ad8cbc49883ab7d1e1051f9c72dff211b45a80e8
[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 <asm/uaccess.h>
38 #include <asm/system.h>
39
40 #include <linux/time.h>
41 #include <asm/semaphore.h>
42
43 #include <linux/vmalloc.h>
44 #include <linux/reiserfs_fs.h>
45
46 #include <linux/kernel.h>
47 #include <linux/errno.h>
48 #include <linux/fcntl.h>
49 #include <linux/stat.h>
50 #include <linux/string.h>
51 #include <linux/smp_lock.h>
52 #include <linux/buffer_head.h>
53 #include <linux/workqueue.h>
54 #include <linux/writeback.h>
55 #include <linux/blkdev.h>
56
57 /* gets a struct reiserfs_journal_list * from a list head */
58 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
59                                j_list))
60 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
61                                j_working_list))
62
63 /* the number of mounted filesystems.  This is used to decide when to
64 ** start and kill the commit workqueue
65 */
66 static int reiserfs_mounted_fs_count;
67
68 static struct workqueue_struct *commit_wq;
69
70 #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
71                                    structs at 4k */
72 #define BUFNR 64                /*read ahead */
73
74 /* cnode stat bits.  Move these into reiserfs_fs.h */
75
76 #define BLOCK_FREED 2           /* this block was freed, and can't be written.  */
77 #define BLOCK_FREED_HOLDER 3    /* this block was freed during this transaction, and can't be written */
78
79 #define BLOCK_NEEDS_FLUSH 4     /* used in flush_journal_list */
80 #define BLOCK_DIRTIED 5
81
82 /* journal list state bits */
83 #define LIST_TOUCHED 1
84 #define LIST_DIRTY   2
85 #define LIST_COMMIT_PENDING  4  /* someone will commit this list */
86
87 /* flags for do_journal_end */
88 #define FLUSH_ALL   1           /* flush commit and real blocks */
89 #define COMMIT_NOW  2           /* end and commit this transaction */
90 #define WAIT        4           /* wait for the log blocks to hit the disk */
91
92 static int do_journal_end(struct reiserfs_transaction_handle *,
93                           struct super_block *, unsigned long nblocks,
94                           int flags);
95 static int flush_journal_list(struct super_block *s,
96                               struct reiserfs_journal_list *jl, int flushall);
97 static int flush_commit_list(struct super_block *s,
98                              struct reiserfs_journal_list *jl, int flushall);
99 static int can_dirty(struct reiserfs_journal_cnode *cn);
100 static int journal_join(struct reiserfs_transaction_handle *th,
101                         struct super_block *p_s_sb, unsigned long nblocks);
102 static int release_journal_dev(struct super_block *super,
103                                struct reiserfs_journal *journal);
104 static int dirty_one_transaction(struct super_block *s,
105                                  struct reiserfs_journal_list *jl);
106 static void flush_async_commits(void *p);
107 static void queue_log_writer(struct super_block *s);
108
109 /* values for join in do_journal_begin_r */
110 enum {
111         JBEGIN_REG = 0,         /* regular journal begin */
112         JBEGIN_JOIN = 1,        /* join the running transaction if at all possible */
113         JBEGIN_ABORT = 2,       /* called from cleanup code, ignores aborted flag */
114 };
115
116 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
117                               struct super_block *p_s_sb,
118                               unsigned long nblocks, int join);
119
120 static void init_journal_hash(struct super_block *p_s_sb)
121 {
122         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
123         memset(journal->j_hash_table, 0,
124                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
125 }
126
127 /*
128 ** clears BH_Dirty and sticks the buffer on the clean list.  Called because I can't allow refile_buffer to
129 ** make schedule happen after I've freed a block.  Look at remove_from_transaction and journal_mark_freed for
130 ** more details.
131 */
132 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
133 {
134         if (bh) {
135                 clear_buffer_dirty(bh);
136                 clear_buffer_journal_test(bh);
137         }
138         return 0;
139 }
140
141 static void disable_barrier(struct super_block *s)
142 {
143         REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH);
144         printk("reiserfs: disabling flush barriers on %s\n",
145                reiserfs_bdevname(s));
146 }
147
148 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
149                                                          *p_s_sb)
150 {
151         struct reiserfs_bitmap_node *bn;
152         static int id;
153
154         bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
155         if (!bn) {
156                 return NULL;
157         }
158         bn->data = kzalloc(p_s_sb->s_blocksize, GFP_NOFS);
159         if (!bn->data) {
160                 kfree(bn);
161                 return NULL;
162         }
163         bn->id = id++;
164         INIT_LIST_HEAD(&bn->list);
165         return bn;
166 }
167
168 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *p_s_sb)
169 {
170         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
171         struct reiserfs_bitmap_node *bn = NULL;
172         struct list_head *entry = journal->j_bitmap_nodes.next;
173
174         journal->j_used_bitmap_nodes++;
175       repeat:
176
177         if (entry != &journal->j_bitmap_nodes) {
178                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
179                 list_del(entry);
180                 memset(bn->data, 0, p_s_sb->s_blocksize);
181                 journal->j_free_bitmap_nodes--;
182                 return bn;
183         }
184         bn = allocate_bitmap_node(p_s_sb);
185         if (!bn) {
186                 yield();
187                 goto repeat;
188         }
189         return bn;
190 }
191 static inline void free_bitmap_node(struct super_block *p_s_sb,
192                                     struct reiserfs_bitmap_node *bn)
193 {
194         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
195         journal->j_used_bitmap_nodes--;
196         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
197                 kfree(bn->data);
198                 kfree(bn);
199         } else {
200                 list_add(&bn->list, &journal->j_bitmap_nodes);
201                 journal->j_free_bitmap_nodes++;
202         }
203 }
204
205 static void allocate_bitmap_nodes(struct super_block *p_s_sb)
206 {
207         int i;
208         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
209         struct reiserfs_bitmap_node *bn = NULL;
210         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
211                 bn = allocate_bitmap_node(p_s_sb);
212                 if (bn) {
213                         list_add(&bn->list, &journal->j_bitmap_nodes);
214                         journal->j_free_bitmap_nodes++;
215                 } else {
216                         break;  // this is ok, we'll try again when more are needed 
217                 }
218         }
219 }
220
221 static int set_bit_in_list_bitmap(struct super_block *p_s_sb, int block,
222                                   struct reiserfs_list_bitmap *jb)
223 {
224         int bmap_nr = block / (p_s_sb->s_blocksize << 3);
225         int bit_nr = block % (p_s_sb->s_blocksize << 3);
226
227         if (!jb->bitmaps[bmap_nr]) {
228                 jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_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 *p_s_sb,
235                                 struct reiserfs_list_bitmap *jb)
236 {
237         int i;
238         if (jb->bitmaps == NULL)
239                 return;
240
241         for (i = 0; i < SB_BMAP_NR(p_s_sb); i++) {
242                 if (jb->bitmaps[i]) {
243                         free_bitmap_node(p_s_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 *p_s_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(p_s_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 *p_s_sb)
268 {
269         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
270         struct list_head *next = journal->j_bitmap_nodes.next;
271         struct reiserfs_bitmap_node *bn;
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 *p_s_sb,
290                                    struct reiserfs_list_bitmap *jb_array,
291                                    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(p_s_sb,
304                                          "clm-2000, unable to 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(p_s_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 *p_s_sb,
322                                                     struct reiserfs_journal_list
323                                                     *jl)
324 {
325         int i, j;
326         struct reiserfs_journal *journal = SB_JOURNAL(p_s_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(p_s_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 *p_s_sb)
382 {
383         struct reiserfs_journal_cnode *cn;
384         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
385
386         reiserfs_check_lock_depth(p_s_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 *p_s_sb,
409                        struct reiserfs_journal_cnode *cn)
410 {
411         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
412
413         reiserfs_check_lock_depth(p_s_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 /* utility function to force a BUG if it is called without the big
433 ** kernel lock held.  caller is the string printed just before calling BUG()
434 */
435 void reiserfs_check_lock_depth(struct super_block *sb, char *caller)
436 {
437 #ifdef CONFIG_SMP
438         if (current->lock_depth < 0) {
439                 reiserfs_panic(sb, "%s called without kernel lock held",
440                                caller);
441         }
442 #else
443         ;
444 #endif
445 }
446
447 /* return a cnode with same dev, block number and size in table, or null if not found */
448 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
449                                                                   super_block
450                                                                   *sb,
451                                                                   struct
452                                                                   reiserfs_journal_cnode
453                                                                   **table,
454                                                                   long bl)
455 {
456         struct reiserfs_journal_cnode *cn;
457         cn = journal_hash(table, sb, bl);
458         while (cn) {
459                 if (cn->blocknr == bl && cn->sb == sb)
460                         return cn;
461                 cn = cn->hnext;
462         }
463         return (struct reiserfs_journal_cnode *)0;
464 }
465
466 /*
467 ** this actually means 'can this block be reallocated yet?'.  If you set search_all, a block can only be allocated
468 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
469 ** being overwritten by a replay after crashing.
470 **
471 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction.  Since deleting
472 ** a block removes it from the current transaction, this case should never happen.  If you don't set search_all, make
473 ** sure you never write the block without logging it.
474 **
475 ** next_zero_bit is a suggestion about the next block to try for find_forward.
476 ** when bl is rejected because it is set in a journal list bitmap, we search
477 ** for the next zero bit in the bitmap that rejected bl.  Then, we return that
478 ** through next_zero_bit for find_forward to try.
479 **
480 ** Just because we return something in next_zero_bit does not mean we won't
481 ** reject it on the next call to reiserfs_in_journal
482 **
483 */
484 int reiserfs_in_journal(struct super_block *p_s_sb,
485                         int bmap_nr, int bit_nr, int search_all,
486                         b_blocknr_t * next_zero_bit)
487 {
488         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
489         struct reiserfs_journal_cnode *cn;
490         struct reiserfs_list_bitmap *jb;
491         int i;
492         unsigned long bl;
493
494         *next_zero_bit = 0;     /* always start this at zero. */
495
496         PROC_INFO_INC(p_s_sb, journal.in_journal);
497         /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
498          ** if we crash before the transaction that freed it commits,  this transaction won't
499          ** have committed either, and the block will never be written
500          */
501         if (search_all) {
502                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
503                         PROC_INFO_INC(p_s_sb, journal.in_journal_bitmap);
504                         jb = journal->j_list_bitmap + i;
505                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
506                             test_bit(bit_nr,
507                                      (unsigned long *)jb->bitmaps[bmap_nr]->
508                                      data)) {
509                                 *next_zero_bit =
510                                     find_next_zero_bit((unsigned long *)
511                                                        (jb->bitmaps[bmap_nr]->
512                                                         data),
513                                                        p_s_sb->s_blocksize << 3,
514                                                        bit_nr + 1);
515                                 return 1;
516                         }
517                 }
518         }
519
520         bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr;
521         /* is it in any old transactions? */
522         if (search_all
523             && (cn =
524                 get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, bl))) {
525                 return 1;
526         }
527
528         /* is it in the current transaction.  This should never happen */
529         if ((cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, bl))) {
530                 BUG();
531                 return 1;
532         }
533
534         PROC_INFO_INC(p_s_sb, journal.in_journal_reusable);
535         /* safe for reuse */
536         return 0;
537 }
538
539 /* insert cn into table
540 */
541 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
542                                        struct reiserfs_journal_cnode *cn)
543 {
544         struct reiserfs_journal_cnode *cn_orig;
545
546         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
547         cn->hnext = cn_orig;
548         cn->hprev = NULL;
549         if (cn_orig) {
550                 cn_orig->hprev = cn;
551         }
552         journal_hash(table, cn->sb, cn->blocknr) = cn;
553 }
554
555 /* lock the current transaction */
556 static inline void lock_journal(struct super_block *p_s_sb)
557 {
558         PROC_INFO_INC(p_s_sb, journal.lock_journal);
559         down(&SB_JOURNAL(p_s_sb)->j_lock);
560 }
561
562 /* unlock the current transaction */
563 static inline void unlock_journal(struct super_block *p_s_sb)
564 {
565         up(&SB_JOURNAL(p_s_sb)->j_lock);
566 }
567
568 static inline void get_journal_list(struct reiserfs_journal_list *jl)
569 {
570         jl->j_refcount++;
571 }
572
573 static inline void put_journal_list(struct super_block *s,
574                                     struct reiserfs_journal_list *jl)
575 {
576         if (jl->j_refcount < 1) {
577                 reiserfs_panic(s, "trans id %lu, refcount at %d",
578                                jl->j_trans_id, jl->j_refcount);
579         }
580         if (--jl->j_refcount == 0)
581                 kfree(jl);
582 }
583
584 /*
585 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
586 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
587 ** transaction.
588 */
589 static void cleanup_freed_for_journal_list(struct super_block *p_s_sb,
590                                            struct reiserfs_journal_list *jl)
591 {
592
593         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
594         if (jb) {
595                 cleanup_bitmap_list(p_s_sb, jb);
596         }
597         jl->j_list_bitmap->journal_list = NULL;
598         jl->j_list_bitmap = NULL;
599 }
600
601 static int journal_list_still_alive(struct super_block *s,
602                                     unsigned long trans_id)
603 {
604         struct reiserfs_journal *journal = SB_JOURNAL(s);
605         struct list_head *entry = &journal->j_journal_list;
606         struct reiserfs_journal_list *jl;
607
608         if (!list_empty(entry)) {
609                 jl = JOURNAL_LIST_ENTRY(entry->next);
610                 if (jl->j_trans_id <= trans_id) {
611                         return 1;
612                 }
613         }
614         return 0;
615 }
616
617 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
618 {
619         char b[BDEVNAME_SIZE];
620
621         if (buffer_journaled(bh)) {
622                 reiserfs_warning(NULL,
623                                  "clm-2084: pinned buffer %lu:%s sent to disk",
624                                  bh->b_blocknr, bdevname(bh->b_bdev, b));
625         }
626         if (uptodate)
627                 set_buffer_uptodate(bh);
628         else
629                 clear_buffer_uptodate(bh);
630         unlock_buffer(bh);
631         put_bh(bh);
632 }
633
634 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
635 {
636         if (uptodate)
637                 set_buffer_uptodate(bh);
638         else
639                 clear_buffer_uptodate(bh);
640         unlock_buffer(bh);
641         put_bh(bh);
642 }
643
644 static void submit_logged_buffer(struct buffer_head *bh)
645 {
646         get_bh(bh);
647         bh->b_end_io = reiserfs_end_buffer_io_sync;
648         clear_buffer_journal_new(bh);
649         clear_buffer_dirty(bh);
650         if (!test_clear_buffer_journal_test(bh))
651                 BUG();
652         if (!buffer_uptodate(bh))
653                 BUG();
654         submit_bh(WRITE, bh);
655 }
656
657 static void submit_ordered_buffer(struct buffer_head *bh)
658 {
659         get_bh(bh);
660         bh->b_end_io = reiserfs_end_ordered_io;
661         clear_buffer_dirty(bh);
662         if (!buffer_uptodate(bh))
663                 BUG();
664         submit_bh(WRITE, bh);
665 }
666
667 static int submit_barrier_buffer(struct buffer_head *bh)
668 {
669         get_bh(bh);
670         bh->b_end_io = reiserfs_end_ordered_io;
671         clear_buffer_dirty(bh);
672         if (!buffer_uptodate(bh))
673                 BUG();
674         return submit_bh(WRITE_BARRIER, bh);
675 }
676
677 static void check_barrier_completion(struct super_block *s,
678                                      struct buffer_head *bh)
679 {
680         if (buffer_eopnotsupp(bh)) {
681                 clear_buffer_eopnotsupp(bh);
682                 disable_barrier(s);
683                 set_buffer_uptodate(bh);
684                 set_buffer_dirty(bh);
685                 sync_dirty_buffer(bh);
686         }
687 }
688
689 #define CHUNK_SIZE 32
690 struct buffer_chunk {
691         struct buffer_head *bh[CHUNK_SIZE];
692         int nr;
693 };
694
695 static void write_chunk(struct buffer_chunk *chunk)
696 {
697         int i;
698         get_fs_excl();
699         for (i = 0; i < chunk->nr; i++) {
700                 submit_logged_buffer(chunk->bh[i]);
701         }
702         chunk->nr = 0;
703         put_fs_excl();
704 }
705
706 static void write_ordered_chunk(struct buffer_chunk *chunk)
707 {
708         int i;
709         get_fs_excl();
710         for (i = 0; i < chunk->nr; i++) {
711                 submit_ordered_buffer(chunk->bh[i]);
712         }
713         chunk->nr = 0;
714         put_fs_excl();
715 }
716
717 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
718                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
719 {
720         int ret = 0;
721         BUG_ON(chunk->nr >= CHUNK_SIZE);
722         chunk->bh[chunk->nr++] = bh;
723         if (chunk->nr >= CHUNK_SIZE) {
724                 ret = 1;
725                 if (lock)
726                         spin_unlock(lock);
727                 fn(chunk);
728                 if (lock)
729                         spin_lock(lock);
730         }
731         return ret;
732 }
733
734 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
735 static struct reiserfs_jh *alloc_jh(void)
736 {
737         struct reiserfs_jh *jh;
738         while (1) {
739                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
740                 if (jh) {
741                         atomic_inc(&nr_reiserfs_jh);
742                         return jh;
743                 }
744                 yield();
745         }
746 }
747
748 /*
749  * we want to free the jh when the buffer has been written
750  * and waited on
751  */
752 void reiserfs_free_jh(struct buffer_head *bh)
753 {
754         struct reiserfs_jh *jh;
755
756         jh = bh->b_private;
757         if (jh) {
758                 bh->b_private = NULL;
759                 jh->bh = NULL;
760                 list_del_init(&jh->list);
761                 kfree(jh);
762                 if (atomic_read(&nr_reiserfs_jh) <= 0)
763                         BUG();
764                 atomic_dec(&nr_reiserfs_jh);
765                 put_bh(bh);
766         }
767 }
768
769 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
770                            int tail)
771 {
772         struct reiserfs_jh *jh;
773
774         if (bh->b_private) {
775                 spin_lock(&j->j_dirty_buffers_lock);
776                 if (!bh->b_private) {
777                         spin_unlock(&j->j_dirty_buffers_lock);
778                         goto no_jh;
779                 }
780                 jh = bh->b_private;
781                 list_del_init(&jh->list);
782         } else {
783               no_jh:
784                 get_bh(bh);
785                 jh = alloc_jh();
786                 spin_lock(&j->j_dirty_buffers_lock);
787                 /* buffer must be locked for __add_jh, should be able to have
788                  * two adds at the same time
789                  */
790                 BUG_ON(bh->b_private);
791                 jh->bh = bh;
792                 bh->b_private = jh;
793         }
794         jh->jl = j->j_current_jl;
795         if (tail)
796                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
797         else {
798                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
799         }
800         spin_unlock(&j->j_dirty_buffers_lock);
801         return 0;
802 }
803
804 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
805 {
806         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
807 }
808 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
809 {
810         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
811 }
812
813 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
814 static int write_ordered_buffers(spinlock_t * lock,
815                                  struct reiserfs_journal *j,
816                                  struct reiserfs_journal_list *jl,
817                                  struct list_head *list)
818 {
819         struct buffer_head *bh;
820         struct reiserfs_jh *jh;
821         int ret = j->j_errno;
822         struct buffer_chunk chunk;
823         struct list_head tmp;
824         INIT_LIST_HEAD(&tmp);
825
826         chunk.nr = 0;
827         spin_lock(lock);
828         while (!list_empty(list)) {
829                 jh = JH_ENTRY(list->next);
830                 bh = jh->bh;
831                 get_bh(bh);
832                 if (test_set_buffer_locked(bh)) {
833                         if (!buffer_dirty(bh)) {
834                                 list_move(&jh->list, &tmp);
835                                 goto loop_next;
836                         }
837                         spin_unlock(lock);
838                         if (chunk.nr)
839                                 write_ordered_chunk(&chunk);
840                         wait_on_buffer(bh);
841                         cond_resched();
842                         spin_lock(lock);
843                         goto loop_next;
844                 }
845                 /* in theory, dirty non-uptodate buffers should never get here,
846                  * but the upper layer io error paths still have a few quirks.
