2 * Write ahead logging implementation copyright Chris Mason 2000
4 * The background commits make this code very interrelated, and
5 * overly complex. I need to rethink things a bit....The major players:
7 * journal_begin -- call with the number of blocks you expect to log.
8 * If the current transaction is too
9 * old, it will block until the current transaction is
10 * finished, and then start a new one.
11 * Usually, your transaction will get joined in with
12 * previous ones for speed.
14 * journal_join -- same as journal_begin, but won't block on the current
15 * transaction regardless of age. Don't ever call
16 * this. Ever. There are only two places it should be
17 * called from, and they are both inside this file.
19 * journal_mark_dirty -- adds blocks into this transaction. clears any flags
20 * that might make them get sent to disk
21 * and then marks them BH_JDirty. Puts the buffer head
22 * into the current transaction hash.
24 * journal_end -- if the current transaction is batchable, it does nothing
25 * otherwise, it could do an async/synchronous commit, or
26 * a full flush of all log and real blocks in the
29 * flush_old_commits -- if the current transaction is too old, it is ended and
30 * commit blocks are sent to disk. Forces commit blocks
31 * to disk for all backgrounded commits that have been
33 * -- Note, if you call this as an immediate flush from
34 * from within kupdate, it will ignore the immediate flag
37 #include <linux/time.h>
38 #include <linux/semaphore.h>
39 #include <linux/vmalloc.h>
41 #include <linux/kernel.h>
42 #include <linux/errno.h>
43 #include <linux/fcntl.h>
44 #include <linux/stat.h>
45 #include <linux/string.h>
46 #include <linux/buffer_head.h>
47 #include <linux/workqueue.h>
48 #include <linux/writeback.h>
49 #include <linux/blkdev.h>
50 #include <linux/backing-dev.h>
51 #include <linux/uaccess.h>
52 #include <linux/slab.h>
55 /* gets a struct reiserfs_journal_list * from a list head */
56 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
58 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
61 /* must be correct to keep the desc and commit structs at 4k */
62 #define JOURNAL_TRANS_HALF 1018
63 #define BUFNR 64 /*read ahead */
65 /* cnode stat bits. Move these into reiserfs_fs.h */
67 /* this block was freed, and can't be written. */
69 /* this block was freed during this transaction, and can't be written */
70 #define BLOCK_FREED_HOLDER 3
72 /* used in flush_journal_list */
73 #define BLOCK_NEEDS_FLUSH 4
74 #define BLOCK_DIRTIED 5
76 /* journal list state bits */
77 #define LIST_TOUCHED 1
79 #define LIST_COMMIT_PENDING 4 /* someone will commit this list */
81 /* flags for do_journal_end */
82 #define FLUSH_ALL 1 /* flush commit and real blocks */
83 #define COMMIT_NOW 2 /* end and commit this transaction */
84 #define WAIT 4 /* wait for the log blocks to hit the disk */
86 static int do_journal_end(struct reiserfs_transaction_handle *, int flags);
87 static int flush_journal_list(struct super_block *s,
88 struct reiserfs_journal_list *jl, int flushall);
89 static int flush_commit_list(struct super_block *s,
90 struct reiserfs_journal_list *jl, int flushall);
91 static int can_dirty(struct reiserfs_journal_cnode *cn);
92 static int journal_join(struct reiserfs_transaction_handle *th,
93 struct super_block *sb);
94 static void release_journal_dev(struct super_block *super,
95 struct reiserfs_journal *journal);
96 static int dirty_one_transaction(struct super_block *s,
97 struct reiserfs_journal_list *jl);
98 static void flush_async_commits(struct work_struct *work);
99 static void queue_log_writer(struct super_block *s);
101 /* values for join in do_journal_begin_r */
103 JBEGIN_REG = 0, /* regular journal begin */
104 /* join the running transaction if at all possible */
106 /* called from cleanup code, ignores aborted flag */
110 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
111 struct super_block *sb,
112 unsigned long nblocks, int join);
114 static void init_journal_hash(struct super_block *sb)
116 struct reiserfs_journal *journal = SB_JOURNAL(sb);
117 memset(journal->j_hash_table, 0,
118 JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
122 * clears BH_Dirty and sticks the buffer on the clean list. Called because
123 * I can't allow refile_buffer to make schedule happen after I've freed a
124 * block. Look at remove_from_transaction and journal_mark_freed for
127 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
130 clear_buffer_dirty(bh);
131 clear_buffer_journal_test(bh);
136 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
139 struct reiserfs_bitmap_node *bn;
142 bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
146 bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
152 INIT_LIST_HEAD(&bn->list);
156 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
158 struct reiserfs_journal *journal = SB_JOURNAL(sb);
159 struct reiserfs_bitmap_node *bn = NULL;
160 struct list_head *entry = journal->j_bitmap_nodes.next;
162 journal->j_used_bitmap_nodes++;
165 if (entry != &journal->j_bitmap_nodes) {
166 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
168 memset(bn->data, 0, sb->s_blocksize);
169 journal->j_free_bitmap_nodes--;
172 bn = allocate_bitmap_node(sb);
179 static inline void free_bitmap_node(struct super_block *sb,
180 struct reiserfs_bitmap_node *bn)
182 struct reiserfs_journal *journal = SB_JOURNAL(sb);
183 journal->j_used_bitmap_nodes--;
184 if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
188 list_add(&bn->list, &journal->j_bitmap_nodes);
189 journal->j_free_bitmap_nodes++;
193 static void allocate_bitmap_nodes(struct super_block *sb)
196 struct reiserfs_journal *journal = SB_JOURNAL(sb);
197 struct reiserfs_bitmap_node *bn = NULL;
198 for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
199 bn = allocate_bitmap_node(sb);
201 list_add(&bn->list, &journal->j_bitmap_nodes);
202 journal->j_free_bitmap_nodes++;
204 /* this is ok, we'll try again when more are needed */
210 static int set_bit_in_list_bitmap(struct super_block *sb,
212 struct reiserfs_list_bitmap *jb)
214 unsigned int bmap_nr = block / (sb->s_blocksize << 3);
215 unsigned int bit_nr = block % (sb->s_blocksize << 3);
217 if (!jb->bitmaps[bmap_nr]) {
218 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
220 set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
224 static void cleanup_bitmap_list(struct super_block *sb,
225 struct reiserfs_list_bitmap *jb)
228 if (jb->bitmaps == NULL)
231 for (i = 0; i < reiserfs_bmap_count(sb); i++) {
232 if (jb->bitmaps[i]) {
233 free_bitmap_node(sb, jb->bitmaps[i]);
234 jb->bitmaps[i] = NULL;
240 * only call this on FS unmount.
242 static int free_list_bitmaps(struct super_block *sb,
243 struct reiserfs_list_bitmap *jb_array)
246 struct reiserfs_list_bitmap *jb;
247 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
249 jb->journal_list = NULL;
250 cleanup_bitmap_list(sb, jb);
257 static int free_bitmap_nodes(struct super_block *sb)
259 struct reiserfs_journal *journal = SB_JOURNAL(sb);
260 struct list_head *next = journal->j_bitmap_nodes.next;
261 struct reiserfs_bitmap_node *bn;
263 while (next != &journal->j_bitmap_nodes) {
264 bn = list_entry(next, struct reiserfs_bitmap_node, list);
268 next = journal->j_bitmap_nodes.next;
269 journal->j_free_bitmap_nodes--;
276 * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
277 * jb_array is the array to be filled in.
279 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
280 struct reiserfs_list_bitmap *jb_array,
281 unsigned int bmap_nr)
285 struct reiserfs_list_bitmap *jb;
286 int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
288 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
290 jb->journal_list = NULL;
291 jb->bitmaps = vzalloc(mem);
293 reiserfs_warning(sb, "clm-2000", "unable to "
294 "allocate bitmaps for journal lists");
300 free_list_bitmaps(sb, jb_array);
307 * find an available list bitmap. If you can't find one, flush a commit list
310 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
311 struct reiserfs_journal_list
315 struct reiserfs_journal *journal = SB_JOURNAL(sb);
316 struct reiserfs_list_bitmap *jb = NULL;
318 for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
319 i = journal->j_list_bitmap_index;
320 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
321 jb = journal->j_list_bitmap + i;
322 if (journal->j_list_bitmap[i].journal_list) {
323 flush_commit_list(sb,
324 journal->j_list_bitmap[i].
326 if (!journal->j_list_bitmap[i].journal_list) {
333 /* double check to make sure if flushed correctly */
334 if (jb->journal_list)
336 jb->journal_list = jl;
341 * allocates a new chunk of X nodes, and links them all together as a list.
342 * Uses the cnode->next and cnode->prev pointers
343 * returns NULL on failure
345 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
347 struct reiserfs_journal_cnode *head;
349 if (num_cnodes <= 0) {
352 head = vzalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
357 head[0].next = head + 1;
358 for (i = 1; i < num_cnodes; i++) {
359 head[i].prev = head + (i - 1);
360 head[i].next = head + (i + 1); /* if last one, overwrite it after the if */
362 head[num_cnodes - 1].next = NULL;
366 /* pulls a cnode off the free list, or returns NULL on failure */
367 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
369 struct reiserfs_journal_cnode *cn;
370 struct reiserfs_journal *journal = SB_JOURNAL(sb);
372 reiserfs_check_lock_depth(sb, "get_cnode");
374 if (journal->j_cnode_free <= 0) {
377 journal->j_cnode_used++;
378 journal->j_cnode_free--;
379 cn = journal->j_cnode_free_list;
384 cn->next->prev = NULL;
386 journal->j_cnode_free_list = cn->next;
387 memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
392 * returns a cnode to the free list
394 static void free_cnode(struct super_block *sb,
395 struct reiserfs_journal_cnode *cn)
397 struct reiserfs_journal *journal = SB_JOURNAL(sb);
399 reiserfs_check_lock_depth(sb, "free_cnode");
401 journal->j_cnode_used--;
402 journal->j_cnode_free++;
403 /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
404 cn->next = journal->j_cnode_free_list;
405 if (journal->j_cnode_free_list) {
406 journal->j_cnode_free_list->prev = cn;
408 cn->prev = NULL; /* not needed with the memset, but I might kill the memset, and forget to do this */
409 journal->j_cnode_free_list = cn;
412 static void clear_prepared_bits(struct buffer_head *bh)
414 clear_buffer_journal_prepared(bh);
415 clear_buffer_journal_restore_dirty(bh);
419 * return a cnode with same dev, block number and size in table,
420 * or null if not found
422 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
426 reiserfs_journal_cnode
430 struct reiserfs_journal_cnode *cn;
431 cn = journal_hash(table, sb, bl);
433 if (cn->blocknr == bl && cn->sb == sb)
437 return (struct reiserfs_journal_cnode *)0;
441 * this actually means 'can this block be reallocated yet?'. If you set
442 * search_all, a block can only be allocated if it is not in the current
443 * transaction, was not freed by the current transaction, and has no chance
444 * of ever being overwritten by a replay after crashing.
446 * If you don't set search_all, a block can only be allocated if it is not
447 * in the current transaction. Since deleting a block removes it from the
448 * current transaction, this case should never happen. If you don't set
449 * search_all, make sure you never write the block without logging it.
