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("clm-2000, unable to allocate bitmaps for journal lists\n") ;
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("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("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(char *caller) {
427 if (current->lock_depth < 0) {
428 printk("%s called without kernel lock held\n", caller) ;
436 /* return a cnode with same dev, block number and size in table, or null if not found */
437 static inline struct reiserfs_journal_cnode *
438 get_journal_hash_dev(struct super_block *sb,
439 struct reiserfs_journal_cnode **table,
442 struct reiserfs_journal_cnode *cn ;
443 cn = journal_hash(table, sb, bl) ;
445 if (cn->blocknr == bl && cn->sb == sb)
449 return (struct reiserfs_journal_cnode *)0 ;
452 /* returns a cnode with same size, block number and dev as bh in the current transaction hash. NULL if not found */
453 static inline struct reiserfs_journal_cnode *get_journal_hash(struct super_block *p_s_sb, struct buffer_head *bh) {
454 struct reiserfs_journal_cnode *cn ;
456 cn = get_journal_hash_dev(p_s_sb, SB_JOURNAL(p_s_sb)->j_hash_table, bh->b_blocknr);
459 return (struct reiserfs_journal_cnode *)0 ;
465 ** this actually means 'can this block be reallocated yet?'. If you set search_all, a block can only be allocated
466 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
467 ** being overwritten by a replay after crashing.
469 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction. Since deleting
470 ** a block removes it from the current transaction, this case should never happen. If you don't set search_all, make
471 ** sure you never write the block without logging it.
473 ** next_zero_bit is a suggestion about the next block to try for find_forward.
474 ** when bl is rejected because it is set in a journal list bitmap, we search
475 ** for the next zero bit in the bitmap that rejected bl. Then, we return that
476 ** through next_zero_bit for find_forward to try.
478 ** Just because we return something in next_zero_bit does not mean we won't
479 ** reject it on the next call to reiserfs_in_journal
482 int reiserfs_in_journal(struct super_block *p_s_sb,
483 int bmap_nr, int bit_nr, int search_all,
484 b_blocknr_t *next_zero_bit) {
485 struct reiserfs_journal_cnode *cn ;
486 struct reiserfs_list_bitmap *jb ;
490 *next_zero_bit = 0 ; /* always start this at zero. */
492 PROC_INFO_INC( p_s_sb, journal.in_journal );
493 /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
494 ** if we crash before the transaction that freed it commits, this transaction won't
495 ** have committed either, and the block will never be written
498 for (i = 0 ; i < JOURNAL_NUM_BITMAPS ; i++) {
499 PROC_INFO_INC( p_s_sb, journal.in_journal_bitmap );
500 jb = SB_JOURNAL(p_s_sb)->j_list_bitmap + i ;
501 if (jb->journal_list && jb->bitmaps[bmap_nr] &&
502 test_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data)) {
503 *next_zero_bit = find_next_zero_bit((unsigned long *)
504 (jb->bitmaps[bmap_nr]->data),
505 p_s_sb->s_blocksize << 3, bit_nr+1) ;
511 bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr;
512 /* is it in any old transactions? */
513 if (search_all && (cn = get_journal_hash_dev(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_hash_table, bl))) {
517 /* is it in the current transaction. This should never happen */
518 if ((cn = get_journal_hash_dev(p_s_sb, SB_JOURNAL(p_s_sb)->j_hash_table, bl))) {
523 PROC_INFO_INC( p_s_sb, journal.in_journal_reusable );
528 /* insert cn into table
530 inline void insert_journal_hash(struct reiserfs_journal_cnode **table, struct reiserfs_journal_cnode *cn) {
531 struct reiserfs_journal_cnode *cn_orig ;
533 cn_orig = journal_hash(table, cn->sb, cn->blocknr) ;
534 cn->hnext = cn_orig ;
537 cn_orig->hprev = cn ;
539 journal_hash(table, cn->sb, cn->blocknr) = cn ;
542 /* lock the current transaction */
543 inline static void lock_journal(struct super_block *p_s_sb) {
544 PROC_INFO_INC( p_s_sb, journal.lock_journal );
545 down(&SB_JOURNAL(p_s_sb)->j_lock);
548 /* unlock the current transaction */
549 inline static void unlock_journal(struct super_block *p_s_sb) {
550 up(&SB_JOURNAL(p_s_sb)->j_lock);
553 static inline void get_journal_list(struct reiserfs_journal_list *jl)
558 static inline void put_journal_list(struct super_block *s,
559 struct reiserfs_journal_list *jl)
561 if (jl->j_refcount < 1) {
562 printk("trans id %lu, refcount at %d\n", jl->j_trans_id,
566 if (--jl->j_refcount == 0)
567 reiserfs_kfree(jl, sizeof(struct reiserfs_journal_list), s);
571 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
572 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
575 static void cleanup_freed_for_journal_list(struct super_block *p_s_sb, struct reiserfs_journal_list *jl) {
577 struct reiserfs_list_bitmap *jb = jl->j_list_bitmap ;
579 cleanup_bitmap_list(p_s_sb, jb) ;
581 jl->j_list_bitmap->journal_list = NULL ;
582 jl->j_list_bitmap = NULL ;
585 static int journal_list_still_alive(struct super_block *s,
586 unsigned long trans_id)
588 struct list_head *entry = &SB_JOURNAL(s)->j_journal_list;
589 struct reiserfs_journal_list *jl;
591 if (!list_empty(entry)) {
592 jl = JOURNAL_LIST_ENTRY(entry->next);
593 if (jl->j_trans_id <= trans_id) {
600 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate) {
601 char b[BDEVNAME_SIZE];
603 if (buffer_journaled(bh)) {
604 reiserfs_warning("clm-2084: pinned buffer %lu:%s sent to disk\n",
605 bh->b_blocknr, bdevname(bh->b_bdev, b)) ;
608 set_buffer_uptodate(bh) ;
610 clear_buffer_uptodate(bh) ;
615 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate) {
617 set_buffer_uptodate(bh) ;
619 clear_buffer_uptodate(bh) ;
624 static void submit_logged_buffer(struct buffer_head *bh) {
626 bh->b_end_io = reiserfs_end_buffer_io_sync ;
627 mark_buffer_notjournal_new(bh) ;
628 clear_buffer_dirty(bh) ;
629 if (!test_and_clear_bit(BH_JTest, &bh->b_state))
631 if (!buffer_uptodate(bh))
633 submit_bh(WRITE, bh) ;
636 static void submit_ordered_buffer(struct buffer_head *bh) {
638 bh->b_end_io = reiserfs_end_ordered_io;
639 clear_buffer_dirty(bh) ;
640 if (!buffer_uptodate(bh))
642 submit_bh(WRITE, bh) ;
645 #define CHUNK_SIZE 32
646 struct buffer_chunk {
647 struct buffer_head *bh[CHUNK_SIZE];
651 static void write_chunk(struct buffer_chunk *chunk) {
653 for (i = 0; i < chunk->nr ; i++) {
654 submit_logged_buffer(chunk->bh[i]) ;
659 static void write_ordered_chunk(struct buffer_chunk *chunk) {
661 for (i = 0; i < chunk->nr ; i++) {
662 submit_ordered_buffer(chunk->bh[i]) ;
667 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
669 void (fn)(struct buffer_chunk *))
672 if (chunk->nr >= CHUNK_SIZE)
674 chunk->bh[chunk->nr++] = bh;
675 if (chunk->nr >= CHUNK_SIZE) {
687 atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
688 static struct reiserfs_jh *alloc_jh(void) {
689 struct reiserfs_jh *jh;
691 jh = kmalloc(sizeof(*jh), GFP_NOFS);
693 atomic_inc(&nr_reiserfs_jh);
701 * we want to free the jh when the buffer has been written
704 void reiserfs_free_jh(struct buffer_head *bh) {
705 struct reiserfs_jh *jh;
709 bh->b_private = NULL;
711 list_del_init(&jh->list);
713 if (atomic_read(&nr_reiserfs_jh) <= 0)
715 atomic_dec(&nr_reiserfs_jh);
720 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
723 struct reiserfs_jh *jh;
726 spin_lock(&j->j_dirty_buffers_lock);
727 if (!bh->b_private) {
728 spin_unlock(&j->j_dirty_buffers_lock);
732 list_del_init(&jh->list);
737 spin_lock(&j->j_dirty_buffers_lock);
738 /* buffer must be locked for __add_jh, should be able to have
739 * two adds at the same time
746 jh->jl = j->j_current_jl;
748 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
750 list_add_tail(&jh->list, &jh->jl->j_bh_list);
752 spin_unlock(&j->j_dirty_buffers_lock);
756 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh) {
757 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
759 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh) {
760 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
763 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
764 static int write_ordered_buffers(spinlock_t *lock,
765 struct reiserfs_journal *j,
766 struct reiserfs_journal_list *jl,
767 struct list_head *list)
769 struct buffer_head *bh;
770 struct reiserfs_jh *jh;
772 struct buffer_chunk chunk;
773 struct list_head tmp;
774 INIT_LIST_HEAD(&tmp);
778 while(!list_empty(list)) {
779 jh = JH_ENTRY(list->next);
782 if (test_set_buffer_locked(bh)) {
783 if (!buffer_dirty(bh)) {
784 list_del_init(&jh->list);
785 list_add(&jh->list, &tmp);
790 write_ordered_chunk(&chunk);
796 if (buffer_dirty(bh)) {
797 list_del_init(&jh->list);
798 list_add(&jh->list, &tmp);
799 add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
801 reiserfs_free_jh(bh);
806 cond_resched_lock(lock);
810 write_ordered_chunk(&chunk);
813 while(!list_empty(&tmp)) {
814 jh = JH_ENTRY(tmp.prev);
817 reiserfs_free_jh(bh);
819 if (buffer_locked(bh)) {
824 if (!buffer_uptodate(bh))
827 cond_resched_lock(lock);
833 static int flush_older_commits(struct super_block *s, struct reiserfs_journal_list *jl) {
834 struct reiserfs_journal_list *other_jl;
835 struct reiserfs_journal_list *first_jl;
836 struct list_head *entry;
837 unsigned long trans_id = jl->j_trans_id;
838 unsigned long other_trans_id;
839 unsigned long first_trans_id;
843 * first we walk backwards to find the oldest uncommitted transation
846 entry = jl->j_list.prev;
848 other_jl = JOURNAL_LIST_ENTRY(entry);
849 if (entry == &SB_JOURNAL(s)->j_journal_list ||
850 atomic_read(&other_jl->j_older_commits_done))
854 entry = other_jl->j_list.prev;
857 /* if we didn't find any older uncommitted transactions, return now */
858 if (first_jl == jl) {
862 first_trans_id = first_jl->j_trans_id;
864 entry = &first_jl->j_list;
866 other_jl = JOURNAL_LIST_ENTRY(entry);
867 other_trans_id = other_jl->j_trans_id;
869 if (other_trans_id < trans_id) {
870 if (atomic_read(&other_jl->j_commit_left) != 0) {
871 flush_commit_list(s, other_jl, 0);
873 /* list we were called with is gone, return */
874 if (!journal_list_still_alive(s, trans_id))
877 /* the one we just flushed is gone, this means all
878 * older lists are also gone, so first_jl is no longer
879 * valid either. Go back to the beginning.
881 if (!journal_list_still_alive(s, other_trans_id)) {
886 if (entry == &SB_JOURNAL(s)->j_journal_list)
894 int reiserfs_async_progress_wait(struct super_block *s) {
896 struct reiserfs_journal *j = SB_JOURNAL(s);
897 if (atomic_read(&j->j_async_throttle))
898 blk_congestion_wait(WRITE, HZ/10);
903 ** if this journal list still has commit blocks unflushed, send them to disk.
