da01f497180a165d163935c4744ef82521bf9151
[sfrench/cifs-2.6.git] / fs / reiserfs / journal.c
1 /*
2  * Write ahead logging implementation copyright Chris Mason 2000
3  *
4  * The background commits make this code very interrelated, and
5  * overly complex.  I need to rethink things a bit....The major players:
6  *
7  * journal_begin -- call with the number of blocks you expect to log.
8  *                  If the current transaction is too
9  *                  old, it will block until the current transaction is
10  *                  finished, and then start a new one.
11  *                  Usually, your transaction will get joined in with
12  *                  previous ones for speed.
13  *
14  * journal_join  -- same as journal_begin, but won't block on the current
15  *                  transaction regardless of age.  Don't ever call
16  *                  this.  Ever.  There are only two places it should be
17  *                  called from, and they are both inside this file.
18  *
19  * journal_mark_dirty -- adds blocks into this transaction.  clears any flags
20  *                       that might make them get sent to disk
21  *                       and then marks them BH_JDirty.  Puts the buffer head
22  *                       into the current transaction hash.
23  *
24  * journal_end -- if the current transaction is batchable, it does nothing
25  *                   otherwise, it could do an async/synchronous commit, or
26  *                   a full flush of all log and real blocks in the
27  *                   transaction.
28  *
29  * flush_old_commits -- if the current transaction is too old, it is ended and
30  *                      commit blocks are sent to disk.  Forces commit blocks
31  *                      to disk for all backgrounded commits that have been
32  *                      around too long.
33  *                   -- Note, if you call this as an immediate flush from
34  *                      from within kupdate, it will ignore the immediate flag
35  */
36
37 #include <linux/time.h>
38 #include <linux/semaphore.h>
39 #include <linux/vmalloc.h>
40 #include "reiserfs.h"
41 #include <linux/kernel.h>
42 #include <linux/errno.h>
43 #include <linux/fcntl.h>
44 #include <linux/stat.h>
45 #include <linux/string.h>
46 #include <linux/buffer_head.h>
47 #include <linux/workqueue.h>
48 #include <linux/writeback.h>
49 #include <linux/blkdev.h>
50 #include <linux/backing-dev.h>
51 #include <linux/uaccess.h>
52 #include <linux/slab.h>
53
54
55 /* gets a struct reiserfs_journal_list * from a list head */
56 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
57                                j_list))
58 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
59                                j_working_list))
60
61 /* must be correct to keep the desc and commit structs at 4k */
62 #define JOURNAL_TRANS_HALF 1018
63 #define BUFNR 64                /*read ahead */
64
65 /* cnode stat bits.  Move these into reiserfs_fs.h */
66
67 /* this block was freed, and can't be written.  */
68 #define BLOCK_FREED 2
69 /* this block was freed during this transaction, and can't be written */
70 #define BLOCK_FREED_HOLDER 3
71
72 /* used in flush_journal_list */
73 #define BLOCK_NEEDS_FLUSH 4
74 #define BLOCK_DIRTIED 5
75
76 /* journal list state bits */
77 #define LIST_TOUCHED 1
78 #define LIST_DIRTY   2
79 #define LIST_COMMIT_PENDING  4  /* someone will commit this list */
80
81 /* flags for do_journal_end */
82 #define FLUSH_ALL   1           /* flush commit and real blocks */
83 #define COMMIT_NOW  2           /* end and commit this transaction */
84 #define WAIT        4           /* wait for the log blocks to hit the disk */
85
86 static int do_journal_end(struct reiserfs_transaction_handle *, int flags);
87 static int flush_journal_list(struct super_block *s,
88                               struct reiserfs_journal_list *jl, int flushall);
89 static int flush_commit_list(struct super_block *s,
90                              struct reiserfs_journal_list *jl, int flushall);
91 static int can_dirty(struct reiserfs_journal_cnode *cn);
92 static int journal_join(struct reiserfs_transaction_handle *th,
93                         struct super_block *sb);
94 static void release_journal_dev(struct super_block *super,
95                                struct reiserfs_journal *journal);
96 static int dirty_one_transaction(struct super_block *s,
97                                  struct reiserfs_journal_list *jl);
98 static void flush_async_commits(struct work_struct *work);
99 static void queue_log_writer(struct super_block *s);
100
101 /* values for join in do_journal_begin_r */
102 enum {
103         JBEGIN_REG = 0,         /* regular journal begin */
104         /* join the running transaction if at all possible */
105         JBEGIN_JOIN = 1,
106         /* called from cleanup code, ignores aborted flag */
107         JBEGIN_ABORT = 2,
108 };
109
110 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
111                               struct super_block *sb,
112                               unsigned long nblocks, int join);
113
114 static void init_journal_hash(struct super_block *sb)
115 {
116         struct reiserfs_journal *journal = SB_JOURNAL(sb);
117         memset(journal->j_hash_table, 0,
118                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
119 }
120
121 /*
122  * clears BH_Dirty and sticks the buffer on the clean list.  Called because
123  * I can't allow refile_buffer to make schedule happen after I've freed a
124  * block.  Look at remove_from_transaction and journal_mark_freed for
125  * more details.
126  */
127 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
128 {
129         if (bh) {
130                 clear_buffer_dirty(bh);
131                 clear_buffer_journal_test(bh);
132         }
133         return 0;
134 }
135
136 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
137                                                          *sb)
138 {
139         struct reiserfs_bitmap_node *bn;
140         static int id;
141
142         bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
143         if (!bn) {
144                 return NULL;
145         }
146         bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
147         if (!bn->data) {
148                 kfree(bn);
149                 return NULL;
150         }
151         bn->id = id++;
152         INIT_LIST_HEAD(&bn->list);
153         return bn;
154 }
155
156 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
157 {
158         struct reiserfs_journal *journal = SB_JOURNAL(sb);
159         struct reiserfs_bitmap_node *bn = NULL;
160         struct list_head *entry = journal->j_bitmap_nodes.next;
161
162         journal->j_used_bitmap_nodes++;
163 repeat:
164
165         if (entry != &journal->j_bitmap_nodes) {
166                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
167                 list_del(entry);
168                 memset(bn->data, 0, sb->s_blocksize);
169                 journal->j_free_bitmap_nodes--;
170                 return bn;
171         }
172         bn = allocate_bitmap_node(sb);
173         if (!bn) {
174                 yield();
175                 goto repeat;
176         }
177         return bn;
178 }
179 static inline void free_bitmap_node(struct super_block *sb,
180                                     struct reiserfs_bitmap_node *bn)
181 {
182         struct reiserfs_journal *journal = SB_JOURNAL(sb);
183         journal->j_used_bitmap_nodes--;
184         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
185                 kfree(bn->data);
186                 kfree(bn);
187         } else {
188                 list_add(&bn->list, &journal->j_bitmap_nodes);
189                 journal->j_free_bitmap_nodes++;
190         }
191 }
192
193 static void allocate_bitmap_nodes(struct super_block *sb)
194 {
195         int i;
196         struct reiserfs_journal *journal = SB_JOURNAL(sb);
197         struct reiserfs_bitmap_node *bn = NULL;
198         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
199                 bn = allocate_bitmap_node(sb);
200                 if (bn) {
201                         list_add(&bn->list, &journal->j_bitmap_nodes);
202                         journal->j_free_bitmap_nodes++;
203                 } else {
204                         /* this is ok, we'll try again when more are needed */
205                         break;
206                 }
207         }
208 }
209
210 static int set_bit_in_list_bitmap(struct super_block *sb,
211                                   b_blocknr_t block,
212                                   struct reiserfs_list_bitmap *jb)
213 {
214         unsigned int bmap_nr = block / (sb->s_blocksize << 3);
215         unsigned int bit_nr = block % (sb->s_blocksize << 3);
216
217         if (!jb->bitmaps[bmap_nr]) {
218                 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
219         }
220         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
221         return 0;
222 }
223
224 static void cleanup_bitmap_list(struct super_block *sb,
225                                 struct reiserfs_list_bitmap *jb)
226 {
227         int i;
228         if (jb->bitmaps == NULL)
229                 return;
230
231         for (i = 0; i < reiserfs_bmap_count(sb); i++) {
232                 if (jb->bitmaps[i]) {
233                         free_bitmap_node(sb, jb->bitmaps[i]);
234                         jb->bitmaps[i] = NULL;
235                 }
236         }
237 }
238
239 /*
240  * only call this on FS unmount.
241  */
242 static int free_list_bitmaps(struct super_block *sb,
243                              struct reiserfs_list_bitmap *jb_array)
244 {
245         int i;
246         struct reiserfs_list_bitmap *jb;
247         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
248                 jb = jb_array + i;
249                 jb->journal_list = NULL;
250                 cleanup_bitmap_list(sb, jb);
251                 vfree(jb->bitmaps);
252                 jb->bitmaps = NULL;
253         }
254         return 0;
255 }
256
257 static int free_bitmap_nodes(struct super_block *sb)
258 {
259         struct reiserfs_journal *journal = SB_JOURNAL(sb);
260         struct list_head *next = journal->j_bitmap_nodes.next;
261         struct reiserfs_bitmap_node *bn;
262
263         while (next != &journal->j_bitmap_nodes) {
264                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
265                 list_del(next);
266                 kfree(bn->data);
267                 kfree(bn);
268                 next = journal->j_bitmap_nodes.next;
269                 journal->j_free_bitmap_nodes--;
270         }
271
272         return 0;
273 }
274
275 /*
276  * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
277  * jb_array is the array to be filled in.
278  */
279 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
280                                    struct reiserfs_list_bitmap *jb_array,
281                                    unsigned int bmap_nr)
282 {
283         int i;
284         int failed = 0;
285         struct reiserfs_list_bitmap *jb;
286         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
287
288         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
289                 jb = jb_array + i;
290                 jb->journal_list = NULL;
291                 jb->bitmaps = vzalloc(mem);
292                 if (!jb->bitmaps) {
293                         reiserfs_warning(sb, "clm-2000", "unable to "
294                                          "allocate bitmaps for journal lists");
295                         failed = 1;
296                         break;
297                 }
298         }
299         if (failed) {
300                 free_list_bitmaps(sb, jb_array);
301                 return -1;
302         }
303         return 0;
304 }
305
306 /*
307  * find an available list bitmap.  If you can't find one, flush a commit list
308  * and try again
309  */
310 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
311                                                     struct reiserfs_journal_list
312                                                     *jl)
313 {
314         int i, j;
315         struct reiserfs_journal *journal = SB_JOURNAL(sb);
316         struct reiserfs_list_bitmap *jb = NULL;
317
318         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
319                 i = journal->j_list_bitmap_index;
320                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
321                 jb = journal->j_list_bitmap + i;
322                 if (journal->j_list_bitmap[i].journal_list) {
323                         flush_commit_list(sb,
324                                           journal->j_list_bitmap[i].
325                                           journal_list, 1);
326                         if (!journal->j_list_bitmap[i].journal_list) {
327                                 break;
328                         }
329                 } else {
330                         break;
331                 }
332         }
333         /* double check to make sure if flushed correctly */
334         if (jb->journal_list)
335                 return NULL;
336         jb->journal_list = jl;
337         return jb;
338 }
339
340 /*
341  * allocates a new chunk of X nodes, and links them all together as a list.
342  * Uses the cnode->next and cnode->prev pointers
343  * returns NULL on failure
344  */
345 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
346 {
347         struct reiserfs_journal_cnode *head;
348         int i;
349         if (num_cnodes <= 0) {
350                 return NULL;
351         }
352         head = vzalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
353         if (!head) {
354                 return NULL;
355         }
356         head[0].prev = NULL;
357         head[0].next = head + 1;
358         for (i = 1; i < num_cnodes; i++) {
359                 head[i].prev = head + (i - 1);
360                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
361         }
362         head[num_cnodes - 1].next = NULL;
363         return head;
364 }
365
366 /* pulls a cnode off the free list, or returns NULL on failure */
367 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
368 {
369         struct reiserfs_journal_cnode *cn;
370         struct reiserfs_journal *journal = SB_JOURNAL(sb);
371
372         reiserfs_check_lock_depth(sb, "get_cnode");
373
374         if (journal->j_cnode_free <= 0) {
375                 return NULL;
376         }
377         journal->j_cnode_used++;
378         journal->j_cnode_free--;
379         cn = journal->j_cnode_free_list;
380         if (!cn) {
381                 return cn;
382         }
383         if (cn->next) {
384                 cn->next->prev = NULL;
385         }
386         journal->j_cnode_free_list = cn->next;
387         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
388         return cn;
389 }
390
391 /*
392  * returns a cnode to the free list
393  */
394 static void free_cnode(struct super_block *sb,
395                        struct reiserfs_journal_cnode *cn)
396 {
397         struct reiserfs_journal *journal = SB_JOURNAL(sb);
398
399         reiserfs_check_lock_depth(sb, "free_cnode");
400
401         journal->j_cnode_used--;
402         journal->j_cnode_free++;
403         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
404         cn->next = journal->j_cnode_free_list;
405         if (journal->j_cnode_free_list) {
406                 journal->j_cnode_free_list->prev = cn;
407         }
408         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
409         journal->j_cnode_free_list = cn;
410 }
411
412 static void clear_prepared_bits(struct buffer_head *bh)
413 {
414         clear_buffer_journal_prepared(bh);
415         clear_buffer_journal_restore_dirty(bh);
416 }
417
418 /*
419  * return a cnode with same dev, block number and size in table,
420  * or null if not found
421  */
422 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
423                                                                   super_block
424                                                                   *sb,
425                                                                   struct
426                                                                   reiserfs_journal_cnode
427                                                                   **table,
428                                                                   long bl)
429 {
430         struct reiserfs_journal_cnode *cn;
431         cn = journal_hash(table, sb, bl);
432         while (cn) {
433                 if (cn->blocknr == bl && cn->sb == sb)
434                         return cn;
435                 cn = cn->hnext;
436         }
437         return (struct reiserfs_journal_cnode *)0;
438 }
439
440 /*
441  * this actually means 'can this block be reallocated yet?'.  If you set
442  * search_all, a block can only be allocated if it is not in the current
443  * transaction, was not freed by the current transaction, and has no chance
444  * of ever being overwritten by a replay after crashing.
445  *
446  * If you don't set search_all, a block can only be allocated if it is not
447  * in the current transaction.  Since deleting a block removes it from the
448  * current transaction, this case should never happen.  If you don't set
449  * search_all, make sure you never write the block without logging it.
450  *
451  * next_zero_bit is a suggestion about the next block to try for find_forward.
452  * when bl is rejected because it is set in a journal list bitmap, we search
453  * for the next zero bit in the bitmap that rejected bl.  Then, we return
454  * that through next_zero_bit for find_forward to try.