451 * next_zero_bit is a suggestion about the next block to try for find_forward.
452 * when bl is rejected because it is set in a journal list bitmap, we search
453 * for the next zero bit in the bitmap that rejected bl. Then, we return
454 * that through next_zero_bit for find_forward to try.
456 * Just because we return something in next_zero_bit does not mean we won't
457 * reject it on the next call to reiserfs_in_journal
459 int reiserfs_in_journal(struct super_block *sb,
460 unsigned int bmap_nr, int bit_nr, int search_all,
461 b_blocknr_t * next_zero_bit)
463 struct reiserfs_journal *journal = SB_JOURNAL(sb);
464 struct reiserfs_journal_cnode *cn;
465 struct reiserfs_list_bitmap *jb;
469 *next_zero_bit = 0; /* always start this at zero. */
471 PROC_INFO_INC(sb, journal.in_journal);
473 * If we aren't doing a search_all, this is a metablock, and it
474 * will be logged before use. if we crash before the transaction
475 * that freed it commits, this transaction won't have committed
476 * either, and the block will never be written
479 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
480 PROC_INFO_INC(sb, journal.in_journal_bitmap);
481 jb = journal->j_list_bitmap + i;
482 if (jb->journal_list && jb->bitmaps[bmap_nr] &&
484 (unsigned long *)jb->bitmaps[bmap_nr]->
487 find_next_zero_bit((unsigned long *)
488 (jb->bitmaps[bmap_nr]->
490 sb->s_blocksize << 3,
497 bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
498 /* is it in any old transactions? */
501 get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
505 /* is it in the current transaction. This should never happen */
506 if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
511 PROC_INFO_INC(sb, journal.in_journal_reusable);
516 /* insert cn into table */
517 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
518 struct reiserfs_journal_cnode *cn)
520 struct reiserfs_journal_cnode *cn_orig;
522 cn_orig = journal_hash(table, cn->sb, cn->blocknr);
528 journal_hash(table, cn->sb, cn->blocknr) = cn;
531 /* lock the current transaction */
532 static inline void lock_journal(struct super_block *sb)
534 PROC_INFO_INC(sb, journal.lock_journal);
536 reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
539 /* unlock the current transaction */
540 static inline void unlock_journal(struct super_block *sb)
542 mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
545 static inline void get_journal_list(struct reiserfs_journal_list *jl)
550 static inline void put_journal_list(struct super_block *s,
551 struct reiserfs_journal_list *jl)
553 if (jl->j_refcount < 1) {
554 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
555 jl->j_trans_id, jl->j_refcount);
557 if (--jl->j_refcount == 0)
562 * this used to be much more involved, and I'm keeping it just in case
563 * things get ugly again. it gets called by flush_commit_list, and
564 * cleans up any data stored about blocks freed during a transaction.
566 static void cleanup_freed_for_journal_list(struct super_block *sb,
567 struct reiserfs_journal_list *jl)
570 struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
572 cleanup_bitmap_list(sb, jb);
574 jl->j_list_bitmap->journal_list = NULL;
575 jl->j_list_bitmap = NULL;
578 static int journal_list_still_alive(struct super_block *s,
579 unsigned int trans_id)
581 struct reiserfs_journal *journal = SB_JOURNAL(s);
582 struct list_head *entry = &journal->j_journal_list;
583 struct reiserfs_journal_list *jl;
585 if (!list_empty(entry)) {
586 jl = JOURNAL_LIST_ENTRY(entry->next);
587 if (jl->j_trans_id <= trans_id) {
595 * If page->mapping was null, we failed to truncate this page for
596 * some reason. Most likely because it was truncated after being
597 * logged via data=journal.
599 * This does a check to see if the buffer belongs to one of these
600 * lost pages before doing the final put_bh. If page->mapping was
601 * null, it tries to free buffers on the page, which should make the
602 * final page_cache_release drop the page from the lru.
604 static void release_buffer_page(struct buffer_head *bh)
606 struct page *page = bh->b_page;
607 if (!page->mapping && trylock_page(page)) {
608 page_cache_get(page);
611 try_to_free_buffers(page);
613 page_cache_release(page);
619 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
621 char b[BDEVNAME_SIZE];
623 if (buffer_journaled(bh)) {
624 reiserfs_warning(NULL, "clm-2084",
625 "pinned buffer %lu:%s sent to disk",
626 bh->b_blocknr, bdevname(bh->b_bdev, b));
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(WRITE, 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(WRITE, 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(WRITE, 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(WRITE, 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, WRITE_FLUSH_FUA);
1118 sync_dirty_buffer(jl->j_commit_bh);
1119 reiserfs_write_lock_nested(s, depth);
1123 * If there was a write error in the journal - we can't commit this
1124 * transaction - it will be invalid and, if successful, will just end
1125 * up propagating the write error out to the filesystem.
1127 if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1128 #ifdef CONFIG_REISERFS_CHECK
1129 reiserfs_warning(s, "journal-615", "buffer write failed");
1133 bforget(jl->j_commit_bh);
1134 if (journal->j_last_commit_id != 0 &&
1135 (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1136 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1137 journal->j_last_commit_id, jl->j_trans_id);
1139 journal->j_last_commit_id = jl->j_trans_id;
1142 * now, every commit block is on the disk. It is safe to allow
1143 * blocks freed during this transaction to be reallocated
1145 cleanup_freed_for_journal_list(s, jl);
1147 retval = retval ? retval : journal->j_errno;
1149 /* mark the metadata dirty */
1151 dirty_one_transaction(s, jl);
1152 atomic_dec(&jl->j_commit_left);
1155 atomic_set(&jl->j_older_commits_done, 1);
1157 mutex_unlock(&jl->j_commit_mutex);
1159 put_journal_list(s, jl);
1162 reiserfs_abort(s, retval, "Journal write error in %s",
1168 * flush_journal_list frequently needs to find a newer transaction for a
1169 * given block. This does that, or returns NULL if it can't find anything
1171 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1172 reiserfs_journal_cnode
1175 struct super_block *sb = cn->sb;
1176 b_blocknr_t blocknr = cn->blocknr;
1180 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1188 static void remove_journal_hash(struct super_block *,
1189 struct reiserfs_journal_cnode **,
1190 struct reiserfs_journal_list *, unsigned long,
1194 * once all the real blocks have been flushed, it is safe to remove them
1195 * from the journal list for this transaction. Aside from freeing the
1196 * cnode, this also allows the block to be reallocated for data blocks
1197 * if it had been deleted.
1199 static void remove_all_from_journal_list(struct super_block *sb,
1200 struct reiserfs_journal_list *jl,
1203 struct reiserfs_journal *journal = SB_JOURNAL(sb);
1204 struct reiserfs_journal_cnode *cn, *last;
1205 cn = jl->j_realblock;
1208 * which is better, to lock once around the whole loop, or
1209 * to lock for each call to remove_journal_hash?
1212 if (cn->blocknr != 0) {
1214 reiserfs_warning(sb, "reiserfs-2201",
1215 "block %u, bh is %d, state %ld",
1216 cn->blocknr, cn->bh ? 1 : 0,
1220 remove_journal_hash(sb, journal->j_list_hash_table,
1221 jl, cn->blocknr, 1);
1225 free_cnode(sb, last);
1227 jl->j_realblock = NULL;
1231 * if this timestamp is greater than the timestamp we wrote last to the
1232 * header block, write it to the header block. once this is done, I can
1233 * safely say the log area for this transaction won't ever be replayed,
1234 * and I can start releasing blocks in this transaction for reuse as data
1235 * blocks. called by flush_journal_list, before it calls
1236 * remove_all_from_journal_list
1238 static int _update_journal_header_block(struct super_block *sb,
1239 unsigned long offset,
1240 unsigned int trans_id)
1242 struct reiserfs_journal_header *jh;
1243 struct reiserfs_journal *journal = SB_JOURNAL(sb);
1246 if (reiserfs_is_journal_aborted(journal))
1249 if (trans_id >= journal->j_last_flush_trans_id) {
1250 if (buffer_locked((journal->j_header_bh))) {
1251 depth = reiserfs_write_unlock_nested(sb);
1252 __wait_on_buffer(journal->j_header_bh);
1253 reiserfs_write_lock_nested(sb, depth);
1254 if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1255 #ifdef CONFIG_REISERFS_CHECK
1256 reiserfs_warning(sb, "journal-699",
1257 "buffer write failed");
1262 journal->j_last_flush_trans_id = trans_id;
1263 journal->j_first_unflushed_offset = offset;
1264 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1266 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1267 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1268 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1270 set_buffer_dirty(journal->j_header_bh);
1271 depth = reiserfs_write_unlock_nested(sb);
1273 if (reiserfs_barrier_flush(sb))
1274 __sync_dirty_buffer(journal->j_header_bh, WRITE_FLUSH_FUA);
1276 sync_dirty_buffer(journal->j_header_bh);
1278 reiserfs_write_lock_nested(sb, depth);
1279 if (!buffer_uptodate(journal->j_header_bh)) {
1280 reiserfs_warning(sb, "journal-837",
1281 "IO error during journal replay");
1288 static int update_journal_header_block(struct super_block *sb,
1289 unsigned long offset,
1290 unsigned int trans_id)
1292 return _update_journal_header_block(sb, offset, trans_id);
1296 ** flush any and all journal lists older than you are
1297 ** can only be called from flush_journal_list
1299 static int flush_older_journal_lists(struct super_block *sb,
1300 struct reiserfs_journal_list *jl)
1302 struct list_head *entry;
1303 struct reiserfs_journal_list *other_jl;
1304 struct reiserfs_journal *journal = SB_JOURNAL(sb);
1305 unsigned int trans_id = jl->j_trans_id;
1308 * we know we are the only ones flushing things, no extra race
1309 * protection is required.
