~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/reiserfs/journal.c

Version: ~ [ linux-5.8-rc4 ] ~ [ linux-5.7.7 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.50 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.131 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.187 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.229 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.229 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp