File: /usr/src/linux/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     ** The commit thread -- a writer process for async commits.  It allows a 
37     **                      a process to request a log flush on a task queue.
38     **                      the commit will happen once the commit thread wakes up.
39     **                      The benefit here is the writer (with whatever
40     **                      related locks it has) doesn't have to wait for the
41     **                      log blocks to hit disk if it doesn't want to.
42     */
43     
44     #include <linux/config.h>
45     #include <asm/uaccess.h>
46     #include <asm/system.h>
47     
48     #include <linux/sched.h>
49     #include <asm/semaphore.h>
50     
51     #include <linux/vmalloc.h>
52     #include <linux/reiserfs_fs.h>
53     
54     #include <linux/kernel.h>
55     #include <linux/errno.h>
56     #include <linux/fcntl.h>
57     #include <linux/locks.h>
58     #include <linux/stat.h>
59     #include <linux/string.h>
60     #include <linux/smp_lock.h>
61     
62     /* the number of mounted filesystems.  This is used to decide when to
63     ** start and kill the commit thread
64     */
65     static int reiserfs_mounted_fs_count = 0 ;
66     
67     /* wake this up when you add something to the commit thread task queue */
68     DECLARE_WAIT_QUEUE_HEAD(reiserfs_commit_thread_wait) ;
69     
70     /* wait on this if you need to be sure you task queue entries have been run */
71     static DECLARE_WAIT_QUEUE_HEAD(reiserfs_commit_thread_done) ;
72     DECLARE_TASK_QUEUE(reiserfs_commit_thread_tq) ;
73     
74     #define JOURNAL_TRANS_HALF 1018   /* must be correct to keep the desc and commit structs at 4k */
75     
76     /* cnode stat bits.  Move these into reiserfs_fs.h */
77     
78     #define BLOCK_FREED 2		/* this block was freed, and can't be written.  */
79     #define BLOCK_FREED_HOLDER 3    /* this block was freed during this transaction, and can't be written */
80     
81     #define BLOCK_NEEDS_FLUSH 4	/* used in flush_journal_list */
82     
83     /* flags for do_journal_end */
84     #define FLUSH_ALL   1		/* flush commit and real blocks */
85     #define COMMIT_NOW  2		/* end and commit this transaction */
86     #define WAIT        4		/* wait for the log blocks to hit the disk*/
87     
88     /* state bits for the journal */
89     #define WRITERS_BLOCKED 1      /* set when new writers not allowed */
90     
91     static int do_journal_end(struct reiserfs_transaction_handle *,struct super_block *,unsigned long nblocks,int flags) ;
92     static int flush_journal_list(struct super_block *s, struct reiserfs_journal_list *jl, int flushall) ;
93     static int flush_commit_list(struct super_block *s, struct reiserfs_journal_list *jl, int flushall)  ;
94     static int can_dirty(struct reiserfs_journal_cnode *cn) ;
95     
96     static void init_journal_hash(struct super_block *p_s_sb) {
97       memset(SB_JOURNAL(p_s_sb)->j_hash_table, 0, JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)) ;
98     }
99     
100     /*
101     ** clears BH_Dirty and sticks the buffer on the clean list.  Called because I can't allow refile_buffer to
102     ** make schedule happen after I've freed a block.  Look at remove_from_transaction and journal_mark_freed for
103     ** more details.
104     */
105     static int reiserfs_clean_and_file_buffer(struct buffer_head *bh) {
106       if (bh) {
107         clear_bit(BH_Dirty, &bh->b_state) ;
108         refile_buffer(bh) ;
109       }
110       return 0 ;
111     }
112     
113     static struct reiserfs_bitmap_node *
114     allocate_bitmap_node(struct super_block *p_s_sb) {
115       struct reiserfs_bitmap_node *bn ;
116       static int id = 0 ;
117     
118       bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS) ;
119       if (!bn) {
120         return NULL ;
121       }
122       bn->data = kmalloc(p_s_sb->s_blocksize, GFP_NOFS) ;
123       if (!bn->data) {
124         kfree(bn) ;
125         return NULL ;
126       }
127       bn->id = id++ ;
128       memset(bn->data, 0, p_s_sb->s_blocksize) ;
129       INIT_LIST_HEAD(&bn->list) ;
130       return bn ;
131     }
132     
133     static struct reiserfs_bitmap_node *
134     get_bitmap_node(struct super_block *p_s_sb) {
135       struct reiserfs_bitmap_node *bn = NULL;
136       struct list_head *entry = SB_JOURNAL(p_s_sb)->j_bitmap_nodes.next ;
137     
138       SB_JOURNAL(p_s_sb)->j_used_bitmap_nodes++ ;
139     repeat:
140     
141       if(entry != &SB_JOURNAL(p_s_sb)->j_bitmap_nodes) {
142         bn = list_entry(entry, struct reiserfs_bitmap_node, list) ;
143         list_del(entry) ;
144         memset(bn->data, 0, p_s_sb->s_blocksize) ;
145         SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes-- ;
146         return bn ;
147       }
148       bn = allocate_bitmap_node(p_s_sb) ;
149       if (!bn) {
150         current->policy = SCHED_YIELD ;
151         schedule() ;
152         goto repeat ;
153       }
154       return bn ;
155     }
156     static inline void free_bitmap_node(struct super_block *p_s_sb,
157                                         struct reiserfs_bitmap_node *bn) {
158       SB_JOURNAL(p_s_sb)->j_used_bitmap_nodes-- ;
159       if (SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
160         kfree(bn->data) ;
161         kfree(bn) ;
162       } else {
163         list_add(&bn->list, &SB_JOURNAL(p_s_sb)->j_bitmap_nodes) ;
164         SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes++ ;
165       }
166     }
167     
168     static void allocate_bitmap_nodes(struct super_block *p_s_sb) {
169       int i ;
170       struct reiserfs_bitmap_node *bn = NULL ;
171       for (i = 0 ; i < REISERFS_MIN_BITMAP_NODES ; i++) {
172         bn = allocate_bitmap_node(p_s_sb) ;
173         if (bn) {
174           list_add(&bn->list, &SB_JOURNAL(p_s_sb)->j_bitmap_nodes) ;
175           SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes++ ;
176         } else {
177           break ; // this is ok, we'll try again when more are needed 
178         }
179       }
180     }
181     
182     static int set_bit_in_list_bitmap(struct super_block *p_s_sb, int block,
183                                       struct reiserfs_list_bitmap *jb) {
184       int bmap_nr = block / (p_s_sb->s_blocksize << 3) ;
185       int bit_nr = block % (p_s_sb->s_blocksize << 3) ;
186     
187       if (!jb->bitmaps[bmap_nr]) {
188         jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb) ;
189       }
190       set_bit(bit_nr, jb->bitmaps[bmap_nr]->data) ;
191       return 0 ;
192     }
193     
194     static void cleanup_bitmap_list(struct super_block *p_s_sb,
195                                     struct reiserfs_list_bitmap *jb) {
196       int i;
197       for (i = 0 ; i < SB_BMAP_NR(p_s_sb) ; i++) {
198         if (jb->bitmaps[i]) {
199           free_bitmap_node(p_s_sb, jb->bitmaps[i]) ;
200           jb->bitmaps[i] = NULL ;
201         }
202       }
203     }
204     
205     /*
206     ** only call this on FS unmount.
207     */
208     static int free_list_bitmaps(struct super_block *p_s_sb,
209                                  struct reiserfs_list_bitmap *jb_array) {
210       int i ;
211       struct reiserfs_list_bitmap *jb ;
212       for (i = 0 ; i < JOURNAL_NUM_BITMAPS ; i++) {
213         jb = jb_array + i ;
214         jb->journal_list = NULL ;
215         cleanup_bitmap_list(p_s_sb, jb) ;
216         vfree(jb->bitmaps) ;
217         jb->bitmaps = NULL ;
218       }
219       return 0;
220     }
221     
222     static int free_bitmap_nodes(struct super_block *p_s_sb) {
223       struct list_head *next = SB_JOURNAL(p_s_sb)->j_bitmap_nodes.next ;
224       struct reiserfs_bitmap_node *bn ;
225     
226       while(next != &SB_JOURNAL(p_s_sb)->j_bitmap_nodes) {
227         bn = list_entry(next, struct reiserfs_bitmap_node, list) ;
228         list_del(next) ;
229         kfree(bn->data) ;
230         kfree(bn) ;
231         next = SB_JOURNAL(p_s_sb)->j_bitmap_nodes.next ;
232         SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes-- ;
233       }
234     
235       return 0 ;
236     }
237     
238     /*
239     ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps. 
240     ** jb_array is the array to be filled in.
241     */
242     int reiserfs_allocate_list_bitmaps(struct super_block *p_s_sb,
243                                        struct reiserfs_list_bitmap *jb_array,
244     				   int bmap_nr) {
245       int i ;
246       int failed = 0 ;
247       struct reiserfs_list_bitmap *jb ;
248       int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *) ;
249     
250       for (i = 0 ; i < JOURNAL_NUM_BITMAPS ; i++) {
251         jb = jb_array + i ;
252         jb->journal_list = NULL ;
253         jb->bitmaps = vmalloc( mem ) ;
254         if (!jb->bitmaps) {
255           reiserfs_warning("clm-2000, unable to allocate bitmaps for journal lists\n") ;
256           failed = 1;   
257           break ;
258         }
259         memset(jb->bitmaps, 0, mem) ;
260       }
261       if (failed) {
262         free_list_bitmaps(p_s_sb, jb_array) ;
263         return -1 ;
264       }
265       return 0 ;
266     }
267     
268     /*
269     ** find an available list bitmap.  If you can't find one, flush a commit list 
270     ** and try again
271     */
272     static struct reiserfs_list_bitmap *
273     get_list_bitmap(struct super_block *p_s_sb, struct reiserfs_journal_list *jl) {
274       int i,j ; 
275       struct reiserfs_list_bitmap *jb = NULL ;
276     
277       for (j = 0 ; j < (JOURNAL_NUM_BITMAPS * 3) ; j++) {
278         i = SB_JOURNAL(p_s_sb)->j_list_bitmap_index ;
279         SB_JOURNAL(p_s_sb)->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS ;
280         jb = SB_JOURNAL(p_s_sb)->j_list_bitmap + i ;
281         if (SB_JOURNAL(p_s_sb)->j_list_bitmap[i].journal_list) {
282           flush_commit_list(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_bitmap[i].journal_list, 1) ;
283           if (!SB_JOURNAL(p_s_sb)->j_list_bitmap[i].journal_list) {
284     	break ;
285           }
286         } else {
287           break ;
288         }
289       }
290       if (jb->journal_list) { /* double check to make sure if flushed correctly */
291         return NULL ;
292       }
293       jb->journal_list = jl ;
294       return jb ;
295     }
296     
297     /* 
298     ** allocates a new chunk of X nodes, and links them all together as a list.
299     ** Uses the cnode->next and cnode->prev pointers
300     ** returns NULL on failure
301     */
302     static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes) {
303       struct reiserfs_journal_cnode *head ;
304       int i ;
305       if (num_cnodes <= 0) {
306         return NULL ;
307       }
308       head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode)) ;
309       if (!head) {
310         return NULL ;
311       }
312       memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode)) ;
313       head[0].prev = NULL ;
314       head[0].next = head + 1 ;
315       for (i = 1 ; i < num_cnodes; i++) {
316         head[i].prev = head + (i - 1) ;
317         head[i].next = head + (i + 1) ; /* if last one, overwrite it after the if */
318       }
319       head[num_cnodes -1].next = NULL ;
320       return head ;
321     }
322     
323     /*
324     ** pulls a cnode off the free list, or returns NULL on failure 
325     */
326     static struct reiserfs_journal_cnode *get_cnode(struct super_block *p_s_sb) {
327       struct reiserfs_journal_cnode *cn ;
328     
329       reiserfs_check_lock_depth("get_cnode") ;
330     
331       if (SB_JOURNAL(p_s_sb)->j_cnode_free <= 0) {
332         return NULL ;
333       }
334       SB_JOURNAL(p_s_sb)->j_cnode_used++ ;
335       SB_JOURNAL(p_s_sb)->j_cnode_free-- ;
336       cn = SB_JOURNAL(p_s_sb)->j_cnode_free_list ;
337       if (!cn) {
338         return cn ;
339       }
340       if (cn->next) {
341         cn->next->prev = NULL ;
342       }
343       SB_JOURNAL(p_s_sb)->j_cnode_free_list = cn->next ;
344       memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ;
345       return cn ;
346     }
347     
348     /*
349     ** returns a cnode to the free list 
350     */
351     static void free_cnode(struct super_block *p_s_sb, struct reiserfs_journal_cnode *cn) {
352     
353       reiserfs_check_lock_depth("free_cnode") ;
354     
355       SB_JOURNAL(p_s_sb)->j_cnode_used-- ;
356       SB_JOURNAL(p_s_sb)->j_cnode_free++ ;
357       /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
358       cn->next = SB_JOURNAL(p_s_sb)->j_cnode_free_list ;
359       if (SB_JOURNAL(p_s_sb)->j_cnode_free_list) {
360         SB_JOURNAL(p_s_sb)->j_cnode_free_list->prev = cn ;
361       }
362       cn->prev = NULL ; /* not needed with the memset, but I might kill the memset, and forget to do this */
363       SB_JOURNAL(p_s_sb)->j_cnode_free_list = cn ;
364     }
365     
366     static int clear_prepared_bits(struct buffer_head *bh) {
367       clear_bit(BH_JPrepared, &bh->b_state) ;
368       return 0 ;
369     }
370     
371     /* buffer is in current transaction */
372     inline int buffer_journaled(struct buffer_head *bh) {
373       if (bh)
374         return test_bit(BH_JDirty, &bh->b_state) ;
375       else
376         return 0 ;
377     }
378     
379     /* disk block was taken off free list before being in a finished transation, or written to disk
380     ** journal_new blocks can be reused immediately, for any purpose
381     */ 
382     inline int buffer_journal_new(struct buffer_head *bh) {
383       if (bh) 
384         return test_bit(BH_JNew, &bh->b_state) ;
385       else
386         return 0 ;
387     }
388     
389     inline int mark_buffer_journal_new(struct buffer_head *bh) {
390       if (bh) {
391         set_bit(BH_JNew, &bh->b_state) ;
392       }
393       return 0 ;
394     }
395     
396     inline int mark_buffer_not_journaled(struct buffer_head *bh) {
397       if (bh) 
398         clear_bit(BH_JDirty, &bh->b_state) ;
399       return 0 ;
400     }
401     
402     /* utility function to force a BUG if it is called without the big
403     ** kernel lock held.  caller is the string printed just before calling BUG()
404     */
405     void reiserfs_check_lock_depth(char *caller) {
406     #ifdef CONFIG_SMP
407       if (current->lock_depth < 0) {
408         printk("%s called without kernel lock held\n", caller) ;
409         show_reiserfs_locks() ;
410         BUG() ;
411       }
412     #else
413       ;
414     #endif
415     }
416     
417     /* return a cnode with same dev, block number and size in table, or null if not found */
418     static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct reiserfs_journal_cnode **table,
419                                        				  kdev_t dev,long bl,int size) {
420       struct reiserfs_journal_cnode *cn ;
421       cn = journal_hash(table, dev, bl) ;
422       while(cn) {
423         if ((cn->blocknr == bl) && (cn->dev == dev))
424           return cn ;
425         cn = cn->hnext ;
426       }
427       return (struct reiserfs_journal_cnode *)0 ;
428     }
429     
430     /* returns a cnode with same size, block number and dev as bh in the current transaction hash.  NULL if not found */
431     static inline struct reiserfs_journal_cnode *get_journal_hash(struct super_block *p_s_sb, struct buffer_head *bh) {
432       struct reiserfs_journal_cnode *cn ;
433       if (bh) {
434         cn =  get_journal_hash_dev(SB_JOURNAL(p_s_sb)->j_hash_table, bh->b_dev, bh->b_blocknr, bh->b_size) ;
435       }
436       else {
437         return (struct reiserfs_journal_cnode *)0 ;
438       }
439       return cn ;
440     }
441     
442     /* once upon a time, the journal would deadlock.  a lot.  Now, when
443     ** CONFIG_REISERFS_CHECK is defined, anytime someone enters a
444     ** transaction, it pushes itself into this ugly static list, and pops
445     ** itself off before calling journal_end.  I made a SysRq key to dump
446     ** the list, and tell me what the writers are when I'm deadlocked.  */
447     
448     				/* are you depending on the compiler
449                                        to optimize this function away
450                                        everywhere it is called? It is not
451                                        obvious how this works, but I
452                                        suppose debugging code need not be
453                                        clear.  -Hans */
454     static char *journal_writers[512] ;
455     int push_journal_writer(char *s) {
456     #ifdef CONFIG_REISERFS_CHECK
457       int i ;
458       for (i = 0 ; i < 512 ; i++) {
459         if (!journal_writers[i]) {
460           journal_writers[i] = s ;
461           return i ;
462         }
463       }
464       return -1 ;
465     #else
466       return 0 ;
467     #endif
468     }
469     int pop_journal_writer(int index) {
470     #ifdef CONFIG_REISERFS_CHECK
471       if (index >= 0) {
472         journal_writers[index] = NULL ;
473       }
474     #endif
475       return 0 ;
476     }
477     
478     int dump_journal_writers(void) {
479       int i ;
480       for (i = 0 ; i < 512 ; i++) {
481         if (journal_writers[i]) {
482           printk("%d: %s\n", i, journal_writers[i]) ;
483         }
484       }
485       return 0 ;
486     }
487     
488     /*
489     ** this actually means 'can this block be reallocated yet?'.  If you set search_all, a block can only be allocated
490     ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
491     ** being overwritten by a replay after crashing.