1312 entry = journal->j_journal_list.next;
1314 if (entry == &journal->j_journal_list)
1316 other_jl = JOURNAL_LIST_ENTRY(entry);
1317 if (other_jl->j_trans_id < trans_id) {
1318 BUG_ON(other_jl->j_refcount <= 0);
1319 /* do not flush all */
1320 flush_journal_list(sb, other_jl, 0);
1322 /* other_jl is now deleted from the list */
1328 static void del_from_work_list(struct super_block *s,
1329 struct reiserfs_journal_list *jl)
1331 struct reiserfs_journal *journal = SB_JOURNAL(s);
1332 if (!list_empty(&jl->j_working_list)) {
1333 list_del_init(&jl->j_working_list);
1334 journal->j_num_work_lists--;
1339 * flush a journal list, both commit and real blocks
1341 * always set flushall to 1, unless you are calling from inside
1342 * flush_journal_list
1344 * IMPORTANT. This can only be called while there are no journal writers,
1345 * and the journal is locked. That means it can only be called from
1346 * do_journal_end, or by journal_release
1348 static int flush_journal_list(struct super_block *s,
1349 struct reiserfs_journal_list *jl, int flushall)
1351 struct reiserfs_journal_list *pjl;
1352 struct reiserfs_journal_cnode *cn, *last;
1356 struct buffer_head *saved_bh;
1357 unsigned long j_len_saved = jl->j_len;
1358 struct reiserfs_journal *journal = SB_JOURNAL(s);
1362 BUG_ON(j_len_saved <= 0);
1364 if (atomic_read(&journal->j_wcount) != 0) {
1365 reiserfs_warning(s, "clm-2048", "called with wcount %d",
1366 atomic_read(&journal->j_wcount));
1369 /* if flushall == 0, the lock is already held */
1371 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1372 } else if (mutex_trylock(&journal->j_flush_mutex)) {
1377 if (j_len_saved > journal->j_trans_max) {
1378 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1379 j_len_saved, jl->j_trans_id);
1383 /* if all the work is already done, get out of here */
1384 if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1385 atomic_read(&jl->j_commit_left) <= 0) {
1386 goto flush_older_and_return;
1390 * start by putting the commit list on disk. This will also flush
1391 * the commit lists of any olders transactions
1393 flush_commit_list(s, jl, 1);
1395 if (!(jl->j_state & LIST_DIRTY)
1396 && !reiserfs_is_journal_aborted(journal))
1399 /* are we done now? */
1400 if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1401 atomic_read(&jl->j_commit_left) <= 0) {
1402 goto flush_older_and_return;
1406 * loop through each cnode, see if we need to write it,
1407 * or wait on a more recent transaction, or just ignore it
1409 if (atomic_read(&journal->j_wcount) != 0) {
1410 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1413 cn = jl->j_realblock;
1418 /* blocknr of 0 is no longer in the hash, ignore it */
1419 if (cn->blocknr == 0) {
1424 * This transaction failed commit.
1425 * Don't write out to the disk
1427 if (!(jl->j_state & LIST_DIRTY))
1430 pjl = find_newer_jl_for_cn(cn);
1432 * the order is important here. We check pjl to make sure we
1433 * don't clear BH_JDirty_wait if we aren't the one writing this
1436 if (!pjl && cn->bh) {
1440 * we do this to make sure nobody releases the
1441 * buffer while we are working with it
1445 if (buffer_journal_dirty(saved_bh)) {
1446 BUG_ON(!can_dirty(cn));
1449 } else if (can_dirty(cn)) {
1451 * everything with !pjl && jwait
1452 * should be writable
1459 * if someone has this block in a newer transaction, just make
1460 * sure they are committed, and don't try writing it to disk
1463 if (atomic_read(&pjl->j_commit_left))
1464 flush_commit_list(s, pjl, 1);
1469 * bh == NULL when the block got to disk on its own, OR,
1470 * the block got freed in a future transaction
1472 if (saved_bh == NULL) {
1477 * this should never happen. kupdate_one_transaction has
1478 * this list locked while it works, so we should never see a
1479 * buffer here that is not marked JDirty_wait
1481 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1482 reiserfs_warning(s, "journal-813",
1483 "BAD! buffer %llu %cdirty %cjwait, "
1484 "not in a newer tranasction",
1485 (unsigned long long)saved_bh->
1486 b_blocknr, was_dirty ? ' ' : '!',
1487 was_jwait ? ' ' : '!');
1491 * we inc again because saved_bh gets decremented
1495 set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1496 lock_buffer(saved_bh);
1497 BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1498 if (buffer_dirty(saved_bh))
1499 submit_logged_buffer(saved_bh);
1501 unlock_buffer(saved_bh);
1504 reiserfs_warning(s, "clm-2082",
1505 "Unable to flush buffer %llu in %s",
1506 (unsigned long long)saved_bh->
1507 b_blocknr, __func__);
1514 * we incremented this to keep others from
1515 * taking the buffer head away
1518 if (atomic_read(&saved_bh->b_count) < 0) {
1519 reiserfs_warning(s, "journal-945",
1520 "saved_bh->b_count < 0");
1525 cn = jl->j_realblock;
1527 if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1529 reiserfs_panic(s, "journal-1011",
1533 depth = reiserfs_write_unlock_nested(s);
1534 __wait_on_buffer(cn->bh);
1535 reiserfs_write_lock_nested(s, depth);
1538 reiserfs_panic(s, "journal-1012",
1541 if (unlikely(!buffer_uptodate(cn->bh))) {
1542 #ifdef CONFIG_REISERFS_CHECK
1543 reiserfs_warning(s, "journal-949",
1544 "buffer write failed");
1549 * note, we must clear the JDirty_wait bit
1550 * after the up to date check, otherwise we
1551 * race against our flushpage routine
1553 BUG_ON(!test_clear_buffer_journal_dirty
1556 /* drop one ref for us */
1558 /* drop one ref for journal_mark_dirty */
1559 release_buffer_page(cn->bh);
1566 reiserfs_abort(s, -EIO,
1567 "Write error while pushing transaction to disk in %s",
1569 flush_older_and_return:
1572 * before we can update the journal header block, we _must_ flush all
1573 * real blocks from all older transactions to disk. This is because
1574 * once the header block is updated, this transaction will not be
1575 * replayed after a crash
1578 flush_older_journal_lists(s, jl);
1581 err = journal->j_errno;
1583 * before we can remove everything from the hash tables for this
1584 * transaction, we must make sure it can never be replayed
1586 * since we are only called from do_journal_end, we know for sure there
1587 * are no allocations going on while we are flushing journal lists. So,
1588 * we only need to update the journal header block for the last list
1591 if (!err && flushall) {
1593 update_journal_header_block(s,
1594 (jl->j_start + jl->j_len +
1595 2) % SB_ONDISK_JOURNAL_SIZE(s),
1598 reiserfs_abort(s, -EIO,
1599 "Write error while updating journal header in %s",
1602 remove_all_from_journal_list(s, jl, 0);
1603 list_del_init(&jl->j_list);
1604 journal->j_num_lists--;
1605 del_from_work_list(s, jl);
1607 if (journal->j_last_flush_id != 0 &&
1608 (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1609 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1610 journal->j_last_flush_id, jl->j_trans_id);
1612 journal->j_last_flush_id = jl->j_trans_id;
1615 * not strictly required since we are freeing the list, but it should
1616 * help find code using dead lists later on
1619 atomic_set(&jl->j_nonzerolen, 0);
1621 jl->j_realblock = NULL;
1622 jl->j_commit_bh = NULL;
1625 put_journal_list(s, jl);
1627 mutex_unlock(&journal->j_flush_mutex);
1631 static int write_one_transaction(struct super_block *s,
1632 struct reiserfs_journal_list *jl,
1633 struct buffer_chunk *chunk)
1635 struct reiserfs_journal_cnode *cn;
1638 jl->j_state |= LIST_TOUCHED;
1639 del_from_work_list(s, jl);
1640 if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1644 cn = jl->j_realblock;
1647 * if the blocknr == 0, this has been cleared from the hash,
1650 if (cn->blocknr == 0) {
1653 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1654 struct buffer_head *tmp_bh;
1656 * we can race against journal_mark_freed when we try
1657 * to lock_buffer(cn->bh), so we have to inc the buffer
1658 * count, and recheck things after locking
1662 lock_buffer(tmp_bh);
1663 if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1664 if (!buffer_journal_dirty(tmp_bh) ||
1665 buffer_journal_prepared(tmp_bh))
1667 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1670 /* note, cn->bh might be null now */
1671 unlock_buffer(tmp_bh);
1682 /* used by flush_commit_list */
1683 static int dirty_one_transaction(struct super_block *s,
1684 struct reiserfs_journal_list *jl)
1686 struct reiserfs_journal_cnode *cn;
1687 struct reiserfs_journal_list *pjl;
1690 jl->j_state |= LIST_DIRTY;
1691 cn = jl->j_realblock;
1694 * look for a more recent transaction that logged this
1695 * buffer. Only the most recent transaction with a buffer in
1696 * it is allowed to send that buffer to disk
1698 pjl = find_newer_jl_for_cn(cn);
1699 if (!pjl && cn->blocknr && cn->bh
1700 && buffer_journal_dirty(cn->bh)) {
1701 BUG_ON(!can_dirty(cn));
1703 * if the buffer is prepared, it will either be logged
1704 * or restored. If restored, we need to make sure
1705 * it actually gets marked dirty
1707 clear_buffer_journal_new(cn->bh);
1708 if (buffer_journal_prepared(cn->bh)) {
1709 set_buffer_journal_restore_dirty(cn->bh);
1711 set_buffer_journal_test(cn->bh);
1712 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 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->s_flags & MS_RDONLY)) {
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 MS_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(get_desc_trans_len(desc) *
2195 sizeof(struct buffer_head *), GFP_NOFS);
2196 real_blocks = kmalloc(get_desc_trans_len(desc) *
2197 sizeof(struct buffer_head *), GFP_NOFS);
2198 if (!log_blocks || !real_blocks) {
2203 reiserfs_warning(sb, "journal-1169",
2204 "kmalloc failed, unable to mount FS");
2207 /* get all the buffer heads */
2208 trans_half = journal_trans_half(sb->s_blocksize);
2209 for (i = 0; i < get_desc_trans_len(desc); i++) {
2212 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2214 i) % SB_ONDISK_JOURNAL_SIZE(sb));
2215 if (i < trans_half) {
2218 le32_to_cpu(desc->j_realblock[i]));
2222 le32_to_cpu(commit->
2223 j_realblock[i - trans_half]));
2225 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2226 reiserfs_warning(sb, "journal-1207",
2227 "REPLAY FAILURE fsck required! "
2228 "Block to replay is outside of "
2232 /* make sure we don't try to replay onto log or reserved area */
2233 if (is_block_in_log_or_reserved_area
2234 (sb, real_blocks[i]->b_blocknr)) {
2235 reiserfs_warning(sb, "journal-1204",
2236 "REPLAY FAILURE fsck required! "
2237 "Trying to replay onto a log block");
2239 brelse_array(log_blocks, i);
2240 brelse_array(real_blocks, i);
2248 /* read in the log blocks, memcpy to the corresponding real block */
2249 ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2250 for (i = 0; i < get_desc_trans_len(desc); i++) {
2252 wait_on_buffer(log_blocks[i]);
2253 if (!buffer_uptodate(log_blocks[i])) {
2254 reiserfs_warning(sb, "journal-1212",
2255 "REPLAY FAILURE fsck required! "
2256 "buffer write failed");
2257 brelse_array(log_blocks + i,
2258 get_desc_trans_len(desc) - i);
2259 brelse_array(real_blocks, get_desc_trans_len(desc));
2266 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2267 real_blocks[i]->b_size);
2268 set_buffer_uptodate(real_blocks[i]);
2269 brelse(log_blocks[i]);
2271 /* flush out the real blocks */
2272 for (i = 0; i < get_desc_trans_len(desc); i++) {
2273 set_buffer_dirty(real_blocks[i]);
2274 write_dirty_buffer(real_blocks[i], WRITE);
2276 for (i = 0; i < get_desc_trans_len(desc); i++) {
2277 wait_on_buffer(real_blocks[i]);
2278 if (!buffer_uptodate(real_blocks[i])) {
2279 reiserfs_warning(sb, "journal-1226",
2280 "REPLAY FAILURE, fsck required! "
2281 "buffer write failed");
2282 brelse_array(real_blocks + i,
2283 get_desc_trans_len(desc) - i);
2290 brelse(real_blocks[i]);
2293 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2294 ((trans_offset + get_desc_trans_len(desc) +
2295 2) % SB_ONDISK_JOURNAL_SIZE(sb));
2296 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2297 "journal-1095: setting journal " "start to offset %ld",
2298 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2301 * init starting values for the first transaction, in case
2302 * this is the last transaction to be replayed.
