1 // SPDX-License-Identifier: GPL-2.0
3 * Write ahead logging implementation copyright Chris Mason 2000
5 * The background commits make this code very interrelated, and
6 * overly complex. I need to rethink things a bit....The major players:
8 * journal_begin -- call with the number of blocks you expect to log.
9 * If the current transaction is too
10 * old, it will block until the current transaction is
11 * finished, and then start a new one.
12 * Usually, your transaction will get joined in with
13 * previous ones for speed.
15 * journal_join -- same as journal_begin, but won't block on the current
16 * transaction regardless of age. Don't ever call
17 * this. Ever. There are only two places it should be
18 * called from, and they are both inside this file.
20 * journal_mark_dirty -- adds blocks into this transaction. clears any flags
21 * that might make them get sent to disk
22 * and then marks them BH_JDirty. Puts the buffer head
23 * into the current transaction hash.
25 * journal_end -- if the current transaction is batchable, it does nothing
26 * otherwise, it could do an async/synchronous commit, or
27 * a full flush of all log and real blocks in the
30 * flush_old_commits -- if the current transaction is too old, it is ended and
31 * commit blocks are sent to disk. Forces commit blocks
32 * to disk for all backgrounded commits that have been
34 * -- Note, if you call this as an immediate flush from
35 * from within kupdate, it will ignore the immediate flag
38 #include <linux/time.h>
39 #include <linux/semaphore.h>
40 #include <linux/vmalloc.h>
42 #include <linux/kernel.h>
43 #include <linux/errno.h>
44 #include <linux/fcntl.h>
45 #include <linux/stat.h>
46 #include <linux/string.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>
53 #include <linux/slab.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 /* must be correct to keep the desc and commit structs at 4k */
63 #define JOURNAL_TRANS_HALF 1018
64 #define BUFNR 64 /*read ahead */
66 /* cnode stat bits. Move these into reiserfs_fs.h */
68 /* this block was freed, and can't be written. */
70 /* this block was freed during this transaction, and can't be written */
71 #define BLOCK_FREED_HOLDER 3
73 /* used in flush_journal_list */
74 #define BLOCK_NEEDS_FLUSH 4
75 #define BLOCK_DIRTIED 5
77 /* journal list state bits */
78 #define LIST_TOUCHED 1
80 #define LIST_COMMIT_PENDING 4 /* someone will commit this list */
82 /* flags for do_journal_end */
83 #define FLUSH_ALL 1 /* flush commit and real blocks */
84 #define COMMIT_NOW 2 /* end and commit this transaction */
85 #define WAIT 4 /* wait for the log blocks to hit the disk */
87 static int do_journal_end(struct reiserfs_transaction_handle *, int flags);
88 static int flush_journal_list(struct super_block *s,
89 struct reiserfs_journal_list *jl, int flushall);
90 static int flush_commit_list(struct super_block *s,
91 struct reiserfs_journal_list *jl, int flushall);
92 static int can_dirty(struct reiserfs_journal_cnode *cn);
93 static int journal_join(struct reiserfs_transaction_handle *th,
94 struct super_block *sb);
95 static void release_journal_dev(struct super_block *super,
96 struct reiserfs_journal *journal);
97 static void dirty_one_transaction(struct super_block *s,
98 struct reiserfs_journal_list *jl);
99 static void flush_async_commits(struct work_struct *work);
100 static void queue_log_writer(struct super_block *s);
102 /* values for join in do_journal_begin_r */
104 JBEGIN_REG = 0, /* regular journal begin */
105 /* join the running transaction if at all possible */
107 /* called from cleanup code, ignores aborted flag */
111 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
112 struct super_block *sb,
113 unsigned long nblocks, int join);
115 static void init_journal_hash(struct super_block *sb)
117 struct reiserfs_journal *journal = SB_JOURNAL(sb);
118 memset(journal->j_hash_table, 0,
119 JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
123 * clears BH_Dirty and sticks the buffer on the clean list. Called because
124 * I can't allow refile_buffer to make schedule happen after I've freed a
125 * block. Look at remove_from_transaction and journal_mark_freed for
128 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
131 clear_buffer_dirty(bh);
132 clear_buffer_journal_test(bh);
137 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
140 struct reiserfs_bitmap_node *bn;
143 bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
147 bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
153 INIT_LIST_HEAD(&bn->list);
157 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
159 struct reiserfs_journal *journal = SB_JOURNAL(sb);
160 struct reiserfs_bitmap_node *bn = NULL;
161 struct list_head *entry = journal->j_bitmap_nodes.next;
163 journal->j_used_bitmap_nodes++;
166 if (entry != &journal->j_bitmap_nodes) {
167 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
169 memset(bn->data, 0, sb->s_blocksize);
170 journal->j_free_bitmap_nodes--;
173 bn = allocate_bitmap_node(sb);
180 static inline void free_bitmap_node(struct super_block *sb,
181 struct reiserfs_bitmap_node *bn)
183 struct reiserfs_journal *journal = SB_JOURNAL(sb);
184 journal->j_used_bitmap_nodes--;
185 if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
189 list_add(&bn->list, &journal->j_bitmap_nodes);
190 journal->j_free_bitmap_nodes++;
194 static void allocate_bitmap_nodes(struct super_block *sb)
197 struct reiserfs_journal *journal = SB_JOURNAL(sb);
198 struct reiserfs_bitmap_node *bn = NULL;
199 for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
200 bn = allocate_bitmap_node(sb);
202 list_add(&bn->list, &journal->j_bitmap_nodes);
203 journal->j_free_bitmap_nodes++;
205 /* this is ok, we'll try again when more are needed */
211 static int set_bit_in_list_bitmap(struct super_block *sb,
213 struct reiserfs_list_bitmap *jb)
215 unsigned int bmap_nr = block / (sb->s_blocksize << 3);
216 unsigned int bit_nr = block % (sb->s_blocksize << 3);
218 if (!jb->bitmaps[bmap_nr]) {
219 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
221 set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
225 static void cleanup_bitmap_list(struct super_block *sb,
226 struct reiserfs_list_bitmap *jb)
229 if (jb->bitmaps == NULL)
232 for (i = 0; i < reiserfs_bmap_count(sb); i++) {
233 if (jb->bitmaps[i]) {
234 free_bitmap_node(sb, jb->bitmaps[i]);
235 jb->bitmaps[i] = NULL;
241 * only call this on FS unmount.
243 static int free_list_bitmaps(struct super_block *sb,
244 struct reiserfs_list_bitmap *jb_array)
247 struct reiserfs_list_bitmap *jb;
248 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
250 jb->journal_list = NULL;
251 cleanup_bitmap_list(sb, jb);
258 static int free_bitmap_nodes(struct super_block *sb)
260 struct reiserfs_journal *journal = SB_JOURNAL(sb);
261 struct list_head *next = journal->j_bitmap_nodes.next;
262 struct reiserfs_bitmap_node *bn;
264 while (next != &journal->j_bitmap_nodes) {
265 bn = list_entry(next, struct reiserfs_bitmap_node, list);
269 next = journal->j_bitmap_nodes.next;
270 journal->j_free_bitmap_nodes--;
277 * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
278 * jb_array is the array to be filled in.
280 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
281 struct reiserfs_list_bitmap *jb_array,
282 unsigned int bmap_nr)
286 struct reiserfs_list_bitmap *jb;
287 int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
289 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
291 jb->journal_list = NULL;
292 jb->bitmaps = vzalloc(mem);
294 reiserfs_warning(sb, "clm-2000", "unable to "
295 "allocate bitmaps for journal lists");
301 free_list_bitmaps(sb, jb_array);
308 * find an available list bitmap. If you can't find one, flush a commit list
311 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
312 struct reiserfs_journal_list
316 struct reiserfs_journal *journal = SB_JOURNAL(sb);
317 struct reiserfs_list_bitmap *jb = NULL;
319 for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
320 i = journal->j_list_bitmap_index;
321 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
322 jb = journal->j_list_bitmap + i;
323 if (journal->j_list_bitmap[i].journal_list) {
324 flush_commit_list(sb,
325 journal->j_list_bitmap[i].
327 if (!journal->j_list_bitmap[i].journal_list) {
334 /* double check to make sure if flushed correctly */
335 if (jb->journal_list)
337 jb->journal_list = jl;
342 * allocates a new chunk of X nodes, and links them all together as a list.
343 * Uses the cnode->next and cnode->prev pointers
344 * returns NULL on failure
346 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
348 struct reiserfs_journal_cnode *head;
350 if (num_cnodes <= 0) {
353 head = vzalloc(array_size(num_cnodes,
354 sizeof(struct reiserfs_journal_cnode)));
359 head[0].next = head + 1;
360 for (i = 1; i < num_cnodes; i++) {
361 head[i].prev = head + (i - 1);
362 head[i].next = head + (i + 1); /* if last one, overwrite it after the if */
364 head[num_cnodes - 1].next = NULL;
368 /* pulls a cnode off the free list, or returns NULL on failure */
369 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
371 struct reiserfs_journal_cnode *cn;
372 struct reiserfs_journal *journal = SB_JOURNAL(sb);
374 reiserfs_check_lock_depth(sb, "get_cnode");
376 if (journal->j_cnode_free <= 0) {
379 journal->j_cnode_used++;
380 journal->j_cnode_free--;
381 cn = journal->j_cnode_free_list;
386 cn->next->prev = NULL;
388 journal->j_cnode_free_list = cn->next;
389 memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
394 * returns a cnode to the free list
396 static void free_cnode(struct super_block *sb,
397 struct reiserfs_journal_cnode *cn)
399 struct reiserfs_journal *journal = SB_JOURNAL(sb);
401 reiserfs_check_lock_depth(sb, "free_cnode");
403 journal->j_cnode_used--;
404 journal->j_cnode_free++;
405 /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
406 cn->next = journal->j_cnode_free_list;
407 if (journal->j_cnode_free_list) {
408 journal->j_cnode_free_list->prev = cn;
410 cn->prev = NULL; /* not needed with the memset, but I might kill the memset, and forget to do this */
411 journal->j_cnode_free_list = cn;
414 static void clear_prepared_bits(struct buffer_head *bh)
416 clear_buffer_journal_prepared(bh);
417 clear_buffer_journal_restore_dirty(bh);
421 * return a cnode with same dev, block number and size in table,
422 * or null if not found
424 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
428 reiserfs_journal_cnode
432 struct reiserfs_journal_cnode *cn;
433 cn = journal_hash(table, sb, bl);
435 if (cn->blocknr == bl && cn->sb == sb)
439 return (struct reiserfs_journal_cnode *)0;
443 * this actually means 'can this block be reallocated yet?'. If you set
444 * search_all, a block can only be allocated if it is not in the current
445 * transaction, was not freed by the current transaction, and has no chance
446 * of ever being overwritten by a replay after crashing.
448 * If you don't set search_all, a block can only be allocated if it is not
449 * in the current transaction. Since deleting a block removes it from the
450 * current transaction, this case should never happen. If you don't set
451 * search_all, make sure you never write the block without logging it.
453 * next_zero_bit is a suggestion about the next block to try for find_forward.
454 * when bl is rejected because it is set in a journal list bitmap, we search
455 * for the next zero bit in the bitmap that rejected bl. Then, we return
456 * that through next_zero_bit for find_forward to try.
458 * Just because we return something in next_zero_bit does not mean we won't
459 * reject it on the next call to reiserfs_in_journal
461 int reiserfs_in_journal(struct super_block *sb,
462 unsigned int bmap_nr, int bit_nr, int search_all,
463 b_blocknr_t * next_zero_bit)
465 struct reiserfs_journal *journal = SB_JOURNAL(sb);
466 struct reiserfs_journal_cnode *cn;
467 struct reiserfs_list_bitmap *jb;
471 *next_zero_bit = 0; /* always start this at zero. */
473 PROC_INFO_INC(sb, journal.in_journal);
475 * If we aren't doing a search_all, this is a metablock, and it
476 * will be logged before use. if we crash before the transaction
477 * that freed it commits, this transaction won't have committed
478 * either, and the block will never be written
481 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
482 PROC_INFO_INC(sb, journal.in_journal_bitmap);
483 jb = journal->j_list_bitmap + i;
484 if (jb->journal_list && jb->bitmaps[bmap_nr] &&
486 (unsigned long *)jb->bitmaps[bmap_nr]->
489 find_next_zero_bit((unsigned long *)
490 (jb->bitmaps[bmap_nr]->
492 sb->s_blocksize << 3,
499 bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
500 /* is it in any old transactions? */
503 get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
507 /* is it in the current transaction. This should never happen */
508 if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
513 PROC_INFO_INC(sb, journal.in_journal_reusable);
518 /* insert cn into table */
519 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
520 struct reiserfs_journal_cnode *cn)
522 struct reiserfs_journal_cnode *cn_orig;
524 cn_orig = journal_hash(table, cn->sb, cn->blocknr);
530 journal_hash(table, cn->sb, cn->blocknr) = cn;
533 /* lock the current transaction */
534 static inline void lock_journal(struct super_block *sb)
536 PROC_INFO_INC(sb, journal.lock_journal);
538 reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
541 /* unlock the current transaction */
542 static inline void unlock_journal(struct super_block *sb)
544 mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
547 static inline void get_journal_list(struct reiserfs_journal_list *jl)
552 static inline void put_journal_list(struct super_block *s,
553 struct reiserfs_journal_list *jl)
555 if (jl->j_refcount < 1) {
556 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
557 jl->j_trans_id, jl->j_refcount);
559 if (--jl->j_refcount == 0)
564 * this used to be much more involved, and I'm keeping it just in case
565 * things get ugly again. it gets called by flush_commit_list, and
566 * cleans up any data stored about blocks freed during a transaction.
