90622200b39c0622e0f159d423c929a036d76257
[linux-2.6.git] / fs / reiserfs / journal.c
1 /*
2 ** Write ahead logging implementation copyright Chris Mason 2000
3 **
4 ** The background commits make this code very interelated, and
5 ** overly complex.  I need to rethink things a bit....The major players:
6 **
7 ** journal_begin -- call with the number of blocks you expect to log.
8 **                  If the current transaction is too
9 **                  old, it will block until the current transaction is
10 **                  finished, and then start a new one.
11 **                  Usually, your transaction will get joined in with
12 **                  previous ones for speed.
13 **
14 ** journal_join  -- same as journal_begin, but won't block on the current
15 **                  transaction regardless of age.  Don't ever call
16 **                  this.  Ever.  There are only two places it should be
17 **                  called from, and they are both inside this file.
18 **
19 ** journal_mark_dirty -- adds blocks into this transaction.  clears any flags
20 **                       that might make them get sent to disk
21 **                       and then marks them BH_JDirty.  Puts the buffer head
22 **                       into the current transaction hash.
23 **
24 ** journal_end -- if the current transaction is batchable, it does nothing
25 **                   otherwise, it could do an async/synchronous commit, or
26 **                   a full flush of all log and real blocks in the
27 **                   transaction.
28 **
29 ** flush_old_commits -- if the current transaction is too old, it is ended and
30 **                      commit blocks are sent to disk.  Forces commit blocks
31 **                      to disk for all backgrounded commits that have been
32 **                      around too long.
33 **                   -- Note, if you call this as an immediate flush from
34 **                      from within kupdate, it will ignore the immediate flag
35 */
36
37 #include <linux/time.h>
38 #include <linux/semaphore.h>
39 #include <linux/vmalloc.h>
40 #include <linux/reiserfs_fs.h>
41 #include <linux/kernel.h>
42 #include <linux/errno.h>
43 #include <linux/fcntl.h>
44 #include <linux/stat.h>
45 #include <linux/string.h>
46 #include <linux/smp_lock.h>
47 #include <linux/buffer_head.h>
48 #include <linux/workqueue.h>
49 #include <linux/writeback.h>
50 #include <linux/blkdev.h>
51 #include <linux/backing-dev.h>
52 #include <linux/uaccess.h>
53
54 #include <asm/system.h>
55
56 /* gets a struct reiserfs_journal_list * from a list head */
57 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
58                                j_list))
59 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
60                                j_working_list))
61
62 /* the number of mounted filesystems.  This is used to decide when to
63 ** start and kill the commit workqueue
64 */
65 static int reiserfs_mounted_fs_count;
66
67 static struct workqueue_struct *commit_wq;
68
69 #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
70                                    structs at 4k */
71 #define BUFNR 64                /*read ahead */
72
73 /* cnode stat bits.  Move these into reiserfs_fs.h */
74
75 #define BLOCK_FREED 2           /* this block was freed, and can't be written.  */
76 #define BLOCK_FREED_HOLDER 3    /* this block was freed during this transaction, and can't be written */
77
78 #define BLOCK_NEEDS_FLUSH 4     /* used in flush_journal_list */
79 #define BLOCK_DIRTIED 5
80
81 /* journal list state bits */
82 #define LIST_TOUCHED 1
83 #define LIST_DIRTY   2
84 #define LIST_COMMIT_PENDING  4  /* someone will commit this list */
85
86 /* flags for do_journal_end */
87 #define FLUSH_ALL   1           /* flush commit and real blocks */
88 #define COMMIT_NOW  2           /* end and commit this transaction */
89 #define WAIT        4           /* wait for the log blocks to hit the disk */
90
91 static int do_journal_end(struct reiserfs_transaction_handle *,
92                           struct super_block *, unsigned long nblocks,
93                           int flags);
94 static int flush_journal_list(struct super_block *s,
95                               struct reiserfs_journal_list *jl, int flushall);
96 static int flush_commit_list(struct super_block *s,
97                              struct reiserfs_journal_list *jl, int flushall);
98 static int can_dirty(struct reiserfs_journal_cnode *cn);
99 static int journal_join(struct reiserfs_transaction_handle *th,
100                         struct super_block *sb, unsigned long nblocks);
101 static int release_journal_dev(struct super_block *super,
102                                struct reiserfs_journal *journal);
103 static int dirty_one_transaction(struct super_block *s,
104                                  struct reiserfs_journal_list *jl);
105 static void flush_async_commits(struct work_struct *work);
106 static void queue_log_writer(struct super_block *s);
107
108 /* values for join in do_journal_begin_r */
109 enum {
110         JBEGIN_REG = 0,         /* regular journal begin */
111         JBEGIN_JOIN = 1,        /* join the running transaction if at all possible */
112         JBEGIN_ABORT = 2,       /* called from cleanup code, ignores aborted flag */
113 };
114
115 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
116                               struct super_block *sb,
117                               unsigned long nblocks, int join);
118
119 static void init_journal_hash(struct super_block *sb)
120 {
121         struct reiserfs_journal *journal = SB_JOURNAL(sb);
122         memset(journal->j_hash_table, 0,
123                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
124 }
125
126 /*
127 ** clears BH_Dirty and sticks the buffer on the clean list.  Called because I can't allow refile_buffer to
128 ** make schedule happen after I've freed a block.  Look at remove_from_transaction and journal_mark_freed for
129 ** more details.
130 */
131 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
132 {
133         if (bh) {
134                 clear_buffer_dirty(bh);
135                 clear_buffer_journal_test(bh);
136         }
137         return 0;
138 }
139
140 static void disable_barrier(struct super_block *s)
141 {
142         REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH);
143         printk("reiserfs: disabling flush barriers on %s\n",
144                reiserfs_bdevname(s));
145 }
146
147 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
148                                                          *sb)
149 {
150         struct reiserfs_bitmap_node *bn;
151         static int id;
152
153         bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
154         if (!bn) {
155                 return NULL;
156         }
157         bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
158         if (!bn->data) {
159                 kfree(bn);
160                 return NULL;
161         }
162         bn->id = id++;
163         INIT_LIST_HEAD(&bn->list);
164         return bn;
165 }
166
167 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
168 {
169         struct reiserfs_journal *journal = SB_JOURNAL(sb);
170         struct reiserfs_bitmap_node *bn = NULL;
171         struct list_head *entry = journal->j_bitmap_nodes.next;
172
173         journal->j_used_bitmap_nodes++;
174       repeat:
175
176         if (entry != &journal->j_bitmap_nodes) {
177                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
178                 list_del(entry);
179                 memset(bn->data, 0, sb->s_blocksize);
180                 journal->j_free_bitmap_nodes--;
181                 return bn;
182         }
183         bn = allocate_bitmap_node(sb);
184         if (!bn) {
185                 yield();
186                 goto repeat;
187         }
188         return bn;
189 }
190 static inline void free_bitmap_node(struct super_block *sb,
191                                     struct reiserfs_bitmap_node *bn)
192 {
193         struct reiserfs_journal *journal = SB_JOURNAL(sb);
194         journal->j_used_bitmap_nodes--;
195         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
196                 kfree(bn->data);
197                 kfree(bn);
198         } else {
199                 list_add(&bn->list, &journal->j_bitmap_nodes);
200                 journal->j_free_bitmap_nodes++;
201         }
202 }
203
204 static void allocate_bitmap_nodes(struct super_block *sb)
205 {
206         int i;
207         struct reiserfs_journal *journal = SB_JOURNAL(sb);
208         struct reiserfs_bitmap_node *bn = NULL;
209         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
210                 bn = allocate_bitmap_node(sb);
211                 if (bn) {
212                         list_add(&bn->list, &journal->j_bitmap_nodes);
213                         journal->j_free_bitmap_nodes++;
214                 } else {
215                         break;  /* this is ok, we'll try again when more are needed */
216                 }
217         }
218 }
219
220 static int set_bit_in_list_bitmap(struct super_block *sb,
221                                   b_blocknr_t block,
222                                   struct reiserfs_list_bitmap *jb)
223 {
224         unsigned int bmap_nr = block / (sb->s_blocksize << 3);
225         unsigned int bit_nr = block % (sb->s_blocksize << 3);
226
227         if (!jb->bitmaps[bmap_nr]) {
228                 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
229         }
230         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
231         return 0;
232 }
233
234 static void cleanup_bitmap_list(struct super_block *sb,
235                                 struct reiserfs_list_bitmap *jb)
236 {
237         int i;
238         if (jb->bitmaps == NULL)
239                 return;
240
241         for (i = 0; i < reiserfs_bmap_count(sb); i++) {
242                 if (jb->bitmaps[i]) {
243                         free_bitmap_node(sb, jb->bitmaps[i]);
244                         jb->bitmaps[i] = NULL;
245                 }
246         }
247 }
248
249 /*
250 ** only call this on FS unmount.
251 */
252 static int free_list_bitmaps(struct super_block *sb,
253                              struct reiserfs_list_bitmap *jb_array)
254 {
255         int i;
256         struct reiserfs_list_bitmap *jb;
257         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
258                 jb = jb_array + i;
259                 jb->journal_list = NULL;
260                 cleanup_bitmap_list(sb, jb);
261                 vfree(jb->bitmaps);
262                 jb->bitmaps = NULL;
263         }
264         return 0;
265 }
266
267 static int free_bitmap_nodes(struct super_block *sb)
268 {
269         struct reiserfs_journal *journal = SB_JOURNAL(sb);
270         struct list_head *next = journal->j_bitmap_nodes.next;
271         struct reiserfs_bitmap_node *bn;
272
273         while (next != &journal->j_bitmap_nodes) {
274                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
275                 list_del(next);
276                 kfree(bn->data);
277                 kfree(bn);
278                 next = journal->j_bitmap_nodes.next;
279                 journal->j_free_bitmap_nodes--;
280         }
281
282         return 0;
283 }
284
285 /*
286 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
287 ** jb_array is the array to be filled in.
288 */
289 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
290                                    struct reiserfs_list_bitmap *jb_array,
291                                    unsigned int bmap_nr)
292 {
293         int i;
294         int failed = 0;
295         struct reiserfs_list_bitmap *jb;
296         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
297
298         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
299                 jb = jb_array + i;
300                 jb->journal_list = NULL;
301                 jb->bitmaps = vmalloc(mem);
302                 if (!jb->bitmaps) {
303                         reiserfs_warning(sb, "clm-2000", "unable to "
304                                          "allocate bitmaps for journal lists");
305                         failed = 1;
306                         break;
307                 }
308                 memset(jb->bitmaps, 0, mem);
309         }
310         if (failed) {
311                 free_list_bitmaps(sb, jb_array);
312                 return -1;
313         }
314         return 0;
315 }
316
317 /*
318 ** find an available list bitmap.  If you can't find one, flush a commit list
319 ** and try again
320 */
321 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
322                                                     struct reiserfs_journal_list
323                                                     *jl)
324 {
325         int i, j;
326         struct reiserfs_journal *journal = SB_JOURNAL(sb);
327         struct reiserfs_list_bitmap *jb = NULL;
328
329         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
330                 i = journal->j_list_bitmap_index;
331                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
332                 jb = journal->j_list_bitmap + i;
333                 if (journal->j_list_bitmap[i].journal_list) {
334                         flush_commit_list(sb,
335                                           journal->j_list_bitmap[i].
336                                           journal_list, 1);
337                         if (!journal->j_list_bitmap[i].journal_list) {
338                                 break;
339                         }
340                 } else {
341                         break;
342                 }
343         }
344         if (jb->journal_list) { /* double check to make sure if flushed correctly */
345                 return NULL;
346         }
347         jb->journal_list = jl;
348         return jb;
349 }
350
351 /*
352 ** allocates a new chunk of X nodes, and links them all together as a list.
353 ** Uses the cnode->next and cnode->prev pointers
354 ** returns NULL on failure
355 */
356 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
357 {
358         struct reiserfs_journal_cnode *head;
359         int i;
360         if (num_cnodes <= 0) {
361                 return NULL;
362         }
363         head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
364         if (!head) {
365                 return NULL;
366         }
367         memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode));
368         head[0].prev = NULL;
369         head[0].next = head + 1;
370         for (i = 1; i < num_cnodes; i++) {
371                 head[i].prev = head + (i - 1);
372                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
373         }
374         head[num_cnodes - 1].next = NULL;
375         return head;
376 }
377
378 /*
379 ** pulls a cnode off the free list, or returns NULL on failure
380 */
381 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
382 {
383         struct reiserfs_journal_cnode *cn;
384         struct reiserfs_journal *journal = SB_JOURNAL(sb);
385
386         reiserfs_check_lock_depth(sb, "get_cnode");
387
388         if (journal->j_cnode_free <= 0) {
389                 return NULL;
390         }
391         journal->j_cnode_used++;
392         journal->j_cnode_free--;
393         cn = journal->j_cnode_free_list;
394         if (!cn) {
395                 return cn;
396         }
397         if (cn->next) {
398                 cn->next->prev = NULL;
399         }
400         journal->j_cnode_free_list = cn->next;
401         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
402         return cn;
403 }
404
405 /*
406 ** returns a cnode to the free list
407 */
408 static void free_cnode(struct super_block *sb,
409                        struct reiserfs_journal_cnode *cn)
410 {
411         struct reiserfs_journal *journal = SB_JOURNAL(sb);
412
413         reiserfs_check_lock_depth(sb, "free_cnode");
414
415         journal->j_cnode_used--;
416         journal->j_cnode_free++;
417         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
418         cn->next = journal->j_cnode_free_list;
419         if (journal->j_cnode_free_list) {
420                 journal->j_cnode_free_list->prev = cn;
421         }
422         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
423         journal->j_cnode_free_list = cn;
424 }
425
426 static void clear_prepared_bits(struct buffer_head *bh)
427 {
428         clear_buffer_journal_prepared(bh);
429         clear_buffer_journal_restore_dirty(bh);
430 }
431
432 /* utility function to force a BUG if it is called without the big
433 ** kernel lock held.  caller is the string printed just before calling BUG()
434 */
435 void reiserfs_check_lock_depth(struct super_block *sb, char *caller)
436 {
437 #ifdef CONFIG_SMP
438         if (current->lock_depth < 0) {
439                 reiserfs_panic(sb, "journal-1", "%s called without kernel "
440                                "lock held", caller);
441         }
442 #else
443         ;
444 #endif
445 }
446
447 /* return a cnode with same dev, block number and size in table, or null if not found */
448 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
449                                                                   super_block
450                                                                   *sb,
451                                                                   struct
452                                                                   reiserfs_journal_cnode
453                                                                   **table,
454                                                                   long bl)
455 {
456         struct reiserfs_journal_cnode *cn;
457         cn = journal_hash(table, sb, bl);
458         while (cn) {
459                 if (cn->blocknr == bl && cn->sb == sb)
460                         return cn;
461                 cn = cn->hnext;
462         }
463         return (struct reiserfs_journal_cnode *)0;
464 }
465
466 /*
467 ** this actually means 'can this block be reallocated yet?'.  If you set search_all, a block can only be allocated
468 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
469 ** being overwritten by a replay after crashing.
470 **
471 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction.  Since deleting
472 ** a block removes it from the current transaction, this case should never happen.  If you don't set search_all, make
473 ** sure you never write the block without logging it.
474 **
475 ** next_zero_bit is a suggestion about the next block to try for find_forward.
476 ** when bl is rejected because it is set in a journal list bitmap, we search
477 ** for the next zero bit in the bitmap that rejected bl.  Then, we return that
478 ** through next_zero_bit for find_forward to try.
479 **
480 ** Just because we return something in next_zero_bit does not mean we won't
481 ** reject it on the next call to reiserfs_in_journal
482 **
483 */
484 int reiserfs_in_journal(struct super_block *sb,
485                         unsigned int bmap_nr, int bit_nr, int search_all,
486                         b_blocknr_t * next_zero_bit)
487 {
488         struct reiserfs_journal *journal = SB_JOURNAL(sb);
489         struct reiserfs_journal_cnode *cn;
490         struct reiserfs_list_bitmap *jb;
491         int i;
492         unsigned long bl;
493
494         *next_zero_bit = 0;     /* always start this at zero. */
495
496         PROC_INFO_INC(sb, journal.in_journal);
497         /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
498          ** if we crash before the transaction that freed it commits,  this transaction won't
499          ** have committed either, and the block will never be written
500          */
501         if (search_all) {
502                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
503                         PROC_INFO_INC(sb, journal.in_journal_bitmap);
504                         jb = journal->j_list_bitmap + i;
505                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
506                             test_bit(bit_nr,
507                                      (unsigned long *)jb->bitmaps[bmap_nr]->
508                                      data)) {
509                                 *next_zero_bit =
510                                     find_next_zero_bit((unsigned long *)
511                                                        (jb->bitmaps[bmap_nr]->
512                                                         data),
513                                                        sb->s_blocksize << 3,
514                                                        bit_nr + 1);
515                                 return 1;
516                         }
517                 }
518         }
519
520         bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
521         /* is it in any old transactions? */
522         if (search_all
523             && (cn =
524                 get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
525                 return 1;
526         }
527
528         /* is it in the current transaction.  This should never happen */
529         if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
530                 BUG();
531                 return 1;
532         }
533
534         PROC_INFO_INC(sb, journal.in_journal_reusable);
535         /* safe for reuse */
536         return 0;
537 }
538
539 /* insert cn into table
540 */
541 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
542                                        struct reiserfs_journal_cnode *cn)
543 {
544         struct reiserfs_journal_cnode *cn_orig;
545
546         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
547         cn->hnext = cn_orig;
548         cn->hprev = NULL;
549         if (cn_orig) {
550                 cn_orig->hprev = cn;
551         }
552         journal_hash(table, cn->sb, cn->blocknr) = cn;
553 }
554
555 /* lock the current transaction */
556 static inline void lock_journal(struct super_block *sb)
557 {
558         PROC_INFO_INC(sb, journal.lock_journal);
559         mutex_lock(&SB_JOURNAL(sb)->j_mutex);
560 }
561
562 /* unlock the current transaction */
563 static inline void unlock_journal(struct super_block *sb)
564 {
565         mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
566 }
567
568 static inline void get_journal_list(struct reiserfs_journal_list *jl)
569 {
570         jl->j_refcount++;
571 }
572
573 static inline void put_journal_list(struct super_block *s,
574                                     struct reiserfs_journal_list *jl)
575 {
576         if (jl->j_refcount < 1) {
577                 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
578                                jl->j_trans_id, jl->j_refcount);
579         }
580         if (--jl->j_refcount == 0)
581                 kfree(jl);
582 }
583
584 /*
585 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
586 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
587 ** transaction.