2304 journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2305 journal->j_last_flush_trans_id = trans_id;
2306 journal->j_trans_id = trans_id + 1;
2307 /* check for trans_id overflow */
2308 if (journal->j_trans_id == 0)
2309 journal->j_trans_id = 10;
2318 * This function reads blocks starting from block and to max_block of bufsize
2319 * size (but no more than BUFNR blocks at a time). This proved to improve
2320 * mounting speed on self-rebuilding raid5 arrays at least.
2321 * Right now it is only used from journal code. But later we might use it
2322 * from other places.
2323 * Note: Do not use journal_getblk/sb_getblk functions here!
2325 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2326 b_blocknr_t block, int bufsize,
2327 b_blocknr_t max_block)
2329 struct buffer_head *bhlist[BUFNR];
2330 unsigned int blocks = BUFNR;
2331 struct buffer_head *bh;
2334 bh = __getblk(dev, block, bufsize);
2335 if (buffer_uptodate(bh))
2338 if (block + BUFNR > max_block) {
2339 blocks = max_block - block;
2343 for (i = 1; i < blocks; i++) {
2344 bh = __getblk(dev, block + i, bufsize);
2345 if (buffer_uptodate(bh)) {
2351 ll_rw_block(READ, j, bhlist);
2352 for (i = 1; i < j; i++)
2356 if (buffer_uptodate(bh))
2363 * read and replay the log
2364 * on a clean unmount, the journal header's next unflushed pointer will be
2365 * to an invalid transaction. This tests that before finding all the
2366 * transactions in the log, which makes normal mount times fast.
2368 * After a crash, this starts with the next unflushed transaction, and
2369 * replays until it finds one too old, or invalid.
2371 * On exit, it sets things up so the first transaction will work correctly.
2372 * NOTE: only called during fs mount
2374 static int journal_read(struct super_block *sb)
2376 struct reiserfs_journal *journal = SB_JOURNAL(sb);
2377 struct reiserfs_journal_desc *desc;
2378 unsigned int oldest_trans_id = 0;
2379 unsigned int oldest_invalid_trans_id = 0;
2381 unsigned long oldest_start = 0;
2382 unsigned long cur_dblock = 0;
2383 unsigned long newest_mount_id = 9;
2384 struct buffer_head *d_bh;
2385 struct reiserfs_journal_header *jh;
2386 int valid_journal_header = 0;
2387 int replay_count = 0;
2388 int continue_replay = 1;
2390 char b[BDEVNAME_SIZE];
2392 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2393 reiserfs_info(sb, "checking transaction log (%s)\n",
2394 bdevname(journal->j_dev_bd, b));
2395 start = get_seconds();
2398 * step 1, read in the journal header block. Check the transaction
2399 * it says is the first unflushed, and if that transaction is not
2400 * valid, replay is done
2402 journal->j_header_bh = journal_bread(sb,
2403 SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2404 + SB_ONDISK_JOURNAL_SIZE(sb));
2405 if (!journal->j_header_bh) {
2408 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2409 if (le32_to_cpu(jh->j_first_unflushed_offset) <
2410 SB_ONDISK_JOURNAL_SIZE(sb)
2411 && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2413 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2414 le32_to_cpu(jh->j_first_unflushed_offset);
2415 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2416 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2417 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2418 "journal-1153: found in "
2419 "header: first_unflushed_offset %d, last_flushed_trans_id "
2420 "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2421 le32_to_cpu(jh->j_last_flush_trans_id));
2422 valid_journal_header = 1;
2425 * now, we try to read the first unflushed offset. If it
2426 * is not valid, there is nothing more we can do, and it
2427 * makes no sense to read through the whole log.
2431 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2432 le32_to_cpu(jh->j_first_unflushed_offset));
2433 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2435 continue_replay = 0;
2438 goto start_log_replay;
2442 * ok, there are transactions that need to be replayed. start
2443 * with the first log block, find all the valid transactions, and
2444 * pick out the oldest.
2446 while (continue_replay
2448 (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2449 SB_ONDISK_JOURNAL_SIZE(sb))) {
2451 * Note that it is required for blocksize of primary fs
2452 * device and journal device to be the same
2455 reiserfs_breada(journal->j_dev_bd, cur_dblock,
2457 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2458 SB_ONDISK_JOURNAL_SIZE(sb));
2460 journal_transaction_is_valid(sb, d_bh,
2461 &oldest_invalid_trans_id,
2464 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2465 if (oldest_start == 0) { /* init all oldest_ values */
2466 oldest_trans_id = get_desc_trans_id(desc);
2467 oldest_start = d_bh->b_blocknr;
2468 newest_mount_id = get_desc_mount_id(desc);
2469 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2470 "journal-1179: Setting "
2471 "oldest_start to offset %llu, trans_id %lu",
2473 SB_ONDISK_JOURNAL_1st_BLOCK
2474 (sb), oldest_trans_id);
2475 } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2476 /* one we just read was older */
2477 oldest_trans_id = get_desc_trans_id(desc);
2478 oldest_start = d_bh->b_blocknr;
2479 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2480 "journal-1180: Resetting "
2481 "oldest_start to offset %lu, trans_id %lu",
2483 SB_ONDISK_JOURNAL_1st_BLOCK
2484 (sb), oldest_trans_id);
2486 if (newest_mount_id < get_desc_mount_id(desc)) {
2487 newest_mount_id = get_desc_mount_id(desc);
2488 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2489 "journal-1299: Setting "
2490 "newest_mount_id to %d",
2491 get_desc_mount_id(desc));
2493 cur_dblock += get_desc_trans_len(desc) + 2;
2501 cur_dblock = oldest_start;
2502 if (oldest_trans_id) {
2503 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2504 "journal-1206: Starting replay "
2505 "from offset %llu, trans_id %lu",
2506 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2511 while (continue_replay && oldest_trans_id > 0) {
2513 journal_read_transaction(sb, cur_dblock, oldest_start,
2514 oldest_trans_id, newest_mount_id);
2517 } else if (ret != 0) {
2521 SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2523 if (cur_dblock == oldest_start)
2527 if (oldest_trans_id == 0) {
2528 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2529 "journal-1225: No valid " "transactions found");
2532 * j_start does not get set correctly if we don't replay any
2533 * transactions. if we had a valid journal_header, set j_start
2534 * to the first unflushed transaction value, copy the trans_id
2537 if (valid_journal_header && replay_count == 0) {
2538 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2539 journal->j_trans_id =
2540 le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2541 /* check for trans_id overflow */
2542 if (journal->j_trans_id == 0)
2543 journal->j_trans_id = 10;
2544 journal->j_last_flush_trans_id =
2545 le32_to_cpu(jh->j_last_flush_trans_id);
2546 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2548 journal->j_mount_id = newest_mount_id + 1;
2550 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2551 "newest_mount_id to %lu", journal->j_mount_id);
2552 journal->j_first_unflushed_offset = journal->j_start;
2553 if (replay_count > 0) {
2555 "replayed %d transactions in %lu seconds\n",
2556 replay_count, get_seconds() - start);
2558 /* needed to satisfy the locking in _update_journal_header_block */
2559 reiserfs_write_lock(sb);
2560 if (!bdev_read_only(sb->s_bdev) &&
2561 _update_journal_header_block(sb, journal->j_start,
2562 journal->j_last_flush_trans_id)) {
2563 reiserfs_write_unlock(sb);
2565 * replay failed, caller must call free_journal_ram and abort
2570 reiserfs_write_unlock(sb);
2574 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2576 struct reiserfs_journal_list *jl;
2577 jl = kzalloc(sizeof(struct reiserfs_journal_list),
2578 GFP_NOFS | __GFP_NOFAIL);
2579 INIT_LIST_HEAD(&jl->j_list);
2580 INIT_LIST_HEAD(&jl->j_working_list);
2581 INIT_LIST_HEAD(&jl->j_tail_bh_list);
2582 INIT_LIST_HEAD(&jl->j_bh_list);
2583 mutex_init(&jl->j_commit_mutex);
2584 SB_JOURNAL(s)->j_num_lists++;
2585 get_journal_list(jl);
2589 static void journal_list_init(struct super_block *sb)
2591 SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2594 static void release_journal_dev(struct super_block *super,
2595 struct reiserfs_journal *journal)
2597 if (journal->j_dev_bd != NULL) {
2598 blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2599 journal->j_dev_bd = NULL;
2603 static int journal_init_dev(struct super_block *super,
2604 struct reiserfs_journal *journal,
2605 const char *jdev_name)
2609 fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
2610 char b[BDEVNAME_SIZE];
2614 journal->j_dev_bd = NULL;
2615 jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2616 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2618 if (bdev_read_only(super->s_bdev))
2619 blkdev_mode = FMODE_READ;
2621 /* there is no "jdev" option and journal is on separate device */
2622 if ((!jdev_name || !jdev_name[0])) {
2623 if (jdev == super->s_dev)
2624 blkdev_mode &= ~FMODE_EXCL;
2625 journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2627 journal->j_dev_mode = blkdev_mode;
2628 if (IS_ERR(journal->j_dev_bd)) {
2629 result = PTR_ERR(journal->j_dev_bd);
2630 journal->j_dev_bd = NULL;
2631 reiserfs_warning(super, "sh-458",
2632 "cannot init journal device '%s': %i",
2633 __bdevname(jdev, b), result);
2635 } else if (jdev != super->s_dev)
2636 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2641 journal->j_dev_mode = blkdev_mode;
2642 journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
2643 if (IS_ERR(journal->j_dev_bd)) {
2644 result = PTR_ERR(journal->j_dev_bd);
2645 journal->j_dev_bd = NULL;
2646 reiserfs_warning(super, "sh-457",
2647 "journal_init_dev: Cannot open '%s': %i",
2652 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2653 reiserfs_info(super,
2654 "journal_init_dev: journal device: %s\n",
2655 bdevname(journal->j_dev_bd, b));
2660 * When creating/tuning a file system user can assign some
2661 * journal params within boundaries which depend on the ratio
2662 * blocksize/standard_blocksize.
2664 * For blocks >= standard_blocksize transaction size should
2665 * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2666 * then JOURNAL_TRANS_MAX_DEFAULT.
2668 * For blocks < standard_blocksize these boundaries should be
2669 * decreased proportionally.
2671 #define REISERFS_STANDARD_BLKSIZE (4096)
2673 static int check_advise_trans_params(struct super_block *sb,
2674 struct reiserfs_journal *journal)
2676 if (journal->j_trans_max) {
2677 /* Non-default journal params. Do sanity check for them. */
2679 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2680 ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2682 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2683 journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2684 SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2685 JOURNAL_MIN_RATIO) {
2686 reiserfs_warning(sb, "sh-462",
2687 "bad transaction max size (%u). "
2688 "FSCK?", journal->j_trans_max);
2691 if (journal->j_max_batch != (journal->j_trans_max) *
2692 JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2693 reiserfs_warning(sb, "sh-463",
2694 "bad transaction max batch (%u). "
2695 "FSCK?", journal->j_max_batch);
2700 * Default journal params.
