are calling from inside
1326 ** flush_journal_list
1328 ** IMPORTANT. This can only be called while there are no journal writers,
1329 ** and the journal is locked. That means it can only be called from
1330 ** do_journal_end, or by journal_release
1332 static int flush_journal_list(struct super_block
*s
,
1333 struct reiserfs_journal_list
*jl
, int flushall
)
1335 struct reiserfs_journal_list
*pjl
;
1336 struct reiserfs_journal_cnode
*cn
, *last
;
1340 struct buffer_head
*saved_bh
;
1341 unsigned long j_len_saved
= jl
->j_len
;
1342 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1345 BUG_ON(j_len_saved
<= 0);
1347 if (atomic_read(&journal
->j_wcount
) != 0) {
1348 reiserfs_warning(s
, "clm-2048", "called with wcount %d",
1349 atomic_read(&journal
->j_wcount
));
1351 BUG_ON(jl
->j_trans_id
== 0);
1353 /* if flushall == 0, the lock is already held */
1355 reiserfs_mutex_lock_safe(&journal
->j_flush_mutex
, s
);
1356 } else if (mutex_trylock(&journal
->j_flush_mutex
)) {
1361 if (j_len_saved
> journal
->j_trans_max
) {
1362 reiserfs_panic(s
, "journal-715", "length is %lu, trans id %lu",
1363 j_len_saved
, jl
->j_trans_id
);
1367 /* if all the work is already done, get out of here */
1368 if (atomic_read(&(jl
->j_nonzerolen
)) <= 0 &&
1369 atomic_read(&(jl
->j_commit_left
)) <= 0) {
1370 goto flush_older_and_return
;
1373 /* start by putting the commit list on disk. This will also flush
1374 ** the commit lists of any olders transactions
1376 flush_commit_list(s
, jl
, 1);
1378 if (!(jl
->j_state
& LIST_DIRTY
)
1379 && !reiserfs_is_journal_aborted(journal
))
1382 /* are we done now? */
1383 if (atomic_read(&(jl
->j_nonzerolen
)) <= 0 &&
1384 atomic_read(&(jl
->j_commit_left
)) <= 0) {
1385 goto flush_older_and_return
;
1388 /* loop through each cnode, see if we need to write it,
1389 ** or wait on a more recent transaction, or just ignore it
1391 if (atomic_read(&(journal
->j_wcount
)) != 0) {
1392 reiserfs_panic(s
, "journal-844", "journal list is flushing, "
1395 cn
= jl
->j_realblock
;
1400 /* blocknr of 0 is no longer in the hash, ignore it */
1401 if (cn
->blocknr
== 0) {
1405 /* This transaction failed commit. Don't write out to the disk */
1406 if (!(jl
->j_state
& LIST_DIRTY
))
1409 pjl
= find_newer_jl_for_cn(cn
);
1410 /* the order is important here. We check pjl to make sure we
1411 ** don't clear BH_JDirty_wait if we aren't the one writing this
1414 if (!pjl
&& cn
->bh
) {
1417 /* we do this to make sure nobody releases the buffer while
1418 ** we are working with it
1422 if (buffer_journal_dirty(saved_bh
)) {
1423 BUG_ON(!can_dirty(cn
));
1426 } else if (can_dirty(cn
)) {
1427 /* everything with !pjl && jwait should be writable */
1432 /* if someone has this block in a newer transaction, just make
1433 ** sure they are committed, and don't try writing it to disk
1436 if (atomic_read(&pjl
->j_commit_left
))
1437 flush_commit_list(s
, pjl
, 1);
1441 /* bh == NULL when the block got to disk on its own, OR,
1442 ** the block got freed in a future transaction
1444 if (saved_bh
== NULL
) {
1448 /* this should never happen. kupdate_one_transaction has this list
1449 ** locked while it works, so we should never see a buffer here that
1450 ** is not marked JDirty_wait
1452 if ((!was_jwait
) && !buffer_locked(saved_bh
)) {
1453 reiserfs_warning(s
, "journal-813",
1454 "BAD! buffer %llu %cdirty %cjwait, "
1455 "not in a newer tranasction",
1456 (unsigned long long)saved_bh
->
1457 b_blocknr
, was_dirty
? ' ' : '!',
1458 was_jwait
? ' ' : '!');
1461 /* we inc again because saved_bh gets decremented at free_cnode */
1463 set_bit(BLOCK_NEEDS_FLUSH
, &cn
->state
);
1464 lock_buffer(saved_bh
);
1465 BUG_ON(cn
->blocknr
!= saved_bh
->b_blocknr
);
1466 if (buffer_dirty(saved_bh
))
1467 submit_logged_buffer(saved_bh
);
1469 unlock_buffer(saved_bh
);
1472 reiserfs_warning(s
, "clm-2082",
1473 "Unable to flush buffer %llu in %s",
1474 (unsigned long long)saved_bh
->
1475 b_blocknr
, __func__
);
1481 /* we incremented this to keep others from taking the buffer head away */
1483 if (atomic_read(&(saved_bh
->b_count
)) < 0) {
1484 reiserfs_warning(s
, "journal-945",
1485 "saved_bh->b_count < 0");
1490 cn
= jl
->j_realblock
;
1492 if (test_bit(BLOCK_NEEDS_FLUSH
, &cn
->state
)) {
1494 reiserfs_panic(s
, "journal-1011",
1498 reiserfs_write_unlock(s
);
1499 wait_on_buffer(cn
->bh
);
1500 reiserfs_write_lock(s
);
1503 reiserfs_panic(s
, "journal-1012",
1506 if (unlikely(!buffer_uptodate(cn
->bh
))) {
1507 #ifdef CONFIG_REISERFS_CHECK
1508 reiserfs_warning(s
, "journal-949",
1509 "buffer write failed");
1513 /* note, we must clear the JDirty_wait bit after the up to date
1514 ** check, otherwise we race against our flushpage routine
1516 BUG_ON(!test_clear_buffer_journal_dirty
1519 /* drop one ref for us */
1521 /* drop one ref for journal_mark_dirty */
1522 release_buffer_page(cn
->bh
);
1529 reiserfs_abort(s
, -EIO
,
1530 "Write error while pushing transaction to disk in %s",
1532 flush_older_and_return
:
1534 /* before we can update the journal header block, we _must_ flush all
1535 ** real blocks from all older transactions to disk. This is because
1536 ** once the header block is updated, this transaction will not be
1537 ** replayed after a crash
1540 flush_older_journal_lists(s
, jl
);
1543 err
= journal
->j_errno
;
1544 /* before we can remove everything from the hash tables for this
1545 ** transaction, we must make sure it can never be replayed
1547 ** since we are only called from do_journal_end, we know for sure there
1548 ** are no allocations going on while we are flushing journal lists. So,
1549 ** we only need to update the journal header block for the last list
1552 if (!err
&& flushall
) {
1554 update_journal_header_block(s
,
1555 (jl
->j_start
+ jl
->j_len
+
1556 2) % SB_ONDISK_JOURNAL_SIZE(s
),
1559 reiserfs_abort(s
, -EIO
,
1560 "Write error while updating journal header in %s",
1563 remove_all_from_journal_list(s
, jl
, 0);
1564 list_del_init(&jl
->j_list
);
1565 journal
->j_num_lists
--;
1566 del_from_work_list(s
, jl
);
1568 if (journal
->j_last_flush_id
!= 0 &&
1569 (jl
->j_trans_id
- journal
->j_last_flush_id
) != 1) {
1570 reiserfs_warning(s
, "clm-2201", "last flush %lu, current %lu",
1571 journal
->j_last_flush_id
, jl
->j_trans_id
);
1573 journal
->j_last_flush_id
= jl
->j_trans_id
;
1575 /* not strictly required since we are freeing the list, but it should
1576 * help find code using dead lists later on
1579 atomic_set(&(jl
->j_nonzerolen
), 0);
1581 jl
->j_realblock
= NULL
;
1582 jl
->j_commit_bh
= NULL
;
1585 put_journal_list(s
, jl
);
1587 mutex_unlock(&journal
->j_flush_mutex
);
1591 static int test_transaction(struct super_block
*s
,
1592 struct reiserfs_journal_list
*jl
)
1594 struct reiserfs_journal_cnode
*cn
;
1596 if (jl
->j_len
== 0 || atomic_read(&jl
->j_nonzerolen
) == 0)
1599 cn
= jl
->j_realblock
;
1601 /* if the blocknr == 0, this has been cleared from the hash,
1604 if (cn
->blocknr
== 0) {
1607 if (cn
->bh
&& !newer_jl_done(cn
))
1616 static int write_one_transaction(struct super_block
*s
,
1617 struct reiserfs_journal_list
*jl
,
1618 struct buffer_chunk
*chunk
)
1620 struct reiserfs_journal_cnode
*cn
;
1623 jl
->j_state
|= LIST_TOUCHED
;
1624 del_from_work_list(s
, jl
);
1625 if (jl
->j_len
== 0 || atomic_read(&jl
->j_nonzerolen
) == 0) {
1629 cn
= jl
->j_realblock
;
1631 /* if the blocknr == 0, this has been cleared from the hash,
1634 if (cn
->blocknr
== 0) {
1637 if (cn
->bh
&& can_dirty(cn
) && buffer_dirty(cn
->bh
)) {
1638 struct buffer_head
*tmp_bh
;
1639 /* we can race against journal_mark_freed when we try
1640 * to lock_buffer(cn->bh), so we have to inc the buffer
1641 * count, and recheck things after locking
1645 lock_buffer(tmp_bh
);
1646 if (cn
->bh
&& can_dirty(cn
) && buffer_dirty(tmp_bh
)) {
1647 if (!buffer_journal_dirty(tmp_bh
) ||
1648 buffer_journal_prepared(tmp_bh
))
1650 add_to_chunk(chunk
, tmp_bh
, NULL
, write_chunk
);
1653 /* note, cn->bh might be null now */
1654 unlock_buffer(tmp_bh
);
1665 /* used by flush_commit_list */
1666 static int dirty_one_transaction(struct super_block
*s
,
1667 struct reiserfs_journal_list
*jl
)
1669 struct reiserfs_journal_cnode
*cn
;
1670 struct reiserfs_journal_list
*pjl
;
1673 jl
->j_state
|= LIST_DIRTY
;
1674 cn
= jl
->j_realblock
;
1676 /* look for a more recent transaction that logged this
1677 ** buffer. Only the most recent transaction with a buffer in
1678 ** it is allowed to send that buffer to disk
1680 pjl
= find_newer_jl_for_cn(cn
);
1681 if (!pjl
&& cn
->blocknr
&& cn
->bh
1682 && buffer_journal_dirty(cn
->bh
)) {
1683 BUG_ON(!can_dirty(cn
));
1684 /* if the buffer is prepared, it will either be logged
1685 * or restored. If restored, we need to make sure
1686 * it actually gets marked dirty
1688 clear_buffer_journal_new(cn
->bh
);
1689 if (buffer_journal_prepared(cn
->bh
)) {
1690 set_buffer_journal_restore_dirty(cn
->bh
);
1692 set_buffer_journal_test(cn
->bh
);
1693 mark_buffer_dirty(cn
->bh
);
1701 static int kupdate_transactions(struct super_block
*s
,
1702 struct reiserfs_journal_list
*jl
,
1703 struct reiserfs_journal_list
**next_jl
,
1704 unsigned int *next_trans_id
,
1705 int num_blocks
, int num_trans
)
1709 int transactions_flushed
= 0;
1710 unsigned int orig_trans_id
= jl
->j_trans_id
;
1711 struct buffer_chunk chunk
;
1712 struct list_head
*entry
;
1713 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1716 reiserfs_mutex_lock_safe(&journal
->j_flush_mutex
, s
);
1717 if (!journal_list_still_alive(s
, orig_trans_id
)) {
1721 /* we've got j_flush_mutex held, nobody is going to delete any
1722 * of these lists out from underneath us
1724 while ((num_trans
&& transactions_flushed
< num_trans
) ||
1725 (!num_trans
&& written
< num_blocks
)) {
1727 if (jl
->j_len
== 0 || (jl
->j_state
& LIST_TOUCHED
) ||
1728 atomic_read(&jl
->j_commit_left
)
1729 || !(jl
->j_state
& LIST_DIRTY
)) {
1730 del_from_work_list(s
, jl
);
1733 ret
= write_one_transaction(s
, jl
, &chunk
);
1737 transactions_flushed
++;
1739 entry
= jl
->j_list
.next
;
1742 if (entry
== &journal
->j_journal_list
) {
1745 jl
= JOURNAL_LIST_ENTRY(entry
);
1747 /* don't bother with older transactions */
1748 if (jl
->j_trans_id
<= orig_trans_id
)
1752 write_chunk(&chunk
);
1756 mutex_unlock(&journal
->j_flush_mutex
);
1760 /* for o_sync and fsync heavy applications, they tend to use
1761 ** all the journa list slots with tiny transactions. These
1762 ** trigger lots and lots of calls to update the header block, which
1763 ** adds seeks and slows things down.