568 static void cleanup_freed_for_journal_list(struct super_block *sb,
569 struct reiserfs_journal_list *jl)
572 struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
574 cleanup_bitmap_list(sb, jb);
576 jl->j_list_bitmap->journal_list = NULL;
577 jl->j_list_bitmap = NULL;
580 static int journal_list_still_alive(struct super_block *s,
581 unsigned int trans_id)
583 struct reiserfs_journal *journal = SB_JOURNAL(s);
584 struct list_head *entry = &journal->j_journal_list;
585 struct reiserfs_journal_list *jl;
587 if (!list_empty(entry)) {
588 jl = JOURNAL_LIST_ENTRY(entry->next);
589 if (jl->j_trans_id <= trans_id) {
597 * If page->mapping was null, we failed to truncate this page for
598 * some reason. Most likely because it was truncated after being
599 * logged via data=journal.
601 * This does a check to see if the buffer belongs to one of these
602 * lost pages before doing the final put_bh. If page->mapping was
603 * null, it tries to free buffers on the page, which should make the
604 * final put_page drop the page from the lru.
606 static void release_buffer_page(struct buffer_head *bh)
608 struct page *page = bh->b_page;
609 if (!page->mapping && trylock_page(page)) {
613 try_to_free_buffers(page);
621 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
623 if (buffer_journaled(bh)) {
624 reiserfs_warning(NULL, "clm-2084",
625 "pinned buffer %lu:%pg sent to disk",
626 bh->b_blocknr, bh->b_bdev);
629 set_buffer_uptodate(bh);
631 clear_buffer_uptodate(bh);
634 release_buffer_page(bh);
637 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
640 set_buffer_uptodate(bh);
642 clear_buffer_uptodate(bh);
647 static void submit_logged_buffer(struct buffer_head *bh)
650 bh->b_end_io = reiserfs_end_buffer_io_sync;
651 clear_buffer_journal_new(bh);
652 clear_buffer_dirty(bh);
653 if (!test_clear_buffer_journal_test(bh))
655 if (!buffer_uptodate(bh))
657 submit_bh(REQ_OP_WRITE, 0, bh);
660 static void submit_ordered_buffer(struct buffer_head *bh)
663 bh->b_end_io = reiserfs_end_ordered_io;
664 clear_buffer_dirty(bh);
665 if (!buffer_uptodate(bh))
667 submit_bh(REQ_OP_WRITE, 0, bh);
670 #define CHUNK_SIZE 32
671 struct buffer_chunk {
672 struct buffer_head *bh[CHUNK_SIZE];
676 static void write_chunk(struct buffer_chunk *chunk)
679 for (i = 0; i < chunk->nr; i++) {
680 submit_logged_buffer(chunk->bh[i]);
685 static void write_ordered_chunk(struct buffer_chunk *chunk)
688 for (i = 0; i < chunk->nr; i++) {
689 submit_ordered_buffer(chunk->bh[i]);
694 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
695 spinlock_t * lock, void (fn) (struct buffer_chunk *))
698 BUG_ON(chunk->nr >= CHUNK_SIZE);
699 chunk->bh[chunk->nr++] = bh;
700 if (chunk->nr >= CHUNK_SIZE) {
713 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
714 static struct reiserfs_jh *alloc_jh(void)
716 struct reiserfs_jh *jh;
718 jh = kmalloc(sizeof(*jh), GFP_NOFS);
720 atomic_inc(&nr_reiserfs_jh);
728 * we want to free the jh when the buffer has been written
731 void reiserfs_free_jh(struct buffer_head *bh)
733 struct reiserfs_jh *jh;
737 bh->b_private = NULL;
739 list_del_init(&jh->list);
741 if (atomic_read(&nr_reiserfs_jh) <= 0)
743 atomic_dec(&nr_reiserfs_jh);
748 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
751 struct reiserfs_jh *jh;
754 spin_lock(&j->j_dirty_buffers_lock);
755 if (!bh->b_private) {
756 spin_unlock(&j->j_dirty_buffers_lock);
760 list_del_init(&jh->list);
765 spin_lock(&j->j_dirty_buffers_lock);
767 * buffer must be locked for __add_jh, should be able to have
768 * two adds at the same time
770 BUG_ON(bh->b_private);
774 jh->jl = j->j_current_jl;
776 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
778 list_add_tail(&jh->list, &jh->jl->j_bh_list);
780 spin_unlock(&j->j_dirty_buffers_lock);
784 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
786 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
788 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
790 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
793 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
794 static int write_ordered_buffers(spinlock_t * lock,
795 struct reiserfs_journal *j,
796 struct reiserfs_journal_list *jl,
797 struct list_head *list)
799 struct buffer_head *bh;
800 struct reiserfs_jh *jh;
801 int ret = j->j_errno;
802 struct buffer_chunk chunk;
803 struct list_head tmp;
804 INIT_LIST_HEAD(&tmp);
808 while (!list_empty(list)) {
809 jh = JH_ENTRY(list->next);
812 if (!trylock_buffer(bh)) {
813 if (!buffer_dirty(bh)) {
814 list_move(&jh->list, &tmp);
819 write_ordered_chunk(&chunk);
826 * in theory, dirty non-uptodate buffers should never get here,
827 * but the upper layer io error paths still have a few quirks.
828 * Handle them here as gracefully as we can
830 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
831 clear_buffer_dirty(bh);
834 if (buffer_dirty(bh)) {
835 list_move(&jh->list, &tmp);
836 add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
838 reiserfs_free_jh(bh);
843 cond_resched_lock(lock);
847 write_ordered_chunk(&chunk);
850 while (!list_empty(&tmp)) {
851 jh = JH_ENTRY(tmp.prev);
854 reiserfs_free_jh(bh);
856 if (buffer_locked(bh)) {
861 if (!buffer_uptodate(bh)) {
865 * ugly interaction with invalidatepage here.
866 * reiserfs_invalidate_page will pin any buffer that has a
867 * valid journal head from an older transaction. If someone
868 * else sets our buffer dirty after we write it in the first
869 * loop, and then someone truncates the page away, nobody
870 * will ever write the buffer. We're safe if we write the
871 * page one last time after freeing the journal header.
873 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
875 ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
879 cond_resched_lock(lock);
885 static int flush_older_commits(struct super_block *s,
886 struct reiserfs_journal_list *jl)
888 struct reiserfs_journal *journal = SB_JOURNAL(s);
889 struct reiserfs_journal_list *other_jl;
890 struct reiserfs_journal_list *first_jl;
891 struct list_head *entry;
892 unsigned int trans_id = jl->j_trans_id;
893 unsigned int other_trans_id;
894 unsigned int first_trans_id;
898 * first we walk backwards to find the oldest uncommitted transation
901 entry = jl->j_list.prev;
903 other_jl = JOURNAL_LIST_ENTRY(entry);
904 if (entry == &journal->j_journal_list ||
905 atomic_read(&other_jl->j_older_commits_done))
909 entry = other_jl->j_list.prev;
912 /* if we didn't find any older uncommitted transactions, return now */
913 if (first_jl == jl) {
917 first_trans_id = first_jl->j_trans_id;
919 entry = &first_jl->j_list;
921 other_jl = JOURNAL_LIST_ENTRY(entry);
922 other_trans_id = other_jl->j_trans_id;
924 if (other_trans_id < trans_id) {
925 if (atomic_read(&other_jl->j_commit_left) != 0) {
926 flush_commit_list(s, other_jl, 0);
928 /* list we were called with is gone, return */
929 if (!journal_list_still_alive(s, trans_id))
933 * the one we just flushed is gone, this means
934 * all older lists are also gone, so first_jl
935 * is no longer valid either. Go back to the
938 if (!journal_list_still_alive
939 (s, other_trans_id)) {
944 if (entry == &journal->j_journal_list)
953 static int reiserfs_async_progress_wait(struct super_block *s)
955 struct reiserfs_journal *j = SB_JOURNAL(s);
957 if (atomic_read(&j->j_async_throttle)) {
960 depth = reiserfs_write_unlock_nested(s);
961 congestion_wait(BLK_RW_ASYNC, HZ / 10);
962 reiserfs_write_lock_nested(s, depth);
969 * if this journal list still has commit blocks unflushed, send them to disk.
971 * log areas must be flushed in order (transaction 2 can't commit before
972 * transaction 1) Before the commit block can by written, every other log
973 * block must be safely on disk
975 static int flush_commit_list(struct super_block *s,
976 struct reiserfs_journal_list *jl, int flushall)
980 struct buffer_head *tbh = NULL;
981 unsigned int trans_id = jl->j_trans_id;
982 struct reiserfs_journal *journal = SB_JOURNAL(s);
987 reiserfs_check_lock_depth(s, "flush_commit_list");
989 if (atomic_read(&jl->j_older_commits_done)) {
994 * before we can put our commit blocks on disk, we have to make
995 * sure everyone older than us is on disk too
997 BUG_ON(jl->j_len <= 0);
998 BUG_ON(trans_id == journal->j_trans_id);
1000 get_journal_list(jl);
1002 if (flush_older_commits(s, jl) == 1) {
1004 * list disappeared during flush_older_commits.
1011 /* make sure nobody is trying to flush this one at the same time */
1012 reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1014 if (!journal_list_still_alive(s, trans_id)) {
1015 mutex_unlock(&jl->j_commit_mutex);
1018 BUG_ON(jl->j_trans_id == 0);
1020 /* this commit is done, exit */
1021 if (atomic_read(&jl->j_commit_left) <= 0) {
1023 atomic_set(&jl->j_older_commits_done, 1);
1025 mutex_unlock(&jl->j_commit_mutex);
1029 if (!list_empty(&jl->j_bh_list)) {
1033 * We might sleep in numerous places inside
1034 * write_ordered_buffers. Relax the write lock.
1036 depth = reiserfs_write_unlock_nested(s);
1037 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1038 journal, jl, &jl->j_bh_list);
1039 if (ret < 0 && retval == 0)
1041 reiserfs_write_lock_nested(s, depth);
1043 BUG_ON(!list_empty(&jl->j_bh_list));
1045 * for the description block and all the log blocks, submit any buffers
1046 * that haven't already reached the disk. Try to write at least 256
1047 * log blocks. later on, we will only wait on blocks that correspond
1048 * to this transaction, but while we're unplugging we might as well
1049 * get a chunk of data on there.
1051 atomic_inc(&journal->j_async_throttle);
1052 write_len = jl->j_len + 1;
1053 if (write_len < 256)
1055 for (i = 0 ; i < write_len ; i++) {
1056 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1057 SB_ONDISK_JOURNAL_SIZE(s);
1058 tbh = journal_find_get_block(s, bn);
1060 if (buffer_dirty(tbh)) {
1061 depth = reiserfs_write_unlock_nested(s);
1062 ll_rw_block(REQ_OP_WRITE, 0, 1, &tbh);
1063 reiserfs_write_lock_nested(s, depth);
1068 atomic_dec(&journal->j_async_throttle);
1070 for (i = 0; i < (jl->j_len + 1); i++) {
1071 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1072 (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1073 tbh = journal_find_get_block(s, bn);
1075 depth = reiserfs_write_unlock_nested(s);
1076 __wait_on_buffer(tbh);
1077 reiserfs_write_lock_nested(s, depth);
1079 * since we're using ll_rw_blk above, it might have skipped
1080 * over a locked buffer. Double check here
1082 /* redundant, sync_dirty_buffer() checks */
1083 if (buffer_dirty(tbh)) {
1084 depth = reiserfs_write_unlock_nested(s);
1085 sync_dirty_buffer(tbh);
1086 reiserfs_write_lock_nested(s, depth);
1088 if (unlikely(!buffer_uptodate(tbh))) {
1089 #ifdef CONFIG_REISERFS_CHECK
1090 reiserfs_warning(s, "journal-601",
1091 "buffer write failed");
1095 /* once for journal_find_get_block */
1097 /* once due to original getblk in do_journal_end */
1099 atomic_dec(&jl->j_commit_left);
1102 BUG_ON(atomic_read(&jl->j_commit_left) != 1);
1105 * If there was a write error in the journal - we can't commit
1106 * this transaction - it will be invalid and, if successful,
1107 * will just end up propagating the write error out to
1110 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1111 if (buffer_dirty(jl->j_commit_bh))
1113 mark_buffer_dirty(jl->j_commit_bh) ;
1114 depth = reiserfs_write_unlock_nested(s);
1115 if (reiserfs_barrier_flush(s))
1116 __sync_dirty_buffer(jl->j_commit_bh,
1117 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1119 sync_dirty_buffer(jl->j_commit_bh);
1120 reiserfs_write_lock_nested(s, depth);
1124 * If there was a write error in the journal - we can't commit this
1125 * transaction - it will be invalid and, if successful, will just end
1126 * up propagating the write error out to the filesystem.
1128 if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1129 #ifdef CONFIG_REISERFS_CHECK
1130 reiserfs_warning(s, "journal-615", "buffer write failed");
1134 bforget(jl->j_commit_bh);
1135 if (journal->j_last_commit_id != 0 &&
1136 (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1137 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1138 journal->j_last_commit_id, jl->j_trans_id);
1140 journal->j_last_commit_id = jl->j_trans_id;
1143 * now, every commit block is on the disk. It is safe to allow
1144 * blocks freed during this transaction to be reallocated
1146 cleanup_freed_for_journal_list(s, jl);
1148 retval = retval ? retval : journal->j_errno;
1150 /* mark the metadata dirty */
1152 dirty_one_transaction(s, jl);
1153 atomic_dec(&jl->j_commit_left);
1156 atomic_set(&jl->j_older_commits_done, 1);
1158 mutex_unlock(&jl->j_commit_mutex);
1160 put_journal_list(s, jl);
1163 reiserfs_abort(s, retval, "Journal write error in %s",
1169 * flush_journal_list frequently needs to find a newer transaction for a
1170 * given block. This does that, or returns NULL if it can't find anything
1172 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1173 reiserfs_journal_cnode
1176 struct super_block *sb = cn->sb;
1177 b_blocknr_t blocknr = cn->blocknr;
1181 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1189 static void remove_journal_hash(struct super_block *,
1190 struct reiserfs_journal_cnode **,
1191 struct reiserfs_journal_list *, unsigned long,
1195 * once all the real blocks have been flushed, it is safe to remove them
1196 * from the journal list for this transaction. Aside from freeing the
1197 * cnode, this also allows the block to be reallocated for data blocks
1198 * if it had been deleted.
