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 *p_s_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 *p_s_sb,
117 unsigned long nblocks, int join);
119 static void init_journal_hash(struct super_block *p_s_sb)
121 struct reiserfs_journal *journal = SB_JOURNAL(p_s_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(p_s_sb->s_blocksize, GFP_NOFS);
163 INIT_LIST_HEAD(&bn->list);
167 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *p_s_sb)
169 struct reiserfs_journal *journal = SB_JOURNAL(p_s_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, p_s_sb->s_blocksize);
180 journal->j_free_bitmap_nodes--;
183 bn = allocate_bitmap_node(p_s_sb);
190 static inline void free_bitmap_node(struct super_block *p_s_sb,
191 struct reiserfs_bitmap_node *bn)
193 struct reiserfs_journal *journal = SB_JOURNAL(p_s_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 *p_s_sb)
207 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
208 struct reiserfs_bitmap_node *bn = NULL;
209 for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
210 bn = allocate_bitmap_node(p_s_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 *p_s_sb,
222 struct reiserfs_list_bitmap *jb)
224 unsigned int bmap_nr = block / (p_s_sb->s_blocksize << 3);
225 unsigned int bit_nr = block % (p_s_sb->s_blocksize << 3);
227 if (!jb->bitmaps[bmap_nr]) {
228 jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb);
230 set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
234 static void cleanup_bitmap_list(struct super_block *p_s_sb,
235 struct reiserfs_list_bitmap *jb)
238 if (jb->bitmaps == NULL)
241 for (i = 0; i < reiserfs_bmap_count(p_s_sb); i++) {
242 if (jb->bitmaps[i]) {
243 free_bitmap_node(p_s_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 *p_s_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(p_s_sb, jb);
267 static int free_bitmap_nodes(struct super_block *p_s_sb)
269 struct reiserfs_journal *journal = SB_JOURNAL(p_s_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 *p_s_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(p_s_sb, "clm-2000", "unable to "
304 "allocate bitmaps for journal lists");
308 memset(jb->bitmaps, 0, mem);
311 free_list_bitmaps(p_s_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 *p_s_sb,
322 struct reiserfs_journal_list
326 struct reiserfs_journal *journal = SB_JOURNAL(p_s_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(p_s_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 *p_s_sb)
383 struct reiserfs_journal_cnode *cn;
384 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
386 reiserfs_check_lock_depth(p_s_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 *p_s_sb,
409 struct reiserfs_journal_cnode *cn)
411 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
413 reiserfs_check_lock_depth(p_s_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 /* utility function to force a BUG if it is called without the big
433 ** kernel lock held. caller is the string printed just before calling BUG()
435 void reiserfs_check_lock_depth(struct super_block *sb, char *caller)
438 if (current->lock_depth < 0) {
439 reiserfs_panic(sb, "%s called without kernel lock held",
447 /* return a cnode with same dev, block number and size in table, or null if not found */
448 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
452 reiserfs_journal_cnode
456 struct reiserfs_journal_cnode *cn;
457 cn = journal_hash(table, sb, bl);
459 if (cn->blocknr == bl && cn->sb == sb)
463 return (struct reiserfs_journal_cnode *)0;
467 ** this actually means 'can this block be reallocated yet?'. If you set search_all, a block can only be allocated
468 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
469 ** being overwritten by a replay after crashing.
471 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction. Since deleting
472 ** a block removes it from the current transaction, this case should never happen. If you don't set search_all, make
473 ** sure you never write the block without logging it.
475 ** next_zero_bit is a suggestion about the next block to try for find_forward.
476 ** when bl is rejected because it is set in a journal list bitmap, we search
477 ** for the next zero bit in the bitmap that rejected bl. Then, we return that
478 ** through next_zero_bit for find_forward to try.
480 ** Just because we return something in next_zero_bit does not mean we won't
481 ** reject it on the next call to reiserfs_in_journal
484 int reiserfs_in_journal(struct super_block *p_s_sb,
485 unsigned int bmap_nr, int bit_nr, int search_all,
486 b_blocknr_t * next_zero_bit)
488 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
489 struct reiserfs_journal_cnode *cn;
490 struct reiserfs_list_bitmap *jb;
494 *next_zero_bit = 0; /* always start this at zero. */
496 PROC_INFO_INC(p_s_sb, journal.in_journal);
497 /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
498 ** if we crash before the transaction that freed it commits, this transaction won't
499 ** have committed either, and the block will never be written
502 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
503 PROC_INFO_INC(p_s_sb, journal.in_journal_bitmap);
504 jb = journal->j_list_bitmap + i;
505 if (jb->journal_list && jb->bitmaps[bmap_nr] &&
507 (unsigned long *)jb->bitmaps[bmap_nr]->
510 find_next_zero_bit((unsigned long *)
511 (jb->bitmaps[bmap_nr]->
513 p_s_sb->s_blocksize << 3,
520 bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr;
521 /* is it in any old transactions? */
524 get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, bl))) {
528 /* is it in the current transaction. This should never happen */
529 if ((cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, bl))) {
534 PROC_INFO_INC(p_s_sb, journal.in_journal_reusable);
539 /* insert cn into table
541 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
542 struct reiserfs_journal_cnode *cn)
544 struct reiserfs_journal_cnode *cn_orig;
546 cn_orig = journal_hash(table, cn->sb, cn->blocknr);
552 journal_hash(table, cn->sb, cn->blocknr) = cn;
555 /* lock the current transaction */
556 static inline void lock_journal(struct super_block *p_s_sb)
558 PROC_INFO_INC(p_s_sb, journal.lock_journal);
559 mutex_lock(&SB_JOURNAL(p_s_sb)->j_mutex);
562 /* unlock the current transaction */
563 static inline void unlock_journal(struct super_block *p_s_sb)
565 mutex_unlock(&SB_JOURNAL(p_s_sb)->j_mutex);
568 static inline void get_journal_list(struct reiserfs_journal_list *jl)
573 static inline void put_journal_list(struct super_block *s,
574 struct reiserfs_journal_list *jl)
576 if (jl->j_refcount < 1) {
577 reiserfs_panic(s, "trans id %u, refcount at %d",
578 jl->j_trans_id, jl->j_refcount);
580 if (--jl->j_refcount == 0)
585 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
586 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
589 static void cleanup_freed_for_journal_list(struct super_block *p_s_sb,
590 struct reiserfs_journal_list *jl)
593 struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
595 cleanup_bitmap_list(p_s_sb, jb);
597 jl->j_list_bitmap->journal_list = NULL;
598 jl->j_list_bitmap = NULL;
601 static int journal_list_still_alive(struct super_block *s,
602 unsigned int trans_id)
604 struct reiserfs_journal *journal = SB_JOURNAL(s);
605 struct list_head *entry = &journal->j_journal_list;
606 struct reiserfs_journal_list *jl;
608 if (!list_empty(entry)) {
609 jl = JOURNAL_LIST_ENTRY(entry->next);
610 if (jl->j_trans_id <= trans_id) {
618 * If page->mapping was null, we failed to truncate this page for
619 * some reason. Most likely because it was truncated after being
620 * logged via data=journal.
622 * This does a check to see if the buffer belongs to one of these
623 * lost pages before doing the final put_bh. If page->mapping was
624 * null, it tries to free buffers on the page, which should make the
625 * final page_cache_release drop the page from the lru.
627 static void release_buffer_page(struct buffer_head *bh)
629 struct page *page = bh->b_page;
630 if (!page->mapping && trylock_page(page)) {
631 page_cache_get(page);
634 try_to_free_buffers(page);
636 page_cache_release(page);
642 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
644 char b[BDEVNAME_SIZE];
646 if (buffer_journaled(bh)) {
647 reiserfs_warning(NULL, "clm-2084",
648 "pinned buffer %lu:%s sent to disk",
649 bh->b_blocknr, bdevname(bh->b_bdev, b));
652 set_buffer_uptodate(bh);
654 clear_buffer_uptodate(bh);
657 release_buffer_page(bh);
660 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
663 set_buffer_uptodate(bh);
665 clear_buffer_uptodate(bh);
670 static void submit_logged_buffer(struct buffer_head *bh)
673 bh->b_end_io = reiserfs_end_buffer_io_sync;
674 clear_buffer_journal_new(bh);
675 clear_buffer_dirty(bh);
676 if (!test_clear_buffer_journal_test(bh))
678 if (!buffer_uptodate(bh))
680 submit_bh(WRITE, bh);
683 static void submit_ordered_buffer(struct buffer_head *bh)
686 bh->b_end_io = reiserfs_end_ordered_io;
687 clear_buffer_dirty(bh);
688 if (!buffer_uptodate(bh))
690 submit_bh(WRITE, bh);
693 static int submit_barrier_buffer(struct buffer_head *bh)
696 bh->b_end_io = reiserfs_end_ordered_io;
697 clear_buffer_dirty(bh);
698 if (!buffer_uptodate(bh))
700 return submit_bh(WRITE_BARRIER, bh);
703 static void check_barrier_completion(struct super_block *s,
704 struct buffer_head *bh)
706 if (buffer_eopnotsupp(bh)) {
707 clear_buffer_eopnotsupp(bh);
709 set_buffer_uptodate(bh);
710 set_buffer_dirty(bh);
711 sync_dirty_buffer(bh);
715 #define CHUNK_SIZE 32
716 struct buffer_chunk {
717 struct buffer_head *bh[CHUNK_SIZE];
721 static void write_chunk(struct buffer_chunk *chunk)
725 for (i = 0; i < chunk->nr; i++) {
726 submit_logged_buffer(chunk->bh[i]);
732 static void write_ordered_chunk(struct buffer_chunk *chunk)
736 for (i = 0; i < chunk->nr; i++) {
737 submit_ordered_buffer(chunk->bh[i]);
743 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
744 spinlock_t * lock, void (fn) (struct buffer_chunk *))
747 BUG_ON(chunk->nr >= CHUNK_SIZE);
748 chunk->bh[chunk->nr++] = bh;
749 if (chunk->nr >= CHUNK_SIZE) {
760 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
761 static struct reiserfs_jh *alloc_jh(void)
763 struct reiserfs_jh *jh;
765 jh = kmalloc(sizeof(*jh), GFP_NOFS);
767 atomic_inc(&nr_reiserfs_jh);
775 * we want to free the jh when the buffer has been written
778 void reiserfs_free_jh(struct buffer_head *bh)
780 struct reiserfs_jh *jh;
784 bh->b_private = NULL;
786 list_del_init(&jh->list);
788 if (atomic_read(&nr_reiserfs_jh) <= 0)
790 atomic_dec(&nr_reiserfs_jh);
795 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
798 struct reiserfs_jh *jh;
801 spin_lock(&j->j_dirty_buffers_lock);
802 if (!bh->b_private) {
803 spin_unlock(&j->j_dirty_buffers_lock);
807 list_del_init(&jh->list);
812 spin_lock(&j->j_dirty_buffers_lock);
813 /* buffer must be locked for __add_jh, should be able to have
814 * two adds at the same time
816 BUG_ON(bh->b_private);
820 jh->jl = j->j_current_jl;
822 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
824 list_add_tail(&jh->list, &jh->jl->j_bh_list);
826 spin_unlock(&j->j_dirty_buffers_lock);
830 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
832 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
834 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
836 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
839 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
840 static int write_ordered_buffers(spinlock_t * lock,
841 struct reiserfs_journal *j,
842 struct reiserfs_journal_list *jl,
843 struct list_head *list)
845 struct buffer_head *bh;
846 struct reiserfs_jh *jh;
847 int ret = j->j_errno;
848 struct buffer_chunk chunk;
849 struct list_head tmp;
850 INIT_LIST_HEAD(&tmp);
854 while (!list_empty(list)) {
855 jh = JH_ENTRY(list->next);
858 if (!trylock_buffer(bh)) {
859 if (!buffer_dirty(bh)) {
860 list_move(&jh->list, &tmp);
865 write_ordered_chunk(&chunk);
871 /* in theory, dirty non-uptodate buffers should never get here,
872 * but the upper layer io error paths still have a few quirks.
873 * Handle them here as gracefully as we can
875 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
876 clear_buffer_dirty(bh);
879 if (buffer_dirty(bh)) {
880 list_move(&jh->list, &tmp);
881 add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
883 reiserfs_free_jh(bh);
888 cond_resched_lock(lock);
892 write_ordered_chunk(&chunk);
895 while (!list_empty(&tmp)) {
896 jh = JH_ENTRY(tmp.prev);
899 reiserfs_free_jh(bh);
901 if (buffer_locked(bh)) {
906 if (!buffer_uptodate(bh)) {
909 /* ugly interaction with invalidatepage here.
910 * reiserfs_invalidate_page will pin any buffer that has a valid
911 * journal head from an older transaction. If someone else sets
912 * our buffer dirty after we write it in the first loop, and
913 * then someone truncates the page away, nobody will ever write
914 * the buffer. We're safe if we write the page one last time
915 * after freeing the journal header.
