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