455  *
456  * Just because we return something in next_zero_bit does not mean we won't
457  * reject it on the next call to reiserfs_in_journal
458  */
459 int reiserfs_in_journal(struct super_block *sb,
460                         unsigned int bmap_nr, int bit_nr, int search_all,
461                         b_blocknr_t * next_zero_bit)
462 {
463         struct reiserfs_journal *journal = SB_JOURNAL(sb);
464         struct reiserfs_journal_cnode *cn;
465         struct reiserfs_list_bitmap *jb;
466         int i;
467         unsigned long bl;
468
469         *next_zero_bit = 0;     /* always start this at zero. */
470
471         PROC_INFO_INC(sb, journal.in_journal);
472         /*
473          * If we aren't doing a search_all, this is a metablock, and it
474          * will be logged before use.  if we crash before the transaction
475          * that freed it commits,  this transaction won't have committed
476          * either, and the block will never be written
477          */
478         if (search_all) {
479                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
480                         PROC_INFO_INC(sb, journal.in_journal_bitmap);
481                         jb = journal->j_list_bitmap + i;
482                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
483                             test_bit(bit_nr,
484                                      (unsigned long *)jb->bitmaps[bmap_nr]->
485                                      data)) {
486                                 *next_zero_bit =
487                                     find_next_zero_bit((unsigned long *)
488                                                        (jb->bitmaps[bmap_nr]->
489                                                         data),
490                                                        sb->s_blocksize << 3,
491                                                        bit_nr + 1);
492                                 return 1;
493                         }
494                 }
495         }
496
497         bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
498         /* is it in any old transactions? */
499         if (search_all
500             && (cn =
501                 get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
502                 return 1;
503         }
504
505         /* is it in the current transaction.  This should never happen */
506         if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
507                 BUG();
508                 return 1;
509         }
510
511         PROC_INFO_INC(sb, journal.in_journal_reusable);
512         /* safe for reuse */
513         return 0;
514 }
515
516 /* insert cn into table */
517 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
518                                        struct reiserfs_journal_cnode *cn)
519 {
520         struct reiserfs_journal_cnode *cn_orig;
521
522         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
523         cn->hnext = cn_orig;
524         cn->hprev = NULL;
525         if (cn_orig) {
526                 cn_orig->hprev = cn;
527         }
528         journal_hash(table, cn->sb, cn->blocknr) = cn;
529 }
530
531 /* lock the current transaction */
532 static inline void lock_journal(struct super_block *sb)
533 {
534         PROC_INFO_INC(sb, journal.lock_journal);
535
536         reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
537 }
538
539 /* unlock the current transaction */
540 static inline void unlock_journal(struct super_block *sb)
541 {
542         mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
543 }
544
545 static inline void get_journal_list(struct reiserfs_journal_list *jl)
546 {
547         jl->j_refcount++;
548 }
549
550 static inline void put_journal_list(struct super_block *s,
551                                     struct reiserfs_journal_list *jl)
552 {
553         if (jl->j_refcount < 1) {
554                 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
555                                jl->j_trans_id, jl->j_refcount);
556         }
557         if (--jl->j_refcount == 0)
558                 kfree(jl);
559 }
560
561 /*
562  * this used to be much more involved, and I'm keeping it just in case
563  * things get ugly again.  it gets called by flush_commit_list, and
564  * cleans up any data stored about blocks freed during a transaction.
565  */
566 static void cleanup_freed_for_journal_list(struct super_block *sb,
567                                            struct reiserfs_journal_list *jl)
568 {
569
570         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
571         if (jb) {
572                 cleanup_bitmap_list(sb, jb);
573         }
574         jl->j_list_bitmap->journal_list = NULL;
575         jl->j_list_bitmap = NULL;
576 }
577
578 static int journal_list_still_alive(struct super_block *s,
579                                     unsigned int trans_id)
580 {
581         struct reiserfs_journal *journal = SB_JOURNAL(s);
582         struct list_head *entry = &journal->j_journal_list;
583         struct reiserfs_journal_list *jl;
584
585         if (!list_empty(entry)) {
586                 jl = JOURNAL_LIST_ENTRY(entry->next);
587                 if (jl->j_trans_id <= trans_id) {
588                         return 1;
589                 }
590         }
591         return 0;
592 }
593
594 /*
595  * If page->mapping was null, we failed to truncate this page for
596  * some reason.  Most likely because it was truncated after being
597  * logged via data=journal.
598  *
599  * This does a check to see if the buffer belongs to one of these
600  * lost pages before doing the final put_bh.  If page->mapping was
601  * null, it tries to free buffers on the page, which should make the
602  * final put_page drop the page from the lru.
603  */
604 static void release_buffer_page(struct buffer_head *bh)
605 {
606         struct page *page = bh->b_page;
607         if (!page->mapping && trylock_page(page)) {
608                 get_page(page);
609                 put_bh(bh);
610                 if (!page->mapping)
611                         try_to_free_buffers(page);
612                 unlock_page(page);
613                 put_page(page);
614         } else {
615                 put_bh(bh);
616         }
617 }
618
619 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
620 {
621         if (buffer_journaled(bh)) {
622                 reiserfs_warning(NULL, "clm-2084",
623                                  "pinned buffer %lu:%pg sent to disk",
624                                  bh->b_blocknr, bh->b_bdev);
625         }
626         if (uptodate)
627                 set_buffer_uptodate(bh);
628         else
629                 clear_buffer_uptodate(bh);
630
631         unlock_buffer(bh);
632         release_buffer_page(bh);
633 }
634
635 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
636 {
637         if (uptodate)
638                 set_buffer_uptodate(bh);
639         else
640                 clear_buffer_uptodate(bh);
641         unlock_buffer(bh);
642         put_bh(bh);
643 }
644
645 static void submit_logged_buffer(struct buffer_head *bh)
646 {
647         get_bh(bh);
648         bh->b_end_io = reiserfs_end_buffer_io_sync;
649         clear_buffer_journal_new(bh);
650         clear_buffer_dirty(bh);
651         if (!test_clear_buffer_journal_test(bh))
652                 BUG();
653         if (!buffer_uptodate(bh))
654                 BUG();
655         submit_bh(REQ_OP_WRITE, 0, bh);
656 }
657
658 static void submit_ordered_buffer(struct buffer_head *bh)
659 {
660         get_bh(bh);
661         bh->b_end_io = reiserfs_end_ordered_io;
662         clear_buffer_dirty(bh);
663         if (!buffer_uptodate(bh))
664                 BUG();
665         submit_bh(REQ_OP_WRITE, 0, bh);
666 }
667
668 #define CHUNK_SIZE 32
669 struct buffer_chunk {
670         struct buffer_head *bh[CHUNK_SIZE];
671         int nr;
672 };
673
674 static void write_chunk(struct buffer_chunk *chunk)
675 {
676         int i;
677         for (i = 0; i < chunk->nr; i++) {
678                 submit_logged_buffer(chunk->bh[i]);
679         }
680         chunk->nr = 0;
681 }
682
683 static void write_ordered_chunk(struct buffer_chunk *chunk)
684 {
685         int i;
686         for (i = 0; i < chunk->nr; i++) {
687                 submit_ordered_buffer(chunk->bh[i]);
688         }
689         chunk->nr = 0;
690 }
691
692 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
693                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
694 {
695         int ret = 0;
696         BUG_ON(chunk->nr >= CHUNK_SIZE);
697         chunk->bh[chunk->nr++] = bh;
698         if (chunk->nr >= CHUNK_SIZE) {
699                 ret = 1;
700                 if (lock) {
701                         spin_unlock(lock);
702                         fn(chunk);
703                         spin_lock(lock);
704                 } else {
705                         fn(chunk);
706                 }
707         }
708         return ret;
709 }
710
711 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
712 static struct reiserfs_jh *alloc_jh(void)
713 {
714         struct reiserfs_jh *jh;
715         while (1) {
716                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
717                 if (jh) {
718                         atomic_inc(&nr_reiserfs_jh);
719                         return jh;
720                 }
721                 yield();
722         }
723 }
724
725 /*
726  * we want to free the jh when the buffer has been written
727  * and waited on
728  */
729 void reiserfs_free_jh(struct buffer_head *bh)
730 {
731         struct reiserfs_jh *jh;
732
733         jh = bh->b_private;
734         if (jh) {
735                 bh->b_private = NULL;
736                 jh->bh = NULL;
737                 list_del_init(&jh->list);
738                 kfree(jh);
739                 if (atomic_read(&nr_reiserfs_jh) <= 0)
740                         BUG();
741                 atomic_dec(&nr_reiserfs_jh);
742                 put_bh(bh);
743         }
744 }
745
746 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
747                            int tail)
748 {
749         struct reiserfs_jh *jh;
750
751         if (bh->b_private) {
752                 spin_lock(&j->j_dirty_buffers_lock);
753                 if (!bh->b_private) {
754                         spin_unlock(&j->j_dirty_buffers_lock);
755                         goto no_jh;
756                 }
757                 jh = bh->b_private;
758                 list_del_init(&jh->list);
759         } else {
760 no_jh:
761                 get_bh(bh);
762                 jh = alloc_jh();
763                 spin_lock(&j->j_dirty_buffers_lock);
764                 /*
765                  * buffer must be locked for __add_jh, should be able to have
766                  * two adds at the same time
767                  */
768                 BUG_ON(bh->b_private);
769                 jh->bh = bh;
770                 bh->b_private = jh;
771         }
772         jh->jl = j->j_current_jl;
773         if (tail)
774                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
775         else {
776                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
777         }
778         spin_unlock(&j->j_dirty_buffers_lock);
779         return 0;
780 }
781
782 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
783 {
784         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
785 }
786 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
787 {
788         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
789 }
790
791 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
792 static int write_ordered_buffers(spinlock_t * lock,
793                                  struct reiserfs_journal *j,
794                                  struct reiserfs_journal_list *jl,
795                                  struct list_head *list)
796 {
797         struct buffer_head *bh;
798         struct reiserfs_jh *jh;
799         int ret = j->j_errno;
800         struct buffer_chunk chunk;
801         struct list_head tmp;
802         INIT_LIST_HEAD(&tmp);
803
804         chunk.nr = 0;
805         spin_lock(lock);
806         while (!list_empty(list)) {
807                 jh = JH_ENTRY(list->next);
808                 bh = jh->bh;
809                 get_bh(bh);
810                 if (!trylock_buffer(bh)) {
811                         if (!buffer_dirty(bh)) {
812                                 list_move(&jh->list, &tmp);
813                                 goto loop_next;
814                         }
815                         spin_unlock(lock);
816                         if (chunk.nr)
817                                 write_ordered_chunk(&chunk);
818                         wait_on_buffer(bh);
819                         cond_resched();
820                         spin_lock(lock);
821                         goto loop_next;
822                 }
823                 /*
824                  * in theory, dirty non-uptodate buffers should never get here,
825                  * but the upper layer io error paths still have a few quirks.
826                  * Handle them here as gracefully as we can
827                  */
828                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
829                         clear_buffer_dirty(bh);
830                         ret = -EIO;
831                 }
832                 if (buffer_dirty(bh)) {
833                         list_move(&jh->list, &tmp);
834                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
835                 } else {
836                         reiserfs_free_jh(bh);
837                         unlock_buffer(bh);
838                 }
839 loop_next:
840                 put_bh(bh);
841                 cond_resched_lock(lock);
842         }
843         if (chunk.nr) {
844                 spin_unlock(lock);
845                 write_ordered_chunk(&chunk);
846                 spin_lock(lock);
847         }
848         while (!list_empty(&tmp)) {
849                 jh = JH_ENTRY(tmp.prev);
850                 bh = jh->bh;
851                 get_bh(bh);
852                 reiserfs_free_jh(bh);
853
854                 if (buffer_locked(bh)) {
855                         spin_unlock(lock);
856                         wait_on_buffer(bh);
857                         spin_lock(lock);
858                 }
859                 if (!buffer_uptodate(bh)) {
860                         ret = -EIO;
861                 }
862                 /*
863                  * ugly interaction with invalidatepage here.
864                  * reiserfs_invalidate_page will pin any buffer that has a
865                  * valid journal head from an older transaction.  If someone
866                  * else sets our buffer dirty after we write it in the first
867                  * loop, and then someone truncates the page away, nobody
868                  * will ever write the buffer. We're safe if we write the
869                  * page one last time after freeing the journal header.
870                  */
871                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
872                         spin_unlock(lock);
873                         ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
874                         spin_lock(lock);
875                 }
876                 put_bh(bh);
877                 cond_resched_lock(lock);
878         }
879         spin_unlock(lock);
880         return ret;
881 }
882
883 static int flush_older_commits(struct super_block *s,
884                                struct reiserfs_journal_list *jl)
885 {
886         struct reiserfs_journal *journal = SB_JOURNAL(s);
887         struct reiserfs_journal_list *other_jl;
888         struct reiserfs_journal_list *first_jl;
889         struct list_head *entry;
890         unsigned int trans_id = jl->j_trans_id;
891         unsigned int other_trans_id;
892         unsigned int first_trans_id;
893
894 find_first:
895         /*
896          * first we walk backwards to find the oldest uncommitted transation
897          */
898         first_jl = jl;
899         entry = jl->j_list.prev;
900         while (1) {
901                 other_jl = JOURNAL_LIST_ENTRY(entry);
902                 if (entry == &journal->j_journal_list ||
903                     atomic_read(&other_jl->j_older_commits_done))
904                         break;
905
906                 first_jl = other_jl;
907                 entry = other_jl->j_list.prev;
908         }
909
910         /* if we didn't find any older uncommitted transactions, return now */
911         if (first_jl == jl) {
912                 return 0;
913         }
914
915         first_trans_id = first_jl->j_trans_id;
916
917         entry = &first_jl->j_list;
918         while (1) {
919                 other_jl = JOURNAL_LIST_ENTRY(entry);
920                 other_trans_id = other_jl->j_trans_id;
921
922                 if (other_trans_id < trans_id) {
923                         if (atomic_read(&other_jl->j_commit_left) != 0) {
924                                 flush_commit_list(s, other_jl, 0);
925
926                                 /* list we were called with is gone, return */
927                                 if (!journal_list_still_alive(s, trans_id))
928                                         return 1;
929
930                                 /*
931                                  * the one we just flushed is gone, this means
932                                  * all older lists are also gone, so first_jl
933                                  * is no longer valid either.  Go back to the
934                                  * beginning.