847                  * Handle them here as gracefully as we can
848                  */
849                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
850                         clear_buffer_dirty(bh);
851                         ret = -EIO;
852                 }
853                 if (buffer_dirty(bh)) {
854                         list_move(&jh->list, &tmp);
855                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
856                 } else {
857                         reiserfs_free_jh(bh);
858                         unlock_buffer(bh);
859                 }
860               loop_next:
861                 put_bh(bh);
862                 cond_resched_lock(lock);
863         }
864         if (chunk.nr) {
865                 spin_unlock(lock);
866                 write_ordered_chunk(&chunk);
867                 spin_lock(lock);
868         }
869         while (!list_empty(&tmp)) {
870                 jh = JH_ENTRY(tmp.prev);
871                 bh = jh->bh;
872                 get_bh(bh);
873                 reiserfs_free_jh(bh);
874
875                 if (buffer_locked(bh)) {
876                         spin_unlock(lock);
877                         wait_on_buffer(bh);
878                         spin_lock(lock);
879                 }
880                 if (!buffer_uptodate(bh)) {
881                         ret = -EIO;
882                 }
883                 /* ugly interaction with invalidatepage here.
884                  * reiserfs_invalidate_page will pin any buffer that has a valid
885                  * journal head from an older transaction.  If someone else sets
886                  * our buffer dirty after we write it in the first loop, and
887                  * then someone truncates the page away, nobody will ever write
888                  * the buffer. We're safe if we write the page one last time
889                  * after freeing the journal header.
890                  */
891                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
892                         spin_unlock(lock);
893                         ll_rw_block(WRITE, 1, &bh);
894                         spin_lock(lock);
895                 }
896                 put_bh(bh);
897                 cond_resched_lock(lock);
898         }
899         spin_unlock(lock);
900         return ret;
901 }
902
903 static int flush_older_commits(struct super_block *s,
904                                struct reiserfs_journal_list *jl)
905 {
906         struct reiserfs_journal *journal = SB_JOURNAL(s);
907         struct reiserfs_journal_list *other_jl;
908         struct reiserfs_journal_list *first_jl;
909         struct list_head *entry;
910         unsigned long trans_id = jl->j_trans_id;
911         unsigned long other_trans_id;
912         unsigned long first_trans_id;
913
914       find_first:
915         /*
916          * first we walk backwards to find the oldest uncommitted transation
917          */
918         first_jl = jl;
919         entry = jl->j_list.prev;
920         while (1) {
921                 other_jl = JOURNAL_LIST_ENTRY(entry);
922                 if (entry == &journal->j_journal_list ||
923                     atomic_read(&other_jl->j_older_commits_done))
924                         break;
925
926                 first_jl = other_jl;
927                 entry = other_jl->j_list.prev;
928         }
929
930         /* if we didn't find any older uncommitted transactions, return now */
931         if (first_jl == jl) {
932                 return 0;
933         }
934
935         first_trans_id = first_jl->j_trans_id;
936
937         entry = &first_jl->j_list;
938         while (1) {
939                 other_jl = JOURNAL_LIST_ENTRY(entry);
940                 other_trans_id = other_jl->j_trans_id;
941
942                 if (other_trans_id < trans_id) {
943                         if (atomic_read(&other_jl->j_commit_left) != 0) {
944                                 flush_commit_list(s, other_jl, 0);
945
946                                 /* list we were called with is gone, return */
947                                 if (!journal_list_still_alive(s, trans_id))
948                                         return 1;
949
950                                 /* the one we just flushed is gone, this means all
951                                  * older lists are also gone, so first_jl is no longer
952                                  * valid either.  Go back to the beginning.
953                                  */
954                                 if (!journal_list_still_alive
955                                     (s, other_trans_id)) {
956                                         goto find_first;
957                                 }
958                         }
959                         entry = entry->next;
960                         if (entry == &journal->j_journal_list)
961                                 return 0;
962                 } else {
963                         return 0;
964                 }
965         }
966         return 0;
967 }
968 int reiserfs_async_progress_wait(struct super_block *s)
969 {
970         DEFINE_WAIT(wait);
971         struct reiserfs_journal *j = SB_JOURNAL(s);
972         if (atomic_read(&j->j_async_throttle))
973                 blk_congestion_wait(WRITE, HZ / 10);
974         return 0;
975 }
976
977 /*
978 ** if this journal list still has commit blocks unflushed, send them to disk.
979 **
980 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
981 ** Before the commit block can by written, every other log block must be safely on disk
982 **
983 */
984 static int flush_commit_list(struct super_block *s,
985                              struct reiserfs_journal_list *jl, int flushall)
986 {
987         int i;
988         int bn;
989         struct buffer_head *tbh = NULL;
990         unsigned long trans_id = jl->j_trans_id;
991         struct reiserfs_journal *journal = SB_JOURNAL(s);
992         int barrier = 0;
993         int retval = 0;
994         int write_len;
995
996         reiserfs_check_lock_depth(s, "flush_commit_list");
997
998         if (atomic_read(&jl->j_older_commits_done)) {
999                 return 0;
1000         }
1001
1002         get_fs_excl();
1003
1004         /* before we can put our commit blocks on disk, we have to make sure everyone older than
1005          ** us is on disk too
1006          */
1007         BUG_ON(jl->j_len <= 0);
1008         BUG_ON(trans_id == journal->j_trans_id);
1009
1010         get_journal_list(jl);
1011         if (flushall) {
1012                 if (flush_older_commits(s, jl) == 1) {
1013                         /* list disappeared during flush_older_commits.  return */
1014                         goto put_jl;
1015                 }
1016         }
1017
1018         /* make sure nobody is trying to flush this one at the same time */
1019         down(&jl->j_commit_lock);
1020         if (!journal_list_still_alive(s, trans_id)) {
1021                 up(&jl->j_commit_lock);
1022                 goto put_jl;
1023         }
1024         BUG_ON(jl->j_trans_id == 0);
1025
1026         /* this commit is done, exit */
1027         if (atomic_read(&(jl->j_commit_left)) <= 0) {
1028                 if (flushall) {
1029                         atomic_set(&(jl->j_older_commits_done), 1);
1030                 }
1031                 up(&jl->j_commit_lock);
1032                 goto put_jl;
1033         }
1034
1035         if (!list_empty(&jl->j_bh_list)) {
1036                 int ret;
1037                 unlock_kernel();
1038                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1039                                             journal, jl, &jl->j_bh_list);
1040                 if (ret < 0 && retval == 0)
1041                         retval = ret;
1042                 lock_kernel();
1043         }
1044         BUG_ON(!list_empty(&jl->j_bh_list));
1045         /*
1046          * for the description block and all the log blocks, submit any buffers
1047          * that haven't already reached the disk.  Try to write at least 256
1048          * log blocks. later on, we will only wait on blocks that correspond
1049          * to this transaction, but while we're unplugging we might as well
1050          * get a chunk of data on there.
1051          */
1052         atomic_inc(&journal->j_async_throttle);
1053         write_len = jl->j_len + 1;
1054         if (write_len < 256)
1055                 write_len = 256;
1056         for (i = 0 ; i < write_len ; i++) {
1057                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1058                     SB_ONDISK_JOURNAL_SIZE(s);
1059                 tbh = journal_find_get_block(s, bn);
1060                 if (tbh) {
1061                         if (buffer_dirty(tbh))
1062                             ll_rw_block(WRITE, 1, &tbh) ;
1063                         put_bh(tbh) ;
1064                 }
1065         }
1066         atomic_dec(&journal->j_async_throttle);
1067
1068         /* We're skipping the commit if there's an error */
1069         if (retval || reiserfs_is_journal_aborted(journal))
1070                 barrier = 0;
1071
1072         /* wait on everything written so far before writing the commit
1073          * if we are in barrier mode, send the commit down now
1074          */
1075         barrier = reiserfs_barrier_flush(s);
1076         if (barrier) {
1077                 int ret;
1078                 lock_buffer(jl->j_commit_bh);
1079                 ret = submit_barrier_buffer(jl->j_commit_bh);
1080                 if (ret == -EOPNOTSUPP) {
1081                         set_buffer_uptodate(jl->j_commit_bh);
1082                         disable_barrier(s);
1083                         barrier = 0;
1084                 }
1085         }
1086         for (i = 0; i < (jl->j_len + 1); i++) {
1087                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1088                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1089                 tbh = journal_find_get_block(s, bn);
1090                 wait_on_buffer(tbh);
1091                 // since we're using ll_rw_blk above, it might have skipped over
1092                 // a locked buffer.  Double check here
1093                 //
1094                 if (buffer_dirty(tbh))  /* redundant, sync_dirty_buffer() checks */
1095                         sync_dirty_buffer(tbh);
1096                 if (unlikely(!buffer_uptodate(tbh))) {
1097 #ifdef CONFIG_REISERFS_CHECK
1098                         reiserfs_warning(s, "journal-601, buffer write failed");
1099 #endif
1100                         retval = -EIO;
1101                 }
1102                 put_bh(tbh);    /* once for journal_find_get_block */
1103                 put_bh(tbh);    /* once due to original getblk in do_journal_end */
1104                 atomic_dec(&(jl->j_commit_left));
1105         }
1106
1107         BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1108
1109         if (!barrier) {
1110                 /* If there was a write error in the journal - we can't commit
1111                  * this transaction - it will be invalid and, if successful,
1112                  * will just end up propogating the write error out to
1113                  * the file system. */
1114                 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1115                         if (buffer_dirty(jl->j_commit_bh))
1116                                 BUG();
1117                         mark_buffer_dirty(jl->j_commit_bh) ;
1118                         sync_dirty_buffer(jl->j_commit_bh) ;
1119                 }
1120         } else
1121                 wait_on_buffer(jl->j_commit_bh);
1122
1123         check_barrier_completion(s, jl->j_commit_bh);
1124
1125         /* If there was a write error in the journal - we can't commit this
1126          * transaction - it will be invalid and, if successful, will just end
1127          * up propogating the write error out to the filesystem. */
1128         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1129 #ifdef CONFIG_REISERFS_CHECK
1130                 reiserfs_warning(s, "journal-615: buffer write failed");
1131 #endif
1132                 retval = -EIO;
1133         }
1134         bforget(jl->j_commit_bh);
1135         if (journal->j_last_commit_id != 0 &&
1136             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1137                 reiserfs_warning(s, "clm-2200: last commit %lu, current %lu",
1138                                  journal->j_last_commit_id, jl->j_trans_id);
1139         }
1140         journal->j_last_commit_id = jl->j_trans_id;
1141
1142         /* now, every commit block is on the disk.  It is safe to allow blocks freed during this transaction to be reallocated */
1143         cleanup_freed_for_journal_list(s, jl);
1144
1145         retval = retval ? retval : journal->j_errno;
1146
1147         /* mark the metadata dirty */
1148         if (!retval)
1149                 dirty_one_transaction(s, jl);
1150         atomic_dec(&(jl->j_commit_left));
1151
1152         if (flushall) {
1153                 atomic_set(&(jl->j_older_commits_done), 1);
1154         }
1155         up(&jl->j_commit_lock);
1156       put_jl:
1157         put_journal_list(s, jl);
1158
1159         if (retval)
1160                 reiserfs_abort(s, retval, "Journal write error in %s",
1161                                __FUNCTION__);
1162         put_fs_excl();
1163         return retval;
1164 }
1165
1166 /*
1167 ** flush_journal_list frequently needs to find a newer transaction for a given block.  This does that, or 
1168 ** returns NULL if it can't find anything 
1169 */
1170 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1171                                                           reiserfs_journal_cnode
1172                                                           *cn)
1173 {
1174         struct super_block *sb = cn->sb;
1175         b_blocknr_t blocknr = cn->blocknr;
1176
1177         cn = cn->hprev;
1178         while (cn) {
1179                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1180                         return cn->jlist;
1181                 }
1182                 cn = cn->hprev;
1183         }
1184         return NULL;
1185 }
1186
1187 static int newer_jl_done(struct reiserfs_journal_cnode *cn)
1188 {
1189         struct super_block *sb = cn->sb;
1190         b_blocknr_t blocknr = cn->blocknr;
1191
1192         cn = cn->hprev;
1193         while (cn) {
1194                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist &&
1195                     atomic_read(&cn->jlist->j_commit_left) != 0)
1196                                     return 0;
1197                 cn = cn->hprev;
1198         }
1199         return 1;
1200 }
1201
1202 static void remove_journal_hash(struct super_block *,
1203                                 struct reiserfs_journal_cnode **,
1204                                 struct reiserfs_journal_list *, unsigned long,
1205                                 int);
1206
1207 /*
1208 ** once all the real blocks have been flushed, it is safe to remove them from the
1209 ** journal list for this transaction.  Aside from freeing the cnode, this also allows the
1210 ** block to be reallocated for data blocks if it had been deleted.
1211 */
1212 static void remove_all_from_journal_list(struct super_block *p_s_sb,
1213                                          struct reiserfs_journal_list *jl,
1214                                          int debug)
1215 {
1216         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1217         struct reiserfs_journal_cnode *cn, *last;
1218         cn = jl->j_realblock;
1219
1220         /* which is better, to lock once around the whole loop, or
1221          ** to lock for each call to remove_journal_hash?
1222          */
1223         while (cn) {
1224                 if (cn->blocknr != 0) {
1225                         if (debug) {
1226                                 reiserfs_warning(p_s_sb,
1227                                                  "block %u, bh is %d, state %ld",
1228                                                  cn->blocknr, cn->bh ? 1 : 0,
1229                                                  cn->state);
1230                         }
1231                         cn->state = 0;
1232                         remove_journal_hash(p_s_sb, journal->j_list_hash_table,
1233                                             jl, cn->blocknr, 1);
1234                 }
1235                 last = cn;
1236                 cn = cn->next;
1237                 free_cnode(p_s_sb, last);
1238         }
1239         jl->j_realblock = NULL;
1240 }
1241
1242 /*
1243 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1244 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1245 ** releasing blocks in this transaction for reuse as data blocks.
1246 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1247 **
1248 */
1249 static int _update_journal_header_block(struct super_block *p_s_sb,
1250                                         unsigned long offset,
1251                                         unsigned long trans_id)
1252 {
1253         struct reiserfs_journal_header *jh;
1254         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1255
1256         if (reiserfs_is_journal_aborted(journal))
1257                 return -EIO;
1258
1259         if (trans_id >= journal->j_last_flush_trans_id) {
1260                 if (buffer_locked((journal->j_header_bh))) {
1261                         wait_on_buffer((journal->j_header_bh));
1262                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1263 #ifdef CONFIG_REISERFS_CHECK
1264                                 reiserfs_warning(p_s_sb,
1265                                                  "journal-699: buffer write failed");
1266 #endif
1267                                 return -EIO;
1268                         }
1269                 }
1270                 journal->j_last_flush_trans_id = trans_id;
1271                 journal->j_first_unflushed_offset = offset;
1272                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1273                                                         b_data);
1274                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1275                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1276                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1277
1278                 if (reiserfs_barrier_flush(p_s_sb)) {
1279                         int ret;
1280                         lock_buffer(journal->j_header_bh);
1281                         ret = submit_barrier_buffer(journal->j_header_bh);
1282                         if (ret == -EOPNOTSUPP) {
1283                                 set_buffer_uptodate(journal->j_header_bh);
1284                                 disable_barrier(p_s_sb);
1285                                 goto sync;
1286                         }
1287                         wait_on_buffer(journal->j_header_bh);
1288                         check_barrier_completion(p_s_sb, journal->j_header_bh);
1289                 } else {
1290                       sync:
1291                         set_buffer_dirty(journal->j_header_bh);
1292                         sync_dirty_buffer(journal->j_header_bh);
1293                 }
1294                 if (!buffer_uptodate(journal->j_header_bh)) {
1295                         reiserfs_warning(p_s_sb,
1296                                          "journal-837: IO error during journal replay");
1297                         return -EIO;
1298                 }
1299         }
1300         return 0;
1301 }
1302
1303 static int update_journal_header_block(struct super_block *p_s_sb,
1304                                        unsigned long offset,
1305                                        unsigned long trans_id)
1306 {
1307         return _update_journal_header_block(p_s_sb, offset, trans_id);
1308 }
1309
1310 /* 
1311 ** flush any and all journal lists older than you are 
1312 ** can only be called from flush_journal_list
1313 */
1314 static int flush_older_journal_lists(struct super_block *p_s_sb,
1315                                      struct reiserfs_journal_list *jl)
1316 {
1317         struct list_head *entry;
1318         struct reiserfs_journal_list *other_jl;
1319         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1320         unsigned long trans_id = jl->j_trans_id;
1321
1322         /* we know we are the only ones flushing things, no extra race
1323          * protection is required.