1765 ** This function tries to clear out a large chunk of the journal lists
1766 ** at once, which makes everything faster since only the newest journal
1767 ** list updates the header block
1769 static int flush_used_journal_lists(struct super_block
*s
,
1770 struct reiserfs_journal_list
*jl
)
1772 unsigned long len
= 0;
1773 unsigned long cur_len
;
1777 struct reiserfs_journal_list
*tjl
;
1778 struct reiserfs_journal_list
*flush_jl
;
1779 unsigned int trans_id
;
1780 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
1782 flush_jl
= tjl
= jl
;
1784 /* in data logging mode, try harder to flush a lot of blocks */
1785 if (reiserfs_data_log(s
))
1787 /* flush for 256 transactions or limit blocks, whichever comes first */
1788 for (i
= 0; i
< 256 && len
< limit
; i
++) {
1789 if (atomic_read(&tjl
->j_commit_left
) ||
1790 tjl
->j_trans_id
< jl
->j_trans_id
) {
1793 cur_len
= atomic_read(&tjl
->j_nonzerolen
);
1795 tjl
->j_state
&= ~LIST_TOUCHED
;
1799 if (tjl
->j_list
.next
== &journal
->j_journal_list
)
1801 tjl
= JOURNAL_LIST_ENTRY(tjl
->j_list
.next
);
1803 /* try to find a group of blocks we can flush across all the
1804 ** transactions, but only bother if we've actually spanned
1805 ** across multiple lists
1807 if (flush_jl
!= jl
) {
1808 ret
= kupdate_transactions(s
, jl
, &tjl
, &trans_id
, len
, i
);
1810 flush_journal_list(s
, flush_jl
, 1);
1815 ** removes any nodes in table with name block and dev as bh.
1816 ** only touchs the hnext and hprev pointers.
1818 void remove_journal_hash(struct super_block
*sb
,
1819 struct reiserfs_journal_cnode
**table
,
1820 struct reiserfs_journal_list
*jl
,
1821 unsigned long block
, int remove_freed
)
1823 struct reiserfs_journal_cnode
*cur
;
1824 struct reiserfs_journal_cnode
**head
;
1826 head
= &(journal_hash(table
, sb
, block
));
1832 if (cur
->blocknr
== block
&& cur
->sb
== sb
1833 && (jl
== NULL
|| jl
== cur
->jlist
)
1834 && (!test_bit(BLOCK_FREED
, &cur
->state
) || remove_freed
)) {
1836 cur
->hnext
->hprev
= cur
->hprev
;
1839 cur
->hprev
->hnext
= cur
->hnext
;
1846 if (cur
->bh
&& cur
->jlist
) /* anybody who clears the cur->bh will also dec the nonzerolen */
1847 atomic_dec(&(cur
->jlist
->j_nonzerolen
));
1855 static void free_journal_ram(struct super_block
*sb
)
1857 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1858 kfree(journal
->j_current_jl
);
1859 journal
->j_num_lists
--;
1861 vfree(journal
->j_cnode_free_orig
);
1862 free_list_bitmaps(sb
, journal
->j_list_bitmap
);
1863 free_bitmap_nodes(sb
); /* must be after free_list_bitmaps */
1864 if (journal
->j_header_bh
) {
1865 brelse(journal
->j_header_bh
);
1867 /* j_header_bh is on the journal dev, make sure not to release the journal
1868 * dev until we brelse j_header_bh
1870 release_journal_dev(sb
, journal
);
1875 ** call on unmount. Only set error to 1 if you haven't made your way out
1876 ** of read_super() yet. Any other caller must keep error at 0.
1878 static int do_journal_release(struct reiserfs_transaction_handle
*th
,
1879 struct super_block
*sb
, int error
)
1881 struct reiserfs_transaction_handle myth
;
1883 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
1885 /* we only want to flush out transactions if we were called with error == 0
1887 if (!error
&& !(sb
->s_flags
& MS_RDONLY
)) {
1888 /* end the current trans */
1889 BUG_ON(!th
->t_trans_id
);
1890 do_journal_end(th
, sb
, 10, FLUSH_ALL
);
1892 /* make sure something gets logged to force our way into the flush code */
1893 if (!journal_join(&myth
, sb
, 1)) {
1894 reiserfs_prepare_for_journal(sb
,
1895 SB_BUFFER_WITH_SB(sb
),
1897 journal_mark_dirty(&myth
, sb
,
1898 SB_BUFFER_WITH_SB(sb
));
1899 do_journal_end(&myth
, sb
, 1, FLUSH_ALL
);
1904 /* this also catches errors during the do_journal_end above */
1905 if (!error
&& reiserfs_is_journal_aborted(journal
)) {
1906 memset(&myth
, 0, sizeof(myth
));
1907 if (!journal_join_abort(&myth
, sb
, 1)) {
1908 reiserfs_prepare_for_journal(sb
,
1909 SB_BUFFER_WITH_SB(sb
),
1911 journal_mark_dirty(&myth
, sb
,
1912 SB_BUFFER_WITH_SB(sb
));
1913 do_journal_end(&myth
, sb
, 1, FLUSH_ALL
);
1917 reiserfs_mounted_fs_count
--;
1918 /* wait for all commits to finish */
1919 cancel_delayed_work(&SB_JOURNAL(sb
)->j_work
);
1922 * We must release the write lock here because
1923 * the workqueue job (flush_async_commit) needs this lock
1925 reiserfs_write_unlock(sb
);
1927 cancel_delayed_work_sync(&REISERFS_SB(sb
)->old_work
);
1928 flush_workqueue(commit_wq
);
1930 if (!reiserfs_mounted_fs_count
) {
1931 destroy_workqueue(commit_wq
);
1935 free_journal_ram(sb
);
1937 reiserfs_write_lock(sb
);
1943 ** call on unmount. flush all journal trans, release all alloc'd ram
1945 int journal_release(struct reiserfs_transaction_handle
*th
,
1946 struct super_block
*sb
)
1948 return do_journal_release(th
, sb
, 0);
1952 ** only call from an error condition inside reiserfs_read_super!
1954 int journal_release_error(struct reiserfs_transaction_handle
*th
,
1955 struct super_block
*sb
)
1957 return do_journal_release(th
, sb
, 1);
1960 /* compares description block with commit block. returns 1 if they differ, 0 if they are the same */
1961 static int journal_compare_desc_commit(struct super_block
*sb
,
1962 struct reiserfs_journal_desc
*desc
,
1963 struct reiserfs_journal_commit
*commit
)
1965 if (get_commit_trans_id(commit
) != get_desc_trans_id(desc
) ||
1966 get_commit_trans_len(commit
) != get_desc_trans_len(desc
) ||
1967 get_commit_trans_len(commit
) > SB_JOURNAL(sb
)->j_trans_max
||
1968 get_commit_trans_len(commit
) <= 0) {
1974 /* returns 0 if it did not find a description block
1975 ** returns -1 if it found a corrupt commit block
1976 ** returns 1 if both desc and commit were valid
1978 static int journal_transaction_is_valid(struct super_block
*sb
,
1979 struct buffer_head
*d_bh
,
1980 unsigned int *oldest_invalid_trans_id
,
1981 unsigned long *newest_mount_id
)
1983 struct reiserfs_journal_desc
*desc
;
1984 struct reiserfs_journal_commit
*commit
;
1985 struct buffer_head
*c_bh
;
1986 unsigned long offset
;
1991 desc
= (struct reiserfs_journal_desc
*)d_bh
->b_data
;
1992 if (get_desc_trans_len(desc
) > 0
1993 && !memcmp(get_journal_desc_magic(d_bh
), JOURNAL_DESC_MAGIC
, 8)) {
1994 if (oldest_invalid_trans_id
&& *oldest_invalid_trans_id
1995 && get_desc_trans_id(desc
) > *oldest_invalid_trans_id
) {
1996 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
1997 "journal-986: transaction "
1998 "is valid returning because trans_id %d is greater than "
1999 "oldest_invalid %lu",
2000 get_desc_trans_id(desc
),
2001 *oldest_invalid_trans_id
);
2005 && *newest_mount_id
> get_desc_mount_id(desc
)) {
2006 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2007 "journal-1087: transaction "
2008 "is valid returning because mount_id %d is less than "
2009 "newest_mount_id %lu",
2010 get_desc_mount_id(desc
),
2014 if (get_desc_trans_len(desc
) > SB_JOURNAL(sb
)->j_trans_max
) {
2015 reiserfs_warning(sb
, "journal-2018",
2016 "Bad transaction length %d "
2017 "encountered, ignoring transaction",
2018 get_desc_trans_len(desc
));
2021 offset
= d_bh
->b_blocknr
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
);
2023 /* ok, we have a journal description block, lets see if the transaction was valid */
2026 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2027 ((offset
+ get_desc_trans_len(desc
) +
2028 1) % SB_ONDISK_JOURNAL_SIZE(sb
)));
2031 commit
= (struct reiserfs_journal_commit
*)c_bh
->b_data
;
2032 if (journal_compare_desc_commit(sb
, desc
, commit
)) {
2033 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2034 "journal_transaction_is_valid, commit offset %ld had bad "
2035 "time %d or length %d",
2037 SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2038 get_commit_trans_id(commit
),
2039 get_commit_trans_len(commit
));
2041 if (oldest_invalid_trans_id
) {
2042 *oldest_invalid_trans_id
=
2043 get_desc_trans_id(desc
);
2044 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2046 "transaction_is_valid setting oldest invalid trans_id "
2048 get_desc_trans_id(desc
));
2053 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2054 "journal-1006: found valid "
2055 "transaction start offset %llu, len %d id %d",
2057 SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2058 get_desc_trans_len(desc
),
2059 get_desc_trans_id(desc
));
2066 static void brelse_array(struct buffer_head
**heads
, int num
)
2069 for (i
= 0; i
< num
; i
++) {
2075 ** given the start, and values for the oldest acceptable transactions,
2076 ** this either reads in a replays a transaction, or returns because the transaction
2077 ** is invalid, or too old.
