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 void disable_barrier(struct super_block *s)
132 REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH);
133 printk("reiserfs: disabling flush barriers on %s\n", reiserfs_bdevname(s));
136 static struct reiserfs_bitmap_node *
137 allocate_bitmap_node(struct super_block *p_s_sb) {
138 struct reiserfs_bitmap_node *bn ;
141 bn = reiserfs_kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS, p_s_sb) ;
145 bn->data = reiserfs_kmalloc(p_s_sb->s_blocksize, GFP_NOFS, p_s_sb) ;
147 reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb) ;
151 memset(bn->data, 0, p_s_sb->s_blocksize) ;
152 INIT_LIST_HEAD(&bn->list) ;
156 static struct reiserfs_bitmap_node *
157 get_bitmap_node(struct super_block *p_s_sb) {
158 struct reiserfs_bitmap_node *bn = NULL;
159 struct list_head *entry = SB_JOURNAL(p_s_sb)->j_bitmap_nodes.next ;
161 SB_JOURNAL(p_s_sb)->j_used_bitmap_nodes++ ;
164 if(entry != &SB_JOURNAL(p_s_sb)->j_bitmap_nodes) {
165 bn = list_entry(entry, struct reiserfs_bitmap_node, list) ;
167 memset(bn->data, 0, p_s_sb->s_blocksize) ;
168 SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes-- ;
171 bn = allocate_bitmap_node(p_s_sb) ;
178 static inline void free_bitmap_node(struct super_block *p_s_sb,
179 struct reiserfs_bitmap_node *bn) {
180 SB_JOURNAL(p_s_sb)->j_used_bitmap_nodes-- ;
181 if (SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
182 reiserfs_kfree(bn->data, p_s_sb->s_blocksize, p_s_sb) ;
183 reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb) ;
185 list_add(&bn->list, &SB_JOURNAL(p_s_sb)->j_bitmap_nodes) ;
186 SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes++ ;
190 static void allocate_bitmap_nodes(struct super_block *p_s_sb) {
192 struct reiserfs_bitmap_node *bn = NULL ;
193 for (i = 0 ; i < REISERFS_MIN_BITMAP_NODES ; i++) {
194 bn = allocate_bitmap_node(p_s_sb) ;
196 list_add(&bn->list, &SB_JOURNAL(p_s_sb)->j_bitmap_nodes) ;
197 SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes++ ;
199 break ; // this is ok, we'll try again when more are needed
204 static int set_bit_in_list_bitmap(struct super_block *p_s_sb, int block,
205 struct reiserfs_list_bitmap *jb) {
206 int bmap_nr = block / (p_s_sb->s_blocksize << 3) ;
207 int bit_nr = block % (p_s_sb->s_blocksize << 3) ;
209 if (!jb->bitmaps[bmap_nr]) {
210 jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb) ;
212 set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data) ;
216 static void cleanup_bitmap_list(struct super_block *p_s_sb,
217 struct reiserfs_list_bitmap *jb) {
219 if (jb->bitmaps == NULL)
222 for (i = 0 ; i < SB_BMAP_NR(p_s_sb) ; i++) {
223 if (jb->bitmaps[i]) {
224 free_bitmap_node(p_s_sb, jb->bitmaps[i]) ;
225 jb->bitmaps[i] = NULL ;
231 ** only call this on FS unmount.
233 static int free_list_bitmaps(struct super_block *p_s_sb,
234 struct reiserfs_list_bitmap *jb_array) {
236 struct reiserfs_list_bitmap *jb ;
237 for (i = 0 ; i < JOURNAL_NUM_BITMAPS ; i++) {
239 jb->journal_list = NULL ;
240 cleanup_bitmap_list(p_s_sb, jb) ;
247 static int free_bitmap_nodes(struct super_block *p_s_sb) {
248 struct list_head *next = SB_JOURNAL(p_s_sb)->j_bitmap_nodes.next ;
249 struct reiserfs_bitmap_node *bn ;
251 while(next != &SB_JOURNAL(p_s_sb)->j_bitmap_nodes) {
252 bn = list_entry(next, struct reiserfs_bitmap_node, list) ;
254 reiserfs_kfree(bn->data, p_s_sb->s_blocksize, p_s_sb) ;
255 reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb) ;
256 next = SB_JOURNAL(p_s_sb)->j_bitmap_nodes.next ;
257 SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes-- ;
264 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
265 ** jb_array is the array to be filled in.
267 int reiserfs_allocate_list_bitmaps(struct super_block *p_s_sb,
268 struct reiserfs_list_bitmap *jb_array,
272 struct reiserfs_list_bitmap *jb ;
273 int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *) ;
275 for (i = 0 ; i < JOURNAL_NUM_BITMAPS ; i++) {
277 jb->journal_list = NULL ;
278 jb->bitmaps = vmalloc( mem ) ;
280 reiserfs_warning(p_s_sb, "clm-2000, unable to allocate bitmaps for journal lists") ;
284 memset(jb->bitmaps, 0, mem) ;
287 free_list_bitmaps(p_s_sb, jb_array) ;
294 ** find an available list bitmap. If you can't find one, flush a commit list
297 static struct reiserfs_list_bitmap *
298 get_list_bitmap(struct super_block *p_s_sb, struct reiserfs_journal_list *jl) {
300 struct reiserfs_list_bitmap *jb = NULL ;
302 for (j = 0 ; j < (JOURNAL_NUM_BITMAPS * 3) ; j++) {
303 i = SB_JOURNAL(p_s_sb)->j_list_bitmap_index ;
304 SB_JOURNAL(p_s_sb)->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS ;
305 jb = SB_JOURNAL(p_s_sb)->j_list_bitmap + i ;
306 if (SB_JOURNAL(p_s_sb)->j_list_bitmap[i].journal_list) {
307 flush_commit_list(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_bitmap[i].journal_list, 1) ;
308 if (!SB_JOURNAL(p_s_sb)->j_list_bitmap[i].journal_list) {
315 if (jb->journal_list) { /* double check to make sure if flushed correctly */
318 jb->journal_list = jl ;
323 ** allocates a new chunk of X nodes, and links them all together as a list.
324 ** Uses the cnode->next and cnode->prev pointers
325 ** returns NULL on failure
327 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes) {
328 struct reiserfs_journal_cnode *head ;
330 if (num_cnodes <= 0) {
333 head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode)) ;
337 memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode)) ;
338 head[0].prev = NULL ;
339 head[0].next = head + 1 ;
340 for (i = 1 ; i < num_cnodes; i++) {
341 head[i].prev = head + (i - 1) ;
342 head[i].next = head + (i + 1) ; /* if last one, overwrite it after the if */
344 head[num_cnodes -1].next = NULL ;
349 ** pulls a cnode off the free list, or returns NULL on failure
351 static struct reiserfs_journal_cnode *get_cnode(struct super_block *p_s_sb) {
352 struct reiserfs_journal_cnode *cn ;
354 reiserfs_check_lock_depth(p_s_sb, "get_cnode") ;
356 if (SB_JOURNAL(p_s_sb)->j_cnode_free <= 0) {
359 SB_JOURNAL(p_s_sb)->j_cnode_used++ ;
360 SB_JOURNAL(p_s_sb)->j_cnode_free-- ;
361 cn = SB_JOURNAL(p_s_sb)->j_cnode_free_list ;
366 cn->next->prev = NULL ;
368 SB_JOURNAL(p_s_sb)->j_cnode_free_list = cn->next ;
369 memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ;
374 ** returns a cnode to the free list
376 static void free_cnode(struct super_block *p_s_sb, struct reiserfs_journal_cnode *cn) {
378 reiserfs_check_lock_depth(p_s_sb, "free_cnode") ;
380 SB_JOURNAL(p_s_sb)->j_cnode_used-- ;
381 SB_JOURNAL(p_s_sb)->j_cnode_free++ ;
382 /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
383 cn->next = SB_JOURNAL(p_s_sb)->j_cnode_free_list ;
384 if (SB_JOURNAL(p_s_sb)->j_cnode_free_list) {
385 SB_JOURNAL(p_s_sb)->j_cnode_free_list->prev = cn ;
387 cn->prev = NULL ; /* not needed with the memset, but I might kill the memset, and forget to do this */
388 SB_JOURNAL(p_s_sb)->j_cnode_free_list = cn ;
391 static int clear_prepared_bits(struct buffer_head *bh) {
392 clear_bit(BH_JPrepared, &bh->b_state) ;
393 clear_bit(BH_JRestore_dirty, &bh->b_state) ;
397 /* buffer is in current transaction */
398 inline int buffer_journaled(const struct buffer_head *bh) {
400 return test_bit(BH_JDirty, &bh->b_state) ;
405 /* disk block was taken off free list before being in a finished transation, or written to disk
406 ** journal_new blocks can be reused immediately, for any purpose
408 inline int buffer_journal_new(const struct buffer_head *bh) {
410 return test_bit(BH_JNew, &bh->b_state) ;
415 inline int mark_buffer_journal_new(struct buffer_head *bh) {
417 set_bit(BH_JNew, &bh->b_state) ;
422 inline int mark_buffer_not_journaled(struct buffer_head *bh) {
424 clear_bit(BH_JDirty, &bh->b_state) ;
428 /* utility function to force a BUG if it is called without the big
429 ** kernel lock held. caller is the string printed just before calling BUG()
431 void reiserfs_check_lock_depth(struct super_block *sb, char *caller) {
433 if (current->lock_depth < 0) {
434 reiserfs_panic (sb, "%s called without kernel lock held", caller) ;
441 /* return a cnode with same dev, block number and size in table, or null if not found */
442 static inline struct reiserfs_journal_cnode *
443 get_journal_hash_dev(struct super_block *sb,
444 struct reiserfs_journal_cnode **table,
447 struct reiserfs_journal_cnode *cn ;
448 cn = journal_hash(table, sb, bl) ;
450 if (cn->blocknr == bl && cn->sb == sb)
454 return (struct reiserfs_journal_cnode *)0 ;
457 /* returns a cnode with same size, block number and dev as bh in the current transaction hash. NULL if not found */
458 static inline struct reiserfs_journal_cnode *get_journal_hash(struct super_block *p_s_sb, struct buffer_head *bh) {
459 struct reiserfs_journal_cnode *cn ;
461 cn = get_journal_hash_dev(p_s_sb, SB_JOURNAL(p_s_sb)->j_hash_table, bh->b_blocknr);
464 return (struct reiserfs_journal_cnode *)0 ;
470 ** this actually means 'can this block be reallocated yet?'. If you set search_all, a block can only be allocated
471 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
472 ** being overwritten by a replay after crashing.
474 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction. Since deleting
475 ** a block removes it from the current transaction, this case should never happen. If you don't set search_all, make
476 ** sure you never write the block without logging it.
478 ** next_zero_bit is a suggestion about the next block to try for find_forward.
479 ** when bl is rejected because it is set in a journal list bitmap, we search
480 ** for the next zero bit in the bitmap that rejected bl. Then, we return that
481 ** through next_zero_bit for find_forward to try.
