2 ** Write ahead logging implementation copyright Chris Mason 2000
4 ** The background commits make this code very interelated, and
5 ** overly complex. I need to rethink things a bit....The major players:
7 ** journal_begin -- call with the number of blocks you expect to log.
8 ** If the current transaction is too
9 ** old, it will block until the current transaction is
10 ** finished, and then start a new one.
11 ** Usually, your transaction will get joined in with
12 ** previous ones for speed.
14 ** journal_join -- same as journal_begin, but won't block on the current
15 ** transaction regardless of age. Don't ever call
16 ** this. Ever. There are only two places it should be
17 ** called from, and they are both inside this file.
19 ** journal_mark_dirty -- adds blocks into this transaction. clears any flags
20 ** that might make them get sent to disk
21 ** and then marks them BH_JDirty. Puts the buffer head
22 ** into the current transaction hash.
24 ** journal_end -- if the current transaction is batchable, it does nothing
25 ** otherwise, it could do an async/synchronous commit, or
26 ** a full flush of all log and real blocks in the
29 ** flush_old_commits -- if the current transaction is too old, it is ended and
30 ** commit blocks are sent to disk. Forces commit blocks
31 ** to disk for all backgrounded commits that have been
33 ** -- Note, if you call this as an immediate flush from
34 ** from within kupdate, it will ignore the immediate flag
37 #include <asm/uaccess.h>
38 #include <asm/system.h>
40 #include <linux/time.h>
41 #include <asm/semaphore.h>
43 #include <linux/vmalloc.h>
44 #include <linux/reiserfs_fs.h>
46 #include <linux/kernel.h>
47 #include <linux/errno.h>
48 #include <linux/fcntl.h>
49 #include <linux/stat.h>
50 #include <linux/string.h>
51 #include <linux/smp_lock.h>
52 #include <linux/buffer_head.h>
53 #include <linux/workqueue.h>
54 #include <linux/writeback.h>
55 #include <linux/blkdev.h>
56 #include <linux/backing-dev.h>
58 /* gets a struct reiserfs_journal_list * from a list head */
59 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
61 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
64 /* the number of mounted filesystems. This is used to decide when to
65 ** start and kill the commit workqueue
67 static int reiserfs_mounted_fs_count;
69 static struct workqueue_struct *commit_wq;
71 #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
73 #define BUFNR 64 /*read ahead */
75 /* cnode stat bits. Move these into reiserfs_fs.h */
77 #define BLOCK_FREED 2 /* this block was freed, and can't be written. */
78 #define BLOCK_FREED_HOLDER 3 /* this block was freed during this transaction, and can't be written */
80 #define BLOCK_NEEDS_FLUSH 4 /* used in flush_journal_list */
81 #define BLOCK_DIRTIED 5
83 /* journal list state bits */
84 #define LIST_TOUCHED 1
86 #define LIST_COMMIT_PENDING 4 /* someone will commit this list */
88 /* flags for do_journal_end */
89 #define FLUSH_ALL 1 /* flush commit and real blocks */
90 #define COMMIT_NOW 2 /* end and commit this transaction */
91 #define WAIT 4 /* wait for the log blocks to hit the disk */
93 static int do_journal_end(struct reiserfs_transaction_handle *,
94 struct super_block *, unsigned long nblocks,
96 static int flush_journal_list(struct super_block *s,
97 struct reiserfs_journal_list *jl, int flushall);
98 static int flush_commit_list(struct super_block *s,
99 struct reiserfs_journal_list *jl, int flushall);
100 static int can_dirty(struct reiserfs_journal_cnode *cn);
101 static int journal_join(struct reiserfs_transaction_handle *th,
102 struct super_block *p_s_sb, unsigned long nblocks);
103 static int release_journal_dev(struct super_block *super,
104 struct reiserfs_journal *journal);
105 static int dirty_one_transaction(struct super_block *s,
106 struct reiserfs_journal_list *jl);
107 static void flush_async_commits(struct work_struct *work);
108 static void queue_log_writer(struct super_block *s);
110 /* values for join in do_journal_begin_r */
112 JBEGIN_REG = 0, /* regular journal begin */
113 JBEGIN_JOIN = 1, /* join the running transaction if at all possible */
114 JBEGIN_ABORT = 2, /* called from cleanup code, ignores aborted flag */
117 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
118 struct super_block *p_s_sb,
119 unsigned long nblocks, int join);
121 static void init_journal_hash(struct super_block *p_s_sb)
123 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
124 memset(journal->j_hash_table, 0,
125 JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
129 ** clears BH_Dirty and sticks the buffer on the clean list. Called because I can't allow refile_buffer to
130 ** make schedule happen after I've freed a block. Look at remove_from_transaction and journal_mark_freed for
133 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
136 clear_buffer_dirty(bh);
137 clear_buffer_journal_test(bh);
142 static void disable_barrier(struct super_block *s)
144 REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH);
145 printk("reiserfs: disabling flush barriers on %s\n",
146 reiserfs_bdevname(s));
149 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
152 struct reiserfs_bitmap_node *bn;
155 bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
159 bn->data = kzalloc(p_s_sb->s_blocksize, GFP_NOFS);
165 INIT_LIST_HEAD(&bn->list);
169 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *p_s_sb)
171 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
172 struct reiserfs_bitmap_node *bn = NULL;
173 struct list_head *entry = journal->j_bitmap_nodes.next;
175 journal->j_used_bitmap_nodes++;
178 if (entry != &journal->j_bitmap_nodes) {
179 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
181 memset(bn->data, 0, p_s_sb->s_blocksize);
182 journal->j_free_bitmap_nodes--;
185 bn = allocate_bitmap_node(p_s_sb);
192 static inline void free_bitmap_node(struct super_block *p_s_sb,
193 struct reiserfs_bitmap_node *bn)
195 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
196 journal->j_used_bitmap_nodes--;
197 if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
201 list_add(&bn->list, &journal->j_bitmap_nodes);
202 journal->j_free_bitmap_nodes++;
206 static void allocate_bitmap_nodes(struct super_block *p_s_sb)
209 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
210 struct reiserfs_bitmap_node *bn = NULL;
211 for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
212 bn = allocate_bitmap_node(p_s_sb);
214 list_add(&bn->list, &journal->j_bitmap_nodes);
215 journal->j_free_bitmap_nodes++;
217 break; // this is ok, we'll try again when more are needed
222 static int set_bit_in_list_bitmap(struct super_block *p_s_sb, int block,
223 struct reiserfs_list_bitmap *jb)
225 int bmap_nr = block / (p_s_sb->s_blocksize << 3);
226 int bit_nr = block % (p_s_sb->s_blocksize << 3);
228 if (!jb->bitmaps[bmap_nr]) {
229 jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb);
231 set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
235 static void cleanup_bitmap_list(struct super_block *p_s_sb,
236 struct reiserfs_list_bitmap *jb)
239 if (jb->bitmaps == NULL)
242 for (i = 0; i < SB_BMAP_NR(p_s_sb); i++) {
243 if (jb->bitmaps[i]) {
244 free_bitmap_node(p_s_sb, jb->bitmaps[i]);
245 jb->bitmaps[i] = NULL;
251 ** only call this on FS unmount.
253 static int free_list_bitmaps(struct super_block *p_s_sb,
254 struct reiserfs_list_bitmap *jb_array)
257 struct reiserfs_list_bitmap *jb;
258 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
260 jb->journal_list = NULL;
261 cleanup_bitmap_list(p_s_sb, jb);
268 static int free_bitmap_nodes(struct super_block *p_s_sb)
270 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
271 struct list_head *next = journal->j_bitmap_nodes.next;
272 struct reiserfs_bitmap_node *bn;
274 while (next != &journal->j_bitmap_nodes) {
275 bn = list_entry(next, struct reiserfs_bitmap_node, list);
279 next = journal->j_bitmap_nodes.next;
280 journal->j_free_bitmap_nodes--;
287 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
288 ** jb_array is the array to be filled in.
290 int reiserfs_allocate_list_bitmaps(struct super_block *p_s_sb,
291 struct reiserfs_list_bitmap *jb_array,
296 struct reiserfs_list_bitmap *jb;
297 int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
299 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
301 jb->journal_list = NULL;
302 jb->bitmaps = vmalloc(mem);
304 reiserfs_warning(p_s_sb,
305 "clm-2000, unable to allocate bitmaps for journal lists");
309 memset(jb->bitmaps, 0, mem);
312 free_list_bitmaps(p_s_sb, jb_array);
319 ** find an available list bitmap. If you can't find one, flush a commit list
322 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *p_s_sb,
323 struct reiserfs_journal_list
327 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
328 struct reiserfs_list_bitmap *jb = NULL;
330 for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
331 i = journal->j_list_bitmap_index;
332 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
333 jb = journal->j_list_bitmap + i;
334 if (journal->j_list_bitmap[i].journal_list) {
335 flush_commit_list(p_s_sb,
336 journal->j_list_bitmap[i].
338 if (!journal->j_list_bitmap[i].journal_list) {
345 if (jb->journal_list) { /* double check to make sure if flushed correctly */
348 jb->journal_list = jl;
353 ** allocates a new chunk of X nodes, and links them all together as a list.
354 ** Uses the cnode->next and cnode->prev pointers
355 ** returns NULL on failure
357 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
359 struct reiserfs_journal_cnode *head;
361 if (num_cnodes <= 0) {
364 head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
368 memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode));
370 head[0].next = head + 1;
371 for (i = 1; i < num_cnodes; i++) {
372 head[i].prev = head + (i - 1);
373 head[i].next = head + (i + 1); /* if last one, overwrite it after the if */
375 head[num_cnodes - 1].next = NULL;
380 ** pulls a cnode off the free list, or returns NULL on failure
382 static struct reiserfs_journal_cnode *get_cnode(struct super_block *p_s_sb)
384 struct reiserfs_journal_cnode *cn;
385 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
387 reiserfs_check_lock_depth(p_s_sb, "get_cnode");
389 if (journal->j_cnode_free <= 0) {
392 journal->j_cnode_used++;
393 journal->j_cnode_free--;
394 cn = journal->j_cnode_free_list;
399 cn->next->prev = NULL;
401 journal->j_cnode_free_list = cn->next;
402 memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
407 ** returns a cnode to the free list
409 static void free_cnode(struct super_block *p_s_sb,
410 struct reiserfs_journal_cnode *cn)
412 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
414 reiserfs_check_lock_depth(p_s_sb, "free_cnode");
416 journal->j_cnode_used--;
417 journal->j_cnode_free++;
418 /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
419 cn->next = journal->j_cnode_free_list;
420 if (journal->j_cnode_free_list) {
421 journal->j_cnode_free_list->prev = cn;
423 cn->prev = NULL; /* not needed with the memset, but I might kill the memset, and forget to do this */
424 journal->j_cnode_free_list = cn;
427 static void clear_prepared_bits(struct buffer_head *bh)
429 clear_buffer_journal_prepared(bh);
430 clear_buffer_journal_restore_dirty(bh);
433 /* utility function to force a BUG if it is called without the big
434 ** kernel lock held. caller is the string printed just before calling BUG()
436 void reiserfs_check_lock_depth(struct super_block *sb, char *caller)
439 if (current->lock_depth < 0) {
440 reiserfs_panic(sb, "%s called without kernel lock held",
448 /* return a cnode with same dev, block number and size in table, or null if not found */
449 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
453 reiserfs_journal_cnode
457 struct reiserfs_journal_cnode *cn;
458 cn = journal_hash(table, sb, bl);
460 if (cn->blocknr == bl && cn->sb == sb)
464 return (struct reiserfs_journal_cnode *)0;
468 ** this actually means 'can this block be reallocated yet?'. If you set search_all, a block can only be allocated
469 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
470 ** being overwritten by a replay after crashing.
472 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction. Since deleting
473 ** a block removes it from the current transaction, this case should never happen. If you don't set search_all, make
474 ** sure you never write the block without logging it.
476 ** next_zero_bit is a suggestion about the next block to try for find_forward.
477 ** when bl is rejected because it is set in a journal list bitmap, we search
478 ** for the next zero bit in the bitmap that rejected bl. Then, we return that
479 ** through next_zero_bit for find_forward to try.