1324          */
1325       restart:
1326         entry = journal->j_journal_list.next;
1327         /* Did we wrap? */
1328         if (entry == &journal->j_journal_list)
1329                 return 0;
1330         other_jl = JOURNAL_LIST_ENTRY(entry);
1331         if (other_jl->j_trans_id < trans_id) {
1332                 BUG_ON(other_jl->j_refcount <= 0);
1333                 /* do not flush all */
1334                 flush_journal_list(p_s_sb, other_jl, 0);
1335
1336                 /* other_jl is now deleted from the list */
1337                 goto restart;
1338         }
1339         return 0;
1340 }
1341
1342 static void del_from_work_list(struct super_block *s,
1343                                struct reiserfs_journal_list *jl)
1344 {
1345         struct reiserfs_journal *journal = SB_JOURNAL(s);
1346         if (!list_empty(&jl->j_working_list)) {
1347                 list_del_init(&jl->j_working_list);
1348                 journal->j_num_work_lists--;
1349         }
1350 }
1351
1352 /* flush a journal list, both commit and real blocks
1353 **
1354 ** always set flushall to 1, unless you are calling from inside
1355 ** flush_journal_list
1356 **
1357 ** IMPORTANT.  This can only be called while there are no journal writers, 
1358 ** and the journal is locked.  That means it can only be called from 
1359 ** do_journal_end, or by journal_release
1360 */
1361 static int flush_journal_list(struct super_block *s,
1362                               struct reiserfs_journal_list *jl, int flushall)
1363 {
1364         struct reiserfs_journal_list *pjl;
1365         struct reiserfs_journal_cnode *cn, *last;
1366         int count;
1367         int was_jwait = 0;
1368         int was_dirty = 0;
1369         struct buffer_head *saved_bh;
1370         unsigned long j_len_saved = jl->j_len;
1371         struct reiserfs_journal *journal = SB_JOURNAL(s);
1372         int err = 0;
1373
1374         BUG_ON(j_len_saved <= 0);
1375
1376         if (atomic_read(&journal->j_wcount) != 0) {
1377                 reiserfs_warning(s,
1378                                  "clm-2048: flush_journal_list called with wcount %d",
1379                                  atomic_read(&journal->j_wcount));
1380         }
1381         BUG_ON(jl->j_trans_id == 0);
1382
1383         /* if flushall == 0, the lock is already held */
1384         if (flushall) {
1385                 down(&journal->j_flush_sem);
1386         } else if (!down_trylock(&journal->j_flush_sem)) {
1387                 BUG();
1388         }
1389
1390         count = 0;
1391         if (j_len_saved > journal->j_trans_max) {
1392                 reiserfs_panic(s,
1393                                "journal-715: flush_journal_list, length is %lu, trans id %lu\n",
1394                                j_len_saved, jl->j_trans_id);
1395                 return 0;
1396         }
1397
1398         get_fs_excl();
1399
1400         /* if all the work is already done, get out of here */
1401         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1402             atomic_read(&(jl->j_commit_left)) <= 0) {
1403                 goto flush_older_and_return;
1404         }
1405
1406         /* start by putting the commit list on disk.  This will also flush 
1407          ** the commit lists of any olders transactions
1408          */
1409         flush_commit_list(s, jl, 1);
1410
1411         if (!(jl->j_state & LIST_DIRTY)
1412             && !reiserfs_is_journal_aborted(journal))
1413                 BUG();
1414
1415         /* are we done now? */
1416         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1417             atomic_read(&(jl->j_commit_left)) <= 0) {
1418                 goto flush_older_and_return;
1419         }
1420
1421         /* loop through each cnode, see if we need to write it, 
1422          ** or wait on a more recent transaction, or just ignore it 
1423          */
1424         if (atomic_read(&(journal->j_wcount)) != 0) {
1425                 reiserfs_panic(s,
1426                                "journal-844: panic journal list is flushing, wcount is not 0\n");
1427         }
1428         cn = jl->j_realblock;
1429         while (cn) {
1430                 was_jwait = 0;
1431                 was_dirty = 0;
1432                 saved_bh = NULL;
1433                 /* blocknr of 0 is no longer in the hash, ignore it */
1434                 if (cn->blocknr == 0) {
1435                         goto free_cnode;
1436                 }
1437
1438                 /* This transaction failed commit. Don't write out to the disk */
1439                 if (!(jl->j_state & LIST_DIRTY))
1440                         goto free_cnode;
1441
1442                 pjl = find_newer_jl_for_cn(cn);
1443                 /* the order is important here.  We check pjl to make sure we
1444                  ** don't clear BH_JDirty_wait if we aren't the one writing this
1445                  ** block to disk
1446                  */
1447                 if (!pjl && cn->bh) {
1448                         saved_bh = cn->bh;
1449
1450                         /* we do this to make sure nobody releases the buffer while 
1451                          ** we are working with it 
1452                          */
1453                         get_bh(saved_bh);
1454
1455                         if (buffer_journal_dirty(saved_bh)) {
1456                                 BUG_ON(!can_dirty(cn));
1457                                 was_jwait = 1;
1458                                 was_dirty = 1;
1459                         } else if (can_dirty(cn)) {
1460                                 /* everything with !pjl && jwait should be writable */
1461                                 BUG();
1462                         }
1463                 }
1464
1465                 /* if someone has this block in a newer transaction, just make
1466                  ** sure they are commited, and don't try writing it to disk
1467                  */
1468                 if (pjl) {
1469                         if (atomic_read(&pjl->j_commit_left))
1470                                 flush_commit_list(s, pjl, 1);
1471                         goto free_cnode;
1472                 }
1473
1474                 /* bh == NULL when the block got to disk on its own, OR, 
1475                  ** the block got freed in a future transaction 
1476                  */
1477                 if (saved_bh == NULL) {
1478                         goto free_cnode;
1479                 }
1480
1481                 /* this should never happen.  kupdate_one_transaction has this list
1482                  ** locked while it works, so we should never see a buffer here that
1483                  ** is not marked JDirty_wait
1484                  */
1485                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1486                         reiserfs_warning(s,
1487                                          "journal-813: BAD! buffer %llu %cdirty %cjwait, "
1488                                          "not in a newer tranasction",
1489                                          (unsigned long long)saved_bh->
1490                                          b_blocknr, was_dirty ? ' ' : '!',
1491                                          was_jwait ? ' ' : '!');
1492                 }
1493                 if (was_dirty) {
1494                         /* we inc again because saved_bh gets decremented at free_cnode */
1495                         get_bh(saved_bh);
1496                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1497                         lock_buffer(saved_bh);
1498                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1499                         if (buffer_dirty(saved_bh))
1500                                 submit_logged_buffer(saved_bh);
1501                         else
1502                                 unlock_buffer(saved_bh);
1503                         count++;
1504                 } else {
1505                         reiserfs_warning(s,
1506                                          "clm-2082: Unable to flush buffer %llu in %s",
1507                                          (unsigned long long)saved_bh->
1508                                          b_blocknr, __FUNCTION__);
1509                 }
1510               free_cnode:
1511                 last = cn;
1512                 cn = cn->next;
1513                 if (saved_bh) {
1514                         /* we incremented this to keep others from taking the buffer head away */
1515                         put_bh(saved_bh);
1516                         if (atomic_read(&(saved_bh->b_count)) < 0) {
1517                                 reiserfs_warning(s,
1518                                                  "journal-945: saved_bh->b_count < 0");
1519                         }
1520                 }
1521         }
1522         if (count > 0) {
1523                 cn = jl->j_realblock;
1524                 while (cn) {
1525                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1526                                 if (!cn->bh) {
1527                                         reiserfs_panic(s,
1528                                                        "journal-1011: cn->bh is NULL\n");
1529                                 }
1530                                 wait_on_buffer(cn->bh);
1531                                 if (!cn->bh) {
1532                                         reiserfs_panic(s,
1533                                                        "journal-1012: cn->bh is NULL\n");
1534                                 }
1535                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1536 #ifdef CONFIG_REISERFS_CHECK
1537                                         reiserfs_warning(s,
1538                                                          "journal-949: buffer write failed\n");
1539 #endif
1540                                         err = -EIO;
1541                                 }
1542                                 /* note, we must clear the JDirty_wait bit after the up to date
1543                                  ** check, otherwise we race against our flushpage routine
1544                                  */
1545                                 BUG_ON(!test_clear_buffer_journal_dirty
1546                                        (cn->bh));
1547
1548                                 /* undo the inc from journal_mark_dirty */
1549                                 put_bh(cn->bh);
1550                                 brelse(cn->bh);
1551                         }
1552                         cn = cn->next;
1553                 }
1554         }
1555
1556         if (err)
1557                 reiserfs_abort(s, -EIO,
1558                                "Write error while pushing transaction to disk in %s",
1559                                __FUNCTION__);
1560       flush_older_and_return:
1561
1562         /* before we can update the journal header block, we _must_ flush all 
1563          ** real blocks from all older transactions to disk.  This is because
1564          ** once the header block is updated, this transaction will not be
1565          ** replayed after a crash
1566          */
1567         if (flushall) {
1568                 flush_older_journal_lists(s, jl);
1569         }
1570
1571         err = journal->j_errno;
1572         /* before we can remove everything from the hash tables for this 
1573          ** transaction, we must make sure it can never be replayed
1574          **
1575          ** since we are only called from do_journal_end, we know for sure there
1576          ** are no allocations going on while we are flushing journal lists.  So,
1577          ** we only need to update the journal header block for the last list
1578          ** being flushed
1579          */
1580         if (!err && flushall) {
1581                 err =
1582                     update_journal_header_block(s,
1583                                                 (jl->j_start + jl->j_len +
1584                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1585                                                 jl->j_trans_id);
1586                 if (err)
1587                         reiserfs_abort(s, -EIO,
1588                                        "Write error while updating journal header in %s",
1589                                        __FUNCTION__);
1590         }
1591         remove_all_from_journal_list(s, jl, 0);
1592         list_del_init(&jl->j_list);
1593         journal->j_num_lists--;
1594         del_from_work_list(s, jl);
1595
1596         if (journal->j_last_flush_id != 0 &&
1597             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1598                 reiserfs_warning(s, "clm-2201: last flush %lu, current %lu",
1599                                  journal->j_last_flush_id, jl->j_trans_id);
1600         }
1601         journal->j_last_flush_id = jl->j_trans_id;
1602
1603         /* not strictly required since we are freeing the list, but it should
1604          * help find code using dead lists later on
1605          */
1606         jl->j_len = 0;
1607         atomic_set(&(jl->j_nonzerolen), 0);
1608         jl->j_start = 0;
1609         jl->j_realblock = NULL;
1610         jl->j_commit_bh = NULL;
1611         jl->j_trans_id = 0;
1612         jl->j_state = 0;
1613         put_journal_list(s, jl);
1614         if (flushall)
1615                 up(&journal->j_flush_sem);
1616         put_fs_excl();
1617         return err;
1618 }
1619
1620 static int test_transaction(struct super_block *s,
1621                             struct reiserfs_journal_list *jl)
1622 {
1623         struct reiserfs_journal_cnode *cn;
1624
1625         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0)
1626                 return 1;
1627
1628         cn = jl->j_realblock;
1629         while (cn) {
1630                 /* if the blocknr == 0, this has been cleared from the hash,
1631                  ** skip it
1632                  */
1633                 if (cn->blocknr == 0) {
1634                         goto next;
1635                 }
1636                 if (cn->bh && !newer_jl_done(cn))
1637                         return 0;
1638               next:
1639                 cn = cn->next;
1640                 cond_resched();
1641         }
1642         return 0;
1643 }
1644
1645 static int write_one_transaction(struct super_block *s,
1646                                  struct reiserfs_journal_list *jl,
1647                                  struct buffer_chunk *chunk)
1648 {
1649         struct reiserfs_journal_cnode *cn;
1650         int ret = 0;
1651
1652         jl->j_state |= LIST_TOUCHED;
1653         del_from_work_list(s, jl);
1654         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1655                 return 0;
1656         }
1657
1658         cn = jl->j_realblock;
1659         while (cn) {
1660                 /* if the blocknr == 0, this has been cleared from the hash,
1661                  ** skip it
1662                  */
1663                 if (cn->blocknr == 0) {
1664                         goto next;
1665                 }
1666                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1667                         struct buffer_head *tmp_bh;
1668                         /* we can race against journal_mark_freed when we try
1669                          * to lock_buffer(cn->bh), so we have to inc the buffer
1670                          * count, and recheck things after locking
1671                          */
1672                         tmp_bh = cn->bh;
1673                         get_bh(tmp_bh);
1674                         lock_buffer(tmp_bh);
1675                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1676                                 if (!buffer_journal_dirty(tmp_bh) ||
1677                                     buffer_journal_prepared(tmp_bh))
1678                                         BUG();
1679                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1680                                 ret++;
1681                         } else {
1682                                 /* note, cn->bh might be null now */
1683                                 unlock_buffer(tmp_bh);
1684                         }
1685                         put_bh(tmp_bh);
1686                 }
1687               next:
1688                 cn = cn->next;
1689                 cond_resched();
1690         }
1691         return ret;
1692 }
1693
1694 /* used by flush_commit_list */
1695 static int dirty_one_transaction(struct super_block *s,
1696                                  struct reiserfs_journal_list *jl)
1697 {
1698         struct reiserfs_journal_cnode *cn;
1699         struct reiserfs_journal_list *pjl;
1700         int ret = 0;
1701
1702         jl->j_state |= LIST_DIRTY;
1703         cn = jl->j_realblock;
1704         while (cn) {
1705                 /* look for a more recent transaction that logged this
1706                  ** buffer.  Only the most recent transaction with a buffer in
1707                  ** it is allowed to send that buffer to disk
1708                  */
1709                 pjl = find_newer_jl_for_cn(cn);
1710                 if (!pjl && cn->blocknr && cn->bh
1711                     && buffer_journal_dirty(cn->bh)) {
1712                         BUG_ON(!can_dirty(cn));
1713                         /* if the buffer is prepared, it will either be logged
1714                          * or restored.  If restored, we need to make sure
1715                          * it actually gets marked dirty
1716                          */
1717                         clear_buffer_journal_new(cn->bh);
1718                         if (buffer_journal_prepared(cn->bh)) {
1719                                 set_buffer_journal_restore_dirty(cn->bh);
1720                         } else {
1721                                 set_buffer_journal_test(cn->bh);
1722                                 mark_buffer_dirty(cn->bh);
1723                         }
1724                 }
1725                 cn = cn->next;
1726         }
1727         return ret;
1728 }
1729
1730 static int kupdate_transactions(struct super_block *s,
1731                                 struct reiserfs_journal_list *jl,
1732                                 struct reiserfs_journal_list **next_jl,
1733                                 unsigned long *next_trans_id,
1734                                 int num_blocks, int num_trans)
1735 {
1736         int ret = 0;
1737         int written = 0;
1738         int transactions_flushed = 0;
1739         unsigned long orig_trans_id = jl->j_trans_id;
1740         struct buffer_chunk chunk;
1741         struct list_head *entry;
1742         struct reiserfs_journal *journal = SB_JOURNAL(s);
1743         chunk.nr = 0;
1744
1745         down(&journal->j_flush_sem);
1746         if (!journal_list_still_alive(s, orig_trans_id)) {
1747                 goto done;
1748         }
1749
1750         /* we've got j_flush_sem held, nobody is going to delete any
1751          * of these lists out from underneath us
1752          */
1753         while ((num_trans && transactions_flushed < num_trans) ||
1754                (!num_trans && written < num_blocks)) {
1755
1756                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1757                     atomic_read(&jl->j_commit_left)
1758                     || !(jl->j_state & LIST_DIRTY)) {
1759                         del_from_work_list(s, jl);
1760                         break;
1761                 }
1762                 ret = write_one_transaction(s, jl, &chunk);
1763
1764                 if (ret < 0)
1765                         goto done;
1766                 transactions_flushed++;
1767                 written += ret;
1768                 entry = jl->j_list.next;
1769
1770                 /* did we wrap? */
1771                 if (entry == &journal->j_journal_list) {
1772                         break;
1773                 }
1774                 jl = JOURNAL_LIST_ENTRY(entry);
1775
1776                 /* don't bother with older transactions */
1777                 if (jl->j_trans_id <= orig_trans_id)
1778                         break;
1779         }
1780         if (chunk.nr) {
1781                 write_chunk(&chunk);
1782         }
1783
1784       done:
1785         up(&journal->j_flush_sem);
1786         return ret;
1787 }
1788
1789 /* for o_sync and fsync heavy applications, they tend to use
1790 ** all the journa list slots with tiny transactions.  These
1791 ** trigger lots and lots of calls to update the header block, which
1792 ** adds seeks and slows things down.