2079 static int journal_read_transaction(struct super_block
*sb
,
2080 unsigned long cur_dblock
,
2081 unsigned long oldest_start
,
2082 unsigned int oldest_trans_id
,
2083 unsigned long newest_mount_id
)
2085 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
2086 struct reiserfs_journal_desc
*desc
;
2087 struct reiserfs_journal_commit
*commit
;
2088 unsigned int trans_id
= 0;
2089 struct buffer_head
*c_bh
;
2090 struct buffer_head
*d_bh
;
2091 struct buffer_head
**log_blocks
= NULL
;
2092 struct buffer_head
**real_blocks
= NULL
;
2093 unsigned int trans_offset
;
2097 d_bh
= journal_bread(sb
, cur_dblock
);
2100 desc
= (struct reiserfs_journal_desc
*)d_bh
->b_data
;
2101 trans_offset
= d_bh
->b_blocknr
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
);
2102 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
, "journal-1037: "
2103 "journal_read_transaction, offset %llu, len %d mount_id %d",
2104 d_bh
->b_blocknr
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2105 get_desc_trans_len(desc
), get_desc_mount_id(desc
));
2106 if (get_desc_trans_id(desc
) < oldest_trans_id
) {
2107 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
, "journal-1039: "
2108 "journal_read_trans skipping because %lu is too old",
2110 SB_ONDISK_JOURNAL_1st_BLOCK(sb
));
2114 if (get_desc_mount_id(desc
) != newest_mount_id
) {
2115 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
, "journal-1146: "
2116 "journal_read_trans skipping because %d is != "
2117 "newest_mount_id %lu", get_desc_mount_id(desc
),
2122 c_bh
= journal_bread(sb
, SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2123 ((trans_offset
+ get_desc_trans_len(desc
) + 1) %
2124 SB_ONDISK_JOURNAL_SIZE(sb
)));
2129 commit
= (struct reiserfs_journal_commit
*)c_bh
->b_data
;
2130 if (journal_compare_desc_commit(sb
, desc
, commit
)) {
2131 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2132 "journal_read_transaction, "
2133 "commit offset %llu had bad time %d or length %d",
2135 SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2136 get_commit_trans_id(commit
),
2137 get_commit_trans_len(commit
));
2143 if (bdev_read_only(sb
->s_bdev
)) {
2144 reiserfs_warning(sb
, "clm-2076",
2145 "device is readonly, unable to replay log");
2151 trans_id
= get_desc_trans_id(desc
);
2152 /* now we know we've got a good transaction, and it was inside the valid time ranges */
2153 log_blocks
= kmalloc(get_desc_trans_len(desc
) *
2154 sizeof(struct buffer_head
*), GFP_NOFS
);
2155 real_blocks
= kmalloc(get_desc_trans_len(desc
) *
2156 sizeof(struct buffer_head
*), GFP_NOFS
);
2157 if (!log_blocks
|| !real_blocks
) {
2162 reiserfs_warning(sb
, "journal-1169",
2163 "kmalloc failed, unable to mount FS");
2166 /* get all the buffer heads */
2167 trans_half
= journal_trans_half(sb
->s_blocksize
);
2168 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2171 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2173 i
) % SB_ONDISK_JOURNAL_SIZE(sb
));
2174 if (i
< trans_half
) {
2177 le32_to_cpu(desc
->j_realblock
[i
]));
2181 le32_to_cpu(commit
->
2182 j_realblock
[i
- trans_half
]));
2184 if (real_blocks
[i
]->b_blocknr
> SB_BLOCK_COUNT(sb
)) {
2185 reiserfs_warning(sb
, "journal-1207",
2186 "REPLAY FAILURE fsck required! "
2187 "Block to replay is outside of "
2191 /* make sure we don't try to replay onto log or reserved area */
2192 if (is_block_in_log_or_reserved_area
2193 (sb
, real_blocks
[i
]->b_blocknr
)) {
2194 reiserfs_warning(sb
, "journal-1204",
2195 "REPLAY FAILURE fsck required! "
2196 "Trying to replay onto a log block");
2198 brelse_array(log_blocks
, i
);
2199 brelse_array(real_blocks
, i
);
2207 /* read in the log blocks, memcpy to the corresponding real block */
2208 ll_rw_block(READ
, get_desc_trans_len(desc
), log_blocks
);
2209 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2211 reiserfs_write_unlock(sb
);
2212 wait_on_buffer(log_blocks
[i
]);
2213 reiserfs_write_lock(sb
);
2215 if (!buffer_uptodate(log_blocks
[i
])) {
2216 reiserfs_warning(sb
, "journal-1212",
2217 "REPLAY FAILURE fsck required! "
2218 "buffer write failed");
2219 brelse_array(log_blocks
+ i
,
2220 get_desc_trans_len(desc
) - i
);
2221 brelse_array(real_blocks
, get_desc_trans_len(desc
));
2228 memcpy(real_blocks
[i
]->b_data
, log_blocks
[i
]->b_data
,
2229 real_blocks
[i
]->b_size
);
2230 set_buffer_uptodate(real_blocks
[i
]);
2231 brelse(log_blocks
[i
]);
2233 /* flush out the real blocks */
2234 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2235 set_buffer_dirty(real_blocks
[i
]);
2236 write_dirty_buffer(real_blocks
[i
], WRITE
);
2238 for (i
= 0; i
< get_desc_trans_len(desc
); i
++) {
2239 wait_on_buffer(real_blocks
[i
]);
2240 if (!buffer_uptodate(real_blocks
[i
])) {
2241 reiserfs_warning(sb
, "journal-1226",
2242 "REPLAY FAILURE, fsck required! "
2243 "buffer write failed");
2244 brelse_array(real_blocks
+ i
,
2245 get_desc_trans_len(desc
) - i
);
2252 brelse(real_blocks
[i
]);
2255 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2256 ((trans_offset
+ get_desc_trans_len(desc
) +
2257 2) % SB_ONDISK_JOURNAL_SIZE(sb
));
2258 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2259 "journal-1095: setting journal " "start to offset %ld",
2260 cur_dblock
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
));
2262 /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2263 journal
->j_start
= cur_dblock
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
);
2264 journal
->j_last_flush_trans_id
= trans_id
;
2265 journal
->j_trans_id
= trans_id
+ 1;
2266 /* check for trans_id overflow */
2267 if (journal
->j_trans_id
== 0)
2268 journal
->j_trans_id
= 10;
2276 /* This function reads blocks starting from block and to max_block of bufsize
2277 size (but no more than BUFNR blocks at a time). This proved to improve
2278 mounting speed on self-rebuilding raid5 arrays at least.
2279 Right now it is only used from journal code. But later we might use it
2281 Note: Do not use journal_getblk/sb_getblk functions here! */
2282 static struct buffer_head
*reiserfs_breada(struct block_device
*dev
,
2283 b_blocknr_t block
, int bufsize
,
2284 b_blocknr_t max_block
)
2286 struct buffer_head
*bhlist
[BUFNR
];
2287 unsigned int blocks
= BUFNR
;
2288 struct buffer_head
*bh
;
2291 bh
= __getblk(dev
, block
, bufsize
);
2292 if (buffer_uptodate(bh
))
2295 if (block
+ BUFNR
> max_block
) {
2296 blocks
= max_block
- block
;
2300 for (i
= 1; i
< blocks
; i
++) {
2301 bh
= __getblk(dev
, block
+ i
, bufsize
);
2302 if (buffer_uptodate(bh
)) {
2308 ll_rw_block(READ
, j
, bhlist
);
2309 for (i
= 1; i
< j
; i
++)
2313 if (buffer_uptodate(bh
))
2320 ** read and replay the log
2321 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2322 ** transaction. This tests that before finding all the transactions in the log, which makes normal mount times fast.
2324 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2326 ** On exit, it sets things up so the first transaction will work correctly.
2328 static int journal_read(struct super_block
*sb
)
2330 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
2331 struct reiserfs_journal_desc
*desc
;
2332 unsigned int oldest_trans_id
= 0;
2333 unsigned int oldest_invalid_trans_id
= 0;
2335 unsigned long oldest_start
= 0;
2336 unsigned long cur_dblock
= 0;
2337 unsigned long newest_mount_id
= 9;
2338 struct buffer_head
*d_bh
;
2339 struct reiserfs_journal_header
*jh
;
2340 int valid_journal_header
= 0;
2341 int replay_count
= 0;
2342 int continue_replay
= 1;
2344 char b
[BDEVNAME_SIZE
];
2346 cur_dblock
= SB_ONDISK_JOURNAL_1st_BLOCK(sb
);
2347 reiserfs_info(sb
, "checking transaction log (%s)\n",
2348 bdevname(journal
->j_dev_bd
, b
));
2349 start
= get_seconds();
2351 /* step 1, read in the journal header block. Check the transaction it says
2352 ** is the first unflushed, and if that transaction is not valid,
2355 journal
->j_header_bh
= journal_bread(sb
,
2356 SB_ONDISK_JOURNAL_1st_BLOCK(sb
)
2357 + SB_ONDISK_JOURNAL_SIZE(sb
));
2358 if (!journal
->j_header_bh
) {
2361 jh
= (struct reiserfs_journal_header
*)(journal
->j_header_bh
->b_data
);
2362 if (le32_to_cpu(jh
->j_first_unflushed_offset
) <
2363 SB_ONDISK_JOURNAL_SIZE(sb
)
2364 && le32_to_cpu(jh
->j_last_flush_trans_id
) > 0) {
2366 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2367 le32_to_cpu(jh
->j_first_unflushed_offset
);
2368 oldest_trans_id
= le32_to_cpu(jh
->j_last_flush_trans_id
) + 1;
2369 newest_mount_id
= le32_to_cpu(jh
->j_mount_id
);
2370 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2371 "journal-1153: found in "
2372 "header: first_unflushed_offset %d, last_flushed_trans_id "
2373 "%lu", le32_to_cpu(jh
->j_first_unflushed_offset
),
2374 le32_to_cpu(jh
->j_last_flush_trans_id
));
2375 valid_journal_header
= 1;
2377 /* now, we try to read the first unflushed offset. If it is not valid,
2378 ** there is nothing more we can do, and it makes no sense to read
2379 ** through the whole log.
2383 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2384 le32_to_cpu(jh
->j_first_unflushed_offset
));
2385 ret
= journal_transaction_is_valid(sb
, d_bh
, NULL
, NULL
);
2387 continue_replay
= 0;
2390 goto start_log_replay
;
2393 /* ok, there are transactions that need to be replayed. start with the first log block, find
2394 ** all the valid transactions, and pick out the oldest.
2396 while (continue_replay
2398 (SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2399 SB_ONDISK_JOURNAL_SIZE(sb
))) {
2400 /* Note that it is required for blocksize of primary fs device and journal
2401 device to be the same */
2403 reiserfs_breada(journal
->j_dev_bd
, cur_dblock
,
2405 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2406 SB_ONDISK_JOURNAL_SIZE(sb
));
2408 journal_transaction_is_valid(sb
, d_bh
,
2409 &oldest_invalid_trans_id
,
2412 desc
= (struct reiserfs_journal_desc
*)d_bh
->b_data
;
2413 if (oldest_start
== 0) { /* init all oldest_ values */
2414 oldest_trans_id
= get_desc_trans_id(desc
);
2415 oldest_start
= d_bh
->b_blocknr
;
2416 newest_mount_id
= get_desc_mount_id(desc
);
2417 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2418 "journal-1179: Setting "
2419 "oldest_start to offset %llu, trans_id %lu",
2421 SB_ONDISK_JOURNAL_1st_BLOCK
2422 (sb
), oldest_trans_id
);
2423 } else if (oldest_trans_id
> get_desc_trans_id(desc
)) {
2424 /* one we just read was older */
2425 oldest_trans_id
= get_desc_trans_id(desc
);
2426 oldest_start
= d_bh
->b_blocknr
;
2427 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2428 "journal-1180: Resetting "
2429 "oldest_start to offset %lu, trans_id %lu",
2431 SB_ONDISK_JOURNAL_1st_BLOCK
2432 (sb
), oldest_trans_id
);
2434 if (newest_mount_id
< get_desc_mount_id(desc
)) {
2435 newest_mount_id
= get_desc_mount_id(desc
);
2436 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2437 "journal-1299: Setting "
2438 "newest_mount_id to %d",
2439 get_desc_mount_id(desc
));
2441 cur_dblock
+= get_desc_trans_len(desc
) + 2;
2449 cur_dblock
= oldest_start
;
2450 if (oldest_trans_id
) {
2451 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2452 "journal-1206: Starting replay "
2453 "from offset %llu, trans_id %lu",
2454 cur_dblock
- SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2459 while (continue_replay
&& oldest_trans_id
> 0) {
2461 journal_read_transaction(sb
, cur_dblock
, oldest_start
,
2462 oldest_trans_id
, newest_mount_id
);
2465 } else if (ret
!= 0) {
2469 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) + journal
->j_start
;
2471 if (cur_dblock
== oldest_start
)
2475 if (oldest_trans_id
== 0) {
2476 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
2477 "journal-1225: No valid " "transactions found");
2479 /* j_start does not get set correctly if we don't replay any transactions.