481 ** Just because we return something in next_zero_bit does not mean we won't
482 ** reject it on the next call to reiserfs_in_journal
485 int reiserfs_in_journal(struct super_block *p_s_sb,
486 int bmap_nr, int bit_nr, int search_all,
487 b_blocknr_t * next_zero_bit)
489 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
490 struct reiserfs_journal_cnode *cn;
491 struct reiserfs_list_bitmap *jb;
495 *next_zero_bit = 0; /* always start this at zero. */
497 PROC_INFO_INC(p_s_sb, journal.in_journal);
498 /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
499 ** if we crash before the transaction that freed it commits, this transaction won't
500 ** have committed either, and the block will never be written
503 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
504 PROC_INFO_INC(p_s_sb, journal.in_journal_bitmap);
505 jb = journal->j_list_bitmap + i;
506 if (jb->journal_list && jb->bitmaps[bmap_nr] &&
508 (unsigned long *)jb->bitmaps[bmap_nr]->
511 find_next_zero_bit((unsigned long *)
512 (jb->bitmaps[bmap_nr]->
514 p_s_sb->s_blocksize << 3,
521 bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr;
522 /* is it in any old transactions? */
525 get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, bl))) {
529 /* is it in the current transaction. This should never happen */
530 if ((cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, bl))) {
535 PROC_INFO_INC(p_s_sb, journal.in_journal_reusable);
540 /* insert cn into table
542 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
543 struct reiserfs_journal_cnode *cn)
545 struct reiserfs_journal_cnode *cn_orig;
547 cn_orig = journal_hash(table, cn->sb, cn->blocknr);
553 journal_hash(table, cn->sb, cn->blocknr) = cn;
556 /* lock the current transaction */
557 static inline void lock_journal(struct super_block *p_s_sb)
559 PROC_INFO_INC(p_s_sb, journal.lock_journal);
560 down(&SB_JOURNAL(p_s_sb)->j_lock);
563 /* unlock the current transaction */
564 static inline void unlock_journal(struct super_block *p_s_sb)
566 up(&SB_JOURNAL(p_s_sb)->j_lock);
569 static inline void get_journal_list(struct reiserfs_journal_list *jl)
574 static inline void put_journal_list(struct super_block *s,
575 struct reiserfs_journal_list *jl)
577 if (jl->j_refcount < 1) {
578 reiserfs_panic(s, "trans id %lu, refcount at %d",
579 jl->j_trans_id, jl->j_refcount);
581 if (--jl->j_refcount == 0)
586 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
587 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
590 static void cleanup_freed_for_journal_list(struct super_block *p_s_sb,
591 struct reiserfs_journal_list *jl)
594 struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
596 cleanup_bitmap_list(p_s_sb, jb);
598 jl->j_list_bitmap->journal_list = NULL;
599 jl->j_list_bitmap = NULL;
602 static int journal_list_still_alive(struct super_block *s,
603 unsigned long trans_id)
605 struct reiserfs_journal *journal = SB_JOURNAL(s);
606 struct list_head *entry = &journal->j_journal_list;
607 struct reiserfs_journal_list *jl;
609 if (!list_empty(entry)) {
610 jl = JOURNAL_LIST_ENTRY(entry->next);
611 if (jl->j_trans_id <= trans_id) {
618 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
620 char b[BDEVNAME_SIZE];
622 if (buffer_journaled(bh)) {
623 reiserfs_warning(NULL,
624 "clm-2084: pinned buffer %lu:%s sent to disk",
625 bh->b_blocknr, bdevname(bh->b_bdev, b));
628 set_buffer_uptodate(bh);
630 clear_buffer_uptodate(bh);
635 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
638 set_buffer_uptodate(bh);
640 clear_buffer_uptodate(bh);
645 static void submit_logged_buffer(struct buffer_head *bh)
648 bh->b_end_io = reiserfs_end_buffer_io_sync;
649 clear_buffer_journal_new(bh);
650 clear_buffer_dirty(bh);
651 if (!test_clear_buffer_journal_test(bh))
653 if (!buffer_uptodate(bh))
655 submit_bh(WRITE, bh);
658 static void submit_ordered_buffer(struct buffer_head *bh)
661 bh->b_end_io = reiserfs_end_ordered_io;
662 clear_buffer_dirty(bh);
663 if (!buffer_uptodate(bh))
665 submit_bh(WRITE, bh);
668 static int submit_barrier_buffer(struct buffer_head *bh)
671 bh->b_end_io = reiserfs_end_ordered_io;
672 clear_buffer_dirty(bh);
673 if (!buffer_uptodate(bh))
675 return submit_bh(WRITE_BARRIER, bh);
678 static void check_barrier_completion(struct super_block *s,
679 struct buffer_head *bh)
681 if (buffer_eopnotsupp(bh)) {
682 clear_buffer_eopnotsupp(bh);
684 set_buffer_uptodate(bh);
685 set_buffer_dirty(bh);
686 sync_dirty_buffer(bh);
690 #define CHUNK_SIZE 32
691 struct buffer_chunk {
692 struct buffer_head *bh[CHUNK_SIZE];
696 static void write_chunk(struct buffer_chunk *chunk)
700 for (i = 0; i < chunk->nr; i++) {
701 submit_logged_buffer(chunk->bh[i]);
707 static void write_ordered_chunk(struct buffer_chunk *chunk)
711 for (i = 0; i < chunk->nr; i++) {
712 submit_ordered_buffer(chunk->bh[i]);
718 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
719 spinlock_t * lock, void (fn) (struct buffer_chunk *))
722 BUG_ON(chunk->nr >= CHUNK_SIZE);
723 chunk->bh[chunk->nr++] = bh;
724 if (chunk->nr >= CHUNK_SIZE) {
735 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
736 static struct reiserfs_jh *alloc_jh(void)
738 struct reiserfs_jh *jh;
740 jh = kmalloc(sizeof(*jh), GFP_NOFS);
742 atomic_inc(&nr_reiserfs_jh);
750 * we want to free the jh when the buffer has been written
753 void reiserfs_free_jh(struct buffer_head *bh)
755 struct reiserfs_jh *jh;
759 bh->b_private = NULL;
761 list_del_init(&jh->list);
763 if (atomic_read(&nr_reiserfs_jh) <= 0)
765 atomic_dec(&nr_reiserfs_jh);
770 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
773 struct reiserfs_jh *jh;
776 spin_lock(&j->j_dirty_buffers_lock);
777 if (!bh->b_private) {
778 spin_unlock(&j->j_dirty_buffers_lock);
782 list_del_init(&jh->list);
787 spin_lock(&j->j_dirty_buffers_lock);
788 /* buffer must be locked for __add_jh, should be able to have
789 * two adds at the same time
791 BUG_ON(bh->b_private);
795 jh->jl = j->j_current_jl;
797 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
799 list_add_tail(&jh->list, &jh->jl->j_bh_list);
801 spin_unlock(&j->j_dirty_buffers_lock);
805 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
807 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
809 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
811 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
814 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
815 static int write_ordered_buffers(spinlock_t * lock,
816 struct reiserfs_journal *j,
817 struct reiserfs_journal_list *jl,
818 struct list_head *list)
820 struct buffer_head *bh;
821 struct reiserfs_jh *jh;
822 int ret = j->j_errno;
823 struct buffer_chunk chunk;
824 struct list_head tmp;
825 INIT_LIST_HEAD(&tmp);
829 while (!list_empty(list)) {
830 jh = JH_ENTRY(list->next);
833 if (test_set_buffer_locked(bh)) {
834 if (!buffer_dirty(bh)) {
835 list_move(&jh->list, &tmp);
840 write_ordered_chunk(&chunk);
846 /* in theory, dirty non-uptodate buffers should never get here,
847 * but the upper layer io error paths still have a few quirks.
848 * Handle them here as gracefully as we can
850 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
851 clear_buffer_dirty(bh);
854 if (buffer_dirty(bh)) {
855 list_move(&jh->list, &tmp);
856 add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
858 reiserfs_free_jh(bh);
863 cond_resched_lock(lock);
867 write_ordered_chunk(&chunk);
870 while (!list_empty(&tmp)) {
871 jh = JH_ENTRY(tmp.prev);
874 reiserfs_free_jh(bh);
876 if (buffer_locked(bh)) {
881 if (!buffer_uptodate(bh)) {
884 /* ugly interaction with invalidatepage here.
885 * reiserfs_invalidate_page will pin any buffer that has a valid
886 * journal head from an older transaction. If someone else sets
887 * our buffer dirty after we write it in the first loop, and
888 * then someone truncates the page away, nobody will ever write
889 * the buffer. We're safe if we write the page one last time
890 * after freeing the journal header.
892 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
894 ll_rw_block(WRITE, 1, &bh);
898 cond_resched_lock(lock);
904 static int flush_older_commits(struct super_block *s,
905 struct reiserfs_journal_list *jl)
907 struct reiserfs_journal *journal = SB_JOURNAL(s);
908 struct reiserfs_journal_list *other_jl;
909 struct reiserfs_journal_list *first_jl;
910 struct list_head *entry;
911 unsigned long trans_id = jl->j_trans_id;
912 unsigned long other_trans_id;
913 unsigned long first_trans_id;
917 * first we walk backwards to find the oldest uncommitted transation
920 entry = jl->j_list.prev;
922 other_jl = JOURNAL_LIST_ENTRY(entry);
923 if (entry == &journal->j_journal_list ||
924 atomic_read(&other_jl->j_older_commits_done))
928 entry = other_jl->j_list.prev;
931 /* if we didn't find any older uncommitted transactions, return now */
932 if (first_jl == jl) {
936 first_trans_id = first_jl->j_trans_id;
938 entry = &first_jl->j_list;
940 other_jl = JOURNAL_LIST_ENTRY(entry);
941 other_trans_id = other_jl->j_trans_id;
943 if (other_trans_id < trans_id) {
944 if (atomic_read(&other_jl->j_commit_left) != 0) {
945 flush_commit_list(s, other_jl, 0);
947 /* list we were called with is gone, return */
948 if (!journal_list_still_alive(s, trans_id))
951 /* the one we just flushed is gone, this means all
952 * older lists are also gone, so first_jl is no longer
953 * valid either. Go back to the beginning.
955 if (!journal_list_still_alive
956 (s, other_trans_id)) {
961 if (entry == &journal->j_journal_list)
970 static int reiserfs_async_progress_wait(struct super_block *s)
973 struct reiserfs_journal *j = SB_JOURNAL(s);
974 if (atomic_read(&j->j_async_throttle))
975 congestion_wait(WRITE, HZ / 10);
980 ** if this journal list still has commit blocks unflushed, send them to disk.
982 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
983 ** Before the commit block can by written, every other log block must be safely on disk
986 static int flush_commit_list(struct super_block *s,
987 struct reiserfs_journal_list *jl, int flushall)
991 struct buffer_head *tbh = NULL;
992 unsigned long trans_id = jl->j_trans_id;
993 struct reiserfs_journal *journal = SB_JOURNAL(s);
998 reiserfs_check_lock_depth(s, "flush_commit_list");
1000 if (atomic_read(&jl->j_older_commits_done)) {
1006 /* before we can put our commit blocks on disk, we have to make sure everyone older than
1007 ** us is on disk too
1009 BUG_ON(jl->j_len <= 0);
1010 BUG_ON(trans_id == journal->j_trans_id);
1012 get_journal_list(jl);
1014 if (flush_older_commits(s, jl) == 1) {
1015 /* list disappeared during flush_older_commits. return */
1020 /* make sure nobody is trying to flush this one at the same time */
1021 down(&jl->j_commit_lock);
1022 if (!journal_list_still_alive(s, trans_id)) {
1023 up(&jl->j_commit_lock);
1026 BUG_ON(jl->j_trans_id == 0);
1028 /* this commit is done, exit */
1029 if (atomic_read(&(jl->j_commit_left)) <= 0) {
1031 atomic_set(&(jl->j_older_commits_done), 1);
1033 up(&jl->j_commit_lock);
1037 if (!list_empty(&jl->j_bh_list)) {
1040 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1041 journal, jl, &jl->j_bh_list);
1042 if (ret < 0 && retval == 0)
1046 BUG_ON(!list_empty(&jl->j_bh_list));
1048 * for the description block and all the log blocks, submit any buffers
1049 * that haven't already reached the disk. Try to write at least 256
1050 * log blocks. later on, we will only wait on blocks that correspond
1051 * to this transaction, but while we're unplugging we might as well
1052 * get a chunk of data on there.
1054 atomic_inc(&journal->j_async_throttle);
1055 write_len = jl->j_len + 1;
1056 if (write_len < 256)
1058 for (i = 0 ; i < write_len ; i++) {
1059 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1060 SB_ONDISK_JOURNAL_SIZE(s);
1061 tbh = journal_find_get_block(s, bn);
1063 if (buffer_dirty(tbh))
1064 ll_rw_block(WRITE, 1, &tbh) ;
1068 atomic_dec(&journal->j_async_throttle);
1070 /* We're skipping the commit if there's an error */
1071 if (retval || reiserfs_is_journal_aborted(journal))
1074 /* wait on everything written so far before writing the commit
1075 * if we are in barrier mode, send the commit down now
1077 barrier = reiserfs_barrier_flush(s);
1080 lock_buffer(jl->j_commit_bh);
1081 ret = submit_barrier_buffer(jl->j_commit_bh);
1082 if (ret == -EOPNOTSUPP) {
1083 set_buffer_uptodate(jl->j_commit_bh);
1088 for (i = 0; i < (jl->j_len + 1); i++) {
1089 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1090 (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1091 tbh = journal_find_get_block(s, bn);
1092 wait_on_buffer(tbh);
1093 // since we're using ll_rw_blk above, it might have skipped over
1094 // a locked buffer. Double check here
1096 if (buffer_dirty(tbh)) /* redundant, sync_dirty_buffer() checks */
1097 sync_dirty_buffer(tbh);
1098 if (unlikely(!buffer_uptodate(tbh))) {
1099 #ifdef CONFIG_REISERFS_CHECK
1100 reiserfs_warning(s, "journal-601, buffer write failed");
1104 put_bh(tbh); /* once for journal_find_get_block */
1105 put_bh(tbh); /* once due to original getblk in do_journal_end */
1106 atomic_dec(&(jl->j_commit_left));
1109 BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1112 /* If there was a write error in the journal - we can't commit
1113 * this transaction - it will be invalid and, if successful,
1114 * will just end up propagating the write error out to
1115 * the file system. */
1116 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1117 if (buffer_dirty(jl->j_commit_bh))
1119 mark_buffer_dirty(jl->j_commit_bh) ;
1120 sync_dirty_buffer(jl->j_commit_bh) ;
1123 wait_on_buffer(jl->j_commit_bh);
1125 check_barrier_completion(s, jl->j_commit_bh);
1127 /* If there was a write error in the journal - we can't commit this
1128 * transaction - it will be invalid and, if successful, will just end
1129 * up propagating the write error out to the filesystem. */
1130 if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1131 #ifdef CONFIG_REISERFS_CHECK
1132 reiserfs_warning(s, "journal-615: buffer write failed");
1136 bforget(jl->j_commit_bh);
1137 if (journal->j_last_commit_id != 0 &&
1138 (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1139 reiserfs_warning(s, "clm-2200: last commit %lu, current %lu",
1140 journal->j_last_commit_id, jl->j_trans_id);
1142 journal->j_last_commit_id = jl->j_trans_id;
1144 /* now, every commit block is on the disk. It is safe to allow 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 up(&jl->j_commit_lock);
1159 put_journal_list(s, jl);
1162 reiserfs_abort(s, retval, "Journal write error in %s",
1169 ** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or
1170 ** returns NULL if it can't find anything
1172 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1173 reiserfs_journal_cnode
1176 struct super_block *sb = cn->sb;
1177 b_blocknr_t blocknr = cn->blocknr;
1181 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1189 static int newer_jl_done(struct reiserfs_journal_cnode *cn)
1191 struct super_block *sb = cn->sb;
1192 b_blocknr_t blocknr = cn->blocknr;
1196 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist &&
1197 atomic_read(&cn->jlist->j_commit_left) != 0)
1204 static void remove_journal_hash(struct super_block *,
1205 struct reiserfs_journal_cnode **,
1206 struct reiserfs_journal_list *, unsigned long,
1210 ** once all the real blocks have been flushed, it is safe to remove them from the
1211 ** journal list for this transaction. Aside from freeing the cnode, this also allows the
1212 ** block to be reallocated for data blocks if it had been deleted.