1200 static void remove_all_from_journal_list(struct super_block *sb,
1201 struct reiserfs_journal_list *jl,
1204 struct reiserfs_journal *journal = SB_JOURNAL(sb);
1205 struct reiserfs_journal_cnode *cn, *last;
1206 cn = jl->j_realblock;
1209 * which is better, to lock once around the whole loop, or
1210 * to lock for each call to remove_journal_hash?
1213 if (cn->blocknr != 0) {
1215 reiserfs_warning(sb, "reiserfs-2201",
1216 "block %u, bh is %d, state %ld",
1217 cn->blocknr, cn->bh ? 1 : 0,
1221 remove_journal_hash(sb, journal->j_list_hash_table,
1222 jl, cn->blocknr, 1);
1226 free_cnode(sb, last);
1228 jl->j_realblock = NULL;
1232 * if this timestamp is greater than the timestamp we wrote last to the
1233 * header block, write it to the header block. once this is done, I can
1234 * safely say the log area for this transaction won't ever be replayed,
1235 * and I can start releasing blocks in this transaction for reuse as data
1236 * blocks. called by flush_journal_list, before it calls
1237 * remove_all_from_journal_list
1239 static int _update_journal_header_block(struct super_block *sb,
1240 unsigned long offset,
1241 unsigned int trans_id)
1243 struct reiserfs_journal_header *jh;
1244 struct reiserfs_journal *journal = SB_JOURNAL(sb);
1247 if (reiserfs_is_journal_aborted(journal))
1250 if (trans_id >= journal->j_last_flush_trans_id) {
1251 if (buffer_locked((journal->j_header_bh))) {
1252 depth = reiserfs_write_unlock_nested(sb);
1253 __wait_on_buffer(journal->j_header_bh);
1254 reiserfs_write_lock_nested(sb, depth);
1255 if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1256 #ifdef CONFIG_REISERFS_CHECK
1257 reiserfs_warning(sb, "journal-699",
1258 "buffer write failed");
1263 journal->j_last_flush_trans_id = trans_id;
1264 journal->j_first_unflushed_offset = offset;
1265 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1267 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1268 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1269 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1271 set_buffer_dirty(journal->j_header_bh);
1272 depth = reiserfs_write_unlock_nested(sb);
1274 if (reiserfs_barrier_flush(sb))
1275 __sync_dirty_buffer(journal->j_header_bh,
1276 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1278 sync_dirty_buffer(journal->j_header_bh);
1280 reiserfs_write_lock_nested(sb, depth);
1281 if (!buffer_uptodate(journal->j_header_bh)) {
1282 reiserfs_warning(sb, "journal-837",
1283 "IO error during journal replay");
1290 static int update_journal_header_block(struct super_block *sb,
1291 unsigned long offset,
1292 unsigned int trans_id)
1294 return _update_journal_header_block(sb, offset, trans_id);
1298 ** flush any and all journal lists older than you are
1299 ** can only be called from flush_journal_list
1301 static int flush_older_journal_lists(struct super_block *sb,
1302 struct reiserfs_journal_list *jl)
1304 struct list_head *entry;
1305 struct reiserfs_journal_list *other_jl;
1306 struct reiserfs_journal *journal = SB_JOURNAL(sb);
1307 unsigned int trans_id = jl->j_trans_id;
1310 * we know we are the only ones flushing things, no extra race
1311 * protection is required.
1314 entry = journal->j_journal_list.next;
1316 if (entry == &journal->j_journal_list)
1318 other_jl = JOURNAL_LIST_ENTRY(entry);
1319 if (other_jl->j_trans_id < trans_id) {
1320 BUG_ON(other_jl->j_refcount <= 0);
1321 /* do not flush all */
1322 flush_journal_list(sb, other_jl, 0);
1324 /* other_jl is now deleted from the list */
1330 static void del_from_work_list(struct super_block *s,
1331 struct reiserfs_journal_list *jl)
1333 struct reiserfs_journal *journal = SB_JOURNAL(s);
1334 if (!list_empty(&jl->j_working_list)) {
1335 list_del_init(&jl->j_working_list);
1336 journal->j_num_work_lists--;
1341 * flush a journal list, both commit and real blocks
1343 * always set flushall to 1, unless you are calling from inside
1344 * flush_journal_list
1346 * IMPORTANT. This can only be called while there are no journal writers,
1347 * and the journal is locked. That means it can only be called from
1348 * do_journal_end, or by journal_release
1350 static int flush_journal_list(struct super_block *s,
1351 struct reiserfs_journal_list *jl, int flushall)
1353 struct reiserfs_journal_list *pjl;
1354 struct reiserfs_journal_cnode *cn, *last;
1358 struct buffer_head *saved_bh;
1359 unsigned long j_len_saved = jl->j_len;
1360 struct reiserfs_journal *journal = SB_JOURNAL(s);
1364 BUG_ON(j_len_saved <= 0);
1366 if (atomic_read(&journal->j_wcount) != 0) {
1367 reiserfs_warning(s, "clm-2048", "called with wcount %d",
1368 atomic_read(&journal->j_wcount));
1371 /* if flushall == 0, the lock is already held */
1373 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1374 } else if (mutex_trylock(&journal->j_flush_mutex)) {
1379 if (j_len_saved > journal->j_trans_max) {
1380 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1381 j_len_saved, jl->j_trans_id);
1385 /* if all the work is already done, get out of here */
1386 if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1387 atomic_read(&jl->j_commit_left) <= 0) {
1388 goto flush_older_and_return;
1392 * start by putting the commit list on disk. This will also flush
1393 * the commit lists of any olders transactions
1395 flush_commit_list(s, jl, 1);
1397 if (!(jl->j_state & LIST_DIRTY)
1398 && !reiserfs_is_journal_aborted(journal))
1401 /* are we done now? */
1402 if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1403 atomic_read(&jl->j_commit_left) <= 0) {
1404 goto flush_older_and_return;
1408 * loop through each cnode, see if we need to write it,
1409 * or wait on a more recent transaction, or just ignore it
1411 if (atomic_read(&journal->j_wcount) != 0) {
1412 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1415 cn = jl->j_realblock;
1420 /* blocknr of 0 is no longer in the hash, ignore it */
1421 if (cn->blocknr == 0) {
1426 * This transaction failed commit.
1427 * Don't write out to the disk
1429 if (!(jl->j_state & LIST_DIRTY))
1432 pjl = find_newer_jl_for_cn(cn);
1434 * the order is important here. We check pjl to make sure we
1435 * don't clear BH_JDirty_wait if we aren't the one writing this
1438 if (!pjl && cn->bh) {
1442 * we do this to make sure nobody releases the
1443 * buffer while we are working with it
1447 if (buffer_journal_dirty(saved_bh)) {
1448 BUG_ON(!can_dirty(cn));
1451 } else if (can_dirty(cn)) {
1453 * everything with !pjl && jwait
1454 * should be writable
1461 * if someone has this block in a newer transaction, just make
1462 * sure they are committed, and don't try writing it to disk
1465 if (atomic_read(&pjl->j_commit_left))
1466 flush_commit_list(s, pjl, 1);
1471 * bh == NULL when the block got to disk on its own, OR,
1472 * the block got freed in a future transaction
1474 if (saved_bh == NULL) {
1479 * this should never happen. kupdate_one_transaction has
1480 * this list locked while it works, so we should never see a
1481 * buffer here that is not marked JDirty_wait
1483 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1484 reiserfs_warning(s, "journal-813",
1485 "BAD! buffer %llu %cdirty %cjwait, "
1486 "not in a newer transaction",
1487 (unsigned long long)saved_bh->
1488 b_blocknr, was_dirty ? ' ' : '!',
1489 was_jwait ? ' ' : '!');
1493 * we inc again because saved_bh gets decremented
1497 set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1498 lock_buffer(saved_bh);
1499 BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1500 if (buffer_dirty(saved_bh))
1501 submit_logged_buffer(saved_bh);
1503 unlock_buffer(saved_bh);
1506 reiserfs_warning(s, "clm-2082",
1507 "Unable to flush buffer %llu in %s",
1508 (unsigned long long)saved_bh->
1509 b_blocknr, __func__);
1516 * we incremented this to keep others from
1517 * taking the buffer head away
1520 if (atomic_read(&saved_bh->b_count) < 0) {
1521 reiserfs_warning(s, "journal-945",
1522 "saved_bh->b_count < 0");
1527 cn = jl->j_realblock;
1529 if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1531 reiserfs_panic(s, "journal-1011",
1535 depth = reiserfs_write_unlock_nested(s);
1536 __wait_on_buffer(cn->bh);
1537 reiserfs_write_lock_nested(s, depth);
1540 reiserfs_panic(s, "journal-1012",
1543 if (unlikely(!buffer_uptodate(cn->bh))) {
1544 #ifdef CONFIG_REISERFS_CHECK
1545 reiserfs_warning(s, "journal-949",
1546 "buffer write failed");
1551 * note, we must clear the JDirty_wait bit
1552 * after the up to date check, otherwise we
1553 * race against our flushpage routine
1555 BUG_ON(!test_clear_buffer_journal_dirty
1558 /* drop one ref for us */
1560 /* drop one ref for journal_mark_dirty */
1561 release_buffer_page(cn->bh);
1568 reiserfs_abort(s, -EIO,
1569 "Write error while pushing transaction to disk in %s",
1571 flush_older_and_return:
1574 * before we can update the journal header block, we _must_ flush all
1575 * real blocks from all older transactions to disk. This is because
1576 * once the header block is updated, this transaction will not be
1577 * replayed after a crash
1580 flush_older_journal_lists(s, jl);
1583 err = journal->j_errno;
1585 * before we can remove everything from the hash tables for this
1586 * transaction, we must make sure it can never be replayed
1588 * since we are only called from do_journal_end, we know for sure there
1589 * are no allocations going on while we are flushing journal lists. So,
1590 * we only need to update the journal header block for the last list
1593 if (!err && flushall) {
1595 update_journal_header_block(s,
1596 (jl->j_start + jl->j_len +
1597 2) % SB_ONDISK_JOURNAL_SIZE(s),
1600 reiserfs_abort(s, -EIO,
1601 "Write error while updating journal header in %s",
1604 remove_all_from_journal_list(s, jl, 0);
1605 list_del_init(&jl->j_list);
1606 journal->j_num_lists--;
1607 del_from_work_list(s, jl);
1609 if (journal->j_last_flush_id != 0 &&
1610 (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1611 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1612 journal->j_last_flush_id, jl->j_trans_id);
1614 journal->j_last_flush_id = jl->j_trans_id;
1617 * not strictly required since we are freeing the list, but it should
1618 * help find code using dead lists later on
1621 atomic_set(&jl->j_nonzerolen, 0);
1623 jl->j_realblock = NULL;
1624 jl->j_commit_bh = NULL;
1627 put_journal_list(s, jl);
1629 mutex_unlock(&journal->j_flush_mutex);
1633 static int write_one_transaction(struct super_block *s,
1634 struct reiserfs_journal_list *jl,
1635 struct buffer_chunk *chunk)
1637 struct reiserfs_journal_cnode *cn;
1640 jl->j_state |= LIST_TOUCHED;
1641 del_from_work_list(s, jl);
1642 if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1646 cn = jl->j_realblock;
1649 * if the blocknr == 0, this has been cleared from the hash,
1652 if (cn->blocknr == 0) {
1655 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1656 struct buffer_head *tmp_bh;
1658 * we can race against journal_mark_freed when we try
1659 * to lock_buffer(cn->bh), so we have to inc the buffer
1660 * count, and recheck things after locking
1664 lock_buffer(tmp_bh);
1665 if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1666 if (!buffer_journal_dirty(tmp_bh) ||
1667 buffer_journal_prepared(tmp_bh))
1669 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1672 /* note, cn->bh might be null now */
1673 unlock_buffer(tmp_bh);
1684 /* used by flush_commit_list */
1685 static void dirty_one_transaction(struct super_block *s,
1686 struct reiserfs_journal_list *jl)
1688 struct reiserfs_journal_cnode *cn;
1689 struct reiserfs_journal_list *pjl;
1691 jl->j_state |= LIST_DIRTY;
1692 cn = jl->j_realblock;
1695 * look for a more recent transaction that logged this
1696 * buffer. Only the most recent transaction with a buffer in
1697 * it is allowed to send that buffer to disk
1699 pjl = find_newer_jl_for_cn(cn);
1700 if (!pjl && cn->blocknr && cn->bh
1701 && buffer_journal_dirty(cn->bh)) {
1702 BUG_ON(!can_dirty(cn));
1704 * if the buffer is prepared, it will either be logged
1705 * or restored. If restored, we need to make sure
1706 * it actually gets marked dirty
1708 clear_buffer_journal_new(cn->bh);
1709 if (buffer_journal_prepared(cn->bh)) {
1710 set_buffer_journal_restore_dirty(cn->bh);
1712 set_buffer_journal_test(cn->bh);
1713 mark_buffer_dirty(cn->bh);
1720 static int kupdate_transactions(struct super_block *s,
1721 struct reiserfs_journal_list *jl,
1722 struct reiserfs_journal_list **next_jl,
1723 unsigned int *next_trans_id,
1724 int num_blocks, int num_trans)
1728 int transactions_flushed = 0;
1729 unsigned int orig_trans_id = jl->j_trans_id;
1730 struct buffer_chunk chunk;
1731 struct list_head *entry;
1732 struct reiserfs_journal *journal = SB_JOURNAL(s);
1735 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1736 if (!journal_list_still_alive(s, orig_trans_id)) {
1741 * we've got j_flush_mutex held, nobody is going to delete any
1742 * of these lists out from underneath us
1744 while ((num_trans && transactions_flushed < num_trans) ||
1745 (!num_trans && written < num_blocks)) {
1747 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1748 atomic_read(&jl->j_commit_left)
1749 || !(jl->j_state & LIST_DIRTY)) {
1750 del_from_work_list(s, jl);
1753 ret = write_one_transaction(s, jl, &chunk);
1757 transactions_flushed++;
1759 entry = jl->j_list.next;
1762 if (entry == &journal->j_journal_list) {
1765 jl = JOURNAL_LIST_ENTRY(entry);
1767 /* don't bother with older transactions */
1768 if (jl->j_trans_id <= orig_trans_id)
1772 write_chunk(&chunk);
1776 mutex_unlock(&journal->j_flush_mutex);
1781 * for o_sync and fsync heavy applications, they tend to use
1782 * all the journa list slots with tiny transactions. These
1783 * trigger lots and lots of calls to update the header block, which
1784 * adds seeks and slows things down.
