xref: /openbmc/linux/fs/reiserfs/journal.c (revision 5bd8e16d)
1 /*
2 ** Write ahead logging implementation copyright Chris Mason 2000
3 **
4 ** The background commits make this code very interrelated, and
5 ** overly complex.  I need to rethink things a bit....The major players:
6 **
7 ** journal_begin -- call with the number of blocks you expect to log.
8 **                  If the current transaction is too
9 ** 		    old, it will block until the current transaction is
10 ** 		    finished, and then start a new one.
11 **		    Usually, your transaction will get joined in with
12 **                  previous ones for speed.
13 **
14 ** journal_join  -- same as journal_begin, but won't block on the current
15 **                  transaction regardless of age.  Don't ever call
16 **                  this.  Ever.  There are only two places it should be
17 **                  called from, and they are both inside this file.
18 **
19 ** journal_mark_dirty -- adds blocks into this transaction.  clears any flags
20 **                       that might make them get sent to disk
21 **                       and then marks them BH_JDirty.  Puts the buffer head
22 **                       into the current transaction hash.
23 **
24 ** journal_end -- if the current transaction is batchable, it does nothing
25 **                   otherwise, it could do an async/synchronous commit, or
26 **                   a full flush of all log and real blocks in the
27 **                   transaction.
28 **
29 ** flush_old_commits -- if the current transaction is too old, it is ended and
30 **                      commit blocks are sent to disk.  Forces commit blocks
31 **                      to disk for all backgrounded commits that have been
32 **                      around too long.
33 **		     -- Note, if you call this as an immediate flush from
34 **		        from within kupdate, it will ignore the immediate flag
35 */
36 
37 #include <linux/time.h>
38 #include <linux/semaphore.h>
39 #include <linux/vmalloc.h>
40 #include "reiserfs.h"
41 #include <linux/kernel.h>
42 #include <linux/errno.h>
43 #include <linux/fcntl.h>
44 #include <linux/stat.h>
45 #include <linux/string.h>
46 #include <linux/buffer_head.h>
47 #include <linux/workqueue.h>
48 #include <linux/writeback.h>
49 #include <linux/blkdev.h>
50 #include <linux/backing-dev.h>
51 #include <linux/uaccess.h>
52 #include <linux/slab.h>
53 
54 
55 /* gets a struct reiserfs_journal_list * from a list head */
56 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
57                                j_list))
58 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
59                                j_working_list))
60 
61 /* the number of mounted filesystems.  This is used to decide when to
62 ** start and kill the commit workqueue
63 */
64 static int reiserfs_mounted_fs_count;
65 
66 static struct workqueue_struct *commit_wq;
67 
68 #define JOURNAL_TRANS_HALF 1018	/* must be correct to keep the desc and commit
69 				   structs at 4k */
70 #define BUFNR 64		/*read ahead */
71 
72 /* cnode stat bits.  Move these into reiserfs_fs.h */
73 
74 #define BLOCK_FREED 2		/* this block was freed, and can't be written.  */
75 #define BLOCK_FREED_HOLDER 3	/* this block was freed during this transaction, and can't be written */
76 
77 #define BLOCK_NEEDS_FLUSH 4	/* used in flush_journal_list */
78 #define BLOCK_DIRTIED 5
79 
80 /* journal list state bits */
81 #define LIST_TOUCHED 1
82 #define LIST_DIRTY   2
83 #define LIST_COMMIT_PENDING  4	/* someone will commit this list */
84 
85 /* flags for do_journal_end */
86 #define FLUSH_ALL   1		/* flush commit and real blocks */
87 #define COMMIT_NOW  2		/* end and commit this transaction */
88 #define WAIT        4		/* wait for the log blocks to hit the disk */
89 
90 static int do_journal_end(struct reiserfs_transaction_handle *,
91 			  struct super_block *, unsigned long nblocks,
92 			  int flags);
93 static int flush_journal_list(struct super_block *s,
94 			      struct reiserfs_journal_list *jl, int flushall);
95 static int flush_commit_list(struct super_block *s,
96 			     struct reiserfs_journal_list *jl, int flushall);
97 static int can_dirty(struct reiserfs_journal_cnode *cn);
98 static int journal_join(struct reiserfs_transaction_handle *th,
99 			struct super_block *sb, unsigned long nblocks);
100 static void release_journal_dev(struct super_block *super,
101 			       struct reiserfs_journal *journal);
102 static int dirty_one_transaction(struct super_block *s,
103 				 struct reiserfs_journal_list *jl);
104 static void flush_async_commits(struct work_struct *work);
105 static void queue_log_writer(struct super_block *s);
106 
107 /* values for join in do_journal_begin_r */
108 enum {
109 	JBEGIN_REG = 0,		/* regular journal begin */
110 	JBEGIN_JOIN = 1,	/* join the running transaction if at all possible */
111 	JBEGIN_ABORT = 2,	/* called from cleanup code, ignores aborted flag */
112 };
113 
114 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
115 			      struct super_block *sb,
116 			      unsigned long nblocks, int join);
117 
118 static void init_journal_hash(struct super_block *sb)
119 {
120 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
121 	memset(journal->j_hash_table, 0,
122 	       JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
123 }
124 
125 /*
126 ** clears BH_Dirty and sticks the buffer on the clean list.  Called because I can't allow refile_buffer to
127 ** make schedule happen after I've freed a block.  Look at remove_from_transaction and journal_mark_freed for
128 ** more details.