1214 static void remove_all_from_journal_list(struct super_block *p_s_sb,
1215 struct reiserfs_journal_list *jl,
1218 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1219 struct reiserfs_journal_cnode *cn, *last;
1220 cn = jl->j_realblock;
1222 /* which is better, to lock once around the whole loop, or
1223 ** to lock for each call to remove_journal_hash?
1226 if (cn->blocknr != 0) {
1228 reiserfs_warning(p_s_sb,
1229 "block %u, bh is %d, state %ld",
1230 cn->blocknr, cn->bh ? 1 : 0,
1234 remove_journal_hash(p_s_sb, journal->j_list_hash_table,
1235 jl, cn->blocknr, 1);
1239 free_cnode(p_s_sb, last);
1241 jl->j_realblock = NULL;
1245 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1246 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1247 ** releasing blocks in this transaction for reuse as data blocks.
1248 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1251 static int _update_journal_header_block(struct super_block *p_s_sb,
1252 unsigned long offset,
1253 unsigned long trans_id)
1255 struct reiserfs_journal_header *jh;
1256 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1258 if (reiserfs_is_journal_aborted(journal))
1261 if (trans_id >= journal->j_last_flush_trans_id) {
1262 if (buffer_locked((journal->j_header_bh))) {
1263 wait_on_buffer((journal->j_header_bh));
1264 if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1265 #ifdef CONFIG_REISERFS_CHECK
1266 reiserfs_warning(p_s_sb,
1267 "journal-699: buffer write failed");
1272 journal->j_last_flush_trans_id = trans_id;
1273 journal->j_first_unflushed_offset = offset;
1274 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1276 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1277 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1278 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1280 if (reiserfs_barrier_flush(p_s_sb)) {
1282 lock_buffer(journal->j_header_bh);
1283 ret = submit_barrier_buffer(journal->j_header_bh);
1284 if (ret == -EOPNOTSUPP) {
1285 set_buffer_uptodate(journal->j_header_bh);
1286 disable_barrier(p_s_sb);
1289 wait_on_buffer(journal->j_header_bh);
1290 check_barrier_completion(p_s_sb, journal->j_header_bh);
1293 set_buffer_dirty(journal->j_header_bh);
1294 sync_dirty_buffer(journal->j_header_bh);
1296 if (!buffer_uptodate(journal->j_header_bh)) {
1297 reiserfs_warning(p_s_sb,
1298 "journal-837: IO error during journal replay");
1305 static int update_journal_header_block(struct super_block *p_s_sb,
1306 unsigned long offset,
1307 unsigned long trans_id)
1309 return _update_journal_header_block(p_s_sb, offset, trans_id);
1313 ** flush any and all journal lists older than you are
1314 ** can only be called from flush_journal_list
1316 static int flush_older_journal_lists(struct super_block *p_s_sb,
1317 struct reiserfs_journal_list *jl)
1319 struct list_head *entry;
1320 struct reiserfs_journal_list *other_jl;
1321 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1322 unsigned long trans_id = jl->j_trans_id;
1324 /* we know we are the only ones flushing things, no extra race
1325 * protection is required.
1328 entry = journal->j_journal_list.next;
1330 if (entry == &journal->j_journal_list)
1332 other_jl = JOURNAL_LIST_ENTRY(entry);
1333 if (other_jl->j_trans_id < trans_id) {
1334 BUG_ON(other_jl->j_refcount <= 0);
1335 /* do not flush all */
1336 flush_journal_list(p_s_sb, other_jl, 0);
1338 /* other_jl is now deleted from the list */
1344 static void del_from_work_list(struct super_block *s,
1345 struct reiserfs_journal_list *jl)
1347 struct reiserfs_journal *journal = SB_JOURNAL(s);
1348 if (!list_empty(&jl->j_working_list)) {
1349 list_del_init(&jl->j_working_list);
1350 journal->j_num_work_lists--;
1354 /* flush a journal list, both commit and real blocks
1356 ** always set flushall to 1, unless you are calling from inside
1357 ** flush_journal_list
1359 ** IMPORTANT. This can only be called while there are no journal writers,
1360 ** and the journal is locked. That means it can only be called from
1361 ** do_journal_end, or by journal_release
1363 static int flush_journal_list(struct super_block *s,
1364 struct reiserfs_journal_list *jl, int flushall)
1366 struct reiserfs_journal_list *pjl;
1367 struct reiserfs_journal_cnode *cn, *last;
1371 struct buffer_head *saved_bh;
1372 unsigned long j_len_saved = jl->j_len;
1373 struct reiserfs_journal *journal = SB_JOURNAL(s);
1376 BUG_ON(j_len_saved <= 0);
1378 if (atomic_read(&journal->j_wcount) != 0) {
1380 "clm-2048: flush_journal_list called with wcount %d",
1381 atomic_read(&journal->j_wcount));
1383 BUG_ON(jl->j_trans_id == 0);
1385 /* if flushall == 0, the lock is already held */
1387 down(&journal->j_flush_sem);
1388 } else if (!down_trylock(&journal->j_flush_sem)) {
1393 if (j_len_saved > journal->j_trans_max) {
1395 "journal-715: flush_journal_list, length is %lu, trans id %lu\n",
1396 j_len_saved, jl->j_trans_id);
1402 /* if all the work is already done, get out of here */
1403 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1404 atomic_read(&(jl->j_commit_left)) <= 0) {
1405 goto flush_older_and_return;
1408 /* start by putting the commit list on disk. This will also flush
1409 ** the commit lists of any olders transactions
1411 flush_commit_list(s, jl, 1);
1413 if (!(jl->j_state & LIST_DIRTY)
1414 && !reiserfs_is_journal_aborted(journal))
1417 /* are we done now? */
1418 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1419 atomic_read(&(jl->j_commit_left)) <= 0) {
1420 goto flush_older_and_return;
1423 /* loop through each cnode, see if we need to write it,
1424 ** or wait on a more recent transaction, or just ignore it
1426 if (atomic_read(&(journal->j_wcount)) != 0) {
1428 "journal-844: panic journal list is flushing, wcount is not 0\n");
1430 cn = jl->j_realblock;
1435 /* blocknr of 0 is no longer in the hash, ignore it */
1436 if (cn->blocknr == 0) {
1440 /* This transaction failed commit. Don't write out to the disk */
1441 if (!(jl->j_state & LIST_DIRTY))
1444 pjl = find_newer_jl_for_cn(cn);
1445 /* the order is important here. We check pjl to make sure we
1446 ** don't clear BH_JDirty_wait if we aren't the one writing this
1449 if (!pjl && cn->bh) {
1452 /* we do this to make sure nobody releases the buffer while
1453 ** we are working with it
1457 if (buffer_journal_dirty(saved_bh)) {
1458 BUG_ON(!can_dirty(cn));
1461 } else if (can_dirty(cn)) {
1462 /* everything with !pjl && jwait should be writable */
1467 /* if someone has this block in a newer transaction, just make
1468 ** sure they are committed, and don't try writing it to disk
1471 if (atomic_read(&pjl->j_commit_left))
1472 flush_commit_list(s, pjl, 1);
1476 /* bh == NULL when the block got to disk on its own, OR,
1477 ** the block got freed in a future transaction
1479 if (saved_bh == NULL) {
1483 /* this should never happen. kupdate_one_transaction has this list
1484 ** locked while it works, so we should never see a buffer here that
1485 ** is not marked JDirty_wait
1487 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1489 "journal-813: BAD! buffer %llu %cdirty %cjwait, "
1490 "not in a newer tranasction",
1491 (unsigned long long)saved_bh->
1492 b_blocknr, was_dirty ? ' ' : '!',
1493 was_jwait ? ' ' : '!');
1496 /* we inc again because saved_bh gets decremented at free_cnode */
1498 set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1499 lock_buffer(saved_bh);
1500 BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1501 if (buffer_dirty(saved_bh))
1502 submit_logged_buffer(saved_bh);
1504 unlock_buffer(saved_bh);
1508 "clm-2082: Unable to flush buffer %llu in %s",
1509 (unsigned long long)saved_bh->
1510 b_blocknr, __FUNCTION__);
1516 /* we incremented this to keep others from taking the buffer head away */
1518 if (atomic_read(&(saved_bh->b_count)) < 0) {
1520 "journal-945: saved_bh->b_count < 0");
1525 cn = jl->j_realblock;
1527 if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1530 "journal-1011: cn->bh is NULL\n");
1532 wait_on_buffer(cn->bh);
1535 "journal-1012: cn->bh is NULL\n");
1537 if (unlikely(!buffer_uptodate(cn->bh))) {
1538 #ifdef CONFIG_REISERFS_CHECK
1540 "journal-949: buffer write failed\n");
1544 /* note, we must clear the JDirty_wait bit after the up to date
1545 ** check, otherwise we race against our flushpage routine
1547 BUG_ON(!test_clear_buffer_journal_dirty
1550 /* undo the inc from journal_mark_dirty */
1559 reiserfs_abort(s, -EIO,
1560 "Write error while pushing transaction to disk in %s",
1562 flush_older_and_return:
1564 /* before we can update the journal header block, we _must_ flush all
1565 ** real blocks from all older transactions to disk. This is because
1566 ** once the header block is updated, this transaction will not be
1567 ** replayed after a crash
1570 flush_older_journal_lists(s, jl);
1573 err = journal->j_errno;
1574 /* before we can remove everything from the hash tables for this
1575 ** transaction, we must make sure it can never be replayed
1577 ** since we are only called from do_journal_end, we know for sure there
1578 ** are no allocations going on while we are flushing journal lists. So,
1579 ** we only need to update the journal header block for the last list
1582 if (!err && flushall) {
1584 update_journal_header_block(s,
1585 (jl->j_start + jl->j_len +
1586 2) % SB_ONDISK_JOURNAL_SIZE(s),
1589 reiserfs_abort(s, -EIO,
1590 "Write error while updating journal header in %s",
1593 remove_all_from_journal_list(s, jl, 0);
1594 list_del_init(&jl->j_list);
1595 journal->j_num_lists--;
1596 del_from_work_list(s, jl);
1598 if (journal->j_last_flush_id != 0 &&
1599 (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1600 reiserfs_warning(s, "clm-2201: last flush %lu, current %lu",
1601 journal->j_last_flush_id, jl->j_trans_id);
1603 journal->j_last_flush_id = jl->j_trans_id;
1605 /* not strictly required since we are freeing the list, but it should
1606 * help find code using dead lists later on
1609 atomic_set(&(jl->j_nonzerolen), 0);
1611 jl->j_realblock = NULL;
1612 jl->j_commit_bh = NULL;
1615 put_journal_list(s, jl);
1617 up(&journal->j_flush_sem);
1622 static int test_transaction(struct super_block *s,
1623 struct reiserfs_journal_list *jl)
1625 struct reiserfs_journal_cnode *cn;
1627 if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0)
1630 cn = jl->j_realblock;
1632 /* if the blocknr == 0, this has been cleared from the hash,
1635 if (cn->blocknr == 0) {
1638 if (cn->bh && !newer_jl_done(cn))
1647 static int write_one_transaction(struct super_block *s,
1648 struct reiserfs_journal_list *jl,
1649 struct buffer_chunk *chunk)
1651 struct reiserfs_journal_cnode *cn;
1654 jl->j_state |= LIST_TOUCHED;
1655 del_from_work_list(s, jl);
1656 if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1660 cn = jl->j_realblock;
1662 /* if the blocknr == 0, this has been cleared from the hash,
1665 if (cn->blocknr == 0) {
1668 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1669 struct buffer_head *tmp_bh;
1670 /* we can race against journal_mark_freed when we try
1671 * to lock_buffer(cn->bh), so we have to inc the buffer
1672 * count, and recheck things after locking
1676 lock_buffer(tmp_bh);
1677 if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1678 if (!buffer_journal_dirty(tmp_bh) ||
1679 buffer_journal_prepared(tmp_bh))
1681 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1684 /* note, cn->bh might be null now */
1685 unlock_buffer(tmp_bh);
1696 /* used by flush_commit_list */
1697 static int dirty_one_transaction(struct super_block *s,
1698 struct reiserfs_journal_list *jl)
1700 struct reiserfs_journal_cnode *cn;
1701 struct reiserfs_journal_list *pjl;
1704 jl->j_state |= LIST_DIRTY;
1705 cn = jl->j_realblock;
1707 /* look for a more recent transaction that logged this
1708 ** buffer. Only the most recent transaction with a buffer in
1709 ** it is allowed to send that buffer to disk
1711 pjl = find_newer_jl_for_cn(cn);
1712 if (!pjl && cn->blocknr && cn->bh
1713 && buffer_journal_dirty(cn->bh)) {
1714 BUG_ON(!can_dirty(cn));
1715 /* if the buffer is prepared, it will either be logged
1716 * or restored. If restored, we need to make sure
1717 * it actually gets marked dirty
1719 clear_buffer_journal_new(cn->bh);
1720 if (buffer_journal_prepared(cn->bh)) {
1721 set_buffer_journal_restore_dirty(cn->bh);
1723 set_buffer_journal_test(cn->bh);
1724 mark_buffer_dirty(cn->bh);
1732 static int kupdate_transactions(struct super_block *s,
1733 struct reiserfs_journal_list *jl,
1734 struct reiserfs_journal_list **next_jl,
1735 unsigned long *next_trans_id,
1736 int num_blocks, int num_trans)
1740 int transactions_flushed = 0;
1741 unsigned long orig_trans_id = jl->j_trans_id;
1742 struct buffer_chunk chunk;
1743 struct list_head *entry;
1744 struct reiserfs_journal *journal = SB_JOURNAL(s);
1747 down(&journal->j_flush_sem);
1748 if (!journal_list_still_alive(s, orig_trans_id)) {
1752 /* we've got j_flush_sem held, nobody is going to delete any
1753 * of these lists out from underneath us
1755 while ((num_trans && transactions_flushed < num_trans) ||
1756 (!num_trans && written < num_blocks)) {
1758 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1759 atomic_read(&jl->j_commit_left)
1760 || !(jl->j_state & LIST_DIRTY)) {
1761 del_from_work_list(s, jl);
1764 ret = write_one_transaction(s, jl, &chunk);
1768 transactions_flushed++;
1770 entry = jl->j_list.next;
1773 if (entry == &journal->j_journal_list) {
1776 jl = JOURNAL_LIST_ENTRY(entry);
1778 /* don't bother with older transactions */
1779 if (jl->j_trans_id <= orig_trans_id)
1783 write_chunk(&chunk);
1787 up(&journal->j_flush_sem);
1791 /* for o_sync and fsync heavy applications, they tend to use
1792 ** all the journa list slots with tiny transactions. These
1793 ** trigger lots and lots of calls to update the header block, which
1794 ** adds seeks and slows things down.