588 */
589 static void cleanup_freed_for_journal_list(struct super_block *sb,
590                                            struct reiserfs_journal_list *jl)
591 {
592
593         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
594         if (jb) {
595                 cleanup_bitmap_list(sb, jb);
596         }
597         jl->j_list_bitmap->journal_list = NULL;
598         jl->j_list_bitmap = NULL;
599 }
600
601 static int journal_list_still_alive(struct super_block *s,
602                                     unsigned int trans_id)
603 {
604         struct reiserfs_journal *journal = SB_JOURNAL(s);
605         struct list_head *entry = &journal->j_journal_list;
606         struct reiserfs_journal_list *jl;
607
608         if (!list_empty(entry)) {
609                 jl = JOURNAL_LIST_ENTRY(entry->next);
610                 if (jl->j_trans_id <= trans_id) {
611                         return 1;
612                 }
613         }
614         return 0;
615 }
616
617 /*
618  * If page->mapping was null, we failed to truncate this page for
619  * some reason.  Most likely because it was truncated after being
620  * logged via data=journal.
621  *
622  * This does a check to see if the buffer belongs to one of these
623  * lost pages before doing the final put_bh.  If page->mapping was
624  * null, it tries to free buffers on the page, which should make the
625  * final page_cache_release drop the page from the lru.
626  */
627 static void release_buffer_page(struct buffer_head *bh)
628 {
629         struct page *page = bh->b_page;
630         if (!page->mapping && trylock_page(page)) {
631                 page_cache_get(page);
632                 put_bh(bh);
633                 if (!page->mapping)
634                         try_to_free_buffers(page);
635                 unlock_page(page);
636                 page_cache_release(page);
637         } else {
638                 put_bh(bh);
639         }
640 }
641
642 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
643 {
644         char b[BDEVNAME_SIZE];
645
646         if (buffer_journaled(bh)) {
647                 reiserfs_warning(NULL, "clm-2084",
648                                  "pinned buffer %lu:%s sent to disk",
649                                  bh->b_blocknr, bdevname(bh->b_bdev, b));
650         }
651         if (uptodate)
652                 set_buffer_uptodate(bh);
653         else
654                 clear_buffer_uptodate(bh);
655
656         unlock_buffer(bh);
657         release_buffer_page(bh);
658 }
659
660 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
661 {
662         if (uptodate)
663                 set_buffer_uptodate(bh);
664         else
665                 clear_buffer_uptodate(bh);
666         unlock_buffer(bh);
667         put_bh(bh);
668 }
669
670 static void submit_logged_buffer(struct buffer_head *bh)
671 {
672         get_bh(bh);
673         bh->b_end_io = reiserfs_end_buffer_io_sync;
674         clear_buffer_journal_new(bh);
675         clear_buffer_dirty(bh);
676         if (!test_clear_buffer_journal_test(bh))
677                 BUG();
678         if (!buffer_uptodate(bh))
679                 BUG();
680         submit_bh(WRITE, bh);
681 }
682
683 static void submit_ordered_buffer(struct buffer_head *bh)
684 {
685         get_bh(bh);
686         bh->b_end_io = reiserfs_end_ordered_io;
687         clear_buffer_dirty(bh);
688         if (!buffer_uptodate(bh))
689                 BUG();
690         submit_bh(WRITE, bh);
691 }
692
693 static int submit_barrier_buffer(struct buffer_head *bh)
694 {
695         get_bh(bh);
696         bh->b_end_io = reiserfs_end_ordered_io;
697         clear_buffer_dirty(bh);
698         if (!buffer_uptodate(bh))
699                 BUG();
700         return submit_bh(WRITE_BARRIER, bh);
701 }
702
703 static void check_barrier_completion(struct super_block *s,
704                                      struct buffer_head *bh)
705 {
706         if (buffer_eopnotsupp(bh)) {
707                 clear_buffer_eopnotsupp(bh);
708                 disable_barrier(s);
709                 set_buffer_uptodate(bh);
710                 set_buffer_dirty(bh);
711                 sync_dirty_buffer(bh);
712         }
713 }
714
715 #define CHUNK_SIZE 32
716 struct buffer_chunk {
717         struct buffer_head *bh[CHUNK_SIZE];
718         int nr;
719 };
720
721 static void write_chunk(struct buffer_chunk *chunk)
722 {
723         int i;
724         get_fs_excl();
725         for (i = 0; i < chunk->nr; i++) {
726                 submit_logged_buffer(chunk->bh[i]);
727         }
728         chunk->nr = 0;
729         put_fs_excl();
730 }
731
732 static void write_ordered_chunk(struct buffer_chunk *chunk)
733 {
734         int i;
735         get_fs_excl();
736         for (i = 0; i < chunk->nr; i++) {
737                 submit_ordered_buffer(chunk->bh[i]);
738         }
739         chunk->nr = 0;
740         put_fs_excl();
741 }
742
743 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
744                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
745 {
746         int ret = 0;
747         BUG_ON(chunk->nr >= CHUNK_SIZE);
748         chunk->bh[chunk->nr++] = bh;
749         if (chunk->nr >= CHUNK_SIZE) {
750                 ret = 1;
751                 if (lock)
752                         spin_unlock(lock);
753                 fn(chunk);
754                 if (lock)
755                         spin_lock(lock);
756         }
757         return ret;
758 }
759
760 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
761 static struct reiserfs_jh *alloc_jh(void)
762 {
763         struct reiserfs_jh *jh;
764         while (1) {
765                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
766                 if (jh) {
767                         atomic_inc(&nr_reiserfs_jh);
768                         return jh;
769                 }
770                 yield();
771         }
772 }
773
774 /*
775  * we want to free the jh when the buffer has been written
776  * and waited on
777  */
778 void reiserfs_free_jh(struct buffer_head *bh)
779 {
780         struct reiserfs_jh *jh;
781
782         jh = bh->b_private;
783         if (jh) {
784                 bh->b_private = NULL;
785                 jh->bh = NULL;
786                 list_del_init(&jh->list);
787                 kfree(jh);
788                 if (atomic_read(&nr_reiserfs_jh) <= 0)
789                         BUG();
790                 atomic_dec(&nr_reiserfs_jh);
791                 put_bh(bh);
792         }
793 }
794
795 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
796                            int tail)
797 {
798         struct reiserfs_jh *jh;
799
800         if (bh->b_private) {
801                 spin_lock(&j->j_dirty_buffers_lock);
802                 if (!bh->b_private) {
803                         spin_unlock(&j->j_dirty_buffers_lock);
804                         goto no_jh;
805                 }
806                 jh = bh->b_private;
807                 list_del_init(&jh->list);
808         } else {
809               no_jh:
810                 get_bh(bh);
811                 jh = alloc_jh();
812                 spin_lock(&j->j_dirty_buffers_lock);
813                 /* buffer must be locked for __add_jh, should be able to have
814                  * two adds at the same time
815                  */
816                 BUG_ON(bh->b_private);
817                 jh->bh = bh;
818                 bh->b_private = jh;
819         }
820         jh->jl = j->j_current_jl;
821         if (tail)
822                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
823         else {
824                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
825         }
826         spin_unlock(&j->j_dirty_buffers_lock);
827         return 0;
828 }
829
830 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
831 {
832         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
833 }
834 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
835 {
836         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
837 }
838
839 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
840 static int write_ordered_buffers(spinlock_t * lock,
841                                  struct reiserfs_journal *j,
842                                  struct reiserfs_journal_list *jl,
843                                  struct list_head *list)
844 {
845         struct buffer_head *bh;
846         struct reiserfs_jh *jh;
847         int ret = j->j_errno;
848         struct buffer_chunk chunk;
849         struct list_head tmp;
850         INIT_LIST_HEAD(&tmp);
851
852         chunk.nr = 0;
853         spin_lock(lock);
854         while (!list_empty(list)) {
855                 jh = JH_ENTRY(list->next);
856                 bh = jh->bh;
857                 get_bh(bh);
858                 if (!trylock_buffer(bh)) {
859                         if (!buffer_dirty(bh)) {
860                                 list_move(&jh->list, &tmp);
861                                 goto loop_next;
862                         }
863                         spin_unlock(lock);
864                         if (chunk.nr)
865                                 write_ordered_chunk(&chunk);
866                         wait_on_buffer(bh);
867                         cond_resched();
868                         spin_lock(lock);
869                         goto loop_next;
870                 }
871                 /* in theory, dirty non-uptodate buffers should never get here,
872                  * but the upper layer io error paths still have a few quirks.
873                  * Handle them here as gracefully as we can
874                  */
875                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
876                         clear_buffer_dirty(bh);
877                         ret = -EIO;
878                 }
879                 if (buffer_dirty(bh)) {
880                         list_move(&jh->list, &tmp);
881                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
882                 } else {
883                         reiserfs_free_jh(bh);
884                         unlock_buffer(bh);
885                 }
886               loop_next:
887                 put_bh(bh);
888                 cond_resched_lock(lock);
889         }
890         if (chunk.nr) {
891                 spin_unlock(lock);
892                 write_ordered_chunk(&chunk);
893                 spin_lock(lock);
894         }
895         while (!list_empty(&tmp)) {
896                 jh = JH_ENTRY(tmp.prev);
897                 bh = jh->bh;
898                 get_bh(bh);
899                 reiserfs_free_jh(bh);
900
901                 if (buffer_locked(bh)) {
902                         spin_unlock(lock);
903                         wait_on_buffer(bh);
904                         spin_lock(lock);
905                 }
906                 if (!buffer_uptodate(bh)) {
907                         ret = -EIO;
908                 }
909                 /* ugly interaction with invalidatepage here.
910                  * reiserfs_invalidate_page will pin any buffer that has a valid
911                  * journal head from an older transaction.  If someone else sets
912                  * our buffer dirty after we write it in the first loop, and
913                  * then someone truncates the page away, nobody will ever write
914                  * the buffer. We're safe if we write the page one last time
915                  * after freeing the journal header.
916                  */
917                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
918                         spin_unlock(lock);
919                         ll_rw_block(WRITE, 1, &bh);
920                         spin_lock(lock);
921                 }
922                 put_bh(bh);
923                 cond_resched_lock(lock);
924         }
925         spin_unlock(lock);
926         return ret;
927 }
928
929 static int flush_older_commits(struct super_block *s,
930                                struct reiserfs_journal_list *jl)
931 {
932         struct reiserfs_journal *journal = SB_JOURNAL(s);
933         struct reiserfs_journal_list *other_jl;
934         struct reiserfs_journal_list *first_jl;
935         struct list_head *entry;
936         unsigned int trans_id = jl->j_trans_id;
937         unsigned int other_trans_id;
938         unsigned int first_trans_id;
939
940       find_first:
941         /*
942          * first we walk backwards to find the oldest uncommitted transation
943          */
944         first_jl = jl;
945         entry = jl->j_list.prev;
946         while (1) {
947                 other_jl = JOURNAL_LIST_ENTRY(entry);
948                 if (entry == &journal->j_journal_list ||
949                     atomic_read(&other_jl->j_older_commits_done))
950                         break;
951
952                 first_jl = other_jl;
953                 entry = other_jl->j_list.prev;
954         }
955
956         /* if we didn't find any older uncommitted transactions, return now */
957         if (first_jl == jl) {
958                 return 0;
959         }
960
961         first_trans_id = first_jl->j_trans_id;
962
963         entry = &first_jl->j_list;
964         while (1) {
965                 other_jl = JOURNAL_LIST_ENTRY(entry);
966                 other_trans_id = other_jl->j_trans_id;
967
968                 if (other_trans_id < trans_id) {
969                         if (atomic_read(&other_jl->j_commit_left) != 0) {
970                                 flush_commit_list(s, other_jl, 0);
971
972                                 /* list we were called with is gone, return */
973                                 if (!journal_list_still_alive(s, trans_id))
974                                         return 1;
975
976                                 /* the one we just flushed is gone, this means all
977                                  * older lists are also gone, so first_jl is no longer
978                                  * valid either.  Go back to the beginning.
979                                  */
980                                 if (!journal_list_still_alive
981                                     (s, other_trans_id)) {
982                                         goto find_first;
983                                 }
984                         }
985                         entry = entry->next;
986                         if (entry == &journal->j_journal_list)
987                                 return 0;
988                 } else {
989                         return 0;
990                 }
991         }
992         return 0;
993 }
994
995 static int reiserfs_async_progress_wait(struct super_block *s)
996 {
997         DEFINE_WAIT(wait);
998         struct reiserfs_journal *j = SB_JOURNAL(s);
999         if (atomic_read(&j->j_async_throttle))
1000                 congestion_wait(BLK_RW_ASYNC, HZ / 10);
1001         return 0;
1002 }
1003
1004 /*
1005 ** if this journal list still has commit blocks unflushed, send them to disk.
1006 **
1007 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
1008 ** Before the commit block can by written, every other log block must be safely on disk
1009 **
1010 */
1011 static int flush_commit_list(struct super_block *s,
1012                              struct reiserfs_journal_list *jl, int flushall)
1013 {
1014         int i;
1015         b_blocknr_t bn;
1016         struct buffer_head *tbh = NULL;
1017         unsigned int trans_id = jl->j_trans_id;
1018         struct reiserfs_journal *journal = SB_JOURNAL(s);
1019         int barrier = 0;
1020         int retval = 0;
1021         int write_len;
1022
1023         reiserfs_check_lock_depth(s, "flush_commit_list");
1024
1025         if (atomic_read(&jl->j_older_commits_done)) {
1026                 return 0;
1027         }
1028
1029         get_fs_excl();
1030
1031         /* before we can put our commit blocks on disk, we have to make sure everyone older than
1032          ** us is on disk too
1033          */
1034         BUG_ON(jl->j_len <= 0);
1035         BUG_ON(trans_id == journal->j_trans_id);
1036
1037         get_journal_list(jl);
1038         if (flushall) {
1039                 if (flush_older_commits(s, jl) == 1) {
1040                         /* list disappeared during flush_older_commits.  return */
1041                         goto put_jl;
1042                 }
1043         }
1044
1045         /* make sure nobody is trying to flush this one at the same time */
1046         mutex_lock(&jl->j_commit_mutex);
1047         if (!journal_list_still_alive(s, trans_id)) {
1048                 mutex_unlock(&jl->j_commit_mutex);
1049                 goto put_jl;
1050         }
1051         BUG_ON(jl->j_trans_id == 0);
1052
1053         /* this commit is done, exit */
1054         if (atomic_read(&(jl->j_commit_left)) <= 0) {
1055                 if (flushall) {
1056                         atomic_set(&(jl->j_older_commits_done), 1);
1057                 }
1058                 mutex_unlock(&jl->j_commit_mutex);
1059                 goto put_jl;
1060         }
1061
1062         if (!list_empty(&jl->j_bh_list)) {
1063                 int ret;
1064                 unlock_kernel();
1065                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1066                                             journal, jl, &jl->j_bh_list);
1067                 if (ret < 0 && retval == 0)
1068                         retval = ret;
1069                 lock_kernel();
1070         }
1071         BUG_ON(!list_empty(&jl->j_bh_list));
1072         /*
1073          * for the description block and all the log blocks, submit any buffers
1074          * that haven't already reached the disk.  Try to write at least 256
1075          * log blocks. later on, we will only wait on blocks that correspond
1076          * to this transaction, but while we're unplugging we might as well
1077          * get a chunk of data on there.
1078          */
1079         atomic_inc(&journal->j_async_throttle);
1080         write_len = jl->j_len + 1;
1081         if (write_len < 256)
1082                 write_len = 256;
1083         for (i = 0 ; i < write_len ; i++) {
1084                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1085                     SB_ONDISK_JOURNAL_SIZE(s);
1086                 tbh = journal_find_get_block(s, bn);
1087                 if (tbh) {
1088                         if (buffer_dirty(tbh))
1089                             ll_rw_block(WRITE, 1, &tbh) ;
1090                         put_bh(tbh) ;
1091                 }
1092         }
1093         atomic_dec(&journal->j_async_throttle);
1094
1095         /* We're skipping the commit if there's an error */
1096         if (retval || reiserfs_is_journal_aborted(journal))
1097                 barrier = 0;
1098
1099         /* wait on everything written so far before writing the commit
1100          * if we are in barrier mode, send the commit down now
1101          */
1102         barrier = reiserfs_barrier_flush(s);
1103         if (barrier) {
1104                 int ret;
1105                 lock_buffer(jl->j_commit_bh);
1106                 ret = submit_barrier_buffer(jl->j_commit_bh);
1107                 if (ret == -EOPNOTSUPP) {
1108                         set_buffer_uptodate(jl->j_commit_bh);
1109                         disable_barrier(s);
1110                         barrier = 0;
1111                 }
1112         }
1113         for (i = 0; i < (jl->j_len + 1); i++) {
1114                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1115                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1116                 tbh = journal_find_get_block(s, bn);
1117                 wait_on_buffer(tbh);
1118                 // since we're using ll_rw_blk above, it might have skipped over
1119                 // a locked buffer.  Double check here
1120                 //
1121                 if (buffer_dirty(tbh))  /* redundant, sync_dirty_buffer() checks */
1122                         sync_dirty_buffer(tbh);
1123                 if (unlikely(!buffer_uptodate(tbh))) {
1124 #ifdef CONFIG_REISERFS_CHECK
1125                         reiserfs_warning(s, "journal-601",
1126                                          "buffer write failed");
1127 #endif
1128                         retval = -EIO;
1129                 }
1130                 put_bh(tbh);    /* once for journal_find_get_block */
1131                 put_bh(tbh);    /* once due to original getblk in do_journal_end */
1132                 atomic_dec(&(jl->j_commit_left));
1133         }
1134
1135         BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1136
1137         if (!barrier) {
1138                 /* If there was a write error in the journal - we can't commit
1139                  * this transaction - it will be invalid and, if successful,
1140                  * will just end up propagating the write error out to
1141                  * the file system. */
1142                 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1143                         if (buffer_dirty(jl->j_commit_bh))
1144                                 BUG();
1145                         mark_buffer_dirty(jl->j_commit_bh) ;
1146                         sync_dirty_buffer(jl->j_commit_bh) ;
1147                 }
1148         } else
1149                 wait_on_buffer(jl->j_commit_bh);
1150
1151         check_barrier_completion(s, jl->j_commit_bh);
1152
1153         /* If there was a write error in the journal - we can't commit this
1154          * transaction - it will be invalid and, if successful, will just end
1155          * up propagating the write error out to the filesystem. */
1156         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1157 #ifdef CONFIG_REISERFS_CHECK
1158                 reiserfs_warning(s, "journal-615", "buffer write failed");
1159 #endif
1160                 retval = -EIO;
1161         }
1162         bforget(jl->j_commit_bh);
1163         if (journal->j_last_commit_id != 0 &&
1164             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1165                 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1166                                  journal->j_last_commit_id, jl->j_trans_id);
1167         }
1168         journal->j_last_commit_id = jl->j_trans_id;
1169
1170         /* now, every commit block is on the disk.  It is safe to allow blocks freed during this transaction to be reallocated */
1171         cleanup_freed_for_journal_list(s, jl);
1172
1173         retval = retval ? retval : journal->j_errno;
1174
1175         /* mark the metadata dirty */
1176         if (!retval)
1177                 dirty_one_transaction(s, jl);
1178         atomic_dec(&(jl->j_commit_left));
1179
1180         if (flushall) {
1181                 atomic_set(&(jl->j_older_commits_done), 1);
1182         }
1183         mutex_unlock(&jl->j_commit_mutex);
1184       put_jl:
1185         put_journal_list(s, jl);
1186
1187         if (retval)
1188                 reiserfs_abort(s, retval, "Journal write error in %s",
1189                                __func__);
1190         put_fs_excl();
1191         return retval;
1192 }
1193
1194 /*
1195 ** flush_journal_list frequently needs to find a newer transaction for a given block.  This does that, or
1196 ** returns NULL if it can't find anything
1197 */
1198 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1199                                                           reiserfs_journal_cnode
1200                                                           *cn)
1201 {
1202         struct super_block *sb = cn->sb;
1203         b_blocknr_t blocknr = cn->blocknr;
1204
1205         cn = cn->hprev;
1206         while (cn) {
1207                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1208                         return cn->jlist;
1209                 }
1210                 cn = cn->hprev;
1211         }
1212         return NULL;
1213 }
1214
1215 static int newer_jl_done(struct reiserfs_journal_cnode *cn)
1216 {
1217         struct super_block *sb = cn->sb;
1218         b_blocknr_t blocknr = cn->blocknr;
1219
1220         cn = cn->hprev;
1221         while (cn) {
1222                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist &&
1223                     atomic_read(&cn->jlist->j_commit_left) != 0)
1224                                     return 0;
1225                 cn = cn->hprev;
1226         }
1227         return 1;
1228 }
1229
1230 static void remove_journal_hash(struct super_block *,
1231                                 struct reiserfs_journal_cnode **,
1232                                 struct reiserfs_journal_list *, unsigned long,
1233                                 int);
1234
1235 /*
1236 ** once all the real blocks have been flushed, it is safe to remove them from the
1237 ** journal list for this transaction.  Aside from freeing the cnode, this also allows the
1238 ** block to be reallocated for data blocks if it had been deleted.
1239 */
1240 static void remove_all_from_journal_list(struct super_block *sb,
1241                                          struct reiserfs_journal_list *jl,
1242                                          int debug)
1243 {
1244         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1245         struct reiserfs_journal_cnode *cn, *last;
1246         cn = jl->j_realblock;
1247
1248         /* which is better, to lock once around the whole loop, or
1249          ** to lock for each call to remove_journal_hash?
1250          */
1251         while (cn) {
1252                 if (cn->blocknr != 0) {
1253                         if (debug) {
1254                                 reiserfs_warning(sb, "reiserfs-2201",
1255                                                  "block %u, bh is %d, state %ld",
1256                                                  cn->blocknr, cn->bh ? 1 : 0,
1257                                                  cn->state);
1258                         }
1259                         cn->state = 0;
1260                         remove_journal_hash(sb, journal->j_list_hash_table,
1261                                             jl, cn->blocknr, 1);
1262                 }
1263                 last = cn;
1264                 cn = cn->next;
1265                 free_cnode(sb, last);
1266         }
1267         jl->j_realblock = NULL;
1268 }
1269
1270 /*
1271 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1272 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1273 ** releasing blocks in this transaction for reuse as data blocks.
1274 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1275 **
1276 */
1277 static int _update_journal_header_block(struct super_block *sb,
1278                                         unsigned long offset,
1279                                         unsigned int trans_id)
1280 {
1281         struct reiserfs_journal_header *jh;
1282         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1283
1284         if (reiserfs_is_journal_aborted(journal))
1285                 return -EIO;
1286
1287         if (trans_id >= journal->j_last_flush_trans_id) {
1288                 if (buffer_locked((journal->j_header_bh))) {
1289                         wait_on_buffer((journal->j_header_bh));
1290                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1291 #ifdef CONFIG_REISERFS_CHECK
1292                                 reiserfs_warning(sb, "journal-699",
1293                                                  "buffer write failed");
1294 #endif
1295                                 return -EIO;
1296                         }
1297                 }
1298                 journal->j_last_flush_trans_id = trans_id;
1299                 journal->j_first_unflushed_offset = offset;
1300                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1301                                                         b_data);
1302                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1303                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1304                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1305
1306                 if (reiserfs_barrier_flush(sb)) {
1307                         int ret;
1308                         lock_buffer(journal->j_header_bh);
1309                         ret = submit_barrier_buffer(journal->j_header_bh);
1310                         if (ret == -EOPNOTSUPP) {
1311                                 set_buffer_uptodate(journal->j_header_bh);
1312                                 disable_barrier(sb);
1313                                 goto sync;
1314                         }
1315                         wait_on_buffer(journal->j_header_bh);
1316                         check_barrier_completion(sb, journal->j_header_bh);
1317                 } else {
1318                       sync:
1319                         set_buffer_dirty(journal->j_header_bh);
1320                         sync_dirty_buffer(journal->j_header_bh);
1321                 }
1322                 if (!buffer_uptodate(journal->j_header_bh)) {
1323                         reiserfs_warning(sb, "journal-837",
1324                                          "IO error during journal replay");
1325                         return -EIO;
1326                 }
1327         }
1328         return 0;
1329 }
1330
1331 static int update_journal_header_block(struct super_block *sb,
1332                                        unsigned long offset,
1333                                        unsigned int trans_id)
1334 {
1335         return _update_journal_header_block(sb, offset, trans_id);
1336 }
1337
1338 /*
1339 ** flush any and all journal lists older than you are
1340 ** can only be called from flush_journal_list
1341 */
1342 static int flush_older_journal_lists(struct super_block *sb,
1343                                      struct reiserfs_journal_list *jl)
1344 {
1345         struct list_head *entry;
1346         struct reiserfs_journal_list *other_jl;
1347         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1348         unsigned int trans_id = jl->j_trans_id;
1349
1350         /* we know we are the only ones flushing things, no extra race
1351          * protection is required.
1352          */
1353       restart:
1354         entry = journal->j_journal_list.next;
1355         /* Did we wrap? */
1356         if (entry == &journal->j_journal_list)
1357                 return 0;
1358         other_jl = JOURNAL_LIST_ENTRY(entry);
1359         if (other_jl->j_trans_id < trans_id) {
1360                 BUG_ON(other_jl->j_refcount <= 0);
1361                 /* do not flush all */
1362                 flush_journal_list(sb, other_jl, 0);
1363
1364                 /* other_jl is now deleted from the list */
1365                 goto restart;
1366         }
1367         return 0;
1368 }
1369
1370 static void del_from_work_list(struct super_block *s,
1371                                struct reiserfs_journal_list *jl)
1372 {
1373         struct reiserfs_journal *journal = SB_JOURNAL(s);
1374         if (!list_empty(&jl->j_working_list)) {
1375                 list_del_init(&jl->j_working_list);
1376                 journal->j_num_work_lists--;
1377         }
1378 }
1379
1380 /* flush a journal list, both commit and real blocks
1381 **
1382 ** always set flushall to 1, unless you are calling from inside
1383 ** flush_journal_list
1384 **
1385 ** IMPORTANT.  This can only be called while there are no journal writers,
1386 ** and the journal is locked.  That means it can only be called from
1387 ** do_journal_end, or by journal_release
1388 */
1389 static int flush_journal_list(struct super_block *s,
1390                               struct reiserfs_journal_list *jl, int flushall)
1391 {
1392         struct reiserfs_journal_list *pjl;
1393         struct reiserfs_journal_cnode *cn, *last;
1394         int count;
1395         int was_jwait = 0;
1396         int was_dirty = 0;
1397         struct buffer_head *saved_bh;
1398         unsigned long j_len_saved = jl->j_len;
1399         struct reiserfs_journal *journal = SB_JOURNAL(s);
1400         int err = 0;
1401
1402         BUG_ON(j_len_saved <= 0);
1403
1404         if (atomic_read(&journal->j_wcount) != 0) {
1405                 reiserfs_warning(s, "clm-2048", "called with wcount %d",
1406                                  atomic_read(&journal->j_wcount));
1407         }
1408         BUG_ON(jl->j_trans_id == 0);
1409
1410         /* if flushall == 0, the lock is already held */
1411         if (flushall) {
1412                 mutex_lock(&journal->j_flush_mutex);
1413         } else if (mutex_trylock(&journal->j_flush_mutex)) {
1414                 BUG();
1415         }
1416
1417         count = 0;
1418         if (j_len_saved > journal->j_trans_max) {
1419                 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1420                                j_len_saved, jl->j_trans_id);
1421                 return 0;
1422         }
1423
1424         get_fs_excl();
1425
1426         /* if all the work is already done, get out of here */
1427         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1428             atomic_read(&(jl->j_commit_left)) <= 0) {
1429                 goto flush_older_and_return;
1430         }
1431
1432         /* start by putting the commit list on disk.  This will also flush
1433          ** the commit lists of any olders transactions
1434          */
1435         flush_commit_list(s, jl, 1);
1436
1437         if (!(jl->j_state & LIST_DIRTY)
1438             && !reiserfs_is_journal_aborted(journal))
1439                 BUG();
1440
1441         /* are we done now? */
1442         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1443             atomic_read(&(jl->j_commit_left)) <= 0) {
1444                 goto flush_older_and_return;
1445         }
1446
1447         /* loop through each cnode, see if we need to write it,
1448          ** or wait on a more recent transaction, or just ignore it
1449          */
1450         if (atomic_read(&(journal->j_wcount)) != 0) {
1451                 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1452                                "wcount is not 0");
1453         }
1454         cn = jl->j_realblock;
1455         while (cn) {
1456                 was_jwait = 0;
1457                 was_dirty = 0;
1458                 saved_bh = NULL;
1459                 /* blocknr of 0 is no longer in the hash, ignore it */
1460                 if (cn->blocknr == 0) {
1461                         goto free_cnode;
1462                 }
1463
1464                 /* This transaction failed commit. Don't write out to the disk */
1465                 if (!(jl->j_state & LIST_DIRTY))
1466                         goto free_cnode;
1467
1468                 pjl = find_newer_jl_for_cn(cn);
1469                 /* the order is important here.  We check pjl to make sure we
1470                  ** don't clear BH_JDirty_wait if we aren't the one writing this
1471                  ** block to disk
1472                  */
1473                 if (!pjl && cn->bh) {
1474                         saved_bh = cn->bh;
1475
1476                         /* we do this to make sure nobody releases the buffer while
1477                          ** we are working with it
1478                          */
1479                         get_bh(saved_bh);
1480
1481                         if (buffer_journal_dirty(saved_bh)) {
1482                                 BUG_ON(!can_dirty(cn));
1483                                 was_jwait = 1;
1484                                 was_dirty = 1;
1485                         } else if (can_dirty(cn)) {
1486                                 /* everything with !pjl && jwait should be writable */
1487                                 BUG();
1488                         }
1489                 }
1490
1491                 /* if someone has this block in a newer transaction, just make
1492                  ** sure they are committed, and don't try writing it to disk
1493                  */
1494                 if (pjl) {
1495                         if (atomic_read(&pjl->j_commit_left))
1496                                 flush_commit_list(s, pjl, 1);
1497                         goto free_cnode;
1498                 }
1499
1500                 /* bh == NULL when the block got to disk on its own, OR,
1501                  ** the block got freed in a future transaction
1502                  */
1503                 if (saved_bh == NULL) {
1504                         goto free_cnode;
1505                 }
1506
1507                 /* this should never happen.  kupdate_one_transaction has this list
1508                  ** locked while it works, so we should never see a buffer here that
1509                  ** is not marked JDirty_wait
1510                  */
1511                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1512                         reiserfs_warning(s, "journal-813",
1513                                          "BAD! buffer %llu %cdirty %cjwait, "
1514                                          "not in a newer tranasction",
1515                                          (unsigned long long)saved_bh->
1516                                          b_blocknr, was_dirty ? ' ' : '!',
1517                                          was_jwait ? ' ' : '!');
1518                 }
1519                 if (was_dirty) {
1520                         /* we inc again because saved_bh gets decremented at free_cnode */
1521                         get_bh(saved_bh);
1522                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1523                         lock_buffer(saved_bh);
1524                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1525                         if (buffer_dirty(saved_bh))
1526                                 submit_logged_buffer(saved_bh);
1527                         else
1528                                 unlock_buffer(saved_bh);
1529                         count++;
1530                 } else {
1531                         reiserfs_warning(s, "clm-2082",
1532                                          "Unable to flush buffer %llu in %s",
1533                                          (unsigned long long)saved_bh->
1534                                          b_blocknr, __func__);
1535                 }
1536               free_cnode:
1537                 last = cn;
1538                 cn = cn->next;
1539                 if (saved_bh) {
1540                         /* we incremented this to keep others from taking the buffer head away */
1541                         put_bh(saved_bh);
1542                         if (atomic_read(&(saved_bh->b_count)) < 0) {
1543                                 reiserfs_warning(s, "journal-945",
1544                                                  "saved_bh->b_count < 0");
1545                         }
1546                 }
1547         }
1548         if (count > 0) {
1549                 cn = jl->j_realblock;
1550                 while (cn) {
1551                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1552                                 if (!cn->bh) {
1553                                         reiserfs_panic(s, "journal-1011",
1554                                                        "cn->bh is NULL");
1555                                 }
1556                                 wait_on_buffer(cn->bh);
1557                                 if (!cn->bh) {
1558                                         reiserfs_panic(s, "journal-1012",
1559                                                        "cn->bh is NULL");
1560                                 }
1561                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1562 #ifdef CONFIG_REISERFS_CHECK
1563                                         reiserfs_warning(s, "journal-949",
1564                                                          "buffer write failed");
1565 #endif
1566                                         err = -EIO;
1567                                 }
1568                                 /* note, we must clear the JDirty_wait bit after the up to date
1569                                  ** check, otherwise we race against our flushpage routine
1570                                  */
1571                                 BUG_ON(!test_clear_buffer_journal_dirty
1572                                        (cn->bh));
1573
1574                                 /* drop one ref for us */
1575                                 put_bh(cn->bh);
1576                                 /* drop one ref for journal_mark_dirty */
1577                                 release_buffer_page(cn->bh);
1578                         }
1579                         cn = cn->next;
1580                 }
1581         }
1582
1583         if (err)
1584                 reiserfs_abort(s, -EIO,
1585                                "Write error while pushing transaction to disk in %s",
1586                                __func__);
1587       flush_older_and_return:
1588
1589         /* before we can update the journal header block, we _must_ flush all
1590          ** real blocks from all older transactions to disk.  This is because
1591          ** once the header block is updated, this transaction will not be
1592          ** replayed after a crash
1593          */
1594         if (flushall) {
1595                 flush_older_journal_lists(s, jl);
1596         }
1597
1598         err = journal->j_errno;
1599         /* before we can remove everything from the hash tables for this
1600          ** transaction, we must make sure it can never be replayed
1601          **
1602          ** since we are only called from do_journal_end, we know for sure there
1603          ** are no allocations going on while we are flushing journal lists.  So,
1604          ** we only need to update the journal header block for the last list
1605          ** being flushed
1606          */
1607         if (!err && flushall) {
1608                 err =
1609                     update_journal_header_block(s,
1610                                                 (jl->j_start + jl->j_len +
1611                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1612                                                 jl->j_trans_id);
1613                 if (err)
1614                         reiserfs_abort(s, -EIO,
1615                                        "Write error while updating journal header in %s",
1616                                        __func__);
1617         }
1618         remove_all_from_journal_list(s, jl, 0);
1619         list_del_init(&jl->j_list);
1620         journal->j_num_lists--;
1621         del_from_work_list(s, jl);
1622
1623         if (journal->j_last_flush_id != 0 &&
1624             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1625                 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1626                                  journal->j_last_flush_id, jl->j_trans_id);
1627         }
1628         journal->j_last_flush_id = jl->j_trans_id;
1629
1630         /* not strictly required since we are freeing the list, but it should
1631          * help find code using dead lists later on
1632          */
1633         jl->j_len = 0;
1634         atomic_set(&(jl->j_nonzerolen), 0);
1635         jl->j_start = 0;
1636         jl->j_realblock = NULL;
1637         jl->j_commit_bh = NULL;
1638         jl->j_trans_id = 0;
1639         jl->j_state = 0;
1640         put_journal_list(s, jl);
1641         if (flushall)
1642                 mutex_unlock(&journal->j_flush_mutex);
1643         put_fs_excl();
1644         return err;
1645 }
1646
1647 static int test_transaction(struct super_block *s,
1648                             struct reiserfs_journal_list *jl)
1649 {
1650         struct reiserfs_journal_cnode *cn;
1651
1652         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0)
1653                 return 1;
1654
1655         cn = jl->j_realblock;
1656         while (cn) {
1657                 /* if the blocknr == 0, this has been cleared from the hash,
1658                  ** skip it
1659                  */
1660                 if (cn->blocknr == 0) {
1661                         goto next;
1662                 }
1663                 if (cn->bh && !newer_jl_done(cn))
1664                         return 0;
1665               next:
1666                 cn = cn->next;
1667                 cond_resched();
1668         }
1669         return 0;
1670 }
1671
1672 static int write_one_transaction(struct super_block *s,
1673                                  struct reiserfs_journal_list *jl,
1674                                  struct buffer_chunk *chunk)
1675 {
1676         struct reiserfs_journal_cnode *cn;
1677         int ret = 0;
1678
1679         jl->j_state |= LIST_TOUCHED;
1680         del_from_work_list(s, jl);
1681         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1682                 return 0;
1683         }
1684
1685         cn = jl->j_realblock;
1686         while (cn) {
1687                 /* if the blocknr == 0, this has been cleared from the hash,
1688                  ** skip it
1689                  */
1690                 if (cn->blocknr == 0) {
1691                         goto next;
1692                 }
1693                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1694                         struct buffer_head *tmp_bh;
1695                         /* we can race against journal_mark_freed when we try
1696                          * to lock_buffer(cn->bh), so we have to inc the buffer
1697                          * count, and recheck things after locking
1698                          */
1699                         tmp_bh = cn->bh;
1700                         get_bh(tmp_bh);
1701                         lock_buffer(tmp_bh);
1702                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1703                                 if (!buffer_journal_dirty(tmp_bh) ||
1704                                     buffer_journal_prepared(tmp_bh))
1705                                         BUG();
1706                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1707                                 ret++;
1708                         } else {
1709                                 /* note, cn->bh might be null now */
1710                                 unlock_buffer(tmp_bh);
1711                         }
1712                         put_bh(tmp_bh);
1713                 }
1714               next:
1715                 cn = cn->next;
1716                 cond_resched();
1717         }
1718         return ret;
1719 }
1720
1721 /* used by flush_commit_list */
1722 static int dirty_one_transaction(struct super_block *s,
1723                                  struct reiserfs_journal_list *jl)
1724 {
1725         struct reiserfs_journal_cnode *cn;
1726         struct reiserfs_journal_list *pjl;
1727         int ret = 0;
1728
1729         jl->j_state |= LIST_DIRTY;
1730         cn = jl->j_realblock;
1731         while (cn) {
1732                 /* look for a more recent transaction that logged this
1733                  ** buffer.  Only the most recent transaction with a buffer in
1734                  ** it is allowed to send that buffer to disk
1735                  */
1736                 pjl = find_newer_jl_for_cn(cn);
1737                 if (!pjl && cn->blocknr && cn->bh
1738                     && buffer_journal_dirty(cn->bh)) {
1739                         BUG_ON(!can_dirty(cn));
1740                         /* if the buffer is prepared, it will either be logged
1741                          * or restored.  If restored, we need to make sure
1742                          * it actually gets marked dirty
1743                          */
1744                         clear_buffer_journal_new(cn->bh);
1745                         if (buffer_journal_prepared(cn->bh)) {
1746                                 set_buffer_journal_restore_dirty(cn->bh);
1747                         } else {
1748                                 set_buffer_journal_test(cn->bh);
1749                                 mark_buffer_dirty(cn->bh);
1750                         }
1751                 }
1752                 cn = cn->next;
1753         }
1754         return ret;
1755 }
1756
1757 static int kupdate_transactions(struct super_block *s,
1758                                 struct reiserfs_journal_list *jl,
1759                                 struct reiserfs_journal_list **next_jl,
1760                                 unsigned int *next_trans_id,
1761                                 int num_blocks, int num_trans)
1762 {
1763         int ret = 0;
1764         int written = 0;
1765         int transactions_flushed = 0;
1766         unsigned int orig_trans_id = jl->j_trans_id;
1767         struct buffer_chunk chunk;
1768         struct list_head *entry;
1769         struct reiserfs_journal *journal = SB_JOURNAL(s);
1770         chunk.nr = 0;
1771
1772         mutex_lock(&journal->j_flush_mutex);
1773         if (!journal_list_still_alive(s, orig_trans_id)) {
1774                 goto done;
1775         }
1776
1777         /* we've got j_flush_mutex held, nobody is going to delete any
1778          * of these lists out from underneath us
1779          */
1780         while ((num_trans && transactions_flushed < num_trans) ||
1781                (!num_trans && written < num_blocks)) {
1782
1783                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1784                     atomic_read(&jl->j_commit_left)
1785                     || !(jl->j_state & LIST_DIRTY)) {
1786                         del_from_work_list(s, jl);
1787                         break;
1788                 }
1789                 ret = write_one_transaction(s, jl, &chunk);
1790
1791                 if (ret < 0)
1792                         goto done;
1793                 transactions_flushed++;
1794                 written += ret;
1795                 entry = jl->j_list.next;
1796
1797                 /* did we wrap? */
1798                 if (entry == &journal->j_journal_list) {
1799                         break;
1800                 }
1801                 jl = JOURNAL_LIST_ENTRY(entry);
1802
1803                 /* don't bother with older transactions */
1804                 if (jl->j_trans_id <= orig_trans_id)
1805                         break;
1806         }
1807         if (chunk.nr) {
1808                 write_chunk(&chunk);
1809         }
1810
1811       done:
1812         mutex_unlock(&journal->j_flush_mutex);
1813         return ret;
1814 }
1815
1816 /* for o_sync and fsync heavy applications, they tend to use
1817 ** all the journa list slots with tiny transactions.  These
1818 ** trigger lots and lots of calls to update the header block, which
1819 ** adds seeks and slows things down.
