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/time.h>
38 #include <linux/semaphore.h>
39 #include <linux/vmalloc.h>
40 #include <linux/reiserfs_fs.h>
41 #include <linux/kernel.h>
42 #include <linux/errno.h>
43 #include <linux/fcntl.h>
44 #include <linux/stat.h>
45 #include <linux/string.h>
46 #include <linux/smp_lock.h>
47 #include <linux/buffer_head.h>
48 #include <linux/workqueue.h>
49 #include <linux/writeback.h>
50 #include <linux/blkdev.h>
51 #include <linux/backing-dev.h>
52 #include <linux/uaccess.h>
54 #include <asm/system.h>
56 /* gets a struct reiserfs_journal_list * from a list head */
57 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
59 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
62 /* the number of mounted filesystems. This is used to decide when to
63 ** start and kill the commit workqueue
65 static int reiserfs_mounted_fs_count;
67 static struct workqueue_struct *commit_wq;
69 #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
71 #define BUFNR 64 /*read ahead */
73 /* cnode stat bits. Move these into reiserfs_fs.h */
75 #define BLOCK_FREED 2 /* this block was freed, and can't be written. */
76 #define BLOCK_FREED_HOLDER 3 /* this block was freed during this transaction, and can't be written */
78 #define BLOCK_NEEDS_FLUSH 4 /* used in flush_journal_list */
79 #define BLOCK_DIRTIED 5
81 /* journal list state bits */
82 #define LIST_TOUCHED 1
84 #define LIST_COMMIT_PENDING 4 /* someone will commit this list */
86 /* flags for do_journal_end */
87 #define FLUSH_ALL 1 /* flush commit and real blocks */
88 #define COMMIT_NOW 2 /* end and commit this transaction */
89 #define WAIT 4 /* wait for the log blocks to hit the disk */
91 static int do_journal_end(struct reiserfs_transaction_handle *,
92 struct super_block *, unsigned long nblocks,
94 static int flush_journal_list(struct super_block *s,
95 struct reiserfs_journal_list *jl, int flushall);
96 static int flush_commit_list(struct super_block *s,
97 struct reiserfs_journal_list *jl, int flushall);
98 static int can_dirty(struct reiserfs_journal_cnode *cn);
99 static int journal_join(struct reiserfs_transaction_handle *th,
100 struct super_block *sb, unsigned long nblocks);
101 static int release_journal_dev(struct super_block *super,
102 struct reiserfs_journal *journal);
103 static int dirty_one_transaction(struct super_block *s,
104 struct reiserfs_journal_list *jl);
105 static void flush_async_commits(struct work_struct *work);
106 static void queue_log_writer(struct super_block *s);
108 /* values for join in do_journal_begin_r */
110 JBEGIN_REG = 0, /* regular journal begin */
111 JBEGIN_JOIN = 1, /* join the running transaction if at all possible */
112 JBEGIN_ABORT = 2, /* called from cleanup code, ignores aborted flag */
115 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
116 struct super_block *sb,
117 unsigned long nblocks, int join);
119 static void init_journal_hash(struct super_block *sb)
121 struct reiserfs_journal *journal = SB_JOURNAL(sb);
122 memset(journal->j_hash_table, 0,
123 JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
127 ** clears BH_Dirty and sticks the buffer on the clean list. Called because I can't allow refile_buffer to
128 ** make schedule happen after I've freed a block. Look at remove_from_transaction and journal_mark_freed for
131 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
134 clear_buffer_dirty(bh);
135 clear_buffer_journal_test(bh);
140 static void disable_barrier(struct super_block *s)
142 REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH);
143 printk("reiserfs: disabling flush barriers on %s\n",
144 reiserfs_bdevname(s));
147 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
150 struct reiserfs_bitmap_node *bn;
153 bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
157 bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
163 INIT_LIST_HEAD(&bn->list);
167 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
169 struct reiserfs_journal *journal = SB_JOURNAL(sb);
170 struct reiserfs_bitmap_node *bn = NULL;
171 struct list_head *entry = journal->j_bitmap_nodes.next;
173 journal->j_used_bitmap_nodes++;
176 if (entry != &journal->j_bitmap_nodes) {
177 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
179 memset(bn->data, 0, sb->s_blocksize);
180 journal->j_free_bitmap_nodes--;
183 bn = allocate_bitmap_node(sb);
190 static inline void free_bitmap_node(struct super_block *sb,
191 struct reiserfs_bitmap_node *bn)
193 struct reiserfs_journal *journal = SB_JOURNAL(sb);
194 journal->j_used_bitmap_nodes--;
195 if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
199 list_add(&bn->list, &journal->j_bitmap_nodes);
200 journal->j_free_bitmap_nodes++;
204 static void allocate_bitmap_nodes(struct super_block *sb)
207 struct reiserfs_journal *journal = SB_JOURNAL(sb);
208 struct reiserfs_bitmap_node *bn = NULL;
209 for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
210 bn = allocate_bitmap_node(sb);
212 list_add(&bn->list, &journal->j_bitmap_nodes);
213 journal->j_free_bitmap_nodes++;
215 break; /* this is ok, we'll try again when more are needed */
220 static int set_bit_in_list_bitmap(struct super_block *sb,
222 struct reiserfs_list_bitmap *jb)
224 unsigned int bmap_nr = block / (sb->s_blocksize << 3);
225 unsigned int bit_nr = block % (sb->s_blocksize << 3);
227 if (!jb->bitmaps[bmap_nr]) {
228 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
230 set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
234 static void cleanup_bitmap_list(struct super_block *sb,
235 struct reiserfs_list_bitmap *jb)
238 if (jb->bitmaps == NULL)
241 for (i = 0; i < reiserfs_bmap_count(sb); i++) {
242 if (jb->bitmaps[i]) {
243 free_bitmap_node(sb, jb->bitmaps[i]);
244 jb->bitmaps[i] = NULL;
250 ** only call this on FS unmount.
252 static int free_list_bitmaps(struct super_block *sb,
253 struct reiserfs_list_bitmap *jb_array)
256 struct reiserfs_list_bitmap *jb;
257 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
259 jb->journal_list = NULL;
260 cleanup_bitmap_list(sb, jb);
267 static int free_bitmap_nodes(struct super_block *sb)
269 struct reiserfs_journal *journal = SB_JOURNAL(sb);
270 struct list_head *next = journal->j_bitmap_nodes.next;
271 struct reiserfs_bitmap_node *bn;
273 while (next != &journal->j_bitmap_nodes) {
274 bn = list_entry(next, struct reiserfs_bitmap_node, list);
278 next = journal->j_bitmap_nodes.next;
279 journal->j_free_bitmap_nodes--;
286 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
287 ** jb_array is the array to be filled in.
289 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
290 struct reiserfs_list_bitmap *jb_array,
291 unsigned int bmap_nr)
295 struct reiserfs_list_bitmap *jb;
296 int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
298 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
300 jb->journal_list = NULL;
301 jb->bitmaps = vmalloc(mem);
303 reiserfs_warning(sb, "clm-2000", "unable to "
304 "allocate bitmaps for journal lists");
308 memset(jb->bitmaps, 0, mem);
311 free_list_bitmaps(sb, jb_array);
318 ** find an available list bitmap. If you can't find one, flush a commit list
321 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
322 struct reiserfs_journal_list
326 struct reiserfs_journal *journal = SB_JOURNAL(sb);
327 struct reiserfs_list_bitmap *jb = NULL;
329 for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
330 i = journal->j_list_bitmap_index;
331 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
332 jb = journal->j_list_bitmap + i;
333 if (journal->j_list_bitmap[i].journal_list) {
334 flush_commit_list(sb,
335 journal->j_list_bitmap[i].
337 if (!journal->j_list_bitmap[i].journal_list) {
344 if (jb->journal_list) { /* double check to make sure if flushed correctly */
347 jb->journal_list = jl;
352 ** allocates a new chunk of X nodes, and links them all together as a list.
353 ** Uses the cnode->next and cnode->prev pointers
354 ** returns NULL on failure
356 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
358 struct reiserfs_journal_cnode *head;
360 if (num_cnodes <= 0) {
363 head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
367 memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode));
369 head[0].next = head + 1;
370 for (i = 1; i < num_cnodes; i++) {
371 head[i].prev = head + (i - 1);
372 head[i].next = head + (i + 1); /* if last one, overwrite it after the if */
374 head[num_cnodes - 1].next = NULL;
379 ** pulls a cnode off the free list, or returns NULL on failure
381 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
383 struct reiserfs_journal_cnode *cn;
384 struct reiserfs_journal *journal = SB_JOURNAL(sb);
386 reiserfs_check_lock_depth(sb, "get_cnode");
388 if (journal->j_cnode_free <= 0) {
391 journal->j_cnode_used++;
392 journal->j_cnode_free--;
393 cn = journal->j_cnode_free_list;
398 cn->next->prev = NULL;
400 journal->j_cnode_free_list = cn->next;
401 memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
406 ** returns a cnode to the free list
408 static void free_cnode(struct super_block *sb,
409 struct reiserfs_journal_cnode *cn)
411 struct reiserfs_journal *journal = SB_JOURNAL(sb);
413 reiserfs_check_lock_depth(sb, "free_cnode");
415 journal->j_cnode_used--;
416 journal->j_cnode_free++;
417 /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
418 cn->next = journal->j_cnode_free_list;
419 if (journal->j_cnode_free_list) {
420 journal->j_cnode_free_list->prev = cn;
422 cn->prev = NULL; /* not needed with the memset, but I might kill the memset, and forget to do this */
423 journal->j_cnode_free_list = cn;
426 static void clear_prepared_bits(struct buffer_head *bh)
428 clear_buffer_journal_prepared(bh);
429 clear_buffer_journal_restore_dirty(bh);
432 /* return a cnode with same dev, block number and size in table, or null if not found */
433 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
437 reiserfs_journal_cnode
441 struct reiserfs_journal_cnode *cn;
442 cn = journal_hash(table, sb, bl);
444 if (cn->blocknr == bl && cn->sb == sb)
448 return (struct reiserfs_journal_cnode *)0;
452 ** this actually means 'can this block be reallocated yet?'. If you set search_all, a block can only be allocated
453 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
454 ** being overwritten by a replay after crashing.
456 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction. Since deleting
457 ** a block removes it from the current transaction, this case should never happen. If you don't set search_all, make
458 ** sure you never write the block without logging it.
460 ** next_zero_bit is a suggestion about the next block to try for find_forward.
461 ** when bl is rejected because it is set in a journal list bitmap, we search
462 ** for the next zero bit in the bitmap that rejected bl. Then, we return that
463 ** through next_zero_bit for find_forward to try.
465 ** Just because we return something in next_zero_bit does not mean we won't
466 ** reject it on the next call to reiserfs_in_journal
469 int reiserfs_in_journal(struct super_block *sb,
470 unsigned int bmap_nr, int bit_nr, int search_all,
471 b_blocknr_t * next_zero_bit)
473 struct reiserfs_journal *journal = SB_JOURNAL(sb);
474 struct reiserfs_journal_cnode *cn;
475 struct reiserfs_list_bitmap *jb;
479 *next_zero_bit = 0; /* always start this at zero. */
481 PROC_INFO_INC(sb, journal.in_journal);
482 /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
483 ** if we crash before the transaction that freed it commits, this transaction won't
484 ** have committed either, and the block will never be written
487 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
488 PROC_INFO_INC(sb, journal.in_journal_bitmap);
489 jb = journal->j_list_bitmap + i;
490 if (jb->journal_list && jb->bitmaps[bmap_nr] &&
492 (unsigned long *)jb->bitmaps[bmap_nr]->
495 find_next_zero_bit((unsigned long *)
496 (jb->bitmaps[bmap_nr]->
498 sb->s_blocksize << 3,
505 bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
506 /* is it in any old transactions? */
509 get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
513 /* is it in the current transaction. This should never happen */
514 if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
519 PROC_INFO_INC(sb, journal.in_journal_reusable);
524 /* insert cn into table
526 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
527 struct reiserfs_journal_cnode *cn)
529 struct reiserfs_journal_cnode *cn_orig;
531 cn_orig = journal_hash(table, cn->sb, cn->blocknr);
537 journal_hash(table, cn->sb, cn->blocknr) = cn;
541 * Several mutexes depend on the write lock.
542 * However sometimes we want to relax the write lock while we hold
543 * these mutexes, according to the release/reacquire on schedule()
544 * properties of the Bkl that were used.
545 * Reiserfs performances and locking were based on this scheme.
546 * Now that the write lock is a mutex and not the bkl anymore, doing so
547 * may result in a deadlock:
549 * A acquire write_lock
550 * A acquire j_commit_mutex
551 * A release write_lock and wait for something
552 * B acquire write_lock
553 * B can't acquire j_commit_mutex and sleep
554 * A can't acquire write lock anymore
557 * What we do here is avoiding such deadlock by playing the same game
558 * than the Bkl: if we can't acquire a mutex that depends on the write lock,
559 * we release the write lock, wait a bit and then retry.
561 * The mutexes concerned by this hack are:
562 * - The commit mutex of a journal list
566 static inline void reiserfs_mutex_lock_safe(struct mutex *m,
567 struct super_block *s)
569 reiserfs_write_unlock(s);
571 reiserfs_write_lock(s);
574 /* lock the current transaction */
575 static inline void lock_journal(struct super_block *sb)
577 PROC_INFO_INC(sb, journal.lock_journal);
579 reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
582 /* unlock the current transaction */
583 static inline void unlock_journal(struct super_block *sb)
585 mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
588 static inline void get_journal_list(struct reiserfs_journal_list *jl)
593 static inline void put_journal_list(struct super_block *s,
594 struct reiserfs_journal_list *jl)
596 if (jl->j_refcount < 1) {
597 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
598 jl->j_trans_id, jl->j_refcount);
600 if (--jl->j_refcount == 0)
605 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
606 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
609 static void cleanup_freed_for_journal_list(struct super_block *sb,
610 struct reiserfs_journal_list *jl)
613 struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
615 cleanup_bitmap_list(sb, jb);
617 jl->j_list_bitmap->journal_list = NULL;
618 jl->j_list_bitmap = NULL;
621 static int journal_list_still_alive(struct super_block *s,
622 unsigned int trans_id)
624 struct reiserfs_journal *journal = SB_JOURNAL(s);
625 struct list_head *entry = &journal->j_journal_list;
626 struct reiserfs_journal_list *jl;
628 if (!list_empty(entry)) {
629 jl = JOURNAL_LIST_ENTRY(entry->next);
630 if (jl->j_trans_id <= trans_id) {
638 * If page->mapping was null, we failed to truncate this page for
639 * some reason. Most likely because it was truncated after being
640 * logged via data=journal.