483 ** Just because we return something in next_zero_bit does not mean we won't
484 ** reject it on the next call to reiserfs_in_journal
487 int reiserfs_in_journal(struct super_block *p_s_sb,
488 int bmap_nr, int bit_nr, int search_all,
489 b_blocknr_t *next_zero_bit) {
490 struct reiserfs_journal_cnode *cn ;
491 struct reiserfs_list_bitmap *jb ;
495 *next_zero_bit = 0 ; /* always start this at zero. */
497 PROC_INFO_INC( p_s_sb, journal.in_journal );
498 /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
499 ** if we crash before the transaction that freed it commits, this transaction won't
500 ** have committed either, and the block will never be written
503 for (i = 0 ; i < JOURNAL_NUM_BITMAPS ; i++) {
504 PROC_INFO_INC( p_s_sb, journal.in_journal_bitmap );
505 jb = SB_JOURNAL(p_s_sb)->j_list_bitmap + i ;
506 if (jb->journal_list && jb->bitmaps[bmap_nr] &&
507 test_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data)) {
508 *next_zero_bit = find_next_zero_bit((unsigned long *)
509 (jb->bitmaps[bmap_nr]->data),
510 p_s_sb->s_blocksize << 3, bit_nr+1) ;
516 bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr;
517 /* is it in any old transactions? */
518 if (search_all && (cn = get_journal_hash_dev(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_hash_table, bl))) {
522 /* is it in the current transaction. This should never happen */
523 if ((cn = get_journal_hash_dev(p_s_sb, SB_JOURNAL(p_s_sb)->j_hash_table, bl))) {
528 PROC_INFO_INC( p_s_sb, journal.in_journal_reusable );
533 /* insert cn into table
535 inline void insert_journal_hash(struct reiserfs_journal_cnode **table, struct reiserfs_journal_cnode *cn) {
536 struct reiserfs_journal_cnode *cn_orig ;
538 cn_orig = journal_hash(table, cn->sb, cn->blocknr) ;
539 cn->hnext = cn_orig ;
542 cn_orig->hprev = cn ;
544 journal_hash(table, cn->sb, cn->blocknr) = cn ;
547 /* lock the current transaction */
548 inline static void lock_journal(struct super_block *p_s_sb) {
549 PROC_INFO_INC( p_s_sb, journal.lock_journal );
550 down(&SB_JOURNAL(p_s_sb)->j_lock);
553 /* unlock the current transaction */
554 inline static void unlock_journal(struct super_block *p_s_sb) {
555 up(&SB_JOURNAL(p_s_sb)->j_lock);
558 static inline void get_journal_list(struct reiserfs_journal_list *jl)
563 static inline void put_journal_list(struct super_block *s,
564 struct reiserfs_journal_list *jl)
566 if (jl->j_refcount < 1) {
567 reiserfs_panic (s, "trans id %lu, refcount at %d", jl->j_trans_id,
570 if (--jl->j_refcount == 0)
571 reiserfs_kfree(jl, sizeof(struct reiserfs_journal_list), s);
575 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
576 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
579 static void cleanup_freed_for_journal_list(struct super_block *p_s_sb, struct reiserfs_journal_list *jl) {
581 struct reiserfs_list_bitmap *jb = jl->j_list_bitmap ;
583 cleanup_bitmap_list(p_s_sb, jb) ;
585 jl->j_list_bitmap->journal_list = NULL ;
586 jl->j_list_bitmap = NULL ;
589 static int journal_list_still_alive(struct super_block *s,
590 unsigned long trans_id)
592 struct list_head *entry = &SB_JOURNAL(s)->j_journal_list;
593 struct reiserfs_journal_list *jl;
595 if (!list_empty(entry)) {
596 jl = JOURNAL_LIST_ENTRY(entry->next);
597 if (jl->j_trans_id <= trans_id) {
604 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate) {
605 char b[BDEVNAME_SIZE];
607 if (buffer_journaled(bh)) {
608 reiserfs_warning(NULL, "clm-2084: pinned buffer %lu:%s sent to disk",
609 bh->b_blocknr, bdevname(bh->b_bdev, b)) ;
612 set_buffer_uptodate(bh) ;
614 clear_buffer_uptodate(bh) ;
619 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate) {
621 set_buffer_uptodate(bh) ;
623 clear_buffer_uptodate(bh) ;
628 static void submit_logged_buffer(struct buffer_head *bh) {
630 bh->b_end_io = reiserfs_end_buffer_io_sync ;
631 mark_buffer_notjournal_new(bh) ;
632 clear_buffer_dirty(bh) ;
633 if (!test_and_clear_bit(BH_JTest, &bh->b_state))
635 if (!buffer_uptodate(bh))
637 submit_bh(WRITE, bh) ;
640 static void submit_ordered_buffer(struct buffer_head *bh) {
642 bh->b_end_io = reiserfs_end_ordered_io;
643 clear_buffer_dirty(bh) ;
644 if (!buffer_uptodate(bh))
646 submit_bh(WRITE, bh) ;
649 static int submit_barrier_buffer(struct buffer_head *bh) {
651 bh->b_end_io = reiserfs_end_ordered_io;
652 clear_buffer_dirty(bh) ;
653 if (!buffer_uptodate(bh))
655 return submit_bh(WRITE_BARRIER, bh) ;
658 static void check_barrier_completion(struct super_block *s,
659 struct buffer_head *bh) {
660 if (buffer_eopnotsupp(bh)) {
661 clear_buffer_eopnotsupp(bh);
663 set_buffer_uptodate(bh);
664 set_buffer_dirty(bh);
665 sync_dirty_buffer(bh);
669 #define CHUNK_SIZE 32
670 struct buffer_chunk {
671 struct buffer_head *bh[CHUNK_SIZE];
675 static void write_chunk(struct buffer_chunk *chunk) {
677 for (i = 0; i < chunk->nr ; i++) {
678 submit_logged_buffer(chunk->bh[i]) ;
683 static void write_ordered_chunk(struct buffer_chunk *chunk) {
685 for (i = 0; i < chunk->nr ; i++) {
686 submit_ordered_buffer(chunk->bh[i]) ;
691 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
693 void (fn)(struct buffer_chunk *))
696 if (chunk->nr >= CHUNK_SIZE)
698 chunk->bh[chunk->nr++] = bh;
699 if (chunk->nr >= CHUNK_SIZE) {
711 atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
712 static struct reiserfs_jh *alloc_jh(void) {
713 struct reiserfs_jh *jh;
715 jh = kmalloc(sizeof(*jh), GFP_NOFS);
717 atomic_inc(&nr_reiserfs_jh);
725 * we want to free the jh when the buffer has been written
728 void reiserfs_free_jh(struct buffer_head *bh) {
729 struct reiserfs_jh *jh;
733 bh->b_private = NULL;
735 list_del_init(&jh->list);
737 if (atomic_read(&nr_reiserfs_jh) <= 0)
739 atomic_dec(&nr_reiserfs_jh);
744 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
747 struct reiserfs_jh *jh;
750 spin_lock(&j->j_dirty_buffers_lock);
751 if (!bh->b_private) {
752 spin_unlock(&j->j_dirty_buffers_lock);
756 list_del_init(&jh->list);
761 spin_lock(&j->j_dirty_buffers_lock);
762 /* buffer must be locked for __add_jh, should be able to have
763 * two adds at the same time
770 jh->jl = j->j_current_jl;
772 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
774 list_add_tail(&jh->list, &jh->jl->j_bh_list);
776 spin_unlock(&j->j_dirty_buffers_lock);
780 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh) {
781 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
783 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh) {
784 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
787 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
788 static int write_ordered_buffers(spinlock_t *lock,
789 struct reiserfs_journal *j,
790 struct reiserfs_journal_list *jl,
791 struct list_head *list)
793 struct buffer_head *bh;
794 struct reiserfs_jh *jh;
796 struct buffer_chunk chunk;
797 struct list_head tmp;
798 INIT_LIST_HEAD(&tmp);
802 while(!list_empty(list)) {
803 jh = JH_ENTRY(list->next);
806 if (test_set_buffer_locked(bh)) {
807 if (!buffer_dirty(bh)) {
808 list_del_init(&jh->list);
809 list_add(&jh->list, &tmp);
814 write_ordered_chunk(&chunk);
820 if (buffer_dirty(bh)) {
821 list_del_init(&jh->list);
822 list_add(&jh->list, &tmp);
823 add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
825 reiserfs_free_jh(bh);
830 cond_resched_lock(lock);
834 write_ordered_chunk(&chunk);
837 while(!list_empty(&tmp)) {
838 jh = JH_ENTRY(tmp.prev);
841 reiserfs_free_jh(bh);
843 if (buffer_locked(bh)) {
848 if (!buffer_uptodate(bh))
851 cond_resched_lock(lock);
857 static int flush_older_commits(struct super_block *s, struct reiserfs_journal_list *jl) {
858 struct reiserfs_journal_list *other_jl;
859 struct reiserfs_journal_list *first_jl;
860 struct list_head *entry;
861 unsigned long trans_id = jl->j_trans_id;
862 unsigned long other_trans_id;
863 unsigned long first_trans_id;
867 * first we walk backwards to find the oldest uncommitted transation
870 entry = jl->j_list.prev;
872 other_jl = JOURNAL_LIST_ENTRY(entry);
873 if (entry == &SB_JOURNAL(s)->j_journal_list ||
874 atomic_read(&other_jl->j_older_commits_done))
878 entry = other_jl->j_list.prev;
881 /* if we didn't find any older uncommitted transactions, return now */
882 if (first_jl == jl) {
886 first_trans_id = first_jl->j_trans_id;
888 entry = &first_jl->j_list;
890 other_jl = JOURNAL_LIST_ENTRY(entry);
891 other_trans_id = other_jl->j_trans_id;
893 if (other_trans_id < trans_id) {
894 if (atomic_read(&other_jl->j_commit_left) != 0) {
895 flush_commit_list(s, other_jl, 0);
897 /* list we were called with is gone, return */
898 if (!journal_list_still_alive(s, trans_id))
901 /* the one we just flushed is gone, this means all
902 * older lists are also gone, so first_jl is no longer
903 * valid either. Go back to the beginning.
905 if (!journal_list_still_alive(s, other_trans_id)) {
910 if (entry == &SB_JOURNAL(s)->j_journal_list)
918 int reiserfs_async_progress_wait(struct super_block *s) {
920 struct reiserfs_journal *j = SB_JOURNAL(s);
921 if (atomic_read(&j->j_async_throttle))
922 blk_congestion_wait(WRITE, HZ/10);
927 ** if this journal list still has commit blocks unflushed, send them to disk.
929 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
930 ** Before the commit block can by written, every other log block must be safely on disk
933 static int flush_commit_list(struct super_block *s, struct reiserfs_journal_list *jl, int flushall) {
936 struct buffer_head *tbh = NULL ;
937 unsigned long trans_id = jl->j_trans_id;
940 reiserfs_check_lock_depth(s, "flush_commit_list") ;
942 if (atomic_read(&jl->j_older_commits_done)) {
946 /* before we can put our commit blocks on disk, we have to make sure everyone older than
951 if (trans_id == SB_JOURNAL(s)->j_trans_id)
954 get_journal_list(jl);
956 if (flush_older_commits(s, jl) == 1) {
957 /* list disappeared during flush_older_commits. return */
962 /* make sure nobody is trying to flush this one at the same time */
963 down(&jl->j_commit_lock);
964 if (!journal_list_still_alive(s, trans_id)) {
965 up(&jl->j_commit_lock);
968 if (jl->j_trans_id == 0)
971 /* this commit is done, exit */
972 if (atomic_read(&(jl->j_commit_left)) <= 0) {
974 atomic_set(&(jl->j_older_commits_done), 1) ;
976 up(&jl->j_commit_lock);
980 if (!list_empty(&jl->j_bh_list)) {
982 write_ordered_buffers(&SB_JOURNAL(s)->j_dirty_buffers_lock,
983 SB_JOURNAL(s), jl, &jl->j_bh_list);
986 if (!list_empty(&jl->j_bh_list))
989 * for the description block and all the log blocks, submit any buffers
990 * that haven't already reached the disk
992 atomic_inc(&SB_JOURNAL(s)->j_async_throttle);
993 for (i = 0 ; i < (jl->j_len + 1) ; i++) {
994 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start+i) %
995 SB_ONDISK_JOURNAL_SIZE(s);
996 tbh = journal_find_get_block(s, bn) ;
997 if (buffer_dirty(tbh))
998 ll_rw_block(WRITE, 1, &tbh) ;
1001 atomic_dec(&SB_JOURNAL(s)->j_async_throttle);
1003 /* wait on everything written so far before writing the commit
1004 * if we are in barrier mode, send the commit down now
1006 barrier = reiserfs_barrier_flush(s);
1009 lock_buffer(jl->j_commit_bh);
1010 ret = submit_barrier_buffer(jl->j_commit_bh);
1011 if (ret == -EOPNOTSUPP) {
1012 set_buffer_uptodate(jl->j_commit_bh);
1017 for (i = 0 ; i < (jl->j_len + 1) ; i++) {
1018 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1019 (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s) ;
1020 tbh = journal_find_get_block(s, bn) ;
1021 wait_on_buffer(tbh) ;
1022 // since we're using ll_rw_blk above, it might have skipped over
1023 // a locked buffer. Double check here
1025 if (buffer_dirty(tbh))
1026 sync_dirty_buffer(tbh);
1027 if (!buffer_uptodate(tbh)) {
1028 reiserfs_panic(s, "journal-601, buffer write failed\n") ;
1030 put_bh(tbh) ; /* once for journal_find_get_block */
1031 put_bh(tbh) ; /* once due to original getblk in do_journal_end */
1032 atomic_dec(&(jl->j_commit_left)) ;
1035 if (atomic_read(&(jl->j_commit_left)) != 1)
1039 if (buffer_dirty(jl->j_commit_bh))
1041 mark_buffer_dirty(jl->j_commit_bh) ;
1042 sync_dirty_buffer(jl->j_commit_bh) ;
1044 wait_on_buffer(jl->j_commit_bh);
1046 check_barrier_completion(s, jl->j_commit_bh);
1047 if (!buffer_uptodate(jl->j_commit_bh)) {
1048 reiserfs_panic(s, "journal-615: buffer write failed\n") ;
1050 bforget(jl->j_commit_bh) ;
1051 if (SB_JOURNAL(s)->j_last_commit_id != 0 &&
1052 (jl->j_trans_id - SB_JOURNAL(s)->j_last_commit_id) != 1) {
1053 reiserfs_warning(s, "clm-2200: last commit %lu, current %lu",
1054 SB_JOURNAL(s)->j_last_commit_id,
1057 SB_JOURNAL(s)->j_last_commit_id = jl->j_trans_id;
1059 /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */
1060 cleanup_freed_for_journal_list(s, jl) ;
1062 /* mark the metadata dirty */
1063 dirty_one_transaction(s, jl);
1064 atomic_dec(&(jl->j_commit_left)) ;
1067 atomic_set(&(jl->j_older_commits_done), 1) ;
1069 up(&jl->j_commit_lock);
1071 put_journal_list(s, jl);
1076 ** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or
1077 ** returns NULL if it can't find anything
1079 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct reiserfs_journal_cnode *cn) {
1080 struct super_block *sb = cn->sb;
1081 b_blocknr_t blocknr = cn->blocknr ;
1085 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1093 void remove_journal_hash(struct super_block *, struct reiserfs_journal_cnode **,
1094 struct reiserfs_journal_list *, unsigned long, int);
1097 ** once all the real blocks have been flushed, it is safe to remove them from the
1098 ** journal list for this transaction. Aside from freeing the cnode, this also allows the
1099 ** block to be reallocated for data blocks if it had been deleted.
1101 static void remove_all_from_journal_list(struct super_block *p_s_sb, struct reiserfs_journal_list *jl, int debug) {
1102 struct reiserfs_journal_cnode *cn, *last ;
1103 cn = jl->j_realblock ;
1105 /* which is better, to lock once around the whole loop, or
1106 ** to lock for each call to remove_journal_hash?
1109 if (cn->blocknr != 0) {
1111 reiserfs_warning (p_s_sb, "block %u, bh is %d, state %ld", cn->blocknr,
1112 cn->bh ? 1: 0, cn->state) ;
1115 remove_journal_hash(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_hash_table, jl, cn->blocknr, 1) ;
1119 free_cnode(p_s_sb, last) ;
1121 jl->j_realblock = NULL ;
1125 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1126 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1127 ** releasing blocks in this transaction for reuse as data blocks.
1128 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1131 static int _update_journal_header_block(struct super_block *p_s_sb, unsigned long offset, unsigned long trans_id) {
1132 struct reiserfs_journal_header *jh ;
1133 if (trans_id >= SB_JOURNAL(p_s_sb)->j_last_flush_trans_id) {
1134 if (buffer_locked((SB_JOURNAL(p_s_sb)->j_header_bh))) {
1135 wait_on_buffer((SB_JOURNAL(p_s_sb)->j_header_bh)) ;
1136 if (!buffer_uptodate(SB_JOURNAL(p_s_sb)->j_header_bh)) {
1137 reiserfs_panic(p_s_sb, "journal-699: buffer write failed\n") ;
1140 SB_JOURNAL(p_s_sb)->j_last_flush_trans_id = trans_id ;
1141 SB_JOURNAL(p_s_sb)->j_first_unflushed_offset = offset ;
1142 jh = (struct reiserfs_journal_header *)(SB_JOURNAL(p_s_sb)->j_header_bh->b_data) ;
1143 jh->j_last_flush_trans_id = cpu_to_le32(trans_id) ;
1144 jh->j_first_unflushed_offset = cpu_to_le32(offset) ;
1145 jh->j_mount_id = cpu_to_le32(SB_JOURNAL(p_s_sb)->j_mount_id) ;
1147 if (reiserfs_barrier_flush(p_s_sb)) {
1149 lock_buffer(SB_JOURNAL(p_s_sb)->j_header_bh);
1150 ret = submit_barrier_buffer(SB_JOURNAL(p_s_sb)->j_header_bh);
1151 if (ret == -EOPNOTSUPP) {
1152 set_buffer_uptodate(SB_JOURNAL(p_s_sb)->j_header_bh);
1153 disable_barrier(p_s_sb);
1156 wait_on_buffer(SB_JOURNAL(p_s_sb)->j_header_bh);
1157 check_barrier_completion(p_s_sb, SB_JOURNAL(p_s_sb)->j_header_bh);
1160 set_buffer_dirty(SB_JOURNAL(p_s_sb)->j_header_bh) ;
1161 sync_dirty_buffer(SB_JOURNAL(p_s_sb)->j_header_bh) ;
1163 if (!buffer_uptodate(SB_JOURNAL(p_s_sb)->j_header_bh)) {
1164 reiserfs_warning (p_s_sb, "journal-837: IO error during journal replay");
1171 static int update_journal_header_block(struct super_block *p_s_sb,
1172 unsigned long offset,
1173 unsigned long trans_id) {
1174 if (_update_journal_header_block(p_s_sb, offset, trans_id)) {
1175 reiserfs_panic(p_s_sb, "journal-712: buffer write failed\n") ;
1180 ** flush any and all journal lists older than you are
1181 ** can only be called from flush_journal_list
1183 static int flush_older_journal_lists(struct super_block *p_s_sb,
1184 struct reiserfs_journal_list *jl)
1186 struct list_head *entry;
1187 struct reiserfs_journal_list *other_jl ;
1188 unsigned long trans_id = jl->j_trans_id;
1190 /* we know we are the only ones flushing things, no extra race
1191 * protection is required.