2480 ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2481 ** copy the trans_id from the header
2483 if (valid_journal_header
&& replay_count
== 0) {
2484 journal
->j_start
= le32_to_cpu(jh
->j_first_unflushed_offset
);
2485 journal
->j_trans_id
=
2486 le32_to_cpu(jh
->j_last_flush_trans_id
) + 1;
2487 /* check for trans_id overflow */
2488 if (journal
->j_trans_id
== 0)
2489 journal
->j_trans_id
= 10;
2490 journal
->j_last_flush_trans_id
=
2491 le32_to_cpu(jh
->j_last_flush_trans_id
);
2492 journal
->j_mount_id
= le32_to_cpu(jh
->j_mount_id
) + 1;
2494 journal
->j_mount_id
= newest_mount_id
+ 1;
2496 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
, "journal-1299: Setting "
2497 "newest_mount_id to %lu", journal
->j_mount_id
);
2498 journal
->j_first_unflushed_offset
= journal
->j_start
;
2499 if (replay_count
> 0) {
2501 "replayed %d transactions in %lu seconds\n",
2502 replay_count
, get_seconds() - start
);
2504 if (!bdev_read_only(sb
->s_bdev
) &&
2505 _update_journal_header_block(sb
, journal
->j_start
,
2506 journal
->j_last_flush_trans_id
)) {
2507 /* replay failed, caller must call free_journal_ram and abort
2515 static struct reiserfs_journal_list
*alloc_journal_list(struct super_block
*s
)
2517 struct reiserfs_journal_list
*jl
;
2518 jl
= kzalloc(sizeof(struct reiserfs_journal_list
),
2519 GFP_NOFS
| __GFP_NOFAIL
);
2520 INIT_LIST_HEAD(&jl
->j_list
);
2521 INIT_LIST_HEAD(&jl
->j_working_list
);
2522 INIT_LIST_HEAD(&jl
->j_tail_bh_list
);
2523 INIT_LIST_HEAD(&jl
->j_bh_list
);
2524 mutex_init(&jl
->j_commit_mutex
);
2525 SB_JOURNAL(s
)->j_num_lists
++;
2526 get_journal_list(jl
);
2530 static void journal_list_init(struct super_block
*sb
)
2532 SB_JOURNAL(sb
)->j_current_jl
= alloc_journal_list(sb
);
2535 static int release_journal_dev(struct super_block
*super
,
2536 struct reiserfs_journal
*journal
)
2542 if (journal
->j_dev_bd
!= NULL
) {
2543 result
= blkdev_put(journal
->j_dev_bd
, journal
->j_dev_mode
);
2544 journal
->j_dev_bd
= NULL
;
2548 reiserfs_warning(super
, "sh-457",
2549 "Cannot release journal device: %i", result
);
2554 static int journal_init_dev(struct super_block
*super
,
2555 struct reiserfs_journal
*journal
,
2556 const char *jdev_name
)
2560 fmode_t blkdev_mode
= FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
;
2561 char b
[BDEVNAME_SIZE
];
2565 journal
->j_dev_bd
= NULL
;
2566 jdev
= SB_ONDISK_JOURNAL_DEVICE(super
) ?
2567 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super
)) : super
->s_dev
;
2569 if (bdev_read_only(super
->s_bdev
))
2570 blkdev_mode
= FMODE_READ
;
2572 /* there is no "jdev" option and journal is on separate device */
2573 if ((!jdev_name
|| !jdev_name
[0])) {
2574 if (jdev
== super
->s_dev
)
2575 blkdev_mode
&= ~FMODE_EXCL
;
2576 journal
->j_dev_bd
= blkdev_get_by_dev(jdev
, blkdev_mode
,
2578 journal
->j_dev_mode
= blkdev_mode
;
2579 if (IS_ERR(journal
->j_dev_bd
)) {
2580 result
= PTR_ERR(journal
->j_dev_bd
);
2581 journal
->j_dev_bd
= NULL
;
2582 reiserfs_warning(super
, "sh-458",
2583 "cannot init journal device '%s': %i",
2584 __bdevname(jdev
, b
), result
);
2586 } else if (jdev
!= super
->s_dev
)
2587 set_blocksize(journal
->j_dev_bd
, super
->s_blocksize
);
2592 journal
->j_dev_mode
= blkdev_mode
;
2593 journal
->j_dev_bd
= blkdev_get_by_path(jdev_name
, blkdev_mode
, journal
);
2594 if (IS_ERR(journal
->j_dev_bd
)) {
2595 result
= PTR_ERR(journal
->j_dev_bd
);
2596 journal
->j_dev_bd
= NULL
;
2597 reiserfs_warning(super
,
2598 "journal_init_dev: Cannot open '%s': %i",
2603 set_blocksize(journal
->j_dev_bd
, super
->s_blocksize
);
2604 reiserfs_info(super
,
2605 "journal_init_dev: journal device: %s\n",
2606 bdevname(journal
->j_dev_bd
, b
));
2611 * When creating/tuning a file system user can assign some
2612 * journal params within boundaries which depend on the ratio
2613 * blocksize/standard_blocksize.
2615 * For blocks >= standard_blocksize transaction size should
2616 * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2617 * then JOURNAL_TRANS_MAX_DEFAULT.
2619 * For blocks < standard_blocksize these boundaries should be
2620 * decreased proportionally.
2622 #define REISERFS_STANDARD_BLKSIZE (4096)
2624 static int check_advise_trans_params(struct super_block
*sb
,
2625 struct reiserfs_journal
*journal
)
2627 if (journal
->j_trans_max
) {
2628 /* Non-default journal params.
2629 Do sanity check for them. */
2631 if (sb
->s_blocksize
< REISERFS_STANDARD_BLKSIZE
)
2632 ratio
= REISERFS_STANDARD_BLKSIZE
/ sb
->s_blocksize
;
2634 if (journal
->j_trans_max
> JOURNAL_TRANS_MAX_DEFAULT
/ ratio
||
2635 journal
->j_trans_max
< JOURNAL_TRANS_MIN_DEFAULT
/ ratio
||
2636 SB_ONDISK_JOURNAL_SIZE(sb
) / journal
->j_trans_max
<
2637 JOURNAL_MIN_RATIO
) {
2638 reiserfs_warning(sb
, "sh-462",
2639 "bad transaction max size (%u). "
2640 "FSCK?", journal
->j_trans_max
);
2643 if (journal
->j_max_batch
!= (journal
->j_trans_max
) *
2644 JOURNAL_MAX_BATCH_DEFAULT
/JOURNAL_TRANS_MAX_DEFAULT
) {
2645 reiserfs_warning(sb
, "sh-463",
2646 "bad transaction max batch (%u). "
2647 "FSCK?", journal
->j_max_batch
);
2651 /* Default journal params.
2652 The file system was created by old version
2653 of mkreiserfs, so some fields contain zeros,
2654 and we need to advise proper values for them */
2655 if (sb
->s_blocksize
!= REISERFS_STANDARD_BLKSIZE
) {
2656 reiserfs_warning(sb
, "sh-464", "bad blocksize (%u)",
2660 journal
->j_trans_max
= JOURNAL_TRANS_MAX_DEFAULT
;
2661 journal
->j_max_batch
= JOURNAL_MAX_BATCH_DEFAULT
;
2662 journal
->j_max_commit_age
= JOURNAL_MAX_COMMIT_AGE
;
2668 ** must be called once on fs mount. calls journal_read for you
2670 int journal_init(struct super_block
*sb
, const char *j_dev_name
,
2671 int old_format
, unsigned int commit_max_age
)
2673 int num_cnodes
= SB_ONDISK_JOURNAL_SIZE(sb
) * 2;
2674 struct buffer_head
*bhjh
;
2675 struct reiserfs_super_block
*rs
;
2676 struct reiserfs_journal_header
*jh
;
2677 struct reiserfs_journal
*journal
;
2678 struct reiserfs_journal_list
*jl
;
2679 char b
[BDEVNAME_SIZE
];
2682 journal
= SB_JOURNAL(sb
) = vzalloc(sizeof(struct reiserfs_journal
));
2684 reiserfs_warning(sb
, "journal-1256",
2685 "unable to get memory for journal structure");
2688 INIT_LIST_HEAD(&journal
->j_bitmap_nodes
);
2689 INIT_LIST_HEAD(&journal
->j_prealloc_list
);
2690 INIT_LIST_HEAD(&journal
->j_working_list
);
2691 INIT_LIST_HEAD(&journal
->j_journal_list
);
2692 journal
->j_persistent_trans
= 0;
2693 if (reiserfs_allocate_list_bitmaps(sb
, journal
->j_list_bitmap
,
2694 reiserfs_bmap_count(sb
)))
2695 goto free_and_return
;
2697 allocate_bitmap_nodes(sb
);
2699 /* reserved for journal area support */
2700 SB_JOURNAL_1st_RESERVED_BLOCK(sb
) = (old_format
?
2701 REISERFS_OLD_DISK_OFFSET_IN_BYTES
2703 reiserfs_bmap_count(sb
) +
2705 REISERFS_DISK_OFFSET_IN_BYTES
/
2706 sb
->s_blocksize
+ 2);
2708 /* Sanity check to see is the standard journal fitting within first bitmap
2709 (actual for small blocksizes) */
2710 if (!SB_ONDISK_JOURNAL_DEVICE(sb
) &&
2711 (SB_JOURNAL_1st_RESERVED_BLOCK(sb
) +
2712 SB_ONDISK_JOURNAL_SIZE(sb
) > sb
->s_blocksize
* 8)) {
2713 reiserfs_warning(sb
, "journal-1393",
2714 "journal does not fit for area addressed "
2715 "by first of bitmap blocks. It starts at "
2716 "%u and its size is %u. Block size %ld",
2717 SB_JOURNAL_1st_RESERVED_BLOCK(sb
),
2718 SB_ONDISK_JOURNAL_SIZE(sb
),
2720 goto free_and_return
;
2723 if (journal_init_dev(sb
, journal
, j_dev_name
) != 0) {
2724 reiserfs_warning(sb
, "sh-462",
2725 "unable to initialize jornal device");
2726 goto free_and_return
;
2729 rs
= SB_DISK_SUPER_BLOCK(sb
);
2731 /* read journal header */
2732 bhjh
= journal_bread(sb
,
2733 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
2734 SB_ONDISK_JOURNAL_SIZE(sb
));
2736 reiserfs_warning(sb
, "sh-459",
2737 "unable to read journal header");
2738 goto free_and_return
;
2740 jh
= (struct reiserfs_journal_header
*)(bhjh
->b_data
);
2742 /* make sure that journal matches to the super block */
2743 if (is_reiserfs_jr(rs
)
2744 && (le32_to_cpu(jh
->jh_journal
.jp_journal_magic
) !=
2745 sb_jp_journal_magic(rs
))) {
2746 reiserfs_warning(sb
, "sh-460",
2747 "journal header magic %x (device %s) does "
2748 "not match to magic found in super block %x",
2749 jh
->jh_journal
.jp_journal_magic
,
2750 bdevname(journal
->j_dev_bd
, b
),
2751 sb_jp_journal_magic(rs
));
2753 goto free_and_return
;
2756 journal
->j_trans_max
= le32_to_cpu(jh
->jh_journal
.jp_journal_trans_max
);
2757 journal
->j_max_batch
= le32_to_cpu(jh
->jh_journal
.jp_journal_max_batch
);
2758 journal
->j_max_commit_age
=
2759 le32_to_cpu(jh
->jh_journal
.jp_journal_max_commit_age
);
2760 journal
->j_max_trans_age
= JOURNAL_MAX_TRANS_AGE
;
2762 if (check_advise_trans_params(sb
, journal
) != 0)
2763 goto free_and_return
;
2764 journal
->j_default_max_commit_age
= journal
->j_max_commit_age
;
2766 if (commit_max_age
!= 0) {
2767 journal
->j_max_commit_age
= commit_max_age
;
2768 journal
->j_max_trans_age
= commit_max_age
;
2771 reiserfs_info(sb
, "journal params: device %s, size %u, "
2772 "journal first block %u, max trans len %u, max batch %u, "
2773 "max commit age %u, max trans age %u\n",
2774 bdevname(journal
->j_dev_bd
, b
),
2775 SB_ONDISK_JOURNAL_SIZE(sb
),
2776 SB_ONDISK_JOURNAL_1st_BLOCK(sb
),
2777 journal
->j_trans_max
,
2778 journal
->j_max_batch
,
2779 journal
->j_max_commit_age
, journal
->j_max_trans_age
);
2783 journal
->j_list_bitmap_index
= 0;
2784 journal_list_init(sb
);
2786 memset(journal
->j_list_hash_table
, 0,
2787 JOURNAL_HASH_SIZE
* sizeof(struct reiserfs_journal_cnode
*));
2789 INIT_LIST_HEAD(&journal
->j_dirty_buffers
);
2790 spin_lock_init(&journal
->j_dirty_buffers_lock
);
2792 journal
->j_start
= 0;
2794 journal
->j_len_alloc
= 0;
2795 atomic_set(&(journal
->j_wcount
), 0);
2796 atomic_set(&(journal
->j_async_throttle
), 0);
2797 journal
->j_bcount
= 0;
2798 journal
->j_trans_start_time
= 0;
2799 journal
->j_last
= NULL
;
2800 journal
->j_first
= NULL
;
2801 init_waitqueue_head(&(journal
->j_join_wait
));
2802 mutex_init(&journal
->j_mutex
);
2803 mutex_init(&journal
->j_flush_mutex
);
2805 journal
->j_trans_id
= 10;
2806 journal
->j_mount_id
= 10;
2807 journal
->j_state
= 0;
2808 atomic_set(&(journal
->j_jlock
), 0);
2809 journal
->j_cnode_free_list
= allocate_cnodes(num_cnodes
);
2810 journal
->j_cnode_free_orig
= journal
->j_cnode_free_list
;
2811 journal
->j_cnode_free
= journal
->j_cnode_free_list
? num_cnodes
: 0;
2812 journal
->j_cnode_used
= 0;
2813 journal
->j_must_wait
= 0;
2815 if (journal
->j_cnode_free
== 0) {
2816 reiserfs_warning(sb
, "journal-2004", "Journal cnode memory "
2817 "allocation failed (%ld bytes). Journal is "
2818 "too large for available memory. Usually "
2819 "this is due to a journal that is too large.",
2820 sizeof (struct reiserfs_journal_cnode
) * num_cnodes
);
2821 goto free_and_return
;
2824 init_journal_hash(sb
);
2825 jl
= journal
->j_current_jl
;
2828 * get_list_bitmap() may call flush_commit_list() which
2829 * requires the lock. Calling flush_commit_list() shouldn't happen
2830 * this early but I like to be paranoid.
