reiserfs, kill-the-BKL: fix unsafe j_flush_mutex lock
[linux-2.6.git] / fs / reiserfs / journal.c
1 /*
2 ** Write ahead logging implementation copyright Chris Mason 2000
3 **
4 ** The background commits make this code very interelated, and
5 ** overly complex.  I need to rethink things a bit....The major players:
6 **
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.
13 **
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.
18 **
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.
23 **
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
27 **                   transaction.
28 **
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
32 **                      around too long.
33 **                   -- Note, if you call this as an immediate flush from
34 **                      from within kupdate, it will ignore the immediate flag
35 */
36
37 #include <linux/time.h>
38 #include <linux/semaphore.h>
39 #include <linux/vmalloc.h>
40 #include <linux/reiserfs_fs.h>
41 #include <linux/kernel.h>
42 #include <linux/errno.h>
43 #include <linux/fcntl.h>
44 #include <linux/stat.h>
45 #include <linux/string.h>
46 #include <linux/smp_lock.h>
47 #include <linux/buffer_head.h>
48 #include <linux/workqueue.h>
49 #include <linux/writeback.h>
50 #include <linux/blkdev.h>
51 #include <linux/backing-dev.h>
52 #include <linux/uaccess.h>
53
54 #include <asm/system.h>
55
56 /* gets a struct reiserfs_journal_list * from a list head */
57 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
58                                j_list))
59 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
60                                j_working_list))
61
62 /* the number of mounted filesystems.  This is used to decide when to
63 ** start and kill the commit workqueue
64 */
65 static int reiserfs_mounted_fs_count;
66
67 static struct workqueue_struct *commit_wq;
68
69 #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
70                                    structs at 4k */
71 #define BUFNR 64                /*read ahead */
72
73 /* cnode stat bits.  Move these into reiserfs_fs.h */
74
75 #define BLOCK_FREED 2           /* this block was freed, and can't be written.  */
76 #define BLOCK_FREED_HOLDER 3    /* this block was freed during this transaction, and can't be written */
77
78 #define BLOCK_NEEDS_FLUSH 4     /* used in flush_journal_list */
79 #define BLOCK_DIRTIED 5
80
81 /* journal list state bits */
82 #define LIST_TOUCHED 1
83 #define LIST_DIRTY   2
84 #define LIST_COMMIT_PENDING  4  /* someone will commit this list */
85
86 /* flags for do_journal_end */
87 #define FLUSH_ALL   1           /* flush commit and real blocks */
88 #define COMMIT_NOW  2           /* end and commit this transaction */
89 #define WAIT        4           /* wait for the log blocks to hit the disk */
90
91 static int do_journal_end(struct reiserfs_transaction_handle *,
92                           struct super_block *, unsigned long nblocks,
93                           int flags);
94 static int flush_journal_list(struct super_block *s,
95                               struct reiserfs_journal_list *jl, int flushall);
96 static int flush_commit_list(struct super_block *s,
97                              struct reiserfs_journal_list *jl, int flushall);
98 static int can_dirty(struct reiserfs_journal_cnode *cn);
99 static int journal_join(struct reiserfs_transaction_handle *th,
100                         struct super_block *sb, unsigned long nblocks);
101 static int release_journal_dev(struct super_block *super,
102                                struct reiserfs_journal *journal);
103 static int dirty_one_transaction(struct super_block *s,
104                                  struct reiserfs_journal_list *jl);
105 static void flush_async_commits(struct work_struct *work);
106 static void queue_log_writer(struct super_block *s);
107
108 /* values for join in do_journal_begin_r */
109 enum {
110         JBEGIN_REG = 0,         /* regular journal begin */
111         JBEGIN_JOIN = 1,        /* join the running transaction if at all possible */
112         JBEGIN_ABORT = 2,       /* called from cleanup code, ignores aborted flag */
113 };
114
115 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
116                               struct super_block *sb,
117                               unsigned long nblocks, int join);
118
119 static void init_journal_hash(struct super_block *sb)
120 {
121         struct reiserfs_journal *journal = SB_JOURNAL(sb);
122         memset(journal->j_hash_table, 0,
123                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
124 }
125
126 /*
127 ** clears BH_Dirty and sticks the buffer on the clean list.  Called because I can't allow refile_buffer to
128 ** make schedule happen after I've freed a block.  Look at remove_from_transaction and journal_mark_freed for
129 ** more details.
130 */
131 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
132 {
133         if (bh) {
134                 clear_buffer_dirty(bh);
135                 clear_buffer_journal_test(bh);
136         }
137         return 0;
138 }
139
140 static void disable_barrier(struct super_block *s)
141 {
142         REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH);
143         printk("reiserfs: disabling flush barriers on %s\n",
144                reiserfs_bdevname(s));
145 }
146
147 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
148                                                          *sb)
149 {
150         struct reiserfs_bitmap_node *bn;
151         static int id;
152
153         bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
154         if (!bn) {
155                 return NULL;
156         }
157         bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
158         if (!bn->data) {
159                 kfree(bn);
160                 return NULL;
161         }
162         bn->id = id++;
163         INIT_LIST_HEAD(&bn->list);
164         return bn;
165 }
166
167 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
168 {
169         struct reiserfs_journal *journal = SB_JOURNAL(sb);
170         struct reiserfs_bitmap_node *bn = NULL;
171         struct list_head *entry = journal->j_bitmap_nodes.next;
172
173         journal->j_used_bitmap_nodes++;
174       repeat:
175
176         if (entry != &journal->j_bitmap_nodes) {
177                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
178                 list_del(entry);
179                 memset(bn->data, 0, sb->s_blocksize);
180                 journal->j_free_bitmap_nodes--;
181                 return bn;
182         }
183         bn = allocate_bitmap_node(sb);
184         if (!bn) {
185                 yield();
186                 goto repeat;
187         }
188         return bn;
189 }
190 static inline void free_bitmap_node(struct super_block *sb,
191                                     struct reiserfs_bitmap_node *bn)
192 {
193         struct reiserfs_journal *journal = SB_JOURNAL(sb);
194         journal->j_used_bitmap_nodes--;
195         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
196                 kfree(bn->data);
197                 kfree(bn);
198         } else {
199                 list_add(&bn->list, &journal->j_bitmap_nodes);
200                 journal->j_free_bitmap_nodes++;
201         }
202 }
203
204 static void allocate_bitmap_nodes(struct super_block *sb)
205 {
206         int i;
207         struct reiserfs_journal *journal = SB_JOURNAL(sb);
208         struct reiserfs_bitmap_node *bn = NULL;
209         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
210                 bn = allocate_bitmap_node(sb);
211                 if (bn) {
212                         list_add(&bn->list, &journal->j_bitmap_nodes);
213                         journal->j_free_bitmap_nodes++;
214                 } else {
215                         break;  /* this is ok, we'll try again when more are needed */
216                 }
217         }
218 }
219
220 static int set_bit_in_list_bitmap(struct super_block *sb,
221                                   b_blocknr_t block,
222                                   struct reiserfs_list_bitmap *jb)
223 {
224         unsigned int bmap_nr = block / (sb->s_blocksize << 3);
225         unsigned int bit_nr = block % (sb->s_blocksize << 3);
226
227         if (!jb->bitmaps[bmap_nr]) {
228                 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
229         }
230         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
231         return 0;
232 }
233
234 static void cleanup_bitmap_list(struct super_block *sb,
235                                 struct reiserfs_list_bitmap *jb)
236 {
237         int i;
238         if (jb->bitmaps == NULL)
239                 return;
240
241         for (i = 0; i < reiserfs_bmap_count(sb); i++) {
242                 if (jb->bitmaps[i]) {
243                         free_bitmap_node(sb, jb->bitmaps[i]);
244                         jb->bitmaps[i] = NULL;
245                 }
246         }
247 }
248
249 /*
250 ** only call this on FS unmount.
251 */
252 static int free_list_bitmaps(struct super_block *sb,
253                              struct reiserfs_list_bitmap *jb_array)
254 {
255         int i;
256         struct reiserfs_list_bitmap *jb;
257         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
258                 jb = jb_array + i;
259                 jb->journal_list = NULL;
260                 cleanup_bitmap_list(sb, jb);
261                 vfree(jb->bitmaps);
262                 jb->bitmaps = NULL;
263         }
264         return 0;
265 }
266
267 static int free_bitmap_nodes(struct super_block *sb)
268 {
269         struct reiserfs_journal *journal = SB_JOURNAL(sb);
270         struct list_head *next = journal->j_bitmap_nodes.next;
271         struct reiserfs_bitmap_node *bn;
272
273         while (next != &journal->j_bitmap_nodes) {
274                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
275                 list_del(next);
276                 kfree(bn->data);
277                 kfree(bn);
278                 next = journal->j_bitmap_nodes.next;
279                 journal->j_free_bitmap_nodes--;
280         }
281
282         return 0;
283 }
284
285 /*
286 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
287 ** jb_array is the array to be filled in.
288 */
289 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
290                                    struct reiserfs_list_bitmap *jb_array,
291                                    unsigned int bmap_nr)
292 {
293         int i;
294         int failed = 0;
295         struct reiserfs_list_bitmap *jb;
296         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
297
298         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
299                 jb = jb_array + i;
300                 jb->journal_list = NULL;
301                 jb->bitmaps = vmalloc(mem);
302                 if (!jb->bitmaps) {
303                         reiserfs_warning(sb, "clm-2000", "unable to "
304                                          "allocate bitmaps for journal lists");
305                         failed = 1;
306                         break;
307                 }
308                 memset(jb->bitmaps, 0, mem);
309         }
310         if (failed) {
311                 free_list_bitmaps(sb, jb_array);
312                 return -1;
313         }
314         return 0;
315 }
316
317 /*
318 ** find an available list bitmap.  If you can't find one, flush a commit list
319 ** and try again
320 */
321 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
322                                                     struct reiserfs_journal_list
323                                                     *jl)
324 {
325         int i, j;
326         struct reiserfs_journal *journal = SB_JOURNAL(sb);
327         struct reiserfs_list_bitmap *jb = NULL;
328
329         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
330                 i = journal->j_list_bitmap_index;
331                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
332                 jb = journal->j_list_bitmap + i;
333                 if (journal->j_list_bitmap[i].journal_list) {
334                         flush_commit_list(sb,
335                                           journal->j_list_bitmap[i].
336                                           journal_list, 1);
337                         if (!journal->j_list_bitmap[i].journal_list) {
338                                 break;
339                         }
340                 } else {
341                         break;
342                 }
343         }
344         if (jb->journal_list) { /* double check to make sure if flushed correctly */
345                 return NULL;
346         }
347         jb->journal_list = jl;
348         return jb;
349 }
350
351 /*
352 ** allocates a new chunk of X nodes, and links them all together as a list.
353 ** Uses the cnode->next and cnode->prev pointers
354 ** returns NULL on failure
355 */
356 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
357 {
358         struct reiserfs_journal_cnode *head;
359         int i;
360         if (num_cnodes <= 0) {
361                 return NULL;
362         }
363         head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
364         if (!head) {
365                 return NULL;
366         }
367         memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode));
368         head[0].prev = NULL;
369         head[0].next = head + 1;
370         for (i = 1; i < num_cnodes; i++) {
371                 head[i].prev = head + (i - 1);
372                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
373         }
374         head[num_cnodes - 1].next = NULL;
375         return head;
376 }
377
378 /*
379 ** pulls a cnode off the free list, or returns NULL on failure
380 */
381 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
382 {
383         struct reiserfs_journal_cnode *cn;
384         struct reiserfs_journal *journal = SB_JOURNAL(sb);
385
386         reiserfs_check_lock_depth(sb, "get_cnode");
387
388         if (journal->j_cnode_free <= 0) {
389                 return NULL;
390         }
391         journal->j_cnode_used++;
392         journal->j_cnode_free--;
393         cn = journal->j_cnode_free_list;
394         if (!cn) {
395                 return cn;
396         }
397         if (cn->next) {
398                 cn->next->prev = NULL;
399         }
400         journal->j_cnode_free_list = cn->next;
401         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
402         return cn;
403 }
404
405 /*
406 ** returns a cnode to the free list
407 */
408 static void free_cnode(struct super_block *sb,
409                        struct reiserfs_journal_cnode *cn)
410 {
411         struct reiserfs_journal *journal = SB_JOURNAL(sb);
412
413         reiserfs_check_lock_depth(sb, "free_cnode");
414
415         journal->j_cnode_used--;
416         journal->j_cnode_free++;
417         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
418         cn->next = journal->j_cnode_free_list;
419         if (journal->j_cnode_free_list) {
420                 journal->j_cnode_free_list->prev = cn;
421         }
422         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
423         journal->j_cnode_free_list = cn;
424 }
425
426 static void clear_prepared_bits(struct buffer_head *bh)
427 {
428         clear_buffer_journal_prepared(bh);
429         clear_buffer_journal_restore_dirty(bh);
430 }
431
432 /* return a cnode with same dev, block number and size in table, or null if not found */
433 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
434                                                                   super_block
435                                                                   *sb,
436                                                                   struct
437                                                                   reiserfs_journal_cnode
438                                                                   **table,
439                                                                   long bl)
440 {
441         struct reiserfs_journal_cnode *cn;
442         cn = journal_hash(table, sb, bl);
443         while (cn) {
444                 if (cn->blocknr == bl && cn->sb == sb)
445                         return cn;
446                 cn = cn->hnext;
447         }
448         return (struct reiserfs_journal_cnode *)0;
449 }
450
451 /*
452 ** this actually means 'can this block be reallocated yet?'.  If you set search_all, a block can only be allocated
453 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
454 ** being overwritten by a replay after crashing.
455 **
456 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction.  Since deleting
457 ** a block removes it from the current transaction, this case should never happen.  If you don't set search_all, make
458 ** sure you never write the block without logging it.
459 **
460 ** next_zero_bit is a suggestion about the next block to try for find_forward.
461 ** when bl is rejected because it is set in a journal list bitmap, we search
462 ** for the next zero bit in the bitmap that rejected bl.  Then, we return that
463 ** through next_zero_bit for find_forward to try.
464 **
465 ** Just because we return something in next_zero_bit does not mean we won't
466 ** reject it on the next call to reiserfs_in_journal
467 **
468 */
469 int reiserfs_in_journal(struct super_block *sb,
470                         unsigned int bmap_nr, int bit_nr, int search_all,
471                         b_blocknr_t * next_zero_bit)
472 {
473         struct reiserfs_journal *journal = SB_JOURNAL(sb);
474         struct reiserfs_journal_cnode *cn;
475         struct reiserfs_list_bitmap *jb;
476         int i;
477         unsigned long bl;
478
479         *next_zero_bit = 0;     /* always start this at zero. */
480
481         PROC_INFO_INC(sb, journal.in_journal);
482         /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
483          ** if we crash before the transaction that freed it commits,  this transaction won't
484          ** have committed either, and the block will never be written
485          */
486         if (search_all) {
487                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
488                         PROC_INFO_INC(sb, journal.in_journal_bitmap);
489                         jb = journal->j_list_bitmap + i;
490                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
491                             test_bit(bit_nr,
492                                      (unsigned long *)jb->bitmaps[bmap_nr]->
493                                      data)) {
494                                 *next_zero_bit =
495                                     find_next_zero_bit((unsigned long *)
496                                                        (jb->bitmaps[bmap_nr]->
497                                                         data),
498                                                        sb->s_blocksize << 3,
499                                                        bit_nr + 1);
500                                 return 1;
501                         }
502                 }
503         }
504
505         bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
506         /* is it in any old transactions? */
507         if (search_all
508             && (cn =
509                 get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
510                 return 1;
511         }
512
513         /* is it in the current transaction.  This should never happen */
514         if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
515                 BUG();
516                 return 1;
517         }
518
519         PROC_INFO_INC(sb, journal.in_journal_reusable);
520         /* safe for reuse */
521         return 0;
522 }
523
524 /* insert cn into table
525 */
526 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
527                                        struct reiserfs_journal_cnode *cn)
528 {
529         struct reiserfs_journal_cnode *cn_orig;
530
531         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
532         cn->hnext = cn_orig;
533         cn->hprev = NULL;
534         if (cn_orig) {
535                 cn_orig->hprev = cn;
536         }
537         journal_hash(table, cn->sb, cn->blocknr) = cn;
538 }
539
540 /*
541  * Several mutexes depend on the write lock.
542  * However sometimes we want to relax the write lock while we hold
543  * these mutexes, according to the release/reacquire on schedule()
544  * properties of the Bkl that were used.
545  * Reiserfs performances and locking were based on this scheme.
546  * Now that the write lock is a mutex and not the bkl anymore, doing so
547  * may result in a deadlock:
548  *
549  * A acquire write_lock
550  * A acquire j_commit_mutex
551  * A release write_lock and wait for something
552  * B acquire write_lock
553  * B can't acquire j_commit_mutex and sleep
554  * A can't acquire write lock anymore
555  * deadlock
556  *
557  * What we do here is avoiding such deadlock by playing the same game
558  * than the Bkl: if we can't acquire a mutex that depends on the write lock,
559  * we release the write lock, wait a bit and then retry.
560  *
561  * The mutexes concerned by this hack are:
562  * - The commit mutex of a journal list
563  * - The flush mutex
564  * - The journal lock
565  */
566 static inline void reiserfs_mutex_lock_safe(struct mutex *m,
567                                struct super_block *s)
568 {
569         while (!mutex_trylock(m)) {
570                 reiserfs_write_unlock(s);
571                 schedule();
572                 reiserfs_write_lock(s);
573         }
574 }
575
576 /* lock the current transaction */
577 static inline void lock_journal(struct super_block *sb)
578 {
579         PROC_INFO_INC(sb, journal.lock_journal);
580
581         reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
582 }
583
584 /* unlock the current transaction */
585 static inline void unlock_journal(struct super_block *sb)
586 {
587         mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
588 }
589
590 static inline void get_journal_list(struct reiserfs_journal_list *jl)
591 {
592         jl->j_refcount++;
593 }
594
595 static inline void put_journal_list(struct super_block *s,
596                                     struct reiserfs_journal_list *jl)
597 {
598         if (jl->j_refcount < 1) {
599                 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
600                                jl->j_trans_id, jl->j_refcount);
601         }
602         if (--jl->j_refcount == 0)
603                 kfree(jl);
604 }
605
606 /*
607 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
608 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
609 ** transaction.
610 */
611 static void cleanup_freed_for_journal_list(struct super_block *sb,
612                                            struct reiserfs_journal_list *jl)
613 {
614
615         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
616         if (jb) {
617                 cleanup_bitmap_list(sb, jb);
618         }
619         jl->j_list_bitmap->journal_list = NULL;
620         jl->j_list_bitmap = NULL;
621 }
622
623 static int journal_list_still_alive(struct super_block *s,
624                                     unsigned int trans_id)
625 {
626         struct reiserfs_journal *journal = SB_JOURNAL(s);
627         struct list_head *entry = &journal->j_journal_list;
628         struct reiserfs_journal_list *jl;
629
630         if (!list_empty(entry)) {
631                 jl = JOURNAL_LIST_ENTRY(entry->next);
632                 if (jl->j_trans_id <= trans_id) {
633                         return 1;
634                 }
635         }
636         return 0;
637 }
638
639 /*
640  * If page->mapping was null, we failed to truncate this page for
641  * some reason.  Most likely because it was truncated after being
642  * logged via data=journal.
643  *
644  * This does a check to see if the buffer belongs to one of these
645  * lost pages before doing the final put_bh.  If page->mapping was
646  * null, it tries to free buffers on the page, which should make the
647  * final page_cache_release drop the page from the lru.
648  */
649 static void release_buffer_page(struct buffer_head *bh)
650 {
651         struct page *page = bh->b_page;
652         if (!page->mapping && trylock_page(page)) {
653                 page_cache_get(page);
654                 put_bh(bh);
655                 if (!page->mapping)
656                         try_to_free_buffers(page);
657                 unlock_page(page);
658                 page_cache_release(page);
659         } else {
660                 put_bh(bh);
661         }
662 }
663
664 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
665 {
666         char b[BDEVNAME_SIZE];
667
668         if (buffer_journaled(bh)) {
669                 reiserfs_warning(NULL, "clm-2084",
670                                  "pinned buffer %lu:%s sent to disk",
671                                  bh->b_blocknr, bdevname(bh->b_bdev, b));
672         }
673         if (uptodate)
674                 set_buffer_uptodate(bh);
675         else
676                 clear_buffer_uptodate(bh);
677
678         unlock_buffer(bh);
679         release_buffer_page(bh);
680 }
681
682 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
683 {
684         if (uptodate)
685                 set_buffer_uptodate(bh);
686         else
687                 clear_buffer_uptodate(bh);
688         unlock_buffer(bh);
689         put_bh(bh);
690 }
691
692 static void submit_logged_buffer(struct buffer_head *bh)
693 {
694         get_bh(bh);
695         bh->b_end_io = reiserfs_end_buffer_io_sync;
696         clear_buffer_journal_new(bh);
697         clear_buffer_dirty(bh);
698         if (!test_clear_buffer_journal_test(bh))
699                 BUG();
700         if (!buffer_uptodate(bh))
701                 BUG();
702         submit_bh(WRITE, bh);
703 }
704
705 static void submit_ordered_buffer(struct buffer_head *bh)
706 {
707         get_bh(bh);
708         bh->b_end_io = reiserfs_end_ordered_io;
709         clear_buffer_dirty(bh);
710         if (!buffer_uptodate(bh))
711                 BUG();
712         submit_bh(WRITE, bh);
713 }
714
715 static int submit_barrier_buffer(struct buffer_head *bh)
716 {
717         get_bh(bh);
718         bh->b_end_io = reiserfs_end_ordered_io;
719         clear_buffer_dirty(bh);
720         if (!buffer_uptodate(bh))
721                 BUG();
722         return submit_bh(WRITE_BARRIER, bh);
723 }
724
725 static void check_barrier_completion(struct super_block *s,
726                                      struct buffer_head *bh)
727 {
728         if (buffer_eopnotsupp(bh)) {
729                 clear_buffer_eopnotsupp(bh);
730                 disable_barrier(s);
731                 set_buffer_uptodate(bh);
732                 set_buffer_dirty(bh);
733                 reiserfs_write_unlock(s);
734                 sync_dirty_buffer(bh);
735                 reiserfs_write_lock(s);
736         }
737 }
738
739 #define CHUNK_SIZE 32
740 struct buffer_chunk {
741         struct buffer_head *bh[CHUNK_SIZE];
742         int nr;
743 };
744
745 static void write_chunk(struct buffer_chunk *chunk)
746 {
747         int i;
748         get_fs_excl();
749         for (i = 0; i < chunk->nr; i++) {
750                 submit_logged_buffer(chunk->bh[i]);
751         }
752         chunk->nr = 0;
753         put_fs_excl();
754 }
755
756 static void write_ordered_chunk(struct buffer_chunk *chunk)
757 {
758         int i;
759         get_fs_excl();
760         for (i = 0; i < chunk->nr; i++) {
761                 submit_ordered_buffer(chunk->bh[i]);
762         }
763         chunk->nr = 0;
764         put_fs_excl();
765 }
766
767 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
768                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
769 {
770         int ret = 0;
771         BUG_ON(chunk->nr >= CHUNK_SIZE);
772         chunk->bh[chunk->nr++] = bh;
773         if (chunk->nr >= CHUNK_SIZE) {
774                 ret = 1;
775                 if (lock)
776                         spin_unlock(lock);
777                 fn(chunk);
778                 if (lock)
779                         spin_lock(lock);
780         }
781         return ret;
782 }
783
784 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
785 static struct reiserfs_jh *alloc_jh(void)
786 {
787         struct reiserfs_jh *jh;
788         while (1) {
789                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
790                 if (jh) {
791                         atomic_inc(&nr_reiserfs_jh);
792                         return jh;
793                 }
794                 yield();
795         }
796 }
797
798 /*
799  * we want to free the jh when the buffer has been written
800  * and waited on
801  */
802 void reiserfs_free_jh(struct buffer_head *bh)
803 {
804         struct reiserfs_jh *jh;
805
806         jh = bh->b_private;
807         if (jh) {
808                 bh->b_private = NULL;
809                 jh->bh = NULL;
810                 list_del_init(&jh->list);
811                 kfree(jh);
812                 if (atomic_read(&nr_reiserfs_jh) <= 0)
813                         BUG();
814                 atomic_dec(&nr_reiserfs_jh);
815                 put_bh(bh);
816         }
817 }
818
819 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
820                            int tail)
821 {
822         struct reiserfs_jh *jh;
823
824         if (bh->b_private) {
825                 spin_lock(&j->j_dirty_buffers_lock);
826                 if (!bh->b_private) {
827                         spin_unlock(&j->j_dirty_buffers_lock);
828                         goto no_jh;
829                 }
830                 jh = bh->b_private;
831                 list_del_init(&jh->list);
832         } else {
833               no_jh:
834                 get_bh(bh);
835                 jh = alloc_jh();
836                 spin_lock(&j->j_dirty_buffers_lock);
837                 /* buffer must be locked for __add_jh, should be able to have
838                  * two adds at the same time
839                  */
840                 BUG_ON(bh->b_private);
841                 jh->bh = bh;
842                 bh->b_private = jh;
843         }
844         jh->jl = j->j_current_jl;
845         if (tail)
846                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
847         else {
848                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
849         }
850         spin_unlock(&j->j_dirty_buffers_lock);
851         return 0;
852 }
853
854 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
855 {
856         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
857 }
858 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
859 {
860         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
861 }
862
863 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
864 static int write_ordered_buffers(spinlock_t * lock,
865                                  struct reiserfs_journal *j,
866                                  struct reiserfs_journal_list *jl,
867                                  struct list_head *list)
868 {
869         struct buffer_head *bh;
870         struct reiserfs_jh *jh;
871         int ret = j->j_errno;
872         struct buffer_chunk chunk;
873         struct list_head tmp;
874         INIT_LIST_HEAD(&tmp);
875
876         chunk.nr = 0;
877         spin_lock(lock);
878         while (!list_empty(list)) {
879                 jh = JH_ENTRY(list->next);
880                 bh = jh->bh;
881                 get_bh(bh);
882                 if (!trylock_buffer(bh)) {
883                         if (!buffer_dirty(bh)) {
884                                 list_move(&jh->list, &tmp);
885                                 goto loop_next;
886                         }
887                         spin_unlock(lock);
888                         if (chunk.nr)
889                                 write_ordered_chunk(&chunk);
890                         wait_on_buffer(bh);
891                         cond_resched();
892                         spin_lock(lock);
893                         goto loop_next;
894                 }
895                 /* in theory, dirty non-uptodate buffers should never get here,
896                  * but the upper layer io error paths still have a few quirks.
897                  * Handle them here as gracefully as we can
898                  */
899                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
900                         clear_buffer_dirty(bh);
901                         ret = -EIO;
902                 }
903                 if (buffer_dirty(bh)) {
904                         list_move(&jh->list, &tmp);
905                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
906                 } else {
907                         reiserfs_free_jh(bh);
908                         unlock_buffer(bh);
909                 }
910               loop_next:
911                 put_bh(bh);
912                 cond_resched_lock(lock);
913         }
914         if (chunk.nr) {
915                 spin_unlock(lock);
916                 write_ordered_chunk(&chunk);
917                 spin_lock(lock);
918         }
919         while (!list_empty(&tmp)) {
920                 jh = JH_ENTRY(tmp.prev);
921                 bh = jh->bh;
922                 get_bh(bh);
923                 reiserfs_free_jh(bh);
924
925                 if (buffer_locked(bh)) {
926                         spin_unlock(lock);
927                         wait_on_buffer(bh);
928                         spin_lock(lock);
929                 }
930                 if (!buffer_uptodate(bh)) {
931                         ret = -EIO;
932                 }
933                 /* ugly interaction with invalidatepage here.
934                  * reiserfs_invalidate_page will pin any buffer that has a valid
935                  * journal head from an older transaction.  If someone else sets
936                  * our buffer dirty after we write it in the first loop, and
937                  * then someone truncates the page away, nobody will ever write
938                  * the buffer. We're safe if we write the page one last time
939                  * after freeing the journal header.
940                  */
941                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
942                         spin_unlock(lock);
943                         ll_rw_block(WRITE, 1, &bh);
944                         spin_lock(lock);
945                 }
946                 put_bh(bh);
947                 cond_resched_lock(lock);
948         }
949         spin_unlock(lock);
950         return ret;
951 }
952
953 static int flush_older_commits(struct super_block *s,
954                                struct reiserfs_journal_list *jl)
955 {
956         struct reiserfs_journal *journal = SB_JOURNAL(s);
957         struct reiserfs_journal_list *other_jl;
958         struct reiserfs_journal_list *first_jl;
959         struct list_head *entry;
960         unsigned int trans_id = jl->j_trans_id;
961         unsigned int other_trans_id;
962         unsigned int first_trans_id;
963
964       find_first:
965         /*
966          * first we walk backwards to find the oldest uncommitted transation
967          */
968         first_jl = jl;
969         entry = jl->j_list.prev;
970         while (1) {
971                 other_jl = JOURNAL_LIST_ENTRY(entry);
972                 if (entry == &journal->j_journal_list ||
973                     atomic_read(&other_jl->j_older_commits_done))
974                         break;
975
976                 first_jl = other_jl;
977                 entry = other_jl->j_list.prev;
978         }
979
980         /* if we didn't find any older uncommitted transactions, return now */
981         if (first_jl == jl) {
982                 return 0;
983         }
984
985         first_trans_id = first_jl->j_trans_id;
986
987         entry = &first_jl->j_list;
988         while (1) {
989                 other_jl = JOURNAL_LIST_ENTRY(entry);
990                 other_trans_id = other_jl->j_trans_id;
991
992                 if (other_trans_id < trans_id) {
993                         if (atomic_read(&other_jl->j_commit_left) != 0) {
994                                 flush_commit_list(s, other_jl, 0);
995
996                                 /* list we were called with is gone, return */
997                                 if (!journal_list_still_alive(s, trans_id))
998                                         return 1;
999
1000                                 /* the one we just flushed is gone, this means all
1001                                  * older lists are also gone, so first_jl is no longer
1002                                  * valid either.  Go back to the beginning.
1003                                  */
1004                                 if (!journal_list_still_alive
1005                                     (s, other_trans_id)) {
1006                                         goto find_first;
1007                                 }
1008                         }
1009                         entry = entry->next;
1010                         if (entry == &journal->j_journal_list)
1011                                 return 0;
1012                 } else {
1013                         return 0;
1014                 }
1015         }
1016         return 0;
1017 }
1018
1019 static int reiserfs_async_progress_wait(struct super_block *s)
1020 {
1021         DEFINE_WAIT(wait);
1022         struct reiserfs_journal *j = SB_JOURNAL(s);
1023
1024         if (atomic_read(&j->j_async_throttle)) {
1025                 reiserfs_write_unlock(s);
1026                 congestion_wait(BLK_RW_ASYNC, HZ / 10);
1027                 reiserfs_write_lock(s);
1028         }
1029
1030         return 0;
1031 }
1032
1033 /*
1034 ** if this journal list still has commit blocks unflushed, send them to disk.
1035 **
1036 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
1037 ** Before the commit block can by written, every other log block must be safely on disk
1038 **
1039 */
1040 static int flush_commit_list(struct super_block *s,
1041                              struct reiserfs_journal_list *jl, int flushall)
1042 {
1043         int i;
1044         b_blocknr_t bn;
1045         struct buffer_head *tbh = NULL;
1046         unsigned int trans_id = jl->j_trans_id;
1047         struct reiserfs_journal *journal = SB_JOURNAL(s);
1048         int barrier = 0;
1049         int retval = 0;
1050         int write_len;
1051
1052         reiserfs_check_lock_depth(s, "flush_commit_list");
1053
1054         if (atomic_read(&jl->j_older_commits_done)) {
1055                 return 0;
1056         }
1057
1058         get_fs_excl();
1059
1060         /* before we can put our commit blocks on disk, we have to make sure everyone older than
1061          ** us is on disk too
1062          */
1063         BUG_ON(jl->j_len <= 0);
1064         BUG_ON(trans_id == journal->j_trans_id);
1065
1066         get_journal_list(jl);
1067         if (flushall) {
1068                 if (flush_older_commits(s, jl) == 1) {
1069                         /* list disappeared during flush_older_commits.  return */
1070                         goto put_jl;
1071                 }
1072         }
1073
1074         /* make sure nobody is trying to flush this one at the same time */
1075         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1076
1077         if (!journal_list_still_alive(s, trans_id)) {
1078                 mutex_unlock(&jl->j_commit_mutex);
1079                 goto put_jl;
1080         }
1081         BUG_ON(jl->j_trans_id == 0);
1082
1083         /* this commit is done, exit */
1084         if (atomic_read(&(jl->j_commit_left)) <= 0) {
1085                 if (flushall) {
1086                         atomic_set(&(jl->j_older_commits_done), 1);
1087                 }
1088                 mutex_unlock(&jl->j_commit_mutex);
1089                 goto put_jl;
1090         }
1091
1092         if (!list_empty(&jl->j_bh_list)) {
1093                 int ret;
1094
1095                 /*
1096                  * We might sleep in numerous places inside
1097                  * write_ordered_buffers. Relax the write lock.
1098                  */
1099                 reiserfs_write_unlock(s);
1100                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1101                                             journal, jl, &jl->j_bh_list);
1102                 if (ret < 0 && retval == 0)
1103                         retval = ret;
1104                 reiserfs_write_lock(s);
1105         }
1106         BUG_ON(!list_empty(&jl->j_bh_list));
1107         /*
1108          * for the description block and all the log blocks, submit any buffers
1109          * that haven't already reached the disk.  Try to write at least 256
1110          * log blocks. later on, we will only wait on blocks that correspond
1111          * to this transaction, but while we're unplugging we might as well
1112          * get a chunk of data on there.
1113          */
1114         atomic_inc(&journal->j_async_throttle);
1115         write_len = jl->j_len + 1;
1116         if (write_len < 256)
1117                 write_len = 256;
1118         for (i = 0 ; i < write_len ; i++) {
1119                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1120                     SB_ONDISK_JOURNAL_SIZE(s);
1121                 tbh = journal_find_get_block(s, bn);
1122                 if (tbh) {
1123                         if (buffer_dirty(tbh))
1124                             ll_rw_block(WRITE, 1, &tbh) ;
1125                         put_bh(tbh) ;
1126                 }
1127         }
1128         atomic_dec(&journal->j_async_throttle);
1129
1130         /* We're skipping the commit if there's an error */
1131         if (retval || reiserfs_is_journal_aborted(journal))
1132                 barrier = 0;
1133
1134         /* wait on everything written so far before writing the commit
1135          * if we are in barrier mode, send the commit down now
1136          */
1137         barrier = reiserfs_barrier_flush(s);
1138         if (barrier) {
1139                 int ret;
1140                 lock_buffer(jl->j_commit_bh);
1141                 ret = submit_barrier_buffer(jl->j_commit_bh);
1142                 if (ret == -EOPNOTSUPP) {
1143                         set_buffer_uptodate(jl->j_commit_bh);
1144                         disable_barrier(s);
1145                         barrier = 0;
1146                 }
1147         }
1148         for (i = 0; i < (jl->j_len + 1); i++) {
1149                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1150                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1151                 tbh = journal_find_get_block(s, bn);
1152
1153                 reiserfs_write_unlock(s);
1154                 wait_on_buffer(tbh);
1155                 reiserfs_write_lock(s);
1156                 // since we're using ll_rw_blk above, it might have skipped over
1157                 // a locked buffer.  Double check here
1158                 //
1159                 /* redundant, sync_dirty_buffer() checks */
1160                 if (buffer_dirty(tbh)) {
1161                         reiserfs_write_unlock(s);
1162                         sync_dirty_buffer(tbh);
1163                         reiserfs_write_lock(s);
1164                 }
1165                 if (unlikely(!buffer_uptodate(tbh))) {
1166 #ifdef CONFIG_REISERFS_CHECK
1167                         reiserfs_warning(s, "journal-601",
1168                                          "buffer write failed");
1169 #endif
1170                         retval = -EIO;
1171                 }
1172                 put_bh(tbh);    /* once for journal_find_get_block */
1173                 put_bh(tbh);    /* once due to original getblk in do_journal_end */
1174                 atomic_dec(&(jl->j_commit_left));
1175         }
1176
1177         BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1178
1179         if (!barrier) {
1180                 /* If there was a write error in the journal - we can't commit
1181                  * this transaction - it will be invalid and, if successful,
1182                  * will just end up propagating the write error out to
1183                  * the file system. */
1184                 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1185                         if (buffer_dirty(jl->j_commit_bh))
1186                                 BUG();
1187                         mark_buffer_dirty(jl->j_commit_bh) ;
1188                         reiserfs_write_unlock(s);
1189                         sync_dirty_buffer(jl->j_commit_bh) ;
1190                         reiserfs_write_lock(s);
1191                 }
1192         } else {
1193                 reiserfs_write_unlock(s);
1194                 wait_on_buffer(jl->j_commit_bh);
1195                 reiserfs_write_lock(s);
1196         }
1197
1198         check_barrier_completion(s, jl->j_commit_bh);
1199
1200         /* If there was a write error in the journal - we can't commit this
1201          * transaction - it will be invalid and, if successful, will just end
1202          * up propagating the write error out to the filesystem. */
1203         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1204 #ifdef CONFIG_REISERFS_CHECK
1205                 reiserfs_warning(s, "journal-615", "buffer write failed");
1206 #endif
1207                 retval = -EIO;
1208         }
1209         bforget(jl->j_commit_bh);
1210         if (journal->j_last_commit_id != 0 &&
1211             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1212                 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1213                                  journal->j_last_commit_id, jl->j_trans_id);
1214         }
1215         journal->j_last_commit_id = jl->j_trans_id;
1216
1217         /* now, every commit block is on the disk.  It is safe to allow blocks freed during this transaction to be reallocated */
1218         cleanup_freed_for_journal_list(s, jl);
1219
1220         retval = retval ? retval : journal->j_errno;
1221
1222         /* mark the metadata dirty */
1223         if (!retval)
1224                 dirty_one_transaction(s, jl);
1225         atomic_dec(&(jl->j_commit_left));
1226
1227         if (flushall) {
1228                 atomic_set(&(jl->j_older_commits_done), 1);
1229         }
1230         mutex_unlock(&jl->j_commit_mutex);
1231       put_jl:
1232         put_journal_list(s, jl);
1233
1234         if (retval)
1235                 reiserfs_abort(s, retval, "Journal write error in %s",
1236                                __func__);
1237         put_fs_excl();
1238         return retval;
1239 }
1240
1241 /*
1242 ** flush_journal_list frequently needs to find a newer transaction for a given block.  This does that, or
1243 ** returns NULL if it can't find anything
1244 */
1245 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1246                                                           reiserfs_journal_cnode
1247                                                           *cn)
1248 {
1249         struct super_block *sb = cn->sb;
1250         b_blocknr_t blocknr = cn->blocknr;
1251
1252         cn = cn->hprev;
1253         while (cn) {
1254                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1255                         return cn->jlist;
1256                 }
1257                 cn = cn->hprev;
1258         }
1259         return NULL;
1260 }
1261
1262 static int newer_jl_done(struct reiserfs_journal_cnode *cn)
1263 {
1264         struct super_block *sb = cn->sb;
1265         b_blocknr_t blocknr = cn->blocknr;
1266
1267         cn = cn->hprev;
1268         while (cn) {
1269                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist &&
1270                     atomic_read(&cn->jlist->j_commit_left) != 0)
1271                                     return 0;
1272                 cn = cn->hprev;
1273         }
1274         return 1;
1275 }
1276
1277 static void remove_journal_hash(struct super_block *,
1278                                 struct reiserfs_journal_cnode **,
1279                                 struct reiserfs_journal_list *, unsigned long,
1280                                 int);
1281
1282 /*
1283 ** once all the real blocks have been flushed, it is safe to remove them from the
1284 ** journal list for this transaction.  Aside from freeing the cnode, this also allows the
1285 ** block to be reallocated for data blocks if it had been deleted.