1194 entry = SB_JOURNAL(p_s_sb)->j_journal_list.next;
1195 other_jl = JOURNAL_LIST_ENTRY(entry);
1196 if (other_jl->j_trans_id < trans_id) {
1197 /* do not flush all */
1198 flush_journal_list(p_s_sb, other_jl, 0) ;
1200 /* other_jl is now deleted from the list */
1206 static void del_from_work_list(struct super_block *s,
1207 struct reiserfs_journal_list *jl) {
1208 if (!list_empty(&jl->j_working_list)) {
1209 list_del_init(&jl->j_working_list);
1210 SB_JOURNAL(s)->j_num_work_lists--;
1214 /* flush a journal list, both commit and real blocks
1216 ** always set flushall to 1, unless you are calling from inside
1217 ** flush_journal_list
1219 ** IMPORTANT. This can only be called while there are no journal writers,
1220 ** and the journal is locked. That means it can only be called from
1221 ** do_journal_end, or by journal_release
1223 static int flush_journal_list(struct super_block *s,
1224 struct reiserfs_journal_list *jl, int flushall) {
1225 struct reiserfs_journal_list *pjl ;
1226 struct reiserfs_journal_cnode *cn, *last ;
1230 struct buffer_head *saved_bh ;
1231 unsigned long j_len_saved = jl->j_len ;
1233 if (j_len_saved <= 0) {
1237 if (atomic_read(&SB_JOURNAL(s)->j_wcount) != 0) {
1238 reiserfs_warning(s, "clm-2048: flush_journal_list called with wcount %d",
1239 atomic_read(&SB_JOURNAL(s)->j_wcount)) ;
1241 if (jl->j_trans_id == 0)
1244 /* if flushall == 0, the lock is already held */
1246 down(&SB_JOURNAL(s)->j_flush_sem);
1247 } else if (!down_trylock(&SB_JOURNAL(s)->j_flush_sem)) {
1252 if (j_len_saved > SB_JOURNAL_TRANS_MAX(s)) {
1253 reiserfs_panic(s, "journal-715: flush_journal_list, length is %lu, trans id %lu\n", j_len_saved, jl->j_trans_id);
1257 /* if all the work is already done, get out of here */
1258 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1259 atomic_read(&(jl->j_commit_left)) <= 0) {
1260 goto flush_older_and_return ;
1263 /* start by putting the commit list on disk. This will also flush
1264 ** the commit lists of any olders transactions
1266 flush_commit_list(s, jl, 1) ;
1268 if (!(jl->j_state & LIST_DIRTY))
1271 /* are we done now? */
1272 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1273 atomic_read(&(jl->j_commit_left)) <= 0) {
1274 goto flush_older_and_return ;
1277 /* loop through each cnode, see if we need to write it,
1278 ** or wait on a more recent transaction, or just ignore it
1280 if (atomic_read(&(SB_JOURNAL(s)->j_wcount)) != 0) {
1281 reiserfs_panic(s, "journal-844: panic journal list is flushing, wcount is not 0\n") ;
1283 cn = jl->j_realblock ;
1288 /* blocknr of 0 is no longer in the hash, ignore it */
1289 if (cn->blocknr == 0) {
1292 pjl = find_newer_jl_for_cn(cn) ;
1293 /* the order is important here. We check pjl to make sure we
1294 ** don't clear BH_JDirty_wait if we aren't the one writing this
1297 if (!pjl && cn->bh) {
1300 /* we do this to make sure nobody releases the buffer while
1301 ** we are working with it
1305 if (buffer_journal_dirty(saved_bh)) {
1310 } else if (can_dirty(cn)) {
1311 /* everything with !pjl && jwait should be writable */
1316 /* if someone has this block in a newer transaction, just make
1317 ** sure they are commited, and don't try writing it to disk
1320 if (atomic_read(&pjl->j_commit_left))
1321 flush_commit_list(s, pjl, 1) ;
1325 /* bh == NULL when the block got to disk on its own, OR,
1326 ** the block got freed in a future transaction
1328 if (saved_bh == NULL) {
1332 /* this should never happen. kupdate_one_transaction has this list
1333 ** locked while it works, so we should never see a buffer here that
1334 ** is not marked JDirty_wait
1336 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1337 reiserfs_warning (s, "journal-813: BAD! buffer %llu %cdirty %cjwait, "
1338 "not in a newer tranasction",
1339 (unsigned long long)saved_bh->b_blocknr,
1340 was_dirty ? ' ' : '!', was_jwait ? ' ' : '!') ;
1343 /* we inc again because saved_bh gets decremented at free_cnode */
1345 set_bit(BLOCK_NEEDS_FLUSH, &cn->state) ;
1346 lock_buffer(saved_bh);
1347 if (cn->blocknr != saved_bh->b_blocknr)
1349 if (buffer_dirty(saved_bh))
1350 submit_logged_buffer(saved_bh) ;
1352 unlock_buffer(saved_bh);
1355 reiserfs_warning (s, "clm-2082: Unable to flush buffer %llu in %s",
1356 (unsigned long long)saved_bh->b_blocknr, __FUNCTION__);
1362 /* we incremented this to keep others from taking the buffer head away */
1364 if (atomic_read(&(saved_bh->b_count)) < 0) {
1365 reiserfs_warning (s, "journal-945: saved_bh->b_count < 0");
1370 cn = jl->j_realblock ;
1372 if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1374 reiserfs_panic(s, "journal-1011: cn->bh is NULL\n") ;
1376 wait_on_buffer(cn->bh) ;
1378 reiserfs_panic(s, "journal-1012: cn->bh is NULL\n") ;
1380 if (!buffer_uptodate(cn->bh)) {
1381 reiserfs_panic(s, "journal-949: buffer write failed\n") ;
1383 /* note, we must clear the JDirty_wait bit after the up to date
1384 ** check, otherwise we race against our flushpage routine
1386 if (!test_and_clear_bit(BH_JDirty_wait, &cn->bh->b_state))
1389 /* undo the inc from journal_mark_dirty */
1397 flush_older_and_return:
1398 /* before we can update the journal header block, we _must_ flush all
1399 ** real blocks from all older transactions to disk. This is because
1400 ** once the header block is updated, this transaction will not be
1401 ** replayed after a crash
1404 flush_older_journal_lists(s, jl);
1407 /* before we can remove everything from the hash tables for this
1408 ** transaction, we must make sure it can never be replayed
1410 ** since we are only called from do_journal_end, we know for sure there
1411 ** are no allocations going on while we are flushing journal lists. So,
1412 ** we only need to update the journal header block for the last list
1416 update_journal_header_block(s, (jl->j_start + jl->j_len + 2) % SB_ONDISK_JOURNAL_SIZE(s), jl->j_trans_id) ;
1418 remove_all_from_journal_list(s, jl, 0) ;
1419 list_del(&jl->j_list);
1420 SB_JOURNAL(s)->j_num_lists--;
1421 del_from_work_list(s, jl);
1423 if (SB_JOURNAL(s)->j_last_flush_id != 0 &&
1424 (jl->j_trans_id - SB_JOURNAL(s)->j_last_flush_id) != 1) {
1425 reiserfs_warning(s, "clm-2201: last flush %lu, current %lu",
1426 SB_JOURNAL(s)->j_last_flush_id,
1429 SB_JOURNAL(s)->j_last_flush_id = jl->j_trans_id;
1431 /* not strictly required since we are freeing the list, but it should
1432 * help find code using dead lists later on
1435 atomic_set(&(jl->j_nonzerolen), 0) ;
1437 jl->j_realblock = NULL ;
1438 jl->j_commit_bh = NULL ;
1439 jl->j_trans_id = 0 ;
1441 put_journal_list(s, jl);
1443 up(&SB_JOURNAL(s)->j_flush_sem);
1447 static int write_one_transaction(struct super_block *s,
1448 struct reiserfs_journal_list *jl,
1449 struct buffer_chunk *chunk)
1451 struct reiserfs_journal_cnode *cn;
1454 jl->j_state |= LIST_TOUCHED;
1455 del_from_work_list(s, jl);
1456 if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1460 cn = jl->j_realblock ;
1462 /* if the blocknr == 0, this has been cleared from the hash,
1465 if (cn->blocknr == 0) {
1468 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1469 struct buffer_head *tmp_bh;
1470 /* we can race against journal_mark_freed when we try
1471 * to lock_buffer(cn->bh), so we have to inc the buffer
1472 * count, and recheck things after locking
1476 lock_buffer(tmp_bh);
1477 if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1478 if (!buffer_journal_dirty(tmp_bh) ||
1479 reiserfs_buffer_prepared(tmp_bh))
1481 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1484 /* note, cn->bh might be null now */
1485 unlock_buffer(tmp_bh);
1496 /* used by flush_commit_list */
1497 static int dirty_one_transaction(struct super_block *s,
1498 struct reiserfs_journal_list *jl)
1500 struct reiserfs_journal_cnode *cn;
1501 struct reiserfs_journal_list *pjl;
1504 jl->j_state |= LIST_DIRTY;
1505 cn = jl->j_realblock ;
1507 /* look for a more recent transaction that logged this
1508 ** buffer. Only the most recent transaction with a buffer in
1509 ** it is allowed to send that buffer to disk
1511 pjl = find_newer_jl_for_cn(cn) ;
1512 if (!pjl && cn->blocknr && cn->bh && buffer_journal_dirty(cn->bh))
1516 /* if the buffer is prepared, it will either be logged
1517 * or restored. If restored, we need to make sure
1518 * it actually gets marked dirty
1520 mark_buffer_notjournal_new(cn->bh) ;
1521 if (test_bit(BH_JPrepared, &cn->bh->b_state)) {
1522 set_bit(BH_JRestore_dirty, &cn->bh->b_state);
1524 set_bit(BH_JTest, &cn->bh->b_state);
1525 mark_buffer_dirty(cn->bh);
1533 static int kupdate_transactions(struct super_block *s,
1534 struct reiserfs_journal_list *jl,
1535 struct reiserfs_journal_list **next_jl,
1536 unsigned long *next_trans_id,
1541 int transactions_flushed = 0;
1542 unsigned long orig_trans_id = jl->j_trans_id;
1543 struct buffer_chunk chunk;
1544 struct list_head *entry;
1547 down(&SB_JOURNAL(s)->j_flush_sem);
1548 if (!journal_list_still_alive(s, orig_trans_id)) {
1552 /* we've got j_flush_sem held, nobody is going to delete any
1553 * of these lists out from underneath us
1555 while((num_trans && transactions_flushed < num_trans) ||
1556 (!num_trans && written < num_blocks)) {
1558 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1559 atomic_read(&jl->j_commit_left))
1561 del_from_work_list(s, jl);
1564 ret = write_one_transaction(s, jl, &chunk);
1568 transactions_flushed++;
1570 entry = jl->j_list.next;
1573 if (entry == &SB_JOURNAL(s)->j_journal_list) {
1576 jl = JOURNAL_LIST_ENTRY(entry);
1578 /* don't bother with older transactions */
1579 if (jl->j_trans_id <= orig_trans_id)
1583 write_chunk(&chunk);
1587 up(&SB_JOURNAL(s)->j_flush_sem);
1591 /* for o_sync and fsync heavy applications, they tend to use
1592 ** all the journa list slots with tiny transactions. These
1593 ** trigger lots and lots of calls to update the header block, which
1594 ** adds seeks and slows things down.
1596 ** This function tries to clear out a large chunk of the journal lists
1597 ** at once, which makes everything faster since only the newest journal
1598 ** list updates the header block
1600 static int flush_used_journal_lists(struct super_block *s,
1601 struct reiserfs_journal_list *jl) {
1602 unsigned long len = 0;
1603 unsigned long cur_len;
1607 struct reiserfs_journal_list *tjl;
1608 struct reiserfs_journal_list *flush_jl;
1609 unsigned long trans_id;
1611 flush_jl = tjl = jl;
1613 /* in data logging mode, try harder to flush a lot of blocks */
1614 if (reiserfs_data_log(s))
1616 /* flush for 256 transactions or limit blocks, whichever comes first */
1617 for(i = 0 ; i < 256 && len < limit ; i++) {
1618 if (atomic_read(&tjl->j_commit_left) ||
1619 tjl->j_trans_id < jl->j_trans_id) {
1622 cur_len = atomic_read(&tjl->j_nonzerolen);
1624 tjl->j_state &= ~LIST_TOUCHED;
1628 if (tjl->j_list.next == &SB_JOURNAL(s)->j_journal_list)
1630 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1632 /* try to find a group of blocks we can flush across all the
1633 ** transactions, but only bother if we've actually spanned
1634 ** across multiple lists
1636 if (flush_jl != jl) {
1637 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1639 flush_journal_list(s, flush_jl, 1);
1644 ** removes any nodes in table with name block and dev as bh.
1645 ** only touchs the hnext and hprev pointers.
1647 void remove_journal_hash(struct super_block *sb,
1648 struct reiserfs_journal_cnode **table,
1649 struct reiserfs_journal_list *jl,
1650 unsigned long block, int remove_freed)
1652 struct reiserfs_journal_cnode *cur ;
1653 struct reiserfs_journal_cnode **head ;
1655 head= &(journal_hash(table, sb, block)) ;
1661 if (cur->blocknr == block && cur->sb == sb && (jl == NULL || jl == cur->jlist) &&
1662 (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1664 cur->hnext->hprev = cur->hprev ;
1667 cur->hprev->hnext = cur->hnext ;
1669 *head = cur->hnext ;
1674 if (cur->bh && cur->jlist) /* anybody who clears the cur->bh will also dec the nonzerolen */
1675 atomic_dec(&(cur->jlist->j_nonzerolen)) ;
1683 static void free_journal_ram(struct super_block *p_s_sb) {
1684 reiserfs_kfree(SB_JOURNAL(p_s_sb)->j_current_jl,
1685 sizeof(struct reiserfs_journal_list), p_s_sb);
1686 SB_JOURNAL(p_s_sb)->j_num_lists--;
1688 vfree(SB_JOURNAL(p_s_sb)->j_cnode_free_orig) ;
1689 free_list_bitmaps(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_bitmap) ;
1690 free_bitmap_nodes(p_s_sb) ; /* must be after free_list_bitmaps */
1691 if (SB_JOURNAL(p_s_sb)->j_header_bh) {
1692 brelse(SB_JOURNAL(p_s_sb)->j_header_bh) ;
1694 /* j_header_bh is on the journal dev, make sure not to release the journal
1695 * dev until we brelse j_header_bh
1697 release_journal_dev(p_s_sb, SB_JOURNAL(p_s_sb));
1698 vfree(SB_JOURNAL(p_s_sb)) ;
1702 ** call on unmount. Only set error to 1 if you haven't made your way out
1703 ** of read_super() yet. Any other caller must keep error at 0.