1793 **
1794 ** This function tries to clear out a large chunk of the journal lists
1795 ** at once, which makes everything faster since only the newest journal
1796 ** list updates the header block
1797 */
1798 static int flush_used_journal_lists(struct super_block *s,
1799                                     struct reiserfs_journal_list *jl)
1800 {
1801         unsigned long len = 0;
1802         unsigned long cur_len;
1803         int ret;
1804         int i;
1805         int limit = 256;
1806         struct reiserfs_journal_list *tjl;
1807         struct reiserfs_journal_list *flush_jl;
1808         unsigned long trans_id;
1809         struct reiserfs_journal *journal = SB_JOURNAL(s);
1810
1811         flush_jl = tjl = jl;
1812
1813         /* in data logging mode, try harder to flush a lot of blocks */
1814         if (reiserfs_data_log(s))
1815                 limit = 1024;
1816         /* flush for 256 transactions or limit blocks, whichever comes first */
1817         for (i = 0; i < 256 && len < limit; i++) {
1818                 if (atomic_read(&tjl->j_commit_left) ||
1819                     tjl->j_trans_id < jl->j_trans_id) {
1820                         break;
1821                 }
1822                 cur_len = atomic_read(&tjl->j_nonzerolen);
1823                 if (cur_len > 0) {
1824                         tjl->j_state &= ~LIST_TOUCHED;
1825                 }
1826                 len += cur_len;
1827                 flush_jl = tjl;
1828                 if (tjl->j_list.next == &journal->j_journal_list)
1829                         break;
1830                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1831         }
1832         /* try to find a group of blocks we can flush across all the
1833          ** transactions, but only bother if we've actually spanned
1834          ** across multiple lists
1835          */
1836         if (flush_jl != jl) {
1837                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1838         }
1839         flush_journal_list(s, flush_jl, 1);
1840         return 0;
1841 }
1842
1843 /*
1844 ** removes any nodes in table with name block and dev as bh.
1845 ** only touchs the hnext and hprev pointers.
1846 */
1847 void remove_journal_hash(struct super_block *sb,
1848                          struct reiserfs_journal_cnode **table,
1849                          struct reiserfs_journal_list *jl,
1850                          unsigned long block, int remove_freed)
1851 {
1852         struct reiserfs_journal_cnode *cur;
1853         struct reiserfs_journal_cnode **head;
1854
1855         head = &(journal_hash(table, sb, block));
1856         if (!head) {
1857                 return;
1858         }
1859         cur = *head;
1860         while (cur) {
1861                 if (cur->blocknr == block && cur->sb == sb
1862                     && (jl == NULL || jl == cur->jlist)
1863                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1864                         if (cur->hnext) {
1865                                 cur->hnext->hprev = cur->hprev;
1866                         }
1867                         if (cur->hprev) {
1868                                 cur->hprev->hnext = cur->hnext;
1869                         } else {
1870                                 *head = cur->hnext;
1871                         }
1872                         cur->blocknr = 0;
1873                         cur->sb = NULL;
1874                         cur->state = 0;
1875                         if (cur->bh && cur->jlist)      /* anybody who clears the cur->bh will also dec the nonzerolen */
1876                                 atomic_dec(&(cur->jlist->j_nonzerolen));
1877                         cur->bh = NULL;
1878                         cur->jlist = NULL;
1879                 }
1880                 cur = cur->hnext;
1881         }
1882 }
1883
1884 static void free_journal_ram(struct super_block *p_s_sb)
1885 {
1886         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1887         kfree(journal->j_current_jl);
1888         journal->j_num_lists--;
1889
1890         vfree(journal->j_cnode_free_orig);
1891         free_list_bitmaps(p_s_sb, journal->j_list_bitmap);
1892         free_bitmap_nodes(p_s_sb);      /* must be after free_list_bitmaps */
1893         if (journal->j_header_bh) {
1894                 brelse(journal->j_header_bh);
1895         }
1896         /* j_header_bh is on the journal dev, make sure not to release the journal
1897          * dev until we brelse j_header_bh
1898          */
1899         release_journal_dev(p_s_sb, journal);
1900         vfree(journal);
1901 }
1902
1903 /*
1904 ** call on unmount.  Only set error to 1 if you haven't made your way out
1905 ** of read_super() yet.  Any other caller must keep error at 0.
1906 */
1907 static int do_journal_release(struct reiserfs_transaction_handle *th,
1908                               struct super_block *p_s_sb, int error)
1909 {
1910         struct reiserfs_transaction_handle myth;
1911         int flushed = 0;
1912         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1913
1914         /* we only want to flush out transactions if we were called with error == 0
1915          */
1916         if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
1917                 /* end the current trans */
1918                 BUG_ON(!th->t_trans_id);
1919                 do_journal_end(th, p_s_sb, 10, FLUSH_ALL);
1920
1921                 /* make sure something gets logged to force our way into the flush code */
1922                 if (!journal_join(&myth, p_s_sb, 1)) {
1923                         reiserfs_prepare_for_journal(p_s_sb,
1924                                                      SB_BUFFER_WITH_SB(p_s_sb),
1925                                                      1);
1926                         journal_mark_dirty(&myth, p_s_sb,
1927                                            SB_BUFFER_WITH_SB(p_s_sb));
1928                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1929                         flushed = 1;
1930                 }
1931         }
1932
1933         /* this also catches errors during the do_journal_end above */
1934         if (!error && reiserfs_is_journal_aborted(journal)) {
1935                 memset(&myth, 0, sizeof(myth));
1936                 if (!journal_join_abort(&myth, p_s_sb, 1)) {
1937                         reiserfs_prepare_for_journal(p_s_sb,
1938                                                      SB_BUFFER_WITH_SB(p_s_sb),
1939                                                      1);
1940                         journal_mark_dirty(&myth, p_s_sb,
1941                                            SB_BUFFER_WITH_SB(p_s_sb));
1942                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1943                 }
1944         }
1945
1946         reiserfs_mounted_fs_count--;
1947         /* wait for all commits to finish */
1948         cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work);
1949         flush_workqueue(commit_wq);
1950         if (!reiserfs_mounted_fs_count) {
1951                 destroy_workqueue(commit_wq);
1952                 commit_wq = NULL;
1953         }
1954
1955         free_journal_ram(p_s_sb);
1956
1957         return 0;
1958 }
1959
1960 /*
1961 ** call on unmount.  flush all journal trans, release all alloc'd ram
1962 */
1963 int journal_release(struct reiserfs_transaction_handle *th,
1964                     struct super_block *p_s_sb)
1965 {
1966         return do_journal_release(th, p_s_sb, 0);
1967 }
1968
1969 /*
1970 ** only call from an error condition inside reiserfs_read_super!
1971 */
1972 int journal_release_error(struct reiserfs_transaction_handle *th,
1973                           struct super_block *p_s_sb)
1974 {
1975         return do_journal_release(th, p_s_sb, 1);
1976 }
1977
1978 /* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
1979 static int journal_compare_desc_commit(struct super_block *p_s_sb,
1980                                        struct reiserfs_journal_desc *desc,
1981                                        struct reiserfs_journal_commit *commit)
1982 {
1983         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1984             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1985             get_commit_trans_len(commit) > SB_JOURNAL(p_s_sb)->j_trans_max ||
1986             get_commit_trans_len(commit) <= 0) {
1987                 return 1;
1988         }
1989         return 0;
1990 }
1991
1992 /* returns 0 if it did not find a description block  
1993 ** returns -1 if it found a corrupt commit block
1994 ** returns 1 if both desc and commit were valid 
1995 */
1996 static int journal_transaction_is_valid(struct super_block *p_s_sb,
1997                                         struct buffer_head *d_bh,
1998                                         unsigned long *oldest_invalid_trans_id,
1999                                         unsigned long *newest_mount_id)
2000 {
2001         struct reiserfs_journal_desc *desc;
2002         struct reiserfs_journal_commit *commit;
2003         struct buffer_head *c_bh;
2004         unsigned long offset;
2005
2006         if (!d_bh)
2007                 return 0;
2008
2009         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2010         if (get_desc_trans_len(desc) > 0
2011             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2012                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2013                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2014                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2015                                        "journal-986: transaction "
2016                                        "is valid returning because trans_id %d is greater than "
2017                                        "oldest_invalid %lu",
2018                                        get_desc_trans_id(desc),
2019                                        *oldest_invalid_trans_id);
2020                         return 0;
2021                 }
2022                 if (newest_mount_id
2023                     && *newest_mount_id > get_desc_mount_id(desc)) {
2024                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2025                                        "journal-1087: transaction "
2026                                        "is valid returning because mount_id %d is less than "
2027                                        "newest_mount_id %lu",
2028                                        get_desc_mount_id(desc),
2029                                        *newest_mount_id);
2030                         return -1;
2031                 }
2032                 if (get_desc_trans_len(desc) > SB_JOURNAL(p_s_sb)->j_trans_max) {
2033                         reiserfs_warning(p_s_sb,
2034                                          "journal-2018: Bad transaction length %d encountered, ignoring transaction",
2035                                          get_desc_trans_len(desc));
2036                         return -1;
2037                 }
2038                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2039
2040                 /* ok, we have a journal description block, lets see if the transaction was valid */
2041                 c_bh =
2042                     journal_bread(p_s_sb,
2043                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2044                                   ((offset + get_desc_trans_len(desc) +
2045                                     1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2046                 if (!c_bh)
2047                         return 0;
2048                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2049                 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2050                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2051                                        "journal_transaction_is_valid, commit offset %ld had bad "
2052                                        "time %d or length %d",
2053                                        c_bh->b_blocknr -
2054                                        SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2055                                        get_commit_trans_id(commit),
2056                                        get_commit_trans_len(commit));
2057                         brelse(c_bh);
2058                         if (oldest_invalid_trans_id) {
2059                                 *oldest_invalid_trans_id =
2060                                     get_desc_trans_id(desc);
2061                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2062                                                "journal-1004: "
2063                                                "transaction_is_valid setting oldest invalid trans_id "
2064                                                "to %d",
2065                                                get_desc_trans_id(desc));
2066                         }
2067                         return -1;
2068                 }
2069                 brelse(c_bh);
2070                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2071                                "journal-1006: found valid "
2072                                "transaction start offset %llu, len %d id %d",
2073                                d_bh->b_blocknr -
2074                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2075                                get_desc_trans_len(desc),
2076                                get_desc_trans_id(desc));
2077                 return 1;
2078         } else {
2079                 return 0;
2080         }
2081 }
2082
2083 static void brelse_array(struct buffer_head **heads, int num)
2084 {
2085         int i;
2086         for (i = 0; i < num; i++) {
2087                 brelse(heads[i]);
2088         }
2089 }
2090
2091 /*
2092 ** given the start, and values for the oldest acceptable transactions,
2093 ** this either reads in a replays a transaction, or returns because the transaction
2094 ** is invalid, or too old.
2095 */
2096 static int journal_read_transaction(struct super_block *p_s_sb,
2097                                     unsigned long cur_dblock,
2098                                     unsigned long oldest_start,
2099                                     unsigned long oldest_trans_id,
2100                                     unsigned long newest_mount_id)
2101 {
2102         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2103         struct reiserfs_journal_desc *desc;
2104         struct reiserfs_journal_commit *commit;
2105         unsigned long trans_id = 0;
2106         struct buffer_head *c_bh;
2107         struct buffer_head *d_bh;
2108         struct buffer_head **log_blocks = NULL;
2109         struct buffer_head **real_blocks = NULL;
2110         unsigned long trans_offset;
2111         int i;
2112         int trans_half;
2113
2114         d_bh = journal_bread(p_s_sb, cur_dblock);
2115         if (!d_bh)
2116                 return 1;
2117         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2118         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2119         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
2120                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2121                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2122                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2123         if (get_desc_trans_id(desc) < oldest_trans_id) {
2124                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
2125                                "journal_read_trans skipping because %lu is too old",
2126                                cur_dblock -
2127                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2128                 brelse(d_bh);
2129                 return 1;
2130         }
2131         if (get_desc_mount_id(desc) != newest_mount_id) {
2132                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
2133                                "journal_read_trans skipping because %d is != "
2134                                "newest_mount_id %lu", get_desc_mount_id(desc),
2135                                newest_mount_id);
2136                 brelse(d_bh);
2137                 return 1;
2138         }
2139         c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2140                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2141                               SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2142         if (!c_bh) {
2143                 brelse(d_bh);
2144                 return 1;
2145         }
2146         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2147         if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2148                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2149                                "journal_read_transaction, "
2150                                "commit offset %llu had bad time %d or length %d",
2151                                c_bh->b_blocknr -
2152                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2153                                get_commit_trans_id(commit),
2154                                get_commit_trans_len(commit));
2155                 brelse(c_bh);
2156                 brelse(d_bh);
2157                 return 1;
2158         }
2159         trans_id = get_desc_trans_id(desc);
2160         /* now we know we've got a good transaction, and it was inside the valid time ranges */
2161         log_blocks = kmalloc(get_desc_trans_len(desc) *
2162                              sizeof(struct buffer_head *), GFP_NOFS);
2163         real_blocks = kmalloc(get_desc_trans_len(desc) *
2164                               sizeof(struct buffer_head *), GFP_NOFS);
2165         if (!log_blocks || !real_blocks) {
2166                 brelse(c_bh);
2167                 brelse(d_bh);
2168                 kfree(log_blocks);
2169                 kfree(real_blocks);
2170                 reiserfs_warning(p_s_sb,
2171                                  "journal-1169: kmalloc failed, unable to mount FS");
2172                 return -1;
2173         }
2174         /* get all the buffer heads */
2175         trans_half = journal_trans_half(p_s_sb->s_blocksize);
2176         for (i = 0; i < get_desc_trans_len(desc); i++) {
2177                 log_blocks[i] =
2178                     journal_getblk(p_s_sb,
2179                                    SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2180                                    (trans_offset + 1 +
2181                                     i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2182                 if (i < trans_half) {
2183                         real_blocks[i] =
2184                             sb_getblk(p_s_sb,
2185                                       le32_to_cpu(desc->j_realblock[i]));
2186                 } else {
2187                         real_blocks[i] =
2188                             sb_getblk(p_s_sb,
2189                                       le32_to_cpu(commit->
2190                                                   j_realblock[i - trans_half]));
2191                 }
2192                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb)) {
2193                         reiserfs_warning(p_s_sb,
2194                                          "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem");
2195                         goto abort_replay;
2196                 }
2197                 /* make sure we don't try to replay onto log or reserved area */
2198                 if (is_block_in_log_or_reserved_area
2199                     (p_s_sb, real_blocks[i]->b_blocknr)) {
2200                         reiserfs_warning(p_s_sb,
2201                                          "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block");
2202                       abort_replay:
2203                         brelse_array(log_blocks, i);
2204                         brelse_array(real_blocks, i);
2205                         brelse(c_bh);
2206                         brelse(d_bh);
2207                         kfree(log_blocks);
2208                         kfree(real_blocks);
2209                         return -1;
2210                 }
2211         }
2212         /* read in the log blocks, memcpy to the corresponding real block */
2213         ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2214         for (i = 0; i < get_desc_trans_len(desc); i++) {
2215                 wait_on_buffer(log_blocks[i]);
2216                 if (!buffer_uptodate(log_blocks[i])) {
2217                         reiserfs_warning(p_s_sb,
2218                                          "journal-1212: REPLAY FAILURE fsck required! buffer write failed");
2219                         brelse_array(log_blocks + i,
2220                                      get_desc_trans_len(desc) - i);
2221                         brelse_array(real_blocks, get_desc_trans_len(desc));
2222                         brelse(c_bh);
2223                         brelse(d_bh);
2224                         kfree(log_blocks);
2225                         kfree(real_blocks);
2226                         return -1;
2227                 }
2228                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2229                        real_blocks[i]->b_size);
2230                 set_buffer_uptodate(real_blocks[i]);
2231                 brelse(log_blocks[i]);
2232         }
2233         /* flush out the real blocks */
2234         for (i = 0; i < get_desc_trans_len(desc); i++) {
2235                 set_buffer_dirty(real_blocks[i]);
2236                 ll_rw_block(SWRITE, 1, real_blocks + i);
2237         }
2238         for (i = 0; i < get_desc_trans_len(desc); i++) {
2239                 wait_on_buffer(real_blocks[i]);
2240                 if (!buffer_uptodate(real_blocks[i])) {
2241                         reiserfs_warning(p_s_sb,
2242                                          "journal-1226: REPLAY FAILURE, fsck required! buffer write failed");
2243                         brelse_array(real_blocks + i,
2244                                      get_desc_trans_len(desc) - i);
2245                         brelse(c_bh);
2246                         brelse(d_bh);
2247                         kfree(log_blocks);
2248                         kfree(real_blocks);
2249                         return -1;
2250                 }
2251                 brelse(real_blocks[i]);
2252         }
2253         cur_dblock =
2254             SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2255             ((trans_offset + get_desc_trans_len(desc) +
2256               2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2257         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2258                        "journal-1095: setting journal " "start to offset %ld",
2259                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2260
2261         /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2262         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2263         journal->j_last_flush_trans_id = trans_id;
2264         journal->j_trans_id = trans_id + 1;
2265         /* check for trans_id overflow */
2266         if (journal->j_trans_id == 0)
2267                 journal->j_trans_id = 10;
2268         brelse(c_bh);
2269         brelse(d_bh);
2270         kfree(log_blocks);
2271         kfree(real_blocks);
2272         return 0;
2273 }
2274
2275 /* This function reads blocks starting from block and to max_block of bufsize
2276    size (but no more than BUFNR blocks at a time). This proved to improve
2277    mounting speed on self-rebuilding raid5 arrays at least.
2278    Right now it is only used from journal code. But later we might use it
2279    from other places.