2701 * The file system was created by old version
2702 * of mkreiserfs, so some fields contain zeros,
2703 * and we need to advise proper values for them
2705 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2706 reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2710 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2711 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2712 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2717 /* must be called once on fs mount. calls journal_read for you */
2718 int journal_init(struct super_block *sb, const char *j_dev_name,
2719 int old_format, unsigned int commit_max_age)
2721 int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2722 struct buffer_head *bhjh;
2723 struct reiserfs_super_block *rs;
2724 struct reiserfs_journal_header *jh;
2725 struct reiserfs_journal *journal;
2726 struct reiserfs_journal_list *jl;
2727 char b[BDEVNAME_SIZE];
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;
2774 * Sanity check to see if journal first block is correct.
2775 * If journal first block is invalid it can cause
2776 * zeroing important superblock members.
2778 if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2779 SB_ONDISK_JOURNAL_1st_BLOCK(sb) < SB_JOURNAL_1st_RESERVED_BLOCK(sb)) {
2780 reiserfs_warning(sb, "journal-1393",
2781 "journal 1st super block is invalid: 1st reserved block %d, but actual 1st block is %d",
2782 SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2783 SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2784 goto free_and_return;
2787 if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2788 reiserfs_warning(sb, "sh-462",
2789 "unable to initialize journal device");
2790 goto free_and_return;
2793 rs = SB_DISK_SUPER_BLOCK(sb);
2795 /* read journal header */
2796 bhjh = journal_bread(sb,
2797 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2798 SB_ONDISK_JOURNAL_SIZE(sb));
2800 reiserfs_warning(sb, "sh-459",
2801 "unable to read journal header");
2802 goto free_and_return;
2804 jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2806 /* make sure that journal matches to the super block */
2807 if (is_reiserfs_jr(rs)
2808 && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2809 sb_jp_journal_magic(rs))) {
2810 reiserfs_warning(sb, "sh-460",
2811 "journal header magic %x (device %s) does "
2812 "not match to magic found in super block %x",
2813 jh->jh_journal.jp_journal_magic,
2814 bdevname(journal->j_dev_bd, b),
2815 sb_jp_journal_magic(rs));
2817 goto free_and_return;
2820 journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2821 journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2822 journal->j_max_commit_age =
2823 le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2824 journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2826 if (check_advise_trans_params(sb, journal) != 0)
2827 goto free_and_return;
2828 journal->j_default_max_commit_age = journal->j_max_commit_age;
2830 if (commit_max_age != 0) {
2831 journal->j_max_commit_age = commit_max_age;
2832 journal->j_max_trans_age = commit_max_age;
2835 reiserfs_info(sb, "journal params: device %s, size %u, "
2836 "journal first block %u, max trans len %u, max batch %u, "
2837 "max commit age %u, max trans age %u\n",
2838 bdevname(journal->j_dev_bd, b),
2839 SB_ONDISK_JOURNAL_SIZE(sb),
2840 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2841 journal->j_trans_max,
2842 journal->j_max_batch,
2843 journal->j_max_commit_age, journal->j_max_trans_age);
2847 journal->j_list_bitmap_index = 0;
2848 journal_list_init(sb);
2850 memset(journal->j_list_hash_table, 0,
2851 JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2853 INIT_LIST_HEAD(&journal->j_dirty_buffers);
2854 spin_lock_init(&journal->j_dirty_buffers_lock);
2856 journal->j_start = 0;
2858 journal->j_len_alloc = 0;
2859 atomic_set(&journal->j_wcount, 0);
2860 atomic_set(&journal->j_async_throttle, 0);
2861 journal->j_bcount = 0;
2862 journal->j_trans_start_time = 0;
2863 journal->j_last = NULL;
2864 journal->j_first = NULL;
2865 init_waitqueue_head(&journal->j_join_wait);
2866 mutex_init(&journal->j_mutex);
2867 mutex_init(&journal->j_flush_mutex);
2869 journal->j_trans_id = 10;
2870 journal->j_mount_id = 10;
2871 journal->j_state = 0;
2872 atomic_set(&journal->j_jlock, 0);
2873 journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2874 journal->j_cnode_free_orig = journal->j_cnode_free_list;
2875 journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2876 journal->j_cnode_used = 0;
2877 journal->j_must_wait = 0;
2879 if (journal->j_cnode_free == 0) {
2880 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2881 "allocation failed (%ld bytes). Journal is "
2882 "too large for available memory. Usually "
2883 "this is due to a journal that is too large.",
2884 sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2885 goto free_and_return;
2888 init_journal_hash(sb);
2889 jl = journal->j_current_jl;
2892 * get_list_bitmap() may call flush_commit_list() which
2893 * requires the lock. Calling flush_commit_list() shouldn't happen
2894 * this early but I like to be paranoid.
2896 reiserfs_write_lock(sb);
2897 jl->j_list_bitmap = get_list_bitmap(sb, jl);
2898 reiserfs_write_unlock(sb);
2899 if (!jl->j_list_bitmap) {
2900 reiserfs_warning(sb, "journal-2005",
2901 "get_list_bitmap failed for journal list 0");
2902 goto free_and_return;
2905 ret = journal_read(sb);
2907 reiserfs_warning(sb, "reiserfs-2006",
2908 "Replay Failure, unable to mount");
2909 goto free_and_return;
2912 INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2913 journal->j_work_sb = sb;
2916 free_journal_ram(sb);
2921 * test for a polite end of the current transaction. Used by file_write,
2922 * and should be used by delete to make sure they don't write more than
2923 * can fit inside a single transaction
2925 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2928 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2929 time_t now = get_seconds();
2930 /* cannot restart while nested */
2931 BUG_ON(!th->t_trans_id);
2932 if (th->t_refcount > 1)
2934 if (journal->j_must_wait > 0 ||
2935 (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2936 atomic_read(&journal->j_jlock) ||
2937 (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2938 journal->j_cnode_free < (journal->j_trans_max * 3)) {
2942 journal->j_len_alloc += new_alloc;
2943 th->t_blocks_allocated += new_alloc ;
2947 /* this must be called inside a transaction */
2948 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2950 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2951 BUG_ON(!th->t_trans_id);
2952 journal->j_must_wait = 1;
2953 set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2957 /* this must be called without a transaction started */
2958 void reiserfs_allow_writes(struct super_block *s)
2960 struct reiserfs_journal *journal = SB_JOURNAL(s);
2961 clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2962 wake_up(&journal->j_join_wait);
2965 /* this must be called without a transaction started */
2966 void reiserfs_wait_on_write_block(struct super_block *s)
2968 struct reiserfs_journal *journal = SB_JOURNAL(s);
2969 wait_event(journal->j_join_wait,
2970 !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2973 static void queue_log_writer(struct super_block *s)
2976 struct reiserfs_journal *journal = SB_JOURNAL(s);
2977 set_bit(J_WRITERS_QUEUED, &journal->j_state);
2980 * we don't want to use wait_event here because
2981 * we only want to wait once.
2983 init_waitqueue_entry(&wait, current);
2984 add_wait_queue(&journal->j_join_wait, &wait);
2985 set_current_state(TASK_UNINTERRUPTIBLE);
2986 if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
2987 int depth = reiserfs_write_unlock_nested(s);
2989 reiserfs_write_lock_nested(s, depth);
2991 __set_current_state(TASK_RUNNING);
2992 remove_wait_queue(&journal->j_join_wait, &wait);
2995 static void wake_queued_writers(struct super_block *s)
2997 struct reiserfs_journal *journal = SB_JOURNAL(s);
2998 if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2999 wake_up(&journal->j_join_wait);
3002 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
3004 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3005 unsigned long bcount = journal->j_bcount;
3009 depth = reiserfs_write_unlock_nested(sb);
3010 schedule_timeout_uninterruptible(1);
3011 reiserfs_write_lock_nested(sb, depth);
3013 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
3014 while ((atomic_read(&journal->j_wcount) > 0 ||
3015 atomic_read(&journal->j_jlock)) &&
3016 journal->j_trans_id == trans_id) {
3017 queue_log_writer(sb);
3019 if (journal->j_trans_id != trans_id)
3021 if (bcount == journal->j_bcount)
3023 bcount = journal->j_bcount;
3028 * join == true if you must join an existing transaction.
3029 * join == false if you can deal with waiting for others to finish
3031 * this will block until the transaction is joinable. send the number of
3032 * blocks you expect to use in nblocks.
