2 ** Write ahead logging implementation copyright Chris Mason 2000
4 ** The background commits make this code very interelated, and
5 ** overly complex. I need to rethink things a bit....The major players:
7 ** journal_begin -- call with the number of blocks you expect to log.
8 ** If the current transaction is too
9 ** old, it will block until the current transaction is
10 ** finished, and then start a new one.
11 ** Usually, your transaction will get joined in with
12 ** previous ones for speed.
14 ** journal_join -- same as journal_begin, but won't block on the current
15 ** transaction regardless of age. Don't ever call
16 ** this. Ever. There are only two places it should be
17 ** called from, and they are both inside this file.
19 ** journal_mark_dirty -- adds blocks into this transaction. clears any flags
20 ** that might make them get sent to disk
21 ** and then marks them BH_JDirty. Puts the buffer head
22 ** into the current transaction hash.
24 ** journal_end -- if the current transaction is batchable, it does nothing
25 ** otherwise, it could do an async/synchronous commit, or
26 ** a full flush of all log and real blocks in the
29 ** flush_old_commits -- if the current transaction is too old, it is ended and
30 ** commit blocks are sent to disk. Forces commit blocks
31 ** to disk for all backgrounded commits that have been
33 ** -- Note, if you call this as an immediate flush from
34 ** from within kupdate, it will ignore the immediate flag
37 #include <linux/config.h>
38 #include <asm/uaccess.h>
39 #include <asm/system.h>
41 #include <linux/time.h>
42 #include <asm/semaphore.h>
44 #include <linux/vmalloc.h>
45 #include <linux/reiserfs_fs.h>
47 #include <linux/kernel.h>
48 #include <linux/errno.h>
49 #include <linux/fcntl.h>
50 #include <linux/stat.h>
51 #include <linux/string.h>
52 #include <linux/smp_lock.h>
53 #include <linux/suspend.h>
54 #include <linux/buffer_head.h>
55 #include <linux/workqueue.h>
56 #include <linux/writeback.h>
57 #include <linux/blkdev.h>
60 /* gets a struct reiserfs_journal_list * from a list head */
61 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
63 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
66 /* the number of mounted filesystems. This is used to decide when to
67 ** start and kill the commit workqueue
69 static int reiserfs_mounted_fs_count;
71 static struct workqueue_struct *commit_wq;
73 #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
75 #define BUFNR 64 /*read ahead */
77 /* cnode stat bits. Move these into reiserfs_fs.h */
79 #define BLOCK_FREED 2 /* this block was freed, and can't be written. */
80 #define BLOCK_FREED_HOLDER 3 /* this block was freed during this transaction, and can't be written */
82 #define BLOCK_NEEDS_FLUSH 4 /* used in flush_journal_list */
83 #define BLOCK_DIRTIED 5
86 /* journal list state bits */
87 #define LIST_TOUCHED 1
89 #define LIST_COMMIT_PENDING 4 /* someone will commit this list */
91 /* flags for do_journal_end */
92 #define FLUSH_ALL 1 /* flush commit and real blocks */
93 #define COMMIT_NOW 2 /* end and commit this transaction */
94 #define WAIT 4 /* wait for the log blocks to hit the disk*/
96 /* state bits for the journal */
97 #define WRITERS_BLOCKED 1 /* set when new writers not allowed */
98 #define WRITERS_QUEUED 2 /* set when log is full due to too many
102 static int do_journal_end(struct reiserfs_transaction_handle *,struct super_block *,unsigned long nblocks,int flags) ;
103 static int flush_journal_list(struct super_block *s, struct reiserfs_journal_list *jl, int flushall) ;
104 static int flush_commit_list(struct super_block *s, struct reiserfs_journal_list *jl, int flushall) ;
105 static int can_dirty(struct reiserfs_journal_cnode *cn) ;
106 static int journal_join(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks);
107 static int release_journal_dev( struct super_block *super,
108 struct reiserfs_journal *journal );
109 static int dirty_one_transaction(struct super_block *s,
110 struct reiserfs_journal_list *jl);
111 static void flush_async_commits(void *p);
113 static void init_journal_hash(struct super_block *p_s_sb) {
114 memset(SB_JOURNAL(p_s_sb)->j_hash_table, 0, JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)) ;
118 ** clears BH_Dirty and sticks the buffer on the clean list. Called because I can't allow refile_buffer to
119 ** make schedule happen after I've freed a block. Look at remove_from_transaction and journal_mark_freed for
122 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh) {
124 clear_buffer_dirty(bh);
125 clear_bit(BH_JTest, &bh->b_state);
130 static struct reiserfs_bitmap_node *
131 allocate_bitmap_node(struct super_block *p_s_sb) {
132 struct reiserfs_bitmap_node *bn ;
135 bn = reiserfs_kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS, p_s_sb) ;
139 bn->data = reiserfs_kmalloc(p_s_sb->s_blocksize, GFP_NOFS, p_s_sb) ;
141 reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb) ;
145 memset(bn->data, 0, p_s_sb->s_blocksize) ;
146 INIT_LIST_HEAD(&bn->list) ;
150 static struct reiserfs_bitmap_node *
151 get_bitmap_node(struct super_block *p_s_sb) {
152 struct reiserfs_bitmap_node *bn = NULL;
153 struct list_head *entry = SB_JOURNAL(p_s_sb)->j_bitmap_nodes.next ;
155 SB_JOURNAL(p_s_sb)->j_used_bitmap_nodes++ ;
158 if(entry != &SB_JOURNAL(p_s_sb)->j_bitmap_nodes) {
159 bn = list_entry(entry, struct reiserfs_bitmap_node, list) ;
161 memset(bn->data, 0, p_s_sb->s_blocksize) ;
162 SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes-- ;
165 bn = allocate_bitmap_node(p_s_sb) ;
172 static inline void free_bitmap_node(struct super_block *p_s_sb,
173 struct reiserfs_bitmap_node *bn) {
174 SB_JOURNAL(p_s_sb)->j_used_bitmap_nodes-- ;
175 if (SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
176 reiserfs_kfree(bn->data, p_s_sb->s_blocksize, p_s_sb) ;
177 reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb) ;
179 list_add(&bn->list, &SB_JOURNAL(p_s_sb)->j_bitmap_nodes) ;
180 SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes++ ;
184 static void allocate_bitmap_nodes(struct super_block *p_s_sb) {
186 struct reiserfs_bitmap_node *bn = NULL ;
187 for (i = 0 ; i < REISERFS_MIN_BITMAP_NODES ; i++) {
188 bn = allocate_bitmap_node(p_s_sb) ;
190 list_add(&bn->list, &SB_JOURNAL(p_s_sb)->j_bitmap_nodes) ;
191 SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes++ ;
193 break ; // this is ok, we'll try again when more are needed
198 static int set_bit_in_list_bitmap(struct super_block *p_s_sb, int block,
199 struct reiserfs_list_bitmap *jb) {
200 int bmap_nr = block / (p_s_sb->s_blocksize << 3) ;
201 int bit_nr = block % (p_s_sb->s_blocksize << 3) ;
203 if (!jb->bitmaps[bmap_nr]) {
204 jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb) ;
206 set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data) ;
210 static void cleanup_bitmap_list(struct super_block *p_s_sb,
211 struct reiserfs_list_bitmap *jb) {
213 if (jb->bitmaps == NULL)
216 for (i = 0 ; i < SB_BMAP_NR(p_s_sb) ; i++) {
217 if (jb->bitmaps[i]) {
218 free_bitmap_node(p_s_sb, jb->bitmaps[i]) ;
219 jb->bitmaps[i] = NULL ;
225 ** only call this on FS unmount.
227 static int free_list_bitmaps(struct super_block *p_s_sb,
228 struct reiserfs_list_bitmap *jb_array) {
230 struct reiserfs_list_bitmap *jb ;
231 for (i = 0 ; i < JOURNAL_NUM_BITMAPS ; i++) {
233 jb->journal_list = NULL ;
234 cleanup_bitmap_list(p_s_sb, jb) ;
241 static int free_bitmap_nodes(struct super_block *p_s_sb) {
242 struct list_head *next = SB_JOURNAL(p_s_sb)->j_bitmap_nodes.next ;
243 struct reiserfs_bitmap_node *bn ;
245 while(next != &SB_JOURNAL(p_s_sb)->j_bitmap_nodes) {
246 bn = list_entry(next, struct reiserfs_bitmap_node, list) ;
248 reiserfs_kfree(bn->data, p_s_sb->s_blocksize, p_s_sb) ;
249 reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb) ;
250 next = SB_JOURNAL(p_s_sb)->j_bitmap_nodes.next ;
251 SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes-- ;
258 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
259 ** jb_array is the array to be filled in.
261 int reiserfs_allocate_list_bitmaps(struct super_block *p_s_sb,
262 struct reiserfs_list_bitmap *jb_array,
266 struct reiserfs_list_bitmap *jb ;
267 int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *) ;
269 for (i = 0 ; i < JOURNAL_NUM_BITMAPS ; i++) {
271 jb->journal_list = NULL ;
272 jb->bitmaps = vmalloc( mem ) ;
274 reiserfs_warning(p_s_sb, "clm-2000, unable to allocate bitmaps for journal lists") ;
278 memset(jb->bitmaps, 0, mem) ;
281 free_list_bitmaps(p_s_sb, jb_array) ;
288 ** find an available list bitmap. If you can't find one, flush a commit list
291 static struct reiserfs_list_bitmap *
292 get_list_bitmap(struct super_block *p_s_sb, struct reiserfs_journal_list *jl) {
294 struct reiserfs_list_bitmap *jb = NULL ;
296 for (j = 0 ; j < (JOURNAL_NUM_BITMAPS * 3) ; j++) {
297 i = SB_JOURNAL(p_s_sb)->j_list_bitmap_index ;
298 SB_JOURNAL(p_s_sb)->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS ;
299 jb = SB_JOURNAL(p_s_sb)->j_list_bitmap + i ;
300 if (SB_JOURNAL(p_s_sb)->j_list_bitmap[i].journal_list) {
301 flush_commit_list(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_bitmap[i].journal_list, 1) ;
302 if (!SB_JOURNAL(p_s_sb)->j_list_bitmap[i].journal_list) {
309 if (jb->journal_list) { /* double check to make sure if flushed correctly */
312 jb->journal_list = jl ;
317 ** allocates a new chunk of X nodes, and links them all together as a list.
318 ** Uses the cnode->next and cnode->prev pointers
319 ** returns NULL on failure
321 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes) {
322 struct reiserfs_journal_cnode *head ;
324 if (num_cnodes <= 0) {
327 head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode)) ;
331 memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode)) ;
332 head[0].prev = NULL ;
333 head[0].next = head + 1 ;
334 for (i = 1 ; i < num_cnodes; i++) {
335 head[i].prev = head + (i - 1) ;
336 head[i].next = head + (i + 1) ; /* if last one, overwrite it after the if */
338 head[num_cnodes -1].next = NULL ;
343 ** pulls a cnode off the free list, or returns NULL on failure
345 static struct reiserfs_journal_cnode *get_cnode(struct super_block *p_s_sb) {
346 struct reiserfs_journal_cnode *cn ;
348 reiserfs_check_lock_depth(p_s_sb, "get_cnode") ;
350 if (SB_JOURNAL(p_s_sb)->j_cnode_free <= 0) {
353 SB_JOURNAL(p_s_sb)->j_cnode_used++ ;
354 SB_JOURNAL(p_s_sb)->j_cnode_free-- ;
355 cn = SB_JOURNAL(p_s_sb)->j_cnode_free_list ;
360 cn->next->prev = NULL ;
362 SB_JOURNAL(p_s_sb)->j_cnode_free_list = cn->next ;
363 memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ;
368 ** returns a cnode to the free list
370 static void free_cnode(struct super_block *p_s_sb, struct reiserfs_journal_cnode *cn) {
372 reiserfs_check_lock_depth(p_s_sb, "free_cnode") ;
374 SB_JOURNAL(p_s_sb)->j_cnode_used-- ;
375 SB_JOURNAL(p_s_sb)->j_cnode_free++ ;
376 /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
377 cn->next = SB_JOURNAL(p_s_sb)->j_cnode_free_list ;
378 if (SB_JOURNAL(p_s_sb)->j_cnode_free_list) {
379 SB_JOURNAL(p_s_sb)->j_cnode_free_list->prev = cn ;
381 cn->prev = NULL ; /* not needed with the memset, but I might kill the memset, and forget to do this */
382 SB_JOURNAL(p_s_sb)->j_cnode_free_list = cn ;
385 static int clear_prepared_bits(struct buffer_head *bh) {
386 clear_bit(BH_JPrepared, &bh->b_state) ;
387 clear_bit(BH_JRestore_dirty, &bh->b_state) ;
391 /* buffer is in current transaction */
392 inline int buffer_journaled(const struct buffer_head *bh) {
394 return test_bit(BH_JDirty, &bh->b_state) ;
399 /* disk block was taken off free list before being in a finished transation, or written to disk
400 ** journal_new blocks can be reused immediately, for any purpose
402 inline int buffer_journal_new(const struct buffer_head *bh) {
404 return test_bit(BH_JNew, &bh->b_state) ;
409 inline int mark_buffer_journal_new(struct buffer_head *bh) {
411 set_bit(BH_JNew, &bh->b_state) ;
416 inline int mark_buffer_not_journaled(struct buffer_head *bh) {
418 clear_bit(BH_JDirty, &bh->b_state) ;
422 /* utility function to force a BUG if it is called without the big
423 ** kernel lock held. caller is the string printed just before calling BUG()
425 void reiserfs_check_lock_depth(struct super_block *sb, char *caller) {
427 if (current->lock_depth < 0) {
428 reiserfs_panic (sb, "%s called without kernel lock held", caller) ;
435 /* return a cnode with same dev, block number and size in table, or null if not found */
436 static inline struct reiserfs_journal_cnode *
437 get_journal_hash_dev(struct super_block *sb,
438 struct reiserfs_journal_cnode **table,
441 struct reiserfs_journal_cnode *cn ;
442 cn = journal_hash(table, sb, bl) ;
444 if (cn->blocknr == bl && cn->sb == sb)
448 return (struct reiserfs_journal_cnode *)0 ;
451 /* returns a cnode with same size, block number and dev as bh in the current transaction hash. NULL if not found */
452 static inline struct reiserfs_journal_cnode *get_journal_hash(struct super_block *p_s_sb, struct buffer_head *bh) {
453 struct reiserfs_journal_cnode *cn ;
455 cn = get_journal_hash_dev(p_s_sb, SB_JOURNAL(p_s_sb)->j_hash_table, bh->b_blocknr);
458 return (struct reiserfs_journal_cnode *)0 ;
464 ** this actually means 'can this block be reallocated yet?'. If you set search_all, a block can only be allocated
465 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
466 ** being overwritten by a replay after crashing.
468 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction. Since deleting
469 ** a block removes it from the current transaction, this case should never happen. If you don't set search_all, make
470 ** sure you never write the block without logging it.
472 ** next_zero_bit is a suggestion about the next block to try for find_forward.
473 ** when bl is rejected because it is set in a journal list bitmap, we search
474 ** for the next zero bit in the bitmap that rejected bl. Then, we return that
475 ** through next_zero_bit for find_forward to try.