129 */
130 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
131 {
132 	if (bh) {
133 		clear_buffer_dirty(bh);
134 		clear_buffer_journal_test(bh);
135 	}
136 	return 0;
137 }
138 
139 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
140 							 *sb)
141 {
142 	struct reiserfs_bitmap_node *bn;
143 	static int id;
144 
145 	bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
146 	if (!bn) {
147 		return NULL;
148 	}
149 	bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
150 	if (!bn->data) {
151 		kfree(bn);
152 		return NULL;
153 	}
154 	bn->id = id++;
155 	INIT_LIST_HEAD(&bn->list);
156 	return bn;
157 }
158 
159 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
160 {
161 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
162 	struct reiserfs_bitmap_node *bn = NULL;
163 	struct list_head *entry = journal->j_bitmap_nodes.next;
164 
165 	journal->j_used_bitmap_nodes++;
166       repeat:
167 
168 	if (entry != &journal->j_bitmap_nodes) {
169 		bn = list_entry(entry, struct reiserfs_bitmap_node, list);
170 		list_del(entry);
171 		memset(bn->data, 0, sb->s_blocksize);
172 		journal->j_free_bitmap_nodes--;
173 		return bn;
174 	}
175 	bn = allocate_bitmap_node(sb);
176 	if (!bn) {
177 		yield();
178 		goto repeat;
179 	}
180 	return bn;
181 }
182 static inline void free_bitmap_node(struct super_block *sb,
183 				    struct reiserfs_bitmap_node *bn)
184 {
185 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
186 	journal->j_used_bitmap_nodes--;
187 	if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
188 		kfree(bn->data);
189 		kfree(bn);
190 	} else {
191 		list_add(&bn->list, &journal->j_bitmap_nodes);
192 		journal->j_free_bitmap_nodes++;
193 	}
194 }
195 
196 static void allocate_bitmap_nodes(struct super_block *sb)
197 {
198 	int i;
199 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
200 	struct reiserfs_bitmap_node *bn = NULL;
201 	for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
202 		bn = allocate_bitmap_node(sb);
203 		if (bn) {
204 			list_add(&bn->list, &journal->j_bitmap_nodes);
205 			journal->j_free_bitmap_nodes++;
206 		} else {
207 			break;	/* this is ok, we'll try again when more are needed */
208 		}
209 	}
210 }
211 
212 static int set_bit_in_list_bitmap(struct super_block *sb,
213 				  b_blocknr_t block,
214 				  struct reiserfs_list_bitmap *jb)
215 {
216 	unsigned int bmap_nr = block / (sb->s_blocksize << 3);
217 	unsigned int bit_nr = block % (sb->s_blocksize << 3);
218 
219 	if (!jb->bitmaps[bmap_nr]) {
220 		jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
221 	}
222 	set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
223 	return 0;
224 }
225 
226 static void cleanup_bitmap_list(struct super_block *sb,
227 				struct reiserfs_list_bitmap *jb)
228 {
229 	int i;
230 	if (jb->bitmaps == NULL)
231 		return;
232 
233 	for (i = 0; i < reiserfs_bmap_count(sb); i++) {
234 		if (jb->bitmaps[i]) {
235 			free_bitmap_node(sb, jb->bitmaps[i]);
236 			jb->bitmaps[i] = NULL;
237 		}
238 	}
239 }
240 
241 /*
242 ** only call this on FS unmount.