1796 ** This function tries to clear out a large chunk of the journal lists
1797 ** at once, which makes everything faster since only the newest journal
1798 ** list updates the header block
1800 static int flush_used_journal_lists(struct super_block *s,
1801 struct reiserfs_journal_list *jl)
1803 unsigned long len = 0;
1804 unsigned long cur_len;
1808 struct reiserfs_journal_list *tjl;
1809 struct reiserfs_journal_list *flush_jl;
1810 unsigned long trans_id;
1811 struct reiserfs_journal *journal = SB_JOURNAL(s);
1813 flush_jl = tjl = jl;
1815 /* in data logging mode, try harder to flush a lot of blocks */
1816 if (reiserfs_data_log(s))
1818 /* flush for 256 transactions or limit blocks, whichever comes first */
1819 for (i = 0; i < 256 && len < limit; i++) {
1820 if (atomic_read(&tjl->j_commit_left) ||
1821 tjl->j_trans_id < jl->j_trans_id) {
1824 cur_len = atomic_read(&tjl->j_nonzerolen);
1826 tjl->j_state &= ~LIST_TOUCHED;
1830 if (tjl->j_list.next == &journal->j_journal_list)
1832 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1834 /* try to find a group of blocks we can flush across all the
1835 ** transactions, but only bother if we've actually spanned
1836 ** across multiple lists
1838 if (flush_jl != jl) {
1839 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1841 flush_journal_list(s, flush_jl, 1);
1846 ** removes any nodes in table with name block and dev as bh.
1847 ** only touchs the hnext and hprev pointers.
1849 void remove_journal_hash(struct super_block *sb,
1850 struct reiserfs_journal_cnode **table,
1851 struct reiserfs_journal_list *jl,
1852 unsigned long block, int remove_freed)
1854 struct reiserfs_journal_cnode *cur;
1855 struct reiserfs_journal_cnode **head;
1857 head = &(journal_hash(table, sb, block));
1863 if (cur->blocknr == block && cur->sb == sb
1864 && (jl == NULL || jl == cur->jlist)
1865 && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1867 cur->hnext->hprev = cur->hprev;
1870 cur->hprev->hnext = cur->hnext;
1877 if (cur->bh && cur->jlist) /* anybody who clears the cur->bh will also dec the nonzerolen */
1878 atomic_dec(&(cur->jlist->j_nonzerolen));
1886 static void free_journal_ram(struct super_block *p_s_sb)
1888 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1889 kfree(journal->j_current_jl);
1890 journal->j_num_lists--;
1892 vfree(journal->j_cnode_free_orig);
1893 free_list_bitmaps(p_s_sb, journal->j_list_bitmap);
1894 free_bitmap_nodes(p_s_sb); /* must be after free_list_bitmaps */
1895 if (journal->j_header_bh) {
1896 brelse(journal->j_header_bh);
1898 /* j_header_bh is on the journal dev, make sure not to release the journal
1899 * dev until we brelse j_header_bh
1901 release_journal_dev(p_s_sb, journal);
1906 ** call on unmount. Only set error to 1 if you haven't made your way out
1907 ** of read_super() yet. Any other caller must keep error at 0.
1909 static int do_journal_release(struct reiserfs_transaction_handle *th,
1910 struct super_block *p_s_sb, int error)
1912 struct reiserfs_transaction_handle myth;
1914 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1916 /* we only want to flush out transactions if we were called with error == 0
1918 if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
1919 /* end the current trans */
1920 BUG_ON(!th->t_trans_id);
1921 do_journal_end(th, p_s_sb, 10, FLUSH_ALL);
1923 /* make sure something gets logged to force our way into the flush code */
1924 if (!journal_join(&myth, p_s_sb, 1)) {
1925 reiserfs_prepare_for_journal(p_s_sb,
1926 SB_BUFFER_WITH_SB(p_s_sb),
1928 journal_mark_dirty(&myth, p_s_sb,
1929 SB_BUFFER_WITH_SB(p_s_sb));
1930 do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1935 /* this also catches errors during the do_journal_end above */
1936 if (!error && reiserfs_is_journal_aborted(journal)) {
1937 memset(&myth, 0, sizeof(myth));
1938 if (!journal_join_abort(&myth, p_s_sb, 1)) {
1939 reiserfs_prepare_for_journal(p_s_sb,
1940 SB_BUFFER_WITH_SB(p_s_sb),
1942 journal_mark_dirty(&myth, p_s_sb,
1943 SB_BUFFER_WITH_SB(p_s_sb));
1944 do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1948 reiserfs_mounted_fs_count--;
1949 /* wait for all commits to finish */
1950 cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work);
1951 flush_workqueue(commit_wq);
1952 if (!reiserfs_mounted_fs_count) {
1953 destroy_workqueue(commit_wq);
1957 free_journal_ram(p_s_sb);
1963 ** call on unmount. flush all journal trans, release all alloc'd ram
1965 int journal_release(struct reiserfs_transaction_handle *th,
1966 struct super_block *p_s_sb)
1968 return do_journal_release(th, p_s_sb, 0);
1972 ** only call from an error condition inside reiserfs_read_super!
1974 int journal_release_error(struct reiserfs_transaction_handle *th,
1975 struct super_block *p_s_sb)
1977 return do_journal_release(th, p_s_sb, 1);
1980 /* compares description block with commit block. returns 1 if they differ, 0 if they are the same */
1981 static int journal_compare_desc_commit(struct super_block *p_s_sb,
1982 struct reiserfs_journal_desc *desc,
1983 struct reiserfs_journal_commit *commit)
1985 if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1986 get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1987 get_commit_trans_len(commit) > SB_JOURNAL(p_s_sb)->j_trans_max ||
1988 get_commit_trans_len(commit) <= 0) {
1994 /* returns 0 if it did not find a description block
1995 ** returns -1 if it found a corrupt commit block
1996 ** returns 1 if both desc and commit were valid
1998 static int journal_transaction_is_valid(struct super_block *p_s_sb,
1999 struct buffer_head *d_bh,
2000 unsigned long *oldest_invalid_trans_id,
2001 unsigned long *newest_mount_id)
2003 struct reiserfs_journal_desc *desc;
2004 struct reiserfs_journal_commit *commit;
2005 struct buffer_head *c_bh;
2006 unsigned long offset;
2011 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2012 if (get_desc_trans_len(desc) > 0
2013 && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2014 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2015 && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2016 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2017 "journal-986: transaction "
2018 "is valid returning because trans_id %d is greater than "
2019 "oldest_invalid %lu",
2020 get_desc_trans_id(desc),
2021 *oldest_invalid_trans_id);
2025 && *newest_mount_id > get_desc_mount_id(desc)) {
2026 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2027 "journal-1087: transaction "
2028 "is valid returning because mount_id %d is less than "
2029 "newest_mount_id %lu",
2030 get_desc_mount_id(desc),
2034 if (get_desc_trans_len(desc) > SB_JOURNAL(p_s_sb)->j_trans_max) {
2035 reiserfs_warning(p_s_sb,
2036 "journal-2018: Bad transaction length %d encountered, ignoring transaction",
2037 get_desc_trans_len(desc));
2040 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2042 /* ok, we have a journal description block, lets see if the transaction was valid */
2044 journal_bread(p_s_sb,
2045 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2046 ((offset + get_desc_trans_len(desc) +
2047 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2050 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2051 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2052 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2053 "journal_transaction_is_valid, commit offset %ld had bad "
2054 "time %d or length %d",
2056 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2057 get_commit_trans_id(commit),
2058 get_commit_trans_len(commit));
2060 if (oldest_invalid_trans_id) {
2061 *oldest_invalid_trans_id =
2062 get_desc_trans_id(desc);
2063 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2065 "transaction_is_valid setting oldest invalid trans_id "
2067 get_desc_trans_id(desc));
2072 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2073 "journal-1006: found valid "
2074 "transaction start offset %llu, len %d id %d",
2076 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2077 get_desc_trans_len(desc),
2078 get_desc_trans_id(desc));
2085 static void brelse_array(struct buffer_head **heads, int num)
2088 for (i = 0; i < num; i++) {
2094 ** given the start, and values for the oldest acceptable transactions,
2095 ** this either reads in a replays a transaction, or returns because the transaction
2096 ** is invalid, or too old.
2098 static int journal_read_transaction(struct super_block *p_s_sb,
2099 unsigned long cur_dblock,
2100 unsigned long oldest_start,
2101 unsigned long oldest_trans_id,
2102 unsigned long newest_mount_id)
2104 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2105 struct reiserfs_journal_desc *desc;
2106 struct reiserfs_journal_commit *commit;
2107 unsigned long trans_id = 0;
2108 struct buffer_head *c_bh;
2109 struct buffer_head *d_bh;
2110 struct buffer_head **log_blocks = NULL;
2111 struct buffer_head **real_blocks = NULL;
2112 unsigned long trans_offset;
2116 d_bh = journal_bread(p_s_sb, cur_dblock);
2119 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2120 trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2121 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
2122 "journal_read_transaction, offset %llu, len %d mount_id %d",
2123 d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2124 get_desc_trans_len(desc), get_desc_mount_id(desc));
2125 if (get_desc_trans_id(desc) < oldest_trans_id) {
2126 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
2127 "journal_read_trans skipping because %lu is too old",
2129 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2133 if (get_desc_mount_id(desc) != newest_mount_id) {
2134 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
2135 "journal_read_trans skipping because %d is != "
2136 "newest_mount_id %lu", get_desc_mount_id(desc),
2141 c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2142 ((trans_offset + get_desc_trans_len(desc) + 1) %
2143 SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2148 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2149 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2150 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2151 "journal_read_transaction, "
2152 "commit offset %llu had bad time %d or length %d",
2154 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2155 get_commit_trans_id(commit),
2156 get_commit_trans_len(commit));
2161 trans_id = get_desc_trans_id(desc);
2162 /* now we know we've got a good transaction, and it was inside the valid time ranges */
2163 log_blocks = kmalloc(get_desc_trans_len(desc) *
2164 sizeof(struct buffer_head *), GFP_NOFS);
2165 real_blocks = kmalloc(get_desc_trans_len(desc) *
2166 sizeof(struct buffer_head *), GFP_NOFS);
2167 if (!log_blocks || !real_blocks) {
2172 reiserfs_warning(p_s_sb,
2173 "journal-1169: kmalloc failed, unable to mount FS");
2176 /* get all the buffer heads */
2177 trans_half = journal_trans_half(p_s_sb->s_blocksize);
2178 for (i = 0; i < get_desc_trans_len(desc); i++) {
2180 journal_getblk(p_s_sb,
2181 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2183 i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2184 if (i < trans_half) {
2187 le32_to_cpu(desc->j_realblock[i]));
2191 le32_to_cpu(commit->
2192 j_realblock[i - trans_half]));
2194 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb)) {
2195 reiserfs_warning(p_s_sb,
2196 "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem");
2199 /* make sure we don't try to replay onto log or reserved area */
2200 if (is_block_in_log_or_reserved_area
2201 (p_s_sb, real_blocks[i]->b_blocknr)) {
2202 reiserfs_warning(p_s_sb,
2203 "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block");
2205 brelse_array(log_blocks, i);
2206 brelse_array(real_blocks, i);
2214 /* read in the log blocks, memcpy to the corresponding real block */
2215 ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2216 for (i = 0; i < get_desc_trans_len(desc); i++) {
2217 wait_on_buffer(log_blocks[i]);
2218 if (!buffer_uptodate(log_blocks[i])) {
2219 reiserfs_warning(p_s_sb,
2220 "journal-1212: REPLAY FAILURE fsck required! buffer write failed");
2221 brelse_array(log_blocks + i,
2222 get_desc_trans_len(desc) - i);
2223 brelse_array(real_blocks, get_desc_trans_len(desc));
2230 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2231 real_blocks[i]->b_size);
2232 set_buffer_uptodate(real_blocks[i]);
2233 brelse(log_blocks[i]);
2235 /* flush out the real blocks */
2236 for (i = 0; i < get_desc_trans_len(desc); i++) {
2237 set_buffer_dirty(real_blocks[i]);
2238 ll_rw_block(SWRITE, 1, real_blocks + i);
2240 for (i = 0; i < get_desc_trans_len(desc); i++) {
2241 wait_on_buffer(real_blocks[i]);
2242 if (!buffer_uptodate(real_blocks[i])) {
2243 reiserfs_warning(p_s_sb,
2244 "journal-1226: REPLAY FAILURE, fsck required! buffer write failed");
2245 brelse_array(real_blocks + i,
2246 get_desc_trans_len(desc) - i);
2253 brelse(real_blocks[i]);
2256 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2257 ((trans_offset + get_desc_trans_len(desc) +
2258 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2259 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2260 "journal-1095: setting journal " "start to offset %ld",
2261 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2263 /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2264 journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2265 journal->j_last_flush_trans_id = trans_id;
2266 journal->j_trans_id = trans_id + 1;
2267 /* check for trans_id overflow */
2268 if (journal->j_trans_id == 0)
2269 journal->j_trans_id = 10;
2277 /* This function reads blocks starting from block and to max_block of bufsize
2278 size (but no more than BUFNR blocks at a time). This proved to improve
2279 mounting speed on self-rebuilding raid5 arrays at least.