1820 **
1821 ** This function tries to clear out a large chunk of the journal lists
1822 ** at once, which makes everything faster since only the newest journal
1823 ** list updates the header block
1824 */
1825 static int flush_used_journal_lists(struct super_block *s,
1826                                     struct reiserfs_journal_list *jl)
1827 {
1828         unsigned long len = 0;
1829         unsigned long cur_len;
1830         int ret;
1831         int i;
1832         int limit = 256;
1833         struct reiserfs_journal_list *tjl;
1834         struct reiserfs_journal_list *flush_jl;
1835         unsigned int trans_id;
1836         struct reiserfs_journal *journal = SB_JOURNAL(s);
1837
1838         flush_jl = tjl = jl;
1839
1840         /* in data logging mode, try harder to flush a lot of blocks */
1841         if (reiserfs_data_log(s))
1842                 limit = 1024;
1843         /* flush for 256 transactions or limit blocks, whichever comes first */
1844         for (i = 0; i < 256 && len < limit; i++) {
1845                 if (atomic_read(&tjl->j_commit_left) ||
1846                     tjl->j_trans_id < jl->j_trans_id) {
1847                         break;
1848                 }
1849                 cur_len = atomic_read(&tjl->j_nonzerolen);
1850                 if (cur_len > 0) {
1851                         tjl->j_state &= ~LIST_TOUCHED;
1852                 }
1853                 len += cur_len;
1854                 flush_jl = tjl;
1855                 if (tjl->j_list.next == &journal->j_journal_list)
1856                         break;
1857                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1858         }
1859         /* try to find a group of blocks we can flush across all the
1860          ** transactions, but only bother if we've actually spanned
1861          ** across multiple lists
1862          */
1863         if (flush_jl != jl) {
1864                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1865         }
1866         flush_journal_list(s, flush_jl, 1);
1867         return 0;
1868 }
1869
1870 /*
1871 ** removes any nodes in table with name block and dev as bh.
1872 ** only touchs the hnext and hprev pointers.
1873 */
1874 void remove_journal_hash(struct super_block *sb,
1875                          struct reiserfs_journal_cnode **table,
1876                          struct reiserfs_journal_list *jl,
1877                          unsigned long block, int remove_freed)
1878 {
1879         struct reiserfs_journal_cnode *cur;
1880         struct reiserfs_journal_cnode **head;
1881
1882         head = &(journal_hash(table, sb, block));
1883         if (!head) {
1884                 return;
1885         }
1886         cur = *head;
1887         while (cur) {
1888                 if (cur->blocknr == block && cur->sb == sb
1889                     && (jl == NULL || jl == cur->jlist)
1890                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1891                         if (cur->hnext) {
1892                                 cur->hnext->hprev = cur->hprev;
1893                         }
1894                         if (cur->hprev) {
1895                                 cur->hprev->hnext = cur->hnext;
1896                         } else {
1897                                 *head = cur->hnext;
1898                         }
1899                         cur->blocknr = 0;
1900                         cur->sb = NULL;
1901                         cur->state = 0;
1902                         if (cur->bh && cur->jlist)      /* anybody who clears the cur->bh will also dec the nonzerolen */
1903                                 atomic_dec(&(cur->jlist->j_nonzerolen));
1904                         cur->bh = NULL;
1905                         cur->jlist = NULL;
1906                 }
1907                 cur = cur->hnext;
1908         }
1909 }
1910
1911 static void free_journal_ram(struct super_block *sb)
1912 {
1913         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1914         kfree(journal->j_current_jl);
1915         journal->j_num_lists--;
1916
1917         vfree(journal->j_cnode_free_orig);
1918         free_list_bitmaps(sb, journal->j_list_bitmap);
1919         free_bitmap_nodes(sb);  /* must be after free_list_bitmaps */
1920         if (journal->j_header_bh) {
1921                 brelse(journal->j_header_bh);
1922         }
1923         /* j_header_bh is on the journal dev, make sure not to release the journal
1924          * dev until we brelse j_header_bh
1925          */
1926         release_journal_dev(sb, journal);
1927         vfree(journal);
1928 }
1929
1930 /*
1931 ** call on unmount.  Only set error to 1 if you haven't made your way out
1932 ** of read_super() yet.  Any other caller must keep error at 0.
1933 */
1934 static int do_journal_release(struct reiserfs_transaction_handle *th,
1935                               struct super_block *sb, int error)
1936 {
1937         struct reiserfs_transaction_handle myth;
1938         int flushed = 0;
1939         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1940
1941         /* we only want to flush out transactions if we were called with error == 0
1942          */
1943         if (!error && !(sb->s_flags & MS_RDONLY)) {
1944                 /* end the current trans */
1945                 BUG_ON(!th->t_trans_id);
1946                 do_journal_end(th, sb, 10, FLUSH_ALL);
1947
1948                 /* make sure something gets logged to force our way into the flush code */
1949                 if (!journal_join(&myth, sb, 1)) {
1950                         reiserfs_prepare_for_journal(sb,
1951                                                      SB_BUFFER_WITH_SB(sb),
1952                                                      1);
1953                         journal_mark_dirty(&myth, sb,
1954                                            SB_BUFFER_WITH_SB(sb));
1955                         do_journal_end(&myth, sb, 1, FLUSH_ALL);
1956                         flushed = 1;
1957                 }
1958         }
1959
1960         /* this also catches errors during the do_journal_end above */
1961         if (!error && reiserfs_is_journal_aborted(journal)) {
1962                 memset(&myth, 0, sizeof(myth));
1963                 if (!journal_join_abort(&myth, sb, 1)) {
1964                         reiserfs_prepare_for_journal(sb,
1965                                                      SB_BUFFER_WITH_SB(sb),
1966                                                      1);
1967                         journal_mark_dirty(&myth, sb,
1968                                            SB_BUFFER_WITH_SB(sb));
1969                         do_journal_end(&myth, sb, 1, FLUSH_ALL);
1970                 }
1971         }
1972
1973         reiserfs_mounted_fs_count--;
1974         /* wait for all commits to finish */
1975         cancel_delayed_work(&SB_JOURNAL(sb)->j_work);
1976         flush_workqueue(commit_wq);
1977         if (!reiserfs_mounted_fs_count) {
1978                 destroy_workqueue(commit_wq);
1979                 commit_wq = NULL;
1980         }
1981
1982         free_journal_ram(sb);
1983
1984         return 0;
1985 }
1986
1987 /*
1988 ** call on unmount.  flush all journal trans, release all alloc'd ram
1989 */
1990 int journal_release(struct reiserfs_transaction_handle *th,
1991                     struct super_block *sb)
1992 {
1993         return do_journal_release(th, sb, 0);
1994 }
1995
1996 /*
1997 ** only call from an error condition inside reiserfs_read_super!
1998 */
1999 int journal_release_error(struct reiserfs_transaction_handle *th,
2000                           struct super_block *sb)
2001 {
2002         return do_journal_release(th, sb, 1);
2003 }
2004
2005 /* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
2006 static int journal_compare_desc_commit(struct super_block *sb,
2007                                        struct reiserfs_journal_desc *desc,
2008                                        struct reiserfs_journal_commit *commit)
2009 {
2010         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
2011             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
2012             get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
2013             get_commit_trans_len(commit) <= 0) {
2014                 return 1;
2015         }
2016         return 0;
2017 }
2018
2019 /* returns 0 if it did not find a description block
2020 ** returns -1 if it found a corrupt commit block
2021 ** returns 1 if both desc and commit were valid
2022 */
2023 static int journal_transaction_is_valid(struct super_block *sb,
2024                                         struct buffer_head *d_bh,
2025                                         unsigned int *oldest_invalid_trans_id,
2026                                         unsigned long *newest_mount_id)
2027 {
2028         struct reiserfs_journal_desc *desc;
2029         struct reiserfs_journal_commit *commit;
2030         struct buffer_head *c_bh;
2031         unsigned long offset;
2032
2033         if (!d_bh)
2034                 return 0;
2035
2036         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2037         if (get_desc_trans_len(desc) > 0
2038             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2039                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2040                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2041                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2042                                        "journal-986: transaction "
2043                                        "is valid returning because trans_id %d is greater than "
2044                                        "oldest_invalid %lu",
2045                                        get_desc_trans_id(desc),
2046                                        *oldest_invalid_trans_id);
2047                         return 0;
2048                 }
2049                 if (newest_mount_id
2050                     && *newest_mount_id > get_desc_mount_id(desc)) {
2051                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2052                                        "journal-1087: transaction "
2053                                        "is valid returning because mount_id %d is less than "
2054                                        "newest_mount_id %lu",
2055                                        get_desc_mount_id(desc),
2056                                        *newest_mount_id);
2057                         return -1;
2058                 }
2059                 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2060                         reiserfs_warning(sb, "journal-2018",
2061                                          "Bad transaction length %d "
2062                                          "encountered, ignoring transaction",
2063                                          get_desc_trans_len(desc));
2064                         return -1;
2065                 }
2066                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2067
2068                 /* ok, we have a journal description block, lets see if the transaction was valid */
2069                 c_bh =
2070                     journal_bread(sb,
2071                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2072                                   ((offset + get_desc_trans_len(desc) +
2073                                     1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2074                 if (!c_bh)
2075                         return 0;
2076                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2077                 if (journal_compare_desc_commit(sb, desc, commit)) {
2078                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2079                                        "journal_transaction_is_valid, commit offset %ld had bad "
2080                                        "time %d or length %d",
2081                                        c_bh->b_blocknr -
2082                                        SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2083                                        get_commit_trans_id(commit),
2084                                        get_commit_trans_len(commit));
2085                         brelse(c_bh);
2086                         if (oldest_invalid_trans_id) {
2087                                 *oldest_invalid_trans_id =
2088                                     get_desc_trans_id(desc);
2089                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2090                                                "journal-1004: "
2091                                                "transaction_is_valid setting oldest invalid trans_id "
2092                                                "to %d",
2093                                                get_desc_trans_id(desc));
2094                         }
2095                         return -1;
2096                 }
2097                 brelse(c_bh);
2098                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2099                                "journal-1006: found valid "
2100                                "transaction start offset %llu, len %d id %d",
2101                                d_bh->b_blocknr -
2102                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2103                                get_desc_trans_len(desc),
2104                                get_desc_trans_id(desc));
2105                 return 1;
2106         } else {
2107                 return 0;
2108         }
2109 }
2110
2111 static void brelse_array(struct buffer_head **heads, int num)
2112 {
2113         int i;
2114         for (i = 0; i < num; i++) {
2115                 brelse(heads[i]);
2116         }
2117 }
2118
2119 /*
2120 ** given the start, and values for the oldest acceptable transactions,
2121 ** this either reads in a replays a transaction, or returns because the transaction
2122 ** is invalid, or too old.