1286 */
1287 static void remove_all_from_journal_list(struct super_block *sb,
1288                                          struct reiserfs_journal_list *jl,
1289                                          int debug)
1290 {
1291         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1292         struct reiserfs_journal_cnode *cn, *last;
1293         cn = jl->j_realblock;
1294
1295         /* which is better, to lock once around the whole loop, or
1296          ** to lock for each call to remove_journal_hash?
1297          */
1298         while (cn) {
1299                 if (cn->blocknr != 0) {
1300                         if (debug) {
1301                                 reiserfs_warning(sb, "reiserfs-2201",
1302                                                  "block %u, bh is %d, state %ld",
1303                                                  cn->blocknr, cn->bh ? 1 : 0,
1304                                                  cn->state);
1305                         }
1306                         cn->state = 0;
1307                         remove_journal_hash(sb, journal->j_list_hash_table,
1308                                             jl, cn->blocknr, 1);
1309                 }
1310                 last = cn;
1311                 cn = cn->next;
1312                 free_cnode(sb, last);
1313         }
1314         jl->j_realblock = NULL;
1315 }
1316
1317 /*
1318 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1319 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1320 ** releasing blocks in this transaction for reuse as data blocks.
1321 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1322 **
1323 */
1324 static int _update_journal_header_block(struct super_block *sb,
1325                                         unsigned long offset,
1326                                         unsigned int trans_id)
1327 {
1328         struct reiserfs_journal_header *jh;
1329         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1330
1331         if (reiserfs_is_journal_aborted(journal))
1332                 return -EIO;
1333
1334         if (trans_id >= journal->j_last_flush_trans_id) {
1335                 if (buffer_locked((journal->j_header_bh))) {
1336                         reiserfs_write_unlock(sb);
1337                         wait_on_buffer((journal->j_header_bh));
1338                         reiserfs_write_lock(sb);
1339                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1340 #ifdef CONFIG_REISERFS_CHECK
1341                                 reiserfs_warning(sb, "journal-699",
1342                                                  "buffer write failed");
1343 #endif
1344                                 return -EIO;
1345                         }
1346                 }
1347                 journal->j_last_flush_trans_id = trans_id;
1348                 journal->j_first_unflushed_offset = offset;
1349                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1350                                                         b_data);
1351                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1352                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1353                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1354
1355                 if (reiserfs_barrier_flush(sb)) {
1356                         int ret;
1357                         lock_buffer(journal->j_header_bh);
1358                         ret = submit_barrier_buffer(journal->j_header_bh);
1359                         if (ret == -EOPNOTSUPP) {
1360                                 set_buffer_uptodate(journal->j_header_bh);
1361                                 disable_barrier(sb);
1362                                 goto sync;
1363                         }
1364                         reiserfs_write_unlock(sb);
1365                         wait_on_buffer(journal->j_header_bh);
1366                         reiserfs_write_lock(sb);
1367                         check_barrier_completion(sb, journal->j_header_bh);
1368                 } else {
1369                       sync:
1370                         set_buffer_dirty(journal->j_header_bh);
1371                         reiserfs_write_unlock(sb);
1372                         sync_dirty_buffer(journal->j_header_bh);
1373                         reiserfs_write_lock(sb);
1374                 }
1375                 if (!buffer_uptodate(journal->j_header_bh)) {
1376                         reiserfs_warning(sb, "journal-837",
1377                                          "IO error during journal replay");
1378                         return -EIO;
1379                 }
1380         }
1381         return 0;
1382 }
1383
1384 static int update_journal_header_block(struct super_block *sb,
1385                                        unsigned long offset,
1386                                        unsigned int trans_id)
1387 {
1388         return _update_journal_header_block(sb, offset, trans_id);
1389 }
1390
1391 /*
1392 ** flush any and all journal lists older than you are
1393 ** can only be called from flush_journal_list
1394 */
1395 static int flush_older_journal_lists(struct super_block *sb,
1396                                      struct reiserfs_journal_list *jl)
1397 {
1398         struct list_head *entry;
1399         struct reiserfs_journal_list *other_jl;
1400         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1401         unsigned int trans_id = jl->j_trans_id;
1402
1403         /* we know we are the only ones flushing things, no extra race
1404          * protection is required.
1405          */
1406       restart:
1407         entry = journal->j_journal_list.next;
1408         /* Did we wrap? */
1409         if (entry == &journal->j_journal_list)
1410                 return 0;
1411         other_jl = JOURNAL_LIST_ENTRY(entry);
1412         if (other_jl->j_trans_id < trans_id) {
1413                 BUG_ON(other_jl->j_refcount <= 0);
1414                 /* do not flush all */
1415                 flush_journal_list(sb, other_jl, 0);
1416
1417                 /* other_jl is now deleted from the list */
1418                 goto restart;
1419         }
1420         return 0;
1421 }
1422
1423 static void del_from_work_list(struct super_block *s,
1424                                struct reiserfs_journal_list *jl)
1425 {
1426         struct reiserfs_journal *journal = SB_JOURNAL(s);
1427         if (!list_empty(&jl->j_working_list)) {
1428                 list_del_init(&jl->j_working_list);
1429                 journal->j_num_work_lists--;
1430         }
1431 }
1432
1433 /* flush a journal list, both commit and real blocks
1434 **
1435 ** always set flushall to 1, unless you are calling from inside
1436 ** flush_journal_list
1437 **
1438 ** IMPORTANT.  This can only be called while there are no journal writers,
1439 ** and the journal is locked.  That means it can only be called from
1440 ** do_journal_end, or by journal_release
1441 */
1442 static int flush_journal_list(struct super_block *s,
1443                               struct reiserfs_journal_list *jl, int flushall)
1444 {
1445         struct reiserfs_journal_list *pjl;
1446         struct reiserfs_journal_cnode *cn, *last;
1447         int count;
1448         int was_jwait = 0;
1449         int was_dirty = 0;
1450         struct buffer_head *saved_bh;
1451         unsigned long j_len_saved = jl->j_len;
1452         struct reiserfs_journal *journal = SB_JOURNAL(s);
1453         int err = 0;
1454
1455         BUG_ON(j_len_saved <= 0);
1456
1457         if (atomic_read(&journal->j_wcount) != 0) {
1458                 reiserfs_warning(s, "clm-2048", "called with wcount %d",
1459                                  atomic_read(&journal->j_wcount));
1460         }
1461         BUG_ON(jl->j_trans_id == 0);
1462
1463         /* if flushall == 0, the lock is already held */
1464         if (flushall) {
1465                 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1466         } else if (mutex_trylock(&journal->j_flush_mutex)) {
1467                 BUG();
1468         }
1469
1470         count = 0;
1471         if (j_len_saved > journal->j_trans_max) {
1472                 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1473                                j_len_saved, jl->j_trans_id);
1474                 return 0;
1475         }
1476
1477         get_fs_excl();
1478
1479         /* if all the work is already done, get out of here */
1480         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1481             atomic_read(&(jl->j_commit_left)) <= 0) {
1482                 goto flush_older_and_return;
1483         }
1484
1485         /* start by putting the commit list on disk.  This will also flush
1486          ** the commit lists of any olders transactions
1487          */
1488         flush_commit_list(s, jl, 1);
1489
1490         if (!(jl->j_state & LIST_DIRTY)
1491             && !reiserfs_is_journal_aborted(journal))
1492                 BUG();
1493
1494         /* are we done now? */
1495         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1496             atomic_read(&(jl->j_commit_left)) <= 0) {
1497                 goto flush_older_and_return;
1498         }
1499
1500         /* loop through each cnode, see if we need to write it,
1501          ** or wait on a more recent transaction, or just ignore it
1502          */
1503         if (atomic_read(&(journal->j_wcount)) != 0) {
1504                 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1505                                "wcount is not 0");
1506         }
1507         cn = jl->j_realblock;
1508         while (cn) {
1509                 was_jwait = 0;
1510                 was_dirty = 0;
1511                 saved_bh = NULL;
1512                 /* blocknr of 0 is no longer in the hash, ignore it */
1513                 if (cn->blocknr == 0) {
1514                         goto free_cnode;
1515                 }
1516
1517                 /* This transaction failed commit. Don't write out to the disk */
1518                 if (!(jl->j_state & LIST_DIRTY))
1519                         goto free_cnode;
1520
1521                 pjl = find_newer_jl_for_cn(cn);
1522                 /* the order is important here.  We check pjl to make sure we
1523                  ** don't clear BH_JDirty_wait if we aren't the one writing this
1524                  ** block to disk
1525                  */
1526                 if (!pjl && cn->bh) {
1527                         saved_bh = cn->bh;
1528
1529                         /* we do this to make sure nobody releases the buffer while
1530                          ** we are working with it
1531                          */
1532                         get_bh(saved_bh);
1533
1534                         if (buffer_journal_dirty(saved_bh)) {
1535                                 BUG_ON(!can_dirty(cn));
1536                                 was_jwait = 1;
1537                                 was_dirty = 1;
1538                         } else if (can_dirty(cn)) {
1539                                 /* everything with !pjl && jwait should be writable */
1540                                 BUG();
1541                         }
1542                 }
1543
1544                 /* if someone has this block in a newer transaction, just make
1545                  ** sure they are committed, and don't try writing it to disk
1546                  */
1547                 if (pjl) {
1548                         if (atomic_read(&pjl->j_commit_left))
1549                                 flush_commit_list(s, pjl, 1);
1550                         goto free_cnode;
1551                 }
1552
1553                 /* bh == NULL when the block got to disk on its own, OR,
1554                  ** the block got freed in a future transaction
1555                  */
1556                 if (saved_bh == NULL) {
1557                         goto free_cnode;
1558                 }
1559
1560                 /* this should never happen.  kupdate_one_transaction has this list
1561                  ** locked while it works, so we should never see a buffer here that
1562                  ** is not marked JDirty_wait
1563                  */
1564                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1565                         reiserfs_warning(s, "journal-813",
1566                                          "BAD! buffer %llu %cdirty %cjwait, "
1567                                          "not in a newer tranasction",
1568                                          (unsigned long long)saved_bh->
1569                                          b_blocknr, was_dirty ? ' ' : '!',
1570                                          was_jwait ? ' ' : '!');
1571                 }
1572                 if (was_dirty) {
1573                         /* we inc again because saved_bh gets decremented at free_cnode */
1574                         get_bh(saved_bh);
1575                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1576                         lock_buffer(saved_bh);
1577                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1578                         if (buffer_dirty(saved_bh))
1579                                 submit_logged_buffer(saved_bh);
1580                         else
1581                                 unlock_buffer(saved_bh);
1582                         count++;
1583                 } else {
1584                         reiserfs_warning(s, "clm-2082",
1585                                          "Unable to flush buffer %llu in %s",
1586                                          (unsigned long long)saved_bh->
1587                                          b_blocknr, __func__);
1588                 }
1589               free_cnode:
1590                 last = cn;
1591                 cn = cn->next;
1592                 if (saved_bh) {
1593                         /* we incremented this to keep others from taking the buffer head away */
1594                         put_bh(saved_bh);
1595                         if (atomic_read(&(saved_bh->b_count)) < 0) {
1596                                 reiserfs_warning(s, "journal-945",
1597                                                  "saved_bh->b_count < 0");
1598                         }
1599                 }
1600         }
1601         if (count > 0) {
1602                 cn = jl->j_realblock;
1603                 while (cn) {
1604                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1605                                 if (!cn->bh) {
1606                                         reiserfs_panic(s, "journal-1011",
1607                                                        "cn->bh is NULL");
1608                                 }
1609
1610                                 reiserfs_write_unlock(s);
1611                                 wait_on_buffer(cn->bh);
1612                                 reiserfs_write_lock(s);
1613
1614                                 if (!cn->bh) {
1615                                         reiserfs_panic(s, "journal-1012",
1616                                                        "cn->bh is NULL");
1617                                 }
1618                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1619 #ifdef CONFIG_REISERFS_CHECK
1620                                         reiserfs_warning(s, "journal-949",
1621                                                          "buffer write failed");
1622 #endif
1623                                         err = -EIO;
1624                                 }
1625                                 /* note, we must clear the JDirty_wait bit after the up to date
1626                                  ** check, otherwise we race against our flushpage routine
1627                                  */
1628                                 BUG_ON(!test_clear_buffer_journal_dirty
1629                                        (cn->bh));
1630
1631                                 /* drop one ref for us */
1632                                 put_bh(cn->bh);
1633                                 /* drop one ref for journal_mark_dirty */
1634                                 release_buffer_page(cn->bh);
1635                         }
1636                         cn = cn->next;
1637                 }
1638         }
1639
1640         if (err)
1641                 reiserfs_abort(s, -EIO,
1642                                "Write error while pushing transaction to disk in %s",
1643                                __func__);
1644       flush_older_and_return:
1645
1646         /* before we can update the journal header block, we _must_ flush all
1647          ** real blocks from all older transactions to disk.  This is because
1648          ** once the header block is updated, this transaction will not be
1649          ** replayed after a crash
1650          */
1651         if (flushall) {
1652                 flush_older_journal_lists(s, jl);
1653         }
1654
1655         err = journal->j_errno;
1656         /* before we can remove everything from the hash tables for this
1657          ** transaction, we must make sure it can never be replayed
1658          **
1659          ** since we are only called from do_journal_end, we know for sure there
1660          ** are no allocations going on while we are flushing journal lists.  So,
1661          ** we only need to update the journal header block for the last list
1662          ** being flushed
1663          */
1664         if (!err && flushall) {
1665                 err =
1666                     update_journal_header_block(s,
1667                                                 (jl->j_start + jl->j_len +
1668                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1669                                                 jl->j_trans_id);
1670                 if (err)
1671                         reiserfs_abort(s, -EIO,
1672                                        "Write error while updating journal header in %s",
1673                                        __func__);
1674         }
1675         remove_all_from_journal_list(s, jl, 0);
1676         list_del_init(&jl->j_list);
1677         journal->j_num_lists--;
1678         del_from_work_list(s, jl);
1679
1680         if (journal->j_last_flush_id != 0 &&
1681             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1682                 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1683                                  journal->j_last_flush_id, jl->j_trans_id);
1684         }
1685         journal->j_last_flush_id = jl->j_trans_id;
1686
1687         /* not strictly required since we are freeing the list, but it should
1688          * help find code using dead lists later on
1689          */
1690         jl->j_len = 0;
1691         atomic_set(&(jl->j_nonzerolen), 0);
1692         jl->j_start = 0;
1693         jl->j_realblock = NULL;
1694         jl->j_commit_bh = NULL;
1695         jl->j_trans_id = 0;
1696         jl->j_state = 0;
1697         put_journal_list(s, jl);
1698         if (flushall)
1699                 mutex_unlock(&journal->j_flush_mutex);
1700         put_fs_excl();
1701         return err;
1702 }
1703
1704 static int test_transaction(struct super_block *s,
1705                             struct reiserfs_journal_list *jl)
1706 {
1707         struct reiserfs_journal_cnode *cn;
1708
1709         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0)
1710                 return 1;
1711
1712         cn = jl->j_realblock;
1713         while (cn) {
1714                 /* if the blocknr == 0, this has been cleared from the hash,
1715                  ** skip it
1716                  */
1717                 if (cn->blocknr == 0) {
1718                         goto next;
1719                 }
1720                 if (cn->bh && !newer_jl_done(cn))
1721                         return 0;
1722               next:
1723                 cn = cn->next;
1724                 cond_resched();
1725         }
1726         return 0;
1727 }
1728
1729 static int write_one_transaction(struct super_block *s,
1730                                  struct reiserfs_journal_list *jl,
1731                                  struct buffer_chunk *chunk)
1732 {
1733         struct reiserfs_journal_cnode *cn;
1734         int ret = 0;
1735
1736         jl->j_state |= LIST_TOUCHED;
1737         del_from_work_list(s, jl);
1738         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1739                 return 0;
1740         }
1741
1742         cn = jl->j_realblock;
1743         while (cn) {
1744                 /* if the blocknr == 0, this has been cleared from the hash,
1745                  ** skip it
1746                  */
1747                 if (cn->blocknr == 0) {
1748                         goto next;
1749                 }
1750                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1751                         struct buffer_head *tmp_bh;
1752                         /* we can race against journal_mark_freed when we try
1753                          * to lock_buffer(cn->bh), so we have to inc the buffer
1754                          * count, and recheck things after locking
1755                          */
1756                         tmp_bh = cn->bh;
1757                         get_bh(tmp_bh);
1758                         lock_buffer(tmp_bh);
1759                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1760                                 if (!buffer_journal_dirty(tmp_bh) ||
1761                                     buffer_journal_prepared(tmp_bh))
1762                                         BUG();
1763                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1764                                 ret++;
1765                         } else {
1766                                 /* note, cn->bh might be null now */
1767                                 unlock_buffer(tmp_bh);
1768                         }
1769                         put_bh(tmp_bh);
1770                 }
1771               next:
1772                 cn = cn->next;
1773                 cond_resched();
1774         }
1775         return ret;
1776 }
1777
1778 /* used by flush_commit_list */
1779 static int dirty_one_transaction(struct super_block *s,
1780                                  struct reiserfs_journal_list *jl)
1781 {
1782         struct reiserfs_journal_cnode *cn;
1783         struct reiserfs_journal_list *pjl;
1784         int ret = 0;
1785
1786         jl->j_state |= LIST_DIRTY;
1787         cn = jl->j_realblock;
1788         while (cn) {
1789                 /* look for a more recent transaction that logged this
1790                  ** buffer.  Only the most recent transaction with a buffer in
1791                  ** it is allowed to send that buffer to disk
1792                  */
1793                 pjl = find_newer_jl_for_cn(cn);
1794                 if (!pjl && cn->blocknr && cn->bh
1795                     && buffer_journal_dirty(cn->bh)) {
1796                         BUG_ON(!can_dirty(cn));
1797                         /* if the buffer is prepared, it will either be logged
1798                          * or restored.  If restored, we need to make sure
1799                          * it actually gets marked dirty
1800                          */
1801                         clear_buffer_journal_new(cn->bh);
1802                         if (buffer_journal_prepared(cn->bh)) {
1803                                 set_buffer_journal_restore_dirty(cn->bh);
1804                         } else {
1805                                 set_buffer_journal_test(cn->bh);
1806                                 mark_buffer_dirty(cn->bh);
1807                         }
1808                 }
1809                 cn = cn->next;
1810         }
1811         return ret;
1812 }
1813
1814 static int kupdate_transactions(struct super_block *s,
1815                                 struct reiserfs_journal_list *jl,
1816                                 struct reiserfs_journal_list **next_jl,
1817                                 unsigned int *next_trans_id,
1818                                 int num_blocks, int num_trans)
1819 {
1820         int ret = 0;
1821         int written = 0;
1822         int transactions_flushed = 0;
1823         unsigned int orig_trans_id = jl->j_trans_id;
1824         struct buffer_chunk chunk;
1825         struct list_head *entry;
1826         struct reiserfs_journal *journal = SB_JOURNAL(s);
1827         chunk.nr = 0;
1828
1829         reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1830         if (!journal_list_still_alive(s, orig_trans_id)) {
1831                 goto done;
1832         }
1833
1834         /* we've got j_flush_mutex held, nobody is going to delete any
1835          * of these lists out from underneath us
1836          */
1837         while ((num_trans && transactions_flushed < num_trans) ||
1838                (!num_trans && written < num_blocks)) {
1839
1840                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1841                     atomic_read(&jl->j_commit_left)
1842                     || !(jl->j_state & LIST_DIRTY)) {
1843                         del_from_work_list(s, jl);
1844                         break;
1845                 }
1846                 ret = write_one_transaction(s, jl, &chunk);
1847
1848                 if (ret < 0)
1849                         goto done;
1850                 transactions_flushed++;
1851                 written += ret;
1852                 entry = jl->j_list.next;
1853
1854                 /* did we wrap? */
1855                 if (entry == &journal->j_journal_list) {
1856                         break;
1857                 }
1858                 jl = JOURNAL_LIST_ENTRY(entry);
1859
1860                 /* don't bother with older transactions */
1861                 if (jl->j_trans_id <= orig_trans_id)
1862                         break;
1863         }
1864         if (chunk.nr) {
1865                 write_chunk(&chunk);
1866         }
1867
1868       done:
1869         mutex_unlock(&journal->j_flush_mutex);
1870         return ret;
1871 }
1872
1873 /* for o_sync and fsync heavy applications, they tend to use
1874 ** all the journa list slots with tiny transactions.  These
1875 ** trigger lots and lots of calls to update the header block, which
1876 ** adds seeks and slows things down.