2280 Right now it is only used from journal code. But later we might use it
2282 Note: Do not use journal_getblk/sb_getblk functions here! */
2283 static struct buffer_head *reiserfs_breada(struct block_device *dev, int block,
2284 int bufsize, unsigned int max_block)
2286 struct buffer_head *bhlist[BUFNR];
2287 unsigned int blocks = BUFNR;
2288 struct buffer_head *bh;
2291 bh = __getblk(dev, block, bufsize);
2292 if (buffer_uptodate(bh))
2295 if (block + BUFNR > max_block) {
2296 blocks = max_block - block;
2300 for (i = 1; i < blocks; i++) {
2301 bh = __getblk(dev, block + i, bufsize);
2302 if (buffer_uptodate(bh)) {
2308 ll_rw_block(READ, j, bhlist);
2309 for (i = 1; i < j; i++)
2313 if (buffer_uptodate(bh))
2320 ** read and replay the log
2321 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2322 ** transaction. This tests that before finding all the transactions in the log, which makes normal mount times fast.
2324 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2326 ** On exit, it sets things up so the first transaction will work correctly.
2328 static int journal_read(struct super_block *p_s_sb)
2330 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2331 struct reiserfs_journal_desc *desc;
2332 unsigned long oldest_trans_id = 0;
2333 unsigned long oldest_invalid_trans_id = 0;
2335 unsigned long oldest_start = 0;
2336 unsigned long cur_dblock = 0;
2337 unsigned long newest_mount_id = 9;
2338 struct buffer_head *d_bh;
2339 struct reiserfs_journal_header *jh;
2340 int valid_journal_header = 0;
2341 int replay_count = 0;
2342 int continue_replay = 1;
2344 char b[BDEVNAME_SIZE];
2346 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2347 reiserfs_info(p_s_sb, "checking transaction log (%s)\n",
2348 bdevname(journal->j_dev_bd, b));
2349 start = get_seconds();
2351 /* step 1, read in the journal header block. Check the transaction it says
2352 ** is the first unflushed, and if that transaction is not valid,
2355 journal->j_header_bh = journal_bread(p_s_sb,
2356 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)
2357 + SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2358 if (!journal->j_header_bh) {
2361 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2362 if (le32_to_cpu(jh->j_first_unflushed_offset) <
2363 SB_ONDISK_JOURNAL_SIZE(p_s_sb)
2364 && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2366 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2367 le32_to_cpu(jh->j_first_unflushed_offset);
2368 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2369 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2370 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2371 "journal-1153: found in "
2372 "header: first_unflushed_offset %d, last_flushed_trans_id "
2373 "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2374 le32_to_cpu(jh->j_last_flush_trans_id));
2375 valid_journal_header = 1;
2377 /* now, we try to read the first unflushed offset. If it is not valid,
2378 ** there is nothing more we can do, and it makes no sense to read
2379 ** through the whole log.
2382 journal_bread(p_s_sb,
2383 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2384 le32_to_cpu(jh->j_first_unflushed_offset));
2385 ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL);
2387 continue_replay = 0;
2390 goto start_log_replay;
2393 if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) {
2394 reiserfs_warning(p_s_sb,
2395 "clm-2076: device is readonly, unable to replay log");
2399 /* ok, there are transactions that need to be replayed. start with the first log block, find
2400 ** all the valid transactions, and pick out the oldest.
2402 while (continue_replay
2404 (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2405 SB_ONDISK_JOURNAL_SIZE(p_s_sb))) {
2406 /* Note that it is required for blocksize of primary fs device and journal
2407 device to be the same */
2409 reiserfs_breada(journal->j_dev_bd, cur_dblock,
2410 p_s_sb->s_blocksize,
2411 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2412 SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2414 journal_transaction_is_valid(p_s_sb, d_bh,
2415 &oldest_invalid_trans_id,
2418 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2419 if (oldest_start == 0) { /* init all oldest_ values */
2420 oldest_trans_id = get_desc_trans_id(desc);
2421 oldest_start = d_bh->b_blocknr;
2422 newest_mount_id = get_desc_mount_id(desc);
2423 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2424 "journal-1179: Setting "
2425 "oldest_start to offset %llu, trans_id %lu",
2427 SB_ONDISK_JOURNAL_1st_BLOCK
2428 (p_s_sb), oldest_trans_id);
2429 } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2430 /* one we just read was older */
2431 oldest_trans_id = get_desc_trans_id(desc);
2432 oldest_start = d_bh->b_blocknr;
2433 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2434 "journal-1180: Resetting "
2435 "oldest_start to offset %lu, trans_id %lu",
2437 SB_ONDISK_JOURNAL_1st_BLOCK
2438 (p_s_sb), oldest_trans_id);
2440 if (newest_mount_id < get_desc_mount_id(desc)) {
2441 newest_mount_id = get_desc_mount_id(desc);
2442 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2443 "journal-1299: Setting "
2444 "newest_mount_id to %d",
2445 get_desc_mount_id(desc));
2447 cur_dblock += get_desc_trans_len(desc) + 2;
2455 cur_dblock = oldest_start;
2456 if (oldest_trans_id) {
2457 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2458 "journal-1206: Starting replay "
2459 "from offset %llu, trans_id %lu",
2460 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2465 while (continue_replay && oldest_trans_id > 0) {
2467 journal_read_transaction(p_s_sb, cur_dblock, oldest_start,
2468 oldest_trans_id, newest_mount_id);
2471 } else if (ret != 0) {
2475 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + journal->j_start;
2477 if (cur_dblock == oldest_start)
2481 if (oldest_trans_id == 0) {
2482 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2483 "journal-1225: No valid " "transactions found");
2485 /* j_start does not get set correctly if we don't replay any transactions.
2486 ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2487 ** copy the trans_id from the header
2489 if (valid_journal_header && replay_count == 0) {
2490 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2491 journal->j_trans_id =
2492 le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2493 /* check for trans_id overflow */
2494 if (journal->j_trans_id == 0)
2495 journal->j_trans_id = 10;
2496 journal->j_last_flush_trans_id =
2497 le32_to_cpu(jh->j_last_flush_trans_id);
2498 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2500 journal->j_mount_id = newest_mount_id + 1;
2502 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2503 "newest_mount_id to %lu", journal->j_mount_id);
2504 journal->j_first_unflushed_offset = journal->j_start;
2505 if (replay_count > 0) {
2506 reiserfs_info(p_s_sb,
2507 "replayed %d transactions in %lu seconds\n",
2508 replay_count, get_seconds() - start);
2510 if (!bdev_read_only(p_s_sb->s_bdev) &&
2511 _update_journal_header_block(p_s_sb, journal->j_start,
2512 journal->j_last_flush_trans_id)) {
2513 /* replay failed, caller must call free_journal_ram and abort
2521 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2523 struct reiserfs_journal_list *jl;
2524 jl = kzalloc(sizeof(struct reiserfs_journal_list),
2525 GFP_NOFS | __GFP_NOFAIL);
2526 INIT_LIST_HEAD(&jl->j_list);
2527 INIT_LIST_HEAD(&jl->j_working_list);
2528 INIT_LIST_HEAD(&jl->j_tail_bh_list);
2529 INIT_LIST_HEAD(&jl->j_bh_list);
2530 sema_init(&jl->j_commit_lock, 1);
2531 SB_JOURNAL(s)->j_num_lists++;
2532 get_journal_list(jl);
2536 static void journal_list_init(struct super_block *p_s_sb)
2538 SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
2541 static int release_journal_dev(struct super_block *super,
2542 struct reiserfs_journal *journal)
2548 if (journal->j_dev_file != NULL) {
2549 result = filp_close(journal->j_dev_file, NULL);
2550 journal->j_dev_file = NULL;
2551 journal->j_dev_bd = NULL;
2552 } else if (journal->j_dev_bd != NULL) {
2553 result = blkdev_put(journal->j_dev_bd);
2554 journal->j_dev_bd = NULL;
2558 reiserfs_warning(super,
2559 "sh-457: release_journal_dev: Cannot release journal device: %i",
2565 static int journal_init_dev(struct super_block *super,
2566 struct reiserfs_journal *journal,
2567 const char *jdev_name)
2571 int blkdev_mode = FMODE_READ | FMODE_WRITE;
2572 char b[BDEVNAME_SIZE];
2576 journal->j_dev_bd = NULL;
2577 journal->j_dev_file = NULL;
2578 jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2579 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2581 if (bdev_read_only(super->s_bdev))
2582 blkdev_mode = FMODE_READ;
2584 /* there is no "jdev" option and journal is on separate device */
2585 if ((!jdev_name || !jdev_name[0])) {
2586 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2587 if (IS_ERR(journal->j_dev_bd)) {
2588 result = PTR_ERR(journal->j_dev_bd);
2589 journal->j_dev_bd = NULL;
2590 reiserfs_warning(super, "sh-458: journal_init_dev: "
2591 "cannot init journal device '%s': %i",
2592 __bdevname(jdev, b), result);
2594 } else if (jdev != super->s_dev)
2595 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2599 journal->j_dev_file = filp_open(jdev_name, 0, 0);
2600 if (!IS_ERR(journal->j_dev_file)) {
2601 struct inode *jdev_inode = journal->j_dev_file->f_mapping->host;
2602 if (!S_ISBLK(jdev_inode->i_mode)) {
2603 reiserfs_warning(super, "journal_init_dev: '%s' is "
2604 "not a block device", jdev_name);
2606 release_journal_dev(super, journal);
2609 journal->j_dev_bd = I_BDEV(jdev_inode);
2610 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2611 reiserfs_info(super,
2612 "journal_init_dev: journal device: %s\n",
2613 bdevname(journal->j_dev_bd, b));
2616 result = PTR_ERR(journal->j_dev_file);
2617 journal->j_dev_file = NULL;
2618 reiserfs_warning(super,
2619 "journal_init_dev: Cannot open '%s': %i",
2626 ** must be called once on fs mount. calls journal_read for you
2628 int journal_init(struct super_block *p_s_sb, const char *j_dev_name,
2629 int old_format, unsigned int commit_max_age)
2631 int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2;
2632 struct buffer_head *bhjh;
2633 struct reiserfs_super_block *rs;
2634 struct reiserfs_journal_header *jh;
2635 struct reiserfs_journal *journal;
2636 struct reiserfs_journal_list *jl;
2637 char b[BDEVNAME_SIZE];
2639 journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof(struct reiserfs_journal));
2641 reiserfs_warning(p_s_sb,
2642 "journal-1256: unable to get memory for journal structure");
2645 memset(journal, 0, sizeof(struct reiserfs_journal));
2646 INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2647 INIT_LIST_HEAD(&journal->j_prealloc_list);
2648 INIT_LIST_HEAD(&journal->j_working_list);
2649 INIT_LIST_HEAD(&journal->j_journal_list);
2650 journal->j_persistent_trans = 0;
2651 if (reiserfs_allocate_list_bitmaps(p_s_sb,
2652 journal->j_list_bitmap,
2653 SB_BMAP_NR(p_s_sb)))
2654 goto free_and_return;
2655 allocate_bitmap_nodes(p_s_sb);