492     **
493     ** If you don't set search_all, a block can only be allocated if it is not in the current transaction.  Since deleting
494     ** a block removes it from the current transaction, this case should never happen.  If you don't set search_all, make
495     ** sure you never write the block without logging it.
496     **
497     ** next_zero_bit is a suggestion about the next block to try for find_forward.
498     ** when bl is rejected because it is set in a journal list bitmap, we search
499     ** for the next zero bit in the bitmap that rejected bl.  Then, we return that
500     ** through next_zero_bit for find_forward to try.
501     **
502     ** Just because we return something in next_zero_bit does not mean we won't
503     ** reject it on the next call to reiserfs_in_journal
504     **
505     */
506     int reiserfs_in_journal(struct super_block *p_s_sb, kdev_t dev, 
507                             unsigned long bl, int size, int search_all, 
508     			unsigned long *next_zero_bit) {
509       struct reiserfs_journal_cnode *cn ;
510       struct reiserfs_list_bitmap *jb ;
511       int i ;
512       int bmap_nr = bl / (p_s_sb->s_blocksize << 3) ;
513       int bit_nr = bl % (p_s_sb->s_blocksize << 3) ;
514       int tmp_bit ;
515     
516       *next_zero_bit = 0 ; /* always start this at zero. */
517     
518       /* we aren't logging all blocks are safe for reuse */
519       if (reiserfs_dont_log(p_s_sb)) {
520         return 0 ;
521       }
522     
523       /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
524       ** if we crash before the transaction that freed it commits,  this transaction won't
525       ** have committed either, and the block will never be written
526       */
527       if (search_all) {
528         for (i = 0 ; i < JOURNAL_NUM_BITMAPS ; i++) {
529           jb = SB_JOURNAL(p_s_sb)->j_list_bitmap + i ;
530           if (jb->journal_list && jb->bitmaps[bmap_nr] &&
531               test_bit(bit_nr, jb->bitmaps[bmap_nr]->data)) {
532     	tmp_bit = find_next_zero_bit((unsigned long *)
533     	                             (jb->bitmaps[bmap_nr]->data),
534     	                             p_s_sb->s_blocksize << 3, bit_nr+1) ; 
535     	*next_zero_bit = bmap_nr * (p_s_sb->s_blocksize << 3) + tmp_bit ;
536     	return 1 ;
537           }
538         }
539       }
540     
541       /* is it in any old transactions? */
542       if (search_all && (cn = get_journal_hash_dev(SB_JOURNAL(p_s_sb)->j_list_hash_table, dev,bl,size))) {
543         return 1; 
544       }
545     
546       /* is it in the current transaction.  This should never happen */
547       if ((cn = get_journal_hash_dev(SB_JOURNAL(p_s_sb)->j_hash_table, dev,bl,size))) {
548         return 1; 
549       }
550     
551       /* safe for reuse */
552       return 0 ;
553     }
554     
555     /* insert cn into table
556     */
557     inline void insert_journal_hash(struct reiserfs_journal_cnode **table, struct reiserfs_journal_cnode *cn) {
558       struct reiserfs_journal_cnode *cn_orig ;
559     
560       cn_orig = journal_hash(table, cn->dev, cn->blocknr) ;
561       cn->hnext = cn_orig ;
562       cn->hprev = NULL ;
563       if (cn_orig) {
564         cn_orig->hprev = cn ;
565       }
566       journal_hash(table, cn->dev, cn->blocknr) =  cn ;
567     }
568     
569     /* lock the current transaction */
570     inline static void lock_journal(struct super_block *p_s_sb) {
571       while(atomic_read(&(SB_JOURNAL(p_s_sb)->j_wlock)) > 0) {
572         sleep_on(&(SB_JOURNAL(p_s_sb)->j_wait)) ;
573       }
574       atomic_set(&(SB_JOURNAL(p_s_sb)->j_wlock), 1) ;
575     }
576     
577     /* unlock the current transaction */
578     inline static void unlock_journal(struct super_block *p_s_sb) {
579       atomic_dec(&(SB_JOURNAL(p_s_sb)->j_wlock)) ;
580       wake_up(&(SB_JOURNAL(p_s_sb)->j_wait)) ;
581     }
582     
583     /*
584     ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
585     ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
586     ** transaction.
587     */
588     static void cleanup_freed_for_journal_list(struct super_block *p_s_sb, struct reiserfs_journal_list *jl) {
589     
590       struct reiserfs_list_bitmap *jb = jl->j_list_bitmap ;
591       if (jb) {
592         cleanup_bitmap_list(p_s_sb, jb) ;
593       }
594       jl->j_list_bitmap->journal_list = NULL ;
595       jl->j_list_bitmap = NULL ;
596     }
597     
598     /*
599     ** if this journal list still has commit blocks unflushed, send them to disk.
600     **
601     ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
602     ** Before the commit block can by written, every other log block must be safely on disk
603     **
604     */
605     static int flush_commit_list(struct super_block *s, struct reiserfs_journal_list *jl, int flushall) {
606       int i, count ;
607       int index = 0 ;
608       int bn ;
609       int retry_count = 0 ;
610       int orig_commit_left = 0 ;
611       struct buffer_head *tbh = NULL ;
612       struct reiserfs_journal_list *other_jl ;
613     
614       reiserfs_check_lock_depth("flush_commit_list") ;
615     
616       if (atomic_read(&jl->j_older_commits_done)) {
617         return 0 ;
618       }
619     
620       /* before we can put our commit blocks on disk, we have to make sure everyone older than
621       ** us is on disk too
622       */
623       if (jl->j_len <= 0) {
624         return 0 ;
625       }
626       if (flushall) {
627         /* we _must_ make sure the transactions are committed in order.  Start with the
628         ** index after this one, wrap all the way around 
629         */
630         index = (jl - SB_JOURNAL_LIST(s)) + 1 ;
631         for (i = 0 ; i < JOURNAL_LIST_COUNT ; i++) {
632           other_jl = SB_JOURNAL_LIST(s) + ( (index + i) % JOURNAL_LIST_COUNT) ;
633           if (other_jl && other_jl != jl && other_jl->j_len > 0 && other_jl->j_trans_id > 0 && 
634               other_jl->j_trans_id <= jl->j_trans_id && (atomic_read(&(jl->j_older_commits_done)) == 0)) {
635             flush_commit_list(s, other_jl, 0) ;
636           }
637         }
638       }
639     
640       count = 0 ;
641       /* don't flush the commit list for the current transactoin */
642       if (jl == ((SB_JOURNAL_LIST(s) + SB_JOURNAL_LIST_INDEX(s)))) {
643         return 0 ;
644       }
645     
646       /* make sure nobody is trying to flush this one at the same time */
647       if (atomic_read(&(jl->j_commit_flushing))) {
648         sleep_on(&(jl->j_commit_wait)) ;
649         if (flushall) {
650           atomic_set(&(jl->j_older_commits_done), 1) ;
651         }
652         return 0 ;
653       }
654       
655       /* this commit is done, exit */
656       if (atomic_read(&(jl->j_commit_left)) <= 0) {
657         if (flushall) {
658           atomic_set(&(jl->j_older_commits_done), 1) ;
659         }
660         return 0 ;
661       }
662       /* keeps others from flushing while we are flushing */
663       atomic_set(&(jl->j_commit_flushing), 1) ; 
664     
665     
666       if (jl->j_len > JOURNAL_TRANS_MAX) {
667         reiserfs_panic(s, "journal-512: flush_commit_list: length is %lu, list number %d\n", jl->j_len, jl - SB_JOURNAL_LIST(s)) ;
668         return 0 ;
669       }
670     
671       orig_commit_left = atomic_read(&(jl->j_commit_left)) ; 
672     
673       /* start by checking all the commit blocks in this transaction.  
674       ** Add anyone not on disk into tbh.  Stop checking once commit_left <= 1, because that means we
675       ** only have the commit block left 
676       */
677     retry:
678       count = 0 ;
679       for (i = 0 ; atomic_read(&(jl->j_commit_left)) > 1 && i < (jl->j_len + 1) ; i++) {  /* everything but commit_bh */
680         bn = reiserfs_get_journal_block(s) + (jl->j_start+i) % JOURNAL_BLOCK_COUNT;
681         tbh = get_hash_table(s->s_dev, bn, s->s_blocksize) ;
682     
683     /* kill this sanity check */
684     if (count > (orig_commit_left + 2)) {
685     reiserfs_panic(s, "journal-539: flush_commit_list: BAD count(%d) > orig_commit_left(%d)!\n", count, orig_commit_left) ;
686     }
687         if (tbh) {
688           if (buffer_locked(tbh)) { /* wait on it, redo it just to make sure */
689     	wait_on_buffer(tbh) ;
690     	if (!buffer_uptodate(tbh)) {
691     	  reiserfs_panic(s, "journal-584, buffer write failed\n") ;
692     	}
693           } 
694           if (buffer_dirty(tbh)) {
695     	printk("journal-569: flush_commit_list, block already dirty!\n") ;
696           } else {				
697     	mark_buffer_dirty(tbh) ;
698           }
699           ll_rw_block(WRITE, 1, &tbh) ;
700           count++ ;
701           put_bh(tbh) ; /* once for our get_hash */
702         } 
703       }
704     
705       /* wait on everyone in tbh before writing commit block*/
706       if (count > 0) {
707         for (i = 0 ; atomic_read(&(jl->j_commit_left)) > 1 && 
708                      i < (jl->j_len + 1) ; i++) {  /* everything but commit_bh */
709           bn = reiserfs_get_journal_block(s) + (jl->j_start + i) % JOURNAL_BLOCK_COUNT  ;
710           tbh = get_hash_table(s->s_dev, bn, s->s_blocksize) ;
711     
712           wait_on_buffer(tbh) ;
713           if (!buffer_uptodate(tbh)) {
714     	reiserfs_panic(s, "journal-601, buffer write failed\n") ;
715           }
716           put_bh(tbh) ; /* once for our get_hash */
717           bforget(tbh) ;    /* once due to original getblk in do_journal_end */
718           atomic_dec(&(jl->j_commit_left)) ;
719         }
720       }
721     
722       if (atomic_read(&(jl->j_commit_left)) != 1) { /* just the commit_bh left, flush it without calling getblk for everyone */
723         if (retry_count < 2) {
724           printk("journal-582: flush_commit_list, not all log blocks on disk yet, trying again\n") ;
725           retry_count++ ;
726           goto retry;
727         }
728         reiserfs_panic(s, "journal-563: flush_commit_list: BAD, j_commit_left is %lu, should be 1\n", 
729                            atomic_read(&(jl->j_commit_left)));
730       }
731     
732       mark_buffer_dirty(jl->j_commit_bh) ;
733       ll_rw_block(WRITE, 1, &(jl->j_commit_bh)) ;
734       wait_on_buffer(jl->j_commit_bh) ;
735       if (!buffer_uptodate(jl->j_commit_bh)) {
736         reiserfs_panic(s, "journal-615: buffer write failed\n") ;
737       }
738       atomic_dec(&(jl->j_commit_left)) ;
739       bforget(jl->j_commit_bh) ;
740     
741       /* now, every commit block is on the disk.  It is safe to allow blocks freed during this transaction to be reallocated */
742       cleanup_freed_for_journal_list(s, jl) ;
743     
744       if (flushall) {
745         atomic_set(&(jl->j_older_commits_done), 1) ;
746       }
747       atomic_set(&(jl->j_commit_flushing), 0) ;
748       wake_up(&(jl->j_commit_wait)) ;
749       
750       s->s_dirt = 1 ;
751       return 0 ;
752     }
753     
754     /*
755     ** flush_journal_list frequently needs to find a newer transaction for a given block.  This does that, or 
756     ** returns NULL if it can't find anything 
757     */
758     static struct reiserfs_journal_list *find_newer_jl_for_cn(struct reiserfs_journal_cnode *cn) {
759       kdev_t dev = cn->dev;
760       unsigned long blocknr = cn->blocknr ;
761     
762       cn = cn->hprev ;
763       while(cn) {
764         if (cn->dev == dev && cn->blocknr == blocknr && cn->jlist) {
765           return cn->jlist ;
766         }
767         cn = cn->hprev ;
768       }
769       return NULL ;
770     }
771     
772     
773     /*
774     ** once all the real blocks have been flushed, it is safe to remove them from the
775     ** journal list for this transaction.  Aside from freeing the cnode, this also allows the
776     ** block to be reallocated for data blocks if it had been deleted.
777     */
778     static void remove_all_from_journal_list(struct super_block *p_s_sb, struct reiserfs_journal_list *jl, int debug) {
779       struct buffer_head fake_bh ;
780       struct reiserfs_journal_cnode *cn, *last ;
781       cn = jl->j_realblock ;
782     
783       /* which is better, to lock once around the whole loop, or
784       ** to lock for each call to remove_from_journal_list?
785       */
786       while(cn) {
787         if (cn->blocknr != 0) {
788           if (debug) {
789             printk("block %lu, bh is %d, state %d\n", cn->blocknr, cn->bh ? 1: 0, 
790     	        cn->state) ;
791           }
792           fake_bh.b_blocknr = cn->blocknr ;
793           fake_bh.b_dev = cn->dev ;
794           cn->state = 0 ;
795           remove_from_journal_list(p_s_sb, jl, &fake_bh, 1) ;
796         }
797         last = cn ;
798         cn = cn->next ;
799         free_cnode(p_s_sb, last) ;
800       }
801       jl->j_realblock = NULL ;
802     }
803     
804     /*
805     ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
806     ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
807     ** releasing blocks in this transaction for reuse as data blocks.