905 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
906 ** Before the commit block can by written, every other log block must be safely on disk
909 static int flush_commit_list(struct super_block *s, struct reiserfs_journal_list *jl, int flushall) {
912 struct buffer_head *tbh = NULL ;
913 unsigned long trans_id = jl->j_trans_id;
915 reiserfs_check_lock_depth("flush_commit_list") ;
917 if (atomic_read(&jl->j_older_commits_done)) {
921 /* before we can put our commit blocks on disk, we have to make sure everyone older than
926 if (trans_id == SB_JOURNAL(s)->j_trans_id)
929 get_journal_list(jl);
931 if (flush_older_commits(s, jl) == 1) {
932 /* list disappeared during flush_older_commits. return */
937 /* make sure nobody is trying to flush this one at the same time */
938 down(&jl->j_commit_lock);
939 if (!journal_list_still_alive(s, trans_id)) {
940 up(&jl->j_commit_lock);
943 if (jl->j_trans_id == 0)
946 /* this commit is done, exit */
947 if (atomic_read(&(jl->j_commit_left)) <= 0) {
949 atomic_set(&(jl->j_older_commits_done), 1) ;
951 up(&jl->j_commit_lock);
955 if (!list_empty(&jl->j_bh_list)) {
957 write_ordered_buffers(&SB_JOURNAL(s)->j_dirty_buffers_lock,
958 SB_JOURNAL(s), jl, &jl->j_bh_list);
961 if (!list_empty(&jl->j_bh_list))
964 * for the description block and all the log blocks, submit any buffers
965 * that haven't already reached the disk
967 atomic_inc(&SB_JOURNAL(s)->j_async_throttle);
968 for (i = 0 ; i < (jl->j_len + 1) ; i++) {
969 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start+i) %
970 SB_ONDISK_JOURNAL_SIZE(s);
971 tbh = journal_find_get_block(s, bn) ;
972 if (buffer_dirty(tbh))
973 ll_rw_block(WRITE, 1, &tbh) ;
976 atomic_dec(&SB_JOURNAL(s)->j_async_throttle);
978 /* wait on everything written so far before writing the commit */
979 for (i = 0 ; i < (jl->j_len + 1) ; i++) {
980 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
981 (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s) ;
982 tbh = journal_find_get_block(s, bn) ;
983 wait_on_buffer(tbh) ;
984 // since we're using ll_rw_blk above, it might have skipped over
985 // a locked buffer. Double check here
987 if (buffer_dirty(tbh))
988 sync_dirty_buffer(tbh);
989 if (!buffer_uptodate(tbh)) {
990 reiserfs_panic(s, "journal-601, buffer write failed\n") ;
992 put_bh(tbh) ; /* once for journal_find_get_block */
993 put_bh(tbh) ; /* once due to original getblk in do_journal_end */
994 atomic_dec(&(jl->j_commit_left)) ;
997 if (atomic_read(&(jl->j_commit_left)) != 1)
1000 if (buffer_dirty(jl->j_commit_bh))
1002 mark_buffer_dirty(jl->j_commit_bh) ;
1003 sync_dirty_buffer(jl->j_commit_bh) ;
1004 if (!buffer_uptodate(jl->j_commit_bh)) {
1005 reiserfs_panic(s, "journal-615: buffer write failed\n") ;
1007 bforget(jl->j_commit_bh) ;
1008 if (SB_JOURNAL(s)->j_last_commit_id != 0 &&
1009 (jl->j_trans_id - SB_JOURNAL(s)->j_last_commit_id) != 1) {
1010 reiserfs_warning("clm-2200: last commit %lu, current %lu\n",
1011 SB_JOURNAL(s)->j_last_commit_id,
1014 SB_JOURNAL(s)->j_last_commit_id = jl->j_trans_id;
1016 /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */
1017 cleanup_freed_for_journal_list(s, jl) ;
1019 /* mark the metadata dirty */
1020 dirty_one_transaction(s, jl);
1021 atomic_dec(&(jl->j_commit_left)) ;
1024 atomic_set(&(jl->j_older_commits_done), 1) ;
1026 up(&jl->j_commit_lock);
1028 put_journal_list(s, jl);
1034 ** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or
1035 ** returns NULL if it can't find anything
1037 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct reiserfs_journal_cnode *cn) {
1038 struct super_block *sb = cn->sb;
1039 b_blocknr_t blocknr = cn->blocknr ;
1043 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1051 void remove_journal_hash(struct super_block *, struct reiserfs_journal_cnode **,
1052 struct reiserfs_journal_list *, unsigned long, int);
1055 ** once all the real blocks have been flushed, it is safe to remove them from the
1056 ** journal list for this transaction. Aside from freeing the cnode, this also allows the
1057 ** block to be reallocated for data blocks if it had been deleted.
1059 static void remove_all_from_journal_list(struct super_block *p_s_sb, struct reiserfs_journal_list *jl, int debug) {
1060 struct reiserfs_journal_cnode *cn, *last ;
1061 cn = jl->j_realblock ;
1063 /* which is better, to lock once around the whole loop, or
1064 ** to lock for each call to remove_journal_hash?
1067 if (cn->blocknr != 0) {
1069 printk("block %u, bh is %d, state %ld\n", cn->blocknr, cn->bh ? 1: 0,
1073 remove_journal_hash(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_hash_table, jl, cn->blocknr, 1) ;
1077 free_cnode(p_s_sb, last) ;
1079 jl->j_realblock = NULL ;
1083 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1084 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1085 ** releasing blocks in this transaction for reuse as data blocks.
1086 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1089 static int _update_journal_header_block(struct super_block *p_s_sb, unsigned long offset, unsigned long trans_id) {
1090 struct reiserfs_journal_header *jh ;
1091 if (trans_id >= SB_JOURNAL(p_s_sb)->j_last_flush_trans_id) {
1092 if (buffer_locked((SB_JOURNAL(p_s_sb)->j_header_bh))) {
1093 wait_on_buffer((SB_JOURNAL(p_s_sb)->j_header_bh)) ;
1094 if (!buffer_uptodate(SB_JOURNAL(p_s_sb)->j_header_bh)) {
1095 reiserfs_panic(p_s_sb, "journal-699: buffer write failed\n") ;
1098 SB_JOURNAL(p_s_sb)->j_last_flush_trans_id = trans_id ;
1099 SB_JOURNAL(p_s_sb)->j_first_unflushed_offset = offset ;
1100 jh = (struct reiserfs_journal_header *)(SB_JOURNAL(p_s_sb)->j_header_bh->b_data) ;
1101 jh->j_last_flush_trans_id = cpu_to_le32(trans_id) ;
1102 jh->j_first_unflushed_offset = cpu_to_le32(offset) ;
1103 jh->j_mount_id = cpu_to_le32(SB_JOURNAL(p_s_sb)->j_mount_id) ;
1104 set_buffer_dirty(SB_JOURNAL(p_s_sb)->j_header_bh) ;
1105 sync_dirty_buffer(SB_JOURNAL(p_s_sb)->j_header_bh) ;
1106 if (!buffer_uptodate(SB_JOURNAL(p_s_sb)->j_header_bh)) {
1107 printk( "reiserfs: journal-837: IO error during journal replay\n" );
1114 static int update_journal_header_block(struct super_block *p_s_sb,
1115 unsigned long offset,
1116 unsigned long trans_id) {
1117 if (_update_journal_header_block(p_s_sb, offset, trans_id)) {
1118 reiserfs_panic(p_s_sb, "journal-712: buffer write failed\n") ;
1123 ** flush any and all journal lists older than you are
1124 ** can only be called from flush_journal_list
1126 static int flush_older_journal_lists(struct super_block *p_s_sb,
1127 struct reiserfs_journal_list *jl)
1129 struct list_head *entry;
1130 struct reiserfs_journal_list *other_jl ;
1131 unsigned long trans_id = jl->j_trans_id;
1133 /* we know we are the only ones flushing things, no extra race
1134 * protection is required.
1137 entry = SB_JOURNAL(p_s_sb)->j_journal_list.next;
1138 other_jl = JOURNAL_LIST_ENTRY(entry);
1139 if (other_jl->j_trans_id < trans_id) {
1140 /* do not flush all */
1141 flush_journal_list(p_s_sb, other_jl, 0) ;
1143 /* other_jl is now deleted from the list */
1149 static void del_from_work_list(struct super_block *s,
1150 struct reiserfs_journal_list *jl) {
1151 if (!list_empty(&jl->j_working_list)) {
1152 list_del_init(&jl->j_working_list);
1153 SB_JOURNAL(s)->j_num_work_lists--;
1157 /* flush a journal list, both commit and real blocks
1159 ** always set flushall to 1, unless you are calling from inside
1160 ** flush_journal_list
1162 ** IMPORTANT. This can only be called while there are no journal writers,
1163 ** and the journal is locked. That means it can only be called from
1164 ** do_journal_end, or by journal_release
1166 static int flush_journal_list(struct super_block *s,
1167 struct reiserfs_journal_list *jl, int flushall) {
1168 struct reiserfs_journal_list *pjl ;
1169 struct reiserfs_journal_cnode *cn, *last ;
1173 struct buffer_head *saved_bh ;
1174 unsigned long j_len_saved = jl->j_len ;
1176 if (j_len_saved <= 0) {
1180 if (atomic_read(&SB_JOURNAL(s)->j_wcount) != 0) {
1181 reiserfs_warning("clm-2048: flush_journal_list called with wcount %d\n",
1182 atomic_read(&SB_JOURNAL(s)->j_wcount)) ;
1184 if (jl->j_trans_id == 0)
1187 /* if flushall == 0, the lock is already held */
1189 down(&SB_JOURNAL(s)->j_flush_sem);
1190 } else if (!down_trylock(&SB_JOURNAL(s)->j_flush_sem)) {
1195 if (j_len_saved > SB_JOURNAL_TRANS_MAX(s)) {
1196 reiserfs_panic(s, "journal-715: flush_journal_list, length is %lu, trans id %lu\n", j_len_saved, jl->j_trans_id);
1200 /* if all the work is already done, get out of here */
1201 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1202 atomic_read(&(jl->j_commit_left)) <= 0) {
1203 goto flush_older_and_return ;
1206 /* start by putting the commit list on disk. This will also flush
1207 ** the commit lists of any olders transactions
1209 flush_commit_list(s, jl, 1) ;
1211 if (!(jl->j_state & LIST_DIRTY))
1214 /* are we done now? */
1215 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1216 atomic_read(&(jl->j_commit_left)) <= 0) {
1217 goto flush_older_and_return ;
1220 /* loop through each cnode, see if we need to write it,
1221 ** or wait on a more recent transaction, or just ignore it
1223 if (atomic_read(&(SB_JOURNAL(s)->j_wcount)) != 0) {
1224 reiserfs_panic(s, "journal-844: panic journal list is flushing, wcount is not 0\n") ;
1226 cn = jl->j_realblock ;
1231 /* blocknr of 0 is no longer in the hash, ignore it */
1232 if (cn->blocknr == 0) {
1235 pjl = find_newer_jl_for_cn(cn) ;
1236 /* the order is important here. We check pjl to make sure we
1237 ** don't clear BH_JDirty_wait if we aren't the one writing this
1240 if (!pjl && cn->bh) {
1243 /* we do this to make sure nobody releases the buffer while
1244 ** we are working with it
1248 if (buffer_journal_dirty(saved_bh)) {
1253 } else if (can_dirty(cn)) {
1254 /* everything with !pjl && jwait should be writable */
1259 /* if someone has this block in a newer transaction, just make
1260 ** sure they are commited, and don't try writing it to disk
1263 if (atomic_read(&pjl->j_commit_left))
1264 flush_commit_list(s, pjl, 1) ;
1268 /* bh == NULL when the block got to disk on its own, OR,
1269 ** the block got freed in a future transaction
1271 if (saved_bh == NULL) {
1275 /* this should never happen. kupdate_one_transaction has this list
1276 ** locked while it works, so we should never see a buffer here that
1277 ** is not marked JDirty_wait
1279 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1280 printk("journal-813: BAD! buffer %llu %cdirty %cjwait, not in a newer tranasction\n", (unsigned long long)saved_bh->b_blocknr,
1281 was_dirty ? ' ' : '!', was_jwait ? ' ' : '!') ;
1284 /* we inc again because saved_bh gets decremented at free_cnode */
1286 set_bit(BLOCK_NEEDS_FLUSH, &cn->state) ;
1287 lock_buffer(saved_bh);
1288 if (cn->blocknr != saved_bh->b_blocknr)
1290 if (buffer_dirty(saved_bh))
1291 submit_logged_buffer(saved_bh) ;
1293 unlock_buffer(saved_bh);
1296 printk("clm-2082: Unable to flush buffer %llu in flush_journal_list\n",
1297 (unsigned long long)saved_bh->b_blocknr) ;
1303 /* we incremented this to keep others from taking the buffer head away */
1305 if (atomic_read(&(saved_bh->b_count)) < 0) {
1306 printk("journal-945: saved_bh->b_count < 0") ;
1311 cn = jl->j_realblock ;
1313 if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1315 reiserfs_panic(s, "journal-1011: cn->bh is NULL\n") ;
1317 wait_on_buffer(cn->bh) ;
1319 reiserfs_panic(s, "journal-1012: cn->bh is NULL\n") ;
1321 if (!buffer_uptodate(cn->bh)) {
1322 reiserfs_panic(s, "journal-949: buffer write failed\n") ;
1324 /* note, we must clear the JDirty_wait bit after the up to date
1325 ** check, otherwise we race against our flushpage routine
1327 if (!test_and_clear_bit(BH_JDirty_wait, &cn->bh->b_state))
1330 /* undo the inc from journal_mark_dirty */
1338 flush_older_and_return:
1339 /* before we can update the journal header block, we _must_ flush all
1340 ** real blocks from all older transactions to disk. This is because
1341 ** once the header block is updated, this transaction will not be
1342 ** replayed after a crash
1345 flush_older_journal_lists(s, jl);
1348 /* before we can remove everything from the hash tables for this
1349 ** transaction, we must make sure it can never be replayed
1351 ** since we are only called from do_journal_end, we know for sure there
1352 ** are no allocations going on while we are flushing journal lists. So,
1353 ** we only need to update the journal header block for the last list
1357 update_journal_header_block(s, (jl->j_start + jl->j_len + 2) % SB_ONDISK_JOURNAL_SIZE(s), jl->j_trans_id) ;
1359 remove_all_from_journal_list(s, jl, 0) ;
1360 list_del(&jl->j_list);
1361 SB_JOURNAL(s)->j_num_lists--;
1362 del_from_work_list(s, jl);
1364 if (SB_JOURNAL(s)->j_last_flush_id != 0 &&
1365 (jl->j_trans_id - SB_JOURNAL(s)->j_last_flush_id) != 1) {
1366 reiserfs_warning("clm-2201: last flush %lu, current %lu\n",
1367 SB_JOURNAL(s)->j_last_flush_id,
1370 SB_JOURNAL(s)->j_last_flush_id = jl->j_trans_id;
1372 /* not strictly required since we are freeing the list, but it should
1373 * help find code using dead lists later on
1376 atomic_set(&(jl->j_nonzerolen), 0) ;
1378 jl->j_realblock = NULL ;
1379 jl->j_commit_bh = NULL ;
1380 jl->j_trans_id = 0 ;
1382 put_journal_list(s, jl);
1384 up(&SB_JOURNAL(s)->j_flush_sem);
1388 static int write_one_transaction(struct super_block *s,
1389 struct reiserfs_journal_list *jl,
1390 struct buffer_chunk *chunk)
1392 struct reiserfs_journal_cnode *cn;
1395 jl->j_state |= LIST_TOUCHED;
1396 del_from_work_list(s, jl);
1397 if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1401 cn = jl->j_realblock ;
1403 /* if the blocknr == 0, this has been cleared from the hash,
1406 if (cn->blocknr == 0) {
1409 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1410 struct buffer_head *tmp_bh;
1411 /* we can race against journal_mark_freed when we try
1412 * to lock_buffer(cn->bh), so we have to inc the buffer
1413 * count, and recheck things after locking
1417 lock_buffer(tmp_bh);
1418 if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1419 if (!buffer_journal_dirty(tmp_bh) ||
1420 reiserfs_buffer_prepared(tmp_bh))
1422 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1425 /* note, cn->bh might be null now */
1426 unlock_buffer(tmp_bh);
1437 /* used by flush_commit_list */
1438 static int dirty_one_transaction(struct super_block *s,
1439 struct reiserfs_journal_list *jl)
1441 struct reiserfs_journal_cnode *cn;
1442 struct reiserfs_journal_list *pjl;
1445 jl->j_state |= LIST_DIRTY;
1446 cn = jl->j_realblock ;
1448 /* look for a more recent transaction that logged this
1449 ** buffer. Only the most recent transaction with a buffer in
1450 ** it is allowed to send that buffer to disk
1452 pjl = find_newer_jl_for_cn(cn) ;
1453 if (!pjl && cn->blocknr && cn->bh && buffer_journal_dirty(cn->bh))
1457 /* if the buffer is prepared, it will either be logged
1458 * or restored. If restored, we need to make sure
1459 * it actually gets marked dirty
1461 mark_buffer_notjournal_new(cn->bh) ;
1462 if (test_bit(BH_JPrepared, &cn->bh->b_state)) {
1463 set_bit(BH_JRestore_dirty, &cn->bh->b_state);
1465 set_bit(BH_JTest, &cn->bh->b_state);
1466 mark_buffer_dirty(cn->bh);
1474 static int kupdate_transactions(struct super_block *s,
1475 struct reiserfs_journal_list *jl,
1476 struct reiserfs_journal_list **next_jl,
1477 unsigned long *next_trans_id,
1482 int transactions_flushed = 0;
1483 unsigned long orig_trans_id = jl->j_trans_id;
1484 struct buffer_chunk chunk;
1485 struct list_head *entry;
1488 down(&SB_JOURNAL(s)->j_flush_sem);
1489 if (!journal_list_still_alive(s, orig_trans_id)) {
1493 /* we've got j_flush_sem held, nobody is going to delete any
1494 * of these lists out from underneath us
1496 while((num_trans && transactions_flushed < num_trans) ||
1497 (!num_trans && written < num_blocks)) {
1499 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1500 atomic_read(&jl->j_commit_left))
1502 del_from_work_list(s, jl);
1505 ret = write_one_transaction(s, jl, &chunk);
1509 transactions_flushed++;
1511 entry = jl->j_list.next;
1514 if (entry == &SB_JOURNAL(s)->j_journal_list) {
1517 jl = JOURNAL_LIST_ENTRY(entry);
1519 /* don't bother with older transactions */
1520 if (jl->j_trans_id <= orig_trans_id)
1524 write_chunk(&chunk);
1528 up(&SB_JOURNAL(s)->j_flush_sem);
1532 /* for o_sync and fsync heavy applications, they tend to use
1533 ** all the journa list slots with tiny transactions. These
1534 ** trigger lots and lots of calls to update the header block, which
1535 ** adds seeks and slows things down.
