1 /*
2 ** Write ahead logging implementation copyright Chris Mason 2000
3 **
4 ** The background commits make this code very interelated, and
5 ** overly complex. I need to rethink things a bit....The major players:
6 **
7 ** journal_begin -- call with the number of blocks you expect to log.
8 ** If the current transaction is too
9 ** old, it will block until the current transaction is
10 ** finished, and then start a new one.
11 ** Usually, your transaction will get joined in with
12 ** previous ones for speed.
13 **
14 ** journal_join -- same as journal_begin, but won't block on the current
15 ** transaction regardless of age. Don't ever call
16 ** this. Ever. There are only two places it should be
17 ** called from, and they are both inside this file.
18 **
19 ** journal_mark_dirty -- adds blocks into this transaction. clears any flags
20 ** that might make them get sent to disk
21 ** and then marks them BH_JDirty. Puts the buffer head
22 ** into the current transaction hash.
23 **
24 ** journal_end -- if the current transaction is batchable, it does nothing
25 ** otherwise, it could do an async/synchronous commit, or
26 ** a full flush of all log and real blocks in the
27 ** transaction.
28 **
29 ** flush_old_commits -- if the current transaction is too old, it is ended and
30 ** commit blocks are sent to disk. Forces commit blocks
31 ** to disk for all backgrounded commits that have been
32 ** around too long.
33 ** -- Note, if you call this as an immediate flush from
34 ** from within kupdate, it will ignore the immediate flag
35 **
36 ** The commit thread -- a writer process for async commits. It allows a
37 ** a process to request a log flush on a task queue.
38 ** the commit will happen once the commit thread wakes up.
39 ** The benefit here is the writer (with whatever
40 ** related locks it has) doesn't have to wait for the
41 ** log blocks to hit disk if it doesn't want to.
42 */
43
44 #include <linux/config.h>
45 #include <asm/uaccess.h>
46 #include <asm/system.h>
47
48 #include <linux/sched.h>
49 #include <asm/semaphore.h>
50
51 #include <linux/vmalloc.h>
52 #include <linux/reiserfs_fs.h>
53
54 #include <linux/kernel.h>
55 #include <linux/errno.h>
56 #include <linux/fcntl.h>
57 #include <linux/locks.h>
58 #include <linux/stat.h>
59 #include <linux/string.h>
60 #include <linux/smp_lock.h>
61
62 /* the number of mounted filesystems. This is used to decide when to
63 ** start and kill the commit thread
64 */
65 static int reiserfs_mounted_fs_count = 0 ;
66
67 /* wake this up when you add something to the commit thread task queue */
68 DECLARE_WAIT_QUEUE_HEAD(reiserfs_commit_thread_wait) ;
69
70 /* wait on this if you need to be sure you task queue entries have been run */
71 static DECLARE_WAIT_QUEUE_HEAD(reiserfs_commit_thread_done) ;
72 DECLARE_TASK_QUEUE(reiserfs_commit_thread_tq) ;
73
74 #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
75 structs at 4k */
76 #define BUFNR 64 /*read ahead */
77
78 /* cnode stat bits. Move these into reiserfs_fs.h */
79
80 #define BLOCK_FREED 2 /* this block was freed, and can't be written. */
81 #define BLOCK_FREED_HOLDER 3 /* this block was freed during this transaction, and can't be written */
82
83 #define BLOCK_NEEDS_FLUSH 4 /* used in flush_journal_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 /* state bits for the journal */
91 #define WRITERS_BLOCKED 1 /* set when new writers not allowed */
92
93 static int do_journal_end(struct reiserfs_transaction_handle *,struct super_block *,unsigned long nblocks,int flags) ;
94 static int flush_journal_list(struct super_block *s, struct reiserfs_journal_list *jl, int flushall) ;
95 static int flush_commit_list(struct super_block *s, struct reiserfs_journal_list *jl, int flushall) ;
96 static int can_dirty(struct reiserfs_journal_cnode *cn) ;
97 static int remove_from_journal_list(struct super_block *s, struct reiserfs_journal_list *jl, struct buffer_head *bh, int remove_freed);
98 static int journal_join(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks);
99 static int release_journal_dev( struct super_block *super,
100 struct reiserfs_journal *journal );
101 static void init_journal_hash(struct super_block *p_s_sb) {
102 memset(SB_JOURNAL(p_s_sb)->j_hash_table, 0, JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)) ;
103 }
104
105 /*
106 ** clears BH_Dirty and sticks the buffer on the clean list. Called because I can't allow refile_buffer to
107 ** make schedule happen after I've freed a block. Look at remove_from_transaction and journal_mark_freed for
108 ** more details.
109 */
110 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh) {
111 if (bh) {
112 clear_bit(BH_Dirty, &bh->b_state) ;
113 refile_buffer(bh) ;
114 }
115 return 0 ;
116 }
117
118 static struct reiserfs_bitmap_node *
119 allocate_bitmap_node(struct super_block *p_s_sb) {
120 struct reiserfs_bitmap_node *bn ;
121 static int id = 0 ;
122
123 bn = reiserfs_kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS, p_s_sb) ;
124 if (!bn) {
125 return NULL ;
126 }
127 bn->data = reiserfs_kmalloc(p_s_sb->s_blocksize, GFP_NOFS, p_s_sb) ;
128 if (!bn->data) {
129 reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb) ;
130 return NULL ;
131 }
132 bn->id = id++ ;
133 memset(bn->data, 0, p_s_sb->s_blocksize) ;
134 INIT_LIST_HEAD(&bn->list) ;
135 return bn ;
136 }
137
138 static struct reiserfs_bitmap_node *
139 get_bitmap_node(struct super_block *p_s_sb) {
140 struct reiserfs_bitmap_node *bn = NULL;
141 struct list_head *entry = SB_JOURNAL(p_s_sb)->j_bitmap_nodes.next ;
142
143 SB_JOURNAL(p_s_sb)->j_used_bitmap_nodes++ ;
144 repeat:
145
146 if(entry != &SB_JOURNAL(p_s_sb)->j_bitmap_nodes) {
147 bn = list_entry(entry, struct reiserfs_bitmap_node, list) ;
148 list_del(entry) ;
149 memset(bn->data, 0, p_s_sb->s_blocksize) ;
150 SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes-- ;
151 return bn ;
152 }
153 bn = allocate_bitmap_node(p_s_sb) ;
154 if (!bn) {
155 yield();
156 goto repeat ;
157 }
158 return bn ;
159 }
160 static inline void free_bitmap_node(struct super_block *p_s_sb,
161 struct reiserfs_bitmap_node *bn) {
162 SB_JOURNAL(p_s_sb)->j_used_bitmap_nodes-- ;
163 if (SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
164 reiserfs_kfree(bn->data, p_s_sb->s_blocksize, p_s_sb) ;
165 reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb) ;
166 } else {
167 list_add(&bn->list, &SB_JOURNAL(p_s_sb)->j_bitmap_nodes) ;
168 SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes++ ;
169 }
170 }
171
172 static void allocate_bitmap_nodes(struct super_block *p_s_sb) {
173 int i ;
174 struct reiserfs_bitmap_node *bn = NULL ;
175 for (i = 0 ; i < REISERFS_MIN_BITMAP_NODES ; i++) {
176 bn = allocate_bitmap_node(p_s_sb) ;
177 if (bn) {
178 list_add(&bn->list, &SB_JOURNAL(p_s_sb)->j_bitmap_nodes) ;
179 SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes++ ;
180 } else {
181 break ; // this is ok, we'll try again when more are needed
182 }
183 }
184 }
185
186 static int set_bit_in_list_bitmap(struct super_block *p_s_sb, int block,
187 struct reiserfs_list_bitmap *jb) {
188 int bmap_nr = block / (p_s_sb->s_blocksize << 3) ;
189 int bit_nr = block % (p_s_sb->s_blocksize << 3) ;
190
191 if (!jb->bitmaps[bmap_nr]) {
192 jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb) ;
193 }
194 set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data) ;
195 return 0 ;
196 }
197
198 static void cleanup_bitmap_list(struct super_block *p_s_sb,
199 struct reiserfs_list_bitmap *jb) {
200 int i;
201 for (i = 0 ; i < SB_BMAP_NR(p_s_sb) ; i++) {
202 if (jb->bitmaps[i]) {
203 free_bitmap_node(p_s_sb, jb->bitmaps[i]) ;
204 jb->bitmaps[i] = NULL ;
205 }
206 }
207 }
208
209 /*
210 ** only call this on FS unmount.
211 */
212 static int free_list_bitmaps(struct super_block *p_s_sb,
213 struct reiserfs_list_bitmap *jb_array) {
214 int i ;
215 struct reiserfs_list_bitmap *jb ;
216 for (i = 0 ; i < JOURNAL_NUM_BITMAPS ; i++) {
217 jb = jb_array + i ;
218 jb->journal_list = NULL ;
219 cleanup_bitmap_list(p_s_sb, jb) ;
220 vfree(jb->bitmaps) ;
221 jb->bitmaps = NULL ;
222 }
223 return 0;
224 }
225
226 static int free_bitmap_nodes(struct super_block *p_s_sb) {
227 struct list_head *next = SB_JOURNAL(p_s_sb)->j_bitmap_nodes.next ;
228 struct reiserfs_bitmap_node *bn ;
229
230 while(next != &SB_JOURNAL(p_s_sb)->j_bitmap_nodes) {
231 bn = list_entry(next, struct reiserfs_bitmap_node, list) ;
232 list_del(next) ;
233 reiserfs_kfree(bn->data, p_s_sb->s_blocksize, p_s_sb) ;
234 reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb) ;
235 next = SB_JOURNAL(p_s_sb)->j_bitmap_nodes.next ;
236 SB_JOURNAL(p_s_sb)->j_free_bitmap_nodes-- ;
237 }
238
239 return 0 ;
240 }
241
242 /*
243 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
244 ** jb_array is the array to be filled in.
245 */
246 int reiserfs_allocate_list_bitmaps(struct super_block *p_s_sb,
247 struct reiserfs_list_bitmap *jb_array,
248 int bmap_nr) {
249 int i ;
250 int failed = 0 ;
251 struct reiserfs_list_bitmap *jb ;
252 int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *) ;
253
254 for (i = 0 ; i < JOURNAL_NUM_BITMAPS ; i++) {
255 jb = jb_array + i ;
256 jb->journal_list = NULL ;
257 jb->bitmaps = vmalloc( mem ) ;
258 if (!jb->bitmaps) {
259 reiserfs_warning(p_s_sb, "clm-2000, unable to allocate bitmaps for journal lists\n") ;
260 failed = 1;
261 break ;
262 }
263 memset(jb->bitmaps, 0, mem) ;
264 }
265 if (failed) {
266 free_list_bitmaps(p_s_sb, jb_array) ;
267 return -1 ;
268 }
269 return 0 ;
270 }
271
272 /*
273 ** find an available list bitmap. If you can't find one, flush a commit list
274 ** and try again
275 */
276 static struct reiserfs_list_bitmap *
277 get_list_bitmap(struct super_block *p_s_sb, struct reiserfs_journal_list *jl) {
278 int i,j ;
279 struct reiserfs_list_bitmap *jb = NULL ;
280
281 for (j = 0 ; j < (JOURNAL_NUM_BITMAPS * 3) ; j++) {
282 i = SB_JOURNAL(p_s_sb)->j_list_bitmap_index ;
283 SB_JOURNAL(p_s_sb)->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS ;
284 jb = SB_JOURNAL(p_s_sb)->j_list_bitmap + i ;
285 if (SB_JOURNAL(p_s_sb)->j_list_bitmap[i].journal_list) {
286 flush_commit_list(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_bitmap[i].journal_list, 1) ;
287 if (!SB_JOURNAL(p_s_sb)->j_list_bitmap[i].journal_list) {
288 break ;
289 }
290 } else {
291 break ;
292 }
293 }
294 if (jb->journal_list) { /* double check to make sure if flushed correctly */
295 return NULL ;
296 }
297 jb->journal_list = jl ;
298 return jb ;
299 }
300
301 /*
302 ** allocates a new chunk of X nodes, and links them all together as a list.
303 ** Uses the cnode->next and cnode->prev pointers
304 ** returns NULL on failure
305 */
306 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes) {
307 struct reiserfs_journal_cnode *head ;
308 int i ;
309 if (num_cnodes <= 0) {
310 return NULL ;
311 }
312 head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode)) ;
313 if (!head) {
314 return NULL ;
315 }
316 memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode)) ;
317 head[0].prev = NULL ;
318 head[0].next = head + 1 ;
319 for (i = 1 ; i < num_cnodes; i++) {
320 head[i].prev = head + (i - 1) ;
321 head[i].next = head + (i + 1) ; /* if last one, overwrite it after the if */
322 }
323 head[num_cnodes -1].next = NULL ;
324 return head ;
325 }
326
327 /*
328 ** pulls a cnode off the free list, or returns NULL on failure
329 */
330 static struct reiserfs_journal_cnode *get_cnode(struct super_block *p_s_sb) {
331 struct reiserfs_journal_cnode *cn ;
332
333 reiserfs_check_lock_depth("get_cnode") ;
334
335 if (SB_JOURNAL(p_s_sb)->j_cnode_free <= 0) {
336 return NULL ;
337 }
338 SB_JOURNAL(p_s_sb)->j_cnode_used++ ;
339 SB_JOURNAL(p_s_sb)->j_cnode_free-- ;
340 cn = SB_JOURNAL(p_s_sb)->j_cnode_free_list ;
341 if (!cn) {
342 return cn ;
343 }
344 if (cn->next) {
345 cn->next->prev = NULL ;
346 }
347 SB_JOURNAL(p_s_sb)->j_cnode_free_list = cn->next ;
348 memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ;
349 return cn ;
350 }
351
352 /*
353 ** returns a cnode to the free list
354 */
355 static void free_cnode(struct super_block *p_s_sb, struct reiserfs_journal_cnode *cn) {
356
357 reiserfs_check_lock_depth("free_cnode") ;
358
359 SB_JOURNAL(p_s_sb)->j_cnode_used-- ;
360 SB_JOURNAL(p_s_sb)->j_cnode_free++ ;
361 /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
362 cn->next = SB_JOURNAL(p_s_sb)->j_cnode_free_list ;
363 if (SB_JOURNAL(p_s_sb)->j_cnode_free_list) {
364 SB_JOURNAL(p_s_sb)->j_cnode_free_list->prev = cn ;
365 }
366 cn->prev = NULL ; /* not needed with the memset, but I might kill the memset, and forget to do this */
367 SB_JOURNAL(p_s_sb)->j_cnode_free_list = cn ;
368 }
369
370 static int clear_prepared_bits(struct buffer_head *bh) {
371 clear_bit(BH_JPrepared, &bh->b_state) ;
372 return 0 ;
373 }
374
375 /* buffer is in current transaction */
376 inline int buffer_journaled(const struct buffer_head *bh) {
377 if (bh)
378 return test_bit(BH_JDirty, &((struct buffer_head *)bh)->b_state) ;
379 else
380 return 0 ;
381 }
382
383 /* disk block was taken off free list before being in a finished transation, or written to disk
384 ** journal_new blocks can be reused immediately, for any purpose
385 */
386 inline int buffer_journal_new(const struct buffer_head *bh) {
387 if (bh)
388 return test_bit(BH_JNew, &((struct buffer_head *)bh)->b_state) ;
389 else
390 return 0 ;
391 }
392
393 inline int mark_buffer_journal_new(struct buffer_head *bh) {
394 if (bh) {
395 set_bit(BH_JNew, &bh->b_state) ;
396 }
397 return 0 ;
398 }
399
400 inline int mark_buffer_not_journaled(struct buffer_head *bh) {
401 if (bh)
402 clear_bit(BH_JDirty, &bh->b_state) ;
403 return 0 ;
404 }
405
406 /* utility function to force a BUG if it is called without the big
407 ** kernel lock held. caller is the string printed just before calling BUG()
408 */
409 void reiserfs_check_lock_depth(char *caller) {
410 #ifdef CONFIG_SMP
411 if (current->lock_depth < 0) {
412 printk("%s called without kernel lock held\n", caller) ;
413 show_reiserfs_locks() ;
414 BUG() ;
415 }
416 #else
417 ;
418 #endif
419 }
420
421 /* return a cnode with same dev, block number and size in table, or null if not found */
422 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct reiserfs_journal_cnode **table,
423 kdev_t dev,long bl,int size) {
424 struct reiserfs_journal_cnode *cn ;
425 cn = journal_hash(table, dev, bl) ;
426 while(cn) {
427 if ((cn->blocknr == bl) && (cn->dev == dev))
428 return cn ;
429 cn = cn->hnext ;
430 }
431 return (struct reiserfs_journal_cnode *)0 ;
432 }
433
434 /* returns a cnode with same size, block number and dev as bh in the current transaction hash. NULL if not found */
435 static inline struct reiserfs_journal_cnode *get_journal_hash(struct super_block *p_s_sb, struct buffer_head *bh) {
436 struct reiserfs_journal_cnode *cn ;
437 if (bh) {
438 cn = get_journal_hash_dev(SB_JOURNAL(p_s_sb)->j_hash_table, bh->b_dev, bh->b_blocknr, bh->b_size) ;
439 }
440 else {
441 return (struct reiserfs_journal_cnode *)0 ;
442 }
443 return cn ;
444 }
445
446 /* once upon a time, the journal would deadlock. a lot. Now, when
447 ** CONFIG_REISERFS_CHECK is defined, anytime someone enters a
448 ** transaction, it pushes itself into this ugly static list, and pops
449 ** itself off before calling journal_end. I made a SysRq key to dump
450 ** the list, and tell me what the writers are when I'm deadlocked. */
451
452 /* are you depending on the compiler
453 to optimize this function away
454 everywhere it is called? It is not
455 obvious how this works, but I
456 suppose debugging code need not be
457 clear. -Hans */
458 static char *journal_writers[512] ;
459 int push_journal_writer(char *s) {
460 #ifdef CONFIG_REISERFS_CHECK
461 int i ;
462 for (i = 0 ; i < 512 ; i++) {
463 if (!journal_writers[i]) {
464 journal_writers[i] = s ;
465 return i ;
466 }
467 }
468 return -1 ;
469 #else
470 return 0 ;
471 #endif
472 }
473 int pop_journal_writer(int index) {
474 #ifdef CONFIG_REISERFS_CHECK
475 if (index >= 0) {
476 journal_writers[index] = NULL ;
477 }
478 #endif
479 return 0 ;
480 }
481
482 int dump_journal_writers(void) {
483 int i ;
484 for (i = 0 ; i < 512 ; i++) {
485 if (journal_writers[i]) {
486 printk("%d: %s\n", i, journal_writers[i]) ;
487 }
488 }
489 return 0 ;
490 }
491
492 /*
493 ** this actually means 'can this block be reallocated yet?'. If you set search_all, a block can only be allocated
494 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
495 ** being overwritten by a replay after crashing.