808     ** called by flush_journal_list, before it calls remove_all_from_journal_list
809     **
810     */
811     static int _update_journal_header_block(struct super_block *p_s_sb, unsigned long offset, unsigned long trans_id) {
812       struct reiserfs_journal_header *jh ;
813       if (trans_id >= SB_JOURNAL(p_s_sb)->j_last_flush_trans_id) {
814         if (buffer_locked((SB_JOURNAL(p_s_sb)->j_header_bh)))  {
815           wait_on_buffer((SB_JOURNAL(p_s_sb)->j_header_bh)) ;
816           if (!buffer_uptodate(SB_JOURNAL(p_s_sb)->j_header_bh)) {
817             reiserfs_panic(p_s_sb, "journal-699: buffer write failed\n") ;
818           }
819         }
820         SB_JOURNAL(p_s_sb)->j_last_flush_trans_id = trans_id ;
821         SB_JOURNAL(p_s_sb)->j_first_unflushed_offset = offset ;
822         jh = (struct reiserfs_journal_header *)(SB_JOURNAL(p_s_sb)->j_header_bh->b_data) ;
823         jh->j_last_flush_trans_id = cpu_to_le32(trans_id) ;
824         jh->j_first_unflushed_offset = cpu_to_le32(offset) ;
825         jh->j_mount_id = cpu_to_le32(SB_JOURNAL(p_s_sb)->j_mount_id) ;
826         set_bit(BH_Dirty, &(SB_JOURNAL(p_s_sb)->j_header_bh->b_state)) ;
827         ll_rw_block(WRITE, 1, &(SB_JOURNAL(p_s_sb)->j_header_bh)) ;
828         wait_on_buffer((SB_JOURNAL(p_s_sb)->j_header_bh)) ; 
829         if (!buffer_uptodate(SB_JOURNAL(p_s_sb)->j_header_bh)) {
830           printk( "reiserfs: journal-837: IO error during journal replay\n" );
831           return -EIO ;
832         }
833       }
834       return 0 ;
835     }
836     
837     static int update_journal_header_block(struct super_block *p_s_sb, 
838                                            unsigned long offset, 
839     				       unsigned long trans_id) {
840         if (_update_journal_header_block(p_s_sb, offset, trans_id)) {
841     	reiserfs_panic(p_s_sb, "journal-712: buffer write failed\n") ;
842         }
843         return 0 ;
844     }
845     /* 
846     ** flush any and all journal lists older than you are 
847     ** can only be called from flush_journal_list
848     */
849     static int flush_older_journal_lists(struct super_block *p_s_sb, struct reiserfs_journal_list *jl, unsigned long trans_id) {
850       int i, index ;
851       struct reiserfs_journal_list *other_jl ;
852     
853       index = jl - SB_JOURNAL_LIST(p_s_sb) ;
854       for (i = 0 ; i < JOURNAL_LIST_COUNT ; i++) {
855         other_jl = SB_JOURNAL_LIST(p_s_sb) + ((index + i) % JOURNAL_LIST_COUNT) ;
856         if (other_jl && other_jl->j_len > 0 && 
857             other_jl->j_trans_id > 0 && 
858     	other_jl->j_trans_id < trans_id && 
859             other_jl != jl) {
860           /* do not flush all */
861           flush_journal_list(p_s_sb, other_jl, 0) ; 
862         }
863       }
864       return 0 ;
865     }
866     
867     static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate) {
868         if (buffer_journaled(bh)) {
869             reiserfs_warning("clm-2084: pinned buffer %u:%s sent to disk\n",
870     	                 bh->b_blocknr, kdevname(bh->b_dev)) ;
871         }
872         mark_buffer_uptodate(bh, uptodate) ;
873         unlock_buffer(bh) ;
874         put_bh(bh) ;
875     }
876     static void submit_logged_buffer(struct buffer_head *bh) {
877         lock_buffer(bh) ;
878         get_bh(bh) ;
879         bh->b_end_io = reiserfs_end_buffer_io_sync ;
880         mark_buffer_notjournal_new(bh) ;
881         clear_bit(BH_Dirty, &bh->b_state) ;
882         submit_bh(WRITE, bh) ;
883     }
884     
885     /* flush a journal list, both commit and real blocks
886     **
887     ** always set flushall to 1, unless you are calling from inside
888     ** flush_journal_list
889     **
890     ** IMPORTANT.  This can only be called while there are no journal writers, 
891     ** and the journal is locked.  That means it can only be called from 
892     ** do_journal_end, or by journal_release
893     */
894     static int flush_journal_list(struct super_block *s, 
895                                   struct reiserfs_journal_list *jl, int flushall) {
896       struct reiserfs_journal_list *pjl ;
897       struct reiserfs_journal_cnode *cn, *last ;
898       int count ;
899       int was_jwait = 0 ;
900       int was_dirty = 0 ;
901       struct buffer_head *saved_bh ; 
902       unsigned long j_len_saved = jl->j_len ;
903     
904       if (j_len_saved <= 0) {
905         return 0 ;
906       }
907     
908       if (atomic_read(&SB_JOURNAL(s)->j_wcount) != 0) {
909         reiserfs_warning("clm-2048: flush_journal_list called with wcount %d\n",
910                           atomic_read(&SB_JOURNAL(s)->j_wcount)) ;
911       }
912       /* if someone is getting the commit list, we must wait for them */
913       while (atomic_read(&(jl->j_commit_flushing))) { 
914         sleep_on(&(jl->j_commit_wait)) ;
915       }
916       /* if someone is flushing this list, we must wait for them */
917       while (atomic_read(&(jl->j_flushing))) {
918         sleep_on(&(jl->j_flush_wait)) ;
919       }
920     
921       /* this list is now ours, we can change anything we want */
922       atomic_set(&(jl->j_flushing), 1) ;
923     
924       count = 0 ;
925       if (j_len_saved > JOURNAL_TRANS_MAX) {
926         reiserfs_panic(s, "journal-715: flush_journal_list, length is %lu, list number %d\n", j_len_saved, jl - SB_JOURNAL_LIST(s)) ;
927         atomic_dec(&(jl->j_flushing)) ;
928         return 0 ;
929       }
930     
931       /* if all the work is already done, get out of here */
932       if (atomic_read(&(jl->j_nonzerolen)) <= 0 && 
933           atomic_read(&(jl->j_commit_left)) <= 0) {
934         goto flush_older_and_return ;
935       } 
936     
937       /* start by putting the commit list on disk.  This will also flush 
938       ** the commit lists of any olders transactions
939       */
940       flush_commit_list(s, jl, 1) ;
941     
942       /* are we done now? */
943       if (atomic_read(&(jl->j_nonzerolen)) <= 0 && 
944           atomic_read(&(jl->j_commit_left)) <= 0) {
945         goto flush_older_and_return ;
946       }
947     
948       /* loop through each cnode, see if we need to write it, 
949       ** or wait on a more recent transaction, or just ignore it 
950       */
951       if (atomic_read(&(SB_JOURNAL(s)->j_wcount)) != 0) {
952         reiserfs_panic(s, "journal-844: panic journal list is flushing, wcount is not 0\n") ;
953       }
954       cn = jl->j_realblock ;
955       while(cn) {
956         was_jwait = 0 ;
957         was_dirty = 0 ;
958         saved_bh = NULL ;
959         /* blocknr of 0 is no longer in the hash, ignore it */
960         if (cn->blocknr == 0) {
961           goto free_cnode ;
962         }
963         pjl = find_newer_jl_for_cn(cn) ;
964         /* the order is important here.  We check pjl to make sure we
965         ** don't clear BH_JDirty_wait if we aren't the one writing this
966         ** block to disk
967         */
968         if (!pjl && cn->bh) {
969           saved_bh = cn->bh ;
970     
971           /* we do this to make sure nobody releases the buffer while 
972           ** we are working with it 
973           */
974           get_bh(saved_bh) ;
975     
976           if (buffer_journal_dirty(saved_bh)) {
977             was_jwait = 1 ;
978     	mark_buffer_notjournal_dirty(saved_bh) ;
979             /* undo the inc from journal_mark_dirty */
980     	put_bh(saved_bh) ;
981           }
982           if (can_dirty(cn)) {
983             was_dirty = 1 ;
984           }
985         }
986     
987         /* if someone has this block in a newer transaction, just make
988         ** sure they are commited, and don't try writing it to disk
989         */
990         if (pjl) {
991           flush_commit_list(s, pjl, 1) ;
992           goto free_cnode ;
993         }
994     
995         /* bh == NULL when the block got to disk on its own, OR, 
996         ** the block got freed in a future transaction 
997         */
998         if (saved_bh == NULL) {
999           goto free_cnode ;
1000         }
1001     
1002         /* this should never happen.  kupdate_one_transaction has this list
1003         ** locked while it works, so we should never see a buffer here that
1004         ** is not marked JDirty_wait
1005         */
1006         if ((!was_jwait) && !buffer_locked(saved_bh)) {
1007     printk("journal-813: BAD! buffer %lu %cdirty %cjwait, not in a newer tranasction\n", saved_bh->b_blocknr,
1008             was_dirty ? ' ' : '!', was_jwait ? ' ' : '!') ;
1009         }
1010         /* kupdate_one_transaction waits on the buffers it is writing, so we
1011         ** should never see locked buffers here
1012         */
1013         if (buffer_locked(saved_bh)) {
1014           printk("clm-2083: locked buffer %lu in flush_journal_list\n", 
1015                   saved_bh->b_blocknr) ;
1016           wait_on_buffer(saved_bh) ;
1017           if (!buffer_uptodate(saved_bh)) {
1018             reiserfs_panic(s, "journal-923: buffer write failed\n") ;
1019           }
1020         } 
1021         if (was_dirty) { 
1022           /* we inc again because saved_bh gets decremented at free_cnode */
1023           get_bh(saved_bh) ;
1024           set_bit(BLOCK_NEEDS_FLUSH, &cn->state) ;
1025           submit_logged_buffer(saved_bh) ;
1026           count++ ;
1027         } else {
1028           printk("clm-2082: Unable to flush buffer %lu in flush_journal_list\n",
1029                   saved_bh->b_blocknr) ;
1030         }
1031     free_cnode:
1032         last = cn ;
1033         cn = cn->next ;
1034         if (saved_bh) {
1035           /* we incremented this to keep others from taking the buffer head away */
1036           put_bh(saved_bh) ;
1037           if (atomic_read(&(saved_bh->b_count)) < 0) {
1038             printk("journal-945: saved_bh->b_count < 0") ;
1039           }
1040         }
1041       }
1042       if (count > 0) {
1043         cn = jl->j_realblock ;
1044         while(cn) {
1045           if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1046     	if (!cn->bh) {
1047     	  reiserfs_panic(s, "journal-1011: cn->bh is NULL\n") ;
1048     	}
1049     	wait_on_buffer(cn->bh) ;
1050     	if (!cn->bh) {
1051     	  reiserfs_panic(s, "journal-1012: cn->bh is NULL\n") ;
1052     	}
1053     	if (!buffer_uptodate(cn->bh)) {
1054     	  reiserfs_panic(s, "journal-949: buffer write failed\n") ;
1055     	}
1056     	refile_buffer(cn->bh) ;
1057             brelse(cn->bh) ;
1058           }
1059           cn = cn->next ;
1060         }
1061       }
1062     
1063     flush_older_and_return:
1064       /* before we can update the journal header block, we _must_ flush all 
1065       ** real blocks from all older transactions to disk.  This is because
1066       ** once the header block is updated, this transaction will not be
1067       ** replayed after a crash
1068       */
1069       if (flushall) {
1070         flush_older_journal_lists(s, jl, jl->j_trans_id) ;
1071       } 
1072       
1073       /* before we can remove everything from the hash tables for this 
1074       ** transaction, we must make sure it can never be replayed
1075       **
1076       ** since we are only called from do_journal_end, we know for sure there
1077       ** are no allocations going on while we are flushing journal lists.  So,
1078       ** we only need to update the journal header block for the last list
1079       ** being flushed
1080       */
1081       if (flushall) {
1082         update_journal_header_block(s, (jl->j_start + jl->j_len + 2) % JOURNAL_BLOCK_COUNT, jl->j_trans_id) ;
1083       }
1084       remove_all_from_journal_list(s, jl, 0) ;
1085       jl->j_len = 0 ;
1086       atomic_set(&(jl->j_nonzerolen), 0) ;
1087       jl->j_start = 0 ;
1088       jl->j_realblock = NULL ;
1089       jl->j_commit_bh = NULL ;
1090       jl->j_trans_id = 0 ;
1091       atomic_dec(&(jl->j_flushing)) ;
1092       wake_up(&(jl->j_flush_wait)) ;
1093       return 0 ;
1094     } 
1095     
1096     
1097     static int kupdate_one_transaction(struct super_block *s,
1098                                         struct reiserfs_journal_list *jl) 
1099     {
1100         struct reiserfs_journal_list *pjl ; /* previous list for this cn */
1101         struct reiserfs_journal_cnode *cn, *walk_cn ;
1102         unsigned long blocknr ;
1103         int run = 0 ;
1104         int orig_trans_id = jl->j_trans_id ;
1105         struct buffer_head *saved_bh ; 
1106         int ret = 0 ;
1107     
1108         /* if someone is getting the commit list, we must wait for them */
1109         while (atomic_read(&(jl->j_commit_flushing))) {
1110             sleep_on(&(jl->j_commit_wait)) ;
1111         }
1112         /* if someone is flushing this list, we must wait for them */
1113         while (atomic_read(&(jl->j_flushing))) {
1114             sleep_on(&(jl->j_flush_wait)) ;
1115         }
1116         /* was it flushed while we slept? */
1117         if (jl->j_len <= 0 || jl->j_trans_id != orig_trans_id) {
1118             return 0 ;
1119         }
1120     
1121         /* this list is now ours, we can change anything we want */
1122         atomic_set(&(jl->j_flushing), 1) ;
1123     
1124     loop_start:
1125         cn = jl->j_realblock ;
1126         while(cn) {
1127             saved_bh = NULL ;
1128             /* if the blocknr == 0, this has been cleared from the hash,
1129             ** skip it
1130             */
1131             if (cn->blocknr == 0) {
1132                 goto next ;
1133             }
1134             /* look for a more recent transaction that logged this
1135             ** buffer.  Only the most recent transaction with a buffer in
1136             ** it is allowed to send that buffer to disk
1137             */
1138             pjl = find_newer_jl_for_cn(cn) ;
1139             if (run == 0 && !pjl && cn->bh && buffer_journal_dirty(cn->bh) &&
1140                 can_dirty(cn)) 
1141             {
1142                 if (!test_bit(BH_JPrepared, &cn->bh->b_state)) {
1143                     set_bit(BLOCK_NEEDS_FLUSH, &cn->state) ;
1144     		submit_logged_buffer(cn->bh) ;
1145                 } else {
1146                     /* someone else is using this buffer.  We can't 
1147                     ** send it to disk right now because they might
1148                     ** be changing/logging it.
1149                     */
1150                     ret = 1 ;
1151                 }
1152             } else if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1153                 clear_bit(BLOCK_NEEDS_FLUSH, &cn->state) ;
1154                 if (!pjl && cn->bh) {
1155                     wait_on_buffer(cn->bh) ;
1156                 }
1157                 /* check again, someone could have logged while we scheduled */
1158                 pjl = find_newer_jl_for_cn(cn) ;
1159     
1160                 /* before the JDirty_wait bit is set, the 
1161                 ** buffer is added to the hash list.  So, if we are
1162                 ** run in the middle of a do_journal_end, we will notice
1163                 ** if this buffer was logged and added from the latest
1164                 ** transaction.  In this case, we don't want to decrement
1165                 ** b_count
1166                 */
1167                 if (!pjl && cn->bh && buffer_journal_dirty(cn->bh)) {
1168                     blocknr = cn->blocknr ;
1169                     walk_cn = cn ;
1170                     saved_bh= cn->bh ;
1171                     /* update all older transactions to show this block
1172                     ** was flushed
1173                     */
1174                     mark_buffer_notjournal_dirty(cn->bh) ;
1175                     while(walk_cn) {
1176                         if (walk_cn->bh && walk_cn->blocknr == blocknr && 
1177                              walk_cn->dev == cn->dev) {
1178                             if (walk_cn->jlist) {
1179                                 atomic_dec(&(walk_cn->jlist->j_nonzerolen)) ;
1180                             }
1181                             walk_cn->bh = NULL ;
1182                         }
1183                         walk_cn = walk_cn->hnext ;
1184                     }
1185                     if (atomic_read(&saved_bh->b_count) < 1) {
1186                         reiserfs_warning("clm-2081: bad count on %lu\n", 
1187                                           saved_bh->b_blocknr) ;
1188                     }
1189                     brelse(saved_bh) ;
1190                 }
1191             }
1192             /*
1193             ** if the more recent transaction is committed to the log,
1194             ** this buffer can be considered flushed.  Decrement our
1195             ** counters to reflect one less buffer that needs writing.
1196             **
1197             ** note, this relies on all of the above code being
1198             ** schedule free once pjl comes back non-null.
1199             */
1200             if (pjl && cn->bh && atomic_read(&pjl->j_commit_left) == 0) {
1201                 atomic_dec(&cn->jlist->j_nonzerolen) ;
1202                 cn->bh = NULL ;
1203             } 
1204     next:
1205             cn = cn->next ;
1206         }
1207         /* the first run through the loop sends all the dirty buffers to
1208         ** ll_rw_block.
1209         ** the second run through the loop does all the accounting
1210         */
1211         if (run++ == 0) {
1212             goto loop_start ;
1213         }
1214     
1215         atomic_set(&(jl->j_flushing), 0) ;
1216         wake_up(&(jl->j_flush_wait)) ;
1217         return ret ;
1218     }
1219     /* since we never give dirty buffers to bdflush/kupdate, we have to
1220     ** flush them ourselves.  This runs through the journal lists, finds
1221     ** old metadata in need of flushing and sends it to disk.
1222     ** this does not end transactions, commit anything, or free
1223     ** cnodes.
1224     **
1225     ** returns the highest transaction id that was flushed last time
1226     */
1227     static unsigned long reiserfs_journal_kupdate(struct super_block *s) {
1228         struct reiserfs_journal_list *jl ;
1229         int i ;
1230         int start ;
1231         time_t age ;
1232         int ret = 0 ;
1233     
1234         start = SB_JOURNAL_LIST_INDEX(s) ;
1235     
1236         /* safety check to prevent flush attempts during a mount */
1237         if (start < 0) {
1238             return 0 ;
1239         }
1240         i = (start + 1) % JOURNAL_LIST_COUNT ;
1241         while(i != start) {
1242             jl = SB_JOURNAL_LIST(s) + i  ;
1243             age = CURRENT_TIME - jl->j_timestamp ;
1244             if (jl->j_len > 0 && // age >= (JOURNAL_MAX_COMMIT_AGE * 2) && 
1245                 atomic_read(&(jl->j_nonzerolen)) > 0 &&
1246                 atomic_read(&(jl->j_commit_left)) == 0) {
1247     
1248                 if (jl->j_trans_id == SB_JOURNAL(s)->j_trans_id) {
1249                     break ;
1250                 }
1251                 /* if ret was already 1, we want to preserve that */
1252                 ret |= kupdate_one_transaction(s, jl) ;
1253             } 
1254             if (atomic_read(&(jl->j_nonzerolen)) > 0) {
1255                 ret |= 1 ;
1256             }
1257             i = (i + 1) % JOURNAL_LIST_COUNT ;
1258         }
1259         return ret ;
1260     }
1261     
1262     /*
1263     ** removes any nodes in table with name block and dev as bh.