1877 **
1878 ** This function tries to clear out a large chunk of the journal lists
1879 ** at once, which makes everything faster since only the newest journal
1880 ** list updates the header block
1881 */
1882 static int flush_used_journal_lists(struct super_block *s,
1883                                     struct reiserfs_journal_list *jl)
1884 {
1885         unsigned long len = 0;
1886         unsigned long cur_len;
1887         int ret;
1888         int i;
1889         int limit = 256;
1890         struct reiserfs_journal_list *tjl;
1891         struct reiserfs_journal_list *flush_jl;
1892         unsigned int trans_id;
1893         struct reiserfs_journal *journal = SB_JOURNAL(s);
1894
1895         flush_jl = tjl = jl;
1896
1897         /* in data logging mode, try harder to flush a lot of blocks */
1898         if (reiserfs_data_log(s))
1899                 limit = 1024;
1900         /* flush for 256 transactions or limit blocks, whichever comes first */
1901         for (i = 0; i < 256 && len < limit; i++) {
1902                 if (atomic_read(&tjl->j_commit_left) ||
1903                     tjl->j_trans_id < jl->j_trans_id) {
1904                         break;
1905                 }
1906                 cur_len = atomic_read(&tjl->j_nonzerolen);
1907                 if (cur_len > 0) {
1908                         tjl->j_state &= ~LIST_TOUCHED;
1909                 }
1910                 len += cur_len;
1911                 flush_jl = tjl;
1912                 if (tjl->j_list.next == &journal->j_journal_list)
1913                         break;
1914                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1915         }
1916         /* try to find a group of blocks we can flush across all the
1917          ** transactions, but only bother if we've actually spanned
1918          ** across multiple lists
1919          */
1920         if (flush_jl != jl) {
1921                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1922         }
1923         flush_journal_list(s, flush_jl, 1);
1924         return 0;
1925 }
1926
1927 /*
1928 ** removes any nodes in table with name block and dev as bh.
1929 ** only touchs the hnext and hprev pointers.
1930 */
1931 void remove_journal_hash(struct super_block *sb,
1932                          struct reiserfs_journal_cnode **table,
1933                          struct reiserfs_journal_list *jl,
1934                          unsigned long block, int remove_freed)
1935 {
1936         struct reiserfs_journal_cnode *cur;
1937         struct reiserfs_journal_cnode **head;
1938
1939         head = &(journal_hash(table, sb, block));
1940         if (!head) {
1941                 return;
1942         }
1943         cur = *head;
1944         while (cur) {
1945                 if (cur->blocknr == block && cur->sb == sb
1946                     && (jl == NULL || jl == cur->jlist)
1947                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1948                         if (cur->hnext) {
1949                                 cur->hnext->hprev = cur->hprev;
1950                         }
1951                         if (cur->hprev) {
1952                                 cur->hprev->hnext = cur->hnext;
1953                         } else {
1954                                 *head = cur->hnext;
1955                         }
1956                         cur->blocknr = 0;
1957                         cur->sb = NULL;
1958                         cur->state = 0;
1959                         if (cur->bh && cur->jlist)      /* anybody who clears the cur->bh will also dec the nonzerolen */
1960                                 atomic_dec(&(cur->jlist->j_nonzerolen));
1961                         cur->bh = NULL;
1962                         cur->jlist = NULL;
1963                 }
1964                 cur = cur->hnext;
1965         }
1966 }
1967
1968 static void free_journal_ram(struct super_block *sb)
1969 {
1970         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1971         kfree(journal->j_current_jl);
1972         journal->j_num_lists--;
1973
1974         vfree(journal->j_cnode_free_orig);
1975         free_list_bitmaps(sb, journal->j_list_bitmap);
1976         free_bitmap_nodes(sb);  /* must be after free_list_bitmaps */
1977         if (journal->j_header_bh) {
1978                 brelse(journal->j_header_bh);
1979         }
1980         /* j_header_bh is on the journal dev, make sure not to release the journal
1981          * dev until we brelse j_header_bh
1982          */
1983         release_journal_dev(sb, journal);
1984         vfree(journal);
1985 }
1986
1987 /*
1988 ** call on unmount.  Only set error to 1 if you haven't made your way out
1989 ** of read_super() yet.  Any other caller must keep error at 0.
1990 */
1991 static int do_journal_release(struct reiserfs_transaction_handle *th,
1992                               struct super_block *sb, int error)
1993 {
1994         struct reiserfs_transaction_handle myth;
1995         int flushed = 0;
1996         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1997
1998         /* we only want to flush out transactions if we were called with error == 0
1999          */
2000         if (!error && !(sb->s_flags & MS_RDONLY)) {
2001                 /* end the current trans */
2002                 BUG_ON(!th->t_trans_id);
2003                 do_journal_end(th, sb, 10, FLUSH_ALL);
2004
2005                 /* make sure something gets logged to force our way into the flush code */
2006                 if (!journal_join(&myth, sb, 1)) {
2007                         reiserfs_prepare_for_journal(sb,
2008                                                      SB_BUFFER_WITH_SB(sb),
2009                                                      1);
2010                         journal_mark_dirty(&myth, sb,
2011                                            SB_BUFFER_WITH_SB(sb));
2012                         do_journal_end(&myth, sb, 1, FLUSH_ALL);
2013                         flushed = 1;
2014                 }
2015         }
2016
2017         /* this also catches errors during the do_journal_end above */
2018         if (!error && reiserfs_is_journal_aborted(journal)) {
2019                 memset(&myth, 0, sizeof(myth));
2020                 if (!journal_join_abort(&myth, sb, 1)) {
2021                         reiserfs_prepare_for_journal(sb,
2022                                                      SB_BUFFER_WITH_SB(sb),
2023                                                      1);
2024                         journal_mark_dirty(&myth, sb,
2025                                            SB_BUFFER_WITH_SB(sb));
2026                         do_journal_end(&myth, sb, 1, FLUSH_ALL);
2027                 }
2028         }
2029
2030         reiserfs_mounted_fs_count--;
2031         /* wait for all commits to finish */
2032         cancel_delayed_work(&SB_JOURNAL(sb)->j_work);
2033
2034         /*
2035          * We must release the write lock here because
2036          * the workqueue job (flush_async_commit) needs this lock
2037          */
2038         reiserfs_write_unlock(sb);
2039         flush_workqueue(commit_wq);
2040
2041         if (!reiserfs_mounted_fs_count) {
2042                 destroy_workqueue(commit_wq);
2043                 commit_wq = NULL;
2044         }
2045         reiserfs_write_lock(sb);
2046
2047         free_journal_ram(sb);
2048
2049         return 0;
2050 }
2051
2052 /*
2053 ** call on unmount.  flush all journal trans, release all alloc'd ram
2054 */
2055 int journal_release(struct reiserfs_transaction_handle *th,
2056                     struct super_block *sb)
2057 {
2058         return do_journal_release(th, sb, 0);
2059 }
2060
2061 /*
2062 ** only call from an error condition inside reiserfs_read_super!
2063 */
2064 int journal_release_error(struct reiserfs_transaction_handle *th,
2065                           struct super_block *sb)
2066 {
2067         return do_journal_release(th, sb, 1);
2068 }
2069
2070 /* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
2071 static int journal_compare_desc_commit(struct super_block *sb,
2072                                        struct reiserfs_journal_desc *desc,
2073                                        struct reiserfs_journal_commit *commit)
2074 {
2075         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
2076             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
2077             get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
2078             get_commit_trans_len(commit) <= 0) {
2079                 return 1;
2080         }
2081         return 0;
2082 }
2083
2084 /* returns 0 if it did not find a description block
2085 ** returns -1 if it found a corrupt commit block
2086 ** returns 1 if both desc and commit were valid
2087 */
2088 static int journal_transaction_is_valid(struct super_block *sb,
2089                                         struct buffer_head *d_bh,
2090                                         unsigned int *oldest_invalid_trans_id,
2091                                         unsigned long *newest_mount_id)
2092 {
2093         struct reiserfs_journal_desc *desc;
2094         struct reiserfs_journal_commit *commit;
2095         struct buffer_head *c_bh;
2096         unsigned long offset;
2097
2098         if (!d_bh)
2099                 return 0;
2100
2101         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2102         if (get_desc_trans_len(desc) > 0
2103             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2104                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2105                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2106                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2107                                        "journal-986: transaction "
2108                                        "is valid returning because trans_id %d is greater than "
2109                                        "oldest_invalid %lu",
2110                                        get_desc_trans_id(desc),
2111                                        *oldest_invalid_trans_id);
2112                         return 0;
2113                 }
2114                 if (newest_mount_id
2115                     && *newest_mount_id > get_desc_mount_id(desc)) {
2116                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2117                                        "journal-1087: transaction "
2118                                        "is valid returning because mount_id %d is less than "
2119                                        "newest_mount_id %lu",
2120                                        get_desc_mount_id(desc),
2121                                        *newest_mount_id);
2122                         return -1;
2123                 }
2124                 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2125                         reiserfs_warning(sb, "journal-2018",
2126                                          "Bad transaction length %d "
2127                                          "encountered, ignoring transaction",
2128                                          get_desc_trans_len(desc));
2129                         return -1;
2130                 }
2131                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2132
2133                 /* ok, we have a journal description block, lets see if the transaction was valid */
2134                 c_bh =
2135                     journal_bread(sb,
2136                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2137                                   ((offset + get_desc_trans_len(desc) +
2138                                     1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2139                 if (!c_bh)
2140                         return 0;
2141                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2142                 if (journal_compare_desc_commit(sb, desc, commit)) {
2143                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2144                                        "journal_transaction_is_valid, commit offset %ld had bad "
2145                                        "time %d or length %d",
2146                                        c_bh->b_blocknr -
2147                                        SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2148                                        get_commit_trans_id(commit),
2149                                        get_commit_trans_len(commit));
2150                         brelse(c_bh);
2151                         if (oldest_invalid_trans_id) {
2152                                 *oldest_invalid_trans_id =
2153                                     get_desc_trans_id(desc);
2154                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2155                                                "journal-1004: "
2156                                                "transaction_is_valid setting oldest invalid trans_id "
2157                                                "to %d",
2158                                                get_desc_trans_id(desc));
2159                         }
2160                         return -1;
2161                 }
2162                 brelse(c_bh);
2163                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2164                                "journal-1006: found valid "
2165                                "transaction start offset %llu, len %d id %d",
2166                                d_bh->b_blocknr -
2167                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2168                                get_desc_trans_len(desc),
2169                                get_desc_trans_id(desc));
2170                 return 1;
2171         } else {
2172                 return 0;
2173         }
2174 }
2175
2176 static void brelse_array(struct buffer_head **heads, int num)
2177 {
2178         int i;
2179         for (i = 0; i < num; i++) {
2180                 brelse(heads[i]);
2181         }
2182 }
2183
2184 /*
2185 ** given the start, and values for the oldest acceptable transactions,
2186 ** this either reads in a replays a transaction, or returns because the transaction
2187 ** is invalid, or too old.