935                                  */
936                                 if (!journal_list_still_alive
937                                     (s, other_trans_id)) {
938                                         goto find_first;
939                                 }
940                         }
941                         entry = entry->next;
942                         if (entry == &journal->j_journal_list)
943                                 return 0;
944                 } else {
945                         return 0;
946                 }
947         }
948         return 0;
949 }
950
951 static int reiserfs_async_progress_wait(struct super_block *s)
952 {
953         struct reiserfs_journal *j = SB_JOURNAL(s);
954
955         if (atomic_read(&j->j_async_throttle)) {
956                 int depth;
957
958                 depth = reiserfs_write_unlock_nested(s);
959                 congestion_wait(BLK_RW_ASYNC, HZ / 10);
960                 reiserfs_write_lock_nested(s, depth);
961         }
962
963         return 0;
964 }
965
966 /*
967  * if this journal list still has commit blocks unflushed, send them to disk.
968  *
969  * log areas must be flushed in order (transaction 2 can't commit before
970  * transaction 1) Before the commit block can by written, every other log
971  * block must be safely on disk
972  */
973 static int flush_commit_list(struct super_block *s,
974                              struct reiserfs_journal_list *jl, int flushall)
975 {
976         int i;
977         b_blocknr_t bn;
978         struct buffer_head *tbh = NULL;
979         unsigned int trans_id = jl->j_trans_id;
980         struct reiserfs_journal *journal = SB_JOURNAL(s);
981         int retval = 0;
982         int write_len;
983         int depth;
984
985         reiserfs_check_lock_depth(s, "flush_commit_list");
986
987         if (atomic_read(&jl->j_older_commits_done)) {
988                 return 0;
989         }
990
991         /*
992          * before we can put our commit blocks on disk, we have to make
993          * sure everyone older than us is on disk too
994          */
995         BUG_ON(jl->j_len <= 0);
996         BUG_ON(trans_id == journal->j_trans_id);
997
998         get_journal_list(jl);
999         if (flushall) {
1000                 if (flush_older_commits(s, jl) == 1) {
1001                         /*
1002                          * list disappeared during flush_older_commits.
1003                          * return
1004                          */
1005                         goto put_jl;
1006                 }
1007         }
1008
1009         /* make sure nobody is trying to flush this one at the same time */
1010         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1011
1012         if (!journal_list_still_alive(s, trans_id)) {
1013                 mutex_unlock(&jl->j_commit_mutex);
1014                 goto put_jl;
1015         }
1016         BUG_ON(jl->j_trans_id == 0);
1017
1018         /* this commit is done, exit */
1019         if (atomic_read(&jl->j_commit_left) <= 0) {
1020                 if (flushall) {
1021                         atomic_set(&jl->j_older_commits_done, 1);
1022                 }
1023                 mutex_unlock(&jl->j_commit_mutex);
1024                 goto put_jl;
1025         }
1026
1027         if (!list_empty(&jl->j_bh_list)) {
1028                 int ret;
1029
1030                 /*
1031                  * We might sleep in numerous places inside
1032                  * write_ordered_buffers. Relax the write lock.
1033                  */
1034                 depth = reiserfs_write_unlock_nested(s);
1035                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1036                                             journal, jl, &jl->j_bh_list);
1037                 if (ret < 0 && retval == 0)
1038                         retval = ret;
1039                 reiserfs_write_lock_nested(s, depth);
1040         }
1041         BUG_ON(!list_empty(&jl->j_bh_list));
1042         /*
1043          * for the description block and all the log blocks, submit any buffers
1044          * that haven't already reached the disk.  Try to write at least 256
1045          * log blocks. later on, we will only wait on blocks that correspond
1046          * to this transaction, but while we're unplugging we might as well
1047          * get a chunk of data on there.
1048          */
1049         atomic_inc(&journal->j_async_throttle);
1050         write_len = jl->j_len + 1;
1051         if (write_len < 256)
1052                 write_len = 256;
1053         for (i = 0 ; i < write_len ; i++) {
1054                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1055                     SB_ONDISK_JOURNAL_SIZE(s);
1056                 tbh = journal_find_get_block(s, bn);
1057                 if (tbh) {
1058                         if (buffer_dirty(tbh)) {
1059                             depth = reiserfs_write_unlock_nested(s);
1060                             ll_rw_block(REQ_OP_WRITE, 0, 1, &tbh);
1061                             reiserfs_write_lock_nested(s, depth);
1062                         }
1063                         put_bh(tbh) ;
1064                 }
1065         }
1066         atomic_dec(&journal->j_async_throttle);
1067
1068         for (i = 0; i < (jl->j_len + 1); i++) {
1069                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1070                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1071                 tbh = journal_find_get_block(s, bn);
1072
1073                 depth = reiserfs_write_unlock_nested(s);
1074                 __wait_on_buffer(tbh);
1075                 reiserfs_write_lock_nested(s, depth);
1076                 /*
1077                  * since we're using ll_rw_blk above, it might have skipped
1078                  * over a locked buffer.  Double check here
1079                  */
1080                 /* redundant, sync_dirty_buffer() checks */
1081                 if (buffer_dirty(tbh)) {
1082                         depth = reiserfs_write_unlock_nested(s);
1083                         sync_dirty_buffer(tbh);
1084                         reiserfs_write_lock_nested(s, depth);
1085                 }
1086                 if (unlikely(!buffer_uptodate(tbh))) {
1087 #ifdef CONFIG_REISERFS_CHECK
1088                         reiserfs_warning(s, "journal-601",
1089                                          "buffer write failed");
1090 #endif
1091                         retval = -EIO;
1092                 }
1093                 /* once for journal_find_get_block */
1094                 put_bh(tbh);
1095                 /* once due to original getblk in do_journal_end */
1096                 put_bh(tbh);
1097                 atomic_dec(&jl->j_commit_left);
1098         }
1099
1100         BUG_ON(atomic_read(&jl->j_commit_left) != 1);
1101
1102         /*
1103          * If there was a write error in the journal - we can't commit
1104          * this transaction - it will be invalid and, if successful,
1105          * will just end up propagating the write error out to
1106          * the file system.
1107          */
1108         if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1109                 if (buffer_dirty(jl->j_commit_bh))
1110                         BUG();
1111                 mark_buffer_dirty(jl->j_commit_bh) ;
1112                 depth = reiserfs_write_unlock_nested(s);
1113                 if (reiserfs_barrier_flush(s))
1114                         __sync_dirty_buffer(jl->j_commit_bh,
1115                                         REQ_PREFLUSH | REQ_FUA);
1116                 else
1117                         sync_dirty_buffer(jl->j_commit_bh);
1118                 reiserfs_write_lock_nested(s, depth);
1119         }
1120
1121         /*
1122          * If there was a write error in the journal - we can't commit this
1123          * transaction - it will be invalid and, if successful, will just end
1124          * up propagating the write error out to the filesystem.
1125          */
1126         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1127 #ifdef CONFIG_REISERFS_CHECK
1128                 reiserfs_warning(s, "journal-615", "buffer write failed");
1129 #endif
1130                 retval = -EIO;
1131         }
1132         bforget(jl->j_commit_bh);
1133         if (journal->j_last_commit_id != 0 &&
1134             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1135                 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1136                                  journal->j_last_commit_id, jl->j_trans_id);
1137         }
1138         journal->j_last_commit_id = jl->j_trans_id;
1139
1140         /*
1141          * now, every commit block is on the disk.  It is safe to allow
1142          * blocks freed during this transaction to be reallocated
1143          */
1144         cleanup_freed_for_journal_list(s, jl);
1145
1146         retval = retval ? retval : journal->j_errno;
1147
1148         /* mark the metadata dirty */
1149         if (!retval)
1150                 dirty_one_transaction(s, jl);
1151         atomic_dec(&jl->j_commit_left);
1152
1153         if (flushall) {
1154                 atomic_set(&jl->j_older_commits_done, 1);
1155         }
1156         mutex_unlock(&jl->j_commit_mutex);
1157 put_jl:
1158         put_journal_list(s, jl);
1159
1160         if (retval)
1161                 reiserfs_abort(s, retval, "Journal write error in %s",
1162                                __func__);
1163         return retval;
1164 }
1165
1166 /*
1167  * flush_journal_list frequently needs to find a newer transaction for a
1168  * given block.  This does that, or returns NULL if it can't find anything
1169  */
1170 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1171                                                           reiserfs_journal_cnode
1172                                                           *cn)
1173 {
1174         struct super_block *sb = cn->sb;
1175         b_blocknr_t blocknr = cn->blocknr;
1176
1177         cn = cn->hprev;
1178         while (cn) {
1179                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1180                         return cn->jlist;
1181                 }
1182                 cn = cn->hprev;
1183         }
1184         return NULL;
1185 }
1186
1187 static void remove_journal_hash(struct super_block *,
1188                                 struct reiserfs_journal_cnode **,
1189                                 struct reiserfs_journal_list *, unsigned long,
1190                                 int);
1191
1192 /*
1193  * once all the real blocks have been flushed, it is safe to remove them
1194  * from the journal list for this transaction.  Aside from freeing the
1195  * cnode, this also allows the block to be reallocated for data blocks
1196  * if it had been deleted.
1197  */
1198 static void remove_all_from_journal_list(struct super_block *sb,
1199                                          struct reiserfs_journal_list *jl,
1200                                          int debug)
1201 {
1202         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1203         struct reiserfs_journal_cnode *cn, *last;
1204         cn = jl->j_realblock;
1205
1206         /*
1207          * which is better, to lock once around the whole loop, or
1208          * to lock for each call to remove_journal_hash?
1209          */
1210         while (cn) {
1211                 if (cn->blocknr != 0) {
1212                         if (debug) {
1213                                 reiserfs_warning(sb, "reiserfs-2201",
1214                                                  "block %u, bh is %d, state %ld",
1215                                                  cn->blocknr, cn->bh ? 1 : 0,
1216                                                  cn->state);
1217                         }
1218                         cn->state = 0;
1219                         remove_journal_hash(sb, journal->j_list_hash_table,
1220                                             jl, cn->blocknr, 1);
1221                 }
1222                 last = cn;
1223                 cn = cn->next;
1224                 free_cnode(sb, last);
1225         }
1226         jl->j_realblock = NULL;
1227 }
1228
1229 /*
1230  * if this timestamp is greater than the timestamp we wrote last to the
1231  * header block, write it to the header block.  once this is done, I can
1232  * safely say the log area for this transaction won't ever be replayed,
1233  * and I can start releasing blocks in this transaction for reuse as data
1234  * blocks.  called by flush_journal_list, before it calls
1235  * remove_all_from_journal_list
1236  */
1237 static int _update_journal_header_block(struct super_block *sb,
1238                                         unsigned long offset,
1239                                         unsigned int trans_id)
1240 {
1241         struct reiserfs_journal_header *jh;
1242         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1243         int depth;
1244
1245         if (reiserfs_is_journal_aborted(journal))
1246                 return -EIO;
1247
1248         if (trans_id >= journal->j_last_flush_trans_id) {
1249                 if (buffer_locked((journal->j_header_bh))) {
1250                         depth = reiserfs_write_unlock_nested(sb);
1251                         __wait_on_buffer(journal->j_header_bh);
1252                         reiserfs_write_lock_nested(sb, depth);
1253                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1254 #ifdef CONFIG_REISERFS_CHECK
1255                                 reiserfs_warning(sb, "journal-699",
1256                                                  "buffer write failed");
1257 #endif
1258                                 return -EIO;
1259                         }
1260                 }
1261                 journal->j_last_flush_trans_id = trans_id;
1262                 journal->j_first_unflushed_offset = offset;
1263                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1264                                                         b_data);
1265                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1266                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1267                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1268
1269                 set_buffer_dirty(journal->j_header_bh);
1270                 depth = reiserfs_write_unlock_nested(sb);
1271
1272                 if (reiserfs_barrier_flush(sb))
1273                         __sync_dirty_buffer(journal->j_header_bh,
1274                                         REQ_PREFLUSH | REQ_FUA);
1275                 else
1276                         sync_dirty_buffer(journal->j_header_bh);
1277
1278                 reiserfs_write_lock_nested(sb, depth);
1279                 if (!buffer_uptodate(journal->j_header_bh)) {
1280                         reiserfs_warning(sb, "journal-837",
1281                                          "IO error during journal replay");
1282                         return -EIO;
1283                 }
1284         }
1285         return 0;
1286 }
1287
1288 static int update_journal_header_block(struct super_block *sb,
1289                                        unsigned long offset,
1290                                        unsigned int trans_id)
1291 {
1292         return _update_journal_header_block(sb, offset, trans_id);
1293 }
1294
1295 /*
1296 ** flush any and all journal lists older than you are
1297 ** can only be called from flush_journal_list
1298 */
1299 static int flush_older_journal_lists(struct super_block *sb,
1300                                      struct reiserfs_journal_list *jl)
1301 {
1302         struct list_head *entry;
1303         struct reiserfs_journal_list *other_jl;
1304         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1305         unsigned int trans_id = jl->j_trans_id;
1306
1307         /*
1308          * we know we are the only ones flushing things, no extra race
1309          * protection is required.