1264     ** only touchs the hnext and hprev pointers.
1265     */
1266     void remove_journal_hash(struct reiserfs_journal_cnode **table, struct reiserfs_journal_list *jl,struct buffer_head *bh,
1267                              int remove_freed){
1268       struct reiserfs_journal_cnode *cur ;
1269       struct reiserfs_journal_cnode **head ;
1270     
1271       if (!bh)
1272         return ;
1273     
1274       head= &(journal_hash(table, bh->b_dev, bh->b_blocknr)) ;
1275       if (!head) {
1276         return ;
1277       }
1278       cur = *head ;
1279       while(cur) {
1280         if (cur->blocknr == bh->b_blocknr && cur->dev == bh->b_dev && (jl == NULL || jl == cur->jlist) && 
1281             (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1282           if (cur->hnext) {
1283             cur->hnext->hprev = cur->hprev ;
1284           }
1285           if (cur->hprev) {
1286     	cur->hprev->hnext = cur->hnext ;
1287           } else {
1288     	*head = cur->hnext ;
1289           }
1290           cur->blocknr = 0 ;
1291           cur->dev = 0 ;
1292           cur->state = 0 ;
1293           if (cur->bh && cur->jlist) /* anybody who clears the cur->bh will also dec the nonzerolen */
1294     	atomic_dec(&(cur->jlist->j_nonzerolen)) ;
1295           cur->bh = NULL ;
1296           cur->jlist = NULL ;
1297         } 
1298         cur = cur->hnext ;
1299       }
1300     }
1301     
1302     static void free_journal_ram(struct super_block *p_s_sb) {
1303       vfree(SB_JOURNAL(p_s_sb)->j_cnode_free_orig) ;
1304       free_list_bitmaps(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_bitmap) ;
1305       free_bitmap_nodes(p_s_sb) ; /* must be after free_list_bitmaps */
1306       if (SB_JOURNAL(p_s_sb)->j_header_bh) {
1307         brelse(SB_JOURNAL(p_s_sb)->j_header_bh) ;
1308       }
1309       vfree(SB_JOURNAL(p_s_sb)) ;
1310     }
1311     
1312     /*
1313     ** call on unmount.  Only set error to 1 if you haven't made your way out
1314     ** of read_super() yet.  Any other caller must keep error at 0.
1315     */
1316     static int do_journal_release(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, int error) {
1317       struct reiserfs_transaction_handle myth ;
1318     
1319       /* we only want to flush out transactions if we were called with error == 0
1320       */
1321       if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
1322         /* end the current trans */
1323         do_journal_end(th, p_s_sb,10, FLUSH_ALL) ;
1324     
1325         /* make sure something gets logged to force our way into the flush code */
1326         journal_join(&myth, p_s_sb, 1) ;
1327         reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
1328         journal_mark_dirty(&myth, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
1329         do_journal_end(&myth, p_s_sb,1, FLUSH_ALL) ;
1330       }
1331     
1332       /* we decrement before we wake up, because the commit thread dies off
1333       ** when it has been woken up and the count is <= 0
1334       */
1335       reiserfs_mounted_fs_count-- ;
1336       wake_up(&reiserfs_commit_thread_wait) ;
1337       sleep_on(&reiserfs_commit_thread_done) ;
1338     
1339       free_journal_ram(p_s_sb) ;
1340     
1341       return 0 ;
1342     }
1343     
1344     /*
1345     ** call on unmount.  flush all journal trans, release all alloc'd ram
1346     */
1347     int journal_release(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb) {
1348       return do_journal_release(th, p_s_sb, 0) ;
1349     }
1350     /*
1351     ** only call from an error condition inside reiserfs_read_super!
1352     */
1353     int journal_release_error(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb) {
1354       return do_journal_release(th, p_s_sb, 1) ;
1355     }
1356     
1357     /* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
1358     static int journal_compare_desc_commit(struct super_block *p_s_sb, struct reiserfs_journal_desc *desc, 
1359     			               struct reiserfs_journal_commit *commit) {
1360       if (le32_to_cpu(commit->j_trans_id) != le32_to_cpu(desc->j_trans_id) || 
1361           le32_to_cpu(commit->j_len) != le32_to_cpu(desc->j_len) || 
1362           le32_to_cpu(commit->j_len) > JOURNAL_TRANS_MAX || 
1363           le32_to_cpu(commit->j_len) <= 0 
1364       ) {
1365         return 1 ;
1366       }
1367       return 0 ;
1368     }
1369     /* returns 0 if it did not find a description block  
1370     ** returns -1 if it found a corrupt commit block
1371     ** returns 1 if both desc and commit were valid 
1372     */
1373     static int journal_transaction_is_valid(struct super_block *p_s_sb, struct buffer_head *d_bh, unsigned long *oldest_invalid_trans_id, unsigned long *newest_mount_id) {
1374       struct reiserfs_journal_desc *desc ;
1375       struct reiserfs_journal_commit *commit ;
1376       struct buffer_head *c_bh ;
1377       unsigned long offset ;
1378     
1379       if (!d_bh)
1380           return 0 ;
1381     
1382       desc = (struct reiserfs_journal_desc *)d_bh->b_data ;
1383       if (le32_to_cpu(desc->j_len) > 0 && !memcmp(desc->j_magic, JOURNAL_DESC_MAGIC, 8)) {
1384         if (oldest_invalid_trans_id && *oldest_invalid_trans_id && le32_to_cpu(desc->j_trans_id) > *oldest_invalid_trans_id) {
1385           reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-986: transaction "
1386     	              "is valid returning because trans_id %d is greater than "
1387     		      "oldest_invalid %lu\n", le32_to_cpu(desc->j_trans_id), 
1388     		       *oldest_invalid_trans_id);
1389           return 0 ;
1390         }
1391         if (newest_mount_id && *newest_mount_id > le32_to_cpu(desc->j_mount_id)) {
1392           reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1087: transaction "
1393                          "is valid returning because mount_id %d is less than "
1394     		     "newest_mount_id %lu\n", desc->j_mount_id, 
1395     		     *newest_mount_id) ;
1396           return -1 ;
1397         }
1398         offset = d_bh->b_blocknr - reiserfs_get_journal_block(p_s_sb) ;
1399     
1400         /* ok, we have a journal description block, lets see if the transaction was valid */
1401         c_bh = bread(p_s_sb->s_dev, reiserfs_get_journal_block(p_s_sb) + ((offset + le32_to_cpu(desc->j_len) + 1) % JOURNAL_BLOCK_COUNT), 
1402         		p_s_sb->s_blocksize) ;
1403         if (!c_bh)
1404           return 0 ;
1405         commit = (struct reiserfs_journal_commit *)c_bh->b_data ;
1406         if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
1407           reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 
1408                          "journal_transaction_is_valid, commit offset %ld had bad "
1409     		     "time %d or length %d\n", 
1410     		     c_bh->b_blocknr - reiserfs_get_journal_block(p_s_sb),
1411     		     le32_to_cpu(commit->j_trans_id), 
1412     		     le32_to_cpu(commit->j_len));
1413           brelse(c_bh) ;
1414           if (oldest_invalid_trans_id)
1415             *oldest_invalid_trans_id = le32_to_cpu(desc->j_trans_id) ;
1416     	reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1004: "
1417     	               "transaction_is_valid setting oldest invalid trans_id "
1418     		       "to %d\n", le32_to_cpu(desc->j_trans_id)) ;
1419           return -1; 
1420         }
1421         brelse(c_bh) ;
1422         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1006: found valid "
1423                        "transaction start offset %lu, len %d id %d\n", 
1424     		   d_bh->b_blocknr - reiserfs_get_journal_block(p_s_sb), 
1425     		   le32_to_cpu(desc->j_len), le32_to_cpu(desc->j_trans_id)) ;
1426         return 1 ;
1427       } else {
1428         return 0 ;
1429       }
1430     }
1431     
1432     static void brelse_array(struct buffer_head **heads, int num) {
1433       int i ;
1434       for (i = 0 ; i < num ; i++) {
1435         brelse(heads[i]) ;
1436       }
1437     }
1438     
1439     /*
1440     ** given the start, and values for the oldest acceptable transactions,
1441     ** this either reads in a replays a transaction, or returns because the transaction
1442     ** is invalid, or too old.
1443     */
1444     static int journal_read_transaction(struct super_block *p_s_sb, unsigned long cur_dblock, unsigned long oldest_start, 
1445     				    unsigned long oldest_trans_id, unsigned long newest_mount_id) {
1446       struct reiserfs_journal_desc *desc ;
1447       struct reiserfs_journal_commit *commit ;
1448       unsigned long trans_id = 0 ;
1449       struct buffer_head *c_bh ;
1450       struct buffer_head *d_bh ;
1451       struct buffer_head **log_blocks = NULL ;
1452       struct buffer_head **real_blocks = NULL ;
1453       unsigned long trans_offset ;
1454       int i;
1455     
1456       d_bh = bread(p_s_sb->s_dev, cur_dblock, p_s_sb->s_blocksize) ;
1457       if (!d_bh)
1458         return 1 ;
1459       desc = (struct reiserfs_journal_desc *)d_bh->b_data ;
1460       trans_offset = d_bh->b_blocknr - reiserfs_get_journal_block(p_s_sb) ;
1461       reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
1462                      "journal_read_transaction, offset %lu, len %d mount_id %d\n", 
1463     		 d_bh->b_blocknr - reiserfs_get_journal_block(p_s_sb), 
1464     		 le32_to_cpu(desc->j_len), le32_to_cpu(desc->j_mount_id)) ;
1465       if (le32_to_cpu(desc->j_trans_id) < oldest_trans_id) {
1466         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
1467                        "journal_read_trans skipping because %lu is too old\n", 
1468     		   cur_dblock - reiserfs_get_journal_block(p_s_sb)) ;
1469         brelse(d_bh) ;
1470         return 1 ;
1471       }
1472       if (le32_to_cpu(desc->j_mount_id) != newest_mount_id) {
1473         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
1474                        "journal_read_trans skipping because %d is != "
1475     		   "newest_mount_id %lu\n", le32_to_cpu(desc->j_mount_id), 
1476     		    newest_mount_id) ;
1477         brelse(d_bh) ;
1478         return 1 ;
1479       }
1480       c_bh = bread(p_s_sb->s_dev, reiserfs_get_journal_block(p_s_sb) + ((trans_offset + le32_to_cpu(desc->j_len) + 1) % JOURNAL_BLOCK_COUNT), 
1481         		p_s_sb->s_blocksize) ;
1482       if (!c_bh) {
1483         brelse(d_bh) ;
1484         return 1 ;
1485       }
1486       commit = (struct reiserfs_journal_commit *)c_bh->b_data ;
1487       if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
1488         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal_read_transaction, "
1489                        "commit offset %ld had bad time %d or length %d\n", 
1490     		   c_bh->b_blocknr - reiserfs_get_journal_block(p_s_sb), 
1491     		   le32_to_cpu(commit->j_trans_id), le32_to_cpu(commit->j_len));
1492         brelse(c_bh) ;
1493         brelse(d_bh) ;
1494         return 1; 
1495       }
1496       trans_id = le32_to_cpu(desc->j_trans_id) ;
1497       /* now we know we've got a good transaction, and it was inside the valid time ranges */
1498       log_blocks = kmalloc(le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), GFP_NOFS) ;
1499       real_blocks = kmalloc(le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), GFP_NOFS) ;
1500       if (!log_blocks  || !real_blocks) {
1501         brelse(c_bh) ;
1502         brelse(d_bh) ;
1503         kfree(log_blocks) ;
1504         kfree(real_blocks) ;
1505         reiserfs_warning("journal-1169: kmalloc failed, unable to mount FS\n") ;
1506         return -1 ;
1507       }
1508       /* get all the buffer heads */
1509       for(i = 0 ; i < le32_to_cpu(desc->j_len) ; i++) {
1510         log_blocks[i] = getblk(p_s_sb->s_dev, reiserfs_get_journal_block(p_s_sb) + (trans_offset + 1 + i) % JOURNAL_BLOCK_COUNT, p_s_sb->s_blocksize);
1511         if (i < JOURNAL_TRANS_HALF) {
1512           real_blocks[i] = getblk(p_s_sb->s_dev, le32_to_cpu(desc->j_realblock[i]), p_s_sb->s_blocksize) ;
1513         } else {
1514           real_blocks[i] = getblk(p_s_sb->s_dev, le32_to_cpu(commit->j_realblock[i - JOURNAL_TRANS_HALF]), p_s_sb->s_blocksize) ;
1515         }
1516         if (real_blocks[i]->b_blocknr >= reiserfs_get_journal_block(p_s_sb) &&
1517             real_blocks[i]->b_blocknr < (reiserfs_get_journal_block(p_s_sb)+JOURNAL_BLOCK_COUNT)) {
1518           reiserfs_warning("journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block\n") ;
1519           brelse_array(log_blocks, i) ;
1520           brelse_array(real_blocks, i) ;
1521           brelse(c_bh) ;
1522           brelse(d_bh) ;
1523           kfree(log_blocks) ;
1524           kfree(real_blocks) ;
1525           return -1 ;
1526         }
1527       }
1528       /* read in the log blocks, memcpy to the corresponding real block */
1529       ll_rw_block(READ, le32_to_cpu(desc->j_len), log_blocks) ;
1530       for (i = 0 ; i < le32_to_cpu(desc->j_len) ; i++) {
1531         wait_on_buffer(log_blocks[i]) ;
1532         if (!buffer_uptodate(log_blocks[i])) {
1533           reiserfs_warning("journal-1212: REPLAY FAILURE fsck required! buffer write failed\n") ;
1534           brelse_array(log_blocks + i, le32_to_cpu(desc->j_len) - i) ;
1535           brelse_array(real_blocks, le32_to_cpu(desc->j_len)) ;
1536           brelse(c_bh) ;
1537           brelse(d_bh) ;
1538           kfree(log_blocks) ;
1539           kfree(real_blocks) ;
1540           return -1 ;
1541         }
1542         memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data, real_blocks[i]->b_size) ;
1543         mark_buffer_uptodate(real_blocks[i], 1) ;
1544         brelse(log_blocks[i]) ;
1545       }
1546       /* flush out the real blocks */
1547       for (i = 0 ; i < le32_to_cpu(desc->j_len) ; i++) {
1548         set_bit(BH_Dirty, &(real_blocks[i]->b_state)) ;
1549         ll_rw_block(WRITE, 1, real_blocks + i) ;
1550       }
1551       for (i = 0 ; i < le32_to_cpu(desc->j_len) ; i++) {
1552         wait_on_buffer(real_blocks[i]) ; 
1553         if (!buffer_uptodate(real_blocks[i])) {
1554           reiserfs_warning("journal-1226: REPLAY FAILURE, fsck required! buffer write failed\n") ;
1555           brelse_array(real_blocks + i, le32_to_cpu(desc->j_len) - i) ;
1556           brelse(c_bh) ;
1557           brelse(d_bh) ;
1558           kfree(log_blocks) ;
1559           kfree(real_blocks) ;
1560           return -1 ;
1561         }
1562         brelse(real_blocks[i]) ;
1563       }
1564       cur_dblock = reiserfs_get_journal_block(p_s_sb) + ((trans_offset + le32_to_cpu(desc->j_len) + 2) % JOURNAL_BLOCK_COUNT) ;
1565       reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1095: setting journal "
1566                      "start to offset %ld\n", 
1567     		 cur_dblock - reiserfs_get_journal_block(p_s_sb)) ;
1568       
1569       /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
1570       SB_JOURNAL(p_s_sb)->j_start = cur_dblock - reiserfs_get_journal_block(p_s_sb) ;
1571       SB_JOURNAL(p_s_sb)->j_last_flush_trans_id = trans_id ;
1572       SB_JOURNAL(p_s_sb)->j_trans_id = trans_id + 1;
1573       brelse(c_bh) ;
1574       brelse(d_bh) ;
1575       kfree(log_blocks) ;
1576       kfree(real_blocks) ;
1577       return 0 ;
1578     }
1579     
1580     /*
1581     ** read and replay the log
1582     ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
1583     ** transaction.  This tests that before finding all the transactions in the log, whic makes normal mount times fast.
1584     **
1585     ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
1586     **
1587     ** On exit, it sets things up so the first transaction will work correctly.