1537 ** This function tries to clear out a large chunk of the journal lists
1538 ** at once, which makes everything faster since only the newest journal
1539 ** list updates the header block
1541 static int flush_used_journal_lists(struct super_block *s,
1542 struct reiserfs_journal_list *jl) {
1543 unsigned long len = 0;
1544 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 /* flush for 256 transactions or 256 blocks, whichever comes first */
1554 for(i = 0 ; i < 256 && len < 256 ; i++) {
1555 if (atomic_read(&tjl->j_commit_left) ||
1556 tjl->j_trans_id < jl->j_trans_id) {
1559 cur_len = atomic_read(&tjl->j_nonzerolen);
1561 tjl->j_state &= ~LIST_TOUCHED;
1565 if (tjl->j_list.next == &SB_JOURNAL(s)->j_journal_list)
1567 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1569 /* try to find a group of blocks we can flush across all the
1570 ** transactions, but only bother if we've actually spanned
1571 ** across multiple lists
1573 if (flush_jl != jl) {
1574 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1576 flush_journal_list(s, flush_jl, 1);
1581 ** removes any nodes in table with name block and dev as bh.
1582 ** only touchs the hnext and hprev pointers.
1584 void remove_journal_hash(struct super_block *sb,
1585 struct reiserfs_journal_cnode **table,
1586 struct reiserfs_journal_list *jl,
1587 unsigned long block, int remove_freed)
1589 struct reiserfs_journal_cnode *cur ;
1590 struct reiserfs_journal_cnode **head ;
1592 head= &(journal_hash(table, sb, block)) ;
1598 if (cur->blocknr == block && cur->sb == sb && (jl == NULL || jl == cur->jlist) &&
1599 (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1601 cur->hnext->hprev = cur->hprev ;
1604 cur->hprev->hnext = cur->hnext ;
1606 *head = cur->hnext ;
1611 if (cur->bh && cur->jlist) /* anybody who clears the cur->bh will also dec the nonzerolen */
1612 atomic_dec(&(cur->jlist->j_nonzerolen)) ;
1620 static void free_journal_ram(struct super_block *p_s_sb) {
1621 reiserfs_kfree(SB_JOURNAL(p_s_sb)->j_current_jl,
1622 sizeof(struct reiserfs_journal_list), p_s_sb);
1623 SB_JOURNAL(p_s_sb)->j_num_lists--;
1625 vfree(SB_JOURNAL(p_s_sb)->j_cnode_free_orig) ;
1626 free_list_bitmaps(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_bitmap) ;
1627 free_bitmap_nodes(p_s_sb) ; /* must be after free_list_bitmaps */
1628 if (SB_JOURNAL(p_s_sb)->j_header_bh) {
1629 brelse(SB_JOURNAL(p_s_sb)->j_header_bh) ;
1631 /* j_header_bh is on the journal dev, make sure not to release the journal
1632 * dev until we brelse j_header_bh
1634 release_journal_dev(p_s_sb, SB_JOURNAL(p_s_sb));
1635 vfree(SB_JOURNAL(p_s_sb)) ;
1639 ** call on unmount. Only set error to 1 if you haven't made your way out
1640 ** of read_super() yet. Any other caller must keep error at 0.
1642 static int do_journal_release(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, int error) {
1643 struct reiserfs_transaction_handle myth ;
1645 /* we only want to flush out transactions if we were called with error == 0
1647 if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
1648 /* end the current trans */
1649 do_journal_end(th, p_s_sb,10, FLUSH_ALL) ;
1651 /* make sure something gets logged to force our way into the flush code */
1652 journal_join(&myth, p_s_sb, 1) ;
1653 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
1654 journal_mark_dirty(&myth, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
1655 do_journal_end(&myth, p_s_sb,1, FLUSH_ALL) ;
1658 reiserfs_mounted_fs_count-- ;
1659 /* wait for all commits to finish */
1660 cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work);
1661 flush_workqueue(commit_wq);
1662 if (!reiserfs_mounted_fs_count) {
1663 destroy_workqueue(commit_wq);
1667 free_journal_ram(p_s_sb) ;
1673 ** call on unmount. flush all journal trans, release all alloc'd ram
1675 int journal_release(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb) {
1676 return do_journal_release(th, p_s_sb, 0) ;
1679 ** only call from an error condition inside reiserfs_read_super!
1681 int journal_release_error(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb) {
1682 return do_journal_release(th, p_s_sb, 1) ;
1685 /* compares description block with commit block. returns 1 if they differ, 0 if they are the same */
1686 static int journal_compare_desc_commit(struct super_block *p_s_sb, struct reiserfs_journal_desc *desc,
1687 struct reiserfs_journal_commit *commit) {
1688 if (get_commit_trans_id (commit) != get_desc_trans_id (desc) ||
1689 get_commit_trans_len (commit) != get_desc_trans_len (desc) ||
1690 get_commit_trans_len (commit) > SB_JOURNAL_TRANS_MAX(p_s_sb) ||
1691 get_commit_trans_len (commit) <= 0
1697 /* returns 0 if it did not find a description block
1698 ** returns -1 if it found a corrupt commit block
1699 ** returns 1 if both desc and commit were valid
1701 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) {
1702 struct reiserfs_journal_desc *desc ;
1703 struct reiserfs_journal_commit *commit ;
1704 struct buffer_head *c_bh ;
1705 unsigned long offset ;
1710 desc = (struct reiserfs_journal_desc *)d_bh->b_data ;
1711 if (get_desc_trans_len(desc) > 0 && !memcmp(get_journal_desc_magic (d_bh), JOURNAL_DESC_MAGIC, 8)) {
1712 if (oldest_invalid_trans_id && *oldest_invalid_trans_id && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
1713 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-986: transaction "
1714 "is valid returning because trans_id %d is greater than "
1715 "oldest_invalid %lu\n", get_desc_trans_id(desc),
1716 *oldest_invalid_trans_id);
1719 if (newest_mount_id && *newest_mount_id > get_desc_mount_id (desc)) {
1720 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1087: transaction "
1721 "is valid returning because mount_id %d is less than "
1722 "newest_mount_id %lu\n", get_desc_mount_id (desc),
1726 if ( get_desc_trans_len(desc) > SB_JOURNAL_TRANS_MAX(p_s_sb) ) {
1727 reiserfs_warning("journal-2018: Bad transaction length %d encountered, ignoring transaction\n", get_desc_trans_len(desc));
1730 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ;
1732 /* ok, we have a journal description block, lets see if the transaction was valid */
1733 c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
1734 ((offset + get_desc_trans_len(desc) + 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ;
1737 commit = (struct reiserfs_journal_commit *)c_bh->b_data ;
1738 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
1739 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1740 "journal_transaction_is_valid, commit offset %ld had bad "
1741 "time %d or length %d\n",
1742 c_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1743 get_commit_trans_id (commit),
1744 get_commit_trans_len(commit));
1746 if (oldest_invalid_trans_id) {
1747 *oldest_invalid_trans_id = get_desc_trans_id(desc) ;
1748 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1004: "
1749 "transaction_is_valid setting oldest invalid trans_id "
1750 "to %d\n", get_desc_trans_id(desc)) ;
1755 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1006: found valid "
1756 "transaction start offset %llu, len %d id %d\n",
1757 d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1758 get_desc_trans_len(desc), get_desc_trans_id(desc)) ;
1765 static void brelse_array(struct buffer_head **heads, int num) {
1767 for (i = 0 ; i < num ; i++) {
1773 ** given the start, and values for the oldest acceptable transactions,
1774 ** this either reads in a replays a transaction, or returns because the transaction
1775 ** is invalid, or too old.