243 */
244 static int free_list_bitmaps(struct super_block *sb,
245 			     struct reiserfs_list_bitmap *jb_array)
246 {
247 	int i;
248 	struct reiserfs_list_bitmap *jb;
249 	for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
250 		jb = jb_array + i;
251 		jb->journal_list = NULL;
252 		cleanup_bitmap_list(sb, jb);
253 		vfree(jb->bitmaps);
254 		jb->bitmaps = NULL;
255 	}
256 	return 0;
257 }
258 
259 static int free_bitmap_nodes(struct super_block *sb)
260 {
261 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
262 	struct list_head *next = journal->j_bitmap_nodes.next;
263 	struct reiserfs_bitmap_node *bn;
264 
265 	while (next != &journal->j_bitmap_nodes) {
266 		bn = list_entry(next, struct reiserfs_bitmap_node, list);
267 		list_del(next);
268 		kfree(bn->data);
269 		kfree(bn);
270 		next = journal->j_bitmap_nodes.next;
271 		journal->j_free_bitmap_nodes--;
272 	}
273 
274 	return 0;
275 }
276 
277 /*
278 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
279 ** jb_array is the array to be filled in.
280 */
281 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
282 				   struct reiserfs_list_bitmap *jb_array,
283 				   unsigned int bmap_nr)
284 {
285 	int i;
286 	int failed = 0;
287 	struct reiserfs_list_bitmap *jb;
288 	int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
289 
290 	for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
291 		jb = jb_array + i;
292 		jb->journal_list = NULL;
293 		jb->bitmaps = vzalloc(mem);
294 		if (!jb->bitmaps) {
295 			reiserfs_warning(sb, "clm-2000", "unable to "
296 					 "allocate bitmaps for journal lists");
297 			failed = 1;
298 			break;
299 		}
300 	}
301 	if (failed) {
302 		free_list_bitmaps(sb, jb_array);
303 		return -1;
304 	}
305 	return 0;
306 }
307 
308 /*
309 ** find an available list bitmap.  If you can't find one, flush a commit list
310 ** and try again
311 */
312 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
313 						    struct reiserfs_journal_list
314 						    *jl)
315 {
316 	int i, j;
317 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
318 	struct reiserfs_list_bitmap *jb = NULL;
319 
320 	for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
321 		i = journal->j_list_bitmap_index;
322 		journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
323 		jb = journal->j_list_bitmap + i;
324 		if (journal->j_list_bitmap[i].journal_list) {
325 			flush_commit_list(sb,
326 					  journal->j_list_bitmap[i].
327 					  journal_list, 1);
328 			if (!journal->j_list_bitmap[i].journal_list) {
329 				break;
330 			}
331 		} else {
332 			break;
333 		}
334 	}
335 	if (jb->journal_list) {	/* double check to make sure if flushed correctly */
336 		return NULL;
337 	}
338 	jb->journal_list = jl;
339 	return jb;
340 }
341 
342 /*
343 ** allocates a new chunk of X nodes, and links them all together as a list.