496 **
497 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction. Since deleting
498 ** a block removes it from the current transaction, this case should never happen. If you don't set search_all, make
499 ** sure you never write the block without logging it.
500 **
501 ** next_zero_bit is a suggestion about the next block to try for find_forward.
502 ** when bl is rejected because it is set in a journal list bitmap, we search
503 ** for the next zero bit in the bitmap that rejected bl. Then, we return that
504 ** through next_zero_bit for find_forward to try.
505 **
506 ** Just because we return something in next_zero_bit does not mean we won't
507 ** reject it on the next call to reiserfs_in_journal
508 **
509 */
510 int reiserfs_in_journal(struct super_block *p_s_sb, kdev_t dev,
511 int bmap_nr, int bit_nr, int size, int search_all,
512 unsigned int *next_zero_bit) {
513 struct reiserfs_journal_cnode *cn ;
514 struct reiserfs_list_bitmap *jb ;
515 int i ;
516 unsigned long bl;
517
518 *next_zero_bit = 0 ; /* always start this at zero. */
519
520 /* we aren't logging all blocks are safe for reuse */
521 if (reiserfs_dont_log(p_s_sb)) {
522 return 0 ;
523 }
524
525 PROC_INFO_INC( p_s_sb, journal.in_journal );
526 /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
527 ** if we crash before the transaction that freed it commits, this transaction won't
528 ** have committed either, and the block will never be written
529 */
530 if (search_all) {
531 for (i = 0 ; i < JOURNAL_NUM_BITMAPS ; i++) {
532 PROC_INFO_INC( p_s_sb, journal.in_journal_bitmap );
533 jb = SB_JOURNAL(p_s_sb)->j_list_bitmap + i ;
534 if (jb->journal_list && jb->bitmaps[bmap_nr] &&
535 test_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data)) {
536 *next_zero_bit = find_next_zero_bit((unsigned long *)
537 (jb->bitmaps[bmap_nr]->data),
538 p_s_sb->s_blocksize << 3, bit_nr+1) ;
539 return 1 ;
540 }
541 }
542 }
543
544 bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr;
545 /* is it in any old transactions? */
546 if (search_all && (cn = get_journal_hash_dev(SB_JOURNAL(p_s_sb)->j_list_hash_table, dev,bl,size))) {
547 return 1;
548 }
549
550 /* is it in the current transaction. This should never happen */
551 if ((cn = get_journal_hash_dev(SB_JOURNAL(p_s_sb)->j_hash_table, dev,bl,size))) {
552 return 1;
553 }
554
555 PROC_INFO_INC( p_s_sb, journal.in_journal_reusable );
556 /* safe for reuse */
557 return 0 ;
558 }
559
560 /* insert cn into table
561 */
562 inline void insert_journal_hash(struct reiserfs_journal_cnode **table, struct reiserfs_journal_cnode *cn) {
563 struct reiserfs_journal_cnode *cn_orig ;
564
565 cn_orig = journal_hash(table, cn->dev, cn->blocknr) ;
566 cn->hnext = cn_orig ;
567 cn->hprev = NULL ;
568 if (cn_orig) {
569 cn_orig->hprev = cn ;
570 }
571 journal_hash(table, cn->dev, cn->blocknr) = cn ;
572 }
573
574 /* lock the current transaction */
575 inline static void lock_journal(struct super_block *p_s_sb) {
576 PROC_INFO_INC( p_s_sb, journal.lock_journal );
577 while(atomic_read(&(SB_JOURNAL(p_s_sb)->j_wlock)) > 0) {
578 PROC_INFO_INC( p_s_sb, journal.lock_journal_wait );
579 sleep_on(&(SB_JOURNAL(p_s_sb)->j_wait)) ;
580 }
581 atomic_set(&(SB_JOURNAL(p_s_sb)->j_wlock), 1) ;
582 }
583
584 /* unlock the current transaction */
585 inline static void unlock_journal(struct super_block *p_s_sb) {
586 atomic_dec(&(SB_JOURNAL(p_s_sb)->j_wlock)) ;
587 wake_up(&(SB_JOURNAL(p_s_sb)->j_wait)) ;
588 }
589
590 /*
591 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
592 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
593 ** transaction.
594 */
595 static void cleanup_freed_for_journal_list(struct super_block *p_s_sb, struct reiserfs_journal_list *jl) {
596
597 struct reiserfs_list_bitmap *jb = jl->j_list_bitmap ;
598 if (jb) {
599 cleanup_bitmap_list(p_s_sb, jb) ;
600 }
601 jl->j_list_bitmap->journal_list = NULL ;
602 jl->j_list_bitmap = NULL ;
603 }
604
605 /*
606 ** if this journal list still has commit blocks unflushed, send them to disk.
607 **
608 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
609 ** Before the commit block can by written, every other log block must be safely on disk
610 **
611 */
612 static int flush_commit_list(struct super_block *s, struct reiserfs_journal_list *jl, int flushall) {
613 int i, count ;
614 int index = 0 ;
615 int bn ;
616 int retry_count = 0 ;
617 int orig_commit_left = 0 ;
618 struct buffer_head *tbh = NULL ;
619 struct reiserfs_journal_list *other_jl ;
620
621 reiserfs_check_lock_depth("flush_commit_list") ;
622
623 if (atomic_read(&jl->j_older_commits_done)) {
624 return 0 ;
625 }
626
627 /* before we can put our commit blocks on disk, we have to make sure everyone older than
628 ** us is on disk too
629 */
630 if (jl->j_len <= 0) {
631 return 0 ;
632 }
633 if (flushall) {
634 /* we _must_ make sure the transactions are committed in order. Start with the
635 ** index after this one, wrap all the way around
636 */
637 index = (jl - SB_JOURNAL_LIST(s)) + 1 ;
638 for (i = 0 ; i < JOURNAL_LIST_COUNT ; i++) {
639 other_jl = SB_JOURNAL_LIST(s) + ( (index + i) % JOURNAL_LIST_COUNT) ;
640 if (other_jl && other_jl != jl && other_jl->j_len > 0 && other_jl->j_trans_id > 0 &&
641 other_jl->j_trans_id <= jl->j_trans_id && (atomic_read(&(jl->j_older_commits_done)) == 0)) {
642 flush_commit_list(s, other_jl, 0) ;
643 }
644 }
645 }
646
647 count = 0 ;
648 /* don't flush the commit list for the current transactoin */
649 if (jl == ((SB_JOURNAL_LIST(s) + SB_JOURNAL_LIST_INDEX(s)))) {
650 return 0 ;
651 }
652
653 /* make sure nobody is trying to flush this one at the same time */
654 if (atomic_read(&(jl->j_commit_flushing))) {
655 sleep_on(&(jl->j_commit_wait)) ;
656 if (flushall) {
657 atomic_set(&(jl->j_older_commits_done), 1) ;
658 }
659 return 0 ;
660 }
661
662 /* this commit is done, exit */
663 if (atomic_read(&(jl->j_commit_left)) <= 0) {
664 if (flushall) {
665 atomic_set(&(jl->j_older_commits_done), 1) ;
666 }
667 return 0 ;
668 }
669 /* keeps others from flushing while we are flushing */
670 atomic_set(&(jl->j_commit_flushing), 1) ;
671
672
673 if (jl->j_len > SB_JOURNAL_TRANS_MAX(s)) {
674 reiserfs_panic(s, "journal-512: flush_commit_list: length is %lu, list number %d\n", jl->j_len, jl - SB_JOURNAL_LIST(s)) ;
675 return 0 ;
676 }
677
678 orig_commit_left = atomic_read(&(jl->j_commit_left)) ;
679
680 /* start by checking all the commit blocks in this transaction.
681 ** Add anyone not on disk into tbh. Stop checking once commit_left <= 1, because that means we
682 ** only have the commit block left
683 */
684 retry:
685 count = 0 ;
686 for (i = 0 ; atomic_read(&(jl->j_commit_left)) > 1 && i < (jl->j_len + 1) ; i++) { /* everything but commit_bh */
687 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start+i) % SB_ONDISK_JOURNAL_SIZE(s);
688 tbh = journal_get_hash_table(s, bn) ;
689
690 /* kill this sanity check */
691 if (count > (orig_commit_left + 2)) {
692 reiserfs_panic(s, "journal-539: flush_commit_list: BAD count(%d) > orig_commit_left(%d)!\n", count, orig_commit_left) ;
693 }
694 if (tbh) {
695 if (buffer_locked(tbh)) { /* wait on it, redo it just to make sure */
696 wait_on_buffer(tbh) ;
697 if (!buffer_uptodate(tbh)) {
698 reiserfs_panic(s, "journal-584, buffer write failed\n") ;
699 }
700 }
701 if (buffer_dirty(tbh)) {
702 reiserfs_warning(s, "journal-569: flush_commit_list, block already dirty!\n") ;
703 } else {
704 mark_buffer_dirty(tbh) ;
705 }
706 ll_rw_block(WRITE, 1, &tbh) ;
707 count++ ;
708 put_bh(tbh) ; /* once for our get_hash */
709 }
710 }
711
712 /* wait on everyone in tbh before writing commit block*/
713 if (count > 0) {
714 for (i = 0 ; atomic_read(&(jl->j_commit_left)) > 1 &&
715 i < (jl->j_len + 1) ; i++) { /* everything but commit_bh */
716 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s) ;
717 tbh = journal_get_hash_table(s, bn) ;
718
719 wait_on_buffer(tbh) ;
720 if (!buffer_uptodate(tbh)) {
721 reiserfs_panic(s, "journal-601, buffer write failed\n") ;
722 }
723 put_bh(tbh) ; /* once for our get_hash */
724 bforget(tbh) ; /* once due to original getblk in do_journal_end */
725 atomic_dec(&(jl->j_commit_left)) ;
726 }
727 }
728
729 if (atomic_read(&(jl->j_commit_left)) != 1) { /* just the commit_bh left, flush it without calling getblk for everyone */
730 if (retry_count < 2) {
731 reiserfs_warning(s, "journal-582: flush_commit_list, not all log blocks on disk yet, trying again\n") ;
732 retry_count++ ;
733 goto retry;
734 }
735 reiserfs_panic(s, "journal-563: flush_commit_list: BAD, j_commit_left is %u, should be 1\n",
736 atomic_read(&(jl->j_commit_left)));
737 }
738
739 mark_buffer_dirty(jl->j_commit_bh) ;
740 ll_rw_block(WRITE, 1, &(jl->j_commit_bh)) ;
741 wait_on_buffer(jl->j_commit_bh) ;
742 if (!buffer_uptodate(jl->j_commit_bh)) {
743 reiserfs_panic(s, "journal-615: buffer write failed\n") ;
744 }
745 atomic_dec(&(jl->j_commit_left)) ;
746 bforget(jl->j_commit_bh) ;
747
748 /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */
749 cleanup_freed_for_journal_list(s, jl) ;
750
751 if (flushall) {
752 atomic_set(&(jl->j_older_commits_done), 1) ;
753 }
754 atomic_set(&(jl->j_commit_flushing), 0) ;
755 wake_up(&(jl->j_commit_wait)) ;
756
757 s->s_dirt = 1 ;
758 return 0 ;
759 }
760
761 /*
762 ** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or
763 ** returns NULL if it can't find anything
764 */
765 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct reiserfs_journal_cnode *cn) {
766 kdev_t dev = cn->dev;
767 unsigned long blocknr = cn->blocknr ;
768
769 cn = cn->hprev ;
770 while(cn) {
771 if (cn->dev == dev && cn->blocknr == blocknr && cn->jlist) {
772 return cn->jlist ;
773 }
774 cn = cn->hprev ;
775 }
776 return NULL ;
777 }
778
779
780 /*
781 ** once all the real blocks have been flushed, it is safe to remove them from the
782 ** journal list for this transaction. Aside from freeing the cnode, this also allows the
783 ** block to be reallocated for data blocks if it had been deleted.
784 */
785 static void remove_all_from_journal_list(struct super_block *p_s_sb, struct reiserfs_journal_list *jl, int debug) {
786 struct buffer_head fake_bh ;
787 struct reiserfs_journal_cnode *cn, *last ;
788 cn = jl->j_realblock ;
789
790 /* which is better, to lock once around the whole loop, or
791 ** to lock for each call to remove_from_journal_list?
792 */
793 while(cn) {
794 if (cn->blocknr != 0) {
795 if (debug) {
796 reiserfs_warning(p_s_sb, "block %lu, bh is %d, state %ld\n", cn->blocknr, cn->bh ? 1: 0,
797 cn->state) ;
798 }
799 fake_bh.b_blocknr = cn->blocknr ;
800 fake_bh.b_dev = cn->dev ;
801 cn->state = 0 ;
802 remove_from_journal_list(p_s_sb, jl, &fake_bh, 1) ;
803 }
804 last = cn ;
805 cn = cn->next ;
806 free_cnode(p_s_sb, last) ;
807 }
808 jl->j_realblock = NULL ;
809 }
810
811 /*
812 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
813 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
814 ** releasing blocks in this transaction for reuse as data blocks.