1705 static int do_journal_release(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, int error) {
1706 struct reiserfs_transaction_handle myth ;
1708 /* we only want to flush out transactions if we were called with error == 0
1710 if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
1711 /* end the current trans */
1712 do_journal_end(th, p_s_sb,10, FLUSH_ALL) ;
1714 /* make sure something gets logged to force our way into the flush code */
1715 journal_join(&myth, p_s_sb, 1) ;
1716 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
1717 journal_mark_dirty(&myth, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
1718 do_journal_end(&myth, p_s_sb,1, FLUSH_ALL) ;
1721 reiserfs_mounted_fs_count-- ;
1722 /* wait for all commits to finish */
1723 cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work);
1724 flush_workqueue(commit_wq);
1725 if (!reiserfs_mounted_fs_count) {
1726 destroy_workqueue(commit_wq);
1730 free_journal_ram(p_s_sb) ;
1736 ** call on unmount. flush all journal trans, release all alloc'd ram
1738 int journal_release(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb) {
1739 return do_journal_release(th, p_s_sb, 0) ;
1742 ** only call from an error condition inside reiserfs_read_super!
1744 int journal_release_error(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb) {
1745 return do_journal_release(th, p_s_sb, 1) ;
1748 /* compares description block with commit block. returns 1 if they differ, 0 if they are the same */
1749 static int journal_compare_desc_commit(struct super_block *p_s_sb, struct reiserfs_journal_desc *desc,
1750 struct reiserfs_journal_commit *commit) {
1751 if (get_commit_trans_id (commit) != get_desc_trans_id (desc) ||
1752 get_commit_trans_len (commit) != get_desc_trans_len (desc) ||
1753 get_commit_trans_len (commit) > SB_JOURNAL_TRANS_MAX(p_s_sb) ||
1754 get_commit_trans_len (commit) <= 0
1760 /* returns 0 if it did not find a description block
1761 ** returns -1 if it found a corrupt commit block
1762 ** returns 1 if both desc and commit were valid
1764 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) {
1765 struct reiserfs_journal_desc *desc ;
1766 struct reiserfs_journal_commit *commit ;
1767 struct buffer_head *c_bh ;
1768 unsigned long offset ;
1773 desc = (struct reiserfs_journal_desc *)d_bh->b_data ;
1774 if (get_desc_trans_len(desc) > 0 && !memcmp(get_journal_desc_magic (d_bh), JOURNAL_DESC_MAGIC, 8)) {
1775 if (oldest_invalid_trans_id && *oldest_invalid_trans_id && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
1776 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-986: transaction "
1777 "is valid returning because trans_id %d is greater than "
1778 "oldest_invalid %lu", get_desc_trans_id(desc),
1779 *oldest_invalid_trans_id);
1782 if (newest_mount_id && *newest_mount_id > get_desc_mount_id (desc)) {
1783 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1087: transaction "
1784 "is valid returning because mount_id %d is less than "
1785 "newest_mount_id %lu", get_desc_mount_id (desc),
1789 if ( get_desc_trans_len(desc) > SB_JOURNAL_TRANS_MAX(p_s_sb) ) {
1790 reiserfs_warning(p_s_sb, "journal-2018: Bad transaction length %d encountered, ignoring transaction", get_desc_trans_len(desc));
1793 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ;
1795 /* ok, we have a journal description block, lets see if the transaction was valid */
1796 c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
1797 ((offset + get_desc_trans_len(desc) + 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ;
1800 commit = (struct reiserfs_journal_commit *)c_bh->b_data ;
1801 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
1802 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1803 "journal_transaction_is_valid, commit offset %ld had bad "
1804 "time %d or length %d",
1805 c_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1806 get_commit_trans_id (commit),
1807 get_commit_trans_len(commit));
1809 if (oldest_invalid_trans_id) {
1810 *oldest_invalid_trans_id = get_desc_trans_id(desc) ;
1811 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1004: "
1812 "transaction_is_valid setting oldest invalid trans_id "
1813 "to %d", get_desc_trans_id(desc)) ;
1818 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1006: found valid "
1819 "transaction start offset %llu, len %d id %d",
1820 d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1821 get_desc_trans_len(desc), get_desc_trans_id(desc)) ;
1828 static void brelse_array(struct buffer_head **heads, int num) {
1830 for (i = 0 ; i < num ; i++) {
1836 ** given the start, and values for the oldest acceptable transactions,
1837 ** this either reads in a replays a transaction, or returns because the transaction
1838 ** is invalid, or too old.
1840 static int journal_read_transaction(struct super_block *p_s_sb, unsigned long cur_dblock, unsigned long oldest_start,
1841 unsigned long oldest_trans_id, unsigned long newest_mount_id) {
1842 struct reiserfs_journal_desc *desc ;
1843 struct reiserfs_journal_commit *commit ;
1844 unsigned long trans_id = 0 ;
1845 struct buffer_head *c_bh ;
1846 struct buffer_head *d_bh ;
1847 struct buffer_head **log_blocks = NULL ;
1848 struct buffer_head **real_blocks = NULL ;
1849 unsigned long trans_offset ;
1853 d_bh = journal_bread(p_s_sb, cur_dblock) ;
1856 desc = (struct reiserfs_journal_desc *)d_bh->b_data ;
1857 trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ;
1858 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
1859 "journal_read_transaction, offset %llu, len %d mount_id %d",
1860 d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1861 get_desc_trans_len(desc), get_desc_mount_id(desc)) ;
1862 if (get_desc_trans_id(desc) < oldest_trans_id) {
1863 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
1864 "journal_read_trans skipping because %lu is too old",
1865 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)) ;
1869 if (get_desc_mount_id(desc) != newest_mount_id) {
1870 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
1871 "journal_read_trans skipping because %d is != "
1872 "newest_mount_id %lu", get_desc_mount_id(desc),
1877 c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
1878 ((trans_offset + get_desc_trans_len(desc) + 1) %
1879 SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ;
1884 commit = (struct reiserfs_journal_commit *)c_bh->b_data ;
1885 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
1886 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal_read_transaction, "
1887 "commit offset %llu had bad time %d or length %d",
1888 c_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1889 get_commit_trans_id(commit), get_commit_trans_len(commit));
1894 trans_id = get_desc_trans_id(desc) ;
1895 /* now we know we've got a good transaction, and it was inside the valid time ranges */
1896 log_blocks = reiserfs_kmalloc(get_desc_trans_len(desc) * sizeof(struct buffer_head *), GFP_NOFS, p_s_sb) ;
1897 real_blocks = reiserfs_kmalloc(get_desc_trans_len(desc) * sizeof(struct buffer_head *), GFP_NOFS, p_s_sb) ;
1898 if (!log_blocks || !real_blocks) {
1901 reiserfs_kfree(log_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1902 reiserfs_kfree(real_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1903 reiserfs_warning(p_s_sb, "journal-1169: kmalloc failed, unable to mount FS") ;
1906 /* get all the buffer heads */
1907 trans_half = journal_trans_half (p_s_sb->s_blocksize) ;
1908 for(i = 0 ; i < get_desc_trans_len(desc) ; i++) {
1909 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));
1910 if (i < trans_half) {
1911 real_blocks[i] = sb_getblk(p_s_sb, le32_to_cpu(desc->j_realblock[i])) ;
1913 real_blocks[i] = sb_getblk(p_s_sb, le32_to_cpu(commit->j_realblock[i - trans_half])) ;
1915 if ( real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb) ) {
1916 reiserfs_warning(p_s_sb, "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem");
1919 /* make sure we don't try to replay onto log or reserved area */
1920 if (is_block_in_log_or_reserved_area(p_s_sb, real_blocks[i]->b_blocknr)) {
1921 reiserfs_warning(p_s_sb, "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block") ;
1923 brelse_array(log_blocks, i) ;
1924 brelse_array(real_blocks, i) ;
1927 reiserfs_kfree(log_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1928 reiserfs_kfree(real_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1932 /* read in the log blocks, memcpy to the corresponding real block */
1933 ll_rw_block(READ, get_desc_trans_len(desc), log_blocks) ;
1934 for (i = 0 ; i < get_desc_trans_len(desc) ; i++) {
1935 wait_on_buffer(log_blocks[i]) ;
1936 if (!buffer_uptodate(log_blocks[i])) {
1937 reiserfs_warning(p_s_sb, "journal-1212: REPLAY FAILURE fsck required! buffer write failed") ;
1938 brelse_array(log_blocks + i, get_desc_trans_len(desc) - i) ;
1939 brelse_array(real_blocks, get_desc_trans_len(desc)) ;
1942 reiserfs_kfree(log_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1943 reiserfs_kfree(real_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1946 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data, real_blocks[i]->b_size) ;
1947 set_buffer_uptodate(real_blocks[i]) ;
1948 brelse(log_blocks[i]) ;
1950 /* flush out the real blocks */
1951 for (i = 0 ; i < get_desc_trans_len(desc) ; i++) {
1952 set_buffer_dirty(real_blocks[i]) ;
1953 ll_rw_block(WRITE, 1, real_blocks + i) ;
1955 for (i = 0 ; i < get_desc_trans_len(desc) ; i++) {
1956 wait_on_buffer(real_blocks[i]) ;
1957 if (!buffer_uptodate(real_blocks[i])) {
1958 reiserfs_warning(p_s_sb, "journal-1226: REPLAY FAILURE, fsck required! buffer write failed") ;
1959 brelse_array(real_blocks + i, get_desc_trans_len(desc) - i) ;
1962 reiserfs_kfree(log_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1963 reiserfs_kfree(real_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ;
1966 brelse(real_blocks[i]) ;
1968 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)) ;
1969 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1095: setting journal "
1970 "start to offset %ld",
1971 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)) ;
1973 /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
1974 SB_JOURNAL(p_s_sb)->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ;
1975 SB_JOURNAL(p_s_sb)->j_last_flush_trans_id = trans_id ;
1976 SB_JOURNAL(p_s_sb)->j_trans_id = trans_id + 1;
1979 reiserfs_kfree(log_blocks, le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), p_s_sb) ;
1980 reiserfs_kfree(real_blocks, le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), p_s_sb) ;
1984 /* This function reads blocks starting from block and to max_block of bufsize
1985 size (but no more than BUFNR blocks at a time). This proved to improve
1986 mounting speed on self-rebuilding raid5 arrays at least.
1987 Right now it is only used from journal code. But later we might use it
1989 Note: Do not use journal_getblk/sb_getblk functions here! */
1990 struct buffer_head * reiserfs_breada (struct block_device *dev, int block, int bufsize,
1991 unsigned int max_block)
1993 struct buffer_head * bhlist[BUFNR];
1994 unsigned int blocks = BUFNR;
1995 struct buffer_head * bh;
1998 bh = __getblk (dev, block, bufsize );
1999 if (buffer_uptodate (bh))
2002 if (block + BUFNR > max_block) {
2003 blocks = max_block - block;
2007 for (i = 1; i < blocks; i++) {
2008 bh = __getblk (dev, block + i, bufsize);
2009 if (buffer_uptodate (bh)) {
2013 else bhlist[j++] = bh;
2015 ll_rw_block (READ, j, bhlist);
2016 for(i = 1; i < j; i++)
2019 wait_on_buffer (bh);
2020 if (buffer_uptodate (bh))
2027 ** read and replay the log
2028 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2029 ** transaction. This tests that before finding all the transactions in the log, which makes normal mount times fast.
2031 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2033 ** On exit, it sets things up so the first transaction will work correctly.
2035 static int journal_read(struct super_block *p_s_sb) {
2036 struct reiserfs_journal_desc *desc ;
2037 unsigned long oldest_trans_id = 0;
2038 unsigned long oldest_invalid_trans_id = 0 ;
2040 unsigned long oldest_start = 0;
2041 unsigned long cur_dblock = 0 ;
2042 unsigned long newest_mount_id = 9 ;
2043 struct buffer_head *d_bh ;
2044 struct reiserfs_journal_header *jh ;
2045 int valid_journal_header = 0 ;
2046 int replay_count = 0 ;
2047 int continue_replay = 1 ;
2049 char b[BDEVNAME_SIZE];
2051 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ;
2052 reiserfs_info (p_s_sb, "checking transaction log (%s)\n",
2053 bdevname(SB_JOURNAL(p_s_sb)->j_dev_bd, b));
2054 start = get_seconds();
2056 /* step 1, read in the journal header block. Check the transaction it says
2057 ** is the first unflushed, and if that transaction is not valid,
2060 SB_JOURNAL(p_s_sb)->j_header_bh = journal_bread(p_s_sb,
2061 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2062 SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2063 if (!SB_JOURNAL(p_s_sb)->j_header_bh) {
2066 jh = (struct reiserfs_journal_header *)(SB_JOURNAL(p_s_sb)->j_header_bh->b_data) ;
2067 if (le32_to_cpu(jh->j_first_unflushed_offset) >= 0 &&
2068 le32_to_cpu(jh->j_first_unflushed_offset) < SB_ONDISK_JOURNAL_SIZE(p_s_sb) &&
2069 le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2070 oldest_start = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2071 le32_to_cpu(jh->j_first_unflushed_offset) ;
2072 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2073 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2074 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1153: found in "
2075 "header: first_unflushed_offset %d, last_flushed_trans_id "
2076 "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2077 le32_to_cpu(jh->j_last_flush_trans_id)) ;
2078 valid_journal_header = 1 ;
2080 /* now, we try to read the first unflushed offset. If it is not valid,
2081 ** there is nothing more we can do, and it makes no sense to read
2082 ** through the whole log.
2084 d_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + le32_to_cpu(jh->j_first_unflushed_offset)) ;
2085 ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL) ;
2087 continue_replay = 0 ;
2090 goto start_log_replay;
2093 if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) {
2094 reiserfs_warning (p_s_sb,
2095 "clm-2076: device is readonly, unable to replay log") ;
2099 /* ok, there are transactions that need to be replayed. start with the first log block, find
2100 ** all the valid transactions, and pick out the oldest.
2102 while(continue_replay && cur_dblock < (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_ONDISK_JOURNAL_SIZE(p_s_sb))) {
2103 /* Note that it is required for blocksize of primary fs device and journal
2104 device to be the same */
2105 d_bh = reiserfs_breada(SB_JOURNAL(p_s_sb)->j_dev_bd, cur_dblock, p_s_sb->s_blocksize,
2106 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_ONDISK_JOURNAL_SIZE(p_s_sb)) ;
2107 ret = journal_transaction_is_valid(p_s_sb, d_bh, &oldest_invalid_trans_id, &newest_mount_id) ;
2109 desc = (struct reiserfs_journal_desc *)d_bh->b_data ;
2110 if (oldest_start == 0) { /* init all oldest_ values */
2111 oldest_trans_id = get_desc_trans_id(desc) ;
2112 oldest_start = d_bh->b_blocknr ;
2113 newest_mount_id = get_desc_mount_id(desc) ;
2114 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1179: Setting "
2115 "oldest_start to offset %llu, trans_id %lu",
2116 oldest_start - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2118 } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2119 /* one we just read was older */
2120 oldest_trans_id = get_desc_trans_id(desc) ;
2121 oldest_start = d_bh->b_blocknr ;
2122 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1180: Resetting "
2123 "oldest_start to offset %lu, trans_id %lu",
2124 oldest_start - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2127 if (newest_mount_id < get_desc_mount_id(desc)) {
2128 newest_mount_id = get_desc_mount_id(desc) ;
2129 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2130 "newest_mount_id to %d", get_desc_mount_id(desc));
2132 cur_dblock += get_desc_trans_len(desc) + 2 ;
2140 cur_dblock = oldest_start ;
2141 if (oldest_trans_id) {
2142 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1206: Starting replay "
2143 "from offset %llu, trans_id %lu",
2144 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2149 while(continue_replay && oldest_trans_id > 0) {
2150 ret = journal_read_transaction(p_s_sb, cur_dblock, oldest_start, oldest_trans_id, newest_mount_id) ;
2153 } else if (ret != 0) {
2156 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_JOURNAL(p_s_sb)->j_start ;
2158 if (cur_dblock == oldest_start)
2162 if (oldest_trans_id == 0) {
2163 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1225: No valid "
2164 "transactions found") ;
2166 /* j_start does not get set correctly if we don't replay any transactions.