3034 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3035 struct super_block *sb, unsigned long nblocks,
3038 time_t now = get_seconds();
3039 unsigned int old_trans_id;
3040 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3041 struct reiserfs_transaction_handle myth;
3042 int sched_count = 0;
3046 reiserfs_check_lock_depth(sb, "journal_begin");
3047 BUG_ON(nblocks > journal->j_trans_max);
3049 PROC_INFO_INC(sb, journal.journal_being);
3050 /* set here for journal_join */
3056 if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3058 retval = journal->j_errno;
3061 journal->j_bcount++;
3063 if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3065 depth = reiserfs_write_unlock_nested(sb);
3066 reiserfs_wait_on_write_block(sb);
3067 reiserfs_write_lock_nested(sb, depth);
3068 PROC_INFO_INC(sb, journal.journal_relock_writers);
3071 now = get_seconds();
3074 * if there is no room in the journal OR
3075 * if this transaction is too old, and we weren't called joinable,
3076 * wait for it to finish before beginning we don't sleep if there
3077 * aren't other writers
3080 if ((!join && journal->j_must_wait > 0) ||
3082 && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3083 || (!join && atomic_read(&journal->j_wcount) > 0
3084 && journal->j_trans_start_time > 0
3085 && (now - journal->j_trans_start_time) >
3086 journal->j_max_trans_age) || (!join
3087 && atomic_read(&journal->j_jlock))
3088 || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3090 old_trans_id = journal->j_trans_id;
3091 /* allow others to finish this transaction */
3094 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3095 journal->j_max_batch &&
3096 ((journal->j_len + nblocks + 2) * 100) <
3097 (journal->j_len_alloc * 75)) {
3098 if (atomic_read(&journal->j_wcount) > 10) {
3100 queue_log_writer(sb);
3105 * don't mess with joining the transaction if all we
3106 * have to do is wait for someone else to do a commit
3108 if (atomic_read(&journal->j_jlock)) {
3109 while (journal->j_trans_id == old_trans_id &&
3110 atomic_read(&journal->j_jlock)) {
3111 queue_log_writer(sb);
3115 retval = journal_join(&myth, sb);
3119 /* someone might have ended the transaction while we joined */
3120 if (old_trans_id != journal->j_trans_id) {
3121 retval = do_journal_end(&myth, 0);
3123 retval = do_journal_end(&myth, COMMIT_NOW);
3129 PROC_INFO_INC(sb, journal.journal_relock_wcount);
3132 /* we are the first writer, set trans_id */
3133 if (journal->j_trans_start_time == 0) {
3134 journal->j_trans_start_time = get_seconds();
3136 atomic_inc(&journal->j_wcount);
3137 journal->j_len_alloc += nblocks;
3138 th->t_blocks_logged = 0;
3139 th->t_blocks_allocated = nblocks;
3140 th->t_trans_id = journal->j_trans_id;
3142 INIT_LIST_HEAD(&th->t_list);
3146 memset(th, 0, sizeof(*th));
3148 * Re-set th->t_super, so we can properly keep track of how many
3149 * persistent transactions there are. We need to do this so if this
3150 * call is part of a failed restart_transaction, we can free it later
3156 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3162 struct reiserfs_transaction_handle *th;
3165 * if we're nesting into an existing transaction. It will be
3166 * persistent on its own
3168 if (reiserfs_transaction_running(s)) {
3169 th = current->journal_info;
3171 BUG_ON(th->t_refcount < 2);
3175 th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3178 ret = journal_begin(th, s, nblocks);
3184 SB_JOURNAL(s)->j_persistent_trans++;
3188 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3190 struct super_block *s = th->t_super;
3193 ret = journal_end(th);
3196 if (th->t_refcount == 0) {
3197 SB_JOURNAL(s)->j_persistent_trans--;
3203 static int journal_join(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_JOIN);
3217 int journal_join_abort(struct reiserfs_transaction_handle *th,
3218 struct super_block *sb)
3220 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3223 * this keeps do_journal_end from NULLing out the
3224 * current->journal_info pointer
3226 th->t_handle_save = cur_th;
3227 BUG_ON(cur_th && cur_th->t_refcount > 1);
3228 return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT);
3231 int journal_begin(struct reiserfs_transaction_handle *th,
3232 struct super_block *sb, unsigned long nblocks)
3234 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3237 th->t_handle_save = NULL;
3239 /* we are nesting into the current transaction */
3240 if (cur_th->t_super == sb) {
3241 BUG_ON(!cur_th->t_refcount);
3242 cur_th->t_refcount++;
3243 memcpy(th, cur_th, sizeof(*th));
3244 if (th->t_refcount <= 1)
3245 reiserfs_warning(sb, "reiserfs-2005",
3246 "BAD: refcount <= 1, but "
3247 "journal_info != 0");
3251 * we've ended up with a handle from a different
3252 * filesystem. save it and restore on journal_end.
3253 * This should never really happen...
3255 reiserfs_warning(sb, "clm-2100",
3256 "nesting info a different FS");
3257 th->t_handle_save = current->journal_info;
3258 current->journal_info = th;
3261 current->journal_info = th;
3263 ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3264 BUG_ON(current->journal_info != th);
3267 * I guess this boils down to being the reciprocal of clm-2100 above.
3268 * If do_journal_begin_r fails, we need to put it back, since
3269 * journal_end won't be called to do it. */
3271 current->journal_info = th->t_handle_save;
3273 BUG_ON(!th->t_refcount);
3279 * puts bh into the current transaction. If it was already there, reorders
3280 * removes the old pointers from the hash, and puts new ones in (to make
3281 * sure replay happen in the right order).
3283 * if it was dirty, cleans and files onto the clean list. I can't let it
3284 * be dirty again until the transaction is committed.
3286 * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3288 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3289 struct buffer_head *bh)
3291 struct super_block *sb = th->t_super;
3292 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3293 struct reiserfs_journal_cnode *cn = NULL;
3294 int count_already_incd = 0;
3296 BUG_ON(!th->t_trans_id);
3298 PROC_INFO_INC(sb, journal.mark_dirty);
3299 if (th->t_trans_id != journal->j_trans_id) {
3300 reiserfs_panic(th->t_super, "journal-1577",
3301 "handle trans id %ld != current trans id %ld",
3302 th->t_trans_id, journal->j_trans_id);
3305 prepared = test_clear_buffer_journal_prepared(bh);
3306 clear_buffer_journal_restore_dirty(bh);
3307 /* already in this transaction, we are done */
3308 if (buffer_journaled(bh)) {
3309 PROC_INFO_INC(sb, journal.mark_dirty_already);
3314 * this must be turned into a panic instead of a warning. We can't
3315 * allow a dirty or journal_dirty or locked buffer to be logged, as
3316 * some changes could get to disk too early. NOT GOOD.
3318 if (!prepared || buffer_dirty(bh)) {
3319 reiserfs_warning(sb, "journal-1777",
3320 "buffer %llu bad state "
3321 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3322 (unsigned long long)bh->b_blocknr,
3323 prepared ? ' ' : '!',
3324 buffer_locked(bh) ? ' ' : '!',
3325 buffer_dirty(bh) ? ' ' : '!',
3326 buffer_journal_dirty(bh) ? ' ' : '!');
3329 if (atomic_read(&journal->j_wcount) <= 0) {
3330 reiserfs_warning(sb, "journal-1409",
3331 "returning because j_wcount was %d",
3332 atomic_read(&journal->j_wcount));
3336 * this error means I've screwed up, and we've overflowed
3337 * the transaction. Nothing can be done here, except make the
3338 * FS readonly or panic.
3340 if (journal->j_len >= journal->j_trans_max) {
3341 reiserfs_panic(th->t_super, "journal-1413",
3342 "j_len (%lu) is too big",
3346 if (buffer_journal_dirty(bh)) {
3347 count_already_incd = 1;
3348 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3349 clear_buffer_journal_dirty(bh);
3352 if (journal->j_len > journal->j_len_alloc) {
3353 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3356 set_buffer_journaled(bh);
3358 /* now put this guy on the end */
3362 reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3365 if (th->t_blocks_logged == th->t_blocks_allocated) {
3366 th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3367 journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3369 th->t_blocks_logged++;
3373 cn->blocknr = bh->b_blocknr;
3376 insert_journal_hash(journal->j_hash_table, cn);
3377 if (!count_already_incd) {
3382 cn->prev = journal->j_last;
3384 if (journal->j_last) {
3385 journal->j_last->next = cn;
3386 journal->j_last = cn;
3388 journal->j_first = cn;
3389 journal->j_last = cn;
3391 reiserfs_schedule_old_flush(sb);
3395 int journal_end(struct reiserfs_transaction_handle *th)
3397 struct super_block *sb = th->t_super;
3398 if (!current->journal_info && th->t_refcount > 1)
3399 reiserfs_warning(sb, "REISER-NESTING",
3400 "th NULL, refcount %d", th->t_refcount);
3402 if (!th->t_trans_id) {
3408 if (th->t_refcount > 0) {
3409 struct reiserfs_transaction_handle *cur_th =
3410 current->journal_info;
3413 * we aren't allowed to close a nested transaction on a
3414 * different filesystem from the one in the task struct
3416 BUG_ON(cur_th->t_super != th->t_super);
3419 memcpy(current->journal_info, th, sizeof(*th));
3424 return do_journal_end(th, 0);
3429 * removes from the current transaction, relsing and descrementing any counters.
3430 * also files the removed buffer directly onto the clean list
3432 * called by journal_mark_freed when a block has been deleted
3434 * returns 1 if it cleaned and relsed the buffer. 0 otherwise
3436 static int remove_from_transaction(struct super_block *sb,
3437 b_blocknr_t blocknr, int already_cleaned)
3439 struct buffer_head *bh;
3440 struct reiserfs_journal_cnode *cn;
3441 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3444 cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3445 if (!cn || !cn->bh) {
3450 cn->prev->next = cn->next;
3453 cn->next->prev = cn->prev;
3455 if (cn == journal->j_first) {
3456 journal->j_first = cn->next;
3458 if (cn == journal->j_last) {
3459 journal->j_last = cn->prev;
3462 remove_journal_hash(sb, journal->j_hash_table, NULL,
3464 clear_buffer_journaled(bh); /* don't log this one */
3466 if (!already_cleaned) {
3467 clear_buffer_journal_dirty(bh);
3468 clear_buffer_dirty(bh);
3469 clear_buffer_journal_test(bh);
3471 if (atomic_read(&bh->b_count) < 0) {
3472 reiserfs_warning(sb, "journal-1752",
3478 journal->j_len_alloc--;
3484 * for any cnode in a journal list, it can only be dirtied of all the
3485 * transactions that include it are committed to disk.
3486 * this checks through each transaction, and returns 1 if you are allowed
3487 * to dirty, and 0 if you aren't
3489 * it is called by dirty_journal_list, which is called after
3490 * flush_commit_list has gotten all the log blocks for a given
3491 * transaction on disk
3494 static int can_dirty(struct reiserfs_journal_cnode *cn)
3496 struct super_block *sb = cn->sb;
3497 b_blocknr_t blocknr = cn->blocknr;
3498 struct reiserfs_journal_cnode *cur = cn->hprev;
3502 * first test hprev. These are all newer than cn, so any node here
3503 * with the same block number and dev means this node can't be sent
3504 * to disk right now.
3506 while (cur && can_dirty) {
3507 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3508 cur->blocknr == blocknr) {
3514 * then test hnext. These are all older than cn. As long as they
3515 * are committed to the log, it is safe to write cn to disk
3518 while (cur && can_dirty) {
3519 if (cur->jlist && cur->jlist->j_len > 0 &&
3520 atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh &&
3521 cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3530 * syncs the commit blocks, but does not force the real buffers to disk
3531 * will wait until the current transaction is done/committed before returning
3533 int journal_end_sync(struct reiserfs_transaction_handle *th)
3535 struct super_block *sb = th->t_super;
3536 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3538 BUG_ON(!th->t_trans_id);
3539 /* you can sync while nested, very, very bad */
3540 BUG_ON(th->t_refcount > 1);
3541 if (journal->j_len == 0) {
3542 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3544 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
3546 return do_journal_end(th, COMMIT_NOW | WAIT);
3549 /* writeback the pending async commits to disk */
3550 static void flush_async_commits(struct work_struct *work)
3552 struct reiserfs_journal *journal =
3553 container_of(work, struct reiserfs_journal, j_work.work);
3554 struct super_block *sb = journal->j_work_sb;
3555 struct reiserfs_journal_list *jl;
3556 struct list_head *entry;
3558 reiserfs_write_lock(sb);
3559 if (!list_empty(&journal->j_journal_list)) {
3560 /* last entry is the youngest, commit it and you get everything */
3561 entry = journal->j_journal_list.prev;
3562 jl = JOURNAL_LIST_ENTRY(entry);
3563 flush_commit_list(sb, jl, 1);
3565 reiserfs_write_unlock(sb);
3569 * flushes any old transactions to disk
3570 * ends the current transaction if it is too old
3572 void reiserfs_flush_old_commits(struct super_block *sb)
3575 struct reiserfs_transaction_handle th;
3576 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3578 now = get_seconds();
3580 * safety check so we don't flush while we are replaying the log during
3583 if (list_empty(&journal->j_journal_list))
3587 * check the current transaction. If there are no writers, and it is
3588 * too old, finish it, and force the commit blocks to disk
3590 if (atomic_read(&journal->j_wcount) <= 0 &&
3591 journal->j_trans_start_time > 0 &&
3592 journal->j_len > 0 &&
3593 (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3594 if (!journal_join(&th, sb)) {
3595 reiserfs_prepare_for_journal(sb,
3596 SB_BUFFER_WITH_SB(sb),
3598 journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3601 * we're only being called from kreiserfsd, it makes
3602 * no sense to do an async commit so that kreiserfsd
3605 do_journal_end(&th, COMMIT_NOW | WAIT);
3611 * returns 0 if do_journal_end should return right away, returns 1 if
3612 * do_journal_end should finish the commit
3614 * if the current transaction is too old, but still has writers, this will
3615 * wait on j_join_wait until all the writers are done. By the time it
3616 * wakes up, the transaction it was called has already ended, so it just
3617 * flushes the commit list and returns 0.