642 * This does a check to see if the buffer belongs to one of these
643 * lost pages before doing the final put_bh. If page->mapping was
644 * null, it tries to free buffers on the page, which should make the
645 * final page_cache_release drop the page from the lru.
647 static void release_buffer_page(struct buffer_head *bh)
649 struct page *page = bh->b_page;
650 if (!page->mapping && trylock_page(page)) {
651 page_cache_get(page);
654 try_to_free_buffers(page);
656 page_cache_release(page);
662 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
664 char b[BDEVNAME_SIZE];
666 if (buffer_journaled(bh)) {
667 reiserfs_warning(NULL, "clm-2084",
668 "pinned buffer %lu:%s sent to disk",
669 bh->b_blocknr, bdevname(bh->b_bdev, b));
672 set_buffer_uptodate(bh);
674 clear_buffer_uptodate(bh);
677 release_buffer_page(bh);
680 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
683 set_buffer_uptodate(bh);
685 clear_buffer_uptodate(bh);
690 static void submit_logged_buffer(struct buffer_head *bh)
693 bh->b_end_io = reiserfs_end_buffer_io_sync;
694 clear_buffer_journal_new(bh);
695 clear_buffer_dirty(bh);
696 if (!test_clear_buffer_journal_test(bh))
698 if (!buffer_uptodate(bh))
700 submit_bh(WRITE, bh);
703 static void submit_ordered_buffer(struct buffer_head *bh)
706 bh->b_end_io = reiserfs_end_ordered_io;
707 clear_buffer_dirty(bh);
708 if (!buffer_uptodate(bh))
710 submit_bh(WRITE, bh);
713 static int submit_barrier_buffer(struct buffer_head *bh)
716 bh->b_end_io = reiserfs_end_ordered_io;
717 clear_buffer_dirty(bh);
718 if (!buffer_uptodate(bh))
720 return submit_bh(WRITE_BARRIER, bh);
723 static void check_barrier_completion(struct super_block *s,
724 struct buffer_head *bh)
726 if (buffer_eopnotsupp(bh)) {
727 clear_buffer_eopnotsupp(bh);
729 set_buffer_uptodate(bh);
730 set_buffer_dirty(bh);
731 reiserfs_write_unlock(s);
732 sync_dirty_buffer(bh);
733 reiserfs_write_lock(s);
737 #define CHUNK_SIZE 32
738 struct buffer_chunk {
739 struct buffer_head *bh[CHUNK_SIZE];
743 static void write_chunk(struct buffer_chunk *chunk)
747 for (i = 0; i < chunk->nr; i++) {
748 submit_logged_buffer(chunk->bh[i]);
754 static void write_ordered_chunk(struct buffer_chunk *chunk)
758 for (i = 0; i < chunk->nr; i++) {
759 submit_ordered_buffer(chunk->bh[i]);
765 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
766 spinlock_t * lock, void (fn) (struct buffer_chunk *))
769 BUG_ON(chunk->nr >= CHUNK_SIZE);
770 chunk->bh[chunk->nr++] = bh;
771 if (chunk->nr >= CHUNK_SIZE) {
782 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
783 static struct reiserfs_jh *alloc_jh(void)
785 struct reiserfs_jh *jh;
787 jh = kmalloc(sizeof(*jh), GFP_NOFS);
789 atomic_inc(&nr_reiserfs_jh);
797 * we want to free the jh when the buffer has been written
800 void reiserfs_free_jh(struct buffer_head *bh)
802 struct reiserfs_jh *jh;
806 bh->b_private = NULL;
808 list_del_init(&jh->list);
810 if (atomic_read(&nr_reiserfs_jh) <= 0)
812 atomic_dec(&nr_reiserfs_jh);
817 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
820 struct reiserfs_jh *jh;
823 spin_lock(&j->j_dirty_buffers_lock);
824 if (!bh->b_private) {
825 spin_unlock(&j->j_dirty_buffers_lock);
829 list_del_init(&jh->list);
834 spin_lock(&j->j_dirty_buffers_lock);
835 /* buffer must be locked for __add_jh, should be able to have
836 * two adds at the same time
838 BUG_ON(bh->b_private);
842 jh->jl = j->j_current_jl;
844 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
846 list_add_tail(&jh->list, &jh->jl->j_bh_list);
848 spin_unlock(&j->j_dirty_buffers_lock);
852 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
854 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
856 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
858 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
861 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
862 static int write_ordered_buffers(spinlock_t * lock,
863 struct reiserfs_journal *j,
864 struct reiserfs_journal_list *jl,
865 struct list_head *list)
867 struct buffer_head *bh;
868 struct reiserfs_jh *jh;
869 int ret = j->j_errno;
870 struct buffer_chunk chunk;
871 struct list_head tmp;
872 INIT_LIST_HEAD(&tmp);
876 while (!list_empty(list)) {
877 jh = JH_ENTRY(list->next);
880 if (!trylock_buffer(bh)) {
881 if (!buffer_dirty(bh)) {
882 list_move(&jh->list, &tmp);
887 write_ordered_chunk(&chunk);
893 /* in theory, dirty non-uptodate buffers should never get here,
894 * but the upper layer io error paths still have a few quirks.
895 * Handle them here as gracefully as we can
897 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
898 clear_buffer_dirty(bh);
901 if (buffer_dirty(bh)) {
902 list_move(&jh->list, &tmp);
903 add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
905 reiserfs_free_jh(bh);
910 cond_resched_lock(lock);
914 write_ordered_chunk(&chunk);
917 while (!list_empty(&tmp)) {
918 jh = JH_ENTRY(tmp.prev);
921 reiserfs_free_jh(bh);
923 if (buffer_locked(bh)) {
928 if (!buffer_uptodate(bh)) {
931 /* ugly interaction with invalidatepage here.
932 * reiserfs_invalidate_page will pin any buffer that has a valid
933 * journal head from an older transaction. If someone else sets
934 * our buffer dirty after we write it in the first loop, and
935 * then someone truncates the page away, nobody will ever write
936 * the buffer. We're safe if we write the page one last time
937 * after freeing the journal header.
939 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
941 ll_rw_block(WRITE, 1, &bh);
945 cond_resched_lock(lock);
951 static int flush_older_commits(struct super_block *s,
952 struct reiserfs_journal_list *jl)
954 struct reiserfs_journal *journal = SB_JOURNAL(s);
955 struct reiserfs_journal_list *other_jl;
956 struct reiserfs_journal_list *first_jl;
957 struct list_head *entry;
958 unsigned int trans_id = jl->j_trans_id;
959 unsigned int other_trans_id;
960 unsigned int first_trans_id;
964 * first we walk backwards to find the oldest uncommitted transation
967 entry = jl->j_list.prev;
969 other_jl = JOURNAL_LIST_ENTRY(entry);
970 if (entry == &journal->j_journal_list ||
971 atomic_read(&other_jl->j_older_commits_done))
975 entry = other_jl->j_list.prev;
978 /* if we didn't find any older uncommitted transactions, return now */
979 if (first_jl == jl) {
983 first_trans_id = first_jl->j_trans_id;
985 entry = &first_jl->j_list;
987 other_jl = JOURNAL_LIST_ENTRY(entry);
988 other_trans_id = other_jl->j_trans_id;
990 if (other_trans_id < trans_id) {
991 if (atomic_read(&other_jl->j_commit_left) != 0) {
992 flush_commit_list(s, other_jl, 0);
994 /* list we were called with is gone, return */
995 if (!journal_list_still_alive(s, trans_id))
998 /* the one we just flushed is gone, this means all
999 * older lists are also gone, so first_jl is no longer
1000 * valid either. Go back to the beginning.
1002 if (!journal_list_still_alive
1003 (s, other_trans_id)) {
1007 entry = entry->next;
1008 if (entry == &journal->j_journal_list)
1017 static int reiserfs_async_progress_wait(struct super_block *s)
1020 struct reiserfs_journal *j = SB_JOURNAL(s);
1022 if (atomic_read(&j->j_async_throttle)) {
1023 reiserfs_write_unlock(s);
1024 congestion_wait(BLK_RW_ASYNC, HZ / 10);
1025 reiserfs_write_lock(s);
1032 ** if this journal list still has commit blocks unflushed, send them to disk.
1034 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
1035 ** Before the commit block can by written, every other log block must be safely on disk
1038 static int flush_commit_list(struct super_block *s,
1039 struct reiserfs_journal_list *jl, int flushall)
1043 struct buffer_head *tbh = NULL;
1044 unsigned int trans_id = jl->j_trans_id;
1045 struct reiserfs_journal *journal = SB_JOURNAL(s);
1050 reiserfs_check_lock_depth(s, "flush_commit_list");
1052 if (atomic_read(&jl->j_older_commits_done)) {
1058 /* before we can put our commit blocks on disk, we have to make sure everyone older than
1059 ** us is on disk too
1061 BUG_ON(jl->j_len <= 0);
1062 BUG_ON(trans_id == journal->j_trans_id);
1064 get_journal_list(jl);
1066 if (flush_older_commits(s, jl) == 1) {
1067 /* list disappeared during flush_older_commits. return */
1072 /* make sure nobody is trying to flush this one at the same time */
1073 reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1075 if (!journal_list_still_alive(s, trans_id)) {
1076 mutex_unlock(&jl->j_commit_mutex);
1079 BUG_ON(jl->j_trans_id == 0);
1081 /* this commit is done, exit */
1082 if (atomic_read(&(jl->j_commit_left)) <= 0) {
1084 atomic_set(&(jl->j_older_commits_done), 1);
1086 mutex_unlock(&jl->j_commit_mutex);
1090 if (!list_empty(&jl->j_bh_list)) {
1094 * We might sleep in numerous places inside
1095 * write_ordered_buffers. Relax the write lock.
1097 reiserfs_write_unlock(s);
1098 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1099 journal, jl, &jl->j_bh_list);
1100 if (ret < 0 && retval == 0)
1102 reiserfs_write_lock(s);
1104 BUG_ON(!list_empty(&jl->j_bh_list));
1106 * for the description block and all the log blocks, submit any buffers
1107 * that haven't already reached the disk. Try to write at least 256
1108 * log blocks. later on, we will only wait on blocks that correspond
1109 * to this transaction, but while we're unplugging we might as well
1110 * get a chunk of data on there.
1112 atomic_inc(&journal->j_async_throttle);
1113 write_len = jl->j_len + 1;
1114 if (write_len < 256)
1116 for (i = 0 ; i < write_len ; i++) {
1117 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1118 SB_ONDISK_JOURNAL_SIZE(s);
1119 tbh = journal_find_get_block(s, bn);
1121 if (buffer_dirty(tbh)) {
1122 reiserfs_write_unlock(s);
1123 ll_rw_block(WRITE, 1, &tbh);
1124 reiserfs_write_lock(s);
1129 atomic_dec(&journal->j_async_throttle);
1131 /* We're skipping the commit if there's an error */
1132 if (retval || reiserfs_is_journal_aborted(journal))
1135 /* wait on everything written so far before writing the commit
1136 * if we are in barrier mode, send the commit down now
1138 barrier = reiserfs_barrier_flush(s);
1141 lock_buffer(jl->j_commit_bh);
1142 ret = submit_barrier_buffer(jl->j_commit_bh);
1143 if (ret == -EOPNOTSUPP) {
1144 set_buffer_uptodate(jl->j_commit_bh);
1149 for (i = 0; i < (jl->j_len + 1); i++) {
1150 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1151 (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1152 tbh = journal_find_get_block(s, bn);
1154 reiserfs_write_unlock(s);
1155 wait_on_buffer(tbh);
1156 reiserfs_write_lock(s);
1157 // since we're using ll_rw_blk above, it might have skipped over
1158 // a locked buffer. Double check here
1160 /* redundant, sync_dirty_buffer() checks */
1161 if (buffer_dirty(tbh)) {
1162 reiserfs_write_unlock(s);
1163 sync_dirty_buffer(tbh);
1164 reiserfs_write_lock(s);
1166 if (unlikely(!buffer_uptodate(tbh))) {
1167 #ifdef CONFIG_REISERFS_CHECK
1168 reiserfs_warning(s, "journal-601",
1169 "buffer write failed");
1173 put_bh(tbh); /* once for journal_find_get_block */
1174 put_bh(tbh); /* once due to original getblk in do_journal_end */
1175 atomic_dec(&(jl->j_commit_left));
1178 BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1181 /* If there was a write error in the journal - we can't commit
1182 * this transaction - it will be invalid and, if successful,
1183 * will just end up propagating the write error out to
1184 * the file system. */
1185 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1186 if (buffer_dirty(jl->j_commit_bh))
1188 mark_buffer_dirty(jl->j_commit_bh) ;
1189 reiserfs_write_unlock(s);
1190 sync_dirty_buffer(jl->j_commit_bh) ;
1191 reiserfs_write_lock(s);
1194 reiserfs_write_unlock(s);
1195 wait_on_buffer(jl->j_commit_bh);
1196 reiserfs_write_lock(s);
1199 check_barrier_completion(s, jl->j_commit_bh);
1201 /* If there was a write error in the journal - we can't commit this
1202 * transaction - it will be invalid and, if successful, will just end
1203 * up propagating the write error out to the filesystem. */
1204 if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1205 #ifdef CONFIG_REISERFS_CHECK
1206 reiserfs_warning(s, "journal-615", "buffer write failed");
1210 bforget(jl->j_commit_bh);
1211 if (journal->j_last_commit_id != 0 &&
1212 (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1213 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1214 journal->j_last_commit_id, jl->j_trans_id);
1216 journal->j_last_commit_id = jl->j_trans_id;
1218 /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */
1219 cleanup_freed_for_journal_list(s, jl);
1221 retval = retval ? retval : journal->j_errno;
1223 /* mark the metadata dirty */
1225 dirty_one_transaction(s, jl);
1226 atomic_dec(&(jl->j_commit_left));
1229 atomic_set(&(jl->j_older_commits_done), 1);
1231 mutex_unlock(&jl->j_commit_mutex);
1233 put_journal_list(s, jl);
1236 reiserfs_abort(s, retval, "Journal write error in %s",
1243 ** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or
1244 ** returns NULL if it can't find anything
1246 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1247 reiserfs_journal_cnode
1250 struct super_block *sb = cn->sb;
1251 b_blocknr_t blocknr = cn->blocknr;
1255 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1263 static int newer_jl_done(struct reiserfs_journal_cnode *cn)
1265 struct super_block *sb = cn->sb;
1266 b_blocknr_t blocknr = cn->blocknr;
1270 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist &&
1271 atomic_read(&cn->jlist->j_commit_left) != 0)
1278 static void remove_journal_hash(struct super_block *,
1279 struct reiserfs_journal_cnode **,
1280 struct reiserfs_journal_list *, unsigned long,
1284 ** once all the real blocks have been flushed, it is safe to remove them from the
1285 ** journal list for this transaction. Aside from freeing the cnode, this also allows the
1286 ** block to be reallocated for data blocks if it had been deleted.