917 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
919 ll_rw_block(WRITE, 1, &bh);
923 cond_resched_lock(lock);
929 static int flush_older_commits(struct super_block *s,
930 struct reiserfs_journal_list *jl)
932 struct reiserfs_journal *journal = SB_JOURNAL(s);
933 struct reiserfs_journal_list *other_jl;
934 struct reiserfs_journal_list *first_jl;
935 struct list_head *entry;
936 unsigned int trans_id = jl->j_trans_id;
937 unsigned int other_trans_id;
938 unsigned int first_trans_id;
942 * first we walk backwards to find the oldest uncommitted transation
945 entry = jl->j_list.prev;
947 other_jl = JOURNAL_LIST_ENTRY(entry);
948 if (entry == &journal->j_journal_list ||
949 atomic_read(&other_jl->j_older_commits_done))
953 entry = other_jl->j_list.prev;
956 /* if we didn't find any older uncommitted transactions, return now */
957 if (first_jl == jl) {
961 first_trans_id = first_jl->j_trans_id;
963 entry = &first_jl->j_list;
965 other_jl = JOURNAL_LIST_ENTRY(entry);
966 other_trans_id = other_jl->j_trans_id;
968 if (other_trans_id < trans_id) {
969 if (atomic_read(&other_jl->j_commit_left) != 0) {
970 flush_commit_list(s, other_jl, 0);
972 /* list we were called with is gone, return */
973 if (!journal_list_still_alive(s, trans_id))
976 /* the one we just flushed is gone, this means all
977 * older lists are also gone, so first_jl is no longer
978 * valid either. Go back to the beginning.
980 if (!journal_list_still_alive
981 (s, other_trans_id)) {
986 if (entry == &journal->j_journal_list)
995 static int reiserfs_async_progress_wait(struct super_block *s)
998 struct reiserfs_journal *j = SB_JOURNAL(s);
999 if (atomic_read(&j->j_async_throttle))
1000 congestion_wait(WRITE, HZ / 10);
1005 ** if this journal list still has commit blocks unflushed, send them to disk.
1007 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
1008 ** Before the commit block can by written, every other log block must be safely on disk
1011 static int flush_commit_list(struct super_block *s,
1012 struct reiserfs_journal_list *jl, int flushall)
1016 struct buffer_head *tbh = NULL;
1017 unsigned int trans_id = jl->j_trans_id;
1018 struct reiserfs_journal *journal = SB_JOURNAL(s);
1023 reiserfs_check_lock_depth(s, "flush_commit_list");
1025 if (atomic_read(&jl->j_older_commits_done)) {
1031 /* before we can put our commit blocks on disk, we have to make sure everyone older than
1032 ** us is on disk too
1034 BUG_ON(jl->j_len <= 0);
1035 BUG_ON(trans_id == journal->j_trans_id);
1037 get_journal_list(jl);
1039 if (flush_older_commits(s, jl) == 1) {
1040 /* list disappeared during flush_older_commits. return */
1045 /* make sure nobody is trying to flush this one at the same time */
1046 mutex_lock(&jl->j_commit_mutex);
1047 if (!journal_list_still_alive(s, trans_id)) {
1048 mutex_unlock(&jl->j_commit_mutex);
1051 BUG_ON(jl->j_trans_id == 0);
1053 /* this commit is done, exit */
1054 if (atomic_read(&(jl->j_commit_left)) <= 0) {
1056 atomic_set(&(jl->j_older_commits_done), 1);
1058 mutex_unlock(&jl->j_commit_mutex);
1062 if (!list_empty(&jl->j_bh_list)) {
1065 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1066 journal, jl, &jl->j_bh_list);
1067 if (ret < 0 && retval == 0)
1071 BUG_ON(!list_empty(&jl->j_bh_list));
1073 * for the description block and all the log blocks, submit any buffers
1074 * that haven't already reached the disk. Try to write at least 256
1075 * log blocks. later on, we will only wait on blocks that correspond
1076 * to this transaction, but while we're unplugging we might as well
1077 * get a chunk of data on there.
1079 atomic_inc(&journal->j_async_throttle);
1080 write_len = jl->j_len + 1;
1081 if (write_len < 256)
1083 for (i = 0 ; i < write_len ; i++) {
1084 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1085 SB_ONDISK_JOURNAL_SIZE(s);
1086 tbh = journal_find_get_block(s, bn);
1088 if (buffer_dirty(tbh))
1089 ll_rw_block(WRITE, 1, &tbh) ;
1093 atomic_dec(&journal->j_async_throttle);
1095 /* We're skipping the commit if there's an error */
1096 if (retval || reiserfs_is_journal_aborted(journal))
1099 /* wait on everything written so far before writing the commit
1100 * if we are in barrier mode, send the commit down now
1102 barrier = reiserfs_barrier_flush(s);
1105 lock_buffer(jl->j_commit_bh);
1106 ret = submit_barrier_buffer(jl->j_commit_bh);
1107 if (ret == -EOPNOTSUPP) {
1108 set_buffer_uptodate(jl->j_commit_bh);
1113 for (i = 0; i < (jl->j_len + 1); i++) {
1114 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1115 (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1116 tbh = journal_find_get_block(s, bn);
1117 wait_on_buffer(tbh);
1118 // since we're using ll_rw_blk above, it might have skipped over
1119 // a locked buffer. Double check here
1121 if (buffer_dirty(tbh)) /* redundant, sync_dirty_buffer() checks */
1122 sync_dirty_buffer(tbh);
1123 if (unlikely(!buffer_uptodate(tbh))) {
1124 #ifdef CONFIG_REISERFS_CHECK
1125 reiserfs_warning(s, "journal-601",
1126 "buffer write failed");
1130 put_bh(tbh); /* once for journal_find_get_block */
1131 put_bh(tbh); /* once due to original getblk in do_journal_end */
1132 atomic_dec(&(jl->j_commit_left));
1135 BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1138 /* If there was a write error in the journal - we can't commit
1139 * this transaction - it will be invalid and, if successful,
1140 * will just end up propagating the write error out to
1141 * the file system. */
1142 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1143 if (buffer_dirty(jl->j_commit_bh))
1145 mark_buffer_dirty(jl->j_commit_bh) ;
1146 sync_dirty_buffer(jl->j_commit_bh) ;
1149 wait_on_buffer(jl->j_commit_bh);
1151 check_barrier_completion(s, jl->j_commit_bh);
1153 /* If there was a write error in the journal - we can't commit this
1154 * transaction - it will be invalid and, if successful, will just end
1155 * up propagating the write error out to the filesystem. */
1156 if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1157 #ifdef CONFIG_REISERFS_CHECK
1158 reiserfs_warning(s, "journal-615", "buffer write failed");
1162 bforget(jl->j_commit_bh);
1163 if (journal->j_last_commit_id != 0 &&
1164 (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1165 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1166 journal->j_last_commit_id, jl->j_trans_id);
1168 journal->j_last_commit_id = jl->j_trans_id;
1170 /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */
1171 cleanup_freed_for_journal_list(s, jl);
1173 retval = retval ? retval : journal->j_errno;
1175 /* mark the metadata dirty */
1177 dirty_one_transaction(s, jl);
1178 atomic_dec(&(jl->j_commit_left));
1181 atomic_set(&(jl->j_older_commits_done), 1);
1183 mutex_unlock(&jl->j_commit_mutex);
1185 put_journal_list(s, jl);
1188 reiserfs_abort(s, retval, "Journal write error in %s",
1195 ** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or
1196 ** returns NULL if it can't find anything
1198 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1199 reiserfs_journal_cnode
1202 struct super_block *sb = cn->sb;
1203 b_blocknr_t blocknr = cn->blocknr;
1207 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1215 static int newer_jl_done(struct reiserfs_journal_cnode *cn)
1217 struct super_block *sb = cn->sb;
1218 b_blocknr_t blocknr = cn->blocknr;
1222 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist &&
1223 atomic_read(&cn->jlist->j_commit_left) != 0)
1230 static void remove_journal_hash(struct super_block *,
1231 struct reiserfs_journal_cnode **,
1232 struct reiserfs_journal_list *, unsigned long,
1236 ** once all the real blocks have been flushed, it is safe to remove them from the
1237 ** journal list for this transaction. Aside from freeing the cnode, this also allows the
1238 ** block to be reallocated for data blocks if it had been deleted.
1240 static void remove_all_from_journal_list(struct super_block *p_s_sb,
1241 struct reiserfs_journal_list *jl,
1244 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1245 struct reiserfs_journal_cnode *cn, *last;
1246 cn = jl->j_realblock;
1248 /* which is better, to lock once around the whole loop, or
1249 ** to lock for each call to remove_journal_hash?
1252 if (cn->blocknr != 0) {
1254 reiserfs_warning(p_s_sb, "reiserfs-2201",
1255 "block %u, bh is %d, state %ld",
1256 cn->blocknr, cn->bh ? 1 : 0,
1260 remove_journal_hash(p_s_sb, journal->j_list_hash_table,
1261 jl, cn->blocknr, 1);
1265 free_cnode(p_s_sb, last);
1267 jl->j_realblock = NULL;
1271 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1272 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1273 ** releasing blocks in this transaction for reuse as data blocks.
1274 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1277 static int _update_journal_header_block(struct super_block *p_s_sb,
1278 unsigned long offset,
1279 unsigned int trans_id)
1281 struct reiserfs_journal_header *jh;
1282 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1284 if (reiserfs_is_journal_aborted(journal))
1287 if (trans_id >= journal->j_last_flush_trans_id) {
1288 if (buffer_locked((journal->j_header_bh))) {
1289 wait_on_buffer((journal->j_header_bh));
1290 if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1291 #ifdef CONFIG_REISERFS_CHECK
1292 reiserfs_warning(p_s_sb, "journal-699",
1293 "buffer write failed");
1298 journal->j_last_flush_trans_id = trans_id;
1299 journal->j_first_unflushed_offset = offset;
1300 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1302 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1303 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1304 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1306 if (reiserfs_barrier_flush(p_s_sb)) {
1308 lock_buffer(journal->j_header_bh);
1309 ret = submit_barrier_buffer(journal->j_header_bh);
1310 if (ret == -EOPNOTSUPP) {
1311 set_buffer_uptodate(journal->j_header_bh);
1312 disable_barrier(p_s_sb);
1315 wait_on_buffer(journal->j_header_bh);
1316 check_barrier_completion(p_s_sb, journal->j_header_bh);
1319 set_buffer_dirty(journal->j_header_bh);
1320 sync_dirty_buffer(journal->j_header_bh);
1322 if (!buffer_uptodate(journal->j_header_bh)) {
1323 reiserfs_warning(p_s_sb, "journal-837",
1324 "IO error during journal replay");
1331 static int update_journal_header_block(struct super_block *p_s_sb,
1332 unsigned long offset,
1333 unsigned int trans_id)
1335 return _update_journal_header_block(p_s_sb, offset, trans_id);
1339 ** flush any and all journal lists older than you are
1340 ** can only be called from flush_journal_list
1342 static int flush_older_journal_lists(struct super_block *p_s_sb,
1343 struct reiserfs_journal_list *jl)
1345 struct list_head *entry;
1346 struct reiserfs_journal_list *other_jl;
1347 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1348 unsigned int trans_id = jl->j_trans_id;
1350 /* we know we are the only ones flushing things, no extra race
1351 * protection is required.