2167 ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2168 ** copy the trans_id from the header
2170 if (valid_journal_header && replay_count == 0) {
2171 SB_JOURNAL(p_s_sb)->j_start = le32_to_cpu(jh->j_first_unflushed_offset) ;
2172 SB_JOURNAL(p_s_sb)->j_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2173 SB_JOURNAL(p_s_sb)->j_last_flush_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) ;
2174 SB_JOURNAL(p_s_sb)->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2176 SB_JOURNAL(p_s_sb)->j_mount_id = newest_mount_id + 1 ;
2178 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2179 "newest_mount_id to %lu", SB_JOURNAL(p_s_sb)->j_mount_id) ;
2180 SB_JOURNAL(p_s_sb)->j_first_unflushed_offset = SB_JOURNAL(p_s_sb)->j_start ;
2181 if (replay_count > 0) {
2182 reiserfs_info (p_s_sb, "replayed %d transactions in %lu seconds\n",
2183 replay_count, get_seconds() - start) ;
2185 if (!bdev_read_only(p_s_sb->s_bdev) &&
2186 _update_journal_header_block(p_s_sb, SB_JOURNAL(p_s_sb)->j_start,
2187 SB_JOURNAL(p_s_sb)->j_last_flush_trans_id))
2189 /* replay failed, caller must call free_journal_ram and abort
2197 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2199 struct reiserfs_journal_list *jl;
2201 jl = reiserfs_kmalloc(sizeof(struct reiserfs_journal_list), GFP_NOFS, s);
2206 memset(jl, 0, sizeof(*jl));
2207 INIT_LIST_HEAD(&jl->j_list);
2208 INIT_LIST_HEAD(&jl->j_working_list);
2209 INIT_LIST_HEAD(&jl->j_tail_bh_list);
2210 INIT_LIST_HEAD(&jl->j_bh_list);
2211 sema_init(&jl->j_commit_lock, 1);
2212 SB_JOURNAL(s)->j_num_lists++;
2213 get_journal_list(jl);
2217 static void journal_list_init(struct super_block *p_s_sb) {
2218 SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
2221 static int release_journal_dev( struct super_block *super,
2222 struct reiserfs_journal *journal )
2228 if( journal -> j_dev_file != NULL ) {
2229 result = filp_close( journal -> j_dev_file, NULL );
2230 journal -> j_dev_file = NULL;
2231 journal -> j_dev_bd = NULL;
2232 } else if( journal -> j_dev_bd != NULL ) {
2233 result = blkdev_put( journal -> j_dev_bd );
2234 journal -> j_dev_bd = NULL;
2238 reiserfs_warning(super, "sh-457: release_journal_dev: Cannot release journal device: %i", result );
2243 static int journal_init_dev( struct super_block *super,
2244 struct reiserfs_journal *journal,
2245 const char *jdev_name )
2249 int blkdev_mode = FMODE_READ | FMODE_WRITE;
2250 char b[BDEVNAME_SIZE];
2254 journal -> j_dev_bd = NULL;
2255 journal -> j_dev_file = NULL;
2256 jdev = SB_ONDISK_JOURNAL_DEVICE( super ) ?
2257 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2259 if (bdev_read_only(super->s_bdev))
2260 blkdev_mode = FMODE_READ;
2262 /* there is no "jdev" option and journal is on separate device */
2263 if( ( !jdev_name || !jdev_name[ 0 ] ) ) {
2264 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2265 if (IS_ERR(journal->j_dev_bd)) {
2266 result = PTR_ERR(journal->j_dev_bd);
2267 journal->j_dev_bd = NULL;
2268 reiserfs_warning (super, "sh-458: journal_init_dev: "
2269 "cannot init journal device '%s': %i",
2270 __bdevname(jdev, b), result );
2272 } else if (jdev != super->s_dev)
2273 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2277 journal -> j_dev_file = filp_open( jdev_name, 0, 0 );
2278 if( !IS_ERR( journal -> j_dev_file ) ) {
2279 struct inode *jdev_inode = journal->j_dev_file->f_mapping->host;
2280 if( !S_ISBLK( jdev_inode -> i_mode ) ) {
2281 reiserfs_warning (super, "journal_init_dev: '%s' is "
2282 "not a block device", jdev_name );
2286 journal->j_dev_bd = I_BDEV(jdev_inode);
2287 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2290 result = PTR_ERR( journal -> j_dev_file );
2291 journal -> j_dev_file = NULL;
2292 reiserfs_warning (super,
2293 "journal_init_dev: Cannot open '%s': %i",
2294 jdev_name, result );
2297 release_journal_dev( super, journal );
2299 reiserfs_info(super, "journal_init_dev: journal device: %s\n",
2300 bdevname(journal->j_dev_bd, b));
2305 ** must be called once on fs mount. calls journal_read for you
2307 int journal_init(struct super_block *p_s_sb, const char * j_dev_name, int old_format, unsigned int commit_max_age) {
2308 int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2 ;
2309 struct buffer_head *bhjh;
2310 struct reiserfs_super_block * rs;
2311 struct reiserfs_journal_header *jh;
2312 struct reiserfs_journal *journal;
2313 struct reiserfs_journal_list *jl;
2314 char b[BDEVNAME_SIZE];
2316 journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof (struct reiserfs_journal)) ;
2318 reiserfs_warning (p_s_sb, "journal-1256: unable to get memory for journal structure") ;
2321 memset(journal, 0, sizeof(struct reiserfs_journal)) ;
2322 INIT_LIST_HEAD(&SB_JOURNAL(p_s_sb)->j_bitmap_nodes) ;
2323 INIT_LIST_HEAD (&SB_JOURNAL(p_s_sb)->j_prealloc_list);
2324 INIT_LIST_HEAD(&SB_JOURNAL(p_s_sb)->j_working_list);
2325 INIT_LIST_HEAD(&SB_JOURNAL(p_s_sb)->j_journal_list);
2326 if (reiserfs_allocate_list_bitmaps(p_s_sb,
2327 SB_JOURNAL(p_s_sb)->j_list_bitmap,
2328 SB_BMAP_NR(p_s_sb)))
2329 goto free_and_return ;
2330 allocate_bitmap_nodes(p_s_sb) ;
2332 /* reserved for journal area support */
2333 SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ?
2334 REISERFS_OLD_DISK_OFFSET_IN_BYTES / p_s_sb->s_blocksize +
2335 SB_BMAP_NR(p_s_sb) + 1 :
2336 REISERFS_DISK_OFFSET_IN_BYTES / p_s_sb->s_blocksize + 2);
2338 /* Sanity check to see is the standard journal fitting withing first bitmap
2339 (actual for small blocksizes) */
2340 if ( !SB_ONDISK_JOURNAL_DEVICE( p_s_sb ) &&
2341 (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) + SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8) ) {
2342 reiserfs_warning (p_s_sb, "journal-1393: journal does not fit for area "
2343 "addressed by first of bitmap blocks. It starts at "
2344 "%u and its size is %u. Block size %ld",
2345 SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb),
2346 SB_ONDISK_JOURNAL_SIZE(p_s_sb), p_s_sb->s_blocksize);
2347 goto free_and_return;
2350 if( journal_init_dev( p_s_sb, journal, j_dev_name ) != 0 ) {
2351 reiserfs_warning (p_s_sb, "sh-462: unable to initialize jornal device");
2352 goto free_and_return;
2355 rs = SB_DISK_SUPER_BLOCK(p_s_sb);
2357 /* read journal header */
2358 bhjh = journal_bread(p_s_sb,
2359 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2361 reiserfs_warning (p_s_sb, "sh-459: unable to read journal header");
2362 goto free_and_return;
2364 jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2366 /* make sure that journal matches to the super block */
2367 if (is_reiserfs_jr(rs) && (jh->jh_journal.jp_journal_magic != sb_jp_journal_magic(rs))) {
2368 reiserfs_warning (p_s_sb, "sh-460: journal header magic %x "
2369 "(device %s) does not match to magic found in super "
2371 jh->jh_journal.jp_journal_magic,
2372 bdevname( SB_JOURNAL(p_s_sb)->j_dev_bd, b),
2373 sb_jp_journal_magic(rs));
2375 goto free_and_return;
2378 SB_JOURNAL_TRANS_MAX(p_s_sb) = le32_to_cpu (jh->jh_journal.jp_journal_trans_max);
2379 SB_JOURNAL_MAX_BATCH(p_s_sb) = le32_to_cpu (jh->jh_journal.jp_journal_max_batch);
2380 SB_JOURNAL_MAX_COMMIT_AGE(p_s_sb) = le32_to_cpu (jh->jh_journal.jp_journal_max_commit_age);
2381 SB_JOURNAL_MAX_TRANS_AGE(p_s_sb) = JOURNAL_MAX_TRANS_AGE;
2383 if (SB_JOURNAL_TRANS_MAX(p_s_sb)) {
2384 /* make sure these parameters are available, assign it if they are not */
2385 __u32 initial = SB_JOURNAL_TRANS_MAX(p_s_sb);
2388 if (p_s_sb->s_blocksize < 4096)
2389 ratio = 4096 / p_s_sb->s_blocksize;
2391 if (SB_ONDISK_JOURNAL_SIZE(p_s_sb)/SB_JOURNAL_TRANS_MAX(p_s_sb) < JOURNAL_MIN_RATIO)
2392 SB_JOURNAL_TRANS_MAX(p_s_sb) = SB_ONDISK_JOURNAL_SIZE(p_s_sb) / JOURNAL_MIN_RATIO;
2393 if (SB_JOURNAL_TRANS_MAX(p_s_sb) > JOURNAL_TRANS_MAX_DEFAULT / ratio)
2394 SB_JOURNAL_TRANS_MAX(p_s_sb) = JOURNAL_TRANS_MAX_DEFAULT / ratio;
2395 if (SB_JOURNAL_TRANS_MAX(p_s_sb) < JOURNAL_TRANS_MIN_DEFAULT / ratio)
2396 SB_JOURNAL_TRANS_MAX(p_s_sb) = JOURNAL_TRANS_MIN_DEFAULT / ratio;
2398 if (SB_JOURNAL_TRANS_MAX(p_s_sb) != initial)
2399 reiserfs_warning (p_s_sb, "sh-461: journal_init: wrong transaction max size (%u). Changed to %u",
2400 initial, SB_JOURNAL_TRANS_MAX(p_s_sb));
2402 SB_JOURNAL_MAX_BATCH(p_s_sb) = SB_JOURNAL_TRANS_MAX(p_s_sb)*
2403 JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT;
2406 if (!SB_JOURNAL_TRANS_MAX(p_s_sb)) {
2407 /*we have the file system was created by old version of mkreiserfs
2408 so this field contains zero value */
2409 SB_JOURNAL_TRANS_MAX(p_s_sb) = JOURNAL_TRANS_MAX_DEFAULT ;
2410 SB_JOURNAL_MAX_BATCH(p_s_sb) = JOURNAL_MAX_BATCH_DEFAULT ;
2411 SB_JOURNAL_MAX_COMMIT_AGE(p_s_sb) = JOURNAL_MAX_COMMIT_AGE ;
2413 /* for blocksize >= 4096 - max transaction size is 1024. For block size < 4096
2414 trans max size is decreased proportionally */
2415 if (p_s_sb->s_blocksize < 4096) {
2416 SB_JOURNAL_TRANS_MAX(p_s_sb) /= (4096 / p_s_sb->s_blocksize) ;
2417 SB_JOURNAL_MAX_BATCH(p_s_sb) = (SB_JOURNAL_TRANS_MAX(p_s_sb)) * 9 / 10 ;
2421 SB_JOURNAL_DEFAULT_MAX_COMMIT_AGE(p_s_sb) = SB_JOURNAL_MAX_COMMIT_AGE(p_s_sb);
2423 if (commit_max_age != 0) {
2424 SB_JOURNAL_MAX_COMMIT_AGE(p_s_sb) = commit_max_age;
2425 SB_JOURNAL_MAX_TRANS_AGE(p_s_sb) = commit_max_age;
2428 reiserfs_info (p_s_sb, "journal params: device %s, size %u, "
2429 "journal first block %u, max trans len %u, max batch %u, "
2430 "max commit age %u, max trans age %u\n",
2431 bdevname( SB_JOURNAL(p_s_sb)->j_dev_bd, b),
2432 SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2433 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2434 SB_JOURNAL_TRANS_MAX(p_s_sb),
2435 SB_JOURNAL_MAX_BATCH(p_s_sb),
2436 SB_JOURNAL_MAX_COMMIT_AGE(p_s_sb),
2437 SB_JOURNAL_MAX_TRANS_AGE(p_s_sb));
2441 SB_JOURNAL(p_s_sb)->j_list_bitmap_index = 0 ;
2442 journal_list_init(p_s_sb) ;
2444 memset(SB_JOURNAL(p_s_sb)->j_list_hash_table, 0, JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)) ;
2446 INIT_LIST_HEAD(&SB_JOURNAL(p_s_sb)->j_dirty_buffers) ;
2447 spin_lock_init(&SB_JOURNAL(p_s_sb)->j_dirty_buffers_lock) ;
2449 SB_JOURNAL(p_s_sb)->j_start = 0 ;
2450 SB_JOURNAL(p_s_sb)->j_len = 0 ;
2451 SB_JOURNAL(p_s_sb)->j_len_alloc = 0 ;
2452 atomic_set(&(SB_JOURNAL(p_s_sb)->j_wcount), 0) ;
2453 atomic_set(&(SB_JOURNAL(p_s_sb)->j_async_throttle), 0) ;
2454 SB_JOURNAL(p_s_sb)->j_bcount = 0 ;
2455 SB_JOURNAL(p_s_sb)->j_trans_start_time = 0 ;
2456 SB_JOURNAL(p_s_sb)->j_last = NULL ;
2457 SB_JOURNAL(p_s_sb)->j_first = NULL ;
2458 init_waitqueue_head(&(SB_JOURNAL(p_s_sb)->j_join_wait)) ;
2459 sema_init(&SB_JOURNAL(p_s_sb)->j_lock, 1);
2460 sema_init(&SB_JOURNAL(p_s_sb)->j_flush_sem, 1);
2462 SB_JOURNAL(p_s_sb)->j_trans_id = 10 ;
2463 SB_JOURNAL(p_s_sb)->j_mount_id = 10 ;
2464 SB_JOURNAL(p_s_sb)->j_state = 0 ;
2465 atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 0) ;
2466 SB_JOURNAL(p_s_sb)->j_cnode_free_list = allocate_cnodes(num_cnodes) ;
2467 SB_JOURNAL(p_s_sb)->j_cnode_free_orig = SB_JOURNAL(p_s_sb)->j_cnode_free_list ;
2468 SB_JOURNAL(p_s_sb)->j_cnode_free = SB_JOURNAL(p_s_sb)->j_cnode_free_list ? num_cnodes : 0 ;
2469 SB_JOURNAL(p_s_sb)->j_cnode_used = 0 ;
2470 SB_JOURNAL(p_s_sb)->j_must_wait = 0 ;
2472 init_journal_hash(p_s_sb) ;
2473 jl = SB_JOURNAL(p_s_sb)->j_current_jl;
2474 jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl);
2475 if (!jl->j_list_bitmap) {
2476 reiserfs_warning(p_s_sb, "journal-2005, get_list_bitmap failed for journal list 0") ;
2477 goto free_and_return;
2479 if (journal_read(p_s_sb) < 0) {
2480 reiserfs_warning(p_s_sb, "Replay Failure, unable to mount") ;
2481 goto free_and_return;
2484 reiserfs_mounted_fs_count++ ;
2485 if (reiserfs_mounted_fs_count <= 1)
2486 commit_wq = create_workqueue("reiserfs");
2488 INIT_WORK(&journal->j_work, flush_async_commits, p_s_sb);
2491 free_journal_ram(p_s_sb);
2496 ** test for a polite end of the current transaction. Used by file_write, and should
2497 ** be used by delete to make sure they don't write more than can fit inside a single
2500 int journal_transaction_should_end(struct reiserfs_transaction_handle *th, int new_alloc) {
2501 time_t now = get_seconds() ;
2502 /* cannot restart while nested */
2503 if (th->t_refcount > 1)
2505 if ( SB_JOURNAL(th->t_super)->j_must_wait > 0 ||
2506 (SB_JOURNAL(th->t_super)->j_len_alloc + new_alloc) >= SB_JOURNAL_MAX_BATCH(th->t_super) ||
2507 atomic_read(&(SB_JOURNAL(th->t_super)->j_jlock)) ||
2508 (now - SB_JOURNAL(th->t_super)->j_trans_start_time) > SB_JOURNAL_MAX_TRANS_AGE(th->t_super) ||
2509 SB_JOURNAL(th->t_super)->j_cnode_free < (SB_JOURNAL_TRANS_MAX(th->t_super) * 3)) {
2515 /* this must be called inside a transaction, and requires the
2516 ** kernel_lock to be held
2518 void reiserfs_block_writes(struct reiserfs_transaction_handle *th) {
2519 struct super_block *s = th->t_super ;
2520 SB_JOURNAL(s)->j_must_wait = 1 ;
2521 set_bit(WRITERS_BLOCKED, &SB_JOURNAL(s)->j_state) ;
2525 /* this must be called without a transaction started, and does not
2528 void reiserfs_allow_writes(struct super_block *s) {
2529 clear_bit(WRITERS_BLOCKED, &SB_JOURNAL(s)->j_state) ;
2530 wake_up(&SB_JOURNAL(s)->j_join_wait) ;
2533 /* this must be called without a transaction started, and does not
2536 void reiserfs_wait_on_write_block(struct super_block *s) {
2537 wait_event(SB_JOURNAL(s)->j_join_wait,
2538 !test_bit(WRITERS_BLOCKED, &SB_JOURNAL(s)->j_state)) ;
2541 static void queue_log_writer(struct super_block *s) {
2543 set_bit(WRITERS_QUEUED, &SB_JOURNAL(s)->j_state);
2546 * we don't want to use wait_event here because
2547 * we only want to wait once.