1288 static void remove_all_from_journal_list(struct super_block *sb,
1289 struct reiserfs_journal_list *jl,
1292 struct reiserfs_journal *journal = SB_JOURNAL(sb);
1293 struct reiserfs_journal_cnode *cn, *last;
1294 cn = jl->j_realblock;
1296 /* which is better, to lock once around the whole loop, or
1297 ** to lock for each call to remove_journal_hash?
1300 if (cn->blocknr != 0) {
1302 reiserfs_warning(sb, "reiserfs-2201",
1303 "block %u, bh is %d, state %ld",
1304 cn->blocknr, cn->bh ? 1 : 0,
1308 remove_journal_hash(sb, journal->j_list_hash_table,
1309 jl, cn->blocknr, 1);
1313 free_cnode(sb, last);
1315 jl->j_realblock = NULL;
1319 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1320 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1321 ** releasing blocks in this transaction for reuse as data blocks.
1322 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1325 static int _update_journal_header_block(struct super_block *sb,
1326 unsigned long offset,
1327 unsigned int trans_id)
1329 struct reiserfs_journal_header *jh;
1330 struct reiserfs_journal *journal = SB_JOURNAL(sb);
1332 if (reiserfs_is_journal_aborted(journal))
1335 if (trans_id >= journal->j_last_flush_trans_id) {
1336 if (buffer_locked((journal->j_header_bh))) {
1337 reiserfs_write_unlock(sb);
1338 wait_on_buffer((journal->j_header_bh));
1339 reiserfs_write_lock(sb);
1340 if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1341 #ifdef CONFIG_REISERFS_CHECK
1342 reiserfs_warning(sb, "journal-699",
1343 "buffer write failed");
1348 journal->j_last_flush_trans_id = trans_id;
1349 journal->j_first_unflushed_offset = offset;
1350 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1352 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1353 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1354 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1356 if (reiserfs_barrier_flush(sb)) {
1358 lock_buffer(journal->j_header_bh);
1359 ret = submit_barrier_buffer(journal->j_header_bh);
1360 if (ret == -EOPNOTSUPP) {
1361 set_buffer_uptodate(journal->j_header_bh);
1362 disable_barrier(sb);
1365 reiserfs_write_unlock(sb);
1366 wait_on_buffer(journal->j_header_bh);
1367 reiserfs_write_lock(sb);
1368 check_barrier_completion(sb, journal->j_header_bh);
1371 set_buffer_dirty(journal->j_header_bh);
1372 reiserfs_write_unlock(sb);
1373 sync_dirty_buffer(journal->j_header_bh);
1374 reiserfs_write_lock(sb);
1376 if (!buffer_uptodate(journal->j_header_bh)) {
1377 reiserfs_warning(sb, "journal-837",
1378 "IO error during journal replay");
1385 static int update_journal_header_block(struct super_block *sb,
1386 unsigned long offset,
1387 unsigned int trans_id)
1389 return _update_journal_header_block(sb, offset, trans_id);
1393 ** flush any and all journal lists older than you are
1394 ** can only be called from flush_journal_list
1396 static int flush_older_journal_lists(struct super_block *sb,
1397 struct reiserfs_journal_list *jl)
1399 struct list_head *entry;
1400 struct reiserfs_journal_list *other_jl;
1401 struct reiserfs_journal *journal = SB_JOURNAL(sb);
1402 unsigned int trans_id = jl->j_trans_id;
1404 /* we know we are the only ones flushing things, no extra race
1405 * protection is required.
1408 entry = journal->j_journal_list.next;
1410 if (entry == &journal->j_journal_list)
1412 other_jl = JOURNAL_LIST_ENTRY(entry);
1413 if (other_jl->j_trans_id < trans_id) {
1414 BUG_ON(other_jl->j_refcount <= 0);
1415 /* do not flush all */
1416 flush_journal_list(sb, other_jl, 0);
1418 /* other_jl is now deleted from the list */
1424 static void del_from_work_list(struct super_block *s,
1425 struct reiserfs_journal_list *jl)
1427 struct reiserfs_journal *journal = SB_JOURNAL(s);
1428 if (!list_empty(&jl->j_working_list)) {
1429 list_del_init(&jl->j_working_list);
1430 journal->j_num_work_lists--;
1434 /* flush a journal list, both commit and real blocks
1436 ** always set flushall to 1, unless you are calling from inside
1437 ** flush_journal_list
1439 ** IMPORTANT. This can only be called while there are no journal writers,
1440 ** and the journal is locked. That means it can only be called from
1441 ** do_journal_end, or by journal_release
1443 static int flush_journal_list(struct super_block *s,
1444 struct reiserfs_journal_list *jl, int flushall)
1446 struct reiserfs_journal_list *pjl;
1447 struct reiserfs_journal_cnode *cn, *last;
1451 struct buffer_head *saved_bh;
1452 unsigned long j_len_saved = jl->j_len;
1453 struct reiserfs_journal *journal = SB_JOURNAL(s);
1456 BUG_ON(j_len_saved <= 0);
1458 if (atomic_read(&journal->j_wcount) != 0) {
1459 reiserfs_warning(s, "clm-2048", "called with wcount %d",
1460 atomic_read(&journal->j_wcount));
1462 BUG_ON(jl->j_trans_id == 0);
1464 /* if flushall == 0, the lock is already held */
1466 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1467 } else if (mutex_trylock(&journal->j_flush_mutex)) {
1472 if (j_len_saved > journal->j_trans_max) {
1473 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1474 j_len_saved, jl->j_trans_id);
1480 /* if all the work is already done, get out of here */
1481 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1482 atomic_read(&(jl->j_commit_left)) <= 0) {
1483 goto flush_older_and_return;
1486 /* start by putting the commit list on disk. This will also flush
1487 ** the commit lists of any olders transactions
1489 flush_commit_list(s, jl, 1);
1491 if (!(jl->j_state & LIST_DIRTY)
1492 && !reiserfs_is_journal_aborted(journal))
1495 /* are we done now? */
1496 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1497 atomic_read(&(jl->j_commit_left)) <= 0) {
1498 goto flush_older_and_return;
1501 /* loop through each cnode, see if we need to write it,
1502 ** or wait on a more recent transaction, or just ignore it
1504 if (atomic_read(&(journal->j_wcount)) != 0) {
1505 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1508 cn = jl->j_realblock;
1513 /* blocknr of 0 is no longer in the hash, ignore it */
1514 if (cn->blocknr == 0) {
1518 /* This transaction failed commit. Don't write out to the disk */
1519 if (!(jl->j_state & LIST_DIRTY))
1522 pjl = find_newer_jl_for_cn(cn);
1523 /* the order is important here. We check pjl to make sure we
1524 ** don't clear BH_JDirty_wait if we aren't the one writing this
1527 if (!pjl && cn->bh) {
1530 /* we do this to make sure nobody releases the buffer while
1531 ** we are working with it
1535 if (buffer_journal_dirty(saved_bh)) {
1536 BUG_ON(!can_dirty(cn));
1539 } else if (can_dirty(cn)) {
1540 /* everything with !pjl && jwait should be writable */
1545 /* if someone has this block in a newer transaction, just make
1546 ** sure they are committed, and don't try writing it to disk
1549 if (atomic_read(&pjl->j_commit_left))
1550 flush_commit_list(s, pjl, 1);
1554 /* bh == NULL when the block got to disk on its own, OR,
1555 ** the block got freed in a future transaction
1557 if (saved_bh == NULL) {
1561 /* this should never happen. kupdate_one_transaction has this list
1562 ** locked while it works, so we should never see a buffer here that
1563 ** is not marked JDirty_wait
1565 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1566 reiserfs_warning(s, "journal-813",
1567 "BAD! buffer %llu %cdirty %cjwait, "
1568 "not in a newer tranasction",
1569 (unsigned long long)saved_bh->
1570 b_blocknr, was_dirty ? ' ' : '!',
1571 was_jwait ? ' ' : '!');
1574 /* we inc again because saved_bh gets decremented at free_cnode */
1576 set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1577 lock_buffer(saved_bh);
1578 BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1579 if (buffer_dirty(saved_bh))
1580 submit_logged_buffer(saved_bh);
1582 unlock_buffer(saved_bh);
1585 reiserfs_warning(s, "clm-2082",
1586 "Unable to flush buffer %llu in %s",
1587 (unsigned long long)saved_bh->
1588 b_blocknr, __func__);
1594 /* we incremented this to keep others from taking the buffer head away */
1596 if (atomic_read(&(saved_bh->b_count)) < 0) {
1597 reiserfs_warning(s, "journal-945",
1598 "saved_bh->b_count < 0");
1603 cn = jl->j_realblock;
1605 if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1607 reiserfs_panic(s, "journal-1011",
1611 reiserfs_write_unlock(s);
1612 wait_on_buffer(cn->bh);
1613 reiserfs_write_lock(s);
1616 reiserfs_panic(s, "journal-1012",
1619 if (unlikely(!buffer_uptodate(cn->bh))) {
1620 #ifdef CONFIG_REISERFS_CHECK
1621 reiserfs_warning(s, "journal-949",
1622 "buffer write failed");
1626 /* note, we must clear the JDirty_wait bit after the up to date
1627 ** check, otherwise we race against our flushpage routine
1629 BUG_ON(!test_clear_buffer_journal_dirty
1632 /* drop one ref for us */
1634 /* drop one ref for journal_mark_dirty */
1635 release_buffer_page(cn->bh);
1642 reiserfs_abort(s, -EIO,
1643 "Write error while pushing transaction to disk in %s",
1645 flush_older_and_return:
1647 /* before we can update the journal header block, we _must_ flush all
1648 ** real blocks from all older transactions to disk. This is because
1649 ** once the header block is updated, this transaction will not be
1650 ** replayed after a crash
1653 flush_older_journal_lists(s, jl);
1656 err = journal->j_errno;
1657 /* before we can remove everything from the hash tables for this
1658 ** transaction, we must make sure it can never be replayed
1660 ** since we are only called from do_journal_end, we know for sure there
1661 ** are no allocations going on while we are flushing journal lists. So,
1662 ** we only need to update the journal header block for the last list
1665 if (!err && flushall) {
1667 update_journal_header_block(s,
1668 (jl->j_start + jl->j_len +
1669 2) % SB_ONDISK_JOURNAL_SIZE(s),
1672 reiserfs_abort(s, -EIO,
1673 "Write error while updating journal header in %s",
1676 remove_all_from_journal_list(s, jl, 0);
1677 list_del_init(&jl->j_list);
1678 journal->j_num_lists--;
1679 del_from_work_list(s, jl);
1681 if (journal->j_last_flush_id != 0 &&
1682 (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1683 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1684 journal->j_last_flush_id, jl->j_trans_id);
1686 journal->j_last_flush_id = jl->j_trans_id;
1688 /* not strictly required since we are freeing the list, but it should
1689 * help find code using dead lists later on
1692 atomic_set(&(jl->j_nonzerolen), 0);
1694 jl->j_realblock = NULL;
1695 jl->j_commit_bh = NULL;
1698 put_journal_list(s, jl);
1700 mutex_unlock(&journal->j_flush_mutex);
1705 static int test_transaction(struct super_block *s,
1706 struct reiserfs_journal_list *jl)
1708 struct reiserfs_journal_cnode *cn;
1710 if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0)
1713 cn = jl->j_realblock;
1715 /* if the blocknr == 0, this has been cleared from the hash,
1718 if (cn->blocknr == 0) {
1721 if (cn->bh && !newer_jl_done(cn))
1730 static int write_one_transaction(struct super_block *s,
1731 struct reiserfs_journal_list *jl,
1732 struct buffer_chunk *chunk)
1734 struct reiserfs_journal_cnode *cn;
1737 jl->j_state |= LIST_TOUCHED;
1738 del_from_work_list(s, jl);
1739 if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1743 cn = jl->j_realblock;
1745 /* if the blocknr == 0, this has been cleared from the hash,
1748 if (cn->blocknr == 0) {
1751 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1752 struct buffer_head *tmp_bh;
1753 /* we can race against journal_mark_freed when we try
1754 * to lock_buffer(cn->bh), so we have to inc the buffer
1755 * count, and recheck things after locking
1759 lock_buffer(tmp_bh);
1760 if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1761 if (!buffer_journal_dirty(tmp_bh) ||
1762 buffer_journal_prepared(tmp_bh))
1764 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1767 /* note, cn->bh might be null now */
1768 unlock_buffer(tmp_bh);
1779 /* used by flush_commit_list */
1780 static int dirty_one_transaction(struct super_block *s,
1781 struct reiserfs_journal_list *jl)
1783 struct reiserfs_journal_cnode *cn;
1784 struct reiserfs_journal_list *pjl;
1787 jl->j_state |= LIST_DIRTY;
1788 cn = jl->j_realblock;
1790 /* look for a more recent transaction that logged this
1791 ** buffer. Only the most recent transaction with a buffer in
1792 ** it is allowed to send that buffer to disk
1794 pjl = find_newer_jl_for_cn(cn);
1795 if (!pjl && cn->blocknr && cn->bh
1796 && buffer_journal_dirty(cn->bh)) {
1797 BUG_ON(!can_dirty(cn));
1798 /* if the buffer is prepared, it will either be logged
1799 * or restored. If restored, we need to make sure
1800 * it actually gets marked dirty
1802 clear_buffer_journal_new(cn->bh);
1803 if (buffer_journal_prepared(cn->bh)) {
1804 set_buffer_journal_restore_dirty(cn->bh);
1806 set_buffer_journal_test(cn->bh);
1807 mark_buffer_dirty(cn->bh);
1815 static int kupdate_transactions(struct super_block *s,
1816 struct reiserfs_journal_list *jl,
1817 struct reiserfs_journal_list **next_jl,
1818 unsigned int *next_trans_id,
1819 int num_blocks, int num_trans)
1823 int transactions_flushed = 0;
1824 unsigned int orig_trans_id = jl->j_trans_id;
1825 struct buffer_chunk chunk;
1826 struct list_head *entry;
1827 struct reiserfs_journal *journal = SB_JOURNAL(s);
1830 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1831 if (!journal_list_still_alive(s, orig_trans_id)) {
1835 /* we've got j_flush_mutex held, nobody is going to delete any
1836 * of these lists out from underneath us
1838 while ((num_trans && transactions_flushed < num_trans) ||
1839 (!num_trans && written < num_blocks)) {
1841 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1842 atomic_read(&jl->j_commit_left)
1843 || !(jl->j_state & LIST_DIRTY)) {
1844 del_from_work_list(s, jl);
1847 ret = write_one_transaction(s, jl, &chunk);
1851 transactions_flushed++;
1853 entry = jl->j_list.next;
1856 if (entry == &journal->j_journal_list) {
1859 jl = JOURNAL_LIST_ENTRY(entry);
1861 /* don't bother with older transactions */
1862 if (jl->j_trans_id <= orig_trans_id)
1866 write_chunk(&chunk);
1870 mutex_unlock(&journal->j_flush_mutex);
1874 /* for o_sync and fsync heavy applications, they tend to use
1875 ** all the journa list slots with tiny transactions. These
1876 ** trigger lots and lots of calls to update the header block, which
1877 ** adds seeks and slows things down.