344 ** Uses the cnode->next and cnode->prev pointers
345 ** returns NULL on failure
346 */
347 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
348 {
349 	struct reiserfs_journal_cnode *head;
350 	int i;
351 	if (num_cnodes <= 0) {
352 		return NULL;
353 	}
354 	head = vzalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
355 	if (!head) {
356 		return NULL;
357 	}
358 	head[0].prev = NULL;
359 	head[0].next = head + 1;
360 	for (i = 1; i < num_cnodes; i++) {
361 		head[i].prev = head + (i - 1);
362 		head[i].next = head + (i + 1);	/* if last one, overwrite it after the if */
363 	}
364 	head[num_cnodes - 1].next = NULL;
365 	return head;
366 }
367 
368 /*
369 ** pulls a cnode off the free list, or returns NULL on failure
370 */
371 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
372 {
373 	struct reiserfs_journal_cnode *cn;
374 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
375 
376 	reiserfs_check_lock_depth(sb, "get_cnode");
377 
378 	if (journal->j_cnode_free <= 0) {
379 		return NULL;
380 	}
381 	journal->j_cnode_used++;
382 	journal->j_cnode_free--;
383 	cn = journal->j_cnode_free_list;
384 	if (!cn) {
385 		return cn;
386 	}
387 	if (cn->next) {
388 		cn->next->prev = NULL;
389 	}
390 	journal->j_cnode_free_list = cn->next;
391 	memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
392 	return cn;
393 }
394 
395 /*
396 ** returns a cnode to the free list
397 */
398 static void free_cnode(struct super_block *sb,
399 		       struct reiserfs_journal_cnode *cn)
400 {
401 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
402 
403 	reiserfs_check_lock_depth(sb, "free_cnode");
404 
405 	journal->j_cnode_used--;
406 	journal->j_cnode_free++;
407 	/* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
408 	cn->next = journal->j_cnode_free_list;
409 	if (journal->j_cnode_free_list) {
410 		journal->j_cnode_free_list->prev = cn;
411 	}
412 	cn->prev = NULL;	/* not needed with the memset, but I might kill the memset, and forget to do this */
413 	journal->j_cnode_free_list = cn;
414 }
415 
416 static void clear_prepared_bits(struct buffer_head *bh)
417 {
418 	clear_buffer_journal_prepared(bh);
419 	clear_buffer_journal_restore_dirty(bh);
420 }
421 
422 /* return a cnode with same dev, block number and size in table, or null if not found */
423 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
424 								  super_block
425 								  *sb,
426 								  struct
427 								  reiserfs_journal_cnode
428 								  **table,
429 								  long bl)
430 {
431 	struct reiserfs_journal_cnode *cn;
432 	cn = journal_hash(table, sb, bl);
433 	while (cn) {
434 		if (cn->blocknr == bl && cn->sb == sb)
435 			return cn;
436 		cn = cn->hnext;
437 	}
438 	return (struct reiserfs_journal_cnode *)0;
439 }
440 
441 /*
442 ** this actually means 'can this block be reallocated yet?'.  If you set search_all, a block can only be allocated
443 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
444 ** being overwritten by a replay after crashing.
445 **
446 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction.  Since deleting
447 ** a block removes it from the current transaction, this case should never happen.  If you don't set search_all, make
448 ** sure you never write the block without logging it.
449 **
450 ** next_zero_bit is a suggestion about the next block to try for find_forward.
451 ** when bl is rejected because it is set in a journal list bitmap, we search
452 ** for the next zero bit in the bitmap that rejected bl.  Then, we return that
453 ** through next_zero_bit for find_forward to try.
454 **
455 ** Just because we return something in next_zero_bit does not mean we won't
456 ** reject it on the next call to reiserfs_in_journal
457 **
458 */
459 int reiserfs_in_journal(struct super_block *sb,
460 			unsigned int bmap_nr, int bit_nr, int search_all,
461 			b_blocknr_t * next_zero_bit)
462 {
463 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
464 	struct reiserfs_journal_cnode *cn;
465 	struct reiserfs_list_bitmap *jb;
466 	int i;
467 	unsigned long bl;
468 