2123 */
2124 static int journal_read_transaction(struct super_block *sb,
2125                                     unsigned long cur_dblock,
2126                                     unsigned long oldest_start,
2127                                     unsigned int oldest_trans_id,
2128                                     unsigned long newest_mount_id)
2129 {
2130         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2131         struct reiserfs_journal_desc *desc;
2132         struct reiserfs_journal_commit *commit;
2133         unsigned int trans_id = 0;
2134         struct buffer_head *c_bh;
2135         struct buffer_head *d_bh;
2136         struct buffer_head **log_blocks = NULL;
2137         struct buffer_head **real_blocks = NULL;
2138         unsigned int trans_offset;
2139         int i;
2140         int trans_half;
2141
2142         d_bh = journal_bread(sb, cur_dblock);
2143         if (!d_bh)
2144                 return 1;
2145         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2146         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2147         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2148                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2149                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2150                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2151         if (get_desc_trans_id(desc) < oldest_trans_id) {
2152                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2153                                "journal_read_trans skipping because %lu is too old",
2154                                cur_dblock -
2155                                SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2156                 brelse(d_bh);
2157                 return 1;
2158         }
2159         if (get_desc_mount_id(desc) != newest_mount_id) {
2160                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2161                                "journal_read_trans skipping because %d is != "
2162                                "newest_mount_id %lu", get_desc_mount_id(desc),
2163                                newest_mount_id);
2164                 brelse(d_bh);
2165                 return 1;
2166         }
2167         c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2168                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2169                               SB_ONDISK_JOURNAL_SIZE(sb)));
2170         if (!c_bh) {
2171                 brelse(d_bh);
2172                 return 1;
2173         }
2174         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2175         if (journal_compare_desc_commit(sb, desc, commit)) {
2176                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2177                                "journal_read_transaction, "
2178                                "commit offset %llu had bad time %d or length %d",
2179                                c_bh->b_blocknr -
2180                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2181                                get_commit_trans_id(commit),
2182                                get_commit_trans_len(commit));
2183                 brelse(c_bh);
2184                 brelse(d_bh);
2185                 return 1;
2186         }
2187         trans_id = get_desc_trans_id(desc);
2188         /* now we know we've got a good transaction, and it was inside the valid time ranges */
2189         log_blocks = kmalloc(get_desc_trans_len(desc) *
2190                              sizeof(struct buffer_head *), GFP_NOFS);
2191         real_blocks = kmalloc(get_desc_trans_len(desc) *
2192                               sizeof(struct buffer_head *), GFP_NOFS);
2193         if (!log_blocks || !real_blocks) {
2194                 brelse(c_bh);
2195                 brelse(d_bh);
2196                 kfree(log_blocks);
2197                 kfree(real_blocks);
2198                 reiserfs_warning(sb, "journal-1169",
2199                                  "kmalloc failed, unable to mount FS");
2200                 return -1;
2201         }
2202         /* get all the buffer heads */
2203         trans_half = journal_trans_half(sb->s_blocksize);
2204         for (i = 0; i < get_desc_trans_len(desc); i++) {
2205                 log_blocks[i] =
2206                     journal_getblk(sb,
2207                                    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2208                                    (trans_offset + 1 +
2209                                     i) % SB_ONDISK_JOURNAL_SIZE(sb));
2210                 if (i < trans_half) {
2211                         real_blocks[i] =
2212                             sb_getblk(sb,
2213                                       le32_to_cpu(desc->j_realblock[i]));
2214                 } else {
2215                         real_blocks[i] =
2216                             sb_getblk(sb,
2217                                       le32_to_cpu(commit->
2218                                                   j_realblock[i - trans_half]));
2219                 }
2220                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2221                         reiserfs_warning(sb, "journal-1207",
2222                                          "REPLAY FAILURE fsck required! "
2223                                          "Block to replay is outside of "
2224                                          "filesystem");
2225                         goto abort_replay;
2226                 }
2227                 /* make sure we don't try to replay onto log or reserved area */
2228                 if (is_block_in_log_or_reserved_area
2229                     (sb, real_blocks[i]->b_blocknr)) {
2230                         reiserfs_warning(sb, "journal-1204",
2231                                          "REPLAY FAILURE fsck required! "
2232                                          "Trying to replay onto a log block");
2233                       abort_replay:
2234                         brelse_array(log_blocks, i);
2235                         brelse_array(real_blocks, i);
2236                         brelse(c_bh);
2237                         brelse(d_bh);
2238                         kfree(log_blocks);
2239                         kfree(real_blocks);
2240                         return -1;
2241                 }
2242         }
2243         /* read in the log blocks, memcpy to the corresponding real block */
2244         ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2245         for (i = 0; i < get_desc_trans_len(desc); i++) {
2246                 wait_on_buffer(log_blocks[i]);
2247                 if (!buffer_uptodate(log_blocks[i])) {
2248                         reiserfs_warning(sb, "journal-1212",
2249                                          "REPLAY FAILURE fsck required! "
2250                                          "buffer write failed");
2251                         brelse_array(log_blocks + i,
2252                                      get_desc_trans_len(desc) - i);
2253                         brelse_array(real_blocks, get_desc_trans_len(desc));
2254                         brelse(c_bh);
2255                         brelse(d_bh);
2256                         kfree(log_blocks);
2257                         kfree(real_blocks);
2258                         return -1;
2259                 }
2260                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2261                        real_blocks[i]->b_size);
2262                 set_buffer_uptodate(real_blocks[i]);
2263                 brelse(log_blocks[i]);
2264         }
2265         /* flush out the real blocks */
2266         for (i = 0; i < get_desc_trans_len(desc); i++) {
2267                 set_buffer_dirty(real_blocks[i]);
2268                 ll_rw_block(SWRITE, 1, real_blocks + i);
2269         }
2270         for (i = 0; i < get_desc_trans_len(desc); i++) {
2271                 wait_on_buffer(real_blocks[i]);
2272                 if (!buffer_uptodate(real_blocks[i])) {
2273                         reiserfs_warning(sb, "journal-1226",
2274                                          "REPLAY FAILURE, fsck required! "
2275                                          "buffer write failed");
2276                         brelse_array(real_blocks + i,
2277                                      get_desc_trans_len(desc) - i);
2278                         brelse(c_bh);
2279                         brelse(d_bh);
2280                         kfree(log_blocks);
2281                         kfree(real_blocks);
2282                         return -1;
2283                 }
2284                 brelse(real_blocks[i]);
2285         }
2286         cur_dblock =
2287             SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2288             ((trans_offset + get_desc_trans_len(desc) +
2289               2) % SB_ONDISK_JOURNAL_SIZE(sb));
2290         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2291                        "journal-1095: setting journal " "start to offset %ld",
2292                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2293
2294         /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2295         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2296         journal->j_last_flush_trans_id = trans_id;
2297         journal->j_trans_id = trans_id + 1;
2298         /* check for trans_id overflow */
2299         if (journal->j_trans_id == 0)
2300                 journal->j_trans_id = 10;
2301         brelse(c_bh);
2302         brelse(d_bh);
2303         kfree(log_blocks);
2304         kfree(real_blocks);
2305         return 0;
2306 }
2307
2308 /* This function reads blocks starting from block and to max_block of bufsize
2309    size (but no more than BUFNR blocks at a time). This proved to improve
2310    mounting speed on self-rebuilding raid5 arrays at least.
2311    Right now it is only used from journal code. But later we might use it
2312    from other places.
2313    Note: Do not use journal_getblk/sb_getblk functions here! */
2314 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2315                                            b_blocknr_t block, int bufsize,
2316                                            b_blocknr_t max_block)
2317 {
2318         struct buffer_head *bhlist[BUFNR];
2319         unsigned int blocks = BUFNR;
2320         struct buffer_head *bh;
2321         int i, j;
2322
2323         bh = __getblk(dev, block, bufsize);
2324         if (buffer_uptodate(bh))
2325                 return (bh);
2326
2327         if (block + BUFNR > max_block) {
2328                 blocks = max_block - block;
2329         }
2330         bhlist[0] = bh;
2331         j = 1;
2332         for (i = 1; i < blocks; i++) {
2333                 bh = __getblk(dev, block + i, bufsize);
2334                 if (buffer_uptodate(bh)) {
2335                         brelse(bh);
2336                         break;
2337                 } else
2338                         bhlist[j++] = bh;
2339         }
2340         ll_rw_block(READ, j, bhlist);
2341         for (i = 1; i < j; i++)
2342                 brelse(bhlist[i]);
2343         bh = bhlist[0];
2344         wait_on_buffer(bh);
2345         if (buffer_uptodate(bh))
2346                 return bh;
2347         brelse(bh);
2348         return NULL;
2349 }
2350
2351 /*
2352 ** read and replay the log
2353 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2354 ** transaction.  This tests that before finding all the transactions in the log, which makes normal mount times fast.
2355 **
2356 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2357 **
2358 ** On exit, it sets things up so the first transaction will work correctly.
2359 */
2360 static int journal_read(struct super_block *sb)
2361 {
2362         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2363         struct reiserfs_journal_desc *desc;
2364         unsigned int oldest_trans_id = 0;
2365         unsigned int oldest_invalid_trans_id = 0;
2366         time_t start;
2367         unsigned long oldest_start = 0;
2368         unsigned long cur_dblock = 0;
2369         unsigned long newest_mount_id = 9;
2370         struct buffer_head *d_bh;
2371         struct reiserfs_journal_header *jh;
2372         int valid_journal_header = 0;
2373         int replay_count = 0;
2374         int continue_replay = 1;
2375         int ret;
2376         char b[BDEVNAME_SIZE];
2377
2378         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2379         reiserfs_info(sb, "checking transaction log (%s)\n",
2380                       bdevname(journal->j_dev_bd, b));
2381         start = get_seconds();
2382
2383         /* step 1, read in the journal header block.  Check the transaction it says
2384          ** is the first unflushed, and if that transaction is not valid,
2385          ** replay is done
2386          */
2387         journal->j_header_bh = journal_bread(sb,
2388                                              SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2389                                              + SB_ONDISK_JOURNAL_SIZE(sb));
2390         if (!journal->j_header_bh) {
2391                 return 1;
2392         }
2393         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2394         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2395             SB_ONDISK_JOURNAL_SIZE(sb)
2396             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2397                 oldest_start =
2398                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2399                     le32_to_cpu(jh->j_first_unflushed_offset);
2400                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2401                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2402                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2403                                "journal-1153: found in "
2404                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2405                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2406                                le32_to_cpu(jh->j_last_flush_trans_id));
2407                 valid_journal_header = 1;
2408
2409                 /* now, we try to read the first unflushed offset.  If it is not valid,
2410                  ** there is nothing more we can do, and it makes no sense to read
2411                  ** through the whole log.
2412                  */
2413                 d_bh =
2414                     journal_bread(sb,
2415                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2416                                   le32_to_cpu(jh->j_first_unflushed_offset));
2417                 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2418                 if (!ret) {
2419                         continue_replay = 0;
2420                 }
2421                 brelse(d_bh);
2422                 goto start_log_replay;
2423         }
2424
2425         if (continue_replay && bdev_read_only(sb->s_bdev)) {
2426                 reiserfs_warning(sb, "clm-2076",
2427                                  "device is readonly, unable to replay log");
2428                 return -1;
2429         }
2430
2431         /* ok, there are transactions that need to be replayed.  start with the first log block, find
2432          ** all the valid transactions, and pick out the oldest.
2433          */
2434         while (continue_replay
2435                && cur_dblock <
2436                (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2437                 SB_ONDISK_JOURNAL_SIZE(sb))) {
2438                 /* Note that it is required for blocksize of primary fs device and journal
2439                    device to be the same */
2440                 d_bh =
2441                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2442                                     sb->s_blocksize,
2443                                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2444                                     SB_ONDISK_JOURNAL_SIZE(sb));
2445                 ret =
2446                     journal_transaction_is_valid(sb, d_bh,
2447                                                  &oldest_invalid_trans_id,
2448                                                  &newest_mount_id);
2449                 if (ret == 1) {
2450                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2451                         if (oldest_start == 0) {        /* init all oldest_ values */
2452                                 oldest_trans_id = get_desc_trans_id(desc);
2453                                 oldest_start = d_bh->b_blocknr;
2454                                 newest_mount_id = get_desc_mount_id(desc);
2455                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2456                                                "journal-1179: Setting "
2457                                                "oldest_start to offset %llu, trans_id %lu",
2458                                                oldest_start -
2459                                                SB_ONDISK_JOURNAL_1st_BLOCK
2460                                                (sb), oldest_trans_id);
2461                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2462                                 /* one we just read was older */
2463                                 oldest_trans_id = get_desc_trans_id(desc);
2464                                 oldest_start = d_bh->b_blocknr;
2465                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2466                                                "journal-1180: Resetting "
2467                                                "oldest_start to offset %lu, trans_id %lu",
2468                                                oldest_start -
2469                                                SB_ONDISK_JOURNAL_1st_BLOCK
2470                                                (sb), oldest_trans_id);
2471                         }
2472                         if (newest_mount_id < get_desc_mount_id(desc)) {
2473                                 newest_mount_id = get_desc_mount_id(desc);
2474                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2475                                                "journal-1299: Setting "
2476                                                "newest_mount_id to %d",
2477                                                get_desc_mount_id(desc));
2478                         }
2479                         cur_dblock += get_desc_trans_len(desc) + 2;
2480                 } else {
2481                         cur_dblock++;
2482                 }
2483                 brelse(d_bh);
2484         }
2485
2486       start_log_replay:
2487         cur_dblock = oldest_start;
2488         if (oldest_trans_id) {
2489                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2490                                "journal-1206: Starting replay "
2491                                "from offset %llu, trans_id %lu",
2492                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2493                                oldest_trans_id);
2494
2495         }
2496         replay_count = 0;
2497         while (continue_replay && oldest_trans_id > 0) {
2498                 ret =
2499                     journal_read_transaction(sb, cur_dblock, oldest_start,
2500                                              oldest_trans_id, newest_mount_id);
2501                 if (ret < 0) {
2502                         return ret;
2503                 } else if (ret != 0) {
2504                         break;
2505                 }
2506                 cur_dblock =
2507                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2508                 replay_count++;
2509                 if (cur_dblock == oldest_start)
2510                         break;
2511         }
2512
2513         if (oldest_trans_id == 0) {
2514                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2515                                "journal-1225: No valid " "transactions found");
2516         }
2517         /* j_start does not get set correctly if we don't replay any transactions.
2518          ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2519          ** copy the trans_id from the header
2520          */
2521         if (valid_journal_header && replay_count == 0) {
2522                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2523                 journal->j_trans_id =
2524                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2525                 /* check for trans_id overflow */
2526                 if (journal->j_trans_id == 0)
2527                         journal->j_trans_id = 10;
2528                 journal->j_last_flush_trans_id =
2529                     le32_to_cpu(jh->j_last_flush_trans_id);
2530                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2531         } else {
2532                 journal->j_mount_id = newest_mount_id + 1;
2533         }
2534         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2535                        "newest_mount_id to %lu", journal->j_mount_id);
2536         journal->j_first_unflushed_offset = journal->j_start;
2537         if (replay_count > 0) {
2538                 reiserfs_info(sb,
2539                               "replayed %d transactions in %lu seconds\n",
2540                               replay_count, get_seconds() - start);
2541         }
2542         if (!bdev_read_only(sb->s_bdev) &&
2543             _update_journal_header_block(sb, journal->j_start,
2544                                          journal->j_last_flush_trans_id)) {
2545                 /* replay failed, caller must call free_journal_ram and abort
2546                  ** the mount
2547                  */
2548                 return -1;
2549         }
2550         return 0;
2551 }
2552
2553 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2554 {
2555         struct reiserfs_journal_list *jl;
2556         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2557                      GFP_NOFS | __GFP_NOFAIL);
2558         INIT_LIST_HEAD(&jl->j_list);
2559         INIT_LIST_HEAD(&jl->j_working_list);
2560         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2561         INIT_LIST_HEAD(&jl->j_bh_list);
2562         mutex_init(&jl->j_commit_mutex);
2563         SB_JOURNAL(s)->j_num_lists++;
2564         get_journal_list(jl);
2565         return jl;
2566 }
2567
2568 static void journal_list_init(struct super_block *sb)
2569 {
2570         SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2571 }
2572
2573 static int release_journal_dev(struct super_block *super,
2574                                struct reiserfs_journal *journal)
2575 {
2576         int result;
2577
2578         result = 0;
2579
2580         if (journal->j_dev_bd != NULL) {
2581                 if (journal->j_dev_bd->bd_dev != super->s_dev)
2582                         bd_release(journal->j_dev_bd);
2583                 result = blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2584                 journal->j_dev_bd = NULL;
2585         }
2586
2587         if (result != 0) {
2588                 reiserfs_warning(super, "sh-457",
2589                                  "Cannot release journal device: %i", result);
2590         }
2591         return result;
2592 }
2593
2594 static int journal_init_dev(struct super_block *super,
2595                             struct reiserfs_journal *journal,
2596                             const char *jdev_name)
2597 {
2598         int result;
2599         dev_t jdev;
2600         fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE;
2601         char b[BDEVNAME_SIZE];
2602
2603         result = 0;
2604
2605         journal->j_dev_bd = NULL;
2606         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2607             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2608
2609         if (bdev_read_only(super->s_bdev))
2610                 blkdev_mode = FMODE_READ;
2611
2612         /* there is no "jdev" option and journal is on separate device */
2613         if ((!jdev_name || !jdev_name[0])) {
2614                 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2615                 journal->j_dev_mode = blkdev_mode;
2616                 if (IS_ERR(journal->j_dev_bd)) {
2617                         result = PTR_ERR(journal->j_dev_bd);
2618                         journal->j_dev_bd = NULL;
2619                         reiserfs_warning(super, "sh-458",
2620                                          "cannot init journal device '%s': %i",
2621                                          __bdevname(jdev, b), result);
2622                         return result;
2623                 } else if (jdev != super->s_dev) {
2624                         result = bd_claim(journal->j_dev_bd, journal);
2625                         if (result) {
2626                                 blkdev_put(journal->j_dev_bd, blkdev_mode);
2627                                 return result;
2628                         }
2629
2630                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2631                 }
2632
2633                 return 0;
2634         }
2635
2636         journal->j_dev_mode = blkdev_mode;
2637         journal->j_dev_bd = open_bdev_exclusive(jdev_name,
2638                                                 blkdev_mode, journal);
2639         if (IS_ERR(journal->j_dev_bd)) {
2640                 result = PTR_ERR(journal->j_dev_bd);
2641                 journal->j_dev_bd = NULL;
2642                 reiserfs_warning(super,
2643                                  "journal_init_dev: Cannot open '%s': %i",
2644                                  jdev_name, result);
2645                 return result;
2646         }
2647
2648         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2649         reiserfs_info(super,
2650                       "journal_init_dev: journal device: %s\n",
2651                       bdevname(journal->j_dev_bd, b));
2652         return 0;
2653 }
2654
2655 /**
2656  * When creating/tuning a file system user can assign some
2657  * journal params within boundaries which depend on the ratio
2658  * blocksize/standard_blocksize.
2659  *
2660  * For blocks >= standard_blocksize transaction size should
2661  * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2662  * then JOURNAL_TRANS_MAX_DEFAULT.
2663  *
2664  * For blocks < standard_blocksize these boundaries should be
2665  * decreased proportionally.
2666  */
2667 #define REISERFS_STANDARD_BLKSIZE (4096)
2668
2669 static int check_advise_trans_params(struct super_block *sb,
2670                                      struct reiserfs_journal *journal)
2671 {
2672         if (journal->j_trans_max) {
2673                 /* Non-default journal params.
2674                    Do sanity check for them. */
2675                 int ratio = 1;
2676                 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2677                         ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2678
2679                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2680                     journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2681                     SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2682                     JOURNAL_MIN_RATIO) {
2683                         reiserfs_warning(sb, "sh-462",
2684                                          "bad transaction max size (%u). "
2685                                          "FSCK?", journal->j_trans_max);
2686                         return 1;
2687                 }
2688                 if (journal->j_max_batch != (journal->j_trans_max) *
2689                         JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2690                         reiserfs_warning(sb, "sh-463",
2691                                          "bad transaction max batch (%u). "
2692                                          "FSCK?", journal->j_max_batch);
2693                         return 1;
2694                 }
2695         } else {
2696                 /* Default journal params.