2280    Note: Do not use journal_getblk/sb_getblk functions here! */
2281 static struct buffer_head *reiserfs_breada(struct block_device *dev, int block,
2282                                            int bufsize, unsigned int max_block)
2283 {
2284         struct buffer_head *bhlist[BUFNR];
2285         unsigned int blocks = BUFNR;
2286         struct buffer_head *bh;
2287         int i, j;
2288
2289         bh = __getblk(dev, block, bufsize);
2290         if (buffer_uptodate(bh))
2291                 return (bh);
2292
2293         if (block + BUFNR > max_block) {
2294                 blocks = max_block - block;
2295         }
2296         bhlist[0] = bh;
2297         j = 1;
2298         for (i = 1; i < blocks; i++) {
2299                 bh = __getblk(dev, block + i, bufsize);
2300                 if (buffer_uptodate(bh)) {
2301                         brelse(bh);
2302                         break;
2303                 } else
2304                         bhlist[j++] = bh;
2305         }
2306         ll_rw_block(READ, j, bhlist);
2307         for (i = 1; i < j; i++)
2308                 brelse(bhlist[i]);
2309         bh = bhlist[0];
2310         wait_on_buffer(bh);
2311         if (buffer_uptodate(bh))
2312                 return bh;
2313         brelse(bh);
2314         return NULL;
2315 }
2316
2317 /*
2318 ** read and replay the log
2319 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2320 ** transaction.  This tests that before finding all the transactions in the log, which makes normal mount times fast.
2321 **
2322 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2323 **
2324 ** On exit, it sets things up so the first transaction will work correctly.
2325 */
2326 static int journal_read(struct super_block *p_s_sb)
2327 {
2328         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2329         struct reiserfs_journal_desc *desc;
2330         unsigned long oldest_trans_id = 0;
2331         unsigned long oldest_invalid_trans_id = 0;
2332         time_t start;
2333         unsigned long oldest_start = 0;
2334         unsigned long cur_dblock = 0;
2335         unsigned long newest_mount_id = 9;
2336         struct buffer_head *d_bh;
2337         struct reiserfs_journal_header *jh;
2338         int valid_journal_header = 0;
2339         int replay_count = 0;
2340         int continue_replay = 1;
2341         int ret;
2342         char b[BDEVNAME_SIZE];
2343
2344         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2345         reiserfs_info(p_s_sb, "checking transaction log (%s)\n",
2346                       bdevname(journal->j_dev_bd, b));
2347         start = get_seconds();
2348
2349         /* step 1, read in the journal header block.  Check the transaction it says 
2350          ** is the first unflushed, and if that transaction is not valid, 
2351          ** replay is done
2352          */
2353         journal->j_header_bh = journal_bread(p_s_sb,
2354                                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)
2355                                              + SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2356         if (!journal->j_header_bh) {
2357                 return 1;
2358         }
2359         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2360         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2361             SB_ONDISK_JOURNAL_SIZE(p_s_sb)
2362             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2363                 oldest_start =
2364                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2365                     le32_to_cpu(jh->j_first_unflushed_offset);
2366                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2367                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2368                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2369                                "journal-1153: found in "
2370                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2371                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2372                                le32_to_cpu(jh->j_last_flush_trans_id));
2373                 valid_journal_header = 1;
2374
2375                 /* now, we try to read the first unflushed offset.  If it is not valid, 
2376                  ** there is nothing more we can do, and it makes no sense to read 
2377                  ** through the whole log.
2378                  */
2379                 d_bh =
2380                     journal_bread(p_s_sb,
2381                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2382                                   le32_to_cpu(jh->j_first_unflushed_offset));
2383                 ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL);
2384                 if (!ret) {
2385                         continue_replay = 0;
2386                 }
2387                 brelse(d_bh);
2388                 goto start_log_replay;
2389         }
2390
2391         if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) {
2392                 reiserfs_warning(p_s_sb,
2393                                  "clm-2076: device is readonly, unable to replay log");
2394                 return -1;
2395         }
2396
2397         /* ok, there are transactions that need to be replayed.  start with the first log block, find
2398          ** all the valid transactions, and pick out the oldest.
2399          */
2400         while (continue_replay
2401                && cur_dblock <
2402                (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2403                 SB_ONDISK_JOURNAL_SIZE(p_s_sb))) {
2404                 /* Note that it is required for blocksize of primary fs device and journal
2405                    device to be the same */
2406                 d_bh =
2407                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2408                                     p_s_sb->s_blocksize,
2409                                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2410                                     SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2411                 ret =
2412                     journal_transaction_is_valid(p_s_sb, d_bh,
2413                                                  &oldest_invalid_trans_id,
2414                                                  &newest_mount_id);
2415                 if (ret == 1) {
2416                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2417                         if (oldest_start == 0) {        /* init all oldest_ values */
2418                                 oldest_trans_id = get_desc_trans_id(desc);
2419                                 oldest_start = d_bh->b_blocknr;
2420                                 newest_mount_id = get_desc_mount_id(desc);
2421                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2422                                                "journal-1179: Setting "
2423                                                "oldest_start to offset %llu, trans_id %lu",
2424                                                oldest_start -
2425                                                SB_ONDISK_JOURNAL_1st_BLOCK
2426                                                (p_s_sb), oldest_trans_id);
2427                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2428                                 /* one we just read was older */
2429                                 oldest_trans_id = get_desc_trans_id(desc);
2430                                 oldest_start = d_bh->b_blocknr;
2431                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2432                                                "journal-1180: Resetting "
2433                                                "oldest_start to offset %lu, trans_id %lu",
2434                                                oldest_start -
2435                                                SB_ONDISK_JOURNAL_1st_BLOCK
2436                                                (p_s_sb), oldest_trans_id);
2437                         }
2438                         if (newest_mount_id < get_desc_mount_id(desc)) {
2439                                 newest_mount_id = get_desc_mount_id(desc);
2440                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2441                                                "journal-1299: Setting "
2442                                                "newest_mount_id to %d",
2443                                                get_desc_mount_id(desc));
2444                         }
2445                         cur_dblock += get_desc_trans_len(desc) + 2;
2446                 } else {
2447                         cur_dblock++;
2448                 }
2449                 brelse(d_bh);
2450         }
2451
2452       start_log_replay:
2453         cur_dblock = oldest_start;
2454         if (oldest_trans_id) {
2455                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2456                                "journal-1206: Starting replay "
2457                                "from offset %llu, trans_id %lu",
2458                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2459                                oldest_trans_id);
2460
2461         }
2462         replay_count = 0;
2463         while (continue_replay && oldest_trans_id > 0) {
2464                 ret =
2465                     journal_read_transaction(p_s_sb, cur_dblock, oldest_start,
2466                                              oldest_trans_id, newest_mount_id);
2467                 if (ret < 0) {
2468                         return ret;
2469                 } else if (ret != 0) {
2470                         break;
2471                 }
2472                 cur_dblock =
2473                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + journal->j_start;
2474                 replay_count++;
2475                 if (cur_dblock == oldest_start)
2476                         break;
2477         }
2478
2479         if (oldest_trans_id == 0) {
2480                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2481                                "journal-1225: No valid " "transactions found");
2482         }
2483         /* j_start does not get set correctly if we don't replay any transactions.
2484          ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2485          ** copy the trans_id from the header
2486          */
2487         if (valid_journal_header && replay_count == 0) {
2488                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2489                 journal->j_trans_id =
2490                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2491                 /* check for trans_id overflow */
2492                 if (journal->j_trans_id == 0)
2493                         journal->j_trans_id = 10;
2494                 journal->j_last_flush_trans_id =
2495                     le32_to_cpu(jh->j_last_flush_trans_id);
2496                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2497         } else {
2498                 journal->j_mount_id = newest_mount_id + 1;
2499         }
2500         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2501                        "newest_mount_id to %lu", journal->j_mount_id);
2502         journal->j_first_unflushed_offset = journal->j_start;
2503         if (replay_count > 0) {
2504                 reiserfs_info(p_s_sb,
2505                               "replayed %d transactions in %lu seconds\n",
2506                               replay_count, get_seconds() - start);
2507         }
2508         if (!bdev_read_only(p_s_sb->s_bdev) &&
2509             _update_journal_header_block(p_s_sb, journal->j_start,
2510                                          journal->j_last_flush_trans_id)) {
2511                 /* replay failed, caller must call free_journal_ram and abort
2512                  ** the mount
2513                  */
2514                 return -1;
2515         }
2516         return 0;
2517 }
2518
2519 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2520 {
2521         struct reiserfs_journal_list *jl;
2522         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2523                      GFP_NOFS | __GFP_NOFAIL);
2524         INIT_LIST_HEAD(&jl->j_list);
2525         INIT_LIST_HEAD(&jl->j_working_list);
2526         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2527         INIT_LIST_HEAD(&jl->j_bh_list);
2528         sema_init(&jl->j_commit_lock, 1);
2529         SB_JOURNAL(s)->j_num_lists++;
2530         get_journal_list(jl);
2531         return jl;
2532 }
2533
2534 static void journal_list_init(struct super_block *p_s_sb)
2535 {
2536         SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
2537 }
2538
2539 static int release_journal_dev(struct super_block *super,
2540                                struct reiserfs_journal *journal)
2541 {
2542         int result;
2543
2544         result = 0;
2545
2546         if (journal->j_dev_file != NULL) {
2547                 result = filp_close(journal->j_dev_file, NULL);
2548                 journal->j_dev_file = NULL;
2549                 journal->j_dev_bd = NULL;
2550         } else if (journal->j_dev_bd != NULL) {
2551                 result = blkdev_put(journal->j_dev_bd);
2552                 journal->j_dev_bd = NULL;
2553         }
2554
2555         if (result != 0) {
2556                 reiserfs_warning(super,
2557                                  "sh-457: release_journal_dev: Cannot release journal device: %i",
2558                                  result);
2559         }
2560         return result;
2561 }
2562
2563 static int journal_init_dev(struct super_block *super,
2564                             struct reiserfs_journal *journal,
2565                             const char *jdev_name)
2566 {
2567         int result;
2568         dev_t jdev;
2569         int blkdev_mode = FMODE_READ | FMODE_WRITE;
2570         char b[BDEVNAME_SIZE];
2571
2572         result = 0;
2573
2574         journal->j_dev_bd = NULL;
2575         journal->j_dev_file = NULL;
2576         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2577             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2578
2579         if (bdev_read_only(super->s_bdev))
2580                 blkdev_mode = FMODE_READ;
2581
2582         /* there is no "jdev" option and journal is on separate device */
2583         if ((!jdev_name || !jdev_name[0])) {
2584                 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2585                 if (IS_ERR(journal->j_dev_bd)) {
2586                         result = PTR_ERR(journal->j_dev_bd);
2587                         journal->j_dev_bd = NULL;
2588                         reiserfs_warning(super, "sh-458: journal_init_dev: "
2589                                          "cannot init journal device '%s': %i",
2590                                          __bdevname(jdev, b), result);
2591                         return result;
2592                 } else if (jdev != super->s_dev)
2593                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2594                 return 0;
2595         }
2596
2597         journal->j_dev_file = filp_open(jdev_name, 0, 0);
2598         if (!IS_ERR(journal->j_dev_file)) {
2599                 struct inode *jdev_inode = journal->j_dev_file->f_mapping->host;
2600                 if (!S_ISBLK(jdev_inode->i_mode)) {
2601                         reiserfs_warning(super, "journal_init_dev: '%s' is "
2602                                          "not a block device", jdev_name);
2603                         result = -ENOTBLK;
2604                         release_journal_dev(super, journal);
2605                 } else {
2606                         /* ok */
2607                         journal->j_dev_bd = I_BDEV(jdev_inode);
2608                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2609                         reiserfs_info(super,
2610                                       "journal_init_dev: journal device: %s\n",
2611                                       bdevname(journal->j_dev_bd, b));
2612                 }
2613         } else {
2614                 result = PTR_ERR(journal->j_dev_file);
2615                 journal->j_dev_file = NULL;
2616                 reiserfs_warning(super,
2617                                  "journal_init_dev: Cannot open '%s': %i",
2618                                  jdev_name, result);
2619         }
2620         return result;
2621 }
2622
2623 /*
2624 ** must be called once on fs mount.  calls journal_read for you
2625 */
2626 int journal_init(struct super_block *p_s_sb, const char *j_dev_name,
2627                  int old_format, unsigned int commit_max_age)
2628 {
2629         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2;
2630         struct buffer_head *bhjh;
2631         struct reiserfs_super_block *rs;
2632         struct reiserfs_journal_header *jh;
2633         struct reiserfs_journal *journal;
2634         struct reiserfs_journal_list *jl;
2635         char b[BDEVNAME_SIZE];
2636
2637         journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof(struct reiserfs_journal));
2638         if (!journal) {
2639                 reiserfs_warning(p_s_sb,
2640                                  "journal-1256: unable to get memory for journal structure");
2641                 return 1;
2642         }
2643         memset(journal, 0, sizeof(struct reiserfs_journal));
2644         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2645         INIT_LIST_HEAD(&journal->j_prealloc_list);
2646         INIT_LIST_HEAD(&journal->j_working_list);
2647         INIT_LIST_HEAD(&journal->j_journal_list);
2648         journal->j_persistent_trans = 0;
2649         if (reiserfs_allocate_list_bitmaps(p_s_sb,
2650                                            journal->j_list_bitmap,
2651                                            SB_BMAP_NR(p_s_sb)))
2652                 goto free_and_return;
2653         allocate_bitmap_nodes(p_s_sb);