2657 /* reserved for journal area support */
2658 SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ?
2659 REISERFS_OLD_DISK_OFFSET_IN_BYTES
2660 / p_s_sb->s_blocksize +
2661 SB_BMAP_NR(p_s_sb) +
2663 REISERFS_DISK_OFFSET_IN_BYTES /
2664 p_s_sb->s_blocksize + 2);
2666 /* Sanity check to see is the standard journal fitting withing first bitmap
2667 (actual for small blocksizes) */
2668 if (!SB_ONDISK_JOURNAL_DEVICE(p_s_sb) &&
2669 (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) +
2670 SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8)) {
2671 reiserfs_warning(p_s_sb,
2672 "journal-1393: journal does not fit for area "
2673 "addressed by first of bitmap blocks. It starts at "
2674 "%u and its size is %u. Block size %ld",
2675 SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb),
2676 SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2677 p_s_sb->s_blocksize);
2678 goto free_and_return;
2681 if (journal_init_dev(p_s_sb, journal, j_dev_name) != 0) {
2682 reiserfs_warning(p_s_sb,
2683 "sh-462: unable to initialize jornal device");
2684 goto free_and_return;
2687 rs = SB_DISK_SUPER_BLOCK(p_s_sb);
2689 /* read journal header */
2690 bhjh = journal_bread(p_s_sb,
2691 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2692 SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2694 reiserfs_warning(p_s_sb,
2695 "sh-459: unable to read journal header");
2696 goto free_and_return;
2698 jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2700 /* make sure that journal matches to the super block */
2701 if (is_reiserfs_jr(rs)
2702 && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2703 sb_jp_journal_magic(rs))) {
2704 reiserfs_warning(p_s_sb,
2705 "sh-460: journal header magic %x "
2706 "(device %s) does not match to magic found in super "
2707 "block %x", jh->jh_journal.jp_journal_magic,
2708 bdevname(journal->j_dev_bd, b),
2709 sb_jp_journal_magic(rs));
2711 goto free_and_return;
2714 journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2715 journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2716 journal->j_max_commit_age =
2717 le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2718 journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2720 if (journal->j_trans_max) {
2721 /* make sure these parameters are available, assign it if they are not */
2722 __u32 initial = journal->j_trans_max;
2725 if (p_s_sb->s_blocksize < 4096)
2726 ratio = 4096 / p_s_sb->s_blocksize;
2728 if (SB_ONDISK_JOURNAL_SIZE(p_s_sb) / journal->j_trans_max <
2730 journal->j_trans_max =
2731 SB_ONDISK_JOURNAL_SIZE(p_s_sb) / JOURNAL_MIN_RATIO;
2732 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio)
2733 journal->j_trans_max =
2734 JOURNAL_TRANS_MAX_DEFAULT / ratio;
2735 if (journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio)
2736 journal->j_trans_max =
2737 JOURNAL_TRANS_MIN_DEFAULT / ratio;
2739 if (journal->j_trans_max != initial)
2740 reiserfs_warning(p_s_sb,
2741 "sh-461: journal_init: wrong transaction max size (%u). Changed to %u",
2742 initial, journal->j_trans_max);
2744 journal->j_max_batch = journal->j_trans_max *
2745 JOURNAL_MAX_BATCH_DEFAULT / JOURNAL_TRANS_MAX_DEFAULT;
2748 if (!journal->j_trans_max) {
2749 /*we have the file system was created by old version of mkreiserfs
2750 so this field contains zero value */
2751 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2752 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2753 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2755 /* for blocksize >= 4096 - max transaction size is 1024. For block size < 4096
2756 trans max size is decreased proportionally */
2757 if (p_s_sb->s_blocksize < 4096) {
2758 journal->j_trans_max /= (4096 / p_s_sb->s_blocksize);
2759 journal->j_max_batch = (journal->j_trans_max) * 9 / 10;
2763 journal->j_default_max_commit_age = journal->j_max_commit_age;
2765 if (commit_max_age != 0) {
2766 journal->j_max_commit_age = commit_max_age;
2767 journal->j_max_trans_age = commit_max_age;
2770 reiserfs_info(p_s_sb, "journal params: device %s, size %u, "
2771 "journal first block %u, max trans len %u, max batch %u, "
2772 "max commit age %u, max trans age %u\n",
2773 bdevname(journal->j_dev_bd, b),
2774 SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2775 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2776 journal->j_trans_max,
2777 journal->j_max_batch,
2778 journal->j_max_commit_age, journal->j_max_trans_age);
2782 journal->j_list_bitmap_index = 0;
2783 journal_list_init(p_s_sb);
2785 memset(journal->j_list_hash_table, 0,
2786 JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2788 INIT_LIST_HEAD(&journal->j_dirty_buffers);
2789 spin_lock_init(&journal->j_dirty_buffers_lock);
2791 journal->j_start = 0;
2793 journal->j_len_alloc = 0;
2794 atomic_set(&(journal->j_wcount), 0);
2795 atomic_set(&(journal->j_async_throttle), 0);
2796 journal->j_bcount = 0;
2797 journal->j_trans_start_time = 0;
2798 journal->j_last = NULL;
2799 journal->j_first = NULL;
2800 init_waitqueue_head(&(journal->j_join_wait));
2801 sema_init(&journal->j_lock, 1);
2802 sema_init(&journal->j_flush_sem, 1);
2804 journal->j_trans_id = 10;
2805 journal->j_mount_id = 10;
2806 journal->j_state = 0;
2807 atomic_set(&(journal->j_jlock), 0);
2808 journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2809 journal->j_cnode_free_orig = journal->j_cnode_free_list;
2810 journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2811 journal->j_cnode_used = 0;
2812 journal->j_must_wait = 0;
2814 if (journal->j_cnode_free == 0) {
2815 reiserfs_warning(p_s_sb, "journal-2004: Journal cnode memory "
2816 "allocation failed (%ld bytes). Journal is "
2817 "too large for available memory. Usually "
2818 "this is due to a journal that is too large.",
2819 sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2820 goto free_and_return;
2823 init_journal_hash(p_s_sb);
2824 jl = journal->j_current_jl;
2825 jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl);
2826 if (!jl->j_list_bitmap) {
2827 reiserfs_warning(p_s_sb,
2828 "journal-2005, get_list_bitmap failed for journal list 0");
2829 goto free_and_return;
2831 if (journal_read(p_s_sb) < 0) {
2832 reiserfs_warning(p_s_sb, "Replay Failure, unable to mount");
2833 goto free_and_return;
2836 reiserfs_mounted_fs_count++;
2837 if (reiserfs_mounted_fs_count <= 1)
2838 commit_wq = create_workqueue("reiserfs");
2840 INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2841 journal->j_work_sb = p_s_sb;
2844 free_journal_ram(p_s_sb);
2849 ** test for a polite end of the current transaction. Used by file_write, and should
2850 ** be used by delete to make sure they don't write more than can fit inside a single
2853 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2856 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2857 time_t now = get_seconds();
2858 /* cannot restart while nested */
2859 BUG_ON(!th->t_trans_id);
2860 if (th->t_refcount > 1)
2862 if (journal->j_must_wait > 0 ||
2863 (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2864 atomic_read(&(journal->j_jlock)) ||
2865 (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2866 journal->j_cnode_free < (journal->j_trans_max * 3)) {
2869 /* protected by the BKL here */
2870 journal->j_len_alloc += new_alloc;
2871 th->t_blocks_allocated += new_alloc ;
2875 /* this must be called inside a transaction, and requires the
2876 ** kernel_lock to be held
2878 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2880 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2881 BUG_ON(!th->t_trans_id);
2882 journal->j_must_wait = 1;
2883 set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2887 /* this must be called without a transaction started, and does not
2890 void reiserfs_allow_writes(struct super_block *s)
2892 struct reiserfs_journal *journal = SB_JOURNAL(s);
2893 clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2894 wake_up(&journal->j_join_wait);
2897 /* this must be called without a transaction started, and does not
2900 void reiserfs_wait_on_write_block(struct super_block *s)
2902 struct reiserfs_journal *journal = SB_JOURNAL(s);
2903 wait_event(journal->j_join_wait,
2904 !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2907 static void queue_log_writer(struct super_block *s)
2910 struct reiserfs_journal *journal = SB_JOURNAL(s);
2911 set_bit(J_WRITERS_QUEUED, &journal->j_state);
2914 * we don't want to use wait_event here because
2915 * we only want to wait once.
2917 init_waitqueue_entry(&wait, current);
2918 add_wait_queue(&journal->j_join_wait, &wait);
2919 set_current_state(TASK_UNINTERRUPTIBLE);
2920 if (test_bit(J_WRITERS_QUEUED, &journal->j_state))
2922 __set_current_state(TASK_RUNNING);
2923 remove_wait_queue(&journal->j_join_wait, &wait);
2926 static void wake_queued_writers(struct super_block *s)
2928 struct reiserfs_journal *journal = SB_JOURNAL(s);
2929 if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2930 wake_up(&journal->j_join_wait);
2933 static void let_transaction_grow(struct super_block *sb, unsigned long trans_id)
2935 struct reiserfs_journal *journal = SB_JOURNAL(sb);
2936 unsigned long bcount = journal->j_bcount;
2938 schedule_timeout_uninterruptible(1);
2939 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2940 while ((atomic_read(&journal->j_wcount) > 0 ||
2941 atomic_read(&journal->j_jlock)) &&
2942 journal->j_trans_id == trans_id) {
2943 queue_log_writer(sb);
2945 if (journal->j_trans_id != trans_id)
2947 if (bcount == journal->j_bcount)
2949 bcount = journal->j_bcount;
2953 /* join == true if you must join an existing transaction.
2954 ** join == false if you can deal with waiting for others to finish
2956 ** this will block until the transaction is joinable. send the number of blocks you
2957 ** expect to use in nblocks.
2959 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
2960 struct super_block *p_s_sb, unsigned long nblocks,
2963 time_t now = get_seconds();
2965 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2966 struct reiserfs_transaction_handle myth;
2967 int sched_count = 0;
2970 reiserfs_check_lock_depth(p_s_sb, "journal_begin");
2971 BUG_ON(nblocks > journal->j_trans_max);
2973 PROC_INFO_INC(p_s_sb, journal.journal_being);
2974 /* set here for journal_join */
2976 th->t_super = p_s_sb;
2979 lock_journal(p_s_sb);
2980 if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
2981 unlock_journal(p_s_sb);
2982 retval = journal->j_errno;
2985 journal->j_bcount++;
2987 if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
2988 unlock_journal(p_s_sb);
2989 reiserfs_wait_on_write_block(p_s_sb);
2990 PROC_INFO_INC(p_s_sb, journal.journal_relock_writers);
2993 now = get_seconds();
2995 /* if there is no room in the journal OR
2996 ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
2997 ** we don't sleep if there aren't other writers
3000 if ((!join && journal->j_must_wait > 0) ||
3002 && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3003 || (!join && atomic_read(&journal->j_wcount) > 0
3004 && journal->j_trans_start_time > 0
3005 && (now - journal->j_trans_start_time) >
3006 journal->j_max_trans_age) || (!join
3007 && atomic_read(&journal->j_jlock))
3008 || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3010 old_trans_id = journal->j_trans_id;
3011 unlock_journal(p_s_sb); /* allow others to finish this transaction */
3013 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3014 journal->j_max_batch &&
3015 ((journal->j_len + nblocks + 2) * 100) <
3016 (journal->j_len_alloc * 75)) {
3017 if (atomic_read(&journal->j_wcount) > 10) {
3019 queue_log_writer(p_s_sb);
3023 /* don't mess with joining the transaction if all we have to do is
3024 * wait for someone else to do a commit
3026 if (atomic_read(&journal->j_jlock)) {
3027 while (journal->j_trans_id == old_trans_id &&
3028 atomic_read(&journal->j_jlock)) {
3029 queue_log_writer(p_s_sb);
3033 retval = journal_join(&myth, p_s_sb, 1);
3037 /* someone might have ended the transaction while we joined */
3038 if (old_trans_id != journal->j_trans_id) {
3039 retval = do_journal_end(&myth, p_s_sb, 1, 0);
3041 retval = do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW);
3047 PROC_INFO_INC(p_s_sb, journal.journal_relock_wcount);
3050 /* we are the first writer, set trans_id */
3051 if (journal->j_trans_start_time == 0) {
3052 journal->j_trans_start_time = get_seconds();
3054 atomic_inc(&(journal->j_wcount));
3055 journal->j_len_alloc += nblocks;
3056 th->t_blocks_logged = 0;
3057 th->t_blocks_allocated = nblocks;
3058 th->t_trans_id = journal->j_trans_id;
3059 unlock_journal(p_s_sb);
3060 INIT_LIST_HEAD(&th->t_list);
3065 memset(th, 0, sizeof(*th));
3066 /* Re-set th->t_super, so we can properly keep track of how many
3067 * persistent transactions there are. We need to do this so if this
3068 * call is part of a failed restart_transaction, we can free it later */
3069 th->t_super = p_s_sb;
3073 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3079 struct reiserfs_transaction_handle *th;
3081 /* if we're nesting into an existing transaction. It will be
3082 ** persistent on its own
3084 if (reiserfs_transaction_running(s)) {
3085 th = current->journal_info;
3087 BUG_ON(th->t_refcount < 2);
3091 th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3094 ret = journal_begin(th, s, nblocks);
3100 SB_JOURNAL(s)->j_persistent_trans++;
3104 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3106 struct super_block *s = th->t_super;
3109 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3112 if (th->t_refcount == 0) {
3113 SB_JOURNAL(s)->j_persistent_trans--;
3119 static int journal_join(struct reiserfs_transaction_handle *th,
3120 struct super_block *p_s_sb, unsigned long nblocks)
3122 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3124 /* this keeps do_journal_end from NULLing out the current->journal_info
3127 th->t_handle_save = cur_th;
3128 BUG_ON(cur_th && cur_th->t_refcount > 1);
3129 return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_JOIN);
3132 int journal_join_abort(struct reiserfs_transaction_handle *th,
3133 struct super_block *p_s_sb, unsigned long nblocks)
3135 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3137 /* this keeps do_journal_end from NULLing out the current->journal_info
3140 th->t_handle_save = cur_th;
3141 BUG_ON(cur_th && cur_th->t_refcount > 1);
3142 return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_ABORT);
3145 int journal_begin(struct reiserfs_transaction_handle *th,
3146 struct super_block *p_s_sb, unsigned long nblocks)
3148 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3151 th->t_handle_save = NULL;
3153 /* we are nesting into the current transaction */
3154 if (cur_th->t_super == p_s_sb) {
3155 BUG_ON(!cur_th->t_refcount);
3156 cur_th->t_refcount++;
3157 memcpy(th, cur_th, sizeof(*th));
3158 if (th->t_refcount <= 1)
3159 reiserfs_warning(p_s_sb,
3160 "BAD: refcount <= 1, but journal_info != 0");
3163 /* we've ended up with a handle from a different filesystem.
3164 ** save it and restore on journal_end. This should never
3167 reiserfs_warning(p_s_sb,
3168 "clm-2100: nesting info a different FS");
3169 th->t_handle_save = current->journal_info;
3170 current->journal_info = th;
3173 current->journal_info = th;
3175 ret = do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_REG);
3176 BUG_ON(current->journal_info != th);
3178 /* I guess this boils down to being the reciprocal of clm-2100 above.
3179 * If do_journal_begin_r fails, we need to put it back, since journal_end
3180 * won't be called to do it. */
3182 current->journal_info = th->t_handle_save;
3184 BUG_ON(!th->t_refcount);
3190 ** puts bh into the current transaction. If it was already there, reorders removes the
3191 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3193 ** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the
3194 ** transaction is committed.
3196 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3198 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3199 struct super_block *p_s_sb, struct buffer_head *bh)
3201 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3202 struct reiserfs_journal_cnode *cn = NULL;
3203 int count_already_incd = 0;
3205 BUG_ON(!th->t_trans_id);
3207 PROC_INFO_INC(p_s_sb, journal.mark_dirty);
3208 if (th->t_trans_id != journal->j_trans_id) {
3209 reiserfs_panic(th->t_super,
3210 "journal-1577: handle trans id %ld != current trans id %ld\n",
3211 th->t_trans_id, journal->j_trans_id);
3216 prepared = test_clear_buffer_journal_prepared(bh);
3217 clear_buffer_journal_restore_dirty(bh);
3218 /* already in this transaction, we are done */
3219 if (buffer_journaled(bh)) {
3220 PROC_INFO_INC(p_s_sb, journal.mark_dirty_already);
3224 /* this must be turned into a panic instead of a warning. We can't allow
3225 ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3226 ** could get to disk too early. NOT GOOD.
3228 if (!prepared || buffer_dirty(bh)) {
3229 reiserfs_warning(p_s_sb, "journal-1777: buffer %llu bad state "
3230 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3231 (unsigned long long)bh->b_blocknr,
3232 prepared ? ' ' : '!',
3233 buffer_locked(bh) ? ' ' : '!',
3234 buffer_dirty(bh) ? ' ' : '!',
3235 buffer_journal_dirty(bh) ? ' ' : '!');
3238 if (atomic_read(&(journal->j_wcount)) <= 0) {
3239 reiserfs_warning(p_s_sb,
3240 "journal-1409: journal_mark_dirty returning because j_wcount was %d",
3241 atomic_read(&(journal->j_wcount)));
3244 /* this error means I've screwed up, and we've overflowed the transaction.
3245 ** Nothing can be done here, except make the FS readonly or panic.