2697                    The file system was created by old version
2698                    of mkreiserfs, so some fields contain zeros,
2699                    and we need to advise proper values for them */
2700                 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2701                         reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2702                                          sb->s_blocksize);
2703                         return 1;
2704                 }
2705                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2706                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2707                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2708         }
2709         return 0;
2710 }
2711
2712 /*
2713 ** must be called once on fs mount.  calls journal_read for you
2714 */
2715 int journal_init(struct super_block *sb, const char *j_dev_name,
2716                  int old_format, unsigned int commit_max_age)
2717 {
2718         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2719         struct buffer_head *bhjh;
2720         struct reiserfs_super_block *rs;
2721         struct reiserfs_journal_header *jh;
2722         struct reiserfs_journal *journal;
2723         struct reiserfs_journal_list *jl;
2724         char b[BDEVNAME_SIZE];
2725
2726         journal = SB_JOURNAL(sb) = vmalloc(sizeof(struct reiserfs_journal));
2727         if (!journal) {
2728                 reiserfs_warning(sb, "journal-1256",
2729                                  "unable to get memory for journal structure");
2730                 return 1;
2731         }
2732         memset(journal, 0, sizeof(struct reiserfs_journal));
2733         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2734         INIT_LIST_HEAD(&journal->j_prealloc_list);
2735         INIT_LIST_HEAD(&journal->j_working_list);
2736         INIT_LIST_HEAD(&journal->j_journal_list);
2737         journal->j_persistent_trans = 0;
2738         if (reiserfs_allocate_list_bitmaps(sb,
2739                                            journal->j_list_bitmap,
2740                                            reiserfs_bmap_count(sb)))
2741                 goto free_and_return;
2742         allocate_bitmap_nodes(sb);
2743
2744         /* reserved for journal area support */
2745         SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2746                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2747                                                  / sb->s_blocksize +
2748                                                  reiserfs_bmap_count(sb) +
2749                                                  1 :
2750                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2751                                                  sb->s_blocksize + 2);
2752
2753         /* Sanity check to see is the standard journal fitting withing first bitmap
2754            (actual for small blocksizes) */
2755         if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2756             (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2757              SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2758                 reiserfs_warning(sb, "journal-1393",
2759                                  "journal does not fit for area addressed "
2760                                  "by first of bitmap blocks. It starts at "
2761                                  "%u and its size is %u. Block size %ld",
2762                                  SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2763                                  SB_ONDISK_JOURNAL_SIZE(sb),
2764                                  sb->s_blocksize);
2765                 goto free_and_return;
2766         }
2767
2768         if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2769                 reiserfs_warning(sb, "sh-462",
2770                                  "unable to initialize jornal device");
2771                 goto free_and_return;
2772         }
2773
2774         rs = SB_DISK_SUPER_BLOCK(sb);
2775
2776         /* read journal header */
2777         bhjh = journal_bread(sb,
2778                              SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2779                              SB_ONDISK_JOURNAL_SIZE(sb));
2780         if (!bhjh) {
2781                 reiserfs_warning(sb, "sh-459",
2782                                  "unable to read journal header");
2783                 goto free_and_return;
2784         }
2785         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2786
2787         /* make sure that journal matches to the super block */
2788         if (is_reiserfs_jr(rs)
2789             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2790                 sb_jp_journal_magic(rs))) {
2791                 reiserfs_warning(sb, "sh-460",
2792                                  "journal header magic %x (device %s) does "
2793                                  "not match to magic found in super block %x",
2794                                  jh->jh_journal.jp_journal_magic,
2795                                  bdevname(journal->j_dev_bd, b),
2796                                  sb_jp_journal_magic(rs));
2797                 brelse(bhjh);
2798                 goto free_and_return;
2799         }
2800
2801         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2802         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2803         journal->j_max_commit_age =
2804             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2805         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2806
2807         if (check_advise_trans_params(sb, journal) != 0)
2808                 goto free_and_return;
2809         journal->j_default_max_commit_age = journal->j_max_commit_age;
2810
2811         if (commit_max_age != 0) {
2812                 journal->j_max_commit_age = commit_max_age;
2813                 journal->j_max_trans_age = commit_max_age;
2814         }
2815
2816         reiserfs_info(sb, "journal params: device %s, size %u, "
2817                       "journal first block %u, max trans len %u, max batch %u, "
2818                       "max commit age %u, max trans age %u\n",
2819                       bdevname(journal->j_dev_bd, b),
2820                       SB_ONDISK_JOURNAL_SIZE(sb),
2821                       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2822                       journal->j_trans_max,
2823                       journal->j_max_batch,
2824                       journal->j_max_commit_age, journal->j_max_trans_age);
2825
2826         brelse(bhjh);
2827
2828         journal->j_list_bitmap_index = 0;
2829         journal_list_init(sb);
2830
2831         memset(journal->j_list_hash_table, 0,
2832                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2833
2834         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2835         spin_lock_init(&journal->j_dirty_buffers_lock);
2836
2837         journal->j_start = 0;
2838         journal->j_len = 0;
2839         journal->j_len_alloc = 0;
2840         atomic_set(&(journal->j_wcount), 0);
2841         atomic_set(&(journal->j_async_throttle), 0);
2842         journal->j_bcount = 0;
2843         journal->j_trans_start_time = 0;
2844         journal->j_last = NULL;
2845         journal->j_first = NULL;
2846         init_waitqueue_head(&(journal->j_join_wait));
2847         mutex_init(&journal->j_mutex);
2848         mutex_init(&journal->j_flush_mutex);
2849
2850         journal->j_trans_id = 10;
2851         journal->j_mount_id = 10;
2852         journal->j_state = 0;
2853         atomic_set(&(journal->j_jlock), 0);
2854         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2855         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2856         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2857         journal->j_cnode_used = 0;
2858         journal->j_must_wait = 0;
2859
2860         if (journal->j_cnode_free == 0) {
2861                 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2862                                  "allocation failed (%ld bytes). Journal is "
2863                                  "too large for available memory. Usually "
2864                                  "this is due to a journal that is too large.",
2865                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2866                 goto free_and_return;
2867         }
2868
2869         init_journal_hash(sb);
2870         jl = journal->j_current_jl;
2871         jl->j_list_bitmap = get_list_bitmap(sb, jl);
2872         if (!jl->j_list_bitmap) {
2873                 reiserfs_warning(sb, "journal-2005",
2874                                  "get_list_bitmap failed for journal list 0");
2875                 goto free_and_return;
2876         }
2877         if (journal_read(sb) < 0) {
2878                 reiserfs_warning(sb, "reiserfs-2006",
2879                                  "Replay Failure, unable to mount");
2880                 goto free_and_return;
2881         }
2882
2883         reiserfs_mounted_fs_count++;
2884         if (reiserfs_mounted_fs_count <= 1)
2885                 commit_wq = create_workqueue("reiserfs");
2886
2887         INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2888         journal->j_work_sb = sb;
2889         return 0;
2890       free_and_return:
2891         free_journal_ram(sb);
2892         return 1;
2893 }
2894
2895 /*
2896 ** test for a polite end of the current transaction.  Used by file_write, and should
2897 ** be used by delete to make sure they don't write more than can fit inside a single
2898 ** transaction
2899 */
2900 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2901                                    int new_alloc)
2902 {
2903         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2904         time_t now = get_seconds();
2905         /* cannot restart while nested */
2906         BUG_ON(!th->t_trans_id);
2907         if (th->t_refcount > 1)
2908                 return 0;
2909         if (journal->j_must_wait > 0 ||
2910             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2911             atomic_read(&(journal->j_jlock)) ||
2912             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2913             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2914                 return 1;
2915         }
2916         /* protected by the BKL here */
2917         journal->j_len_alloc += new_alloc;
2918         th->t_blocks_allocated += new_alloc ;
2919         return 0;
2920 }
2921
2922 /* this must be called inside a transaction, and requires the
2923 ** kernel_lock to be held
2924 */
2925 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2926 {
2927         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2928         BUG_ON(!th->t_trans_id);
2929         journal->j_must_wait = 1;
2930         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2931         return;
2932 }
2933
2934 /* this must be called without a transaction started, and does not
2935 ** require BKL
2936 */
2937 void reiserfs_allow_writes(struct super_block *s)
2938 {
2939         struct reiserfs_journal *journal = SB_JOURNAL(s);
2940         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2941         wake_up(&journal->j_join_wait);
2942 }
2943
2944 /* this must be called without a transaction started, and does not
2945 ** require BKL
2946 */
2947 void reiserfs_wait_on_write_block(struct super_block *s)
2948 {
2949         struct reiserfs_journal *journal = SB_JOURNAL(s);
2950         wait_event(journal->j_join_wait,
2951                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2952 }
2953
2954 static void queue_log_writer(struct super_block *s)
2955 {
2956         wait_queue_t wait;
2957         struct reiserfs_journal *journal = SB_JOURNAL(s);
2958         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2959
2960         /*
2961          * we don't want to use wait_event here because
2962          * we only want to wait once.
2963          */
2964         init_waitqueue_entry(&wait, current);
2965         add_wait_queue(&journal->j_join_wait, &wait);
2966         set_current_state(TASK_UNINTERRUPTIBLE);
2967         if (test_bit(J_WRITERS_QUEUED, &journal->j_state))
2968                 schedule();
2969         __set_current_state(TASK_RUNNING);
2970         remove_wait_queue(&journal->j_join_wait, &wait);
2971 }
2972
2973 static void wake_queued_writers(struct super_block *s)
2974 {
2975         struct reiserfs_journal *journal = SB_JOURNAL(s);
2976         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2977                 wake_up(&journal->j_join_wait);
2978 }
2979
2980 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
2981 {
2982         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2983         unsigned long bcount = journal->j_bcount;
2984         while (1) {
2985                 schedule_timeout_uninterruptible(1);
2986                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2987                 while ((atomic_read(&journal->j_wcount) > 0 ||
2988                         atomic_read(&journal->j_jlock)) &&
2989                        journal->j_trans_id == trans_id) {
2990                         queue_log_writer(sb);
2991                 }
2992                 if (journal->j_trans_id != trans_id)
2993                         break;
2994                 if (bcount == journal->j_bcount)
2995                         break;
2996                 bcount = journal->j_bcount;
2997         }
2998 }
2999
3000 /* join == true if you must join an existing transaction.
3001 ** join == false if you can deal with waiting for others to finish
3002 **
3003 ** this will block until the transaction is joinable.  send the number of blocks you
3004 ** expect to use in nblocks.
3005 */
3006 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3007                               struct super_block *sb, unsigned long nblocks,
3008                               int join)
3009 {
3010         time_t now = get_seconds();
3011         unsigned int old_trans_id;
3012         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3013         struct reiserfs_transaction_handle myth;
3014         int sched_count = 0;
3015         int retval;
3016
3017         reiserfs_check_lock_depth(sb, "journal_begin");
3018         BUG_ON(nblocks > journal->j_trans_max);
3019
3020         PROC_INFO_INC(sb, journal.journal_being);
3021         /* set here for journal_join */
3022         th->t_refcount = 1;
3023         th->t_super = sb;
3024
3025       relock:
3026         lock_journal(sb);
3027         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3028                 unlock_journal(sb);
3029                 retval = journal->j_errno;
3030                 goto out_fail;
3031         }
3032         journal->j_bcount++;
3033
3034         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3035                 unlock_journal(sb);
3036                 reiserfs_wait_on_write_block(sb);
3037                 PROC_INFO_INC(sb, journal.journal_relock_writers);
3038                 goto relock;
3039         }
3040         now = get_seconds();
3041
3042         /* if there is no room in the journal OR
3043          ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
3044          ** we don't sleep if there aren't other writers
3045          */
3046
3047         if ((!join && journal->j_must_wait > 0) ||
3048             (!join
3049              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3050             || (!join && atomic_read(&journal->j_wcount) > 0
3051                 && journal->j_trans_start_time > 0
3052                 && (now - journal->j_trans_start_time) >
3053                 journal->j_max_trans_age) || (!join
3054                                               && atomic_read(&journal->j_jlock))
3055             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3056
3057                 old_trans_id = journal->j_trans_id;
3058                 unlock_journal(sb);     /* allow others to finish this transaction */
3059
3060                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3061                     journal->j_max_batch &&
3062                     ((journal->j_len + nblocks + 2) * 100) <
3063                     (journal->j_len_alloc * 75)) {
3064                         if (atomic_read(&journal->j_wcount) > 10) {
3065                                 sched_count++;
3066                                 queue_log_writer(sb);
3067                                 goto relock;
3068                         }
3069                 }
3070                 /* don't mess with joining the transaction if all we have to do is
3071                  * wait for someone else to do a commit
3072                  */
3073                 if (atomic_read(&journal->j_jlock)) {
3074                         while (journal->j_trans_id == old_trans_id &&
3075                                atomic_read(&journal->j_jlock)) {
3076                                 queue_log_writer(sb);
3077                         }
3078                         goto relock;
3079                 }
3080                 retval = journal_join(&myth, sb, 1);
3081                 if (retval)
3082                         goto out_fail;
3083
3084                 /* someone might have ended the transaction while we joined */
3085                 if (old_trans_id != journal->j_trans_id) {
3086                         retval = do_journal_end(&myth, sb, 1, 0);
3087                 } else {
3088                         retval = do_journal_end(&myth, sb, 1, COMMIT_NOW);
3089                 }
3090
3091                 if (retval)
3092                         goto out_fail;
3093
3094                 PROC_INFO_INC(sb, journal.journal_relock_wcount);
3095                 goto relock;
3096         }
3097         /* we are the first writer, set trans_id */
3098         if (journal->j_trans_start_time == 0) {
3099                 journal->j_trans_start_time = get_seconds();
3100         }
3101         atomic_inc(&(journal->j_wcount));
3102         journal->j_len_alloc += nblocks;
3103         th->t_blocks_logged = 0;
3104         th->t_blocks_allocated = nblocks;
3105         th->t_trans_id = journal->j_trans_id;
3106         unlock_journal(sb);
3107         INIT_LIST_HEAD(&th->t_list);
3108         get_fs_excl();
3109         return 0;
3110
3111       out_fail:
3112         memset(th, 0, sizeof(*th));
3113         /* Re-set th->t_super, so we can properly keep track of how many
3114          * persistent transactions there are. We need to do this so if this
3115          * call is part of a failed restart_transaction, we can free it later */
3116         th->t_super = sb;
3117         return retval;
3118 }
3119
3120 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3121                                                                     super_block
3122                                                                     *s,
3123                                                                     int nblocks)
3124 {
3125         int ret;
3126         struct reiserfs_transaction_handle *th;
3127
3128         /* if we're nesting into an existing transaction.  It will be
3129          ** persistent on its own
3130          */
3131         if (reiserfs_transaction_running(s)) {
3132                 th = current->journal_info;
3133                 th->t_refcount++;
3134                 BUG_ON(th->t_refcount < 2);
3135                 
3136                 return th;
3137         }
3138         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3139         if (!th)
3140                 return NULL;
3141         ret = journal_begin(th, s, nblocks);
3142         if (ret) {
3143                 kfree(th);
3144                 return NULL;
3145         }
3146
3147         SB_JOURNAL(s)->j_persistent_trans++;
3148         return th;
3149 }
3150
3151 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3152 {
3153         struct super_block *s = th->t_super;
3154         int ret = 0;
3155         if (th->t_trans_id)
3156                 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3157         else
3158                 ret = -EIO;
3159         if (th->t_refcount == 0) {
3160                 SB_JOURNAL(s)->j_persistent_trans--;
3161                 kfree(th);
3162         }
3163         return ret;
3164 }
3165
3166 static int journal_join(struct reiserfs_transaction_handle *th,
3167                         struct super_block *sb, unsigned long nblocks)
3168 {
3169         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3170
3171         /* this keeps do_journal_end from NULLing out the current->journal_info
3172          ** pointer
3173          */
3174         th->t_handle_save = cur_th;
3175         BUG_ON(cur_th && cur_th->t_refcount > 1);
3176         return do_journal_begin_r(th, sb, nblocks, JBEGIN_JOIN);
3177 }
3178
3179 int journal_join_abort(struct reiserfs_transaction_handle *th,
3180                        struct super_block *sb, unsigned long nblocks)
3181 {
3182         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3183
3184         /* this keeps do_journal_end from NULLing out the current->journal_info
3185          ** pointer
3186          */
3187         th->t_handle_save = cur_th;
3188         BUG_ON(cur_th && cur_th->t_refcount > 1);
3189         return do_journal_begin_r(th, sb, nblocks, JBEGIN_ABORT);
3190 }
3191
3192 int journal_begin(struct reiserfs_transaction_handle *th,
3193                   struct super_block *sb, unsigned long nblocks)
3194 {
3195         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3196         int ret;
3197
3198         th->t_handle_save = NULL;
3199         if (cur_th) {
3200                 /* we are nesting into the current transaction */
3201                 if (cur_th->t_super == sb) {
3202                         BUG_ON(!cur_th->t_refcount);
3203                         cur_th->t_refcount++;
3204                         memcpy(th, cur_th, sizeof(*th));
3205                         if (th->t_refcount <= 1)
3206                                 reiserfs_warning(sb, "reiserfs-2005",
3207                                                  "BAD: refcount <= 1, but "
3208                                                  "journal_info != 0");
3209                         return 0;
3210                 } else {
3211                         /* we've ended up with a handle from a different filesystem.
3212                          ** save it and restore on journal_end.  This should never
3213                          ** really happen...
3214                          */
3215                         reiserfs_warning(sb, "clm-2100",
3216                                          "nesting info a different FS");
3217                         th->t_handle_save = current->journal_info;
3218                         current->journal_info = th;
3219                 }
3220         } else {
3221                 current->journal_info = th;
3222         }
3223         ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3224         BUG_ON(current->journal_info != th);
3225
3226         /* I guess this boils down to being the reciprocal of clm-2100 above.
3227          * If do_journal_begin_r fails, we need to put it back, since journal_end
3228          * won't be called to do it. */
3229         if (ret)
3230                 current->journal_info = th->t_handle_save;
3231         else
3232                 BUG_ON(!th->t_refcount);
3233
3234         return ret;
3235 }
3236
3237 /*
3238 ** puts bh into the current transaction.  If it was already there, reorders removes the
3239 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3240 **
3241 ** if it was dirty, cleans and files onto the clean list.  I can't let it be dirty again until the
3242 ** transaction is committed.
3243 **
3244 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3245 */
3246 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3247                        struct super_block *sb, struct buffer_head *bh)
3248 {
3249         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3250         struct reiserfs_journal_cnode *cn = NULL;
3251         int count_already_incd = 0;
3252         int prepared = 0;
3253         BUG_ON(!th->t_trans_id);
3254
3255         PROC_INFO_INC(sb, journal.mark_dirty);
3256         if (th->t_trans_id != journal->j_trans_id) {
3257                 reiserfs_panic(th->t_super, "journal-1577",
3258                                "handle trans id %ld != current trans id %ld",
3259                                th->t_trans_id, journal->j_trans_id);
3260         }
3261
3262         sb->s_dirt = 1;
3263
3264         prepared = test_clear_buffer_journal_prepared(bh);
3265         clear_buffer_journal_restore_dirty(bh);
3266         /* already in this transaction, we are done */
3267         if (buffer_journaled(bh)) {
3268                 PROC_INFO_INC(sb, journal.mark_dirty_already);
3269                 return 0;
3270         }
3271
3272         /* this must be turned into a panic instead of a warning.  We can't allow
3273          ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3274          ** could get to disk too early.  NOT GOOD.