477 ** Just because we return something in next_zero_bit does not mean we won't
478 ** reject it on the next call to reiserfs_in_journal
481 int reiserfs_in_journal(struct super_block *p_s_sb,
482 int bmap_nr, int bit_nr, int search_all,
483 b_blocknr_t *next_zero_bit) {
484 struct reiserfs_journal_cnode *cn ;
485 struct reiserfs_list_bitmap *jb ;
489 *next_zero_bit = 0 ; /* always start this at zero. */
491 PROC_INFO_INC( p_s_sb, journal.in_journal );
492 /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
493 ** if we crash before the transaction that freed it commits, this transaction won't
494 ** have committed either, and the block will never be written
497 for (i = 0 ; i < JOURNAL_NUM_BITMAPS ; i++) {
498 PROC_INFO_INC( p_s_sb, journal.in_journal_bitmap );
499 jb = SB_JOURNAL(p_s_sb)->j_list_bitmap + i ;
500 if (jb->journal_list && jb->bitmaps[bmap_nr] &&
501 test_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data)) {
502 *next_zero_bit = find_next_zero_bit((unsigned long *)
503 (jb->bitmaps[bmap_nr]->data),
504 p_s_sb->s_blocksize << 3, bit_nr+1) ;
510 bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr;
511 /* is it in any old transactions? */
512 if (search_all && (cn = get_journal_hash_dev(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_hash_table, bl))) {
516 /* is it in the current transaction. This should never happen */
517 if ((cn = get_journal_hash_dev(p_s_sb, SB_JOURNAL(p_s_sb)->j_hash_table, bl))) {
522 PROC_INFO_INC( p_s_sb, journal.in_journal_reusable );
527 /* insert cn into table
529 inline void insert_journal_hash(struct reiserfs_journal_cnode **table, struct reiserfs_journal_cnode *cn) {
530 struct reiserfs_journal_cnode *cn_orig ;
532 cn_orig = journal_hash(table, cn->sb, cn->blocknr) ;
533 cn->hnext = cn_orig ;
536 cn_orig->hprev = cn ;
538 journal_hash(table, cn->sb, cn->blocknr) = cn ;
541 /* lock the current transaction */
542 inline static void lock_journal(struct super_block *p_s_sb) {
543 PROC_INFO_INC( p_s_sb, journal.lock_journal );
544 down(&SB_JOURNAL(p_s_sb)->j_lock);
547 /* unlock the current transaction */
548 inline static void unlock_journal(struct super_block *p_s_sb) {
549 up(&SB_JOURNAL(p_s_sb)->j_lock);
552 static inline void get_journal_list(struct reiserfs_journal_list *jl)
557 static inline void put_journal_list(struct super_block *s,
558 struct reiserfs_journal_list *jl)
560 if (jl->j_refcount < 1) {
561 reiserfs_panic (s, "trans id %lu, refcount at %d", jl->j_trans_id,
564 if (--jl->j_refcount == 0)
565 reiserfs_kfree(jl, sizeof(struct reiserfs_journal_list), s);
569 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
570 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
573 static void cleanup_freed_for_journal_list(struct super_block *p_s_sb, struct reiserfs_journal_list *jl) {
575 struct reiserfs_list_bitmap *jb = jl->j_list_bitmap ;
577 cleanup_bitmap_list(p_s_sb, jb) ;
579 jl->j_list_bitmap->journal_list = NULL ;
580 jl->j_list_bitmap = NULL ;
583 static int journal_list_still_alive(struct super_block *s,
584 unsigned long trans_id)
586 struct list_head *entry = &SB_JOURNAL(s)->j_journal_list;
587 struct reiserfs_journal_list *jl;
589 if (!list_empty(entry)) {
590 jl = JOURNAL_LIST_ENTRY(entry->next);
591 if (jl->j_trans_id <= trans_id) {
598 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate) {
599 char b[BDEVNAME_SIZE];
601 if (buffer_journaled(bh)) {
602 reiserfs_warning(NULL, "clm-2084: pinned buffer %lu:%s sent to disk",
603 bh->b_blocknr, bdevname(bh->b_bdev, b)) ;
606 set_buffer_uptodate(bh) ;
608 clear_buffer_uptodate(bh) ;
613 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate) {
615 set_buffer_uptodate(bh) ;
617 clear_buffer_uptodate(bh) ;
622 static void submit_logged_buffer(struct buffer_head *bh) {
624 bh->b_end_io = reiserfs_end_buffer_io_sync ;
625 mark_buffer_notjournal_new(bh) ;
626 clear_buffer_dirty(bh) ;
627 if (!test_and_clear_bit(BH_JTest, &bh->b_state))
629 if (!buffer_uptodate(bh))
631 submit_bh(WRITE, bh) ;
634 static void submit_ordered_buffer(struct buffer_head *bh) {
636 bh->b_end_io = reiserfs_end_ordered_io;
637 clear_buffer_dirty(bh) ;
638 if (!buffer_uptodate(bh))
640 submit_bh(WRITE, bh) ;
643 #define CHUNK_SIZE 32
644 struct buffer_chunk {
645 struct buffer_head *bh[CHUNK_SIZE];
649 static void write_chunk(struct buffer_chunk *chunk) {
651 for (i = 0; i < chunk->nr ; i++) {
652 submit_logged_buffer(chunk->bh[i]) ;
657 static void write_ordered_chunk(struct buffer_chunk *chunk) {
659 for (i = 0; i < chunk->nr ; i++) {
660 submit_ordered_buffer(chunk->bh[i]) ;
665 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
667 void (fn)(struct buffer_chunk *))
670 if (chunk->nr >= CHUNK_SIZE)
672 chunk->bh[chunk->nr++] = bh;
673 if (chunk->nr >= CHUNK_SIZE) {
685 atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
686 static struct reiserfs_jh *alloc_jh(void) {
687 struct reiserfs_jh *jh;
689 jh = kmalloc(sizeof(*jh), GFP_NOFS);
691 atomic_inc(&nr_reiserfs_jh);
699 * we want to free the jh when the buffer has been written
702 void reiserfs_free_jh(struct buffer_head *bh) {
703 struct reiserfs_jh *jh;
707 bh->b_private = NULL;
709 list_del_init(&jh->list);
711 if (atomic_read(&nr_reiserfs_jh) <= 0)
713 atomic_dec(&nr_reiserfs_jh);
718 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
721 struct reiserfs_jh *jh;
724 spin_lock(&j->j_dirty_buffers_lock);
725 if (!bh->b_private) {
726 spin_unlock(&j->j_dirty_buffers_lock);
730 list_del_init(&jh->list);
735 spin_lock(&j->j_dirty_buffers_lock);
736 /* buffer must be locked for __add_jh, should be able to have
737 * two adds at the same time
744 jh->jl = j->j_current_jl;
746 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
748 list_add_tail(&jh->list, &jh->jl->j_bh_list);
750 spin_unlock(&j->j_dirty_buffers_lock);
754 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh) {
755 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
757 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh) {
758 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
761 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
762 static int write_ordered_buffers(spinlock_t *lock,
763 struct reiserfs_journal *j,
764 struct reiserfs_journal_list *jl,
765 struct list_head *list)
767 struct buffer_head *bh;
768 struct reiserfs_jh *jh;
770 struct buffer_chunk chunk;
771 struct list_head tmp;
772 INIT_LIST_HEAD(&tmp);
776 while(!list_empty(list)) {
777 jh = JH_ENTRY(list->next);
780 if (test_set_buffer_locked(bh)) {
781 if (!buffer_dirty(bh)) {
782 list_del_init(&jh->list);
783 list_add(&jh->list, &tmp);
788 write_ordered_chunk(&chunk);
794 if (buffer_dirty(bh)) {
795 list_del_init(&jh->list);
796 list_add(&jh->list, &tmp);
797 add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
799 reiserfs_free_jh(bh);
804 cond_resched_lock(lock);
808 write_ordered_chunk(&chunk);
811 while(!list_empty(&tmp)) {
812 jh = JH_ENTRY(tmp.prev);
815 reiserfs_free_jh(bh);
817 if (buffer_locked(bh)) {
822 if (!buffer_uptodate(bh))
825 cond_resched_lock(lock);
831 static int flush_older_commits(struct super_block *s, struct reiserfs_journal_list *jl) {
832 struct reiserfs_journal_list *other_jl;
833 struct reiserfs_journal_list *first_jl;
834 struct list_head *entry;
835 unsigned long trans_id = jl->j_trans_id;
836 unsigned long other_trans_id;
837 unsigned long first_trans_id;
841 * first we walk backwards to find the oldest uncommitted transation
844 entry = jl->j_list.prev;
846 other_jl = JOURNAL_LIST_ENTRY(entry);
847 if (entry == &SB_JOURNAL(s)->j_journal_list ||
848 atomic_read(&other_jl->j_older_commits_done))
852 entry = other_jl->j_list.prev;
855 /* if we didn't find any older uncommitted transactions, return now */
856 if (first_jl == jl) {
860 first_trans_id = first_jl->j_trans_id;
862 entry = &first_jl->j_list;
864 other_jl = JOURNAL_LIST_ENTRY(entry);
865 other_trans_id = other_jl->j_trans_id;
867 if (other_trans_id < trans_id) {
868 if (atomic_read(&other_jl->j_commit_left) != 0) {
869 flush_commit_list(s, other_jl, 0);
871 /* list we were called with is gone, return */
872 if (!journal_list_still_alive(s, trans_id))
875 /* the one we just flushed is gone, this means all
876 * older lists are also gone, so first_jl is no longer
877 * valid either. Go back to the beginning.
879 if (!journal_list_still_alive(s, other_trans_id)) {
884 if (entry == &SB_JOURNAL(s)->j_journal_list)
892 int reiserfs_async_progress_wait(struct super_block *s) {
894 struct reiserfs_journal *j = SB_JOURNAL(s);
895 if (atomic_read(&j->j_async_throttle))
896 blk_congestion_wait(WRITE, HZ/10);
901 ** if this journal list still has commit blocks unflushed, send them to disk.
903 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
904 ** Before the commit block can by written, every other log block must be safely on disk
907 static int flush_commit_list(struct super_block *s, struct reiserfs_journal_list *jl, int flushall) {
910 struct buffer_head *tbh = NULL ;
911 unsigned long trans_id = jl->j_trans_id;
913 reiserfs_check_lock_depth(s, "flush_commit_list") ;
915 if (atomic_read(&jl->j_older_commits_done)) {
919 /* before we can put our commit blocks on disk, we have to make sure everyone older than
924 if (trans_id == SB_JOURNAL(s)->j_trans_id)
927 get_journal_list(jl);
929 if (flush_older_commits(s, jl) == 1) {
930 /* list disappeared during flush_older_commits. return */
935 /* make sure nobody is trying to flush this one at the same time */
936 down(&jl->j_commit_lock);
937 if (!journal_list_still_alive(s, trans_id)) {
938 up(&jl->j_commit_lock);
941 if (jl->j_trans_id == 0)
944 /* this commit is done, exit */
945 if (atomic_read(&(jl->j_commit_left)) <= 0) {
947 atomic_set(&(jl->j_older_commits_done), 1) ;
949 up(&jl->j_commit_lock);
953 if (!list_empty(&jl->j_bh_list)) {
955 write_ordered_buffers(&SB_JOURNAL(s)->j_dirty_buffers_lock,
956 SB_JOURNAL(s), jl, &jl->j_bh_list);
959 if (!list_empty(&jl->j_bh_list))
962 * for the description block and all the log blocks, submit any buffers
963 * that haven't already reached the disk
965 atomic_inc(&SB_JOURNAL(s)->j_async_throttle);
966 for (i = 0 ; i < (jl->j_len + 1) ; i++) {
967 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start+i) %
968 SB_ONDISK_JOURNAL_SIZE(s);
969 tbh = journal_find_get_block(s, bn) ;
970 if (buffer_dirty(tbh))
971 ll_rw_block(WRITE, 1, &tbh) ;
974 atomic_dec(&SB_JOURNAL(s)->j_async_throttle);
976 /* wait on everything written so far before writing the commit */
977 for (i = 0 ; i < (jl->j_len + 1) ; i++) {
978 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
979 (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s) ;
980 tbh = journal_find_get_block(s, bn) ;
981 wait_on_buffer(tbh) ;
982 // since we're using ll_rw_blk above, it might have skipped over
983 // a locked buffer. Double check here
985 if (buffer_dirty(tbh))
986 sync_dirty_buffer(tbh);
987 if (!buffer_uptodate(tbh)) {
988 reiserfs_panic(s, "journal-601, buffer write failed\n") ;
990 put_bh(tbh) ; /* once for journal_find_get_block */
991 put_bh(tbh) ; /* once due to original getblk in do_journal_end */
992 atomic_dec(&(jl->j_commit_left)) ;
995 if (atomic_read(&(jl->j_commit_left)) != 1)
998 if (buffer_dirty(jl->j_commit_bh))
1000 mark_buffer_dirty(jl->j_commit_bh) ;
1001 sync_dirty_buffer(jl->j_commit_bh) ;
1002 if (!buffer_uptodate(jl->j_commit_bh)) {
1003 reiserfs_panic(s, "journal-615: buffer write failed\n") ;
1005 bforget(jl->j_commit_bh) ;
1006 if (SB_JOURNAL(s)->j_last_commit_id != 0 &&
1007 (jl->j_trans_id - SB_JOURNAL(s)->j_last_commit_id) != 1) {
1008 reiserfs_warning(s, "clm-2200: last commit %lu, current %lu",
1009 SB_JOURNAL(s)->j_last_commit_id,
1012 SB_JOURNAL(s)->j_last_commit_id = jl->j_trans_id;
1014 /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */
1015 cleanup_freed_for_journal_list(s, jl) ;
1017 /* mark the metadata dirty */
1018 dirty_one_transaction(s, jl);
1019 atomic_dec(&(jl->j_commit_left)) ;
1022 atomic_set(&(jl->j_older_commits_done), 1) ;
1024 up(&jl->j_commit_lock);
1026 put_journal_list(s, jl);
1031 ** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or
1032 ** returns NULL if it can't find anything
1034 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct reiserfs_journal_cnode *cn) {
1035 struct super_block *sb = cn->sb;
1036 b_blocknr_t blocknr = cn->blocknr ;
1040 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1048 void remove_journal_hash(struct super_block *, struct reiserfs_journal_cnode **,
1049 struct reiserfs_journal_list *, unsigned long, int);
1052 ** once all the real blocks have been flushed, it is safe to remove them from the
1053 ** journal list for this transaction. Aside from freeing the cnode, this also allows the
1054 ** block to be reallocated for data blocks if it had been deleted.
1056 static void remove_all_from_journal_list(struct super_block *p_s_sb, struct reiserfs_journal_list *jl, int debug) {
1057 struct reiserfs_journal_cnode *cn, *last ;
1058 cn = jl->j_realblock ;
1060 /* which is better, to lock once around the whole loop, or
1061 ** to lock for each call to remove_journal_hash?
1064 if (cn->blocknr != 0) {
1066 reiserfs_warning (p_s_sb, "block %u, bh is %d, state %ld", cn->blocknr,
1067 cn->bh ? 1: 0, cn->state) ;
1070 remove_journal_hash(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_hash_table, jl, cn->blocknr, 1) ;
1074 free_cnode(p_s_sb, last) ;
1076 jl->j_realblock = NULL ;
1080 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1081 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1082 ** releasing blocks in this transaction for reuse as data blocks.