469 	*next_zero_bit = 0;	/* always start this at zero. */
470 
471 	PROC_INFO_INC(sb, journal.in_journal);
472 	/* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
473 	 ** if we crash before the transaction that freed it commits,  this transaction won't
474 	 ** have committed either, and the block will never be written
475 	 */
476 	if (search_all) {
477 		for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
478 			PROC_INFO_INC(sb, journal.in_journal_bitmap);
479 			jb = journal->j_list_bitmap + i;
480 			if (jb->journal_list && jb->bitmaps[bmap_nr] &&
481 			    test_bit(bit_nr,
482 				     (unsigned long *)jb->bitmaps[bmap_nr]->
483 				     data)) {
484 				*next_zero_bit =
485 				    find_next_zero_bit((unsigned long *)
486 						       (jb->bitmaps[bmap_nr]->
487 							data),
488 						       sb->s_blocksize << 3,
489 						       bit_nr + 1);
490 				return 1;
491 			}
492 		}
493 	}
494 
495 	bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
496 	/* is it in any old transactions? */
497 	if (search_all
498 	    && (cn =
499 		get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
500 		return 1;
501 	}
502 
503 	/* is it in the current transaction.  This should never happen */
504 	if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
505 		BUG();
506 		return 1;
507 	}
508 
509 	PROC_INFO_INC(sb, journal.in_journal_reusable);
510 	/* safe for reuse */
511 	return 0;
512 }
513 
514 /* insert cn into table
515 */
516 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
517 				       struct reiserfs_journal_cnode *cn)
518 {
519 	struct reiserfs_journal_cnode *cn_orig;
520 
521 	cn_orig = journal_hash(table, cn->sb, cn->blocknr);
522 	cn->hnext = cn_orig;
523 	cn->hprev = NULL;
524 	if (cn_orig) {
525 		cn_orig->hprev = cn;
526 	}
527 	journal_hash(table, cn->sb, cn->blocknr) = cn;
528 }
529 
530 /* lock the current transaction */
531 static inline void lock_journal(struct super_block *sb)
532 {
533 	PROC_INFO_INC(sb, journal.lock_journal);
534 
535 	reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
536 }
537 
538 /* unlock the current transaction */
539 static inline void unlock_journal(struct super_block *sb)
540 {
541 	mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
542 }
543 
544 static inline void get_journal_list(struct reiserfs_journal_list *jl)
545 {
546 	jl->j_refcount++;
547 }
548 
549 static inline void put_journal_list(struct super_block *s,
550 				    struct reiserfs_journal_list *jl)
551 {
552 	if (jl->j_refcount < 1) {
553 		reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
554 			       jl->j_trans_id, jl->j_refcount);
555 	}
556 	if (--jl->j_refcount == 0)
557 		kfree(jl);
558 }
559 
560 /*
561 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
562 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
563 ** transaction.
564 */
565 static void cleanup_freed_for_journal_list(struct super_block *sb,
566 					   struct reiserfs_journal_list *jl)
567 {
568 
569 	struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
570 	if (jb) {
571 		cleanup_bitmap_list(sb, jb);
572 	}
573 	jl->j_list_bitmap->journal_list = NULL;
574 	jl->j_list_bitmap = NULL;
575 }
576 
577 static int journal_list_still_alive(struct super_block *s,
578 				    unsigned int trans_id)
579 {
580 	struct reiserfs_journal *journal = SB_JOURNAL(s);
581 	struct list_head *entry = &journal->j_journal_list;
582 	struct reiserfs_journal_list *jl;
583 
584 	if (!list_empty(entry)) {
585 		jl = JOURNAL_LIST_ENTRY(entry->next);
586 		if (jl->j_trans_id <= trans_id) {
587 			return 1;
588 		}
589 	}
590 	return 0;
591 }
592 
593 /*
594  * If page->mapping was null, we failed to truncate this page for
595  * some reason.  Most likely because it was truncated after being
596  * logged via data=journal.
597  *
598  * This does a check to see if the buffer belongs to one of these
599  * lost pages before doing the final put_bh.  If page->mapping was
600  * null, it tries to free buffers on the page, which should make the
601  * final page_cache_release drop the page from the lru.