1310          */
1311 restart:
1312         entry = journal->j_journal_list.next;
1313         /* Did we wrap? */
1314         if (entry == &journal->j_journal_list)
1315                 return 0;
1316         other_jl = JOURNAL_LIST_ENTRY(entry);
1317         if (other_jl->j_trans_id < trans_id) {
1318                 BUG_ON(other_jl->j_refcount <= 0);
1319                 /* do not flush all */
1320                 flush_journal_list(sb, other_jl, 0);
1321
1322                 /* other_jl is now deleted from the list */
1323                 goto restart;
1324         }
1325         return 0;
1326 }
1327
1328 static void del_from_work_list(struct super_block *s,
1329                                struct reiserfs_journal_list *jl)
1330 {
1331         struct reiserfs_journal *journal = SB_JOURNAL(s);
1332         if (!list_empty(&jl->j_working_list)) {
1333                 list_del_init(&jl->j_working_list);
1334                 journal->j_num_work_lists--;
1335         }
1336 }
1337
1338 /*
1339  * flush a journal list, both commit and real blocks
1340  *
1341  * always set flushall to 1, unless you are calling from inside
1342  * flush_journal_list
1343  *
1344  * IMPORTANT.  This can only be called while there are no journal writers,
1345  * and the journal is locked.  That means it can only be called from
1346  * do_journal_end, or by journal_release
1347  */
1348 static int flush_journal_list(struct super_block *s,
1349                               struct reiserfs_journal_list *jl, int flushall)
1350 {
1351         struct reiserfs_journal_list *pjl;
1352         struct reiserfs_journal_cnode *cn, *last;
1353         int count;
1354         int was_jwait = 0;
1355         int was_dirty = 0;
1356         struct buffer_head *saved_bh;
1357         unsigned long j_len_saved = jl->j_len;
1358         struct reiserfs_journal *journal = SB_JOURNAL(s);
1359         int err = 0;
1360         int depth;
1361
1362         BUG_ON(j_len_saved <= 0);
1363
1364         if (atomic_read(&journal->j_wcount) != 0) {
1365                 reiserfs_warning(s, "clm-2048", "called with wcount %d",
1366                                  atomic_read(&journal->j_wcount));
1367         }
1368
1369         /* if flushall == 0, the lock is already held */
1370         if (flushall) {
1371                 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1372         } else if (mutex_trylock(&journal->j_flush_mutex)) {
1373                 BUG();
1374         }
1375
1376         count = 0;
1377         if (j_len_saved > journal->j_trans_max) {
1378                 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1379                                j_len_saved, jl->j_trans_id);
1380                 return 0;
1381         }
1382
1383         /* if all the work is already done, get out of here */
1384         if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1385             atomic_read(&jl->j_commit_left) <= 0) {
1386                 goto flush_older_and_return;
1387         }
1388
1389         /*
1390          * start by putting the commit list on disk.  This will also flush
1391          * the commit lists of any olders transactions
1392          */
1393         flush_commit_list(s, jl, 1);
1394
1395         if (!(jl->j_state & LIST_DIRTY)
1396             && !reiserfs_is_journal_aborted(journal))
1397                 BUG();
1398
1399         /* are we done now? */
1400         if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1401             atomic_read(&jl->j_commit_left) <= 0) {
1402                 goto flush_older_and_return;
1403         }
1404
1405         /*
1406          * loop through each cnode, see if we need to write it,
1407          * or wait on a more recent transaction, or just ignore it
1408          */
1409         if (atomic_read(&journal->j_wcount) != 0) {
1410                 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1411                                "wcount is not 0");
1412         }
1413         cn = jl->j_realblock;
1414         while (cn) {
1415                 was_jwait = 0;
1416                 was_dirty = 0;
1417                 saved_bh = NULL;
1418                 /* blocknr of 0 is no longer in the hash, ignore it */
1419                 if (cn->blocknr == 0) {
1420                         goto free_cnode;
1421                 }
1422
1423                 /*
1424                  * This transaction failed commit.
1425                  * Don't write out to the disk
1426                  */
1427                 if (!(jl->j_state & LIST_DIRTY))
1428                         goto free_cnode;
1429
1430                 pjl = find_newer_jl_for_cn(cn);
1431                 /*
1432                  * the order is important here.  We check pjl to make sure we
1433                  * don't clear BH_JDirty_wait if we aren't the one writing this
1434                  * block to disk
1435                  */
1436                 if (!pjl && cn->bh) {
1437                         saved_bh = cn->bh;
1438
1439                         /*
1440                          * we do this to make sure nobody releases the
1441                          * buffer while we are working with it
1442                          */
1443                         get_bh(saved_bh);
1444
1445                         if (buffer_journal_dirty(saved_bh)) {
1446                                 BUG_ON(!can_dirty(cn));
1447                                 was_jwait = 1;
1448                                 was_dirty = 1;
1449                         } else if (can_dirty(cn)) {
1450                                 /*
1451                                  * everything with !pjl && jwait
1452                                  * should be writable
1453                                  */
1454                                 BUG();
1455                         }
1456                 }
1457
1458                 /*
1459                  * if someone has this block in a newer transaction, just make
1460                  * sure they are committed, and don't try writing it to disk
1461                  */
1462                 if (pjl) {
1463                         if (atomic_read(&pjl->j_commit_left))
1464                                 flush_commit_list(s, pjl, 1);
1465                         goto free_cnode;
1466                 }
1467
1468                 /*
1469                  * bh == NULL when the block got to disk on its own, OR,
1470                  * the block got freed in a future transaction
1471                  */
1472                 if (saved_bh == NULL) {
1473                         goto free_cnode;
1474                 }
1475
1476                 /*
1477                  * this should never happen.  kupdate_one_transaction has
1478                  * this list locked while it works, so we should never see a
1479                  * buffer here that is not marked JDirty_wait
1480                  */
1481                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1482                         reiserfs_warning(s, "journal-813",
1483                                          "BAD! buffer %llu %cdirty %cjwait, "
1484                                          "not in a newer tranasction",
1485                                          (unsigned long long)saved_bh->
1486                                          b_blocknr, was_dirty ? ' ' : '!',
1487                                          was_jwait ? ' ' : '!');
1488                 }
1489                 if (was_dirty) {
1490                         /*
1491                          * we inc again because saved_bh gets decremented
1492                          * at free_cnode
1493                          */
1494                         get_bh(saved_bh);
1495                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1496                         lock_buffer(saved_bh);
1497                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1498                         if (buffer_dirty(saved_bh))
1499                                 submit_logged_buffer(saved_bh);
1500                         else
1501                                 unlock_buffer(saved_bh);
1502                         count++;
1503                 } else {
1504                         reiserfs_warning(s, "clm-2082",
1505                                          "Unable to flush buffer %llu in %s",
1506                                          (unsigned long long)saved_bh->
1507                                          b_blocknr, __func__);
1508                 }
1509 free_cnode:
1510                 last = cn;
1511                 cn = cn->next;
1512                 if (saved_bh) {
1513                         /*
1514                          * we incremented this to keep others from
1515                          * taking the buffer head away
1516                          */
1517                         put_bh(saved_bh);
1518                         if (atomic_read(&saved_bh->b_count) < 0) {
1519                                 reiserfs_warning(s, "journal-945",
1520                                                  "saved_bh->b_count < 0");
1521                         }
1522                 }
1523         }
1524         if (count > 0) {
1525                 cn = jl->j_realblock;
1526                 while (cn) {
1527                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1528                                 if (!cn->bh) {
1529                                         reiserfs_panic(s, "journal-1011",
1530                                                        "cn->bh is NULL");
1531                                 }
1532
1533                                 depth = reiserfs_write_unlock_nested(s);
1534                                 __wait_on_buffer(cn->bh);
1535                                 reiserfs_write_lock_nested(s, depth);
1536
1537                                 if (!cn->bh) {
1538                                         reiserfs_panic(s, "journal-1012",
1539                                                        "cn->bh is NULL");
1540                                 }
1541                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1542 #ifdef CONFIG_REISERFS_CHECK
1543                                         reiserfs_warning(s, "journal-949",
1544                                                          "buffer write failed");
1545 #endif
1546                                         err = -EIO;
1547                                 }
1548                                 /*
1549                                  * note, we must clear the JDirty_wait bit
1550                                  * after the up to date check, otherwise we
1551                                  * race against our flushpage routine
1552                                  */
1553                                 BUG_ON(!test_clear_buffer_journal_dirty
1554                                        (cn->bh));
1555
1556                                 /* drop one ref for us */
1557                                 put_bh(cn->bh);
1558                                 /* drop one ref for journal_mark_dirty */
1559                                 release_buffer_page(cn->bh);
1560                         }
1561                         cn = cn->next;
1562                 }
1563         }
1564
1565         if (err)
1566                 reiserfs_abort(s, -EIO,
1567                                "Write error while pushing transaction to disk in %s",
1568                                __func__);
1569 flush_older_and_return:
1570
1571         /*
1572          * before we can update the journal header block, we _must_ flush all
1573          * real blocks from all older transactions to disk.  This is because
1574          * once the header block is updated, this transaction will not be
1575          * replayed after a crash
1576          */
1577         if (flushall) {
1578                 flush_older_journal_lists(s, jl);
1579         }
1580
1581         err = journal->j_errno;
1582         /*
1583          * before we can remove everything from the hash tables for this
1584          * transaction, we must make sure it can never be replayed
1585          *
1586          * since we are only called from do_journal_end, we know for sure there
1587          * are no allocations going on while we are flushing journal lists.  So,
1588          * we only need to update the journal header block for the last list
1589          * being flushed
1590          */
1591         if (!err && flushall) {
1592                 err =
1593                     update_journal_header_block(s,
1594                                                 (jl->j_start + jl->j_len +
1595                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1596                                                 jl->j_trans_id);
1597                 if (err)
1598                         reiserfs_abort(s, -EIO,
1599                                        "Write error while updating journal header in %s",
1600                                        __func__);
1601         }
1602         remove_all_from_journal_list(s, jl, 0);
1603         list_del_init(&jl->j_list);
1604         journal->j_num_lists--;
1605         del_from_work_list(s, jl);
1606
1607         if (journal->j_last_flush_id != 0 &&
1608             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1609                 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1610                                  journal->j_last_flush_id, jl->j_trans_id);
1611         }
1612         journal->j_last_flush_id = jl->j_trans_id;
1613
1614         /*
1615          * not strictly required since we are freeing the list, but it should
1616          * help find code using dead lists later on
1617          */
1618         jl->j_len = 0;
1619         atomic_set(&jl->j_nonzerolen, 0);
1620         jl->j_start = 0;
1621         jl->j_realblock = NULL;
1622         jl->j_commit_bh = NULL;
1623         jl->j_trans_id = 0;
1624         jl->j_state = 0;
1625         put_journal_list(s, jl);
1626         if (flushall)
1627                 mutex_unlock(&journal->j_flush_mutex);
1628         return err;
1629 }
1630
1631 static int write_one_transaction(struct super_block *s,
1632                                  struct reiserfs_journal_list *jl,
1633                                  struct buffer_chunk *chunk)
1634 {
1635         struct reiserfs_journal_cnode *cn;
1636         int ret = 0;
1637
1638         jl->j_state |= LIST_TOUCHED;
1639         del_from_work_list(s, jl);
1640         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1641                 return 0;
1642         }
1643
1644         cn = jl->j_realblock;
1645         while (cn) {
1646                 /*
1647                  * if the blocknr == 0, this has been cleared from the hash,
1648                  * skip it
1649                  */
1650                 if (cn->blocknr == 0) {
1651                         goto next;
1652                 }
1653                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1654                         struct buffer_head *tmp_bh;
1655                         /*
1656                          * we can race against journal_mark_freed when we try
1657                          * to lock_buffer(cn->bh), so we have to inc the buffer
1658                          * count, and recheck things after locking
1659                          */
1660                         tmp_bh = cn->bh;
1661                         get_bh(tmp_bh);
1662                         lock_buffer(tmp_bh);
1663                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1664                                 if (!buffer_journal_dirty(tmp_bh) ||
1665                                     buffer_journal_prepared(tmp_bh))
1666                                         BUG();
1667                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1668                                 ret++;
1669                         } else {
1670                                 /* note, cn->bh might be null now */
1671                                 unlock_buffer(tmp_bh);
1672                         }
1673                         put_bh(tmp_bh);
1674                 }
1675 next:
1676                 cn = cn->next;
1677                 cond_resched();
1678         }
1679         return ret;
1680 }
1681
1682 /* used by flush_commit_list */
1683 static int dirty_one_transaction(struct super_block *s,
1684                                  struct reiserfs_journal_list *jl)
1685 {
1686         struct reiserfs_journal_cnode *cn;
1687         struct reiserfs_journal_list *pjl;
1688         int ret = 0;
1689
1690         jl->j_state |= LIST_DIRTY;
1691         cn = jl->j_realblock;
1692         while (cn) {
1693                 /*
1694                  * look for a more recent transaction that logged this
1695                  * buffer.  Only the most recent transaction with a buffer in
1696                  * it is allowed to send that buffer to disk
1697                  */
1698                 pjl = find_newer_jl_for_cn(cn);
1699                 if (!pjl && cn->blocknr && cn->bh
1700                     && buffer_journal_dirty(cn->bh)) {
1701                         BUG_ON(!can_dirty(cn));
1702                         /*
1703                          * if the buffer is prepared, it will either be logged
1704                          * or restored.  If restored, we need to make sure
1705                          * it actually gets marked dirty
1706                          */
1707                         clear_buffer_journal_new(cn->bh);
1708                         if (buffer_journal_prepared(cn->bh)) {
1709                                 set_buffer_journal_restore_dirty(cn->bh);
1710                         } else {
1711                                 set_buffer_journal_test(cn->bh);
1712                                 mark_buffer_dirty(cn->bh);
1713                         }
1714                 }
1715                 cn = cn->next;
1716         }
1717         return ret;
1718 }
1719
1720 static int kupdate_transactions(struct super_block *s,
1721                                 struct reiserfs_journal_list *jl,
1722                                 struct reiserfs_journal_list **next_jl,
1723                                 unsigned int *next_trans_id,
1724                                 int num_blocks, int num_trans)
1725 {
1726         int ret = 0;
1727         int written = 0;
1728         int transactions_flushed = 0;
1729         unsigned int orig_trans_id = jl->j_trans_id;
1730         struct buffer_chunk chunk;
1731         struct list_head *entry;
1732         struct reiserfs_journal *journal = SB_JOURNAL(s);
1733         chunk.nr = 0;
1734
1735         reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1736         if (!journal_list_still_alive(s, orig_trans_id)) {
1737                 goto done;
1738         }
1739
1740         /*
1741          * we've got j_flush_mutex held, nobody is going to delete any
1742          * of these lists out from underneath us
1743          */
1744         while ((num_trans && transactions_flushed < num_trans) ||
1745                (!num_trans && written < num_blocks)) {
1746
1747                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1748                     atomic_read(&jl->j_commit_left)
1749                     || !(jl->j_state & LIST_DIRTY)) {
1750                         del_from_work_list(s, jl);
1751                         break;
1752                 }
1753                 ret = write_one_transaction(s, jl, &chunk);
1754
1755                 if (ret < 0)
1756                         goto done;
1757                 transactions_flushed++;
1758                 written += ret;
1759                 entry = jl->j_list.next;
1760
1761                 /* did we wrap? */
1762                 if (entry == &journal->j_journal_list) {
1763                         break;
1764                 }
1765                 jl = JOURNAL_LIST_ENTRY(entry);
1766
1767                 /* don't bother with older transactions */
1768                 if (jl->j_trans_id <= orig_trans_id)
1769                         break;
1770         }
1771         if (chunk.nr) {
1772                 write_chunk(&chunk);
1773         }
1774
1775 done:
1776         mutex_unlock(&journal->j_flush_mutex);
1777         return ret;
1778 }
1779
1780 /*
1781  * for o_sync and fsync heavy applications, they tend to use
1782  * all the journa list slots with tiny transactions.  These
1783  * trigger lots and lots of calls to update the header block, which
1784  * adds seeks and slows things down.