1354 entry = journal->j_journal_list.next;
1356 if (entry == &journal->j_journal_list)
1358 other_jl = JOURNAL_LIST_ENTRY(entry);
1359 if (other_jl->j_trans_id < trans_id) {
1360 BUG_ON(other_jl->j_refcount <= 0);
1361 /* do not flush all */
1362 flush_journal_list(p_s_sb, other_jl, 0);
1364 /* other_jl is now deleted from the list */
1370 static void del_from_work_list(struct super_block *s,
1371 struct reiserfs_journal_list *jl)
1373 struct reiserfs_journal *journal = SB_JOURNAL(s);
1374 if (!list_empty(&jl->j_working_list)) {
1375 list_del_init(&jl->j_working_list);
1376 journal->j_num_work_lists--;
1380 /* flush a journal list, both commit and real blocks
1382 ** always set flushall to 1, unless you are calling from inside
1383 ** flush_journal_list
1385 ** IMPORTANT. This can only be called while there are no journal writers,
1386 ** and the journal is locked. That means it can only be called from
1387 ** do_journal_end, or by journal_release
1389 static int flush_journal_list(struct super_block *s,
1390 struct reiserfs_journal_list *jl, int flushall)
1392 struct reiserfs_journal_list *pjl;
1393 struct reiserfs_journal_cnode *cn, *last;
1397 struct buffer_head *saved_bh;
1398 unsigned long j_len_saved = jl->j_len;
1399 struct reiserfs_journal *journal = SB_JOURNAL(s);
1402 BUG_ON(j_len_saved <= 0);
1404 if (atomic_read(&journal->j_wcount) != 0) {
1405 reiserfs_warning(s, "clm-2048", "called with wcount %d",
1406 atomic_read(&journal->j_wcount));
1408 BUG_ON(jl->j_trans_id == 0);
1410 /* if flushall == 0, the lock is already held */
1412 mutex_lock(&journal->j_flush_mutex);
1413 } else if (mutex_trylock(&journal->j_flush_mutex)) {
1418 if (j_len_saved > journal->j_trans_max) {
1420 "journal-715: flush_journal_list, length is %lu, trans id %lu\n",
1421 j_len_saved, jl->j_trans_id);
1427 /* if all the work is already done, get out of here */
1428 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1429 atomic_read(&(jl->j_commit_left)) <= 0) {
1430 goto flush_older_and_return;
1433 /* start by putting the commit list on disk. This will also flush
1434 ** the commit lists of any olders transactions
1436 flush_commit_list(s, jl, 1);
1438 if (!(jl->j_state & LIST_DIRTY)
1439 && !reiserfs_is_journal_aborted(journal))
1442 /* are we done now? */
1443 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1444 atomic_read(&(jl->j_commit_left)) <= 0) {
1445 goto flush_older_and_return;
1448 /* loop through each cnode, see if we need to write it,
1449 ** or wait on a more recent transaction, or just ignore it
1451 if (atomic_read(&(journal->j_wcount)) != 0) {
1453 "journal-844: panic journal list is flushing, wcount is not 0\n");
1455 cn = jl->j_realblock;
1460 /* blocknr of 0 is no longer in the hash, ignore it */
1461 if (cn->blocknr == 0) {
1465 /* This transaction failed commit. Don't write out to the disk */
1466 if (!(jl->j_state & LIST_DIRTY))
1469 pjl = find_newer_jl_for_cn(cn);
1470 /* the order is important here. We check pjl to make sure we
1471 ** don't clear BH_JDirty_wait if we aren't the one writing this
1474 if (!pjl && cn->bh) {
1477 /* we do this to make sure nobody releases the buffer while
1478 ** we are working with it
1482 if (buffer_journal_dirty(saved_bh)) {
1483 BUG_ON(!can_dirty(cn));
1486 } else if (can_dirty(cn)) {
1487 /* everything with !pjl && jwait should be writable */
1492 /* if someone has this block in a newer transaction, just make
1493 ** sure they are committed, and don't try writing it to disk
1496 if (atomic_read(&pjl->j_commit_left))
1497 flush_commit_list(s, pjl, 1);
1501 /* bh == NULL when the block got to disk on its own, OR,
1502 ** the block got freed in a future transaction
1504 if (saved_bh == NULL) {
1508 /* this should never happen. kupdate_one_transaction has this list
1509 ** locked while it works, so we should never see a buffer here that
1510 ** is not marked JDirty_wait
1512 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1513 reiserfs_warning(s, "journal-813",
1514 "BAD! buffer %llu %cdirty %cjwait, "
1515 "not in a newer tranasction",
1516 (unsigned long long)saved_bh->
1517 b_blocknr, was_dirty ? ' ' : '!',
1518 was_jwait ? ' ' : '!');
1521 /* we inc again because saved_bh gets decremented at free_cnode */
1523 set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1524 lock_buffer(saved_bh);
1525 BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1526 if (buffer_dirty(saved_bh))
1527 submit_logged_buffer(saved_bh);
1529 unlock_buffer(saved_bh);
1532 reiserfs_warning(s, "clm-2082",
1533 "Unable to flush buffer %llu in %s",
1534 (unsigned long long)saved_bh->
1535 b_blocknr, __func__);
1541 /* we incremented this to keep others from taking the buffer head away */
1543 if (atomic_read(&(saved_bh->b_count)) < 0) {
1544 reiserfs_warning(s, "journal-945",
1545 "saved_bh->b_count < 0");
1550 cn = jl->j_realblock;
1552 if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1555 "journal-1011: cn->bh is NULL\n");
1557 wait_on_buffer(cn->bh);
1560 "journal-1012: cn->bh is NULL\n");
1562 if (unlikely(!buffer_uptodate(cn->bh))) {
1563 #ifdef CONFIG_REISERFS_CHECK
1564 reiserfs_warning(s, "journal-949",
1565 "buffer write failed");
1569 /* note, we must clear the JDirty_wait bit after the up to date
1570 ** check, otherwise we race against our flushpage routine
1572 BUG_ON(!test_clear_buffer_journal_dirty
1575 /* drop one ref for us */
1577 /* drop one ref for journal_mark_dirty */
1578 release_buffer_page(cn->bh);
1585 reiserfs_abort(s, -EIO,
1586 "Write error while pushing transaction to disk in %s",
1588 flush_older_and_return:
1590 /* before we can update the journal header block, we _must_ flush all
1591 ** real blocks from all older transactions to disk. This is because
1592 ** once the header block is updated, this transaction will not be
1593 ** replayed after a crash
1596 flush_older_journal_lists(s, jl);
1599 err = journal->j_errno;
1600 /* before we can remove everything from the hash tables for this
1601 ** transaction, we must make sure it can never be replayed
1603 ** since we are only called from do_journal_end, we know for sure there
1604 ** are no allocations going on while we are flushing journal lists. So,
1605 ** we only need to update the journal header block for the last list
1608 if (!err && flushall) {
1610 update_journal_header_block(s,
1611 (jl->j_start + jl->j_len +
1612 2) % SB_ONDISK_JOURNAL_SIZE(s),
1615 reiserfs_abort(s, -EIO,
1616 "Write error while updating journal header in %s",
1619 remove_all_from_journal_list(s, jl, 0);
1620 list_del_init(&jl->j_list);
1621 journal->j_num_lists--;
1622 del_from_work_list(s, jl);
1624 if (journal->j_last_flush_id != 0 &&
1625 (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1626 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1627 journal->j_last_flush_id, jl->j_trans_id);
1629 journal->j_last_flush_id = jl->j_trans_id;
1631 /* not strictly required since we are freeing the list, but it should
1632 * help find code using dead lists later on
1635 atomic_set(&(jl->j_nonzerolen), 0);
1637 jl->j_realblock = NULL;
1638 jl->j_commit_bh = NULL;
1641 put_journal_list(s, jl);
1643 mutex_unlock(&journal->j_flush_mutex);
1648 static int test_transaction(struct super_block *s,
1649 struct reiserfs_journal_list *jl)
1651 struct reiserfs_journal_cnode *cn;
1653 if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0)
1656 cn = jl->j_realblock;
1658 /* if the blocknr == 0, this has been cleared from the hash,
1661 if (cn->blocknr == 0) {
1664 if (cn->bh && !newer_jl_done(cn))
1673 static int write_one_transaction(struct super_block *s,
1674 struct reiserfs_journal_list *jl,
1675 struct buffer_chunk *chunk)
1677 struct reiserfs_journal_cnode *cn;
1680 jl->j_state |= LIST_TOUCHED;
1681 del_from_work_list(s, jl);
1682 if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1686 cn = jl->j_realblock;
1688 /* if the blocknr == 0, this has been cleared from the hash,
1691 if (cn->blocknr == 0) {
1694 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1695 struct buffer_head *tmp_bh;
1696 /* we can race against journal_mark_freed when we try
1697 * to lock_buffer(cn->bh), so we have to inc the buffer
1698 * count, and recheck things after locking
1702 lock_buffer(tmp_bh);
1703 if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1704 if (!buffer_journal_dirty(tmp_bh) ||
1705 buffer_journal_prepared(tmp_bh))
1707 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1710 /* note, cn->bh might be null now */
1711 unlock_buffer(tmp_bh);
1722 /* used by flush_commit_list */
1723 static int dirty_one_transaction(struct super_block *s,
1724 struct reiserfs_journal_list *jl)
1726 struct reiserfs_journal_cnode *cn;
1727 struct reiserfs_journal_list *pjl;
1730 jl->j_state |= LIST_DIRTY;
1731 cn = jl->j_realblock;
1733 /* look for a more recent transaction that logged this
1734 ** buffer. Only the most recent transaction with a buffer in
1735 ** it is allowed to send that buffer to disk
1737 pjl = find_newer_jl_for_cn(cn);
1738 if (!pjl && cn->blocknr && cn->bh
1739 && buffer_journal_dirty(cn->bh)) {
1740 BUG_ON(!can_dirty(cn));
1741 /* if the buffer is prepared, it will either be logged
1742 * or restored. If restored, we need to make sure
1743 * it actually gets marked dirty
1745 clear_buffer_journal_new(cn->bh);
1746 if (buffer_journal_prepared(cn->bh)) {
1747 set_buffer_journal_restore_dirty(cn->bh);
1749 set_buffer_journal_test(cn->bh);
1750 mark_buffer_dirty(cn->bh);
1758 static int kupdate_transactions(struct super_block *s,
1759 struct reiserfs_journal_list *jl,
1760 struct reiserfs_journal_list **next_jl,
1761 unsigned int *next_trans_id,
1762 int num_blocks, int num_trans)
1766 int transactions_flushed = 0;
1767 unsigned int orig_trans_id = jl->j_trans_id;
1768 struct buffer_chunk chunk;
1769 struct list_head *entry;
1770 struct reiserfs_journal *journal = SB_JOURNAL(s);
1773 mutex_lock(&journal->j_flush_mutex);
1774 if (!journal_list_still_alive(s, orig_trans_id)) {
1778 /* we've got j_flush_mutex held, nobody is going to delete any
1779 * of these lists out from underneath us
1781 while ((num_trans && transactions_flushed < num_trans) ||
1782 (!num_trans && written < num_blocks)) {
1784 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1785 atomic_read(&jl->j_commit_left)
1786 || !(jl->j_state & LIST_DIRTY)) {
1787 del_from_work_list(s, jl);
1790 ret = write_one_transaction(s, jl, &chunk);
1794 transactions_flushed++;
1796 entry = jl->j_list.next;
1799 if (entry == &journal->j_journal_list) {
1802 jl = JOURNAL_LIST_ENTRY(entry);
1804 /* don't bother with older transactions */
1805 if (jl->j_trans_id <= orig_trans_id)
1809 write_chunk(&chunk);
1813 mutex_unlock(&journal->j_flush_mutex);
1817 /* for o_sync and fsync heavy applications, they tend to use
1818 ** all the journa list slots with tiny transactions. These
1819 ** trigger lots and lots of calls to update the header block, which
1820 ** adds seeks and slows things down.
1822 ** This function tries to clear out a large chunk of the journal lists
1823 ** at once, which makes everything faster since only the newest journal
1824 ** list updates the header block
1826 static int flush_used_journal_lists(struct super_block *s,
1827 struct reiserfs_journal_list *jl)
1829 unsigned long len = 0;
1830 unsigned long cur_len;
1834 struct reiserfs_journal_list *tjl;
1835 struct reiserfs_journal_list *flush_jl;
1836 unsigned int trans_id;
1837 struct reiserfs_journal *journal = SB_JOURNAL(s);
1839 flush_jl = tjl = jl;
1841 /* in data logging mode, try harder to flush a lot of blocks */
1842 if (reiserfs_data_log(s))
1844 /* flush for 256 transactions or limit blocks, whichever comes first */
1845 for (i = 0; i < 256 && len < limit; i++) {
1846 if (atomic_read(&tjl->j_commit_left) ||
1847 tjl->j_trans_id < jl->j_trans_id) {
1850 cur_len = atomic_read(&tjl->j_nonzerolen);
1852 tjl->j_state &= ~LIST_TOUCHED;
1856 if (tjl->j_list.next == &journal->j_journal_list)
1858 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1860 /* try to find a group of blocks we can flush across all the
1861 ** transactions, but only bother if we've actually spanned
1862 ** across multiple lists
1864 if (flush_jl != jl) {
1865 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1867 flush_journal_list(s, flush_jl, 1);
1872 ** removes any nodes in table with name block and dev as bh.
1873 ** only touchs the hnext and hprev pointers.
1875 void remove_journal_hash(struct super_block *sb,
1876 struct reiserfs_journal_cnode **table,
1877 struct reiserfs_journal_list *jl,
1878 unsigned long block, int remove_freed)
1880 struct reiserfs_journal_cnode *cur;
1881 struct reiserfs_journal_cnode **head;
1883 head = &(journal_hash(table, sb, block));
1889 if (cur->blocknr == block && cur->sb == sb
1890 && (jl == NULL || jl == cur->jlist)
1891 && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1893 cur->hnext->hprev = cur->hprev;
1896 cur->hprev->hnext = cur->hnext;
1903 if (cur->bh && cur->jlist) /* anybody who clears the cur->bh will also dec the nonzerolen */
1904 atomic_dec(&(cur->jlist->j_nonzerolen));
1912 static void free_journal_ram(struct super_block *p_s_sb)
1914 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1915 kfree(journal->j_current_jl);
1916 journal->j_num_lists--;
1918 vfree(journal->j_cnode_free_orig);
1919 free_list_bitmaps(p_s_sb, journal->j_list_bitmap);
1920 free_bitmap_nodes(p_s_sb); /* must be after free_list_bitmaps */
1921 if (journal->j_header_bh) {
1922 brelse(journal->j_header_bh);
1924 /* j_header_bh is on the journal dev, make sure not to release the journal
1925 * dev until we brelse j_header_bh
1927 release_journal_dev(p_s_sb, journal);
1932 ** call on unmount. Only set error to 1 if you haven't made your way out
1933 ** of read_super() yet. Any other caller must keep error at 0.