602  */
603 static void release_buffer_page(struct buffer_head *bh)
604 {
605 	struct page *page = bh->b_page;
606 	if (!page->mapping && trylock_page(page)) {
607 		page_cache_get(page);
608 		put_bh(bh);
609 		if (!page->mapping)
610 			try_to_free_buffers(page);
611 		unlock_page(page);
612 		page_cache_release(page);
613 	} else {
614 		put_bh(bh);
615 	}
616 }
617 
618 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
619 {
620 	char b[BDEVNAME_SIZE];
621 
622 	if (buffer_journaled(bh)) {
623 		reiserfs_warning(NULL, "clm-2084",
624 				 "pinned buffer %lu:%s sent to disk",
625 				 bh->b_blocknr, bdevname(bh->b_bdev, b));
626 	}
627 	if (uptodate)
628 		set_buffer_uptodate(bh);
629 	else
630 		clear_buffer_uptodate(bh);
631 
632 	unlock_buffer(bh);
633 	release_buffer_page(bh);
634 }
635 
636 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
637 {
638 	if (uptodate)
639 		set_buffer_uptodate(bh);
640 	else
641 		clear_buffer_uptodate(bh);
642 	unlock_buffer(bh);
643 	put_bh(bh);
644 }
645 
646 static void submit_logged_buffer(struct buffer_head *bh)
647 {
648 	get_bh(bh);
649 	bh->b_end_io = reiserfs_end_buffer_io_sync;
650 	clear_buffer_journal_new(bh);
651 	clear_buffer_dirty(bh);
652 	if (!test_clear_buffer_journal_test(bh))
653 		BUG();
654 	if (!buffer_uptodate(bh))
655 		BUG();
656 	submit_bh(WRITE, bh);
657 }
658 
659 static void submit_ordered_buffer(struct buffer_head *bh)
660 {
661 	get_bh(bh);
662 	bh->b_end_io = reiserfs_end_ordered_io;
663 	clear_buffer_dirty(bh);
664 	if (!buffer_uptodate(bh))
665 		BUG();
666 	submit_bh(WRITE, bh);
667 }
668 
669 #define CHUNK_SIZE 32
670 struct buffer_chunk {
671 	struct buffer_head *bh[CHUNK_SIZE];
672 	int nr;
673 };
674 
675 static void write_chunk(struct buffer_chunk *chunk)
676 {
677 	int i;
678 	for (i = 0; i < chunk->nr; i++) {
679 		submit_logged_buffer(chunk->bh[i]);
680 	}
681 	chunk->nr = 0;
682 }
683 
684 static void write_ordered_chunk(struct buffer_chunk *chunk)
685 {
686 	int i;
687 	for (i = 0; i < chunk->nr; i++) {
688 		submit_ordered_buffer(chunk->bh[i]);
689 	}
690 	chunk->nr = 0;
691 }
692 
693 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
694 			spinlock_t * lock, void (fn) (struct buffer_chunk *))
695 {
696 	int ret = 0;
697 	BUG_ON(chunk->nr >= CHUNK_SIZE);
698 	chunk->bh[chunk->nr++] = bh;
699 	if (chunk->nr >= CHUNK_SIZE) {
700 		ret = 1;
701 		if (lock)
702 			spin_unlock(lock);
703 		fn(chunk);
704 		if (lock)
705 			spin_lock(lock);
706 	}
707 	return ret;
708 }
709 
710 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
711 static struct reiserfs_jh *alloc_jh(void)
712 {
713 	struct reiserfs_jh *jh;
714 	while (1) {
715 		jh = kmalloc(sizeof(*jh), GFP_NOFS);
716 		if (jh) {
717 			atomic_inc(&nr_reiserfs_jh);
718 			return jh;
719 		}
720 		yield();
721 	}
722 }
723 
724 /*
725  * we want to free the jh when the buffer has been written
726  * and waited on
727  */
728 void reiserfs_free_jh(struct buffer_head *bh)
729 {
730 	struct reiserfs_jh *jh;
731 
732 	jh = bh->b_private;
733 	if (jh) {
734 		bh->b_private = NULL;
735 		jh->bh = NULL;
736 		list_del_init(&jh->list);
737 		kfree(jh);
738 		if (atomic_read(&nr_reiserfs_jh) <= 0)
739 			BUG();
740 		atomic_dec(&nr_reiserfs_jh);
741 		put_bh(bh);
742 	}
743 }
744 
745 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
746 			   int tail)
747 {
748 	struct reiserfs_jh *jh;
749 
750 	if (bh->b_private) {
751 		spin_lock(&j->j_dirty_buffers_lock);
752 		if (!bh->b_private) {
753 			spin_unlock(&j->j_dirty_buffers_lock);
754 			goto no_jh;
755 		}
756 		jh = bh->b_private;
757 		list_del_init(&jh->list);
758 	} else {
759 	      no_jh:
760 		get_bh(bh);
761 		jh = alloc_jh();
762 		spin_lock(&j->j_dirty_buffers_lock);
763 		/* buffer must be locked for __add_jh, should be able to have
764 		 * two adds at the same time
765 		 */
766 		BUG_ON(bh->b_private);
767 		jh->bh = bh;
768 		bh->b_private = jh;
769 	}
770 	jh->jl = j->j_current_jl;