1879 ** This function tries to clear out a large chunk of the journal lists
1880 ** at once, which makes everything faster since only the newest journal
1881 ** list updates the header block
1883 static int flush_used_journal_lists(struct super_block *s,
1884 struct reiserfs_journal_list *jl)
1886 unsigned long len = 0;
1887 unsigned long cur_len;
1891 struct reiserfs_journal_list *tjl;
1892 struct reiserfs_journal_list *flush_jl;
1893 unsigned int trans_id;
1894 struct reiserfs_journal *journal = SB_JOURNAL(s);
1896 flush_jl = tjl = jl;
1898 /* in data logging mode, try harder to flush a lot of blocks */
1899 if (reiserfs_data_log(s))
1901 /* flush for 256 transactions or limit blocks, whichever comes first */
1902 for (i = 0; i < 256 && len < limit; i++) {
1903 if (atomic_read(&tjl->j_commit_left) ||
1904 tjl->j_trans_id < jl->j_trans_id) {
1907 cur_len = atomic_read(&tjl->j_nonzerolen);
1909 tjl->j_state &= ~LIST_TOUCHED;
1913 if (tjl->j_list.next == &journal->j_journal_list)
1915 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1917 /* try to find a group of blocks we can flush across all the
1918 ** transactions, but only bother if we've actually spanned
1919 ** across multiple lists
1921 if (flush_jl != jl) {
1922 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1924 flush_journal_list(s, flush_jl, 1);
1929 ** removes any nodes in table with name block and dev as bh.
1930 ** only touchs the hnext and hprev pointers.
1932 void remove_journal_hash(struct super_block *sb,
1933 struct reiserfs_journal_cnode **table,
1934 struct reiserfs_journal_list *jl,
1935 unsigned long block, int remove_freed)
1937 struct reiserfs_journal_cnode *cur;
1938 struct reiserfs_journal_cnode **head;
1940 head = &(journal_hash(table, sb, block));
1946 if (cur->blocknr == block && cur->sb == sb
1947 && (jl == NULL || jl == cur->jlist)
1948 && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1950 cur->hnext->hprev = cur->hprev;
1953 cur->hprev->hnext = cur->hnext;
1960 if (cur->bh && cur->jlist) /* anybody who clears the cur->bh will also dec the nonzerolen */
1961 atomic_dec(&(cur->jlist->j_nonzerolen));
1969 static void free_journal_ram(struct super_block *sb)
1971 struct reiserfs_journal *journal = SB_JOURNAL(sb);
1972 kfree(journal->j_current_jl);
1973 journal->j_num_lists--;
1975 vfree(journal->j_cnode_free_orig);
1976 free_list_bitmaps(sb, journal->j_list_bitmap);
1977 free_bitmap_nodes(sb); /* must be after free_list_bitmaps */
1978 if (journal->j_header_bh) {
1979 brelse(journal->j_header_bh);
1981 /* j_header_bh is on the journal dev, make sure not to release the journal
1982 * dev until we brelse j_header_bh
1984 release_journal_dev(sb, journal);
1989 ** call on unmount. Only set error to 1 if you haven't made your way out
1990 ** of read_super() yet. Any other caller must keep error at 0.
1992 static int do_journal_release(struct reiserfs_transaction_handle *th,
1993 struct super_block *sb, int error)
1995 struct reiserfs_transaction_handle myth;
1997 struct reiserfs_journal *journal = SB_JOURNAL(sb);
1999 /* we only want to flush out transactions if we were called with error == 0
2001 if (!error && !(sb->s_flags & MS_RDONLY)) {
2002 /* end the current trans */
2003 BUG_ON(!th->t_trans_id);
2004 do_journal_end(th, sb, 10, FLUSH_ALL);
2006 /* make sure something gets logged to force our way into the flush code */
2007 if (!journal_join(&myth, sb, 1)) {
2008 reiserfs_prepare_for_journal(sb,
2009 SB_BUFFER_WITH_SB(sb),
2011 journal_mark_dirty(&myth, sb,
2012 SB_BUFFER_WITH_SB(sb));
2013 do_journal_end(&myth, sb, 1, FLUSH_ALL);
2018 /* this also catches errors during the do_journal_end above */
2019 if (!error && reiserfs_is_journal_aborted(journal)) {
2020 memset(&myth, 0, sizeof(myth));
2021 if (!journal_join_abort(&myth, sb, 1)) {
2022 reiserfs_prepare_for_journal(sb,
2023 SB_BUFFER_WITH_SB(sb),
2025 journal_mark_dirty(&myth, sb,
2026 SB_BUFFER_WITH_SB(sb));
2027 do_journal_end(&myth, sb, 1, FLUSH_ALL);
2031 reiserfs_mounted_fs_count--;
2032 /* wait for all commits to finish */
2033 cancel_delayed_work(&SB_JOURNAL(sb)->j_work);
2036 * We must release the write lock here because
2037 * the workqueue job (flush_async_commit) needs this lock
2039 reiserfs_write_unlock(sb);
2040 flush_workqueue(commit_wq);
2042 if (!reiserfs_mounted_fs_count) {
2043 destroy_workqueue(commit_wq);
2046 reiserfs_write_lock(sb);
2048 free_journal_ram(sb);
2054 ** call on unmount. flush all journal trans, release all alloc'd ram
2056 int journal_release(struct reiserfs_transaction_handle *th,
2057 struct super_block *sb)
2059 return do_journal_release(th, sb, 0);
2063 ** only call from an error condition inside reiserfs_read_super!
2065 int journal_release_error(struct reiserfs_transaction_handle *th,
2066 struct super_block *sb)
2068 return do_journal_release(th, sb, 1);
2071 /* compares description block with commit block. returns 1 if they differ, 0 if they are the same */
2072 static int journal_compare_desc_commit(struct super_block *sb,
2073 struct reiserfs_journal_desc *desc,
2074 struct reiserfs_journal_commit *commit)
2076 if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
2077 get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
2078 get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
2079 get_commit_trans_len(commit) <= 0) {
2085 /* returns 0 if it did not find a description block
2086 ** returns -1 if it found a corrupt commit block
2087 ** returns 1 if both desc and commit were valid
2089 static int journal_transaction_is_valid(struct super_block *sb,
2090 struct buffer_head *d_bh,
2091 unsigned int *oldest_invalid_trans_id,
2092 unsigned long *newest_mount_id)
2094 struct reiserfs_journal_desc *desc;
2095 struct reiserfs_journal_commit *commit;
2096 struct buffer_head *c_bh;
2097 unsigned long offset;
2102 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2103 if (get_desc_trans_len(desc) > 0
2104 && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2105 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2106 && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2107 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2108 "journal-986: transaction "
2109 "is valid returning because trans_id %d is greater than "
2110 "oldest_invalid %lu",
2111 get_desc_trans_id(desc),
2112 *oldest_invalid_trans_id);
2116 && *newest_mount_id > get_desc_mount_id(desc)) {
2117 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2118 "journal-1087: transaction "
2119 "is valid returning because mount_id %d is less than "
2120 "newest_mount_id %lu",
2121 get_desc_mount_id(desc),
2125 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2126 reiserfs_warning(sb, "journal-2018",
2127 "Bad transaction length %d "
2128 "encountered, ignoring transaction",
2129 get_desc_trans_len(desc));
2132 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2134 /* ok, we have a journal description block, lets see if the transaction was valid */
2137 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2138 ((offset + get_desc_trans_len(desc) +
2139 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2142 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2143 if (journal_compare_desc_commit(sb, desc, commit)) {
2144 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2145 "journal_transaction_is_valid, commit offset %ld had bad "
2146 "time %d or length %d",
2148 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2149 get_commit_trans_id(commit),
2150 get_commit_trans_len(commit));
2152 if (oldest_invalid_trans_id) {
2153 *oldest_invalid_trans_id =
2154 get_desc_trans_id(desc);
2155 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2157 "transaction_is_valid setting oldest invalid trans_id "
2159 get_desc_trans_id(desc));
2164 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2165 "journal-1006: found valid "
2166 "transaction start offset %llu, len %d id %d",
2168 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2169 get_desc_trans_len(desc),
2170 get_desc_trans_id(desc));
2177 static void brelse_array(struct buffer_head **heads, int num)
2180 for (i = 0; i < num; i++) {
2186 ** given the start, and values for the oldest acceptable transactions,
2187 ** this either reads in a replays a transaction, or returns because the transaction
2188 ** is invalid, or too old.
2190 static int journal_read_transaction(struct super_block *sb,
2191 unsigned long cur_dblock,
2192 unsigned long oldest_start,
2193 unsigned int oldest_trans_id,
2194 unsigned long newest_mount_id)
2196 struct reiserfs_journal *journal = SB_JOURNAL(sb);
2197 struct reiserfs_journal_desc *desc;
2198 struct reiserfs_journal_commit *commit;
2199 unsigned int trans_id = 0;
2200 struct buffer_head *c_bh;
2201 struct buffer_head *d_bh;
2202 struct buffer_head **log_blocks = NULL;
2203 struct buffer_head **real_blocks = NULL;
2204 unsigned int trans_offset;
2208 d_bh = journal_bread(sb, cur_dblock);
2211 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2212 trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2213 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2214 "journal_read_transaction, offset %llu, len %d mount_id %d",
2215 d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2216 get_desc_trans_len(desc), get_desc_mount_id(desc));
2217 if (get_desc_trans_id(desc) < oldest_trans_id) {
2218 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2219 "journal_read_trans skipping because %lu is too old",
2221 SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2225 if (get_desc_mount_id(desc) != newest_mount_id) {
2226 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2227 "journal_read_trans skipping because %d is != "
2228 "newest_mount_id %lu", get_desc_mount_id(desc),
2233 c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2234 ((trans_offset + get_desc_trans_len(desc) + 1) %
2235 SB_ONDISK_JOURNAL_SIZE(sb)));
2240 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2241 if (journal_compare_desc_commit(sb, desc, commit)) {
2242 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2243 "journal_read_transaction, "
2244 "commit offset %llu had bad time %d or length %d",
2246 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2247 get_commit_trans_id(commit),
2248 get_commit_trans_len(commit));
2253 trans_id = get_desc_trans_id(desc);
2254 /* now we know we've got a good transaction, and it was inside the valid time ranges */
2255 log_blocks = kmalloc(get_desc_trans_len(desc) *
2256 sizeof(struct buffer_head *), GFP_NOFS);
2257 real_blocks = kmalloc(get_desc_trans_len(desc) *
2258 sizeof(struct buffer_head *), GFP_NOFS);
2259 if (!log_blocks || !real_blocks) {
2264 reiserfs_warning(sb, "journal-1169",
2265 "kmalloc failed, unable to mount FS");
2268 /* get all the buffer heads */
2269 trans_half = journal_trans_half(sb->s_blocksize);
2270 for (i = 0; i < get_desc_trans_len(desc); i++) {
2273 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2275 i) % SB_ONDISK_JOURNAL_SIZE(sb));
2276 if (i < trans_half) {
2279 le32_to_cpu(desc->j_realblock[i]));
2283 le32_to_cpu(commit->
2284 j_realblock[i - trans_half]));
2286 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2287 reiserfs_warning(sb, "journal-1207",
2288 "REPLAY FAILURE fsck required! "
2289 "Block to replay is outside of "
2293 /* make sure we don't try to replay onto log or reserved area */
2294 if (is_block_in_log_or_reserved_area
2295 (sb, real_blocks[i]->b_blocknr)) {
2296 reiserfs_warning(sb, "journal-1204",
2297 "REPLAY FAILURE fsck required! "
2298 "Trying to replay onto a log block");
2300 brelse_array(log_blocks, i);
2301 brelse_array(real_blocks, i);
2309 /* read in the log blocks, memcpy to the corresponding real block */
2310 ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2311 for (i = 0; i < get_desc_trans_len(desc); i++) {
2313 reiserfs_write_unlock(sb);
2314 wait_on_buffer(log_blocks[i]);
2315 reiserfs_write_lock(sb);
2317 if (!buffer_uptodate(log_blocks[i])) {
2318 reiserfs_warning(sb, "journal-1212",
2319 "REPLAY FAILURE fsck required! "
2320 "buffer write failed");
2321 brelse_array(log_blocks + i,
2322 get_desc_trans_len(desc) - i);
2323 brelse_array(real_blocks, get_desc_trans_len(desc));
2330 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2331 real_blocks[i]->b_size);
2332 set_buffer_uptodate(real_blocks[i]);
2333 brelse(log_blocks[i]);
2335 /* flush out the real blocks */
2336 for (i = 0; i < get_desc_trans_len(desc); i++) {
2337 set_buffer_dirty(real_blocks[i]);
2338 ll_rw_block(SWRITE, 1, real_blocks + i);
2340 for (i = 0; i < get_desc_trans_len(desc); i++) {
2341 wait_on_buffer(real_blocks[i]);
2342 if (!buffer_uptodate(real_blocks[i])) {
2343 reiserfs_warning(sb, "journal-1226",
2344 "REPLAY FAILURE, fsck required! "
2345 "buffer write failed");
2346 brelse_array(real_blocks + i,
2347 get_desc_trans_len(desc) - i);
2354 brelse(real_blocks[i]);
2357 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2358 ((trans_offset + get_desc_trans_len(desc) +
2359 2) % SB_ONDISK_JOURNAL_SIZE(sb));
2360 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2361 "journal-1095: setting journal " "start to offset %ld",
2362 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2364 /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2365 journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2366 journal->j_last_flush_trans_id = trans_id;
2367 journal->j_trans_id = trans_id + 1;
2368 /* check for trans_id overflow */
2369 if (journal->j_trans_id == 0)
2370 journal->j_trans_id = 10;
2378 /* This function reads blocks starting from block and to max_block of bufsize
2379 size (but no more than BUFNR blocks at a time). This proved to improve
2380 mounting speed on self-rebuilding raid5 arrays at least.