1935 static int do_journal_release(struct reiserfs_transaction_handle *th,
1936 struct super_block *p_s_sb, int error)
1938 struct reiserfs_transaction_handle myth;
1940 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1942 /* we only want to flush out transactions if we were called with error == 0
1944 if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
1945 /* end the current trans */
1946 BUG_ON(!th->t_trans_id);
1947 do_journal_end(th, p_s_sb, 10, FLUSH_ALL);
1949 /* make sure something gets logged to force our way into the flush code */
1950 if (!journal_join(&myth, p_s_sb, 1)) {
1951 reiserfs_prepare_for_journal(p_s_sb,
1952 SB_BUFFER_WITH_SB(p_s_sb),
1954 journal_mark_dirty(&myth, p_s_sb,
1955 SB_BUFFER_WITH_SB(p_s_sb));
1956 do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1961 /* this also catches errors during the do_journal_end above */
1962 if (!error && reiserfs_is_journal_aborted(journal)) {
1963 memset(&myth, 0, sizeof(myth));
1964 if (!journal_join_abort(&myth, p_s_sb, 1)) {
1965 reiserfs_prepare_for_journal(p_s_sb,
1966 SB_BUFFER_WITH_SB(p_s_sb),
1968 journal_mark_dirty(&myth, p_s_sb,
1969 SB_BUFFER_WITH_SB(p_s_sb));
1970 do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1974 reiserfs_mounted_fs_count--;
1975 /* wait for all commits to finish */
1976 cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work);
1977 flush_workqueue(commit_wq);
1978 if (!reiserfs_mounted_fs_count) {
1979 destroy_workqueue(commit_wq);
1983 free_journal_ram(p_s_sb);
1989 ** call on unmount. flush all journal trans, release all alloc'd ram
1991 int journal_release(struct reiserfs_transaction_handle *th,
1992 struct super_block *p_s_sb)
1994 return do_journal_release(th, p_s_sb, 0);
1998 ** only call from an error condition inside reiserfs_read_super!
2000 int journal_release_error(struct reiserfs_transaction_handle *th,
2001 struct super_block *p_s_sb)
2003 return do_journal_release(th, p_s_sb, 1);
2006 /* compares description block with commit block. returns 1 if they differ, 0 if they are the same */
2007 static int journal_compare_desc_commit(struct super_block *p_s_sb,
2008 struct reiserfs_journal_desc *desc,
2009 struct reiserfs_journal_commit *commit)
2011 if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
2012 get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
2013 get_commit_trans_len(commit) > SB_JOURNAL(p_s_sb)->j_trans_max ||
2014 get_commit_trans_len(commit) <= 0) {
2020 /* returns 0 if it did not find a description block
2021 ** returns -1 if it found a corrupt commit block
2022 ** returns 1 if both desc and commit were valid
2024 static int journal_transaction_is_valid(struct super_block *p_s_sb,
2025 struct buffer_head *d_bh,
2026 unsigned int *oldest_invalid_trans_id,
2027 unsigned long *newest_mount_id)
2029 struct reiserfs_journal_desc *desc;
2030 struct reiserfs_journal_commit *commit;
2031 struct buffer_head *c_bh;
2032 unsigned long offset;
2037 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2038 if (get_desc_trans_len(desc) > 0
2039 && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2040 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2041 && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2042 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2043 "journal-986: transaction "
2044 "is valid returning because trans_id %d is greater than "
2045 "oldest_invalid %lu",
2046 get_desc_trans_id(desc),
2047 *oldest_invalid_trans_id);
2051 && *newest_mount_id > get_desc_mount_id(desc)) {
2052 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2053 "journal-1087: transaction "
2054 "is valid returning because mount_id %d is less than "
2055 "newest_mount_id %lu",
2056 get_desc_mount_id(desc),
2060 if (get_desc_trans_len(desc) > SB_JOURNAL(p_s_sb)->j_trans_max) {
2061 reiserfs_warning(p_s_sb, "journal-2018",
2062 "Bad transaction length %d "
2063 "encountered, ignoring transaction",
2064 get_desc_trans_len(desc));
2067 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2069 /* ok, we have a journal description block, lets see if the transaction was valid */
2071 journal_bread(p_s_sb,
2072 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2073 ((offset + get_desc_trans_len(desc) +
2074 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2077 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2078 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2079 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2080 "journal_transaction_is_valid, commit offset %ld had bad "
2081 "time %d or length %d",
2083 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2084 get_commit_trans_id(commit),
2085 get_commit_trans_len(commit));
2087 if (oldest_invalid_trans_id) {
2088 *oldest_invalid_trans_id =
2089 get_desc_trans_id(desc);
2090 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2092 "transaction_is_valid setting oldest invalid trans_id "
2094 get_desc_trans_id(desc));
2099 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2100 "journal-1006: found valid "
2101 "transaction start offset %llu, len %d id %d",
2103 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2104 get_desc_trans_len(desc),
2105 get_desc_trans_id(desc));
2112 static void brelse_array(struct buffer_head **heads, int num)
2115 for (i = 0; i < num; i++) {
2121 ** given the start, and values for the oldest acceptable transactions,
2122 ** this either reads in a replays a transaction, or returns because the transaction
2123 ** is invalid, or too old.
2125 static int journal_read_transaction(struct super_block *p_s_sb,
2126 unsigned long cur_dblock,
2127 unsigned long oldest_start,
2128 unsigned int oldest_trans_id,
2129 unsigned long newest_mount_id)
2131 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2132 struct reiserfs_journal_desc *desc;
2133 struct reiserfs_journal_commit *commit;
2134 unsigned int trans_id = 0;
2135 struct buffer_head *c_bh;
2136 struct buffer_head *d_bh;
2137 struct buffer_head **log_blocks = NULL;
2138 struct buffer_head **real_blocks = NULL;
2139 unsigned int trans_offset;
2143 d_bh = journal_bread(p_s_sb, cur_dblock);
2146 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2147 trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2148 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
2149 "journal_read_transaction, offset %llu, len %d mount_id %d",
2150 d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2151 get_desc_trans_len(desc), get_desc_mount_id(desc));
2152 if (get_desc_trans_id(desc) < oldest_trans_id) {
2153 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
2154 "journal_read_trans skipping because %lu is too old",
2156 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2160 if (get_desc_mount_id(desc) != newest_mount_id) {
2161 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
2162 "journal_read_trans skipping because %d is != "
2163 "newest_mount_id %lu", get_desc_mount_id(desc),
2168 c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2169 ((trans_offset + get_desc_trans_len(desc) + 1) %
2170 SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2175 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2176 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2177 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2178 "journal_read_transaction, "
2179 "commit offset %llu had bad time %d or length %d",
2181 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2182 get_commit_trans_id(commit),
2183 get_commit_trans_len(commit));
2188 trans_id = get_desc_trans_id(desc);
2189 /* now we know we've got a good transaction, and it was inside the valid time ranges */
2190 log_blocks = kmalloc(get_desc_trans_len(desc) *
2191 sizeof(struct buffer_head *), GFP_NOFS);
2192 real_blocks = kmalloc(get_desc_trans_len(desc) *
2193 sizeof(struct buffer_head *), GFP_NOFS);
2194 if (!log_blocks || !real_blocks) {
2199 reiserfs_warning(p_s_sb, "journal-1169",
2200 "kmalloc failed, unable to mount FS");
2203 /* get all the buffer heads */
2204 trans_half = journal_trans_half(p_s_sb->s_blocksize);
2205 for (i = 0; i < get_desc_trans_len(desc); i++) {
2207 journal_getblk(p_s_sb,
2208 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2210 i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2211 if (i < trans_half) {
2214 le32_to_cpu(desc->j_realblock[i]));
2218 le32_to_cpu(commit->
2219 j_realblock[i - trans_half]));
2221 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb)) {
2222 reiserfs_warning(p_s_sb, "journal-1207",
2223 "REPLAY FAILURE fsck required! "
2224 "Block to replay is outside of "
2228 /* make sure we don't try to replay onto log or reserved area */
2229 if (is_block_in_log_or_reserved_area
2230 (p_s_sb, real_blocks[i]->b_blocknr)) {
2231 reiserfs_warning(p_s_sb, "journal-1204",
2232 "REPLAY FAILURE fsck required! "
2233 "Trying to replay onto a log block");
2235 brelse_array(log_blocks, i);
2236 brelse_array(real_blocks, i);
2244 /* read in the log blocks, memcpy to the corresponding real block */
2245 ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2246 for (i = 0; i < get_desc_trans_len(desc); i++) {
2247 wait_on_buffer(log_blocks[i]);
2248 if (!buffer_uptodate(log_blocks[i])) {
2249 reiserfs_warning(p_s_sb, "journal-1212",
2250 "REPLAY FAILURE fsck required! "
2251 "buffer write failed");
2252 brelse_array(log_blocks + i,
2253 get_desc_trans_len(desc) - i);
2254 brelse_array(real_blocks, get_desc_trans_len(desc));
2261 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2262 real_blocks[i]->b_size);
2263 set_buffer_uptodate(real_blocks[i]);
2264 brelse(log_blocks[i]);
2266 /* flush out the real blocks */
2267 for (i = 0; i < get_desc_trans_len(desc); i++) {
2268 set_buffer_dirty(real_blocks[i]);
2269 ll_rw_block(SWRITE, 1, real_blocks + i);
2271 for (i = 0; i < get_desc_trans_len(desc); i++) {
2272 wait_on_buffer(real_blocks[i]);
2273 if (!buffer_uptodate(real_blocks[i])) {
2274 reiserfs_warning(p_s_sb, "journal-1226",
2275 "REPLAY FAILURE, fsck required! "
2276 "buffer write failed");
2277 brelse_array(real_blocks + i,
2278 get_desc_trans_len(desc) - i);
2285 brelse(real_blocks[i]);
2288 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2289 ((trans_offset + get_desc_trans_len(desc) +
2290 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2291 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2292 "journal-1095: setting journal " "start to offset %ld",
2293 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2295 /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2296 journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2297 journal->j_last_flush_trans_id = trans_id;
2298 journal->j_trans_id = trans_id + 1;
2299 /* check for trans_id overflow */
2300 if (journal->j_trans_id == 0)
2301 journal->j_trans_id = 10;
2309 /* This function reads blocks starting from block and to max_block of bufsize
2310 size (but no more than BUFNR blocks at a time). This proved to improve
2311 mounting speed on self-rebuilding raid5 arrays at least.
2312 Right now it is only used from journal code. But later we might use it
2314 Note: Do not use journal_getblk/sb_getblk functions here! */
2315 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2316 b_blocknr_t block, int bufsize,
2317 b_blocknr_t max_block)
2319 struct buffer_head *bhlist[BUFNR];
2320 unsigned int blocks = BUFNR;
2321 struct buffer_head *bh;
2324 bh = __getblk(dev, block, bufsize);
2325 if (buffer_uptodate(bh))
2328 if (block + BUFNR > max_block) {
2329 blocks = max_block - block;
2333 for (i = 1; i < blocks; i++) {
2334 bh = __getblk(dev, block + i, bufsize);
2335 if (buffer_uptodate(bh)) {
2341 ll_rw_block(READ, j, bhlist);
2342 for (i = 1; i < j; i++)
2346 if (buffer_uptodate(bh))
2353 ** read and replay the log
2354 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2355 ** transaction. This tests that before finding all the transactions in the log, which makes normal mount times fast.
2357 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2359 ** On exit, it sets things up so the first transaction will work correctly.
2361 static int journal_read(struct super_block *p_s_sb)
2363 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2364 struct reiserfs_journal_desc *desc;
2365 unsigned int oldest_trans_id = 0;
2366 unsigned int oldest_invalid_trans_id = 0;
2368 unsigned long oldest_start = 0;
2369 unsigned long cur_dblock = 0;
2370 unsigned long newest_mount_id = 9;
2371 struct buffer_head *d_bh;
2372 struct reiserfs_journal_header *jh;
2373 int valid_journal_header = 0;
2374 int replay_count = 0;
2375 int continue_replay = 1;
2377 char b[BDEVNAME_SIZE];
2379 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2380 reiserfs_info(p_s_sb, "checking transaction log (%s)\n",
2381 bdevname(journal->j_dev_bd, b));
2382 start = get_seconds();
2384 /* step 1, read in the journal header block. Check the transaction it says
2385 ** is the first unflushed, and if that transaction is not valid,
2388 journal->j_header_bh = journal_bread(p_s_sb,
2389 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)
2390 + SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2391 if (!journal->j_header_bh) {
2394 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2395 if (le32_to_cpu(jh->j_first_unflushed_offset) <
2396 SB_ONDISK_JOURNAL_SIZE(p_s_sb)
2397 && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2399 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2400 le32_to_cpu(jh->j_first_unflushed_offset);
2401 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2402 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2403 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2404 "journal-1153: found in "
2405 "header: first_unflushed_offset %d, last_flushed_trans_id "
2406 "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2407 le32_to_cpu(jh->j_last_flush_trans_id));
2408 valid_journal_header = 1;
2410 /* now, we try to read the first unflushed offset. If it is not valid,
2411 ** there is nothing more we can do, and it makes no sense to read
2412 ** through the whole log.
2415 journal_bread(p_s_sb,
2416 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2417 le32_to_cpu(jh->j_first_unflushed_offset));
2418 ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL);
2420 continue_replay = 0;
2423 goto start_log_replay;
2426 if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) {
2427 reiserfs_warning(p_s_sb, "clm-2076",
2428 "device is readonly, unable to replay log");
2432 /* ok, there are transactions that need to be replayed. start with the first log block, find
2433 ** all the valid transactions, and pick out the oldest.