1083 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1086 static int _update_journal_header_block(struct super_block *p_s_sb, unsigned long offset, unsigned long trans_id) {
1087 struct reiserfs_journal_header *jh ;
1088 if (trans_id >= SB_JOURNAL(p_s_sb)->j_last_flush_trans_id) {
1089 if (buffer_locked((SB_JOURNAL(p_s_sb)->j_header_bh))) {
1090 wait_on_buffer((SB_JOURNAL(p_s_sb)->j_header_bh)) ;
1091 if (!buffer_uptodate(SB_JOURNAL(p_s_sb)->j_header_bh)) {
1092 reiserfs_panic(p_s_sb, "journal-699: buffer write failed\n") ;
1095 SB_JOURNAL(p_s_sb)->j_last_flush_trans_id = trans_id ;
1096 SB_JOURNAL(p_s_sb)->j_first_unflushed_offset = offset ;
1097 jh = (struct reiserfs_journal_header *)(SB_JOURNAL(p_s_sb)->j_header_bh->b_data) ;
1098 jh->j_last_flush_trans_id = cpu_to_le32(trans_id) ;
1099 jh->j_first_unflushed_offset = cpu_to_le32(offset) ;
1100 jh->j_mount_id = cpu_to_le32(SB_JOURNAL(p_s_sb)->j_mount_id) ;
1101 set_buffer_dirty(SB_JOURNAL(p_s_sb)->j_header_bh) ;
1102 sync_dirty_buffer(SB_JOURNAL(p_s_sb)->j_header_bh) ;
1103 if (!buffer_uptodate(SB_JOURNAL(p_s_sb)->j_header_bh)) {
1104 reiserfs_warning (p_s_sb, "journal-837: IO error during journal replay");
1111 static int update_journal_header_block(struct super_block *p_s_sb,
1112 unsigned long offset,
1113 unsigned long trans_id) {
1114 if (_update_journal_header_block(p_s_sb, offset, trans_id)) {
1115 reiserfs_panic(p_s_sb, "journal-712: buffer write failed\n") ;
1120 ** flush any and all journal lists older than you are
1121 ** can only be called from flush_journal_list
1123 static int flush_older_journal_lists(struct super_block *p_s_sb,
1124 struct reiserfs_journal_list *jl)
1126 struct list_head *entry;
1127 struct reiserfs_journal_list *other_jl ;
1128 unsigned long trans_id = jl->j_trans_id;
1130 /* we know we are the only ones flushing things, no extra race
1131 * protection is required.
1134 entry = SB_JOURNAL(p_s_sb)->j_journal_list.next;
1135 other_jl = JOURNAL_LIST_ENTRY(entry);
1136 if (other_jl->j_trans_id < trans_id) {
1137 /* do not flush all */
1138 flush_journal_list(p_s_sb, other_jl, 0) ;
1140 /* other_jl is now deleted from the list */
1146 static void del_from_work_list(struct super_block *s,
1147 struct reiserfs_journal_list *jl) {
1148 if (!list_empty(&jl->j_working_list)) {
1149 list_del_init(&jl->j_working_list);
1150 SB_JOURNAL(s)->j_num_work_lists--;
1154 /* flush a journal list, both commit and real blocks
1156 ** always set flushall to 1, unless you are calling from inside
1157 ** flush_journal_list
1159 ** IMPORTANT. This can only be called while there are no journal writers,
1160 ** and the journal is locked. That means it can only be called from
1161 ** do_journal_end, or by journal_release
1163 static int flush_journal_list(struct super_block *s,
1164 struct reiserfs_journal_list *jl, int flushall) {
1165 struct reiserfs_journal_list *pjl ;
1166 struct reiserfs_journal_cnode *cn, *last ;
1170 struct buffer_head *saved_bh ;
1171 unsigned long j_len_saved = jl->j_len ;
1173 if (j_len_saved <= 0) {
1177 if (atomic_read(&SB_JOURNAL(s)->j_wcount) != 0) {
1178 reiserfs_warning(s, "clm-2048: flush_journal_list called with wcount %d",
1179 atomic_read(&SB_JOURNAL(s)->j_wcount)) ;
1181 if (jl->j_trans_id == 0)
1184 /* if flushall == 0, the lock is already held */
1186 down(&SB_JOURNAL(s)->j_flush_sem);
1187 } else if (!down_trylock(&SB_JOURNAL(s)->j_flush_sem)) {
1192 if (j_len_saved > SB_JOURNAL_TRANS_MAX(s)) {
1193 reiserfs_panic(s, "journal-715: flush_journal_list, length is %lu, trans id %lu\n", j_len_saved, jl->j_trans_id);
1197 /* if all the work is already done, get out of here */
1198 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1199 atomic_read(&(jl->j_commit_left)) <= 0) {
1200 goto flush_older_and_return ;
1203 /* start by putting the commit list on disk. This will also flush
1204 ** the commit lists of any olders transactions
1206 flush_commit_list(s, jl, 1) ;
1208 if (!(jl->j_state & LIST_DIRTY))
1211 /* are we done now? */
1212 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1213 atomic_read(&(jl->j_commit_left)) <= 0) {
1214 goto flush_older_and_return ;
1217 /* loop through each cnode, see if we need to write it,
1218 ** or wait on a more recent transaction, or just ignore it
1220 if (atomic_read(&(SB_JOURNAL(s)->j_wcount)) != 0) {
1221 reiserfs_panic(s, "journal-844: panic journal list is flushing, wcount is not 0\n") ;
1223 cn = jl->j_realblock ;
1228 /* blocknr of 0 is no longer in the hash, ignore it */
1229 if (cn->blocknr == 0) {
1232 pjl = find_newer_jl_for_cn(cn) ;
1233 /* the order is important here. We check pjl to make sure we
1234 ** don't clear BH_JDirty_wait if we aren't the one writing this
1237 if (!pjl && cn->bh) {
1240 /* we do this to make sure nobody releases the buffer while
1241 ** we are working with it
1245 if (buffer_journal_dirty(saved_bh)) {
1250 } else if (can_dirty(cn)) {
1251 /* everything with !pjl && jwait should be writable */
1256 /* if someone has this block in a newer transaction, just make
1257 ** sure they are commited, and don't try writing it to disk
1260 if (atomic_read(&pjl->j_commit_left))
1261 flush_commit_list(s, pjl, 1) ;
1265 /* bh == NULL when the block got to disk on its own, OR,
1266 ** the block got freed in a future transaction
1268 if (saved_bh == NULL) {
1272 /* this should never happen. kupdate_one_transaction has this list
1273 ** locked while it works, so we should never see a buffer here that
1274 ** is not marked JDirty_wait
1276 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1277 reiserfs_warning (s, "journal-813: BAD! buffer %llu %cdirty %cjwait, "
1278 "not in a newer tranasction",
1279 (unsigned long long)saved_bh->b_blocknr,
1280 was_dirty ? ' ' : '!', was_jwait ? ' ' : '!') ;
1283 /* we inc again because saved_bh gets decremented at free_cnode */
1285 set_bit(BLOCK_NEEDS_FLUSH, &cn->state) ;
1286 lock_buffer(saved_bh);
1287 if (cn->blocknr != saved_bh->b_blocknr)
1289 if (buffer_dirty(saved_bh))
1290 submit_logged_buffer(saved_bh) ;
1292 unlock_buffer(saved_bh);
1295 reiserfs_warning (s, "clm-2082: Unable to flush buffer %llu in %s",
1296 (unsigned long long)saved_bh->b_blocknr, __FUNCTION__);
1302 /* we incremented this to keep others from taking the buffer head away */
1304 if (atomic_read(&(saved_bh->b_count)) < 0) {
1305 reiserfs_warning (s, "journal-945: saved_bh->b_count < 0");
1310 cn = jl->j_realblock ;
1312 if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1314 reiserfs_panic(s, "journal-1011: cn->bh is NULL\n") ;
1316 wait_on_buffer(cn->bh) ;
1318 reiserfs_panic(s, "journal-1012: cn->bh is NULL\n") ;
1320 if (!buffer_uptodate(cn->bh)) {
1321 reiserfs_panic(s, "journal-949: buffer write failed\n") ;
1323 /* note, we must clear the JDirty_wait bit after the up to date
1324 ** check, otherwise we race against our flushpage routine
1326 if (!test_and_clear_bit(BH_JDirty_wait, &cn->bh->b_state))
1329 /* undo the inc from journal_mark_dirty */
1337 flush_older_and_return:
1338 /* before we can update the journal header block, we _must_ flush all
1339 ** real blocks from all older transactions to disk. This is because
1340 ** once the header block is updated, this transaction will not be
1341 ** replayed after a crash
1344 flush_older_journal_lists(s, jl);
1347 /* before we can remove everything from the hash tables for this
1348 ** transaction, we must make sure it can never be replayed
1350 ** since we are only called from do_journal_end, we know for sure there
1351 ** are no allocations going on while we are flushing journal lists. So,
1352 ** we only need to update the journal header block for the last list
1356 update_journal_header_block(s, (jl->j_start + jl->j_len + 2) % SB_ONDISK_JOURNAL_SIZE(s), jl->j_trans_id) ;
1358 remove_all_from_journal_list(s, jl, 0) ;
1359 list_del(&jl->j_list);
1360 SB_JOURNAL(s)->j_num_lists--;
1361 del_from_work_list(s, jl);
1363 if (SB_JOURNAL(s)->j_last_flush_id != 0 &&
1364 (jl->j_trans_id - SB_JOURNAL(s)->j_last_flush_id) != 1) {
1365 reiserfs_warning(s, "clm-2201: last flush %lu, current %lu",
1366 SB_JOURNAL(s)->j_last_flush_id,
1369 SB_JOURNAL(s)->j_last_flush_id = jl->j_trans_id;
1371 /* not strictly required since we are freeing the list, but it should
1372 * help find code using dead lists later on
1375 atomic_set(&(jl->j_nonzerolen), 0) ;
1377 jl->j_realblock = NULL ;
1378 jl->j_commit_bh = NULL ;
1379 jl->j_trans_id = 0 ;
1381 put_journal_list(s, jl);
1383 up(&SB_JOURNAL(s)->j_flush_sem);
1387 static int write_one_transaction(struct super_block *s,
1388 struct reiserfs_journal_list *jl,
1389 struct buffer_chunk *chunk)
1391 struct reiserfs_journal_cnode *cn;
1394 jl->j_state |= LIST_TOUCHED;
1395 del_from_work_list(s, jl);
1396 if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1400 cn = jl->j_realblock ;
1402 /* if the blocknr == 0, this has been cleared from the hash,
1405 if (cn->blocknr == 0) {
1408 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1409 struct buffer_head *tmp_bh;
1410 /* we can race against journal_mark_freed when we try
1411 * to lock_buffer(cn->bh), so we have to inc the buffer
1412 * count, and recheck things after locking
1416 lock_buffer(tmp_bh);
1417 if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1418 if (!buffer_journal_dirty(tmp_bh) ||
1419 reiserfs_buffer_prepared(tmp_bh))
1421 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1424 /* note, cn->bh might be null now */
1425 unlock_buffer(tmp_bh);
1436 /* used by flush_commit_list */
1437 static int dirty_one_transaction(struct super_block *s,
1438 struct reiserfs_journal_list *jl)
1440 struct reiserfs_journal_cnode *cn;
1441 struct reiserfs_journal_list *pjl;
1444 jl->j_state |= LIST_DIRTY;
1445 cn = jl->j_realblock ;
1447 /* look for a more recent transaction that logged this
1448 ** buffer. Only the most recent transaction with a buffer in
1449 ** it is allowed to send that buffer to disk
1451 pjl = find_newer_jl_for_cn(cn) ;
1452 if (!pjl && cn->blocknr && cn->bh && buffer_journal_dirty(cn->bh))
1456 /* if the buffer is prepared, it will either be logged
1457 * or restored. If restored, we need to make sure
1458 * it actually gets marked dirty
1460 mark_buffer_notjournal_new(cn->bh) ;
1461 if (test_bit(BH_JPrepared, &cn->bh->b_state)) {
1462 set_bit(BH_JRestore_dirty, &cn->bh->b_state);
1464 set_bit(BH_JTest, &cn->bh->b_state);
1465 mark_buffer_dirty(cn->bh);
1473 static int kupdate_transactions(struct super_block *s,
1474 struct reiserfs_journal_list *jl,
1475 struct reiserfs_journal_list **next_jl,
1476 unsigned long *next_trans_id,
1481 int transactions_flushed = 0;
1482 unsigned long orig_trans_id = jl->j_trans_id;
1483 struct buffer_chunk chunk;
1484 struct list_head *entry;
1487 down(&SB_JOURNAL(s)->j_flush_sem);
1488 if (!journal_list_still_alive(s, orig_trans_id)) {
1492 /* we've got j_flush_sem held, nobody is going to delete any
1493 * of these lists out from underneath us
1495 while((num_trans && transactions_flushed < num_trans) ||
1496 (!num_trans && written < num_blocks)) {
1498 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1499 atomic_read(&jl->j_commit_left))
1501 del_from_work_list(s, jl);
1504 ret = write_one_transaction(s, jl, &chunk);
1508 transactions_flushed++;
1510 entry = jl->j_list.next;
1513 if (entry == &SB_JOURNAL(s)->j_journal_list) {
1516 jl = JOURNAL_LIST_ENTRY(entry);
1518 /* don't bother with older transactions */
1519 if (jl->j_trans_id <= orig_trans_id)
1523 write_chunk(&chunk);
1527 up(&SB_JOURNAL(s)->j_flush_sem);
1531 /* for o_sync and fsync heavy applications, they tend to use
1532 ** all the journa list slots with tiny transactions. These
1533 ** trigger lots and lots of calls to update the header block, which
1534 ** adds seeks and slows things down.
1536 ** This function tries to clear out a large chunk of the journal lists
1537 ** at once, which makes everything faster since only the newest journal
1538 ** list updates the header block
1540 static int flush_used_journal_lists(struct super_block *s,
1541 struct reiserfs_journal_list *jl) {
1542 unsigned long len = 0;
1543 unsigned long cur_len;
1547 struct reiserfs_journal_list *tjl;
1548 struct reiserfs_journal_list *flush_jl;
1549 unsigned long trans_id;
1551 flush_jl = tjl = jl;
1553 /* in data logging mode, try harder to flush a lot of blocks */
1554 if (reiserfs_data_log(s))
1556 /* flush for 256 transactions or limit blocks, whichever comes first */
1557 for(i = 0 ; i < 256 && len < limit ; i++) {
1558 if (atomic_read(&tjl->j_commit_left) ||
1559 tjl->j_trans_id < jl->j_trans_id) {
1562 cur_len = atomic_read(&tjl->j_nonzerolen);
1564 tjl->j_state &= ~LIST_TOUCHED;
1568 if (tjl->j_list.next == &SB_JOURNAL(s)->j_journal_list)
1570 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1572 /* try to find a group of blocks we can flush across all the
1573 ** transactions, but only bother if we've actually spanned
1574 ** across multiple lists
1576 if (flush_jl != jl) {
1577 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1579 flush_journal_list(s, flush_jl, 1);
1584 ** removes any nodes in table with name block and dev as bh.
1585 ** only touchs the hnext and hprev pointers.