1777 static int journal_read_transaction(struct super_block *p_s_sb, unsigned long cur_dblock, unsigned long oldest_start,
1778 unsigned long oldest_trans_id, unsigned long newest_mount_id) {
1779 struct reiserfs_journal_desc *desc ;
1780 struct reiserfs_journal_commit *commit ;
1781 unsigned long trans_id = 0 ;
1782 struct buffer_head *c_bh ;
1783 struct buffer_head *d_bh ;
1784 struct buffer_head **log_blocks = NULL ;
1785 struct buffer_head **real_blocks = NULL ;
1786 unsigned long trans_offset ;
1790 d_bh = journal_bread(p_s_sb, cur_dblock) ;
1793 desc = (struct reiserfs_journal_desc *)d_bh->b_data ;
1794 trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ;
1795 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
1796 "journal_read_transaction, offset %llu, len %d mount_id %d\n",
1797 d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1798 get_desc_trans_len(desc), get_desc_mount_id(desc)) ;
1799 if (get_desc_trans_id(desc) < oldest_trans_id) {
1800 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
1801 "journal_read_trans skipping because %lu is too old\n",
1802 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)) ;
1806 if (get_desc_mount_id(desc) != newest_mount_id) {
1807 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
1808 "journal_read_trans skipping because %d is != "
1809 "newest_mount_id %lu\n", get_desc_mount_id(desc),
1814 c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
1815 ((trans_offset + get_desc_trans_len(desc) + 1) %
1816 SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ;
1821 commit = (struct reiserfs_journal_commit *)c_bh->b_data ;
1822 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
1823 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal_read_transaction, "
1824 "commit offset %llu had bad time %d or length %d\n",
1825 c_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1826 get_commit_trans_id(commit), get_commit_trans_len(commit));
1831 trans_id = get_desc_trans_id(desc) ;
1832 /* now we know we've got a good transaction, and it was inside the valid time ranges */
1833 log_blocks = reiserfs_kmalloc(get_desc_trans_len(desc) * sizeof(struct buffer_head *), GFP_NOFS, p_s_sb) ;
1834 real_blocks = reiserfs_kmalloc(get_desc_trans_len(desc) * sizeof(struct buffer_head *), GFP_NOFS, p_s_sb) ;
1835 if (!log_blocks || !real_blocks) {
1838 reiserfs_kfree(log_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1839 reiserfs_kfree(real_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1840 reiserfs_warning("journal-1169: kmalloc failed, unable to mount FS\n") ;
1843 /* get all the buffer heads */
1844 trans_half = journal_trans_half (p_s_sb->s_blocksize) ;
1845 for(i = 0 ; i < get_desc_trans_len(desc) ; i++) {
1846 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));
1847 if (i < trans_half) {
1848 real_blocks[i] = sb_getblk(p_s_sb, le32_to_cpu(desc->j_realblock[i])) ;
1850 real_blocks[i] = sb_getblk(p_s_sb, le32_to_cpu(commit->j_realblock[i - trans_half])) ;
1852 if ( real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb) ) {
1853 reiserfs_warning("journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem\n");
1856 /* make sure we don't try to replay onto log or reserved area */
1857 if (is_block_in_log_or_reserved_area(p_s_sb, real_blocks[i]->b_blocknr)) {
1858 reiserfs_warning("journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block\n") ;
1860 brelse_array(log_blocks, i) ;
1861 brelse_array(real_blocks, i) ;
1864 reiserfs_kfree(log_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1865 reiserfs_kfree(real_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1869 /* read in the log blocks, memcpy to the corresponding real block */
1870 ll_rw_block(READ, get_desc_trans_len(desc), log_blocks) ;
1871 for (i = 0 ; i < get_desc_trans_len(desc) ; i++) {
1872 wait_on_buffer(log_blocks[i]) ;
1873 if (!buffer_uptodate(log_blocks[i])) {
1874 reiserfs_warning("journal-1212: REPLAY FAILURE fsck required! buffer write failed\n") ;
1875 brelse_array(log_blocks + i, get_desc_trans_len(desc) - i) ;
1876 brelse_array(real_blocks, get_desc_trans_len(desc)) ;
1879 reiserfs_kfree(log_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1880 reiserfs_kfree(real_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1883 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data, real_blocks[i]->b_size) ;
1884 set_buffer_uptodate(real_blocks[i]) ;
1885 brelse(log_blocks[i]) ;
1887 /* flush out the real blocks */
1888 for (i = 0 ; i < get_desc_trans_len(desc) ; i++) {
1889 set_buffer_dirty(real_blocks[i]) ;
1890 ll_rw_block(WRITE, 1, real_blocks + i) ;
1892 for (i = 0 ; i < get_desc_trans_len(desc) ; i++) {
1893 wait_on_buffer(real_blocks[i]) ;
1894 if (!buffer_uptodate(real_blocks[i])) {
1895 reiserfs_warning("journal-1226: REPLAY FAILURE, fsck required! buffer write failed\n") ;
1896 brelse_array(real_blocks + i, get_desc_trans_len(desc) - i) ;
1899 reiserfs_kfree(log_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1900 reiserfs_kfree(real_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1903 brelse(real_blocks[i]) ;
1905 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)) ;
1906 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1095: setting journal "
1907 "start to offset %ld\n",
1908 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)) ;
1910 /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
1911 SB_JOURNAL(p_s_sb)->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ;
1912 SB_JOURNAL(p_s_sb)->j_last_flush_trans_id = trans_id ;
1913 SB_JOURNAL(p_s_sb)->j_trans_id = trans_id + 1;
1916 reiserfs_kfree(log_blocks, le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), p_s_sb) ;
1917 reiserfs_kfree(real_blocks, le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), p_s_sb) ;
1921 /* This function reads blocks starting from block and to max_block of bufsize
1922 size (but no more than BUFNR blocks at a time). This proved to improve
1923 mounting speed on self-rebuilding raid5 arrays at least.
1924 Right now it is only used from journal code. But later we might use it
1926 Note: Do not use journal_getblk/sb_getblk functions here! */
1927 struct buffer_head * reiserfs_breada (struct block_device *dev, int block, int bufsize,
1928 unsigned int max_block)
1930 struct buffer_head * bhlist[BUFNR];
1931 unsigned int blocks = BUFNR;
1932 struct buffer_head * bh;
1935 bh = __getblk (dev, block, bufsize );
1936 if (buffer_uptodate (bh))
1939 if (block + BUFNR > max_block) {
1940 blocks = max_block - block;
1944 for (i = 1; i < blocks; i++) {
1945 bh = __getblk (dev, block + i, bufsize);
1946 if (buffer_uptodate (bh)) {
1950 else bhlist[j++] = bh;
1952 ll_rw_block (READ, j, bhlist);
1953 for(i = 1; i < j; i++)
1956 wait_on_buffer (bh);
1957 if (buffer_uptodate (bh))
1964 ** read and replay the log
1965 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
1966 ** transaction. This tests that before finding all the transactions in the log, which makes normal mount times fast.
1968 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
1970 ** On exit, it sets things up so the first transaction will work correctly.
1972 static int journal_read(struct super_block *p_s_sb) {
1973 struct reiserfs_journal_desc *desc ;
1974 unsigned long oldest_trans_id = 0;
1975 unsigned long oldest_invalid_trans_id = 0 ;
1977 unsigned long oldest_start = 0;
1978 unsigned long cur_dblock = 0 ;
1979 unsigned long newest_mount_id = 9 ;
1980 struct buffer_head *d_bh ;
1981 struct reiserfs_journal_header *jh ;
1982 int valid_journal_header = 0 ;
1983 int replay_count = 0 ;
1984 int continue_replay = 1 ;
1986 char b[BDEVNAME_SIZE];
1988 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ;
1989 printk("reiserfs: checking transaction log (%s) for (%s)\n",
1990 bdevname(SB_JOURNAL(p_s_sb)->j_dev_bd, b),
1991 reiserfs_bdevname(p_s_sb));
1992 start = get_seconds();
1994 /* step 1, read in the journal header block. Check the transaction it says
1995 ** is the first unflushed, and if that transaction is not valid,
1998 SB_JOURNAL(p_s_sb)->j_header_bh = journal_bread(p_s_sb,
1999 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2000 SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2001 if (!SB_JOURNAL(p_s_sb)->j_header_bh) {
2004 jh = (struct reiserfs_journal_header *)(SB_JOURNAL(p_s_sb)->j_header_bh->b_data) ;
2005 if (le32_to_cpu(jh->j_first_unflushed_offset) >= 0 &&
2006 le32_to_cpu(jh->j_first_unflushed_offset) < SB_ONDISK_JOURNAL_SIZE(p_s_sb) &&
2007 le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2008 oldest_start = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2009 le32_to_cpu(jh->j_first_unflushed_offset) ;
2010 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2011 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2012 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1153: found in "
2013 "header: first_unflushed_offset %d, last_flushed_trans_id "
2014 "%lu\n", le32_to_cpu(jh->j_first_unflushed_offset),
2015 le32_to_cpu(jh->j_last_flush_trans_id)) ;
2016 valid_journal_header = 1 ;
2018 /* now, we try to read the first unflushed offset. If it is not valid,
2019 ** there is nothing more we can do, and it makes no sense to read
2020 ** through the whole log.
2022 d_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + le32_to_cpu(jh->j_first_unflushed_offset)) ;
2023 ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL) ;
2025 continue_replay = 0 ;
2028 goto start_log_replay;
2031 if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) {
2032 printk("clm-2076: device is readonly, unable to replay log\n") ;
2036 /* ok, there are transactions that need to be replayed. start with the first log block, find
2037 ** all the valid transactions, and pick out the oldest.
2039 while(continue_replay && cur_dblock < (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_ONDISK_JOURNAL_SIZE(p_s_sb))) {
2040 /* Note that it is required for blocksize of primary fs device and journal
2041 device to be the same */
2042 d_bh = reiserfs_breada(SB_JOURNAL(p_s_sb)->j_dev_bd, cur_dblock, p_s_sb->s_blocksize,
2043 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_ONDISK_JOURNAL_SIZE(p_s_sb)) ;
2044 ret = journal_transaction_is_valid(p_s_sb, d_bh, &oldest_invalid_trans_id, &newest_mount_id) ;
2046 desc = (struct reiserfs_journal_desc *)d_bh->b_data ;
2047 if (oldest_start == 0) { /* init all oldest_ values */
2048 oldest_trans_id = get_desc_trans_id(desc) ;
2049 oldest_start = d_bh->b_blocknr ;
2050 newest_mount_id = get_desc_mount_id(desc) ;
2051 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1179: Setting "
2052 "oldest_start to offset %llu, trans_id %lu\n",
2053 oldest_start - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2055 } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2056 /* one we just read was older */
2057 oldest_trans_id = get_desc_trans_id(desc) ;
2058 oldest_start = d_bh->b_blocknr ;
2059 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1180: Resetting "
2060 "oldest_start to offset %lu, trans_id %lu\n",
2061 oldest_start - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2064 if (newest_mount_id < get_desc_mount_id(desc)) {
2065 newest_mount_id = get_desc_mount_id(desc) ;
2066 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2067 "newest_mount_id to %d\n", get_desc_mount_id(desc));
2069 cur_dblock += get_desc_trans_len(desc) + 2 ;
2077 cur_dblock = oldest_start ;
2078 if (oldest_trans_id) {
2079 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1206: Starting replay "
2080 "from offset %llu, trans_id %lu\n",
2081 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2086 while(continue_replay && oldest_trans_id > 0) {
2087 ret = journal_read_transaction(p_s_sb, cur_dblock, oldest_start, oldest_trans_id, newest_mount_id) ;
2090 } else if (ret != 0) {
2093 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_JOURNAL(p_s_sb)->j_start ;
2095 if (cur_dblock == oldest_start)
2099 if (oldest_trans_id == 0) {
2100 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1225: No valid "
2101 "transactions found\n") ;
2103 /* j_start does not get set correctly if we don't replay any transactions.
2104 ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2105 ** copy the trans_id from the header
2107 if (valid_journal_header && replay_count == 0) {
2108 SB_JOURNAL(p_s_sb)->j_start = le32_to_cpu(jh->j_first_unflushed_offset) ;
2109 SB_JOURNAL(p_s_sb)->j_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2110 SB_JOURNAL(p_s_sb)->j_last_flush_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) ;
2111 SB_JOURNAL(p_s_sb)->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2113 SB_JOURNAL(p_s_sb)->j_mount_id = newest_mount_id + 1 ;
2115 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2116 "newest_mount_id to %lu\n", SB_JOURNAL(p_s_sb)->j_mount_id) ;
2117 SB_JOURNAL(p_s_sb)->j_first_unflushed_offset = SB_JOURNAL(p_s_sb)->j_start ;
2118 if (replay_count > 0) {
2119 printk("reiserfs: replayed %d transactions in %lu seconds\n", replay_count,
2120 get_seconds() - start) ;
2122 if (!bdev_read_only(p_s_sb->s_bdev) &&
2123 _update_journal_header_block(p_s_sb, SB_JOURNAL(p_s_sb)->j_start,
2124 SB_JOURNAL(p_s_sb)->j_last_flush_trans_id))
2126 /* replay failed, caller must call free_journal_ram and abort
2134 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2136 struct reiserfs_journal_list *jl;
2138 jl = reiserfs_kmalloc(sizeof(struct reiserfs_journal_list), GFP_NOFS, s);
2143 memset(jl, 0, sizeof(*jl));
2144 INIT_LIST_HEAD(&jl->j_list);
2145 INIT_LIST_HEAD(&jl->j_working_list);
2146 INIT_LIST_HEAD(&jl->j_tail_bh_list);
2147 INIT_LIST_HEAD(&jl->j_bh_list);
2148 sema_init(&jl->j_commit_lock, 1);
2149 SB_JOURNAL(s)->j_num_lists++;
2150 get_journal_list(jl);
2154 static void journal_list_init(struct super_block *p_s_sb) {
2155 SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
2158 static int release_journal_dev( struct super_block *super,
2159 struct reiserfs_journal *journal )
2165 if( journal -> j_dev_file != NULL ) {
2166 result = filp_close( journal -> j_dev_file, NULL );
2167 journal -> j_dev_file = NULL;
2168 journal -> j_dev_bd = NULL;
2169 } else if( journal -> j_dev_bd != NULL ) {
2170 result = blkdev_put( journal -> j_dev_bd );
2171 journal -> j_dev_bd = NULL;
2175 reiserfs_warning("sh-457: release_journal_dev: Cannot release journal device: %i\n", result );
2180 static int journal_init_dev( struct super_block *super,
2181 struct reiserfs_journal *journal,
2182 const char *jdev_name )
2186 int blkdev_mode = FMODE_READ | FMODE_WRITE;
2187 char b[BDEVNAME_SIZE];
2191 journal -> j_dev_bd = NULL;
2192 journal -> j_dev_file = NULL;
2193 jdev = SB_ONDISK_JOURNAL_DEVICE( super ) ?