1786 * This function tries to clear out a large chunk of the journal lists
1787 * at once, which makes everything faster since only the newest journal
1788 * list updates the header block
1790 static int flush_used_journal_lists(struct super_block *s,
1791 struct reiserfs_journal_list *jl)
1793 unsigned long len = 0;
1794 unsigned long cur_len;
1798 struct reiserfs_journal_list *tjl;
1799 struct reiserfs_journal_list *flush_jl;
1800 unsigned int trans_id;
1801 struct reiserfs_journal *journal = SB_JOURNAL(s);
1803 flush_jl = tjl = jl;
1805 /* in data logging mode, try harder to flush a lot of blocks */
1806 if (reiserfs_data_log(s))
1808 /* flush for 256 transactions or limit blocks, whichever comes first */
1809 for (i = 0; i < 256 && len < limit; i++) {
1810 if (atomic_read(&tjl->j_commit_left) ||
1811 tjl->j_trans_id < jl->j_trans_id) {
1814 cur_len = atomic_read(&tjl->j_nonzerolen);
1816 tjl->j_state &= ~LIST_TOUCHED;
1820 if (tjl->j_list.next == &journal->j_journal_list)
1822 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1824 get_journal_list(jl);
1825 get_journal_list(flush_jl);
1828 * try to find a group of blocks we can flush across all the
1829 * transactions, but only bother if we've actually spanned
1830 * across multiple lists
1832 if (flush_jl != jl) {
1833 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1835 flush_journal_list(s, flush_jl, 1);
1836 put_journal_list(s, flush_jl);
1837 put_journal_list(s, jl);
1842 * removes any nodes in table with name block and dev as bh.
1843 * only touchs the hnext and hprev pointers.
1845 static void remove_journal_hash(struct super_block *sb,
1846 struct reiserfs_journal_cnode **table,
1847 struct reiserfs_journal_list *jl,
1848 unsigned long block, int remove_freed)
1850 struct reiserfs_journal_cnode *cur;
1851 struct reiserfs_journal_cnode **head;
1853 head = &(journal_hash(table, sb, block));
1859 if (cur->blocknr == block && cur->sb == sb
1860 && (jl == NULL || jl == cur->jlist)
1861 && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1863 cur->hnext->hprev = cur->hprev;
1866 cur->hprev->hnext = cur->hnext;
1874 * anybody who clears the cur->bh will also
1875 * dec the nonzerolen
1877 if (cur->bh && cur->jlist)
1878 atomic_dec(&cur->jlist->j_nonzerolen);
1886 static void free_journal_ram(struct super_block *sb)
1888 struct reiserfs_journal *journal = SB_JOURNAL(sb);
1889 kfree(journal->j_current_jl);
1890 journal->j_num_lists--;
1892 vfree(journal->j_cnode_free_orig);
1893 free_list_bitmaps(sb, journal->j_list_bitmap);
1894 free_bitmap_nodes(sb); /* must be after free_list_bitmaps */
1895 if (journal->j_header_bh) {
1896 brelse(journal->j_header_bh);
1899 * j_header_bh is on the journal dev, make sure
1900 * not to release the journal dev until we brelse j_header_bh
1902 release_journal_dev(sb, journal);
1907 * call on unmount. Only set error to 1 if you haven't made your way out
1908 * of read_super() yet. Any other caller must keep error at 0.
1910 static int do_journal_release(struct reiserfs_transaction_handle *th,
1911 struct super_block *sb, int error)
1913 struct reiserfs_transaction_handle myth;
1915 struct reiserfs_journal *journal = SB_JOURNAL(sb);
1918 * we only want to flush out transactions if we were
1919 * called with error == 0
1921 if (!error && !sb_rdonly(sb)) {
1922 /* end the current trans */
1923 BUG_ON(!th->t_trans_id);
1924 do_journal_end(th, FLUSH_ALL);
1927 * make sure something gets logged to force
1928 * our way into the flush code
1930 if (!journal_join(&myth, sb)) {
1931 reiserfs_prepare_for_journal(sb,
1932 SB_BUFFER_WITH_SB(sb),
1934 journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1935 do_journal_end(&myth, FLUSH_ALL);
1940 /* this also catches errors during the do_journal_end above */
1941 if (!error && reiserfs_is_journal_aborted(journal)) {
1942 memset(&myth, 0, sizeof(myth));
1943 if (!journal_join_abort(&myth, sb)) {
1944 reiserfs_prepare_for_journal(sb,
1945 SB_BUFFER_WITH_SB(sb),
1947 journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1948 do_journal_end(&myth, FLUSH_ALL);
1954 * We must release the write lock here because
1955 * the workqueue job (flush_async_commit) needs this lock
1957 reiserfs_write_unlock(sb);
1960 * Cancel flushing of old commits. Note that neither of these works
1961 * will be requeued because superblock is being shutdown and doesn't
1962 * have SB_ACTIVE set.
1964 reiserfs_cancel_old_flush(sb);
1965 /* wait for all commits to finish */
1966 cancel_delayed_work_sync(&SB_JOURNAL(sb)->j_work);
1968 free_journal_ram(sb);
1970 reiserfs_write_lock(sb);
1975 /* * call on unmount. flush all journal trans, release all alloc'd ram */
1976 int journal_release(struct reiserfs_transaction_handle *th,
1977 struct super_block *sb)
1979 return do_journal_release(th, sb, 0);
1982 /* only call from an error condition inside reiserfs_read_super! */
1983 int journal_release_error(struct reiserfs_transaction_handle *th,
1984 struct super_block *sb)
1986 return do_journal_release(th, sb, 1);
1990 * compares description block with commit block.
1991 * returns 1 if they differ, 0 if they are the same
1993 static int journal_compare_desc_commit(struct super_block *sb,
1994 struct reiserfs_journal_desc *desc,
1995 struct reiserfs_journal_commit *commit)
1997 if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1998 get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1999 get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
2000 get_commit_trans_len(commit) <= 0) {
2007 * returns 0 if it did not find a description block
2008 * returns -1 if it found a corrupt commit block
2009 * returns 1 if both desc and commit were valid
2010 * NOTE: only called during fs mount
2012 static int journal_transaction_is_valid(struct super_block *sb,
2013 struct buffer_head *d_bh,
2014 unsigned int *oldest_invalid_trans_id,
2015 unsigned long *newest_mount_id)
2017 struct reiserfs_journal_desc *desc;
2018 struct reiserfs_journal_commit *commit;
2019 struct buffer_head *c_bh;
2020 unsigned long offset;
2025 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2026 if (get_desc_trans_len(desc) > 0
2027 && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2028 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2029 && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2030 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2031 "journal-986: transaction "
2032 "is valid returning because trans_id %d is greater than "
2033 "oldest_invalid %lu",
2034 get_desc_trans_id(desc),
2035 *oldest_invalid_trans_id);
2039 && *newest_mount_id > get_desc_mount_id(desc)) {
2040 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2041 "journal-1087: transaction "
2042 "is valid returning because mount_id %d is less than "
2043 "newest_mount_id %lu",
2044 get_desc_mount_id(desc),
2048 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2049 reiserfs_warning(sb, "journal-2018",
2050 "Bad transaction length %d "
2051 "encountered, ignoring transaction",
2052 get_desc_trans_len(desc));
2055 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2058 * ok, we have a journal description block,
2059 * let's see if the transaction was valid
2063 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2064 ((offset + get_desc_trans_len(desc) +
2065 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2068 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2069 if (journal_compare_desc_commit(sb, desc, commit)) {
2070 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2071 "journal_transaction_is_valid, commit offset %ld had bad "
2072 "time %d or length %d",
2074 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2075 get_commit_trans_id(commit),
2076 get_commit_trans_len(commit));
2078 if (oldest_invalid_trans_id) {
2079 *oldest_invalid_trans_id =
2080 get_desc_trans_id(desc);
2081 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2083 "transaction_is_valid setting oldest invalid trans_id "
2085 get_desc_trans_id(desc));
2090 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2091 "journal-1006: found valid "
2092 "transaction start offset %llu, len %d id %d",
2094 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2095 get_desc_trans_len(desc),
2096 get_desc_trans_id(desc));
2103 static void brelse_array(struct buffer_head **heads, int num)
2106 for (i = 0; i < num; i++) {
2112 * given the start, and values for the oldest acceptable transactions,
2113 * this either reads in a replays a transaction, or returns because the
2114 * transaction is invalid, or too old.
2115 * NOTE: only called during fs mount
2117 static int journal_read_transaction(struct super_block *sb,
2118 unsigned long cur_dblock,
2119 unsigned long oldest_start,
2120 unsigned int oldest_trans_id,
2121 unsigned long newest_mount_id)
2123 struct reiserfs_journal *journal = SB_JOURNAL(sb);
2124 struct reiserfs_journal_desc *desc;
2125 struct reiserfs_journal_commit *commit;
2126 unsigned int trans_id = 0;
2127 struct buffer_head *c_bh;
2128 struct buffer_head *d_bh;
2129 struct buffer_head **log_blocks = NULL;
2130 struct buffer_head **real_blocks = NULL;
2131 unsigned int trans_offset;
2135 d_bh = journal_bread(sb, cur_dblock);
2138 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2139 trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2140 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2141 "journal_read_transaction, offset %llu, len %d mount_id %d",
2142 d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2143 get_desc_trans_len(desc), get_desc_mount_id(desc));
2144 if (get_desc_trans_id(desc) < oldest_trans_id) {
2145 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2146 "journal_read_trans skipping because %lu is too old",
2148 SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2152 if (get_desc_mount_id(desc) != newest_mount_id) {
2153 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2154 "journal_read_trans skipping because %d is != "
2155 "newest_mount_id %lu", get_desc_mount_id(desc),
2160 c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2161 ((trans_offset + get_desc_trans_len(desc) + 1) %
2162 SB_ONDISK_JOURNAL_SIZE(sb)));
2167 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2168 if (journal_compare_desc_commit(sb, desc, commit)) {
2169 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2170 "journal_read_transaction, "
2171 "commit offset %llu had bad time %d or length %d",
2173 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2174 get_commit_trans_id(commit),
2175 get_commit_trans_len(commit));
2181 if (bdev_read_only(sb->s_bdev)) {
2182 reiserfs_warning(sb, "clm-2076",
2183 "device is readonly, unable to replay log");
2189 trans_id = get_desc_trans_id(desc);
2191 * now we know we've got a good transaction, and it was
2192 * inside the valid time ranges
2194 log_blocks = kmalloc_array(get_desc_trans_len(desc),
2195 sizeof(struct buffer_head *),
2197 real_blocks = kmalloc_array(get_desc_trans_len(desc),
2198 sizeof(struct buffer_head *),
2200 if (!log_blocks || !real_blocks) {
2205 reiserfs_warning(sb, "journal-1169",
2206 "kmalloc failed, unable to mount FS");
2209 /* get all the buffer heads */
2210 trans_half = journal_trans_half(sb->s_blocksize);
2211 for (i = 0; i < get_desc_trans_len(desc); i++) {
2214 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2216 i) % SB_ONDISK_JOURNAL_SIZE(sb));
2217 if (i < trans_half) {
2220 le32_to_cpu(desc->j_realblock[i]));
2224 le32_to_cpu(commit->
2225 j_realblock[i - trans_half]));
2227 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2228 reiserfs_warning(sb, "journal-1207",
2229 "REPLAY FAILURE fsck required! "
2230 "Block to replay is outside of "
2234 /* make sure we don't try to replay onto log or reserved area */
2235 if (is_block_in_log_or_reserved_area
2236 (sb, real_blocks[i]->b_blocknr)) {
2237 reiserfs_warning(sb, "journal-1204",
2238 "REPLAY FAILURE fsck required! "
2239 "Trying to replay onto a log block");
2241 brelse_array(log_blocks, i);
2242 brelse_array(real_blocks, i);
2250 /* read in the log blocks, memcpy to the corresponding real block */
2251 ll_rw_block(REQ_OP_READ, 0, get_desc_trans_len(desc), log_blocks);
2252 for (i = 0; i < get_desc_trans_len(desc); i++) {
2254 wait_on_buffer(log_blocks[i]);
2255 if (!buffer_uptodate(log_blocks[i])) {
2256 reiserfs_warning(sb, "journal-1212",
2257 "REPLAY FAILURE fsck required! "
2258 "buffer write failed");
2259 brelse_array(log_blocks + i,
2260 get_desc_trans_len(desc) - i);
2261 brelse_array(real_blocks, get_desc_trans_len(desc));
2268 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2269 real_blocks[i]->b_size);
2270 set_buffer_uptodate(real_blocks[i]);
2271 brelse(log_blocks[i]);
2273 /* flush out the real blocks */
2274 for (i = 0; i < get_desc_trans_len(desc); i++) {
2275 set_buffer_dirty(real_blocks[i]);
2276 write_dirty_buffer(real_blocks[i], 0);
2278 for (i = 0; i < get_desc_trans_len(desc); i++) {
2279 wait_on_buffer(real_blocks[i]);
2280 if (!buffer_uptodate(real_blocks[i])) {
2281 reiserfs_warning(sb, "journal-1226",
2282 "REPLAY FAILURE, fsck required! "
2283 "buffer write failed");
2284 brelse_array(real_blocks + i,
2285 get_desc_trans_len(desc) - i);
2292 brelse(real_blocks[i]);
2295 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2296 ((trans_offset + get_desc_trans_len(desc) +
2297 2) % SB_ONDISK_JOURNAL_SIZE(sb));
2298 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2299 "journal-1095: setting journal " "start to offset %ld",
2300 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2303 * init starting values for the first transaction, in case
2304 * this is the last transaction to be replayed.