815 ** called by flush_journal_list, before it calls remove_all_from_journal_list
816 **
817 */
818 static int _update_journal_header_block(struct super_block *p_s_sb, unsigned long offset, unsigned long trans_id) {
819 struct reiserfs_journal_header *jh ;
820 if (trans_id >= SB_JOURNAL(p_s_sb)->j_last_flush_trans_id) {
821 if (buffer_locked((SB_JOURNAL(p_s_sb)->j_header_bh))) {
822 wait_on_buffer((SB_JOURNAL(p_s_sb)->j_header_bh)) ;
823 if (!buffer_uptodate(SB_JOURNAL(p_s_sb)->j_header_bh)) {
824 reiserfs_panic(p_s_sb, "journal-699: buffer write failed\n") ;
825 }
826 }
827 SB_JOURNAL(p_s_sb)->j_last_flush_trans_id = trans_id ;
828 SB_JOURNAL(p_s_sb)->j_first_unflushed_offset = offset ;
829 jh = (struct reiserfs_journal_header *)(SB_JOURNAL(p_s_sb)->j_header_bh->b_data) ;
830 jh->j_last_flush_trans_id = cpu_to_le32(trans_id) ;
831 jh->j_first_unflushed_offset = cpu_to_le32(offset) ;
832 jh->j_mount_id = cpu_to_le32(SB_JOURNAL(p_s_sb)->j_mount_id) ;
833 set_bit(BH_Dirty, &(SB_JOURNAL(p_s_sb)->j_header_bh->b_state)) ;
834 ll_rw_block(WRITE, 1, &(SB_JOURNAL(p_s_sb)->j_header_bh)) ;
835 wait_on_buffer((SB_JOURNAL(p_s_sb)->j_header_bh)) ;
836 if (!buffer_uptodate(SB_JOURNAL(p_s_sb)->j_header_bh)) {
837 reiserfs_warning( p_s_sb, "reiserfs: journal-837: IO error during journal replay\n" );
838 return -EIO ;
839 }
840 }
841 return 0 ;
842 }
843
844 static int update_journal_header_block(struct super_block *p_s_sb,
845 unsigned long offset,
846 unsigned long trans_id) {
847 if (_update_journal_header_block(p_s_sb, offset, trans_id)) {
848 reiserfs_panic(p_s_sb, "journal-712: buffer write failed\n") ;
849 }
850 return 0 ;
851 }
852 /*
853 ** flush any and all journal lists older than you are
854 ** can only be called from flush_journal_list
855 */
856 static int flush_older_journal_lists(struct super_block *p_s_sb, struct reiserfs_journal_list *jl, unsigned long trans_id) {
857 int i, index ;
858 struct reiserfs_journal_list *other_jl ;
859
860 index = jl - SB_JOURNAL_LIST(p_s_sb) ;
861 for (i = 0 ; i < JOURNAL_LIST_COUNT ; i++) {
862 other_jl = SB_JOURNAL_LIST(p_s_sb) + ((index + i) % JOURNAL_LIST_COUNT) ;
863 if (other_jl && other_jl->j_len > 0 &&
864 other_jl->j_trans_id > 0 &&
865 other_jl->j_trans_id < trans_id &&
866 other_jl != jl) {
867 /* do not flush all */
868 flush_journal_list(p_s_sb, other_jl, 0) ;
869 }
870 }
871 return 0 ;
872 }
873
874 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate) {
875 if (buffer_journaled(bh)) {
876 reiserfs_warning(NULL, "clm-2084: pinned buffer %lu:%s sent to disk\n",
877 bh->b_blocknr, kdevname(bh->b_dev)) ;
878 }
879 mark_buffer_uptodate(bh, uptodate) ;
880 unlock_buffer(bh) ;
881 put_bh(bh) ;
882 }
883 static void submit_logged_buffer(struct buffer_head *bh) {
884 lock_buffer(bh) ;
885 get_bh(bh) ;
886 bh->b_end_io = reiserfs_end_buffer_io_sync ;
887 mark_buffer_notjournal_new(bh) ;
888 clear_bit(BH_Dirty, &bh->b_state) ;
889 submit_bh(WRITE, bh) ;
890 }
891
892 /* flush a journal list, both commit and real blocks
893 **
894 ** always set flushall to 1, unless you are calling from inside
895 ** flush_journal_list
896 **
897 ** IMPORTANT. This can only be called while there are no journal writers,
898 ** and the journal is locked. That means it can only be called from
899 ** do_journal_end, or by journal_release
900 */
901 static int flush_journal_list(struct super_block *s,
902 struct reiserfs_journal_list *jl, int flushall) {
903 struct reiserfs_journal_list *pjl ;
904 struct reiserfs_journal_cnode *cn, *last ;
905 int count ;
906 int was_jwait = 0 ;
907 int was_dirty = 0 ;
908 struct buffer_head *saved_bh ;
909 unsigned long j_len_saved = jl->j_len ;
910
911 if (j_len_saved <= 0) {
912 return 0 ;
913 }
914
915 if (atomic_read(&SB_JOURNAL(s)->j_wcount) != 0) {
916 reiserfs_warning(s, "clm-2048: flush_journal_list called with wcount %d\n",
917 atomic_read(&SB_JOURNAL(s)->j_wcount)) ;
918 }
919 /* if someone is getting the commit list, we must wait for them */
920 while (atomic_read(&(jl->j_commit_flushing))) {
921 sleep_on(&(jl->j_commit_wait)) ;
922 }
923 /* if someone is flushing this list, we must wait for them */
924 while (atomic_read(&(jl->j_flushing))) {
925 sleep_on(&(jl->j_flush_wait)) ;
926 }
927
928 /* this list is now ours, we can change anything we want */
929 atomic_set(&(jl->j_flushing), 1) ;
930
931 count = 0 ;
932 if (j_len_saved > SB_JOURNAL_TRANS_MAX(s)) {
933 reiserfs_panic(s, "journal-715: flush_journal_list, length is %lu, list number %d\n", j_len_saved, jl - SB_JOURNAL_LIST(s)) ;
934 atomic_dec(&(jl->j_flushing)) ;
935 return 0 ;
936 }
937
938 /* if all the work is already done, get out of here */
939 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
940 atomic_read(&(jl->j_commit_left)) <= 0) {
941 goto flush_older_and_return ;
942 }
943
944 /* start by putting the commit list on disk. This will also flush
945 ** the commit lists of any olders transactions
946 */
947 flush_commit_list(s, jl, 1) ;
948
949 /* are we done now? */
950 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
951 atomic_read(&(jl->j_commit_left)) <= 0) {
952 goto flush_older_and_return ;
953 }
954
955 /* loop through each cnode, see if we need to write it,
956 ** or wait on a more recent transaction, or just ignore it
957 */
958 if (atomic_read(&(SB_JOURNAL(s)->j_wcount)) != 0) {
959 reiserfs_panic(s, "journal-844: panic journal list is flushing, wcount is not 0\n") ;
960 }
961 cn = jl->j_realblock ;
962 while(cn) {
963 was_jwait = 0 ;
964 was_dirty = 0 ;
965 saved_bh = NULL ;
966 /* blocknr of 0 is no longer in the hash, ignore it */
967 if (cn->blocknr == 0) {
968 goto free_cnode ;
969 }
970 pjl = find_newer_jl_for_cn(cn) ;
971 /* the order is important here. We check pjl to make sure we
972 ** don't clear BH_JDirty_wait if we aren't the one writing this
973 ** block to disk
974 */
975 if (!pjl && cn->bh) {
976 saved_bh = cn->bh ;
977
978 /* we do this to make sure nobody releases the buffer while
979 ** we are working with it
980 */
981 get_bh(saved_bh) ;
982
983 if (buffer_journal_dirty(saved_bh)) {
984 was_jwait = 1 ;
985 mark_buffer_notjournal_dirty(saved_bh) ;
986 /* undo the inc from journal_mark_dirty */
987 put_bh(saved_bh) ;
988 }
989 if (can_dirty(cn)) {
990 was_dirty = 1 ;
991 }
992 }
993
994 /* if someone has this block in a newer transaction, just make
995 ** sure they are commited, and don't try writing it to disk
996 */
997 if (pjl) {
998 flush_commit_list(s, pjl, 1) ;
999 goto free_cnode ;
1000 }
1001
1002 /* bh == NULL when the block got to disk on its own, OR,
1003 ** the block got freed in a future transaction
1004 */
1005 if (saved_bh == NULL) {
1006 goto free_cnode ;
1007 }
1008
1009 /* this should never happen. kupdate_one_transaction has this list
1010 ** locked while it works, so we should never see a buffer here that
1011 ** is not marked JDirty_wait
1012 */
1013 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1014 reiserfs_warning(s, "journal-813: BAD! buffer %lu %cdirty %cjwait, not in a newer tranasction\n", saved_bh->b_blocknr,
1015 was_dirty ? ' ' : '!', was_jwait ? ' ' : '!') ;
1016 }
1017 /* kupdate_one_transaction waits on the buffers it is writing, so we
1018 ** should never see locked buffers here
1019 */
1020 if (buffer_locked(saved_bh)) {
1021 reiserfs_warning(s, "clm-2083: locked buffer %lu in flush_journal_list\n",
1022 saved_bh->b_blocknr) ;
1023 wait_on_buffer(saved_bh) ;
1024 if (!buffer_uptodate(saved_bh)) {
1025 reiserfs_panic(s, "journal-923: buffer write failed\n") ;
1026 }
1027 }
1028 if (was_dirty) {
1029 /* we inc again because saved_bh gets decremented at free_cnode */
1030 get_bh(saved_bh) ;
1031 set_bit(BLOCK_NEEDS_FLUSH, &cn->state) ;
1032 submit_logged_buffer(saved_bh) ;
1033 count++ ;
1034 } else {
1035 reiserfs_warning(s, "clm-2082: Unable to flush buffer %lu in flush_journal_list\n",
1036 saved_bh->b_blocknr) ;
1037 }
1038 free_cnode:
1039 last = cn ;
1040 cn = cn->next ;
1041 if (saved_bh) {
1042 /* we incremented this to keep others from taking the buffer head away */
1043 put_bh(saved_bh) ;
1044 if (atomic_read(&(saved_bh->b_count)) < 0) {
1045 reiserfs_warning(s, "journal-945: saved_bh->b_count < 0\n") ;
1046 }
1047 }
1048 }
1049 if (count > 0) {
1050 cn = jl->j_realblock ;
1051 while(cn) {
1052 if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1053 if (!cn->bh) {
1054 reiserfs_panic(s, "journal-1011: cn->bh is NULL\n") ;
1055 }
1056 wait_on_buffer(cn->bh) ;
1057 if (!cn->bh) {
1058 reiserfs_panic(s, "journal-1012: cn->bh is NULL\n") ;
1059 }
1060 if (!buffer_uptodate(cn->bh)) {
1061 reiserfs_panic(s, "journal-949: buffer write failed\n") ;
1062 }
1063 refile_buffer(cn->bh) ;
1064 brelse(cn->bh) ;
1065 }
1066 cn = cn->next ;
1067 }
1068 }
1069
1070 flush_older_and_return:
1071 /* before we can update the journal header block, we _must_ flush all
1072 ** real blocks from all older transactions to disk. This is because
1073 ** once the header block is updated, this transaction will not be
1074 ** replayed after a crash
1075 */
1076 if (flushall) {
1077 flush_older_journal_lists(s, jl, jl->j_trans_id) ;
1078 }
1079
1080 /* before we can remove everything from the hash tables for this
1081 ** transaction, we must make sure it can never be replayed
1082 **
1083 ** since we are only called from do_journal_end, we know for sure there
1084 ** are no allocations going on while we are flushing journal lists. So,
1085 ** we only need to update the journal header block for the last list
1086 ** being flushed
1087 */
1088 if (flushall) {
1089 update_journal_header_block(s, (jl->j_start + jl->j_len + 2) % SB_ONDISK_JOURNAL_SIZE(s), jl->j_trans_id) ;
1090 }
1091 remove_all_from_journal_list(s, jl, 0) ;
1092 jl->j_len = 0 ;
1093 atomic_set(&(jl->j_nonzerolen), 0) ;
1094 jl->j_start = 0 ;
1095 jl->j_realblock = NULL ;
1096 jl->j_commit_bh = NULL ;
1097 jl->j_trans_id = 0 ;
1098 atomic_dec(&(jl->j_flushing)) ;
1099 wake_up(&(jl->j_flush_wait)) ;
1100 return 0 ;
1101 }
1102
1103
1104 static int kupdate_one_transaction(struct super_block *s,
1105 struct reiserfs_journal_list *jl)
1106 {
1107 struct reiserfs_journal_list *pjl ; /* previous list for this cn */
1108 struct reiserfs_journal_cnode *cn, *walk_cn ;
1109 unsigned long blocknr ;
1110 int run = 0 ;
1111 int orig_trans_id = jl->j_trans_id ;
1112 struct buffer_head *saved_bh ;
1113 int ret = 0 ;
1114
1115 /* if someone is getting the commit list, we must wait for them */
1116 while (atomic_read(&(jl->j_commit_flushing))) {
1117 sleep_on(&(jl->j_commit_wait)) ;
1118 }
1119 /* if someone is flushing this list, we must wait for them */
1120 while (atomic_read(&(jl->j_flushing))) {
1121 sleep_on(&(jl->j_flush_wait)) ;
1122 }
1123 /* was it flushed while we slept? */
1124 if (jl->j_len <= 0 || jl->j_trans_id != orig_trans_id) {
1125 return 0 ;
1126 }
1127
1128 /* this list is now ours, we can change anything we want */
1129 atomic_set(&(jl->j_flushing), 1) ;
1130
1131 loop_start:
1132 cn = jl->j_realblock ;
1133 while(cn) {
1134 saved_bh = NULL ;
1135 /* if the blocknr == 0, this has been cleared from the hash,
1136 ** skip it
1137 */
1138 if (cn->blocknr == 0) {
1139 goto next ;
1140 }
1141 /* look for a more recent transaction that logged this
1142 ** buffer. Only the most recent transaction with a buffer in
1143 ** it is allowed to send that buffer to disk
1144 */
1145 pjl = find_newer_jl_for_cn(cn) ;
1146 if (run == 0 && !pjl && cn->bh && buffer_journal_dirty(cn->bh) &&
1147 can_dirty(cn))
1148 {
1149 if (!test_bit(BH_JPrepared, &cn->bh->b_state)) {
1150 set_bit(BLOCK_NEEDS_FLUSH, &cn->state) ;
1151 submit_logged_buffer(cn->bh) ;
1152 } else {
1153 /* someone else is using this buffer. We can't
1154 ** send it to disk right now because they might
1155 ** be changing/logging it.
1156 */
1157 ret = 1 ;
1158 }
1159 } else if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1160 clear_bit(BLOCK_NEEDS_FLUSH, &cn->state) ;
1161 if (!pjl && cn->bh) {
1162 wait_on_buffer(cn->bh) ;
1163 }
1164 /* check again, someone could have logged while we scheduled */
1165 pjl = find_newer_jl_for_cn(cn) ;
1166
1167 /* before the JDirty_wait bit is set, the
1168 ** buffer is added to the hash list. So, if we are
1169 ** run in the middle of a do_journal_end, we will notice
1170 ** if this buffer was logged and added from the latest
1171 ** transaction. In this case, we don't want to decrement
1172 ** b_count
1173 */
1174 if (!pjl && cn->bh && buffer_journal_dirty(cn->bh)) {
1175 blocknr = cn->blocknr ;
1176 walk_cn = cn ;
1177 saved_bh= cn->bh ;
1178 /* update all older transactions to show this block
1179 ** was flushed
1180 */
1181 mark_buffer_notjournal_dirty(cn->bh) ;
1182 while(walk_cn) {
1183 if (walk_cn->bh && walk_cn->blocknr == blocknr &&
1184 walk_cn->dev == cn->dev) {
1185 if (walk_cn->jlist) {
1186 atomic_dec(&(walk_cn->jlist->j_nonzerolen)) ;
1187 }
1188 walk_cn->bh = NULL ;
1189 }
1190 walk_cn = walk_cn->hnext ;
1191 }
1192 if (atomic_read(&saved_bh->b_count) < 1) {
1193 reiserfs_warning(s, "clm-2081: bad count on %lu\n",
1194 saved_bh->b_blocknr) ;
1195 }
1196 brelse(saved_bh) ;
1197 }
1198 }
1199 /*
1200 ** if the more recent transaction is committed to the log,
1201 ** this buffer can be considered flushed. Decrement our
1202 ** counters to reflect one less buffer that needs writing.
1203 **
1204 ** note, this relies on all of the above code being
1205 ** schedule free once pjl comes back non-null.
1206 */
1207 if (pjl && cn->bh && atomic_read(&pjl->j_commit_left) == 0) {
1208 atomic_dec(&cn->jlist->j_nonzerolen) ;
1209 cn->bh = NULL ;
1210 }
1211 next:
1212 cn = cn->next ;
1213 }
1214 /* the first run through the loop sends all the dirty buffers to
1215 ** ll_rw_block.
1216 ** the second run through the loop does all the accounting
1217 */
1218 if (run++ == 0) {
1219 goto loop_start ;
1220 }
1221
1222 atomic_set(&(jl->j_flushing), 0) ;
1223 wake_up(&(jl->j_flush_wait)) ;
1224 return ret ;
1225 }
1226 /* since we never give dirty buffers to bdflush/kupdate, we have to
1227 ** flush them ourselves. This runs through the journal lists, finds
1228 ** old metadata in need of flushing and sends it to disk.
1229 ** this does not end transactions, commit anything, or free
1230 ** cnodes.
1231 **
1232 ** returns the highest transaction id that was flushed last time
1233 */
1234 static unsigned long reiserfs_journal_kupdate(struct super_block *s) {
1235 struct reiserfs_journal_list *jl ;
1236 int i ;
1237 int start ;
1238 time_t age ;
1239 int ret = 0 ;
1240
1241 start = SB_JOURNAL_LIST_INDEX(s) ;
1242
1243 /* safety check to prevent flush attempts during a mount */
1244 if (start < 0) {
1245 return 0 ;
1246 }
1247 i = (start + 1) % JOURNAL_LIST_COUNT ;
1248 while(i != start) {
1249 jl = SB_JOURNAL_LIST(s) + i ;
1250 age = CURRENT_TIME - jl->j_timestamp ;
1251 if (jl->j_len > 0 && // age >= (JOURNAL_MAX_COMMIT_AGE * 2) &&
1252 atomic_read(&(jl->j_nonzerolen)) > 0 &&
1253 atomic_read(&(jl->j_commit_left)) == 0) {
1254
1255 if (jl->j_trans_id == SB_JOURNAL(s)->j_trans_id) {
1256 break ;
1257 }
1258 /* if ret was already 1, we want to preserve that */
1259 ret |= kupdate_one_transaction(s, jl) ;
1260 }
1261 if (atomic_read(&(jl->j_nonzerolen)) > 0) {
1262 ret |= 1 ;
1263 }
1264 i = (i + 1) % JOURNAL_LIST_COUNT ;
1265 }
1266 return ret ;
1267 }
1268
1269 /*
1270 ** removes any nodes in table with name block and dev as bh.
1271 ** only touchs the hnext and hprev pointers.
1272 */
1273 void remove_journal_hash(struct reiserfs_journal_cnode **table, struct reiserfs_journal_list *jl,struct buffer_head *bh,
1274 int remove_freed){
1275 struct reiserfs_journal_cnode *cur ;
1276 struct reiserfs_journal_cnode **head ;
1277
1278 if (!bh)
1279 return ;
1280
1281 head= &(journal_hash(table, bh->b_dev, bh->b_blocknr)) ;
1282 if (!head) {
1283 return ;
1284 }
1285 cur = *head ;
1286 while(cur) {
1287 if (cur->blocknr == bh->b_blocknr && cur->dev == bh->b_dev && (jl == NULL || jl == cur->jlist) &&
1288 (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1289 if (cur->hnext) {
1290 cur->hnext->hprev = cur->hprev ;
1291 }
1292 if (cur->hprev) {
1293 cur->hprev->hnext = cur->hnext ;
1294 } else {
1295 *head = cur->hnext ;
1296 }
1297 cur->blocknr = 0 ;
1298 cur->dev = 0 ;
1299 cur->state = 0 ;
1300 if (cur->bh && cur->jlist) /* anybody who clears the cur->bh will also dec the nonzerolen */
1301 atomic_dec(&(cur->jlist->j_nonzerolen)) ;
1302 cur->bh = NULL ;
1303 cur->jlist = NULL ;
1304 }
1305 cur = cur->hnext ;
1306 }
1307 }
1308
1309 static void free_journal_ram(struct super_block *p_s_sb) {
1310 vfree(SB_JOURNAL(p_s_sb)->j_cnode_free_orig) ;
1311 free_list_bitmaps(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_bitmap) ;
1312 free_bitmap_nodes(p_s_sb) ; /* must be after free_list_bitmaps */
1313 if (SB_JOURNAL(p_s_sb)->j_header_bh) {
1314 brelse(SB_JOURNAL(p_s_sb)->j_header_bh) ;
1315 }
1316 /* j_header_bh is on the journal dev, make sure not to release the journal
1317 * dev until we brelse j_header_bh
1318 */
1319 release_journal_dev(p_s_sb, SB_JOURNAL(p_s_sb));
1320 vfree(SB_JOURNAL(p_s_sb)) ;
1321 }
1322
1323 /*
1324 ** call on unmount. Only set error to 1 if you haven't made your way out
1325 ** of read_super() yet. Any other caller must keep error at 0.
1326 */
1327 static int do_journal_release(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, int error) {
1328 struct reiserfs_transaction_handle myth ;
1329
1330 /* we only want to flush out transactions if we were called with error == 0
1331 */
1332 if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
1333 /* end the current trans */
1334 do_journal_end(th, p_s_sb,10, FLUSH_ALL) ;
1335
1336 /* make sure something gets logged to force our way into the flush code */
1337 journal_join(&myth, p_s_sb, 1) ;
1338 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
1339 journal_mark_dirty(&myth, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
1340 do_journal_end(&myth, p_s_sb,1, FLUSH_ALL) ;
1341 }
1342
1343 /* we decrement before we wake up, because the commit thread dies off
1344 ** when it has been woken up and the count is <= 0
1345 */
1346 reiserfs_mounted_fs_count-- ;
1347 wake_up(&reiserfs_commit_thread_wait) ;
1348 sleep_on(&reiserfs_commit_thread_done) ;
1349
1350 free_journal_ram(p_s_sb) ;
1351
1352 return 0 ;
1353 }
1354
1355 /*
1356 ** call on unmount. flush all journal trans, release all alloc'd ram
1357 */
1358 int journal_release(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb) {
1359 return do_journal_release(th, p_s_sb, 0) ;
1360 }
1361 /*
1362 ** only call from an error condition inside reiserfs_read_super!