1785  *
1786  * This function tries to clear out a large chunk of the journal lists
1787  * at once, which makes everything faster since only the newest journal
1788  * list updates the header block
1789  */
1790 static int flush_used_journal_lists(struct super_block *s,
1791                                     struct reiserfs_journal_list *jl)
1792 {
1793         unsigned long len = 0;
1794         unsigned long cur_len;
1795         int ret;
1796         int i;
1797         int limit = 256;
1798         struct reiserfs_journal_list *tjl;
1799         struct reiserfs_journal_list *flush_jl;
1800         unsigned int trans_id;
1801         struct reiserfs_journal *journal = SB_JOURNAL(s);
1802
1803         flush_jl = tjl = jl;
1804
1805         /* in data logging mode, try harder to flush a lot of blocks */
1806         if (reiserfs_data_log(s))
1807                 limit = 1024;
1808         /* flush for 256 transactions or limit blocks, whichever comes first */
1809         for (i = 0; i < 256 && len < limit; i++) {
1810                 if (atomic_read(&tjl->j_commit_left) ||
1811                     tjl->j_trans_id < jl->j_trans_id) {
1812                         break;
1813                 }
1814                 cur_len = atomic_read(&tjl->j_nonzerolen);
1815                 if (cur_len > 0) {
1816                         tjl->j_state &= ~LIST_TOUCHED;
1817                 }
1818                 len += cur_len;
1819                 flush_jl = tjl;
1820                 if (tjl->j_list.next == &journal->j_journal_list)
1821                         break;
1822                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1823         }
1824         get_journal_list(jl);
1825         get_journal_list(flush_jl);
1826
1827         /*
1828          * try to find a group of blocks we can flush across all the
1829          * transactions, but only bother if we've actually spanned
1830          * across multiple lists
1831          */
1832         if (flush_jl != jl) {
1833                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1834         }
1835         flush_journal_list(s, flush_jl, 1);
1836         put_journal_list(s, flush_jl);
1837         put_journal_list(s, jl);
1838         return 0;
1839 }
1840
1841 /*
1842  * removes any nodes in table with name block and dev as bh.
1843  * only touchs the hnext and hprev pointers.
1844  */
1845 void remove_journal_hash(struct super_block *sb,
1846                          struct reiserfs_journal_cnode **table,
1847                          struct reiserfs_journal_list *jl,
1848                          unsigned long block, int remove_freed)
1849 {
1850         struct reiserfs_journal_cnode *cur;
1851         struct reiserfs_journal_cnode **head;
1852
1853         head = &(journal_hash(table, sb, block));
1854         if (!head) {
1855                 return;
1856         }
1857         cur = *head;
1858         while (cur) {
1859                 if (cur->blocknr == block && cur->sb == sb
1860                     && (jl == NULL || jl == cur->jlist)
1861                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1862                         if (cur->hnext) {
1863                                 cur->hnext->hprev = cur->hprev;
1864                         }
1865                         if (cur->hprev) {
1866                                 cur->hprev->hnext = cur->hnext;
1867                         } else {
1868                                 *head = cur->hnext;
1869                         }
1870                         cur->blocknr = 0;
1871                         cur->sb = NULL;
1872                         cur->state = 0;
1873                         /*
1874                          * anybody who clears the cur->bh will also
1875                          * dec the nonzerolen
1876                          */
1877                         if (cur->bh && cur->jlist)
1878                                 atomic_dec(&cur->jlist->j_nonzerolen);
1879                         cur->bh = NULL;
1880                         cur->jlist = NULL;
1881                 }
1882                 cur = cur->hnext;
1883         }
1884 }
1885
1886 static void free_journal_ram(struct super_block *sb)
1887 {
1888         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1889         kfree(journal->j_current_jl);
1890         journal->j_num_lists--;
1891
1892         vfree(journal->j_cnode_free_orig);
1893         free_list_bitmaps(sb, journal->j_list_bitmap);
1894         free_bitmap_nodes(sb);  /* must be after free_list_bitmaps */
1895         if (journal->j_header_bh) {
1896                 brelse(journal->j_header_bh);
1897         }
1898         /*
1899          * j_header_bh is on the journal dev, make sure
1900          * not to release the journal dev until we brelse j_header_bh
1901          */
1902         release_journal_dev(sb, journal);
1903         vfree(journal);
1904 }
1905
1906 /*
1907  * call on unmount.  Only set error to 1 if you haven't made your way out
1908  * of read_super() yet.  Any other caller must keep error at 0.
1909  */
1910 static int do_journal_release(struct reiserfs_transaction_handle *th,
1911                               struct super_block *sb, int error)
1912 {
1913         struct reiserfs_transaction_handle myth;
1914         int flushed = 0;
1915         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1916
1917         /*
1918          * we only want to flush out transactions if we were
1919          * called with error == 0
1920          */
1921         if (!error && !(sb->s_flags & MS_RDONLY)) {
1922                 /* end the current trans */
1923                 BUG_ON(!th->t_trans_id);
1924                 do_journal_end(th, FLUSH_ALL);
1925
1926                 /*
1927                  * make sure something gets logged to force
1928                  * our way into the flush code
1929                  */
1930                 if (!journal_join(&myth, sb)) {
1931                         reiserfs_prepare_for_journal(sb,
1932                                                      SB_BUFFER_WITH_SB(sb),
1933                                                      1);
1934                         journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1935                         do_journal_end(&myth, FLUSH_ALL);
1936                         flushed = 1;
1937                 }
1938         }
1939
1940         /* this also catches errors during the do_journal_end above */
1941         if (!error && reiserfs_is_journal_aborted(journal)) {
1942                 memset(&myth, 0, sizeof(myth));
1943                 if (!journal_join_abort(&myth, sb)) {
1944                         reiserfs_prepare_for_journal(sb,
1945                                                      SB_BUFFER_WITH_SB(sb),
1946                                                      1);
1947                         journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1948                         do_journal_end(&myth, FLUSH_ALL);
1949                 }
1950         }
1951
1952
1953         /*
1954          * We must release the write lock here because
1955          * the workqueue job (flush_async_commit) needs this lock
1956          */
1957         reiserfs_write_unlock(sb);
1958
1959         /*
1960          * Cancel flushing of old commits. Note that neither of these works
1961          * will be requeued because superblock is being shutdown and doesn't
1962          * have MS_ACTIVE set.
1963          */
1964         reiserfs_cancel_old_flush(sb);
1965         /* wait for all commits to finish */
1966         cancel_delayed_work_sync(&SB_JOURNAL(sb)->j_work);
1967
1968         free_journal_ram(sb);
1969
1970         reiserfs_write_lock(sb);
1971
1972         return 0;
1973 }
1974
1975 /* * call on unmount.  flush all journal trans, release all alloc'd ram */
1976 int journal_release(struct reiserfs_transaction_handle *th,
1977                     struct super_block *sb)
1978 {
1979         return do_journal_release(th, sb, 0);
1980 }
1981
1982 /* only call from an error condition inside reiserfs_read_super!  */
1983 int journal_release_error(struct reiserfs_transaction_handle *th,
1984                           struct super_block *sb)
1985 {
1986         return do_journal_release(th, sb, 1);
1987 }
1988
1989 /*
1990  * compares description block with commit block.
1991  * returns 1 if they differ, 0 if they are the same
1992  */
1993 static int journal_compare_desc_commit(struct super_block *sb,
1994                                        struct reiserfs_journal_desc *desc,
1995                                        struct reiserfs_journal_commit *commit)
1996 {
1997         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1998             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1999             get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
2000             get_commit_trans_len(commit) <= 0) {
2001                 return 1;
2002         }
2003         return 0;
2004 }
2005
2006 /*
2007  * returns 0 if it did not find a description block
2008  * returns -1 if it found a corrupt commit block
2009  * returns 1 if both desc and commit were valid
2010  * NOTE: only called during fs mount
2011  */
2012 static int journal_transaction_is_valid(struct super_block *sb,
2013                                         struct buffer_head *d_bh,
2014                                         unsigned int *oldest_invalid_trans_id,
2015                                         unsigned long *newest_mount_id)
2016 {
2017         struct reiserfs_journal_desc *desc;
2018         struct reiserfs_journal_commit *commit;
2019         struct buffer_head *c_bh;
2020         unsigned long offset;
2021
2022         if (!d_bh)
2023                 return 0;
2024
2025         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2026         if (get_desc_trans_len(desc) > 0
2027             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2028                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2029                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2030                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2031                                        "journal-986: transaction "
2032                                        "is valid returning because trans_id %d is greater than "
2033                                        "oldest_invalid %lu",
2034                                        get_desc_trans_id(desc),
2035                                        *oldest_invalid_trans_id);
2036                         return 0;
2037                 }
2038                 if (newest_mount_id
2039                     && *newest_mount_id > get_desc_mount_id(desc)) {
2040                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2041                                        "journal-1087: transaction "
2042                                        "is valid returning because mount_id %d is less than "
2043                                        "newest_mount_id %lu",
2044                                        get_desc_mount_id(desc),
2045                                        *newest_mount_id);
2046                         return -1;
2047                 }
2048                 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2049                         reiserfs_warning(sb, "journal-2018",
2050                                          "Bad transaction length %d "
2051                                          "encountered, ignoring transaction",
2052                                          get_desc_trans_len(desc));
2053                         return -1;
2054                 }
2055                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2056
2057                 /*
2058                  * ok, we have a journal description block,
2059                  * let's see if the transaction was valid
2060                  */
2061                 c_bh =
2062                     journal_bread(sb,
2063                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2064                                   ((offset + get_desc_trans_len(desc) +
2065                                     1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2066                 if (!c_bh)
2067                         return 0;
2068                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2069                 if (journal_compare_desc_commit(sb, desc, commit)) {
2070                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2071                                        "journal_transaction_is_valid, commit offset %ld had bad "
2072                                        "time %d or length %d",
2073                                        c_bh->b_blocknr -
2074                                        SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2075                                        get_commit_trans_id(commit),
2076                                        get_commit_trans_len(commit));
2077                         brelse(c_bh);
2078                         if (oldest_invalid_trans_id) {
2079                                 *oldest_invalid_trans_id =
2080                                     get_desc_trans_id(desc);
2081                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2082                                                "journal-1004: "
2083                                                "transaction_is_valid setting oldest invalid trans_id "
2084                                                "to %d",
2085                                                get_desc_trans_id(desc));
2086                         }
2087                         return -1;
2088                 }
2089                 brelse(c_bh);
2090                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2091                                "journal-1006: found valid "
2092                                "transaction start offset %llu, len %d id %d",
2093                                d_bh->b_blocknr -
2094                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2095                                get_desc_trans_len(desc),
2096                                get_desc_trans_id(desc));
2097                 return 1;
2098         } else {
2099                 return 0;
2100         }
2101 }
2102
2103 static void brelse_array(struct buffer_head **heads, int num)
2104 {
2105         int i;
2106         for (i = 0; i < num; i++) {
2107                 brelse(heads[i]);
2108         }
2109 }
2110
2111 /*
2112  * given the start, and values for the oldest acceptable transactions,
2113  * this either reads in a replays a transaction, or returns because the
2114  * transaction is invalid, or too old.
2115  * NOTE: only called during fs mount
2116  */
2117 static int journal_read_transaction(struct super_block *sb,
2118                                     unsigned long cur_dblock,
2119                                     unsigned long oldest_start,
2120                                     unsigned int oldest_trans_id,
2121                                     unsigned long newest_mount_id)
2122 {
2123         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2124         struct reiserfs_journal_desc *desc;
2125         struct reiserfs_journal_commit *commit;
2126         unsigned int trans_id = 0;
2127         struct buffer_head *c_bh;
2128         struct buffer_head *d_bh;
2129         struct buffer_head **log_blocks = NULL;
2130         struct buffer_head **real_blocks = NULL;
2131         unsigned int trans_offset;
2132         int i;
2133         int trans_half;
2134
2135         d_bh = journal_bread(sb, cur_dblock);
2136         if (!d_bh)
2137                 return 1;
2138         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2139         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2140         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2141                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2142                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2143                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2144         if (get_desc_trans_id(desc) < oldest_trans_id) {
2145                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2146                                "journal_read_trans skipping because %lu is too old",
2147                                cur_dblock -
2148                                SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2149                 brelse(d_bh);
2150                 return 1;
2151         }
2152         if (get_desc_mount_id(desc) != newest_mount_id) {
2153                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2154                                "journal_read_trans skipping because %d is != "
2155                                "newest_mount_id %lu", get_desc_mount_id(desc),
2156                                newest_mount_id);
2157                 brelse(d_bh);
2158                 return 1;
2159         }
2160         c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2161                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2162                               SB_ONDISK_JOURNAL_SIZE(sb)));
2163         if (!c_bh) {
2164                 brelse(d_bh);
2165                 return 1;
2166         }
2167         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2168         if (journal_compare_desc_commit(sb, desc, commit)) {
2169                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2170                                "journal_read_transaction, "
2171                                "commit offset %llu had bad time %d or length %d",
2172                                c_bh->b_blocknr -
2173                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2174                                get_commit_trans_id(commit),
2175                                get_commit_trans_len(commit));
2176                 brelse(c_bh);
2177                 brelse(d_bh);
2178                 return 1;
2179         }
2180
2181         if (bdev_read_only(sb->s_bdev)) {
2182                 reiserfs_warning(sb, "clm-2076",
2183                                  "device is readonly, unable to replay log");
2184                 brelse(c_bh);
2185                 brelse(d_bh);
2186                 return -EROFS;
2187         }
2188
2189         trans_id = get_desc_trans_id(desc);
2190         /*
2191          * now we know we've got a good transaction, and it was
2192          * inside the valid time ranges
2193          */
2194         log_blocks = kmalloc(get_desc_trans_len(desc) *
2195                              sizeof(struct buffer_head *), GFP_NOFS);
2196         real_blocks = kmalloc(get_desc_trans_len(desc) *
2197                               sizeof(struct buffer_head *), GFP_NOFS);
2198         if (!log_blocks || !real_blocks) {
2199                 brelse(c_bh);
2200                 brelse(d_bh);
2201                 kfree(log_blocks);
2202                 kfree(real_blocks);
2203                 reiserfs_warning(sb, "journal-1169",
2204                                  "kmalloc failed, unable to mount FS");
2205                 return -1;
2206         }
2207         /* get all the buffer heads */
2208         trans_half = journal_trans_half(sb->s_blocksize);
2209         for (i = 0; i < get_desc_trans_len(desc); i++) {
2210                 log_blocks[i] =
2211                     journal_getblk(sb,
2212                                    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2213                                    (trans_offset + 1 +
2214                                     i) % SB_ONDISK_JOURNAL_SIZE(sb));
2215                 if (i < trans_half) {
2216                         real_blocks[i] =
2217                             sb_getblk(sb,
2218                                       le32_to_cpu(desc->j_realblock[i]));
2219                 } else {
2220                         real_blocks[i] =
2221                             sb_getblk(sb,
2222                                       le32_to_cpu(commit->
2223                                                   j_realblock[i - trans_half]));
2224                 }
2225                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2226                         reiserfs_warning(sb, "journal-1207",
2227                                          "REPLAY FAILURE fsck required! "
2228                                          "Block to replay is outside of "
2229                                          "filesystem");
2230                         goto abort_replay;
2231                 }
2232                 /* make sure we don't try to replay onto log or reserved area */
2233                 if (is_block_in_log_or_reserved_area
2234                     (sb, real_blocks[i]->b_blocknr)) {
2235                         reiserfs_warning(sb, "journal-1204",
2236                                          "REPLAY FAILURE fsck required! "
2237                                          "Trying to replay onto a log block");
2238 abort_replay:
2239                         brelse_array(log_blocks, i);
2240                         brelse_array(real_blocks, i);
2241                         brelse(c_bh);
2242                         brelse(d_bh);
2243                         kfree(log_blocks);
2244                         kfree(real_blocks);
2245                         return -1;
2246                 }
2247         }
2248         /* read in the log blocks, memcpy to the corresponding real block */
2249         ll_rw_block(REQ_OP_READ, 0, get_desc_trans_len(desc), log_blocks);
2250         for (i = 0; i < get_desc_trans_len(desc); i++) {
2251
2252                 wait_on_buffer(log_blocks[i]);
2253                 if (!buffer_uptodate(log_blocks[i])) {
2254                         reiserfs_warning(sb, "journal-1212",
2255                                          "REPLAY FAILURE fsck required! "
2256                                          "buffer write failed");
2257                         brelse_array(log_blocks + i,
2258                                      get_desc_trans_len(desc) - i);
2259                         brelse_array(real_blocks, get_desc_trans_len(desc));
2260                         brelse(c_bh);
2261                         brelse(d_bh);
2262                         kfree(log_blocks);
2263                         kfree(real_blocks);
2264                         return -1;
2265                 }
2266                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2267                        real_blocks[i]->b_size);
2268                 set_buffer_uptodate(real_blocks[i]);
2269                 brelse(log_blocks[i]);
2270         }
2271         /* flush out the real blocks */
2272         for (i = 0; i < get_desc_trans_len(desc); i++) {
2273                 set_buffer_dirty(real_blocks[i]);
2274                 write_dirty_buffer(real_blocks[i], 0);
2275         }
2276         for (i = 0; i < get_desc_trans_len(desc); i++) {
2277                 wait_on_buffer(real_blocks[i]);
2278                 if (!buffer_uptodate(real_blocks[i])) {
2279                         reiserfs_warning(sb, "journal-1226",
2280                                          "REPLAY FAILURE, fsck required! "
2281                                          "buffer write failed");
2282                         brelse_array(real_blocks + i,
2283                                      get_desc_trans_len(desc) - i);
2284                         brelse(c_bh);
2285                         brelse(d_bh);
2286                         kfree(log_blocks);
2287                         kfree(real_blocks);
2288                         return -1;
2289                 }
2290                 brelse(real_blocks[i]);
2291         }
2292         cur_dblock =
2293             SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2294             ((trans_offset + get_desc_trans_len(desc) +
2295               2) % SB_ONDISK_JOURNAL_SIZE(sb));
2296         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2297                        "journal-1095: setting journal " "start to offset %ld",
2298                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2299
2300         /*
2301          * init starting values for the first transaction, in case
2302          * this is the last transaction to be replayed.