2832 reiserfs_write_lock(sb
);
2833 jl
->j_list_bitmap
= get_list_bitmap(sb
, jl
);
2834 reiserfs_write_unlock(sb
);
2835 if (!jl
->j_list_bitmap
) {
2836 reiserfs_warning(sb
, "journal-2005",
2837 "get_list_bitmap failed for journal list 0");
2838 goto free_and_return
;
2842 * Journal_read needs to be inspected in order to push down
2843 * the lock further inside (or even remove it).
2845 reiserfs_write_lock(sb
);
2846 ret
= journal_read(sb
);
2847 reiserfs_write_unlock(sb
);
2849 reiserfs_warning(sb
, "reiserfs-2006",
2850 "Replay Failure, unable to mount");
2851 goto free_and_return
;
2854 reiserfs_mounted_fs_count
++;
2855 if (reiserfs_mounted_fs_count
<= 1)
2856 commit_wq
= alloc_workqueue("reiserfs", WQ_MEM_RECLAIM
, 0);
2858 INIT_DELAYED_WORK(&journal
->j_work
, flush_async_commits
);
2859 journal
->j_work_sb
= sb
;
2862 free_journal_ram(sb
);
2867 ** test for a polite end of the current transaction. Used by file_write, and should
2868 ** be used by delete to make sure they don't write more than can fit inside a single
2871 int journal_transaction_should_end(struct reiserfs_transaction_handle
*th
,
2874 struct reiserfs_journal
*journal
= SB_JOURNAL(th
->t_super
);
2875 time_t now
= get_seconds();
2876 /* cannot restart while nested */
2877 BUG_ON(!th
->t_trans_id
);
2878 if (th
->t_refcount
> 1)
2880 if (journal
->j_must_wait
> 0 ||
2881 (journal
->j_len_alloc
+ new_alloc
) >= journal
->j_max_batch
||
2882 atomic_read(&(journal
->j_jlock
)) ||
2883 (now
- journal
->j_trans_start_time
) > journal
->j_max_trans_age
||
2884 journal
->j_cnode_free
< (journal
->j_trans_max
* 3)) {
2888 journal
->j_len_alloc
+= new_alloc
;
2889 th
->t_blocks_allocated
+= new_alloc
;
2893 /* this must be called inside a transaction
2895 void reiserfs_block_writes(struct reiserfs_transaction_handle
*th
)
2897 struct reiserfs_journal
*journal
= SB_JOURNAL(th
->t_super
);
2898 BUG_ON(!th
->t_trans_id
);
2899 journal
->j_must_wait
= 1;
2900 set_bit(J_WRITERS_BLOCKED
, &journal
->j_state
);
2904 /* this must be called without a transaction started
2906 void reiserfs_allow_writes(struct super_block
*s
)
2908 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2909 clear_bit(J_WRITERS_BLOCKED
, &journal
->j_state
);
2910 wake_up(&journal
->j_join_wait
);
2913 /* this must be called without a transaction started
2915 void reiserfs_wait_on_write_block(struct super_block
*s
)
2917 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2918 wait_event(journal
->j_join_wait
,
2919 !test_bit(J_WRITERS_BLOCKED
, &journal
->j_state
));
2922 static void queue_log_writer(struct super_block
*s
)
2925 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2926 set_bit(J_WRITERS_QUEUED
, &journal
->j_state
);
2929 * we don't want to use wait_event here because
2930 * we only want to wait once.
2932 init_waitqueue_entry(&wait
, current
);
2933 add_wait_queue(&journal
->j_join_wait
, &wait
);
2934 set_current_state(TASK_UNINTERRUPTIBLE
);
2935 if (test_bit(J_WRITERS_QUEUED
, &journal
->j_state
)) {
2936 reiserfs_write_unlock(s
);
2938 reiserfs_write_lock(s
);
2940 __set_current_state(TASK_RUNNING
);
2941 remove_wait_queue(&journal
->j_join_wait
, &wait
);
2944 static void wake_queued_writers(struct super_block
*s
)
2946 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
2947 if (test_and_clear_bit(J_WRITERS_QUEUED
, &journal
->j_state
))
2948 wake_up(&journal
->j_join_wait
);
2951 static void let_transaction_grow(struct super_block
*sb
, unsigned int trans_id
)
2953 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
2954 unsigned long bcount
= journal
->j_bcount
;
2956 reiserfs_write_unlock(sb
);
2957 schedule_timeout_uninterruptible(1);
2958 reiserfs_write_lock(sb
);
2959 journal
->j_current_jl
->j_state
|= LIST_COMMIT_PENDING
;
2960 while ((atomic_read(&journal
->j_wcount
) > 0 ||
2961 atomic_read(&journal
->j_jlock
)) &&
2962 journal
->j_trans_id
== trans_id
) {
2963 queue_log_writer(sb
);
2965 if (journal
->j_trans_id
!= trans_id
)
2967 if (bcount
== journal
->j_bcount
)
2969 bcount
= journal
->j_bcount
;
2973 /* join == true if you must join an existing transaction.
2974 ** join == false if you can deal with waiting for others to finish
2976 ** this will block until the transaction is joinable. send the number of blocks you
2977 ** expect to use in nblocks.
2979 static int do_journal_begin_r(struct reiserfs_transaction_handle
*th
,
2980 struct super_block
*sb
, unsigned long nblocks
,
2983 time_t now
= get_seconds();
2984 unsigned int old_trans_id
;
2985 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
2986 struct reiserfs_transaction_handle myth
;
2987 int sched_count
= 0;
2990 reiserfs_check_lock_depth(sb
, "journal_begin");
2991 BUG_ON(nblocks
> journal
->j_trans_max
);
2993 PROC_INFO_INC(sb
, journal
.journal_being
);
2994 /* set here for journal_join */
3000 if (join
!= JBEGIN_ABORT
&& reiserfs_is_journal_aborted(journal
)) {
3002 retval
= journal
->j_errno
;
3005 journal
->j_bcount
++;
3007 if (test_bit(J_WRITERS_BLOCKED
, &journal
->j_state
)) {
3009 reiserfs_write_unlock(sb
);
3010 reiserfs_wait_on_write_block(sb
);
3011 reiserfs_write_lock(sb
);
3012 PROC_INFO_INC(sb
, journal
.journal_relock_writers
);
3015 now
= get_seconds();
3017 /* if there is no room in the journal OR
3018 ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
3019 ** we don't sleep if there aren't other writers
3022 if ((!join
&& journal
->j_must_wait
> 0) ||
3024 && (journal
->j_len_alloc
+ nblocks
+ 2) >= journal
->j_max_batch
)
3025 || (!join
&& atomic_read(&journal
->j_wcount
) > 0
3026 && journal
->j_trans_start_time
> 0
3027 && (now
- journal
->j_trans_start_time
) >
3028 journal
->j_max_trans_age
) || (!join
3029 && atomic_read(&journal
->j_jlock
))
3030 || (!join
&& journal
->j_cnode_free
< (journal
->j_trans_max
* 3))) {
3032 old_trans_id
= journal
->j_trans_id
;
3033 unlock_journal(sb
); /* allow others to finish this transaction */
3035 if (!join
&& (journal
->j_len_alloc
+ nblocks
+ 2) >=
3036 journal
->j_max_batch
&&
3037 ((journal
->j_len
+ nblocks
+ 2) * 100) <
3038 (journal
->j_len_alloc
* 75)) {
3039 if (atomic_read(&journal
->j_wcount
) > 10) {
3041 queue_log_writer(sb
);
3045 /* don't mess with joining the transaction if all we have to do is
3046 * wait for someone else to do a commit
3048 if (atomic_read(&journal
->j_jlock
)) {
3049 while (journal
->j_trans_id
== old_trans_id
&&
3050 atomic_read(&journal
->j_jlock
)) {
3051 queue_log_writer(sb
);
3055 retval
= journal_join(&myth
, sb
, 1);
3059 /* someone might have ended the transaction while we joined */
3060 if (old_trans_id
!= journal
->j_trans_id
) {
3061 retval
= do_journal_end(&myth
, sb
, 1, 0);
3063 retval
= do_journal_end(&myth
, sb
, 1, COMMIT_NOW
);
3069 PROC_INFO_INC(sb
, journal
.journal_relock_wcount
);
3072 /* we are the first writer, set trans_id */
3073 if (journal
->j_trans_start_time
== 0) {
3074 journal
->j_trans_start_time
= get_seconds();
3076 atomic_inc(&(journal
->j_wcount
));
3077 journal
->j_len_alloc
+= nblocks
;
3078 th
->t_blocks_logged
= 0;
3079 th
->t_blocks_allocated
= nblocks
;
3080 th
->t_trans_id
= journal
->j_trans_id
;
3082 INIT_LIST_HEAD(&th
->t_list
);
3086 memset(th
, 0, sizeof(*th
));
3087 /* Re-set th->t_super, so we can properly keep track of how many
3088 * persistent transactions there are. We need to do this so if this
3089 * call is part of a failed restart_transaction, we can free it later */
3094 struct reiserfs_transaction_handle
*reiserfs_persistent_transaction(struct
3100 struct reiserfs_transaction_handle
*th
;
3102 /* if we're nesting into an existing transaction. It will be
3103 ** persistent on its own
3105 if (reiserfs_transaction_running(s
)) {
3106 th
= current
->journal_info
;
3108 BUG_ON(th
->t_refcount
< 2);
3112 th
= kmalloc(sizeof(struct reiserfs_transaction_handle
), GFP_NOFS
);
3115 ret
= journal_begin(th
, s
, nblocks
);
3121 SB_JOURNAL(s
)->j_persistent_trans
++;
3125 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle
*th
)
3127 struct super_block
*s
= th
->t_super
;
3130 ret
= journal_end(th
, th
->t_super
, th
->t_blocks_allocated
);
3133 if (th
->t_refcount
== 0) {
3134 SB_JOURNAL(s
)->j_persistent_trans
--;
3140 static int journal_join(struct reiserfs_transaction_handle
*th
,
3141 struct super_block
*sb
, unsigned long nblocks
)
3143 struct reiserfs_transaction_handle
*cur_th
= current
->journal_info
;
3145 /* this keeps do_journal_end from NULLing out the current->journal_info
3148 th
->t_handle_save
= cur_th
;
3149 BUG_ON(cur_th
&& cur_th
->t_refcount
> 1);
3150 return do_journal_begin_r(th
, sb
, nblocks
, JBEGIN_JOIN
);
3153 int journal_join_abort(struct reiserfs_transaction_handle
*th
,
3154 struct super_block
*sb
, unsigned long nblocks
)
3156 struct reiserfs_transaction_handle
*cur_th
= current
->journal_info
;
3158 /* this keeps do_journal_end from NULLing out the current->journal_info
3161 th
->t_handle_save
= cur_th
;
3162 BUG_ON(cur_th
&& cur_th
->t_refcount
> 1);
3163 return do_journal_begin_r(th
, sb
, nblocks
, JBEGIN_ABORT
);
3166 int journal_begin(struct reiserfs_transaction_handle
*th
,
3167 struct super_block
*sb
, unsigned long nblocks
)
3169 struct reiserfs_transaction_handle
*cur_th
= current
->journal_info
;
3172 th
->t_handle_save
= NULL
;
3174 /* we are nesting into the current transaction */
3175 if (cur_th
->t_super
== sb
) {
3176 BUG_ON(!cur_th
->t_refcount
);
3177 cur_th
->t_refcount
++;
3178 memcpy(th
, cur_th
, sizeof(*th
));
3179 if (th
->t_refcount
<= 1)
3180 reiserfs_warning(sb
, "reiserfs-2005",
3181 "BAD: refcount <= 1, but "
3182 "journal_info != 0");
3185 /* we've ended up with a handle from a different filesystem.