1363 */
1364 int journal_release_error(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb) {
1365 return do_journal_release(th, p_s_sb, 1) ;
1366 }
1367
1368 /* compares description block with commit block. returns 1 if they differ, 0 if they are the same */
1369 static int journal_compare_desc_commit(struct super_block *p_s_sb, struct reiserfs_journal_desc *desc,
1370 struct reiserfs_journal_commit *commit) {
1371 if (le32_to_cpu(commit->j_trans_id) != le32_to_cpu(desc->j_trans_id) ||
1372 le32_to_cpu(commit->j_len) != le32_to_cpu(desc->j_len) ||
1373 le32_to_cpu(commit->j_len) > SB_JOURNAL_TRANS_MAX(p_s_sb) ||
1374 le32_to_cpu(commit->j_len) <= 0
1375 ) {
1376 return 1 ;
1377 }
1378 return 0 ;
1379 }
1380 /* returns 0 if it did not find a description block
1381 ** returns -1 if it found a corrupt commit block
1382 ** returns 1 if both desc and commit were valid
1383 */
1384 static int journal_transaction_is_valid(struct super_block *p_s_sb, struct buffer_head *d_bh, unsigned long *oldest_invalid_trans_id, unsigned long *newest_mount_id) {
1385 struct reiserfs_journal_desc *desc ;
1386 struct reiserfs_journal_commit *commit ;
1387 struct buffer_head *c_bh ;
1388 unsigned long offset ;
1389
1390 if (!d_bh)
1391 return 0 ;
1392
1393 desc = (struct reiserfs_journal_desc *)d_bh->b_data ;
1394 if (le32_to_cpu(desc->j_len) > 0 && !memcmp(desc->j_magic, JOURNAL_DESC_MAGIC, 8)) {
1395 if (oldest_invalid_trans_id && *oldest_invalid_trans_id && le32_to_cpu(desc->j_trans_id) > *oldest_invalid_trans_id) {
1396 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-986: transaction "
1397 "is valid returning because trans_id %d is greater than "
1398 "oldest_invalid %lu\n", le32_to_cpu(desc->j_trans_id),
1399 *oldest_invalid_trans_id);
1400 return 0 ;
1401 }
1402 if (newest_mount_id && *newest_mount_id > le32_to_cpu(desc->j_mount_id)) {
1403 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1087: transaction "
1404 "is valid returning because mount_id %d is less than "
1405 "newest_mount_id %lu\n", desc->j_mount_id,
1406 *newest_mount_id) ;
1407 return -1 ;
1408 }
1409 if ( le32_to_cpu(desc->j_len) > SB_JOURNAL_TRANS_MAX(p_s_sb) ) {
1410 reiserfs_warning(p_s_sb, "journal-2018: Bad transaction length %d encountered, ignoring transaction\n", le32_to_cpu(desc->j_len));
1411 return -1 ;
1412 }
1413 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ;
1414
1415 /* ok, we have a journal description block, lets see if the transaction was valid */
1416 c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
1417 ((offset + le32_to_cpu(desc->j_len) + 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ;
1418 if (!c_bh)
1419 return 0 ;
1420 commit = (struct reiserfs_journal_commit *)c_bh->b_data ;
1421 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
1422 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1423 "journal_transaction_is_valid, commit offset %ld had bad "
1424 "time %d or length %d\n",
1425 c_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1426 le32_to_cpu(commit->j_trans_id),
1427 le32_to_cpu(commit->j_len));
1428 brelse(c_bh) ;
1429 if (oldest_invalid_trans_id) {
1430 *oldest_invalid_trans_id = le32_to_cpu(desc->j_trans_id) ;
1431 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1004: "
1432 "transaction_is_valid setting oldest invalid trans_id "
1433 "to %d\n", le32_to_cpu(desc->j_trans_id)) ;
1434 }
1435 return -1;
1436 }
1437 brelse(c_bh) ;
1438 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1006: found valid "
1439 "transaction start offset %lu, len %d id %d\n",
1440 d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1441 le32_to_cpu(desc->j_len), le32_to_cpu(desc->j_trans_id)) ;
1442 return 1 ;
1443 } else {
1444 return 0 ;
1445 }
1446 }
1447
1448 static void brelse_array(struct buffer_head **heads, int num) {
1449 int i ;
1450 for (i = 0 ; i < num ; i++) {
1451 brelse(heads[i]) ;
1452 }
1453 }
1454
1455 /*
1456 ** given the start, and values for the oldest acceptable transactions,
1457 ** this either reads in a replays a transaction, or returns because the transaction
1458 ** is invalid, or too old.
1459 */
1460 static int journal_read_transaction(struct super_block *p_s_sb, unsigned long cur_dblock, unsigned long oldest_start,
1461 unsigned long oldest_trans_id, unsigned long newest_mount_id) {
1462 struct reiserfs_journal_desc *desc ;
1463 struct reiserfs_journal_commit *commit ;
1464 unsigned long trans_id = 0 ;
1465 struct buffer_head *c_bh ;
1466 struct buffer_head *d_bh ;
1467 struct buffer_head **log_blocks = NULL ;
1468 struct buffer_head **real_blocks = NULL ;
1469 unsigned long trans_offset ;
1470 int i;
1471
1472 d_bh = journal_bread(p_s_sb, cur_dblock) ;
1473 if (!d_bh)
1474 return 1 ;
1475 desc = (struct reiserfs_journal_desc *)d_bh->b_data ;
1476 trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ;
1477 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
1478 "journal_read_transaction, offset %lu, len %d mount_id %d\n",
1479 d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1480 le32_to_cpu(desc->j_len), le32_to_cpu(desc->j_mount_id)) ;
1481 if (le32_to_cpu(desc->j_trans_id) < oldest_trans_id) {
1482 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
1483 "journal_read_trans skipping because %lu is too old\n",
1484 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)) ;
1485 brelse(d_bh) ;
1486 return 1 ;
1487 }
1488 if (le32_to_cpu(desc->j_mount_id) != newest_mount_id) {
1489 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
1490 "journal_read_trans skipping because %d is != "
1491 "newest_mount_id %lu\n", le32_to_cpu(desc->j_mount_id),
1492 newest_mount_id) ;
1493 brelse(d_bh) ;
1494 return 1 ;
1495 }
1496 c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
1497 ((trans_offset + le32_to_cpu(desc->j_len) + 1) %
1498 SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ;
1499 if (!c_bh) {
1500 brelse(d_bh) ;
1501 return 1 ;
1502 }
1503 commit = (struct reiserfs_journal_commit *)c_bh->b_data ;
1504 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
1505 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal_read_transaction, "
1506 "commit offset %ld had bad time %d or length %d\n",
1507 c_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1508 le32_to_cpu(commit->j_trans_id), le32_to_cpu(commit->j_len));
1509 brelse(c_bh) ;
1510 brelse(d_bh) ;
1511 return 1;
1512 }
1513 trans_id = le32_to_cpu(desc->j_trans_id) ;
1514 /* now we know we've got a good transaction, and it was inside the valid time ranges */
1515 log_blocks = reiserfs_kmalloc(le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), GFP_NOFS, p_s_sb) ;
1516 real_blocks = reiserfs_kmalloc(le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), GFP_NOFS, p_s_sb) ;
1517 if (!log_blocks || !real_blocks) {
1518 brelse(c_bh) ;
1519 brelse(d_bh) ;
1520 reiserfs_kfree(log_blocks, le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), p_s_sb) ;
1521 reiserfs_kfree(real_blocks, le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), p_s_sb) ;
1522 reiserfs_warning(p_s_sb, "journal-1169: kmalloc failed, unable to mount FS\n") ;
1523 return -1 ;
1524 }
1525 /* get all the buffer heads */
1526 for(i = 0 ; i < le32_to_cpu(desc->j_len) ; i++) {
1527 log_blocks[i] = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + (trans_offset + 1 + i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
1528 if (i < JOURNAL_TRANS_HALF) {
1529 real_blocks[i] = sb_getblk(p_s_sb, le32_to_cpu(desc->j_realblock[i])) ;
1530 } else {
1531 real_blocks[i] = sb_getblk(p_s_sb, le32_to_cpu(commit->j_realblock[i - JOURNAL_TRANS_HALF])) ;
1532 }
1533 if ( real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb) ) {
1534 reiserfs_warning(p_s_sb, "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem\n");
1535 goto abort_replay;
1536 }
1537 /* make sure we don't try to replay onto log or reserved area */
1538 if (is_block_in_log_or_reserved_area(p_s_sb, real_blocks[i]->b_blocknr)) {
1539 reiserfs_warning(p_s_sb, "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block\n") ;
1540 abort_replay:
1541 brelse_array(log_blocks, i) ;
1542 brelse_array(real_blocks, i) ;
1543 brelse(c_bh) ;
1544 brelse(d_bh) ;
1545 reiserfs_kfree(log_blocks, le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), p_s_sb) ;
1546 reiserfs_kfree(real_blocks, le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), p_s_sb) ;
1547 return -1 ;
1548 }
1549 }
1550 /* read in the log blocks, memcpy to the corresponding real block */
1551 ll_rw_block(READ, le32_to_cpu(desc->j_len), log_blocks) ;
1552 for (i = 0 ; i < le32_to_cpu(desc->j_len) ; i++) {
1553 wait_on_buffer(log_blocks[i]) ;
1554 if (!buffer_uptodate(log_blocks[i])) {
1555 reiserfs_warning(p_s_sb, "journal-1212: REPLAY FAILURE fsck required! buffer write failed\n") ;
1556 brelse_array(log_blocks + i, le32_to_cpu(desc->j_len) - i) ;
1557 brelse_array(real_blocks, le32_to_cpu(desc->j_len)) ;
1558 brelse(c_bh) ;
1559 brelse(d_bh) ;
1560 reiserfs_kfree(log_blocks, le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), p_s_sb) ;
1561 reiserfs_kfree(real_blocks, le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), p_s_sb) ;
1562 return -1 ;
1563 }
1564 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data, real_blocks[i]->b_size) ;
1565 mark_buffer_uptodate(real_blocks[i], 1) ;
1566 brelse(log_blocks[i]) ;
1567 }
1568 /* flush out the real blocks */
1569 for (i = 0 ; i < le32_to_cpu(desc->j_len) ; i++) {
1570 set_bit(BH_Dirty, &(real_blocks[i]->b_state)) ;
1571 ll_rw_block(WRITE, 1, real_blocks + i) ;
1572 }
1573 for (i = 0 ; i < le32_to_cpu(desc->j_len) ; i++) {
1574 wait_on_buffer(real_blocks[i]) ;
1575 if (!buffer_uptodate(real_blocks[i])) {
1576 reiserfs_warning(p_s_sb, "journal-1226: REPLAY FAILURE, fsck required! buffer write failed\n") ;
1577 brelse_array(real_blocks + i, le32_to_cpu(desc->j_len) - i) ;
1578 brelse(c_bh) ;
1579 brelse(d_bh) ;
1580 reiserfs_kfree(log_blocks, le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), p_s_sb) ;
1581 reiserfs_kfree(real_blocks, le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), p_s_sb) ;
1582 return -1 ;
1583 }
1584 brelse(real_blocks[i]) ;
1585 }
1586 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + ((trans_offset + le32_to_cpu(desc->j_len) + 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)) ;
1587 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1095: setting journal "
1588 "start to offset %ld\n",
1589 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)) ;
1590
1591 /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
1592 SB_JOURNAL(p_s_sb)->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ;
1593 SB_JOURNAL(p_s_sb)->j_last_flush_trans_id = trans_id ;
1594 SB_JOURNAL(p_s_sb)->j_trans_id = trans_id + 1;
1595 brelse(c_bh) ;
1596 brelse(d_bh) ;
1597 reiserfs_kfree(log_blocks, le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), p_s_sb) ;
1598 reiserfs_kfree(real_blocks, le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), p_s_sb) ;
1599 return 0 ;
1600 }
1601
1602 /*
1603 ** read and replay the log
1604 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
1605 ** transaction. This tests that before finding all the transactions in the log, whic makes normal mount times fast.
1606 **
1607 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
1608 **
1609 ** On exit, it sets things up so the first transaction will work correctly.
1610 */
1611 struct buffer_head * reiserfs_breada (kdev_t dev, int block, int bufsize,
1612 unsigned int max_block)
1613 {
1614 struct buffer_head * bhlist[BUFNR];
1615 unsigned int blocks = BUFNR;
1616 struct buffer_head * bh;
1617 int i, j;
1618
1619 bh = getblk (dev, block, bufsize);
1620 if (buffer_uptodate (bh))
1621 return (bh);
1622
1623 if (block + BUFNR > max_block) {
1624 blocks = max_block - block;
1625 }
1626 bhlist[0] = bh;
1627 j = 1;
1628 for (i = 1; i < blocks; i++) {
1629 bh = getblk (dev, block + i, bufsize);
1630 if (buffer_uptodate (bh)) {
1631 brelse (bh);
1632 break;
1633 }
1634 else bhlist[j++] = bh;
1635 }
1636 ll_rw_block (READ, j, bhlist);
1637 for(i = 1; i < j; i++)
1638 brelse (bhlist[i]);
1639 bh = bhlist[0];
1640 wait_on_buffer (bh);
1641 if (buffer_uptodate (bh))
1642 return bh;
1643 brelse (bh);
1644 return NULL;
1645 }
1646
1647 static struct buffer_head * journal_breada (struct super_block *p_s_sb, int block)
1648 {
1649 return reiserfs_breada (SB_JOURNAL_DEV(p_s_sb), block, p_s_sb->s_blocksize,
1650 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_ONDISK_JOURNAL_SIZE(p_s_sb));
1651 }
1652
1653 static int journal_read(struct super_block *p_s_sb) {
1654 struct reiserfs_journal_desc *desc ;
1655 unsigned long oldest_trans_id = 0;
1656 unsigned long oldest_invalid_trans_id = 0 ;
1657 time_t start ;
1658 unsigned long oldest_start = 0;
1659 unsigned long cur_dblock = 0 ;
1660 unsigned long newest_mount_id = 9 ;
1661 struct buffer_head *d_bh ;
1662 struct reiserfs_journal_header *jh ;
1663 int valid_journal_header = 0 ;
1664 int replay_count = 0 ;
1665 int continue_replay = 1 ;
1666 int ret ;
1667
1668 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ;
1669 printk("reiserfs: checking transaction log (device %s) ...\n",
1670 bdevname(SB_JOURNAL_DEV(p_s_sb))) ;
1671 printk("for (%s)\n",
1672 bdevname(p_s_sb->s_dev)) ;
1673
1674 start = CURRENT_TIME ;
1675
1676 /* step 1, read in the journal header block. Check the transaction it says
1677 ** is the first unflushed, and if that transaction is not valid,
1678 ** replay is done
1679 */
1680 SB_JOURNAL(p_s_sb)->j_header_bh = journal_bread(p_s_sb,
1681 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
1682 SB_ONDISK_JOURNAL_SIZE(p_s_sb)) ;
1683 if (!SB_JOURNAL(p_s_sb)->j_header_bh) {
1684 return 1 ;
1685 }
1686 jh = (struct reiserfs_journal_header *)(SB_JOURNAL(p_s_sb)->j_header_bh->b_data) ;
1687 if (le32_to_cpu(jh->j_first_unflushed_offset) >= 0 &&
1688 le32_to_cpu(jh->j_first_unflushed_offset) < SB_ONDISK_JOURNAL_SIZE(p_s_sb) &&
1689 le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
1690 oldest_start = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
1691 le32_to_cpu(jh->j_first_unflushed_offset) ;
1692 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
1693 newest_mount_id = le32_to_cpu(jh->j_mount_id);
1694 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1153: found in "
1695 "header: first_unflushed_offset %d, last_flushed_trans_id "
1696 "%lu\n", le32_to_cpu(jh->j_first_unflushed_offset),
1697 le32_to_cpu(jh->j_last_flush_trans_id)) ;
1698 valid_journal_header = 1 ;
1699
1700 /* now, we try to read the first unflushed offset. If it is not valid,
1701 ** there is nothing more we can do, and it makes no sense to read
1702 ** through the whole log.
1703 */
1704 d_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + le32_to_cpu(jh->j_first_unflushed_offset)) ;
1705 ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL) ;
1706 if (!ret) {
1707 continue_replay = 0 ;
1708 }
1709 brelse(d_bh) ;
1710 goto start_log_replay;
1711 }
1712
1713 if (continue_replay && is_read_only(p_s_sb->s_dev)) {
1714 reiserfs_warning(p_s_sb, "clm-2076: device is readonly, unable to replay log\n") ;
1715 return -1 ;
1716 }
1717 if (continue_replay && (p_s_sb->s_flags & MS_RDONLY)) {
1718 printk("Warning, log replay starting on readonly filesystem\n") ;
1719 }
1720
1721 /* ok, there are transactions that need to be replayed. start with the first log block, find
1722 ** all the valid transactions, and pick out the oldest.
1723 */
1724 while(continue_replay && cur_dblock < (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_ONDISK_JOURNAL_SIZE(p_s_sb))) {
1725 d_bh = journal_breada(p_s_sb, cur_dblock) ;
1726 ret = journal_transaction_is_valid(p_s_sb, d_bh, &oldest_invalid_trans_id, &newest_mount_id) ;
1727 if (ret == 1) {
1728 desc = (struct reiserfs_journal_desc *)d_bh->b_data ;
1729 if (oldest_start == 0) { /* init all oldest_ values */
1730 oldest_trans_id = le32_to_cpu(desc->j_trans_id) ;
1731 oldest_start = d_bh->b_blocknr ;
1732 newest_mount_id = le32_to_cpu(desc->j_mount_id) ;
1733 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1179: Setting "
1734 "oldest_start to offset %lu, trans_id %lu\n",
1735 oldest_start - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1736 oldest_trans_id) ;
1737 } else if (oldest_trans_id > le32_to_cpu(desc->j_trans_id)) {
1738 /* one we just read was older */
1739 oldest_trans_id = le32_to_cpu(desc->j_trans_id) ;
1740 oldest_start = d_bh->b_blocknr ;
1741 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1180: Resetting "
1742 "oldest_start to offset %lu, trans_id %lu\n",
1743 oldest_start - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1744 oldest_trans_id) ;
1745 }
1746 if (newest_mount_id < le32_to_cpu(desc->j_mount_id)) {
1747 newest_mount_id = le32_to_cpu(desc->j_mount_id) ;
1748 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
1749 "newest_mount_id to %d\n", le32_to_cpu(desc->j_mount_id));
1750 }
1751 cur_dblock += le32_to_cpu(desc->j_len) + 2 ;
1752 } else {
1753 cur_dblock++ ;
1754 }
1755 brelse(d_bh) ;
1756 }
1757
1758 start_log_replay:
1759 cur_dblock = oldest_start ;
1760 if (oldest_trans_id) {
1761 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1206: Starting replay "
1762 "from offset %lu, trans_id %lu\n",
1763 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1764 oldest_trans_id) ;
1765
1766 }
1767 replay_count = 0 ;
1768 while(continue_replay && oldest_trans_id > 0) {
1769 ret = journal_read_transaction(p_s_sb, cur_dblock, oldest_start, oldest_trans_id, newest_mount_id) ;
1770 if (ret < 0) {
1771 return ret ;
1772 } else if (ret != 0) {
1773 break ;
1774 }
1775 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_JOURNAL(p_s_sb)->j_start ;
1776 replay_count++ ;
1777 if (cur_dblock == oldest_start)
1778 break;
1779 }
1780
1781 if (oldest_trans_id == 0) {
1782 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1225: No valid "
1783 "transactions found\n") ;
1784 }
1785 /* j_start does not get set correctly if we don't replay any transactions.