1587 void remove_journal_hash(struct super_block *sb,
1588 struct reiserfs_journal_cnode **table,
1589 struct reiserfs_journal_list *jl,
1590 unsigned long block, int remove_freed)
1592 struct reiserfs_journal_cnode *cur ;
1593 struct reiserfs_journal_cnode **head ;
1595 head= &(journal_hash(table, sb, block)) ;
1601 if (cur->blocknr == block && cur->sb == sb && (jl == NULL || jl == cur->jlist) &&
1602 (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1604 cur->hnext->hprev = cur->hprev ;
1607 cur->hprev->hnext = cur->hnext ;
1609 *head = cur->hnext ;
1614 if (cur->bh && cur->jlist) /* anybody who clears the cur->bh will also dec the nonzerolen */
1615 atomic_dec(&(cur->jlist->j_nonzerolen)) ;
1623 static void free_journal_ram(struct super_block *p_s_sb) {
1624 reiserfs_kfree(SB_JOURNAL(p_s_sb)->j_current_jl,
1625 sizeof(struct reiserfs_journal_list), p_s_sb);
1626 SB_JOURNAL(p_s_sb)->j_num_lists--;
1628 vfree(SB_JOURNAL(p_s_sb)->j_cnode_free_orig) ;
1629 free_list_bitmaps(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_bitmap) ;
1630 free_bitmap_nodes(p_s_sb) ; /* must be after free_list_bitmaps */
1631 if (SB_JOURNAL(p_s_sb)->j_header_bh) {
1632 brelse(SB_JOURNAL(p_s_sb)->j_header_bh) ;
1634 /* j_header_bh is on the journal dev, make sure not to release the journal
1635 * dev until we brelse j_header_bh
1637 release_journal_dev(p_s_sb, SB_JOURNAL(p_s_sb));
1638 vfree(SB_JOURNAL(p_s_sb)) ;
1642 ** call on unmount. Only set error to 1 if you haven't made your way out
1643 ** of read_super() yet. Any other caller must keep error at 0.
1645 static int do_journal_release(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, int error) {
1646 struct reiserfs_transaction_handle myth ;
1648 /* we only want to flush out transactions if we were called with error == 0
1650 if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
1651 /* end the current trans */
1652 do_journal_end(th, p_s_sb,10, FLUSH_ALL) ;
1654 /* make sure something gets logged to force our way into the flush code */
1655 journal_join(&myth, p_s_sb, 1) ;
1656 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
1657 journal_mark_dirty(&myth, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
1658 do_journal_end(&myth, p_s_sb,1, FLUSH_ALL) ;
1661 reiserfs_mounted_fs_count-- ;
1662 /* wait for all commits to finish */
1663 cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work);
1664 flush_workqueue(commit_wq);
1665 if (!reiserfs_mounted_fs_count) {
1666 destroy_workqueue(commit_wq);
1670 free_journal_ram(p_s_sb) ;
1676 ** call on unmount. flush all journal trans, release all alloc'd ram
1678 int journal_release(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb) {
1679 return do_journal_release(th, p_s_sb, 0) ;
1682 ** only call from an error condition inside reiserfs_read_super!
1684 int journal_release_error(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb) {
1685 return do_journal_release(th, p_s_sb, 1) ;
1688 /* compares description block with commit block. returns 1 if they differ, 0 if they are the same */
1689 static int journal_compare_desc_commit(struct super_block *p_s_sb, struct reiserfs_journal_desc *desc,
1690 struct reiserfs_journal_commit *commit) {
1691 if (get_commit_trans_id (commit) != get_desc_trans_id (desc) ||
1692 get_commit_trans_len (commit) != get_desc_trans_len (desc) ||
1693 get_commit_trans_len (commit) > SB_JOURNAL_TRANS_MAX(p_s_sb) ||
1694 get_commit_trans_len (commit) <= 0
1700 /* returns 0 if it did not find a description block
1701 ** returns -1 if it found a corrupt commit block
1702 ** returns 1 if both desc and commit were valid
1704 static int journal_transaction_is_valid(struct super_block *p_s_sb, struct buffer_head *d_bh, unsigned long *oldest_invalid_trans_id, unsigned long *newest_mount_id) {
1705 struct reiserfs_journal_desc *desc ;
1706 struct reiserfs_journal_commit *commit ;
1707 struct buffer_head *c_bh ;
1708 unsigned long offset ;
1713 desc = (struct reiserfs_journal_desc *)d_bh->b_data ;
1714 if (get_desc_trans_len(desc) > 0 && !memcmp(get_journal_desc_magic (d_bh), JOURNAL_DESC_MAGIC, 8)) {
1715 if (oldest_invalid_trans_id && *oldest_invalid_trans_id && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
1716 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-986: transaction "
1717 "is valid returning because trans_id %d is greater than "
1718 "oldest_invalid %lu", get_desc_trans_id(desc),
1719 *oldest_invalid_trans_id);
1722 if (newest_mount_id && *newest_mount_id > get_desc_mount_id (desc)) {
1723 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1087: transaction "
1724 "is valid returning because mount_id %d is less than "
1725 "newest_mount_id %lu", get_desc_mount_id (desc),
1729 if ( get_desc_trans_len(desc) > SB_JOURNAL_TRANS_MAX(p_s_sb) ) {
1730 reiserfs_warning(p_s_sb, "journal-2018: Bad transaction length %d encountered, ignoring transaction", get_desc_trans_len(desc));
1733 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ;
1735 /* ok, we have a journal description block, lets see if the transaction was valid */
1736 c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
1737 ((offset + get_desc_trans_len(desc) + 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ;
1740 commit = (struct reiserfs_journal_commit *)c_bh->b_data ;
1741 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
1742 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1743 "journal_transaction_is_valid, commit offset %ld had bad "
1744 "time %d or length %d",
1745 c_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1746 get_commit_trans_id (commit),
1747 get_commit_trans_len(commit));
1749 if (oldest_invalid_trans_id) {
1750 *oldest_invalid_trans_id = get_desc_trans_id(desc) ;
1751 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1004: "
1752 "transaction_is_valid setting oldest invalid trans_id "
1753 "to %d", get_desc_trans_id(desc)) ;
1758 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1006: found valid "
1759 "transaction start offset %llu, len %d id %d",
1760 d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1761 get_desc_trans_len(desc), get_desc_trans_id(desc)) ;
1768 static void brelse_array(struct buffer_head **heads, int num) {
1770 for (i = 0 ; i < num ; i++) {
1776 ** given the start, and values for the oldest acceptable transactions,
1777 ** this either reads in a replays a transaction, or returns because the transaction
1778 ** is invalid, or too old.
1780 static int journal_read_transaction(struct super_block *p_s_sb, unsigned long cur_dblock, unsigned long oldest_start,
1781 unsigned long oldest_trans_id, unsigned long newest_mount_id) {
1782 struct reiserfs_journal_desc *desc ;
1783 struct reiserfs_journal_commit *commit ;
1784 unsigned long trans_id = 0 ;
1785 struct buffer_head *c_bh ;
1786 struct buffer_head *d_bh ;
1787 struct buffer_head **log_blocks = NULL ;
1788 struct buffer_head **real_blocks = NULL ;
1789 unsigned long trans_offset ;
1793 d_bh = journal_bread(p_s_sb, cur_dblock) ;
1796 desc = (struct reiserfs_journal_desc *)d_bh->b_data ;
1797 trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ;
1798 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
1799 "journal_read_transaction, offset %llu, len %d mount_id %d",
1800 d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1801 get_desc_trans_len(desc), get_desc_mount_id(desc)) ;
1802 if (get_desc_trans_id(desc) < oldest_trans_id) {
1803 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
1804 "journal_read_trans skipping because %lu is too old",
1805 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)) ;
1809 if (get_desc_mount_id(desc) != newest_mount_id) {
1810 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
1811 "journal_read_trans skipping because %d is != "
1812 "newest_mount_id %lu", get_desc_mount_id(desc),
1817 c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
1818 ((trans_offset + get_desc_trans_len(desc) + 1) %
1819 SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ;
1824 commit = (struct reiserfs_journal_commit *)c_bh->b_data ;
1825 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
1826 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal_read_transaction, "
1827 "commit offset %llu had bad time %d or length %d",
1828 c_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1829 get_commit_trans_id(commit), get_commit_trans_len(commit));
1834 trans_id = get_desc_trans_id(desc) ;
1835 /* now we know we've got a good transaction, and it was inside the valid time ranges */
1836 log_blocks = reiserfs_kmalloc(get_desc_trans_len(desc) * sizeof(struct buffer_head *), GFP_NOFS, p_s_sb) ;
1837 real_blocks = reiserfs_kmalloc(get_desc_trans_len(desc) * sizeof(struct buffer_head *), GFP_NOFS, p_s_sb) ;
1838 if (!log_blocks || !real_blocks) {
1841 reiserfs_kfree(log_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1842 reiserfs_kfree(real_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1843 reiserfs_warning(p_s_sb, "journal-1169: kmalloc failed, unable to mount FS") ;
1846 /* get all the buffer heads */
1847 trans_half = journal_trans_half (p_s_sb->s_blocksize) ;
1848 for(i = 0 ; i < get_desc_trans_len(desc) ; i++) {
1849 log_blocks[i] = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + (trans_offset + 1 + i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
1850 if (i < trans_half) {
1851 real_blocks[i] = sb_getblk(p_s_sb, le32_to_cpu(desc->j_realblock[i])) ;
1853 real_blocks[i] = sb_getblk(p_s_sb, le32_to_cpu(commit->j_realblock[i - trans_half])) ;
1855 if ( real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb) ) {
1856 reiserfs_warning(p_s_sb, "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem");
1859 /* make sure we don't try to replay onto log or reserved area */
1860 if (is_block_in_log_or_reserved_area(p_s_sb, real_blocks[i]->b_blocknr)) {
1861 reiserfs_warning(p_s_sb, "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block") ;
1863 brelse_array(log_blocks, i) ;
1864 brelse_array(real_blocks, i) ;
1867 reiserfs_kfree(log_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1868 reiserfs_kfree(real_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1872 /* read in the log blocks, memcpy to the corresponding real block */
1873 ll_rw_block(READ, get_desc_trans_len(desc), log_blocks) ;
1874 for (i = 0 ; i < get_desc_trans_len(desc) ; i++) {
1875 wait_on_buffer(log_blocks[i]) ;
1876 if (!buffer_uptodate(log_blocks[i])) {
1877 reiserfs_warning(p_s_sb, "journal-1212: REPLAY FAILURE fsck required! buffer write failed") ;
1878 brelse_array(log_blocks + i, get_desc_trans_len(desc) - i) ;
1879 brelse_array(real_blocks, get_desc_trans_len(desc)) ;
1882 reiserfs_kfree(log_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1883 reiserfs_kfree(real_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1886 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data, real_blocks[i]->b_size) ;
1887 set_buffer_uptodate(real_blocks[i]) ;
1888 brelse(log_blocks[i]) ;
1890 /* flush out the real blocks */
1891 for (i = 0 ; i < get_desc_trans_len(desc) ; i++) {
1892 set_buffer_dirty(real_blocks[i]) ;
1893 ll_rw_block(WRITE, 1, real_blocks + i) ;
1895 for (i = 0 ; i < get_desc_trans_len(desc) ; i++) {
1896 wait_on_buffer(real_blocks[i]) ;
1897 if (!buffer_uptodate(real_blocks[i])) {
1898 reiserfs_warning(p_s_sb, "journal-1226: REPLAY FAILURE, fsck required! buffer write failed") ;
1899 brelse_array(real_blocks + i, get_desc_trans_len(desc) - i) ;
1902 reiserfs_kfree(log_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1903 reiserfs_kfree(real_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1906 brelse(real_blocks[i]) ;
1908 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + ((trans_offset + get_desc_trans_len(desc) + 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)) ;
1909 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1095: setting journal "
1910 "start to offset %ld",
1911 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)) ;
1913 /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
1914 SB_JOURNAL(p_s_sb)->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ;
1915 SB_JOURNAL(p_s_sb)->j_last_flush_trans_id = trans_id ;
1916 SB_JOURNAL(p_s_sb)->j_trans_id = trans_id + 1;
1919 reiserfs_kfree(log_blocks, le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), p_s_sb) ;
1920 reiserfs_kfree(real_blocks, le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), p_s_sb) ;
1924 /* This function reads blocks starting from block and to max_block of bufsize
1925 size (but no more than BUFNR blocks at a time). This proved to improve
1926 mounting speed on self-rebuilding raid5 arrays at least.
1927 Right now it is only used from journal code. But later we might use it
1929 Note: Do not use journal_getblk/sb_getblk functions here! */
1930 struct buffer_head * reiserfs_breada (struct block_device *dev, int block, int bufsize,
1931 unsigned int max_block)
1933 struct buffer_head * bhlist[BUFNR];
1934 unsigned int blocks = BUFNR;
1935 struct buffer_head * bh;
1938 bh = __getblk (dev, block, bufsize );
1939 if (buffer_uptodate (bh))
1942 if (block + BUFNR > max_block) {
1943 blocks = max_block - block;
1947 for (i = 1; i < blocks; i++) {
1948 bh = __getblk (dev, block + i, bufsize);
1949 if (buffer_uptodate (bh)) {
1953 else bhlist[j++] = bh;
1955 ll_rw_block (READ, j, bhlist);
1956 for(i = 1; i < j; i++)
1959 wait_on_buffer (bh);
1960 if (buffer_uptodate (bh))
1967 ** read and replay the log
1968 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
1969 ** transaction. This tests that before finding all the transactions in the log, which makes normal mount times fast.
1971 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
1973 ** On exit, it sets things up so the first transaction will work correctly.
1975 static int journal_read(struct super_block *p_s_sb) {
1976 struct reiserfs_journal_desc *desc ;
1977 unsigned long oldest_trans_id = 0;
1978 unsigned long oldest_invalid_trans_id = 0 ;
1980 unsigned long oldest_start = 0;
1981 unsigned long cur_dblock = 0 ;
1982 unsigned long newest_mount_id = 9 ;
1983 struct buffer_head *d_bh ;
1984 struct reiserfs_journal_header *jh ;
1985 int valid_journal_header = 0 ;
1986 int replay_count = 0 ;
1987 int continue_replay = 1 ;
1989 char b[BDEVNAME_SIZE];
1991 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ;
1992 reiserfs_info (p_s_sb, "checking transaction log (%s)\n",
1993 bdevname(SB_JOURNAL(p_s_sb)->j_dev_bd, b));
1994 start = get_seconds();
1996 /* step 1, read in the journal header block. Check the transaction it says
1997 ** is the first unflushed, and if that transaction is not valid,
2000 SB_JOURNAL(p_s_sb)->j_header_bh = journal_bread(p_s_sb,
2001 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2002 SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2003 if (!SB_JOURNAL(p_s_sb)->j_header_bh) {
2006 jh = (struct reiserfs_journal_header *)(SB_JOURNAL(p_s_sb)->j_header_bh->b_data) ;
2007 if (le32_to_cpu(jh->j_first_unflushed_offset) >= 0 &&
2008 le32_to_cpu(jh->j_first_unflushed_offset) < SB_ONDISK_JOURNAL_SIZE(p_s_sb) &&
2009 le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2010 oldest_start = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2011 le32_to_cpu(jh->j_first_unflushed_offset) ;
2012 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2013 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2014 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1153: found in "
2015 "header: first_unflushed_offset %d, last_flushed_trans_id "
2016 "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2017 le32_to_cpu(jh->j_last_flush_trans_id)) ;
2018 valid_journal_header = 1 ;
2020 /* now, we try to read the first unflushed offset. If it is not valid,
2021 ** there is nothing more we can do, and it makes no sense to read
2022 ** through the whole log.
2024 d_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + le32_to_cpu(jh->j_first_unflushed_offset)) ;
2025 ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL) ;
2027 continue_replay = 0 ;
2030 goto start_log_replay;
2033 if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) {
2034 reiserfs_warning (p_s_sb,
2035 "clm-2076: device is readonly, unable to replay log") ;
2039 /* ok, there are transactions that need to be replayed. start with the first log block, find
2040 ** all the valid transactions, and pick out the oldest.