2194 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2196 if (bdev_read_only(super->s_bdev))
2197 blkdev_mode = FMODE_READ;
2199 /* there is no "jdev" option and journal is on separate device */
2200 if( ( !jdev_name || !jdev_name[ 0 ] ) ) {
2201 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2202 if (IS_ERR(journal->j_dev_bd)) {
2203 result = PTR_ERR(journal->j_dev_bd);
2204 journal->j_dev_bd = NULL;
2205 printk( "sh-458: journal_init_dev: cannot init journal device\n '%s': %i",
2206 __bdevname(jdev, b), result );
2208 } else if (jdev != super->s_dev)
2209 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2213 journal -> j_dev_file = filp_open( jdev_name, 0, 0 );
2214 if( !IS_ERR( journal -> j_dev_file ) ) {
2215 struct inode *jdev_inode = journal->j_dev_file->f_mapping->host;
2216 if( !S_ISBLK( jdev_inode -> i_mode ) ) {
2217 printk( "journal_init_dev: '%s' is not a block device\n", jdev_name );
2221 journal->j_dev_bd = I_BDEV(jdev_inode);
2222 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2225 result = PTR_ERR( journal -> j_dev_file );
2226 journal -> j_dev_file = NULL;
2227 printk( "journal_init_dev: Cannot open '%s': %i\n", jdev_name, result );
2230 release_journal_dev( super, journal );
2232 printk( "journal_init_dev: journal device: %s\n",
2233 bdevname(journal->j_dev_bd, b));
2238 ** must be called once on fs mount. calls journal_read for you
2240 int journal_init(struct super_block *p_s_sb, const char * j_dev_name, int old_format, unsigned int commit_max_age) {
2241 int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2 ;
2242 struct buffer_head *bhjh;
2243 struct reiserfs_super_block * rs;
2244 struct reiserfs_journal_header *jh;
2245 struct reiserfs_journal *journal;
2246 struct reiserfs_journal_list *jl;
2247 char b[BDEVNAME_SIZE];
2249 journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof (struct reiserfs_journal)) ;
2251 printk("journal-1256: unable to get memory for journal structure\n") ;
2254 memset(journal, 0, sizeof(struct reiserfs_journal)) ;
2255 INIT_LIST_HEAD(&SB_JOURNAL(p_s_sb)->j_bitmap_nodes) ;
2256 INIT_LIST_HEAD (&SB_JOURNAL(p_s_sb)->j_prealloc_list);
2257 INIT_LIST_HEAD(&SB_JOURNAL(p_s_sb)->j_working_list);
2258 INIT_LIST_HEAD(&SB_JOURNAL(p_s_sb)->j_journal_list);
2259 reiserfs_allocate_list_bitmaps(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_bitmap,
2260 SB_BMAP_NR(p_s_sb)) ;
2261 allocate_bitmap_nodes(p_s_sb) ;
2263 /* reserved for journal area support */
2264 SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ?
2265 REISERFS_OLD_DISK_OFFSET_IN_BYTES / p_s_sb->s_blocksize +
2266 SB_BMAP_NR(p_s_sb) + 1 :
2267 REISERFS_DISK_OFFSET_IN_BYTES / p_s_sb->s_blocksize + 2);
2269 /* Sanity check to see is the standard journal fitting withing first bitmap
2270 (actual for small blocksizes) */
2271 if ( !SB_ONDISK_JOURNAL_DEVICE( p_s_sb ) &&
2272 (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) + SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8) ) {
2273 printk("journal-1393: journal does not fit for area addressed by first of bitmap blocks. "
2274 "It starts at %u and its size is %u. Block size %ld\n",
2275 SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb), SB_ONDISK_JOURNAL_SIZE(p_s_sb), p_s_sb->s_blocksize) ;
2276 goto free_and_return;
2279 if( journal_init_dev( p_s_sb, journal, j_dev_name ) != 0 ) {
2280 printk( "sh-462: unable to initialize jornal device\n");
2281 goto free_and_return;
2284 rs = SB_DISK_SUPER_BLOCK(p_s_sb);
2286 /* read journal header */
2287 bhjh = journal_bread(p_s_sb,
2288 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2290 printk("sh-459: unable to read journal header\n") ;
2291 goto free_and_return;
2293 jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2295 /* make sure that journal matches to the super block */
2296 if (is_reiserfs_jr(rs) && (jh->jh_journal.jp_journal_magic != sb_jp_journal_magic(rs))) {
2297 printk("sh-460: journal header magic %x (device %s) does not match "
2298 "to magic found in super block %x (device %s)\n",
2299 jh->jh_journal.jp_journal_magic,
2300 bdevname( SB_JOURNAL(p_s_sb)->j_dev_bd, b),
2301 sb_jp_journal_magic(rs), reiserfs_bdevname (p_s_sb));
2303 goto free_and_return;
2306 SB_JOURNAL_TRANS_MAX(p_s_sb) = le32_to_cpu (jh->jh_journal.jp_journal_trans_max);
2307 SB_JOURNAL_MAX_BATCH(p_s_sb) = le32_to_cpu (jh->jh_journal.jp_journal_max_batch);
2308 SB_JOURNAL_MAX_COMMIT_AGE(p_s_sb) = le32_to_cpu (jh->jh_journal.jp_journal_max_commit_age);
2309 SB_JOURNAL_MAX_TRANS_AGE(p_s_sb) = JOURNAL_MAX_TRANS_AGE;
2311 if (SB_JOURNAL_TRANS_MAX(p_s_sb)) {
2312 /* make sure these parameters are available, assign it if they are not */
2313 __u32 initial = SB_JOURNAL_TRANS_MAX(p_s_sb);
2316 if (p_s_sb->s_blocksize < 4096)
2317 ratio = 4096 / p_s_sb->s_blocksize;
2319 if (SB_ONDISK_JOURNAL_SIZE(p_s_sb)/SB_JOURNAL_TRANS_MAX(p_s_sb) < JOURNAL_MIN_RATIO)
2320 SB_JOURNAL_TRANS_MAX(p_s_sb) = SB_ONDISK_JOURNAL_SIZE(p_s_sb) / JOURNAL_MIN_RATIO;
2321 if (SB_JOURNAL_TRANS_MAX(p_s_sb) > JOURNAL_TRANS_MAX_DEFAULT / ratio)
2322 SB_JOURNAL_TRANS_MAX(p_s_sb) = JOURNAL_TRANS_MAX_DEFAULT / ratio;
2323 if (SB_JOURNAL_TRANS_MAX(p_s_sb) < JOURNAL_TRANS_MIN_DEFAULT / ratio)
2324 SB_JOURNAL_TRANS_MAX(p_s_sb) = JOURNAL_TRANS_MIN_DEFAULT / ratio;
2326 if (SB_JOURNAL_TRANS_MAX(p_s_sb) != initial)
2327 printk ("sh-461: journal_init: wrong transaction max size (%u). Changed to %u\n",
2328 initial, SB_JOURNAL_TRANS_MAX(p_s_sb));
2330 SB_JOURNAL_MAX_BATCH(p_s_sb) = SB_JOURNAL_TRANS_MAX(p_s_sb)*
2331 JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT;
2334 if (!SB_JOURNAL_TRANS_MAX(p_s_sb)) {
2335 /*we have the file system was created by old version of mkreiserfs
2336 so this field contains zero value */
2337 SB_JOURNAL_TRANS_MAX(p_s_sb) = JOURNAL_TRANS_MAX_DEFAULT ;
2338 SB_JOURNAL_MAX_BATCH(p_s_sb) = JOURNAL_MAX_BATCH_DEFAULT ;
2339 SB_JOURNAL_MAX_COMMIT_AGE(p_s_sb) = JOURNAL_MAX_COMMIT_AGE ;
2341 /* for blocksize >= 4096 - max transaction size is 1024. For block size < 4096
2342 trans max size is decreased proportionally */
2343 if (p_s_sb->s_blocksize < 4096) {
2344 SB_JOURNAL_TRANS_MAX(p_s_sb) /= (4096 / p_s_sb->s_blocksize) ;
2345 SB_JOURNAL_MAX_BATCH(p_s_sb) = (SB_JOURNAL_TRANS_MAX(p_s_sb)) * 9 / 10 ;
2349 SB_JOURNAL_DEFAULT_MAX_COMMIT_AGE(p_s_sb) = SB_JOURNAL_MAX_COMMIT_AGE(p_s_sb);
2351 if (commit_max_age != 0) {
2352 SB_JOURNAL_MAX_COMMIT_AGE(p_s_sb) = commit_max_age;
2353 SB_JOURNAL_MAX_TRANS_AGE(p_s_sb) = commit_max_age;
2356 printk ("Reiserfs journal params: device %s, size %u, "
2357 "journal first block %u, max trans len %u, max batch %u, "
2358 "max commit age %u, max trans age %u\n",
2359 bdevname( SB_JOURNAL(p_s_sb)->j_dev_bd, b),
2360 SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2361 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2362 SB_JOURNAL_TRANS_MAX(p_s_sb),
2363 SB_JOURNAL_MAX_BATCH(p_s_sb),
2364 SB_JOURNAL_MAX_COMMIT_AGE(p_s_sb),
2365 SB_JOURNAL_MAX_TRANS_AGE(p_s_sb));
2369 SB_JOURNAL(p_s_sb)->j_list_bitmap_index = 0 ;
2370 journal_list_init(p_s_sb) ;
2372 memset(SB_JOURNAL(p_s_sb)->j_list_hash_table, 0, JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)) ;
2374 INIT_LIST_HEAD(&SB_JOURNAL(p_s_sb)->j_dirty_buffers) ;
2375 spin_lock_init(&SB_JOURNAL(p_s_sb)->j_dirty_buffers_lock) ;
2377 SB_JOURNAL(p_s_sb)->j_start = 0 ;
2378 SB_JOURNAL(p_s_sb)->j_len = 0 ;
2379 SB_JOURNAL(p_s_sb)->j_len_alloc = 0 ;
2380 atomic_set(&(SB_JOURNAL(p_s_sb)->j_wcount), 0) ;
2381 atomic_set(&(SB_JOURNAL(p_s_sb)->j_async_throttle), 0) ;
2382 SB_JOURNAL(p_s_sb)->j_bcount = 0 ;
2383 SB_JOURNAL(p_s_sb)->j_trans_start_time = 0 ;
2384 SB_JOURNAL(p_s_sb)->j_last = NULL ;
2385 SB_JOURNAL(p_s_sb)->j_first = NULL ;
2386 init_waitqueue_head(&(SB_JOURNAL(p_s_sb)->j_join_wait)) ;
2387 sema_init(&SB_JOURNAL(p_s_sb)->j_lock, 1);
2388 sema_init(&SB_JOURNAL(p_s_sb)->j_flush_sem, 1);
2390 SB_JOURNAL(p_s_sb)->j_trans_id = 10 ;
2391 SB_JOURNAL(p_s_sb)->j_mount_id = 10 ;
2392 SB_JOURNAL(p_s_sb)->j_state = 0 ;
2393 atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 0) ;
2394 SB_JOURNAL(p_s_sb)->j_cnode_free_list = allocate_cnodes(num_cnodes) ;
2395 SB_JOURNAL(p_s_sb)->j_cnode_free_orig = SB_JOURNAL(p_s_sb)->j_cnode_free_list ;
2396 SB_JOURNAL(p_s_sb)->j_cnode_free = SB_JOURNAL(p_s_sb)->j_cnode_free_list ? num_cnodes : 0 ;
2397 SB_JOURNAL(p_s_sb)->j_cnode_used = 0 ;
2398 SB_JOURNAL(p_s_sb)->j_must_wait = 0 ;
2400 init_journal_hash(p_s_sb) ;
2401 jl = SB_JOURNAL(p_s_sb)->j_current_jl;
2402 jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl);
2403 if (!jl->j_list_bitmap) {
2404 reiserfs_warning("journal-2005, get_list_bitmap failed for journal list 0\n") ;
2405 goto free_and_return;
2407 if (journal_read(p_s_sb) < 0) {
2408 reiserfs_warning("Replay Failure, unable to mount\n") ;
2409 goto free_and_return;
2412 reiserfs_mounted_fs_count++ ;
2413 if (reiserfs_mounted_fs_count <= 1)
2414 commit_wq = create_workqueue("reiserfs");
2416 INIT_WORK(&journal->j_work, flush_async_commits, p_s_sb);
2419 free_journal_ram(p_s_sb);
2424 ** test for a polite end of the current transaction. Used by file_write, and should
2425 ** be used by delete to make sure they don't write more than can fit inside a single
2428 int journal_transaction_should_end(struct reiserfs_transaction_handle *th, int new_alloc) {
2429 time_t now = get_seconds() ;
2430 /* cannot restart while nested */
2431 if (th->t_refcount > 1)
2433 if ( SB_JOURNAL(th->t_super)->j_must_wait > 0 ||
2434 (SB_JOURNAL(th->t_super)->j_len_alloc + new_alloc) >= SB_JOURNAL_MAX_BATCH(th->t_super) ||
2435 atomic_read(&(SB_JOURNAL(th->t_super)->j_jlock)) ||
2436 (now - SB_JOURNAL(th->t_super)->j_trans_start_time) > SB_JOURNAL_MAX_TRANS_AGE(th->t_super) ||
2437 SB_JOURNAL(th->t_super)->j_cnode_free < (SB_JOURNAL_TRANS_MAX(th->t_super) * 3)) {
2443 /* this must be called inside a transaction, and requires the
2444 ** kernel_lock to be held
2446 void reiserfs_block_writes(struct reiserfs_transaction_handle *th) {
2447 struct super_block *s = th->t_super ;
2448 SB_JOURNAL(s)->j_must_wait = 1 ;
2449 set_bit(WRITERS_BLOCKED, &SB_JOURNAL(s)->j_state) ;
2453 /* this must be called without a transaction started, and does not
2456 void reiserfs_allow_writes(struct super_block *s) {
2457 clear_bit(WRITERS_BLOCKED, &SB_JOURNAL(s)->j_state) ;
2458 wake_up(&SB_JOURNAL(s)->j_join_wait) ;
2461 /* this must be called without a transaction started, and does not
2464 void reiserfs_wait_on_write_block(struct super_block *s) {
2465 wait_event(SB_JOURNAL(s)->j_join_wait,
2466 !test_bit(WRITERS_BLOCKED, &SB_JOURNAL(s)->j_state)) ;
2469 static void queue_log_writer(struct super_block *s) {
2471 set_bit(WRITERS_QUEUED, &SB_JOURNAL(s)->j_state);
2474 * we don't want to use wait_event here because
2475 * we only want to wait once.