2188 */
2189 static int journal_read_transaction(struct super_block *sb,
2190                                     unsigned long cur_dblock,
2191                                     unsigned long oldest_start,
2192                                     unsigned int oldest_trans_id,
2193                                     unsigned long newest_mount_id)
2194 {
2195         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2196         struct reiserfs_journal_desc *desc;
2197         struct reiserfs_journal_commit *commit;
2198         unsigned int trans_id = 0;
2199         struct buffer_head *c_bh;
2200         struct buffer_head *d_bh;
2201         struct buffer_head **log_blocks = NULL;
2202         struct buffer_head **real_blocks = NULL;
2203         unsigned int trans_offset;
2204         int i;
2205         int trans_half;
2206
2207         d_bh = journal_bread(sb, cur_dblock);
2208         if (!d_bh)
2209                 return 1;
2210         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2211         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2212         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2213                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2214                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2215                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2216         if (get_desc_trans_id(desc) < oldest_trans_id) {
2217                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2218                                "journal_read_trans skipping because %lu is too old",
2219                                cur_dblock -
2220                                SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2221                 brelse(d_bh);
2222                 return 1;
2223         }
2224         if (get_desc_mount_id(desc) != newest_mount_id) {
2225                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2226                                "journal_read_trans skipping because %d is != "
2227                                "newest_mount_id %lu", get_desc_mount_id(desc),
2228                                newest_mount_id);
2229                 brelse(d_bh);
2230                 return 1;
2231         }
2232         c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2233                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2234                               SB_ONDISK_JOURNAL_SIZE(sb)));
2235         if (!c_bh) {
2236                 brelse(d_bh);
2237                 return 1;
2238         }
2239         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2240         if (journal_compare_desc_commit(sb, desc, commit)) {
2241                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2242                                "journal_read_transaction, "
2243                                "commit offset %llu had bad time %d or length %d",
2244                                c_bh->b_blocknr -
2245                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2246                                get_commit_trans_id(commit),
2247                                get_commit_trans_len(commit));
2248                 brelse(c_bh);
2249                 brelse(d_bh);
2250                 return 1;
2251         }
2252         trans_id = get_desc_trans_id(desc);
2253         /* now we know we've got a good transaction, and it was inside the valid time ranges */
2254         log_blocks = kmalloc(get_desc_trans_len(desc) *
2255                              sizeof(struct buffer_head *), GFP_NOFS);
2256         real_blocks = kmalloc(get_desc_trans_len(desc) *
2257                               sizeof(struct buffer_head *), GFP_NOFS);
2258         if (!log_blocks || !real_blocks) {
2259                 brelse(c_bh);
2260                 brelse(d_bh);
2261                 kfree(log_blocks);
2262                 kfree(real_blocks);
2263                 reiserfs_warning(sb, "journal-1169",
2264                                  "kmalloc failed, unable to mount FS");
2265                 return -1;
2266         }
2267         /* get all the buffer heads */
2268         trans_half = journal_trans_half(sb->s_blocksize);
2269         for (i = 0; i < get_desc_trans_len(desc); i++) {
2270                 log_blocks[i] =
2271                     journal_getblk(sb,
2272                                    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2273                                    (trans_offset + 1 +
2274                                     i) % SB_ONDISK_JOURNAL_SIZE(sb));
2275                 if (i < trans_half) {
2276                         real_blocks[i] =
2277                             sb_getblk(sb,
2278                                       le32_to_cpu(desc->j_realblock[i]));
2279                 } else {
2280                         real_blocks[i] =
2281                             sb_getblk(sb,
2282                                       le32_to_cpu(commit->
2283                                                   j_realblock[i - trans_half]));
2284                 }
2285                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2286                         reiserfs_warning(sb, "journal-1207",
2287                                          "REPLAY FAILURE fsck required! "
2288                                          "Block to replay is outside of "
2289                                          "filesystem");
2290                         goto abort_replay;
2291                 }
2292                 /* make sure we don't try to replay onto log or reserved area */
2293                 if (is_block_in_log_or_reserved_area
2294                     (sb, real_blocks[i]->b_blocknr)) {
2295                         reiserfs_warning(sb, "journal-1204",
2296                                          "REPLAY FAILURE fsck required! "
2297                                          "Trying to replay onto a log block");
2298                       abort_replay:
2299                         brelse_array(log_blocks, i);
2300                         brelse_array(real_blocks, i);
2301                         brelse(c_bh);
2302                         brelse(d_bh);
2303                         kfree(log_blocks);
2304                         kfree(real_blocks);
2305                         return -1;
2306                 }
2307         }
2308         /* read in the log blocks, memcpy to the corresponding real block */
2309         ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2310         for (i = 0; i < get_desc_trans_len(desc); i++) {
2311
2312                 reiserfs_write_unlock(sb);
2313                 wait_on_buffer(log_blocks[i]);
2314                 reiserfs_write_lock(sb);
2315
2316                 if (!buffer_uptodate(log_blocks[i])) {
2317                         reiserfs_warning(sb, "journal-1212",
2318                                          "REPLAY FAILURE fsck required! "
2319                                          "buffer write failed");
2320                         brelse_array(log_blocks + i,
2321                                      get_desc_trans_len(desc) - i);
2322                         brelse_array(real_blocks, get_desc_trans_len(desc));
2323                         brelse(c_bh);
2324                         brelse(d_bh);
2325                         kfree(log_blocks);
2326                         kfree(real_blocks);
2327                         return -1;
2328                 }
2329                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2330                        real_blocks[i]->b_size);
2331                 set_buffer_uptodate(real_blocks[i]);
2332                 brelse(log_blocks[i]);
2333         }
2334         /* flush out the real blocks */
2335         for (i = 0; i < get_desc_trans_len(desc); i++) {
2336                 set_buffer_dirty(real_blocks[i]);
2337                 ll_rw_block(SWRITE, 1, real_blocks + i);
2338         }
2339         for (i = 0; i < get_desc_trans_len(desc); i++) {
2340                 wait_on_buffer(real_blocks[i]);
2341                 if (!buffer_uptodate(real_blocks[i])) {
2342                         reiserfs_warning(sb, "journal-1226",
2343                                          "REPLAY FAILURE, fsck required! "
2344                                          "buffer write failed");
2345                         brelse_array(real_blocks + i,
2346                                      get_desc_trans_len(desc) - i);
2347                         brelse(c_bh);
2348                         brelse(d_bh);
2349                         kfree(log_blocks);
2350                         kfree(real_blocks);
2351                         return -1;
2352                 }
2353                 brelse(real_blocks[i]);
2354         }
2355         cur_dblock =
2356             SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2357             ((trans_offset + get_desc_trans_len(desc) +
2358               2) % SB_ONDISK_JOURNAL_SIZE(sb));
2359         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2360                        "journal-1095: setting journal " "start to offset %ld",
2361                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2362
2363         /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2364         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2365         journal->j_last_flush_trans_id = trans_id;
2366         journal->j_trans_id = trans_id + 1;
2367         /* check for trans_id overflow */
2368         if (journal->j_trans_id == 0)
2369                 journal->j_trans_id = 10;
2370         brelse(c_bh);
2371         brelse(d_bh);
2372         kfree(log_blocks);
2373         kfree(real_blocks);
2374         return 0;
2375 }
2376
2377 /* This function reads blocks starting from block and to max_block of bufsize
2378    size (but no more than BUFNR blocks at a time). This proved to improve
2379    mounting speed on self-rebuilding raid5 arrays at least.
2380    Right now it is only used from journal code. But later we might use it
2381    from other places.
2382    Note: Do not use journal_getblk/sb_getblk functions here! */
2383 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2384                                            b_blocknr_t block, int bufsize,
2385                                            b_blocknr_t max_block)
2386 {
2387         struct buffer_head *bhlist[BUFNR];
2388         unsigned int blocks = BUFNR;
2389         struct buffer_head *bh;
2390         int i, j;
2391
2392         bh = __getblk(dev, block, bufsize);
2393         if (buffer_uptodate(bh))
2394                 return (bh);
2395
2396         if (block + BUFNR > max_block) {
2397                 blocks = max_block - block;
2398         }
2399         bhlist[0] = bh;
2400         j = 1;
2401         for (i = 1; i < blocks; i++) {
2402                 bh = __getblk(dev, block + i, bufsize);
2403                 if (buffer_uptodate(bh)) {
2404                         brelse(bh);
2405                         break;
2406                 } else
2407                         bhlist[j++] = bh;
2408         }
2409         ll_rw_block(READ, j, bhlist);
2410         for (i = 1; i < j; i++)
2411                 brelse(bhlist[i]);
2412         bh = bhlist[0];
2413         wait_on_buffer(bh);
2414         if (buffer_uptodate(bh))
2415                 return bh;
2416         brelse(bh);
2417         return NULL;
2418 }
2419
2420 /*
2421 ** read and replay the log
2422 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2423 ** transaction.  This tests that before finding all the transactions in the log, which makes normal mount times fast.
2424 **
2425 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2426 **
2427 ** On exit, it sets things up so the first transaction will work correctly.
2428 */
2429 static int journal_read(struct super_block *sb)
2430 {
2431         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2432         struct reiserfs_journal_desc *desc;
2433         unsigned int oldest_trans_id = 0;
2434         unsigned int oldest_invalid_trans_id = 0;
2435         time_t start;
2436         unsigned long oldest_start = 0;
2437         unsigned long cur_dblock = 0;
2438         unsigned long newest_mount_id = 9;
2439         struct buffer_head *d_bh;
2440         struct reiserfs_journal_header *jh;
2441         int valid_journal_header = 0;
2442         int replay_count = 0;
2443         int continue_replay = 1;
2444         int ret;
2445         char b[BDEVNAME_SIZE];
2446
2447         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2448         reiserfs_info(sb, "checking transaction log (%s)\n",
2449                       bdevname(journal->j_dev_bd, b));
2450         start = get_seconds();
2451
2452         /* step 1, read in the journal header block.  Check the transaction it says
2453          ** is the first unflushed, and if that transaction is not valid,
2454          ** replay is done
2455          */
2456         journal->j_header_bh = journal_bread(sb,
2457                                              SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2458                                              + SB_ONDISK_JOURNAL_SIZE(sb));
2459         if (!journal->j_header_bh) {
2460                 return 1;
2461         }
2462         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2463         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2464             SB_ONDISK_JOURNAL_SIZE(sb)
2465             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2466                 oldest_start =
2467                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2468                     le32_to_cpu(jh->j_first_unflushed_offset);
2469                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2470                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2471                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2472                                "journal-1153: found in "
2473                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2474                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2475                                le32_to_cpu(jh->j_last_flush_trans_id));
2476                 valid_journal_header = 1;
2477
2478                 /* now, we try to read the first unflushed offset.  If it is not valid,
2479                  ** there is nothing more we can do, and it makes no sense to read
2480                  ** through the whole log.
2481                  */
2482                 d_bh =
2483                     journal_bread(sb,
2484                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2485                                   le32_to_cpu(jh->j_first_unflushed_offset));
2486                 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2487                 if (!ret) {
2488                         continue_replay = 0;
2489                 }
2490                 brelse(d_bh);
2491                 goto start_log_replay;
2492         }
2493
2494         if (continue_replay && bdev_read_only(sb->s_bdev)) {
2495                 reiserfs_warning(sb, "clm-2076",
2496                                  "device is readonly, unable to replay log");
2497                 return -1;
2498         }
2499
2500         /* ok, there are transactions that need to be replayed.  start with the first log block, find
2501          ** all the valid transactions, and pick out the oldest.
2502          */
2503         while (continue_replay
2504                && cur_dblock <
2505                (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2506                 SB_ONDISK_JOURNAL_SIZE(sb))) {
2507                 /* Note that it is required for blocksize of primary fs device and journal
2508                    device to be the same */
2509                 d_bh =
2510                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2511                                     sb->s_blocksize,
2512                                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2513                                     SB_ONDISK_JOURNAL_SIZE(sb));
2514                 ret =
2515                     journal_transaction_is_valid(sb, d_bh,
2516                                                  &oldest_invalid_trans_id,
2517                                                  &newest_mount_id);
2518                 if (ret == 1) {
2519                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2520                         if (oldest_start == 0) {        /* init all oldest_ values */
2521                                 oldest_trans_id = get_desc_trans_id(desc);
2522                                 oldest_start = d_bh->b_blocknr;
2523                                 newest_mount_id = get_desc_mount_id(desc);
2524                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2525                                                "journal-1179: Setting "
2526                                                "oldest_start to offset %llu, trans_id %lu",
2527                                                oldest_start -
2528                                                SB_ONDISK_JOURNAL_1st_BLOCK
2529                                                (sb), oldest_trans_id);
2530                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2531                                 /* one we just read was older */
2532                                 oldest_trans_id = get_desc_trans_id(desc);
2533                                 oldest_start = d_bh->b_blocknr;
2534                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2535                                                "journal-1180: Resetting "
2536                                                "oldest_start to offset %lu, trans_id %lu",
2537                                                oldest_start -
2538                                                SB_ONDISK_JOURNAL_1st_BLOCK
2539                                                (sb), oldest_trans_id);
2540                         }
2541                         if (newest_mount_id < get_desc_mount_id(desc)) {
2542                                 newest_mount_id = get_desc_mount_id(desc);
2543                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2544                                                "journal-1299: Setting "
2545                                                "newest_mount_id to %d",
2546                                                get_desc_mount_id(desc));
2547                         }
2548                         cur_dblock += get_desc_trans_len(desc) + 2;
2549                 } else {
2550                         cur_dblock++;
2551                 }
2552                 brelse(d_bh);
2553         }
2554
2555       start_log_replay:
2556         cur_dblock = oldest_start;
2557         if (oldest_trans_id) {
2558                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2559                                "journal-1206: Starting replay "
2560                                "from offset %llu, trans_id %lu",
2561                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2562                                oldest_trans_id);
2563
2564         }
2565         replay_count = 0;
2566         while (continue_replay && oldest_trans_id > 0) {
2567                 ret =
2568                     journal_read_transaction(sb, cur_dblock, oldest_start,
2569                                              oldest_trans_id, newest_mount_id);
2570                 if (ret < 0) {
2571                         return ret;
2572                 } else if (ret != 0) {
2573                         break;
2574                 }
2575                 cur_dblock =
2576                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2577                 replay_count++;
2578                 if (cur_dblock == oldest_start)
2579                         break;
2580         }
2581
2582         if (oldest_trans_id == 0) {
2583                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2584                                "journal-1225: No valid " "transactions found");
2585         }
2586         /* j_start does not get set correctly if we don't replay any transactions.
2587          ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2588          ** copy the trans_id from the header
2589          */
2590         if (valid_journal_header && replay_count == 0) {
2591                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2592                 journal->j_trans_id =
2593                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2594                 /* check for trans_id overflow */
2595                 if (journal->j_trans_id == 0)
2596                         journal->j_trans_id = 10;
2597                 journal->j_last_flush_trans_id =
2598                     le32_to_cpu(jh->j_last_flush_trans_id);
2599                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2600         } else {
2601                 journal->j_mount_id = newest_mount_id + 1;
2602         }
2603         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2604                        "newest_mount_id to %lu", journal->j_mount_id);
2605         journal->j_first_unflushed_offset = journal->j_start;
2606         if (replay_count > 0) {
2607                 reiserfs_info(sb,
2608                               "replayed %d transactions in %lu seconds\n",
2609                               replay_count, get_seconds() - start);
2610         }
2611         if (!bdev_read_only(sb->s_bdev) &&
2612             _update_journal_header_block(sb, journal->j_start,
2613                                          journal->j_last_flush_trans_id)) {
2614                 /* replay failed, caller must call free_journal_ram and abort
2615                  ** the mount
2616                  */
2617                 return -1;
2618         }
2619         return 0;
2620 }
2621
2622 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2623 {
2624         struct reiserfs_journal_list *jl;
2625         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2626                      GFP_NOFS | __GFP_NOFAIL);
2627         INIT_LIST_HEAD(&jl->j_list);
2628         INIT_LIST_HEAD(&jl->j_working_list);
2629         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2630         INIT_LIST_HEAD(&jl->j_bh_list);
2631         mutex_init(&jl->j_commit_mutex);
2632         SB_JOURNAL(s)->j_num_lists++;
2633         get_journal_list(jl);
2634         return jl;
2635 }
2636
2637 static void journal_list_init(struct super_block *sb)
2638 {
2639         SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2640 }
2641
2642 static int release_journal_dev(struct super_block *super,
2643                                struct reiserfs_journal *journal)
2644 {
2645         int result;
2646
2647         result = 0;
2648
2649         if (journal->j_dev_bd != NULL) {
2650                 if (journal->j_dev_bd->bd_dev != super->s_dev)
2651                         bd_release(journal->j_dev_bd);
2652                 result = blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2653                 journal->j_dev_bd = NULL;
2654         }
2655
2656         if (result != 0) {
2657                 reiserfs_warning(super, "sh-457",
2658                                  "Cannot release journal device: %i", result);
2659         }
2660         return result;
2661 }
2662
2663 static int journal_init_dev(struct super_block *super,
2664                             struct reiserfs_journal *journal,
2665                             const char *jdev_name)
2666 {
2667         int result;
2668         dev_t jdev;
2669         fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE;
2670         char b[BDEVNAME_SIZE];
2671
2672         result = 0;
2673
2674         journal->j_dev_bd = NULL;
2675         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2676             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2677
2678         if (bdev_read_only(super->s_bdev))
2679                 blkdev_mode = FMODE_READ;
2680
2681         /* there is no "jdev" option and journal is on separate device */
2682         if ((!jdev_name || !jdev_name[0])) {
2683                 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2684                 journal->j_dev_mode = blkdev_mode;
2685                 if (IS_ERR(journal->j_dev_bd)) {
2686                         result = PTR_ERR(journal->j_dev_bd);
2687                         journal->j_dev_bd = NULL;
2688                         reiserfs_warning(super, "sh-458",
2689                                          "cannot init journal device '%s': %i",
2690                                          __bdevname(jdev, b), result);
2691                         return result;
2692                 } else if (jdev != super->s_dev) {
2693                         result = bd_claim(journal->j_dev_bd, journal);
2694                         if (result) {
2695                                 blkdev_put(journal->j_dev_bd, blkdev_mode);
2696                                 return result;
2697                         }
2698
2699                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2700                 }
2701
2702                 return 0;
2703         }
2704
2705         journal->j_dev_mode = blkdev_mode;
2706         journal->j_dev_bd = open_bdev_exclusive(jdev_name,
2707                                                 blkdev_mode, journal);
2708         if (IS_ERR(journal->j_dev_bd)) {
2709                 result = PTR_ERR(journal->j_dev_bd);
2710                 journal->j_dev_bd = NULL;
2711                 reiserfs_warning(super,
2712                                  "journal_init_dev: Cannot open '%s': %i",
2713                                  jdev_name, result);
2714                 return result;
2715         }
2716
2717         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2718         reiserfs_info(super,
2719                       "journal_init_dev: journal device: %s\n",
2720                       bdevname(journal->j_dev_bd, b));
2721         return 0;
2722 }
2723
2724 /**
2725  * When creating/tuning a file system user can assign some
2726  * journal params within boundaries which depend on the ratio
2727  * blocksize/standard_blocksize.
2728  *
2729  * For blocks >= standard_blocksize transaction size should
2730  * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2731  * then JOURNAL_TRANS_MAX_DEFAULT.
2732  *
2733  * For blocks < standard_blocksize these boundaries should be
2734  * decreased proportionally.
2735  */
2736 #define REISERFS_STANDARD_BLKSIZE (4096)
2737
2738 static int check_advise_trans_params(struct super_block *sb,
2739                                      struct reiserfs_journal *journal)
2740 {
2741         if (journal->j_trans_max) {
2742                 /* Non-default journal params.
2743                    Do sanity check for them. */
2744                 int ratio = 1;
2745                 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2746                         ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2747
2748                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2749                     journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2750                     SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2751                     JOURNAL_MIN_RATIO) {
2752                         reiserfs_warning(sb, "sh-462",
2753                                          "bad transaction max size (%u). "
2754                                          "FSCK?", journal->j_trans_max);
2755                         return 1;
2756                 }
2757                 if (journal->j_max_batch != (journal->j_trans_max) *
2758                         JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2759                         reiserfs_warning(sb, "sh-463",
2760                                          "bad transaction max batch (%u). "
2761                                          "FSCK?", journal->j_max_batch);
2762                         return 1;
2763                 }
2764         } else {
2765                 /* Default journal params.