2042 while(continue_replay && cur_dblock < (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_ONDISK_JOURNAL_SIZE(p_s_sb))) {
2043 /* Note that it is required for blocksize of primary fs device and journal
2044 device to be the same */
2045 d_bh = reiserfs_breada(SB_JOURNAL(p_s_sb)->j_dev_bd, cur_dblock, p_s_sb->s_blocksize,
2046 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_ONDISK_JOURNAL_SIZE(p_s_sb)) ;
2047 ret = journal_transaction_is_valid(p_s_sb, d_bh, &oldest_invalid_trans_id, &newest_mount_id) ;
2049 desc = (struct reiserfs_journal_desc *)d_bh->b_data ;
2050 if (oldest_start == 0) { /* init all oldest_ values */
2051 oldest_trans_id = get_desc_trans_id(desc) ;
2052 oldest_start = d_bh->b_blocknr ;
2053 newest_mount_id = get_desc_mount_id(desc) ;
2054 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1179: Setting "
2055 "oldest_start to offset %llu, trans_id %lu",
2056 oldest_start - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2058 } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2059 /* one we just read was older */
2060 oldest_trans_id = get_desc_trans_id(desc) ;
2061 oldest_start = d_bh->b_blocknr ;
2062 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1180: Resetting "
2063 "oldest_start to offset %lu, trans_id %lu",
2064 oldest_start - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2067 if (newest_mount_id < get_desc_mount_id(desc)) {
2068 newest_mount_id = get_desc_mount_id(desc) ;
2069 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2070 "newest_mount_id to %d", get_desc_mount_id(desc));
2072 cur_dblock += get_desc_trans_len(desc) + 2 ;
2080 cur_dblock = oldest_start ;
2081 if (oldest_trans_id) {
2082 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1206: Starting replay "
2083 "from offset %llu, trans_id %lu",
2084 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2089 while(continue_replay && oldest_trans_id > 0) {
2090 ret = journal_read_transaction(p_s_sb, cur_dblock, oldest_start, oldest_trans_id, newest_mount_id) ;
2093 } else if (ret != 0) {
2096 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_JOURNAL(p_s_sb)->j_start ;
2098 if (cur_dblock == oldest_start)
2102 if (oldest_trans_id == 0) {
2103 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1225: No valid "
2104 "transactions found") ;
2106 /* j_start does not get set correctly if we don't replay any transactions.
2107 ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2108 ** copy the trans_id from the header
2110 if (valid_journal_header && replay_count == 0) {
2111 SB_JOURNAL(p_s_sb)->j_start = le32_to_cpu(jh->j_first_unflushed_offset) ;
2112 SB_JOURNAL(p_s_sb)->j_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2113 SB_JOURNAL(p_s_sb)->j_last_flush_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) ;
2114 SB_JOURNAL(p_s_sb)->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2116 SB_JOURNAL(p_s_sb)->j_mount_id = newest_mount_id + 1 ;
2118 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2119 "newest_mount_id to %lu", SB_JOURNAL(p_s_sb)->j_mount_id) ;
2120 SB_JOURNAL(p_s_sb)->j_first_unflushed_offset = SB_JOURNAL(p_s_sb)->j_start ;
2121 if (replay_count > 0) {
2122 reiserfs_info (p_s_sb, "replayed %d transactions in %lu seconds\n",
2123 replay_count, get_seconds() - start) ;
2125 if (!bdev_read_only(p_s_sb->s_bdev) &&
2126 _update_journal_header_block(p_s_sb, SB_JOURNAL(p_s_sb)->j_start,
2127 SB_JOURNAL(p_s_sb)->j_last_flush_trans_id))
2129 /* replay failed, caller must call free_journal_ram and abort
2137 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2139 struct reiserfs_journal_list *jl;
2141 jl = reiserfs_kmalloc(sizeof(struct reiserfs_journal_list), GFP_NOFS, s);
2146 memset(jl, 0, sizeof(*jl));
2147 INIT_LIST_HEAD(&jl->j_list);
2148 INIT_LIST_HEAD(&jl->j_working_list);
2149 INIT_LIST_HEAD(&jl->j_tail_bh_list);
2150 INIT_LIST_HEAD(&jl->j_bh_list);
2151 sema_init(&jl->j_commit_lock, 1);
2152 SB_JOURNAL(s)->j_num_lists++;
2153 get_journal_list(jl);
2157 static void journal_list_init(struct super_block *p_s_sb) {
2158 SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
2161 static int release_journal_dev( struct super_block *super,
2162 struct reiserfs_journal *journal )
2168 if( journal -> j_dev_file != NULL ) {
2169 result = filp_close( journal -> j_dev_file, NULL );
2170 journal -> j_dev_file = NULL;
2171 journal -> j_dev_bd = NULL;
2172 } else if( journal -> j_dev_bd != NULL ) {
2173 result = blkdev_put( journal -> j_dev_bd );
2174 journal -> j_dev_bd = NULL;
2178 reiserfs_warning(super, "sh-457: release_journal_dev: Cannot release journal device: %i", result );
2183 static int journal_init_dev( struct super_block *super,
2184 struct reiserfs_journal *journal,
2185 const char *jdev_name )
2189 int blkdev_mode = FMODE_READ | FMODE_WRITE;
2190 char b[BDEVNAME_SIZE];
2194 journal -> j_dev_bd = NULL;
2195 journal -> j_dev_file = NULL;
2196 jdev = SB_ONDISK_JOURNAL_DEVICE( super ) ?
2197 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2199 if (bdev_read_only(super->s_bdev))
2200 blkdev_mode = FMODE_READ;
2202 /* there is no "jdev" option and journal is on separate device */
2203 if( ( !jdev_name || !jdev_name[ 0 ] ) ) {
2204 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2205 if (IS_ERR(journal->j_dev_bd)) {
2206 result = PTR_ERR(journal->j_dev_bd);
2207 journal->j_dev_bd = NULL;
2208 reiserfs_warning (super, "sh-458: journal_init_dev: "
2209 "cannot init journal device '%s': %i",
2210 __bdevname(jdev, b), result );
2212 } else if (jdev != super->s_dev)
2213 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2217 journal -> j_dev_file = filp_open( jdev_name, 0, 0 );
2218 if( !IS_ERR( journal -> j_dev_file ) ) {
2219 struct inode *jdev_inode = journal->j_dev_file->f_mapping->host;
2220 if( !S_ISBLK( jdev_inode -> i_mode ) ) {
2221 reiserfs_warning (super, "journal_init_dev: '%s' is "
2222 "not a block device", jdev_name );
2226 journal->j_dev_bd = I_BDEV(jdev_inode);
2227 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2230 result = PTR_ERR( journal -> j_dev_file );
2231 journal -> j_dev_file = NULL;
2232 reiserfs_warning (super,
2233 "journal_init_dev: Cannot open '%s': %i",
2234 jdev_name, result );
2237 release_journal_dev( super, journal );
2239 reiserfs_info(super, "journal_init_dev: journal device: %s\n",
2240 bdevname(journal->j_dev_bd, b));
2245 ** must be called once on fs mount. calls journal_read for you
2247 int journal_init(struct super_block *p_s_sb, const char * j_dev_name, int old_format, unsigned int commit_max_age) {
2248 int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2 ;
2249 struct buffer_head *bhjh;
2250 struct reiserfs_super_block * rs;
2251 struct reiserfs_journal_header *jh;
2252 struct reiserfs_journal *journal;
2253 struct reiserfs_journal_list *jl;
2254 char b[BDEVNAME_SIZE];
2256 journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof (struct reiserfs_journal)) ;
2258 reiserfs_warning (p_s_sb, "journal-1256: unable to get memory for journal structure") ;
2261 memset(journal, 0, sizeof(struct reiserfs_journal)) ;
2262 INIT_LIST_HEAD(&SB_JOURNAL(p_s_sb)->j_bitmap_nodes) ;
2263 INIT_LIST_HEAD (&SB_JOURNAL(p_s_sb)->j_prealloc_list);
2264 INIT_LIST_HEAD(&SB_JOURNAL(p_s_sb)->j_working_list);
2265 INIT_LIST_HEAD(&SB_JOURNAL(p_s_sb)->j_journal_list);
2266 if (reiserfs_allocate_list_bitmaps(p_s_sb,
2267 SB_JOURNAL(p_s_sb)->j_list_bitmap,
2268 SB_BMAP_NR(p_s_sb)))
2269 goto free_and_return ;
2270 allocate_bitmap_nodes(p_s_sb) ;
2272 /* reserved for journal area support */
2273 SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ?
2274 REISERFS_OLD_DISK_OFFSET_IN_BYTES / p_s_sb->s_blocksize +
2275 SB_BMAP_NR(p_s_sb) + 1 :
2276 REISERFS_DISK_OFFSET_IN_BYTES / p_s_sb->s_blocksize + 2);
2278 /* Sanity check to see is the standard journal fitting withing first bitmap
2279 (actual for small blocksizes) */
2280 if ( !SB_ONDISK_JOURNAL_DEVICE( p_s_sb ) &&
2281 (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) + SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8) ) {
2282 reiserfs_warning (p_s_sb, "journal-1393: journal does not fit for area "
2283 "addressed by first of bitmap blocks. It starts at "
2284 "%u and its size is %u. Block size %ld",
2285 SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb),
2286 SB_ONDISK_JOURNAL_SIZE(p_s_sb), p_s_sb->s_blocksize);
2287 goto free_and_return;
2290 if( journal_init_dev( p_s_sb, journal, j_dev_name ) != 0 ) {
2291 reiserfs_warning (p_s_sb, "sh-462: unable to initialize jornal device");
2292 goto free_and_return;
2295 rs = SB_DISK_SUPER_BLOCK(p_s_sb);
2297 /* read journal header */
2298 bhjh = journal_bread(p_s_sb,
2299 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2301 reiserfs_warning (p_s_sb, "sh-459: unable to read journal header");
2302 goto free_and_return;
2304 jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2306 /* make sure that journal matches to the super block */
2307 if (is_reiserfs_jr(rs) && (jh->jh_journal.jp_journal_magic != sb_jp_journal_magic(rs))) {
2308 reiserfs_warning (p_s_sb, "sh-460: journal header magic %x "
2309 "(device %s) does not match to magic found in super "
2311 jh->jh_journal.jp_journal_magic,
2312 bdevname( SB_JOURNAL(p_s_sb)->j_dev_bd, b),
2313 sb_jp_journal_magic(rs));
2315 goto free_and_return;
2318 SB_JOURNAL_TRANS_MAX(p_s_sb) = le32_to_cpu (jh->jh_journal.jp_journal_trans_max);
2319 SB_JOURNAL_MAX_BATCH(p_s_sb) = le32_to_cpu (jh->jh_journal.jp_journal_max_batch);
2320 SB_JOURNAL_MAX_COMMIT_AGE(p_s_sb) = le32_to_cpu (jh->jh_journal.jp_journal_max_commit_age);
2321 SB_JOURNAL_MAX_TRANS_AGE(p_s_sb) = JOURNAL_MAX_TRANS_AGE;
2323 if (SB_JOURNAL_TRANS_MAX(p_s_sb)) {
2324 /* make sure these parameters are available, assign it if they are not */
2325 __u32 initial = SB_JOURNAL_TRANS_MAX(p_s_sb);
2328 if (p_s_sb->s_blocksize < 4096)
2329 ratio = 4096 / p_s_sb->s_blocksize;
2331 if (SB_ONDISK_JOURNAL_SIZE(p_s_sb)/SB_JOURNAL_TRANS_MAX(p_s_sb) < JOURNAL_MIN_RATIO)
2332 SB_JOURNAL_TRANS_MAX(p_s_sb) = SB_ONDISK_JOURNAL_SIZE(p_s_sb) / JOURNAL_MIN_RATIO;
2333 if (SB_JOURNAL_TRANS_MAX(p_s_sb) > JOURNAL_TRANS_MAX_DEFAULT / ratio)
2334 SB_JOURNAL_TRANS_MAX(p_s_sb) = JOURNAL_TRANS_MAX_DEFAULT / ratio;
2335 if (SB_JOURNAL_TRANS_MAX(p_s_sb) < JOURNAL_TRANS_MIN_DEFAULT / ratio)
2336 SB_JOURNAL_TRANS_MAX(p_s_sb) = JOURNAL_TRANS_MIN_DEFAULT / ratio;
2338 if (SB_JOURNAL_TRANS_MAX(p_s_sb) != initial)
2339 reiserfs_warning (p_s_sb, "sh-461: journal_init: wrong transaction max size (%u). Changed to %u",
2340 initial, SB_JOURNAL_TRANS_MAX(p_s_sb));
2342 SB_JOURNAL_MAX_BATCH(p_s_sb) = SB_JOURNAL_TRANS_MAX(p_s_sb)*
2343 JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT;
2346 if (!SB_JOURNAL_TRANS_MAX(p_s_sb)) {
2347 /*we have the file system was created by old version of mkreiserfs
2348 so this field contains zero value */
2349 SB_JOURNAL_TRANS_MAX(p_s_sb) = JOURNAL_TRANS_MAX_DEFAULT ;
2350 SB_JOURNAL_MAX_BATCH(p_s_sb) = JOURNAL_MAX_BATCH_DEFAULT ;
2351 SB_JOURNAL_MAX_COMMIT_AGE(p_s_sb) = JOURNAL_MAX_COMMIT_AGE ;
2353 /* for blocksize >= 4096 - max transaction size is 1024. For block size < 4096
2354 trans max size is decreased proportionally */
2355 if (p_s_sb->s_blocksize < 4096) {
2356 SB_JOURNAL_TRANS_MAX(p_s_sb) /= (4096 / p_s_sb->s_blocksize) ;
2357 SB_JOURNAL_MAX_BATCH(p_s_sb) = (SB_JOURNAL_TRANS_MAX(p_s_sb)) * 9 / 10 ;
2361 SB_JOURNAL_DEFAULT_MAX_COMMIT_AGE(p_s_sb) = SB_JOURNAL_MAX_COMMIT_AGE(p_s_sb);
2363 if (commit_max_age != 0) {
2364 SB_JOURNAL_MAX_COMMIT_AGE(p_s_sb) = commit_max_age;
2365 SB_JOURNAL_MAX_TRANS_AGE(p_s_sb) = commit_max_age;
2368 reiserfs_info (p_s_sb, "journal params: device %s, size %u, "
2369 "journal first block %u, max trans len %u, max batch %u, "
2370 "max commit age %u, max trans age %u\n",
2371 bdevname( SB_JOURNAL(p_s_sb)->j_dev_bd, b),
2372 SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2373 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2374 SB_JOURNAL_TRANS_MAX(p_s_sb),
2375 SB_JOURNAL_MAX_BATCH(p_s_sb),
2376 SB_JOURNAL_MAX_COMMIT_AGE(p_s_sb),
2377 SB_JOURNAL_MAX_TRANS_AGE(p_s_sb));
2381 SB_JOURNAL(p_s_sb)->j_list_bitmap_index = 0 ;
2382 journal_list_init(p_s_sb) ;
2384 memset(SB_JOURNAL(p_s_sb)->j_list_hash_table, 0, JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)) ;
2386 INIT_LIST_HEAD(&SB_JOURNAL(p_s_sb)->j_dirty_buffers) ;
2387 spin_lock_init(&SB_JOURNAL(p_s_sb)->j_dirty_buffers_lock) ;
2389 SB_JOURNAL(p_s_sb)->j_start = 0 ;
2390 SB_JOURNAL(p_s_sb)->j_len = 0 ;
2391 SB_JOURNAL(p_s_sb)->j_len_alloc = 0 ;
2392 atomic_set(&(SB_JOURNAL(p_s_sb)->j_wcount), 0) ;
2393 atomic_set(&(SB_JOURNAL(p_s_sb)->j_async_throttle), 0) ;
2394 SB_JOURNAL(p_s_sb)->j_bcount = 0 ;
2395 SB_JOURNAL(p_s_sb)->j_trans_start_time = 0 ;
2396 SB_JOURNAL(p_s_sb)->j_last = NULL ;
2397 SB_JOURNAL(p_s_sb)->j_first = NULL ;
2398 init_waitqueue_head(&(SB_JOURNAL(p_s_sb)->j_join_wait)) ;
2399 sema_init(&SB_JOURNAL(p_s_sb)->j_lock, 1);
2400 sema_init(&SB_JOURNAL(p_s_sb)->j_flush_sem, 1);
2402 SB_JOURNAL(p_s_sb)->j_trans_id = 10 ;
2403 SB_JOURNAL(p_s_sb)->j_mount_id = 10 ;
2404 SB_JOURNAL(p_s_sb)->j_state = 0 ;
2405 atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 0) ;
2406 SB_JOURNAL(p_s_sb)->j_cnode_free_list = allocate_cnodes(num_cnodes) ;
2407 SB_JOURNAL(p_s_sb)->j_cnode_free_orig = SB_JOURNAL(p_s_sb)->j_cnode_free_list ;
2408 SB_JOURNAL(p_s_sb)->j_cnode_free = SB_JOURNAL(p_s_sb)->j_cnode_free_list ? num_cnodes : 0 ;
2409 SB_JOURNAL(p_s_sb)->j_cnode_used = 0 ;
2410 SB_JOURNAL(p_s_sb)->j_must_wait = 0 ;
2412 init_journal_hash(p_s_sb) ;
2413 jl = SB_JOURNAL(p_s_sb)->j_current_jl;
2414 jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl);
2415 if (!jl->j_list_bitmap) {
2416 reiserfs_warning(p_s_sb, "journal-2005, get_list_bitmap failed for journal list 0") ;
2417 goto free_and_return;
2419 if (journal_read(p_s_sb) < 0) {
2420 reiserfs_warning(p_s_sb, "Replay Failure, unable to mount") ;
2421 goto free_and_return;
2424 reiserfs_mounted_fs_count++ ;
2425 if (reiserfs_mounted_fs_count <= 1)
2426 commit_wq = create_workqueue("reiserfs");
2428 INIT_WORK(&journal->j_work, flush_async_commits, p_s_sb);
2431 free_journal_ram(p_s_sb);
2436 ** test for a polite end of the current transaction. Used by file_write, and should
2437 ** be used by delete to make sure they don't write more than can fit inside a single
2440 int journal_transaction_should_end(struct reiserfs_transaction_handle *th, int new_alloc) {
2441 time_t now = get_seconds() ;
2442 /* cannot restart while nested */
2443 if (th->t_refcount > 1)
2445 if ( SB_JOURNAL(th->t_super)->j_must_wait > 0 ||
2446 (SB_JOURNAL(th->t_super)->j_len_alloc + new_alloc) >= SB_JOURNAL_MAX_BATCH(th->t_super) ||
2447 atomic_read(&(SB_JOURNAL(th->t_super)->j_jlock)) ||
2448 (now - SB_JOURNAL(th->t_super)->j_trans_start_time) > SB_JOURNAL_MAX_TRANS_AGE(th->t_super) ||
2449 SB_JOURNAL(th->t_super)->j_cnode_free < (SB_JOURNAL_TRANS_MAX(th->t_super) * 3)) {
2455 /* this must be called inside a transaction, and requires the
2456 ** kernel_lock to be held
2458 void reiserfs_block_writes(struct reiserfs_transaction_handle *th) {
2459 struct super_block *s = th->t_super ;
2460 SB_JOURNAL(s)->j_must_wait = 1 ;
2461 set_bit(WRITERS_BLOCKED, &SB_JOURNAL(s)->j_state) ;
2465 /* this must be called without a transaction started, and does not
2468 void reiserfs_allow_writes(struct super_block *s) {
2469 clear_bit(WRITERS_BLOCKED, &SB_JOURNAL(s)->j_state) ;
2470 wake_up(&SB_JOURNAL(s)->j_join_wait) ;
2473 /* this must be called without a transaction started, and does not
2476 void reiserfs_wait_on_write_block(struct super_block *s) {
2477 wait_event(SB_JOURNAL(s)->j_join_wait,
2478 !test_bit(WRITERS_BLOCKED, &SB_JOURNAL(s)->j_state)) ;
2481 static void queue_log_writer(struct super_block *s) {
2483 set_bit(WRITERS_QUEUED, &SB_JOURNAL(s)->j_state);
2486 * we don't want to use wait_event here because
2487 * we only want to wait once.