3275          */
3276         if (!prepared || buffer_dirty(bh)) {
3277                 reiserfs_warning(sb, "journal-1777",
3278                                  "buffer %llu bad state "
3279                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3280                                  (unsigned long long)bh->b_blocknr,
3281                                  prepared ? ' ' : '!',
3282                                  buffer_locked(bh) ? ' ' : '!',
3283                                  buffer_dirty(bh) ? ' ' : '!',
3284                                  buffer_journal_dirty(bh) ? ' ' : '!');
3285         }
3286
3287         if (atomic_read(&(journal->j_wcount)) <= 0) {
3288                 reiserfs_warning(sb, "journal-1409",
3289                                  "returning because j_wcount was %d",
3290                                  atomic_read(&(journal->j_wcount)));
3291                 return 1;
3292         }
3293         /* this error means I've screwed up, and we've overflowed the transaction.
3294          ** Nothing can be done here, except make the FS readonly or panic.
3295          */
3296         if (journal->j_len >= journal->j_trans_max) {
3297                 reiserfs_panic(th->t_super, "journal-1413",
3298                                "j_len (%lu) is too big",
3299                                journal->j_len);
3300         }
3301
3302         if (buffer_journal_dirty(bh)) {
3303                 count_already_incd = 1;
3304                 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3305                 clear_buffer_journal_dirty(bh);
3306         }
3307
3308         if (journal->j_len > journal->j_len_alloc) {
3309                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3310         }
3311
3312         set_buffer_journaled(bh);
3313
3314         /* now put this guy on the end */
3315         if (!cn) {
3316                 cn = get_cnode(sb);
3317                 if (!cn) {
3318                         reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3319                 }
3320
3321                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3322                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3323                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3324                 }
3325                 th->t_blocks_logged++;
3326                 journal->j_len++;
3327
3328                 cn->bh = bh;
3329                 cn->blocknr = bh->b_blocknr;
3330                 cn->sb = sb;
3331                 cn->jlist = NULL;
3332                 insert_journal_hash(journal->j_hash_table, cn);
3333                 if (!count_already_incd) {
3334                         get_bh(bh);
3335                 }
3336         }
3337         cn->next = NULL;
3338         cn->prev = journal->j_last;
3339         cn->bh = bh;
3340         if (journal->j_last) {
3341                 journal->j_last->next = cn;
3342                 journal->j_last = cn;
3343         } else {
3344                 journal->j_first = cn;
3345                 journal->j_last = cn;
3346         }
3347         return 0;
3348 }
3349
3350 int journal_end(struct reiserfs_transaction_handle *th,
3351                 struct super_block *sb, unsigned long nblocks)
3352 {
3353         if (!current->journal_info && th->t_refcount > 1)
3354                 reiserfs_warning(sb, "REISER-NESTING",
3355                                  "th NULL, refcount %d", th->t_refcount);
3356
3357         if (!th->t_trans_id) {
3358                 WARN_ON(1);
3359                 return -EIO;
3360         }
3361
3362         th->t_refcount--;
3363         if (th->t_refcount > 0) {
3364                 struct reiserfs_transaction_handle *cur_th =
3365                     current->journal_info;
3366
3367                 /* we aren't allowed to close a nested transaction on a different
3368                  ** filesystem from the one in the task struct
3369                  */
3370                 BUG_ON(cur_th->t_super != th->t_super);
3371
3372                 if (th != cur_th) {
3373                         memcpy(current->journal_info, th, sizeof(*th));
3374                         th->t_trans_id = 0;
3375                 }
3376                 return 0;
3377         } else {
3378                 return do_journal_end(th, sb, nblocks, 0);
3379         }
3380 }
3381
3382 /* removes from the current transaction, relsing and descrementing any counters.
3383 ** also files the removed buffer directly onto the clean list
3384 **
3385 ** called by journal_mark_freed when a block has been deleted
3386 **
3387 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3388 */
3389 static int remove_from_transaction(struct super_block *sb,
3390                                    b_blocknr_t blocknr, int already_cleaned)
3391 {
3392         struct buffer_head *bh;
3393         struct reiserfs_journal_cnode *cn;
3394         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3395         int ret = 0;
3396
3397         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3398         if (!cn || !cn->bh) {
3399                 return ret;
3400         }
3401         bh = cn->bh;
3402         if (cn->prev) {
3403                 cn->prev->next = cn->next;
3404         }
3405         if (cn->next) {
3406                 cn->next->prev = cn->prev;
3407         }
3408         if (cn == journal->j_first) {
3409                 journal->j_first = cn->next;
3410         }
3411         if (cn == journal->j_last) {
3412                 journal->j_last = cn->prev;
3413         }
3414         if (bh)
3415                 remove_journal_hash(sb, journal->j_hash_table, NULL,
3416                                     bh->b_blocknr, 0);
3417         clear_buffer_journaled(bh);     /* don't log this one */
3418
3419         if (!already_cleaned) {
3420                 clear_buffer_journal_dirty(bh);
3421                 clear_buffer_dirty(bh);
3422                 clear_buffer_journal_test(bh);
3423                 put_bh(bh);
3424                 if (atomic_read(&(bh->b_count)) < 0) {
3425                         reiserfs_warning(sb, "journal-1752",
3426                                          "b_count < 0");
3427                 }
3428                 ret = 1;
3429         }
3430         journal->j_len--;
3431         journal->j_len_alloc--;
3432         free_cnode(sb, cn);
3433         return ret;
3434 }
3435
3436 /*
3437 ** for any cnode in a journal list, it can only be dirtied of all the
3438 ** transactions that include it are committed to disk.
3439 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3440 ** and 0 if you aren't
3441 **
3442 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3443 ** blocks for a given transaction on disk
3444 **
3445 */
3446 static int can_dirty(struct reiserfs_journal_cnode *cn)
3447 {
3448         struct super_block *sb = cn->sb;
3449         b_blocknr_t blocknr = cn->blocknr;
3450         struct reiserfs_journal_cnode *cur = cn->hprev;
3451         int can_dirty = 1;
3452
3453         /* first test hprev.  These are all newer than cn, so any node here
3454          ** with the same block number and dev means this node can't be sent
3455          ** to disk right now.
3456          */
3457         while (cur && can_dirty) {
3458                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3459                     cur->blocknr == blocknr) {
3460                         can_dirty = 0;
3461                 }
3462                 cur = cur->hprev;
3463         }
3464         /* then test hnext.  These are all older than cn.  As long as they
3465          ** are committed to the log, it is safe to write cn to disk
3466          */
3467         cur = cn->hnext;
3468         while (cur && can_dirty) {
3469                 if (cur->jlist && cur->jlist->j_len > 0 &&
3470                     atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3471                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3472                         can_dirty = 0;
3473                 }
3474                 cur = cur->hnext;
3475         }
3476         return can_dirty;
3477 }
3478
3479 /* syncs the commit blocks, but does not force the real buffers to disk
3480 ** will wait until the current transaction is done/committed before returning
3481 */
3482 int journal_end_sync(struct reiserfs_transaction_handle *th,
3483                      struct super_block *sb, unsigned long nblocks)
3484 {
3485         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3486
3487         BUG_ON(!th->t_trans_id);
3488         /* you can sync while nested, very, very bad */
3489         BUG_ON(th->t_refcount > 1);
3490         if (journal->j_len == 0) {
3491                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3492                                              1);
3493                 journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
3494         }
3495         return do_journal_end(th, sb, nblocks, COMMIT_NOW | WAIT);
3496 }
3497
3498 /*
3499 ** writeback the pending async commits to disk
3500 */
3501 static void flush_async_commits(struct work_struct *work)
3502 {
3503         struct reiserfs_journal *journal =
3504                 container_of(work, struct reiserfs_journal, j_work.work);
3505         struct super_block *sb = journal->j_work_sb;
3506         struct reiserfs_journal_list *jl;
3507         struct list_head *entry;
3508
3509         lock_kernel();
3510         if (!list_empty(&journal->j_journal_list)) {
3511                 /* last entry is the youngest, commit it and you get everything */
3512                 entry = journal->j_journal_list.prev;
3513                 jl = JOURNAL_LIST_ENTRY(entry);
3514                 flush_commit_list(sb, jl, 1);
3515         }
3516         unlock_kernel();
3517 }
3518
3519 /*
3520 ** flushes any old transactions to disk
3521 ** ends the current transaction if it is too old
3522 */
3523 int reiserfs_flush_old_commits(struct super_block *sb)
3524 {
3525         time_t now;
3526         struct reiserfs_transaction_handle th;
3527         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3528
3529         now = get_seconds();
3530         /* safety check so we don't flush while we are replaying the log during
3531          * mount
3532          */
3533         if (list_empty(&journal->j_journal_list)) {
3534                 return 0;
3535         }
3536
3537         /* check the current transaction.  If there are no writers, and it is
3538          * too old, finish it, and force the commit blocks to disk
3539          */
3540         if (atomic_read(&journal->j_wcount) <= 0 &&
3541             journal->j_trans_start_time > 0 &&
3542             journal->j_len > 0 &&
3543             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3544                 if (!journal_join(&th, sb, 1)) {
3545                         reiserfs_prepare_for_journal(sb,
3546                                                      SB_BUFFER_WITH_SB(sb),
3547                                                      1);
3548                         journal_mark_dirty(&th, sb,
3549                                            SB_BUFFER_WITH_SB(sb));
3550
3551                         /* we're only being called from kreiserfsd, it makes no sense to do
3552                          ** an async commit so that kreiserfsd can do it later
3553                          */
3554                         do_journal_end(&th, sb, 1, COMMIT_NOW | WAIT);
3555                 }
3556         }
3557         return sb->s_dirt;
3558 }
3559
3560 /*
3561 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3562 **
3563 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3564 ** the writers are done.  By the time it wakes up, the transaction it was called has already ended, so it just
3565 ** flushes the commit list and returns 0.
3566 **
3567 ** Won't batch when flush or commit_now is set.  Also won't batch when others are waiting on j_join_wait.
3568 **
3569 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3570 */
3571 static int check_journal_end(struct reiserfs_transaction_handle *th,
3572                              struct super_block *sb, unsigned long nblocks,
3573                              int flags)
3574 {
3575
3576         time_t now;
3577         int flush = flags & FLUSH_ALL;
3578         int commit_now = flags & COMMIT_NOW;
3579         int wait_on_commit = flags & WAIT;
3580         struct reiserfs_journal_list *jl;
3581         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3582
3583         BUG_ON(!th->t_trans_id);
3584
3585         if (th->t_trans_id != journal->j_trans_id) {
3586                 reiserfs_panic(th->t_super, "journal-1577",
3587                                "handle trans id %ld != current trans id %ld",
3588                                th->t_trans_id, journal->j_trans_id);
3589         }
3590
3591         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3592         if (atomic_read(&(journal->j_wcount)) > 0) {    /* <= 0 is allowed.  unmounting might not call begin */
3593                 atomic_dec(&(journal->j_wcount));
3594         }
3595
3596         /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3597          ** will be dealt with by next transaction that actually writes something, but should be taken
3598          ** care of in this trans
3599          */
3600         BUG_ON(journal->j_len == 0);
3601
3602         /* if wcount > 0, and we are called to with flush or commit_now,
3603          ** we wait on j_join_wait.  We will wake up when the last writer has
3604          ** finished the transaction, and started it on its way to the disk.
3605          ** Then, we flush the commit or journal list, and just return 0
3606          ** because the rest of journal end was already done for this transaction.
3607          */
3608         if (atomic_read(&(journal->j_wcount)) > 0) {
3609                 if (flush || commit_now) {
3610                         unsigned trans_id;
3611
3612                         jl = journal->j_current_jl;
3613                         trans_id = jl->j_trans_id;
3614                         if (wait_on_commit)
3615                                 jl->j_state |= LIST_COMMIT_PENDING;
3616                         atomic_set(&(journal->j_jlock), 1);
3617                         if (flush) {
3618                                 journal->j_next_full_flush = 1;
3619                         }
3620                         unlock_journal(sb);
3621
3622                         /* sleep while the current transaction is still j_jlocked */
3623                         while (journal->j_trans_id == trans_id) {
3624                                 if (atomic_read(&journal->j_jlock)) {
3625                                         queue_log_writer(sb);
3626                                 } else {
3627                                         lock_journal(sb);
3628                                         if (journal->j_trans_id == trans_id) {
3629                                                 atomic_set(&(journal->j_jlock),
3630                                                            1);
3631                                         }
3632                                         unlock_journal(sb);
3633                                 }
3634                         }
3635                         BUG_ON(journal->j_trans_id == trans_id);
3636                         
3637                         if (commit_now
3638                             && journal_list_still_alive(sb, trans_id)
3639                             && wait_on_commit) {
3640                                 flush_commit_list(sb, jl, 1);
3641                         }
3642                         return 0;
3643                 }
3644                 unlock_journal(sb);
3645                 return 0;
3646         }
3647
3648         /* deal with old transactions where we are the last writers */
3649         now = get_seconds();
3650         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3651                 commit_now = 1;
3652                 journal->j_next_async_flush = 1;
3653         }
3654         /* don't batch when someone is waiting on j_join_wait */
3655         /* don't batch when syncing the commit or flushing the whole trans */
3656         if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3657             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3658             && journal->j_len_alloc < journal->j_max_batch
3659             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3660                 journal->j_bcount++;
3661                 unlock_journal(sb);
3662                 return 0;
3663         }
3664
3665         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3666                 reiserfs_panic(sb, "journal-003",
3667                                "j_start (%ld) is too high",
3668                                journal->j_start);
3669         }
3670         return 1;
3671 }
3672
3673 /*
3674 ** Does all the work that makes deleting blocks safe.
3675 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3676 **
3677 ** otherwise:
3678 ** set a bit for the block in the journal bitmap.  That will prevent it from being allocated for unformatted nodes
3679 ** before this transaction has finished.
3680 **
3681 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.  That will prevent any old transactions with
3682 ** this block from trying to flush to the real location.  Since we aren't removing the cnode from the journal_list_hash,
3683 ** the block can't be reallocated yet.
3684 **
3685 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3686 */
3687 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3688                        struct super_block *sb, b_blocknr_t blocknr)
3689 {
3690         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3691         struct reiserfs_journal_cnode *cn = NULL;
3692         struct buffer_head *bh = NULL;
3693         struct reiserfs_list_bitmap *jb = NULL;
3694         int cleaned = 0;
3695         BUG_ON(!th->t_trans_id);
3696
3697         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3698         if (cn && cn->bh) {
3699                 bh = cn->bh;
3700                 get_bh(bh);
3701         }
3702         /* if it is journal new, we just remove it from this transaction */
3703         if (bh && buffer_journal_new(bh)) {
3704                 clear_buffer_journal_new(bh);
3705                 clear_prepared_bits(bh);
3706                 reiserfs_clean_and_file_buffer(bh);
3707                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3708         } else {
3709                 /* set the bit for this block in the journal bitmap for this transaction */
3710                 jb = journal->j_current_jl->j_list_bitmap;
3711                 if (!jb) {
3712                         reiserfs_panic(sb, "journal-1702",
3713                                        "journal_list_bitmap is NULL");
3714                 }
3715                 set_bit_in_list_bitmap(sb, blocknr, jb);
3716
3717                 /* Note, the entire while loop is not allowed to schedule.  */
3718
3719                 if (bh) {
3720                         clear_prepared_bits(bh);
3721                         reiserfs_clean_and_file_buffer(bh);
3722                 }
3723                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3724
3725                 /* find all older transactions with this block, make sure they don't try to write it out */
3726                 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3727                                           blocknr);
3728                 while (cn) {
3729                         if (sb == cn->sb && blocknr == cn->blocknr) {
3730                                 set_bit(BLOCK_FREED, &cn->state);
3731                                 if (cn->bh) {
3732                                         if (!cleaned) {
3733                                                 /* remove_from_transaction will brelse the buffer if it was 
3734                                                  ** in the current trans
3735                                                  */
3736                                                 clear_buffer_journal_dirty(cn->
3737                                                                            bh);
3738                                                 clear_buffer_dirty(cn->bh);
3739                                                 clear_buffer_journal_test(cn->
3740                                                                           bh);
3741                                                 cleaned = 1;
3742                                                 put_bh(cn->bh);
3743                                                 if (atomic_read
3744                                                     (&(cn->bh->b_count)) < 0) {
3745                                                         reiserfs_warning(sb,
3746                                                                  "journal-2138",
3747                                                                  "cn->bh->b_count < 0");
3748                                                 }
3749                                         }
3750                                         if (cn->jlist) {        /* since we are clearing the bh, we MUST dec nonzerolen */
3751                                                 atomic_dec(&
3752                                                            (cn->jlist->
3753                                                             j_nonzerolen));
3754                                         }
3755                                         cn->bh = NULL;
3756                                 }
3757                         }
3758                         cn = cn->hnext;
3759                 }
3760         }
3761
3762         if (bh)
3763                 release_buffer_page(bh); /* get_hash grabs the buffer */
3764         return 0;
3765 }
3766
3767 void reiserfs_update_inode_transaction(struct inode *inode)
3768 {
3769         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3770         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3771         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3772 }
3773
3774 /*
3775  * returns -1 on error, 0 if no commits/barriers were done and 1
3776  * if a transaction was actually committed and the barrier was done
3777  */
3778 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3779                              struct reiserfs_journal_list *jl)
3780 {
3781         struct reiserfs_transaction_handle th;
3782         struct super_block *sb = inode->i_sb;
3783         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3784         int ret = 0;
3785
3786         /* is it from the current transaction, or from an unknown transaction? */
3787         if (id == journal->j_trans_id) {
3788                 jl = journal->j_current_jl;
3789                 /* try to let other writers come in and grow this transaction */
3790                 let_transaction_grow(sb, id);
3791                 if (journal->j_trans_id != id) {
3792                         goto flush_commit_only;
3793                 }
3794
3795                 ret = journal_begin(&th, sb, 1);
3796                 if (ret)
3797                         return ret;
3798
3799                 /* someone might have ended this transaction while we joined */
3800                 if (journal->j_trans_id != id) {
3801                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3802                                                      1);
3803                         journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3804                         ret = journal_end(&th, sb, 1);
3805                         goto flush_commit_only;
3806                 }
3807
3808                 ret = journal_end_sync(&th, sb, 1);
3809                 if (!ret)
3810                         ret = 1;
3811
3812         } else {
3813                 /* this gets tricky, we have to make sure the journal list in
3814                  * the inode still exists.  We know the list is still around
3815                  * if we've got a larger transaction id than the oldest list
3816                  */
3817               flush_commit_only:
3818                 if (journal_list_still_alive(inode->i_sb, id)) {
3819                         /*
3820                          * we only set ret to 1 when we know for sure
3821                          * the barrier hasn't been started yet on the commit
3822                          * block.