2306 journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2307 journal->j_last_flush_trans_id = trans_id;
2308 journal->j_trans_id = trans_id + 1;
2309 /* check for trans_id overflow */
2310 if (journal->j_trans_id == 0)
2311 journal->j_trans_id = 10;
2320 * This function reads blocks starting from block and to max_block of bufsize
2321 * size (but no more than BUFNR blocks at a time). This proved to improve
2322 * mounting speed on self-rebuilding raid5 arrays at least.
2323 * Right now it is only used from journal code. But later we might use it
2324 * from other places.
2325 * Note: Do not use journal_getblk/sb_getblk functions here!
2327 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2328 b_blocknr_t block, int bufsize,
2329 b_blocknr_t max_block)
2331 struct buffer_head *bhlist[BUFNR];
2332 unsigned int blocks = BUFNR;
2333 struct buffer_head *bh;
2336 bh = __getblk(dev, block, bufsize);
2337 if (buffer_uptodate(bh))
2340 if (block + BUFNR > max_block) {
2341 blocks = max_block - block;
2345 for (i = 1; i < blocks; i++) {
2346 bh = __getblk(dev, block + i, bufsize);
2347 if (buffer_uptodate(bh)) {
2353 ll_rw_block(REQ_OP_READ, 0, j, bhlist);
2354 for (i = 1; i < j; i++)
2358 if (buffer_uptodate(bh))
2365 * read and replay the log
2366 * on a clean unmount, the journal header's next unflushed pointer will be
2367 * to an invalid transaction. This tests that before finding all the
2368 * transactions in the log, which makes normal mount times fast.
2370 * After a crash, this starts with the next unflushed transaction, and
2371 * replays until it finds one too old, or invalid.
2373 * On exit, it sets things up so the first transaction will work correctly.
2374 * NOTE: only called during fs mount
2376 static int journal_read(struct super_block *sb)
2378 struct reiserfs_journal *journal = SB_JOURNAL(sb);
2379 struct reiserfs_journal_desc *desc;
2380 unsigned int oldest_trans_id = 0;
2381 unsigned int oldest_invalid_trans_id = 0;
2383 unsigned long oldest_start = 0;
2384 unsigned long cur_dblock = 0;
2385 unsigned long newest_mount_id = 9;
2386 struct buffer_head *d_bh;
2387 struct reiserfs_journal_header *jh;
2388 int valid_journal_header = 0;
2389 int replay_count = 0;
2390 int continue_replay = 1;
2393 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2394 reiserfs_info(sb, "checking transaction log (%pg)\n",
2396 start = ktime_get_seconds();
2399 * step 1, read in the journal header block. Check the transaction
2400 * it says is the first unflushed, and if that transaction is not
2401 * valid, replay is done
2403 journal->j_header_bh = journal_bread(sb,
2404 SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2405 + SB_ONDISK_JOURNAL_SIZE(sb));
2406 if (!journal->j_header_bh) {
2409 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2410 if (le32_to_cpu(jh->j_first_unflushed_offset) <
2411 SB_ONDISK_JOURNAL_SIZE(sb)
2412 && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2414 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2415 le32_to_cpu(jh->j_first_unflushed_offset);
2416 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2417 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2418 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2419 "journal-1153: found in "
2420 "header: first_unflushed_offset %d, last_flushed_trans_id "
2421 "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2422 le32_to_cpu(jh->j_last_flush_trans_id));
2423 valid_journal_header = 1;
2426 * now, we try to read the first unflushed offset. If it
2427 * is not valid, there is nothing more we can do, and it
2428 * makes no sense to read through the whole log.
2432 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2433 le32_to_cpu(jh->j_first_unflushed_offset));
2434 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2436 continue_replay = 0;
2439 goto start_log_replay;
2443 * ok, there are transactions that need to be replayed. start
2444 * with the first log block, find all the valid transactions, and
2445 * pick out the oldest.
2447 while (continue_replay
2449 (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2450 SB_ONDISK_JOURNAL_SIZE(sb))) {
2452 * Note that it is required for blocksize of primary fs
2453 * device and journal device to be the same
2456 reiserfs_breada(journal->j_dev_bd, cur_dblock,
2458 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2459 SB_ONDISK_JOURNAL_SIZE(sb));
2461 journal_transaction_is_valid(sb, d_bh,
2462 &oldest_invalid_trans_id,
2465 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2466 if (oldest_start == 0) { /* init all oldest_ values */
2467 oldest_trans_id = get_desc_trans_id(desc);
2468 oldest_start = d_bh->b_blocknr;
2469 newest_mount_id = get_desc_mount_id(desc);
2470 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2471 "journal-1179: Setting "
2472 "oldest_start to offset %llu, trans_id %lu",
2474 SB_ONDISK_JOURNAL_1st_BLOCK
2475 (sb), oldest_trans_id);
2476 } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2477 /* one we just read was older */
2478 oldest_trans_id = get_desc_trans_id(desc);
2479 oldest_start = d_bh->b_blocknr;
2480 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2481 "journal-1180: Resetting "
2482 "oldest_start to offset %lu, trans_id %lu",
2484 SB_ONDISK_JOURNAL_1st_BLOCK
2485 (sb), oldest_trans_id);
2487 if (newest_mount_id < get_desc_mount_id(desc)) {
2488 newest_mount_id = get_desc_mount_id(desc);
2489 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2490 "journal-1299: Setting "
2491 "newest_mount_id to %d",
2492 get_desc_mount_id(desc));
2494 cur_dblock += get_desc_trans_len(desc) + 2;
2502 cur_dblock = oldest_start;
2503 if (oldest_trans_id) {
2504 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2505 "journal-1206: Starting replay "
2506 "from offset %llu, trans_id %lu",
2507 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2512 while (continue_replay && oldest_trans_id > 0) {
2514 journal_read_transaction(sb, cur_dblock, oldest_start,
2515 oldest_trans_id, newest_mount_id);
2518 } else if (ret != 0) {
2522 SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2524 if (cur_dblock == oldest_start)
2528 if (oldest_trans_id == 0) {
2529 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2530 "journal-1225: No valid " "transactions found");
2533 * j_start does not get set correctly if we don't replay any
2534 * transactions. if we had a valid journal_header, set j_start
2535 * to the first unflushed transaction value, copy the trans_id
2538 if (valid_journal_header && replay_count == 0) {
2539 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2540 journal->j_trans_id =
2541 le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2542 /* check for trans_id overflow */
2543 if (journal->j_trans_id == 0)
2544 journal->j_trans_id = 10;
2545 journal->j_last_flush_trans_id =
2546 le32_to_cpu(jh->j_last_flush_trans_id);
2547 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2549 journal->j_mount_id = newest_mount_id + 1;
2551 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2552 "newest_mount_id to %lu", journal->j_mount_id);
2553 journal->j_first_unflushed_offset = journal->j_start;
2554 if (replay_count > 0) {
2556 "replayed %d transactions in %lu seconds\n",
2557 replay_count, ktime_get_seconds() - start);
2559 /* needed to satisfy the locking in _update_journal_header_block */
2560 reiserfs_write_lock(sb);
2561 if (!bdev_read_only(sb->s_bdev) &&
2562 _update_journal_header_block(sb, journal->j_start,
2563 journal->j_last_flush_trans_id)) {
2564 reiserfs_write_unlock(sb);
2566 * replay failed, caller must call free_journal_ram and abort
2571 reiserfs_write_unlock(sb);
2575 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2577 struct reiserfs_journal_list *jl;
2578 jl = kzalloc(sizeof(struct reiserfs_journal_list),
2579 GFP_NOFS | __GFP_NOFAIL);
2580 INIT_LIST_HEAD(&jl->j_list);
2581 INIT_LIST_HEAD(&jl->j_working_list);
2582 INIT_LIST_HEAD(&jl->j_tail_bh_list);
2583 INIT_LIST_HEAD(&jl->j_bh_list);
2584 mutex_init(&jl->j_commit_mutex);
2585 SB_JOURNAL(s)->j_num_lists++;
2586 get_journal_list(jl);
2590 static void journal_list_init(struct super_block *sb)
2592 SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2595 static void release_journal_dev(struct super_block *super,
2596 struct reiserfs_journal *journal)
2598 if (journal->j_dev_bd != NULL) {
2599 blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2600 journal->j_dev_bd = NULL;
2604 static int journal_init_dev(struct super_block *super,
2605 struct reiserfs_journal *journal,
2606 const char *jdev_name)
2610 fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
2611 char b[BDEVNAME_SIZE];
2615 journal->j_dev_bd = NULL;
2616 jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2617 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2619 if (bdev_read_only(super->s_bdev))
2620 blkdev_mode = FMODE_READ;
2622 /* there is no "jdev" option and journal is on separate device */
2623 if ((!jdev_name || !jdev_name[0])) {
2624 if (jdev == super->s_dev)
2625 blkdev_mode &= ~FMODE_EXCL;
2626 journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2628 journal->j_dev_mode = blkdev_mode;
2629 if (IS_ERR(journal->j_dev_bd)) {
2630 result = PTR_ERR(journal->j_dev_bd);
2631 journal->j_dev_bd = NULL;
2632 reiserfs_warning(super, "sh-458",
2633 "cannot init journal device '%s': %i",
2634 __bdevname(jdev, b), result);
2636 } else if (jdev != super->s_dev)
2637 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2642 journal->j_dev_mode = blkdev_mode;
2643 journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
2644 if (IS_ERR(journal->j_dev_bd)) {
2645 result = PTR_ERR(journal->j_dev_bd);
2646 journal->j_dev_bd = NULL;
2647 reiserfs_warning(super, "sh-457",
2648 "journal_init_dev: Cannot open '%s': %i",
2653 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2654 reiserfs_info(super,
2655 "journal_init_dev: journal device: %pg\n",
2661 * When creating/tuning a file system user can assign some
2662 * journal params within boundaries which depend on the ratio
2663 * blocksize/standard_blocksize.
2665 * For blocks >= standard_blocksize transaction size should
2666 * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2667 * then JOURNAL_TRANS_MAX_DEFAULT.
2669 * For blocks < standard_blocksize these boundaries should be
2670 * decreased proportionally.
2672 #define REISERFS_STANDARD_BLKSIZE (4096)
2674 static int check_advise_trans_params(struct super_block *sb,
2675 struct reiserfs_journal *journal)
2677 if (journal->j_trans_max) {
2678 /* Non-default journal params. Do sanity check for them. */
2680 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2681 ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2683 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2684 journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2685 SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2686 JOURNAL_MIN_RATIO) {
2687 reiserfs_warning(sb, "sh-462",
2688 "bad transaction max size (%u). "
2689 "FSCK?", journal->j_trans_max);
2692 if (journal->j_max_batch != (journal->j_trans_max) *
2693 JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2694 reiserfs_warning(sb, "sh-463",
2695 "bad transaction max batch (%u). "
2696 "FSCK?", journal->j_max_batch);
2701 * Default journal params.
2702 * The file system was created by old version
2703 * of mkreiserfs, so some fields contain zeros,
2704 * and we need to advise proper values for them
2706 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2707 reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2711 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2712 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2713 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2718 /* must be called once on fs mount. calls journal_read for you */
2719 int journal_init(struct super_block *sb, const char *j_dev_name,
2720 int old_format, unsigned int commit_max_age)
2722 int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2723 struct buffer_head *bhjh;
2724 struct reiserfs_super_block *rs;
2725 struct reiserfs_journal_header *jh;
2726 struct reiserfs_journal *journal;
2727 struct reiserfs_journal_list *jl;
2730 journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal));
2732 reiserfs_warning(sb, "journal-1256",
2733 "unable to get memory for journal structure");
2736 INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2737 INIT_LIST_HEAD(&journal->j_prealloc_list);
2738 INIT_LIST_HEAD(&journal->j_working_list);
2739 INIT_LIST_HEAD(&journal->j_journal_list);
2740 journal->j_persistent_trans = 0;
2741 if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2742 reiserfs_bmap_count(sb)))
2743 goto free_and_return;
2745 allocate_bitmap_nodes(sb);
2747 /* reserved for journal area support */
2748 SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2749 REISERFS_OLD_DISK_OFFSET_IN_BYTES
2751 reiserfs_bmap_count(sb) +
2753 REISERFS_DISK_OFFSET_IN_BYTES /
2754 sb->s_blocksize + 2);
2757 * Sanity check to see is the standard journal fitting
2758 * within first bitmap (actual for small blocksizes)
2760 if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2761 (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2762 SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2763 reiserfs_warning(sb, "journal-1393",
2764 "journal does not fit for area addressed "
2765 "by first of bitmap blocks. It starts at "
2766 "%u and its size is %u. Block size %ld",
2767 SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2768 SB_ONDISK_JOURNAL_SIZE(sb),
2770 goto free_and_return;
2773 if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2774 reiserfs_warning(sb, "sh-462",
2775 "unable to initialize journal device");
2776 goto free_and_return;
2779 rs = SB_DISK_SUPER_BLOCK(sb);
2781 /* read journal header */
2782 bhjh = journal_bread(sb,
2783 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2784 SB_ONDISK_JOURNAL_SIZE(sb));
2786 reiserfs_warning(sb, "sh-459",
2787 "unable to read journal header");
2788 goto free_and_return;
2790 jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2792 /* make sure that journal matches to the super block */
2793 if (is_reiserfs_jr(rs)
2794 && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2795 sb_jp_journal_magic(rs))) {
2796 reiserfs_warning(sb, "sh-460",
2797 "journal header magic %x (device %pg) does "
2798 "not match to magic found in super block %x",
2799 jh->jh_journal.jp_journal_magic,
2801 sb_jp_journal_magic(rs));
2803 goto free_and_return;
2806 journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2807 journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2808 journal->j_max_commit_age =
2809 le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2810 journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2812 if (check_advise_trans_params(sb, journal) != 0)
2813 goto free_and_return;
2814 journal->j_default_max_commit_age = journal->j_max_commit_age;
2816 if (commit_max_age != 0) {
2817 journal->j_max_commit_age = commit_max_age;
2818 journal->j_max_trans_age = commit_max_age;
2821 reiserfs_info(sb, "journal params: device %pg, size %u, "
2822 "journal first block %u, max trans len %u, max batch %u, "
2823 "max commit age %u, max trans age %u\n",
2825 SB_ONDISK_JOURNAL_SIZE(sb),
2826 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2827 journal->j_trans_max,
2828 journal->j_max_batch,
2829 journal->j_max_commit_age, journal->j_max_trans_age);
2833 journal->j_list_bitmap_index = 0;
2834 journal_list_init(sb);
2836 memset(journal->j_list_hash_table, 0,
2837 JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2839 INIT_LIST_HEAD(&journal->j_dirty_buffers);
2840 spin_lock_init(&journal->j_dirty_buffers_lock);
2842 journal->j_start = 0;
2844 journal->j_len_alloc = 0;
2845 atomic_set(&journal->j_wcount, 0);
2846 atomic_set(&journal->j_async_throttle, 0);
2847 journal->j_bcount = 0;
2848 journal->j_trans_start_time = 0;
2849 journal->j_last = NULL;
2850 journal->j_first = NULL;
2851 init_waitqueue_head(&journal->j_join_wait);
2852 mutex_init(&journal->j_mutex);
2853 mutex_init(&journal->j_flush_mutex);
2855 journal->j_trans_id = 10;
2856 journal->j_mount_id = 10;
2857 journal->j_state = 0;
2858 atomic_set(&journal->j_jlock, 0);
2859 journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2860 journal->j_cnode_free_orig = journal->j_cnode_free_list;
2861 journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2862 journal->j_cnode_used = 0;
2863 journal->j_must_wait = 0;
2865 if (journal->j_cnode_free == 0) {
2866 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2867 "allocation failed (%ld bytes). Journal is "
2868 "too large for available memory. Usually "
2869 "this is due to a journal that is too large.",
2870 sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2871 goto free_and_return;
2874 init_journal_hash(sb);
2875 jl = journal->j_current_jl;
2878 * get_list_bitmap() may call flush_commit_list() which
2879 * requires the lock. Calling flush_commit_list() shouldn't happen
2880 * this early but I like to be paranoid.