2381 Right now it is only used from journal code. But later we might use it
2383 Note: Do not use journal_getblk/sb_getblk functions here! */
2384 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2385 b_blocknr_t block, int bufsize,
2386 b_blocknr_t max_block)
2388 struct buffer_head *bhlist[BUFNR];
2389 unsigned int blocks = BUFNR;
2390 struct buffer_head *bh;
2393 bh = __getblk(dev, block, bufsize);
2394 if (buffer_uptodate(bh))
2397 if (block + BUFNR > max_block) {
2398 blocks = max_block - block;
2402 for (i = 1; i < blocks; i++) {
2403 bh = __getblk(dev, block + i, bufsize);
2404 if (buffer_uptodate(bh)) {
2410 ll_rw_block(READ, j, bhlist);
2411 for (i = 1; i < j; i++)
2415 if (buffer_uptodate(bh))
2422 ** read and replay the log
2423 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2424 ** transaction. This tests that before finding all the transactions in the log, which makes normal mount times fast.
2426 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2428 ** On exit, it sets things up so the first transaction will work correctly.
2430 static int journal_read(struct super_block *sb)
2432 struct reiserfs_journal *journal = SB_JOURNAL(sb);
2433 struct reiserfs_journal_desc *desc;
2434 unsigned int oldest_trans_id = 0;
2435 unsigned int oldest_invalid_trans_id = 0;
2437 unsigned long oldest_start = 0;
2438 unsigned long cur_dblock = 0;
2439 unsigned long newest_mount_id = 9;
2440 struct buffer_head *d_bh;
2441 struct reiserfs_journal_header *jh;
2442 int valid_journal_header = 0;
2443 int replay_count = 0;
2444 int continue_replay = 1;
2446 char b[BDEVNAME_SIZE];
2448 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2449 reiserfs_info(sb, "checking transaction log (%s)\n",
2450 bdevname(journal->j_dev_bd, b));
2451 start = get_seconds();
2453 /* step 1, read in the journal header block. Check the transaction it says
2454 ** is the first unflushed, and if that transaction is not valid,
2457 journal->j_header_bh = journal_bread(sb,
2458 SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2459 + SB_ONDISK_JOURNAL_SIZE(sb));
2460 if (!journal->j_header_bh) {
2463 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2464 if (le32_to_cpu(jh->j_first_unflushed_offset) <
2465 SB_ONDISK_JOURNAL_SIZE(sb)
2466 && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2468 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2469 le32_to_cpu(jh->j_first_unflushed_offset);
2470 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2471 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2472 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2473 "journal-1153: found in "
2474 "header: first_unflushed_offset %d, last_flushed_trans_id "
2475 "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2476 le32_to_cpu(jh->j_last_flush_trans_id));
2477 valid_journal_header = 1;
2479 /* now, we try to read the first unflushed offset. If it is not valid,
2480 ** there is nothing more we can do, and it makes no sense to read
2481 ** through the whole log.
2485 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2486 le32_to_cpu(jh->j_first_unflushed_offset));
2487 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2489 continue_replay = 0;
2492 goto start_log_replay;
2495 if (continue_replay && bdev_read_only(sb->s_bdev)) {
2496 reiserfs_warning(sb, "clm-2076",
2497 "device is readonly, unable to replay log");
2501 /* ok, there are transactions that need to be replayed. start with the first log block, find
2502 ** all the valid transactions, and pick out the oldest.
2504 while (continue_replay
2506 (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2507 SB_ONDISK_JOURNAL_SIZE(sb))) {
2508 /* Note that it is required for blocksize of primary fs device and journal
2509 device to be the same */
2511 reiserfs_breada(journal->j_dev_bd, cur_dblock,
2513 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2514 SB_ONDISK_JOURNAL_SIZE(sb));
2516 journal_transaction_is_valid(sb, d_bh,
2517 &oldest_invalid_trans_id,
2520 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2521 if (oldest_start == 0) { /* init all oldest_ values */
2522 oldest_trans_id = get_desc_trans_id(desc);
2523 oldest_start = d_bh->b_blocknr;
2524 newest_mount_id = get_desc_mount_id(desc);
2525 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2526 "journal-1179: Setting "
2527 "oldest_start to offset %llu, trans_id %lu",
2529 SB_ONDISK_JOURNAL_1st_BLOCK
2530 (sb), oldest_trans_id);
2531 } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2532 /* one we just read was older */
2533 oldest_trans_id = get_desc_trans_id(desc);
2534 oldest_start = d_bh->b_blocknr;
2535 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2536 "journal-1180: Resetting "
2537 "oldest_start to offset %lu, trans_id %lu",
2539 SB_ONDISK_JOURNAL_1st_BLOCK
2540 (sb), oldest_trans_id);
2542 if (newest_mount_id < get_desc_mount_id(desc)) {
2543 newest_mount_id = get_desc_mount_id(desc);
2544 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2545 "journal-1299: Setting "
2546 "newest_mount_id to %d",
2547 get_desc_mount_id(desc));
2549 cur_dblock += get_desc_trans_len(desc) + 2;
2557 cur_dblock = oldest_start;
2558 if (oldest_trans_id) {
2559 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2560 "journal-1206: Starting replay "
2561 "from offset %llu, trans_id %lu",
2562 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2567 while (continue_replay && oldest_trans_id > 0) {
2569 journal_read_transaction(sb, cur_dblock, oldest_start,
2570 oldest_trans_id, newest_mount_id);
2573 } else if (ret != 0) {
2577 SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2579 if (cur_dblock == oldest_start)
2583 if (oldest_trans_id == 0) {
2584 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2585 "journal-1225: No valid " "transactions found");
2587 /* j_start does not get set correctly if we don't replay any transactions.
2588 ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2589 ** copy the trans_id from the header
2591 if (valid_journal_header && replay_count == 0) {
2592 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2593 journal->j_trans_id =
2594 le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2595 /* check for trans_id overflow */
2596 if (journal->j_trans_id == 0)
2597 journal->j_trans_id = 10;
2598 journal->j_last_flush_trans_id =
2599 le32_to_cpu(jh->j_last_flush_trans_id);
2600 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2602 journal->j_mount_id = newest_mount_id + 1;
2604 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2605 "newest_mount_id to %lu", journal->j_mount_id);
2606 journal->j_first_unflushed_offset = journal->j_start;
2607 if (replay_count > 0) {
2609 "replayed %d transactions in %lu seconds\n",
2610 replay_count, get_seconds() - start);
2612 if (!bdev_read_only(sb->s_bdev) &&
2613 _update_journal_header_block(sb, journal->j_start,
2614 journal->j_last_flush_trans_id)) {
2615 /* replay failed, caller must call free_journal_ram and abort
2623 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2625 struct reiserfs_journal_list *jl;
2626 jl = kzalloc(sizeof(struct reiserfs_journal_list),
2627 GFP_NOFS | __GFP_NOFAIL);
2628 INIT_LIST_HEAD(&jl->j_list);
2629 INIT_LIST_HEAD(&jl->j_working_list);
2630 INIT_LIST_HEAD(&jl->j_tail_bh_list);
2631 INIT_LIST_HEAD(&jl->j_bh_list);
2632 mutex_init(&jl->j_commit_mutex);
2633 SB_JOURNAL(s)->j_num_lists++;
2634 get_journal_list(jl);
2638 static void journal_list_init(struct super_block *sb)
2640 SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2643 static int release_journal_dev(struct super_block *super,
2644 struct reiserfs_journal *journal)
2650 if (journal->j_dev_bd != NULL) {
2651 if (journal->j_dev_bd->bd_dev != super->s_dev)
2652 bd_release(journal->j_dev_bd);
2653 result = blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2654 journal->j_dev_bd = NULL;
2658 reiserfs_warning(super, "sh-457",
2659 "Cannot release journal device: %i", result);
2664 static int journal_init_dev(struct super_block *super,
2665 struct reiserfs_journal *journal,
2666 const char *jdev_name)
2670 fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE;
2671 char b[BDEVNAME_SIZE];
2675 journal->j_dev_bd = NULL;
2676 jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2677 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2679 if (bdev_read_only(super->s_bdev))
2680 blkdev_mode = FMODE_READ;
2682 /* there is no "jdev" option and journal is on separate device */
2683 if ((!jdev_name || !jdev_name[0])) {
2684 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2685 journal->j_dev_mode = blkdev_mode;
2686 if (IS_ERR(journal->j_dev_bd)) {
2687 result = PTR_ERR(journal->j_dev_bd);
2688 journal->j_dev_bd = NULL;
2689 reiserfs_warning(super, "sh-458",
2690 "cannot init journal device '%s': %i",
2691 __bdevname(jdev, b), result);
2693 } else if (jdev != super->s_dev) {
2694 result = bd_claim(journal->j_dev_bd, journal);
2696 blkdev_put(journal->j_dev_bd, blkdev_mode);
2700 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2706 journal->j_dev_mode = blkdev_mode;
2707 journal->j_dev_bd = open_bdev_exclusive(jdev_name,
2708 blkdev_mode, journal);
2709 if (IS_ERR(journal->j_dev_bd)) {
2710 result = PTR_ERR(journal->j_dev_bd);
2711 journal->j_dev_bd = NULL;
2712 reiserfs_warning(super,
2713 "journal_init_dev: Cannot open '%s': %i",
2718 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2719 reiserfs_info(super,
2720 "journal_init_dev: journal device: %s\n",
2721 bdevname(journal->j_dev_bd, b));
2726 * When creating/tuning a file system user can assign some
2727 * journal params within boundaries which depend on the ratio
2728 * blocksize/standard_blocksize.
2730 * For blocks >= standard_blocksize transaction size should
2731 * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2732 * then JOURNAL_TRANS_MAX_DEFAULT.
2734 * For blocks < standard_blocksize these boundaries should be
2735 * decreased proportionally.
2737 #define REISERFS_STANDARD_BLKSIZE (4096)
2739 static int check_advise_trans_params(struct super_block *sb,
2740 struct reiserfs_journal *journal)
2742 if (journal->j_trans_max) {
2743 /* Non-default journal params.
2744 Do sanity check for them. */
2746 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2747 ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2749 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2750 journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2751 SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2752 JOURNAL_MIN_RATIO) {
2753 reiserfs_warning(sb, "sh-462",
2754 "bad transaction max size (%u). "
2755 "FSCK?", journal->j_trans_max);
2758 if (journal->j_max_batch != (journal->j_trans_max) *
2759 JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2760 reiserfs_warning(sb, "sh-463",
2761 "bad transaction max batch (%u). "
2762 "FSCK?", journal->j_max_batch);
2766 /* Default journal params.