2477 init_waitqueue_entry(&wait, current);
2478 add_wait_queue(&SB_JOURNAL(s)->j_join_wait, &wait);
2479 set_current_state(TASK_UNINTERRUPTIBLE);
2480 if (test_bit(WRITERS_QUEUED, &SB_JOURNAL(s)->j_state))
2482 current->state = TASK_RUNNING;
2483 remove_wait_queue(&SB_JOURNAL(s)->j_join_wait, &wait);
2486 static void wake_queued_writers(struct super_block *s) {
2487 if (test_and_clear_bit(WRITERS_QUEUED, &SB_JOURNAL(s)->j_state))
2488 wake_up(&SB_JOURNAL(s)->j_join_wait);
2491 static void let_transaction_grow(struct super_block *sb,
2492 unsigned long trans_id)
2494 unsigned long bcount = SB_JOURNAL(sb)->j_bcount;
2496 set_current_state(TASK_UNINTERRUPTIBLE);
2497 schedule_timeout(1);
2498 SB_JOURNAL(sb)->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2499 while ((atomic_read(&SB_JOURNAL(sb)->j_wcount) > 0 ||
2500 atomic_read(&SB_JOURNAL(sb)->j_jlock)) &&
2501 SB_JOURNAL(sb)->j_trans_id == trans_id) {
2502 queue_log_writer(sb);
2504 if (SB_JOURNAL(sb)->j_trans_id != trans_id)
2506 if (bcount == SB_JOURNAL(sb)->j_bcount)
2508 bcount = SB_JOURNAL(sb)->j_bcount;
2512 /* join == true if you must join an existing transaction.
2513 ** join == false if you can deal with waiting for others to finish
2515 ** this will block until the transaction is joinable. send the number of blocks you
2516 ** expect to use in nblocks.
2518 static int do_journal_begin_r(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb,unsigned long nblocks,int join) {
2519 time_t now = get_seconds() ;
2521 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2522 struct reiserfs_transaction_handle myth;
2523 int sched_count = 0;
2525 reiserfs_check_lock_depth("journal_begin") ;
2526 RFALSE( p_s_sb->s_flags & MS_RDONLY,
2527 "clm-2078: calling journal_begin on readonly FS") ;
2529 PROC_INFO_INC( p_s_sb, journal.journal_being );
2530 /* set here for journal_join */
2532 th->t_super = p_s_sb ;
2535 lock_journal(p_s_sb) ;
2536 journal->j_bcount++;
2538 if (test_bit(WRITERS_BLOCKED, &journal->j_state)) {
2539 unlock_journal(p_s_sb) ;
2540 reiserfs_wait_on_write_block(p_s_sb) ;
2541 PROC_INFO_INC( p_s_sb, journal.journal_relock_writers );
2544 now = get_seconds();
2546 /* if there is no room in the journal OR
2547 ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
2548 ** we don't sleep if there aren't other writers
2551 if ( (!join && journal->j_must_wait > 0) ||
2552 ( !join && (journal->j_len_alloc + nblocks + 2) >= SB_JOURNAL_MAX_BATCH(p_s_sb)) ||
2553 (!join && atomic_read(&journal->j_wcount) > 0 && journal->j_trans_start_time > 0 &&
2554 (now - journal->j_trans_start_time) > SB_JOURNAL_MAX_TRANS_AGE(p_s_sb)) ||
2555 (!join && atomic_read(&journal->j_jlock)) ||
2556 (!join && journal->j_cnode_free < (SB_JOURNAL_TRANS_MAX(p_s_sb) * 3))) {
2558 old_trans_id = journal->j_trans_id;
2559 unlock_journal(p_s_sb) ; /* allow others to finish this transaction */
2561 if (!join && (journal->j_len_alloc + nblocks + 2) >=
2562 SB_JOURNAL_MAX_BATCH(p_s_sb) &&
2563 ((journal->j_len + nblocks + 2) * 100) < (journal->j_len_alloc * 75))
2565 if (atomic_read(&journal->j_wcount) > 10) {
2567 queue_log_writer(p_s_sb);
2571 /* don't mess with joining the transaction if all we have to do is
2572 * wait for someone else to do a commit
2574 if (atomic_read(&journal->j_jlock)) {
2575 while (journal->j_trans_id == old_trans_id &&
2576 atomic_read(&journal->j_jlock)) {
2577 queue_log_writer(p_s_sb);
2581 journal_join(&myth, p_s_sb, 1) ;
2583 /* someone might have ended the transaction while we joined */
2584 if (old_trans_id != SB_JOURNAL(p_s_sb)->j_trans_id) {
2585 do_journal_end(&myth, p_s_sb, 1, 0) ;
2587 do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW) ;
2590 PROC_INFO_INC( p_s_sb, journal.journal_relock_wcount );
2593 /* we are the first writer, set trans_id */
2594 if (journal->j_trans_start_time == 0) {
2595 journal->j_trans_start_time = get_seconds();
2597 atomic_inc(&(journal->j_wcount)) ;
2598 journal->j_len_alloc += nblocks ;
2599 th->t_blocks_logged = 0 ;
2600 th->t_blocks_allocated = nblocks ;
2601 th->t_trans_id = journal->j_trans_id ;
2602 unlock_journal(p_s_sb) ;
2606 struct reiserfs_transaction_handle *
2607 reiserfs_persistent_transaction(struct super_block *s, int nblocks) {
2609 struct reiserfs_transaction_handle *th ;
2611 /* if we're nesting into an existing transaction. It will be
2612 ** persistent on its own
2614 if (reiserfs_transaction_running(s)) {
2615 th = current->journal_info ;
2617 if (th->t_refcount < 2) {
2622 th = reiserfs_kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS, s) ;
2625 ret = journal_begin(th, s, nblocks) ;
2627 reiserfs_kfree(th, sizeof(struct reiserfs_transaction_handle), s) ;
2634 reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th) {
2635 struct super_block *s = th->t_super;
2637 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
2638 if (th->t_refcount == 0)
2639 reiserfs_kfree(th, sizeof(struct reiserfs_transaction_handle), s) ;
2643 static int journal_join(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) {
2644 struct reiserfs_transaction_handle *cur_th = current->journal_info;
2646 /* this keeps do_journal_end from NULLing out the current->journal_info
2649 th->t_handle_save = cur_th ;
2650 if (cur_th && cur_th->t_refcount > 1) {
2653 return do_journal_begin_r(th, p_s_sb, nblocks, 1) ;
2656 int journal_begin(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb, unsigned long nblocks) {
2657 struct reiserfs_transaction_handle *cur_th = current->journal_info ;
2660 th->t_handle_save = NULL ;
2662 /* we are nesting into the current transaction */
2663 if (cur_th->t_super == p_s_sb) {
2664 cur_th->t_refcount++ ;
2665 memcpy(th, cur_th, sizeof(*th));
2666 if (th->t_refcount <= 1)
2667 printk("BAD: refcount <= 1, but journal_info != 0\n");
2670 /* we've ended up with a handle from a different filesystem.
2671 ** save it and restore on journal_end. This should never
2674 reiserfs_warning("clm-2100: nesting info a different FS\n") ;
2675 th->t_handle_save = current->journal_info ;
2676 current->journal_info = th;
2679 current->journal_info = th;
2681 ret = do_journal_begin_r(th, p_s_sb, nblocks, 0) ;
2682 if (current->journal_info != th)
2688 ** puts bh into the current transaction. If it was already there, reorders removes the
2689 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
2691 ** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the
2692 ** transaction is committed.
2694 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
2696 int journal_mark_dirty(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, struct buffer_head *bh) {
2697 struct reiserfs_journal_cnode *cn = NULL;
2698 int count_already_incd = 0 ;
2701 PROC_INFO_INC( p_s_sb, journal.mark_dirty );
2702 if (th->t_trans_id != SB_JOURNAL(p_s_sb)->j_trans_id) {
2703 reiserfs_panic(th->t_super, "journal-1577: handle trans id %ld != current trans id %ld\n",
2704 th->t_trans_id, SB_JOURNAL(p_s_sb)->j_trans_id);
2708 prepared = test_and_clear_bit(BH_JPrepared, &bh->b_state) ;
2709 clear_bit(BH_JRestore_dirty, &bh->b_state);
2710 /* already in this transaction, we are done */
2711 if (buffer_journaled(bh)) {
2712 PROC_INFO_INC( p_s_sb, journal.mark_dirty_already );
2716 /* this must be turned into a panic instead of a warning. We can't allow
2717 ** a dirty or journal_dirty or locked buffer to be logged, as some changes
2718 ** could get to disk too early. NOT GOOD.
2720 if (!prepared || buffer_dirty(bh)) {
2721 printk("journal-1777: buffer %llu bad state %cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT\n", (unsigned long long)bh->b_blocknr, prepared ? ' ' : '!',
2722 buffer_locked(bh) ? ' ' : '!',
2723 buffer_dirty(bh) ? ' ' : '!',
2724 buffer_journal_dirty(bh) ? ' ' : '!') ;
2727 if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) <= 0) {
2728 printk("journal-1409: journal_mark_dirty returning because j_wcount was %d\n", atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount))) ;
2731 /* this error means I've screwed up, and we've overflowed the transaction.
2732 ** Nothing can be done here, except make the FS readonly or panic.
2734 if (SB_JOURNAL(p_s_sb)->j_len >= SB_JOURNAL_TRANS_MAX(p_s_sb)) {
2735 reiserfs_panic(th->t_super, "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n", SB_JOURNAL(p_s_sb)->j_len) ;
2738 if (buffer_journal_dirty(bh)) {
2739 count_already_incd = 1 ;
2740 PROC_INFO_INC( p_s_sb, journal.mark_dirty_notjournal );
2741 mark_buffer_notjournal_dirty(bh) ;
2744 if (SB_JOURNAL(p_s_sb)->j_len > SB_JOURNAL(p_s_sb)->j_len_alloc) {
2745 SB_JOURNAL(p_s_sb)->j_len_alloc = SB_JOURNAL(p_s_sb)->j_len + JOURNAL_PER_BALANCE_CNT ;
2748 set_bit(BH_JDirty, &bh->b_state) ;
2750 /* now put this guy on the end */
2752 cn = get_cnode(p_s_sb) ;
2754 reiserfs_panic(p_s_sb, "get_cnode failed!\n");
2757 if (th->t_blocks_logged == th->t_blocks_allocated) {
2758 th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT ;
2759 SB_JOURNAL(p_s_sb)->j_len_alloc += JOURNAL_PER_BALANCE_CNT ;
2761 th->t_blocks_logged++ ;
2762 SB_JOURNAL(p_s_sb)->j_len++ ;
2765 cn->blocknr = bh->b_blocknr ;
2768 insert_journal_hash(SB_JOURNAL(p_s_sb)->j_hash_table, cn) ;
2769 if (!count_already_incd) {
2774 cn->prev = SB_JOURNAL(p_s_sb)->j_last ;
2776 if (SB_JOURNAL(p_s_sb)->j_last) {
2777 SB_JOURNAL(p_s_sb)->j_last->next = cn ;
2778 SB_JOURNAL(p_s_sb)->j_last = cn ;
2780 SB_JOURNAL(p_s_sb)->j_first = cn ;
2781 SB_JOURNAL(p_s_sb)->j_last = cn ;
2786 int journal_end(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) {
2787 if (!current->journal_info && th->t_refcount > 1)
2788 printk("REISER-NESTING: th NULL, refcount %d\n", th->t_refcount);
2791 if (th->t_refcount > 0) {
2792 struct reiserfs_transaction_handle *cur_th = current->journal_info ;
2794 /* we aren't allowed to close a nested transaction on a different
2795 ** filesystem from the one in the task struct
2797 if (cur_th->t_super != th->t_super)
2801 memcpy(current->journal_info, th, sizeof(*th));
2806 return do_journal_end(th, p_s_sb, nblocks, 0) ;
2810 /* removes from the current transaction, relsing and descrementing any counters.
2811 ** also files the removed buffer directly onto the clean list
2813 ** called by journal_mark_freed when a block has been deleted
2815 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
2817 static int remove_from_transaction(struct super_block *p_s_sb, b_blocknr_t blocknr, int already_cleaned) {
2818 struct buffer_head *bh ;
2819 struct reiserfs_journal_cnode *cn ;
2822 cn = get_journal_hash_dev(p_s_sb, SB_JOURNAL(p_s_sb)->j_hash_table, blocknr) ;
2823 if (!cn || !cn->bh) {
2828 cn->prev->next = cn->next ;
2831 cn->next->prev = cn->prev ;
2833 if (cn == SB_JOURNAL(p_s_sb)->j_first) {
2834 SB_JOURNAL(p_s_sb)->j_first = cn->next ;
2836 if (cn == SB_JOURNAL(p_s_sb)->j_last) {
2837 SB_JOURNAL(p_s_sb)->j_last = cn->prev ;
2840 remove_journal_hash(p_s_sb, SB_JOURNAL(p_s_sb)->j_hash_table, NULL, bh->b_blocknr, 0) ;
2841 mark_buffer_not_journaled(bh) ; /* don't log this one */
2843 if (!already_cleaned) {
2844 mark_buffer_notjournal_dirty(bh) ;
2846 if (atomic_read(&(bh->b_count)) < 0) {
2847 printk("journal-1752: remove from trans, b_count < 0\n") ;
2851 SB_JOURNAL(p_s_sb)->j_len-- ;
2852 SB_JOURNAL(p_s_sb)->j_len_alloc-- ;
2853 free_cnode(p_s_sb, cn) ;
2858 ** for any cnode in a journal list, it can only be dirtied of all the
2859 ** transactions that include it are commited to disk.
2860 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
2861 ** and 0 if you aren't
2863 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
2864 ** blocks for a given transaction on disk
2867 static int can_dirty(struct reiserfs_journal_cnode *cn) {
2868 struct super_block *sb = cn->sb;
2869 b_blocknr_t blocknr = cn->blocknr ;
2870 struct reiserfs_journal_cnode *cur = cn->hprev ;
2873 /* first test hprev. These are all newer than cn, so any node here
2874 ** with the same block number and dev means this node can't be sent
2875 ** to disk right now.