2489 init_waitqueue_entry(&wait, current);
2490 add_wait_queue(&SB_JOURNAL(s)->j_join_wait, &wait);
2491 set_current_state(TASK_UNINTERRUPTIBLE);
2492 if (test_bit(WRITERS_QUEUED, &SB_JOURNAL(s)->j_state))
2494 current->state = TASK_RUNNING;
2495 remove_wait_queue(&SB_JOURNAL(s)->j_join_wait, &wait);
2498 static void wake_queued_writers(struct super_block *s) {
2499 if (test_and_clear_bit(WRITERS_QUEUED, &SB_JOURNAL(s)->j_state))
2500 wake_up(&SB_JOURNAL(s)->j_join_wait);
2503 static void let_transaction_grow(struct super_block *sb,
2504 unsigned long trans_id)
2506 unsigned long bcount = SB_JOURNAL(sb)->j_bcount;
2508 set_current_state(TASK_UNINTERRUPTIBLE);
2509 schedule_timeout(1);
2510 SB_JOURNAL(sb)->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2511 while ((atomic_read(&SB_JOURNAL(sb)->j_wcount) > 0 ||
2512 atomic_read(&SB_JOURNAL(sb)->j_jlock)) &&
2513 SB_JOURNAL(sb)->j_trans_id == trans_id) {
2514 queue_log_writer(sb);
2516 if (SB_JOURNAL(sb)->j_trans_id != trans_id)
2518 if (bcount == SB_JOURNAL(sb)->j_bcount)
2520 bcount = SB_JOURNAL(sb)->j_bcount;
2524 /* join == true if you must join an existing transaction.
2525 ** join == false if you can deal with waiting for others to finish
2527 ** this will block until the transaction is joinable. send the number of blocks you
2528 ** expect to use in nblocks.
2530 static int do_journal_begin_r(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb,unsigned long nblocks,int join) {
2531 time_t now = get_seconds() ;
2533 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2534 struct reiserfs_transaction_handle myth;
2535 int sched_count = 0;
2537 reiserfs_check_lock_depth(p_s_sb, "journal_begin") ;
2538 RFALSE( p_s_sb->s_flags & MS_RDONLY,
2539 "clm-2078: calling journal_begin on readonly FS") ;
2541 PROC_INFO_INC( p_s_sb, journal.journal_being );
2542 /* set here for journal_join */
2544 th->t_super = p_s_sb ;
2547 lock_journal(p_s_sb) ;
2548 journal->j_bcount++;
2550 if (test_bit(WRITERS_BLOCKED, &journal->j_state)) {
2551 unlock_journal(p_s_sb) ;
2552 reiserfs_wait_on_write_block(p_s_sb) ;
2553 PROC_INFO_INC( p_s_sb, journal.journal_relock_writers );
2556 now = get_seconds();
2558 /* if there is no room in the journal OR
2559 ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
2560 ** we don't sleep if there aren't other writers
2563 if ( (!join && journal->j_must_wait > 0) ||
2564 ( !join && (journal->j_len_alloc + nblocks + 2) >= SB_JOURNAL_MAX_BATCH(p_s_sb)) ||
2565 (!join && atomic_read(&journal->j_wcount) > 0 && journal->j_trans_start_time > 0 &&
2566 (now - journal->j_trans_start_time) > SB_JOURNAL_MAX_TRANS_AGE(p_s_sb)) ||
2567 (!join && atomic_read(&journal->j_jlock)) ||
2568 (!join && journal->j_cnode_free < (SB_JOURNAL_TRANS_MAX(p_s_sb) * 3))) {
2570 old_trans_id = journal->j_trans_id;
2571 unlock_journal(p_s_sb) ; /* allow others to finish this transaction */
2573 if (!join && (journal->j_len_alloc + nblocks + 2) >=
2574 SB_JOURNAL_MAX_BATCH(p_s_sb) &&
2575 ((journal->j_len + nblocks + 2) * 100) < (journal->j_len_alloc * 75))
2577 if (atomic_read(&journal->j_wcount) > 10) {
2579 queue_log_writer(p_s_sb);
2583 /* don't mess with joining the transaction if all we have to do is
2584 * wait for someone else to do a commit
2586 if (atomic_read(&journal->j_jlock)) {
2587 while (journal->j_trans_id == old_trans_id &&
2588 atomic_read(&journal->j_jlock)) {
2589 queue_log_writer(p_s_sb);
2593 journal_join(&myth, p_s_sb, 1) ;
2595 /* someone might have ended the transaction while we joined */
2596 if (old_trans_id != SB_JOURNAL(p_s_sb)->j_trans_id) {
2597 do_journal_end(&myth, p_s_sb, 1, 0) ;
2599 do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW) ;
2602 PROC_INFO_INC( p_s_sb, journal.journal_relock_wcount );
2605 /* we are the first writer, set trans_id */
2606 if (journal->j_trans_start_time == 0) {
2607 journal->j_trans_start_time = get_seconds();
2609 atomic_inc(&(journal->j_wcount)) ;
2610 journal->j_len_alloc += nblocks ;
2611 th->t_blocks_logged = 0 ;
2612 th->t_blocks_allocated = nblocks ;
2613 th->t_trans_id = journal->j_trans_id ;
2614 unlock_journal(p_s_sb) ;
2618 struct reiserfs_transaction_handle *
2619 reiserfs_persistent_transaction(struct super_block *s, int nblocks) {
2621 struct reiserfs_transaction_handle *th ;
2623 /* if we're nesting into an existing transaction. It will be
2624 ** persistent on its own
2626 if (reiserfs_transaction_running(s)) {
2627 th = current->journal_info ;
2629 if (th->t_refcount < 2) {
2634 th = reiserfs_kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS, s) ;
2637 ret = journal_begin(th, s, nblocks) ;
2639 reiserfs_kfree(th, sizeof(struct reiserfs_transaction_handle), s) ;
2646 reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th) {
2647 struct super_block *s = th->t_super;
2649 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
2650 if (th->t_refcount == 0)
2651 reiserfs_kfree(th, sizeof(struct reiserfs_transaction_handle), s) ;
2655 static int journal_join(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) {
2656 struct reiserfs_transaction_handle *cur_th = current->journal_info;
2658 /* this keeps do_journal_end from NULLing out the current->journal_info
2661 th->t_handle_save = cur_th ;
2662 if (cur_th && cur_th->t_refcount > 1) {
2665 return do_journal_begin_r(th, p_s_sb, nblocks, 1) ;
2668 int journal_begin(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb, unsigned long nblocks) {
2669 struct reiserfs_transaction_handle *cur_th = current->journal_info ;
2672 th->t_handle_save = NULL ;
2674 /* we are nesting into the current transaction */
2675 if (cur_th->t_super == p_s_sb) {
2676 cur_th->t_refcount++ ;
2677 memcpy(th, cur_th, sizeof(*th));
2678 if (th->t_refcount <= 1)
2679 reiserfs_warning (p_s_sb, "BAD: refcount <= 1, but journal_info != 0");
2682 /* we've ended up with a handle from a different filesystem.
2683 ** save it and restore on journal_end. This should never
2686 reiserfs_warning(p_s_sb, "clm-2100: nesting info a different FS") ;
2687 th->t_handle_save = current->journal_info ;
2688 current->journal_info = th;
2691 current->journal_info = th;
2693 ret = do_journal_begin_r(th, p_s_sb, nblocks, 0) ;
2694 if (current->journal_info != th)
2700 ** puts bh into the current transaction. If it was already there, reorders removes the
2701 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
2703 ** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the
2704 ** transaction is committed.
2706 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
2708 int journal_mark_dirty(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, struct buffer_head *bh) {
2709 struct reiserfs_journal_cnode *cn = NULL;
2710 int count_already_incd = 0 ;
2713 PROC_INFO_INC( p_s_sb, journal.mark_dirty );
2714 if (th->t_trans_id != SB_JOURNAL(p_s_sb)->j_trans_id) {
2715 reiserfs_panic(th->t_super, "journal-1577: handle trans id %ld != current trans id %ld\n",
2716 th->t_trans_id, SB_JOURNAL(p_s_sb)->j_trans_id);
2720 prepared = test_and_clear_bit(BH_JPrepared, &bh->b_state) ;
2721 clear_bit(BH_JRestore_dirty, &bh->b_state);
2722 /* already in this transaction, we are done */
2723 if (buffer_journaled(bh)) {
2724 PROC_INFO_INC( p_s_sb, journal.mark_dirty_already );
2728 /* this must be turned into a panic instead of a warning. We can't allow
2729 ** a dirty or journal_dirty or locked buffer to be logged, as some changes
2730 ** could get to disk too early. NOT GOOD.
2732 if (!prepared || buffer_dirty(bh)) {
2733 reiserfs_warning (p_s_sb, "journal-1777: buffer %llu bad state "
2734 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
2735 (unsigned long long)bh->b_blocknr, prepared ? ' ' : '!',
2736 buffer_locked(bh) ? ' ' : '!',
2737 buffer_dirty(bh) ? ' ' : '!',
2738 buffer_journal_dirty(bh) ? ' ' : '!') ;
2741 if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) <= 0) {
2742 reiserfs_warning (p_s_sb, "journal-1409: journal_mark_dirty returning because j_wcount was %d", atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount))) ;
2745 /* this error means I've screwed up, and we've overflowed the transaction.
2746 ** Nothing can be done here, except make the FS readonly or panic.
2748 if (SB_JOURNAL(p_s_sb)->j_len >= SB_JOURNAL_TRANS_MAX(p_s_sb)) {
2749 reiserfs_panic(th->t_super, "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n", SB_JOURNAL(p_s_sb)->j_len) ;
2752 if (buffer_journal_dirty(bh)) {
2753 count_already_incd = 1 ;
2754 PROC_INFO_INC( p_s_sb, journal.mark_dirty_notjournal );
2755 mark_buffer_notjournal_dirty(bh) ;
2758 if (SB_JOURNAL(p_s_sb)->j_len > SB_JOURNAL(p_s_sb)->j_len_alloc) {
2759 SB_JOURNAL(p_s_sb)->j_len_alloc = SB_JOURNAL(p_s_sb)->j_len + JOURNAL_PER_BALANCE_CNT ;
2762 set_bit(BH_JDirty, &bh->b_state) ;
2764 /* now put this guy on the end */
2766 cn = get_cnode(p_s_sb) ;
2768 reiserfs_panic(p_s_sb, "get_cnode failed!\n");
2771 if (th->t_blocks_logged == th->t_blocks_allocated) {
2772 th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT ;
2773 SB_JOURNAL(p_s_sb)->j_len_alloc += JOURNAL_PER_BALANCE_CNT ;
2775 th->t_blocks_logged++ ;
2776 SB_JOURNAL(p_s_sb)->j_len++ ;
2779 cn->blocknr = bh->b_blocknr ;
2782 insert_journal_hash(SB_JOURNAL(p_s_sb)->j_hash_table, cn) ;
2783 if (!count_already_incd) {
2788 cn->prev = SB_JOURNAL(p_s_sb)->j_last ;
2790 if (SB_JOURNAL(p_s_sb)->j_last) {
2791 SB_JOURNAL(p_s_sb)->j_last->next = cn ;
2792 SB_JOURNAL(p_s_sb)->j_last = cn ;
2794 SB_JOURNAL(p_s_sb)->j_first = cn ;
2795 SB_JOURNAL(p_s_sb)->j_last = cn ;
2800 int journal_end(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) {
2801 if (!current->journal_info && th->t_refcount > 1)
2802 reiserfs_warning (p_s_sb, "REISER-NESTING: th NULL, refcount %d",
2806 if (th->t_refcount > 0) {
2807 struct reiserfs_transaction_handle *cur_th = current->journal_info ;
2809 /* we aren't allowed to close a nested transaction on a different
2810 ** filesystem from the one in the task struct
2812 if (cur_th->t_super != th->t_super)
2816 memcpy(current->journal_info, th, sizeof(*th));
2821 return do_journal_end(th, p_s_sb, nblocks, 0) ;
2825 /* removes from the current transaction, relsing and descrementing any counters.