2767 The file system was created by old version
2768 of mkreiserfs, so some fields contain zeros,
2769 and we need to advise proper values for them */
2770 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2771 reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2775 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2776 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2777 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2783 ** must be called once on fs mount. calls journal_read for you
2785 int journal_init(struct super_block *sb, const char *j_dev_name,
2786 int old_format, unsigned int commit_max_age)
2788 int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2789 struct buffer_head *bhjh;
2790 struct reiserfs_super_block *rs;
2791 struct reiserfs_journal_header *jh;
2792 struct reiserfs_journal *journal;
2793 struct reiserfs_journal_list *jl;
2794 char b[BDEVNAME_SIZE];
2796 journal = SB_JOURNAL(sb) = vmalloc(sizeof(struct reiserfs_journal));
2798 reiserfs_warning(sb, "journal-1256",
2799 "unable to get memory for journal structure");
2802 memset(journal, 0, sizeof(struct reiserfs_journal));
2803 INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2804 INIT_LIST_HEAD(&journal->j_prealloc_list);
2805 INIT_LIST_HEAD(&journal->j_working_list);
2806 INIT_LIST_HEAD(&journal->j_journal_list);
2807 journal->j_persistent_trans = 0;
2808 if (reiserfs_allocate_list_bitmaps(sb,
2809 journal->j_list_bitmap,
2810 reiserfs_bmap_count(sb)))
2811 goto free_and_return;
2812 allocate_bitmap_nodes(sb);
2814 /* reserved for journal area support */
2815 SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2816 REISERFS_OLD_DISK_OFFSET_IN_BYTES
2818 reiserfs_bmap_count(sb) +
2820 REISERFS_DISK_OFFSET_IN_BYTES /
2821 sb->s_blocksize + 2);
2823 /* Sanity check to see is the standard journal fitting withing first bitmap
2824 (actual for small blocksizes) */
2825 if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2826 (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2827 SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2828 reiserfs_warning(sb, "journal-1393",
2829 "journal does not fit for area addressed "
2830 "by first of bitmap blocks. It starts at "
2831 "%u and its size is %u. Block size %ld",
2832 SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2833 SB_ONDISK_JOURNAL_SIZE(sb),
2835 goto free_and_return;
2838 if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2839 reiserfs_warning(sb, "sh-462",
2840 "unable to initialize jornal device");
2841 goto free_and_return;
2844 rs = SB_DISK_SUPER_BLOCK(sb);
2846 /* read journal header */
2847 bhjh = journal_bread(sb,
2848 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2849 SB_ONDISK_JOURNAL_SIZE(sb));
2851 reiserfs_warning(sb, "sh-459",
2852 "unable to read journal header");
2853 goto free_and_return;
2855 jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2857 /* make sure that journal matches to the super block */
2858 if (is_reiserfs_jr(rs)
2859 && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2860 sb_jp_journal_magic(rs))) {
2861 reiserfs_warning(sb, "sh-460",
2862 "journal header magic %x (device %s) does "
2863 "not match to magic found in super block %x",
2864 jh->jh_journal.jp_journal_magic,
2865 bdevname(journal->j_dev_bd, b),
2866 sb_jp_journal_magic(rs));
2868 goto free_and_return;
2871 journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2872 journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2873 journal->j_max_commit_age =
2874 le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2875 journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2877 if (check_advise_trans_params(sb, journal) != 0)
2878 goto free_and_return;
2879 journal->j_default_max_commit_age = journal->j_max_commit_age;
2881 if (commit_max_age != 0) {
2882 journal->j_max_commit_age = commit_max_age;
2883 journal->j_max_trans_age = commit_max_age;
2886 reiserfs_info(sb, "journal params: device %s, size %u, "
2887 "journal first block %u, max trans len %u, max batch %u, "
2888 "max commit age %u, max trans age %u\n",
2889 bdevname(journal->j_dev_bd, b),
2890 SB_ONDISK_JOURNAL_SIZE(sb),
2891 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2892 journal->j_trans_max,
2893 journal->j_max_batch,
2894 journal->j_max_commit_age, journal->j_max_trans_age);
2898 journal->j_list_bitmap_index = 0;
2899 journal_list_init(sb);
2901 memset(journal->j_list_hash_table, 0,
2902 JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2904 INIT_LIST_HEAD(&journal->j_dirty_buffers);
2905 spin_lock_init(&journal->j_dirty_buffers_lock);
2907 journal->j_start = 0;
2909 journal->j_len_alloc = 0;
2910 atomic_set(&(journal->j_wcount), 0);
2911 atomic_set(&(journal->j_async_throttle), 0);
2912 journal->j_bcount = 0;
2913 journal->j_trans_start_time = 0;
2914 journal->j_last = NULL;
2915 journal->j_first = NULL;
2916 init_waitqueue_head(&(journal->j_join_wait));
2917 mutex_init(&journal->j_mutex);
2918 mutex_init(&journal->j_flush_mutex);
2920 journal->j_trans_id = 10;
2921 journal->j_mount_id = 10;
2922 journal->j_state = 0;
2923 atomic_set(&(journal->j_jlock), 0);
2924 journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2925 journal->j_cnode_free_orig = journal->j_cnode_free_list;
2926 journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2927 journal->j_cnode_used = 0;
2928 journal->j_must_wait = 0;
2930 if (journal->j_cnode_free == 0) {
2931 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2932 "allocation failed (%ld bytes). Journal is "
2933 "too large for available memory. Usually "
2934 "this is due to a journal that is too large.",
2935 sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2936 goto free_and_return;
2939 init_journal_hash(sb);
2940 jl = journal->j_current_jl;
2941 jl->j_list_bitmap = get_list_bitmap(sb, jl);
2942 if (!jl->j_list_bitmap) {
2943 reiserfs_warning(sb, "journal-2005",
2944 "get_list_bitmap failed for journal list 0");
2945 goto free_and_return;
2947 if (journal_read(sb) < 0) {
2948 reiserfs_warning(sb, "reiserfs-2006",
2949 "Replay Failure, unable to mount");
2950 goto free_and_return;
2953 reiserfs_mounted_fs_count++;
2954 if (reiserfs_mounted_fs_count <= 1)
2955 commit_wq = create_workqueue("reiserfs");
2957 INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2958 journal->j_work_sb = sb;
2961 free_journal_ram(sb);
2966 ** test for a polite end of the current transaction. Used by file_write, and should
2967 ** be used by delete to make sure they don't write more than can fit inside a single
2970 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2973 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2974 time_t now = get_seconds();
2975 /* cannot restart while nested */
2976 BUG_ON(!th->t_trans_id);
2977 if (th->t_refcount > 1)
2979 if (journal->j_must_wait > 0 ||
2980 (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2981 atomic_read(&(journal->j_jlock)) ||
2982 (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2983 journal->j_cnode_free < (journal->j_trans_max * 3)) {
2986 /* protected by the BKL here */
2987 journal->j_len_alloc += new_alloc;
2988 th->t_blocks_allocated += new_alloc ;
2992 /* this must be called inside a transaction, and requires the
2993 ** kernel_lock to be held
2995 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2997 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2998 BUG_ON(!th->t_trans_id);
2999 journal->j_must_wait = 1;
3000 set_bit(J_WRITERS_BLOCKED, &journal->j_state);
3004 /* this must be called without a transaction started, and does not
3007 void reiserfs_allow_writes(struct super_block *s)
3009 struct reiserfs_journal *journal = SB_JOURNAL(s);
3010 clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
3011 wake_up(&journal->j_join_wait);
3014 /* this must be called without a transaction started, and does not
3017 void reiserfs_wait_on_write_block(struct super_block *s)
3019 struct reiserfs_journal *journal = SB_JOURNAL(s);
3020 wait_event(journal->j_join_wait,
3021 !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
3024 static void queue_log_writer(struct super_block *s)
3027 struct reiserfs_journal *journal = SB_JOURNAL(s);
3028 set_bit(J_WRITERS_QUEUED, &journal->j_state);
3031 * we don't want to use wait_event here because
3032 * we only want to wait once.
3034 init_waitqueue_entry(&wait, current);
3035 add_wait_queue(&journal->j_join_wait, &wait);
3036 set_current_state(TASK_UNINTERRUPTIBLE);
3037 if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
3038 reiserfs_write_unlock(s);
3040 reiserfs_write_lock(s);
3042 __set_current_state(TASK_RUNNING);
3043 remove_wait_queue(&journal->j_join_wait, &wait);
3046 static void wake_queued_writers(struct super_block *s)
3048 struct reiserfs_journal *journal = SB_JOURNAL(s);
3049 if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
3050 wake_up(&journal->j_join_wait);
3053 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
3055 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3056 unsigned long bcount = journal->j_bcount;
3058 reiserfs_write_unlock(sb);
3059 schedule_timeout_uninterruptible(1);
3060 reiserfs_write_lock(sb);
3061 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
3062 while ((atomic_read(&journal->j_wcount) > 0 ||
3063 atomic_read(&journal->j_jlock)) &&
3064 journal->j_trans_id == trans_id) {
3065 queue_log_writer(sb);
3067 if (journal->j_trans_id != trans_id)
3069 if (bcount == journal->j_bcount)
3071 bcount = journal->j_bcount;
3075 /* join == true if you must join an existing transaction.
3076 ** join == false if you can deal with waiting for others to finish
3078 ** this will block until the transaction is joinable. send the number of blocks you
3079 ** expect to use in nblocks.
3081 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3082 struct super_block *sb, unsigned long nblocks,
3085 time_t now = get_seconds();
3086 unsigned int old_trans_id;
3087 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3088 struct reiserfs_transaction_handle myth;
3089 int sched_count = 0;
3092 reiserfs_check_lock_depth(sb, "journal_begin");
3093 BUG_ON(nblocks > journal->j_trans_max);
3095 PROC_INFO_INC(sb, journal.journal_being);
3096 /* set here for journal_join */
3102 if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3104 retval = journal->j_errno;
3107 journal->j_bcount++;
3109 if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3111 reiserfs_write_unlock(sb);
3112 reiserfs_wait_on_write_block(sb);
3113 reiserfs_write_lock(sb);
3114 PROC_INFO_INC(sb, journal.journal_relock_writers);
3117 now = get_seconds();
3119 /* if there is no room in the journal OR
3120 ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
3121 ** we don't sleep if there aren't other writers
3124 if ((!join && journal->j_must_wait > 0) ||
3126 && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3127 || (!join && atomic_read(&journal->j_wcount) > 0
3128 && journal->j_trans_start_time > 0
3129 && (now - journal->j_trans_start_time) >
3130 journal->j_max_trans_age) || (!join
3131 && atomic_read(&journal->j_jlock))
3132 || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3134 old_trans_id = journal->j_trans_id;
3135 unlock_journal(sb); /* allow others to finish this transaction */
3137 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3138 journal->j_max_batch &&
3139 ((journal->j_len + nblocks + 2) * 100) <
3140 (journal->j_len_alloc * 75)) {
3141 if (atomic_read(&journal->j_wcount) > 10) {
3143 queue_log_writer(sb);
3147 /* don't mess with joining the transaction if all we have to do is
3148 * wait for someone else to do a commit
3150 if (atomic_read(&journal->j_jlock)) {
3151 while (journal->j_trans_id == old_trans_id &&
3152 atomic_read(&journal->j_jlock)) {
3153 queue_log_writer(sb);
3157 retval = journal_join(&myth, sb, 1);
3161 /* someone might have ended the transaction while we joined */
3162 if (old_trans_id != journal->j_trans_id) {
3163 retval = do_journal_end(&myth, sb, 1, 0);
3165 retval = do_journal_end(&myth, sb, 1, COMMIT_NOW);
3171 PROC_INFO_INC(sb, journal.journal_relock_wcount);
3174 /* we are the first writer, set trans_id */
3175 if (journal->j_trans_start_time == 0) {
3176 journal->j_trans_start_time = get_seconds();
3178 atomic_inc(&(journal->j_wcount));
3179 journal->j_len_alloc += nblocks;
3180 th->t_blocks_logged = 0;
3181 th->t_blocks_allocated = nblocks;
3182 th->t_trans_id = journal->j_trans_id;
3184 INIT_LIST_HEAD(&th->t_list);
3189 memset(th, 0, sizeof(*th));
3190 /* Re-set th->t_super, so we can properly keep track of how many
3191 * persistent transactions there are. We need to do this so if this
3192 * call is part of a failed restart_transaction, we can free it later */
3197 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3203 struct reiserfs_transaction_handle *th;
3205 /* if we're nesting into an existing transaction. It will be
3206 ** persistent on its own
3208 if (reiserfs_transaction_running(s)) {
3209 th = current->journal_info;
3211 BUG_ON(th->t_refcount < 2);
3215 th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3218 ret = journal_begin(th, s, nblocks);
3224 SB_JOURNAL(s)->j_persistent_trans++;
3228 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3230 struct super_block *s = th->t_super;
3233 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3236 if (th->t_refcount == 0) {
3237 SB_JOURNAL(s)->j_persistent_trans--;
3243 static int journal_join(struct reiserfs_transaction_handle *th,
3244 struct super_block *sb, unsigned long nblocks)
3246 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3248 /* this keeps do_journal_end from NULLing out the current->journal_info
3251 th->t_handle_save = cur_th;
3252 BUG_ON(cur_th && cur_th->t_refcount > 1);
3253 return do_journal_begin_r(th, sb, nblocks, JBEGIN_JOIN);
3256 int journal_join_abort(struct reiserfs_transaction_handle *th,
3257 struct super_block *sb, unsigned long nblocks)
3259 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3261 /* this keeps do_journal_end from NULLing out the current->journal_info
3264 th->t_handle_save = cur_th;
3265 BUG_ON(cur_th && cur_th->t_refcount > 1);
3266 return do_journal_begin_r(th, sb, nblocks, JBEGIN_ABORT);
3269 int journal_begin(struct reiserfs_transaction_handle *th,
3270 struct super_block *sb, unsigned long nblocks)
3272 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3275 th->t_handle_save = NULL;
3277 /* we are nesting into the current transaction */
3278 if (cur_th->t_super == sb) {
3279 BUG_ON(!cur_th->t_refcount);
3280 cur_th->t_refcount++;
3281 memcpy(th, cur_th, sizeof(*th));
3282 if (th->t_refcount <= 1)
3283 reiserfs_warning(sb, "reiserfs-2005",
3284 "BAD: refcount <= 1, but "
3285 "journal_info != 0");
3288 /* we've ended up with a handle from a different filesystem.
3289 ** save it and restore on journal_end. This should never
3292 reiserfs_warning(sb, "clm-2100",
3293 "nesting info a different FS");
3294 th->t_handle_save = current->journal_info;
3295 current->journal_info = th;
3298 current->journal_info = th;
3300 ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3301 BUG_ON(current->journal_info != th);
3303 /* I guess this boils down to being the reciprocal of clm-2100 above.
3304 * If do_journal_begin_r fails, we need to put it back, since journal_end
3305 * won't be called to do it. */
3307 current->journal_info = th->t_handle_save;
3309 BUG_ON(!th->t_refcount);
3315 ** puts bh into the current transaction. If it was already there, reorders removes the
3316 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3318 ** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the
3319 ** transaction is committed.
3321 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3323 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3324 struct super_block *sb, struct buffer_head *bh)
3326 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3327 struct reiserfs_journal_cnode *cn = NULL;
3328 int count_already_incd = 0;
3330 BUG_ON(!th->t_trans_id);
3332 PROC_INFO_INC(sb, journal.mark_dirty);
3333 if (th->t_trans_id != journal->j_trans_id) {
3334 reiserfs_panic(th->t_super, "journal-1577",
3335 "handle trans id %ld != current trans id %ld",
3336 th->t_trans_id, journal->j_trans_id);
3341 prepared = test_clear_buffer_journal_prepared(bh);
3342 clear_buffer_journal_restore_dirty(bh);
3343 /* already in this transaction, we are done */
3344 if (buffer_journaled(bh)) {
3345 PROC_INFO_INC(sb, journal.mark_dirty_already);
3349 /* this must be turned into a panic instead of a warning. We can't allow
3350 ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3351 ** could get to disk too early. NOT GOOD.