1786 ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
1787 ** copy the trans_id from the header
1788 */
1789 if (valid_journal_header && replay_count == 0) {
1790 SB_JOURNAL(p_s_sb)->j_start = le32_to_cpu(jh->j_first_unflushed_offset) ;
1791 SB_JOURNAL(p_s_sb)->j_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
1792 SB_JOURNAL(p_s_sb)->j_last_flush_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) ;
1793 SB_JOURNAL(p_s_sb)->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
1794 } else {
1795 SB_JOURNAL(p_s_sb)->j_mount_id = newest_mount_id + 1 ;
1796 }
1797 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
1798 "newest_mount_id to %lu\n", SB_JOURNAL(p_s_sb)->j_mount_id) ;
1799 SB_JOURNAL(p_s_sb)->j_first_unflushed_offset = SB_JOURNAL(p_s_sb)->j_start ;
1800 if (replay_count > 0) {
1801 printk("reiserfs: replayed %d transactions in %lu seconds\n", replay_count,
1802 CURRENT_TIME - start) ;
1803 }
1804 if (!is_read_only(p_s_sb->s_dev) &&
1805 _update_journal_header_block(p_s_sb, SB_JOURNAL(p_s_sb)->j_start,
1806 SB_JOURNAL(p_s_sb)->j_last_flush_trans_id))
1807 {
1808 /* replay failed, caller must call free_journal_ram and abort
1809 ** the mount
1810 */
1811 return -1 ;
1812 }
1813 return 0 ;
1814 }
1815
1816
1817 struct reiserfs_journal_commit_task {
1818 struct super_block *p_s_sb ;
1819 int jindex ;
1820 int wake_on_finish ; /* if this is one, we wake the task_done queue, if it
1821 ** is zero, we free the whole struct on finish
1822 */
1823 struct reiserfs_journal_commit_task *self ;
1824 struct wait_queue *task_done ;
1825 struct tq_struct task ;
1826 } ;
1827
1828 static void reiserfs_journal_commit_task_func(struct reiserfs_journal_commit_task *ct) {
1829
1830 struct reiserfs_journal_list *jl ;
1831 jl = SB_JOURNAL_LIST(ct->p_s_sb) + ct->jindex ;
1832
1833 flush_commit_list(ct->p_s_sb, SB_JOURNAL_LIST(ct->p_s_sb) + ct->jindex, 1) ;
1834
1835 if (jl->j_len > 0 && atomic_read(&(jl->j_nonzerolen)) > 0 &&
1836 atomic_read(&(jl->j_commit_left)) == 0) {
1837 kupdate_one_transaction(ct->p_s_sb, jl) ;
1838 }
1839 reiserfs_kfree(ct->self, sizeof(struct reiserfs_journal_commit_task), ct->p_s_sb) ;
1840 }
1841
1842 static void setup_commit_task_arg(struct reiserfs_journal_commit_task *ct,
1843 struct super_block *p_s_sb,
1844 int jindex) {
1845 if (!ct) {
1846 reiserfs_panic(NULL, "journal-1360: setup_commit_task_arg called with NULL struct\n") ;
1847 }
1848 ct->p_s_sb = p_s_sb ;
1849 ct->jindex = jindex ;
1850 ct->task_done = NULL ;
1851 INIT_LIST_HEAD(&ct->task.list) ;
1852 ct->task.sync = 0 ;
1853 ct->task.routine = (void *)(void *)reiserfs_journal_commit_task_func ;
1854 ct->self = ct ;
1855 ct->task.data = (void *)ct ;
1856 }
1857
1858 static void commit_flush_async(struct super_block *p_s_sb, int jindex) {
1859 struct reiserfs_journal_commit_task *ct ;
1860 /* using GFP_NOFS, GFP_KERNEL could try to flush inodes, which will try
1861 ** to start/join a transaction, which will deadlock
1862 */
1863 ct = reiserfs_kmalloc(sizeof(struct reiserfs_journal_commit_task), GFP_NOFS, p_s_sb) ;
1864 if (ct) {
1865 setup_commit_task_arg(ct, p_s_sb, jindex) ;
1866 queue_task(&(ct->task), &reiserfs_commit_thread_tq);
1867 wake_up(&reiserfs_commit_thread_wait) ;
1868 } else {
1869 #ifdef CONFIG_REISERFS_CHECK
1870 reiserfs_warning(p_s_sb, "journal-1540: kmalloc failed, doing sync commit\n") ;
1871 #endif
1872 flush_commit_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + jindex, 1) ;
1873 }
1874 }
1875
1876 /*
1877 ** this is the commit thread. It is started with kernel_thread on
1878 ** FS mount, and journal_release() waits for it to exit.
1879 **
1880 ** It could do a periodic commit, but there is a lot code for that
1881 ** elsewhere right now, and I only wanted to implement this little
1882 ** piece for starters.
1883 **
1884 ** All we do here is sleep on the j_commit_thread_wait wait queue, and
1885 ** then run the per filesystem commit task queue when we wakeup.
1886 */
1887 static int reiserfs_journal_commit_thread(void *nullp) {
1888
1889 daemonize() ;
1890
1891 spin_lock_irq(¤t->sigmask_lock);
1892 sigfillset(¤t->blocked);
1893 recalc_sigpending(current);
1894 spin_unlock_irq(¤t->sigmask_lock);
1895
1896 sprintf(current->comm, "kreiserfsd") ;
1897 lock_kernel() ;
1898 while(1) {
1899
1900 while(TQ_ACTIVE(reiserfs_commit_thread_tq)) {
1901 run_task_queue(&reiserfs_commit_thread_tq) ;
1902 }
1903
1904 /* if there aren't any more filesystems left, break */
1905 if (reiserfs_mounted_fs_count <= 0) {
1906 run_task_queue(&reiserfs_commit_thread_tq) ;
1907 break ;
1908 }
1909 wake_up(&reiserfs_commit_thread_done) ;
1910 interruptible_sleep_on_timeout(&reiserfs_commit_thread_wait, 5 * HZ) ;
1911 }
1912 unlock_kernel() ;
1913 wake_up(&reiserfs_commit_thread_done) ;
1914 return 0 ;
1915 }
1916
1917 static void journal_list_init(struct super_block *p_s_sb) {
1918 int i ;
1919 for (i = 0 ; i < JOURNAL_LIST_COUNT ; i++) {
1920 init_waitqueue_head(&(SB_JOURNAL_LIST(p_s_sb)[i].j_commit_wait)) ;
1921 init_waitqueue_head(&(SB_JOURNAL_LIST(p_s_sb)[i].j_flush_wait)) ;
1922 }
1923 }
1924
1925 static int release_journal_dev( struct super_block *super,
1926 struct reiserfs_journal *journal )
1927 {
1928 int result;
1929
1930 result = 0;
1931
1932 if( journal -> j_dev_bd != NULL && journal->j_dev_bd != super->s_bdev) {
1933 result = blkdev_put( journal -> j_dev_bd, BDEV_FS );
1934 journal -> j_dev_bd = NULL;
1935 }
1936 if( journal -> j_dev_file != NULL ) {
1937 result = filp_close( journal -> j_dev_file, NULL );
1938 journal -> j_dev_file = NULL;
1939 }
1940 if( result != 0 ) {
1941 reiserfs_warning(super, "release_journal_dev: Cannot release journal device: %i", result );
1942 }
1943 return result;
1944 }
1945
1946 static int journal_init_dev( struct super_block *super,
1947 struct reiserfs_journal *journal,
1948 const char *jdev_name )
1949 {
1950 int result;
1951 kdev_t jdev;
1952 int blkdev_mode = FMODE_READ | FMODE_WRITE;
1953
1954 result = 0;
1955
1956 journal -> j_dev_bd = NULL;
1957 journal -> j_dev_file = NULL;
1958 jdev = SB_JOURNAL_DEV( super ) =
1959 SB_ONDISK_JOURNAL_DEVICE( super ) ?
1960 to_kdev_t(SB_ONDISK_JOURNAL_DEVICE( super )) : super -> s_dev;
1961
1962 /* there is no "jdev" option */
1963
1964 if (is_read_only(super->s_dev))
1965 blkdev_mode = FMODE_READ;
1966
1967 if( ( !jdev_name || !jdev_name[ 0 ] ) ) {
1968
1969 /* don't add an extra reference to the device when
1970 * the log is on the same disk as the FS. It makes the
1971 * raid code unhappy
1972 */
1973 if (jdev == super->s_dev) {
1974 journal->j_dev_bd = super->s_bdev;
1975 return 0;
1976 }
1977 journal -> j_dev_bd = bdget( kdev_t_to_nr( jdev ) );
1978 if( journal -> j_dev_bd ) {
1979 result = blkdev_get( journal -> j_dev_bd,
1980 blkdev_mode, 0, BDEV_FS );
1981 if (result) {
1982 bdput(journal->j_dev_bd);
1983 journal->j_dev_bd = NULL;
1984 }
1985 } else {
1986 result = -ENOMEM;
1987 }
1988 if( result != 0 )
1989 printk( "journal_init_dev: cannot init journal device\n '%s': %i",
1990 kdevname( jdev ), result );
1991
1992 return result;
1993 }
1994
1995 /* "jdev" option has been found */
1996
1997 journal -> j_dev_file = filp_open( jdev_name, 0, 0 );
1998 if( !IS_ERR( journal -> j_dev_file ) ) {
1999 struct inode *jdev_inode;
2000
2001 jdev_inode = journal -> j_dev_file -> f_dentry -> d_inode;
2002 journal -> j_dev_bd = jdev_inode -> i_bdev;
2003 if( !S_ISBLK( jdev_inode -> i_mode ) ) {
2004 printk( "journal_init_dev: '%s' is not a block device", jdev_name );
2005 result = -ENOTBLK;
2006 } else if( journal -> j_dev_file -> f_vfsmnt -> mnt_flags & MNT_NODEV) {
2007 printk( "journal_init_dev: Cannot use devices on '%s'", jdev_name );
2008 result = -EACCES;
2009 } else if( jdev_inode -> i_bdev == NULL ) {
2010 printk( "journal_init_dev: bdev unintialized for '%s'", jdev_name );
2011 result = -ENOMEM;
2012 } else if( ( result = blkdev_get( jdev_inode -> i_bdev,
2013 blkdev_mode,
2014 0, BDEV_FS ) ) != 0 ) {
2015 journal -> j_dev_bd = NULL;
2016 printk( "journal_init_dev: Cannot load device '%s': %i", jdev_name,
2017 result );
2018 } else
2019 /* ok */
2020 SB_JOURNAL_DEV( super ) =
2021 to_kdev_t( jdev_inode -> i_bdev -> bd_dev );
2022 } else {
2023 result = PTR_ERR( journal -> j_dev_file );
2024 journal -> j_dev_file = NULL;
2025 printk( "journal_init_dev: Cannot open '%s': %i", jdev_name, result );
2026 }
2027 if( result != 0 ) {
2028 release_journal_dev( super, journal );
2029 }
2030 printk( "journal_init_dev: journal device: %s", kdevname( SB_JOURNAL_DEV( super ) ) );
2031 return result;
2032 }
2033
2034 /*
2035 ** must be called once on fs mount. calls journal_read for you
2036 */
2037 int journal_init(struct super_block *p_s_sb, const char * j_dev_name,
2038 int old_format) {
2039 int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2 ;
2040 struct buffer_head *bhjh;
2041 struct reiserfs_super_block * rs;
2042 struct reiserfs_journal_header *jh;
2043 struct reiserfs_journal *journal;
2044
2045 if (sizeof(struct reiserfs_journal_commit) != 4096 ||
2046 sizeof(struct reiserfs_journal_desc) != 4096) {
2047 reiserfs_warning(p_s_sb, "journal-1249: commit or desc struct not 4096 %Zd %Zd\n",
2048 sizeof(struct reiserfs_journal_commit),
2049 sizeof(struct reiserfs_journal_desc)) ;
2050 return 1 ;
2051 }
2052
2053 if ( SB_ONDISK_JOURNAL_SIZE(p_s_sb) < 512 ) {
2054 reiserfs_warning(p_s_sb, "Journal size %d is less than 512+1 blocks, which unsupported\n", SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2055 return 1 ;
2056 }
2057
2058 journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof (struct reiserfs_journal)) ;
2059 if (!journal) {
2060 reiserfs_warning(p_s_sb, "journal-1256: unable to get memory for journal structure\n") ;
2061 return 1 ;
2062 }
2063 memset(journal, 0, sizeof(struct reiserfs_journal)) ;
2064 INIT_LIST_HEAD(&SB_JOURNAL(p_s_sb)->j_bitmap_nodes) ;
2065 INIT_LIST_HEAD (&SB_JOURNAL(p_s_sb)->j_prealloc_list);
2066
2067 reiserfs_allocate_list_bitmaps(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_bitmap,
2068 SB_BMAP_NR(p_s_sb)) ;
2069 allocate_bitmap_nodes(p_s_sb) ;
2070
2071 /* reserved for journal area support */
2072 SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ?
2073 REISERFS_OLD_DISK_OFFSET_IN_BYTES /
2074 p_s_sb->s_blocksize +
2075 SB_BMAP_NR(p_s_sb) + 1 :
2076 REISERFS_DISK_OFFSET_IN_BYTES /
2077 p_s_sb->s_blocksize + 2);
2078
2079 if( journal_init_dev( p_s_sb, journal, j_dev_name ) != 0 ) {
2080 reiserfs_warning(p_s_sb, "journal-1259: unable to initialize jornal device\n");
2081 goto free_and_return;
2082 }
2083
2084 rs = SB_DISK_SUPER_BLOCK(p_s_sb);
2085
2086 /* read journal header */
2087 bhjh = journal_bread (p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2088 SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2089 if (!bhjh) {
2090 reiserfs_warning(p_s_sb, "journal-459: unable to read journal header\n") ;
2091 goto free_and_return;
2092 }
2093 jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2094
2095 /* make sure that journal matches to the super block */
2096 if (is_reiserfs_jr(rs) &&
2097 jh->jh_journal.jp_journal_magic != sb_jp_journal_magic(rs)) {
2098 char jname[ 32 ];
2099 char fname[ 32 ];
2100
2101 strcpy( jname, kdevname( SB_JOURNAL_DEV(p_s_sb) ) );
2102 strcpy( fname, kdevname( p_s_sb->s_dev ) );
2103 printk("journal-460: journal header magic %x (device %s) does not "
2104 "match magic found in super block %x (device %s)\n",
2105 jh->jh_journal.jp_journal_magic, jname,
2106 sb_jp_journal_magic(rs), fname);
2107 brelse (bhjh);
2108 goto free_and_return;
2109 }
2110
2111 SB_JOURNAL_TRANS_MAX(p_s_sb) = le32_to_cpu (jh->jh_journal.jp_journal_trans_max);
2112 SB_JOURNAL_MAX_BATCH(p_s_sb) = le32_to_cpu (jh->jh_journal.jp_journal_max_batch);
2113 SB_JOURNAL_MAX_COMMIT_AGE(p_s_sb) = le32_to_cpu (jh->jh_journal.jp_journal_max_commit_age);
2114 SB_JOURNAL_MAX_TRANS_AGE(p_s_sb) = JOURNAL_MAX_TRANS_AGE;
2115
2116 if (SB_JOURNAL_TRANS_MAX(p_s_sb)) {
2117 /* make sure these parameters are available, assign if they are not */
2118 __u32 initial = SB_JOURNAL_TRANS_MAX(p_s_sb);
2119 __u32 ratio = 1;
2120
2121 if (p_s_sb->s_blocksize < 4096)
2122 ratio = 4096 / p_s_sb->s_blocksize;
2123
2124 if (SB_ONDISK_JOURNAL_SIZE(p_s_sb)/SB_JOURNAL_TRANS_MAX(p_s_sb) <
2125 JOURNAL_MIN_RATIO)
2126 {
2127 SB_JOURNAL_TRANS_MAX(p_s_sb) = SB_ONDISK_JOURNAL_SIZE(p_s_sb) /
2128 JOURNAL_MIN_RATIO;
2129 }
2130 if (SB_JOURNAL_TRANS_MAX(p_s_sb) > JOURNAL_TRANS_MAX_DEFAULT / ratio)
2131 SB_JOURNAL_TRANS_MAX(p_s_sb) = JOURNAL_TRANS_MAX_DEFAULT / ratio;
2132 if (SB_JOURNAL_TRANS_MAX(p_s_sb) < JOURNAL_TRANS_MIN_DEFAULT / ratio)
2133 SB_JOURNAL_TRANS_MAX(p_s_sb) = JOURNAL_TRANS_MIN_DEFAULT / ratio;
2134
2135 if (SB_JOURNAL_TRANS_MAX(p_s_sb) != initial) {
2136 printk ("reiserfs warning: wrong transaction max size (%u). "
2137 "Changed to %u\n", initial, SB_JOURNAL_TRANS_MAX(p_s_sb));
2138 }
2139 SB_JOURNAL_MAX_BATCH(p_s_sb) = SB_JOURNAL_TRANS_MAX(p_s_sb) *
2140 JOURNAL_MAX_BATCH_DEFAULT /
2141 JOURNAL_TRANS_MAX_DEFAULT;
2142 }
2143
2144 if (!SB_JOURNAL_TRANS_MAX(p_s_sb)) {
2145 /*we have the file system was created by old version of mkreiserfs
2146 so this field contains zero value */
2147 SB_JOURNAL_TRANS_MAX(p_s_sb) = JOURNAL_TRANS_MAX_DEFAULT ;
2148 SB_JOURNAL_MAX_BATCH(p_s_sb) = JOURNAL_MAX_BATCH_DEFAULT ;
2149 SB_JOURNAL_MAX_COMMIT_AGE(p_s_sb) = JOURNAL_MAX_COMMIT_AGE ;
2150
2151 /* for blocksize >= 4096 - max transaction size is 1024. For
2152 block size < 4096 trans max size is decreased proportionally */
2153 if (p_s_sb->s_blocksize < 4096) {
2154 SB_JOURNAL_TRANS_MAX(p_s_sb) /= (4096 / p_s_sb->s_blocksize) ;
2155 SB_JOURNAL_MAX_BATCH(p_s_sb) = SB_JOURNAL_TRANS_MAX(p_s_sb)*9 / 10;
2156 }
2157 }
2158
2159 brelse (bhjh);
2160
2161 SB_JOURNAL(p_s_sb)->j_list_bitmap_index = 0 ;
2162 SB_JOURNAL_LIST_INDEX(p_s_sb) = -10000 ; /* make sure flush_old_commits does not try to flush a list while replay is on */
2163
2164 /* clear out the journal list array */
2165 memset(SB_JOURNAL_LIST(p_s_sb), 0,
2166 sizeof(struct reiserfs_journal_list) * JOURNAL_LIST_COUNT) ;
2167
2168 journal_list_init(p_s_sb) ;
2169
2170 memset(SB_JOURNAL(p_s_sb)->j_list_hash_table, 0,
2171 JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)) ;
2172 memset(journal_writers, 0, sizeof(char *) * 512) ; /* debug code */
2173
2174 INIT_LIST_HEAD(&(SB_JOURNAL(p_s_sb)->j_dirty_buffers)) ;
2175
2176 SB_JOURNAL(p_s_sb)->j_start = 0 ;
2177 SB_JOURNAL(p_s_sb)->j_len = 0 ;
2178 SB_JOURNAL(p_s_sb)->j_len_alloc = 0 ;
2179 atomic_set(&(SB_JOURNAL(p_s_sb)->j_wcount), 0) ;
2180 SB_JOURNAL(p_s_sb)->j_bcount = 0 ;
2181 SB_JOURNAL(p_s_sb)->j_trans_start_time = 0 ;
2182 SB_JOURNAL(p_s_sb)->j_last = NULL ;
2183 SB_JOURNAL(p_s_sb)->j_first = NULL ;
2184 init_waitqueue_head(&(SB_JOURNAL(p_s_sb)->j_join_wait)) ;
2185 init_waitqueue_head(&(SB_JOURNAL(p_s_sb)->j_wait)) ;
2186
2187 SB_JOURNAL(p_s_sb)->j_trans_id = 10 ;
2188 SB_JOURNAL(p_s_sb)->j_mount_id = 10 ;
2189 SB_JOURNAL(p_s_sb)->j_state = 0 ;
2190 atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 0) ;
2191 atomic_set(&(SB_JOURNAL(p_s_sb)->j_wlock), 0) ;