2303          */
2304         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2305         journal->j_last_flush_trans_id = trans_id;
2306         journal->j_trans_id = trans_id + 1;
2307         /* check for trans_id overflow */
2308         if (journal->j_trans_id == 0)
2309                 journal->j_trans_id = 10;
2310         brelse(c_bh);
2311         brelse(d_bh);
2312         kfree(log_blocks);
2313         kfree(real_blocks);
2314         return 0;
2315 }
2316
2317 /*
2318  * This function reads blocks starting from block and to max_block of bufsize
2319  * size (but no more than BUFNR blocks at a time). This proved to improve
2320  * mounting speed on self-rebuilding raid5 arrays at least.
2321  * Right now it is only used from journal code. But later we might use it
2322  * from other places.
2323  * Note: Do not use journal_getblk/sb_getblk functions here!
2324  */
2325 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2326                                            b_blocknr_t block, int bufsize,
2327                                            b_blocknr_t max_block)
2328 {
2329         struct buffer_head *bhlist[BUFNR];
2330         unsigned int blocks = BUFNR;
2331         struct buffer_head *bh;
2332         int i, j;
2333
2334         bh = __getblk(dev, block, bufsize);
2335         if (buffer_uptodate(bh))
2336                 return (bh);
2337
2338         if (block + BUFNR > max_block) {
2339                 blocks = max_block - block;
2340         }
2341         bhlist[0] = bh;
2342         j = 1;
2343         for (i = 1; i < blocks; i++) {
2344                 bh = __getblk(dev, block + i, bufsize);
2345                 if (buffer_uptodate(bh)) {
2346                         brelse(bh);
2347                         break;
2348                 } else
2349                         bhlist[j++] = bh;
2350         }
2351         ll_rw_block(REQ_OP_READ, 0, j, bhlist);
2352         for (i = 1; i < j; i++)
2353                 brelse(bhlist[i]);
2354         bh = bhlist[0];
2355         wait_on_buffer(bh);
2356         if (buffer_uptodate(bh))
2357                 return bh;
2358         brelse(bh);
2359         return NULL;
2360 }
2361
2362 /*
2363  * read and replay the log
2364  * on a clean unmount, the journal header's next unflushed pointer will be
2365  * to an invalid transaction.  This tests that before finding all the
2366  * transactions in the log, which makes normal mount times fast.
2367  *
2368  * After a crash, this starts with the next unflushed transaction, and
2369  * replays until it finds one too old, or invalid.
2370  *
2371  * On exit, it sets things up so the first transaction will work correctly.
2372  * NOTE: only called during fs mount
2373  */
2374 static int journal_read(struct super_block *sb)
2375 {
2376         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2377         struct reiserfs_journal_desc *desc;
2378         unsigned int oldest_trans_id = 0;
2379         unsigned int oldest_invalid_trans_id = 0;
2380         time_t start;
2381         unsigned long oldest_start = 0;
2382         unsigned long cur_dblock = 0;
2383         unsigned long newest_mount_id = 9;
2384         struct buffer_head *d_bh;
2385         struct reiserfs_journal_header *jh;
2386         int valid_journal_header = 0;
2387         int replay_count = 0;
2388         int continue_replay = 1;
2389         int ret;
2390
2391         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2392         reiserfs_info(sb, "checking transaction log (%pg)\n",
2393                       journal->j_dev_bd);
2394         start = get_seconds();
2395
2396         /*
2397          * step 1, read in the journal header block.  Check the transaction
2398          * it says is the first unflushed, and if that transaction is not
2399          * valid, replay is done
2400          */
2401         journal->j_header_bh = journal_bread(sb,
2402                                              SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2403                                              + SB_ONDISK_JOURNAL_SIZE(sb));
2404         if (!journal->j_header_bh) {
2405                 return 1;
2406         }
2407         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2408         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2409             SB_ONDISK_JOURNAL_SIZE(sb)
2410             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2411                 oldest_start =
2412                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2413                     le32_to_cpu(jh->j_first_unflushed_offset);
2414                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2415                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2416                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2417                                "journal-1153: found in "
2418                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2419                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2420                                le32_to_cpu(jh->j_last_flush_trans_id));
2421                 valid_journal_header = 1;
2422
2423                 /*
2424                  * now, we try to read the first unflushed offset.  If it
2425                  * is not valid, there is nothing more we can do, and it
2426                  * makes no sense to read through the whole log.
2427                  */
2428                 d_bh =
2429                     journal_bread(sb,
2430                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2431                                   le32_to_cpu(jh->j_first_unflushed_offset));
2432                 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2433                 if (!ret) {
2434                         continue_replay = 0;
2435                 }
2436                 brelse(d_bh);
2437                 goto start_log_replay;
2438         }
2439
2440         /*
2441          * ok, there are transactions that need to be replayed.  start
2442          * with the first log block, find all the valid transactions, and
2443          * pick out the oldest.
2444          */
2445         while (continue_replay
2446                && cur_dblock <
2447                (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2448                 SB_ONDISK_JOURNAL_SIZE(sb))) {
2449                 /*
2450                  * Note that it is required for blocksize of primary fs
2451                  * device and journal device to be the same
2452                  */
2453                 d_bh =
2454                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2455                                     sb->s_blocksize,
2456                                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2457                                     SB_ONDISK_JOURNAL_SIZE(sb));
2458                 ret =
2459                     journal_transaction_is_valid(sb, d_bh,
2460                                                  &oldest_invalid_trans_id,
2461                                                  &newest_mount_id);
2462                 if (ret == 1) {
2463                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2464                         if (oldest_start == 0) {        /* init all oldest_ values */
2465                                 oldest_trans_id = get_desc_trans_id(desc);
2466                                 oldest_start = d_bh->b_blocknr;
2467                                 newest_mount_id = get_desc_mount_id(desc);
2468                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2469                                                "journal-1179: Setting "
2470                                                "oldest_start to offset %llu, trans_id %lu",
2471                                                oldest_start -
2472                                                SB_ONDISK_JOURNAL_1st_BLOCK
2473                                                (sb), oldest_trans_id);
2474                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2475                                 /* one we just read was older */
2476                                 oldest_trans_id = get_desc_trans_id(desc);
2477                                 oldest_start = d_bh->b_blocknr;
2478                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2479                                                "journal-1180: Resetting "
2480                                                "oldest_start to offset %lu, trans_id %lu",
2481                                                oldest_start -
2482                                                SB_ONDISK_JOURNAL_1st_BLOCK
2483                                                (sb), oldest_trans_id);
2484                         }
2485                         if (newest_mount_id < get_desc_mount_id(desc)) {
2486                                 newest_mount_id = get_desc_mount_id(desc);
2487                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2488                                                "journal-1299: Setting "
2489                                                "newest_mount_id to %d",
2490                                                get_desc_mount_id(desc));
2491                         }
2492                         cur_dblock += get_desc_trans_len(desc) + 2;
2493                 } else {
2494                         cur_dblock++;
2495                 }
2496                 brelse(d_bh);
2497         }
2498
2499 start_log_replay:
2500         cur_dblock = oldest_start;
2501         if (oldest_trans_id) {
2502                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2503                                "journal-1206: Starting replay "
2504                                "from offset %llu, trans_id %lu",
2505                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2506                                oldest_trans_id);
2507
2508         }
2509         replay_count = 0;
2510         while (continue_replay && oldest_trans_id > 0) {
2511                 ret =
2512                     journal_read_transaction(sb, cur_dblock, oldest_start,
2513                                              oldest_trans_id, newest_mount_id);
2514                 if (ret < 0) {
2515                         return ret;
2516                 } else if (ret != 0) {
2517                         break;
2518                 }
2519                 cur_dblock =
2520                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2521                 replay_count++;
2522                 if (cur_dblock == oldest_start)
2523                         break;
2524         }
2525
2526         if (oldest_trans_id == 0) {
2527                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2528                                "journal-1225: No valid " "transactions found");
2529         }
2530         /*
2531          * j_start does not get set correctly if we don't replay any
2532          * transactions.  if we had a valid journal_header, set j_start
2533          * to the first unflushed transaction value, copy the trans_id
2534          * from the header
2535          */
2536         if (valid_journal_header && replay_count == 0) {
2537                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2538                 journal->j_trans_id =
2539                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2540                 /* check for trans_id overflow */
2541                 if (journal->j_trans_id == 0)
2542                         journal->j_trans_id = 10;
2543                 journal->j_last_flush_trans_id =
2544                     le32_to_cpu(jh->j_last_flush_trans_id);
2545                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2546         } else {
2547                 journal->j_mount_id = newest_mount_id + 1;
2548         }
2549         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2550                        "newest_mount_id to %lu", journal->j_mount_id);
2551         journal->j_first_unflushed_offset = journal->j_start;
2552         if (replay_count > 0) {
2553                 reiserfs_info(sb,
2554                               "replayed %d transactions in %lu seconds\n",
2555                               replay_count, get_seconds() - start);
2556         }
2557         /* needed to satisfy the locking in _update_journal_header_block */
2558         reiserfs_write_lock(sb);
2559         if (!bdev_read_only(sb->s_bdev) &&
2560             _update_journal_header_block(sb, journal->j_start,
2561                                          journal->j_last_flush_trans_id)) {
2562                 reiserfs_write_unlock(sb);
2563                 /*
2564                  * replay failed, caller must call free_journal_ram and abort
2565                  * the mount
2566                  */
2567                 return -1;
2568         }
2569         reiserfs_write_unlock(sb);
2570         return 0;
2571 }
2572
2573 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2574 {
2575         struct reiserfs_journal_list *jl;
2576         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2577                      GFP_NOFS | __GFP_NOFAIL);
2578         INIT_LIST_HEAD(&jl->j_list);
2579         INIT_LIST_HEAD(&jl->j_working_list);
2580         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2581         INIT_LIST_HEAD(&jl->j_bh_list);
2582         mutex_init(&jl->j_commit_mutex);
2583         SB_JOURNAL(s)->j_num_lists++;
2584         get_journal_list(jl);
2585         return jl;
2586 }
2587
2588 static void journal_list_init(struct super_block *sb)
2589 {
2590         SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2591 }
2592
2593 static void release_journal_dev(struct super_block *super,
2594                                struct reiserfs_journal *journal)
2595 {
2596         if (journal->j_dev_bd != NULL) {
2597                 blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2598                 journal->j_dev_bd = NULL;
2599         }
2600 }
2601
2602 static int journal_init_dev(struct super_block *super,
2603                             struct reiserfs_journal *journal,
2604                             const char *jdev_name)
2605 {
2606         int result;
2607         dev_t jdev;
2608         fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
2609         char b[BDEVNAME_SIZE];
2610
2611         result = 0;
2612
2613         journal->j_dev_bd = NULL;
2614         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2615             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2616
2617         if (bdev_read_only(super->s_bdev))
2618                 blkdev_mode = FMODE_READ;
2619
2620         /* there is no "jdev" option and journal is on separate device */
2621         if ((!jdev_name || !jdev_name[0])) {
2622                 if (jdev == super->s_dev)
2623                         blkdev_mode &= ~FMODE_EXCL;
2624                 journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2625                                                       journal);
2626                 journal->j_dev_mode = blkdev_mode;
2627                 if (IS_ERR(journal->j_dev_bd)) {
2628                         result = PTR_ERR(journal->j_dev_bd);
2629                         journal->j_dev_bd = NULL;
2630                         reiserfs_warning(super, "sh-458",
2631                                          "cannot init journal device '%s': %i",
2632                                          __bdevname(jdev, b), result);
2633                         return result;
2634                 } else if (jdev != super->s_dev)
2635                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2636
2637                 return 0;
2638         }
2639
2640         journal->j_dev_mode = blkdev_mode;
2641         journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
2642         if (IS_ERR(journal->j_dev_bd)) {
2643                 result = PTR_ERR(journal->j_dev_bd);
2644                 journal->j_dev_bd = NULL;
2645                 reiserfs_warning(super,
2646                                  "journal_init_dev: Cannot open '%s': %i",
2647                                  jdev_name, result);
2648                 return result;
2649         }
2650
2651         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2652         reiserfs_info(super,
2653                       "journal_init_dev: journal device: %pg\n",
2654                       journal->j_dev_bd);
2655         return 0;
2656 }
2657
2658 /*
2659  * When creating/tuning a file system user can assign some
2660  * journal params within boundaries which depend on the ratio
2661  * blocksize/standard_blocksize.