2435 while (continue_replay
2437 (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2438 SB_ONDISK_JOURNAL_SIZE(p_s_sb))) {
2439 /* Note that it is required for blocksize of primary fs device and journal
2440 device to be the same */
2442 reiserfs_breada(journal->j_dev_bd, cur_dblock,
2443 p_s_sb->s_blocksize,
2444 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2445 SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2447 journal_transaction_is_valid(p_s_sb, d_bh,
2448 &oldest_invalid_trans_id,
2451 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2452 if (oldest_start == 0) { /* init all oldest_ values */
2453 oldest_trans_id = get_desc_trans_id(desc);
2454 oldest_start = d_bh->b_blocknr;
2455 newest_mount_id = get_desc_mount_id(desc);
2456 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2457 "journal-1179: Setting "
2458 "oldest_start to offset %llu, trans_id %lu",
2460 SB_ONDISK_JOURNAL_1st_BLOCK
2461 (p_s_sb), oldest_trans_id);
2462 } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2463 /* one we just read was older */
2464 oldest_trans_id = get_desc_trans_id(desc);
2465 oldest_start = d_bh->b_blocknr;
2466 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2467 "journal-1180: Resetting "
2468 "oldest_start to offset %lu, trans_id %lu",
2470 SB_ONDISK_JOURNAL_1st_BLOCK
2471 (p_s_sb), oldest_trans_id);
2473 if (newest_mount_id < get_desc_mount_id(desc)) {
2474 newest_mount_id = get_desc_mount_id(desc);
2475 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2476 "journal-1299: Setting "
2477 "newest_mount_id to %d",
2478 get_desc_mount_id(desc));
2480 cur_dblock += get_desc_trans_len(desc) + 2;
2488 cur_dblock = oldest_start;
2489 if (oldest_trans_id) {
2490 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2491 "journal-1206: Starting replay "
2492 "from offset %llu, trans_id %lu",
2493 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2498 while (continue_replay && oldest_trans_id > 0) {
2500 journal_read_transaction(p_s_sb, cur_dblock, oldest_start,
2501 oldest_trans_id, newest_mount_id);
2504 } else if (ret != 0) {
2508 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + journal->j_start;
2510 if (cur_dblock == oldest_start)
2514 if (oldest_trans_id == 0) {
2515 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2516 "journal-1225: No valid " "transactions found");
2518 /* j_start does not get set correctly if we don't replay any transactions.
2519 ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2520 ** copy the trans_id from the header
2522 if (valid_journal_header && replay_count == 0) {
2523 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2524 journal->j_trans_id =
2525 le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2526 /* check for trans_id overflow */
2527 if (journal->j_trans_id == 0)
2528 journal->j_trans_id = 10;
2529 journal->j_last_flush_trans_id =
2530 le32_to_cpu(jh->j_last_flush_trans_id);
2531 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2533 journal->j_mount_id = newest_mount_id + 1;
2535 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2536 "newest_mount_id to %lu", journal->j_mount_id);
2537 journal->j_first_unflushed_offset = journal->j_start;
2538 if (replay_count > 0) {
2539 reiserfs_info(p_s_sb,
2540 "replayed %d transactions in %lu seconds\n",
2541 replay_count, get_seconds() - start);
2543 if (!bdev_read_only(p_s_sb->s_bdev) &&
2544 _update_journal_header_block(p_s_sb, journal->j_start,
2545 journal->j_last_flush_trans_id)) {
2546 /* replay failed, caller must call free_journal_ram and abort
2554 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2556 struct reiserfs_journal_list *jl;
2557 jl = kzalloc(sizeof(struct reiserfs_journal_list),
2558 GFP_NOFS | __GFP_NOFAIL);
2559 INIT_LIST_HEAD(&jl->j_list);
2560 INIT_LIST_HEAD(&jl->j_working_list);
2561 INIT_LIST_HEAD(&jl->j_tail_bh_list);
2562 INIT_LIST_HEAD(&jl->j_bh_list);
2563 mutex_init(&jl->j_commit_mutex);
2564 SB_JOURNAL(s)->j_num_lists++;
2565 get_journal_list(jl);
2569 static void journal_list_init(struct super_block *p_s_sb)
2571 SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
2574 static int release_journal_dev(struct super_block *super,
2575 struct reiserfs_journal *journal)
2581 if (journal->j_dev_bd != NULL) {
2582 if (journal->j_dev_bd->bd_dev != super->s_dev)
2583 bd_release(journal->j_dev_bd);
2584 result = blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2585 journal->j_dev_bd = NULL;
2589 reiserfs_warning(super, "sh-457",
2590 "Cannot release journal device: %i", result);
2595 static int journal_init_dev(struct super_block *super,
2596 struct reiserfs_journal *journal,
2597 const char *jdev_name)
2601 fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE;
2602 char b[BDEVNAME_SIZE];
2606 journal->j_dev_bd = NULL;
2607 jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2608 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2610 if (bdev_read_only(super->s_bdev))
2611 blkdev_mode = FMODE_READ;
2613 /* there is no "jdev" option and journal is on separate device */
2614 if ((!jdev_name || !jdev_name[0])) {
2615 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2616 journal->j_dev_mode = blkdev_mode;
2617 if (IS_ERR(journal->j_dev_bd)) {
2618 result = PTR_ERR(journal->j_dev_bd);
2619 journal->j_dev_bd = NULL;
2620 reiserfs_warning(super, "sh-458",
2621 "cannot init journal device '%s': %i",
2622 __bdevname(jdev, b), result);
2624 } else if (jdev != super->s_dev) {
2625 result = bd_claim(journal->j_dev_bd, journal);
2627 blkdev_put(journal->j_dev_bd, blkdev_mode);
2631 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2637 journal->j_dev_mode = blkdev_mode;
2638 journal->j_dev_bd = open_bdev_exclusive(jdev_name,
2639 blkdev_mode, journal);
2640 if (IS_ERR(journal->j_dev_bd)) {
2641 result = PTR_ERR(journal->j_dev_bd);
2642 journal->j_dev_bd = NULL;
2643 reiserfs_warning(super,
2644 "journal_init_dev: Cannot open '%s': %i",
2649 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2650 reiserfs_info(super,
2651 "journal_init_dev: journal device: %s\n",
2652 bdevname(journal->j_dev_bd, b));
2657 * When creating/tuning a file system user can assign some
2658 * journal params within boundaries which depend on the ratio
2659 * blocksize/standard_blocksize.
2661 * For blocks >= standard_blocksize transaction size should
2662 * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2663 * then JOURNAL_TRANS_MAX_DEFAULT.
2665 * For blocks < standard_blocksize these boundaries should be
2666 * decreased proportionally.
2668 #define REISERFS_STANDARD_BLKSIZE (4096)
2670 static int check_advise_trans_params(struct super_block *p_s_sb,
2671 struct reiserfs_journal *journal)
2673 if (journal->j_trans_max) {
2674 /* Non-default journal params.
2675 Do sanity check for them. */
2677 if (p_s_sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2678 ratio = REISERFS_STANDARD_BLKSIZE / p_s_sb->s_blocksize;
2680 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2681 journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2682 SB_ONDISK_JOURNAL_SIZE(p_s_sb) / journal->j_trans_max <
2683 JOURNAL_MIN_RATIO) {
2684 reiserfs_warning(p_s_sb, "sh-462",
2685 "bad transaction max size (%u). "
2686 "FSCK?", journal->j_trans_max);
2689 if (journal->j_max_batch != (journal->j_trans_max) *
2690 JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2691 reiserfs_warning(p_s_sb, "sh-463",
2692 "bad transaction max batch (%u). "
2693 "FSCK?", journal->j_max_batch);
2697 /* Default journal params.
2698 The file system was created by old version
2699 of mkreiserfs, so some fields contain zeros,
2700 and we need to advise proper values for them */
2701 if (p_s_sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2702 reiserfs_warning(p_s_sb, "sh-464", "bad blocksize (%u)",
2703 p_s_sb->s_blocksize);
2706 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2707 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2708 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2714 ** must be called once on fs mount. calls journal_read for you
2716 int journal_init(struct super_block *p_s_sb, const char *j_dev_name,
2717 int old_format, unsigned int commit_max_age)
2719 int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2;
2720 struct buffer_head *bhjh;
2721 struct reiserfs_super_block *rs;
2722 struct reiserfs_journal_header *jh;
2723 struct reiserfs_journal *journal;
2724 struct reiserfs_journal_list *jl;
2725 char b[BDEVNAME_SIZE];
2727 journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof(struct reiserfs_journal));
2729 reiserfs_warning(p_s_sb, "journal-1256",
2730 "unable to get memory for journal structure");
2733 memset(journal, 0, sizeof(struct reiserfs_journal));
2734 INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2735 INIT_LIST_HEAD(&journal->j_prealloc_list);
2736 INIT_LIST_HEAD(&journal->j_working_list);
2737 INIT_LIST_HEAD(&journal->j_journal_list);
2738 journal->j_persistent_trans = 0;
2739 if (reiserfs_allocate_list_bitmaps(p_s_sb,
2740 journal->j_list_bitmap,
2741 reiserfs_bmap_count(p_s_sb)))
2742 goto free_and_return;
2743 allocate_bitmap_nodes(p_s_sb);
2745 /* reserved for journal area support */
2746 SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ?
2747 REISERFS_OLD_DISK_OFFSET_IN_BYTES
2748 / p_s_sb->s_blocksize +
2749 reiserfs_bmap_count(p_s_sb) +
2751 REISERFS_DISK_OFFSET_IN_BYTES /
2752 p_s_sb->s_blocksize + 2);
2754 /* Sanity check to see is the standard journal fitting withing first bitmap
2755 (actual for small blocksizes) */
2756 if (!SB_ONDISK_JOURNAL_DEVICE(p_s_sb) &&
2757 (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) +
2758 SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8)) {
2759 reiserfs_warning(p_s_sb, "journal-1393",
2760 "journal does not fit for area addressed "
2761 "by first of bitmap blocks. It starts at "
2762 "%u and its size is %u. Block size %ld",
2763 SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb),
2764 SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2765 p_s_sb->s_blocksize);
2766 goto free_and_return;
2769 if (journal_init_dev(p_s_sb, journal, j_dev_name) != 0) {
2770 reiserfs_warning(p_s_sb, "sh-462",
2771 "unable to initialize jornal device");
2772 goto free_and_return;
2775 rs = SB_DISK_SUPER_BLOCK(p_s_sb);
2777 /* read journal header */
2778 bhjh = journal_bread(p_s_sb,
2779 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2780 SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2782 reiserfs_warning(p_s_sb, "sh-459",
2783 "unable to read journal header");
2784 goto free_and_return;
2786 jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2788 /* make sure that journal matches to the super block */
2789 if (is_reiserfs_jr(rs)
2790 && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2791 sb_jp_journal_magic(rs))) {
2792 reiserfs_warning(p_s_sb, "sh-460",
2793 "journal header magic %x (device %s) does "
2794 "not match to magic found in super block %x",
2795 jh->jh_journal.jp_journal_magic,
2796 bdevname(journal->j_dev_bd, b),
2797 sb_jp_journal_magic(rs));
2799 goto free_and_return;
2802 journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2803 journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2804 journal->j_max_commit_age =
2805 le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2806 journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2808 if (check_advise_trans_params(p_s_sb, journal) != 0)
2809 goto free_and_return;
2810 journal->j_default_max_commit_age = journal->j_max_commit_age;
2812 if (commit_max_age != 0) {
2813 journal->j_max_commit_age = commit_max_age;
2814 journal->j_max_trans_age = commit_max_age;
2817 reiserfs_info(p_s_sb, "journal params: device %s, size %u, "
2818 "journal first block %u, max trans len %u, max batch %u, "
2819 "max commit age %u, max trans age %u\n",
2820 bdevname(journal->j_dev_bd, b),
2821 SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2822 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2823 journal->j_trans_max,
2824 journal->j_max_batch,
2825 journal->j_max_commit_age, journal->j_max_trans_age);
2829 journal->j_list_bitmap_index = 0;
2830 journal_list_init(p_s_sb);
2832 memset(journal->j_list_hash_table, 0,
2833 JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2835 INIT_LIST_HEAD(&journal->j_dirty_buffers);
2836 spin_lock_init(&journal->j_dirty_buffers_lock);
2838 journal->j_start = 0;
2840 journal->j_len_alloc = 0;
2841 atomic_set(&(journal->j_wcount), 0);
2842 atomic_set(&(journal->j_async_throttle), 0);
2843 journal->j_bcount = 0;
2844 journal->j_trans_start_time = 0;
2845 journal->j_last = NULL;
2846 journal->j_first = NULL;
2847 init_waitqueue_head(&(journal->j_join_wait));
2848 mutex_init(&journal->j_mutex);
2849 mutex_init(&journal->j_flush_mutex);
2851 journal->j_trans_id = 10;
2852 journal->j_mount_id = 10;
2853 journal->j_state = 0;
2854 atomic_set(&(journal->j_jlock), 0);
2855 journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2856 journal->j_cnode_free_orig = journal->j_cnode_free_list;
2857 journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2858 journal->j_cnode_used = 0;
2859 journal->j_must_wait = 0;
2861 if (journal->j_cnode_free == 0) {
2862 reiserfs_warning(p_s_sb, "journal-2004", "Journal cnode memory "
2863 "allocation failed (%ld bytes). Journal is "
2864 "too large for available memory. Usually "
2865 "this is due to a journal that is too large.",
2866 sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2867 goto free_and_return;
2870 init_journal_hash(p_s_sb);
2871 jl = journal->j_current_jl;
2872 jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl);
2873 if (!jl->j_list_bitmap) {
2874 reiserfs_warning(p_s_sb, "journal-2005",
2875 "get_list_bitmap failed for journal list 0");
2876 goto free_and_return;
2878 if (journal_read(p_s_sb) < 0) {
2879 reiserfs_warning(p_s_sb, "reiserfs-2006",
2880 "Replay Failure, unable to mount");
2881 goto free_and_return;
2884 reiserfs_mounted_fs_count++;
2885 if (reiserfs_mounted_fs_count <= 1)
2886 commit_wq = create_workqueue("reiserfs");
2888 INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2889 journal->j_work_sb = p_s_sb;
2892 free_journal_ram(p_s_sb);
2897 ** test for a polite end of the current transaction. Used by file_write, and should
2898 ** be used by delete to make sure they don't write more than can fit inside a single
2901 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2904 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2905 time_t now = get_seconds();
2906 /* cannot restart while nested */
2907 BUG_ON(!th->t_trans_id);
2908 if (th->t_refcount > 1)
2910 if (journal->j_must_wait > 0 ||
2911 (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2912 atomic_read(&(journal->j_jlock)) ||
2913 (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2914 journal->j_cnode_free < (journal->j_trans_max * 3)) {
2917 /* protected by the BKL here */
2918 journal->j_len_alloc += new_alloc;
2919 th->t_blocks_allocated += new_alloc ;
2923 /* this must be called inside a transaction, and requires the
2924 ** kernel_lock to be held
2926 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2928 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2929 BUG_ON(!th->t_trans_id);
2930 journal->j_must_wait = 1;
2931 set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2935 /* this must be called without a transaction started, and does not
2938 void reiserfs_allow_writes(struct super_block *s)
2940 struct reiserfs_journal *journal = SB_JOURNAL(s);
2941 clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2942 wake_up(&journal->j_join_wait);
2945 /* this must be called without a transaction started, and does not
2948 void reiserfs_wait_on_write_block(struct super_block *s)
2950 struct reiserfs_journal *journal = SB_JOURNAL(s);
2951 wait_event(journal->j_join_wait,
2952 !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2955 static void queue_log_writer(struct super_block *s)
2958 struct reiserfs_journal *journal = SB_JOURNAL(s);
2959 set_bit(J_WRITERS_QUEUED, &journal->j_state);
2962 * we don't want to use wait_event here because
2963 * we only want to wait once.
