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