3823                          */
3824                         if (atomic_read(&jl->j_commit_left) > 1)
3825                                 ret = 1;
3826                         flush_commit_list(sb, jl, 1);
3827                         if (journal->j_errno)
3828                                 ret = journal->j_errno;
3829                 }
3830         }
3831         /* otherwise the list is gone, and long since committed */
3832         return ret;
3833 }
3834
3835 int reiserfs_commit_for_inode(struct inode *inode)
3836 {
3837         unsigned int id = REISERFS_I(inode)->i_trans_id;
3838         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3839
3840         /* for the whole inode, assume unset id means it was
3841          * changed in the current transaction.  More conservative
3842          */
3843         if (!id || !jl) {
3844                 reiserfs_update_inode_transaction(inode);
3845                 id = REISERFS_I(inode)->i_trans_id;
3846                 /* jl will be updated in __commit_trans_jl */
3847         }
3848
3849         return __commit_trans_jl(inode, id, jl);
3850 }
3851
3852 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3853                                       struct buffer_head *bh)
3854 {
3855         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3856         PROC_INFO_INC(sb, journal.restore_prepared);
3857         if (!bh) {
3858                 return;
3859         }
3860         if (test_clear_buffer_journal_restore_dirty(bh) &&
3861             buffer_journal_dirty(bh)) {
3862                 struct reiserfs_journal_cnode *cn;
3863                 cn = get_journal_hash_dev(sb,
3864                                           journal->j_list_hash_table,
3865                                           bh->b_blocknr);
3866                 if (cn && can_dirty(cn)) {
3867                         set_buffer_journal_test(bh);
3868                         mark_buffer_dirty(bh);
3869                 }
3870         }
3871         clear_buffer_journal_prepared(bh);
3872 }
3873
3874 extern struct tree_balance *cur_tb;
3875 /*
3876 ** before we can change a metadata block, we have to make sure it won't
3877 ** be written to disk while we are altering it.  So, we must:
3878 ** clean it
3879 ** wait on it.
3880 **
3881 */
3882 int reiserfs_prepare_for_journal(struct super_block *sb,
3883                                  struct buffer_head *bh, int wait)
3884 {
3885         PROC_INFO_INC(sb, journal.prepare);
3886
3887         if (!trylock_buffer(bh)) {
3888                 if (!wait)
3889                         return 0;
3890                 lock_buffer(bh);
3891         }
3892         set_buffer_journal_prepared(bh);
3893         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3894                 clear_buffer_journal_test(bh);
3895                 set_buffer_journal_restore_dirty(bh);
3896         }
3897         unlock_buffer(bh);
3898         return 1;
3899 }
3900
3901 static void flush_old_journal_lists(struct super_block *s)
3902 {
3903         struct reiserfs_journal *journal = SB_JOURNAL(s);
3904         struct reiserfs_journal_list *jl;
3905         struct list_head *entry;
3906         time_t now = get_seconds();
3907
3908         while (!list_empty(&journal->j_journal_list)) {
3909                 entry = journal->j_journal_list.next;
3910                 jl = JOURNAL_LIST_ENTRY(entry);
3911                 /* this check should always be run, to send old lists to disk */
3912                 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) &&
3913                     atomic_read(&jl->j_commit_left) == 0 &&
3914                     test_transaction(s, jl)) {
3915                         flush_used_journal_lists(s, jl);
3916                 } else {
3917                         break;
3918                 }
3919         }
3920 }
3921
3922 /*
3923 ** long and ugly.  If flush, will not return until all commit
3924 ** blocks and all real buffers in the trans are on disk.
3925 ** If no_async, won't return until all commit blocks are on disk.
3926 **
3927 ** keep reading, there are comments as you go along
3928 **
3929 ** If the journal is aborted, we just clean up. Things like flushing
3930 ** journal lists, etc just won't happen.
3931 */
3932 static int do_journal_end(struct reiserfs_transaction_handle *th,
3933                           struct super_block *sb, unsigned long nblocks,
3934                           int flags)
3935 {
3936         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3937         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3938         struct reiserfs_journal_cnode *last_cn = NULL;
3939         struct reiserfs_journal_desc *desc;
3940         struct reiserfs_journal_commit *commit;
3941         struct buffer_head *c_bh;       /* commit bh */
3942         struct buffer_head *d_bh;       /* desc bh */
3943         int cur_write_start = 0;        /* start index of current log write */
3944         int old_start;
3945         int i;
3946         int flush;
3947         int wait_on_commit;
3948         struct reiserfs_journal_list *jl, *temp_jl;
3949         struct list_head *entry, *safe;
3950         unsigned long jindex;
3951         unsigned int commit_trans_id;
3952         int trans_half;
3953
3954         BUG_ON(th->t_refcount > 1);
3955         BUG_ON(!th->t_trans_id);
3956
3957         /* protect flush_older_commits from doing mistakes if the
3958            transaction ID counter gets overflowed.  */
3959         if (th->t_trans_id == ~0U)
3960                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
3961         flush = flags & FLUSH_ALL;
3962         wait_on_commit = flags & WAIT;
3963
3964         put_fs_excl();
3965         current->journal_info = th->t_handle_save;
3966         reiserfs_check_lock_depth(sb, "journal end");
3967         if (journal->j_len == 0) {
3968                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3969                                              1);
3970                 journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
3971         }
3972
3973         lock_journal(sb);
3974         if (journal->j_next_full_flush) {
3975                 flags |= FLUSH_ALL;
3976                 flush = 1;
3977         }
3978         if (journal->j_next_async_flush) {
3979                 flags |= COMMIT_NOW | WAIT;
3980                 wait_on_commit = 1;
3981         }
3982
3983         /* check_journal_end locks the journal, and unlocks if it does not return 1
3984          ** it tells us if we should continue with the journal_end, or just return
3985          */
3986         if (!check_journal_end(th, sb, nblocks, flags)) {
3987                 sb->s_dirt = 1;
3988                 wake_queued_writers(sb);
3989                 reiserfs_async_progress_wait(sb);
3990                 goto out;
3991         }
3992
3993         /* check_journal_end might set these, check again */
3994         if (journal->j_next_full_flush) {
3995                 flush = 1;
3996         }
3997
3998         /*
3999          ** j must wait means we have to flush the log blocks, and the real blocks for
4000          ** this transaction
4001          */
4002         if (journal->j_must_wait > 0) {
4003                 flush = 1;
4004         }
4005 #ifdef REISERFS_PREALLOCATE
4006         /* quota ops might need to nest, setup the journal_info pointer for them
4007          * and raise the refcount so that it is > 0. */
4008         current->journal_info = th;
4009         th->t_refcount++;
4010         reiserfs_discard_all_prealloc(th);      /* it should not involve new blocks into
4011                                                  * the transaction */
4012         th->t_refcount--;
4013         current->journal_info = th->t_handle_save;
4014 #endif
4015
4016         /* setup description block */
4017         d_bh =
4018             journal_getblk(sb,
4019                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4020                            journal->j_start);
4021         set_buffer_uptodate(d_bh);
4022         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4023         memset(d_bh->b_data, 0, d_bh->b_size);
4024         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4025         set_desc_trans_id(desc, journal->j_trans_id);
4026
4027         /* setup commit block.  Don't write (keep it clean too) this one until after everyone else is written */
4028         c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4029                               ((journal->j_start + journal->j_len +
4030                                 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4031         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4032         memset(c_bh->b_data, 0, c_bh->b_size);
4033         set_commit_trans_id(commit, journal->j_trans_id);
4034         set_buffer_uptodate(c_bh);
4035
4036         /* init this journal list */
4037         jl = journal->j_current_jl;
4038
4039         /* we lock the commit before doing anything because
4040          * we want to make sure nobody tries to run flush_commit_list until
4041          * the new transaction is fully setup, and we've already flushed the
4042          * ordered bh list
4043          */
4044         mutex_lock(&jl->j_commit_mutex);
4045
4046         /* save the transaction id in case we need to commit it later */
4047         commit_trans_id = jl->j_trans_id;
4048
4049         atomic_set(&jl->j_older_commits_done, 0);
4050         jl->j_trans_id = journal->j_trans_id;
4051         jl->j_timestamp = journal->j_trans_start_time;
4052         jl->j_commit_bh = c_bh;
4053         jl->j_start = journal->j_start;
4054         jl->j_len = journal->j_len;
4055         atomic_set(&jl->j_nonzerolen, journal->j_len);
4056         atomic_set(&jl->j_commit_left, journal->j_len + 2);
4057         jl->j_realblock = NULL;
4058
4059         /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4060          **  for each real block, add it to the journal list hash,
4061          ** copy into real block index array in the commit or desc block
4062          */
4063         trans_half = journal_trans_half(sb->s_blocksize);
4064         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4065                 if (buffer_journaled(cn->bh)) {
4066                         jl_cn = get_cnode(sb);
4067                         if (!jl_cn) {
4068                                 reiserfs_panic(sb, "journal-1676",
4069                                                "get_cnode returned NULL");
4070                         }
4071                         if (i == 0) {
4072                                 jl->j_realblock = jl_cn;
4073                         }
4074                         jl_cn->prev = last_cn;
4075                         jl_cn->next = NULL;
4076                         if (last_cn) {
4077                                 last_cn->next = jl_cn;
4078                         }
4079                         last_cn = jl_cn;
4080                         /* make sure the block we are trying to log is not a block
4081                            of journal or reserved area */
4082
4083                         if (is_block_in_log_or_reserved_area
4084                             (sb, cn->bh->b_blocknr)) {
4085                                 reiserfs_panic(sb, "journal-2332",
4086                                                "Trying to log block %lu, "
4087                                                "which is a log block",
4088                                                cn->bh->b_blocknr);
4089                         }
4090                         jl_cn->blocknr = cn->bh->b_blocknr;
4091                         jl_cn->state = 0;
4092                         jl_cn->sb = sb;
4093                         jl_cn->bh = cn->bh;
4094                         jl_cn->jlist = jl;
4095                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4096                         if (i < trans_half) {
4097                                 desc->j_realblock[i] =
4098                                     cpu_to_le32(cn->bh->b_blocknr);
4099                         } else {
4100                                 commit->j_realblock[i - trans_half] =
4101                                     cpu_to_le32(cn->bh->b_blocknr);
4102                         }
4103                 } else {
4104                         i--;
4105                 }
4106         }
4107         set_desc_trans_len(desc, journal->j_len);
4108         set_desc_mount_id(desc, journal->j_mount_id);
4109         set_desc_trans_id(desc, journal->j_trans_id);
4110         set_commit_trans_len(commit, journal->j_len);
4111
4112         /* special check in case all buffers in the journal were marked for not logging */
4113         BUG_ON(journal->j_len == 0);
4114
4115         /* we're about to dirty all the log blocks, mark the description block
4116          * dirty now too.  Don't mark the commit block dirty until all the
4117          * others are on disk
4118          */
4119         mark_buffer_dirty(d_bh);
4120
4121         /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4122         cur_write_start = journal->j_start;
4123         cn = journal->j_first;
4124         jindex = 1;             /* start at one so we don't get the desc again */
4125         while (cn) {
4126                 clear_buffer_journal_new(cn->bh);
4127                 /* copy all the real blocks into log area.  dirty log blocks */
4128                 if (buffer_journaled(cn->bh)) {
4129                         struct buffer_head *tmp_bh;
4130                         char *addr;
4131                         struct page *page;
4132                         tmp_bh =
4133                             journal_getblk(sb,
4134                                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4135                                            ((cur_write_start +
4136                                              jindex) %
4137                                             SB_ONDISK_JOURNAL_SIZE(sb)));
4138                         set_buffer_uptodate(tmp_bh);
4139                         page = cn->bh->b_page;
4140                         addr = kmap(page);
4141                         memcpy(tmp_bh->b_data,
4142                                addr + offset_in_page(cn->bh->b_data),
4143                                cn->bh->b_size);
4144                         kunmap(page);
4145                         mark_buffer_dirty(tmp_bh);
4146                         jindex++;
4147                         set_buffer_journal_dirty(cn->bh);
4148                         clear_buffer_journaled(cn->bh);
4149                 } else {
4150                         /* JDirty cleared sometime during transaction.  don't log this one */
4151                         reiserfs_warning(sb, "journal-2048",
4152                                          "BAD, buffer in journal hash, "
4153                                          "but not JDirty!");
4154                         brelse(cn->bh);
4155                 }
4156                 next = cn->next;
4157                 free_cnode(sb, cn);
4158                 cn = next;
4159                 cond_resched();
4160         }
4161
4162         /* we are done  with both the c_bh and d_bh, but
4163          ** c_bh must be written after all other commit blocks,
4164          ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4165          */
4166
4167         journal->j_current_jl = alloc_journal_list(sb);
4168
4169         /* now it is safe to insert this transaction on the main list */
4170         list_add_tail(&jl->j_list, &journal->j_journal_list);
4171         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4172         journal->j_num_work_lists++;
4173
4174         /* reset journal values for the next transaction */
4175         old_start = journal->j_start;
4176         journal->j_start =
4177             (journal->j_start + journal->j_len +
4178              2) % SB_ONDISK_JOURNAL_SIZE(sb);
4179         atomic_set(&(journal->j_wcount), 0);
4180         journal->j_bcount = 0;
4181         journal->j_last = NULL;
4182         journal->j_first = NULL;
4183         journal->j_len = 0;
4184         journal->j_trans_start_time = 0;
4185         /* check for trans_id overflow */
4186         if (++journal->j_trans_id == 0)
4187                 journal->j_trans_id = 10;
4188         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4189         journal->j_must_wait = 0;
4190         journal->j_len_alloc = 0;
4191         journal->j_next_full_flush = 0;
4192         journal->j_next_async_flush = 0;
4193         init_journal_hash(sb);
4194
4195         // make sure reiserfs_add_jh sees the new current_jl before we
4196         // write out the tails
4197         smp_mb();
4198
4199         /* tail conversion targets have to hit the disk before we end the
4200          * transaction.  Otherwise a later transaction might repack the tail
4201          * before this transaction commits, leaving the data block unflushed and
4202          * clean, if we crash before the later transaction commits, the data block
4203          * is lost.
4204          */
4205         if (!list_empty(&jl->j_tail_bh_list)) {
4206                 unlock_kernel();
4207                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4208                                       journal, jl, &jl->j_tail_bh_list);
4209                 lock_kernel();
4210         }
4211         BUG_ON(!list_empty(&jl->j_tail_bh_list));
4212         mutex_unlock(&jl->j_commit_mutex);
4213
4214         /* honor the flush wishes from the caller, simple commits can
4215          ** be done outside the journal lock, they are done below
4216          **
4217          ** if we don't flush the commit list right now, we put it into
4218          ** the work queue so the people waiting on the async progress work
4219          ** queue don't wait for this proc to flush journal lists and such.
4220          */
4221         if (flush) {
4222                 flush_commit_list(sb, jl, 1);
4223                 flush_journal_list(sb, jl, 1);
4224         } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4225                 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4226
4227         /* if the next transaction has any chance of wrapping, flush
4228          ** transactions that might get overwritten.  If any journal lists are very
4229          ** old flush them as well.
4230          */
4231       first_jl:
4232         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4233                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4234                 if (journal->j_start <= temp_jl->j_start) {
4235                         if ((journal->j_start + journal->j_trans_max + 1) >=
4236                             temp_jl->j_start) {
4237                                 flush_used_journal_lists(sb, temp_jl);
4238                                 goto first_jl;
4239                         } else if ((journal->j_start +
4240                                     journal->j_trans_max + 1) <
4241                                    SB_ONDISK_JOURNAL_SIZE(sb)) {
4242                                 /* if we don't cross into the next transaction and we don't
4243                                  * wrap, there is no way we can overlap any later transactions
4244                                  * break now
4245                                  */
4246                                 break;
4247                         }
4248                 } else if ((journal->j_start +
4249                             journal->j_trans_max + 1) >
4250                            SB_ONDISK_JOURNAL_SIZE(sb)) {
4251                         if (((journal->j_start + journal->j_trans_max + 1) %
4252                              SB_ONDISK_JOURNAL_SIZE(sb)) >=
4253                             temp_jl->j_start) {
4254                                 flush_used_journal_lists(sb, temp_jl);
4255                                 goto first_jl;
4256                         } else {
4257                                 /* we don't overlap anything from out start to the end of the
4258                                  * log, and our wrapped portion doesn't overlap anything at
4259                                  * the start of the log.  We can break
4260                                  */
4261                                 break;
4262                         }
4263                 }
4264         }
4265         flush_old_journal_lists(sb);
4266
4267         journal->j_current_jl->j_list_bitmap =
4268             get_list_bitmap(sb, journal->j_current_jl);
4269
4270         if (!(journal->j_current_jl->j_list_bitmap)) {
4271                 reiserfs_panic(sb, "journal-1996",
4272                                "could not get a list bitmap");
4273         }
4274
4275         atomic_set(&(journal->j_jlock), 0);
4276         unlock_journal(sb);
4277         /* wake up any body waiting to join. */
4278         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4279         wake_up(&(journal->j_join_wait));
4280
4281         if (!flush && wait_on_commit &&
4282             journal_list_still_alive(sb, commit_trans_id)) {
4283                 flush_commit_list(sb, jl, 1);
4284         }
4285       out:
4286         reiserfs_check_lock_depth(sb, "journal end2");
4287
4288         memset(th, 0, sizeof(*th));
4289         /* Re-set th->t_super, so we can properly keep track of how many
4290          * persistent transactions there are. We need to do this so if this
4291          * call is part of a failed restart_transaction, we can free it later */
4292         th->t_super = sb;
4293
4294         return journal->j_errno;
4295 }
4296
4297 /* Send the file system read only and refuse new transactions */
4298 void reiserfs_abort_journal(struct super_block *sb, int errno)
4299 {
4300         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4301         if (test_bit(J_ABORTED, &journal->j_state))
4302                 return;
4303
4304         if (!journal->j_errno)
4305                 journal->j_errno = errno;
4306
4307         sb->s_flags |= MS_RDONLY;
4308         set_bit(J_ABORTED, &journal->j_state);
4309
4310 #ifdef CONFIG_REISERFS_CHECK
4311         dump_stack();
4312 #endif
4313 }
4314