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