2766                    The file system was created by old version
2767                    of mkreiserfs, so some fields contain zeros,
2768                    and we need to advise proper values for them */
2769                 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2770                         reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2771                                          sb->s_blocksize);
2772                         return 1;
2773                 }
2774                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2775                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2776                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2777         }
2778         return 0;
2779 }
2780
2781 /*
2782 ** must be called once on fs mount.  calls journal_read for you
2783 */
2784 int journal_init(struct super_block *sb, const char *j_dev_name,
2785                  int old_format, unsigned int commit_max_age)
2786 {
2787         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2788         struct buffer_head *bhjh;
2789         struct reiserfs_super_block *rs;
2790         struct reiserfs_journal_header *jh;
2791         struct reiserfs_journal *journal;
2792         struct reiserfs_journal_list *jl;
2793         char b[BDEVNAME_SIZE];
2794
2795         journal = SB_JOURNAL(sb) = vmalloc(sizeof(struct reiserfs_journal));
2796         if (!journal) {
2797                 reiserfs_warning(sb, "journal-1256",
2798                                  "unable to get memory for journal structure");
2799                 return 1;
2800         }
2801         memset(journal, 0, sizeof(struct reiserfs_journal));
2802         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2803         INIT_LIST_HEAD(&journal->j_prealloc_list);
2804         INIT_LIST_HEAD(&journal->j_working_list);
2805         INIT_LIST_HEAD(&journal->j_journal_list);
2806         journal->j_persistent_trans = 0;
2807         if (reiserfs_allocate_list_bitmaps(sb,
2808                                            journal->j_list_bitmap,
2809                                            reiserfs_bmap_count(sb)))
2810                 goto free_and_return;
2811         allocate_bitmap_nodes(sb);
2812
2813         /* reserved for journal area support */
2814         SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2815                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2816                                                  / sb->s_blocksize +
2817                                                  reiserfs_bmap_count(sb) +
2818                                                  1 :
2819                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2820                                                  sb->s_blocksize + 2);
2821
2822         /* Sanity check to see is the standard journal fitting withing first bitmap
2823            (actual for small blocksizes) */
2824         if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2825             (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2826              SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2827                 reiserfs_warning(sb, "journal-1393",
2828                                  "journal does not fit for area addressed "
2829                                  "by first of bitmap blocks. It starts at "
2830                                  "%u and its size is %u. Block size %ld",
2831                                  SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2832                                  SB_ONDISK_JOURNAL_SIZE(sb),
2833                                  sb->s_blocksize);
2834                 goto free_and_return;
2835         }
2836
2837         if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2838                 reiserfs_warning(sb, "sh-462",
2839                                  "unable to initialize jornal device");
2840                 goto free_and_return;
2841         }
2842
2843         rs = SB_DISK_SUPER_BLOCK(sb);
2844
2845         /* read journal header */
2846         bhjh = journal_bread(sb,
2847                              SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2848                              SB_ONDISK_JOURNAL_SIZE(sb));
2849         if (!bhjh) {
2850                 reiserfs_warning(sb, "sh-459",
2851                                  "unable to read journal header");
2852                 goto free_and_return;
2853         }
2854         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2855
2856         /* make sure that journal matches to the super block */
2857         if (is_reiserfs_jr(rs)
2858             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2859                 sb_jp_journal_magic(rs))) {
2860                 reiserfs_warning(sb, "sh-460",
2861                                  "journal header magic %x (device %s) does "
2862                                  "not match to magic found in super block %x",
2863                                  jh->jh_journal.jp_journal_magic,
2864                                  bdevname(journal->j_dev_bd, b),
2865                                  sb_jp_journal_magic(rs));
2866                 brelse(bhjh);
2867                 goto free_and_return;
2868         }
2869
2870         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2871         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2872         journal->j_max_commit_age =
2873             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2874         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2875
2876         if (check_advise_trans_params(sb, journal) != 0)
2877                 goto free_and_return;
2878         journal->j_default_max_commit_age = journal->j_max_commit_age;
2879
2880         if (commit_max_age != 0) {
2881                 journal->j_max_commit_age = commit_max_age;
2882                 journal->j_max_trans_age = commit_max_age;
2883         }
2884
2885         reiserfs_info(sb, "journal params: device %s, size %u, "
2886                       "journal first block %u, max trans len %u, max batch %u, "
2887                       "max commit age %u, max trans age %u\n",
2888                       bdevname(journal->j_dev_bd, b),
2889                       SB_ONDISK_JOURNAL_SIZE(sb),
2890                       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2891                       journal->j_trans_max,
2892                       journal->j_max_batch,
2893                       journal->j_max_commit_age, journal->j_max_trans_age);
2894
2895         brelse(bhjh);
2896
2897         journal->j_list_bitmap_index = 0;
2898         journal_list_init(sb);
2899
2900         memset(journal->j_list_hash_table, 0,
2901                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2902
2903         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2904         spin_lock_init(&journal->j_dirty_buffers_lock);
2905
2906         journal->j_start = 0;
2907         journal->j_len = 0;
2908         journal->j_len_alloc = 0;
2909         atomic_set(&(journal->j_wcount), 0);
2910         atomic_set(&(journal->j_async_throttle), 0);
2911         journal->j_bcount = 0;
2912         journal->j_trans_start_time = 0;
2913         journal->j_last = NULL;
2914         journal->j_first = NULL;
2915         init_waitqueue_head(&(journal->j_join_wait));
2916         mutex_init(&journal->j_mutex);
2917         mutex_init(&journal->j_flush_mutex);
2918
2919         journal->j_trans_id = 10;
2920         journal->j_mount_id = 10;
2921         journal->j_state = 0;
2922         atomic_set(&(journal->j_jlock), 0);
2923         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2924         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2925         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2926         journal->j_cnode_used = 0;
2927         journal->j_must_wait = 0;
2928
2929         if (journal->j_cnode_free == 0) {
2930                 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2931                                  "allocation failed (%ld bytes). Journal is "
2932                                  "too large for available memory. Usually "
2933                                  "this is due to a journal that is too large.",
2934                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2935                 goto free_and_return;
2936         }
2937
2938         init_journal_hash(sb);
2939         jl = journal->j_current_jl;
2940         jl->j_list_bitmap = get_list_bitmap(sb, jl);
2941         if (!jl->j_list_bitmap) {
2942                 reiserfs_warning(sb, "journal-2005",
2943                                  "get_list_bitmap failed for journal list 0");
2944                 goto free_and_return;
2945         }
2946         if (journal_read(sb) < 0) {
2947                 reiserfs_warning(sb, "reiserfs-2006",
2948                                  "Replay Failure, unable to mount");
2949                 goto free_and_return;
2950         }
2951
2952         reiserfs_mounted_fs_count++;
2953         if (reiserfs_mounted_fs_count <= 1)
2954                 commit_wq = create_workqueue("reiserfs");
2955
2956         INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2957         journal->j_work_sb = sb;
2958         return 0;
2959       free_and_return:
2960         free_journal_ram(sb);
2961         return 1;
2962 }
2963
2964 /*
2965 ** test for a polite end of the current transaction.  Used by file_write, and should
2966 ** be used by delete to make sure they don't write more than can fit inside a single
2967 ** transaction
2968 */
2969 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2970                                    int new_alloc)
2971 {
2972         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2973         time_t now = get_seconds();
2974         /* cannot restart while nested */
2975         BUG_ON(!th->t_trans_id);
2976         if (th->t_refcount > 1)
2977                 return 0;
2978         if (journal->j_must_wait > 0 ||
2979             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2980             atomic_read(&(journal->j_jlock)) ||
2981             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2982             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2983                 return 1;
2984         }
2985         /* protected by the BKL here */
2986         journal->j_len_alloc += new_alloc;
2987         th->t_blocks_allocated += new_alloc ;
2988         return 0;
2989 }
2990
2991 /* this must be called inside a transaction, and requires the
2992 ** kernel_lock to be held
2993 */
2994 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2995 {
2996         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2997         BUG_ON(!th->t_trans_id);
2998         journal->j_must_wait = 1;
2999         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
3000         return;
3001 }
3002
3003 /* this must be called without a transaction started, and does not
3004 ** require BKL
3005 */
3006 void reiserfs_allow_writes(struct super_block *s)
3007 {
3008         struct reiserfs_journal *journal = SB_JOURNAL(s);
3009         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
3010         wake_up(&journal->j_join_wait);
3011 }
3012
3013 /* this must be called without a transaction started, and does not
3014 ** require BKL
3015 */
3016 void reiserfs_wait_on_write_block(struct super_block *s)
3017 {
3018         struct reiserfs_journal *journal = SB_JOURNAL(s);
3019         wait_event(journal->j_join_wait,
3020                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
3021 }
3022
3023 static void queue_log_writer(struct super_block *s)
3024 {
3025         wait_queue_t wait;
3026         struct reiserfs_journal *journal = SB_JOURNAL(s);
3027         set_bit(J_WRITERS_QUEUED, &journal->j_state);
3028
3029         /*
3030          * we don't want to use wait_event here because
3031          * we only want to wait once.
3032          */
3033         init_waitqueue_entry(&wait, current);
3034         add_wait_queue(&journal->j_join_wait, &wait);
3035         set_current_state(TASK_UNINTERRUPTIBLE);
3036         if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
3037                 reiserfs_write_unlock(s);
3038                 schedule();
3039                 reiserfs_write_lock(s);
3040         }
3041         __set_current_state(TASK_RUNNING);
3042         remove_wait_queue(&journal->j_join_wait, &wait);
3043 }
3044
3045 static void wake_queued_writers(struct super_block *s)
3046 {
3047         struct reiserfs_journal *journal = SB_JOURNAL(s);
3048         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
3049                 wake_up(&journal->j_join_wait);
3050 }
3051
3052 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
3053 {
3054         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3055         unsigned long bcount = journal->j_bcount;
3056         while (1) {
3057                 reiserfs_write_unlock(sb);
3058                 schedule_timeout_uninterruptible(1);
3059                 reiserfs_write_lock(sb);
3060                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
3061                 while ((atomic_read(&journal->j_wcount) > 0 ||
3062                         atomic_read(&journal->j_jlock)) &&
3063                        journal->j_trans_id == trans_id) {
3064                         queue_log_writer(sb);
3065                 }
3066                 if (journal->j_trans_id != trans_id)
3067                         break;
3068                 if (bcount == journal->j_bcount)
3069                         break;
3070                 bcount = journal->j_bcount;
3071         }
3072 }
3073
3074 /* join == true if you must join an existing transaction.
3075 ** join == false if you can deal with waiting for others to finish
3076 **
3077 ** this will block until the transaction is joinable.  send the number of blocks you
3078 ** expect to use in nblocks.
3079 */
3080 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3081                               struct super_block *sb, unsigned long nblocks,
3082                               int join)
3083 {
3084         time_t now = get_seconds();
3085         unsigned int old_trans_id;
3086         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3087         struct reiserfs_transaction_handle myth;
3088         int sched_count = 0;
3089         int retval;
3090
3091         reiserfs_check_lock_depth(sb, "journal_begin");
3092         BUG_ON(nblocks > journal->j_trans_max);
3093
3094         PROC_INFO_INC(sb, journal.journal_being);
3095         /* set here for journal_join */
3096         th->t_refcount = 1;
3097         th->t_super = sb;
3098
3099       relock:
3100         lock_journal(sb);
3101         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3102                 unlock_journal(sb);
3103                 retval = journal->j_errno;
3104                 goto out_fail;
3105         }
3106         journal->j_bcount++;
3107
3108         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3109                 unlock_journal(sb);
3110                 reiserfs_write_unlock(sb);
3111                 reiserfs_wait_on_write_block(sb);
3112                 reiserfs_write_lock(sb);
3113                 PROC_INFO_INC(sb, journal.journal_relock_writers);
3114                 goto relock;
3115         }
3116         now = get_seconds();
3117
3118         /* if there is no room in the journal OR
3119          ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
3120          ** we don't sleep if there aren't other writers
3121          */
3122
3123         if ((!join && journal->j_must_wait > 0) ||
3124             (!join
3125              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3126             || (!join && atomic_read(&journal->j_wcount) > 0
3127                 && journal->j_trans_start_time > 0
3128                 && (now - journal->j_trans_start_time) >
3129                 journal->j_max_trans_age) || (!join
3130                                               && atomic_read(&journal->j_jlock))
3131             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3132
3133                 old_trans_id = journal->j_trans_id;
3134                 unlock_journal(sb);     /* allow others to finish this transaction */
3135
3136                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3137                     journal->j_max_batch &&
3138                     ((journal->j_len + nblocks + 2) * 100) <
3139                     (journal->j_len_alloc * 75)) {
3140                         if (atomic_read(&journal->j_wcount) > 10) {
3141                                 sched_count++;
3142                                 queue_log_writer(sb);
3143                                 goto relock;
3144                         }
3145                 }
3146                 /* don't mess with joining the transaction if all we have to do is
3147                  * wait for someone else to do a commit
3148                  */
3149                 if (atomic_read(&journal->j_jlock)) {
3150                         while (journal->j_trans_id == old_trans_id &&
3151                                atomic_read(&journal->j_jlock)) {
3152                                 queue_log_writer(sb);
3153                         }
3154                         goto relock;
3155                 }
3156                 retval = journal_join(&myth, sb, 1);
3157                 if (retval)
3158                         goto out_fail;
3159
3160                 /* someone might have ended the transaction while we joined */
3161                 if (old_trans_id != journal->j_trans_id) {
3162                         retval = do_journal_end(&myth, sb, 1, 0);
3163                 } else {
3164                         retval = do_journal_end(&myth, sb, 1, COMMIT_NOW);
3165                 }
3166
3167                 if (retval)
3168                         goto out_fail;
3169
3170                 PROC_INFO_INC(sb, journal.journal_relock_wcount);
3171                 goto relock;
3172         }
3173         /* we are the first writer, set trans_id */
3174         if (journal->j_trans_start_time == 0) {
3175                 journal->j_trans_start_time = get_seconds();
3176         }
3177         atomic_inc(&(journal->j_wcount));
3178         journal->j_len_alloc += nblocks;
3179         th->t_blocks_logged = 0;
3180         th->t_blocks_allocated = nblocks;
3181         th->t_trans_id = journal->j_trans_id;
3182         unlock_journal(sb);
3183         INIT_LIST_HEAD(&th->t_list);
3184         get_fs_excl();
3185         return 0;
3186
3187       out_fail:
3188         memset(th, 0, sizeof(*th));
3189         /* Re-set th->t_super, so we can properly keep track of how many
3190          * persistent transactions there are. We need to do this so if this
3191          * call is part of a failed restart_transaction, we can free it later */
3192         th->t_super = sb;
3193         return retval;
3194 }
3195
3196 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3197                                                                     super_block
3198                                                                     *s,
3199                                                                     int nblocks)
3200 {
3201         int ret;
3202         struct reiserfs_transaction_handle *th;
3203
3204         /* if we're nesting into an existing transaction.  It will be
3205          ** persistent on its own
3206          */
3207         if (reiserfs_transaction_running(s)) {
3208                 th = current->journal_info;
3209                 th->t_refcount++;
3210                 BUG_ON(th->t_refcount < 2);
3211                 
3212                 return th;
3213         }
3214         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3215         if (!th)
3216                 return NULL;
3217         ret = journal_begin(th, s, nblocks);
3218         if (ret) {
3219                 kfree(th);
3220                 return NULL;
3221         }
3222
3223         SB_JOURNAL(s)->j_persistent_trans++;
3224         return th;
3225 }
3226
3227 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3228 {
3229         struct super_block *s = th->t_super;
3230         int ret = 0;
3231         if (th->t_trans_id)
3232                 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3233         else
3234                 ret = -EIO;
3235         if (th->t_refcount == 0) {
3236                 SB_JOURNAL(s)->j_persistent_trans--;
3237                 kfree(th);
3238         }
3239         return ret;
3240 }
3241
3242 static int journal_join(struct reiserfs_transaction_handle *th,
3243                         struct super_block *sb, unsigned long nblocks)
3244 {
3245         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3246
3247         /* this keeps do_journal_end from NULLing out the current->journal_info
3248          ** pointer
3249          */
3250         th->t_handle_save = cur_th;
3251         BUG_ON(cur_th && cur_th->t_refcount > 1);
3252         return do_journal_begin_r(th, sb, nblocks, JBEGIN_JOIN);
3253 }
3254
3255 int journal_join_abort(struct reiserfs_transaction_handle *th,
3256                        struct super_block *sb, unsigned long nblocks)
3257 {
3258         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3259
3260         /* this keeps do_journal_end from NULLing out the current->journal_info
3261          ** pointer
3262          */
3263         th->t_handle_save = cur_th;
3264         BUG_ON(cur_th && cur_th->t_refcount > 1);
3265         return do_journal_begin_r(th, sb, nblocks, JBEGIN_ABORT);
3266 }
3267
3268 int journal_begin(struct reiserfs_transaction_handle *th,
3269                   struct super_block *sb, unsigned long nblocks)
3270 {
3271         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3272         int ret;
3273
3274         th->t_handle_save = NULL;
3275         if (cur_th) {
3276                 /* we are nesting into the current transaction */
3277                 if (cur_th->t_super == sb) {
3278                         BUG_ON(!cur_th->t_refcount);
3279                         cur_th->t_refcount++;
3280                         memcpy(th, cur_th, sizeof(*th));
3281                         if (th->t_refcount <= 1)
3282                                 reiserfs_warning(sb, "reiserfs-2005",
3283                                                  "BAD: refcount <= 1, but "
3284                                                  "journal_info != 0");
3285                         return 0;
3286                 } else {
3287                         /* we've ended up with a handle from a different filesystem.
3288                          ** save it and restore on journal_end.  This should never
3289                          ** really happen...
3290                          */
3291                         reiserfs_warning(sb, "clm-2100",
3292                                          "nesting info a different FS");
3293                         th->t_handle_save = current->journal_info;
3294                         current->journal_info = th;
3295                 }
3296         } else {
3297                 current->journal_info = th;
3298         }
3299         ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3300         BUG_ON(current->journal_info != th);
3301
3302         /* I guess this boils down to being the reciprocal of clm-2100 above.
3303          * If do_journal_begin_r fails, we need to put it back, since journal_end
3304          * won't be called to do it. */
3305         if (ret)
3306                 current->journal_info = th->t_handle_save;
3307         else
3308                 BUG_ON(!th->t_refcount);
3309
3310         return ret;
3311 }
3312
3313 /*
3314 ** puts bh into the current transaction.  If it was already there, reorders removes the
3315 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3316 **
3317 ** if it was dirty, cleans and files onto the clean list.  I can't let it be dirty again until the
3318 ** transaction is committed.
3319 **
3320 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3321 */
3322 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3323                        struct super_block *sb, struct buffer_head *bh)
3324 {
3325         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3326         struct reiserfs_journal_cnode *cn = NULL;
3327         int count_already_incd = 0;
3328         int prepared = 0;
3329         BUG_ON(!th->t_trans_id);
3330
3331         PROC_INFO_INC(sb, journal.mark_dirty);
3332         if (th->t_trans_id != journal->j_trans_id) {
3333                 reiserfs_panic(th->t_super, "journal-1577",
3334                                "handle trans id %ld != current trans id %ld",
3335                                th->t_trans_id, journal->j_trans_id);
3336         }
3337
3338         sb->s_dirt = 1;
3339
3340         prepared = test_clear_buffer_journal_prepared(bh);
3341         clear_buffer_journal_restore_dirty(bh);
3342         /* already in this transaction, we are done */
3343         if (buffer_journaled(bh)) {
3344                 PROC_INFO_INC(sb, journal.mark_dirty_already);
3345                 return 0;
3346         }
3347
3348         /* this must be turned into a panic instead of a warning.  We can't allow
3349          ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3350          ** could get to disk too early.  NOT GOOD.