3353 if (!prepared || buffer_dirty(bh)) {
3354 reiserfs_warning(sb, "journal-1777",
3355 "buffer %llu bad state "
3356 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3357 (unsigned long long)bh->b_blocknr,
3358 prepared ? ' ' : '!',
3359 buffer_locked(bh) ? ' ' : '!',
3360 buffer_dirty(bh) ? ' ' : '!',
3361 buffer_journal_dirty(bh) ? ' ' : '!');
3364 if (atomic_read(&(journal->j_wcount)) <= 0) {
3365 reiserfs_warning(sb, "journal-1409",
3366 "returning because j_wcount was %d",
3367 atomic_read(&(journal->j_wcount)));
3370 /* this error means I've screwed up, and we've overflowed the transaction.
3371 ** Nothing can be done here, except make the FS readonly or panic.
3373 if (journal->j_len >= journal->j_trans_max) {
3374 reiserfs_panic(th->t_super, "journal-1413",
3375 "j_len (%lu) is too big",
3379 if (buffer_journal_dirty(bh)) {
3380 count_already_incd = 1;
3381 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3382 clear_buffer_journal_dirty(bh);
3385 if (journal->j_len > journal->j_len_alloc) {
3386 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3389 set_buffer_journaled(bh);
3391 /* now put this guy on the end */
3395 reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3398 if (th->t_blocks_logged == th->t_blocks_allocated) {
3399 th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3400 journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3402 th->t_blocks_logged++;
3406 cn->blocknr = bh->b_blocknr;
3409 insert_journal_hash(journal->j_hash_table, cn);
3410 if (!count_already_incd) {
3415 cn->prev = journal->j_last;
3417 if (journal->j_last) {
3418 journal->j_last->next = cn;
3419 journal->j_last = cn;
3421 journal->j_first = cn;
3422 journal->j_last = cn;
3427 int journal_end(struct reiserfs_transaction_handle *th,
3428 struct super_block *sb, unsigned long nblocks)
3430 if (!current->journal_info && th->t_refcount > 1)
3431 reiserfs_warning(sb, "REISER-NESTING",
3432 "th NULL, refcount %d", th->t_refcount);
3434 if (!th->t_trans_id) {
3440 if (th->t_refcount > 0) {
3441 struct reiserfs_transaction_handle *cur_th =
3442 current->journal_info;
3444 /* we aren't allowed to close a nested transaction on a different
3445 ** filesystem from the one in the task struct
3447 BUG_ON(cur_th->t_super != th->t_super);
3450 memcpy(current->journal_info, th, sizeof(*th));
3455 return do_journal_end(th, sb, nblocks, 0);
3459 /* removes from the current transaction, relsing and descrementing any counters.
3460 ** also files the removed buffer directly onto the clean list
3462 ** called by journal_mark_freed when a block has been deleted
3464 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3466 static int remove_from_transaction(struct super_block *sb,
3467 b_blocknr_t blocknr, int already_cleaned)
3469 struct buffer_head *bh;
3470 struct reiserfs_journal_cnode *cn;
3471 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3474 cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3475 if (!cn || !cn->bh) {
3480 cn->prev->next = cn->next;
3483 cn->next->prev = cn->prev;
3485 if (cn == journal->j_first) {
3486 journal->j_first = cn->next;
3488 if (cn == journal->j_last) {
3489 journal->j_last = cn->prev;
3492 remove_journal_hash(sb, journal->j_hash_table, NULL,
3494 clear_buffer_journaled(bh); /* don't log this one */
3496 if (!already_cleaned) {
3497 clear_buffer_journal_dirty(bh);
3498 clear_buffer_dirty(bh);
3499 clear_buffer_journal_test(bh);
3501 if (atomic_read(&(bh->b_count)) < 0) {
3502 reiserfs_warning(sb, "journal-1752",
3508 journal->j_len_alloc--;
3514 ** for any cnode in a journal list, it can only be dirtied of all the
3515 ** transactions that include it are committed to disk.
3516 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3517 ** and 0 if you aren't
3519 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3520 ** blocks for a given transaction on disk
3523 static int can_dirty(struct reiserfs_journal_cnode *cn)
3525 struct super_block *sb = cn->sb;
3526 b_blocknr_t blocknr = cn->blocknr;
3527 struct reiserfs_journal_cnode *cur = cn->hprev;
3530 /* first test hprev. These are all newer than cn, so any node here
3531 ** with the same block number and dev means this node can't be sent
3532 ** to disk right now.
3534 while (cur && can_dirty) {
3535 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3536 cur->blocknr == blocknr) {
3541 /* then test hnext. These are all older than cn. As long as they
3542 ** are committed to the log, it is safe to write cn to disk
3545 while (cur && can_dirty) {
3546 if (cur->jlist && cur->jlist->j_len > 0 &&
3547 atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3548 cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3556 /* syncs the commit blocks, but does not force the real buffers to disk
3557 ** will wait until the current transaction is done/committed before returning
3559 int journal_end_sync(struct reiserfs_transaction_handle *th,
3560 struct super_block *sb, unsigned long nblocks)
3562 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3564 BUG_ON(!th->t_trans_id);
3565 /* you can sync while nested, very, very bad */
3566 BUG_ON(th->t_refcount > 1);
3567 if (journal->j_len == 0) {
3568 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3570 journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
3572 return do_journal_end(th, sb, nblocks, COMMIT_NOW | WAIT);
3576 ** writeback the pending async commits to disk
3578 static void flush_async_commits(struct work_struct *work)
3580 struct reiserfs_journal *journal =
3581 container_of(work, struct reiserfs_journal, j_work.work);
3582 struct super_block *sb = journal->j_work_sb;
3583 struct reiserfs_journal_list *jl;
3584 struct list_head *entry;
3586 reiserfs_write_lock(sb);
3587 if (!list_empty(&journal->j_journal_list)) {
3588 /* last entry is the youngest, commit it and you get everything */
3589 entry = journal->j_journal_list.prev;
3590 jl = JOURNAL_LIST_ENTRY(entry);
3591 flush_commit_list(sb, jl, 1);
3593 reiserfs_write_unlock(sb);
3597 ** flushes any old transactions to disk
3598 ** ends the current transaction if it is too old
3600 int reiserfs_flush_old_commits(struct super_block *sb)
3603 struct reiserfs_transaction_handle th;
3604 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3606 now = get_seconds();
3607 /* safety check so we don't flush while we are replaying the log during
3610 if (list_empty(&journal->j_journal_list)) {
3614 /* check the current transaction. If there are no writers, and it is
3615 * too old, finish it, and force the commit blocks to disk
3617 if (atomic_read(&journal->j_wcount) <= 0 &&
3618 journal->j_trans_start_time > 0 &&
3619 journal->j_len > 0 &&
3620 (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3621 if (!journal_join(&th, sb, 1)) {
3622 reiserfs_prepare_for_journal(sb,
3623 SB_BUFFER_WITH_SB(sb),
3625 journal_mark_dirty(&th, sb,
3626 SB_BUFFER_WITH_SB(sb));
3628 /* we're only being called from kreiserfsd, it makes no sense to do
3629 ** an async commit so that kreiserfsd can do it later
3631 do_journal_end(&th, sb, 1, COMMIT_NOW | WAIT);
3638 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3640 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3641 ** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just
3642 ** flushes the commit list and returns 0.
3644 ** Won't batch when flush or commit_now is set. Also won't batch when others are waiting on j_join_wait.
3646 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3648 static int check_journal_end(struct reiserfs_transaction_handle *th,
3649 struct super_block *sb, unsigned long nblocks,
3654 int flush = flags & FLUSH_ALL;
3655 int commit_now = flags & COMMIT_NOW;
3656 int wait_on_commit = flags & WAIT;
3657 struct reiserfs_journal_list *jl;
3658 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3660 BUG_ON(!th->t_trans_id);
3662 if (th->t_trans_id != journal->j_trans_id) {
3663 reiserfs_panic(th->t_super, "journal-1577",
3664 "handle trans id %ld != current trans id %ld",
3665 th->t_trans_id, journal->j_trans_id);
3668 journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3669 if (atomic_read(&(journal->j_wcount)) > 0) { /* <= 0 is allowed. unmounting might not call begin */
3670 atomic_dec(&(journal->j_wcount));
3673 /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3674 ** will be dealt with by next transaction that actually writes something, but should be taken
3675 ** care of in this trans
3677 BUG_ON(journal->j_len == 0);
3679 /* if wcount > 0, and we are called to with flush or commit_now,
3680 ** we wait on j_join_wait. We will wake up when the last writer has
3681 ** finished the transaction, and started it on its way to the disk.
3682 ** Then, we flush the commit or journal list, and just return 0
3683 ** because the rest of journal end was already done for this transaction.
3685 if (atomic_read(&(journal->j_wcount)) > 0) {
3686 if (flush || commit_now) {
3689 jl = journal->j_current_jl;
3690 trans_id = jl->j_trans_id;
3692 jl->j_state |= LIST_COMMIT_PENDING;
3693 atomic_set(&(journal->j_jlock), 1);
3695 journal->j_next_full_flush = 1;
3699 /* sleep while the current transaction is still j_jlocked */
3700 while (journal->j_trans_id == trans_id) {
3701 if (atomic_read(&journal->j_jlock)) {
3702 queue_log_writer(sb);
3705 if (journal->j_trans_id == trans_id) {
3706 atomic_set(&(journal->j_jlock),
3712 BUG_ON(journal->j_trans_id == trans_id);
3715 && journal_list_still_alive(sb, trans_id)
3716 && wait_on_commit) {
3717 flush_commit_list(sb, jl, 1);
3725 /* deal with old transactions where we are the last writers */
3726 now = get_seconds();
3727 if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3729 journal->j_next_async_flush = 1;
3731 /* don't batch when someone is waiting on j_join_wait */
3732 /* don't batch when syncing the commit or flushing the whole trans */
3733 if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3734 && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3735 && journal->j_len_alloc < journal->j_max_batch
3736 && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3737 journal->j_bcount++;
3742 if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3743 reiserfs_panic(sb, "journal-003",
3744 "j_start (%ld) is too high",
3751 ** Does all the work that makes deleting blocks safe.
3752 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3755 ** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes
3756 ** before this transaction has finished.
3758 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with
3759 ** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash,
3760 ** the block can't be reallocated yet.
3762 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3764 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3765 struct super_block *sb, b_blocknr_t blocknr)
3767 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3768 struct reiserfs_journal_cnode *cn = NULL;
3769 struct buffer_head *bh = NULL;
3770 struct reiserfs_list_bitmap *jb = NULL;
3772 BUG_ON(!th->t_trans_id);
3774 cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3779 /* if it is journal new, we just remove it from this transaction */
3780 if (bh && buffer_journal_new(bh)) {
3781 clear_buffer_journal_new(bh);
3782 clear_prepared_bits(bh);
3783 reiserfs_clean_and_file_buffer(bh);
3784 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3786 /* set the bit for this block in the journal bitmap for this transaction */
3787 jb = journal->j_current_jl->j_list_bitmap;
3789 reiserfs_panic(sb, "journal-1702",
3790 "journal_list_bitmap is NULL");
3792 set_bit_in_list_bitmap(sb, blocknr, jb);
3794 /* Note, the entire while loop is not allowed to schedule. */
3797 clear_prepared_bits(bh);
3798 reiserfs_clean_and_file_buffer(bh);
3800 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3802 /* find all older transactions with this block, make sure they don't try to write it out */
3803 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3806 if (sb == cn->sb && blocknr == cn->blocknr) {
3807 set_bit(BLOCK_FREED, &cn->state);
3810 /* remove_from_transaction will brelse the buffer if it was
3811 ** in the current trans
3813 clear_buffer_journal_dirty(cn->
3815 clear_buffer_dirty(cn->bh);
3816 clear_buffer_journal_test(cn->
3821 (&(cn->bh->b_count)) < 0) {
3822 reiserfs_warning(sb,
3824 "cn->bh->b_count < 0");
3827 if (cn->jlist) { /* since we are clearing the bh, we MUST dec nonzerolen */
3840 release_buffer_page(bh); /* get_hash grabs the buffer */
3844 void reiserfs_update_inode_transaction(struct inode *inode)
3846 struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3847 REISERFS_I(inode)->i_jl = journal->j_current_jl;
3848 REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3852 * returns -1 on error, 0 if no commits/barriers were done and 1
3853 * if a transaction was actually committed and the barrier was done
3855 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3856 struct reiserfs_journal_list *jl)
3858 struct reiserfs_transaction_handle th;
3859 struct super_block *sb = inode->i_sb;
3860 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3863 /* is it from the current transaction, or from an unknown transaction? */
3864 if (id == journal->j_trans_id) {
3865 jl = journal->j_current_jl;
3866 /* try to let other writers come in and grow this transaction */
3867 let_transaction_grow(sb, id);
3868 if (journal->j_trans_id != id) {
3869 goto flush_commit_only;
3872 ret = journal_begin(&th, sb, 1);
3876 /* someone might have ended this transaction while we joined */
3877 if (journal->j_trans_id != id) {
3878 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3880 journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3881 ret = journal_end(&th, sb, 1);
3882 goto flush_commit_only;
3885 ret = journal_end_sync(&th, sb, 1);
3890 /* this gets tricky, we have to make sure the journal list in
3891 * the inode still exists. We know the list is still around
3892 * if we've got a larger transaction id than the oldest list
3895 if (journal_list_still_alive(inode->i_sb, id)) {
3897 * we only set ret to 1 when we know for sure
3898 * the barrier hasn't been started yet on the commit
3901 if (atomic_read(&jl->j_commit_left) > 1)
3903 flush_commit_list(sb, jl, 1);
3904 if (journal->j_errno)
3905 ret = journal->j_errno;
3908 /* otherwise the list is gone, and long since committed */
3912 int reiserfs_commit_for_inode(struct inode *inode)
3914 unsigned int id = REISERFS_I(inode)->i_trans_id;
3915 struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3917 /* for the whole inode, assume unset id means it was
3918 * changed in the current transaction. More conservative
3921 reiserfs_update_inode_transaction(inode);
3922 id = REISERFS_I(inode)->i_trans_id;
3923 /* jl will be updated in __commit_trans_jl */
3926 return __commit_trans_jl(inode, id, jl);
3929 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3930 struct buffer_head *bh)
3932 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3933 PROC_INFO_INC(sb, journal.restore_prepared);
3937 if (test_clear_buffer_journal_restore_dirty(bh) &&
3938 buffer_journal_dirty(bh)) {
3939 struct reiserfs_journal_cnode *cn;
3940 cn = get_journal_hash_dev(sb,
3941 journal->j_list_hash_table,
3943 if (cn && can_dirty(cn)) {
3944 set_buffer_journal_test(bh);
3945 mark_buffer_dirty(bh);
3948 clear_buffer_journal_prepared(bh);
3951 extern struct tree_balance *cur_tb;
3953 ** before we can change a metadata block, we have to make sure it won't
3954 ** be written to disk while we are altering it. So, we must:
3959 int reiserfs_prepare_for_journal(struct super_block *sb,
3960 struct buffer_head *bh, int wait)
3962 PROC_INFO_INC(sb, journal.prepare);
3964 if (!trylock_buffer(bh)) {
3969 set_buffer_journal_prepared(bh);
3970 if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3971 clear_buffer_journal_test(bh);
3972 set_buffer_journal_restore_dirty(bh);
3978 static void flush_old_journal_lists(struct super_block *s)
3980 struct reiserfs_journal *journal = SB_JOURNAL(s);
3981 struct reiserfs_journal_list *jl;
3982 struct list_head *entry;
3983 time_t now = get_seconds();
3985 while (!list_empty(&journal->j_journal_list)) {
3986 entry = journal->j_journal_list.next;
3987 jl = JOURNAL_LIST_ENTRY(entry);
3988 /* this check should always be run, to send old lists to disk */
3989 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) &&
3990 atomic_read(&jl->j_commit_left) == 0 &&
3991 test_transaction(s, jl)) {
3992 flush_used_journal_lists(s, jl);
4000 ** long and ugly. If flush, will not return until all commit
4001 ** blocks and all real buffers in the trans are on disk.
4002 ** If no_async, won't return until all commit blocks are on disk.
4004 ** keep reading, there are comments as you go along
4006 ** If the journal is aborted, we just clean up. Things like flushing
4007 ** journal lists, etc just won't happen.
4009 static int do_journal_end(struct reiserfs_transaction_handle *th,
4010 struct super_block *sb, unsigned long nblocks,
4013 struct reiserfs_journal *journal = SB_JOURNAL(sb);
4014 struct reiserfs_journal_cnode *cn, *next, *jl_cn;
4015 struct reiserfs_journal_cnode *last_cn = NULL;
4016 struct reiserfs_journal_desc *desc;
4017 struct reiserfs_journal_commit *commit;
4018 struct buffer_head *c_bh; /* commit bh */
4019 struct buffer_head *d_bh; /* desc bh */
4020 int cur_write_start = 0; /* start index of current log write */
4025 struct reiserfs_journal_list *jl, *temp_jl;
4026 struct list_head *entry, *safe;
4027 unsigned long jindex;
4028 unsigned int commit_trans_id;
4031 BUG_ON(th->t_refcount > 1);
4032 BUG_ON(!th->t_trans_id);
4034 /* protect flush_older_commits from doing mistakes if the
4035 transaction ID counter gets overflowed. */
4036 if (th->t_trans_id == ~0U)
4037 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
4038 flush = flags & FLUSH_ALL;
4039 wait_on_commit = flags & WAIT;
4042 current->journal_info = th->t_handle_save;
4043 reiserfs_check_lock_depth(sb, "journal end");
4044 if (journal->j_len == 0) {
4045 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
4047 journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
4051 if (journal->j_next_full_flush) {
4055 if (journal->j_next_async_flush) {
4056 flags |= COMMIT_NOW | WAIT;
4060 /* check_journal_end locks the journal, and unlocks if it does not return 1
4061 ** it tells us if we should continue with the journal_end, or just return
4063 if (!check_journal_end(th, sb, nblocks, flags)) {
4065 wake_queued_writers(sb);
4066 reiserfs_async_progress_wait(sb);
4070 /* check_journal_end might set these, check again */
4071 if (journal->j_next_full_flush) {
4076 ** j must wait means we have to flush the log blocks, and the real blocks for
4079 if (journal->j_must_wait > 0) {
4082 #ifdef REISERFS_PREALLOCATE
4083 /* quota ops might need to nest, setup the journal_info pointer for them
4084 * and raise the refcount so that it is > 0. */
4085 current->journal_info = th;
4087 reiserfs_discard_all_prealloc(th); /* it should not involve new blocks into
4088 * the transaction */
4090 current->journal_info = th->t_handle_save;
4093 /* setup description block */
4096 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4098 set_buffer_uptodate(d_bh);
4099 desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4100 memset(d_bh->b_data, 0, d_bh->b_size);
4101 memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4102 set_desc_trans_id(desc, journal->j_trans_id);
4104 /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */
4105 c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4106 ((journal->j_start + journal->j_len +
4107 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4108 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4109 memset(c_bh->b_data, 0, c_bh->b_size);
4110 set_commit_trans_id(commit, journal->j_trans_id);
4111 set_buffer_uptodate(c_bh);
4113 /* init this journal list */
4114 jl = journal->j_current_jl;
4116 /* we lock the commit before doing anything because
4117 * we want to make sure nobody tries to run flush_commit_list until
4118 * the new transaction is fully setup, and we've already flushed the
4121 reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4123 /* save the transaction id in case we need to commit it later */
4124 commit_trans_id = jl->j_trans_id;
4126 atomic_set(&jl->j_older_commits_done, 0);
4127 jl->j_trans_id = journal->j_trans_id;
4128 jl->j_timestamp = journal->j_trans_start_time;
4129 jl->j_commit_bh = c_bh;
4130 jl->j_start = journal->j_start;
4131 jl->j_len = journal->j_len;
4132 atomic_set(&jl->j_nonzerolen, journal->j_len);
4133 atomic_set(&jl->j_commit_left, journal->j_len + 2);
4134 jl->j_realblock = NULL;
4136 /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4137 ** for each real block, add it to the journal list hash,
4138 ** copy into real block index array in the commit or desc block
4140 trans_half = journal_trans_half(sb->s_blocksize);
4141 for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4142 if (buffer_journaled(cn->bh)) {
4143 jl_cn = get_cnode(sb);
4145 reiserfs_panic(sb, "journal-1676",
4146 "get_cnode returned NULL");
4149 jl->j_realblock = jl_cn;
4151 jl_cn->prev = last_cn;
4154 last_cn->next = jl_cn;
4157 /* make sure the block we are trying to log is not a block
4158 of journal or reserved area */
4160 if (is_block_in_log_or_reserved_area
4161 (sb, cn->bh->b_blocknr)) {
4162 reiserfs_panic(sb, "journal-2332",
4163 "Trying to log block %lu, "
4164 "which is a log block",
4167 jl_cn->blocknr = cn->bh->b_blocknr;
4172 insert_journal_hash(journal->j_list_hash_table, jl_cn);
4173 if (i < trans_half) {
4174 desc->j_realblock[i] =
4175 cpu_to_le32(cn->bh->b_blocknr);
4177 commit->j_realblock[i - trans_half] =
4178 cpu_to_le32(cn->bh->b_blocknr);
4184 set_desc_trans_len(desc, journal->j_len);
4185 set_desc_mount_id(desc, journal->j_mount_id);
4186 set_desc_trans_id(desc, journal->j_trans_id);
4187 set_commit_trans_len(commit, journal->j_len);
4189 /* special check in case all buffers in the journal were marked for not logging */
4190 BUG_ON(journal->j_len == 0);
4192 /* we're about to dirty all the log blocks, mark the description block
4193 * dirty now too. Don't mark the commit block dirty until all the
4194 * others are on disk
4196 mark_buffer_dirty(d_bh);
4198 /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4199 cur_write_start = journal->j_start;
4200 cn = journal->j_first;
4201 jindex = 1; /* start at one so we don't get the desc again */
4203 clear_buffer_journal_new(cn->bh);
4204 /* copy all the real blocks into log area. dirty log blocks */
4205 if (buffer_journaled(cn->bh)) {
4206 struct buffer_head *tmp_bh;
4211 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4214 SB_ONDISK_JOURNAL_SIZE(sb)));
4215 set_buffer_uptodate(tmp_bh);
4216 page = cn->bh->b_page;
4218 memcpy(tmp_bh->b_data,
4219 addr + offset_in_page(cn->bh->b_data),
4222 mark_buffer_dirty(tmp_bh);
4224 set_buffer_journal_dirty(cn->bh);
4225 clear_buffer_journaled(cn->bh);
4227 /* JDirty cleared sometime during transaction. don't log this one */
4228 reiserfs_warning(sb, "journal-2048",
4229 "BAD, buffer in journal hash, "
4236 reiserfs_write_unlock(sb);
4238 reiserfs_write_lock(sb);
4241 /* we are done with both the c_bh and d_bh, but
4242 ** c_bh must be written after all other commit blocks,
4243 ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4246 journal->j_current_jl = alloc_journal_list(sb);
4248 /* now it is safe to insert this transaction on the main list */
4249 list_add_tail(&jl->j_list, &journal->j_journal_list);
4250 list_add_tail(&jl->j_working_list, &journal->j_working_list);
4251 journal->j_num_work_lists++;
4253 /* reset journal values for the next transaction */
4254 old_start = journal->j_start;
4256 (journal->j_start + journal->j_len +
4257 2) % SB_ONDISK_JOURNAL_SIZE(sb);
4258 atomic_set(&(journal->j_wcount), 0);
4259 journal->j_bcount = 0;
4260 journal->j_last = NULL;
4261 journal->j_first = NULL;
4263 journal->j_trans_start_time = 0;
4264 /* check for trans_id overflow */
4265 if (++journal->j_trans_id == 0)
4266 journal->j_trans_id = 10;
4267 journal->j_current_jl->j_trans_id = journal->j_trans_id;
4268 journal->j_must_wait = 0;
4269 journal->j_len_alloc = 0;
4270 journal->j_next_full_flush = 0;
4271 journal->j_next_async_flush = 0;
4272 init_journal_hash(sb);
4274 // make sure reiserfs_add_jh sees the new current_jl before we
4275 // write out the tails
4278 /* tail conversion targets have to hit the disk before we end the
4279 * transaction. Otherwise a later transaction might repack the tail
4280 * before this transaction commits, leaving the data block unflushed and
4281 * clean, if we crash before the later transaction commits, the data block
4284 if (!list_empty(&jl->j_tail_bh_list)) {
4285 reiserfs_write_unlock(sb);
4286 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4287 journal, jl, &jl->j_tail_bh_list);
4288 reiserfs_write_lock(sb);
4290 BUG_ON(!list_empty(&jl->j_tail_bh_list));
4291 mutex_unlock(&jl->j_commit_mutex);
4293 /* honor the flush wishes from the caller, simple commits can
4294 ** be done outside the journal lock, they are done below
4296 ** if we don't flush the commit list right now, we put it into
4297 ** the work queue so the people waiting on the async progress work
4298 ** queue don't wait for this proc to flush journal lists and such.
4301 flush_commit_list(sb, jl, 1);
4302 flush_journal_list(sb, jl, 1);
4303 } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4304 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4306 /* if the next transaction has any chance of wrapping, flush
4307 ** transactions that might get overwritten. If any journal lists are very
4308 ** old flush them as well.
4311 list_for_each_safe(entry, safe, &journal->j_journal_list) {
4312 temp_jl = JOURNAL_LIST_ENTRY(entry);
4313 if (journal->j_start <= temp_jl->j_start) {
4314 if ((journal->j_start + journal->j_trans_max + 1) >=
4316 flush_used_journal_lists(sb, temp_jl);
4318 } else if ((journal->j_start +
4319 journal->j_trans_max + 1) <
4320 SB_ONDISK_JOURNAL_SIZE(sb)) {
4321 /* if we don't cross into the next transaction and we don't
4322 * wrap, there is no way we can overlap any later transactions
4327 } else if ((journal->j_start +
4328 journal->j_trans_max + 1) >
4329 SB_ONDISK_JOURNAL_SIZE(sb)) {
4330 if (((journal->j_start + journal->j_trans_max + 1) %
4331 SB_ONDISK_JOURNAL_SIZE(sb)) >=
4333 flush_used_journal_lists(sb, temp_jl);
4336 /* we don't overlap anything from out start to the end of the
4337 * log, and our wrapped portion doesn't overlap anything at
4338 * the start of the log. We can break
4344 flush_old_journal_lists(sb);
4346 journal->j_current_jl->j_list_bitmap =
4347 get_list_bitmap(sb, journal->j_current_jl);
4349 if (!(journal->j_current_jl->j_list_bitmap)) {
4350 reiserfs_panic(sb, "journal-1996",
4351 "could not get a list bitmap");
4354 atomic_set(&(journal->j_jlock), 0);
4356 /* wake up any body waiting to join. */
4357 clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4358 wake_up(&(journal->j_join_wait));
4360 if (!flush && wait_on_commit &&
4361 journal_list_still_alive(sb, commit_trans_id)) {
4362 flush_commit_list(sb, jl, 1);
4365 reiserfs_check_lock_depth(sb, "journal end2");
4367 memset(th, 0, sizeof(*th));
4368 /* Re-set th->t_super, so we can properly keep track of how many
4369 * persistent transactions there are. We need to do this so if this
4370 * call is part of a failed restart_transaction, we can free it later */
4373 return journal->j_errno;
4376 /* Send the file system read only and refuse new transactions */
4377 void reiserfs_abort_journal(struct super_block *sb, int errno)
4379 struct reiserfs_journal *journal = SB_JOURNAL(sb);
4380 if (test_bit(J_ABORTED, &journal->j_state))
4383 if (!journal->j_errno)
4384 journal->j_errno = errno;
4386 sb->s_flags |= MS_RDONLY;
4387 set_bit(J_ABORTED, &journal->j_state);
4389 #ifdef CONFIG_REISERFS_CHECK