3247 if (journal->j_len >= journal->j_trans_max) {
3248 reiserfs_panic(th->t_super,
3249 "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n",
3253 if (buffer_journal_dirty(bh)) {
3254 count_already_incd = 1;
3255 PROC_INFO_INC(p_s_sb, journal.mark_dirty_notjournal);
3256 clear_buffer_journal_dirty(bh);
3259 if (journal->j_len > journal->j_len_alloc) {
3260 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3263 set_buffer_journaled(bh);
3265 /* now put this guy on the end */
3267 cn = get_cnode(p_s_sb);
3269 reiserfs_panic(p_s_sb, "get_cnode failed!\n");
3272 if (th->t_blocks_logged == th->t_blocks_allocated) {
3273 th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3274 journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3276 th->t_blocks_logged++;
3280 cn->blocknr = bh->b_blocknr;
3283 insert_journal_hash(journal->j_hash_table, cn);
3284 if (!count_already_incd) {
3289 cn->prev = journal->j_last;
3291 if (journal->j_last) {
3292 journal->j_last->next = cn;
3293 journal->j_last = cn;
3295 journal->j_first = cn;
3296 journal->j_last = cn;
3301 int journal_end(struct reiserfs_transaction_handle *th,
3302 struct super_block *p_s_sb, unsigned long nblocks)
3304 if (!current->journal_info && th->t_refcount > 1)
3305 reiserfs_warning(p_s_sb, "REISER-NESTING: th NULL, refcount %d",
3308 if (!th->t_trans_id) {
3314 if (th->t_refcount > 0) {
3315 struct reiserfs_transaction_handle *cur_th =
3316 current->journal_info;
3318 /* we aren't allowed to close a nested transaction on a different
3319 ** filesystem from the one in the task struct
3321 BUG_ON(cur_th->t_super != th->t_super);
3324 memcpy(current->journal_info, th, sizeof(*th));
3329 return do_journal_end(th, p_s_sb, nblocks, 0);
3333 /* removes from the current transaction, relsing and descrementing any counters.
3334 ** also files the removed buffer directly onto the clean list
3336 ** called by journal_mark_freed when a block has been deleted
3338 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3340 static int remove_from_transaction(struct super_block *p_s_sb,
3341 b_blocknr_t blocknr, int already_cleaned)
3343 struct buffer_head *bh;
3344 struct reiserfs_journal_cnode *cn;
3345 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3348 cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3349 if (!cn || !cn->bh) {
3354 cn->prev->next = cn->next;
3357 cn->next->prev = cn->prev;
3359 if (cn == journal->j_first) {
3360 journal->j_first = cn->next;
3362 if (cn == journal->j_last) {
3363 journal->j_last = cn->prev;
3366 remove_journal_hash(p_s_sb, journal->j_hash_table, NULL,
3368 clear_buffer_journaled(bh); /* don't log this one */
3370 if (!already_cleaned) {
3371 clear_buffer_journal_dirty(bh);
3372 clear_buffer_dirty(bh);
3373 clear_buffer_journal_test(bh);
3375 if (atomic_read(&(bh->b_count)) < 0) {
3376 reiserfs_warning(p_s_sb,
3377 "journal-1752: remove from trans, b_count < 0");
3382 journal->j_len_alloc--;
3383 free_cnode(p_s_sb, cn);
3388 ** for any cnode in a journal list, it can only be dirtied of all the
3389 ** transactions that include it are committed to disk.
3390 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3391 ** and 0 if you aren't
3393 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3394 ** blocks for a given transaction on disk
3397 static int can_dirty(struct reiserfs_journal_cnode *cn)
3399 struct super_block *sb = cn->sb;
3400 b_blocknr_t blocknr = cn->blocknr;
3401 struct reiserfs_journal_cnode *cur = cn->hprev;
3404 /* first test hprev. These are all newer than cn, so any node here
3405 ** with the same block number and dev means this node can't be sent
3406 ** to disk right now.
3408 while (cur && can_dirty) {
3409 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3410 cur->blocknr == blocknr) {
3415 /* then test hnext. These are all older than cn. As long as they
3416 ** are committed to the log, it is safe to write cn to disk
3419 while (cur && can_dirty) {
3420 if (cur->jlist && cur->jlist->j_len > 0 &&
3421 atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3422 cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3430 /* syncs the commit blocks, but does not force the real buffers to disk
3431 ** will wait until the current transaction is done/committed before returning
3433 int journal_end_sync(struct reiserfs_transaction_handle *th,
3434 struct super_block *p_s_sb, unsigned long nblocks)
3436 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3438 BUG_ON(!th->t_trans_id);
3439 /* you can sync while nested, very, very bad */
3440 BUG_ON(th->t_refcount > 1);
3441 if (journal->j_len == 0) {
3442 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3444 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3446 return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT);
3450 ** writeback the pending async commits to disk
3452 static void flush_async_commits(struct work_struct *work)
3454 struct reiserfs_journal *journal =
3455 container_of(work, struct reiserfs_journal, j_work.work);
3456 struct super_block *p_s_sb = journal->j_work_sb;
3457 struct reiserfs_journal_list *jl;
3458 struct list_head *entry;
3461 if (!list_empty(&journal->j_journal_list)) {
3462 /* last entry is the youngest, commit it and you get everything */
3463 entry = journal->j_journal_list.prev;
3464 jl = JOURNAL_LIST_ENTRY(entry);
3465 flush_commit_list(p_s_sb, jl, 1);
3471 ** flushes any old transactions to disk
3472 ** ends the current transaction if it is too old
3474 int reiserfs_flush_old_commits(struct super_block *p_s_sb)
3477 struct reiserfs_transaction_handle th;
3478 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3480 now = get_seconds();
3481 /* safety check so we don't flush while we are replaying the log during
3484 if (list_empty(&journal->j_journal_list)) {
3488 /* check the current transaction. If there are no writers, and it is
3489 * too old, finish it, and force the commit blocks to disk
3491 if (atomic_read(&journal->j_wcount) <= 0 &&
3492 journal->j_trans_start_time > 0 &&
3493 journal->j_len > 0 &&
3494 (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3495 if (!journal_join(&th, p_s_sb, 1)) {
3496 reiserfs_prepare_for_journal(p_s_sb,
3497 SB_BUFFER_WITH_SB(p_s_sb),
3499 journal_mark_dirty(&th, p_s_sb,
3500 SB_BUFFER_WITH_SB(p_s_sb));
3502 /* we're only being called from kreiserfsd, it makes no sense to do
3503 ** an async commit so that kreiserfsd can do it later
3505 do_journal_end(&th, p_s_sb, 1, COMMIT_NOW | WAIT);
3508 return p_s_sb->s_dirt;
3512 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3514 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3515 ** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just
3516 ** flushes the commit list and returns 0.
3518 ** Won't batch when flush or commit_now is set. Also won't batch when others are waiting on j_join_wait.
3520 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3522 static int check_journal_end(struct reiserfs_transaction_handle *th,
3523 struct super_block *p_s_sb, unsigned long nblocks,
3528 int flush = flags & FLUSH_ALL;
3529 int commit_now = flags & COMMIT_NOW;
3530 int wait_on_commit = flags & WAIT;
3531 struct reiserfs_journal_list *jl;
3532 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3534 BUG_ON(!th->t_trans_id);
3536 if (th->t_trans_id != journal->j_trans_id) {
3537 reiserfs_panic(th->t_super,
3538 "journal-1577: handle trans id %ld != current trans id %ld\n",
3539 th->t_trans_id, journal->j_trans_id);
3542 journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3543 if (atomic_read(&(journal->j_wcount)) > 0) { /* <= 0 is allowed. unmounting might not call begin */
3544 atomic_dec(&(journal->j_wcount));
3547 /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3548 ** will be dealt with by next transaction that actually writes something, but should be taken
3549 ** care of in this trans
3551 BUG_ON(journal->j_len == 0);
3553 /* if wcount > 0, and we are called to with flush or commit_now,
3554 ** we wait on j_join_wait. We will wake up when the last writer has
3555 ** finished the transaction, and started it on its way to the disk.
3556 ** Then, we flush the commit or journal list, and just return 0
3557 ** because the rest of journal end was already done for this transaction.
3559 if (atomic_read(&(journal->j_wcount)) > 0) {
3560 if (flush || commit_now) {
3563 jl = journal->j_current_jl;
3564 trans_id = jl->j_trans_id;
3566 jl->j_state |= LIST_COMMIT_PENDING;
3567 atomic_set(&(journal->j_jlock), 1);
3569 journal->j_next_full_flush = 1;
3571 unlock_journal(p_s_sb);
3573 /* sleep while the current transaction is still j_jlocked */
3574 while (journal->j_trans_id == trans_id) {
3575 if (atomic_read(&journal->j_jlock)) {
3576 queue_log_writer(p_s_sb);
3578 lock_journal(p_s_sb);
3579 if (journal->j_trans_id == trans_id) {
3580 atomic_set(&(journal->j_jlock),
3583 unlock_journal(p_s_sb);
3586 BUG_ON(journal->j_trans_id == trans_id);
3589 && journal_list_still_alive(p_s_sb, trans_id)
3590 && wait_on_commit) {
3591 flush_commit_list(p_s_sb, jl, 1);
3595 unlock_journal(p_s_sb);
3599 /* deal with old transactions where we are the last writers */
3600 now = get_seconds();
3601 if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3603 journal->j_next_async_flush = 1;
3605 /* don't batch when someone is waiting on j_join_wait */
3606 /* don't batch when syncing the commit or flushing the whole trans */
3607 if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3608 && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3609 && journal->j_len_alloc < journal->j_max_batch
3610 && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3611 journal->j_bcount++;
3612 unlock_journal(p_s_sb);
3616 if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
3617 reiserfs_panic(p_s_sb,
3618 "journal-003: journal_end: j_start (%ld) is too high\n",
3625 ** Does all the work that makes deleting blocks safe.
3626 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3629 ** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes
3630 ** before this transaction has finished.
3632 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with
3633 ** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash,
3634 ** the block can't be reallocated yet.
3636 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3638 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3639 struct super_block *p_s_sb, b_blocknr_t blocknr)
3641 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3642 struct reiserfs_journal_cnode *cn = NULL;
3643 struct buffer_head *bh = NULL;
3644 struct reiserfs_list_bitmap *jb = NULL;
3646 BUG_ON(!th->t_trans_id);
3648 cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3653 /* if it is journal new, we just remove it from this transaction */
3654 if (bh && buffer_journal_new(bh)) {
3655 clear_buffer_journal_new(bh);
3656 clear_prepared_bits(bh);
3657 reiserfs_clean_and_file_buffer(bh);
3658 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3660 /* set the bit for this block in the journal bitmap for this transaction */
3661 jb = journal->j_current_jl->j_list_bitmap;
3663 reiserfs_panic(p_s_sb,
3664 "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n");
3666 set_bit_in_list_bitmap(p_s_sb, blocknr, jb);
3668 /* Note, the entire while loop is not allowed to schedule. */
3671 clear_prepared_bits(bh);
3672 reiserfs_clean_and_file_buffer(bh);
3674 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3676 /* find all older transactions with this block, make sure they don't try to write it out */
3677 cn = get_journal_hash_dev(p_s_sb, journal->j_list_hash_table,
3680 if (p_s_sb == cn->sb && blocknr == cn->blocknr) {
3681 set_bit(BLOCK_FREED, &cn->state);
3684 /* remove_from_transaction will brelse the buffer if it was
3685 ** in the current trans
3687 clear_buffer_journal_dirty(cn->
3689 clear_buffer_dirty(cn->bh);
3690 clear_buffer_journal_test(cn->
3695 (&(cn->bh->b_count)) < 0) {
3696 reiserfs_warning(p_s_sb,
3697 "journal-2138: cn->bh->b_count < 0");
3700 if (cn->jlist) { /* since we are clearing the bh, we MUST dec nonzerolen */
3713 put_bh(bh); /* get_hash grabs the buffer */
3714 if (atomic_read(&(bh->b_count)) < 0) {
3715 reiserfs_warning(p_s_sb,
3716 "journal-2165: bh->b_count < 0");
3722 void reiserfs_update_inode_transaction(struct inode *inode)
3724 struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3725 REISERFS_I(inode)->i_jl = journal->j_current_jl;
3726 REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3730 * returns -1 on error, 0 if no commits/barriers were done and 1
3731 * if a transaction was actually committed and the barrier was done
3733 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3734 struct reiserfs_journal_list *jl)
3736 struct reiserfs_transaction_handle th;
3737 struct super_block *sb = inode->i_sb;
3738 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3741 /* is it from the current transaction, or from an unknown transaction? */
3742 if (id == journal->j_trans_id) {
3743 jl = journal->j_current_jl;
3744 /* try to let other writers come in and grow this transaction */
3745 let_transaction_grow(sb, id);
3746 if (journal->j_trans_id != id) {
3747 goto flush_commit_only;
3750 ret = journal_begin(&th, sb, 1);
3754 /* someone might have ended this transaction while we joined */
3755 if (journal->j_trans_id != id) {
3756 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3758 journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3759 ret = journal_end(&th, sb, 1);
3760 goto flush_commit_only;
3763 ret = journal_end_sync(&th, sb, 1);
3768 /* this gets tricky, we have to make sure the journal list in
3769 * the inode still exists. We know the list is still around
3770 * if we've got a larger transaction id than the oldest list
3773 if (journal_list_still_alive(inode->i_sb, id)) {
3775 * we only set ret to 1 when we know for sure
3776 * the barrier hasn't been started yet on the commit
3779 if (atomic_read(&jl->j_commit_left) > 1)
3781 flush_commit_list(sb, jl, 1);
3782 if (journal->j_errno)
3783 ret = journal->j_errno;
3786 /* otherwise the list is gone, and long since committed */
3790 int reiserfs_commit_for_inode(struct inode *inode)
3792 unsigned long id = REISERFS_I(inode)->i_trans_id;
3793 struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3795 /* for the whole inode, assume unset id means it was
3796 * changed in the current transaction. More conservative
3799 reiserfs_update_inode_transaction(inode);
3800 id = REISERFS_I(inode)->i_trans_id;
3801 /* jl will be updated in __commit_trans_jl */
3804 return __commit_trans_jl(inode, id, jl);
3807 void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb,
3808 struct buffer_head *bh)
3810 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3811 PROC_INFO_INC(p_s_sb, journal.restore_prepared);
3815 if (test_clear_buffer_journal_restore_dirty(bh) &&
3816 buffer_journal_dirty(bh)) {
3817 struct reiserfs_journal_cnode *cn;
3818 cn = get_journal_hash_dev(p_s_sb,
3819 journal->j_list_hash_table,
3821 if (cn && can_dirty(cn)) {
3822 set_buffer_journal_test(bh);
3823 mark_buffer_dirty(bh);
3826 clear_buffer_journal_prepared(bh);
3829 extern struct tree_balance *cur_tb;
3831 ** before we can change a metadata block, we have to make sure it won't
3832 ** be written to disk while we are altering it. So, we must:
3837 int reiserfs_prepare_for_journal(struct super_block *p_s_sb,
3838 struct buffer_head *bh, int wait)
3840 PROC_INFO_INC(p_s_sb, journal.prepare);
3842 if (test_set_buffer_locked(bh)) {
3847 set_buffer_journal_prepared(bh);
3848 if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3849 clear_buffer_journal_test(bh);
3850 set_buffer_journal_restore_dirty(bh);
3856 static void flush_old_journal_lists(struct super_block *s)
3858 struct reiserfs_journal *journal = SB_JOURNAL(s);
3859 struct reiserfs_journal_list *jl;
3860 struct list_head *entry;
3861 time_t now = get_seconds();
3863 while (!list_empty(&journal->j_journal_list)) {
3864 entry = journal->j_journal_list.next;
3865 jl = JOURNAL_LIST_ENTRY(entry);
3866 /* this check should always be run, to send old lists to disk */
3867 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) &&
3868 atomic_read(&jl->j_commit_left) == 0 &&
3869 test_transaction(s, jl)) {
3870 flush_used_journal_lists(s, jl);
3878 ** long and ugly. If flush, will not return until all commit
3879 ** blocks and all real buffers in the trans are on disk.