3351          */
3352         if (!prepared || buffer_dirty(bh)) {
3353                 reiserfs_warning(sb, "journal-1777",
3354                                  "buffer %llu bad state "
3355                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3356                                  (unsigned long long)bh->b_blocknr,
3357                                  prepared ? ' ' : '!',
3358                                  buffer_locked(bh) ? ' ' : '!',
3359                                  buffer_dirty(bh) ? ' ' : '!',
3360                                  buffer_journal_dirty(bh) ? ' ' : '!');
3361         }
3362
3363         if (atomic_read(&(journal->j_wcount)) <= 0) {
3364                 reiserfs_warning(sb, "journal-1409",
3365                                  "returning because j_wcount was %d",
3366                                  atomic_read(&(journal->j_wcount)));
3367                 return 1;
3368         }
3369         /* this error means I've screwed up, and we've overflowed the transaction.
3370          ** Nothing can be done here, except make the FS readonly or panic.
3371          */
3372         if (journal->j_len >= journal->j_trans_max) {
3373                 reiserfs_panic(th->t_super, "journal-1413",
3374                                "j_len (%lu) is too big",
3375                                journal->j_len);
3376         }
3377
3378         if (buffer_journal_dirty(bh)) {
3379                 count_already_incd = 1;
3380                 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3381                 clear_buffer_journal_dirty(bh);
3382         }
3383
3384         if (journal->j_len > journal->j_len_alloc) {
3385                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3386         }
3387
3388         set_buffer_journaled(bh);
3389
3390         /* now put this guy on the end */
3391         if (!cn) {
3392                 cn = get_cnode(sb);
3393                 if (!cn) {
3394                         reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3395                 }
3396
3397                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3398                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3399                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3400                 }
3401                 th->t_blocks_logged++;
3402                 journal->j_len++;
3403
3404                 cn->bh = bh;
3405                 cn->blocknr = bh->b_blocknr;
3406                 cn->sb = sb;
3407                 cn->jlist = NULL;
3408                 insert_journal_hash(journal->j_hash_table, cn);
3409                 if (!count_already_incd) {
3410                         get_bh(bh);
3411                 }
3412         }
3413         cn->next = NULL;
3414         cn->prev = journal->j_last;
3415         cn->bh = bh;
3416         if (journal->j_last) {
3417                 journal->j_last->next = cn;
3418                 journal->j_last = cn;
3419         } else {
3420                 journal->j_first = cn;
3421                 journal->j_last = cn;
3422         }
3423         return 0;
3424 }
3425
3426 int journal_end(struct reiserfs_transaction_handle *th,
3427                 struct super_block *sb, unsigned long nblocks)
3428 {
3429         if (!current->journal_info && th->t_refcount > 1)
3430                 reiserfs_warning(sb, "REISER-NESTING",
3431                                  "th NULL, refcount %d", th->t_refcount);
3432
3433         if (!th->t_trans_id) {
3434                 WARN_ON(1);
3435                 return -EIO;
3436         }
3437
3438         th->t_refcount--;
3439         if (th->t_refcount > 0) {
3440                 struct reiserfs_transaction_handle *cur_th =
3441                     current->journal_info;
3442
3443                 /* we aren't allowed to close a nested transaction on a different
3444                  ** filesystem from the one in the task struct
3445                  */
3446                 BUG_ON(cur_th->t_super != th->t_super);
3447
3448                 if (th != cur_th) {
3449                         memcpy(current->journal_info, th, sizeof(*th));
3450                         th->t_trans_id = 0;
3451                 }
3452                 return 0;
3453         } else {
3454                 return do_journal_end(th, sb, nblocks, 0);
3455         }
3456 }
3457
3458 /* removes from the current transaction, relsing and descrementing any counters.
3459 ** also files the removed buffer directly onto the clean list
3460 **
3461 ** called by journal_mark_freed when a block has been deleted
3462 **
3463 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3464 */
3465 static int remove_from_transaction(struct super_block *sb,
3466                                    b_blocknr_t blocknr, int already_cleaned)
3467 {
3468         struct buffer_head *bh;
3469         struct reiserfs_journal_cnode *cn;
3470         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3471         int ret = 0;
3472
3473         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3474         if (!cn || !cn->bh) {
3475                 return ret;
3476         }
3477         bh = cn->bh;
3478         if (cn->prev) {
3479                 cn->prev->next = cn->next;
3480         }
3481         if (cn->next) {
3482                 cn->next->prev = cn->prev;
3483         }
3484         if (cn == journal->j_first) {
3485                 journal->j_first = cn->next;
3486         }
3487         if (cn == journal->j_last) {
3488                 journal->j_last = cn->prev;
3489         }
3490         if (bh)
3491                 remove_journal_hash(sb, journal->j_hash_table, NULL,
3492                                     bh->b_blocknr, 0);
3493         clear_buffer_journaled(bh);     /* don't log this one */
3494
3495         if (!already_cleaned) {
3496                 clear_buffer_journal_dirty(bh);
3497                 clear_buffer_dirty(bh);
3498                 clear_buffer_journal_test(bh);
3499                 put_bh(bh);
3500                 if (atomic_read(&(bh->b_count)) < 0) {
3501                         reiserfs_warning(sb, "journal-1752",
3502                                          "b_count < 0");
3503                 }
3504                 ret = 1;
3505         }
3506         journal->j_len--;
3507         journal->j_len_alloc--;
3508         free_cnode(sb, cn);
3509         return ret;
3510 }
3511
3512 /*
3513 ** for any cnode in a journal list, it can only be dirtied of all the
3514 ** transactions that include it are committed to disk.
3515 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3516 ** and 0 if you aren't
3517 **
3518 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3519 ** blocks for a given transaction on disk
3520 **
3521 */
3522 static int can_dirty(struct reiserfs_journal_cnode *cn)
3523 {
3524         struct super_block *sb = cn->sb;
3525         b_blocknr_t blocknr = cn->blocknr;
3526         struct reiserfs_journal_cnode *cur = cn->hprev;
3527         int can_dirty = 1;
3528
3529         /* first test hprev.  These are all newer than cn, so any node here
3530          ** with the same block number and dev means this node can't be sent
3531          ** to disk right now.
3532          */
3533         while (cur && can_dirty) {
3534                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3535                     cur->blocknr == blocknr) {
3536                         can_dirty = 0;
3537                 }
3538                 cur = cur->hprev;
3539         }
3540         /* then test hnext.  These are all older than cn.  As long as they
3541          ** are committed to the log, it is safe to write cn to disk
3542          */
3543         cur = cn->hnext;
3544         while (cur && can_dirty) {
3545                 if (cur->jlist && cur->jlist->j_len > 0 &&
3546                     atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3547                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3548                         can_dirty = 0;
3549                 }
3550                 cur = cur->hnext;
3551         }
3552         return can_dirty;
3553 }
3554
3555 /* syncs the commit blocks, but does not force the real buffers to disk
3556 ** will wait until the current transaction is done/committed before returning
3557 */
3558 int journal_end_sync(struct reiserfs_transaction_handle *th,
3559                      struct super_block *sb, unsigned long nblocks)
3560 {
3561         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3562
3563         BUG_ON(!th->t_trans_id);
3564         /* you can sync while nested, very, very bad */
3565         BUG_ON(th->t_refcount > 1);
3566         if (journal->j_len == 0) {
3567                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3568                                              1);
3569                 journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
3570         }
3571         return do_journal_end(th, sb, nblocks, COMMIT_NOW | WAIT);
3572 }
3573
3574 /*
3575 ** writeback the pending async commits to disk
3576 */
3577 static void flush_async_commits(struct work_struct *work)
3578 {
3579         struct reiserfs_journal *journal =
3580                 container_of(work, struct reiserfs_journal, j_work.work);
3581         struct super_block *sb = journal->j_work_sb;
3582         struct reiserfs_journal_list *jl;
3583         struct list_head *entry;
3584
3585         reiserfs_write_lock(sb);
3586         if (!list_empty(&journal->j_journal_list)) {
3587                 /* last entry is the youngest, commit it and you get everything */
3588                 entry = journal->j_journal_list.prev;
3589                 jl = JOURNAL_LIST_ENTRY(entry);
3590                 flush_commit_list(sb, jl, 1);
3591         }
3592         reiserfs_write_unlock(sb);
3593 }
3594
3595 /*
3596 ** flushes any old transactions to disk
3597 ** ends the current transaction if it is too old
3598 */
3599 int reiserfs_flush_old_commits(struct super_block *sb)
3600 {
3601         time_t now;
3602         struct reiserfs_transaction_handle th;
3603         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3604
3605         now = get_seconds();
3606         /* safety check so we don't flush while we are replaying the log during
3607          * mount
3608          */
3609         if (list_empty(&journal->j_journal_list)) {
3610                 return 0;
3611         }
3612
3613         /* check the current transaction.  If there are no writers, and it is
3614          * too old, finish it, and force the commit blocks to disk
3615          */
3616         if (atomic_read(&journal->j_wcount) <= 0 &&
3617             journal->j_trans_start_time > 0 &&
3618             journal->j_len > 0 &&
3619             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3620                 if (!journal_join(&th, sb, 1)) {
3621                         reiserfs_prepare_for_journal(sb,
3622                                                      SB_BUFFER_WITH_SB(sb),
3623                                                      1);
3624                         journal_mark_dirty(&th, sb,
3625                                            SB_BUFFER_WITH_SB(sb));
3626
3627                         /* we're only being called from kreiserfsd, it makes no sense to do
3628                          ** an async commit so that kreiserfsd can do it later
3629                          */
3630                         do_journal_end(&th, sb, 1, COMMIT_NOW | WAIT);
3631                 }
3632         }
3633         return sb->s_dirt;
3634 }
3635
3636 /*
3637 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3638 **
3639 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3640 ** the writers are done.  By the time it wakes up, the transaction it was called has already ended, so it just
3641 ** flushes the commit list and returns 0.
3642 **
3643 ** Won't batch when flush or commit_now is set.  Also won't batch when others are waiting on j_join_wait.
3644 **
3645 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3646 */
3647 static int check_journal_end(struct reiserfs_transaction_handle *th,
3648                              struct super_block *sb, unsigned long nblocks,
3649                              int flags)
3650 {
3651
3652         time_t now;
3653         int flush = flags & FLUSH_ALL;
3654         int commit_now = flags & COMMIT_NOW;
3655         int wait_on_commit = flags & WAIT;
3656         struct reiserfs_journal_list *jl;
3657         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3658
3659         BUG_ON(!th->t_trans_id);
3660
3661         if (th->t_trans_id != journal->j_trans_id) {
3662                 reiserfs_panic(th->t_super, "journal-1577",
3663                                "handle trans id %ld != current trans id %ld",
3664                                th->t_trans_id, journal->j_trans_id);
3665         }
3666
3667         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3668         if (atomic_read(&(journal->j_wcount)) > 0) {    /* <= 0 is allowed.  unmounting might not call begin */
3669                 atomic_dec(&(journal->j_wcount));
3670         }
3671
3672         /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3673          ** will be dealt with by next transaction that actually writes something, but should be taken
3674          ** care of in this trans
3675          */
3676         BUG_ON(journal->j_len == 0);
3677
3678         /* if wcount > 0, and we are called to with flush or commit_now,
3679          ** we wait on j_join_wait.  We will wake up when the last writer has
3680          ** finished the transaction, and started it on its way to the disk.
3681          ** Then, we flush the commit or journal list, and just return 0
3682          ** because the rest of journal end was already done for this transaction.
3683          */
3684         if (atomic_read(&(journal->j_wcount)) > 0) {
3685                 if (flush || commit_now) {
3686                         unsigned trans_id;
3687
3688                         jl = journal->j_current_jl;
3689                         trans_id = jl->j_trans_id;
3690                         if (wait_on_commit)
3691                                 jl->j_state |= LIST_COMMIT_PENDING;
3692                         atomic_set(&(journal->j_jlock), 1);
3693                         if (flush) {
3694                                 journal->j_next_full_flush = 1;
3695                         }
3696                         unlock_journal(sb);
3697
3698                         /* sleep while the current transaction is still j_jlocked */
3699                         while (journal->j_trans_id == trans_id) {
3700                                 if (atomic_read(&journal->j_jlock)) {
3701                                         queue_log_writer(sb);
3702                                 } else {
3703                                         lock_journal(sb);
3704                                         if (journal->j_trans_id == trans_id) {
3705                                                 atomic_set(&(journal->j_jlock),
3706                                                            1);
3707                                         }
3708                                         unlock_journal(sb);
3709                                 }
3710                         }
3711                         BUG_ON(journal->j_trans_id == trans_id);
3712                         
3713                         if (commit_now
3714                             && journal_list_still_alive(sb, trans_id)
3715                             && wait_on_commit) {
3716                                 flush_commit_list(sb, jl, 1);
3717                         }
3718                         return 0;
3719                 }
3720                 unlock_journal(sb);
3721                 return 0;
3722         }
3723
3724         /* deal with old transactions where we are the last writers */
3725         now = get_seconds();
3726         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3727                 commit_now = 1;
3728                 journal->j_next_async_flush = 1;
3729         }
3730         /* don't batch when someone is waiting on j_join_wait */
3731         /* don't batch when syncing the commit or flushing the whole trans */
3732         if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3733             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3734             && journal->j_len_alloc < journal->j_max_batch
3735             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3736                 journal->j_bcount++;
3737                 unlock_journal(sb);
3738                 return 0;
3739         }
3740
3741         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3742                 reiserfs_panic(sb, "journal-003",
3743                                "j_start (%ld) is too high",
3744                                journal->j_start);
3745         }
3746         return 1;
3747 }
3748
3749 /*
3750 ** Does all the work that makes deleting blocks safe.
3751 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3752 **
3753 ** otherwise:
3754 ** set a bit for the block in the journal bitmap.  That will prevent it from being allocated for unformatted nodes
3755 ** before this transaction has finished.
3756 **
3757 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.  That will prevent any old transactions with
3758 ** this block from trying to flush to the real location.  Since we aren't removing the cnode from the journal_list_hash,
3759 ** the block can't be reallocated yet.
3760 **
3761 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3762 */
3763 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3764                        struct super_block *sb, b_blocknr_t blocknr)
3765 {
3766         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3767         struct reiserfs_journal_cnode *cn = NULL;
3768         struct buffer_head *bh = NULL;
3769         struct reiserfs_list_bitmap *jb = NULL;
3770         int cleaned = 0;
3771         BUG_ON(!th->t_trans_id);
3772
3773         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3774         if (cn && cn->bh) {
3775                 bh = cn->bh;
3776                 get_bh(bh);
3777         }
3778         /* if it is journal new, we just remove it from this transaction */
3779         if (bh && buffer_journal_new(bh)) {
3780                 clear_buffer_journal_new(bh);
3781                 clear_prepared_bits(bh);
3782                 reiserfs_clean_and_file_buffer(bh);
3783                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3784         } else {
3785                 /* set the bit for this block in the journal bitmap for this transaction */
3786                 jb = journal->j_current_jl->j_list_bitmap;
3787                 if (!jb) {
3788                         reiserfs_panic(sb, "journal-1702",
3789                                        "journal_list_bitmap is NULL");
3790                 }
3791                 set_bit_in_list_bitmap(sb, blocknr, jb);
3792
3793                 /* Note, the entire while loop is not allowed to schedule.  */
3794
3795                 if (bh) {
3796                         clear_prepared_bits(bh);
3797                         reiserfs_clean_and_file_buffer(bh);
3798                 }
3799                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3800
3801                 /* find all older transactions with this block, make sure they don't try to write it out */
3802                 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3803                                           blocknr);
3804                 while (cn) {
3805                         if (sb == cn->sb && blocknr == cn->blocknr) {
3806                                 set_bit(BLOCK_FREED, &cn->state);
3807                                 if (cn->bh) {
3808                                         if (!cleaned) {
3809                                                 /* remove_from_transaction will brelse the buffer if it was 
3810                                                  ** in the current trans
3811                                                  */
3812                                                 clear_buffer_journal_dirty(cn->
3813                                                                            bh);
3814                                                 clear_buffer_dirty(cn->bh);
3815                                                 clear_buffer_journal_test(cn->
3816                                                                           bh);
3817                                                 cleaned = 1;
3818                                                 put_bh(cn->bh);
3819                                                 if (atomic_read
3820                                                     (&(cn->bh->b_count)) < 0) {
3821                                                         reiserfs_warning(sb,
3822                                                                  "journal-2138",
3823                                                                  "cn->bh->b_count < 0");
3824                                                 }
3825                                         }
3826                                         if (cn->jlist) {        /* since we are clearing the bh, we MUST dec nonzerolen */
3827                                                 atomic_dec(&
3828                                                            (cn->jlist->
3829                                                             j_nonzerolen));
3830                                         }
3831                                         cn->bh = NULL;
3832                                 }
3833                         }
3834                         cn = cn->hnext;
3835                 }
3836         }
3837
3838         if (bh)
3839                 release_buffer_page(bh); /* get_hash grabs the buffer */
3840         return 0;
3841 }
3842
3843 void reiserfs_update_inode_transaction(struct inode *inode)
3844 {
3845         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3846         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3847         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3848 }
3849
3850 /*
3851  * returns -1 on error, 0 if no commits/barriers were done and 1
3852  * if a transaction was actually committed and the barrier was done
3853  */
3854 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3855                              struct reiserfs_journal_list *jl)
3856 {
3857         struct reiserfs_transaction_handle th;
3858         struct super_block *sb = inode->i_sb;
3859         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3860         int ret = 0;
3861
3862         /* is it from the current transaction, or from an unknown transaction? */
3863         if (id == journal->j_trans_id) {
3864                 jl = journal->j_current_jl;
3865                 /* try to let other writers come in and grow this transaction */
3866                 let_transaction_grow(sb, id);
3867                 if (journal->j_trans_id != id) {
3868                         goto flush_commit_only;
3869                 }
3870
3871                 ret = journal_begin(&th, sb, 1);
3872                 if (ret)
3873                         return ret;
3874
3875                 /* someone might have ended this transaction while we joined */
3876                 if (journal->j_trans_id != id) {
3877                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3878                                                      1);
3879                         journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3880                         ret = journal_end(&th, sb, 1);
3881                         goto flush_commit_only;
3882                 }
3883
3884                 ret = journal_end_sync(&th, sb, 1);
3885                 if (!ret)
3886                         ret = 1;
3887
3888         } else {
3889                 /* this gets tricky, we have to make sure the journal list in
3890                  * the inode still exists.  We know the list is still around
3891                  * if we've got a larger transaction id than the oldest list
3892                  */
3893               flush_commit_only:
3894                 if (journal_list_still_alive(inode->i_sb, id)) {
3895                         /*
3896                          * we only set ret to 1 when we know for sure
3897                          * the barrier hasn't been started yet on the commit
3898                          * block.