2654
2655         /* reserved for journal area support */
2656         SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ?
2657                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2658                                                  / p_s_sb->s_blocksize +
2659                                                  SB_BMAP_NR(p_s_sb) +
2660                                                  1 :
2661                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2662                                                  p_s_sb->s_blocksize + 2);
2663
2664         /* Sanity check to see is the standard journal fitting withing first bitmap
2665            (actual for small blocksizes) */
2666         if (!SB_ONDISK_JOURNAL_DEVICE(p_s_sb) &&
2667             (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) +
2668              SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8)) {
2669                 reiserfs_warning(p_s_sb,
2670                                  "journal-1393: journal does not fit for area "
2671                                  "addressed by first of bitmap blocks. It starts at "
2672                                  "%u and its size is %u. Block size %ld",
2673                                  SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb),
2674                                  SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2675                                  p_s_sb->s_blocksize);
2676                 goto free_and_return;
2677         }
2678
2679         if (journal_init_dev(p_s_sb, journal, j_dev_name) != 0) {
2680                 reiserfs_warning(p_s_sb,
2681                                  "sh-462: unable to initialize jornal device");
2682                 goto free_and_return;
2683         }
2684
2685         rs = SB_DISK_SUPER_BLOCK(p_s_sb);
2686
2687         /* read journal header */
2688         bhjh = journal_bread(p_s_sb,
2689                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2690                              SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2691         if (!bhjh) {
2692                 reiserfs_warning(p_s_sb,
2693                                  "sh-459: unable to read journal header");
2694                 goto free_and_return;
2695         }
2696         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2697
2698         /* make sure that journal matches to the super block */
2699         if (is_reiserfs_jr(rs)
2700             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2701                 sb_jp_journal_magic(rs))) {
2702                 reiserfs_warning(p_s_sb,
2703                                  "sh-460: journal header magic %x "
2704                                  "(device %s) does not match to magic found in super "
2705                                  "block %x", jh->jh_journal.jp_journal_magic,
2706                                  bdevname(journal->j_dev_bd, b),
2707                                  sb_jp_journal_magic(rs));
2708                 brelse(bhjh);
2709                 goto free_and_return;
2710         }
2711
2712         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2713         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2714         journal->j_max_commit_age =
2715             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2716         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2717
2718         if (journal->j_trans_max) {
2719                 /* make sure these parameters are available, assign it if they are not */
2720                 __u32 initial = journal->j_trans_max;
2721                 __u32 ratio = 1;
2722
2723                 if (p_s_sb->s_blocksize < 4096)
2724                         ratio = 4096 / p_s_sb->s_blocksize;
2725
2726                 if (SB_ONDISK_JOURNAL_SIZE(p_s_sb) / journal->j_trans_max <
2727                     JOURNAL_MIN_RATIO)
2728                         journal->j_trans_max =
2729                             SB_ONDISK_JOURNAL_SIZE(p_s_sb) / JOURNAL_MIN_RATIO;
2730                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio)
2731                         journal->j_trans_max =
2732                             JOURNAL_TRANS_MAX_DEFAULT / ratio;
2733                 if (journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio)
2734                         journal->j_trans_max =
2735                             JOURNAL_TRANS_MIN_DEFAULT / ratio;
2736
2737                 if (journal->j_trans_max != initial)
2738                         reiserfs_warning(p_s_sb,
2739                                          "sh-461: journal_init: wrong transaction max size (%u). Changed to %u",
2740                                          initial, journal->j_trans_max);
2741
2742                 journal->j_max_batch = journal->j_trans_max *
2743                     JOURNAL_MAX_BATCH_DEFAULT / JOURNAL_TRANS_MAX_DEFAULT;
2744         }
2745
2746         if (!journal->j_trans_max) {
2747                 /*we have the file system was created by old version of mkreiserfs 
2748                    so this field contains zero value */
2749                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2750                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2751                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2752
2753                 /* for blocksize >= 4096 - max transaction size is 1024. For block size < 4096
2754                    trans max size is decreased proportionally */
2755                 if (p_s_sb->s_blocksize < 4096) {
2756                         journal->j_trans_max /= (4096 / p_s_sb->s_blocksize);
2757                         journal->j_max_batch = (journal->j_trans_max) * 9 / 10;
2758                 }
2759         }
2760
2761         journal->j_default_max_commit_age = journal->j_max_commit_age;
2762
2763         if (commit_max_age != 0) {
2764                 journal->j_max_commit_age = commit_max_age;
2765                 journal->j_max_trans_age = commit_max_age;
2766         }
2767
2768         reiserfs_info(p_s_sb, "journal params: device %s, size %u, "
2769                       "journal first block %u, max trans len %u, max batch %u, "
2770                       "max commit age %u, max trans age %u\n",
2771                       bdevname(journal->j_dev_bd, b),
2772                       SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2773                       SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2774                       journal->j_trans_max,
2775                       journal->j_max_batch,
2776                       journal->j_max_commit_age, journal->j_max_trans_age);
2777
2778         brelse(bhjh);
2779
2780         journal->j_list_bitmap_index = 0;
2781         journal_list_init(p_s_sb);
2782
2783         memset(journal->j_list_hash_table, 0,
2784                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2785
2786         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2787         spin_lock_init(&journal->j_dirty_buffers_lock);
2788
2789         journal->j_start = 0;
2790         journal->j_len = 0;
2791         journal->j_len_alloc = 0;
2792         atomic_set(&(journal->j_wcount), 0);
2793         atomic_set(&(journal->j_async_throttle), 0);
2794         journal->j_bcount = 0;
2795         journal->j_trans_start_time = 0;
2796         journal->j_last = NULL;
2797         journal->j_first = NULL;
2798         init_waitqueue_head(&(journal->j_join_wait));
2799         sema_init(&journal->j_lock, 1);
2800         sema_init(&journal->j_flush_sem, 1);
2801
2802         journal->j_trans_id = 10;
2803         journal->j_mount_id = 10;
2804         journal->j_state = 0;
2805         atomic_set(&(journal->j_jlock), 0);
2806         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2807         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2808         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2809         journal->j_cnode_used = 0;
2810         journal->j_must_wait = 0;
2811
2812         if (journal->j_cnode_free == 0) {
2813                 reiserfs_warning(p_s_sb, "journal-2004: Journal cnode memory "
2814                                  "allocation failed (%ld bytes). Journal is "
2815                                  "too large for available memory. Usually "
2816                                  "this is due to a journal that is too large.",
2817                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2818                 goto free_and_return;
2819         }
2820
2821         init_journal_hash(p_s_sb);
2822         jl = journal->j_current_jl;
2823         jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl);
2824         if (!jl->j_list_bitmap) {
2825                 reiserfs_warning(p_s_sb,
2826                                  "journal-2005, get_list_bitmap failed for journal list 0");
2827                 goto free_and_return;
2828         }
2829         if (journal_read(p_s_sb) < 0) {
2830                 reiserfs_warning(p_s_sb, "Replay Failure, unable to mount");
2831                 goto free_and_return;
2832         }
2833
2834         reiserfs_mounted_fs_count++;
2835         if (reiserfs_mounted_fs_count <= 1)
2836                 commit_wq = create_workqueue("reiserfs");
2837
2838         INIT_WORK(&journal->j_work, flush_async_commits, p_s_sb);
2839         return 0;
2840       free_and_return:
2841         free_journal_ram(p_s_sb);
2842         return 1;
2843 }
2844
2845 /*
2846 ** test for a polite end of the current transaction.  Used by file_write, and should
2847 ** be used by delete to make sure they don't write more than can fit inside a single
2848 ** transaction
2849 */
2850 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2851                                    int new_alloc)
2852 {
2853         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2854         time_t now = get_seconds();
2855         /* cannot restart while nested */
2856         BUG_ON(!th->t_trans_id);
2857         if (th->t_refcount > 1)
2858                 return 0;
2859         if (journal->j_must_wait > 0 ||
2860             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2861             atomic_read(&(journal->j_jlock)) ||
2862             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2863             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2864                 return 1;
2865         }
2866         /* protected by the BKL here */
2867         journal->j_len_alloc += new_alloc;
2868         th->t_blocks_allocated += new_alloc ;
2869         return 0;
2870 }
2871
2872 /* this must be called inside a transaction, and requires the 
2873 ** kernel_lock to be held
2874 */
2875 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2876 {
2877         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2878         BUG_ON(!th->t_trans_id);
2879         journal->j_must_wait = 1;
2880         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2881         return;
2882 }
2883
2884 /* this must be called without a transaction started, and does not
2885 ** require BKL
2886 */
2887 void reiserfs_allow_writes(struct super_block *s)
2888 {
2889         struct reiserfs_journal *journal = SB_JOURNAL(s);
2890         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2891         wake_up(&journal->j_join_wait);
2892 }
2893
2894 /* this must be called without a transaction started, and does not
2895 ** require BKL
2896 */
2897 void reiserfs_wait_on_write_block(struct super_block *s)
2898 {
2899         struct reiserfs_journal *journal = SB_JOURNAL(s);
2900         wait_event(journal->j_join_wait,
2901                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2902 }
2903
2904 static void queue_log_writer(struct super_block *s)
2905 {
2906         wait_queue_t wait;
2907         struct reiserfs_journal *journal = SB_JOURNAL(s);
2908         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2909
2910         /*
2911          * we don't want to use wait_event here because
2912          * we only want to wait once.
2913          */
2914         init_waitqueue_entry(&wait, current);
2915         add_wait_queue(&journal->j_join_wait, &wait);
2916         set_current_state(TASK_UNINTERRUPTIBLE);
2917         if (test_bit(J_WRITERS_QUEUED, &journal->j_state))
2918                 schedule();
2919         current->state = TASK_RUNNING;
2920         remove_wait_queue(&journal->j_join_wait, &wait);
2921 }
2922
2923 static void wake_queued_writers(struct super_block *s)
2924 {
2925         struct reiserfs_journal *journal = SB_JOURNAL(s);
2926         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2927                 wake_up(&journal->j_join_wait);
2928 }
2929
2930 static void let_transaction_grow(struct super_block *sb, unsigned long trans_id)
2931 {
2932         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2933         unsigned long bcount = journal->j_bcount;
2934         while (1) {
2935                 schedule_timeout_uninterruptible(1);
2936                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2937                 while ((atomic_read(&journal->j_wcount) > 0 ||
2938                         atomic_read(&journal->j_jlock)) &&
2939                        journal->j_trans_id == trans_id) {
2940                         queue_log_writer(sb);
2941                 }
2942                 if (journal->j_trans_id != trans_id)
2943                         break;
2944                 if (bcount == journal->j_bcount)
2945                         break;
2946                 bcount = journal->j_bcount;
2947         }
2948 }
2949
2950 /* join == true if you must join an existing transaction.
2951 ** join == false if you can deal with waiting for others to finish
2952 **
2953 ** this will block until the transaction is joinable.  send the number of blocks you
2954 ** expect to use in nblocks.
2955 */
2956 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
2957                               struct super_block *p_s_sb, unsigned long nblocks,
2958                               int join)
2959 {
2960         time_t now = get_seconds();
2961         int old_trans_id;
2962         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2963         struct reiserfs_transaction_handle myth;
2964         int sched_count = 0;
2965         int retval;
2966
2967         reiserfs_check_lock_depth(p_s_sb, "journal_begin");
2968         BUG_ON(nblocks > journal->j_trans_max);
2969
2970         PROC_INFO_INC(p_s_sb, journal.journal_being);
2971         /* set here for journal_join */
2972         th->t_refcount = 1;
2973         th->t_super = p_s_sb;
2974
2975       relock:
2976         lock_journal(p_s_sb);
2977         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
2978                 unlock_journal(p_s_sb);
2979                 retval = journal->j_errno;
2980                 goto out_fail;
2981         }
2982         journal->j_bcount++;
2983
2984         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
2985                 unlock_journal(p_s_sb);
2986                 reiserfs_wait_on_write_block(p_s_sb);
2987                 PROC_INFO_INC(p_s_sb, journal.journal_relock_writers);
2988                 goto relock;
2989         }
2990         now = get_seconds();
2991
2992         /* if there is no room in the journal OR
2993          ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning 
2994          ** we don't sleep if there aren't other writers
2995          */
2996
2997         if ((!join && journal->j_must_wait > 0) ||
2998             (!join
2999              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3000             || (!join && atomic_read(&journal->j_wcount) > 0
3001                 && journal->j_trans_start_time > 0
3002                 && (now - journal->j_trans_start_time) >
3003                 journal->j_max_trans_age) || (!join
3004                                               && atomic_read(&journal->j_jlock))
3005             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3006
3007                 old_trans_id = journal->j_trans_id;
3008                 unlock_journal(p_s_sb); /* allow others to finish this transaction */
3009
3010                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3011                     journal->j_max_batch &&
3012                     ((journal->j_len + nblocks + 2) * 100) <
3013                     (journal->j_len_alloc * 75)) {
3014                         if (atomic_read(&journal->j_wcount) > 10) {
3015                                 sched_count++;
3016                                 queue_log_writer(p_s_sb);
3017                                 goto relock;
3018                         }
3019                 }
3020                 /* don't mess with joining the transaction if all we have to do is
3021                  * wait for someone else to do a commit
3022                  */
3023                 if (atomic_read(&journal->j_jlock)) {
3024                         while (journal->j_trans_id == old_trans_id &&
3025                                atomic_read(&journal->j_jlock)) {
3026                                 queue_log_writer(p_s_sb);
3027                         }
3028                         goto relock;
3029                 }
3030                 retval = journal_join(&myth, p_s_sb, 1);
3031                 if (retval)
3032                         goto out_fail;
3033
3034                 /* someone might have ended the transaction while we joined */
3035                 if (old_trans_id != journal->j_trans_id) {
3036                         retval = do_journal_end(&myth, p_s_sb, 1, 0);
3037                 } else {
3038                         retval = do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW);
3039                 }
3040
3041                 if (retval)
3042                         goto out_fail;
3043
3044                 PROC_INFO_INC(p_s_sb, journal.journal_relock_wcount);
3045                 goto relock;
3046         }
3047         /* we are the first writer, set trans_id */
3048         if (journal->j_trans_start_time == 0) {
3049                 journal->j_trans_start_time = get_seconds();
3050         }
3051         atomic_inc(&(journal->j_wcount));
3052         journal->j_len_alloc += nblocks;
3053         th->t_blocks_logged = 0;
3054         th->t_blocks_allocated = nblocks;
3055         th->t_trans_id = journal->j_trans_id;
3056         unlock_journal(p_s_sb);
3057         INIT_LIST_HEAD(&th->t_list);
3058         get_fs_excl();
3059         return 0;
3060
3061       out_fail:
3062         memset(th, 0, sizeof(*th));
3063         /* Re-set th->t_super, so we can properly keep track of how many
3064          * persistent transactions there are. We need to do this so if this
3065          * call is part of a failed restart_transaction, we can free it later */
3066         th->t_super = p_s_sb;
3067         return retval;
3068 }
3069
3070 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3071                                                                     super_block
3072                                                                     *s,
3073                                                                     int nblocks)
3074 {
3075         int ret;
3076         struct reiserfs_transaction_handle *th;
3077
3078         /* if we're nesting into an existing transaction.  It will be
3079          ** persistent on its own
3080          */
3081         if (reiserfs_transaction_running(s)) {
3082                 th = current->journal_info;
3083                 th->t_refcount++;
3084                 BUG_ON(th->t_refcount < 2);
3085                 
3086                 return th;
3087         }
3088         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3089         if (!th)
3090                 return NULL;
3091         ret = journal_begin(th, s, nblocks);
3092         if (ret) {
3093                 kfree(th);
3094                 return NULL;
3095         }
3096
3097         SB_JOURNAL(s)->j_persistent_trans++;
3098         return th;
3099 }
3100
3101 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3102 {
3103         struct super_block *s = th->t_super;
3104         int ret = 0;
3105         if (th->t_trans_id)
3106                 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3107         else
3108                 ret = -EIO;
3109         if (th->t_refcount == 0) {
3110                 SB_JOURNAL(s)->j_persistent_trans--;
3111                 kfree(th);
3112         }
3113         return ret;
3114 }
3115
3116 static int journal_join(struct reiserfs_transaction_handle *th,
3117                         struct super_block *p_s_sb, unsigned long nblocks)
3118 {
3119         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3120
3121         /* this keeps do_journal_end from NULLing out the current->journal_info
3122          ** pointer
3123          */
3124         th->t_handle_save = cur_th;
3125         BUG_ON(cur_th && cur_th->t_refcount > 1);
3126         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_JOIN);
3127 }
3128
3129 int journal_join_abort(struct reiserfs_transaction_handle *th,
3130                        struct super_block *p_s_sb, unsigned long nblocks)
3131 {
3132         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3133
3134         /* this keeps do_journal_end from NULLing out the current->journal_info
3135          ** pointer
3136          */
3137         th->t_handle_save = cur_th;
3138         BUG_ON(cur_th && cur_th->t_refcount > 1);
3139         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_ABORT);
3140 }
3141
3142 int journal_begin(struct reiserfs_transaction_handle *th,
3143                   struct super_block *p_s_sb, unsigned long nblocks)
3144 {
3145         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3146         int ret;
3147
3148         th->t_handle_save = NULL;
3149         if (cur_th) {
3150                 /* we are nesting into the current transaction */
3151                 if (cur_th->t_super == p_s_sb) {
3152                         BUG_ON(!cur_th->t_refcount);
3153                         cur_th->t_refcount++;
3154                         memcpy(th, cur_th, sizeof(*th));
3155                         if (th->t_refcount <= 1)
3156                                 reiserfs_warning(p_s_sb,
3157                                                  "BAD: refcount <= 1, but journal_info != 0");
3158                         return 0;
3159                 } else {
3160                         /* we've ended up with a handle from a different filesystem.
3161                          ** save it and restore on journal_end.  This should never
3162                          ** really happen...
3163                          */
3164                         reiserfs_warning(p_s_sb,
3165                                          "clm-2100: nesting info a different FS");
3166                         th->t_handle_save = current->journal_info;
3167                         current->journal_info = th;
3168                 }
3169         } else {
3170                 current->journal_info = th;
3171         }
3172         ret = do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_REG);
3173         BUG_ON(current->journal_info != th);
3174
3175         /* I guess this boils down to being the reciprocal of clm-2100 above.
3176          * If do_journal_begin_r fails, we need to put it back, since journal_end
3177          * won't be called to do it. */
3178         if (ret)
3179                 current->journal_info = th->t_handle_save;
3180         else
3181                 BUG_ON(!th->t_refcount);
3182
3183         return ret;
3184 }
3185
3186 /*
3187 ** puts bh into the current transaction.  If it was already there, reorders removes the
3188 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3189 **
3190 ** if it was dirty, cleans and files onto the clean list.  I can't let it be dirty again until the
3191 ** transaction is committed.
3192 ** 
3193 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3194 */
3195 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3196                        struct super_block *p_s_sb, struct buffer_head *bh)
3197 {
3198         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3199         struct reiserfs_journal_cnode *cn = NULL;
3200         int count_already_incd = 0;
3201         int prepared = 0;
3202         BUG_ON(!th->t_trans_id);
3203
3204         PROC_INFO_INC(p_s_sb, journal.mark_dirty);
3205         if (th->t_trans_id != journal->j_trans_id) {
3206                 reiserfs_panic(th->t_super,
3207                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3208                                th->t_trans_id, journal->j_trans_id);
3209         }
3210
3211         p_s_sb->s_dirt = 1;
3212
3213         prepared = test_clear_buffer_journal_prepared(bh);
3214         clear_buffer_journal_restore_dirty(bh);
3215         /* already in this transaction, we are done */
3216         if (buffer_journaled(bh)) {
3217                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_already);
3218                 return 0;
3219         }
3220
3221         /* this must be turned into a panic instead of a warning.  We can't allow
3222          ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3223          ** could get to disk too early.  NOT GOOD.
3224          */
3225         if (!prepared || buffer_dirty(bh)) {
3226                 reiserfs_warning(p_s_sb, "journal-1777: buffer %llu bad state "
3227                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3228                                  (unsigned long long)bh->b_blocknr,
3229                                  prepared ? ' ' : '!',
3230                                  buffer_locked(bh) ? ' ' : '!',
3231                                  buffer_dirty(bh) ? ' ' : '!',
3232                                  buffer_journal_dirty(bh) ? ' ' : '!');
3233         }
3234
3235         if (atomic_read(&(journal->j_wcount)) <= 0) {
3236                 reiserfs_warning(p_s_sb,
3237                                  "journal-1409: journal_mark_dirty returning because j_wcount was %d",
3238                                  atomic_read(&(journal->j_wcount)));
3239                 return 1;
3240         }
3241         /* this error means I've screwed up, and we've overflowed the transaction.  
3242          ** Nothing can be done here, except make the FS readonly or panic.