2882 reiserfs_write_lock(sb);
2883 jl->j_list_bitmap = get_list_bitmap(sb, jl);
2884 reiserfs_write_unlock(sb);
2885 if (!jl->j_list_bitmap) {
2886 reiserfs_warning(sb, "journal-2005",
2887 "get_list_bitmap failed for journal list 0");
2888 goto free_and_return;
2891 ret = journal_read(sb);
2893 reiserfs_warning(sb, "reiserfs-2006",
2894 "Replay Failure, unable to mount");
2895 goto free_and_return;
2898 INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2899 journal->j_work_sb = sb;
2902 free_journal_ram(sb);
2907 * test for a polite end of the current transaction. Used by file_write,
2908 * and should be used by delete to make sure they don't write more than
2909 * can fit inside a single transaction
2911 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2914 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2915 time64_t now = ktime_get_seconds();
2916 /* cannot restart while nested */
2917 BUG_ON(!th->t_trans_id);
2918 if (th->t_refcount > 1)
2920 if (journal->j_must_wait > 0 ||
2921 (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2922 atomic_read(&journal->j_jlock) ||
2923 (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2924 journal->j_cnode_free < (journal->j_trans_max * 3)) {
2928 journal->j_len_alloc += new_alloc;
2929 th->t_blocks_allocated += new_alloc ;
2933 /* this must be called inside a transaction */
2934 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2936 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2937 BUG_ON(!th->t_trans_id);
2938 journal->j_must_wait = 1;
2939 set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2943 /* this must be called without a transaction started */
2944 void reiserfs_allow_writes(struct super_block *s)
2946 struct reiserfs_journal *journal = SB_JOURNAL(s);
2947 clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2948 wake_up(&journal->j_join_wait);
2951 /* this must be called without a transaction started */
2952 void reiserfs_wait_on_write_block(struct super_block *s)
2954 struct reiserfs_journal *journal = SB_JOURNAL(s);
2955 wait_event(journal->j_join_wait,
2956 !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2959 static void queue_log_writer(struct super_block *s)
2961 wait_queue_entry_t wait;
2962 struct reiserfs_journal *journal = SB_JOURNAL(s);
2963 set_bit(J_WRITERS_QUEUED, &journal->j_state);
2966 * we don't want to use wait_event here because
2967 * we only want to wait once.
2969 init_waitqueue_entry(&wait, current);
2970 add_wait_queue(&journal->j_join_wait, &wait);
2971 set_current_state(TASK_UNINTERRUPTIBLE);
2972 if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
2973 int depth = reiserfs_write_unlock_nested(s);
2975 reiserfs_write_lock_nested(s, depth);
2977 __set_current_state(TASK_RUNNING);
2978 remove_wait_queue(&journal->j_join_wait, &wait);
2981 static void wake_queued_writers(struct super_block *s)
2983 struct reiserfs_journal *journal = SB_JOURNAL(s);
2984 if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2985 wake_up(&journal->j_join_wait);
2988 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
2990 struct reiserfs_journal *journal = SB_JOURNAL(sb);
2991 unsigned long bcount = journal->j_bcount;
2995 depth = reiserfs_write_unlock_nested(sb);
2996 schedule_timeout_uninterruptible(1);
2997 reiserfs_write_lock_nested(sb, depth);
2999 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
3000 while ((atomic_read(&journal->j_wcount) > 0 ||
3001 atomic_read(&journal->j_jlock)) &&
3002 journal->j_trans_id == trans_id) {
3003 queue_log_writer(sb);
3005 if (journal->j_trans_id != trans_id)
3007 if (bcount == journal->j_bcount)
3009 bcount = journal->j_bcount;
3014 * join == true if you must join an existing transaction.
3015 * join == false if you can deal with waiting for others to finish
3017 * this will block until the transaction is joinable. send the number of
3018 * blocks you expect to use in nblocks.
3020 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3021 struct super_block *sb, unsigned long nblocks,
3024 time64_t now = ktime_get_seconds();
3025 unsigned int old_trans_id;
3026 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3027 struct reiserfs_transaction_handle myth;
3028 int sched_count = 0;
3032 reiserfs_check_lock_depth(sb, "journal_begin");
3033 BUG_ON(nblocks > journal->j_trans_max);
3035 PROC_INFO_INC(sb, journal.journal_being);
3036 /* set here for journal_join */
3042 if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3044 retval = journal->j_errno;
3047 journal->j_bcount++;
3049 if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3051 depth = reiserfs_write_unlock_nested(sb);
3052 reiserfs_wait_on_write_block(sb);
3053 reiserfs_write_lock_nested(sb, depth);
3054 PROC_INFO_INC(sb, journal.journal_relock_writers);
3057 now = ktime_get_seconds();
3060 * if there is no room in the journal OR
3061 * if this transaction is too old, and we weren't called joinable,
3062 * wait for it to finish before beginning we don't sleep if there
3063 * aren't other writers
3066 if ((!join && journal->j_must_wait > 0) ||
3068 && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3069 || (!join && atomic_read(&journal->j_wcount) > 0
3070 && journal->j_trans_start_time > 0
3071 && (now - journal->j_trans_start_time) >
3072 journal->j_max_trans_age) || (!join
3073 && atomic_read(&journal->j_jlock))
3074 || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3076 old_trans_id = journal->j_trans_id;
3077 /* allow others to finish this transaction */
3080 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3081 journal->j_max_batch &&
3082 ((journal->j_len + nblocks + 2) * 100) <
3083 (journal->j_len_alloc * 75)) {
3084 if (atomic_read(&journal->j_wcount) > 10) {
3086 queue_log_writer(sb);
3091 * don't mess with joining the transaction if all we
3092 * have to do is wait for someone else to do a commit
3094 if (atomic_read(&journal->j_jlock)) {
3095 while (journal->j_trans_id == old_trans_id &&
3096 atomic_read(&journal->j_jlock)) {
3097 queue_log_writer(sb);
3101 retval = journal_join(&myth, sb);
3105 /* someone might have ended the transaction while we joined */
3106 if (old_trans_id != journal->j_trans_id) {
3107 retval = do_journal_end(&myth, 0);
3109 retval = do_journal_end(&myth, COMMIT_NOW);
3115 PROC_INFO_INC(sb, journal.journal_relock_wcount);
3118 /* we are the first writer, set trans_id */
3119 if (journal->j_trans_start_time == 0) {
3120 journal->j_trans_start_time = ktime_get_seconds();
3122 atomic_inc(&journal->j_wcount);
3123 journal->j_len_alloc += nblocks;
3124 th->t_blocks_logged = 0;
3125 th->t_blocks_allocated = nblocks;
3126 th->t_trans_id = journal->j_trans_id;
3128 INIT_LIST_HEAD(&th->t_list);
3132 memset(th, 0, sizeof(*th));
3134 * Re-set th->t_super, so we can properly keep track of how many
3135 * persistent transactions there are. We need to do this so if this
3136 * call is part of a failed restart_transaction, we can free it later
3142 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3148 struct reiserfs_transaction_handle *th;
3151 * if we're nesting into an existing transaction. It will be
3152 * persistent on its own
3154 if (reiserfs_transaction_running(s)) {
3155 th = current->journal_info;
3157 BUG_ON(th->t_refcount < 2);
3161 th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3164 ret = journal_begin(th, s, nblocks);
3170 SB_JOURNAL(s)->j_persistent_trans++;
3174 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3176 struct super_block *s = th->t_super;
3179 ret = journal_end(th);
3182 if (th->t_refcount == 0) {
3183 SB_JOURNAL(s)->j_persistent_trans--;
3189 static int journal_join(struct reiserfs_transaction_handle *th,
3190 struct super_block *sb)
3192 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3195 * this keeps do_journal_end from NULLing out the
3196 * current->journal_info pointer
3198 th->t_handle_save = cur_th;
3199 BUG_ON(cur_th && cur_th->t_refcount > 1);
3200 return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN);
3203 int journal_join_abort(struct reiserfs_transaction_handle *th,
3204 struct super_block *sb)
3206 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3209 * this keeps do_journal_end from NULLing out the
3210 * current->journal_info pointer
3212 th->t_handle_save = cur_th;
3213 BUG_ON(cur_th && cur_th->t_refcount > 1);
3214 return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT);
3217 int journal_begin(struct reiserfs_transaction_handle *th,
3218 struct super_block *sb, unsigned long nblocks)
3220 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3223 th->t_handle_save = NULL;
3225 /* we are nesting into the current transaction */
3226 if (cur_th->t_super == sb) {
3227 BUG_ON(!cur_th->t_refcount);
3228 cur_th->t_refcount++;
3229 memcpy(th, cur_th, sizeof(*th));
3230 if (th->t_refcount <= 1)
3231 reiserfs_warning(sb, "reiserfs-2005",
3232 "BAD: refcount <= 1, but "
3233 "journal_info != 0");
3237 * we've ended up with a handle from a different
3238 * filesystem. save it and restore on journal_end.
3239 * This should never really happen...
3241 reiserfs_warning(sb, "clm-2100",
3242 "nesting info a different FS");
3243 th->t_handle_save = current->journal_info;
3244 current->journal_info = th;
3247 current->journal_info = th;
3249 ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3250 BUG_ON(current->journal_info != th);
3253 * I guess this boils down to being the reciprocal of clm-2100 above.
3254 * If do_journal_begin_r fails, we need to put it back, since
3255 * journal_end won't be called to do it. */
3257 current->journal_info = th->t_handle_save;
3259 BUG_ON(!th->t_refcount);
3265 * puts bh into the current transaction. If it was already there, reorders
3266 * removes the old pointers from the hash, and puts new ones in (to make
3267 * sure replay happen in the right order).
3269 * if it was dirty, cleans and files onto the clean list. I can't let it
3270 * be dirty again until the transaction is committed.
3272 * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3274 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3275 struct buffer_head *bh)
3277 struct super_block *sb = th->t_super;
3278 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3279 struct reiserfs_journal_cnode *cn = NULL;
3280 int count_already_incd = 0;
3282 BUG_ON(!th->t_trans_id);
3284 PROC_INFO_INC(sb, journal.mark_dirty);
3285 if (th->t_trans_id != journal->j_trans_id) {
3286 reiserfs_panic(th->t_super, "journal-1577",
3287 "handle trans id %ld != current trans id %ld",
3288 th->t_trans_id, journal->j_trans_id);
3291 prepared = test_clear_buffer_journal_prepared(bh);
3292 clear_buffer_journal_restore_dirty(bh);
3293 /* already in this transaction, we are done */
3294 if (buffer_journaled(bh)) {
3295 PROC_INFO_INC(sb, journal.mark_dirty_already);
3300 * this must be turned into a panic instead of a warning. We can't
3301 * allow a dirty or journal_dirty or locked buffer to be logged, as
3302 * some changes could get to disk too early. NOT GOOD.
3304 if (!prepared || buffer_dirty(bh)) {
3305 reiserfs_warning(sb, "journal-1777",
3306 "buffer %llu bad state "
3307 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3308 (unsigned long long)bh->b_blocknr,
3309 prepared ? ' ' : '!',
3310 buffer_locked(bh) ? ' ' : '!',
3311 buffer_dirty(bh) ? ' ' : '!',
3312 buffer_journal_dirty(bh) ? ' ' : '!');
3315 if (atomic_read(&journal->j_wcount) <= 0) {
3316 reiserfs_warning(sb, "journal-1409",
3317 "returning because j_wcount was %d",
3318 atomic_read(&journal->j_wcount));
3322 * this error means I've screwed up, and we've overflowed
3323 * the transaction. Nothing can be done here, except make the
3324 * FS readonly or panic.