2549 init_waitqueue_entry(&wait, current);
2550 add_wait_queue(&SB_JOURNAL(s)->j_join_wait, &wait);
2551 set_current_state(TASK_UNINTERRUPTIBLE);
2552 if (test_bit(WRITERS_QUEUED, &SB_JOURNAL(s)->j_state))
2554 current->state = TASK_RUNNING;
2555 remove_wait_queue(&SB_JOURNAL(s)->j_join_wait, &wait);
2558 static void wake_queued_writers(struct super_block *s) {
2559 if (test_and_clear_bit(WRITERS_QUEUED, &SB_JOURNAL(s)->j_state))
2560 wake_up(&SB_JOURNAL(s)->j_join_wait);
2563 static void let_transaction_grow(struct super_block *sb,
2564 unsigned long trans_id)
2566 unsigned long bcount = SB_JOURNAL(sb)->j_bcount;
2568 set_current_state(TASK_UNINTERRUPTIBLE);
2569 schedule_timeout(1);
2570 SB_JOURNAL(sb)->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2571 while ((atomic_read(&SB_JOURNAL(sb)->j_wcount) > 0 ||
2572 atomic_read(&SB_JOURNAL(sb)->j_jlock)) &&
2573 SB_JOURNAL(sb)->j_trans_id == trans_id) {
2574 queue_log_writer(sb);
2576 if (SB_JOURNAL(sb)->j_trans_id != trans_id)
2578 if (bcount == SB_JOURNAL(sb)->j_bcount)
2580 bcount = SB_JOURNAL(sb)->j_bcount;
2584 /* join == true if you must join an existing transaction.
2585 ** join == false if you can deal with waiting for others to finish
2587 ** this will block until the transaction is joinable. send the number of blocks you
2588 ** expect to use in nblocks.
2590 static int do_journal_begin_r(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb,unsigned long nblocks,int join) {
2591 time_t now = get_seconds() ;
2593 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2594 struct reiserfs_transaction_handle myth;
2595 int sched_count = 0;
2597 reiserfs_check_lock_depth(p_s_sb, "journal_begin") ;
2598 RFALSE( p_s_sb->s_flags & MS_RDONLY,
2599 "clm-2078: calling journal_begin on readonly FS") ;
2601 PROC_INFO_INC( p_s_sb, journal.journal_being );
2602 /* set here for journal_join */
2604 th->t_super = p_s_sb ;
2607 lock_journal(p_s_sb) ;
2608 journal->j_bcount++;
2610 if (test_bit(WRITERS_BLOCKED, &journal->j_state)) {
2611 unlock_journal(p_s_sb) ;
2612 reiserfs_wait_on_write_block(p_s_sb) ;
2613 PROC_INFO_INC( p_s_sb, journal.journal_relock_writers );
2616 now = get_seconds();
2618 /* if there is no room in the journal OR
2619 ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
2620 ** we don't sleep if there aren't other writers
2623 if ( (!join && journal->j_must_wait > 0) ||
2624 ( !join && (journal->j_len_alloc + nblocks + 2) >= SB_JOURNAL_MAX_BATCH(p_s_sb)) ||
2625 (!join && atomic_read(&journal->j_wcount) > 0 && journal->j_trans_start_time > 0 &&
2626 (now - journal->j_trans_start_time) > SB_JOURNAL_MAX_TRANS_AGE(p_s_sb)) ||
2627 (!join && atomic_read(&journal->j_jlock)) ||
2628 (!join && journal->j_cnode_free < (SB_JOURNAL_TRANS_MAX(p_s_sb) * 3))) {
2630 old_trans_id = journal->j_trans_id;
2631 unlock_journal(p_s_sb) ; /* allow others to finish this transaction */
2633 if (!join && (journal->j_len_alloc + nblocks + 2) >=
2634 SB_JOURNAL_MAX_BATCH(p_s_sb) &&
2635 ((journal->j_len + nblocks + 2) * 100) < (journal->j_len_alloc * 75))
2637 if (atomic_read(&journal->j_wcount) > 10) {
2639 queue_log_writer(p_s_sb);
2643 /* don't mess with joining the transaction if all we have to do is
2644 * wait for someone else to do a commit
2646 if (atomic_read(&journal->j_jlock)) {
2647 while (journal->j_trans_id == old_trans_id &&
2648 atomic_read(&journal->j_jlock)) {
2649 queue_log_writer(p_s_sb);
2653 journal_join(&myth, p_s_sb, 1) ;
2655 /* someone might have ended the transaction while we joined */
2656 if (old_trans_id != SB_JOURNAL(p_s_sb)->j_trans_id) {
2657 do_journal_end(&myth, p_s_sb, 1, 0) ;
2659 do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW) ;
2662 PROC_INFO_INC( p_s_sb, journal.journal_relock_wcount );
2665 /* we are the first writer, set trans_id */
2666 if (journal->j_trans_start_time == 0) {
2667 journal->j_trans_start_time = get_seconds();
2669 atomic_inc(&(journal->j_wcount)) ;
2670 journal->j_len_alloc += nblocks ;
2671 th->t_blocks_logged = 0 ;
2672 th->t_blocks_allocated = nblocks ;
2673 th->t_trans_id = journal->j_trans_id ;
2674 unlock_journal(p_s_sb) ;
2678 struct reiserfs_transaction_handle *
2679 reiserfs_persistent_transaction(struct super_block *s, int nblocks) {
2681 struct reiserfs_transaction_handle *th ;
2683 /* if we're nesting into an existing transaction. It will be
2684 ** persistent on its own
2686 if (reiserfs_transaction_running(s)) {
2687 th = current->journal_info ;
2689 if (th->t_refcount < 2) {
2694 th = reiserfs_kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS, s) ;
2697 ret = journal_begin(th, s, nblocks) ;
2699 reiserfs_kfree(th, sizeof(struct reiserfs_transaction_handle), s) ;
2706 reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th) {
2707 struct super_block *s = th->t_super;
2709 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
2710 if (th->t_refcount == 0)
2711 reiserfs_kfree(th, sizeof(struct reiserfs_transaction_handle), s) ;
2715 static int journal_join(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) {
2716 struct reiserfs_transaction_handle *cur_th = current->journal_info;
2718 /* this keeps do_journal_end from NULLing out the current->journal_info
2721 th->t_handle_save = cur_th ;
2722 if (cur_th && cur_th->t_refcount > 1) {
2725 return do_journal_begin_r(th, p_s_sb, nblocks, 1) ;
2728 int journal_begin(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb, unsigned long nblocks) {
2729 struct reiserfs_transaction_handle *cur_th = current->journal_info ;
2732 th->t_handle_save = NULL ;
2734 /* we are nesting into the current transaction */
2735 if (cur_th->t_super == p_s_sb) {
2736 cur_th->t_refcount++ ;
2737 memcpy(th, cur_th, sizeof(*th));
2738 if (th->t_refcount <= 1)
2739 reiserfs_warning (p_s_sb, "BAD: refcount <= 1, but journal_info != 0");
2742 /* we've ended up with a handle from a different filesystem.
2743 ** save it and restore on journal_end. This should never
2746 reiserfs_warning(p_s_sb, "clm-2100: nesting info a different FS") ;
2747 th->t_handle_save = current->journal_info ;
2748 current->journal_info = th;
2751 current->journal_info = th;
2753 ret = do_journal_begin_r(th, p_s_sb, nblocks, 0) ;
2754 if (current->journal_info != th)
2760 ** puts bh into the current transaction. If it was already there, reorders removes the
2761 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
2763 ** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the
2764 ** transaction is committed.
2766 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
2768 int journal_mark_dirty(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, struct buffer_head *bh) {
2769 struct reiserfs_journal_cnode *cn = NULL;
2770 int count_already_incd = 0 ;
2773 PROC_INFO_INC( p_s_sb, journal.mark_dirty );
2774 if (th->t_trans_id != SB_JOURNAL(p_s_sb)->j_trans_id) {
2775 reiserfs_panic(th->t_super, "journal-1577: handle trans id %ld != current trans id %ld\n",
2776 th->t_trans_id, SB_JOURNAL(p_s_sb)->j_trans_id);
2780 prepared = test_and_clear_bit(BH_JPrepared, &bh->b_state) ;
2781 clear_bit(BH_JRestore_dirty, &bh->b_state);
2782 /* already in this transaction, we are done */
2783 if (buffer_journaled(bh)) {
2784 PROC_INFO_INC( p_s_sb, journal.mark_dirty_already );
2788 /* this must be turned into a panic instead of a warning. We can't allow
2789 ** a dirty or journal_dirty or locked buffer to be logged, as some changes
2790 ** could get to disk too early. NOT GOOD.
2792 if (!prepared || buffer_dirty(bh)) {
2793 reiserfs_warning (p_s_sb, "journal-1777: buffer %llu bad state "
2794 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
2795 (unsigned long long)bh->b_blocknr, prepared ? ' ' : '!',
2796 buffer_locked(bh) ? ' ' : '!',
2797 buffer_dirty(bh) ? ' ' : '!',
2798 buffer_journal_dirty(bh) ? ' ' : '!') ;
2801 if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) <= 0) {
2802 reiserfs_warning (p_s_sb, "journal-1409: journal_mark_dirty returning because j_wcount was %d", atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount))) ;
2805 /* this error means I've screwed up, and we've overflowed the transaction.
2806 ** Nothing can be done here, except make the FS readonly or panic.
2808 if (SB_JOURNAL(p_s_sb)->j_len >= SB_JOURNAL_TRANS_MAX(p_s_sb)) {
2809 reiserfs_panic(th->t_super, "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n", SB_JOURNAL(p_s_sb)->j_len) ;
2812 if (buffer_journal_dirty(bh)) {
2813 count_already_incd = 1 ;
2814 PROC_INFO_INC( p_s_sb, journal.mark_dirty_notjournal );
2815 mark_buffer_notjournal_dirty(bh) ;
2818 if (SB_JOURNAL(p_s_sb)->j_len > SB_JOURNAL(p_s_sb)->j_len_alloc) {
2819 SB_JOURNAL(p_s_sb)->j_len_alloc = SB_JOURNAL(p_s_sb)->j_len + JOURNAL_PER_BALANCE_CNT ;
2822 set_bit(BH_JDirty, &bh->b_state) ;
2824 /* now put this guy on the end */
2826 cn = get_cnode(p_s_sb) ;
2828 reiserfs_panic(p_s_sb, "get_cnode failed!\n");
2831 if (th->t_blocks_logged == th->t_blocks_allocated) {
2832 th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT ;
2833 SB_JOURNAL(p_s_sb)->j_len_alloc += JOURNAL_PER_BALANCE_CNT ;
2835 th->t_blocks_logged++ ;
2836 SB_JOURNAL(p_s_sb)->j_len++ ;
2839 cn->blocknr = bh->b_blocknr ;
2842 insert_journal_hash(SB_JOURNAL(p_s_sb)->j_hash_table, cn) ;
2843 if (!count_already_incd) {
2848 cn->prev = SB_JOURNAL(p_s_sb)->j_last ;
2850 if (SB_JOURNAL(p_s_sb)->j_last) {
2851 SB_JOURNAL(p_s_sb)->j_last->next = cn ;
2852 SB_JOURNAL(p_s_sb)->j_last = cn ;
2854 SB_JOURNAL(p_s_sb)->j_first = cn ;
2855 SB_JOURNAL(p_s_sb)->j_last = cn ;
2860 int journal_end(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) {
2861 if (!current->journal_info && th->t_refcount > 1)
2862 reiserfs_warning (p_s_sb, "REISER-NESTING: th NULL, refcount %d",
2866 if (th->t_refcount > 0) {
2867 struct reiserfs_transaction_handle *cur_th = current->journal_info ;
2869 /* we aren't allowed to close a nested transaction on a different
2870 ** filesystem from the one in the task struct
2872 if (cur_th->t_super != th->t_super)
2876 memcpy(current->journal_info, th, sizeof(*th));
2881 return do_journal_end(th, p_s_sb, nblocks, 0) ;
2885 /* removes from the current transaction, relsing and descrementing any counters.