3899                          */
3900                         if (atomic_read(&jl->j_commit_left) > 1)
3901                                 ret = 1;
3902                         flush_commit_list(sb, jl, 1);
3903                         if (journal->j_errno)
3904                                 ret = journal->j_errno;
3905                 }
3906         }
3907         /* otherwise the list is gone, and long since committed */
3908         return ret;
3909 }
3910
3911 int reiserfs_commit_for_inode(struct inode *inode)
3912 {
3913         unsigned int id = REISERFS_I(inode)->i_trans_id;
3914         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3915
3916         /* for the whole inode, assume unset id means it was
3917          * changed in the current transaction.  More conservative
3918          */
3919         if (!id || !jl) {
3920                 reiserfs_update_inode_transaction(inode);
3921                 id = REISERFS_I(inode)->i_trans_id;
3922                 /* jl will be updated in __commit_trans_jl */
3923         }
3924
3925         return __commit_trans_jl(inode, id, jl);
3926 }
3927
3928 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3929                                       struct buffer_head *bh)
3930 {
3931         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3932         PROC_INFO_INC(sb, journal.restore_prepared);
3933         if (!bh) {
3934                 return;
3935         }
3936         if (test_clear_buffer_journal_restore_dirty(bh) &&
3937             buffer_journal_dirty(bh)) {
3938                 struct reiserfs_journal_cnode *cn;
3939                 cn = get_journal_hash_dev(sb,
3940                                           journal->j_list_hash_table,
3941                                           bh->b_blocknr);
3942                 if (cn && can_dirty(cn)) {
3943                         set_buffer_journal_test(bh);
3944                         mark_buffer_dirty(bh);
3945                 }
3946         }
3947         clear_buffer_journal_prepared(bh);
3948 }
3949
3950 extern struct tree_balance *cur_tb;
3951 /*
3952 ** before we can change a metadata block, we have to make sure it won't
3953 ** be written to disk while we are altering it.  So, we must:
3954 ** clean it
3955 ** wait on it.
3956 **
3957 */
3958 int reiserfs_prepare_for_journal(struct super_block *sb,
3959                                  struct buffer_head *bh, int wait)
3960 {
3961         PROC_INFO_INC(sb, journal.prepare);
3962
3963         if (!trylock_buffer(bh)) {
3964                 if (!wait)
3965                         return 0;
3966                 lock_buffer(bh);
3967         }
3968         set_buffer_journal_prepared(bh);
3969         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3970                 clear_buffer_journal_test(bh);
3971                 set_buffer_journal_restore_dirty(bh);
3972         }
3973         unlock_buffer(bh);
3974         return 1;
3975 }
3976
3977 static void flush_old_journal_lists(struct super_block *s)
3978 {
3979         struct reiserfs_journal *journal = SB_JOURNAL(s);
3980         struct reiserfs_journal_list *jl;
3981         struct list_head *entry;
3982         time_t now = get_seconds();
3983
3984         while (!list_empty(&journal->j_journal_list)) {
3985                 entry = journal->j_journal_list.next;
3986                 jl = JOURNAL_LIST_ENTRY(entry);
3987                 /* this check should always be run, to send old lists to disk */
3988                 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) &&
3989                     atomic_read(&jl->j_commit_left) == 0 &&
3990                     test_transaction(s, jl)) {
3991                         flush_used_journal_lists(s, jl);
3992                 } else {
3993                         break;
3994                 }
3995         }
3996 }
3997
3998 /*
3999 ** long and ugly.  If flush, will not return until all commit
4000 ** blocks and all real buffers in the trans are on disk.
4001 ** If no_async, won't return until all commit blocks are on disk.
4002 **
4003 ** keep reading, there are comments as you go along
4004 **
4005 ** If the journal is aborted, we just clean up. Things like flushing
4006 ** journal lists, etc just won't happen.
4007 */
4008 static int do_journal_end(struct reiserfs_transaction_handle *th,
4009                           struct super_block *sb, unsigned long nblocks,
4010                           int flags)
4011 {
4012         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4013         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
4014         struct reiserfs_journal_cnode *last_cn = NULL;
4015         struct reiserfs_journal_desc *desc;
4016         struct reiserfs_journal_commit *commit;
4017         struct buffer_head *c_bh;       /* commit bh */
4018         struct buffer_head *d_bh;       /* desc bh */
4019         int cur_write_start = 0;        /* start index of current log write */
4020         int old_start;
4021         int i;
4022         int flush;
4023         int wait_on_commit;
4024         struct reiserfs_journal_list *jl, *temp_jl;
4025         struct list_head *entry, *safe;
4026         unsigned long jindex;
4027         unsigned int commit_trans_id;
4028         int trans_half;
4029
4030         BUG_ON(th->t_refcount > 1);
4031         BUG_ON(!th->t_trans_id);
4032
4033         /* protect flush_older_commits from doing mistakes if the
4034            transaction ID counter gets overflowed.  */
4035         if (th->t_trans_id == ~0U)
4036                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
4037         flush = flags & FLUSH_ALL;
4038         wait_on_commit = flags & WAIT;
4039
4040         put_fs_excl();
4041         current->journal_info = th->t_handle_save;
4042         reiserfs_check_lock_depth(sb, "journal end");
4043         if (journal->j_len == 0) {
4044                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
4045                                              1);
4046                 journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
4047         }
4048
4049         lock_journal(sb);
4050         if (journal->j_next_full_flush) {
4051                 flags |= FLUSH_ALL;
4052                 flush = 1;
4053         }
4054         if (journal->j_next_async_flush) {
4055                 flags |= COMMIT_NOW | WAIT;
4056                 wait_on_commit = 1;
4057         }
4058
4059         /* check_journal_end locks the journal, and unlocks if it does not return 1
4060          ** it tells us if we should continue with the journal_end, or just return
4061          */
4062         if (!check_journal_end(th, sb, nblocks, flags)) {
4063                 sb->s_dirt = 1;
4064                 wake_queued_writers(sb);
4065                 reiserfs_async_progress_wait(sb);
4066                 goto out;
4067         }
4068
4069         /* check_journal_end might set these, check again */
4070         if (journal->j_next_full_flush) {
4071                 flush = 1;
4072         }
4073
4074         /*
4075          ** j must wait means we have to flush the log blocks, and the real blocks for
4076          ** this transaction
4077          */
4078         if (journal->j_must_wait > 0) {
4079                 flush = 1;
4080         }
4081 #ifdef REISERFS_PREALLOCATE
4082         /* quota ops might need to nest, setup the journal_info pointer for them
4083          * and raise the refcount so that it is > 0. */
4084         current->journal_info = th;
4085         th->t_refcount++;
4086         reiserfs_discard_all_prealloc(th);      /* it should not involve new blocks into
4087                                                  * the transaction */
4088         th->t_refcount--;
4089         current->journal_info = th->t_handle_save;
4090 #endif
4091
4092         /* setup description block */
4093         d_bh =
4094             journal_getblk(sb,
4095                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4096                            journal->j_start);
4097         set_buffer_uptodate(d_bh);
4098         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4099         memset(d_bh->b_data, 0, d_bh->b_size);
4100         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4101         set_desc_trans_id(desc, journal->j_trans_id);
4102
4103         /* setup commit block.  Don't write (keep it clean too) this one until after everyone else is written */
4104         c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4105                               ((journal->j_start + journal->j_len +
4106                                 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4107         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4108         memset(c_bh->b_data, 0, c_bh->b_size);
4109         set_commit_trans_id(commit, journal->j_trans_id);
4110         set_buffer_uptodate(c_bh);
4111
4112         /* init this journal list */
4113         jl = journal->j_current_jl;
4114
4115         /* we lock the commit before doing anything because
4116          * we want to make sure nobody tries to run flush_commit_list until
4117          * the new transaction is fully setup, and we've already flushed the
4118          * ordered bh list
4119          */
4120         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4121
4122         /* save the transaction id in case we need to commit it later */
4123         commit_trans_id = jl->j_trans_id;
4124
4125         atomic_set(&jl->j_older_commits_done, 0);
4126         jl->j_trans_id = journal->j_trans_id;
4127         jl->j_timestamp = journal->j_trans_start_time;
4128         jl->j_commit_bh = c_bh;
4129         jl->j_start = journal->j_start;
4130         jl->j_len = journal->j_len;
4131         atomic_set(&jl->j_nonzerolen, journal->j_len);
4132         atomic_set(&jl->j_commit_left, journal->j_len + 2);
4133         jl->j_realblock = NULL;
4134
4135         /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4136          **  for each real block, add it to the journal list hash,
4137          ** copy into real block index array in the commit or desc block
4138          */
4139         trans_half = journal_trans_half(sb->s_blocksize);
4140         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4141                 if (buffer_journaled(cn->bh)) {
4142                         jl_cn = get_cnode(sb);
4143                         if (!jl_cn) {
4144                                 reiserfs_panic(sb, "journal-1676",
4145                                                "get_cnode returned NULL");
4146                         }
4147                         if (i == 0) {
4148                                 jl->j_realblock = jl_cn;
4149                         }
4150                         jl_cn->prev = last_cn;
4151                         jl_cn->next = NULL;
4152                         if (last_cn) {
4153                                 last_cn->next = jl_cn;
4154                         }
4155                         last_cn = jl_cn;
4156                         /* make sure the block we are trying to log is not a block
4157                            of journal or reserved area */
4158
4159                         if (is_block_in_log_or_reserved_area
4160                             (sb, cn->bh->b_blocknr)) {
4161                                 reiserfs_panic(sb, "journal-2332",
4162                                                "Trying to log block %lu, "
4163                                                "which is a log block",
4164                                                cn->bh->b_blocknr);
4165                         }
4166                         jl_cn->blocknr = cn->bh->b_blocknr;
4167                         jl_cn->state = 0;
4168                         jl_cn->sb = sb;
4169                         jl_cn->bh = cn->bh;
4170                         jl_cn->jlist = jl;
4171                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4172                         if (i < trans_half) {
4173                                 desc->j_realblock[i] =
4174                                     cpu_to_le32(cn->bh->b_blocknr);
4175                         } else {
4176                                 commit->j_realblock[i - trans_half] =
4177                                     cpu_to_le32(cn->bh->b_blocknr);
4178                         }
4179                 } else {
4180                         i--;
4181                 }
4182         }
4183         set_desc_trans_len(desc, journal->j_len);
4184         set_desc_mount_id(desc, journal->j_mount_id);
4185         set_desc_trans_id(desc, journal->j_trans_id);
4186         set_commit_trans_len(commit, journal->j_len);
4187
4188         /* special check in case all buffers in the journal were marked for not logging */
4189         BUG_ON(journal->j_len == 0);
4190
4191         /* we're about to dirty all the log blocks, mark the description block
4192          * dirty now too.  Don't mark the commit block dirty until all the
4193          * others are on disk
4194          */
4195         mark_buffer_dirty(d_bh);
4196
4197         /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4198         cur_write_start = journal->j_start;
4199         cn = journal->j_first;
4200         jindex = 1;             /* start at one so we don't get the desc again */
4201         while (cn) {
4202                 clear_buffer_journal_new(cn->bh);
4203                 /* copy all the real blocks into log area.  dirty log blocks */
4204                 if (buffer_journaled(cn->bh)) {
4205                         struct buffer_head *tmp_bh;
4206                         char *addr;
4207                         struct page *page;
4208                         tmp_bh =
4209                             journal_getblk(sb,
4210                                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4211                                            ((cur_write_start +
4212                                              jindex) %
4213                                             SB_ONDISK_JOURNAL_SIZE(sb)));
4214                         set_buffer_uptodate(tmp_bh);
4215                         page = cn->bh->b_page;
4216                         addr = kmap(page);
4217                         memcpy(tmp_bh->b_data,
4218                                addr + offset_in_page(cn->bh->b_data),
4219                                cn->bh->b_size);
4220                         kunmap(page);
4221                         mark_buffer_dirty(tmp_bh);
4222                         jindex++;
4223                         set_buffer_journal_dirty(cn->bh);
4224                         clear_buffer_journaled(cn->bh);
4225                 } else {
4226                         /* JDirty cleared sometime during transaction.  don't log this one */
4227                         reiserfs_warning(sb, "journal-2048",
4228                                          "BAD, buffer in journal hash, "
4229                                          "but not JDirty!");
4230                         brelse(cn->bh);
4231                 }
4232                 next = cn->next;
4233                 free_cnode(sb, cn);
4234                 cn = next;
4235                 cond_resched();
4236         }
4237
4238         /* we are done  with both the c_bh and d_bh, but
4239          ** c_bh must be written after all other commit blocks,
4240          ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4241          */
4242
4243         journal->j_current_jl = alloc_journal_list(sb);
4244
4245         /* now it is safe to insert this transaction on the main list */
4246         list_add_tail(&jl->j_list, &journal->j_journal_list);
4247         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4248         journal->j_num_work_lists++;
4249
4250         /* reset journal values for the next transaction */
4251         old_start = journal->j_start;
4252         journal->j_start =
4253             (journal->j_start + journal->j_len +
4254              2) % SB_ONDISK_JOURNAL_SIZE(sb);
4255         atomic_set(&(journal->j_wcount), 0);
4256         journal->j_bcount = 0;
4257         journal->j_last = NULL;
4258         journal->j_first = NULL;
4259         journal->j_len = 0;
4260         journal->j_trans_start_time = 0;
4261         /* check for trans_id overflow */
4262         if (++journal->j_trans_id == 0)
4263                 journal->j_trans_id = 10;
4264         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4265         journal->j_must_wait = 0;
4266         journal->j_len_alloc = 0;
4267         journal->j_next_full_flush = 0;
4268         journal->j_next_async_flush = 0;
4269         init_journal_hash(sb);
4270
4271         // make sure reiserfs_add_jh sees the new current_jl before we
4272         // write out the tails
4273         smp_mb();
4274
4275         /* tail conversion targets have to hit the disk before we end the
4276          * transaction.  Otherwise a later transaction might repack the tail
4277          * before this transaction commits, leaving the data block unflushed and
4278          * clean, if we crash before the later transaction commits, the data block
4279          * is lost.
4280          */
4281         if (!list_empty(&jl->j_tail_bh_list)) {
4282                 reiserfs_write_unlock(sb);
4283                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4284                                       journal, jl, &jl->j_tail_bh_list);
4285                 reiserfs_write_lock(sb);
4286         }
4287         BUG_ON(!list_empty(&jl->j_tail_bh_list));
4288         mutex_unlock(&jl->j_commit_mutex);
4289
4290         /* honor the flush wishes from the caller, simple commits can
4291          ** be done outside the journal lock, they are done below
4292          **
4293          ** if we don't flush the commit list right now, we put it into
4294          ** the work queue so the people waiting on the async progress work
4295          ** queue don't wait for this proc to flush journal lists and such.
4296          */
4297         if (flush) {
4298                 flush_commit_list(sb, jl, 1);
4299                 flush_journal_list(sb, jl, 1);
4300         } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4301                 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4302
4303         /* if the next transaction has any chance of wrapping, flush
4304          ** transactions that might get overwritten.  If any journal lists are very
4305          ** old flush them as well.
4306          */
4307       first_jl:
4308         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4309                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4310                 if (journal->j_start <= temp_jl->j_start) {
4311                         if ((journal->j_start + journal->j_trans_max + 1) >=
4312                             temp_jl->j_start) {
4313                                 flush_used_journal_lists(sb, temp_jl);
4314                                 goto first_jl;
4315                         } else if ((journal->j_start +
4316                                     journal->j_trans_max + 1) <
4317                                    SB_ONDISK_JOURNAL_SIZE(sb)) {
4318                                 /* if we don't cross into the next transaction and we don't
4319                                  * wrap, there is no way we can overlap any later transactions
4320                                  * break now
4321                                  */
4322                                 break;
4323                         }
4324                 } else if ((journal->j_start +
4325                             journal->j_trans_max + 1) >
4326                            SB_ONDISK_JOURNAL_SIZE(sb)) {
4327                         if (((journal->j_start + journal->j_trans_max + 1) %
4328                              SB_ONDISK_JOURNAL_SIZE(sb)) >=
4329                             temp_jl->j_start) {
4330                                 flush_used_journal_lists(sb, temp_jl);
4331                                 goto first_jl;
4332                         } else {
4333                                 /* we don't overlap anything from out start to the end of the
4334                                  * log, and our wrapped portion doesn't overlap anything at
4335                                  * the start of the log.  We can break
4336                                  */
4337                                 break;
4338                         }
4339                 }
4340         }
4341         flush_old_journal_lists(sb);
4342
4343         journal->j_current_jl->j_list_bitmap =
4344             get_list_bitmap(sb, journal->j_current_jl);
4345
4346         if (!(journal->j_current_jl->j_list_bitmap)) {
4347                 reiserfs_panic(sb, "journal-1996",
4348                                "could not get a list bitmap");
4349         }
4350
4351         atomic_set(&(journal->j_jlock), 0);
4352         unlock_journal(sb);
4353         /* wake up any body waiting to join. */
4354         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4355         wake_up(&(journal->j_join_wait));
4356
4357         if (!flush && wait_on_commit &&
4358             journal_list_still_alive(sb, commit_trans_id)) {
4359                 flush_commit_list(sb, jl, 1);
4360         }
4361       out:
4362         reiserfs_check_lock_depth(sb, "journal end2");
4363
4364         memset(th, 0, sizeof(*th));
4365         /* Re-set th->t_super, so we can properly keep track of how many
4366          * persistent transactions there are. We need to do this so if this
4367          * call is part of a failed restart_transaction, we can free it later */
4368         th->t_super = sb;
4369
4370         return journal->j_errno;
4371 }
4372
4373 /* Send the file system read only and refuse new transactions */
4374 void reiserfs_abort_journal(struct super_block *sb, int errno)
4375 {
4376         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4377         if (test_bit(J_ABORTED, &journal->j_state))
4378                 return;
4379
4380         if (!journal->j_errno)
4381                 journal->j_errno = errno;
4382
4383         sb->s_flags |= MS_RDONLY;
4384         set_bit(J_ABORTED, &journal->j_state);
4385
4386 #ifdef CONFIG_REISERFS_CHECK
4387         dump_stack();
4388 #endif
4389 }
4390