3326 if (journal->j_len >= journal->j_trans_max) {
3327 reiserfs_panic(th->t_super, "journal-1413",
3328 "j_len (%lu) is too big",
3332 if (buffer_journal_dirty(bh)) {
3333 count_already_incd = 1;
3334 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3335 clear_buffer_journal_dirty(bh);
3338 if (journal->j_len > journal->j_len_alloc) {
3339 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3342 set_buffer_journaled(bh);
3344 /* now put this guy on the end */
3348 reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3351 if (th->t_blocks_logged == th->t_blocks_allocated) {
3352 th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3353 journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3355 th->t_blocks_logged++;
3359 cn->blocknr = bh->b_blocknr;
3362 insert_journal_hash(journal->j_hash_table, cn);
3363 if (!count_already_incd) {
3368 cn->prev = journal->j_last;
3370 if (journal->j_last) {
3371 journal->j_last->next = cn;
3372 journal->j_last = cn;
3374 journal->j_first = cn;
3375 journal->j_last = cn;
3377 reiserfs_schedule_old_flush(sb);
3381 int journal_end(struct reiserfs_transaction_handle *th)
3383 struct super_block *sb = th->t_super;
3384 if (!current->journal_info && th->t_refcount > 1)
3385 reiserfs_warning(sb, "REISER-NESTING",
3386 "th NULL, refcount %d", th->t_refcount);
3388 if (!th->t_trans_id) {
3394 if (th->t_refcount > 0) {
3395 struct reiserfs_transaction_handle *cur_th =
3396 current->journal_info;
3399 * we aren't allowed to close a nested transaction on a
3400 * different filesystem from the one in the task struct
3402 BUG_ON(cur_th->t_super != th->t_super);
3405 memcpy(current->journal_info, th, sizeof(*th));
3410 return do_journal_end(th, 0);
3415 * removes from the current transaction, relsing and descrementing any counters.
3416 * also files the removed buffer directly onto the clean list
3418 * called by journal_mark_freed when a block has been deleted
3420 * returns 1 if it cleaned and relsed the buffer. 0 otherwise
3422 static int remove_from_transaction(struct super_block *sb,
3423 b_blocknr_t blocknr, int already_cleaned)
3425 struct buffer_head *bh;
3426 struct reiserfs_journal_cnode *cn;
3427 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3430 cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3431 if (!cn || !cn->bh) {
3436 cn->prev->next = cn->next;
3439 cn->next->prev = cn->prev;
3441 if (cn == journal->j_first) {
3442 journal->j_first = cn->next;
3444 if (cn == journal->j_last) {
3445 journal->j_last = cn->prev;
3448 remove_journal_hash(sb, journal->j_hash_table, NULL,
3450 clear_buffer_journaled(bh); /* don't log this one */
3452 if (!already_cleaned) {
3453 clear_buffer_journal_dirty(bh);
3454 clear_buffer_dirty(bh);
3455 clear_buffer_journal_test(bh);
3457 if (atomic_read(&bh->b_count) < 0) {
3458 reiserfs_warning(sb, "journal-1752",
3464 journal->j_len_alloc--;
3470 * for any cnode in a journal list, it can only be dirtied of all the
3471 * transactions that include it are committed to disk.
3472 * this checks through each transaction, and returns 1 if you are allowed
3473 * to dirty, and 0 if you aren't
3475 * it is called by dirty_journal_list, which is called after
3476 * flush_commit_list has gotten all the log blocks for a given
3477 * transaction on disk
3480 static int can_dirty(struct reiserfs_journal_cnode *cn)
3482 struct super_block *sb = cn->sb;
3483 b_blocknr_t blocknr = cn->blocknr;
3484 struct reiserfs_journal_cnode *cur = cn->hprev;
3488 * first test hprev. These are all newer than cn, so any node here
3489 * with the same block number and dev means this node can't be sent
3490 * to disk right now.
3492 while (cur && can_dirty) {
3493 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3494 cur->blocknr == blocknr) {
3500 * then test hnext. These are all older than cn. As long as they
3501 * are committed to the log, it is safe to write cn to disk
3504 while (cur && can_dirty) {
3505 if (cur->jlist && cur->jlist->j_len > 0 &&
3506 atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh &&
3507 cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3516 * syncs the commit blocks, but does not force the real buffers to disk
3517 * will wait until the current transaction is done/committed before returning
3519 int journal_end_sync(struct reiserfs_transaction_handle *th)
3521 struct super_block *sb = th->t_super;
3522 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3524 BUG_ON(!th->t_trans_id);
3525 /* you can sync while nested, very, very bad */
3526 BUG_ON(th->t_refcount > 1);
3527 if (journal->j_len == 0) {
3528 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3530 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
3532 return do_journal_end(th, COMMIT_NOW | WAIT);
3535 /* writeback the pending async commits to disk */
3536 static void flush_async_commits(struct work_struct *work)
3538 struct reiserfs_journal *journal =
3539 container_of(work, struct reiserfs_journal, j_work.work);
3540 struct super_block *sb = journal->j_work_sb;
3541 struct reiserfs_journal_list *jl;
3542 struct list_head *entry;
3544 reiserfs_write_lock(sb);
3545 if (!list_empty(&journal->j_journal_list)) {
3546 /* last entry is the youngest, commit it and you get everything */
3547 entry = journal->j_journal_list.prev;
3548 jl = JOURNAL_LIST_ENTRY(entry);
3549 flush_commit_list(sb, jl, 1);
3551 reiserfs_write_unlock(sb);
3555 * flushes any old transactions to disk
3556 * ends the current transaction if it is too old
3558 void reiserfs_flush_old_commits(struct super_block *sb)
3561 struct reiserfs_transaction_handle th;
3562 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3564 now = ktime_get_seconds();
3566 * safety check so we don't flush while we are replaying the log during
3569 if (list_empty(&journal->j_journal_list))
3573 * check the current transaction. If there are no writers, and it is
3574 * too old, finish it, and force the commit blocks to disk
3576 if (atomic_read(&journal->j_wcount) <= 0 &&
3577 journal->j_trans_start_time > 0 &&
3578 journal->j_len > 0 &&
3579 (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3580 if (!journal_join(&th, sb)) {
3581 reiserfs_prepare_for_journal(sb,
3582 SB_BUFFER_WITH_SB(sb),
3584 journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3587 * we're only being called from kreiserfsd, it makes
3588 * no sense to do an async commit so that kreiserfsd
3591 do_journal_end(&th, COMMIT_NOW | WAIT);
3597 * returns 0 if do_journal_end should return right away, returns 1 if
3598 * do_journal_end should finish the commit
3600 * if the current transaction is too old, but still has writers, this will
3601 * wait on j_join_wait until all the writers are done. By the time it
3602 * wakes up, the transaction it was called has already ended, so it just
3603 * flushes the commit list and returns 0.
3605 * Won't batch when flush or commit_now is set. Also won't batch when
3606 * others are waiting on j_join_wait.
3608 * Note, we can't allow the journal_end to proceed while there are still
3609 * writers in the log.
3611 static int check_journal_end(struct reiserfs_transaction_handle *th, int flags)
3615 int flush = flags & FLUSH_ALL;
3616 int commit_now = flags & COMMIT_NOW;
3617 int wait_on_commit = flags & WAIT;
3618 struct reiserfs_journal_list *jl;
3619 struct super_block *sb = th->t_super;
3620 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3622 BUG_ON(!th->t_trans_id);
3624 if (th->t_trans_id != journal->j_trans_id) {
3625 reiserfs_panic(th->t_super, "journal-1577",
3626 "handle trans id %ld != current trans id %ld",
3627 th->t_trans_id, journal->j_trans_id);
3630 journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3631 /* <= 0 is allowed. unmounting might not call begin */
3632 if (atomic_read(&journal->j_wcount) > 0)
3633 atomic_dec(&journal->j_wcount);
3636 * BUG, deal with case where j_len is 0, but people previously
3637 * freed blocks need to be released will be dealt with by next
3638 * transaction that actually writes something, but should be taken
3639 * care of in this trans
3641 BUG_ON(journal->j_len == 0);
3644 * if wcount > 0, and we are called to with flush or commit_now,
3645 * we wait on j_join_wait. We will wake up when the last writer has
3646 * finished the transaction, and started it on its way to the disk.
3647 * Then, we flush the commit or journal list, and just return 0
3648 * because the rest of journal end was already done for this
3651 if (atomic_read(&journal->j_wcount) > 0) {
3652 if (flush || commit_now) {
3655 jl = journal->j_current_jl;
3656 trans_id = jl->j_trans_id;
3658 jl->j_state |= LIST_COMMIT_PENDING;
3659 atomic_set(&journal->j_jlock, 1);
3661 journal->j_next_full_flush = 1;
3666 * sleep while the current transaction is
3669 while (journal->j_trans_id == trans_id) {
3670 if (atomic_read(&journal->j_jlock)) {
3671 queue_log_writer(sb);
3674 if (journal->j_trans_id == trans_id) {
3675 atomic_set(&journal->j_jlock,
3681 BUG_ON(journal->j_trans_id == trans_id);
3684 && journal_list_still_alive(sb, trans_id)
3685 && wait_on_commit) {
3686 flush_commit_list(sb, jl, 1);
3694 /* deal with old transactions where we are the last writers */
3695 now = ktime_get_seconds();
3696 if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3698 journal->j_next_async_flush = 1;
3700 /* don't batch when someone is waiting on j_join_wait */
3701 /* don't batch when syncing the commit or flushing the whole trans */
3702 if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock))
3703 && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3704 && journal->j_len_alloc < journal->j_max_batch
3705 && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3706 journal->j_bcount++;
3711 if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3712 reiserfs_panic(sb, "journal-003",
3713 "j_start (%ld) is too high",
3720 * Does all the work that makes deleting blocks safe.
3721 * when deleting a block mark BH_JNew, just remove it from the current
3722 * transaction, clean it's buffer_head and move on.
3725 * set a bit for the block in the journal bitmap. That will prevent it from
3726 * being allocated for unformatted nodes before this transaction has finished.
3728 * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.
3729 * That will prevent any old transactions with this block from trying to flush
3730 * to the real location. Since we aren't removing the cnode from the
3731 * journal_list_hash, *the block can't be reallocated yet.
3733 * Then remove it from the current transaction, decrementing any counters and
3734 * filing it on the clean list.
3736 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3737 struct super_block *sb, b_blocknr_t blocknr)
3739 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3740 struct reiserfs_journal_cnode *cn = NULL;
3741 struct buffer_head *bh = NULL;
3742 struct reiserfs_list_bitmap *jb = NULL;
3744 BUG_ON(!th->t_trans_id);
3746 cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3751 /* if it is journal new, we just remove it from this transaction */
3752 if (bh && buffer_journal_new(bh)) {
3753 clear_buffer_journal_new(bh);
3754 clear_prepared_bits(bh);
3755 reiserfs_clean_and_file_buffer(bh);
3756 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3759 * set the bit for this block in the journal bitmap
3760 * for this transaction
3762 jb = journal->j_current_jl->j_list_bitmap;
3764 reiserfs_panic(sb, "journal-1702",
3765 "journal_list_bitmap is NULL");
3767 set_bit_in_list_bitmap(sb, blocknr, jb);
3769 /* Note, the entire while loop is not allowed to schedule. */
3772 clear_prepared_bits(bh);
3773 reiserfs_clean_and_file_buffer(bh);
3775 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3778 * find all older transactions with this block,
3779 * make sure they don't try to write it out
3781 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3784 if (sb == cn->sb && blocknr == cn->blocknr) {
3785 set_bit(BLOCK_FREED, &cn->state);
3788 * remove_from_transaction will brelse
3789 * the buffer if it was in the current
3793 clear_buffer_journal_dirty(cn->
3795 clear_buffer_dirty(cn->bh);
3796 clear_buffer_journal_test(cn->
3801 (&cn->bh->b_count) < 0) {
3802 reiserfs_warning(sb,
3804 "cn->bh->b_count < 0");
3808 * since we are clearing the bh,
3809 * we MUST dec nonzerolen
3812 atomic_dec(&cn->jlist->
3823 release_buffer_page(bh); /* get_hash grabs the buffer */
3827 void reiserfs_update_inode_transaction(struct inode *inode)
3829 struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3830 REISERFS_I(inode)->i_jl = journal->j_current_jl;
3831 REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3835 * returns -1 on error, 0 if no commits/barriers were done and 1
3836 * if a transaction was actually committed and the barrier was done
3838 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3839 struct reiserfs_journal_list *jl)
3841 struct reiserfs_transaction_handle th;
3842 struct super_block *sb = inode->i_sb;
3843 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3847 * is it from the current transaction,
3848 * or from an unknown transaction?