2965 init_waitqueue_entry(&wait, current);
2966 add_wait_queue(&journal->j_join_wait, &wait);
2967 set_current_state(TASK_UNINTERRUPTIBLE);
2968 if (test_bit(J_WRITERS_QUEUED, &journal->j_state))
2970 __set_current_state(TASK_RUNNING);
2971 remove_wait_queue(&journal->j_join_wait, &wait);
2974 static void wake_queued_writers(struct super_block *s)
2976 struct reiserfs_journal *journal = SB_JOURNAL(s);
2977 if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2978 wake_up(&journal->j_join_wait);
2981 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
2983 struct reiserfs_journal *journal = SB_JOURNAL(sb);
2984 unsigned long bcount = journal->j_bcount;
2986 schedule_timeout_uninterruptible(1);
2987 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2988 while ((atomic_read(&journal->j_wcount) > 0 ||
2989 atomic_read(&journal->j_jlock)) &&
2990 journal->j_trans_id == trans_id) {
2991 queue_log_writer(sb);
2993 if (journal->j_trans_id != trans_id)
2995 if (bcount == journal->j_bcount)
2997 bcount = journal->j_bcount;
3001 /* join == true if you must join an existing transaction.
3002 ** join == false if you can deal with waiting for others to finish
3004 ** this will block until the transaction is joinable. send the number of blocks you
3005 ** expect to use in nblocks.
3007 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3008 struct super_block *p_s_sb, unsigned long nblocks,
3011 time_t now = get_seconds();
3012 unsigned int old_trans_id;
3013 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3014 struct reiserfs_transaction_handle myth;
3015 int sched_count = 0;
3018 reiserfs_check_lock_depth(p_s_sb, "journal_begin");
3019 BUG_ON(nblocks > journal->j_trans_max);
3021 PROC_INFO_INC(p_s_sb, journal.journal_being);
3022 /* set here for journal_join */
3024 th->t_super = p_s_sb;
3027 lock_journal(p_s_sb);
3028 if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3029 unlock_journal(p_s_sb);
3030 retval = journal->j_errno;
3033 journal->j_bcount++;
3035 if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3036 unlock_journal(p_s_sb);
3037 reiserfs_wait_on_write_block(p_s_sb);
3038 PROC_INFO_INC(p_s_sb, journal.journal_relock_writers);
3041 now = get_seconds();
3043 /* if there is no room in the journal OR
3044 ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
3045 ** we don't sleep if there aren't other writers
3048 if ((!join && journal->j_must_wait > 0) ||
3050 && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3051 || (!join && atomic_read(&journal->j_wcount) > 0
3052 && journal->j_trans_start_time > 0
3053 && (now - journal->j_trans_start_time) >
3054 journal->j_max_trans_age) || (!join
3055 && atomic_read(&journal->j_jlock))
3056 || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3058 old_trans_id = journal->j_trans_id;
3059 unlock_journal(p_s_sb); /* allow others to finish this transaction */
3061 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3062 journal->j_max_batch &&
3063 ((journal->j_len + nblocks + 2) * 100) <
3064 (journal->j_len_alloc * 75)) {
3065 if (atomic_read(&journal->j_wcount) > 10) {
3067 queue_log_writer(p_s_sb);
3071 /* don't mess with joining the transaction if all we have to do is
3072 * wait for someone else to do a commit
3074 if (atomic_read(&journal->j_jlock)) {
3075 while (journal->j_trans_id == old_trans_id &&
3076 atomic_read(&journal->j_jlock)) {
3077 queue_log_writer(p_s_sb);
3081 retval = journal_join(&myth, p_s_sb, 1);
3085 /* someone might have ended the transaction while we joined */
3086 if (old_trans_id != journal->j_trans_id) {
3087 retval = do_journal_end(&myth, p_s_sb, 1, 0);
3089 retval = do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW);
3095 PROC_INFO_INC(p_s_sb, journal.journal_relock_wcount);
3098 /* we are the first writer, set trans_id */
3099 if (journal->j_trans_start_time == 0) {
3100 journal->j_trans_start_time = get_seconds();
3102 atomic_inc(&(journal->j_wcount));
3103 journal->j_len_alloc += nblocks;
3104 th->t_blocks_logged = 0;
3105 th->t_blocks_allocated = nblocks;
3106 th->t_trans_id = journal->j_trans_id;
3107 unlock_journal(p_s_sb);
3108 INIT_LIST_HEAD(&th->t_list);
3113 memset(th, 0, sizeof(*th));
3114 /* Re-set th->t_super, so we can properly keep track of how many
3115 * persistent transactions there are. We need to do this so if this
3116 * call is part of a failed restart_transaction, we can free it later */
3117 th->t_super = p_s_sb;
3121 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3127 struct reiserfs_transaction_handle *th;
3129 /* if we're nesting into an existing transaction. It will be
3130 ** persistent on its own
3132 if (reiserfs_transaction_running(s)) {
3133 th = current->journal_info;
3135 BUG_ON(th->t_refcount < 2);
3139 th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3142 ret = journal_begin(th, s, nblocks);
3148 SB_JOURNAL(s)->j_persistent_trans++;
3152 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3154 struct super_block *s = th->t_super;
3157 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3160 if (th->t_refcount == 0) {
3161 SB_JOURNAL(s)->j_persistent_trans--;
3167 static int journal_join(struct reiserfs_transaction_handle *th,
3168 struct super_block *p_s_sb, unsigned long nblocks)
3170 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3172 /* this keeps do_journal_end from NULLing out the current->journal_info
3175 th->t_handle_save = cur_th;
3176 BUG_ON(cur_th && cur_th->t_refcount > 1);
3177 return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_JOIN);
3180 int journal_join_abort(struct reiserfs_transaction_handle *th,
3181 struct super_block *p_s_sb, unsigned long nblocks)
3183 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3185 /* this keeps do_journal_end from NULLing out the current->journal_info
3188 th->t_handle_save = cur_th;
3189 BUG_ON(cur_th && cur_th->t_refcount > 1);
3190 return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_ABORT);
3193 int journal_begin(struct reiserfs_transaction_handle *th,
3194 struct super_block *p_s_sb, unsigned long nblocks)
3196 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3199 th->t_handle_save = NULL;
3201 /* we are nesting into the current transaction */
3202 if (cur_th->t_super == p_s_sb) {
3203 BUG_ON(!cur_th->t_refcount);
3204 cur_th->t_refcount++;
3205 memcpy(th, cur_th, sizeof(*th));
3206 if (th->t_refcount <= 1)
3207 reiserfs_warning(p_s_sb, "reiserfs-2005",
3208 "BAD: refcount <= 1, but "
3209 "journal_info != 0");
3212 /* we've ended up with a handle from a different filesystem.
3213 ** save it and restore on journal_end. This should never
3216 reiserfs_warning(p_s_sb, "clm-2100",
3217 "nesting info a different FS");
3218 th->t_handle_save = current->journal_info;
3219 current->journal_info = th;
3222 current->journal_info = th;
3224 ret = do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_REG);
3225 BUG_ON(current->journal_info != th);
3227 /* I guess this boils down to being the reciprocal of clm-2100 above.
3228 * If do_journal_begin_r fails, we need to put it back, since journal_end
3229 * won't be called to do it. */
3231 current->journal_info = th->t_handle_save;
3233 BUG_ON(!th->t_refcount);
3239 ** puts bh into the current transaction. If it was already there, reorders removes the
3240 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3242 ** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the
3243 ** transaction is committed.
3245 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3247 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3248 struct super_block *p_s_sb, struct buffer_head *bh)
3250 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3251 struct reiserfs_journal_cnode *cn = NULL;
3252 int count_already_incd = 0;
3254 BUG_ON(!th->t_trans_id);
3256 PROC_INFO_INC(p_s_sb, journal.mark_dirty);
3257 if (th->t_trans_id != journal->j_trans_id) {
3258 reiserfs_panic(th->t_super,
3259 "journal-1577: handle trans id %ld != current trans id %ld\n",
3260 th->t_trans_id, journal->j_trans_id);
3265 prepared = test_clear_buffer_journal_prepared(bh);
3266 clear_buffer_journal_restore_dirty(bh);
3267 /* already in this transaction, we are done */
3268 if (buffer_journaled(bh)) {
3269 PROC_INFO_INC(p_s_sb, journal.mark_dirty_already);
3273 /* this must be turned into a panic instead of a warning. We can't allow
3274 ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3275 ** could get to disk too early. NOT GOOD.
3277 if (!prepared || buffer_dirty(bh)) {
3278 reiserfs_warning(p_s_sb, "journal-1777",
3279 "buffer %llu bad state "
3280 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3281 (unsigned long long)bh->b_blocknr,
3282 prepared ? ' ' : '!',
3283 buffer_locked(bh) ? ' ' : '!',
3284 buffer_dirty(bh) ? ' ' : '!',
3285 buffer_journal_dirty(bh) ? ' ' : '!');
3288 if (atomic_read(&(journal->j_wcount)) <= 0) {
3289 reiserfs_warning(p_s_sb, "journal-1409",
3290 "returning because j_wcount was %d",
3291 atomic_read(&(journal->j_wcount)));
3294 /* this error means I've screwed up, and we've overflowed the transaction.
3295 ** Nothing can be done here, except make the FS readonly or panic.
3297 if (journal->j_len >= journal->j_trans_max) {
3298 reiserfs_panic(th->t_super,
3299 "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n",
3303 if (buffer_journal_dirty(bh)) {
3304 count_already_incd = 1;
3305 PROC_INFO_INC(p_s_sb, journal.mark_dirty_notjournal);
3306 clear_buffer_journal_dirty(bh);
3309 if (journal->j_len > journal->j_len_alloc) {
3310 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3313 set_buffer_journaled(bh);
3315 /* now put this guy on the end */
3317 cn = get_cnode(p_s_sb);
3319 reiserfs_panic(p_s_sb, "get_cnode failed!\n");
3322 if (th->t_blocks_logged == th->t_blocks_allocated) {
3323 th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3324 journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3326 th->t_blocks_logged++;
3330 cn->blocknr = bh->b_blocknr;
3333 insert_journal_hash(journal->j_hash_table, cn);
3334 if (!count_already_incd) {
3339 cn->prev = journal->j_last;
3341 if (journal->j_last) {
3342 journal->j_last->next = cn;
3343 journal->j_last = cn;
3345 journal->j_first = cn;
3346 journal->j_last = cn;
3351 int journal_end(struct reiserfs_transaction_handle *th,
3352 struct super_block *p_s_sb, unsigned long nblocks)
3354 if (!current->journal_info && th->t_refcount > 1)
3355 reiserfs_warning(p_s_sb, "REISER-NESTING",
3356 "th NULL, refcount %d", th->t_refcount);
3358 if (!th->t_trans_id) {
3364 if (th->t_refcount > 0) {
3365 struct reiserfs_transaction_handle *cur_th =
3366 current->journal_info;
3368 /* we aren't allowed to close a nested transaction on a different
3369 ** filesystem from the one in the task struct
3371 BUG_ON(cur_th->t_super != th->t_super);
3374 memcpy(current->journal_info, th, sizeof(*th));
3379 return do_journal_end(th, p_s_sb, nblocks, 0);
3383 /* removes from the current transaction, relsing and descrementing any counters.