2886 ** also files the removed buffer directly onto the clean list
2888 ** called by journal_mark_freed when a block has been deleted
2890 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
2892 static int remove_from_transaction(struct super_block *p_s_sb, b_blocknr_t blocknr, int already_cleaned) {
2893 struct buffer_head *bh ;
2894 struct reiserfs_journal_cnode *cn ;
2897 cn = get_journal_hash_dev(p_s_sb, SB_JOURNAL(p_s_sb)->j_hash_table, blocknr) ;
2898 if (!cn || !cn->bh) {
2903 cn->prev->next = cn->next ;
2906 cn->next->prev = cn->prev ;
2908 if (cn == SB_JOURNAL(p_s_sb)->j_first) {
2909 SB_JOURNAL(p_s_sb)->j_first = cn->next ;
2911 if (cn == SB_JOURNAL(p_s_sb)->j_last) {
2912 SB_JOURNAL(p_s_sb)->j_last = cn->prev ;
2915 remove_journal_hash(p_s_sb, SB_JOURNAL(p_s_sb)->j_hash_table, NULL, bh->b_blocknr, 0) ;
2916 mark_buffer_not_journaled(bh) ; /* don't log this one */
2918 if (!already_cleaned) {
2919 mark_buffer_notjournal_dirty(bh) ;
2921 if (atomic_read(&(bh->b_count)) < 0) {
2922 reiserfs_warning (p_s_sb, "journal-1752: remove from trans, b_count < 0");
2926 SB_JOURNAL(p_s_sb)->j_len-- ;
2927 SB_JOURNAL(p_s_sb)->j_len_alloc-- ;
2928 free_cnode(p_s_sb, cn) ;
2933 ** for any cnode in a journal list, it can only be dirtied of all the
2934 ** transactions that include it are commited to disk.
2935 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
2936 ** and 0 if you aren't
2938 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
2939 ** blocks for a given transaction on disk
2942 static int can_dirty(struct reiserfs_journal_cnode *cn) {
2943 struct super_block *sb = cn->sb;
2944 b_blocknr_t blocknr = cn->blocknr ;
2945 struct reiserfs_journal_cnode *cur = cn->hprev ;
2948 /* first test hprev. These are all newer than cn, so any node here
2949 ** with the same block number and dev means this node can't be sent
2950 ** to disk right now.
2952 while(cur && can_dirty) {
2953 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
2954 cur->blocknr == blocknr) {
2959 /* then test hnext. These are all older than cn. As long as they
2960 ** are committed to the log, it is safe to write cn to disk
2963 while(cur && can_dirty) {
2964 if (cur->jlist && cur->jlist->j_len > 0 &&
2965 atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
2966 cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
2974 /* syncs the commit blocks, but does not force the real buffers to disk
2975 ** will wait until the current transaction is done/commited before returning
2977 int journal_end_sync(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) {
2979 /* you can sync while nested, very, very bad */
2980 if (th->t_refcount > 1) {
2983 if (SB_JOURNAL(p_s_sb)->j_len == 0) {
2984 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
2985 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
2987 return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT) ;
2991 ** writeback the pending async commits to disk
2993 static void flush_async_commits(void *p) {
2994 struct super_block *p_s_sb = p;
2995 struct reiserfs_journal_list *jl;
2996 struct list_head *entry;
2999 if (!list_empty(&SB_JOURNAL(p_s_sb)->j_journal_list)) {
3000 /* last entry is the youngest, commit it and you get everything */
3001 entry = SB_JOURNAL(p_s_sb)->j_journal_list.prev;
3002 jl = JOURNAL_LIST_ENTRY(entry);
3003 flush_commit_list(p_s_sb, jl, 1);
3007 * this is a little racey, but there's no harm in missing
3008 * the filemap_fdata_write
3010 if (!atomic_read(&SB_JOURNAL(p_s_sb)->j_async_throttle)) {
3011 atomic_inc(&SB_JOURNAL(p_s_sb)->j_async_throttle);
3012 filemap_fdatawrite(p_s_sb->s_bdev->bd_inode->i_mapping);
3013 atomic_dec(&SB_JOURNAL(p_s_sb)->j_async_throttle);
3018 ** flushes any old transactions to disk
3019 ** ends the current transaction if it is too old
3021 int reiserfs_flush_old_commits(struct super_block *p_s_sb) {
3023 struct reiserfs_transaction_handle th ;
3025 now = get_seconds();
3026 /* safety check so we don't flush while we are replaying the log during
3029 if (list_empty(&SB_JOURNAL(p_s_sb)->j_journal_list)) {
3033 /* check the current transaction. If there are no writers, and it is
3034 * too old, finish it, and force the commit blocks to disk
3036 if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) <= 0 &&
3037 SB_JOURNAL(p_s_sb)->j_trans_start_time > 0 &&
3038 SB_JOURNAL(p_s_sb)->j_len > 0 &&
3039 (now - SB_JOURNAL(p_s_sb)->j_trans_start_time) >
3040 SB_JOURNAL_MAX_TRANS_AGE(p_s_sb))
3042 journal_join(&th, p_s_sb, 1) ;
3043 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
3044 journal_mark_dirty(&th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
3046 /* we're only being called from kreiserfsd, it makes no sense to do
3047 ** an async commit so that kreiserfsd can do it later
3049 do_journal_end(&th, p_s_sb,1, COMMIT_NOW | WAIT) ;
3051 return p_s_sb->s_dirt;
3055 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3057 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3058 ** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just
3059 ** flushes the commit list and returns 0.
3061 ** Won't batch when flush or commit_now is set. Also won't batch when others are waiting on j_join_wait.
3063 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3065 static int check_journal_end(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb,
3066 unsigned long nblocks, int flags) {
3069 int flush = flags & FLUSH_ALL ;
3070 int commit_now = flags & COMMIT_NOW ;
3071 int wait_on_commit = flags & WAIT ;
3072 struct reiserfs_journal_list *jl;
3074 if (th->t_trans_id != SB_JOURNAL(p_s_sb)->j_trans_id) {
3075 reiserfs_panic(th->t_super, "journal-1577: handle trans id %ld != current trans id %ld\n",
3076 th->t_trans_id, SB_JOURNAL(p_s_sb)->j_trans_id);
3079 SB_JOURNAL(p_s_sb)->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged) ;
3080 if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) > 0) { /* <= 0 is allowed. unmounting might not call begin */
3081 atomic_dec(&(SB_JOURNAL(p_s_sb)->j_wcount)) ;
3084 /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3085 ** will be dealt with by next transaction that actually writes something, but should be taken
3086 ** care of in this trans
3088 if (SB_JOURNAL(p_s_sb)->j_len == 0) {
3091 /* if wcount > 0, and we are called to with flush or commit_now,
3092 ** we wait on j_join_wait. We will wake up when the last writer has
3093 ** finished the transaction, and started it on its way to the disk.
3094 ** Then, we flush the commit or journal list, and just return 0
3095 ** because the rest of journal end was already done for this transaction.
3097 if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) > 0) {
3098 if (flush || commit_now) {
3101 jl = SB_JOURNAL(p_s_sb)->j_current_jl;
3102 trans_id = jl->j_trans_id;
3104 jl->j_state |= LIST_COMMIT_PENDING;
3105 atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 1) ;
3107 SB_JOURNAL(p_s_sb)->j_next_full_flush = 1 ;
3109 unlock_journal(p_s_sb) ;
3111 /* sleep while the current transaction is still j_jlocked */
3112 while(SB_JOURNAL(p_s_sb)->j_trans_id == trans_id) {
3113 if (atomic_read(&SB_JOURNAL(p_s_sb)->j_jlock)) {
3114 queue_log_writer(p_s_sb);
3116 lock_journal(p_s_sb);
3117 if (SB_JOURNAL(p_s_sb)->j_trans_id == trans_id) {
3118 atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 1) ;
3120 unlock_journal(p_s_sb);
3123 if (SB_JOURNAL(p_s_sb)->j_trans_id == trans_id) {
3126 if (commit_now && journal_list_still_alive(p_s_sb, trans_id) &&
3129 flush_commit_list(p_s_sb, jl, 1) ;
3133 unlock_journal(p_s_sb) ;
3137 /* deal with old transactions where we are the last writers */
3138 now = get_seconds();
3139 if ((now - SB_JOURNAL(p_s_sb)->j_trans_start_time) > SB_JOURNAL_MAX_TRANS_AGE(p_s_sb)) {
3141 SB_JOURNAL(p_s_sb)->j_next_async_flush = 1 ;
3143 /* don't batch when someone is waiting on j_join_wait */
3144 /* don't batch when syncing the commit or flushing the whole trans */
3145 if (!(SB_JOURNAL(p_s_sb)->j_must_wait > 0) && !(atomic_read(&(SB_JOURNAL(p_s_sb)->j_jlock))) && !flush && !commit_now &&
3146 (SB_JOURNAL(p_s_sb)->j_len < SB_JOURNAL_MAX_BATCH(p_s_sb)) &&
3147 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)) {
3148 SB_JOURNAL(p_s_sb)->j_bcount++ ;
3149 unlock_journal(p_s_sb) ;
3153 if (SB_JOURNAL(p_s_sb)->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
3154 reiserfs_panic(p_s_sb, "journal-003: journal_end: j_start (%ld) is too high\n", SB_JOURNAL(p_s_sb)->j_start) ;
3160 ** Does all the work that makes deleting blocks safe.
3161 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3164 ** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes
3165 ** before this transaction has finished.
3167 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with
3168 ** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash,
3169 ** the block can't be reallocated yet.
3171 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3173 int journal_mark_freed(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, b_blocknr_t blocknr) {
3174 struct reiserfs_journal_cnode *cn = NULL ;
3175 struct buffer_head *bh = NULL ;
3176 struct reiserfs_list_bitmap *jb = NULL ;
3179 cn = get_journal_hash_dev(p_s_sb, SB_JOURNAL(p_s_sb)->j_hash_table, blocknr);
3184 /* if it is journal new, we just remove it from this transaction */
3185 if (bh && buffer_journal_new(bh)) {
3186 mark_buffer_notjournal_new(bh) ;
3187 clear_prepared_bits(bh) ;
3188 reiserfs_clean_and_file_buffer(bh) ;
3189 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned) ;
3191 /* set the bit for this block in the journal bitmap for this transaction */
3192 jb = SB_JOURNAL(p_s_sb)->j_current_jl->j_list_bitmap;
3194 reiserfs_panic(p_s_sb, "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n") ;
3196 set_bit_in_list_bitmap(p_s_sb, blocknr, jb) ;
3198 /* Note, the entire while loop is not allowed to schedule. */
3201 clear_prepared_bits(bh) ;
3202 reiserfs_clean_and_file_buffer(bh) ;
3204 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned) ;
3206 /* find all older transactions with this block, make sure they don't try to write it out */
3207 cn = get_journal_hash_dev(p_s_sb,SB_JOURNAL(p_s_sb)->j_list_hash_table, blocknr) ;
3209 if (p_s_sb == cn->sb && blocknr == cn->blocknr) {
3210 set_bit(BLOCK_FREED, &cn->state) ;
3213 /* remove_from_transaction will brelse the buffer if it was
3214 ** in the current trans
3216 mark_buffer_notjournal_dirty(cn->bh) ;
3219 if (atomic_read(&(cn->bh->b_count)) < 0) {
3220 reiserfs_warning (p_s_sb, "journal-2138: cn->bh->b_count < 0");
3223 if (cn->jlist) { /* since we are clearing the bh, we MUST dec nonzerolen */
3224 atomic_dec(&(cn->jlist->j_nonzerolen)) ;
3234 put_bh(bh) ; /* get_hash grabs the buffer */
3235 if (atomic_read(&(bh->b_count)) < 0) {
3236 reiserfs_warning (p_s_sb, "journal-2165: bh->b_count < 0");
3242 void reiserfs_update_inode_transaction(struct inode *inode) {
3243 REISERFS_I(inode)->i_jl = SB_JOURNAL(inode->i_sb)->j_current_jl;
3244 REISERFS_I(inode)->i_trans_id = SB_JOURNAL(inode->i_sb)->j_trans_id ;
3248 * returns -1 on error, 0 if no commits/barriers were done and 1
3249 * if a transaction was actually committed and the barrier was done
3251 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3252 struct reiserfs_journal_list *jl)
3254 struct reiserfs_transaction_handle th ;
3255 struct super_block *sb = inode->i_sb ;
3258 /* is it from the current transaction, or from an unknown transaction? */
3259 if (id == SB_JOURNAL(sb)->j_trans_id) {
3260 jl = SB_JOURNAL(sb)->j_current_jl;
3261 /* try to let other writers come in and grow this transaction */
3262 let_transaction_grow(sb, id);
3263 if (SB_JOURNAL(sb)->j_trans_id != id) {
3264 goto flush_commit_only;
3267 journal_begin(&th, sb, 1) ;
3269 /* someone might have ended this transaction while we joined */
3270 if (SB_JOURNAL(sb)->j_trans_id != id) {
3271 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 1) ;
3272 journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb)) ;
3273 journal_end(&th, sb, 1) ;
3274 goto flush_commit_only;
3277 journal_end_sync(&th, sb, 1) ;
3281 /* this gets tricky, we have to make sure the journal list in
3282 * the inode still exists. We know the list is still around
3283 * if we've got a larger transaction id than the oldest list
3286 if (journal_list_still_alive(inode->i_sb, id)) {
3288 * we only set ret to 1 when we know for sure
3289 * the barrier hasn't been started yet on the commit
3292 if (atomic_read(&jl->j_commit_left) > 1)
3294 flush_commit_list(sb, jl, 1) ;
3297 /* otherwise the list is gone, and long since committed */
3301 int reiserfs_commit_for_inode(struct inode *inode) {
3302 unsigned long id = REISERFS_I(inode)->i_trans_id;
3303 struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3305 /* for the whole inode, assume unset id means it was
3306 * changed in the current transaction. More conservative
3309 reiserfs_update_inode_transaction(inode) ;
3310 id = REISERFS_I(inode)->i_trans_id;
3311 /* jl will be updated in __commit_trans_jl */
3314 return __commit_trans_jl(inode, id, jl);
3317 void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb,
3318 struct buffer_head *bh) {
3319 PROC_INFO_INC( p_s_sb, journal.restore_prepared );
3323 if (test_and_clear_bit(BH_JRestore_dirty, &bh->b_state) &&
3324 buffer_journal_dirty(bh)) {
3325 struct reiserfs_journal_cnode *cn;
3326 cn = get_journal_hash_dev(p_s_sb,
3327 SB_JOURNAL(p_s_sb)->j_list_hash_table,
3329 if (cn && can_dirty(cn)) {
3330 set_bit(BH_JTest, &bh->b_state);
3331 mark_buffer_dirty(bh);
3334 clear_bit(BH_JPrepared, &bh->b_state) ;
3337 extern struct tree_balance *cur_tb ;
3339 ** before we can change a metadata block, we have to make sure it won't
3340 ** be written to disk while we are altering it. So, we must:
3345 int reiserfs_prepare_for_journal(struct super_block *p_s_sb,
3346 struct buffer_head *bh, int wait) {
3347 PROC_INFO_INC( p_s_sb, journal.prepare );
3349 if (test_set_buffer_locked(bh)) {
3354 set_bit(BH_JPrepared, &bh->b_state);
3355 if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3356 clear_bit(BH_JTest, &bh->b_state);
3357 set_bit(BH_JRestore_dirty, &bh->b_state);
3363 static void flush_old_journal_lists(struct super_block *s) {
3364 struct reiserfs_journal_list *jl;
3365 struct list_head *entry;
3366 time_t now = get_seconds();
3368 while(!list_empty(&SB_JOURNAL(s)->j_journal_list)) {
3369 entry = SB_JOURNAL(s)->j_journal_list.next;
3370 jl = JOURNAL_LIST_ENTRY(entry);
3371 /* this check should always be run, to send old lists to disk */
3372 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4))) {
3373 flush_used_journal_lists(s, jl);
3381 ** long and ugly. If flush, will not return until all commit
3382 ** blocks and all real buffers in the trans are on disk.