2662  *
2663  * For blocks >= standard_blocksize transaction size should
2664  * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2665  * then JOURNAL_TRANS_MAX_DEFAULT.
2666  *
2667  * For blocks < standard_blocksize these boundaries should be
2668  * decreased proportionally.
2669  */
2670 #define REISERFS_STANDARD_BLKSIZE (4096)
2671
2672 static int check_advise_trans_params(struct super_block *sb,
2673                                      struct reiserfs_journal *journal)
2674 {
2675         if (journal->j_trans_max) {
2676                 /* Non-default journal params.  Do sanity check for them. */
2677                 int ratio = 1;
2678                 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2679                         ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2680
2681                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2682                     journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2683                     SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2684                     JOURNAL_MIN_RATIO) {
2685                         reiserfs_warning(sb, "sh-462",
2686                                          "bad transaction max size (%u). "
2687                                          "FSCK?", journal->j_trans_max);
2688                         return 1;
2689                 }
2690                 if (journal->j_max_batch != (journal->j_trans_max) *
2691                         JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2692                         reiserfs_warning(sb, "sh-463",
2693                                          "bad transaction max batch (%u). "
2694                                          "FSCK?", journal->j_max_batch);
2695                         return 1;
2696                 }
2697         } else {
2698                 /*
2699                  * Default journal params.
2700                  * The file system was created by old version
2701                  * of mkreiserfs, so some fields contain zeros,
2702                  * and we need to advise proper values for them
2703                  */
2704                 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2705                         reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2706                                          sb->s_blocksize);
2707                         return 1;
2708                 }
2709                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2710                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2711                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2712         }
2713         return 0;
2714 }
2715
2716 /* must be called once on fs mount.  calls journal_read for you */
2717 int journal_init(struct super_block *sb, const char *j_dev_name,
2718                  int old_format, unsigned int commit_max_age)
2719 {
2720         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2721         struct buffer_head *bhjh;
2722         struct reiserfs_super_block *rs;
2723         struct reiserfs_journal_header *jh;
2724         struct reiserfs_journal *journal;
2725         struct reiserfs_journal_list *jl;
2726         int ret;
2727
2728         journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal));
2729         if (!journal) {
2730                 reiserfs_warning(sb, "journal-1256",
2731                                  "unable to get memory for journal structure");
2732                 return 1;
2733         }
2734         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2735         INIT_LIST_HEAD(&journal->j_prealloc_list);
2736         INIT_LIST_HEAD(&journal->j_working_list);
2737         INIT_LIST_HEAD(&journal->j_journal_list);
2738         journal->j_persistent_trans = 0;
2739         if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2740                                            reiserfs_bmap_count(sb)))
2741                 goto free_and_return;
2742
2743         allocate_bitmap_nodes(sb);
2744
2745         /* reserved for journal area support */
2746         SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2747                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2748                                                  / sb->s_blocksize +
2749                                                  reiserfs_bmap_count(sb) +
2750                                                  1 :
2751                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2752                                                  sb->s_blocksize + 2);
2753
2754         /*
2755          * Sanity check to see is the standard journal fitting
2756          * within first bitmap (actual for small blocksizes)
2757          */
2758         if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2759             (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2760              SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2761                 reiserfs_warning(sb, "journal-1393",
2762                                  "journal does not fit for area addressed "
2763                                  "by first of bitmap blocks. It starts at "
2764                                  "%u and its size is %u. Block size %ld",
2765                                  SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2766                                  SB_ONDISK_JOURNAL_SIZE(sb),
2767                                  sb->s_blocksize);
2768                 goto free_and_return;
2769         }
2770
2771         if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2772                 reiserfs_warning(sb, "sh-462",
2773                                  "unable to initialize journal device");
2774                 goto free_and_return;
2775         }
2776
2777         rs = SB_DISK_SUPER_BLOCK(sb);
2778
2779         /* read journal header */
2780         bhjh = journal_bread(sb,
2781                              SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2782                              SB_ONDISK_JOURNAL_SIZE(sb));
2783         if (!bhjh) {
2784                 reiserfs_warning(sb, "sh-459",
2785                                  "unable to read journal header");
2786                 goto free_and_return;
2787         }
2788         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2789
2790         /* make sure that journal matches to the super block */
2791         if (is_reiserfs_jr(rs)
2792             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2793                 sb_jp_journal_magic(rs))) {
2794                 reiserfs_warning(sb, "sh-460",
2795                                  "journal header magic %x (device %pg) does "
2796                                  "not match to magic found in super block %x",
2797                                  jh->jh_journal.jp_journal_magic,
2798                                  journal->j_dev_bd,
2799                                  sb_jp_journal_magic(rs));
2800                 brelse(bhjh);
2801                 goto free_and_return;
2802         }
2803
2804         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2805         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2806         journal->j_max_commit_age =
2807             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2808         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2809
2810         if (check_advise_trans_params(sb, journal) != 0)
2811                 goto free_and_return;
2812         journal->j_default_max_commit_age = journal->j_max_commit_age;
2813
2814         if (commit_max_age != 0) {
2815                 journal->j_max_commit_age = commit_max_age;
2816                 journal->j_max_trans_age = commit_max_age;
2817         }
2818
2819         reiserfs_info(sb, "journal params: device %pg, size %u, "
2820                       "journal first block %u, max trans len %u, max batch %u, "
2821                       "max commit age %u, max trans age %u\n",
2822                       journal->j_dev_bd,
2823                       SB_ONDISK_JOURNAL_SIZE(sb),
2824                       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2825                       journal->j_trans_max,
2826                       journal->j_max_batch,
2827                       journal->j_max_commit_age, journal->j_max_trans_age);
2828
2829         brelse(bhjh);
2830
2831         journal->j_list_bitmap_index = 0;
2832         journal_list_init(sb);
2833
2834         memset(journal->j_list_hash_table, 0,
2835                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2836
2837         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2838         spin_lock_init(&journal->j_dirty_buffers_lock);
2839
2840         journal->j_start = 0;
2841         journal->j_len = 0;
2842         journal->j_len_alloc = 0;
2843         atomic_set(&journal->j_wcount, 0);
2844         atomic_set(&journal->j_async_throttle, 0);
2845         journal->j_bcount = 0;
2846         journal->j_trans_start_time = 0;
2847         journal->j_last = NULL;
2848         journal->j_first = NULL;
2849         init_waitqueue_head(&journal->j_join_wait);
2850         mutex_init(&journal->j_mutex);
2851         mutex_init(&journal->j_flush_mutex);
2852
2853         journal->j_trans_id = 10;
2854         journal->j_mount_id = 10;
2855         journal->j_state = 0;
2856         atomic_set(&journal->j_jlock, 0);
2857         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2858         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2859         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2860         journal->j_cnode_used = 0;
2861         journal->j_must_wait = 0;
2862
2863         if (journal->j_cnode_free == 0) {
2864                 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2865                                  "allocation failed (%ld bytes). Journal is "
2866                                  "too large for available memory. Usually "
2867                                  "this is due to a journal that is too large.",
2868                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2869                 goto free_and_return;
2870         }
2871
2872         init_journal_hash(sb);
2873         jl = journal->j_current_jl;
2874
2875         /*
2876          * get_list_bitmap() may call flush_commit_list() which
2877          * requires the lock. Calling flush_commit_list() shouldn't happen
2878          * this early but I like to be paranoid.
2879          */
2880         reiserfs_write_lock(sb);
2881         jl->j_list_bitmap = get_list_bitmap(sb, jl);
2882         reiserfs_write_unlock(sb);
2883         if (!jl->j_list_bitmap) {
2884                 reiserfs_warning(sb, "journal-2005",
2885                                  "get_list_bitmap failed for journal list 0");
2886                 goto free_and_return;
2887         }
2888
2889         ret = journal_read(sb);
2890         if (ret < 0) {
2891                 reiserfs_warning(sb, "reiserfs-2006",
2892                                  "Replay Failure, unable to mount");
2893                 goto free_and_return;
2894         }
2895
2896         INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2897         journal->j_work_sb = sb;
2898         return 0;
2899 free_and_return:
2900         free_journal_ram(sb);
2901         return 1;
2902 }
2903
2904 /*
2905  * test for a polite end of the current transaction.  Used by file_write,
2906  * and should be used by delete to make sure they don't write more than
2907  * can fit inside a single transaction
2908  */
2909 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2910                                    int new_alloc)
2911 {
2912         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2913         time_t now = get_seconds();
2914         /* cannot restart while nested */
2915         BUG_ON(!th->t_trans_id);
2916         if (th->t_refcount > 1)
2917                 return 0;
2918         if (journal->j_must_wait > 0 ||
2919             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2920             atomic_read(&journal->j_jlock) ||
2921             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2922             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2923                 return 1;
2924         }
2925
2926         journal->j_len_alloc += new_alloc;
2927         th->t_blocks_allocated += new_alloc ;
2928         return 0;
2929 }
2930
2931 /* this must be called inside a transaction */
2932 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2933 {
2934         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2935         BUG_ON(!th->t_trans_id);
2936         journal->j_must_wait = 1;
2937         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2938         return;
2939 }
2940
2941 /* this must be called without a transaction started */
2942 void reiserfs_allow_writes(struct super_block *s)
2943 {
2944         struct reiserfs_journal *journal = SB_JOURNAL(s);
2945         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2946         wake_up(&journal->j_join_wait);
2947 }
2948
2949 /* this must be called without a transaction started */
2950 void reiserfs_wait_on_write_block(struct super_block *s)
2951 {
2952         struct reiserfs_journal *journal = SB_JOURNAL(s);
2953         wait_event(journal->j_join_wait,
2954                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2955 }
2956
2957 static void queue_log_writer(struct super_block *s)
2958 {
2959         wait_queue_t wait;
2960         struct reiserfs_journal *journal = SB_JOURNAL(s);
2961         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2962
2963         /*
2964          * we don't want to use wait_event here because
2965          * we only want to wait once.
2966          */
2967         init_waitqueue_entry(&wait, current);
2968         add_wait_queue(&journal->j_join_wait, &wait);
2969         set_current_state(TASK_UNINTERRUPTIBLE);
2970         if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
2971                 int depth = reiserfs_write_unlock_nested(s);
2972                 schedule();
2973                 reiserfs_write_lock_nested(s, depth);
2974         }
2975         __set_current_state(TASK_RUNNING);
2976         remove_wait_queue(&journal->j_join_wait, &wait);
2977 }
2978
2979 static void wake_queued_writers(struct super_block *s)
2980 {
2981         struct reiserfs_journal *journal = SB_JOURNAL(s);
2982         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2983                 wake_up(&journal->j_join_wait);
2984 }
2985
2986 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
2987 {
2988         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2989         unsigned long bcount = journal->j_bcount;
2990         while (1) {
2991                 int depth;
2992
2993                 depth = reiserfs_write_unlock_nested(sb);
2994                 schedule_timeout_uninterruptible(1);
2995                 reiserfs_write_lock_nested(sb, depth);
2996
2997                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2998                 while ((atomic_read(&journal->j_wcount) > 0 ||
2999                         atomic_read(&journal->j_jlock)) &&
3000                        journal->j_trans_id == trans_id) {
3001                         queue_log_writer(sb);
3002                 }
3003                 if (journal->j_trans_id != trans_id)
3004                         break;
3005                 if (bcount == journal->j_bcount)
3006                         break;
3007                 bcount = journal->j_bcount;
3008         }
3009 }
3010
3011 /*
3012  * join == true if you must join an existing transaction.
3013  * join == false if you can deal with waiting for others to finish
3014  *
3015  * this will block until the transaction is joinable.  send the number of
3016  * blocks you expect to use in nblocks.