2877 while(cur && can_dirty) {
2878 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
2879 cur->blocknr == blocknr) {
2884 /* then test hnext. These are all older than cn. As long as they
2885 ** are committed to the log, it is safe to write cn to disk
2888 while(cur && can_dirty) {
2889 if (cur->jlist && cur->jlist->j_len > 0 &&
2890 atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
2891 cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
2899 /* syncs the commit blocks, but does not force the real buffers to disk
2900 ** will wait until the current transaction is done/commited before returning
2902 int journal_end_sync(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) {
2904 /* you can sync while nested, very, very bad */
2905 if (th->t_refcount > 1) {
2908 if (SB_JOURNAL(p_s_sb)->j_len == 0) {
2909 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
2910 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
2912 return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT) ;
2916 ** writeback the pending async commits to disk
2918 static void flush_async_commits(void *p) {
2919 struct super_block *p_s_sb = p;
2920 struct reiserfs_journal_list *jl;
2921 struct list_head *entry;
2924 if (!list_empty(&SB_JOURNAL(p_s_sb)->j_journal_list)) {
2925 /* last entry is the youngest, commit it and you get everything */
2926 entry = SB_JOURNAL(p_s_sb)->j_journal_list.prev;
2927 jl = JOURNAL_LIST_ENTRY(entry);
2928 flush_commit_list(p_s_sb, jl, 1);
2932 * this is a little racey, but there's no harm in missing
2933 * the filemap_fdata_write
2935 if (!atomic_read(&SB_JOURNAL(p_s_sb)->j_async_throttle)) {
2936 atomic_inc(&SB_JOURNAL(p_s_sb)->j_async_throttle);
2937 filemap_fdatawrite(p_s_sb->s_bdev->bd_inode->i_mapping);
2938 atomic_dec(&SB_JOURNAL(p_s_sb)->j_async_throttle);
2943 ** flushes any old transactions to disk
2944 ** ends the current transaction if it is too old
2946 int reiserfs_flush_old_commits(struct super_block *p_s_sb) {
2948 struct reiserfs_transaction_handle th ;
2950 now = get_seconds();
2951 /* safety check so we don't flush while we are replaying the log during
2954 if (list_empty(&SB_JOURNAL(p_s_sb)->j_journal_list)) {
2958 /* check the current transaction. If there are no writers, and it is
2959 * too old, finish it, and force the commit blocks to disk
2961 if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) <= 0 &&
2962 SB_JOURNAL(p_s_sb)->j_trans_start_time > 0 &&
2963 SB_JOURNAL(p_s_sb)->j_len > 0 &&
2964 (now - SB_JOURNAL(p_s_sb)->j_trans_start_time) >
2965 SB_JOURNAL_MAX_TRANS_AGE(p_s_sb))
2967 journal_join(&th, p_s_sb, 1) ;
2968 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
2969 journal_mark_dirty(&th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
2971 /* we're only being called from kreiserfsd, it makes no sense to do
2972 ** an async commit so that kreiserfsd can do it later
2974 do_journal_end(&th, p_s_sb,1, COMMIT_NOW | WAIT) ;
2976 return p_s_sb->s_dirt;
2980 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
2982 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
2983 ** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just
2984 ** flushes the commit list and returns 0.
2986 ** Won't batch when flush or commit_now is set. Also won't batch when others are waiting on j_join_wait.
2988 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
2990 static int check_journal_end(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb,
2991 unsigned long nblocks, int flags) {
2994 int flush = flags & FLUSH_ALL ;
2995 int commit_now = flags & COMMIT_NOW ;
2996 int wait_on_commit = flags & WAIT ;
2997 struct reiserfs_journal_list *jl;
2999 if (th->t_trans_id != SB_JOURNAL(p_s_sb)->j_trans_id) {
3000 reiserfs_panic(th->t_super, "journal-1577: handle trans id %ld != current trans id %ld\n",
3001 th->t_trans_id, SB_JOURNAL(p_s_sb)->j_trans_id);
3004 SB_JOURNAL(p_s_sb)->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged) ;
3005 if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) > 0) { /* <= 0 is allowed. unmounting might not call begin */
3006 atomic_dec(&(SB_JOURNAL(p_s_sb)->j_wcount)) ;
3009 /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3010 ** will be dealt with by next transaction that actually writes something, but should be taken
3011 ** care of in this trans
3013 if (SB_JOURNAL(p_s_sb)->j_len == 0) {
3016 /* if wcount > 0, and we are called to with flush or commit_now,
3017 ** we wait on j_join_wait. We will wake up when the last writer has
3018 ** finished the transaction, and started it on its way to the disk.
3019 ** Then, we flush the commit or journal list, and just return 0
3020 ** because the rest of journal end was already done for this transaction.
3022 if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) > 0) {
3023 if (flush || commit_now) {
3026 jl = SB_JOURNAL(p_s_sb)->j_current_jl;
3027 trans_id = jl->j_trans_id;
3029 jl->j_state |= LIST_COMMIT_PENDING;
3030 atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 1) ;
3032 SB_JOURNAL(p_s_sb)->j_next_full_flush = 1 ;
3034 unlock_journal(p_s_sb) ;
3036 /* sleep while the current transaction is still j_jlocked */
3037 while(SB_JOURNAL(p_s_sb)->j_trans_id == trans_id) {
3038 if (atomic_read(&SB_JOURNAL(p_s_sb)->j_jlock)) {
3039 queue_log_writer(p_s_sb);
3041 lock_journal(p_s_sb);
3042 if (SB_JOURNAL(p_s_sb)->j_trans_id == trans_id) {
3043 atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 1) ;
3045 unlock_journal(p_s_sb);
3048 if (SB_JOURNAL(p_s_sb)->j_trans_id == trans_id) {
3051 if (commit_now && journal_list_still_alive(p_s_sb, trans_id) &&
3054 flush_commit_list(p_s_sb, jl, 1) ;
3058 unlock_journal(p_s_sb) ;
3062 /* deal with old transactions where we are the last writers */
3063 now = get_seconds();
3064 if ((now - SB_JOURNAL(p_s_sb)->j_trans_start_time) > SB_JOURNAL_MAX_TRANS_AGE(p_s_sb)) {
3066 SB_JOURNAL(p_s_sb)->j_next_async_flush = 1 ;
3068 /* don't batch when someone is waiting on j_join_wait */
3069 /* don't batch when syncing the commit or flushing the whole trans */
3070 if (!(SB_JOURNAL(p_s_sb)->j_must_wait > 0) && !(atomic_read(&(SB_JOURNAL(p_s_sb)->j_jlock))) && !flush && !commit_now &&
3071 (SB_JOURNAL(p_s_sb)->j_len < SB_JOURNAL_MAX_BATCH(p_s_sb)) &&
3072 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)) {
3073 SB_JOURNAL(p_s_sb)->j_bcount++ ;
3074 unlock_journal(p_s_sb) ;
3078 if (SB_JOURNAL(p_s_sb)->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
3079 reiserfs_panic(p_s_sb, "journal-003: journal_end: j_start (%ld) is too high\n", SB_JOURNAL(p_s_sb)->j_start) ;
3085 ** Does all the work that makes deleting blocks safe.
3086 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3089 ** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes
3090 ** before this transaction has finished.
3092 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with
3093 ** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash,
3094 ** the block can't be reallocated yet.
3096 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3098 int journal_mark_freed(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, b_blocknr_t blocknr) {
3099 struct reiserfs_journal_cnode *cn = NULL ;
3100 struct buffer_head *bh = NULL ;
3101 struct reiserfs_list_bitmap *jb = NULL ;
3104 cn = get_journal_hash_dev(p_s_sb, SB_JOURNAL(p_s_sb)->j_hash_table, blocknr);
3109 /* if it is journal new, we just remove it from this transaction */
3110 if (bh && buffer_journal_new(bh)) {
3111 mark_buffer_notjournal_new(bh) ;
3112 clear_prepared_bits(bh) ;
3113 reiserfs_clean_and_file_buffer(bh) ;
3114 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned) ;
3116 /* set the bit for this block in the journal bitmap for this transaction */
3117 jb = SB_JOURNAL(p_s_sb)->j_current_jl->j_list_bitmap;
3119 reiserfs_panic(p_s_sb, "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n") ;
3121 set_bit_in_list_bitmap(p_s_sb, blocknr, jb) ;
3123 /* Note, the entire while loop is not allowed to schedule. */
3126 clear_prepared_bits(bh) ;
3127 reiserfs_clean_and_file_buffer(bh) ;
3129 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned) ;
3131 /* find all older transactions with this block, make sure they don't try to write it out */
3132 cn = get_journal_hash_dev(p_s_sb,SB_JOURNAL(p_s_sb)->j_list_hash_table, blocknr) ;
3134 if (p_s_sb == cn->sb && blocknr == cn->blocknr) {
3135 set_bit(BLOCK_FREED, &cn->state) ;
3138 /* remove_from_transaction will brelse the buffer if it was
3139 ** in the current trans
3141 mark_buffer_notjournal_dirty(cn->bh) ;
3144 if (atomic_read(&(cn->bh->b_count)) < 0) {
3145 printk("journal-2138: cn->bh->b_count < 0\n") ;
3148 if (cn->jlist) { /* since we are clearing the bh, we MUST dec nonzerolen */
3149 atomic_dec(&(cn->jlist->j_nonzerolen)) ;
3159 put_bh(bh) ; /* get_hash grabs the buffer */
3160 if (atomic_read(&(bh->b_count)) < 0) {
3161 printk("journal-2165: bh->b_count < 0\n") ;
3167 void reiserfs_update_inode_transaction(struct inode *inode) {
3168 REISERFS_I(inode)->i_jl = SB_JOURNAL(inode->i_sb)->j_current_jl;
3169 REISERFS_I(inode)->i_trans_id = SB_JOURNAL(inode->i_sb)->j_trans_id ;
3172 static void __commit_trans_jl(struct inode *inode, unsigned long id,
3173 struct reiserfs_journal_list *jl)
3175 struct reiserfs_transaction_handle th ;
3176 struct super_block *sb = inode->i_sb ;
3178 /* is it from the current transaction, or from an unknown transaction? */
3179 if (id == SB_JOURNAL(sb)->j_trans_id) {
3180 jl = SB_JOURNAL(sb)->j_current_jl;
3181 /* try to let other writers come in and grow this transaction */
3182 let_transaction_grow(sb, id);
3183 if (SB_JOURNAL(sb)->j_trans_id != id) {
3184 goto flush_commit_only;
3187 journal_begin(&th, sb, 1) ;
3189 /* someone might have ended this transaction while we joined */
3190 if (SB_JOURNAL(sb)->j_trans_id != id) {
3191 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 1) ;
3192 journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb)) ;
3193 journal_end(&th, sb, 1) ;
3194 goto flush_commit_only;
3197 journal_end_sync(&th, sb, 1) ;
3200 /* this gets tricky, we have to make sure the journal list in
3201 * the inode still exists. We know the list is still around
3202 * if we've got a larger transaction id than the oldest list
3205 if (journal_list_still_alive(inode->i_sb, id)) {
3206 flush_commit_list(sb, jl, 1) ;
3209 /* otherwise the list is gone, and long since committed */
3212 void reiserfs_commit_for_inode(struct inode *inode) {
3213 unsigned long id = REISERFS_I(inode)->i_trans_id;
3214 struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3216 /* for the whole inode, assume unset id means it was
3217 * changed in the current transaction. More conservative
3220 reiserfs_update_inode_transaction(inode) ;
3221 id = REISERFS_I(inode)->i_trans_id;
3222 /* jl will be updated in __commit_trans_jl */
3225 __commit_trans_jl(inode, id, jl);
3228 void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb,
3229 struct buffer_head *bh) {
3230 PROC_INFO_INC( p_s_sb, journal.restore_prepared );
3234 if (test_and_clear_bit(BH_JRestore_dirty, &bh->b_state) &&
3235 buffer_journal_dirty(bh)) {
3236 struct reiserfs_journal_cnode *cn;
3237 cn = get_journal_hash_dev(p_s_sb,
3238 SB_JOURNAL(p_s_sb)->j_list_hash_table,
3240 if (cn && can_dirty(cn)) {
3241 set_bit(BH_JTest, &bh->b_state);
3242 mark_buffer_dirty(bh);
3245 clear_bit(BH_JPrepared, &bh->b_state) ;
3248 extern struct tree_balance *cur_tb ;
3250 ** before we can change a metadata block, we have to make sure it won't
3251 ** be written to disk while we are altering it. So, we must:
3256 int reiserfs_prepare_for_journal(struct super_block *p_s_sb,
3257 struct buffer_head *bh, int wait) {
3258 PROC_INFO_INC( p_s_sb, journal.prepare );
3260 if (test_set_buffer_locked(bh)) {
3265 set_bit(BH_JPrepared, &bh->b_state);
3266 if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3267 clear_bit(BH_JTest, &bh->b_state);
3268 set_bit(BH_JRestore_dirty, &bh->b_state);
3274 static void flush_old_journal_lists(struct super_block *s) {
3275 struct reiserfs_journal_list *jl;
3276 struct list_head *entry;
3277 time_t now = get_seconds();
3279 while(!list_empty(&SB_JOURNAL(s)->j_journal_list)) {
3280 entry = SB_JOURNAL(s)->j_journal_list.next;
3281 jl = JOURNAL_LIST_ENTRY(entry);
3282 /* this check should always be run, to send old lists to disk */
3283 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4))) {
3284 flush_used_journal_lists(s, jl);
3292 ** long and ugly. If flush, will not return until all commit
3293 ** blocks and all real buffers in the trans are on disk.