3243          */
3244         if (journal->j_len >= journal->j_trans_max) {
3245                 reiserfs_panic(th->t_super,
3246                                "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n",
3247                                journal->j_len);
3248         }
3249
3250         if (buffer_journal_dirty(bh)) {
3251                 count_already_incd = 1;
3252                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_notjournal);
3253                 clear_buffer_journal_dirty(bh);
3254         }
3255
3256         if (journal->j_len > journal->j_len_alloc) {
3257                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3258         }
3259
3260         set_buffer_journaled(bh);
3261
3262         /* now put this guy on the end */
3263         if (!cn) {
3264                 cn = get_cnode(p_s_sb);
3265                 if (!cn) {
3266                         reiserfs_panic(p_s_sb, "get_cnode failed!\n");
3267                 }
3268
3269                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3270                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3271                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3272                 }
3273                 th->t_blocks_logged++;
3274                 journal->j_len++;
3275
3276                 cn->bh = bh;
3277                 cn->blocknr = bh->b_blocknr;
3278                 cn->sb = p_s_sb;
3279                 cn->jlist = NULL;
3280                 insert_journal_hash(journal->j_hash_table, cn);
3281                 if (!count_already_incd) {
3282                         get_bh(bh);
3283                 }
3284         }
3285         cn->next = NULL;
3286         cn->prev = journal->j_last;
3287         cn->bh = bh;
3288         if (journal->j_last) {
3289                 journal->j_last->next = cn;
3290                 journal->j_last = cn;
3291         } else {
3292                 journal->j_first = cn;
3293                 journal->j_last = cn;
3294         }
3295         return 0;
3296 }
3297
3298 int journal_end(struct reiserfs_transaction_handle *th,
3299                 struct super_block *p_s_sb, unsigned long nblocks)
3300 {
3301         if (!current->journal_info && th->t_refcount > 1)
3302                 reiserfs_warning(p_s_sb, "REISER-NESTING: th NULL, refcount %d",
3303                                  th->t_refcount);
3304
3305         if (!th->t_trans_id) {
3306                 WARN_ON(1);
3307                 return -EIO;
3308         }
3309
3310         th->t_refcount--;
3311         if (th->t_refcount > 0) {
3312                 struct reiserfs_transaction_handle *cur_th =
3313                     current->journal_info;
3314
3315                 /* we aren't allowed to close a nested transaction on a different
3316                  ** filesystem from the one in the task struct
3317                  */
3318                 BUG_ON(cur_th->t_super != th->t_super);
3319
3320                 if (th != cur_th) {
3321                         memcpy(current->journal_info, th, sizeof(*th));
3322                         th->t_trans_id = 0;
3323                 }
3324                 return 0;
3325         } else {
3326                 return do_journal_end(th, p_s_sb, nblocks, 0);
3327         }
3328 }
3329
3330 /* removes from the current transaction, relsing and descrementing any counters.  
3331 ** also files the removed buffer directly onto the clean list
3332 **
3333 ** called by journal_mark_freed when a block has been deleted
3334 **
3335 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3336 */
3337 static int remove_from_transaction(struct super_block *p_s_sb,
3338                                    b_blocknr_t blocknr, int already_cleaned)
3339 {
3340         struct buffer_head *bh;
3341         struct reiserfs_journal_cnode *cn;
3342         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3343         int ret = 0;
3344
3345         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3346         if (!cn || !cn->bh) {
3347                 return ret;
3348         }
3349         bh = cn->bh;
3350         if (cn->prev) {
3351                 cn->prev->next = cn->next;
3352         }
3353         if (cn->next) {
3354                 cn->next->prev = cn->prev;
3355         }
3356         if (cn == journal->j_first) {
3357                 journal->j_first = cn->next;
3358         }
3359         if (cn == journal->j_last) {
3360                 journal->j_last = cn->prev;
3361         }
3362         if (bh)
3363                 remove_journal_hash(p_s_sb, journal->j_hash_table, NULL,
3364                                     bh->b_blocknr, 0);
3365         clear_buffer_journaled(bh);     /* don't log this one */
3366
3367         if (!already_cleaned) {
3368                 clear_buffer_journal_dirty(bh);
3369                 clear_buffer_dirty(bh);
3370                 clear_buffer_journal_test(bh);
3371                 put_bh(bh);
3372                 if (atomic_read(&(bh->b_count)) < 0) {
3373                         reiserfs_warning(p_s_sb,
3374                                          "journal-1752: remove from trans, b_count < 0");
3375                 }
3376                 ret = 1;
3377         }
3378         journal->j_len--;
3379         journal->j_len_alloc--;
3380         free_cnode(p_s_sb, cn);
3381         return ret;
3382 }
3383
3384 /*
3385 ** for any cnode in a journal list, it can only be dirtied of all the
3386 ** transactions that include it are commited to disk.
3387 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3388 ** and 0 if you aren't
3389 **
3390 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3391 ** blocks for a given transaction on disk
3392 **
3393 */
3394 static int can_dirty(struct reiserfs_journal_cnode *cn)
3395 {
3396         struct super_block *sb = cn->sb;
3397         b_blocknr_t blocknr = cn->blocknr;
3398         struct reiserfs_journal_cnode *cur = cn->hprev;
3399         int can_dirty = 1;
3400
3401         /* first test hprev.  These are all newer than cn, so any node here
3402          ** with the same block number and dev means this node can't be sent
3403          ** to disk right now.
3404          */
3405         while (cur && can_dirty) {
3406                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3407                     cur->blocknr == blocknr) {
3408                         can_dirty = 0;
3409                 }
3410                 cur = cur->hprev;
3411         }
3412         /* then test hnext.  These are all older than cn.  As long as they
3413          ** are committed to the log, it is safe to write cn to disk
3414          */
3415         cur = cn->hnext;
3416         while (cur && can_dirty) {
3417                 if (cur->jlist && cur->jlist->j_len > 0 &&
3418                     atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3419                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3420                         can_dirty = 0;
3421                 }
3422                 cur = cur->hnext;
3423         }
3424         return can_dirty;
3425 }
3426
3427 /* syncs the commit blocks, but does not force the real buffers to disk
3428 ** will wait until the current transaction is done/commited before returning 
3429 */
3430 int journal_end_sync(struct reiserfs_transaction_handle *th,
3431                      struct super_block *p_s_sb, unsigned long nblocks)
3432 {
3433         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3434
3435         BUG_ON(!th->t_trans_id);
3436         /* you can sync while nested, very, very bad */
3437         BUG_ON(th->t_refcount > 1);
3438         if (journal->j_len == 0) {
3439                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3440                                              1);
3441                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3442         }
3443         return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT);
3444 }
3445
3446 /*
3447 ** writeback the pending async commits to disk
3448 */
3449 static void flush_async_commits(void *p)
3450 {
3451         struct super_block *p_s_sb = p;
3452         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3453         struct reiserfs_journal_list *jl;
3454         struct list_head *entry;
3455
3456         lock_kernel();
3457         if (!list_empty(&journal->j_journal_list)) {
3458                 /* last entry is the youngest, commit it and you get everything */
3459                 entry = journal->j_journal_list.prev;
3460                 jl = JOURNAL_LIST_ENTRY(entry);
3461                 flush_commit_list(p_s_sb, jl, 1);
3462         }
3463         unlock_kernel();
3464 }
3465
3466 /*
3467 ** flushes any old transactions to disk
3468 ** ends the current transaction if it is too old
3469 */
3470 int reiserfs_flush_old_commits(struct super_block *p_s_sb)
3471 {
3472         time_t now;
3473         struct reiserfs_transaction_handle th;
3474         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3475
3476         now = get_seconds();
3477         /* safety check so we don't flush while we are replaying the log during
3478          * mount
3479          */
3480         if (list_empty(&journal->j_journal_list)) {
3481                 return 0;
3482         }
3483
3484         /* check the current transaction.  If there are no writers, and it is
3485          * too old, finish it, and force the commit blocks to disk
3486          */
3487         if (atomic_read(&journal->j_wcount) <= 0 &&
3488             journal->j_trans_start_time > 0 &&
3489             journal->j_len > 0 &&
3490             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3491                 if (!journal_join(&th, p_s_sb, 1)) {
3492                         reiserfs_prepare_for_journal(p_s_sb,
3493                                                      SB_BUFFER_WITH_SB(p_s_sb),
3494                                                      1);
3495                         journal_mark_dirty(&th, p_s_sb,
3496                                            SB_BUFFER_WITH_SB(p_s_sb));
3497
3498                         /* we're only being called from kreiserfsd, it makes no sense to do
3499                          ** an async commit so that kreiserfsd can do it later
3500                          */
3501                         do_journal_end(&th, p_s_sb, 1, COMMIT_NOW | WAIT);
3502                 }
3503         }
3504         return p_s_sb->s_dirt;
3505 }
3506
3507 /*
3508 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3509 ** 
3510 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all 
3511 ** the writers are done.  By the time it wakes up, the transaction it was called has already ended, so it just
3512 ** flushes the commit list and returns 0.
3513 **
3514 ** Won't batch when flush or commit_now is set.  Also won't batch when others are waiting on j_join_wait.
3515 ** 
3516 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3517 */
3518 static int check_journal_end(struct reiserfs_transaction_handle *th,
3519                              struct super_block *p_s_sb, unsigned long nblocks,
3520                              int flags)
3521 {
3522
3523         time_t now;
3524         int flush = flags & FLUSH_ALL;
3525         int commit_now = flags & COMMIT_NOW;
3526         int wait_on_commit = flags & WAIT;
3527         struct reiserfs_journal_list *jl;
3528         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3529
3530         BUG_ON(!th->t_trans_id);
3531
3532         if (th->t_trans_id != journal->j_trans_id) {
3533                 reiserfs_panic(th->t_super,
3534                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3535                                th->t_trans_id, journal->j_trans_id);
3536         }
3537
3538         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3539         if (atomic_read(&(journal->j_wcount)) > 0) {    /* <= 0 is allowed.  unmounting might not call begin */
3540                 atomic_dec(&(journal->j_wcount));
3541         }
3542
3543         /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released 
3544          ** will be dealt with by next transaction that actually writes something, but should be taken
3545          ** care of in this trans
3546          */
3547         BUG_ON(journal->j_len == 0);
3548
3549         /* if wcount > 0, and we are called to with flush or commit_now,
3550          ** we wait on j_join_wait.  We will wake up when the last writer has
3551          ** finished the transaction, and started it on its way to the disk.
3552          ** Then, we flush the commit or journal list, and just return 0 
3553          ** because the rest of journal end was already done for this transaction.
3554          */
3555         if (atomic_read(&(journal->j_wcount)) > 0) {
3556                 if (flush || commit_now) {
3557                         unsigned trans_id;
3558
3559                         jl = journal->j_current_jl;
3560                         trans_id = jl->j_trans_id;
3561                         if (wait_on_commit)
3562                                 jl->j_state |= LIST_COMMIT_PENDING;
3563                         atomic_set(&(journal->j_jlock), 1);
3564                         if (flush) {
3565                                 journal->j_next_full_flush = 1;
3566                         }
3567                         unlock_journal(p_s_sb);
3568
3569                         /* sleep while the current transaction is still j_jlocked */
3570                         while (journal->j_trans_id == trans_id) {
3571                                 if (atomic_read(&journal->j_jlock)) {
3572                                         queue_log_writer(p_s_sb);
3573                                 } else {
3574                                         lock_journal(p_s_sb);
3575                                         if (journal->j_trans_id == trans_id) {
3576                                                 atomic_set(&(journal->j_jlock),
3577                                                            1);
3578                                         }
3579                                         unlock_journal(p_s_sb);
3580                                 }
3581                         }
3582                         BUG_ON(journal->j_trans_id == trans_id);
3583                         
3584                         if (commit_now
3585                             && journal_list_still_alive(p_s_sb, trans_id)
3586                             && wait_on_commit) {
3587                                 flush_commit_list(p_s_sb, jl, 1);
3588                         }
3589                         return 0;
3590                 }
3591                 unlock_journal(p_s_sb);
3592                 return 0;
3593         }
3594
3595         /* deal with old transactions where we are the last writers */
3596         now = get_seconds();
3597         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3598                 commit_now = 1;
3599                 journal->j_next_async_flush = 1;
3600         }
3601         /* don't batch when someone is waiting on j_join_wait */
3602         /* don't batch when syncing the commit or flushing the whole trans */
3603         if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3604             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3605             && journal->j_len_alloc < journal->j_max_batch
3606             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3607                 journal->j_bcount++;
3608                 unlock_journal(p_s_sb);
3609                 return 0;
3610         }
3611
3612         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
3613                 reiserfs_panic(p_s_sb,
3614                                "journal-003: journal_end: j_start (%ld) is too high\n",
3615                                journal->j_start);
3616         }
3617         return 1;
3618 }
3619
3620 /*
3621 ** Does all the work that makes deleting blocks safe.
3622 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3623 ** 
3624 ** otherwise:
3625 ** set a bit for the block in the journal bitmap.  That will prevent it from being allocated for unformatted nodes
3626 ** before this transaction has finished.
3627 **
3628 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.  That will prevent any old transactions with
3629 ** this block from trying to flush to the real location.  Since we aren't removing the cnode from the journal_list_hash,
3630 ** the block can't be reallocated yet.
3631 **
3632 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3633 */
3634 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3635                        struct super_block *p_s_sb, b_blocknr_t blocknr)
3636 {
3637         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3638         struct reiserfs_journal_cnode *cn = NULL;
3639         struct buffer_head *bh = NULL;
3640         struct reiserfs_list_bitmap *jb = NULL;
3641         int cleaned = 0;
3642         BUG_ON(!th->t_trans_id);
3643
3644         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3645         if (cn && cn->bh) {
3646                 bh = cn->bh;
3647                 get_bh(bh);
3648         }
3649         /* if it is journal new, we just remove it from this transaction */
3650         if (bh && buffer_journal_new(bh)) {
3651                 clear_buffer_journal_new(bh);
3652                 clear_prepared_bits(bh);
3653                 reiserfs_clean_and_file_buffer(bh);
3654                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3655         } else {
3656                 /* set the bit for this block in the journal bitmap for this transaction */
3657                 jb = journal->j_current_jl->j_list_bitmap;
3658                 if (!jb) {
3659                         reiserfs_panic(p_s_sb,
3660                                        "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n");
3661                 }
3662                 set_bit_in_list_bitmap(p_s_sb, blocknr, jb);
3663
3664                 /* Note, the entire while loop is not allowed to schedule.  */
3665
3666                 if (bh) {
3667                         clear_prepared_bits(bh);
3668                         reiserfs_clean_and_file_buffer(bh);
3669                 }
3670                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3671
3672                 /* find all older transactions with this block, make sure they don't try to write it out */
3673                 cn = get_journal_hash_dev(p_s_sb, journal->j_list_hash_table,
3674                                           blocknr);
3675                 while (cn) {
3676                         if (p_s_sb == cn->sb && blocknr == cn->blocknr) {
3677                                 set_bit(BLOCK_FREED, &cn->state);
3678                                 if (cn->bh) {
3679                                         if (!cleaned) {
3680                                                 /* remove_from_transaction will brelse the buffer if it was 
3681                                                  ** in the current trans
3682                                                  */
3683                                                 clear_buffer_journal_dirty(cn->
3684                                                                            bh);
3685                                                 clear_buffer_dirty(cn->bh);
3686                                                 clear_buffer_journal_test(cn->
3687                                                                           bh);
3688                                                 cleaned = 1;
3689                                                 put_bh(cn->bh);
3690                                                 if (atomic_read
3691                                                     (&(cn->bh->b_count)) < 0) {
3692                                                         reiserfs_warning(p_s_sb,
3693                                                                          "journal-2138: cn->bh->b_count < 0");
3694                                                 }
3695                                         }
3696                                         if (cn->jlist) {        /* since we are clearing the bh, we MUST dec nonzerolen */
3697                                                 atomic_dec(&
3698                                                            (cn->jlist->
3699                                                             j_nonzerolen));
3700                                         }
3701                                         cn->bh = NULL;
3702                                 }
3703                         }
3704                         cn = cn->hnext;
3705                 }
3706         }
3707
3708         if (bh) {
3709                 put_bh(bh);     /* get_hash grabs the buffer */
3710                 if (atomic_read(&(bh->b_count)) < 0) {
3711                         reiserfs_warning(p_s_sb,
3712                                          "journal-2165: bh->b_count < 0");
3713                 }
3714         }
3715         return 0;
3716 }
3717
3718 void reiserfs_update_inode_transaction(struct inode *inode)
3719 {
3720         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3721         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3722         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3723 }
3724
3725 /*
3726  * returns -1 on error, 0 if no commits/barriers were done and 1
3727  * if a transaction was actually committed and the barrier was done
3728  */
3729 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3730                              struct reiserfs_journal_list *jl)
3731 {
3732         struct reiserfs_transaction_handle th;
3733         struct super_block *sb = inode->i_sb;
3734         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3735         int ret = 0;
3736
3737         /* is it from the current transaction, or from an unknown transaction? */
3738         if (id == journal->j_trans_id) {
3739                 jl = journal->j_current_jl;
3740                 /* try to let other writers come in and grow this transaction */
3741                 let_transaction_grow(sb, id);
3742                 if (journal->j_trans_id != id) {
3743                         goto flush_commit_only;
3744                 }
3745
3746                 ret = journal_begin(&th, sb, 1);
3747                 if (ret)
3748                         return ret;
3749
3750                 /* someone might have ended this transaction while we joined */
3751                 if (journal->j_trans_id != id) {
3752                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3753                                                      1);
3754                         journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3755                         ret = journal_end(&th, sb, 1);
3756                         goto flush_commit_only;
3757                 }
3758
3759                 ret = journal_end_sync(&th, sb, 1);
3760                 if (!ret)
3761                         ret = 1;
3762
3763         } else {
3764                 /* this gets tricky, we have to make sure the journal list in
3765                  * the inode still exists.  We know the list is still around
3766                  * if we've got a larger transaction id than the oldest list
3767                  */
3768               flush_commit_only:
3769                 if (journal_list_still_alive(inode->i_sb, id)) {
3770                         /*
3771                          * we only set ret to 1 when we know for sure
3772                          * the barrier hasn't been started yet on the commit
3773                          * block.