3383 ** If no_async, won't return until all commit blocks are on disk.
3385 ** keep reading, there are comments as you go along
3387 static int do_journal_end(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb, unsigned long nblocks,
3389 struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3390 struct reiserfs_journal_cnode *last_cn = NULL;
3391 struct reiserfs_journal_desc *desc ;
3392 struct reiserfs_journal_commit *commit ;
3393 struct buffer_head *c_bh ; /* commit bh */
3394 struct buffer_head *d_bh ; /* desc bh */
3395 int cur_write_start = 0 ; /* start index of current log write */
3398 int flush = flags & FLUSH_ALL ;
3399 int wait_on_commit = flags & WAIT ;
3400 struct reiserfs_journal_list *jl, *temp_jl;
3401 struct list_head *entry, *safe;
3402 unsigned long jindex;
3403 unsigned long commit_trans_id;
3406 if (th->t_refcount > 1)
3409 current->journal_info = th->t_handle_save;
3410 reiserfs_check_lock_depth(p_s_sb, "journal end");
3411 if (SB_JOURNAL(p_s_sb)->j_len == 0) {
3412 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
3413 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
3416 lock_journal(p_s_sb) ;
3417 if (SB_JOURNAL(p_s_sb)->j_next_full_flush) {
3418 flags |= FLUSH_ALL ;
3421 if (SB_JOURNAL(p_s_sb)->j_next_async_flush) {
3422 flags |= COMMIT_NOW | WAIT;
3426 /* check_journal_end locks the journal, and unlocks if it does not return 1
3427 ** it tells us if we should continue with the journal_end, or just return
3429 if (!check_journal_end(th, p_s_sb, nblocks, flags)) {
3431 wake_queued_writers(p_s_sb);
3432 reiserfs_async_progress_wait(p_s_sb);
3436 /* check_journal_end might set these, check again */
3437 if (SB_JOURNAL(p_s_sb)->j_next_full_flush) {
3442 ** j must wait means we have to flush the log blocks, and the real blocks for
3445 if (SB_JOURNAL(p_s_sb)->j_must_wait > 0) {
3449 #ifdef REISERFS_PREALLOCATE
3450 /* quota ops might need to nest, setup the journal_info pointer for them */
3451 current->journal_info = th ;
3452 reiserfs_discard_all_prealloc(th); /* it should not involve new blocks into
3453 * the transaction */
3454 current->journal_info = th->t_handle_save ;
3457 /* setup description block */
3458 d_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_JOURNAL(p_s_sb)->j_start) ;
3459 set_buffer_uptodate(d_bh);
3460 desc = (struct reiserfs_journal_desc *)(d_bh)->b_data ;
3461 memset(d_bh->b_data, 0, d_bh->b_size) ;
3462 memcpy(get_journal_desc_magic (d_bh), JOURNAL_DESC_MAGIC, 8) ;
3463 set_desc_trans_id(desc, SB_JOURNAL(p_s_sb)->j_trans_id) ;
3465 /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */
3466 c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3467 ((SB_JOURNAL(p_s_sb)->j_start + SB_JOURNAL(p_s_sb)->j_len + 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ;
3468 commit = (struct reiserfs_journal_commit *)c_bh->b_data ;
3469 memset(c_bh->b_data, 0, c_bh->b_size) ;
3470 set_commit_trans_id(commit, SB_JOURNAL(p_s_sb)->j_trans_id) ;
3471 set_buffer_uptodate(c_bh) ;
3473 /* init this journal list */
3474 jl = SB_JOURNAL(p_s_sb)->j_current_jl;
3476 /* we lock the commit before doing anything because
3477 * we want to make sure nobody tries to run flush_commit_list until
3478 * the new transaction is fully setup, and we've already flushed the
3481 down(&jl->j_commit_lock);
3483 /* save the transaction id in case we need to commit it later */
3484 commit_trans_id = jl->j_trans_id;
3486 atomic_set(&jl->j_older_commits_done, 0) ;
3487 jl->j_trans_id = SB_JOURNAL(p_s_sb)->j_trans_id ;
3488 jl->j_timestamp = SB_JOURNAL(p_s_sb)->j_trans_start_time ;
3489 jl->j_commit_bh = c_bh ;
3490 jl->j_start = SB_JOURNAL(p_s_sb)->j_start ;
3491 jl->j_len = SB_JOURNAL(p_s_sb)->j_len ;
3492 atomic_set(&jl->j_nonzerolen, SB_JOURNAL(p_s_sb)->j_len) ;
3493 atomic_set(&jl->j_commit_left, SB_JOURNAL(p_s_sb)->j_len + 2);
3494 jl->j_realblock = NULL ;
3496 /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
3497 ** for each real block, add it to the journal list hash,
3498 ** copy into real block index array in the commit or desc block
3500 trans_half = journal_trans_half(p_s_sb->s_blocksize);
3501 for (i = 0, cn = SB_JOURNAL(p_s_sb)->j_first ; cn ; cn = cn->next, i++) {
3502 if (test_bit(BH_JDirty, &cn->bh->b_state) ) {
3503 jl_cn = get_cnode(p_s_sb) ;
3505 reiserfs_panic(p_s_sb, "journal-1676, get_cnode returned NULL\n") ;
3508 jl->j_realblock = jl_cn ;
3510 jl_cn->prev = last_cn ;
3511 jl_cn->next = NULL ;
3513 last_cn->next = jl_cn ;
3516 /* make sure the block we are trying to log is not a block
3517 of journal or reserved area */
3519 if (is_block_in_log_or_reserved_area(p_s_sb, cn->bh->b_blocknr)) {
3520 reiserfs_panic(p_s_sb, "journal-2332: Trying to log block %lu, which is a log block\n", cn->bh->b_blocknr) ;
3522 jl_cn->blocknr = cn->bh->b_blocknr ;
3525 jl_cn->bh = cn->bh ;
3527 insert_journal_hash(SB_JOURNAL(p_s_sb)->j_list_hash_table, jl_cn) ;
3528 if (i < trans_half) {
3529 desc->j_realblock[i] = cpu_to_le32(cn->bh->b_blocknr) ;
3531 commit->j_realblock[i - trans_half] = cpu_to_le32(cn->bh->b_blocknr) ;
3537 set_desc_trans_len(desc, SB_JOURNAL(p_s_sb)->j_len) ;
3538 set_desc_mount_id(desc, SB_JOURNAL(p_s_sb)->j_mount_id) ;
3539 set_desc_trans_id(desc, SB_JOURNAL(p_s_sb)->j_trans_id) ;
3540 set_commit_trans_len(commit, SB_JOURNAL(p_s_sb)->j_len);
3542 /* special check in case all buffers in the journal were marked for not logging */
3543 if (SB_JOURNAL(p_s_sb)->j_len == 0) {
3547 /* we're about to dirty all the log blocks, mark the description block
3548 * dirty now too. Don't mark the commit block dirty until all the
3549 * others are on disk
3551 mark_buffer_dirty(d_bh);
3553 /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
3554 cur_write_start = SB_JOURNAL(p_s_sb)->j_start ;
3555 cn = SB_JOURNAL(p_s_sb)->j_first ;
3556 jindex = 1 ; /* start at one so we don't get the desc again */
3558 clear_bit(BH_JNew, &(cn->bh->b_state)) ;
3559 /* copy all the real blocks into log area. dirty log blocks */
3560 if (test_bit(BH_JDirty, &cn->bh->b_state)) {
3561 struct buffer_head *tmp_bh ;
3564 tmp_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3565 ((cur_write_start + jindex) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ;
3566 set_buffer_uptodate(tmp_bh);
3567 page = cn->bh->b_page;
3569 memcpy(tmp_bh->b_data, addr + offset_in_page(cn->bh->b_data),
3572 mark_buffer_dirty(tmp_bh);
3574 set_bit(BH_JDirty_wait, &(cn->bh->b_state)) ;
3575 clear_bit(BH_JDirty, &(cn->bh->b_state)) ;
3577 /* JDirty cleared sometime during transaction. don't log this one */
3578 reiserfs_warning(p_s_sb, "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!") ;
3582 free_cnode(p_s_sb, cn) ;
3587 /* we are done with both the c_bh and d_bh, but
3588 ** c_bh must be written after all other commit blocks,
3589 ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
3592 SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
3594 /* now it is safe to insert this transaction on the main list */
3595 list_add_tail(&jl->j_list, &SB_JOURNAL(p_s_sb)->j_journal_list);
3596 list_add_tail(&jl->j_working_list, &SB_JOURNAL(p_s_sb)->j_working_list);
3597 SB_JOURNAL(p_s_sb)->j_num_work_lists++;
3599 /* reset journal values for the next transaction */
3600 old_start = SB_JOURNAL(p_s_sb)->j_start ;
3601 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);
3602 atomic_set(&(SB_JOURNAL(p_s_sb)->j_wcount), 0) ;
3603 SB_JOURNAL(p_s_sb)->j_bcount = 0 ;
3604 SB_JOURNAL(p_s_sb)->j_last = NULL ;
3605 SB_JOURNAL(p_s_sb)->j_first = NULL ;
3606 SB_JOURNAL(p_s_sb)->j_len = 0 ;
3607 SB_JOURNAL(p_s_sb)->j_trans_start_time = 0 ;
3608 SB_JOURNAL(p_s_sb)->j_trans_id++ ;
3609 SB_JOURNAL(p_s_sb)->j_current_jl->j_trans_id = SB_JOURNAL(p_s_sb)->j_trans_id;
3610 SB_JOURNAL(p_s_sb)->j_must_wait = 0 ;
3611 SB_JOURNAL(p_s_sb)->j_len_alloc = 0 ;
3612 SB_JOURNAL(p_s_sb)->j_next_full_flush = 0 ;
3613 SB_JOURNAL(p_s_sb)->j_next_async_flush = 0 ;
3614 init_journal_hash(p_s_sb) ;
3616 // make sure reiserfs_add_jh sees the new current_jl before we
3617 // write out the tails
3620 /* tail conversion targets have to hit the disk before we end the
3621 * transaction. Otherwise a later transaction might repack the tail
3622 * before this transaction commits, leaving the data block unflushed and
3623 * clean, if we crash before the later transaction commits, the data block
3626 if (!list_empty(&jl->j_tail_bh_list)) {
3628 write_ordered_buffers(&SB_JOURNAL(p_s_sb)->j_dirty_buffers_lock,
3629 SB_JOURNAL(p_s_sb), jl, &jl->j_tail_bh_list);
3632 if (!list_empty(&jl->j_tail_bh_list))
3634 up(&jl->j_commit_lock);
3636 /* honor the flush wishes from the caller, simple commits can
3637 ** be done outside the journal lock, they are done below
3639 ** if we don't flush the commit list right now, we put it into
3640 ** the work queue so the people waiting on the async progress work
3641 ** queue don't wait for this proc to flush journal lists and such.
3644 flush_commit_list(p_s_sb, jl, 1) ;
3645 flush_journal_list(p_s_sb, jl, 1) ;
3646 } else if (!(jl->j_state & LIST_COMMIT_PENDING))
3647 queue_delayed_work(commit_wq, &SB_JOURNAL(p_s_sb)->j_work, HZ/10);
3650 /* if the next transaction has any chance of wrapping, flush
3651 ** transactions that might get overwritten. If any journal lists are very
3652 ** old flush them as well.
3655 list_for_each_safe(entry, safe, &SB_JOURNAL(p_s_sb)->j_journal_list) {
3656 temp_jl = JOURNAL_LIST_ENTRY(entry);
3657 if (SB_JOURNAL(p_s_sb)->j_start <= temp_jl->j_start) {
3658 if ((SB_JOURNAL(p_s_sb)->j_start + SB_JOURNAL_TRANS_MAX(p_s_sb) + 1) >=
3661 flush_used_journal_lists(p_s_sb, temp_jl);
3663 } else if ((SB_JOURNAL(p_s_sb)->j_start +
3664 SB_JOURNAL_TRANS_MAX(p_s_sb) + 1) <
3665 SB_ONDISK_JOURNAL_SIZE(p_s_sb))
3667 /* if we don't cross into the next transaction and we don't
3668 * wrap, there is no way we can overlap any later transactions
3673 } else if ((SB_JOURNAL(p_s_sb)->j_start +
3674 SB_JOURNAL_TRANS_MAX(p_s_sb) + 1) >
3675 SB_ONDISK_JOURNAL_SIZE(p_s_sb))
3677 if (((SB_JOURNAL(p_s_sb)->j_start + SB_JOURNAL_TRANS_MAX(p_s_sb) + 1) %
3678 SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >= temp_jl->j_start)
3680 flush_used_journal_lists(p_s_sb, temp_jl);
3683 /* we don't overlap anything from out start to the end of the
3684 * log, and our wrapped portion doesn't overlap anything at
3685 * the start of the log. We can break
3691 flush_old_journal_lists(p_s_sb);
3693 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) ;
3695 if (!(SB_JOURNAL(p_s_sb)->j_current_jl->j_list_bitmap)) {
3696 reiserfs_panic(p_s_sb, "journal-1996: do_journal_end, could not get a list bitmap\n") ;
3699 atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 0) ;
3700 unlock_journal(p_s_sb) ;
3701 /* wake up any body waiting to join. */
3702 clear_bit(WRITERS_QUEUED, &SB_JOURNAL(p_s_sb)->j_state);
3703 wake_up(&(SB_JOURNAL(p_s_sb)->j_join_wait)) ;
3705 if (!flush && wait_on_commit &&
3706 journal_list_still_alive(p_s_sb, commit_trans_id)) {
3707 flush_commit_list(p_s_sb, jl, 1) ;
3710 reiserfs_check_lock_depth(p_s_sb, "journal end2");