2192 SB_JOURNAL(p_s_sb)->j_cnode_free_list = allocate_cnodes(num_cnodes) ;
2193 SB_JOURNAL(p_s_sb)->j_cnode_free_orig = SB_JOURNAL(p_s_sb)->j_cnode_free_list ;
2194 SB_JOURNAL(p_s_sb)->j_cnode_free = SB_JOURNAL(p_s_sb)->j_cnode_free_list ?
2195 num_cnodes : 0 ;
2196 SB_JOURNAL(p_s_sb)->j_cnode_used = 0 ;
2197 SB_JOURNAL(p_s_sb)->j_must_wait = 0 ;
2198 init_journal_hash(p_s_sb) ;
2199 SB_JOURNAL_LIST(p_s_sb)[0].j_list_bitmap = get_list_bitmap(p_s_sb, SB_JOURNAL_LIST(p_s_sb)) ;
2200 if (!(SB_JOURNAL_LIST(p_s_sb)[0].j_list_bitmap)) {
2201 reiserfs_warning(p_s_sb, "journal-2005, get_list_bitmap failed for journal list 0\n") ;
2202 goto free_and_return;
2203 }
2204 if (journal_read(p_s_sb) < 0) {
2205 reiserfs_warning(p_s_sb, "Replay Failure, unable to mount\n") ;
2206 goto free_and_return;
2207 }
2208 /* once the read is done, we can set this where it belongs */
2209 SB_JOURNAL_LIST_INDEX(p_s_sb) = 0 ;
2210
2211 if (reiserfs_dont_log (p_s_sb))
2212 return 0;
2213
2214 reiserfs_mounted_fs_count++ ;
2215 if (reiserfs_mounted_fs_count <= 1) {
2216 kernel_thread((void *)(void *)reiserfs_journal_commit_thread, NULL,
2217 CLONE_FS | CLONE_FILES | CLONE_VM) ;
2218 }
2219 return 0 ;
2220
2221 free_and_return:
2222 free_journal_ram(p_s_sb);
2223 return 1;
2224 }
2225
2226 /*
2227 ** test for a polite end of the current transaction. Used by file_write, and should
2228 ** be used by delete to make sure they don't write more than can fit inside a single
2229 ** transaction
2230 */
2231 int journal_transaction_should_end(struct reiserfs_transaction_handle *th, int new_alloc) {
2232 time_t now = CURRENT_TIME ;
2233 if (reiserfs_dont_log(th->t_super))
2234 return 0 ;
2235 if ( SB_JOURNAL(th->t_super)->j_must_wait > 0 ||
2236 (SB_JOURNAL(th->t_super)->j_len_alloc + new_alloc) >= SB_JOURNAL_MAX_BATCH(th->t_super) ||
2237 atomic_read(&(SB_JOURNAL(th->t_super)->j_jlock)) ||
2238 (now - SB_JOURNAL(th->t_super)->j_trans_start_time) > SB_JOURNAL_MAX_TRANS_AGE(th->t_super) ||
2239 SB_JOURNAL(th->t_super)->j_cnode_free < (SB_JOURNAL_TRANS_MAX(th->t_super) * 3)) {
2240 return 1 ;
2241 }
2242 return 0 ;
2243 }
2244
2245 /* this must be called inside a transaction, and requires the
2246 ** kernel_lock to be held
2247 */
2248 void reiserfs_block_writes(struct reiserfs_transaction_handle *th) {
2249 struct super_block *s = th->t_super ;
2250 SB_JOURNAL(s)->j_must_wait = 1 ;
2251 set_bit(WRITERS_BLOCKED, &SB_JOURNAL(s)->j_state) ;
2252 return ;
2253 }
2254
2255 /* this must be called without a transaction started, and does not
2256 ** require BKL
2257 */
2258 void reiserfs_allow_writes(struct super_block *s) {
2259 clear_bit(WRITERS_BLOCKED, &SB_JOURNAL(s)->j_state) ;
2260 wake_up(&SB_JOURNAL(s)->j_join_wait) ;
2261 }
2262
2263 /* this must be called without a transaction started, and does not
2264 ** require BKL
2265 */
2266 void reiserfs_wait_on_write_block(struct super_block *s) {
2267 wait_event(SB_JOURNAL(s)->j_join_wait,
2268 !test_bit(WRITERS_BLOCKED, &SB_JOURNAL(s)->j_state)) ;
2269 }
2270
2271 /* join == true if you must join an existing transaction.
2272 ** join == false if you can deal with waiting for others to finish
2273 **
2274 ** this will block until the transaction is joinable. send the number of blocks you
2275 ** expect to use in nblocks.
2276 */
2277 static int do_journal_begin_r(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb,unsigned long nblocks,int join) {
2278 time_t now = CURRENT_TIME ;
2279 int old_trans_id ;
2280
2281 reiserfs_check_lock_depth("journal_begin") ;
2282 RFALSE( p_s_sb->s_flags & MS_RDONLY,
2283 "clm-2078: calling journal_begin on readonly FS") ;
2284
2285 if (reiserfs_dont_log(p_s_sb)) {
2286 th->t_super = p_s_sb ; /* others will check this for the don't log flag */
2287 return 0 ;
2288 }
2289 PROC_INFO_INC( p_s_sb, journal.journal_being );
2290
2291 relock:
2292 lock_journal(p_s_sb) ;
2293
2294 if (test_bit(WRITERS_BLOCKED, &SB_JOURNAL(p_s_sb)->j_state)) {
2295 unlock_journal(p_s_sb) ;
2296 reiserfs_wait_on_write_block(p_s_sb) ;
2297 PROC_INFO_INC( p_s_sb, journal.journal_relock_writers );
2298 goto relock ;
2299 }
2300
2301 /* if there is no room in the journal OR
2302 ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
2303 ** we don't sleep if there aren't other writers
2304 */
2305
2306 if ( (!join && SB_JOURNAL(p_s_sb)->j_must_wait > 0) ||
2307 ( !join && (SB_JOURNAL(p_s_sb)->j_len_alloc + nblocks + 2) >= SB_JOURNAL_MAX_BATCH(p_s_sb)) ||
2308 (!join && atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) > 0 && SB_JOURNAL(p_s_sb)->j_trans_start_time > 0 &&
2309 (now - SB_JOURNAL(p_s_sb)->j_trans_start_time) > SB_JOURNAL_MAX_TRANS_AGE(p_s_sb)) ||
2310 (!join && atomic_read(&(SB_JOURNAL(p_s_sb)->j_jlock)) ) ||
2311 (!join && SB_JOURNAL(p_s_sb)->j_cnode_free < (SB_JOURNAL_TRANS_MAX(p_s_sb) * 3))) {
2312
2313 unlock_journal(p_s_sb) ; /* allow others to finish this transaction */
2314
2315 /* if writer count is 0, we can just force this transaction to end, and start
2316 ** a new one afterwards.
2317 */
2318 if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) <= 0) {
2319 struct reiserfs_transaction_handle myth ;
2320 journal_join(&myth, p_s_sb, 1) ;
2321 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
2322 journal_mark_dirty(&myth, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
2323 do_journal_end(&myth, p_s_sb,1,COMMIT_NOW) ;
2324 } else {
2325 /* but if the writer count isn't zero, we have to wait for the current writers to finish.
2326 ** They won't batch on transaction end once we set j_jlock
2327 */
2328 atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 1) ;
2329 old_trans_id = SB_JOURNAL(p_s_sb)->j_trans_id ;
2330 while(atomic_read(&(SB_JOURNAL(p_s_sb)->j_jlock)) &&
2331 SB_JOURNAL(p_s_sb)->j_trans_id == old_trans_id) {
2332 sleep_on(&(SB_JOURNAL(p_s_sb)->j_join_wait)) ;
2333 }
2334 }
2335 PROC_INFO_INC( p_s_sb, journal.journal_relock_wcount );
2336 goto relock ;
2337 }
2338
2339 if (SB_JOURNAL(p_s_sb)->j_trans_start_time == 0) { /* we are the first writer, set trans_id */
2340 SB_JOURNAL(p_s_sb)->j_trans_start_time = now ;
2341 }
2342 atomic_inc(&(SB_JOURNAL(p_s_sb)->j_wcount)) ;
2343 SB_JOURNAL(p_s_sb)->j_len_alloc += nblocks ;
2344 th->t_blocks_logged = 0 ;
2345 th->t_blocks_allocated = nblocks ;
2346 th->t_super = p_s_sb ;
2347 th->t_trans_id = SB_JOURNAL(p_s_sb)->j_trans_id ;
2348 th->t_caller = "Unknown" ;
2349 unlock_journal(p_s_sb) ;
2350 p_s_sb->s_dirt = 1;
2351 return 0 ;
2352 }
2353
2354
2355 static int journal_join(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) {
2356 return do_journal_begin_r(th, p_s_sb, nblocks, 1) ;
2357 }
2358
2359 int journal_begin(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb, unsigned long nblocks) {
2360 return do_journal_begin_r(th, p_s_sb, nblocks, 0) ;
2361 }
2362
2363 /* not used at all */
2364 int journal_prepare(struct super_block * p_s_sb, struct buffer_head *bh) {
2365 return 0 ;
2366 }
2367
2368 /*
2369 ** puts bh into the current transaction. If it was already there, reorders removes the
2370 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
2371 **
2372 ** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the
2373 ** transaction is committed.
2374 **
2375 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
2376 */
2377 int journal_mark_dirty(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, struct buffer_head *bh) {
2378 struct reiserfs_journal_cnode *cn = NULL;
2379 int count_already_incd = 0 ;
2380 int prepared = 0 ;
2381
2382 PROC_INFO_INC( p_s_sb, journal.mark_dirty );
2383 if (reiserfs_dont_log(th->t_super)) {
2384 mark_buffer_dirty(bh) ;
2385 return 0 ;
2386 }
2387
2388 if (th->t_trans_id != SB_JOURNAL(p_s_sb)->j_trans_id) {
2389 reiserfs_panic(th->t_super, "journal-1577: handle trans id %ld != current trans id %ld\n",
2390 th->t_trans_id, SB_JOURNAL(p_s_sb)->j_trans_id);
2391 }
2392 p_s_sb->s_dirt = 1 ;
2393
2394 prepared = test_and_clear_bit(BH_JPrepared, &bh->b_state) ;
2395 /* already in this transaction, we are done */
2396 if (buffer_journaled(bh)) {
2397 PROC_INFO_INC( p_s_sb, journal.mark_dirty_already );
2398 return 0 ;
2399 }
2400
2401 /* this must be turned into a panic instead of a warning. We can't allow
2402 ** a dirty or journal_dirty or locked buffer to be logged, as some changes
2403 ** could get to disk too early. NOT GOOD.
2404 */
2405 if (!prepared || buffer_locked(bh)) {
2406 reiserfs_warning(p_s_sb, "journal-1777: buffer %lu bad state %cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT\n", bh->b_blocknr, prepared ? ' ' : '!',
2407 buffer_locked(bh) ? ' ' : '!',
2408 buffer_dirty(bh) ? ' ' : '!',
2409 buffer_journal_dirty(bh) ? ' ' : '!') ;
2410 show_reiserfs_locks() ;
2411 }
2412 count_already_incd = clear_prepared_bits(bh) ;
2413
2414 if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) <= 0) {
2415 reiserfs_warning(p_s_sb, "journal-1409: journal_mark_dirty returning because j_wcount was %d\n", atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount))) ;
2416 return 1 ;
2417 }
2418 /* this error means I've screwed up, and we've overflowed the transaction.
2419 ** Nothing can be done here, except make the FS readonly or panic.
2420 */
2421 if (SB_JOURNAL(p_s_sb)->j_len >= SB_JOURNAL_TRANS_MAX(p_s_sb)) {
2422 reiserfs_panic(th->t_super, "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n", SB_JOURNAL(p_s_sb)->j_len) ;
2423 }
2424
2425 if (buffer_journal_dirty(bh)) {
2426 count_already_incd = 1 ;
2427 PROC_INFO_INC( p_s_sb, journal.mark_dirty_notjournal );
2428 mark_buffer_notjournal_dirty(bh) ;
2429 }
2430
2431 if (buffer_dirty(bh)) {
2432 clear_bit(BH_Dirty, &bh->b_state) ;
2433 }
2434
2435 if (buffer_journaled(bh)) { /* must double check after getting lock */
2436 goto done ;
2437 }
2438
2439 if (SB_JOURNAL(p_s_sb)->j_len > SB_JOURNAL(p_s_sb)->j_len_alloc) {
2440 SB_JOURNAL(p_s_sb)->j_len_alloc = SB_JOURNAL(p_s_sb)->j_len + JOURNAL_PER_BALANCE_CNT ;
2441 }
2442
2443 set_bit(BH_JDirty, &bh->b_state) ;
2444
2445 /* now put this guy on the end */
2446 if (!cn) {
2447 cn = get_cnode(p_s_sb) ;
2448 if (!cn) {
2449 reiserfs_panic(p_s_sb, "get_cnode failed!\n");
2450 }
2451
2452 if (th->t_blocks_logged == th->t_blocks_allocated) {
2453 th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT ;
2454 SB_JOURNAL(p_s_sb)->j_len_alloc += JOURNAL_PER_BALANCE_CNT ;
2455 }
2456 th->t_blocks_logged++ ;
2457 SB_JOURNAL(p_s_sb)->j_len++ ;
2458
2459 cn->bh = bh ;
2460 cn->blocknr = bh->b_blocknr ;
2461 cn->dev = bh->b_dev ;
2462 cn->jlist = NULL ;
2463 insert_journal_hash(SB_JOURNAL(p_s_sb)->j_hash_table, cn) ;
2464 if (!count_already_incd) {
2465 get_bh(bh) ;
2466 }
2467 }
2468 cn->next = NULL ;
2469 cn->prev = SB_JOURNAL(p_s_sb)->j_last ;
2470 cn->bh = bh ;
2471 if (SB_JOURNAL(p_s_sb)->j_last) {
2472 SB_JOURNAL(p_s_sb)->j_last->next = cn ;
2473 SB_JOURNAL(p_s_sb)->j_last = cn ;
2474 } else {
2475 SB_JOURNAL(p_s_sb)->j_first = cn ;
2476 SB_JOURNAL(p_s_sb)->j_last = cn ;
2477 }
2478 done:
2479 return 0 ;
2480 }
2481
2482 /*
2483 ** if buffer already in current transaction, do a journal_mark_dirty
2484 ** otherwise, just mark it dirty and move on. Used for writes to meta blocks
2485 ** that don't need journaling
2486 */
2487 int journal_mark_dirty_nolog(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, struct buffer_head *bh) {
2488 if (reiserfs_dont_log(th->t_super) || buffer_journaled(bh) ||
2489 buffer_journal_dirty(bh)) {
2490 return journal_mark_dirty(th, p_s_sb, bh) ;
2491 }
2492 if (get_journal_hash_dev(SB_JOURNAL(p_s_sb)->j_list_hash_table, bh->b_dev,bh->b_blocknr,bh->b_size)) {
2493 return journal_mark_dirty(th, p_s_sb, bh) ;
2494 }
2495 mark_buffer_dirty(bh) ;
2496 return 0 ;
2497 }
2498
2499 int journal_end(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) {
2500 return do_journal_end(th, p_s_sb, nblocks, 0) ;
2501 }
2502
2503 /* removes from the current transaction, relsing and descrementing any counters.