3186 ** save it and restore on journal_end. This should never
3189 reiserfs_warning(sb
, "clm-2100",
3190 "nesting info a different FS");
3191 th
->t_handle_save
= current
->journal_info
;
3192 current
->journal_info
= th
;
3195 current
->journal_info
= th
;
3197 ret
= do_journal_begin_r(th
, sb
, nblocks
, JBEGIN_REG
);
3198 BUG_ON(current
->journal_info
!= th
);
3200 /* I guess this boils down to being the reciprocal of clm-2100 above.
3201 * If do_journal_begin_r fails, we need to put it back, since journal_end
3202 * won't be called to do it. */
3204 current
->journal_info
= th
->t_handle_save
;
3206 BUG_ON(!th
->t_refcount
);
3212 ** puts bh into the current transaction. If it was already there, reorders removes the
3213 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3215 ** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the
3216 ** transaction is committed.
3218 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3220 int journal_mark_dirty(struct reiserfs_transaction_handle
*th
,
3221 struct super_block
*sb
, struct buffer_head
*bh
)
3223 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3224 struct reiserfs_journal_cnode
*cn
= NULL
;
3225 int count_already_incd
= 0;
3227 BUG_ON(!th
->t_trans_id
);
3229 PROC_INFO_INC(sb
, journal
.mark_dirty
);
3230 if (th
->t_trans_id
!= journal
->j_trans_id
) {
3231 reiserfs_panic(th
->t_super
, "journal-1577",
3232 "handle trans id %ld != current trans id %ld",
3233 th
->t_trans_id
, journal
->j_trans_id
);
3236 prepared
= test_clear_buffer_journal_prepared(bh
);
3237 clear_buffer_journal_restore_dirty(bh
);
3238 /* already in this transaction, we are done */
3239 if (buffer_journaled(bh
)) {
3240 PROC_INFO_INC(sb
, journal
.mark_dirty_already
);
3244 /* this must be turned into a panic instead of a warning. We can't allow
3245 ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3246 ** could get to disk too early. NOT GOOD.
3248 if (!prepared
|| buffer_dirty(bh
)) {
3249 reiserfs_warning(sb
, "journal-1777",
3250 "buffer %llu bad state "
3251 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3252 (unsigned long long)bh
->b_blocknr
,
3253 prepared
? ' ' : '!',
3254 buffer_locked(bh
) ? ' ' : '!',
3255 buffer_dirty(bh
) ? ' ' : '!',
3256 buffer_journal_dirty(bh
) ? ' ' : '!');
3259 if (atomic_read(&(journal
->j_wcount
)) <= 0) {
3260 reiserfs_warning(sb
, "journal-1409",
3261 "returning because j_wcount was %d",
3262 atomic_read(&(journal
->j_wcount
)));
3265 /* this error means I've screwed up, and we've overflowed the transaction.
3266 ** Nothing can be done here, except make the FS readonly or panic.
3268 if (journal
->j_len
>= journal
->j_trans_max
) {
3269 reiserfs_panic(th
->t_super
, "journal-1413",
3270 "j_len (%lu) is too big",
3274 if (buffer_journal_dirty(bh
)) {
3275 count_already_incd
= 1;
3276 PROC_INFO_INC(sb
, journal
.mark_dirty_notjournal
);
3277 clear_buffer_journal_dirty(bh
);
3280 if (journal
->j_len
> journal
->j_len_alloc
) {
3281 journal
->j_len_alloc
= journal
->j_len
+ JOURNAL_PER_BALANCE_CNT
;
3284 set_buffer_journaled(bh
);
3286 /* now put this guy on the end */
3290 reiserfs_panic(sb
, "journal-4", "get_cnode failed!");
3293 if (th
->t_blocks_logged
== th
->t_blocks_allocated
) {
3294 th
->t_blocks_allocated
+= JOURNAL_PER_BALANCE_CNT
;
3295 journal
->j_len_alloc
+= JOURNAL_PER_BALANCE_CNT
;
3297 th
->t_blocks_logged
++;
3301 cn
->blocknr
= bh
->b_blocknr
;
3304 insert_journal_hash(journal
->j_hash_table
, cn
);
3305 if (!count_already_incd
) {
3310 cn
->prev
= journal
->j_last
;
3312 if (journal
->j_last
) {
3313 journal
->j_last
->next
= cn
;
3314 journal
->j_last
= cn
;
3316 journal
->j_first
= cn
;
3317 journal
->j_last
= cn
;
3319 reiserfs_schedule_old_flush(sb
);
3323 int journal_end(struct reiserfs_transaction_handle
*th
,
3324 struct super_block
*sb
, unsigned long nblocks
)
3326 if (!current
->journal_info
&& th
->t_refcount
> 1)
3327 reiserfs_warning(sb
, "REISER-NESTING",
3328 "th NULL, refcount %d", th
->t_refcount
);
3330 if (!th
->t_trans_id
) {
3336 if (th
->t_refcount
> 0) {
3337 struct reiserfs_transaction_handle
*cur_th
=
3338 current
->journal_info
;
3340 /* we aren't allowed to close a nested transaction on a different
3341 ** filesystem from the one in the task struct
3343 BUG_ON(cur_th
->t_super
!= th
->t_super
);
3346 memcpy(current
->journal_info
, th
, sizeof(*th
));
3351 return do_journal_end(th
, sb
, nblocks
, 0);
3355 /* removes from the current transaction, relsing and descrementing any counters.
3356 ** also files the removed buffer directly onto the clean list
3358 ** called by journal_mark_freed when a block has been deleted
3360 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3362 static int remove_from_transaction(struct super_block
*sb
,
3363 b_blocknr_t blocknr
, int already_cleaned
)
3365 struct buffer_head
*bh
;
3366 struct reiserfs_journal_cnode
*cn
;
3367 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3370 cn
= get_journal_hash_dev(sb
, journal
->j_hash_table
, blocknr
);
3371 if (!cn
|| !cn
->bh
) {
3376 cn
->prev
->next
= cn
->next
;
3379 cn
->next
->prev
= cn
->prev
;
3381 if (cn
== journal
->j_first
) {
3382 journal
->j_first
= cn
->next
;
3384 if (cn
== journal
->j_last
) {
3385 journal
->j_last
= cn
->prev
;
3388 remove_journal_hash(sb
, journal
->j_hash_table
, NULL
,
3390 clear_buffer_journaled(bh
); /* don't log this one */
3392 if (!already_cleaned
) {
3393 clear_buffer_journal_dirty(bh
);
3394 clear_buffer_dirty(bh
);
3395 clear_buffer_journal_test(bh
);
3397 if (atomic_read(&(bh
->b_count
)) < 0) {
3398 reiserfs_warning(sb
, "journal-1752",
3404 journal
->j_len_alloc
--;
3410 ** for any cnode in a journal list, it can only be dirtied of all the
3411 ** transactions that include it are committed to disk.
3412 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3413 ** and 0 if you aren't
3415 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3416 ** blocks for a given transaction on disk
3419 static int can_dirty(struct reiserfs_journal_cnode
*cn
)
3421 struct super_block
*sb
= cn
->sb
;
3422 b_blocknr_t blocknr
= cn
->blocknr
;
3423 struct reiserfs_journal_cnode
*cur
= cn
->hprev
;
3426 /* first test hprev. These are all newer than cn, so any node here
3427 ** with the same block number and dev means this node can't be sent
3428 ** to disk right now.
3430 while (cur
&& can_dirty
) {
3431 if (cur
->jlist
&& cur
->bh
&& cur
->blocknr
&& cur
->sb
== sb
&&
3432 cur
->blocknr
== blocknr
) {
3437 /* then test hnext. These are all older than cn. As long as they
3438 ** are committed to the log, it is safe to write cn to disk
3441 while (cur
&& can_dirty
) {
3442 if (cur
->jlist
&& cur
->jlist
->j_len
> 0 &&
3443 atomic_read(&(cur
->jlist
->j_commit_left
)) > 0 && cur
->bh
&&
3444 cur
->blocknr
&& cur
->sb
== sb
&& cur
->blocknr
== blocknr
) {
3452 /* syncs the commit blocks, but does not force the real buffers to disk
3453 ** will wait until the current transaction is done/committed before returning
3455 int journal_end_sync(struct reiserfs_transaction_handle
*th
,
3456 struct super_block
*sb
, unsigned long nblocks
)
3458 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3460 BUG_ON(!th
->t_trans_id
);
3461 /* you can sync while nested, very, very bad */
3462 BUG_ON(th
->t_refcount
> 1);
3463 if (journal
->j_len
== 0) {
3464 reiserfs_prepare_for_journal(sb
, SB_BUFFER_WITH_SB(sb
),
3466 journal_mark_dirty(th
, sb
, SB_BUFFER_WITH_SB(sb
));
3468 return do_journal_end(th
, sb
, nblocks
, COMMIT_NOW
| WAIT
);
3472 ** writeback the pending async commits to disk
3474 static void flush_async_commits(struct work_struct
*work
)
3476 struct reiserfs_journal
*journal
=
3477 container_of(work
, struct reiserfs_journal
, j_work
.work
);
3478 struct super_block
*sb
= journal
->j_work_sb
;
3479 struct reiserfs_journal_list
*jl
;
3480 struct list_head
*entry
;
3482 reiserfs_write_lock(sb
);
3483 if (!list_empty(&journal
->j_journal_list
)) {
3484 /* last entry is the youngest, commit it and you get everything */
3485 entry
= journal
->j_journal_list
.prev
;
3486 jl
= JOURNAL_LIST_ENTRY(entry
);
3487 flush_commit_list(sb
, jl
, 1);
3489 reiserfs_write_unlock(sb
);
3493 ** flushes any old transactions to disk
3494 ** ends the current transaction if it is too old
3496 void reiserfs_flush_old_commits(struct super_block
*sb
)
3499 struct reiserfs_transaction_handle th
;
3500 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3502 now
= get_seconds();
3503 /* safety check so we don't flush while we are replaying the log during
3506 if (list_empty(&journal
->j_journal_list
))
3509 /* check the current transaction. If there are no writers, and it is
3510 * too old, finish it, and force the commit blocks to disk
3512 if (atomic_read(&journal
->j_wcount
) <= 0 &&
3513 journal
->j_trans_start_time
> 0 &&
3514 journal
->j_len
> 0 &&
3515 (now
- journal
->j_trans_start_time
) > journal
->j_max_trans_age
) {
3516 if (!journal_join(&th
, sb
, 1)) {
3517 reiserfs_prepare_for_journal(sb
,
3518 SB_BUFFER_WITH_SB(sb
),
3520 journal_mark_dirty(&th
, sb
,
3521 SB_BUFFER_WITH_SB(sb
));
3523 /* we're only being called from kreiserfsd, it makes no sense to do
3524 ** an async commit so that kreiserfsd can do it later
3526 do_journal_end(&th
, sb
, 1, COMMIT_NOW
| WAIT
);
3532 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3534 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3535 ** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just
3536 ** flushes the commit list and returns 0.
3538 ** Won't batch when flush or commit_now is set. Also won't batch when others are waiting on j_join_wait.