3774                          */
3775                         if (atomic_read(&jl->j_commit_left) > 1)
3776                                 ret = 1;
3777                         flush_commit_list(sb, jl, 1);
3778                         if (journal->j_errno)
3779                                 ret = journal->j_errno;
3780                 }
3781         }
3782         /* otherwise the list is gone, and long since committed */
3783         return ret;
3784 }
3785
3786 int reiserfs_commit_for_inode(struct inode *inode)
3787 {
3788         unsigned long id = REISERFS_I(inode)->i_trans_id;
3789         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3790
3791         /* for the whole inode, assume unset id means it was
3792          * changed in the current transaction.  More conservative
3793          */
3794         if (!id || !jl) {
3795                 reiserfs_update_inode_transaction(inode);
3796                 id = REISERFS_I(inode)->i_trans_id;
3797                 /* jl will be updated in __commit_trans_jl */
3798         }
3799
3800         return __commit_trans_jl(inode, id, jl);
3801 }
3802
3803 void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb,
3804                                       struct buffer_head *bh)
3805 {
3806         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3807         PROC_INFO_INC(p_s_sb, journal.restore_prepared);
3808         if (!bh) {
3809                 return;
3810         }
3811         if (test_clear_buffer_journal_restore_dirty(bh) &&
3812             buffer_journal_dirty(bh)) {
3813                 struct reiserfs_journal_cnode *cn;
3814                 cn = get_journal_hash_dev(p_s_sb,
3815                                           journal->j_list_hash_table,
3816                                           bh->b_blocknr);
3817                 if (cn && can_dirty(cn)) {
3818                         set_buffer_journal_test(bh);
3819                         mark_buffer_dirty(bh);
3820                 }
3821         }
3822         clear_buffer_journal_prepared(bh);
3823 }
3824
3825 extern struct tree_balance *cur_tb;
3826 /*
3827 ** before we can change a metadata block, we have to make sure it won't
3828 ** be written to disk while we are altering it.  So, we must:
3829 ** clean it
3830 ** wait on it.
3831 ** 
3832 */
3833 int reiserfs_prepare_for_journal(struct super_block *p_s_sb,
3834                                  struct buffer_head *bh, int wait)
3835 {
3836         PROC_INFO_INC(p_s_sb, journal.prepare);
3837
3838         if (test_set_buffer_locked(bh)) {
3839                 if (!wait)
3840                         return 0;
3841                 lock_buffer(bh);
3842         }
3843         set_buffer_journal_prepared(bh);
3844         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3845                 clear_buffer_journal_test(bh);
3846                 set_buffer_journal_restore_dirty(bh);
3847         }
3848         unlock_buffer(bh);
3849         return 1;
3850 }
3851
3852 static void flush_old_journal_lists(struct super_block *s)
3853 {
3854         struct reiserfs_journal *journal = SB_JOURNAL(s);
3855         struct reiserfs_journal_list *jl;
3856         struct list_head *entry;
3857         time_t now = get_seconds();
3858
3859         while (!list_empty(&journal->j_journal_list)) {
3860                 entry = journal->j_journal_list.next;
3861                 jl = JOURNAL_LIST_ENTRY(entry);
3862                 /* this check should always be run, to send old lists to disk */
3863                 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) &&
3864                     atomic_read(&jl->j_commit_left) == 0 &&
3865                     test_transaction(s, jl)) {
3866                         flush_used_journal_lists(s, jl);
3867                 } else {
3868                         break;
3869                 }
3870         }
3871 }
3872
3873 /* 
3874 ** long and ugly.  If flush, will not return until all commit
3875 ** blocks and all real buffers in the trans are on disk.
3876 ** If no_async, won't return until all commit blocks are on disk.
3877 **
3878 ** keep reading, there are comments as you go along
3879 **
3880 ** If the journal is aborted, we just clean up. Things like flushing
3881 ** journal lists, etc just won't happen.
3882 */
3883 static int do_journal_end(struct reiserfs_transaction_handle *th,
3884                           struct super_block *p_s_sb, unsigned long nblocks,
3885                           int flags)
3886 {
3887         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3888         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3889         struct reiserfs_journal_cnode *last_cn = NULL;
3890         struct reiserfs_journal_desc *desc;
3891         struct reiserfs_journal_commit *commit;
3892         struct buffer_head *c_bh;       /* commit bh */
3893         struct buffer_head *d_bh;       /* desc bh */
3894         int cur_write_start = 0;        /* start index of current log write */
3895         int old_start;
3896         int i;
3897         int flush;
3898         int wait_on_commit;
3899         struct reiserfs_journal_list *jl, *temp_jl;
3900         struct list_head *entry, *safe;
3901         unsigned long jindex;
3902         unsigned long commit_trans_id;
3903         int trans_half;
3904
3905         BUG_ON(th->t_refcount > 1);
3906         BUG_ON(!th->t_trans_id);
3907
3908         /* protect flush_older_commits from doing mistakes if the
3909            transaction ID counter gets overflowed.  */
3910         if (th->t_trans_id == ~0UL)
3911                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
3912         flush = flags & FLUSH_ALL;
3913         wait_on_commit = flags & WAIT;
3914
3915         put_fs_excl();
3916         current->journal_info = th->t_handle_save;
3917         reiserfs_check_lock_depth(p_s_sb, "journal end");
3918         if (journal->j_len == 0) {
3919                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3920                                              1);
3921                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3922         }
3923
3924         lock_journal(p_s_sb);
3925         if (journal->j_next_full_flush) {
3926                 flags |= FLUSH_ALL;
3927                 flush = 1;
3928         }
3929         if (journal->j_next_async_flush) {
3930                 flags |= COMMIT_NOW | WAIT;
3931                 wait_on_commit = 1;
3932         }
3933
3934         /* check_journal_end locks the journal, and unlocks if it does not return 1 
3935          ** it tells us if we should continue with the journal_end, or just return
3936          */
3937         if (!check_journal_end(th, p_s_sb, nblocks, flags)) {
3938                 p_s_sb->s_dirt = 1;
3939                 wake_queued_writers(p_s_sb);
3940                 reiserfs_async_progress_wait(p_s_sb);
3941                 goto out;
3942         }
3943
3944         /* check_journal_end might set these, check again */
3945         if (journal->j_next_full_flush) {
3946                 flush = 1;
3947         }
3948
3949         /*
3950          ** j must wait means we have to flush the log blocks, and the real blocks for
3951          ** this transaction
3952          */
3953         if (journal->j_must_wait > 0) {
3954                 flush = 1;
3955         }
3956 #ifdef REISERFS_PREALLOCATE
3957         /* quota ops might need to nest, setup the journal_info pointer for them
3958          * and raise the refcount so that it is > 0. */
3959         current->journal_info = th;
3960         th->t_refcount++;
3961         reiserfs_discard_all_prealloc(th);      /* it should not involve new blocks into
3962                                                  * the transaction */
3963         th->t_refcount--;
3964         current->journal_info = th->t_handle_save;
3965 #endif
3966
3967         /* setup description block */
3968         d_bh =
3969             journal_getblk(p_s_sb,
3970                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3971                            journal->j_start);
3972         set_buffer_uptodate(d_bh);
3973         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
3974         memset(d_bh->b_data, 0, d_bh->b_size);
3975         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
3976         set_desc_trans_id(desc, journal->j_trans_id);
3977
3978         /* setup commit block.  Don't write (keep it clean too) this one until after everyone else is written */
3979         c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3980                               ((journal->j_start + journal->j_len +
3981                                 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
3982         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
3983         memset(c_bh->b_data, 0, c_bh->b_size);
3984         set_commit_trans_id(commit, journal->j_trans_id);
3985         set_buffer_uptodate(c_bh);
3986
3987         /* init this journal list */
3988         jl = journal->j_current_jl;
3989
3990         /* we lock the commit before doing anything because
3991          * we want to make sure nobody tries to run flush_commit_list until
3992          * the new transaction is fully setup, and we've already flushed the
3993          * ordered bh list
3994          */
3995         down(&jl->j_commit_lock);
3996
3997         /* save the transaction id in case we need to commit it later */
3998         commit_trans_id = jl->j_trans_id;
3999
4000         atomic_set(&jl->j_older_commits_done, 0);
4001         jl->j_trans_id = journal->j_trans_id;
4002         jl->j_timestamp = journal->j_trans_start_time;
4003         jl->j_commit_bh = c_bh;
4004         jl->j_start = journal->j_start;
4005         jl->j_len = journal->j_len;
4006         atomic_set(&jl->j_nonzerolen, journal->j_len);
4007         atomic_set(&jl->j_commit_left, journal->j_len + 2);
4008         jl->j_realblock = NULL;
4009
4010         /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4011          **  for each real block, add it to the journal list hash,
4012          ** copy into real block index array in the commit or desc block
4013          */
4014         trans_half = journal_trans_half(p_s_sb->s_blocksize);
4015         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4016                 if (buffer_journaled(cn->bh)) {
4017                         jl_cn = get_cnode(p_s_sb);
4018                         if (!jl_cn) {
4019                                 reiserfs_panic(p_s_sb,
4020                                                "journal-1676, get_cnode returned NULL\n");
4021                         }
4022                         if (i == 0) {
4023                                 jl->j_realblock = jl_cn;
4024                         }
4025                         jl_cn->prev = last_cn;
4026                         jl_cn->next = NULL;
4027                         if (last_cn) {
4028                                 last_cn->next = jl_cn;
4029                         }
4030                         last_cn = jl_cn;
4031                         /* make sure the block we are trying to log is not a block 
4032                            of journal or reserved area */
4033
4034                         if (is_block_in_log_or_reserved_area
4035                             (p_s_sb, cn->bh->b_blocknr)) {
4036                                 reiserfs_panic(p_s_sb,
4037                                                "journal-2332: Trying to log block %lu, which is a log block\n",
4038                                                cn->bh->b_blocknr);
4039                         }
4040                         jl_cn->blocknr = cn->bh->b_blocknr;
4041                         jl_cn->state = 0;
4042                         jl_cn->sb = p_s_sb;
4043                         jl_cn->bh = cn->bh;
4044                         jl_cn->jlist = jl;
4045                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4046                         if (i < trans_half) {
4047                                 desc->j_realblock[i] =
4048                                     cpu_to_le32(cn->bh->b_blocknr);
4049                         } else {
4050                                 commit->j_realblock[i - trans_half] =
4051                                     cpu_to_le32(cn->bh->b_blocknr);
4052                         }
4053                 } else {
4054                         i--;
4055                 }
4056         }
4057         set_desc_trans_len(desc, journal->j_len);
4058         set_desc_mount_id(desc, journal->j_mount_id);
4059         set_desc_trans_id(desc, journal->j_trans_id);
4060         set_commit_trans_len(commit, journal->j_len);
4061
4062         /* special check in case all buffers in the journal were marked for not logging */
4063         BUG_ON(journal->j_len == 0);
4064
4065         /* we're about to dirty all the log blocks, mark the description block
4066          * dirty now too.  Don't mark the commit block dirty until all the
4067          * others are on disk
4068          */
4069         mark_buffer_dirty(d_bh);
4070
4071         /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4072         cur_write_start = journal->j_start;
4073         cn = journal->j_first;
4074         jindex = 1;             /* start at one so we don't get the desc again */
4075         while (cn) {
4076                 clear_buffer_journal_new(cn->bh);
4077                 /* copy all the real blocks into log area.  dirty log blocks */
4078                 if (buffer_journaled(cn->bh)) {
4079                         struct buffer_head *tmp_bh;
4080                         char *addr;
4081                         struct page *page;
4082                         tmp_bh =
4083                             journal_getblk(p_s_sb,
4084                                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4085                                            ((cur_write_start +
4086                                              jindex) %
4087                                             SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
4088                         set_buffer_uptodate(tmp_bh);
4089                         page = cn->bh->b_page;
4090                         addr = kmap(page);
4091                         memcpy(tmp_bh->b_data,
4092                                addr + offset_in_page(cn->bh->b_data),
4093                                cn->bh->b_size);
4094                         kunmap(page);
4095                         mark_buffer_dirty(tmp_bh);
4096                         jindex++;
4097                         set_buffer_journal_dirty(cn->bh);
4098                         clear_buffer_journaled(cn->bh);
4099                 } else {
4100                         /* JDirty cleared sometime during transaction.  don't log this one */
4101                         reiserfs_warning(p_s_sb,
4102                                          "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!");
4103                         brelse(cn->bh);
4104                 }
4105                 next = cn->next;
4106                 free_cnode(p_s_sb, cn);
4107                 cn = next;
4108                 cond_resched();
4109         }
4110
4111         /* we are done  with both the c_bh and d_bh, but
4112          ** c_bh must be written after all other commit blocks,
4113          ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4114          */
4115
4116         journal->j_current_jl = alloc_journal_list(p_s_sb);
4117
4118         /* now it is safe to insert this transaction on the main list */
4119         list_add_tail(&jl->j_list, &journal->j_journal_list);
4120         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4121         journal->j_num_work_lists++;
4122
4123         /* reset journal values for the next transaction */
4124         old_start = journal->j_start;
4125         journal->j_start =
4126             (journal->j_start + journal->j_len +
4127              2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb);
4128         atomic_set(&(journal->j_wcount), 0);
4129         journal->j_bcount = 0;
4130         journal->j_last = NULL;
4131         journal->j_first = NULL;
4132         journal->j_len = 0;
4133         journal->j_trans_start_time = 0;
4134         /* check for trans_id overflow */
4135         if (++journal->j_trans_id == 0)
4136                 journal->j_trans_id = 10;
4137         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4138         journal->j_must_wait = 0;
4139         journal->j_len_alloc = 0;
4140         journal->j_next_full_flush = 0;
4141         journal->j_next_async_flush = 0;
4142         init_journal_hash(p_s_sb);
4143
4144         // make sure reiserfs_add_jh sees the new current_jl before we
4145         // write out the tails
4146         smp_mb();
4147
4148         /* tail conversion targets have to hit the disk before we end the
4149          * transaction.  Otherwise a later transaction might repack the tail
4150          * before this transaction commits, leaving the data block unflushed and
4151          * clean, if we crash before the later transaction commits, the data block
4152          * is lost.
4153          */
4154         if (!list_empty(&jl->j_tail_bh_list)) {
4155                 unlock_kernel();
4156                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4157                                       journal, jl, &jl->j_tail_bh_list);
4158                 lock_kernel();
4159         }
4160         BUG_ON(!list_empty(&jl->j_tail_bh_list));
4161         up(&jl->j_commit_lock);
4162
4163         /* honor the flush wishes from the caller, simple commits can
4164          ** be done outside the journal lock, they are done below
4165          **
4166          ** if we don't flush the commit list right now, we put it into
4167          ** the work queue so the people waiting on the async progress work
4168          ** queue don't wait for this proc to flush journal lists and such.
4169          */
4170         if (flush) {
4171                 flush_commit_list(p_s_sb, jl, 1);
4172                 flush_journal_list(p_s_sb, jl, 1);
4173         } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4174                 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4175
4176         /* if the next transaction has any chance of wrapping, flush 
4177          ** transactions that might get overwritten.  If any journal lists are very 
4178          ** old flush them as well.  
4179          */
4180       first_jl:
4181         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4182                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4183                 if (journal->j_start <= temp_jl->j_start) {
4184                         if ((journal->j_start + journal->j_trans_max + 1) >=
4185                             temp_jl->j_start) {
4186                                 flush_used_journal_lists(p_s_sb, temp_jl);
4187                                 goto first_jl;
4188                         } else if ((journal->j_start +
4189                                     journal->j_trans_max + 1) <
4190                                    SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4191                                 /* if we don't cross into the next transaction and we don't
4192                                  * wrap, there is no way we can overlap any later transactions
4193                                  * break now
4194                                  */
4195                                 break;
4196                         }
4197                 } else if ((journal->j_start +
4198                             journal->j_trans_max + 1) >
4199                            SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4200                         if (((journal->j_start + journal->j_trans_max + 1) %
4201                              SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >=
4202                             temp_jl->j_start) {
4203                                 flush_used_journal_lists(p_s_sb, temp_jl);
4204                                 goto first_jl;
4205                         } else {
4206                                 /* we don't overlap anything from out start to the end of the
4207                                  * log, and our wrapped portion doesn't overlap anything at
4208                                  * the start of the log.  We can break
4209                                  */
4210                                 break;
4211                         }
4212                 }
4213         }
4214         flush_old_journal_lists(p_s_sb);
4215
4216         journal->j_current_jl->j_list_bitmap =
4217             get_list_bitmap(p_s_sb, journal->j_current_jl);
4218
4219         if (!(journal->j_current_jl->j_list_bitmap)) {
4220                 reiserfs_panic(p_s_sb,
4221                                "journal-1996: do_journal_end, could not get a list bitmap\n");
4222         }
4223
4224         atomic_set(&(journal->j_jlock), 0);
4225         unlock_journal(p_s_sb);
4226         /* wake up any body waiting to join. */
4227         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4228         wake_up(&(journal->j_join_wait));
4229
4230         if (!flush && wait_on_commit &&
4231             journal_list_still_alive(p_s_sb, commit_trans_id)) {
4232                 flush_commit_list(p_s_sb, jl, 1);
4233         }
4234       out:
4235         reiserfs_check_lock_depth(p_s_sb, "journal end2");
4236
4237         memset(th, 0, sizeof(*th));
4238         /* Re-set th->t_super, so we can properly keep track of how many
4239          * persistent transactions there are. We need to do this so if this
4240          * call is part of a failed restart_transaction, we can free it later */
4241         th->t_super = p_s_sb;
4242
4243         return journal->j_errno;
4244 }
4245
4246 static void __reiserfs_journal_abort_hard(struct super_block *sb)
4247 {
4248         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4249         if (test_bit(J_ABORTED, &journal->j_state))
4250                 return;
4251
4252         printk(KERN_CRIT "REISERFS: Aborting journal for filesystem on %s\n",
4253                reiserfs_bdevname(sb));
4254
4255         sb->s_flags |= MS_RDONLY;
4256         set_bit(J_ABORTED, &journal->j_state);
4257
4258 #ifdef CONFIG_REISERFS_CHECK
4259         dump_stack();
4260 #endif
4261 }
4262
4263 static void __reiserfs_journal_abort_soft(struct super_block *sb, int errno)
4264 {
4265         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4266         if (test_bit(J_ABORTED, &journal->j_state))
4267                 return;
4268
4269         if (!journal->j_errno)
4270                 journal->j_errno = errno;
4271
4272         __reiserfs_journal_abort_hard(sb);
4273 }
4274
4275 void reiserfs_journal_abort(struct super_block *sb, int errno)
4276 {
4277         return __reiserfs_journal_abort_soft(sb, errno);
4278 }