1588     */
1589     static int journal_read(struct super_block *p_s_sb) {
1590       struct reiserfs_journal_desc *desc ;
1591       unsigned long last_flush_trans_id = 0 ;
1592       unsigned long oldest_trans_id = 0;
1593       unsigned long oldest_invalid_trans_id = 0 ;
1594       time_t start ;
1595       unsigned long last_flush_start = 0;
1596       unsigned long oldest_start = 0;
1597       unsigned long cur_dblock = 0 ;
1598       unsigned long newest_mount_id = 9 ;
1599       struct buffer_head *d_bh ;
1600       struct reiserfs_journal_header *jh ;
1601       int valid_journal_header = 0 ;
1602       int replay_count = 0 ;
1603       int continue_replay = 1 ;
1604       int ret ;
1605     
1606       cur_dblock = reiserfs_get_journal_block(p_s_sb) ;
1607       printk("reiserfs: checking transaction log (device %s) ...\n",
1608               kdevname(p_s_sb->s_dev)) ;
1609       start = CURRENT_TIME ;
1610     
1611       /* step 1, read in the journal header block.  Check the transaction it says 
1612       ** is the first unflushed, and if that transaction is not valid, 
1613       ** replay is done
1614       */
1615       SB_JOURNAL(p_s_sb)->j_header_bh = bread(p_s_sb->s_dev, 
1616                                               reiserfs_get_journal_block(p_s_sb) + 
1617     					  JOURNAL_BLOCK_COUNT, 
1618     					  p_s_sb->s_blocksize) ;
1619       if (!SB_JOURNAL(p_s_sb)->j_header_bh) {
1620         return 1 ;
1621       }
1622       jh = (struct reiserfs_journal_header *)(SB_JOURNAL(p_s_sb)->j_header_bh->b_data) ;
1623       if (le32_to_cpu(jh->j_first_unflushed_offset) >= 0 && 
1624           le32_to_cpu(jh->j_first_unflushed_offset) < JOURNAL_BLOCK_COUNT &&
1625           le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
1626         last_flush_start = reiserfs_get_journal_block(p_s_sb) + 
1627                            le32_to_cpu(jh->j_first_unflushed_offset) ;
1628         last_flush_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) ;
1629         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1153: found in "
1630                        "header: first_unflushed_offset %d, last_flushed_trans_id "
1631     		   "%lu\n", le32_to_cpu(jh->j_first_unflushed_offset), 
1632     		   last_flush_trans_id) ;
1633         valid_journal_header = 1 ;
1634     
1635         /* now, we try to read the first unflushed offset.  If it is not valid, 
1636         ** there is nothing more we can do, and it makes no sense to read 
1637         ** through the whole log.
1638         */
1639         d_bh = bread(p_s_sb->s_dev, reiserfs_get_journal_block(p_s_sb) + le32_to_cpu(jh->j_first_unflushed_offset), p_s_sb->s_blocksize) ;
1640         ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL) ;
1641         if (!ret) {
1642           continue_replay = 0 ;
1643         }
1644         brelse(d_bh) ;
1645       }
1646     
1647       if (continue_replay && is_read_only(p_s_sb->s_dev)) {
1648         printk("clm-2076: device is readonly, unable to replay log\n") ;
1649         return -1 ;
1650       }
1651       if (continue_replay && (p_s_sb->s_flags & MS_RDONLY)) {
1652         printk("Warning, log replay starting on readonly filesystem\n") ;    
1653       }
1654     
1655       /* ok, there are transactions that need to be replayed.  start with the first log block, find
1656       ** all the valid transactions, and pick out the oldest.
1657       */
1658       while(continue_replay && cur_dblock < (reiserfs_get_journal_block(p_s_sb) + JOURNAL_BLOCK_COUNT)) {
1659         d_bh = bread(p_s_sb->s_dev, cur_dblock, p_s_sb->s_blocksize) ;
1660         ret = journal_transaction_is_valid(p_s_sb, d_bh, &oldest_invalid_trans_id, &newest_mount_id) ;
1661         if (ret == 1) {
1662           desc = (struct reiserfs_journal_desc *)d_bh->b_data ;
1663           if (oldest_start == 0) { /* init all oldest_ values */
1664             oldest_trans_id = le32_to_cpu(desc->j_trans_id) ;
1665     	oldest_start = d_bh->b_blocknr ;
1666     	newest_mount_id = le32_to_cpu(desc->j_mount_id) ;
1667     	reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1179: Setting "
1668     	               "oldest_start to offset %lu, trans_id %lu\n", 
1669     		       oldest_start - reiserfs_get_journal_block(p_s_sb), 
1670     		       oldest_trans_id) ;
1671           } else if (oldest_trans_id > le32_to_cpu(desc->j_trans_id)) { 
1672             /* one we just read was older */
1673             oldest_trans_id = le32_to_cpu(desc->j_trans_id) ;
1674     	oldest_start = d_bh->b_blocknr ;
1675     	reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1180: Resetting "
1676     	               "oldest_start to offset %lu, trans_id %lu\n", 
1677     			oldest_start - reiserfs_get_journal_block(p_s_sb), 
1678     			oldest_trans_id) ;
1679           }
1680           if (newest_mount_id < le32_to_cpu(desc->j_mount_id)) {
1681             newest_mount_id = le32_to_cpu(desc->j_mount_id) ;
1682     	reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
1683     	              "newest_mount_id to %d\n", le32_to_cpu(desc->j_mount_id));
1684           }
1685           cur_dblock += le32_to_cpu(desc->j_len) + 2 ;
1686         } 
1687         else {
1688           cur_dblock++ ;
1689         }
1690         brelse(d_bh) ;
1691       }
1692       /* step three, starting at the oldest transaction, replay */
1693       if (last_flush_start > 0) {
1694         oldest_start = last_flush_start ;
1695         oldest_trans_id = last_flush_trans_id ;
1696       } 
1697       cur_dblock = oldest_start ;
1698       if (oldest_trans_id)  {
1699         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1206: Starting replay "
1700                        "from offset %lu, trans_id %lu\n", 
1701     		   cur_dblock - reiserfs_get_journal_block(p_s_sb), 
1702     		   oldest_trans_id) ;
1703     
1704       }
1705       replay_count = 0 ;
1706       while(continue_replay && oldest_trans_id > 0) {
1707         ret = journal_read_transaction(p_s_sb, cur_dblock, oldest_start, oldest_trans_id, newest_mount_id) ;
1708         if (ret < 0) {
1709           return ret ;
1710         } else if (ret != 0) {
1711           break ;
1712         }
1713         cur_dblock = reiserfs_get_journal_block(p_s_sb) + SB_JOURNAL(p_s_sb)->j_start ;
1714         replay_count++ ;
1715       }
1716     
1717       if (oldest_trans_id == 0) {
1718         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1225: No valid "
1719                        "transactions found\n") ;
1720       }
1721       /* j_start does not get set correctly if we don't replay any transactions.
1722       ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
1723       ** copy the trans_id from the header
1724       */
1725       if (valid_journal_header && replay_count == 0) { 
1726         SB_JOURNAL(p_s_sb)->j_start = le32_to_cpu(jh->j_first_unflushed_offset) ;
1727         SB_JOURNAL(p_s_sb)->j_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
1728         SB_JOURNAL(p_s_sb)->j_last_flush_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) ;
1729         SB_JOURNAL(p_s_sb)->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
1730       } else {
1731         SB_JOURNAL(p_s_sb)->j_mount_id = newest_mount_id + 1 ;
1732       }
1733       reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
1734                      "newest_mount_id to %lu\n", SB_JOURNAL(p_s_sb)->j_mount_id) ;
1735       SB_JOURNAL(p_s_sb)->j_first_unflushed_offset = SB_JOURNAL(p_s_sb)->j_start ; 
1736       if (replay_count > 0) {
1737         printk("reiserfs: replayed %d transactions in %lu seconds\n", replay_count, 
1738     	    CURRENT_TIME - start) ;
1739       }
1740       if (!is_read_only(p_s_sb->s_dev) && 
1741            _update_journal_header_block(p_s_sb, SB_JOURNAL(p_s_sb)->j_start, 
1742                                        SB_JOURNAL(p_s_sb)->j_last_flush_trans_id))
1743       {
1744           /* replay failed, caller must call free_journal_ram and abort
1745           ** the mount
1746           */
1747           return -1 ;
1748       }
1749       return 0 ;
1750     }
1751     
1752     
1753     struct reiserfs_journal_commit_task {
1754       struct super_block *p_s_sb ;
1755       int jindex ;
1756       int wake_on_finish ; /* if this is one, we wake the task_done queue, if it
1757                            ** is zero, we free the whole struct on finish
1758     		       */
1759       struct reiserfs_journal_commit_task *self ;
1760       struct wait_queue *task_done ;
1761       struct tq_struct task ;
1762     } ;
1763     
1764     static void reiserfs_journal_commit_task_func(struct reiserfs_journal_commit_task *ct) {
1765     
1766       struct reiserfs_journal_list *jl ;
1767       jl = SB_JOURNAL_LIST(ct->p_s_sb) + ct->jindex ;
1768     
1769       flush_commit_list(ct->p_s_sb, SB_JOURNAL_LIST(ct->p_s_sb) + ct->jindex, 1) ; 
1770       if (jl->j_len > 0 && atomic_read(&(jl->j_nonzerolen)) > 0 && 
1771           atomic_read(&(jl->j_commit_left)) == 0) {
1772         kupdate_one_transaction(ct->p_s_sb, jl) ;
1773       }
1774       kfree(ct->self) ;
1775     }
1776     
1777     static void setup_commit_task_arg(struct reiserfs_journal_commit_task *ct,
1778                                       struct super_block *p_s_sb, 
1779     				  int jindex) {
1780       if (!ct) {
1781         reiserfs_panic(NULL, "journal-1360: setup_commit_task_arg called with NULL struct\n") ;
1782       }
1783       ct->p_s_sb = p_s_sb ;
1784       ct->jindex = jindex ;
1785       ct->task_done = NULL ;
1786       INIT_LIST_HEAD(&ct->task.list) ;
1787       ct->task.sync = 0 ;
1788       ct->task.routine = (void *)(void *)reiserfs_journal_commit_task_func ; 
1789       ct->self = ct ;
1790       ct->task.data = (void *)ct ;
1791     }
1792     
1793     static void commit_flush_async(struct super_block *p_s_sb, int jindex) {
1794       struct reiserfs_journal_commit_task *ct ;
1795       /* using GFP_NOFS, GFP_KERNEL could try to flush inodes, which will try
1796       ** to start/join a transaction, which will deadlock
1797       */
1798       ct = kmalloc(sizeof(struct reiserfs_journal_commit_task), GFP_NOFS) ;
1799       if (ct) {
1800         setup_commit_task_arg(ct, p_s_sb, jindex) ;
1801         queue_task(&(ct->task), &reiserfs_commit_thread_tq);
1802         wake_up(&reiserfs_commit_thread_wait) ;
1803       } else {
1804     #ifdef CONFIG_REISERFS_CHECK
1805         reiserfs_warning("journal-1540: kmalloc failed, doing sync commit\n") ;
1806     #endif
1807         flush_commit_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + jindex, 1) ;
1808       }
1809     }
1810     
1811     /*
1812     ** this is the commit thread.  It is started with kernel_thread on
1813     ** FS mount, and journal_release() waits for it to exit.
1814     **
1815     ** It could do a periodic commit, but there is a lot code for that
1816     ** elsewhere right now, and I only wanted to implement this little
1817     ** piece for starters.
1818     **
1819     ** All we do here is sleep on the j_commit_thread_wait wait queue, and
1820     ** then run the per filesystem commit task queue when we wakeup.
1821     */
1822     static int reiserfs_journal_commit_thread(void *nullp) {
1823     
1824       daemonize() ;
1825     
1826       spin_lock_irq(&current->sigmask_lock);
1827       sigfillset(&current->blocked);
1828       recalc_sigpending(current);
1829       spin_unlock_irq(&current->sigmask_lock);
1830     
1831       sprintf(current->comm, "kreiserfsd") ;
1832       lock_kernel() ;
1833       while(1) {
1834     
1835         while(TQ_ACTIVE(reiserfs_commit_thread_tq)) {
1836           run_task_queue(&reiserfs_commit_thread_tq) ;
1837         }
1838     
1839         /* if there aren't any more filesystems left, break */
1840         if (reiserfs_mounted_fs_count <= 0) {
1841           run_task_queue(&reiserfs_commit_thread_tq) ;
1842           break ;
1843         }
1844         wake_up(&reiserfs_commit_thread_done) ;
1845         interruptible_sleep_on_timeout(&reiserfs_commit_thread_wait, 5) ;
1846       }
1847       unlock_kernel() ;
1848       wake_up(&reiserfs_commit_thread_done) ;
1849       return 0 ;
1850     }
1851     
1852     static void journal_list_init(struct super_block *p_s_sb) {
1853       int i ;
1854       for (i = 0 ; i < JOURNAL_LIST_COUNT ; i++) {
1855         init_waitqueue_head(&(SB_JOURNAL_LIST(p_s_sb)[i].j_commit_wait)) ;
1856         init_waitqueue_head(&(SB_JOURNAL_LIST(p_s_sb)[i].j_flush_wait)) ;
1857       }
1858     }
1859     
1860     /*
1861     ** must be called once on fs mount.  calls journal_read for you
1862     */
1863     int journal_init(struct super_block *p_s_sb) {
1864       int num_cnodes = JOURNAL_BLOCK_COUNT * 2 ;
1865     
1866       if (sizeof(struct reiserfs_journal_commit) != 4096 ||
1867           sizeof(struct reiserfs_journal_desc) != 4096
1868          ) {
1869         printk("journal-1249: commit or desc struct not 4096 %Zd %Zd\n", sizeof(struct reiserfs_journal_commit), 
1870             sizeof(struct reiserfs_journal_desc)) ;
1871         return 1 ;
1872       }
1873       /* sanity check to make sure they don't overflow the journal */
1874       if (JOURNAL_BLOCK_COUNT > reiserfs_get_journal_orig_size(p_s_sb)) {
1875         printk("journal-1393: current JOURNAL_BLOCK_COUNT (%d) is too big.  This FS was created with a journal size of %lu blocks\n",
1876                 JOURNAL_BLOCK_COUNT, reiserfs_get_journal_orig_size(p_s_sb)) ;
1877         return 1 ;
1878       }
1879       SB_JOURNAL(p_s_sb) = vmalloc(sizeof (struct reiserfs_journal)) ;
1880     
1881       if (!SB_JOURNAL(p_s_sb)) {
1882         printk("journal-1256: unable to get memory for journal structure\n") ;
1883         return 1 ;
1884       }
1885       memset(SB_JOURNAL(p_s_sb), 0, sizeof(struct reiserfs_journal)) ;
1886     
1887       SB_JOURNAL(p_s_sb)->j_list_bitmap_index = 0 ;
1888       SB_JOURNAL_LIST_INDEX(p_s_sb) = -10000 ; /* make sure flush_old_commits does not try to flush a list while replay is on */
1889     
1890       /* clear out the journal list array */
1891       memset(SB_JOURNAL_LIST(p_s_sb), 0, sizeof(struct reiserfs_journal_list) * JOURNAL_LIST_COUNT) ; 
1892       journal_list_init(p_s_sb) ;
1893     
1894       memset(SB_JOURNAL(p_s_sb)->j_list_hash_table, 0, JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)) ;
1895       memset(journal_writers, 0, sizeof(char *) * 512) ; /* debug code */
1896     
1897       INIT_LIST_HEAD(&SB_JOURNAL(p_s_sb)->j_bitmap_nodes) ;
1898       INIT_LIST_HEAD(&(SB_JOURNAL(p_s_sb)->j_dummy_inode.i_dirty_buffers)) ;
1899       reiserfs_allocate_list_bitmaps(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_bitmap, 
1900                                      SB_BMAP_NR(p_s_sb)) ;
1901       allocate_bitmap_nodes(p_s_sb) ;
1902     
1903       SB_JOURNAL(p_s_sb)->j_start = 0 ;
1904       SB_JOURNAL(p_s_sb)->j_len = 0 ;
1905       SB_JOURNAL(p_s_sb)->j_len_alloc = 0 ;
1906       atomic_set(&(SB_JOURNAL(p_s_sb)->j_wcount), 0) ;
1907       SB_JOURNAL(p_s_sb)->j_bcount = 0 ;	  
1908       SB_JOURNAL(p_s_sb)->j_trans_start_time = 0 ;	  
1909       SB_JOURNAL(p_s_sb)->j_last = NULL ;	  
1910       SB_JOURNAL(p_s_sb)->j_first = NULL ;     
1911       init_waitqueue_head(&(SB_JOURNAL(p_s_sb)->j_join_wait)) ;
1912       init_waitqueue_head(&(SB_JOURNAL(p_s_sb)->j_wait)) ; 
1913     
1914       SB_JOURNAL(p_s_sb)->j_trans_id = 10 ;  
1915       SB_JOURNAL(p_s_sb)->j_mount_id = 10 ; 
1916       SB_JOURNAL(p_s_sb)->j_state = 0 ;
1917       atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 0) ;
1918       atomic_set(&(SB_JOURNAL(p_s_sb)->j_wlock), 0) ;
1919       SB_JOURNAL(p_s_sb)->j_cnode_free_list = allocate_cnodes(num_cnodes) ;
1920       SB_JOURNAL(p_s_sb)->j_cnode_free_orig = SB_JOURNAL(p_s_sb)->j_cnode_free_list ;
1921       SB_JOURNAL(p_s_sb)->j_cnode_free = SB_JOURNAL(p_s_sb)->j_cnode_free_list ? num_cnodes : 0 ;
1922       SB_JOURNAL(p_s_sb)->j_cnode_used = 0 ;
1923       SB_JOURNAL(p_s_sb)->j_must_wait = 0 ;
1924       init_journal_hash(p_s_sb) ;
1925       SB_JOURNAL_LIST(p_s_sb)[0].j_list_bitmap = get_list_bitmap(p_s_sb, SB_JOURNAL_LIST(p_s_sb)) ;
1926       if (!(SB_JOURNAL_LIST(p_s_sb)[0].j_list_bitmap)) {
1927         reiserfs_warning("journal-2005, get_list_bitmap failed for journal list 0\n") ;
1928         return 1 ;
1929       }
1930       if (journal_read(p_s_sb) < 0) {
1931         reiserfs_warning("Replay Failure, unable to mount\n") ;
1932         free_journal_ram(p_s_sb) ;
1933         return 1 ;
1934       }
1935       SB_JOURNAL_LIST_INDEX(p_s_sb) = 0 ; /* once the read is done, we can set this
1936                                              where it belongs */
1937     
1938       INIT_LIST_HEAD (&SB_JOURNAL(p_s_sb)->j_prealloc_list);
1939       
1940       if (reiserfs_dont_log (p_s_sb))
1941         return 0;
1942     
1943       reiserfs_mounted_fs_count++ ;
1944       if (reiserfs_mounted_fs_count <= 1) {
1945         kernel_thread((void *)(void *)reiserfs_journal_commit_thread, NULL,
1946                       CLONE_FS | CLONE_FILES | CLONE_VM) ;
1947       }
1948       return 0 ;
1949     }
1950     
1951     /*
1952     ** test for a polite end of the current transaction.  Used by file_write, and should
1953     ** be used by delete to make sure they don't write more than can fit inside a single
1954     ** transaction
1955     */
1956     int journal_transaction_should_end(struct reiserfs_transaction_handle *th, int new_alloc) {
1957       time_t now = CURRENT_TIME ;
1958       if (reiserfs_dont_log(th->t_super)) 
1959         return 0 ;
1960       if ( SB_JOURNAL(th->t_super)->j_must_wait > 0 ||
1961            (SB_JOURNAL(th->t_super)->j_len_alloc + new_alloc) >= JOURNAL_MAX_BATCH || 
1962            atomic_read(&(SB_JOURNAL(th->t_super)->j_jlock)) ||
1963           (now - SB_JOURNAL(th->t_super)->j_trans_start_time) > JOURNAL_MAX_TRANS_AGE ||
1964            SB_JOURNAL(th->t_super)->j_cnode_free < (JOURNAL_TRANS_MAX * 3)) { 
1965         return 1 ;
1966       }
1967       return 0 ;
1968     }
1969     
1970     /* this must be called inside a transaction, and requires the 
1971     ** kernel_lock to be held
1972     */
1973     void reiserfs_block_writes(struct reiserfs_transaction_handle *th) {
1974         struct super_block *s = th->t_super ;
1975         SB_JOURNAL(s)->j_must_wait = 1 ;
1976         set_bit(WRITERS_BLOCKED, &SB_JOURNAL(s)->j_state) ;
1977         return ;
1978     }
1979     
1980     /* this must be called without a transaction started, and does not
1981     ** require BKL
1982     */
1983     void reiserfs_allow_writes(struct super_block *s) {
1984         clear_bit(WRITERS_BLOCKED, &SB_JOURNAL(s)->j_state) ;
1985         wake_up(&SB_JOURNAL(s)->j_join_wait) ;
1986     }
1987     
1988     /* this must be called without a transaction started, and does not
1989     ** require BKL
1990     */
1991     void reiserfs_wait_on_write_block(struct super_block *s) {
1992         wait_event(SB_JOURNAL(s)->j_join_wait, 
1993                    !test_bit(WRITERS_BLOCKED, &SB_JOURNAL(s)->j_state)) ;
1994     }
1995     
1996     /* join == true if you must join an existing transaction.