3850 if (id == journal->j_trans_id) {
3851 jl = journal->j_current_jl;
3853 * try to let other writers come in and
3854 * grow this transaction
3856 let_transaction_grow(sb, id);
3857 if (journal->j_trans_id != id) {
3858 goto flush_commit_only;
3861 ret = journal_begin(&th, sb, 1);
3865 /* someone might have ended this transaction while we joined */
3866 if (journal->j_trans_id != id) {
3867 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3869 journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3870 ret = journal_end(&th);
3871 goto flush_commit_only;
3874 ret = journal_end_sync(&th);
3880 * this gets tricky, we have to make sure the journal list in
3881 * the inode still exists. We know the list is still around
3882 * if we've got a larger transaction id than the oldest list
3885 if (journal_list_still_alive(inode->i_sb, id)) {
3887 * we only set ret to 1 when we know for sure
3888 * the barrier hasn't been started yet on the commit
3891 if (atomic_read(&jl->j_commit_left) > 1)
3893 flush_commit_list(sb, jl, 1);
3894 if (journal->j_errno)
3895 ret = journal->j_errno;
3898 /* otherwise the list is gone, and long since committed */
3902 int reiserfs_commit_for_inode(struct inode *inode)
3904 unsigned int id = REISERFS_I(inode)->i_trans_id;
3905 struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3908 * for the whole inode, assume unset id means it was
3909 * changed in the current transaction. More conservative
3912 reiserfs_update_inode_transaction(inode);
3913 id = REISERFS_I(inode)->i_trans_id;
3914 /* jl will be updated in __commit_trans_jl */
3917 return __commit_trans_jl(inode, id, jl);
3920 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3921 struct buffer_head *bh)
3923 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3924 PROC_INFO_INC(sb, journal.restore_prepared);
3928 if (test_clear_buffer_journal_restore_dirty(bh) &&
3929 buffer_journal_dirty(bh)) {
3930 struct reiserfs_journal_cnode *cn;
3931 reiserfs_write_lock(sb);
3932 cn = get_journal_hash_dev(sb,
3933 journal->j_list_hash_table,
3935 if (cn && can_dirty(cn)) {
3936 set_buffer_journal_test(bh);
3937 mark_buffer_dirty(bh);
3939 reiserfs_write_unlock(sb);
3941 clear_buffer_journal_prepared(bh);
3944 extern struct tree_balance *cur_tb;
3946 * before we can change a metadata block, we have to make sure it won't
3947 * be written to disk while we are altering it. So, we must:
3951 int reiserfs_prepare_for_journal(struct super_block *sb,
3952 struct buffer_head *bh, int wait)
3954 PROC_INFO_INC(sb, journal.prepare);
3956 if (!trylock_buffer(bh)) {
3961 set_buffer_journal_prepared(bh);
3962 if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3963 clear_buffer_journal_test(bh);
3964 set_buffer_journal_restore_dirty(bh);
3971 * long and ugly. If flush, will not return until all commit
3972 * blocks and all real buffers in the trans are on disk.
3973 * If no_async, won't return until all commit blocks are on disk.
3975 * keep reading, there are comments as you go along
3977 * If the journal is aborted, we just clean up. Things like flushing
3978 * journal lists, etc just won't happen.
3980 static int do_journal_end(struct reiserfs_transaction_handle *th, int flags)
3982 struct super_block *sb = th->t_super;
3983 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3984 struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3985 struct reiserfs_journal_cnode *last_cn = NULL;
3986 struct reiserfs_journal_desc *desc;
3987 struct reiserfs_journal_commit *commit;
3988 struct buffer_head *c_bh; /* commit bh */
3989 struct buffer_head *d_bh; /* desc bh */
3990 int cur_write_start = 0; /* start index of current log write */
3995 struct reiserfs_journal_list *jl, *temp_jl;
3996 struct list_head *entry, *safe;
3997 unsigned long jindex;
3998 unsigned int commit_trans_id;
4002 BUG_ON(th->t_refcount > 1);
4003 BUG_ON(!th->t_trans_id);
4004 BUG_ON(!th->t_super);
4007 * protect flush_older_commits from doing mistakes if the
4008 * transaction ID counter gets overflowed.
4010 if (th->t_trans_id == ~0U)
4011 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
4012 flush = flags & FLUSH_ALL;
4013 wait_on_commit = flags & WAIT;
4015 current->journal_info = th->t_handle_save;
4016 reiserfs_check_lock_depth(sb, "journal end");
4017 if (journal->j_len == 0) {
4018 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
4020 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
4024 if (journal->j_next_full_flush) {
4028 if (journal->j_next_async_flush) {
4029 flags |= COMMIT_NOW | WAIT;
4034 * check_journal_end locks the journal, and unlocks if it does
4035 * not return 1 it tells us if we should continue with the
4036 * journal_end, or just return
4038 if (!check_journal_end(th, flags)) {
4039 reiserfs_schedule_old_flush(sb);
4040 wake_queued_writers(sb);
4041 reiserfs_async_progress_wait(sb);
4045 /* check_journal_end might set these, check again */
4046 if (journal->j_next_full_flush) {
4051 * j must wait means we have to flush the log blocks, and the
4052 * real blocks for this transaction
4054 if (journal->j_must_wait > 0) {
4057 #ifdef REISERFS_PREALLOCATE
4059 * quota ops might need to nest, setup the journal_info pointer
4060 * for them and raise the refcount so that it is > 0.
4062 current->journal_info = th;
4065 /* it should not involve new blocks into the transaction */
4066 reiserfs_discard_all_prealloc(th);
4069 current->journal_info = th->t_handle_save;
4072 /* setup description block */
4075 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4077 set_buffer_uptodate(d_bh);
4078 desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4079 memset(d_bh->b_data, 0, d_bh->b_size);
4080 memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4081 set_desc_trans_id(desc, journal->j_trans_id);
4084 * setup commit block. Don't write (keep it clean too) this one
4085 * until after everyone else is written
4087 c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4088 ((journal->j_start + journal->j_len +
4089 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4090 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4091 memset(c_bh->b_data, 0, c_bh->b_size);
4092 set_commit_trans_id(commit, journal->j_trans_id);
4093 set_buffer_uptodate(c_bh);
4095 /* init this journal list */
4096 jl = journal->j_current_jl;
4099 * we lock the commit before doing anything because
4100 * we want to make sure nobody tries to run flush_commit_list until
4101 * the new transaction is fully setup, and we've already flushed the
4104 reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4106 /* save the transaction id in case we need to commit it later */
4107 commit_trans_id = jl->j_trans_id;
4109 atomic_set(&jl->j_older_commits_done, 0);
4110 jl->j_trans_id = journal->j_trans_id;
4111 jl->j_timestamp = journal->j_trans_start_time;
4112 jl->j_commit_bh = c_bh;
4113 jl->j_start = journal->j_start;
4114 jl->j_len = journal->j_len;
4115 atomic_set(&jl->j_nonzerolen, journal->j_len);
4116 atomic_set(&jl->j_commit_left, journal->j_len + 2);
4117 jl->j_realblock = NULL;
4120 * The ENTIRE FOR LOOP MUST not cause schedule to occur.
4121 * for each real block, add it to the journal list hash,
4122 * copy into real block index array in the commit or desc block
4124 trans_half = journal_trans_half(sb->s_blocksize);
4125 for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4126 if (buffer_journaled(cn->bh)) {
4127 jl_cn = get_cnode(sb);
4129 reiserfs_panic(sb, "journal-1676",
4130 "get_cnode returned NULL");
4133 jl->j_realblock = jl_cn;
4135 jl_cn->prev = last_cn;
4138 last_cn->next = jl_cn;
4142 * make sure the block we are trying to log
4143 * is not a block of journal or reserved area
4145 if (is_block_in_log_or_reserved_area
4146 (sb, cn->bh->b_blocknr)) {
4147 reiserfs_panic(sb, "journal-2332",
4148 "Trying to log block %lu, "
4149 "which is a log block",
4152 jl_cn->blocknr = cn->bh->b_blocknr;
4157 insert_journal_hash(journal->j_list_hash_table, jl_cn);
4158 if (i < trans_half) {
4159 desc->j_realblock[i] =
4160 cpu_to_le32(cn->bh->b_blocknr);
4162 commit->j_realblock[i - trans_half] =
4163 cpu_to_le32(cn->bh->b_blocknr);
4169 set_desc_trans_len(desc, journal->j_len);
4170 set_desc_mount_id(desc, journal->j_mount_id);
4171 set_desc_trans_id(desc, journal->j_trans_id);
4172 set_commit_trans_len(commit, journal->j_len);
4175 * special check in case all buffers in the journal
4176 * were marked for not logging
4178 BUG_ON(journal->j_len == 0);
4181 * we're about to dirty all the log blocks, mark the description block
4182 * dirty now too. Don't mark the commit block dirty until all the
4183 * others are on disk
4185 mark_buffer_dirty(d_bh);
4188 * first data block is j_start + 1, so add one to
4189 * cur_write_start wherever you use it
4191 cur_write_start = journal->j_start;
4192 cn = journal->j_first;
4193 jindex = 1; /* start at one so we don't get the desc again */
4195 clear_buffer_journal_new(cn->bh);
4196 /* copy all the real blocks into log area. dirty log blocks */
4197 if (buffer_journaled(cn->bh)) {
4198 struct buffer_head *tmp_bh;
4203 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4206 SB_ONDISK_JOURNAL_SIZE(sb)));
4207 set_buffer_uptodate(tmp_bh);
4208 page = cn->bh->b_page;
4210 memcpy(tmp_bh->b_data,
4211 addr + offset_in_page(cn->bh->b_data),
4214 mark_buffer_dirty(tmp_bh);
4216 set_buffer_journal_dirty(cn->bh);
4217 clear_buffer_journaled(cn->bh);
4220 * JDirty cleared sometime during transaction.
4221 * don't log this one
4223 reiserfs_warning(sb, "journal-2048",
4224 "BAD, buffer in journal hash, "
4231 reiserfs_cond_resched(sb);
4235 * we are done with both the c_bh and d_bh, but
4236 * c_bh must be written after all other commit blocks,
4237 * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4240 journal->j_current_jl = alloc_journal_list(sb);
4242 /* now it is safe to insert this transaction on the main list */
4243 list_add_tail(&jl->j_list, &journal->j_journal_list);
4244 list_add_tail(&jl->j_working_list, &journal->j_working_list);
4245 journal->j_num_work_lists++;
4247 /* reset journal values for the next transaction */
4248 old_start = journal->j_start;
4250 (journal->j_start + journal->j_len +
4251 2) % SB_ONDISK_JOURNAL_SIZE(sb);
4252 atomic_set(&journal->j_wcount, 0);
4253 journal->j_bcount = 0;
4254 journal->j_last = NULL;
4255 journal->j_first = NULL;
4257 journal->j_trans_start_time = 0;
4258 /* check for trans_id overflow */
4259 if (++journal->j_trans_id == 0)
4260 journal->j_trans_id = 10;
4261 journal->j_current_jl->j_trans_id = journal->j_trans_id;
4262 journal->j_must_wait = 0;
4263 journal->j_len_alloc = 0;
4264 journal->j_next_full_flush = 0;
4265 journal->j_next_async_flush = 0;
4266 init_journal_hash(sb);
4269 * make sure reiserfs_add_jh sees the new current_jl before we
4270 * write out the tails
4275 * tail conversion targets have to hit the disk before we end the
4276 * transaction. Otherwise a later transaction might repack the tail
4277 * before this transaction commits, leaving the data block unflushed
4278 * and clean, if we crash before the later transaction commits, the
4279 * data block is lost.
4281 if (!list_empty(&jl->j_tail_bh_list)) {
4282 depth = reiserfs_write_unlock_nested(sb);
4283 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4284 journal, jl, &jl->j_tail_bh_list);
4285 reiserfs_write_lock_nested(sb, depth);
4287 BUG_ON(!list_empty(&jl->j_tail_bh_list));
4288 mutex_unlock(&jl->j_commit_mutex);
4291 * honor the flush wishes from the caller, simple commits can
4292 * be done outside the journal lock, they are done below
4294 * if we don't flush the commit list right now, we put it into
4295 * the work queue so the people waiting on the async progress work
4296 * queue don't wait for this proc to flush journal lists and such.
4299 flush_commit_list(sb, jl, 1);
4300 flush_journal_list(sb, jl, 1);
4301 } else if (!(jl->j_state & LIST_COMMIT_PENDING)) {
4303 * Avoid queueing work when sb is being shut down. Transaction
4304 * will be flushed on journal shutdown.
4306 if (sb->s_flags & SB_ACTIVE)
4307 queue_delayed_work(REISERFS_SB(sb)->commit_wq,
4308 &journal->j_work, HZ / 10);
4312 * if the next transaction has any chance of wrapping, flush
4313 * transactions that might get overwritten. If any journal lists
4314 * are very old flush them as well.
4317 list_for_each_safe(entry, safe, &journal->j_journal_list) {
4318 temp_jl = JOURNAL_LIST_ENTRY(entry);
4319 if (journal->j_start <= temp_jl->j_start) {
4320 if ((journal->j_start + journal->j_trans_max + 1) >=
4322 flush_used_journal_lists(sb, temp_jl);
4324 } else if ((journal->j_start +
4325 journal->j_trans_max + 1) <
4326 SB_ONDISK_JOURNAL_SIZE(sb)) {
4328 * if we don't cross into the next
4329 * transaction and we don't wrap, there is
4330 * no way we can overlap any later transactions
4335 } else if ((journal->j_start +
4336 journal->j_trans_max + 1) >
4337 SB_ONDISK_JOURNAL_SIZE(sb)) {
4338 if (((journal->j_start + journal->j_trans_max + 1) %
4339 SB_ONDISK_JOURNAL_SIZE(sb)) >=
4341 flush_used_journal_lists(sb, temp_jl);
4345 * we don't overlap anything from out start
4346 * to the end of the log, and our wrapped
4347 * portion doesn't overlap anything at
4348 * the start of the log. We can break
4355 journal->j_current_jl->j_list_bitmap =
4356 get_list_bitmap(sb, journal->j_current_jl);
4358 if (!(journal->j_current_jl->j_list_bitmap)) {
4359 reiserfs_panic(sb, "journal-1996",
4360 "could not get a list bitmap");
4363 atomic_set(&journal->j_jlock, 0);
4365 /* wake up any body waiting to join. */
4366 clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4367 wake_up(&journal->j_join_wait);
4369 if (!flush && wait_on_commit &&
4370 journal_list_still_alive(sb, commit_trans_id)) {
4371 flush_commit_list(sb, jl, 1);
4374 reiserfs_check_lock_depth(sb, "journal end2");
4376 memset(th, 0, sizeof(*th));
4378 * Re-set th->t_super, so we can properly keep track of how many
4379 * persistent transactions there are. We need to do this so if this
4380 * call is part of a failed restart_transaction, we can free it later
4384 return journal->j_errno;
4387 /* Send the file system read only and refuse new transactions */
4388 void reiserfs_abort_journal(struct super_block *sb, int errno)
4390 struct reiserfs_journal *journal = SB_JOURNAL(sb);
4391 if (test_bit(J_ABORTED, &journal->j_state))
4394 if (!journal->j_errno)
4395 journal->j_errno = errno;
4397 sb->s_flags |= SB_RDONLY;
4398 set_bit(J_ABORTED, &journal->j_state);
4400 #ifdef CONFIG_REISERFS_CHECK