2826 ** also files the removed buffer directly onto the clean list
2828 ** called by journal_mark_freed when a block has been deleted
2830 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
2832 static int remove_from_transaction(struct super_block *p_s_sb, b_blocknr_t blocknr, int already_cleaned) {
2833 struct buffer_head *bh ;
2834 struct reiserfs_journal_cnode *cn ;
2837 cn = get_journal_hash_dev(p_s_sb, SB_JOURNAL(p_s_sb)->j_hash_table, blocknr) ;
2838 if (!cn || !cn->bh) {
2843 cn->prev->next = cn->next ;
2846 cn->next->prev = cn->prev ;
2848 if (cn == SB_JOURNAL(p_s_sb)->j_first) {
2849 SB_JOURNAL(p_s_sb)->j_first = cn->next ;
2851 if (cn == SB_JOURNAL(p_s_sb)->j_last) {
2852 SB_JOURNAL(p_s_sb)->j_last = cn->prev ;
2855 remove_journal_hash(p_s_sb, SB_JOURNAL(p_s_sb)->j_hash_table, NULL, bh->b_blocknr, 0) ;
2856 mark_buffer_not_journaled(bh) ; /* don't log this one */
2858 if (!already_cleaned) {
2859 mark_buffer_notjournal_dirty(bh) ;
2861 if (atomic_read(&(bh->b_count)) < 0) {
2862 reiserfs_warning (p_s_sb, "journal-1752: remove from trans, b_count < 0");
2866 SB_JOURNAL(p_s_sb)->j_len-- ;
2867 SB_JOURNAL(p_s_sb)->j_len_alloc-- ;
2868 free_cnode(p_s_sb, cn) ;
2873 ** for any cnode in a journal list, it can only be dirtied of all the
2874 ** transactions that include it are commited to disk.
2875 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
2876 ** and 0 if you aren't
2878 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
2879 ** blocks for a given transaction on disk
2882 static int can_dirty(struct reiserfs_journal_cnode *cn) {
2883 struct super_block *sb = cn->sb;
2884 b_blocknr_t blocknr = cn->blocknr ;
2885 struct reiserfs_journal_cnode *cur = cn->hprev ;
2888 /* first test hprev. These are all newer than cn, so any node here
2889 ** with the same block number and dev means this node can't be sent
2890 ** to disk right now.
2892 while(cur && can_dirty) {
2893 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
2894 cur->blocknr == blocknr) {
2899 /* then test hnext. These are all older than cn. As long as they
2900 ** are committed to the log, it is safe to write cn to disk
2903 while(cur && can_dirty) {
2904 if (cur->jlist && cur->jlist->j_len > 0 &&
2905 atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
2906 cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
2914 /* syncs the commit blocks, but does not force the real buffers to disk
2915 ** will wait until the current transaction is done/commited before returning
2917 int journal_end_sync(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) {
2919 /* you can sync while nested, very, very bad */
2920 if (th->t_refcount > 1) {
2923 if (SB_JOURNAL(p_s_sb)->j_len == 0) {
2924 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
2925 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
2927 return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT) ;
2931 ** writeback the pending async commits to disk
2933 static void flush_async_commits(void *p) {
2934 struct super_block *p_s_sb = p;
2935 struct reiserfs_journal_list *jl;
2936 struct list_head *entry;
2939 if (!list_empty(&SB_JOURNAL(p_s_sb)->j_journal_list)) {
2940 /* last entry is the youngest, commit it and you get everything */
2941 entry = SB_JOURNAL(p_s_sb)->j_journal_list.prev;
2942 jl = JOURNAL_LIST_ENTRY(entry);
2943 flush_commit_list(p_s_sb, jl, 1);
2947 * this is a little racey, but there's no harm in missing
2948 * the filemap_fdata_write
2950 if (!atomic_read(&SB_JOURNAL(p_s_sb)->j_async_throttle)) {
2951 atomic_inc(&SB_JOURNAL(p_s_sb)->j_async_throttle);
2952 filemap_fdatawrite(p_s_sb->s_bdev->bd_inode->i_mapping);
2953 atomic_dec(&SB_JOURNAL(p_s_sb)->j_async_throttle);
2958 ** flushes any old transactions to disk
2959 ** ends the current transaction if it is too old
2961 int reiserfs_flush_old_commits(struct super_block *p_s_sb) {
2963 struct reiserfs_transaction_handle th ;
2965 now = get_seconds();
2966 /* safety check so we don't flush while we are replaying the log during
2969 if (list_empty(&SB_JOURNAL(p_s_sb)->j_journal_list)) {
2973 /* check the current transaction. If there are no writers, and it is
2974 * too old, finish it, and force the commit blocks to disk
2976 if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) <= 0 &&
2977 SB_JOURNAL(p_s_sb)->j_trans_start_time > 0 &&
2978 SB_JOURNAL(p_s_sb)->j_len > 0 &&
2979 (now - SB_JOURNAL(p_s_sb)->j_trans_start_time) >
2980 SB_JOURNAL_MAX_TRANS_AGE(p_s_sb))
2982 journal_join(&th, p_s_sb, 1) ;
2983 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
2984 journal_mark_dirty(&th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
2986 /* we're only being called from kreiserfsd, it makes no sense to do
2987 ** an async commit so that kreiserfsd can do it later
2989 do_journal_end(&th, p_s_sb,1, COMMIT_NOW | WAIT) ;
2991 return p_s_sb->s_dirt;
2995 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
2997 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
2998 ** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just
2999 ** flushes the commit list and returns 0.
3001 ** Won't batch when flush or commit_now is set. Also won't batch when others are waiting on j_join_wait.
3003 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3005 static int check_journal_end(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb,
3006 unsigned long nblocks, int flags) {
3009 int flush = flags & FLUSH_ALL ;
3010 int commit_now = flags & COMMIT_NOW ;
3011 int wait_on_commit = flags & WAIT ;
3012 struct reiserfs_journal_list *jl;
3014 if (th->t_trans_id != SB_JOURNAL(p_s_sb)->j_trans_id) {
3015 reiserfs_panic(th->t_super, "journal-1577: handle trans id %ld != current trans id %ld\n",
3016 th->t_trans_id, SB_JOURNAL(p_s_sb)->j_trans_id);
3019 SB_JOURNAL(p_s_sb)->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged) ;
3020 if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) > 0) { /* <= 0 is allowed. unmounting might not call begin */
3021 atomic_dec(&(SB_JOURNAL(p_s_sb)->j_wcount)) ;
3024 /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3025 ** will be dealt with by next transaction that actually writes something, but should be taken
3026 ** care of in this trans
3028 if (SB_JOURNAL(p_s_sb)->j_len == 0) {
3031 /* if wcount > 0, and we are called to with flush or commit_now,
3032 ** we wait on j_join_wait. We will wake up when the last writer has
3033 ** finished the transaction, and started it on its way to the disk.
3034 ** Then, we flush the commit or journal list, and just return 0
3035 ** because the rest of journal end was already done for this transaction.
3037 if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) > 0) {
3038 if (flush || commit_now) {
3041 jl = SB_JOURNAL(p_s_sb)->j_current_jl;
3042 trans_id = jl->j_trans_id;
3044 jl->j_state |= LIST_COMMIT_PENDING;
3045 atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 1) ;
3047 SB_JOURNAL(p_s_sb)->j_next_full_flush = 1 ;
3049 unlock_journal(p_s_sb) ;
3051 /* sleep while the current transaction is still j_jlocked */
3052 while(SB_JOURNAL(p_s_sb)->j_trans_id == trans_id) {
3053 if (atomic_read(&SB_JOURNAL(p_s_sb)->j_jlock)) {
3054 queue_log_writer(p_s_sb);
3056 lock_journal(p_s_sb);
3057 if (SB_JOURNAL(p_s_sb)->j_trans_id == trans_id) {
3058 atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 1) ;
3060 unlock_journal(p_s_sb);
3063 if (SB_JOURNAL(p_s_sb)->j_trans_id == trans_id) {
3066 if (commit_now && journal_list_still_alive(p_s_sb, trans_id) &&
3069 flush_commit_list(p_s_sb, jl, 1) ;
3073 unlock_journal(p_s_sb) ;
3077 /* deal with old transactions where we are the last writers */
3078 now = get_seconds();
3079 if ((now - SB_JOURNAL(p_s_sb)->j_trans_start_time) > SB_JOURNAL_MAX_TRANS_AGE(p_s_sb)) {
3081 SB_JOURNAL(p_s_sb)->j_next_async_flush = 1 ;
3083 /* don't batch when someone is waiting on j_join_wait */
3084 /* don't batch when syncing the commit or flushing the whole trans */
3085 if (!(SB_JOURNAL(p_s_sb)->j_must_wait > 0) && !(atomic_read(&(SB_JOURNAL(p_s_sb)->j_jlock))) && !flush && !commit_now &&
3086 (SB_JOURNAL(p_s_sb)->j_len < SB_JOURNAL_MAX_BATCH(p_s_sb)) &&
3087 SB_JOURNAL(p_s_sb)->j_len_alloc < SB_JOURNAL_MAX_BATCH(p_s_sb) && SB_JOURNAL(p_s_sb)->j_cnode_free > (SB_JOURNAL_TRANS_MAX(p_s_sb) * 3)) {
3088 SB_JOURNAL(p_s_sb)->j_bcount++ ;
3089 unlock_journal(p_s_sb) ;
3093 if (SB_JOURNAL(p_s_sb)->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
3094 reiserfs_panic(p_s_sb, "journal-003: journal_end: j_start (%ld) is too high\n", SB_JOURNAL(p_s_sb)->j_start) ;
3100 ** Does all the work that makes deleting blocks safe.
3101 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3104 ** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes
3105 ** before this transaction has finished.
3107 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with
3108 ** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash,
3109 ** the block can't be reallocated yet.
3111 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3113 int journal_mark_freed(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, b_blocknr_t blocknr) {
3114 struct reiserfs_journal_cnode *cn = NULL ;
3115 struct buffer_head *bh = NULL ;
3116 struct reiserfs_list_bitmap *jb = NULL ;
3119 cn = get_journal_hash_dev(p_s_sb, SB_JOURNAL(p_s_sb)->j_hash_table, blocknr);
3124 /* if it is journal new, we just remove it from this transaction */
3125 if (bh && buffer_journal_new(bh)) {
3126 mark_buffer_notjournal_new(bh) ;
3127 clear_prepared_bits(bh) ;
3128 reiserfs_clean_and_file_buffer(bh) ;
3129 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned) ;
3131 /* set the bit for this block in the journal bitmap for this transaction */
3132 jb = SB_JOURNAL(p_s_sb)->j_current_jl->j_list_bitmap;
3134 reiserfs_panic(p_s_sb, "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n") ;
3136 set_bit_in_list_bitmap(p_s_sb, blocknr, jb) ;
3138 /* Note, the entire while loop is not allowed to schedule. */
3141 clear_prepared_bits(bh) ;
3142 reiserfs_clean_and_file_buffer(bh) ;
3144 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned) ;
3146 /* find all older transactions with this block, make sure they don't try to write it out */
3147 cn = get_journal_hash_dev(p_s_sb,SB_JOURNAL(p_s_sb)->j_list_hash_table, blocknr) ;
3149 if (p_s_sb == cn->sb && blocknr == cn->blocknr) {
3150 set_bit(BLOCK_FREED, &cn->state) ;
3153 /* remove_from_transaction will brelse the buffer if it was
3154 ** in the current trans
3156 mark_buffer_notjournal_dirty(cn->bh) ;
3159 if (atomic_read(&(cn->bh->b_count)) < 0) {
3160 reiserfs_warning (p_s_sb, "journal-2138: cn->bh->b_count < 0");
3163 if (cn->jlist) { /* since we are clearing the bh, we MUST dec nonzerolen */
3164 atomic_dec(&(cn->jlist->j_nonzerolen)) ;
3174 put_bh(bh) ; /* get_hash grabs the buffer */
3175 if (atomic_read(&(bh->b_count)) < 0) {
3176 reiserfs_warning (p_s_sb, "journal-2165: bh->b_count < 0");
3182 void reiserfs_update_inode_transaction(struct inode *inode) {
3183 REISERFS_I(inode)->i_jl = SB_JOURNAL(inode->i_sb)->j_current_jl;
3184 REISERFS_I(inode)->i_trans_id = SB_JOURNAL(inode->i_sb)->j_trans_id ;
3187 static void __commit_trans_jl(struct inode *inode, unsigned long id,
3188 struct reiserfs_journal_list *jl)
3190 struct reiserfs_transaction_handle th ;
3191 struct super_block *sb = inode->i_sb ;
3193 /* is it from the current transaction, or from an unknown transaction? */
3194 if (id == SB_JOURNAL(sb)->j_trans_id) {
3195 jl = SB_JOURNAL(sb)->j_current_jl;
3196 /* try to let other writers come in and grow this transaction */
3197 let_transaction_grow(sb, id);
3198 if (SB_JOURNAL(sb)->j_trans_id != id) {
3199 goto flush_commit_only;
3202 journal_begin(&th, sb, 1) ;
3204 /* someone might have ended this transaction while we joined */
3205 if (SB_JOURNAL(sb)->j_trans_id != id) {
3206 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 1) ;
3207 journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb)) ;
3208 journal_end(&th, sb, 1) ;
3209 goto flush_commit_only;
3212 journal_end_sync(&th, sb, 1) ;
3215 /* this gets tricky, we have to make sure the journal list in
3216 * the inode still exists. We know the list is still around
3217 * if we've got a larger transaction id than the oldest list
3220 if (journal_list_still_alive(inode->i_sb, id)) {
3221 flush_commit_list(sb, jl, 1) ;
3224 /* otherwise the list is gone, and long since committed */
3227 void reiserfs_commit_for_inode(struct inode *inode) {
3228 unsigned long id = REISERFS_I(inode)->i_trans_id;
3229 struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3231 /* for the whole inode, assume unset id means it was
3232 * changed in the current transaction. More conservative
3235 reiserfs_update_inode_transaction(inode) ;
3236 id = REISERFS_I(inode)->i_trans_id;
3237 /* jl will be updated in __commit_trans_jl */
3240 __commit_trans_jl(inode, id, jl);
3243 void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb,
3244 struct buffer_head *bh) {
3245 PROC_INFO_INC( p_s_sb, journal.restore_prepared );
3249 if (test_and_clear_bit(BH_JRestore_dirty, &bh->b_state) &&
3250 buffer_journal_dirty(bh)) {
3251 struct reiserfs_journal_cnode *cn;
3252 cn = get_journal_hash_dev(p_s_sb,
3253 SB_JOURNAL(p_s_sb)->j_list_hash_table,
3255 if (cn && can_dirty(cn)) {
3256 set_bit(BH_JTest, &bh->b_state);
3257 mark_buffer_dirty(bh);
3260 clear_bit(BH_JPrepared, &bh->b_state) ;
3263 extern struct tree_balance *cur_tb ;
3265 ** before we can change a metadata block, we have to make sure it won't
3266 ** be written to disk while we are altering it. So, we must:
3271 int reiserfs_prepare_for_journal(struct super_block *p_s_sb,
3272 struct buffer_head *bh, int wait) {
3273 PROC_INFO_INC( p_s_sb, journal.prepare );
3275 if (test_set_buffer_locked(bh)) {
3280 set_bit(BH_JPrepared, &bh->b_state);
3281 if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3282 clear_bit(BH_JTest, &bh->b_state);
3283 set_bit(BH_JRestore_dirty, &bh->b_state);
3289 static void flush_old_journal_lists(struct super_block *s) {
3290 struct reiserfs_journal_list *jl;
3291 struct list_head *entry;
3292 time_t now = get_seconds();
3294 while(!list_empty(&SB_JOURNAL(s)->j_journal_list)) {
3295 entry = SB_JOURNAL(s)->j_journal_list.next;
3296 jl = JOURNAL_LIST_ENTRY(entry);
3297 /* this check should always be run, to send old lists to disk */
3298 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4))) {
3299 flush_used_journal_lists(s, jl);
3307 ** long and ugly. If flush, will not return until all commit
3308 ** blocks and all real buffers in the trans are on disk.
3309 ** If no_async, won't return until all commit blocks are on disk.
3311 ** keep reading, there are comments as you go along
3313 static int do_journal_end(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb, unsigned long nblocks,
3315 struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3316 struct reiserfs_journal_cnode *last_cn = NULL;
3317 struct reiserfs_journal_desc *desc ;
3318 struct reiserfs_journal_commit *commit ;
3319 struct buffer_head *c_bh ; /* commit bh */
3320 struct buffer_head *d_bh ; /* desc bh */
3321 int cur_write_start = 0 ; /* start index of current log write */
3324 int flush = flags & FLUSH_ALL ;
3325 int wait_on_commit = flags & WAIT ;
3326 struct reiserfs_journal_list *jl, *temp_jl;
3327 struct list_head *entry, *safe;
3328 unsigned long jindex;
3329 unsigned long commit_trans_id;
3332 if (th->t_refcount > 1)
3335 current->journal_info = th->t_handle_save;
3336 reiserfs_check_lock_depth(p_s_sb, "journal end");
3337 if (SB_JOURNAL(p_s_sb)->j_len == 0) {
3338 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
3339 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
3342 lock_journal(p_s_sb) ;
3343 if (SB_JOURNAL(p_s_sb)->j_next_full_flush) {
3344 flags |= FLUSH_ALL ;
3347 if (SB_JOURNAL(p_s_sb)->j_next_async_flush) {
3348 flags |= COMMIT_NOW | WAIT;
3352 /* check_journal_end locks the journal, and unlocks if it does not return 1
3353 ** it tells us if we should continue with the journal_end, or just return
3355 if (!check_journal_end(th, p_s_sb, nblocks, flags)) {
3357 wake_queued_writers(p_s_sb);
3358 reiserfs_async_progress_wait(p_s_sb);
3362 /* check_journal_end might set these, check again */
3363 if (SB_JOURNAL(p_s_sb)->j_next_full_flush) {
3368 ** j must wait means we have to flush the log blocks, and the real blocks for
3371 if (SB_JOURNAL(p_s_sb)->j_must_wait > 0) {
3375 #ifdef REISERFS_PREALLOCATE
3376 /* quota ops might need to nest, setup the journal_info pointer for them */
3377 current->journal_info = th ;
3378 reiserfs_discard_all_prealloc(th); /* it should not involve new blocks into
3379 * the transaction */
3380 current->journal_info = th->t_handle_save ;
3383 /* setup description block */
3384 d_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_JOURNAL(p_s_sb)->j_start) ;
3385 set_buffer_uptodate(d_bh);
3386 desc = (struct reiserfs_journal_desc *)(d_bh)->b_data ;
3387 memset(d_bh->b_data, 0, d_bh->b_size) ;
3388 memcpy(get_journal_desc_magic (d_bh), JOURNAL_DESC_MAGIC, 8) ;
3389 set_desc_trans_id(desc, SB_JOURNAL(p_s_sb)->j_trans_id) ;
3391 /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */
3392 c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3393 ((SB_JOURNAL(p_s_sb)->j_start + SB_JOURNAL(p_s_sb)->j_len + 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ;
3394 commit = (struct reiserfs_journal_commit *)c_bh->b_data ;
3395 memset(c_bh->b_data, 0, c_bh->b_size) ;
3396 set_commit_trans_id(commit, SB_JOURNAL(p_s_sb)->j_trans_id) ;
3397 set_buffer_uptodate(c_bh) ;
3399 /* init this journal list */
3400 jl = SB_JOURNAL(p_s_sb)->j_current_jl;
3402 /* we lock the commit before doing anything because
3403 * we want to make sure nobody tries to run flush_commit_list until
3404 * the new transaction is fully setup, and we've already flushed the
3407 down(&jl->j_commit_lock);
3409 /* save the transaction id in case we need to commit it later */
3410 commit_trans_id = jl->j_trans_id;
3412 atomic_set(&jl->j_older_commits_done, 0) ;
3413 jl->j_trans_id = SB_JOURNAL(p_s_sb)->j_trans_id ;
3414 jl->j_timestamp = SB_JOURNAL(p_s_sb)->j_trans_start_time ;
3415 jl->j_commit_bh = c_bh ;
3416 jl->j_start = SB_JOURNAL(p_s_sb)->j_start ;
3417 jl->j_len = SB_JOURNAL(p_s_sb)->j_len ;
3418 atomic_set(&jl->j_nonzerolen, SB_JOURNAL(p_s_sb)->j_len) ;
3419 atomic_set(&jl->j_commit_left, SB_JOURNAL(p_s_sb)->j_len + 2);
3420 jl->j_realblock = NULL ;
3422 /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
3423 ** for each real block, add it to the journal list hash,
3424 ** copy into real block index array in the commit or desc block
3426 trans_half = journal_trans_half(p_s_sb->s_blocksize);
3427 for (i = 0, cn = SB_JOURNAL(p_s_sb)->j_first ; cn ; cn = cn->next, i++) {
3428 if (test_bit(BH_JDirty, &cn->bh->b_state) ) {
3429 jl_cn = get_cnode(p_s_sb) ;
3431 reiserfs_panic(p_s_sb, "journal-1676, get_cnode returned NULL\n") ;
3434 jl->j_realblock = jl_cn ;
3436 jl_cn->prev = last_cn ;
3437 jl_cn->next = NULL ;
3439 last_cn->next = jl_cn ;
3442 /* make sure the block we are trying to log is not a block
3443 of journal or reserved area */
3445 if (is_block_in_log_or_reserved_area(p_s_sb, cn->bh->b_blocknr)) {
3446 reiserfs_panic(p_s_sb, "journal-2332: Trying to log block %lu, which is a log block\n", cn->bh->b_blocknr) ;
3448 jl_cn->blocknr = cn->bh->b_blocknr ;
3451 jl_cn->bh = cn->bh ;
3453 insert_journal_hash(SB_JOURNAL(p_s_sb)->j_list_hash_table, jl_cn) ;
3454 if (i < trans_half) {
3455 desc->j_realblock[i] = cpu_to_le32(cn->bh->b_blocknr) ;
3457 commit->j_realblock[i - trans_half] = cpu_to_le32(cn->bh->b_blocknr) ;
3463 set_desc_trans_len(desc, SB_JOURNAL(p_s_sb)->j_len) ;
3464 set_desc_mount_id(desc, SB_JOURNAL(p_s_sb)->j_mount_id) ;
3465 set_desc_trans_id(desc, SB_JOURNAL(p_s_sb)->j_trans_id) ;
3466 set_commit_trans_len(commit, SB_JOURNAL(p_s_sb)->j_len);
3468 /* special check in case all buffers in the journal were marked for not logging */
3469 if (SB_JOURNAL(p_s_sb)->j_len == 0) {
3473 /* we're about to dirty all the log blocks, mark the description block
3474 * dirty now too. Don't mark the commit block dirty until all the
3475 * others are on disk
3477 mark_buffer_dirty(d_bh);
3479 /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
3480 cur_write_start = SB_JOURNAL(p_s_sb)->j_start ;
3481 cn = SB_JOURNAL(p_s_sb)->j_first ;
3482 jindex = 1 ; /* start at one so we don't get the desc again */
3484 clear_bit(BH_JNew, &(cn->bh->b_state)) ;
3485 /* copy all the real blocks into log area. dirty log blocks */
3486 if (test_bit(BH_JDirty, &cn->bh->b_state)) {
3487 struct buffer_head *tmp_bh ;
3490 tmp_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3491 ((cur_write_start + jindex) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ;
3492 set_buffer_uptodate(tmp_bh);
3493 page = cn->bh->b_page;
3495 memcpy(tmp_bh->b_data, addr + offset_in_page(cn->bh->b_data),
3498 mark_buffer_dirty(tmp_bh);
3500 set_bit(BH_JDirty_wait, &(cn->bh->b_state)) ;
3501 clear_bit(BH_JDirty, &(cn->bh->b_state)) ;
3503 /* JDirty cleared sometime during transaction. don't log this one */
3504 reiserfs_warning(p_s_sb, "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!") ;
3508 free_cnode(p_s_sb, cn) ;
3513 /* we are done with both the c_bh and d_bh, but
3514 ** c_bh must be written after all other commit blocks,
3515 ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
3518 SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
3520 /* now it is safe to insert this transaction on the main list */
3521 list_add_tail(&jl->j_list, &SB_JOURNAL(p_s_sb)->j_journal_list);
3522 list_add_tail(&jl->j_working_list, &SB_JOURNAL(p_s_sb)->j_working_list);
3523 SB_JOURNAL(p_s_sb)->j_num_work_lists++;
3525 /* reset journal values for the next transaction */
3526 old_start = SB_JOURNAL(p_s_sb)->j_start ;
3527 SB_JOURNAL(p_s_sb)->j_start = (SB_JOURNAL(p_s_sb)->j_start + SB_JOURNAL(p_s_sb)->j_len + 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb);
3528 atomic_set(&(SB_JOURNAL(p_s_sb)->j_wcount), 0) ;
3529 SB_JOURNAL(p_s_sb)->j_bcount = 0 ;
3530 SB_JOURNAL(p_s_sb)->j_last = NULL ;
3531 SB_JOURNAL(p_s_sb)->j_first = NULL ;
3532 SB_JOURNAL(p_s_sb)->j_len = 0 ;
3533 SB_JOURNAL(p_s_sb)->j_trans_start_time = 0 ;
3534 SB_JOURNAL(p_s_sb)->j_trans_id++ ;
3535 SB_JOURNAL(p_s_sb)->j_current_jl->j_trans_id = SB_JOURNAL(p_s_sb)->j_trans_id;
3536 SB_JOURNAL(p_s_sb)->j_must_wait = 0 ;
3537 SB_JOURNAL(p_s_sb)->j_len_alloc = 0 ;
3538 SB_JOURNAL(p_s_sb)->j_next_full_flush = 0 ;
3539 SB_JOURNAL(p_s_sb)->j_next_async_flush = 0 ;
3540 init_journal_hash(p_s_sb) ;
3542 // make sure reiserfs_add_jh sees the new current_jl before we
3543 // write out the tails
3546 /* tail conversion targets have to hit the disk before we end the
3547 * transaction. Otherwise a later transaction might repack the tail
3548 * before this transaction commits, leaving the data block unflushed and
3549 * clean, if we crash before the later transaction commits, the data block
3552 if (!list_empty(&jl->j_tail_bh_list)) {
3554 write_ordered_buffers(&SB_JOURNAL(p_s_sb)->j_dirty_buffers_lock,
3555 SB_JOURNAL(p_s_sb), jl, &jl->j_tail_bh_list);
3558 if (!list_empty(&jl->j_tail_bh_list))
3560 up(&jl->j_commit_lock);
3562 /* honor the flush wishes from the caller, simple commits can
3563 ** be done outside the journal lock, they are done below
3565 ** if we don't flush the commit list right now, we put it into
3566 ** the work queue so the people waiting on the async progress work
3567 ** queue don't wait for this proc to flush journal lists and such.
3570 flush_commit_list(p_s_sb, jl, 1) ;
3571 flush_journal_list(p_s_sb, jl, 1) ;
3572 } else if (!(jl->j_state & LIST_COMMIT_PENDING))
3573 queue_delayed_work(commit_wq, &SB_JOURNAL(p_s_sb)->j_work, HZ/10);
3576 /* if the next transaction has any chance of wrapping, flush
3577 ** transactions that might get overwritten. If any journal lists are very
3578 ** old flush them as well.
3581 list_for_each_safe(entry, safe, &SB_JOURNAL(p_s_sb)->j_journal_list) {
3582 temp_jl = JOURNAL_LIST_ENTRY(entry);
3583 if (SB_JOURNAL(p_s_sb)->j_start <= temp_jl->j_start) {
3584 if ((SB_JOURNAL(p_s_sb)->j_start + SB_JOURNAL_TRANS_MAX(p_s_sb) + 1) >=
3587 flush_used_journal_lists(p_s_sb, temp_jl);
3589 } else if ((SB_JOURNAL(p_s_sb)->j_start +
3590 SB_JOURNAL_TRANS_MAX(p_s_sb) + 1) <
3591 SB_ONDISK_JOURNAL_SIZE(p_s_sb))
3593 /* if we don't cross into the next transaction and we don't
3594 * wrap, there is no way we can overlap any later transactions
3599 } else if ((SB_JOURNAL(p_s_sb)->j_start +
3600 SB_JOURNAL_TRANS_MAX(p_s_sb) + 1) >
3601 SB_ONDISK_JOURNAL_SIZE(p_s_sb))
3603 if (((SB_JOURNAL(p_s_sb)->j_start + SB_JOURNAL_TRANS_MAX(p_s_sb) + 1) %
3604 SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >= temp_jl->j_start)
3606 flush_used_journal_lists(p_s_sb, temp_jl);
3609 /* we don't overlap anything from out start to the end of the
3610 * log, and our wrapped portion doesn't overlap anything at
3611 * the start of the log. We can break
3617 flush_old_journal_lists(p_s_sb);
3619 SB_JOURNAL(p_s_sb)->j_current_jl->j_list_bitmap = get_list_bitmap(p_s_sb, SB_JOURNAL(p_s_sb)->j_current_jl) ;
3621 if (!(SB_JOURNAL(p_s_sb)->j_current_jl->j_list_bitmap)) {
3622 reiserfs_panic(p_s_sb, "journal-1996: do_journal_end, could not get a list bitmap\n") ;
3625 atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 0) ;
3626 unlock_journal(p_s_sb) ;
3627 /* wake up any body waiting to join. */
3628 clear_bit(WRITERS_QUEUED, &SB_JOURNAL(p_s_sb)->j_state);
3629 wake_up(&(SB_JOURNAL(p_s_sb)->j_join_wait)) ;
3631 if (!flush && wait_on_commit &&
3632 journal_list_still_alive(p_s_sb, commit_trans_id)) {
3633 flush_commit_list(p_s_sb, jl, 1) ;
3636 reiserfs_check_lock_depth(p_s_sb, "journal end2");