3294 ** If no_async, won't return until all commit blocks are on disk.
3296 ** keep reading, there are comments as you go along
3298 static int do_journal_end(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb, unsigned long nblocks,
3300 struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3301 struct reiserfs_journal_cnode *last_cn = NULL;
3302 struct reiserfs_journal_desc *desc ;
3303 struct reiserfs_journal_commit *commit ;
3304 struct buffer_head *c_bh ; /* commit bh */
3305 struct buffer_head *d_bh ; /* desc bh */
3306 int cur_write_start = 0 ; /* start index of current log write */
3309 int flush = flags & FLUSH_ALL ;
3310 int wait_on_commit = flags & WAIT ;
3311 struct reiserfs_journal_list *jl, *temp_jl;
3312 struct list_head *entry, *safe;
3313 unsigned long jindex;
3314 unsigned long commit_trans_id;
3317 if (th->t_refcount > 1)
3320 current->journal_info = th->t_handle_save;
3321 reiserfs_check_lock_depth("journal end");
3322 if (SB_JOURNAL(p_s_sb)->j_len == 0) {
3323 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
3324 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
3327 lock_journal(p_s_sb) ;
3328 if (SB_JOURNAL(p_s_sb)->j_next_full_flush) {
3329 flags |= FLUSH_ALL ;
3332 if (SB_JOURNAL(p_s_sb)->j_next_async_flush) {
3333 flags |= COMMIT_NOW | WAIT;
3337 /* check_journal_end locks the journal, and unlocks if it does not return 1
3338 ** it tells us if we should continue with the journal_end, or just return
3340 if (!check_journal_end(th, p_s_sb, nblocks, flags)) {
3342 wake_queued_writers(p_s_sb);
3343 reiserfs_async_progress_wait(p_s_sb);
3347 /* check_journal_end might set these, check again */
3348 if (SB_JOURNAL(p_s_sb)->j_next_full_flush) {
3353 ** j must wait means we have to flush the log blocks, and the real blocks for
3356 if (SB_JOURNAL(p_s_sb)->j_must_wait > 0) {
3360 #ifdef REISERFS_PREALLOCATE
3361 /* quota ops might need to nest, setup the journal_info pointer for them */
3362 current->journal_info = th ;
3363 reiserfs_discard_all_prealloc(th); /* it should not involve new blocks into
3364 * the transaction */
3365 current->journal_info = th->t_handle_save ;
3368 /* setup description block */
3369 d_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_JOURNAL(p_s_sb)->j_start) ;
3370 set_buffer_uptodate(d_bh);
3371 desc = (struct reiserfs_journal_desc *)(d_bh)->b_data ;
3372 memset(d_bh->b_data, 0, d_bh->b_size) ;
3373 memcpy(get_journal_desc_magic (d_bh), JOURNAL_DESC_MAGIC, 8) ;
3374 set_desc_trans_id(desc, SB_JOURNAL(p_s_sb)->j_trans_id) ;
3376 /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */
3377 c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3378 ((SB_JOURNAL(p_s_sb)->j_start + SB_JOURNAL(p_s_sb)->j_len + 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ;
3379 commit = (struct reiserfs_journal_commit *)c_bh->b_data ;
3380 memset(c_bh->b_data, 0, c_bh->b_size) ;
3381 set_commit_trans_id(commit, SB_JOURNAL(p_s_sb)->j_trans_id) ;
3382 set_buffer_uptodate(c_bh) ;
3384 /* init this journal list */
3385 jl = SB_JOURNAL(p_s_sb)->j_current_jl;
3387 /* we lock the commit before doing anything because
3388 * we want to make sure nobody tries to run flush_commit_list until
3389 * the new transaction is fully setup, and we've already flushed the
3392 down(&jl->j_commit_lock);
3394 /* save the transaction id in case we need to commit it later */
3395 commit_trans_id = jl->j_trans_id;
3397 atomic_set(&jl->j_older_commits_done, 0) ;
3398 jl->j_trans_id = SB_JOURNAL(p_s_sb)->j_trans_id ;
3399 jl->j_timestamp = SB_JOURNAL(p_s_sb)->j_trans_start_time ;
3400 jl->j_commit_bh = c_bh ;
3401 jl->j_start = SB_JOURNAL(p_s_sb)->j_start ;
3402 jl->j_len = SB_JOURNAL(p_s_sb)->j_len ;
3403 atomic_set(&jl->j_nonzerolen, SB_JOURNAL(p_s_sb)->j_len) ;
3404 atomic_set(&jl->j_commit_left, SB_JOURNAL(p_s_sb)->j_len + 2);
3405 jl->j_realblock = NULL ;
3407 /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
3408 ** for each real block, add it to the journal list hash,
3409 ** copy into real block index array in the commit or desc block
3411 trans_half = journal_trans_half(p_s_sb->s_blocksize);
3412 for (i = 0, cn = SB_JOURNAL(p_s_sb)->j_first ; cn ; cn = cn->next, i++) {
3413 if (test_bit(BH_JDirty, &cn->bh->b_state) ) {
3414 jl_cn = get_cnode(p_s_sb) ;
3416 reiserfs_panic(p_s_sb, "journal-1676, get_cnode returned NULL\n") ;
3419 jl->j_realblock = jl_cn ;
3421 jl_cn->prev = last_cn ;
3422 jl_cn->next = NULL ;
3424 last_cn->next = jl_cn ;
3427 /* make sure the block we are trying to log is not a block
3428 of journal or reserved area */
3430 if (is_block_in_log_or_reserved_area(p_s_sb, cn->bh->b_blocknr)) {
3431 reiserfs_panic(p_s_sb, "journal-2332: Trying to log block %lu, which is a log block\n", cn->bh->b_blocknr) ;
3433 jl_cn->blocknr = cn->bh->b_blocknr ;
3436 jl_cn->bh = cn->bh ;
3438 insert_journal_hash(SB_JOURNAL(p_s_sb)->j_list_hash_table, jl_cn) ;
3439 if (i < trans_half) {
3440 desc->j_realblock[i] = cpu_to_le32(cn->bh->b_blocknr) ;
3442 commit->j_realblock[i - trans_half] = cpu_to_le32(cn->bh->b_blocknr) ;
3448 set_desc_trans_len(desc, SB_JOURNAL(p_s_sb)->j_len) ;
3449 set_desc_mount_id(desc, SB_JOURNAL(p_s_sb)->j_mount_id) ;
3450 set_desc_trans_id(desc, SB_JOURNAL(p_s_sb)->j_trans_id) ;
3451 set_commit_trans_len(commit, SB_JOURNAL(p_s_sb)->j_len);
3453 /* special check in case all buffers in the journal were marked for not logging */
3454 if (SB_JOURNAL(p_s_sb)->j_len == 0) {
3458 /* we're about to dirty all the log blocks, mark the description block
3459 * dirty now too. Don't mark the commit block dirty until all the
3460 * others are on disk
3462 mark_buffer_dirty(d_bh);
3464 /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
3465 cur_write_start = SB_JOURNAL(p_s_sb)->j_start ;
3466 cn = SB_JOURNAL(p_s_sb)->j_first ;
3467 jindex = 1 ; /* start at one so we don't get the desc again */
3469 clear_bit(BH_JNew, &(cn->bh->b_state)) ;
3470 /* copy all the real blocks into log area. dirty log blocks */
3471 if (test_bit(BH_JDirty, &cn->bh->b_state)) {
3472 struct buffer_head *tmp_bh ;
3473 tmp_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3474 ((cur_write_start + jindex) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ;
3475 set_buffer_uptodate(tmp_bh);
3476 memcpy(tmp_bh->b_data, cn->bh->b_data, cn->bh->b_size) ;
3477 mark_buffer_dirty(tmp_bh);
3479 set_bit(BH_JDirty_wait, &(cn->bh->b_state)) ;
3480 clear_bit(BH_JDirty, &(cn->bh->b_state)) ;
3482 /* JDirty cleared sometime during transaction. don't log this one */
3483 reiserfs_warning("journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!\n") ;
3487 free_cnode(p_s_sb, cn) ;
3492 /* we are done with both the c_bh and d_bh, but
3493 ** c_bh must be written after all other commit blocks,
3494 ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
3497 SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
3499 /* now it is safe to insert this transaction on the main list */
3500 list_add_tail(&jl->j_list, &SB_JOURNAL(p_s_sb)->j_journal_list);
3501 list_add_tail(&jl->j_working_list, &SB_JOURNAL(p_s_sb)->j_working_list);
3502 SB_JOURNAL(p_s_sb)->j_num_work_lists++;
3504 /* reset journal values for the next transaction */
3505 old_start = SB_JOURNAL(p_s_sb)->j_start ;
3506 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);
3507 atomic_set(&(SB_JOURNAL(p_s_sb)->j_wcount), 0) ;
3508 SB_JOURNAL(p_s_sb)->j_bcount = 0 ;
3509 SB_JOURNAL(p_s_sb)->j_last = NULL ;
3510 SB_JOURNAL(p_s_sb)->j_first = NULL ;
3511 SB_JOURNAL(p_s_sb)->j_len = 0 ;
3512 SB_JOURNAL(p_s_sb)->j_trans_start_time = 0 ;
3513 SB_JOURNAL(p_s_sb)->j_trans_id++ ;
3514 SB_JOURNAL(p_s_sb)->j_current_jl->j_trans_id = SB_JOURNAL(p_s_sb)->j_trans_id;
3515 SB_JOURNAL(p_s_sb)->j_must_wait = 0 ;
3516 SB_JOURNAL(p_s_sb)->j_len_alloc = 0 ;
3517 SB_JOURNAL(p_s_sb)->j_next_full_flush = 0 ;
3518 SB_JOURNAL(p_s_sb)->j_next_async_flush = 0 ;
3519 init_journal_hash(p_s_sb) ;
3521 // make sure reiserfs_add_jh sees the new current_jl before we
3522 // write out the tails
3525 /* tail conversion targets have to hit the disk before we end the
3526 * transaction. Otherwise a later transaction might repack the tail
3527 * before this transaction commits, leaving the data block unflushed and
3528 * clean, if we crash before the later transaction commits, the data block
3531 if (!list_empty(&jl->j_tail_bh_list)) {
3533 write_ordered_buffers(&SB_JOURNAL(p_s_sb)->j_dirty_buffers_lock,
3534 SB_JOURNAL(p_s_sb), jl, &jl->j_tail_bh_list);
3537 if (!list_empty(&jl->j_tail_bh_list))
3539 up(&jl->j_commit_lock);
3541 /* honor the flush wishes from the caller, simple commits can
3542 ** be done outside the journal lock, they are done below
3544 ** if we don't flush the commit list right now, we put it into
3545 ** the work queue so the people waiting on the async progress work
3546 ** queue don't wait for this proc to flush journal lists and such.
3549 flush_commit_list(p_s_sb, jl, 1) ;
3550 flush_journal_list(p_s_sb, jl, 1) ;
3551 } else if (!(jl->j_state & LIST_COMMIT_PENDING))
3552 queue_delayed_work(commit_wq, &SB_JOURNAL(p_s_sb)->j_work, HZ/10);
3555 /* if the next transaction has any chance of wrapping, flush
3556 ** transactions that might get overwritten. If any journal lists are very
3557 ** old flush them as well.
3560 list_for_each_safe(entry, safe, &SB_JOURNAL(p_s_sb)->j_journal_list) {
3561 temp_jl = JOURNAL_LIST_ENTRY(entry);
3562 if (SB_JOURNAL(p_s_sb)->j_start <= temp_jl->j_start) {
3563 if ((SB_JOURNAL(p_s_sb)->j_start + SB_JOURNAL_TRANS_MAX(p_s_sb) + 1) >=
3566 flush_used_journal_lists(p_s_sb, temp_jl);
3568 } else if ((SB_JOURNAL(p_s_sb)->j_start +
3569 SB_JOURNAL_TRANS_MAX(p_s_sb) + 1) <
3570 SB_ONDISK_JOURNAL_SIZE(p_s_sb))
3572 /* if we don't cross into the next transaction and we don't
3573 * wrap, there is no way we can overlap any later transactions
3578 } else if ((SB_JOURNAL(p_s_sb)->j_start +
3579 SB_JOURNAL_TRANS_MAX(p_s_sb) + 1) >
3580 SB_ONDISK_JOURNAL_SIZE(p_s_sb))
3582 if (((SB_JOURNAL(p_s_sb)->j_start + SB_JOURNAL_TRANS_MAX(p_s_sb) + 1) %
3583 SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >= temp_jl->j_start)
3585 flush_used_journal_lists(p_s_sb, temp_jl);
3588 /* we don't overlap anything from out start to the end of the
3589 * log, and our wrapped portion doesn't overlap anything at
3590 * the start of the log. We can break
3596 flush_old_journal_lists(p_s_sb);
3598 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) ;
3600 if (!(SB_JOURNAL(p_s_sb)->j_current_jl->j_list_bitmap)) {
3601 reiserfs_panic(p_s_sb, "journal-1996: do_journal_end, could not get a list bitmap\n") ;
3604 atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 0) ;
3605 unlock_journal(p_s_sb) ;
3606 /* wake up any body waiting to join. */
3607 clear_bit(WRITERS_QUEUED, &SB_JOURNAL(p_s_sb)->j_state);
3608 wake_up(&(SB_JOURNAL(p_s_sb)->j_join_wait)) ;
3610 if (!flush && wait_on_commit &&
3611 journal_list_still_alive(p_s_sb, commit_trans_id)) {
3612 flush_commit_list(p_s_sb, jl, 1) ;
3615 reiserfs_check_lock_depth("journal end2");