1997     ** join == false if you can deal with waiting for others to finish
1998     **
1999     ** this will block until the transaction is joinable.  send the number of blocks you
2000     ** expect to use in nblocks.
2001     */
2002     static int do_journal_begin_r(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb,unsigned long nblocks,int join) {
2003       time_t now = CURRENT_TIME ;
2004       int old_trans_id  ;
2005     
2006       reiserfs_check_lock_depth("journal_begin") ;
2007     #ifdef CONFIG_REISERFS_CHECK
2008       if (p_s_sb->s_flags & MS_RDONLY) {
2009         printk("clm-2078: calling journal_begin on readonly FS\n") ;
2010         BUG() ;
2011       }
2012     #endif
2013     
2014       if (reiserfs_dont_log(p_s_sb)) {
2015         th->t_super = p_s_sb ; /* others will check this for the don't log flag */
2016         return 0 ;
2017       }
2018     
2019     relock:
2020       lock_journal(p_s_sb) ;
2021     
2022       if (test_bit(WRITERS_BLOCKED, &SB_JOURNAL(p_s_sb)->j_state)) {
2023         unlock_journal(p_s_sb) ;
2024         reiserfs_wait_on_write_block(p_s_sb) ;
2025         goto relock ;
2026       }
2027     
2028       /* if there is no room in the journal OR
2029       ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning 
2030       ** we don't sleep if there aren't other writers
2031       */
2032     
2033     
2034       if (  (!join && SB_JOURNAL(p_s_sb)->j_must_wait > 0) ||
2035          ( !join && (SB_JOURNAL(p_s_sb)->j_len_alloc + nblocks + 2) >= JOURNAL_MAX_BATCH) || 
2036          (!join && atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) > 0 && SB_JOURNAL(p_s_sb)->j_trans_start_time > 0 && 
2037           (now - SB_JOURNAL(p_s_sb)->j_trans_start_time) > JOURNAL_MAX_TRANS_AGE) ||
2038          (!join && atomic_read(&(SB_JOURNAL(p_s_sb)->j_jlock)) ) ||
2039          (!join && SB_JOURNAL(p_s_sb)->j_cnode_free < (JOURNAL_TRANS_MAX * 3))) {
2040     
2041         unlock_journal(p_s_sb) ; /* allow others to finish this transaction */
2042     
2043         /* if writer count is 0, we can just force this transaction to end, and start
2044         ** a new one afterwards.
2045         */
2046         if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) <= 0) {
2047           struct reiserfs_transaction_handle myth ;
2048           journal_join(&myth, p_s_sb, 1) ;
2049           reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
2050           journal_mark_dirty(&myth, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
2051           do_journal_end(&myth, p_s_sb,1,COMMIT_NOW) ;
2052         } else {
2053           /* but if the writer count isn't zero, we have to wait for the current writers to finish.
2054           ** They won't batch on transaction end once we set j_jlock
2055           */
2056           atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 1) ;
2057           old_trans_id = SB_JOURNAL(p_s_sb)->j_trans_id ;
2058           while(atomic_read(&(SB_JOURNAL(p_s_sb)->j_jlock)) &&
2059                 SB_JOURNAL(p_s_sb)->j_trans_id == old_trans_id) {
2060     	sleep_on(&(SB_JOURNAL(p_s_sb)->j_join_wait)) ;
2061           }
2062         }
2063         goto relock ;
2064       }
2065     
2066       if (SB_JOURNAL(p_s_sb)->j_trans_start_time == 0) { /* we are the first writer, set trans_id */
2067         SB_JOURNAL(p_s_sb)->j_trans_start_time = now ;
2068       }
2069       atomic_inc(&(SB_JOURNAL(p_s_sb)->j_wcount)) ;
2070       SB_JOURNAL(p_s_sb)->j_len_alloc += nblocks ;
2071       th->t_blocks_logged = 0 ;
2072       th->t_blocks_allocated = nblocks ;
2073       th->t_super = p_s_sb ;
2074       th->t_trans_id = SB_JOURNAL(p_s_sb)->j_trans_id ;
2075       th->t_caller = "Unknown" ;
2076       unlock_journal(p_s_sb) ;
2077       p_s_sb->s_dirt = 1; 
2078       return 0 ;
2079     }
2080     
2081     
2082     int journal_join(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) {
2083       return do_journal_begin_r(th, p_s_sb, nblocks, 1) ;
2084     }
2085     
2086     int journal_begin(struct reiserfs_transaction_handle *th, struct super_block  * p_s_sb, unsigned long nblocks) {
2087       return do_journal_begin_r(th, p_s_sb, nblocks, 0) ;
2088     }
2089     
2090     /* not used at all */
2091     int journal_prepare(struct super_block  * p_s_sb, struct buffer_head *bh) {
2092       return 0 ;
2093     }
2094     
2095     /*
2096     ** puts bh into the current transaction.  If it was already there, reorders removes the
2097     ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
2098     **
2099     ** if it was dirty, cleans and files onto the clean list.  I can't let it be dirty again until the
2100     ** transaction is committed.
2101     ** 
2102     ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
2103     */
2104     int journal_mark_dirty(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, struct buffer_head *bh) {
2105       struct reiserfs_journal_cnode *cn = NULL;
2106       int count_already_incd = 0 ;
2107       int prepared = 0 ;
2108     
2109       if (reiserfs_dont_log(th->t_super)) {
2110         mark_buffer_dirty(bh) ;
2111         return 0 ;
2112       }
2113     
2114       if (th->t_trans_id != SB_JOURNAL(p_s_sb)->j_trans_id) {
2115         reiserfs_panic(th->t_super, "journal-1577: handle trans id %d != current trans id %d\n", 
2116                        th->t_trans_id, SB_JOURNAL(p_s_sb)->j_trans_id);
2117       }
2118       p_s_sb->s_dirt = 1 ;
2119     
2120       prepared = test_and_clear_bit(BH_JPrepared, &bh->b_state) ;
2121       /* already in this transaction, we are done */
2122       if (buffer_journaled(bh)) {
2123         return 0 ;
2124       }
2125     
2126       /* this must be turned into a panic instead of a warning.  We can't allow
2127       ** a dirty or journal_dirty or locked buffer to be logged, as some changes
2128       ** could get to disk too early.  NOT GOOD.
2129       */
2130       if (!prepared || buffer_locked(bh)) {
2131         printk("journal-1777: buffer %lu bad state %cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT\n", bh->b_blocknr, prepared ? ' ' : '!', 
2132                                 buffer_locked(bh) ? ' ' : '!',
2133     			    buffer_dirty(bh) ? ' ' : '!',
2134     			    buffer_journal_dirty(bh) ? ' ' : '!') ;
2135         show_reiserfs_locks() ;
2136       }
2137       count_already_incd = clear_prepared_bits(bh) ;
2138     
2139       if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) <= 0) {
2140         printk("journal-1409: journal_mark_dirty returning because j_wcount was %d\n", atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount))) ;
2141         return 1 ;
2142       }
2143       /* this error means I've screwed up, and we've overflowed the transaction.  
2144       ** Nothing can be done here, except make the FS readonly or panic.
2145       */ 
2146       if (SB_JOURNAL(p_s_sb)->j_len >= JOURNAL_TRANS_MAX) { 
2147         reiserfs_panic(th->t_super, "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n", SB_JOURNAL(p_s_sb)->j_len) ;
2148       }
2149     
2150       if (buffer_journal_dirty(bh)) {
2151         count_already_incd = 1 ;
2152         mark_buffer_notjournal_dirty(bh) ;
2153       }
2154     
2155       if (buffer_dirty(bh)) {
2156         clear_bit(BH_Dirty, &bh->b_state) ;
2157       }
2158     
2159       if (buffer_journaled(bh)) { /* must double check after getting lock */
2160         goto done ;
2161       }
2162     
2163       if (SB_JOURNAL(p_s_sb)->j_len > SB_JOURNAL(p_s_sb)->j_len_alloc) {
2164         SB_JOURNAL(p_s_sb)->j_len_alloc = SB_JOURNAL(p_s_sb)->j_len + JOURNAL_PER_BALANCE_CNT ;
2165       }
2166     
2167       set_bit(BH_JDirty, &bh->b_state) ;
2168     
2169       /* now put this guy on the end */
2170       if (!cn) {
2171         cn = get_cnode(p_s_sb) ;
2172         if (!cn) {
2173           reiserfs_panic(p_s_sb, "get_cnode failed!\n"); 
2174         }
2175     
2176         if (th->t_blocks_logged == th->t_blocks_allocated) {
2177           th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT ;
2178           SB_JOURNAL(p_s_sb)->j_len_alloc += JOURNAL_PER_BALANCE_CNT ;
2179         }
2180         th->t_blocks_logged++ ;
2181         SB_JOURNAL(p_s_sb)->j_len++ ;
2182     
2183         cn->bh = bh ;
2184         cn->blocknr = bh->b_blocknr ;
2185         cn->dev = bh->b_dev ;
2186         cn->jlist = NULL ;
2187         insert_journal_hash(SB_JOURNAL(p_s_sb)->j_hash_table, cn) ;
2188         if (!count_already_incd) {
2189           get_bh(bh) ;
2190         }
2191       }
2192       cn->next = NULL ;
2193       cn->prev = SB_JOURNAL(p_s_sb)->j_last ;
2194       cn->bh = bh ;
2195       if (SB_JOURNAL(p_s_sb)->j_last) {
2196         SB_JOURNAL(p_s_sb)->j_last->next = cn ;
2197         SB_JOURNAL(p_s_sb)->j_last = cn ;
2198       } else {
2199         SB_JOURNAL(p_s_sb)->j_first = cn ;
2200         SB_JOURNAL(p_s_sb)->j_last = cn ;
2201       }
2202     done:
2203       return 0 ;
2204     }
2205     
2206     /*
2207     ** if buffer already in current transaction, do a journal_mark_dirty
2208     ** otherwise, just mark it dirty and move on.  Used for writes to meta blocks
2209     ** that don't need journaling
2210     */
2211     int journal_mark_dirty_nolog(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, struct buffer_head *bh) {
2212       if (reiserfs_dont_log(th->t_super) || buffer_journaled(bh) || 
2213           buffer_journal_dirty(bh)) {
2214         return journal_mark_dirty(th, p_s_sb, bh) ;
2215       }
2216       if (get_journal_hash_dev(SB_JOURNAL(p_s_sb)->j_list_hash_table, bh->b_dev,bh->b_blocknr,bh->b_size)) {
2217         return journal_mark_dirty(th, p_s_sb, bh) ;
2218       }
2219       mark_buffer_dirty(bh) ;
2220       return 0 ;
2221     }
2222     
2223     int journal_end(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) {
2224       return do_journal_end(th, p_s_sb, nblocks, 0) ;
2225     }
2226     
2227     /* removes from the current transaction, relsing and descrementing any counters.  
2228     ** also files the removed buffer directly onto the clean list
2229     **
2230     ** called by journal_mark_freed when a block has been deleted
2231     **
2232     ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
2233     */
2234     int remove_from_transaction(struct super_block *p_s_sb, unsigned long blocknr, int already_cleaned) {
2235       struct buffer_head *bh ;
2236       struct reiserfs_journal_cnode *cn ;
2237       int ret = 0;
2238     
2239       cn = get_journal_hash_dev(SB_JOURNAL(p_s_sb)->j_hash_table, p_s_sb->s_dev, blocknr, p_s_sb->s_blocksize) ;
2240       if (!cn || !cn->bh) {
2241         return ret ;
2242       }
2243       bh = cn->bh ;
2244       if (cn->prev) {
2245         cn->prev->next = cn->next ;
2246       }
2247       if (cn->next) {
2248         cn->next->prev = cn->prev ;
2249       }
2250       if (cn == SB_JOURNAL(p_s_sb)->j_first) {
2251         SB_JOURNAL(p_s_sb)->j_first = cn->next ;  
2252       }
2253       if (cn == SB_JOURNAL(p_s_sb)->j_last) {
2254         SB_JOURNAL(p_s_sb)->j_last = cn->prev ;
2255       }
2256       remove_journal_hash(SB_JOURNAL(p_s_sb)->j_hash_table, NULL, bh, 0) ; 
2257       mark_buffer_not_journaled(bh) ; /* don't log this one */
2258     
2259       if (!already_cleaned) {
2260         mark_buffer_notjournal_dirty(bh) ; 
2261         put_bh(bh) ;
2262         if (atomic_read(&(bh->b_count)) < 0) {
2263           printk("journal-1752: remove from trans, b_count < 0\n") ;
2264         }
2265         if (!buffer_locked(bh)) reiserfs_clean_and_file_buffer(bh) ; 
2266         ret = 1 ;
2267       }
2268       SB_JOURNAL(p_s_sb)->j_len-- ;
2269       SB_JOURNAL(p_s_sb)->j_len_alloc-- ;
2270       free_cnode(p_s_sb, cn) ;
2271       return ret ;
2272     }
2273     
2274     /* removes from a specific journal list hash */
2275     int remove_from_journal_list(struct super_block *s, struct reiserfs_journal_list *jl, struct buffer_head *bh, int remove_freed) {
2276       remove_journal_hash(SB_JOURNAL(s)->j_list_hash_table, jl, bh, remove_freed) ;
2277       return 0 ;
2278     }
2279     
2280     /*
2281     ** for any cnode in a journal list, it can only be dirtied of all the
2282     ** transactions that include it are commited to disk.