2504 ** also files the removed buffer directly onto the clean list
2505 **
2506 ** called by journal_mark_freed when a block has been deleted
2507 **
2508 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
2509 */
2510 static int remove_from_transaction(struct super_block *p_s_sb, unsigned long blocknr, int already_cleaned) {
2511 struct buffer_head *bh ;
2512 struct reiserfs_journal_cnode *cn ;
2513 int ret = 0;
2514
2515 cn = get_journal_hash_dev(SB_JOURNAL(p_s_sb)->j_hash_table, p_s_sb->s_dev, blocknr, p_s_sb->s_blocksize) ;
2516 if (!cn || !cn->bh) {
2517 return ret ;
2518 }
2519 bh = cn->bh ;
2520 if (cn->prev) {
2521 cn->prev->next = cn->next ;
2522 }
2523 if (cn->next) {
2524 cn->next->prev = cn->prev ;
2525 }
2526 if (cn == SB_JOURNAL(p_s_sb)->j_first) {
2527 SB_JOURNAL(p_s_sb)->j_first = cn->next ;
2528 }
2529 if (cn == SB_JOURNAL(p_s_sb)->j_last) {
2530 SB_JOURNAL(p_s_sb)->j_last = cn->prev ;
2531 }
2532 remove_journal_hash(SB_JOURNAL(p_s_sb)->j_hash_table, NULL, bh, 0) ;
2533 mark_buffer_not_journaled(bh) ; /* don't log this one */
2534
2535 if (!already_cleaned) {
2536 mark_buffer_notjournal_dirty(bh) ;
2537 put_bh(bh) ;
2538 if (atomic_read(&(bh->b_count)) < 0) {
2539 reiserfs_warning(p_s_sb, "journal-1752: remove from trans, b_count < 0\n") ;
2540 }
2541 if (!buffer_locked(bh)) reiserfs_clean_and_file_buffer(bh) ;
2542 ret = 1 ;
2543 }
2544 SB_JOURNAL(p_s_sb)->j_len-- ;
2545 SB_JOURNAL(p_s_sb)->j_len_alloc-- ;
2546 free_cnode(p_s_sb, cn) ;
2547 return ret ;
2548 }
2549
2550 /* removes from a specific journal list hash */
2551 static int remove_from_journal_list(struct super_block *s, struct reiserfs_journal_list *jl, struct buffer_head *bh, int remove_freed) {
2552 remove_journal_hash(SB_JOURNAL(s)->j_list_hash_table, jl, bh, remove_freed) ;
2553 return 0 ;
2554 }
2555
2556 /*
2557 ** for any cnode in a journal list, it can only be dirtied of all the
2558 ** transactions that include it are commited to disk.
2559 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
2560 ** and 0 if you aren't
2561 **
2562 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
2563 ** blocks for a given transaction on disk
2564 **
2565 */
2566 static int can_dirty(struct reiserfs_journal_cnode *cn) {
2567 kdev_t dev = cn->dev ;
2568 unsigned long blocknr = cn->blocknr ;
2569 struct reiserfs_journal_cnode *cur = cn->hprev ;
2570 int can_dirty = 1 ;
2571
2572 /* first test hprev. These are all newer than cn, so any node here
2573 ** with the name block number and dev means this node can't be sent
2574 ** to disk right now.
2575 */
2576 while(cur && can_dirty) {
2577 if (cur->jlist && cur->bh && cur->blocknr && cur->dev == dev &&
2578 cur->blocknr == blocknr) {
2579 can_dirty = 0 ;
2580 }
2581 cur = cur->hprev ;
2582 }
2583 /* then test hnext. These are all older than cn. As long as they
2584 ** are committed to the log, it is safe to write cn to disk
2585 */
2586 cur = cn->hnext ;
2587 while(cur && can_dirty) {
2588 if (cur->jlist && cur->jlist->j_len > 0 &&
2589 atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
2590 cur->blocknr && cur->dev == dev && cur->blocknr == blocknr) {
2591 can_dirty = 0 ;
2592 }
2593 cur = cur->hnext ;
2594 }
2595 return can_dirty ;
2596 }
2597
2598 /* syncs the commit blocks, but does not force the real buffers to disk
2599 ** will wait until the current transaction is done/commited before returning
2600 */
2601 int journal_end_sync(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) {
2602
2603 if (SB_JOURNAL(p_s_sb)->j_len == 0) {
2604 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
2605 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
2606 }
2607 return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT) ;
2608 }
2609
2610 int show_reiserfs_locks(void) {
2611
2612 dump_journal_writers() ;
2613 return 0 ;
2614 }
2615
2616 /*
2617 ** used to get memory back from async commits that are floating around
2618 ** and to reclaim any blocks deleted but unusable because their commits
2619 ** haven't hit disk yet. called from bitmap.c
2620 **
2621 ** if it starts flushing things, it ors SCHEDULE_OCCURRED into repeat.
2622 ** note, this is just if schedule has a chance of occuring. I need to
2623 ** change flush_commit_lists to have a repeat parameter too.
2624 **
2625 */
2626 void flush_async_commits(struct super_block *p_s_sb) {
2627 int i ;
2628
2629 for (i = 0 ; i < JOURNAL_LIST_COUNT ; i++) {
2630 if (i != SB_JOURNAL_LIST_INDEX(p_s_sb)) {
2631 flush_commit_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + i, 1) ;
2632 }
2633 }
2634 }
2635
2636 /*
2637 ** flushes any old transactions to disk
2638 ** ends the current transaction if it is too old
2639 **
2640 ** also calls flush_journal_list with old_only == 1, which allows me to reclaim
2641 ** memory and such from the journal lists whose real blocks are all on disk.
2642 **
2643 ** called by sync_dev_journal from buffer.c
2644 */
2645 int flush_old_commits(struct super_block *p_s_sb, int immediate) {
2646 int i ;
2647 int count = 0;
2648 int start ;
2649 time_t now ;
2650 struct reiserfs_transaction_handle th ;
2651
2652 start = SB_JOURNAL_LIST_INDEX(p_s_sb) ;
2653 now = CURRENT_TIME ;
2654
2655 /* safety check so we don't flush while we are replaying the log during mount */
2656 if (SB_JOURNAL_LIST_INDEX(p_s_sb) < 0) {
2657 return 0 ;
2658 }
2659 /* starting with oldest, loop until we get to the start */
2660 i = (SB_JOURNAL_LIST_INDEX(p_s_sb) + 1) % JOURNAL_LIST_COUNT ;
2661 while(i != start) {
2662 if (SB_JOURNAL_LIST(p_s_sb)[i].j_len > 0 && ((now - SB_JOURNAL_LIST(p_s_sb)[i].j_timestamp) > SB_JOURNAL_MAX_COMMIT_AGE(p_s_sb) ||
2663 immediate)) {
2664 /* we have to check again to be sure the current transaction did not change */
2665 if (i != SB_JOURNAL_LIST_INDEX(p_s_sb)) {
2666 flush_commit_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + i, 1) ;
2667 }
2668 }
2669 i = (i + 1) % JOURNAL_LIST_COUNT ;
2670 count++ ;
2671 }
2672 /* now, check the current transaction. If there are no writers, and it is too old, finish it, and
2673 ** force the commit blocks to disk
2674 */
2675 if (!immediate && atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) <= 0 &&
2676 SB_JOURNAL(p_s_sb)->j_trans_start_time > 0 &&
2677 SB_JOURNAL(p_s_sb)->j_len > 0 &&
2678 (now - SB_JOURNAL(p_s_sb)->j_trans_start_time) > SB_JOURNAL_MAX_TRANS_AGE(p_s_sb)) {
2679 journal_join(&th, p_s_sb, 1) ;
2680 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
2681 journal_mark_dirty(&th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
2682 do_journal_end(&th, p_s_sb,1, COMMIT_NOW) ;
2683 } else if (immediate) { /* belongs above, but I wanted this to be very explicit as a special case. If they say to
2684 flush, we must be sure old transactions hit the disk too. */
2685 journal_join(&th, p_s_sb, 1) ;
2686 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
2687 journal_mark_dirty(&th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
2688 do_journal_end(&th, p_s_sb,1, COMMIT_NOW | WAIT) ;
2689 }
2690 reiserfs_journal_kupdate(p_s_sb) ;
2691 return 0 ;
2692 }
2693
2694 /*
2695 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
2696 **
2697 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
2698 ** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just
2699 ** flushes the commit list and returns 0.
2700 **
2701 ** Won't batch when flush or commit_now is set. Also won't batch when others are waiting on j_join_wait.
2702 **
2703 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
2704 */
2705 static int check_journal_end(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb,
2706 unsigned long nblocks, int flags) {
2707
2708 time_t now ;
2709 int flush = flags & FLUSH_ALL ;
2710 int commit_now = flags & COMMIT_NOW ;
2711 int wait_on_commit = flags & WAIT ;
2712
2713 if (th->t_trans_id != SB_JOURNAL(p_s_sb)->j_trans_id) {
2714 reiserfs_panic(th->t_super, "journal-1577: handle trans id %ld != current trans id %ld\n",
2715 th->t_trans_id, SB_JOURNAL(p_s_sb)->j_trans_id);
2716 }
2717
2718 SB_JOURNAL(p_s_sb)->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged) ;
2719 if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) > 0) { /* <= 0 is allowed. unmounting might not call begin */
2720 atomic_dec(&(SB_JOURNAL(p_s_sb)->j_wcount)) ;
2721 }
2722
2723 /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
2724 ** will be dealt with by next transaction that actually writes something, but should be taken
2725 ** care of in this trans
2726 */
2727 if (SB_JOURNAL(p_s_sb)->j_len == 0) {
2728 int wcount = atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) ;
2729 unlock_journal(p_s_sb) ;
2730 if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_jlock)) > 0 && wcount <= 0) {
2731 atomic_dec(&(SB_JOURNAL(p_s_sb)->j_jlock)) ;
2732 wake_up(&(SB_JOURNAL(p_s_sb)->j_join_wait)) ;
2733 }
2734 return 0 ;
2735 }
2736 /* if wcount > 0, and we are called to with flush or commit_now,
2737 ** we wait on j_join_wait. We will wake up when the last writer has
2738 ** finished the transaction, and started it on its way to the disk.
2739 ** Then, we flush the commit or journal list, and just return 0
2740 ** because the rest of journal end was already done for this transaction.
2741 */
2742 if (atomic_read(&(SB_JOURNAL(p_s_sb)->j_wcount)) > 0) {
2743 if (flush || commit_now) {
2744 int orig_jindex = SB_JOURNAL_LIST_INDEX(p_s_sb) ;
2745 atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 1) ;
2746 if (flush) {
2747 SB_JOURNAL(p_s_sb)->j_next_full_flush = 1 ;
2748 }
2749 unlock_journal(p_s_sb) ;
2750 /* sleep while the current transaction is still j_jlocked */
2751 while(atomic_read(&(SB_JOURNAL(p_s_sb)->j_jlock)) &&
2752 SB_JOURNAL(p_s_sb)->j_trans_id == th->t_trans_id) {
2753 sleep_on(&(SB_JOURNAL(p_s_sb)->j_join_wait)) ;
2754 }
2755 if (commit_now) {
2756 if (wait_on_commit) {
2757 flush_commit_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + orig_jindex, 1) ;
2758 } else {
2759 commit_flush_async(p_s_sb, orig_jindex) ;
2760 }
2761 }
2762 return 0 ;
2763 }
2764 unlock_journal(p_s_sb) ;
2765 return 0 ;
2766 }
2767
2768 /* deal with old transactions where we are the last writers */
2769 now = CURRENT_TIME ;
2770 if ((now - SB_JOURNAL(p_s_sb)->j_trans_start_time) > SB_JOURNAL_MAX_TRANS_AGE(p_s_sb)) {
2771 commit_now = 1 ;
2772 SB_JOURNAL(p_s_sb)->j_next_async_flush = 1 ;
2773 }
2774 /* don't batch when someone is waiting on j_join_wait */
2775 /* don't batch when syncing the commit or flushing the whole trans */
2776 if (!(SB_JOURNAL(p_s_sb)->j_must_wait > 0) && !(atomic_read(&(SB_JOURNAL(p_s_sb)->j_jlock))) && !flush && !commit_now &&
2777 (SB_JOURNAL(p_s_sb)->j_len < SB_JOURNAL_MAX_BATCH(p_s_sb)) &&
2778 SB_JOURNAL(p_s_sb)->j_len_alloc < SB_JOURNAL_MAX_BATCH(p_s_sb) && SB_JOURNAL(p_s_sb)->j_cnode_free > (SB_JOURNAL_TRANS_MAX(p_s_sb) * 3)) {
2779 SB_JOURNAL(p_s_sb)->j_bcount++ ;
2780 unlock_journal(p_s_sb) ;
2781 return 0 ;
2782 }
2783
2784 if (SB_JOURNAL(p_s_sb)->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
2785 reiserfs_panic(p_s_sb, "journal-003: journal_end: j_start (%ld) is too high\n", SB_JOURNAL(p_s_sb)->j_start) ;
2786 }
2787 return 1 ;
2788 }
2789
2790 /*
2791 ** Does all the work that makes deleting blocks safe.
2792 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
2793 **
2794 ** otherwise:
2795 ** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes
2796 ** before this transaction has finished.
2797 **
2798 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with
2799 ** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash,
2800 ** the block can't be reallocated yet.
2801 **
2802 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
2803 */
2804 int journal_mark_freed(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long blocknr) {
2805 struct reiserfs_journal_cnode *cn = NULL ;
2806 struct buffer_head *bh = NULL ;
2807 struct reiserfs_list_bitmap *jb = NULL ;
2808 int cleaned = 0 ;
2809
2810 if (reiserfs_dont_log(th->t_super)) {
2811 bh = sb_get_hash_table(p_s_sb, blocknr) ;
2812 if (bh && buffer_dirty (bh)) {
2813 reiserfs_warning (p_s_sb, "journal_mark_freed(dont_log): dirty buffer on hash list: %lx %ld\n", bh->b_state, blocknr);
2814 BUG ();
2815 }
2816 brelse (bh);
2817 return 0 ;
2818 }
2819 bh = sb_get_hash_table(p_s_sb, blocknr) ;
2820 /* if it is journal new, we just remove it from this transaction */
2821 if (bh && buffer_journal_new(bh)) {
2822 mark_buffer_notjournal_new(bh) ;
2823 clear_prepared_bits(bh) ;
2824 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned) ;
2825 } else {
2826 /* set the bit for this block in the journal bitmap for this transaction */
2827 jb = SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_list_bitmap ;
2828 if (!jb) {
2829 reiserfs_panic(p_s_sb, "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n") ;
2830 }
2831 set_bit_in_list_bitmap(p_s_sb, blocknr, jb) ;
2832
2833 /* Note, the entire while loop is not allowed to schedule. */
2834
2835 if (bh) {
2836 clear_prepared_bits(bh) ;
2837 }
2838 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned) ;
2839
2840 /* find all older transactions with this block, make sure they don't try to write it out */
2841 cn = get_journal_hash_dev(SB_JOURNAL(p_s_sb)->j_list_hash_table, p_s_sb->s_dev, blocknr, p_s_sb->s_blocksize) ;
2842 while (cn) {
2843 if (p_s_sb->s_dev == cn->dev && blocknr == cn->blocknr) {
2844 set_bit(BLOCK_FREED, &cn->state) ;
2845 if (cn->bh) {
2846 if (!cleaned) {
2847 /* remove_from_transaction will brelse the buffer if it was
2848 ** in the current trans
2849 */
2850 mark_buffer_notjournal_dirty(cn->bh) ;
2851 cleaned = 1 ;
2852 put_bh(cn->bh) ;
2853 if (atomic_read(&(cn->bh->b_count)) < 0) {
2854 reiserfs_warning(p_s_sb, "journal-2138: cn->bh->b_count < 0\n") ;
2855 }
2856 }
2857 if (cn->jlist) { /* since we are clearing the bh, we MUST dec nonzerolen */
2858 atomic_dec(&(cn->jlist->j_nonzerolen)) ;
2859 }
2860 cn->bh = NULL ;
2861 }
2862 }
2863 cn = cn->hnext ;
2864 }
2865 }
2866
2867 if (bh) {
2868 reiserfs_clean_and_file_buffer(bh) ;
2869 put_bh(bh) ; /* get_hash grabs the buffer */
2870 if (atomic_read(&(bh->b_count)) < 0) {
2871 reiserfs_warning(p_s_sb, "journal-2165: bh->b_count < 0\n") ;
2872 }
2873 }
2874 return 0 ;
2875 }
2876
2877 void reiserfs_update_inode_transaction(struct inode *inode) {
2878
2879 inode->u.reiserfs_i.i_trans_index = SB_JOURNAL_LIST_INDEX(inode->i_sb);
2880
2881 inode->u.reiserfs_i.i_trans_id = SB_JOURNAL(inode->i_sb)->j_trans_id ;
2882 }
2883
2884 void reiserfs_update_tail_transaction(struct inode *inode) {
2885
2886 inode->u.reiserfs_i.i_tail_trans_index = SB_JOURNAL_LIST_INDEX(inode->i_sb);
2887
2888 inode->u.reiserfs_i.i_tail_trans_id = SB_JOURNAL(inode->i_sb)->j_trans_id ;
2889 }
2890
2891 static void __commit_trans_index(struct inode *inode, unsigned long id,
2892 unsigned long index)
2893 {
2894 struct reiserfs_journal_list *jl ;
2895 struct reiserfs_transaction_handle th ;
2896 struct super_block *sb = inode->i_sb ;
2897
2898 jl = SB_JOURNAL_LIST(sb) + index;
2899
2900 /* is it from the current transaction, or from an unknown transaction? */
2901 if (id == SB_JOURNAL(sb)->j_trans_id) {
2902 journal_join(&th, sb, 1) ;
2903 journal_end_sync(&th, sb, 1) ;
2904 } else if (jl->j_trans_id == id) {
2905 flush_commit_list(sb, jl, 1) ;
2906 }
2907 /* if the transaction id does not match, this list is long since flushed
2908 ** and we don't have to do anything here
2909 */
2910 }
2911 void reiserfs_commit_for_tail(struct inode *inode) {
2912 unsigned long id = inode->u.reiserfs_i.i_tail_trans_id;
2913 unsigned long index = inode->u.reiserfs_i.i_tail_trans_index;
2914
2915 /* for tails, if this info is unset there's nothing to commit */
2916 if (id && index)
2917 __commit_trans_index(inode, id, index);
2918 }
2919 void reiserfs_commit_for_inode(struct inode *inode) {
2920 unsigned long id = inode->u.reiserfs_i.i_trans_id;
2921 unsigned long index = inode->u.reiserfs_i.i_trans_index;
2922
2923 /* for the whole inode, assume unset id or index means it was
2924 * changed in the current transaction. More conservative
2925 */
2926 if (!id || !index)
2927 reiserfs_update_inode_transaction(inode) ;
2928
2929 __commit_trans_index(inode, id, index);
2930 }
2931
2932 void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb,
2933 struct buffer_head *bh) {
2934 PROC_INFO_INC( p_s_sb, journal.restore_prepared );
2935 if (reiserfs_dont_log (p_s_sb))
2936 return;
2937
2938 if (!bh) {
2939 return ;
2940 }
2941 clear_bit(BH_JPrepared, &bh->b_state) ;
2942 }
2943
2944 extern struct tree_balance *cur_tb ;
2945 /*
2946 ** before we can change a metadata block, we have to make sure it won't
2947 ** be written to disk while we are altering it. So, we must:
2948 ** clean it
2949 ** wait on it.
2950 **
2951 */
2952 void reiserfs_prepare_for_journal(struct super_block *p_s_sb,
2953 struct buffer_head *bh, int wait) {
2954 int retry_count = 0 ;
2955
2956 PROC_INFO_INC( p_s_sb, journal.prepare );
2957 if (reiserfs_dont_log (p_s_sb))
2958 return;
2959
2960 while(!test_bit(BH_JPrepared, &bh->b_state) ||
2961 (wait && buffer_locked(bh))) {
2962 if (buffer_journaled(bh)) {
2963 set_bit(BH_JPrepared, &bh->b_state) ;
2964 return ;
2965 }
2966 set_bit(BH_JPrepared, &bh->b_state) ;
2967 if (wait) {
2968 RFALSE( buffer_locked(bh) && cur_tb != NULL,
2969 "waiting while do_balance was running\n") ;
2970 wait_on_buffer(bh) ;
2971 }
2972 PROC_INFO_INC( p_s_sb, journal.prepare_retry );
2973 retry_count++ ;
2974 }
2975 }
2976
2977 /*
2978 ** long and ugly. If flush, will not return until all commit
2979 ** blocks and all real buffers in the trans are on disk.
2980 ** If no_async, won't return until all commit blocks are on disk.
2981 **
2982 ** keep reading, there are comments as you go along
2983 */
2984 static int do_journal_end(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb, unsigned long nblocks,
2985 int flags) {
2986 struct reiserfs_journal_cnode *cn, *next, *jl_cn;
2987 struct reiserfs_journal_cnode *last_cn = NULL;
2988 struct reiserfs_journal_desc *desc ;
2989 struct reiserfs_journal_commit *commit ;
2990 struct buffer_head *c_bh ; /* commit bh */
2991 struct buffer_head *d_bh ; /* desc bh */
2992 int cur_write_start = 0 ; /* start index of current log write */
2993 int cur_blocks_left = 0 ; /* number of journal blocks left to write */
2994 int old_start ;
2995 int i ;
2996 int jindex ;
2997 int orig_jindex ;
2998 int flush = flags & FLUSH_ALL ;
2999 int commit_now = flags & COMMIT_NOW ;
3000 int wait_on_commit = flags & WAIT ;
3001 struct reiserfs_super_block *rs ;
3002
3003 if (reiserfs_dont_log(th->t_super)) {
3004 return 0 ;
3005 }
3006
3007 lock_journal(p_s_sb) ;
3008 if (SB_JOURNAL(p_s_sb)->j_next_full_flush) {
3009 flags |= FLUSH_ALL ;
3010 flush = 1 ;
3011 }
3012 if (SB_JOURNAL(p_s_sb)->j_next_async_flush) {
3013 flags |= COMMIT_NOW ;
3014 commit_now = 1 ;
3015 }
3016
3017 /* check_journal_end locks the journal, and unlocks if it does not return 1
3018 ** it tells us if we should continue with the journal_end, or just return
3019 */
3020 if (!check_journal_end(th, p_s_sb, nblocks, flags)) {
3021 return 0 ;
3022 }
3023
3024 /* check_journal_end might set these, check again */
3025 if (SB_JOURNAL(p_s_sb)->j_next_full_flush) {
3026 flush = 1 ;
3027 }
3028 if (SB_JOURNAL(p_s_sb)->j_next_async_flush) {
3029 commit_now = 1 ;
3030 }
3031 /*
3032 ** j must wait means we have to flush the log blocks, and the real blocks for
3033 ** this transaction
3034 */
3035 if (SB_JOURNAL(p_s_sb)->j_must_wait > 0) {
3036 flush = 1 ;
3037 }
3038
3039 #ifdef REISERFS_PREALLOCATE
3040 reiserfs_discard_all_prealloc(th); /* it should not involve new blocks into
3041 * the transaction */
3042 #endif
3043
3044 rs = SB_DISK_SUPER_BLOCK(p_s_sb) ;
3045 /* setup description block */
3046 d_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_JOURNAL(p_s_sb)->j_start) ;
3047 mark_buffer_uptodate(d_bh, 1) ;
3048 desc = (struct reiserfs_journal_desc *)(d_bh)->b_data ;
3049 memset(desc, 0, sizeof(struct reiserfs_journal_desc)) ;
3050 memcpy(desc->j_magic, JOURNAL_DESC_MAGIC, 8) ;
3051 desc->j_trans_id = cpu_to_le32(SB_JOURNAL(p_s_sb)->j_trans_id) ;
3052
3053 /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */
3054 c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3055 ((SB_JOURNAL(p_s_sb)->j_start + SB_JOURNAL(p_s_sb)->j_len + 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ;
3056 commit = (struct reiserfs_journal_commit *)c_bh->b_data ;
3057 memset(commit, 0, sizeof(struct reiserfs_journal_commit)) ;
3058 commit->j_trans_id = cpu_to_le32(SB_JOURNAL(p_s_sb)->j_trans_id) ;
3059 mark_buffer_uptodate(c_bh, 1) ;
3060
3061 /* init this journal list */
3062 atomic_set(&(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_older_commits_done), 0) ;
3063 SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_trans_id = SB_JOURNAL(p_s_sb)->j_trans_id ;
3064 SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_timestamp = SB_JOURNAL(p_s_sb)->j_trans_start_time ;
3065 SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_commit_bh = c_bh ;
3066 SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_start = SB_JOURNAL(p_s_sb)->j_start ;
3067 SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_len = SB_JOURNAL(p_s_sb)->j_len ;
3068 atomic_set(&(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_nonzerolen), SB_JOURNAL(p_s_sb)->j_len) ;
3069 atomic_set(&(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_commit_left), SB_JOURNAL(p_s_sb)->j_len + 2);
3070 SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_realblock = NULL ;
3071 atomic_set(&(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_commit_flushing), 1) ;
3072 atomic_set(&(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_flushing), 1) ;
3073
3074 /* which is faster, locking/unlocking at the start and end of the for
3075 ** or locking once per iteration around the insert_journal_hash?
3076 ** eitherway, we are write locking insert_journal_hash. The ENTIRE FOR
3077 ** LOOP MUST not cause schedule to occur.
3078 */
3079
3080 /* for each real block, add it to the journal list hash,
3081 ** copy into real block index array in the commit or desc block
3082 */
3083 for (i = 0, cn = SB_JOURNAL(p_s_sb)->j_first ; cn ; cn = cn->next, i++) {
3084 if (test_bit(BH_JDirty, &cn->bh->b_state) ) {
3085 jl_cn = get_cnode(p_s_sb) ;
3086 if (!jl_cn) {
3087 reiserfs_panic(p_s_sb, "journal-1676, get_cnode returned NULL\n") ;
3088 }
3089 if (i == 0) {
3090 SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_realblock = jl_cn ;
3091 }
3092 jl_cn->prev = last_cn ;
3093 jl_cn->next = NULL ;
3094 if (last_cn) {
3095 last_cn->next = jl_cn ;
3096 }
3097 last_cn = jl_cn ;
3098 /* make sure the block we are trying to log is not a block
3099 of journal or reserved area */
3100
3101 if (is_block_in_log_or_reserved_area(p_s_sb, cn->bh->b_blocknr)) {
3102 reiserfs_panic(p_s_sb, "journal-2332: Trying to log block %lu, which is a log block\n", cn->bh->b_blocknr) ;
3103 }
3104 jl_cn->blocknr = cn->bh->b_blocknr ;
3105 jl_cn->state = 0 ;
3106 jl_cn->dev = cn->bh->b_dev ;
3107 jl_cn->bh = cn->bh ;
3108 jl_cn->jlist = SB_JOURNAL_LIST(p_s_sb) + SB_JOURNAL_LIST_INDEX(p_s_sb) ;
3109 insert_journal_hash(SB_JOURNAL(p_s_sb)->j_list_hash_table, jl_cn) ;
3110 if (i < JOURNAL_TRANS_HALF) {
3111 desc->j_realblock[i] = cpu_to_le32(cn->bh->b_blocknr) ;
3112 } else {
3113 commit->j_realblock[i - JOURNAL_TRANS_HALF] = cpu_to_le32(cn->bh->b_blocknr) ;
3114 }
3115 } else {
3116 i-- ;
3117 }
3118 }
3119
3120 desc->j_len = cpu_to_le32(SB_JOURNAL(p_s_sb)->j_len) ;
3121 desc->j_mount_id = cpu_to_le32(SB_JOURNAL(p_s_sb)->j_mount_id) ;
3122 desc->j_trans_id = cpu_to_le32(SB_JOURNAL(p_s_sb)->j_trans_id) ;
3123 commit->j_len = cpu_to_le32(SB_JOURNAL(p_s_sb)->j_len) ;
3124
3125 /* special check in case all buffers in the journal were marked for not logging */
3126 if (SB_JOURNAL(p_s_sb)->j_len == 0) {
3127 brelse(d_bh) ;
3128 brelse(c_bh) ;
3129 unlock_journal(p_s_sb) ;
3130 reiserfs_warning(p_s_sb, "journal-2020: do_journal_end: BAD desc->j_len is ZERO\n") ;
3131 atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 0) ;
3132 wake_up(&(SB_JOURNAL(p_s_sb)->j_join_wait)) ;
3133 return 0 ;
3134 }
3135
3136 /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
3137 cur_write_start = SB_JOURNAL(p_s_sb)->j_start ;
3138 cur_blocks_left = SB_JOURNAL(p_s_sb)->j_len ;
3139 cn = SB_JOURNAL(p_s_sb)->j_first ;
3140 jindex = 1 ; /* start at one so we don't get the desc again */
3141 while(cur_blocks_left > 0) {
3142 /* copy all the real blocks into log area. dirty log blocks */
3143 if (test_bit(BH_JDirty, &cn->bh->b_state)) {
3144 struct buffer_head *tmp_bh ;
3145 tmp_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3146 ((cur_write_start + jindex) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ;
3147 mark_buffer_uptodate(tmp_bh, 1) ;
3148 memcpy(tmp_bh->b_data, cn->bh->b_data, cn->bh->b_size) ;
3149 jindex++ ;
3150 } else {
3151 /* JDirty cleared sometime during transaction. don't log this one */
3152 reiserfs_warning(p_s_sb, "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!\n") ;
3153 }
3154 cn = cn->next ;
3155 cur_blocks_left-- ;
3156 }
3157
3158 /* we are done with both the c_bh and d_bh, but
3159 ** c_bh must be written after all other commit blocks,
3160 ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
3161 */
3162
3163 /* now loop through and mark all buffers from this transaction as JDirty_wait
3164 ** clear the JDirty bit, clear BH_JNew too.
3165 ** if they weren't JDirty, they weren't logged, just relse them and move on
3166 */
3167 cn = SB_JOURNAL(p_s_sb)->j_first ;
3168 while(cn) {
3169 clear_bit(BH_JNew, &(cn->bh->b_state)) ;
3170 if (test_bit(BH_JDirty, &(cn->bh->b_state))) {
3171 set_bit(BH_JDirty_wait, &(cn->bh->b_state)) ;
3172 clear_bit(BH_JDirty, &(cn->bh->b_state)) ;
3173 } else {
3174 brelse(cn->bh) ;
3175 }
3176 next = cn->next ;
3177 free_cnode(p_s_sb, cn) ;
3178 cn = next ;
3179 }
3180
3181 /* unlock the journal list for committing and flushing */
3182 atomic_set(&(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_commit_flushing), 0) ;
3183 atomic_set(&(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_flushing), 0) ;
3184
3185 orig_jindex = SB_JOURNAL_LIST_INDEX(p_s_sb) ;
3186 jindex = (SB_JOURNAL_LIST_INDEX(p_s_sb) + 1) % JOURNAL_LIST_COUNT ;
3187 SB_JOURNAL_LIST_INDEX(p_s_sb) = jindex ;
3188
3189 /* write any buffers that must hit disk before this commit is done */
3190 fsync_buffers_list(&(SB_JOURNAL(p_s_sb)->j_dirty_buffers)) ;
3191
3192 /* honor the flush and async wishes from the caller */
3193 if (flush) {
3194
3195 flush_commit_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + orig_jindex, 1) ;
3196 flush_journal_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + orig_jindex , 1) ;
3197 } else if (commit_now) {
3198 if (wait_on_commit) {
3199 flush_commit_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + orig_jindex, 1) ;
3200 } else {
3201 commit_flush_async(p_s_sb, orig_jindex) ;
3202 }
3203 }
3204
3205 /* reset journal values for the next transaction */
3206 old_start = SB_JOURNAL(p_s_sb)->j_start ;
3207 SB_JOURNAL(p_s_sb)->j_start = (SB_JOURNAL(p_s_sb)->j_start + SB_JOURNAL(p_s_sb)->j_len + 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb);
3208 atomic_set(&(SB_JOURNAL(p_s_sb)->j_wcount), 0) ;
3209 SB_JOURNAL(p_s_sb)->j_bcount = 0 ;
3210 SB_JOURNAL(p_s_sb)->j_last = NULL ;
3211 SB_JOURNAL(p_s_sb)->j_first = NULL ;
3212 SB_JOURNAL(p_s_sb)->j_len = 0 ;
3213 SB_JOURNAL(p_s_sb)->j_trans_start_time = 0 ;
3214 SB_JOURNAL(p_s_sb)->j_trans_id++ ;
3215 SB_JOURNAL(p_s_sb)->j_must_wait = 0 ;
3216 SB_JOURNAL(p_s_sb)->j_len_alloc = 0 ;
3217 SB_JOURNAL(p_s_sb)->j_next_full_flush = 0 ;
3218 SB_JOURNAL(p_s_sb)->j_next_async_flush = 0 ;
3219 init_journal_hash(p_s_sb) ;
3220
3221 /* if the next transaction has any chance of wrapping, flush
3222 ** transactions that might get overwritten. If any journal lists are very
3223 ** old flush them as well.
3224 */
3225 for (i = 0 ; i < JOURNAL_LIST_COUNT ; i++) {
3226 jindex = i ;
3227 if (SB_JOURNAL_LIST(p_s_sb)[jindex].j_len > 0 && SB_JOURNAL(p_s_sb)->j_start <= SB_JOURNAL_LIST(p_s_sb)[jindex].j_start) {
3228 if ((SB_JOURNAL(p_s_sb)->j_start + SB_JOURNAL_TRANS_MAX(p_s_sb) + 1) >= SB_JOURNAL_LIST(p_s_sb)[jindex].j_start) {
3229 flush_journal_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + jindex, 1) ;
3230 }
3231 } else if (SB_JOURNAL_LIST(p_s_sb)[jindex].j_len > 0 &&
3232 (SB_JOURNAL(p_s_sb)->j_start + SB_JOURNAL_TRANS_MAX(p_s_sb) + 1) > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
3233 if (((SB_JOURNAL(p_s_sb)->j_start + SB_JOURNAL_TRANS_MAX(p_s_sb) + 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >=
3234 SB_JOURNAL_LIST(p_s_sb)[jindex].j_start) {
3235 flush_journal_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + jindex, 1 ) ;
3236 }
3237 }
3238 /* this check should always be run, to send old lists to disk */
3239 if (SB_JOURNAL_LIST(p_s_sb)[jindex].j_len > 0 &&
3240 SB_JOURNAL_LIST(p_s_sb)[jindex].j_timestamp <
3241 (CURRENT_TIME - (SB_JOURNAL_MAX_TRANS_AGE(p_s_sb) * 4))) {
3242 flush_journal_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + jindex, 1 ) ;
3243 }
3244 }
3245
3246 /* if the next journal_list is still in use, flush it */
3247 if (SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_len != 0) {
3248 flush_journal_list(p_s_sb, SB_JOURNAL_LIST(p_s_sb) + SB_JOURNAL_LIST_INDEX(p_s_sb), 1) ;
3249 }
3250
3251 /* we don't want anyone flushing the new transaction's list */
3252 atomic_set(&(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_commit_flushing), 1) ;
3253 atomic_set(&(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_flushing), 1) ;
3254 SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_list_bitmap = get_list_bitmap(p_s_sb, SB_JOURNAL_LIST(p_s_sb) +
3255 SB_JOURNAL_LIST_INDEX(p_s_sb)) ;
3256
3257 if (!(SB_JOURNAL_LIST(p_s_sb)[SB_JOURNAL_LIST_INDEX(p_s_sb)].j_list_bitmap)) {
3258 reiserfs_panic(p_s_sb, "journal-1996: do_journal_end, could not get a list bitmap\n") ;
3259 }
3260 unlock_journal(p_s_sb) ;
3261 atomic_set(&(SB_JOURNAL(p_s_sb)->j_jlock), 0) ;
3262 /* wake up any body waiting to join. */
3263 wake_up(&(SB_JOURNAL(p_s_sb)->j_join_wait)) ;
3264 return 0 ;
3265 }
3266
3267
3268
Cache object: 65f8838c615999d1b736fa41bcdb2695
|