3540 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3542 static int check_journal_end(struct reiserfs_transaction_handle
*th
,
3543 struct super_block
*sb
, unsigned long nblocks
,
3548 int flush
= flags
& FLUSH_ALL
;
3549 int commit_now
= flags
& COMMIT_NOW
;
3550 int wait_on_commit
= flags
& WAIT
;
3551 struct reiserfs_journal_list
*jl
;
3552 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3554 BUG_ON(!th
->t_trans_id
);
3556 if (th
->t_trans_id
!= journal
->j_trans_id
) {
3557 reiserfs_panic(th
->t_super
, "journal-1577",
3558 "handle trans id %ld != current trans id %ld",
3559 th
->t_trans_id
, journal
->j_trans_id
);
3562 journal
->j_len_alloc
-= (th
->t_blocks_allocated
- th
->t_blocks_logged
);
3563 if (atomic_read(&(journal
->j_wcount
)) > 0) { /* <= 0 is allowed. unmounting might not call begin */
3564 atomic_dec(&(journal
->j_wcount
));
3567 /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3568 ** will be dealt with by next transaction that actually writes something, but should be taken
3569 ** care of in this trans
3571 BUG_ON(journal
->j_len
== 0);
3573 /* if wcount > 0, and we are called to with flush or commit_now,
3574 ** we wait on j_join_wait. We will wake up when the last writer has
3575 ** finished the transaction, and started it on its way to the disk.
3576 ** Then, we flush the commit or journal list, and just return 0
3577 ** because the rest of journal end was already done for this transaction.
3579 if (atomic_read(&(journal
->j_wcount
)) > 0) {
3580 if (flush
|| commit_now
) {
3583 jl
= journal
->j_current_jl
;
3584 trans_id
= jl
->j_trans_id
;
3586 jl
->j_state
|= LIST_COMMIT_PENDING
;
3587 atomic_set(&(journal
->j_jlock
), 1);
3589 journal
->j_next_full_flush
= 1;
3593 /* sleep while the current transaction is still j_jlocked */
3594 while (journal
->j_trans_id
== trans_id
) {
3595 if (atomic_read(&journal
->j_jlock
)) {
3596 queue_log_writer(sb
);
3599 if (journal
->j_trans_id
== trans_id
) {
3600 atomic_set(&(journal
->j_jlock
),
3606 BUG_ON(journal
->j_trans_id
== trans_id
);
3609 && journal_list_still_alive(sb
, trans_id
)
3610 && wait_on_commit
) {
3611 flush_commit_list(sb
, jl
, 1);
3619 /* deal with old transactions where we are the last writers */
3620 now
= get_seconds();
3621 if ((now
- journal
->j_trans_start_time
) > journal
->j_max_trans_age
) {
3623 journal
->j_next_async_flush
= 1;
3625 /* don't batch when someone is waiting on j_join_wait */
3626 /* don't batch when syncing the commit or flushing the whole trans */
3627 if (!(journal
->j_must_wait
> 0) && !(atomic_read(&(journal
->j_jlock
)))
3628 && !flush
&& !commit_now
&& (journal
->j_len
< journal
->j_max_batch
)
3629 && journal
->j_len_alloc
< journal
->j_max_batch
3630 && journal
->j_cnode_free
> (journal
->j_trans_max
* 3)) {
3631 journal
->j_bcount
++;
3636 if (journal
->j_start
> SB_ONDISK_JOURNAL_SIZE(sb
)) {
3637 reiserfs_panic(sb
, "journal-003",
3638 "j_start (%ld) is too high",
3645 ** Does all the work that makes deleting blocks safe.
3646 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3649 ** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes
3650 ** before this transaction has finished.
3652 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with
3653 ** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash,
3654 ** the block can't be reallocated yet.
3656 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3658 int journal_mark_freed(struct reiserfs_transaction_handle
*th
,
3659 struct super_block
*sb
, b_blocknr_t blocknr
)
3661 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3662 struct reiserfs_journal_cnode
*cn
= NULL
;
3663 struct buffer_head
*bh
= NULL
;
3664 struct reiserfs_list_bitmap
*jb
= NULL
;
3666 BUG_ON(!th
->t_trans_id
);
3668 cn
= get_journal_hash_dev(sb
, journal
->j_hash_table
, blocknr
);
3673 /* if it is journal new, we just remove it from this transaction */
3674 if (bh
&& buffer_journal_new(bh
)) {
3675 clear_buffer_journal_new(bh
);
3676 clear_prepared_bits(bh
);
3677 reiserfs_clean_and_file_buffer(bh
);
3678 cleaned
= remove_from_transaction(sb
, blocknr
, cleaned
);
3680 /* set the bit for this block in the journal bitmap for this transaction */
3681 jb
= journal
->j_current_jl
->j_list_bitmap
;
3683 reiserfs_panic(sb
, "journal-1702",
3684 "journal_list_bitmap is NULL");
3686 set_bit_in_list_bitmap(sb
, blocknr
, jb
);
3688 /* Note, the entire while loop is not allowed to schedule. */
3691 clear_prepared_bits(bh
);
3692 reiserfs_clean_and_file_buffer(bh
);
3694 cleaned
= remove_from_transaction(sb
, blocknr
, cleaned
);
3696 /* find all older transactions with this block, make sure they don't try to write it out */
3697 cn
= get_journal_hash_dev(sb
, journal
->j_list_hash_table
,
3700 if (sb
== cn
->sb
&& blocknr
== cn
->blocknr
) {
3701 set_bit(BLOCK_FREED
, &cn
->state
);
3704 /* remove_from_transaction will brelse the buffer if it was
3705 ** in the current trans
3707 clear_buffer_journal_dirty(cn
->
3709 clear_buffer_dirty(cn
->bh
);
3710 clear_buffer_journal_test(cn
->
3715 (&(cn
->bh
->b_count
)) < 0) {
3716 reiserfs_warning(sb
,
3718 "cn->bh->b_count < 0");
3721 if (cn
->jlist
) { /* since we are clearing the bh, we MUST dec nonzerolen */
3734 release_buffer_page(bh
); /* get_hash grabs the buffer */
3738 void reiserfs_update_inode_transaction(struct inode
*inode
)
3740 struct reiserfs_journal
*journal
= SB_JOURNAL(inode
->i_sb
);
3741 REISERFS_I(inode
)->i_jl
= journal
->j_current_jl
;
3742 REISERFS_I(inode
)->i_trans_id
= journal
->j_trans_id
;
3746 * returns -1 on error, 0 if no commits/barriers were done and 1
3747 * if a transaction was actually committed and the barrier was done
3749 static int __commit_trans_jl(struct inode
*inode
, unsigned long id
,
3750 struct reiserfs_journal_list
*jl
)
3752 struct reiserfs_transaction_handle th
;
3753 struct super_block
*sb
= inode
->i_sb
;
3754 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3757 /* is it from the current transaction, or from an unknown transaction? */
3758 if (id
== journal
->j_trans_id
) {
3759 jl
= journal
->j_current_jl
;
3760 /* try to let other writers come in and grow this transaction */
3761 let_transaction_grow(sb
, id
);
3762 if (journal
->j_trans_id
!= id
) {
3763 goto flush_commit_only
;
3766 ret
= journal_begin(&th
, sb
, 1);
3770 /* someone might have ended this transaction while we joined */
3771 if (journal
->j_trans_id
!= id
) {
3772 reiserfs_prepare_for_journal(sb
, SB_BUFFER_WITH_SB(sb
),
3774 journal_mark_dirty(&th
, sb
, SB_BUFFER_WITH_SB(sb
));
3775 ret
= journal_end(&th
, sb
, 1);
3776 goto flush_commit_only
;
3779 ret
= journal_end_sync(&th
, sb
, 1);
3784 /* this gets tricky, we have to make sure the journal list in
3785 * the inode still exists. We know the list is still around
3786 * if we've got a larger transaction id than the oldest list
3789 if (journal_list_still_alive(inode
->i_sb
, id
)) {
3791 * we only set ret to 1 when we know for sure
3792 * the barrier hasn't been started yet on the commit
3795 if (atomic_read(&jl
->j_commit_left
) > 1)
3797 flush_commit_list(sb
, jl
, 1);
3798 if (journal
->j_errno
)
3799 ret
= journal
->j_errno
;
3802 /* otherwise the list is gone, and long since committed */
3806 int reiserfs_commit_for_inode(struct inode
*inode
)
3808 unsigned int id
= REISERFS_I(inode
)->i_trans_id
;
3809 struct reiserfs_journal_list
*jl
= REISERFS_I(inode
)->i_jl
;
3811 /* for the whole inode, assume unset id means it was
3812 * changed in the current transaction. More conservative
3815 reiserfs_update_inode_transaction(inode
);
3816 id
= REISERFS_I(inode
)->i_trans_id
;
3817 /* jl will be updated in __commit_trans_jl */
3820 return __commit_trans_jl(inode
, id
, jl
);
3823 void reiserfs_restore_prepared_buffer(struct super_block
*sb
,
3824 struct buffer_head
*bh
)
3826 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3827 PROC_INFO_INC(sb
, journal
.restore_prepared
);
3831 if (test_clear_buffer_journal_restore_dirty(bh
) &&
3832 buffer_journal_dirty(bh
)) {
3833 struct reiserfs_journal_cnode
*cn
;
3834 cn
= get_journal_hash_dev(sb
,
3835 journal
->j_list_hash_table
,
3837 if (cn
&& can_dirty(cn
)) {
3838 set_buffer_journal_test(bh
);
3839 mark_buffer_dirty(bh
);
3842 clear_buffer_journal_prepared(bh
);
3845 extern struct tree_balance
*cur_tb
;
3847 ** before we can change a metadata block, we have to make sure it won't
3848 ** be written to disk while we are altering it. So, we must:
3853 int reiserfs_prepare_for_journal(struct super_block
*sb
,
3854 struct buffer_head
*bh
, int wait
)
3856 PROC_INFO_INC(sb
, journal
.prepare
);
3858 if (!trylock_buffer(bh
)) {
3863 set_buffer_journal_prepared(bh
);
3864 if (test_clear_buffer_dirty(bh
) && buffer_journal_dirty(bh
)) {
3865 clear_buffer_journal_test(bh
);
3866 set_buffer_journal_restore_dirty(bh
);
3872 static void flush_old_journal_lists(struct super_block
*s
)
3874 struct reiserfs_journal
*journal
= SB_JOURNAL(s
);
3875 struct reiserfs_journal_list
*jl
;
3876 struct list_head
*entry
;
3877 time_t now
= get_seconds();
3879 while (!list_empty(&journal
->j_journal_list
)) {
3880 entry
= journal
->j_journal_list
.next
;
3881 jl
= JOURNAL_LIST_ENTRY(entry
);
3882 /* this check should always be run, to send old lists to disk */
3883 if (jl
->j_timestamp
< (now
- (JOURNAL_MAX_TRANS_AGE
* 4)) &&
3884 atomic_read(&jl
->j_commit_left
) == 0 &&
3885 test_transaction(s
, jl
)) {
3886 flush_used_journal_lists(s
, jl
);
3894 ** long and ugly. If flush, will not return until all commit
3895 ** blocks and all real buffers in the trans are on disk.
3896 ** If no_async, won't return until all commit blocks are on disk.
3898 ** keep reading, there are comments as you go along
3900 ** If the journal is aborted, we just clean up. Things like flushing
3901 ** journal lists, etc just won't happen.