2283     ** this checks through each transaction, and returns 1 if you are allowed to dirty,
2284     ** and 0 if you aren't
2285     **
2286     ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
2287     ** blocks for a given transaction on disk
2288     **
2289     */
2290     static int can_dirty(struct reiserfs_journal_cnode *cn) {
2291       kdev_t dev = cn->dev ;
2292       unsigned long blocknr = cn->blocknr  ;
2293       struct reiserfs_journal_cnode *cur = cn->hprev ;
2294       int can_dirty = 1 ;
2295       
2296       /* first test hprev.  These are all newer than cn, so any node here
2297       ** with the name block number and dev means this node can't be sent
2298       ** to disk right now.
2299       */
2300       while(cur && can_dirty) {
2301         if (cur->jlist && cur->bh && cur->blocknr && cur->dev == dev && 
2302             cur->blocknr == blocknr) {
2303           can_dirty = 0 ;
2304         }
2305         cur = cur->hprev ;
2306       }
2307       /* then test hnext.  These are all older than cn.  As long as they
2308       ** are committed to the log, it is safe to write cn to disk
2309       */
2310       cur = cn->hnext ;
2311       while(cur && can_dirty) {
2312         if (cur->jlist && cur->jlist->j_len > 0 && 
2313             atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh && 
2314             cur->blocknr && cur->dev == dev && cur->blocknr == blocknr) {
2315           can_dirty = 0 ;
2316         }
2317         cur = cur->hnext ;
2318       }
2319       return can_dirty ;
2320     }
2321     
2322     /* syncs the commit blocks, but does not force the real buffers to disk
2323     ** will wait until the current transaction is done/commited before returning 
2324     */
2325     int journal_end_sync(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) {
2326       return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT) ;
2327     }
2328     
2329     int show_reiserfs_locks(void) {
2330     
2331       dump_journal_writers() ;
2332     #if 0 /* debugging code for when we are compiled static don't delete */
2333       p_s_sb = sb_entry(super_blocks.next);
2334       while (p_s_sb != sb_entry(&super_blocks)) {
2335         if (reiserfs_is_super(p_s_sb)) {
2336     printk("journal lock is %d, join lock is %d, writers %d must wait is %d\n", 
2337             atomic_read(&(SB_JOURNAL(p_s_sb)->j_wlock)),
2338             atomic_read(&(SB_JOURNAL(p_s_sb)->j_jlock)),
2339     	atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)),
2340     	SB_JOURNAL(p_s_sb)->j_must_wait) ;
2341     	printk("used cnodes %d, free cnodes %d\n", SB_JOURNAL(p_s_sb)->j_cnode_used, SB_JOURNAL(p_s_sb)->j_cnode_free) ;
2342         }
2343         p_s_sb = sb_entry(p_s_sb->s_list.next);
2344       }
2345     #endif
2346       return 0 ;
2347     }
2348     
2349     /*
2350     ** used to get memory back from async commits that are floating around
2351     ** and to reclaim any blocks deleted but unusable because their commits
2352     ** haven't hit disk yet.  called from bitmap.c
2353     **
2354     ** if it starts flushing things, it ors SCHEDULE_OCCURRED into repeat.
2355     ** note, this is just if schedule has a chance of occuring.  I need to 
2356     ** change flush_commit_lists to have a repeat parameter too.
2357     **
2358     */
2359     void flush_async_commits(struct super_block *p_s_sb) {
2360       int i ;
2361     
2362       for (i = 0 ; i < JOURNAL_LIST_COUNT ; i++) {
2363         if (i != SB_JOURNAL_LIST_INDEX(p_s_sb)) {
2364           flush_commit_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + i, 1) ; 
2365         }
2366       }
2367     }
2368     
2369     /*
2370     ** flushes any old transactions to disk
2371     ** ends the current transaction if it is too old
2372     **
2373     ** also calls flush_journal_list with old_only == 1, which allows me to reclaim
2374     ** memory and such from the journal lists whose real blocks are all on disk.
2375     **
2376     ** called by sync_dev_journal from buffer.c
2377     */
2378     int flush_old_commits(struct super_block *p_s_sb, int immediate) {
2379       int i ;
2380       int count = 0;
2381       int start ; 
2382       time_t now ; 
2383       struct reiserfs_transaction_handle th ; 
2384     
2385       start =  SB_JOURNAL_LIST_INDEX(p_s_sb) ;
2386       now = CURRENT_TIME ;
2387     
2388       /* safety check so we don't flush while we are replaying the log during mount */
2389       if (SB_JOURNAL_LIST_INDEX(p_s_sb) < 0) {
2390         return 0  ;
2391       }
2392       /* starting with oldest, loop until we get to the start */
2393       i = (SB_JOURNAL_LIST_INDEX(p_s_sb) + 1) % JOURNAL_LIST_COUNT ;
2394       while(i != start) {
2395         if (SB_JOURNAL_LIST(p_s_sb)[i].j_len > 0 && ((now - SB_JOURNAL_LIST(p_s_sb)[i].j_timestamp) > JOURNAL_MAX_COMMIT_AGE ||
2396            immediate)) {
2397           /* we have to check again to be sure the current transaction did not change */
2398           if (i != SB_JOURNAL_LIST_INDEX(p_s_sb))  {
2399     	flush_commit_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + i, 1) ;
2400           }
2401         }
2402         i = (i + 1) % JOURNAL_LIST_COUNT ;
2403         count++ ;
2404       }
2405       /* now, check the current transaction.  If there are no writers, and it is too old, finish it, and
2406       ** force the commit blocks to disk
2407       */
2408       if (!immediate && atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) <= 0 &&  
2409          SB_JOURNAL(p_s_sb)->j_trans_start_time > 0 && 
2410          SB_JOURNAL(p_s_sb)->j_len > 0 && 
2411          (now - SB_JOURNAL(p_s_sb)->j_trans_start_time) > JOURNAL_MAX_TRANS_AGE) {
2412         journal_join(&th, p_s_sb, 1) ;
2413         reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
2414         journal_mark_dirty(&th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
2415         do_journal_end(&th, p_s_sb,1, COMMIT_NOW) ;
2416       } else if (immediate) { /* belongs above, but I wanted this to be very explicit as a special case.  If they say to 
2417                                  flush, we must be sure old transactions hit the disk too. */
2418         journal_join(&th, p_s_sb, 1) ;
2419         reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
2420         journal_mark_dirty(&th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
2421         do_journal_end(&th, p_s_sb,1, COMMIT_NOW | WAIT) ;
2422       }
2423       reiserfs_journal_kupdate(p_s_sb) ;
2424       return 0 ;
2425     }
2426     
2427     /*
2428     ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
2429     ** 
2430     ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all 
2431     ** the writers are done.  By the time it wakes up, the transaction it was called has already ended, so it just
2432     ** flushes the commit list and returns 0.
2433     **
2434     ** Won't batch when flush or commit_now is set.  Also won't batch when others are waiting on j_join_wait.
2435     ** 
2436     ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
2437     */
2438     static int check_journal_end(struct reiserfs_transaction_handle *th, struct super_block  * p_s_sb, 
2439                                  unsigned long nblocks, int flags) {
2440     
2441       time_t now ;
2442       int flush = flags & FLUSH_ALL ;
2443       int commit_now = flags & COMMIT_NOW ;
2444       int wait_on_commit = flags & WAIT ;
2445     
2446       if (th->t_trans_id != SB_JOURNAL(p_s_sb)->j_trans_id) {
2447         reiserfs_panic(th->t_super, "journal-1577: handle trans id %d != current trans id %d\n", 
2448                        th->t_trans_id, SB_JOURNAL(p_s_sb)->j_trans_id);
2449       }
2450     
2451       SB_JOURNAL(p_s_sb)->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged) ;
2452       if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) > 0) { /* <= 0 is allowed.  unmounting might not call begin */
2453         atomic_dec(&(SB_JOURNAL(p_s_sb)->j_wcount)) ;
2454       }
2455     
2456       /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released 
2457       ** will be dealt with by next transaction that actually writes something, but should be taken
2458       ** care of in this trans
2459       */
2460       if (SB_JOURNAL(p_s_sb)->j_len == 0) {
2461         int wcount = atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) ;
2462         unlock_journal(p_s_sb) ;
2463         if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_jlock))  > 0 && wcount <= 0) {
2464           atomic_dec(&(SB_JOURNAL(p_s_sb)->j_jlock)) ;
2465           wake_up(&(SB_JOURNAL(p_s_sb)->j_join_wait)) ;
2466         }
2467         return 0 ;
2468       }
2469       /* if wcount > 0, and we are called to with flush or commit_now,
2470       ** we wait on j_join_wait.  We will wake up when the last writer has
2471       ** finished the transaction, and started it on its way to the disk.
2472       ** Then, we flush the commit or journal list, and just return 0 
2473       ** because the rest of journal end was already done for this transaction.
2474       */
2475       if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) > 0) {
2476         if (flush || commit_now) {
2477           int orig_jindex = SB_JOURNAL_LIST_INDEX(p_s_sb) ;
2478           atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 1) ;
2479           if (flush) {
2480             SB_JOURNAL(p_s_sb)->j_next_full_flush = 1 ;
2481           }
2482           unlock_journal(p_s_sb) ;
2483           /* sleep while the current transaction is still j_jlocked */
2484           while(atomic_read(&(SB_JOURNAL(p_s_sb)->j_jlock)) && 
2485                 SB_JOURNAL(p_s_sb)->j_trans_id == th->t_trans_id) {
2486     	sleep_on(&(SB_JOURNAL(p_s_sb)->j_join_wait)) ;
2487           }
2488           if (commit_now) {
2489     	if (wait_on_commit) {
2490     	  flush_commit_list(p_s_sb,  SB_JOURNAL_LIST(p_s_sb) + orig_jindex, 1) ;
2491     	} else {
2492     	  commit_flush_async(p_s_sb, orig_jindex) ; 
2493     	}
2494           }
2495           return 0 ;
2496         } 
2497         unlock_journal(p_s_sb) ;
2498         return 0 ;
2499       }
2500     
2501       /* deal with old transactions where we are the last writers */
2502       now = CURRENT_TIME ;
2503       if ((now - SB_JOURNAL(p_s_sb)->j_trans_start_time) > JOURNAL_MAX_TRANS_AGE) {
2504         commit_now = 1 ;
2505         SB_JOURNAL(p_s_sb)->j_next_async_flush = 1 ;
2506       }
2507       /* don't batch when someone is waiting on j_join_wait */
2508       /* don't batch when syncing the commit or flushing the whole trans */
2509       if (!(SB_JOURNAL(p_s_sb)->j_must_wait > 0) && !(atomic_read(&(SB_JOURNAL(p_s_sb)->j_jlock))) && !flush && !commit_now && 
2510           (SB_JOURNAL(p_s_sb)->j_len < JOURNAL_MAX_BATCH)  && 
2511           SB_JOURNAL(p_s_sb)->j_len_alloc < JOURNAL_MAX_BATCH && SB_JOURNAL(p_s_sb)->j_cnode_free > (JOURNAL_TRANS_MAX * 3)) {
2512         SB_JOURNAL(p_s_sb)->j_bcount++ ;
2513         unlock_journal(p_s_sb) ;
2514         return 0 ;
2515       }
2516     
2517       if (SB_JOURNAL(p_s_sb)->j_start > JOURNAL_BLOCK_COUNT) {
2518         reiserfs_panic(p_s_sb, "journal-003: journal_end: j_start (%d) is too high\n", SB_JOURNAL(p_s_sb)->j_start) ;
2519       }
2520       return 1 ;
2521     }
2522     
2523     /*
2524     ** Does all the work that makes deleting blocks safe.
2525     ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
2526     ** 
2527     ** otherwise:
2528     ** set a bit for the block in the journal bitmap.  That will prevent it from being allocated for unformatted nodes
2529     ** before this transaction has finished.
2530     **
2531     ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.  That will prevent any old transactions with
2532     ** this block from trying to flush to the real location.  Since we aren't removing the cnode from the journal_list_hash,
2533     ** the block can't be reallocated yet.
2534     **
2535     ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
2536     */
2537     int journal_mark_freed(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long blocknr) {
2538       struct reiserfs_journal_cnode *cn = NULL ;
2539       struct buffer_head *bh = NULL ;
2540       struct reiserfs_list_bitmap *jb = NULL ;
2541       int cleaned = 0 ;
2542       
2543       if (reiserfs_dont_log(th->t_super)) {
2544         bh = get_hash_table(p_s_sb->s_dev, blocknr, p_s_sb->s_blocksize) ;
2545         if (bh && buffer_dirty (bh)) {
2546           printk ("journal_mark_freed(dont_log): dirty buffer on hash list: %lx %ld\n", bh->b_state, blocknr);
2547           BUG ();
2548         }
2549         brelse (bh);
2550         return 0 ;
2551       }
2552       bh = get_hash_table(p_s_sb->s_dev, blocknr, p_s_sb->s_blocksize) ;
2553       /* if it is journal new, we just remove it from this transaction */
2554       if (bh && buffer_journal_new(bh)) {
2555         mark_buffer_notjournal_new(bh) ;
2556         clear_prepared_bits(bh) ;
2557         cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned) ;
2558       } else {
2559         /* set the bit for this block in the journal bitmap for this transaction */
2560         jb = SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_list_bitmap ;
2561         if (!jb) {
2562           reiserfs_panic(p_s_sb, "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n") ;
2563         }
2564         set_bit_in_list_bitmap(p_s_sb, blocknr, jb) ;
2565     
2566         /* Note, the entire while loop is not allowed to schedule.  */
2567     
2568         if (bh) {
2569           clear_prepared_bits(bh) ;
2570         }
2571         cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned) ;
2572     
2573         /* find all older transactions with this block, make sure they don't try to write it out */
2574         cn = get_journal_hash_dev(SB_JOURNAL(p_s_sb)->j_list_hash_table, p_s_sb->s_dev, blocknr, p_s_sb->s_blocksize) ;
2575         while (cn) {
2576           if (p_s_sb->s_dev == cn->dev && blocknr == cn->blocknr) {
2577     	set_bit(BLOCK_FREED, &cn->state) ;
2578     	if (cn->bh) {
2579     	  if (!cleaned) {
2580     	    /* remove_from_transaction will brelse the buffer if it was 
2581     	    ** in the current trans
2582     	    */
2583     	    mark_buffer_notjournal_dirty(cn->bh) ;
2584     	    cleaned = 1 ;
2585     	    put_bh(cn->bh) ;
2586     	    if (atomic_read(&(cn->bh->b_count)) < 0) {
2587     	      printk("journal-2138: cn->bh->b_count < 0\n") ;
2588     	    }
2589     	  }
2590     	  if (cn->jlist) { /* since we are clearing the bh, we MUST dec nonzerolen */
2591     	    atomic_dec(&(cn->jlist->j_nonzerolen)) ;
2592     	  }
2593     	  cn->bh = NULL ; 
2594     	} 
2595           }
2596           cn = cn->hnext ;
2597         }
2598       }
2599     
2600       if (bh) {
2601         reiserfs_clean_and_file_buffer(bh) ;
2602         put_bh(bh) ; /* get_hash grabs the buffer */
2603         if (atomic_read(&(bh->b_count)) < 0) {
2604           printk("journal-2165: bh->b_count < 0\n") ;
2605         }
2606       }
2607       return 0 ;
2608     }
2609     
2610     void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb, 
2611                                           struct buffer_head *bh) {
2612       if (reiserfs_dont_log (p_s_sb))
2613         return;
2614     
2615       if (!bh) {
2616         return ;
2617       }
2618       clear_bit(BH_JPrepared, &bh->b_state) ;
2619     }
2620     
2621     extern struct tree_balance *cur_tb ;
2622     /*
2623     ** before we can change a metadata block, we have to make sure it won't
2624     ** be written to disk while we are altering it.  So, we must:
2625     ** clean it
2626     ** wait on it.