3017 */
3018 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3019                               struct super_block *sb, unsigned long nblocks,
3020                               int join)
3021 {
3022         time_t now = get_seconds();
3023         unsigned int old_trans_id;
3024         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3025         struct reiserfs_transaction_handle myth;
3026         int sched_count = 0;
3027         int retval;
3028         int depth;
3029
3030         reiserfs_check_lock_depth(sb, "journal_begin");
3031         BUG_ON(nblocks > journal->j_trans_max);
3032
3033         PROC_INFO_INC(sb, journal.journal_being);
3034         /* set here for journal_join */
3035         th->t_refcount = 1;
3036         th->t_super = sb;
3037
3038 relock:
3039         lock_journal(sb);
3040         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3041                 unlock_journal(sb);
3042                 retval = journal->j_errno;
3043                 goto out_fail;
3044         }
3045         journal->j_bcount++;
3046
3047         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3048                 unlock_journal(sb);
3049                 depth = reiserfs_write_unlock_nested(sb);
3050                 reiserfs_wait_on_write_block(sb);
3051                 reiserfs_write_lock_nested(sb, depth);
3052                 PROC_INFO_INC(sb, journal.journal_relock_writers);
3053                 goto relock;
3054         }
3055         now = get_seconds();
3056
3057         /*
3058          * if there is no room in the journal OR
3059          * if this transaction is too old, and we weren't called joinable,
3060          * wait for it to finish before beginning we don't sleep if there
3061          * aren't other writers
3062          */
3063
3064         if ((!join && journal->j_must_wait > 0) ||
3065             (!join
3066              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3067             || (!join && atomic_read(&journal->j_wcount) > 0
3068                 && journal->j_trans_start_time > 0
3069                 && (now - journal->j_trans_start_time) >
3070                 journal->j_max_trans_age) || (!join
3071                                               && atomic_read(&journal->j_jlock))
3072             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3073
3074                 old_trans_id = journal->j_trans_id;
3075                 /* allow others to finish this transaction */
3076                 unlock_journal(sb);
3077
3078                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3079                     journal->j_max_batch &&
3080                     ((journal->j_len + nblocks + 2) * 100) <
3081                     (journal->j_len_alloc * 75)) {
3082                         if (atomic_read(&journal->j_wcount) > 10) {
3083                                 sched_count++;
3084                                 queue_log_writer(sb);
3085                                 goto relock;
3086                         }
3087                 }
3088                 /*
3089                  * don't mess with joining the transaction if all we
3090                  * have to do is wait for someone else to do a commit
3091                  */
3092                 if (atomic_read(&journal->j_jlock)) {
3093                         while (journal->j_trans_id == old_trans_id &&
3094                                atomic_read(&journal->j_jlock)) {
3095                                 queue_log_writer(sb);
3096                         }
3097                         goto relock;
3098                 }
3099                 retval = journal_join(&myth, sb);
3100                 if (retval)
3101                         goto out_fail;
3102
3103                 /* someone might have ended the transaction while we joined */
3104                 if (old_trans_id != journal->j_trans_id) {
3105                         retval = do_journal_end(&myth, 0);
3106                 } else {
3107                         retval = do_journal_end(&myth, COMMIT_NOW);
3108                 }
3109
3110                 if (retval)
3111                         goto out_fail;
3112
3113                 PROC_INFO_INC(sb, journal.journal_relock_wcount);
3114                 goto relock;
3115         }
3116         /* we are the first writer, set trans_id */
3117         if (journal->j_trans_start_time == 0) {
3118                 journal->j_trans_start_time = get_seconds();
3119         }
3120         atomic_inc(&journal->j_wcount);
3121         journal->j_len_alloc += nblocks;
3122         th->t_blocks_logged = 0;
3123         th->t_blocks_allocated = nblocks;
3124         th->t_trans_id = journal->j_trans_id;
3125         unlock_journal(sb);
3126         INIT_LIST_HEAD(&th->t_list);
3127         return 0;
3128
3129 out_fail:
3130         memset(th, 0, sizeof(*th));
3131         /*
3132          * Re-set th->t_super, so we can properly keep track of how many
3133          * persistent transactions there are. We need to do this so if this
3134          * call is part of a failed restart_transaction, we can free it later
3135          */
3136         th->t_super = sb;
3137         return retval;
3138 }
3139
3140 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3141                                                                     super_block
3142                                                                     *s,
3143                                                                     int nblocks)
3144 {
3145         int ret;
3146         struct reiserfs_transaction_handle *th;
3147
3148         /*
3149          * if we're nesting into an existing transaction.  It will be
3150          * persistent on its own
3151          */
3152         if (reiserfs_transaction_running(s)) {
3153                 th = current->journal_info;
3154                 th->t_refcount++;
3155                 BUG_ON(th->t_refcount < 2);
3156
3157                 return th;
3158         }
3159         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3160         if (!th)
3161                 return NULL;
3162         ret = journal_begin(th, s, nblocks);
3163         if (ret) {
3164                 kfree(th);
3165                 return NULL;
3166         }
3167
3168         SB_JOURNAL(s)->j_persistent_trans++;
3169         return th;
3170 }
3171
3172 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3173 {
3174         struct super_block *s = th->t_super;
3175         int ret = 0;
3176         if (th->t_trans_id)
3177                 ret = journal_end(th);
3178         else
3179                 ret = -EIO;
3180         if (th->t_refcount == 0) {
3181                 SB_JOURNAL(s)->j_persistent_trans--;
3182                 kfree(th);
3183         }
3184         return ret;
3185 }
3186
3187 static int journal_join(struct reiserfs_transaction_handle *th,
3188                         struct super_block *sb)
3189 {
3190         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3191
3192         /*
3193          * this keeps do_journal_end from NULLing out the
3194          * current->journal_info pointer
3195          */
3196         th->t_handle_save = cur_th;
3197         BUG_ON(cur_th && cur_th->t_refcount > 1);
3198         return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN);
3199 }
3200
3201 int journal_join_abort(struct reiserfs_transaction_handle *th,
3202                        struct super_block *sb)
3203 {
3204         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3205
3206         /*
3207          * this keeps do_journal_end from NULLing out the
3208          * current->journal_info pointer
3209          */
3210         th->t_handle_save = cur_th;
3211         BUG_ON(cur_th && cur_th->t_refcount > 1);
3212         return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT);
3213 }
3214
3215 int journal_begin(struct reiserfs_transaction_handle *th,
3216                   struct super_block *sb, unsigned long nblocks)
3217 {
3218         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3219         int ret;
3220
3221         th->t_handle_save = NULL;
3222         if (cur_th) {
3223                 /* we are nesting into the current transaction */
3224                 if (cur_th->t_super == sb) {
3225                         BUG_ON(!cur_th->t_refcount);
3226                         cur_th->t_refcount++;
3227                         memcpy(th, cur_th, sizeof(*th));
3228                         if (th->t_refcount <= 1)
3229                                 reiserfs_warning(sb, "reiserfs-2005",
3230                                                  "BAD: refcount <= 1, but "
3231                                                  "journal_info != 0");
3232                         return 0;
3233                 } else {
3234                         /*
3235                          * we've ended up with a handle from a different
3236                          * filesystem.  save it and restore on journal_end.
3237                          * This should never really happen...
3238                          */
3239                         reiserfs_warning(sb, "clm-2100",
3240                                          "nesting info a different FS");
3241                         th->t_handle_save = current->journal_info;
3242                         current->journal_info = th;
3243                 }
3244         } else {
3245                 current->journal_info = th;
3246         }
3247         ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3248         BUG_ON(current->journal_info != th);
3249
3250         /*
3251          * I guess this boils down to being the reciprocal of clm-2100 above.
3252          * If do_journal_begin_r fails, we need to put it back, since
3253          * journal_end won't be called to do it. */
3254         if (ret)
3255                 current->journal_info = th->t_handle_save;
3256         else
3257                 BUG_ON(!th->t_refcount);
3258
3259         return ret;
3260 }
3261
3262 /*
3263  * puts bh into the current transaction.  If it was already there, reorders
3264  * removes the old pointers from the hash, and puts new ones in (to make
3265  * sure replay happen in the right order).
3266  *
3267  * if it was dirty, cleans and files onto the clean list.  I can't let it
3268  * be dirty again until the transaction is committed.
3269  *
3270  * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3271  */
3272 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3273                        struct buffer_head *bh)
3274 {
3275         struct super_block *sb = th->t_super;
3276         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3277         struct reiserfs_journal_cnode *cn = NULL;
3278         int count_already_incd = 0;
3279         int prepared = 0;
3280         BUG_ON(!th->t_trans_id);
3281
3282         PROC_INFO_INC(sb, journal.mark_dirty);
3283         if (th->t_trans_id != journal->j_trans_id) {
3284                 reiserfs_panic(th->t_super, "journal-1577",
3285                                "handle trans id %ld != current trans id %ld",
3286                                th->t_trans_id, journal->j_trans_id);
3287         }
3288
3289         prepared = test_clear_buffer_journal_prepared(bh);
3290         clear_buffer_journal_restore_dirty(bh);
3291         /* already in this transaction, we are done */
3292         if (buffer_journaled(bh)) {
3293                 PROC_INFO_INC(sb, journal.mark_dirty_already);
3294                 return 0;
3295         }
3296
3297         /*
3298          * this must be turned into a panic instead of a warning.  We can't
3299          * allow a dirty or journal_dirty or locked buffer to be logged, as
3300          * some changes could get to disk too early.  NOT GOOD.
3301          */
3302         if (!prepared || buffer_dirty(bh)) {
3303                 reiserfs_warning(sb, "journal-1777",
3304                                  "buffer %llu bad state "
3305                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3306                                  (unsigned long long)bh->b_blocknr,
3307                                  prepared ? ' ' : '!',
3308                                  buffer_locked(bh) ? ' ' : '!',
3309                                  buffer_dirty(bh) ? ' ' : '!',
3310                                  buffer_journal_dirty(bh) ? ' ' : '!');
3311         }
3312
3313         if (atomic_read(&journal->j_wcount) <= 0) {
3314                 reiserfs_warning(sb, "journal-1409",
3315                                  "returning because j_wcount was %d",
3316                                  atomic_read(&journal->j_wcount));
3317                 return 1;
3318         }
3319         /*
3320          * this error means I've screwed up, and we've overflowed
3321          * the transaction.  Nothing can be done here, except make the
3322          * FS readonly or panic.
3323          */
3324         if (journal->j_len >= journal->j_trans_max) {
3325                 reiserfs_panic(th->t_super, "journal-1413",
3326                                "j_len (%lu) is too big",
3327                                journal->j_len);
3328         }
3329
3330         if (buffer_journal_dirty(bh)) {
3331                 count_already_incd = 1;
3332                 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3333                 clear_buffer_journal_dirty(bh);
3334         }
3335
3336         if (journal->j_len > journal->j_len_alloc) {
3337                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3338         }
3339
3340         set_buffer_journaled(bh);
3341
3342         /* now put this guy on the end */
3343         if (!cn) {
3344                 cn = get_cnode(sb);
3345                 if (!cn) {
3346                         reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3347                 }
3348
3349                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3350                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3351                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3352                 }
3353                 th->t_blocks_logged++;
3354                 journal->j_len++;
3355
3356                 cn->bh = bh;
3357                 cn->blocknr = bh->b_blocknr;
3358                 cn->sb = sb;
3359                 cn->jlist = NULL;
3360                 insert_journal_hash(journal->j_hash_table, cn);
3361                 if (!count_already_incd) {
3362                         get_bh(bh);
3363                 }
3364         }
3365         cn->next = NULL;
3366         cn->prev = journal->j_last;
3367         cn->bh = bh;
3368         if (journal->j_last) {
3369                 journal->j_last->next = cn;
3370                 journal->j_last = cn;
3371         } else {
3372                 journal->j_first = cn;
3373                 journal->j_last = cn;
3374         }
3375         reiserfs_schedule_old_flush(sb);
3376         return 0;
3377 }
3378
3379 int journal_end(struct reiserfs_transaction_handle *th)
3380 {
3381         struct super_block *sb = th->t_super;
3382         if (!current->journal_info && th->t_refcount > 1)
3383                 reiserfs_warning(sb, "REISER-NESTING",
3384                                  "th NULL, refcount %d", th->t_refcount);
3385
3386         if (!th->t_trans_id) {
3387                 WARN_ON(1);
3388                 return -EIO;
3389         }
3390
3391         th->t_refcount--;
3392         if (th->t_refcount > 0) {
3393                 struct reiserfs_transaction_handle *cur_th =
3394                     current->journal_info;
3395
3396                 /*
3397                  * we aren't allowed to close a nested transaction on a
3398                  * different filesystem from the one in the task struct
3399                  */
3400                 BUG_ON(cur_th->t_super != th->t_super);
3401
3402                 if (th != cur_th) {
3403                         memcpy(current->journal_info, th, sizeof(*th));
3404                         th->t_trans_id = 0;
3405                 }
3406                 return 0;
3407         } else {
3408                 return do_journal_end(th, 0);
3409         }
3410 }
3411
3412 /*
3413  * removes from the current transaction, relsing and descrementing any counters.
3414  * also files the removed buffer directly onto the clean list
3415  *
3416  * called by journal_mark_freed when a block has been deleted
3417  *
3418  * returns 1 if it cleaned and relsed the buffer. 0 otherwise
3419  */
3420 static int remove_from_transaction(struct super_block *sb,
3421                                    b_blocknr_t blocknr, int already_cleaned)
3422 {
3423         struct buffer_head *bh;
3424         struct reiserfs_journal_cnode *cn;
3425         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3426         int ret = 0;
3427
3428         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3429         if (!cn || !cn->bh) {
3430                 return ret;
3431         }
3432         bh = cn->bh;
3433         if (cn->prev) {
3434                 cn->prev->next = cn->next;
3435         }
3436         if (cn->next) {
3437                 cn->next->prev = cn->prev;
3438         }
3439         if (cn == journal->j_first) {
3440                 journal->j_first = cn->next;
3441         }
3442         if (cn == journal->j_last) {
3443                 journal->j_last = cn->prev;
3444         }
3445         if (bh)
3446                 remove_journal_hash(sb, journal->j_hash_table, NULL,
3447                                     bh->b_blocknr, 0);
3448         clear_buffer_journaled(bh);     /* don't log this one */
3449
3450         if (!already_cleaned) {
3451                 clear_buffer_journal_dirty(bh);
3452                 clear_buffer_dirty(bh);
3453                 clear_buffer_journal_test(bh);
3454                 put_bh(bh);
3455                 if (atomic_read(&bh->b_count) < 0) {
3456                         reiserfs_warning(sb, "journal-1752",
3457                                          "b_count < 0");
3458                 }
3459                 ret = 1;
3460         }
3461         journal->j_len--;
3462         journal->j_len_alloc--;
3463         free_cnode(sb, cn);
3464         return ret;
3465 }
3466
3467 /*
3468  * for any cnode in a journal list, it can only be dirtied of all the
3469  * transactions that include it are committed to disk.
3470  * this checks through each transaction, and returns 1 if you are allowed
3471  * to dirty, and 0 if you aren't
3472  *
3473  * it is called by dirty_journal_list, which is called after
3474  * flush_commit_list has gotten all the log blocks for a given
3475  * transaction on disk
3476  *
3477  */
3478 static int can_dirty(struct reiserfs_journal_cnode *cn)
3479 {
3480         struct super_block *sb = cn->sb;
3481         b_blocknr_t blocknr = cn->blocknr;
3482         struct reiserfs_journal_cnode *cur = cn->hprev;
3483         int can_dirty = 1;
3484
3485         /*
3486          * first test hprev.  These are all newer than cn, so any node here
3487          * with the same block number and dev means this node can't be sent
3488          * to disk right now.
3489          */
3490         while (cur && can_dirty) {
3491                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3492                     cur->blocknr == blocknr) {
3493                         can_dirty = 0;
3494                 }
3495                 cur = cur->hprev;
3496         }
3497         /*
3498          * then test hnext.  These are all older than cn.  As long as they
3499          * are committed to the log, it is safe to write cn to disk
3500          */
3501         cur = cn->hnext;
3502         while (cur && can_dirty) {
3503                 if (cur->jlist && cur->jlist->j_len > 0 &&
3504                     atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh &&
3505                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3506                         can_dirty = 0;
3507                 }
3508                 cur = cur->hnext;
3509         }
3510         return can_dirty;
3511 }
3512
3513 /*
3514  * syncs the commit blocks, but does not force the real buffers to disk
3515  * will wait until the current transaction is done/committed before returning