3384 ** also files the removed buffer directly onto the clean list
3386 ** called by journal_mark_freed when a block has been deleted
3388 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3390 static int remove_from_transaction(struct super_block *p_s_sb,
3391 b_blocknr_t blocknr, int already_cleaned)
3393 struct buffer_head *bh;
3394 struct reiserfs_journal_cnode *cn;
3395 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3398 cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3399 if (!cn || !cn->bh) {
3404 cn->prev->next = cn->next;
3407 cn->next->prev = cn->prev;
3409 if (cn == journal->j_first) {
3410 journal->j_first = cn->next;
3412 if (cn == journal->j_last) {
3413 journal->j_last = cn->prev;
3416 remove_journal_hash(p_s_sb, journal->j_hash_table, NULL,
3418 clear_buffer_journaled(bh); /* don't log this one */
3420 if (!already_cleaned) {
3421 clear_buffer_journal_dirty(bh);
3422 clear_buffer_dirty(bh);
3423 clear_buffer_journal_test(bh);
3425 if (atomic_read(&(bh->b_count)) < 0) {
3426 reiserfs_warning(p_s_sb, "journal-1752",
3432 journal->j_len_alloc--;
3433 free_cnode(p_s_sb, cn);
3438 ** for any cnode in a journal list, it can only be dirtied of all the
3439 ** transactions that include it are committed to disk.
3440 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3441 ** and 0 if you aren't
3443 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3444 ** blocks for a given transaction on disk
3447 static int can_dirty(struct reiserfs_journal_cnode *cn)
3449 struct super_block *sb = cn->sb;
3450 b_blocknr_t blocknr = cn->blocknr;
3451 struct reiserfs_journal_cnode *cur = cn->hprev;
3454 /* first test hprev. These are all newer than cn, so any node here
3455 ** with the same block number and dev means this node can't be sent
3456 ** to disk right now.
3458 while (cur && can_dirty) {
3459 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3460 cur->blocknr == blocknr) {
3465 /* then test hnext. These are all older than cn. As long as they
3466 ** are committed to the log, it is safe to write cn to disk
3469 while (cur && can_dirty) {
3470 if (cur->jlist && cur->jlist->j_len > 0 &&
3471 atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3472 cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3480 /* syncs the commit blocks, but does not force the real buffers to disk
3481 ** will wait until the current transaction is done/committed before returning
3483 int journal_end_sync(struct reiserfs_transaction_handle *th,
3484 struct super_block *p_s_sb, unsigned long nblocks)
3486 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3488 BUG_ON(!th->t_trans_id);
3489 /* you can sync while nested, very, very bad */
3490 BUG_ON(th->t_refcount > 1);
3491 if (journal->j_len == 0) {
3492 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3494 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3496 return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT);
3500 ** writeback the pending async commits to disk
3502 static void flush_async_commits(struct work_struct *work)
3504 struct reiserfs_journal *journal =
3505 container_of(work, struct reiserfs_journal, j_work.work);
3506 struct super_block *p_s_sb = journal->j_work_sb;
3507 struct reiserfs_journal_list *jl;
3508 struct list_head *entry;
3511 if (!list_empty(&journal->j_journal_list)) {
3512 /* last entry is the youngest, commit it and you get everything */
3513 entry = journal->j_journal_list.prev;
3514 jl = JOURNAL_LIST_ENTRY(entry);
3515 flush_commit_list(p_s_sb, jl, 1);
3521 ** flushes any old transactions to disk
3522 ** ends the current transaction if it is too old
3524 int reiserfs_flush_old_commits(struct super_block *p_s_sb)
3527 struct reiserfs_transaction_handle th;
3528 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3530 now = get_seconds();
3531 /* safety check so we don't flush while we are replaying the log during
3534 if (list_empty(&journal->j_journal_list)) {
3538 /* check the current transaction. If there are no writers, and it is
3539 * too old, finish it, and force the commit blocks to disk
3541 if (atomic_read(&journal->j_wcount) <= 0 &&
3542 journal->j_trans_start_time > 0 &&
3543 journal->j_len > 0 &&
3544 (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3545 if (!journal_join(&th, p_s_sb, 1)) {
3546 reiserfs_prepare_for_journal(p_s_sb,
3547 SB_BUFFER_WITH_SB(p_s_sb),
3549 journal_mark_dirty(&th, p_s_sb,
3550 SB_BUFFER_WITH_SB(p_s_sb));
3552 /* we're only being called from kreiserfsd, it makes no sense to do
3553 ** an async commit so that kreiserfsd can do it later
3555 do_journal_end(&th, p_s_sb, 1, COMMIT_NOW | WAIT);
3558 return p_s_sb->s_dirt;
3562 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3564 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3565 ** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just
3566 ** flushes the commit list and returns 0.
3568 ** Won't batch when flush or commit_now is set. Also won't batch when others are waiting on j_join_wait.
3570 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3572 static int check_journal_end(struct reiserfs_transaction_handle *th,
3573 struct super_block *p_s_sb, unsigned long nblocks,
3578 int flush = flags & FLUSH_ALL;
3579 int commit_now = flags & COMMIT_NOW;
3580 int wait_on_commit = flags & WAIT;
3581 struct reiserfs_journal_list *jl;
3582 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3584 BUG_ON(!th->t_trans_id);
3586 if (th->t_trans_id != journal->j_trans_id) {
3587 reiserfs_panic(th->t_super,
3588 "journal-1577: handle trans id %ld != current trans id %ld\n",
3589 th->t_trans_id, journal->j_trans_id);
3592 journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3593 if (atomic_read(&(journal->j_wcount)) > 0) { /* <= 0 is allowed. unmounting might not call begin */
3594 atomic_dec(&(journal->j_wcount));
3597 /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3598 ** will be dealt with by next transaction that actually writes something, but should be taken
3599 ** care of in this trans
3601 BUG_ON(journal->j_len == 0);
3603 /* if wcount > 0, and we are called to with flush or commit_now,
3604 ** we wait on j_join_wait. We will wake up when the last writer has
3605 ** finished the transaction, and started it on its way to the disk.
3606 ** Then, we flush the commit or journal list, and just return 0
3607 ** because the rest of journal end was already done for this transaction.
3609 if (atomic_read(&(journal->j_wcount)) > 0) {
3610 if (flush || commit_now) {
3613 jl = journal->j_current_jl;
3614 trans_id = jl->j_trans_id;
3616 jl->j_state |= LIST_COMMIT_PENDING;
3617 atomic_set(&(journal->j_jlock), 1);
3619 journal->j_next_full_flush = 1;
3621 unlock_journal(p_s_sb);
3623 /* sleep while the current transaction is still j_jlocked */
3624 while (journal->j_trans_id == trans_id) {
3625 if (atomic_read(&journal->j_jlock)) {
3626 queue_log_writer(p_s_sb);
3628 lock_journal(p_s_sb);
3629 if (journal->j_trans_id == trans_id) {
3630 atomic_set(&(journal->j_jlock),
3633 unlock_journal(p_s_sb);
3636 BUG_ON(journal->j_trans_id == trans_id);
3639 && journal_list_still_alive(p_s_sb, trans_id)
3640 && wait_on_commit) {
3641 flush_commit_list(p_s_sb, jl, 1);
3645 unlock_journal(p_s_sb);
3649 /* deal with old transactions where we are the last writers */
3650 now = get_seconds();
3651 if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3653 journal->j_next_async_flush = 1;
3655 /* don't batch when someone is waiting on j_join_wait */
3656 /* don't batch when syncing the commit or flushing the whole trans */
3657 if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3658 && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3659 && journal->j_len_alloc < journal->j_max_batch
3660 && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3661 journal->j_bcount++;
3662 unlock_journal(p_s_sb);
3666 if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
3667 reiserfs_panic(p_s_sb,
3668 "journal-003: journal_end: j_start (%ld) is too high\n",
3675 ** Does all the work that makes deleting blocks safe.
3676 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3679 ** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes
3680 ** before this transaction has finished.
3682 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with
3683 ** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash,
3684 ** the block can't be reallocated yet.
3686 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3688 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3689 struct super_block *p_s_sb, b_blocknr_t blocknr)
3691 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3692 struct reiserfs_journal_cnode *cn = NULL;
3693 struct buffer_head *bh = NULL;
3694 struct reiserfs_list_bitmap *jb = NULL;
3696 BUG_ON(!th->t_trans_id);
3698 cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3703 /* if it is journal new, we just remove it from this transaction */
3704 if (bh && buffer_journal_new(bh)) {
3705 clear_buffer_journal_new(bh);
3706 clear_prepared_bits(bh);
3707 reiserfs_clean_and_file_buffer(bh);
3708 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3710 /* set the bit for this block in the journal bitmap for this transaction */
3711 jb = journal->j_current_jl->j_list_bitmap;
3713 reiserfs_panic(p_s_sb,
3714 "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n");
3716 set_bit_in_list_bitmap(p_s_sb, blocknr, jb);
3718 /* Note, the entire while loop is not allowed to schedule. */
3721 clear_prepared_bits(bh);
3722 reiserfs_clean_and_file_buffer(bh);
3724 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3726 /* find all older transactions with this block, make sure they don't try to write it out */
3727 cn = get_journal_hash_dev(p_s_sb, journal->j_list_hash_table,
3730 if (p_s_sb == cn->sb && blocknr == cn->blocknr) {
3731 set_bit(BLOCK_FREED, &cn->state);
3734 /* remove_from_transaction will brelse the buffer if it was
3735 ** in the current trans
3737 clear_buffer_journal_dirty(cn->
3739 clear_buffer_dirty(cn->bh);
3740 clear_buffer_journal_test(cn->
3745 (&(cn->bh->b_count)) < 0) {
3746 reiserfs_warning(p_s_sb,
3748 "cn->bh->b_count < 0");
3751 if (cn->jlist) { /* since we are clearing the bh, we MUST dec nonzerolen */
3764 release_buffer_page(bh); /* get_hash grabs the buffer */
3768 void reiserfs_update_inode_transaction(struct inode *inode)
3770 struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3771 REISERFS_I(inode)->i_jl = journal->j_current_jl;
3772 REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3776 * returns -1 on error, 0 if no commits/barriers were done and 1
3777 * if a transaction was actually committed and the barrier was done
3779 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3780 struct reiserfs_journal_list *jl)
3782 struct reiserfs_transaction_handle th;
3783 struct super_block *sb = inode->i_sb;
3784 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3787 /* is it from the current transaction, or from an unknown transaction? */
3788 if (id == journal->j_trans_id) {
3789 jl = journal->j_current_jl;
3790 /* try to let other writers come in and grow this transaction */
3791 let_transaction_grow(sb, id);
3792 if (journal->j_trans_id != id) {
3793 goto flush_commit_only;
3796 ret = journal_begin(&th, sb, 1);
3800 /* someone might have ended this transaction while we joined */
3801 if (journal->j_trans_id != id) {
3802 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3804 journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3805 ret = journal_end(&th, sb, 1);
3806 goto flush_commit_only;
3809 ret = journal_end_sync(&th, sb, 1);
3814 /* this gets tricky, we have to make sure the journal list in
3815 * the inode still exists. We know the list is still around
3816 * if we've got a larger transaction id than the oldest list
3819 if (journal_list_still_alive(inode->i_sb, id)) {
3821 * we only set ret to 1 when we know for sure
3822 * the barrier hasn't been started yet on the commit
3825 if (atomic_read(&jl->j_commit_left) > 1)
3827 flush_commit_list(sb, jl, 1);
3828 if (journal->j_errno)
3829 ret = journal->j_errno;
3832 /* otherwise the list is gone, and long since committed */
3836 int reiserfs_commit_for_inode(struct inode *inode)
3838 unsigned int id = REISERFS_I(inode)->i_trans_id;
3839 struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3841 /* for the whole inode, assume unset id means it was
3842 * changed in the current transaction. More conservative
3845 reiserfs_update_inode_transaction(inode);
3846 id = REISERFS_I(inode)->i_trans_id;
3847 /* jl will be updated in __commit_trans_jl */
3850 return __commit_trans_jl(inode, id, jl);
3853 void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb,
3854 struct buffer_head *bh)
3856 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3857 PROC_INFO_INC(p_s_sb, journal.restore_prepared);
3861 if (test_clear_buffer_journal_restore_dirty(bh) &&
3862 buffer_journal_dirty(bh)) {
3863 struct reiserfs_journal_cnode *cn;
3864 cn = get_journal_hash_dev(p_s_sb,
3865 journal->j_list_hash_table,
3867 if (cn && can_dirty(cn)) {
3868 set_buffer_journal_test(bh);
3869 mark_buffer_dirty(bh);
3872 clear_buffer_journal_prepared(bh);
3875 extern struct tree_balance *cur_tb;
3877 ** before we can change a metadata block, we have to make sure it won't
3878 ** be written to disk while we are altering it. So, we must:
3883 int reiserfs_prepare_for_journal(struct super_block *p_s_sb,
3884 struct buffer_head *bh, int wait)
3886 PROC_INFO_INC(p_s_sb, journal.prepare);
3888 if (!trylock_buffer(bh)) {
3893 set_buffer_journal_prepared(bh);
3894 if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3895 clear_buffer_journal_test(bh);
3896 set_buffer_journal_restore_dirty(bh);
3902 static void flush_old_journal_lists(struct super_block *s)
3904 struct reiserfs_journal *journal = SB_JOURNAL(s);
3905 struct reiserfs_journal_list *jl;
3906 struct list_head *entry;
3907 time_t now = get_seconds();
3909 while (!list_empty(&journal->j_journal_list)) {
3910 entry = journal->j_journal_list.next;
3911 jl = JOURNAL_LIST_ENTRY(entry);
3912 /* this check should always be run, to send old lists to disk */
3913 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) &&
3914 atomic_read(&jl->j_commit_left) == 0 &&
3915 test_transaction(s, jl)) {
3916 flush_used_journal_lists(s, jl);
3924 ** long and ugly. If flush, will not return until all commit
3925 ** blocks and all real buffers in the trans are on disk.