2627     ** 
2628     */
2629     void reiserfs_prepare_for_journal(struct super_block *p_s_sb, 
2630                                       struct buffer_head *bh, int wait) {
2631       int retry_count = 0 ;
2632     
2633       if (reiserfs_dont_log (p_s_sb))
2634         return;
2635     
2636       while(!test_bit(BH_JPrepared, &bh->b_state) ||
2637             (wait && buffer_locked(bh))) {
2638         if (buffer_journaled(bh)) {
2639           set_bit(BH_JPrepared, &bh->b_state) ;
2640           return ;
2641         }
2642         set_bit(BH_JPrepared, &bh->b_state) ;
2643         if (wait) {
2644     #ifdef CONFIG_REISERFS_CHECK
2645           if (buffer_locked(bh) && cur_tb != NULL) {
2646     	printk("reiserfs_prepare_for_journal, waiting while do_balance was running\n") ;
2647             BUG() ;
2648           }
2649     #endif
2650           wait_on_buffer(bh) ;
2651         }
2652         retry_count++ ;
2653       }
2654     }
2655     
2656     /*
2657     ** long and ugly.  If flush, will not return until all commit
2658     ** blocks and all real buffers in the trans are on disk.
2659     ** If no_async, won't return until all commit blocks are on disk.
2660     **
2661     ** keep reading, there are comments as you go along
2662     */
2663     static int do_journal_end(struct reiserfs_transaction_handle *th, struct super_block  * p_s_sb, unsigned long nblocks, 
2664     		          int flags) {
2665       struct reiserfs_journal_cnode *cn, *next, *jl_cn; 
2666       struct reiserfs_journal_cnode *last_cn = NULL;
2667       struct reiserfs_journal_desc *desc ; 
2668       struct reiserfs_journal_commit *commit ; 
2669       struct buffer_head *c_bh ; /* commit bh */
2670       struct buffer_head *d_bh ; /* desc bh */
2671       int cur_write_start = 0 ; /* start index of current log write */
2672       int cur_blocks_left = 0 ; /* number of journal blocks left to write */
2673       int old_start ;
2674       int i ;
2675       int jindex ;
2676       int orig_jindex ;
2677       int flush = flags & FLUSH_ALL ;
2678       int commit_now = flags & COMMIT_NOW ;
2679       int wait_on_commit = flags & WAIT ;
2680       struct reiserfs_super_block *rs ; 
2681     
2682       if (reiserfs_dont_log(th->t_super)) {
2683         return 0 ;
2684       }
2685     
2686       lock_journal(p_s_sb) ;
2687       if (SB_JOURNAL(p_s_sb)->j_next_full_flush) {
2688         flags |= FLUSH_ALL ;
2689         flush = 1 ;
2690       }
2691       if (SB_JOURNAL(p_s_sb)->j_next_async_flush) {
2692         flags |= COMMIT_NOW ;
2693         commit_now = 1 ;
2694       }
2695     
2696       /* check_journal_end locks the journal, and unlocks if it does not return 1 
2697       ** it tells us if we should continue with the journal_end, or just return
2698       */
2699       if (!check_journal_end(th, p_s_sb, nblocks, flags)) {
2700         return 0 ;
2701       }
2702     
2703       /* check_journal_end might set these, check again */
2704       if (SB_JOURNAL(p_s_sb)->j_next_full_flush) {
2705         flush = 1 ;
2706       }
2707       if (SB_JOURNAL(p_s_sb)->j_next_async_flush) {
2708         commit_now = 1 ;
2709       }
2710       /*
2711       ** j must wait means we have to flush the log blocks, and the real blocks for
2712       ** this transaction
2713       */
2714       if (SB_JOURNAL(p_s_sb)->j_must_wait > 0) {
2715         flush = 1 ;
2716       }
2717     
2718     #ifdef REISERFS_PREALLOCATE
2719       reiserfs_discard_all_prealloc(th); /* it should not involve new blocks into
2720     				      * the transaction */
2721     #endif
2722       
2723       rs = SB_DISK_SUPER_BLOCK(p_s_sb) ;
2724       /* setup description block */
2725       d_bh = getblk(p_s_sb->s_dev, reiserfs_get_journal_block(p_s_sb) + SB_JOURNAL(p_s_sb)->j_start, p_s_sb->s_blocksize) ; 
2726       mark_buffer_uptodate(d_bh, 1) ;
2727       desc = (struct reiserfs_journal_desc *)(d_bh)->b_data ;
2728       memset(desc, 0, sizeof(struct reiserfs_journal_desc)) ;
2729       memcpy(desc->j_magic, JOURNAL_DESC_MAGIC, 8) ;
2730       desc->j_trans_id = cpu_to_le32(SB_JOURNAL(p_s_sb)->j_trans_id) ;
2731     
2732       /* setup commit block.  Don't write (keep it clean too) this one until after everyone else is written */
2733       c_bh = getblk(p_s_sb->s_dev,  reiserfs_get_journal_block(p_s_sb) + 
2734       				        ((SB_JOURNAL(p_s_sb)->j_start + SB_JOURNAL(p_s_sb)->j_len + 1) % JOURNAL_BLOCK_COUNT), 
2735     					 p_s_sb->s_blocksize) ;
2736       commit = (struct reiserfs_journal_commit *)c_bh->b_data ;
2737       memset(commit, 0, sizeof(struct reiserfs_journal_commit)) ;
2738       commit->j_trans_id = cpu_to_le32(SB_JOURNAL(p_s_sb)->j_trans_id) ;
2739       mark_buffer_uptodate(c_bh, 1) ;
2740     
2741       /* init this journal list */
2742       atomic_set(&(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_older_commits_done), 0) ;
2743       SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_trans_id = SB_JOURNAL(p_s_sb)->j_trans_id ;
2744       SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_timestamp = SB_JOURNAL(p_s_sb)->j_trans_start_time ;
2745       SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_commit_bh = c_bh ;
2746       SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_start = SB_JOURNAL(p_s_sb)->j_start ;
2747       SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_len = SB_JOURNAL(p_s_sb)->j_len ;  
2748       atomic_set(&(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_nonzerolen), SB_JOURNAL(p_s_sb)->j_len) ;
2749       atomic_set(&(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_commit_left), SB_JOURNAL(p_s_sb)->j_len + 2);
2750       SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_realblock = NULL ;
2751       atomic_set(&(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_commit_flushing), 1) ;
2752       atomic_set(&(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_flushing), 1) ;
2753     
2754       /* which is faster, locking/unlocking at the start and end of the for
2755       ** or locking once per iteration around the insert_journal_hash?
2756       ** eitherway, we are write locking insert_journal_hash.  The ENTIRE FOR
2757       ** LOOP MUST not cause schedule to occur.
2758       */
2759     
2760       /* for each real block, add it to the journal list hash,
2761       ** copy into real block index array in the commit or desc block
2762       */
2763       for (i = 0, cn = SB_JOURNAL(p_s_sb)->j_first ; cn ; cn = cn->next, i++) {
2764         if (test_bit(BH_JDirty, &cn->bh->b_state) ) {
2765           jl_cn = get_cnode(p_s_sb) ;
2766           if (!jl_cn) {
2767             reiserfs_panic(p_s_sb, "journal-1676, get_cnode returned NULL\n") ;
2768           }
2769           if (i == 0) {
2770             SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_realblock = jl_cn ;
2771           }
2772           jl_cn->prev = last_cn ;
2773           jl_cn->next = NULL ;
2774           if (last_cn) {
2775             last_cn->next = jl_cn ;
2776           }
2777           last_cn = jl_cn ;
2778           if (cn->bh->b_blocknr >= reiserfs_get_journal_block(p_s_sb) &&
2779               cn->bh->b_blocknr < (reiserfs_get_journal_block(p_s_sb) + JOURNAL_BLOCK_COUNT)) {
2780             reiserfs_panic(p_s_sb, "journal-2332: Trying to log block %lu, which is a log block\n", cn->bh->b_blocknr) ;
2781           }
2782           jl_cn->blocknr = cn->bh->b_blocknr ; 
2783           jl_cn->state = 0 ;
2784           jl_cn->dev = cn->bh->b_dev ; 
2785           jl_cn->bh = cn->bh ;
2786           jl_cn->jlist = SB_JOURNAL_LIST(p_s_sb) + SB_JOURNAL_LIST_INDEX(p_s_sb) ;
2787           insert_journal_hash(SB_JOURNAL(p_s_sb)->j_list_hash_table, jl_cn) ; 
2788           if (i < JOURNAL_TRANS_HALF) {
2789     	desc->j_realblock[i] = cpu_to_le32(cn->bh->b_blocknr) ;
2790           } else {
2791     	commit->j_realblock[i - JOURNAL_TRANS_HALF] = cpu_to_le32(cn->bh->b_blocknr) ;
2792           }
2793         } else {
2794           i-- ;
2795         }
2796       }
2797     
2798       desc->j_len = cpu_to_le32(SB_JOURNAL(p_s_sb)->j_len)  ;
2799       desc->j_mount_id = cpu_to_le32(SB_JOURNAL(p_s_sb)->j_mount_id) ;
2800       desc->j_trans_id = cpu_to_le32(SB_JOURNAL(p_s_sb)->j_trans_id) ;
2801       commit->j_len = cpu_to_le32(SB_JOURNAL(p_s_sb)->j_len) ;
2802     
2803       /* special check in case all buffers in the journal were marked for not logging */
2804       if (SB_JOURNAL(p_s_sb)->j_len == 0) {
2805         brelse(d_bh) ;
2806         brelse(c_bh) ;
2807         unlock_journal(p_s_sb) ;
2808     printk("journal-2020: do_journal_end: BAD desc->j_len is ZERO\n") ;
2809         atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 0) ;
2810         wake_up(&(SB_JOURNAL(p_s_sb)->j_join_wait)) ;
2811         return 0 ;
2812       }
2813     
2814       /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
2815       cur_write_start = SB_JOURNAL(p_s_sb)->j_start ;
2816       cur_blocks_left = SB_JOURNAL(p_s_sb)->j_len  ;
2817       cn = SB_JOURNAL(p_s_sb)->j_first ;
2818       jindex = 1 ; /* start at one so we don't get the desc again */
2819       while(cur_blocks_left > 0) {
2820         /* copy all the real blocks into log area.  dirty log blocks */
2821         if (test_bit(BH_JDirty, &cn->bh->b_state)) {
2822           struct buffer_head *tmp_bh ;
2823           tmp_bh = getblk(p_s_sb->s_dev, reiserfs_get_journal_block(p_s_sb) + 
2824     		     ((cur_write_start + jindex) % JOURNAL_BLOCK_COUNT), 
2825     				       p_s_sb->s_blocksize) ;
2826           mark_buffer_uptodate(tmp_bh, 1) ;
2827           memcpy(tmp_bh->b_data, cn->bh->b_data, cn->bh->b_size) ;  
2828           jindex++ ;
2829         } else {
2830           /* JDirty cleared sometime during transaction.  don't log this one */
2831           printk("journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!\n") ;
2832         }
2833         cn = cn->next ;
2834         cur_blocks_left-- ;
2835       }
2836     
2837       /* we are done  with both the c_bh and d_bh, but
2838       ** c_bh must be written after all other commit blocks,
2839       ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
2840       */
2841     
2842       /* now loop through and mark all buffers from this transaction as JDirty_wait
2843       ** clear the JDirty bit, clear BH_JNew too.  
2844       ** if they weren't JDirty, they weren't logged, just relse them and move on
2845       */
2846       cn = SB_JOURNAL(p_s_sb)->j_first ; 
2847       while(cn) {
2848         clear_bit(BH_JNew, &(cn->bh->b_state)) ;
2849         if (test_bit(BH_JDirty, &(cn->bh->b_state))) {
2850           set_bit(BH_JDirty_wait, &(cn->bh->b_state)) ; 
2851           clear_bit(BH_JDirty, &(cn->bh->b_state)) ;
2852         } else {
2853           brelse(cn->bh) ;
2854         }
2855         next = cn->next ;
2856         free_cnode(p_s_sb, cn) ;
2857         cn = next ;
2858       }
2859     
2860       /* unlock the journal list for committing and flushing */
2861       atomic_set(&(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_commit_flushing), 0) ;
2862       atomic_set(&(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_flushing), 0) ;
2863     
2864       orig_jindex = SB_JOURNAL_LIST_INDEX(p_s_sb) ;
2865       jindex = (SB_JOURNAL_LIST_INDEX(p_s_sb) + 1) % JOURNAL_LIST_COUNT ; 
2866       SB_JOURNAL_LIST_INDEX(p_s_sb) = jindex ;
2867     
2868       /* write any buffers that must hit disk before this commit is done */
2869       fsync_inode_buffers(&(SB_JOURNAL(p_s_sb)->j_dummy_inode)) ;
2870     
2871       /* honor the flush and async wishes from the caller */
2872       if (flush) {
2873       
2874         flush_commit_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + orig_jindex, 1) ;
2875         flush_journal_list(p_s_sb,  SB_JOURNAL_LIST(p_s_sb) + orig_jindex , 1) ;  
2876       } else if (commit_now) {
2877         if (wait_on_commit) {
2878           flush_commit_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + orig_jindex, 1) ;
2879         } else {
2880           commit_flush_async(p_s_sb, orig_jindex) ; 
2881         }
2882       }
2883     
2884       /* reset journal values for the next transaction */
2885       old_start = SB_JOURNAL(p_s_sb)->j_start ;
2886       SB_JOURNAL(p_s_sb)->j_start = (SB_JOURNAL(p_s_sb)->j_start + SB_JOURNAL(p_s_sb)->j_len + 2) % JOURNAL_BLOCK_COUNT;
2887       atomic_set(&(SB_JOURNAL(p_s_sb)->j_wcount), 0) ;
2888       SB_JOURNAL(p_s_sb)->j_bcount = 0 ;
2889       SB_JOURNAL(p_s_sb)->j_last = NULL ;
2890       SB_JOURNAL(p_s_sb)->j_first = NULL ;
2891       SB_JOURNAL(p_s_sb)->j_len = 0 ;
2892       SB_JOURNAL(p_s_sb)->j_trans_start_time = 0 ;
2893       SB_JOURNAL(p_s_sb)->j_trans_id++ ;
2894       SB_JOURNAL(p_s_sb)->j_must_wait = 0 ;
2895       SB_JOURNAL(p_s_sb)->j_len_alloc = 0 ;
2896       SB_JOURNAL(p_s_sb)->j_next_full_flush = 0 ;
2897       SB_JOURNAL(p_s_sb)->j_next_async_flush = 0 ;
2898       init_journal_hash(p_s_sb) ; 
2899     
2900       /* if the next transaction has any chance of wrapping, flush 
2901       ** transactions that might get overwritten.  If any journal lists are very 
2902       ** old flush them as well.  
2903       */
2904       for (i = 0 ; i < JOURNAL_LIST_COUNT ; i++) {
2905         jindex = i ;
2906         if (SB_JOURNAL_LIST(p_s_sb)[jindex].j_len > 0 && SB_JOURNAL(p_s_sb)->j_start <= SB_JOURNAL_LIST(p_s_sb)[jindex].j_start) {
2907           if ((SB_JOURNAL(p_s_sb)->j_start + JOURNAL_TRANS_MAX + 1) >= SB_JOURNAL_LIST(p_s_sb)[jindex].j_start) {
2908     	flush_journal_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + jindex, 1) ; 
2909           }
2910         } else if (SB_JOURNAL_LIST(p_s_sb)[jindex].j_len > 0 && 
2911                   (SB_JOURNAL(p_s_sb)->j_start + JOURNAL_TRANS_MAX + 1) > JOURNAL_BLOCK_COUNT) {
2912           if (((SB_JOURNAL(p_s_sb)->j_start + JOURNAL_TRANS_MAX + 1) % JOURNAL_BLOCK_COUNT) >= 
2913                 SB_JOURNAL_LIST(p_s_sb)[jindex].j_start) {
2914     	flush_journal_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + jindex, 1 ) ; 
2915           }
2916         } 
2917         /* this check should always be run, to send old lists to disk */
2918         if (SB_JOURNAL_LIST(p_s_sb)[jindex].j_len > 0 && 
2919                   SB_JOURNAL_LIST(p_s_sb)[jindex].j_timestamp < 
2920     	      (CURRENT_TIME - (JOURNAL_MAX_TRANS_AGE * 4))) {
2921     	flush_journal_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + jindex, 1 ) ; 
2922         }
2923       }
2924     
2925       /* if the next journal_list is still in use, flush it */
2926       if (SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_len != 0) {
2927         flush_journal_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + SB_JOURNAL_LIST_INDEX(p_s_sb), 1) ; 
2928       }
2929     
2930       /* we don't want anyone flushing the new transaction's list */
2931       atomic_set(&(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_commit_flushing), 1) ;
2932       atomic_set(&(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_flushing), 1) ;
2933       SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_list_bitmap = get_list_bitmap(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + 
2934     											 SB_JOURNAL_LIST_INDEX(p_s_sb)) ;
2935     
2936       if (!(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_list_bitmap)) {
2937         reiserfs_panic(p_s_sb, "journal-1996: do_journal_end, could not get a list bitmap\n") ;
2938       }
2939       unlock_journal(p_s_sb) ;
2940       atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 0) ;
2941       /* wake up any body waiting to join. */
2942       wake_up(&(SB_JOURNAL(p_s_sb)->j_join_wait)) ;
2943       return 0 ;
2944     }
2945     
2946     
2947     
2948