3880 ** If no_async, won't return until all commit blocks are on disk.
3882 ** keep reading, there are comments as you go along
3884 ** If the journal is aborted, we just clean up. Things like flushing
3885 ** journal lists, etc just won't happen.
3887 static int do_journal_end(struct reiserfs_transaction_handle *th,
3888 struct super_block *p_s_sb, unsigned long nblocks,
3891 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3892 struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3893 struct reiserfs_journal_cnode *last_cn = NULL;
3894 struct reiserfs_journal_desc *desc;
3895 struct reiserfs_journal_commit *commit;
3896 struct buffer_head *c_bh; /* commit bh */
3897 struct buffer_head *d_bh; /* desc bh */
3898 int cur_write_start = 0; /* start index of current log write */
3903 struct reiserfs_journal_list *jl, *temp_jl;
3904 struct list_head *entry, *safe;
3905 unsigned long jindex;
3906 unsigned long commit_trans_id;
3909 BUG_ON(th->t_refcount > 1);
3910 BUG_ON(!th->t_trans_id);
3912 /* protect flush_older_commits from doing mistakes if the
3913 transaction ID counter gets overflowed. */
3914 if (th->t_trans_id == ~0UL)
3915 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
3916 flush = flags & FLUSH_ALL;
3917 wait_on_commit = flags & WAIT;
3920 current->journal_info = th->t_handle_save;
3921 reiserfs_check_lock_depth(p_s_sb, "journal end");
3922 if (journal->j_len == 0) {
3923 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3925 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3928 lock_journal(p_s_sb);
3929 if (journal->j_next_full_flush) {
3933 if (journal->j_next_async_flush) {
3934 flags |= COMMIT_NOW | WAIT;
3938 /* check_journal_end locks the journal, and unlocks if it does not return 1
3939 ** it tells us if we should continue with the journal_end, or just return
3941 if (!check_journal_end(th, p_s_sb, nblocks, flags)) {
3943 wake_queued_writers(p_s_sb);
3944 reiserfs_async_progress_wait(p_s_sb);
3948 /* check_journal_end might set these, check again */
3949 if (journal->j_next_full_flush) {
3954 ** j must wait means we have to flush the log blocks, and the real blocks for
3957 if (journal->j_must_wait > 0) {
3960 #ifdef REISERFS_PREALLOCATE
3961 /* quota ops might need to nest, setup the journal_info pointer for them
3962 * and raise the refcount so that it is > 0. */
3963 current->journal_info = th;
3965 reiserfs_discard_all_prealloc(th); /* it should not involve new blocks into
3966 * the transaction */
3968 current->journal_info = th->t_handle_save;
3971 /* setup description block */
3973 journal_getblk(p_s_sb,
3974 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3976 set_buffer_uptodate(d_bh);
3977 desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
3978 memset(d_bh->b_data, 0, d_bh->b_size);
3979 memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
3980 set_desc_trans_id(desc, journal->j_trans_id);
3982 /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */
3983 c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3984 ((journal->j_start + journal->j_len +
3985 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
3986 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
3987 memset(c_bh->b_data, 0, c_bh->b_size);
3988 set_commit_trans_id(commit, journal->j_trans_id);
3989 set_buffer_uptodate(c_bh);
3991 /* init this journal list */
3992 jl = journal->j_current_jl;
3994 /* we lock the commit before doing anything because
3995 * we want to make sure nobody tries to run flush_commit_list until
3996 * the new transaction is fully setup, and we've already flushed the
3999 down(&jl->j_commit_lock);
4001 /* save the transaction id in case we need to commit it later */
4002 commit_trans_id = jl->j_trans_id;
4004 atomic_set(&jl->j_older_commits_done, 0);
4005 jl->j_trans_id = journal->j_trans_id;
4006 jl->j_timestamp = journal->j_trans_start_time;
4007 jl->j_commit_bh = c_bh;
4008 jl->j_start = journal->j_start;
4009 jl->j_len = journal->j_len;
4010 atomic_set(&jl->j_nonzerolen, journal->j_len);
4011 atomic_set(&jl->j_commit_left, journal->j_len + 2);
4012 jl->j_realblock = NULL;
4014 /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4015 ** for each real block, add it to the journal list hash,
4016 ** copy into real block index array in the commit or desc block
4018 trans_half = journal_trans_half(p_s_sb->s_blocksize);
4019 for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4020 if (buffer_journaled(cn->bh)) {
4021 jl_cn = get_cnode(p_s_sb);
4023 reiserfs_panic(p_s_sb,
4024 "journal-1676, get_cnode returned NULL\n");
4027 jl->j_realblock = jl_cn;
4029 jl_cn->prev = last_cn;
4032 last_cn->next = jl_cn;
4035 /* make sure the block we are trying to log is not a block
4036 of journal or reserved area */
4038 if (is_block_in_log_or_reserved_area
4039 (p_s_sb, cn->bh->b_blocknr)) {
4040 reiserfs_panic(p_s_sb,
4041 "journal-2332: Trying to log block %lu, which is a log block\n",
4044 jl_cn->blocknr = cn->bh->b_blocknr;
4049 insert_journal_hash(journal->j_list_hash_table, jl_cn);
4050 if (i < trans_half) {
4051 desc->j_realblock[i] =
4052 cpu_to_le32(cn->bh->b_blocknr);
4054 commit->j_realblock[i - trans_half] =
4055 cpu_to_le32(cn->bh->b_blocknr);
4061 set_desc_trans_len(desc, journal->j_len);
4062 set_desc_mount_id(desc, journal->j_mount_id);
4063 set_desc_trans_id(desc, journal->j_trans_id);
4064 set_commit_trans_len(commit, journal->j_len);
4066 /* special check in case all buffers in the journal were marked for not logging */
4067 BUG_ON(journal->j_len == 0);
4069 /* we're about to dirty all the log blocks, mark the description block
4070 * dirty now too. Don't mark the commit block dirty until all the
4071 * others are on disk
4073 mark_buffer_dirty(d_bh);
4075 /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4076 cur_write_start = journal->j_start;
4077 cn = journal->j_first;
4078 jindex = 1; /* start at one so we don't get the desc again */
4080 clear_buffer_journal_new(cn->bh);
4081 /* copy all the real blocks into log area. dirty log blocks */
4082 if (buffer_journaled(cn->bh)) {
4083 struct buffer_head *tmp_bh;
4087 journal_getblk(p_s_sb,
4088 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4091 SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
4092 set_buffer_uptodate(tmp_bh);
4093 page = cn->bh->b_page;
4095 memcpy(tmp_bh->b_data,
4096 addr + offset_in_page(cn->bh->b_data),
4099 mark_buffer_dirty(tmp_bh);
4101 set_buffer_journal_dirty(cn->bh);
4102 clear_buffer_journaled(cn->bh);
4104 /* JDirty cleared sometime during transaction. don't log this one */
4105 reiserfs_warning(p_s_sb,
4106 "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!");
4110 free_cnode(p_s_sb, cn);
4115 /* we are done with both the c_bh and d_bh, but
4116 ** c_bh must be written after all other commit blocks,
4117 ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4120 journal->j_current_jl = alloc_journal_list(p_s_sb);
4122 /* now it is safe to insert this transaction on the main list */
4123 list_add_tail(&jl->j_list, &journal->j_journal_list);
4124 list_add_tail(&jl->j_working_list, &journal->j_working_list);
4125 journal->j_num_work_lists++;
4127 /* reset journal values for the next transaction */
4128 old_start = journal->j_start;
4130 (journal->j_start + journal->j_len +
4131 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb);
4132 atomic_set(&(journal->j_wcount), 0);
4133 journal->j_bcount = 0;
4134 journal->j_last = NULL;
4135 journal->j_first = NULL;
4137 journal->j_trans_start_time = 0;
4138 /* check for trans_id overflow */
4139 if (++journal->j_trans_id == 0)
4140 journal->j_trans_id = 10;
4141 journal->j_current_jl->j_trans_id = journal->j_trans_id;
4142 journal->j_must_wait = 0;
4143 journal->j_len_alloc = 0;
4144 journal->j_next_full_flush = 0;
4145 journal->j_next_async_flush = 0;
4146 init_journal_hash(p_s_sb);
4148 // make sure reiserfs_add_jh sees the new current_jl before we
4149 // write out the tails
4152 /* tail conversion targets have to hit the disk before we end the
4153 * transaction. Otherwise a later transaction might repack the tail
4154 * before this transaction commits, leaving the data block unflushed and
4155 * clean, if we crash before the later transaction commits, the data block
4158 if (!list_empty(&jl->j_tail_bh_list)) {
4160 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4161 journal, jl, &jl->j_tail_bh_list);
4164 BUG_ON(!list_empty(&jl->j_tail_bh_list));
4165 up(&jl->j_commit_lock);
4167 /* honor the flush wishes from the caller, simple commits can
4168 ** be done outside the journal lock, they are done below
4170 ** if we don't flush the commit list right now, we put it into
4171 ** the work queue so the people waiting on the async progress work
4172 ** queue don't wait for this proc to flush journal lists and such.
4175 flush_commit_list(p_s_sb, jl, 1);
4176 flush_journal_list(p_s_sb, jl, 1);
4177 } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4178 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4180 /* if the next transaction has any chance of wrapping, flush
4181 ** transactions that might get overwritten. If any journal lists are very
4182 ** old flush them as well.
4185 list_for_each_safe(entry, safe, &journal->j_journal_list) {
4186 temp_jl = JOURNAL_LIST_ENTRY(entry);
4187 if (journal->j_start <= temp_jl->j_start) {
4188 if ((journal->j_start + journal->j_trans_max + 1) >=
4190 flush_used_journal_lists(p_s_sb, temp_jl);
4192 } else if ((journal->j_start +
4193 journal->j_trans_max + 1) <
4194 SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4195 /* if we don't cross into the next transaction and we don't
4196 * wrap, there is no way we can overlap any later transactions
4201 } else if ((journal->j_start +
4202 journal->j_trans_max + 1) >
4203 SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4204 if (((journal->j_start + journal->j_trans_max + 1) %
4205 SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >=
4207 flush_used_journal_lists(p_s_sb, temp_jl);
4210 /* we don't overlap anything from out start to the end of the
4211 * log, and our wrapped portion doesn't overlap anything at
4212 * the start of the log. We can break
4218 flush_old_journal_lists(p_s_sb);
4220 journal->j_current_jl->j_list_bitmap =
4221 get_list_bitmap(p_s_sb, journal->j_current_jl);
4223 if (!(journal->j_current_jl->j_list_bitmap)) {
4224 reiserfs_panic(p_s_sb,
4225 "journal-1996: do_journal_end, could not get a list bitmap\n");
4228 atomic_set(&(journal->j_jlock), 0);
4229 unlock_journal(p_s_sb);
4230 /* wake up any body waiting to join. */
4231 clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4232 wake_up(&(journal->j_join_wait));
4234 if (!flush && wait_on_commit &&
4235 journal_list_still_alive(p_s_sb, commit_trans_id)) {
4236 flush_commit_list(p_s_sb, jl, 1);
4239 reiserfs_check_lock_depth(p_s_sb, "journal end2");
4241 memset(th, 0, sizeof(*th));
4242 /* Re-set th->t_super, so we can properly keep track of how many
4243 * persistent transactions there are. We need to do this so if this
4244 * call is part of a failed restart_transaction, we can free it later */
4245 th->t_super = p_s_sb;
4247 return journal->j_errno;
4250 static void __reiserfs_journal_abort_hard(struct super_block *sb)
4252 struct reiserfs_journal *journal = SB_JOURNAL(sb);
4253 if (test_bit(J_ABORTED, &journal->j_state))
4256 printk(KERN_CRIT "REISERFS: Aborting journal for filesystem on %s\n",
4257 reiserfs_bdevname(sb));
4259 sb->s_flags |= MS_RDONLY;
4260 set_bit(J_ABORTED, &journal->j_state);
4262 #ifdef CONFIG_REISERFS_CHECK
4267 static void __reiserfs_journal_abort_soft(struct super_block *sb, int errno)
4269 struct reiserfs_journal *journal = SB_JOURNAL(sb);
4270 if (test_bit(J_ABORTED, &journal->j_state))
4273 if (!journal->j_errno)
4274 journal->j_errno = errno;
4276 __reiserfs_journal_abort_hard(sb);
4279 void reiserfs_journal_abort(struct super_block *sb, int errno)
4281 return __reiserfs_journal_abort_soft(sb, errno);