3926 ** If no_async, won't return until all commit blocks are on disk.
3928 ** keep reading, there are comments as you go along
3930 ** If the journal is aborted, we just clean up. Things like flushing
3931 ** journal lists, etc just won't happen.
3933 static int do_journal_end(struct reiserfs_transaction_handle *th,
3934 struct super_block *p_s_sb, unsigned long nblocks,
3937 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3938 struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3939 struct reiserfs_journal_cnode *last_cn = NULL;
3940 struct reiserfs_journal_desc *desc;
3941 struct reiserfs_journal_commit *commit;
3942 struct buffer_head *c_bh; /* commit bh */
3943 struct buffer_head *d_bh; /* desc bh */
3944 int cur_write_start = 0; /* start index of current log write */
3949 struct reiserfs_journal_list *jl, *temp_jl;
3950 struct list_head *entry, *safe;
3951 unsigned long jindex;
3952 unsigned int commit_trans_id;
3955 BUG_ON(th->t_refcount > 1);
3956 BUG_ON(!th->t_trans_id);
3958 /* protect flush_older_commits from doing mistakes if the
3959 transaction ID counter gets overflowed. */
3960 if (th->t_trans_id == ~0U)
3961 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
3962 flush = flags & FLUSH_ALL;
3963 wait_on_commit = flags & WAIT;
3966 current->journal_info = th->t_handle_save;
3967 reiserfs_check_lock_depth(p_s_sb, "journal end");
3968 if (journal->j_len == 0) {
3969 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3971 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3974 lock_journal(p_s_sb);
3975 if (journal->j_next_full_flush) {
3979 if (journal->j_next_async_flush) {
3980 flags |= COMMIT_NOW | WAIT;
3984 /* check_journal_end locks the journal, and unlocks if it does not return 1
3985 ** it tells us if we should continue with the journal_end, or just return
3987 if (!check_journal_end(th, p_s_sb, nblocks, flags)) {
3989 wake_queued_writers(p_s_sb);
3990 reiserfs_async_progress_wait(p_s_sb);
3994 /* check_journal_end might set these, check again */
3995 if (journal->j_next_full_flush) {
4000 ** j must wait means we have to flush the log blocks, and the real blocks for
4003 if (journal->j_must_wait > 0) {
4006 #ifdef REISERFS_PREALLOCATE
4007 /* quota ops might need to nest, setup the journal_info pointer for them
4008 * and raise the refcount so that it is > 0. */
4009 current->journal_info = th;
4011 reiserfs_discard_all_prealloc(th); /* it should not involve new blocks into
4012 * the transaction */
4014 current->journal_info = th->t_handle_save;
4017 /* setup description block */
4019 journal_getblk(p_s_sb,
4020 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4022 set_buffer_uptodate(d_bh);
4023 desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4024 memset(d_bh->b_data, 0, d_bh->b_size);
4025 memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4026 set_desc_trans_id(desc, journal->j_trans_id);
4028 /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */
4029 c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4030 ((journal->j_start + journal->j_len +
4031 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
4032 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4033 memset(c_bh->b_data, 0, c_bh->b_size);
4034 set_commit_trans_id(commit, journal->j_trans_id);
4035 set_buffer_uptodate(c_bh);
4037 /* init this journal list */
4038 jl = journal->j_current_jl;
4040 /* we lock the commit before doing anything because
4041 * we want to make sure nobody tries to run flush_commit_list until
4042 * the new transaction is fully setup, and we've already flushed the
4045 mutex_lock(&jl->j_commit_mutex);
4047 /* save the transaction id in case we need to commit it later */
4048 commit_trans_id = jl->j_trans_id;
4050 atomic_set(&jl->j_older_commits_done, 0);
4051 jl->j_trans_id = journal->j_trans_id;
4052 jl->j_timestamp = journal->j_trans_start_time;
4053 jl->j_commit_bh = c_bh;
4054 jl->j_start = journal->j_start;
4055 jl->j_len = journal->j_len;
4056 atomic_set(&jl->j_nonzerolen, journal->j_len);
4057 atomic_set(&jl->j_commit_left, journal->j_len + 2);
4058 jl->j_realblock = NULL;
4060 /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4061 ** for each real block, add it to the journal list hash,
4062 ** copy into real block index array in the commit or desc block
4064 trans_half = journal_trans_half(p_s_sb->s_blocksize);
4065 for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4066 if (buffer_journaled(cn->bh)) {
4067 jl_cn = get_cnode(p_s_sb);
4069 reiserfs_panic(p_s_sb,
4070 "journal-1676, get_cnode returned NULL\n");
4073 jl->j_realblock = jl_cn;
4075 jl_cn->prev = last_cn;
4078 last_cn->next = jl_cn;
4081 /* make sure the block we are trying to log is not a block
4082 of journal or reserved area */
4084 if (is_block_in_log_or_reserved_area
4085 (p_s_sb, cn->bh->b_blocknr)) {
4086 reiserfs_panic(p_s_sb,
4087 "journal-2332: Trying to log block %lu, which is a log block\n",
4090 jl_cn->blocknr = cn->bh->b_blocknr;
4095 insert_journal_hash(journal->j_list_hash_table, jl_cn);
4096 if (i < trans_half) {
4097 desc->j_realblock[i] =
4098 cpu_to_le32(cn->bh->b_blocknr);
4100 commit->j_realblock[i - trans_half] =
4101 cpu_to_le32(cn->bh->b_blocknr);
4107 set_desc_trans_len(desc, journal->j_len);
4108 set_desc_mount_id(desc, journal->j_mount_id);
4109 set_desc_trans_id(desc, journal->j_trans_id);
4110 set_commit_trans_len(commit, journal->j_len);
4112 /* special check in case all buffers in the journal were marked for not logging */
4113 BUG_ON(journal->j_len == 0);
4115 /* we're about to dirty all the log blocks, mark the description block
4116 * dirty now too. Don't mark the commit block dirty until all the
4117 * others are on disk
4119 mark_buffer_dirty(d_bh);
4121 /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4122 cur_write_start = journal->j_start;
4123 cn = journal->j_first;
4124 jindex = 1; /* start at one so we don't get the desc again */
4126 clear_buffer_journal_new(cn->bh);
4127 /* copy all the real blocks into log area. dirty log blocks */
4128 if (buffer_journaled(cn->bh)) {
4129 struct buffer_head *tmp_bh;
4133 journal_getblk(p_s_sb,
4134 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4137 SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
4138 set_buffer_uptodate(tmp_bh);
4139 page = cn->bh->b_page;
4141 memcpy(tmp_bh->b_data,
4142 addr + offset_in_page(cn->bh->b_data),
4145 mark_buffer_dirty(tmp_bh);
4147 set_buffer_journal_dirty(cn->bh);
4148 clear_buffer_journaled(cn->bh);
4150 /* JDirty cleared sometime during transaction. don't log this one */
4151 reiserfs_warning(p_s_sb, "journal-2048",
4152 "BAD, buffer in journal hash, "
4157 free_cnode(p_s_sb, cn);
4162 /* we are done with both the c_bh and d_bh, but
4163 ** c_bh must be written after all other commit blocks,
4164 ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4167 journal->j_current_jl = alloc_journal_list(p_s_sb);
4169 /* now it is safe to insert this transaction on the main list */
4170 list_add_tail(&jl->j_list, &journal->j_journal_list);
4171 list_add_tail(&jl->j_working_list, &journal->j_working_list);
4172 journal->j_num_work_lists++;
4174 /* reset journal values for the next transaction */
4175 old_start = journal->j_start;
4177 (journal->j_start + journal->j_len +
4178 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb);
4179 atomic_set(&(journal->j_wcount), 0);
4180 journal->j_bcount = 0;
4181 journal->j_last = NULL;
4182 journal->j_first = NULL;
4184 journal->j_trans_start_time = 0;
4185 /* check for trans_id overflow */
4186 if (++journal->j_trans_id == 0)
4187 journal->j_trans_id = 10;
4188 journal->j_current_jl->j_trans_id = journal->j_trans_id;
4189 journal->j_must_wait = 0;
4190 journal->j_len_alloc = 0;
4191 journal->j_next_full_flush = 0;
4192 journal->j_next_async_flush = 0;
4193 init_journal_hash(p_s_sb);
4195 // make sure reiserfs_add_jh sees the new current_jl before we
4196 // write out the tails
4199 /* tail conversion targets have to hit the disk before we end the
4200 * transaction. Otherwise a later transaction might repack the tail
4201 * before this transaction commits, leaving the data block unflushed and
4202 * clean, if we crash before the later transaction commits, the data block
4205 if (!list_empty(&jl->j_tail_bh_list)) {
4207 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4208 journal, jl, &jl->j_tail_bh_list);
4211 BUG_ON(!list_empty(&jl->j_tail_bh_list));
4212 mutex_unlock(&jl->j_commit_mutex);
4214 /* honor the flush wishes from the caller, simple commits can
4215 ** be done outside the journal lock, they are done below
4217 ** if we don't flush the commit list right now, we put it into
4218 ** the work queue so the people waiting on the async progress work
4219 ** queue don't wait for this proc to flush journal lists and such.
4222 flush_commit_list(p_s_sb, jl, 1);
4223 flush_journal_list(p_s_sb, jl, 1);
4224 } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4225 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4227 /* if the next transaction has any chance of wrapping, flush
4228 ** transactions that might get overwritten. If any journal lists are very
4229 ** old flush them as well.
4232 list_for_each_safe(entry, safe, &journal->j_journal_list) {
4233 temp_jl = JOURNAL_LIST_ENTRY(entry);
4234 if (journal->j_start <= temp_jl->j_start) {
4235 if ((journal->j_start + journal->j_trans_max + 1) >=
4237 flush_used_journal_lists(p_s_sb, temp_jl);
4239 } else if ((journal->j_start +
4240 journal->j_trans_max + 1) <
4241 SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4242 /* if we don't cross into the next transaction and we don't
4243 * wrap, there is no way we can overlap any later transactions
4248 } else if ((journal->j_start +
4249 journal->j_trans_max + 1) >
4250 SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4251 if (((journal->j_start + journal->j_trans_max + 1) %
4252 SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >=
4254 flush_used_journal_lists(p_s_sb, temp_jl);
4257 /* we don't overlap anything from out start to the end of the
4258 * log, and our wrapped portion doesn't overlap anything at
4259 * the start of the log. We can break
4265 flush_old_journal_lists(p_s_sb);
4267 journal->j_current_jl->j_list_bitmap =
4268 get_list_bitmap(p_s_sb, journal->j_current_jl);
4270 if (!(journal->j_current_jl->j_list_bitmap)) {
4271 reiserfs_panic(p_s_sb,
4272 "journal-1996: do_journal_end, could not get a list bitmap\n");
4275 atomic_set(&(journal->j_jlock), 0);
4276 unlock_journal(p_s_sb);
4277 /* wake up any body waiting to join. */
4278 clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4279 wake_up(&(journal->j_join_wait));
4281 if (!flush && wait_on_commit &&
4282 journal_list_still_alive(p_s_sb, commit_trans_id)) {
4283 flush_commit_list(p_s_sb, jl, 1);
4286 reiserfs_check_lock_depth(p_s_sb, "journal end2");
4288 memset(th, 0, sizeof(*th));
4289 /* Re-set th->t_super, so we can properly keep track of how many
4290 * persistent transactions there are. We need to do this so if this
4291 * call is part of a failed restart_transaction, we can free it later */
4292 th->t_super = p_s_sb;
4294 return journal->j_errno;
4297 static void __reiserfs_journal_abort_hard(struct super_block *sb)
4299 struct reiserfs_journal *journal = SB_JOURNAL(sb);
4300 if (test_bit(J_ABORTED, &journal->j_state))
4303 printk(KERN_CRIT "REISERFS: Aborting journal for filesystem on %s\n",
4304 reiserfs_bdevname(sb));
4306 sb->s_flags |= MS_RDONLY;
4307 set_bit(J_ABORTED, &journal->j_state);
4309 #ifdef CONFIG_REISERFS_CHECK
4314 static void __reiserfs_journal_abort_soft(struct super_block *sb, int errno)
4316 struct reiserfs_journal *journal = SB_JOURNAL(sb);
4317 if (test_bit(J_ABORTED, &journal->j_state))
4320 if (!journal->j_errno)
4321 journal->j_errno = errno;
4323 __reiserfs_journal_abort_hard(sb);
4326 void reiserfs_journal_abort(struct super_block *sb, int errno)
4328 __reiserfs_journal_abort_soft(sb, errno);