3619 * Won't batch when flush or commit_now is set. Also won't batch when
3620 * others are waiting on j_join_wait.
3622 * Note, we can't allow the journal_end to proceed while there are still
3623 * writers in the log.
3625 static int check_journal_end(struct reiserfs_transaction_handle *th, int flags)
3629 int flush = flags & FLUSH_ALL;
3630 int commit_now = flags & COMMIT_NOW;
3631 int wait_on_commit = flags & WAIT;
3632 struct reiserfs_journal_list *jl;
3633 struct super_block *sb = th->t_super;
3634 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3636 BUG_ON(!th->t_trans_id);
3638 if (th->t_trans_id != journal->j_trans_id) {
3639 reiserfs_panic(th->t_super, "journal-1577",
3640 "handle trans id %ld != current trans id %ld",
3641 th->t_trans_id, journal->j_trans_id);
3644 journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3645 /* <= 0 is allowed. unmounting might not call begin */
3646 if (atomic_read(&journal->j_wcount) > 0)
3647 atomic_dec(&journal->j_wcount);
3650 * BUG, deal with case where j_len is 0, but people previously
3651 * freed blocks need to be released will be dealt with by next
3652 * transaction that actually writes something, but should be taken
3653 * care of in this trans
3655 BUG_ON(journal->j_len == 0);
3658 * if wcount > 0, and we are called to with flush or commit_now,
3659 * we wait on j_join_wait. We will wake up when the last writer has
3660 * finished the transaction, and started it on its way to the disk.
3661 * Then, we flush the commit or journal list, and just return 0
3662 * because the rest of journal end was already done for this
3665 if (atomic_read(&journal->j_wcount) > 0) {
3666 if (flush || commit_now) {
3669 jl = journal->j_current_jl;
3670 trans_id = jl->j_trans_id;
3672 jl->j_state |= LIST_COMMIT_PENDING;
3673 atomic_set(&journal->j_jlock, 1);
3675 journal->j_next_full_flush = 1;
3680 * sleep while the current transaction is
3683 while (journal->j_trans_id == trans_id) {
3684 if (atomic_read(&journal->j_jlock)) {
3685 queue_log_writer(sb);
3688 if (journal->j_trans_id == trans_id) {
3689 atomic_set(&journal->j_jlock,
3695 BUG_ON(journal->j_trans_id == trans_id);
3698 && journal_list_still_alive(sb, trans_id)
3699 && wait_on_commit) {
3700 flush_commit_list(sb, jl, 1);
3708 /* deal with old transactions where we are the last writers */
3709 now = get_seconds();
3710 if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3712 journal->j_next_async_flush = 1;
3714 /* don't batch when someone is waiting on j_join_wait */
3715 /* don't batch when syncing the commit or flushing the whole trans */
3716 if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock))
3717 && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3718 && journal->j_len_alloc < journal->j_max_batch
3719 && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3720 journal->j_bcount++;
3725 if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3726 reiserfs_panic(sb, "journal-003",
3727 "j_start (%ld) is too high",
3734 * Does all the work that makes deleting blocks safe.
3735 * when deleting a block mark BH_JNew, just remove it from the current
3736 * transaction, clean it's buffer_head and move on.
3739 * set a bit for the block in the journal bitmap. That will prevent it from
3740 * being allocated for unformatted nodes before this transaction has finished.
3742 * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.
3743 * That will prevent any old transactions with this block from trying to flush
3744 * to the real location. Since we aren't removing the cnode from the
3745 * journal_list_hash, *the block can't be reallocated yet.
3747 * Then remove it from the current transaction, decrementing any counters and
3748 * filing it on the clean list.
3750 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3751 struct super_block *sb, b_blocknr_t blocknr)
3753 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3754 struct reiserfs_journal_cnode *cn = NULL;
3755 struct buffer_head *bh = NULL;
3756 struct reiserfs_list_bitmap *jb = NULL;
3758 BUG_ON(!th->t_trans_id);
3760 cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3765 /* if it is journal new, we just remove it from this transaction */
3766 if (bh && buffer_journal_new(bh)) {
3767 clear_buffer_journal_new(bh);
3768 clear_prepared_bits(bh);
3769 reiserfs_clean_and_file_buffer(bh);
3770 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3773 * set the bit for this block in the journal bitmap
3774 * for this transaction
3776 jb = journal->j_current_jl->j_list_bitmap;
3778 reiserfs_panic(sb, "journal-1702",
3779 "journal_list_bitmap is NULL");
3781 set_bit_in_list_bitmap(sb, blocknr, jb);
3783 /* Note, the entire while loop is not allowed to schedule. */
3786 clear_prepared_bits(bh);
3787 reiserfs_clean_and_file_buffer(bh);
3789 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3792 * find all older transactions with this block,
3793 * make sure they don't try to write it out
3795 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3798 if (sb == cn->sb && blocknr == cn->blocknr) {
3799 set_bit(BLOCK_FREED, &cn->state);
3802 * remove_from_transaction will brelse
3803 * the buffer if it was in the current
3807 clear_buffer_journal_dirty(cn->
3809 clear_buffer_dirty(cn->bh);
3810 clear_buffer_journal_test(cn->
3815 (&cn->bh->b_count) < 0) {
3816 reiserfs_warning(sb,
3818 "cn->bh->b_count < 0");
3822 * since we are clearing the bh,
3823 * we MUST dec nonzerolen
3826 atomic_dec(&cn->jlist->
3837 release_buffer_page(bh); /* get_hash grabs the buffer */
3841 void reiserfs_update_inode_transaction(struct inode *inode)
3843 struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3844 REISERFS_I(inode)->i_jl = journal->j_current_jl;
3845 REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3849 * returns -1 on error, 0 if no commits/barriers were done and 1
3850 * if a transaction was actually committed and the barrier was done
3852 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3853 struct reiserfs_journal_list *jl)
3855 struct reiserfs_transaction_handle th;
3856 struct super_block *sb = inode->i_sb;
3857 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3861 * is it from the current transaction,
3862 * or from an unknown transaction?
3864 if (id == journal->j_trans_id) {
3865 jl = journal->j_current_jl;
3867 * try to let other writers come in and
3868 * grow this transaction
3870 let_transaction_grow(sb, id);
3871 if (journal->j_trans_id != id) {
3872 goto flush_commit_only;
3875 ret = journal_begin(&th, sb, 1);
3879 /* someone might have ended this transaction while we joined */
3880 if (journal->j_trans_id != id) {
3881 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3883 journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3884 ret = journal_end(&th);
3885 goto flush_commit_only;
3888 ret = journal_end_sync(&th);
3894 * this gets tricky, we have to make sure the journal list in
3895 * the inode still exists. We know the list is still around
3896 * if we've got a larger transaction id than the oldest list
3899 if (journal_list_still_alive(inode->i_sb, id)) {
3901 * we only set ret to 1 when we know for sure
3902 * the barrier hasn't been started yet on the commit
3905 if (atomic_read(&jl->j_commit_left) > 1)
3907 flush_commit_list(sb, jl, 1);
3908 if (journal->j_errno)
3909 ret = journal->j_errno;
3912 /* otherwise the list is gone, and long since committed */
3916 int reiserfs_commit_for_inode(struct inode *inode)
3918 unsigned int id = REISERFS_I(inode)->i_trans_id;
3919 struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3922 * for the whole inode, assume unset id means it was
3923 * changed in the current transaction. More conservative
3926 reiserfs_update_inode_transaction(inode);
3927 id = REISERFS_I(inode)->i_trans_id;
3928 /* jl will be updated in __commit_trans_jl */
3931 return __commit_trans_jl(inode, id, jl);
3934 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3935 struct buffer_head *bh)
3937 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3938 PROC_INFO_INC(sb, journal.restore_prepared);
3942 if (test_clear_buffer_journal_restore_dirty(bh) &&
3943 buffer_journal_dirty(bh)) {
3944 struct reiserfs_journal_cnode *cn;
3945 reiserfs_write_lock(sb);
3946 cn = get_journal_hash_dev(sb,
3947 journal->j_list_hash_table,
3949 if (cn && can_dirty(cn)) {
3950 set_buffer_journal_test(bh);
3951 mark_buffer_dirty(bh);
3953 reiserfs_write_unlock(sb);
3955 clear_buffer_journal_prepared(bh);
3958 extern struct tree_balance *cur_tb;
3960 * before we can change a metadata block, we have to make sure it won't
3961 * be written to disk while we are altering it. So, we must:
3965 int reiserfs_prepare_for_journal(struct super_block *sb,
3966 struct buffer_head *bh, int wait)
3968 PROC_INFO_INC(sb, journal.prepare);
3970 if (!trylock_buffer(bh)) {
3975 set_buffer_journal_prepared(bh);
3976 if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3977 clear_buffer_journal_test(bh);
3978 set_buffer_journal_restore_dirty(bh);
3985 * long and ugly. If flush, will not return until all commit
3986 * blocks and all real buffers in the trans are on disk.
3987 * If no_async, won't return until all commit blocks are on disk.
3989 * keep reading, there are comments as you go along
3991 * If the journal is aborted, we just clean up. Things like flushing
3992 * journal lists, etc just won't happen.
3994 static int do_journal_end(struct reiserfs_transaction_handle *th, int flags)
3996 struct super_block *sb = th->t_super;
3997 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3998 struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3999 struct reiserfs_journal_cnode *last_cn = NULL;
4000 struct reiserfs_journal_desc *desc;
4001 struct reiserfs_journal_commit *commit;
4002 struct buffer_head *c_bh; /* commit bh */
4003 struct buffer_head *d_bh; /* desc bh */
4004 int cur_write_start = 0; /* start index of current log write */
4009 struct reiserfs_journal_list *jl, *temp_jl;
4010 struct list_head *entry, *safe;
4011 unsigned long jindex;
4012 unsigned int commit_trans_id;
4016 BUG_ON(th->t_refcount > 1);
4017 BUG_ON(!th->t_trans_id);
4018 BUG_ON(!th->t_super);
4021 * protect flush_older_commits from doing mistakes if the
4022 * transaction ID counter gets overflowed.