3903 static int do_journal_end(struct reiserfs_transaction_handle
*th
,
3904 struct super_block
*sb
, unsigned long nblocks
,
3907 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
3908 struct reiserfs_journal_cnode
*cn
, *next
, *jl_cn
;
3909 struct reiserfs_journal_cnode
*last_cn
= NULL
;
3910 struct reiserfs_journal_desc
*desc
;
3911 struct reiserfs_journal_commit
*commit
;
3912 struct buffer_head
*c_bh
; /* commit bh */
3913 struct buffer_head
*d_bh
; /* desc bh */
3914 int cur_write_start
= 0; /* start index of current log write */
3919 struct reiserfs_journal_list
*jl
, *temp_jl
;
3920 struct list_head
*entry
, *safe
;
3921 unsigned long jindex
;
3922 unsigned int commit_trans_id
;
3925 BUG_ON(th
->t_refcount
> 1);
3926 BUG_ON(!th
->t_trans_id
);
3928 /* protect flush_older_commits from doing mistakes if the
3929 transaction ID counter gets overflowed. */
3930 if (th
->t_trans_id
== ~0U)
3931 flags
|= FLUSH_ALL
| COMMIT_NOW
| WAIT
;
3932 flush
= flags
& FLUSH_ALL
;
3933 wait_on_commit
= flags
& WAIT
;
3935 current
->journal_info
= th
->t_handle_save
;
3936 reiserfs_check_lock_depth(sb
, "journal end");
3937 if (journal
->j_len
== 0) {
3938 reiserfs_prepare_for_journal(sb
, SB_BUFFER_WITH_SB(sb
),
3940 journal_mark_dirty(th
, sb
, SB_BUFFER_WITH_SB(sb
));
3944 if (journal
->j_next_full_flush
) {
3948 if (journal
->j_next_async_flush
) {
3949 flags
|= COMMIT_NOW
| WAIT
;
3953 /* check_journal_end locks the journal, and unlocks if it does not return 1
3954 ** it tells us if we should continue with the journal_end, or just return
3956 if (!check_journal_end(th
, sb
, nblocks
, flags
)) {
3957 reiserfs_schedule_old_flush(sb
);
3958 wake_queued_writers(sb
);
3959 reiserfs_async_progress_wait(sb
);
3963 /* check_journal_end might set these, check again */
3964 if (journal
->j_next_full_flush
) {
3969 ** j must wait means we have to flush the log blocks, and the real blocks for
3972 if (journal
->j_must_wait
> 0) {
3975 #ifdef REISERFS_PREALLOCATE
3976 /* quota ops might need to nest, setup the journal_info pointer for them
3977 * and raise the refcount so that it is > 0. */
3978 current
->journal_info
= th
;
3980 reiserfs_discard_all_prealloc(th
); /* it should not involve new blocks into
3981 * the transaction */
3983 current
->journal_info
= th
->t_handle_save
;
3986 /* setup description block */
3989 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
3991 set_buffer_uptodate(d_bh
);
3992 desc
= (struct reiserfs_journal_desc
*)(d_bh
)->b_data
;
3993 memset(d_bh
->b_data
, 0, d_bh
->b_size
);
3994 memcpy(get_journal_desc_magic(d_bh
), JOURNAL_DESC_MAGIC
, 8);
3995 set_desc_trans_id(desc
, journal
->j_trans_id
);
3997 /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */
3998 c_bh
= journal_getblk(sb
, SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
3999 ((journal
->j_start
+ journal
->j_len
+
4000 1) % SB_ONDISK_JOURNAL_SIZE(sb
)));
4001 commit
= (struct reiserfs_journal_commit
*)c_bh
->b_data
;
4002 memset(c_bh
->b_data
, 0, c_bh
->b_size
);
4003 set_commit_trans_id(commit
, journal
->j_trans_id
);
4004 set_buffer_uptodate(c_bh
);
4006 /* init this journal list */
4007 jl
= journal
->j_current_jl
;
4009 /* we lock the commit before doing anything because
4010 * we want to make sure nobody tries to run flush_commit_list until
4011 * the new transaction is fully setup, and we've already flushed the
4014 reiserfs_mutex_lock_safe(&jl
->j_commit_mutex
, sb
);
4016 /* save the transaction id in case we need to commit it later */
4017 commit_trans_id
= jl
->j_trans_id
;
4019 atomic_set(&jl
->j_older_commits_done
, 0);
4020 jl
->j_trans_id
= journal
->j_trans_id
;
4021 jl
->j_timestamp
= journal
->j_trans_start_time
;
4022 jl
->j_commit_bh
= c_bh
;
4023 jl
->j_start
= journal
->j_start
;
4024 jl
->j_len
= journal
->j_len
;
4025 atomic_set(&jl
->j_nonzerolen
, journal
->j_len
);
4026 atomic_set(&jl
->j_commit_left
, journal
->j_len
+ 2);
4027 jl
->j_realblock
= NULL
;
4029 /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4030 ** for each real block, add it to the journal list hash,
4031 ** copy into real block index array in the commit or desc block
4033 trans_half
= journal_trans_half(sb
->s_blocksize
);
4034 for (i
= 0, cn
= journal
->j_first
; cn
; cn
= cn
->next
, i
++) {
4035 if (buffer_journaled(cn
->bh
)) {
4036 jl_cn
= get_cnode(sb
);
4038 reiserfs_panic(sb
, "journal-1676",
4039 "get_cnode returned NULL");
4042 jl
->j_realblock
= jl_cn
;
4044 jl_cn
->prev
= last_cn
;
4047 last_cn
->next
= jl_cn
;
4050 /* make sure the block we are trying to log is not a block
4051 of journal or reserved area */
4053 if (is_block_in_log_or_reserved_area
4054 (sb
, cn
->bh
->b_blocknr
)) {
4055 reiserfs_panic(sb
, "journal-2332",
4056 "Trying to log block %lu, "
4057 "which is a log block",
4060 jl_cn
->blocknr
= cn
->bh
->b_blocknr
;
4065 insert_journal_hash(journal
->j_list_hash_table
, jl_cn
);
4066 if (i
< trans_half
) {
4067 desc
->j_realblock
[i
] =
4068 cpu_to_le32(cn
->bh
->b_blocknr
);
4070 commit
->j_realblock
[i
- trans_half
] =
4071 cpu_to_le32(cn
->bh
->b_blocknr
);
4077 set_desc_trans_len(desc
, journal
->j_len
);
4078 set_desc_mount_id(desc
, journal
->j_mount_id
);
4079 set_desc_trans_id(desc
, journal
->j_trans_id
);
4080 set_commit_trans_len(commit
, journal
->j_len
);
4082 /* special check in case all buffers in the journal were marked for not logging */
4083 BUG_ON(journal
->j_len
== 0);
4085 /* we're about to dirty all the log blocks, mark the description block
4086 * dirty now too. Don't mark the commit block dirty until all the
4087 * others are on disk
4089 mark_buffer_dirty(d_bh
);
4091 /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4092 cur_write_start
= journal
->j_start
;
4093 cn
= journal
->j_first
;
4094 jindex
= 1; /* start at one so we don't get the desc again */
4096 clear_buffer_journal_new(cn
->bh
);
4097 /* copy all the real blocks into log area. dirty log blocks */
4098 if (buffer_journaled(cn
->bh
)) {
4099 struct buffer_head
*tmp_bh
;
4104 SB_ONDISK_JOURNAL_1st_BLOCK(sb
) +
4107 SB_ONDISK_JOURNAL_SIZE(sb
)));
4108 set_buffer_uptodate(tmp_bh
);
4109 page
= cn
->bh
->b_page
;
4111 memcpy(tmp_bh
->b_data
,
4112 addr
+ offset_in_page(cn
->bh
->b_data
),
4115 mark_buffer_dirty(tmp_bh
);
4117 set_buffer_journal_dirty(cn
->bh
);
4118 clear_buffer_journaled(cn
->bh
);
4120 /* JDirty cleared sometime during transaction. don't log this one */
4121 reiserfs_warning(sb
, "journal-2048",
4122 "BAD, buffer in journal hash, "
4129 reiserfs_write_unlock(sb
);
4131 reiserfs_write_lock(sb
);
4134 /* we are done with both the c_bh and d_bh, but
4135 ** c_bh must be written after all other commit blocks,
4136 ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4139 journal
->j_current_jl
= alloc_journal_list(sb
);
4141 /* now it is safe to insert this transaction on the main list */
4142 list_add_tail(&jl
->j_list
, &journal
->j_journal_list
);
4143 list_add_tail(&jl
->j_working_list
, &journal
->j_working_list
);
4144 journal
->j_num_work_lists
++;
4146 /* reset journal values for the next transaction */
4147 old_start
= journal
->j_start
;
4149 (journal
->j_start
+ journal
->j_len
+
4150 2) % SB_ONDISK_JOURNAL_SIZE(sb
);
4151 atomic_set(&(journal
->j_wcount
), 0);
4152 journal
->j_bcount
= 0;
4153 journal
->j_last
= NULL
;
4154 journal
->j_first
= NULL
;
4156 journal
->j_trans_start_time
= 0;
4157 /* check for trans_id overflow */
4158 if (++journal
->j_trans_id
== 0)
4159 journal
->j_trans_id
= 10;
4160 journal
->j_current_jl
->j_trans_id
= journal
->j_trans_id
;
4161 journal
->j_must_wait
= 0;
4162 journal
->j_len_alloc
= 0;
4163 journal
->j_next_full_flush
= 0;
4164 journal
->j_next_async_flush
= 0;
4165 init_journal_hash(sb
);
4167 // make sure reiserfs_add_jh sees the new current_jl before we
4168 // write out the tails
4171 /* tail conversion targets have to hit the disk before we end the
4172 * transaction. Otherwise a later transaction might repack the tail
4173 * before this transaction commits, leaving the data block unflushed and
4174 * clean, if we crash before the later transaction commits, the data block
4177 if (!list_empty(&jl
->j_tail_bh_list
)) {
4178 reiserfs_write_unlock(sb
);
4179 write_ordered_buffers(&journal
->j_dirty_buffers_lock
,
4180 journal
, jl
, &jl
->j_tail_bh_list
);
4181 reiserfs_write_lock(sb
);
4183 BUG_ON(!list_empty(&jl
->j_tail_bh_list
));
4184 mutex_unlock(&jl
->j_commit_mutex
);
4186 /* honor the flush wishes from the caller, simple commits can
4187 ** be done outside the journal lock, they are done below
4189 ** if we don't flush the commit list right now, we put it into
4190 ** the work queue so the people waiting on the async progress work
4191 ** queue don't wait for this proc to flush journal lists and such.
4194 flush_commit_list(sb
, jl
, 1);
4195 flush_journal_list(sb
, jl
, 1);
4196 } else if (!(jl
->j_state
& LIST_COMMIT_PENDING
))
4197 queue_delayed_work(commit_wq
, &journal
->j_work
, HZ
/ 10);
4199 /* if the next transaction has any chance of wrapping, flush
4200 ** transactions that might get overwritten. If any journal lists are very
4201 ** old flush them as well.
4204 list_for_each_safe(entry
, safe
, &journal
->j_journal_list
) {
4205 temp_jl
= JOURNAL_LIST_ENTRY(entry
);
4206 if (journal
->j_start
<= temp_jl
->j_start
) {
4207 if ((journal
->j_start
+ journal
->j_trans_max
+ 1) >=
4209 flush_used_journal_lists(sb
, temp_jl
);
4211 } else if ((journal
->j_start
+
4212 journal
->j_trans_max
+ 1) <
4213 SB_ONDISK_JOURNAL_SIZE(sb
)) {
4214 /* if we don't cross into the next transaction and we don't
4215 * wrap, there is no way we can overlap any later transactions
4220 } else if ((journal
->j_start
+
4221 journal
->j_trans_max
+ 1) >
4222 SB_ONDISK_JOURNAL_SIZE(sb
)) {
4223 if (((journal
->j_start
+ journal
->j_trans_max
+ 1) %
4224 SB_ONDISK_JOURNAL_SIZE(sb
)) >=
4226 flush_used_journal_lists(sb
, temp_jl
);
4229 /* we don't overlap anything from out start to the end of the
4230 * log, and our wrapped portion doesn't overlap anything at
4231 * the start of the log. We can break
4237 flush_old_journal_lists(sb
);
4239 journal
->j_current_jl
->j_list_bitmap
=
4240 get_list_bitmap(sb
, journal
->j_current_jl
);
4242 if (!(journal
->j_current_jl
->j_list_bitmap
)) {
4243 reiserfs_panic(sb
, "journal-1996",
4244 "could not get a list bitmap");
4247 atomic_set(&(journal
->j_jlock
), 0);
4249 /* wake up any body waiting to join. */
4250 clear_bit(J_WRITERS_QUEUED
, &journal
->j_state
);
4251 wake_up(&(journal
->j_join_wait
));
4253 if (!flush
&& wait_on_commit
&&
4254 journal_list_still_alive(sb
, commit_trans_id
)) {
4255 flush_commit_list(sb
, jl
, 1);
4258 reiserfs_check_lock_depth(sb
, "journal end2");
4260 memset(th
, 0, sizeof(*th
));
4261 /* Re-set th->t_super, so we can properly keep track of how many
4262 * persistent transactions there are. We need to do this so if this
4263 * call is part of a failed restart_transaction, we can free it later */
4266 return journal
->j_errno
;
4269 /* Send the file system read only and refuse new transactions */
4270 void reiserfs_abort_journal(struct super_block
*sb
, int errno
)
4272 struct reiserfs_journal
*journal
= SB_JOURNAL(sb
);
4273 if (test_bit(J_ABORTED
, &journal
->j_state
))
4276 if (!journal
->j_errno
)
4277 journal
->j_errno
= errno
;
4279 sb
->s_flags
|= MS_RDONLY
;
4280 set_bit(J_ABORTED
, &journal
->j_state
);
4282 #ifdef CONFIG_REISERFS_CHECK