4024 if (th->t_trans_id == ~0U)
4025 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
4026 flush = flags & FLUSH_ALL;
4027 wait_on_commit = flags & WAIT;
4029 current->journal_info = th->t_handle_save;
4030 reiserfs_check_lock_depth(sb, "journal end");
4031 if (journal->j_len == 0) {
4032 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
4034 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
4038 if (journal->j_next_full_flush) {
4042 if (journal->j_next_async_flush) {
4043 flags |= COMMIT_NOW | WAIT;
4048 * check_journal_end locks the journal, and unlocks if it does
4049 * not return 1 it tells us if we should continue with the
4050 * journal_end, or just return
4052 if (!check_journal_end(th, flags)) {
4053 reiserfs_schedule_old_flush(sb);
4054 wake_queued_writers(sb);
4055 reiserfs_async_progress_wait(sb);
4059 /* check_journal_end might set these, check again */
4060 if (journal->j_next_full_flush) {
4065 * j must wait means we have to flush the log blocks, and the
4066 * real blocks for this transaction
4068 if (journal->j_must_wait > 0) {
4071 #ifdef REISERFS_PREALLOCATE
4073 * quota ops might need to nest, setup the journal_info pointer
4074 * for them and raise the refcount so that it is > 0.
4076 current->journal_info = th;
4079 /* it should not involve new blocks into the transaction */
4080 reiserfs_discard_all_prealloc(th);
4083 current->journal_info = th->t_handle_save;
4086 /* setup description block */
4089 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4091 set_buffer_uptodate(d_bh);
4092 desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4093 memset(d_bh->b_data, 0, d_bh->b_size);
4094 memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4095 set_desc_trans_id(desc, journal->j_trans_id);
4098 * setup commit block. Don't write (keep it clean too) this one
4099 * until after everyone else is written
4101 c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4102 ((journal->j_start + journal->j_len +
4103 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4104 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4105 memset(c_bh->b_data, 0, c_bh->b_size);
4106 set_commit_trans_id(commit, journal->j_trans_id);
4107 set_buffer_uptodate(c_bh);
4109 /* init this journal list */
4110 jl = journal->j_current_jl;
4113 * we lock the commit before doing anything because
4114 * we want to make sure nobody tries to run flush_commit_list until
4115 * the new transaction is fully setup, and we've already flushed the
4118 reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4120 /* save the transaction id in case we need to commit it later */
4121 commit_trans_id = jl->j_trans_id;
4123 atomic_set(&jl->j_older_commits_done, 0);
4124 jl->j_trans_id = journal->j_trans_id;
4125 jl->j_timestamp = journal->j_trans_start_time;
4126 jl->j_commit_bh = c_bh;
4127 jl->j_start = journal->j_start;
4128 jl->j_len = journal->j_len;
4129 atomic_set(&jl->j_nonzerolen, journal->j_len);
4130 atomic_set(&jl->j_commit_left, journal->j_len + 2);
4131 jl->j_realblock = NULL;
4134 * The ENTIRE FOR LOOP MUST not cause schedule to occur.
4135 * for each real block, add it to the journal list hash,
4136 * copy into real block index array in the commit or desc block
4138 trans_half = journal_trans_half(sb->s_blocksize);
4139 for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4140 if (buffer_journaled(cn->bh)) {
4141 jl_cn = get_cnode(sb);
4143 reiserfs_panic(sb, "journal-1676",
4144 "get_cnode returned NULL");
4147 jl->j_realblock = jl_cn;
4149 jl_cn->prev = last_cn;
4152 last_cn->next = jl_cn;
4156 * make sure the block we are trying to log
4157 * is not a block of journal or reserved area
4159 if (is_block_in_log_or_reserved_area
4160 (sb, cn->bh->b_blocknr)) {
4161 reiserfs_panic(sb, "journal-2332",
4162 "Trying to log block %lu, "
4163 "which is a log block",
4166 jl_cn->blocknr = cn->bh->b_blocknr;
4171 insert_journal_hash(journal->j_list_hash_table, jl_cn);
4172 if (i < trans_half) {
4173 desc->j_realblock[i] =
4174 cpu_to_le32(cn->bh->b_blocknr);
4176 commit->j_realblock[i - trans_half] =
4177 cpu_to_le32(cn->bh->b_blocknr);
4183 set_desc_trans_len(desc, journal->j_len);
4184 set_desc_mount_id(desc, journal->j_mount_id);
4185 set_desc_trans_id(desc, journal->j_trans_id);
4186 set_commit_trans_len(commit, journal->j_len);
4189 * special check in case all buffers in the journal
4190 * were marked for not logging
4192 BUG_ON(journal->j_len == 0);
4195 * we're about to dirty all the log blocks, mark the description block
4196 * dirty now too. Don't mark the commit block dirty until all the
4197 * others are on disk
4199 mark_buffer_dirty(d_bh);
4202 * first data block is j_start + 1, so add one to
4203 * cur_write_start wherever you use it
4205 cur_write_start = journal->j_start;
4206 cn = journal->j_first;
4207 jindex = 1; /* start at one so we don't get the desc again */
4209 clear_buffer_journal_new(cn->bh);
4210 /* copy all the real blocks into log area. dirty log blocks */
4211 if (buffer_journaled(cn->bh)) {
4212 struct buffer_head *tmp_bh;
4217 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4220 SB_ONDISK_JOURNAL_SIZE(sb)));
4221 set_buffer_uptodate(tmp_bh);
4222 page = cn->bh->b_page;
4224 memcpy(tmp_bh->b_data,
4225 addr + offset_in_page(cn->bh->b_data),
4228 mark_buffer_dirty(tmp_bh);
4230 set_buffer_journal_dirty(cn->bh);
4231 clear_buffer_journaled(cn->bh);
4234 * JDirty cleared sometime during transaction.
4235 * don't log this one
4237 reiserfs_warning(sb, "journal-2048",
4238 "BAD, buffer in journal hash, "
4245 reiserfs_cond_resched(sb);
4249 * we are done with both the c_bh and d_bh, but
4250 * c_bh must be written after all other commit blocks,
4251 * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4254 journal->j_current_jl = alloc_journal_list(sb);
4256 /* now it is safe to insert this transaction on the main list */
4257 list_add_tail(&jl->j_list, &journal->j_journal_list);
4258 list_add_tail(&jl->j_working_list, &journal->j_working_list);
4259 journal->j_num_work_lists++;
4261 /* reset journal values for the next transaction */
4262 old_start = journal->j_start;
4264 (journal->j_start + journal->j_len +
4265 2) % SB_ONDISK_JOURNAL_SIZE(sb);
4266 atomic_set(&journal->j_wcount, 0);
4267 journal->j_bcount = 0;
4268 journal->j_last = NULL;
4269 journal->j_first = NULL;
4271 journal->j_trans_start_time = 0;
4272 /* check for trans_id overflow */
4273 if (++journal->j_trans_id == 0)
4274 journal->j_trans_id = 10;
4275 journal->j_current_jl->j_trans_id = journal->j_trans_id;
4276 journal->j_must_wait = 0;
4277 journal->j_len_alloc = 0;
4278 journal->j_next_full_flush = 0;
4279 journal->j_next_async_flush = 0;
4280 init_journal_hash(sb);
4283 * make sure reiserfs_add_jh sees the new current_jl before we
4284 * write out the tails
4289 * tail conversion targets have to hit the disk before we end the
4290 * transaction. Otherwise a later transaction might repack the tail
4291 * before this transaction commits, leaving the data block unflushed
4292 * and clean, if we crash before the later transaction commits, the
4293 * data block is lost.
4295 if (!list_empty(&jl->j_tail_bh_list)) {
4296 depth = reiserfs_write_unlock_nested(sb);
4297 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4298 journal, jl, &jl->j_tail_bh_list);
4299 reiserfs_write_lock_nested(sb, depth);
4301 BUG_ON(!list_empty(&jl->j_tail_bh_list));
4302 mutex_unlock(&jl->j_commit_mutex);
4305 * honor the flush wishes from the caller, simple commits can
4306 * be done outside the journal lock, they are done below
4308 * if we don't flush the commit list right now, we put it into
4309 * the work queue so the people waiting on the async progress work
4310 * queue don't wait for this proc to flush journal lists and such.
4313 flush_commit_list(sb, jl, 1);
4314 flush_journal_list(sb, jl, 1);
4315 } else if (!(jl->j_state & LIST_COMMIT_PENDING)) {
4317 * Avoid queueing work when sb is being shut down. Transaction
4318 * will be flushed on journal shutdown.
4320 if (sb->s_flags & MS_ACTIVE)
4321 queue_delayed_work(REISERFS_SB(sb)->commit_wq,
4322 &journal->j_work, HZ / 10);
4326 * if the next transaction has any chance of wrapping, flush
4327 * transactions that might get overwritten. If any journal lists
4328 * are very old flush them as well.
4331 list_for_each_safe(entry, safe, &journal->j_journal_list) {
4332 temp_jl = JOURNAL_LIST_ENTRY(entry);
4333 if (journal->j_start <= temp_jl->j_start) {
4334 if ((journal->j_start + journal->j_trans_max + 1) >=
4336 flush_used_journal_lists(sb, temp_jl);
4338 } else if ((journal->j_start +
4339 journal->j_trans_max + 1) <
4340 SB_ONDISK_JOURNAL_SIZE(sb)) {
4342 * if we don't cross into the next
4343 * transaction and we don't wrap, there is
4344 * no way we can overlap any later transactions
4349 } else if ((journal->j_start +
4350 journal->j_trans_max + 1) >
4351 SB_ONDISK_JOURNAL_SIZE(sb)) {
4352 if (((journal->j_start + journal->j_trans_max + 1) %
4353 SB_ONDISK_JOURNAL_SIZE(sb)) >=
4355 flush_used_journal_lists(sb, temp_jl);
4359 * we don't overlap anything from out start
4360 * to the end of the log, and our wrapped
4361 * portion doesn't overlap anything at
4362 * the start of the log. We can break
4369 journal->j_current_jl->j_list_bitmap =
4370 get_list_bitmap(sb, journal->j_current_jl);
4372 if (!(journal->j_current_jl->j_list_bitmap)) {
4373 reiserfs_panic(sb, "journal-1996",
4374 "could not get a list bitmap");
4377 atomic_set(&journal->j_jlock, 0);
4379 /* wake up any body waiting to join. */
4380 clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4381 wake_up(&journal->j_join_wait);
4383 if (!flush && wait_on_commit &&
4384 journal_list_still_alive(sb, commit_trans_id)) {
4385 flush_commit_list(sb, jl, 1);
4388 reiserfs_check_lock_depth(sb, "journal end2");
4390 memset(th, 0, sizeof(*th));
4392 * Re-set th->t_super, so we can properly keep track of how many
4393 * persistent transactions there are. We need to do this so if this
4394 * call is part of a failed restart_transaction, we can free it later
4398 return journal->j_errno;
4401 /* Send the file system read only and refuse new transactions */
4402 void reiserfs_abort_journal(struct super_block *sb, int errno)
4404 struct reiserfs_journal *journal = SB_JOURNAL(sb);
4405 if (test_bit(J_ABORTED, &journal->j_state))
4408 if (!journal->j_errno)
4409 journal->j_errno = errno;
4411 sb->s_flags |= MS_RDONLY;
4412 set_bit(J_ABORTED, &journal->j_state);
4414 #ifdef CONFIG_REISERFS_CHECK