771 	if (tail)
772 		list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
773 	else {
774 		list_add_tail(&jh->list, &jh->jl->j_bh_list);
775 	}
776 	spin_unlock(&j->j_dirty_buffers_lock);
777 	return 0;
778 }
779 
780 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
781 {
782 	return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
783 }
784 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
785 {
786 	return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
787 }
788 
789 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
790 static int write_ordered_buffers(spinlock_t * lock,
791 				 struct reiserfs_journal *j,
792 				 struct reiserfs_journal_list *jl,
793 				 struct list_head *list)
794 {
795 	struct buffer_head *bh;
796 	struct reiserfs_jh *jh;
797 	int ret = j->j_errno;
798 	struct buffer_chunk chunk;
799 	struct list_head tmp;
800 	INIT_LIST_HEAD(&tmp);
801 
802 	chunk.nr = 0;
803 	spin_lock(lock);
804 	while (!list_empty(list)) {
805 		jh = JH_ENTRY(list->next);
806 		bh = jh->bh;
807 		get_bh(bh);
808 		if (!trylock_buffer(bh)) {
809 			if (!buffer_dirty(bh)) {
810 				list_move(&jh->list, &tmp);
811 				goto loop_next;
812 			}
813 			spin_unlock(lock);
814 			if (chunk.nr)
815 				write_ordered_chunk(&chunk);
816 			wait_on_buffer(bh);
817 			cond_resched();
818 			spin_lock(lock);
819 			goto loop_next;
820 		}
821 		/* in theory, dirty non-uptodate buffers should never get here,
822 		 * but the upper layer io error paths still have a few quirks.
823 		 * Handle them here as gracefully as we can
824 		 */
825 		if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
826 			clear_buffer_dirty(bh);
827 			ret = -EIO;
828 		}
829 		if (buffer_dirty(bh)) {
830 			list_move(&jh->list, &tmp);
831 			add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
832 		} else {
833 			reiserfs_free_jh(bh);
834 			unlock_buffer(bh);
835 		}
836 	      loop_next:
837 		put_bh(bh);
838 		cond_resched_lock(lock);
839 	}
840 	if (chunk.nr) {
841 		spin_unlock(lock);
842 		write_ordered_chunk(&chunk);
843 		spin_lock(lock);
844 	}
845 	while (!list_empty(&tmp)) {
846 		jh = JH_ENTRY(tmp.prev);
847 		bh = jh->bh;
848 		get_bh(bh);
849 		reiserfs_free_jh(bh);
850 
851 		if (buffer_locked(bh)) {
852 			spin_unlock(lock);
853 			wait_on_buffer(bh);
854 			spin_lock(lock);
855 		}
856 		if (!buffer_uptodate(bh)) {
857 			ret = -EIO;
858 		}
859 		/* ugly interaction with invalidatepage here.
860 		 * reiserfs_invalidate_page will pin any buffer that has a valid
861 		 * journal head from an older transaction.  If someone else sets
862 		 * our buffer dirty after we write it in the first loop, and
863 		 * then someone truncates the page away, nobody will ever write
864 		 * the buffer. We're safe if we write the page one last time
865 		 * after freeing the journal header.
866 		 */
867 		if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
868 			spin_unlock(lock);
869 			ll_rw_block(WRITE, 1, &bh);
870 			spin_lock(lock);
871 		}
872 		put_bh(bh);
873 		cond_resched_lock(lock);
874 	}
875 	spin_unlock(lock);
876 	return ret;
877 }
878 
879 static int flush_older_commits(struct super_block *s,
880 			       struct reiserfs_journal_list *jl)
881 {
882 	struct reiserfs_journal *journal = SB_JOURNAL(s);
883 	struct reiserfs_journal_list *other_jl;
884 	struct reiserfs_journal_list *first_jl;
885 	struct list_head *entry;
886 	unsigned int trans_id = jl->j_trans_id;
887 	unsigned int other_trans_id;
888 	unsigned int first_trans_id;
889 
890       find_first:
891 	/*
892 	 * first we walk backwards to find the oldest uncommitted transation
893 	 */
894 	first_jl = jl;
895 	entry = jl->j_list.prev;
896 	while (1) {
897 		other_jl = JOURNAL_LIST_ENTRY(entry);
898 		if (entry == &journal->j_journal_list ||
899 		    atomic_read(&other_jl->j_older_commits_done))
900 			break;
901 
902 		first_jl = other_jl;
903 		entry = other_jl->j_list.prev;
904 	}
905 
906 	/* if we didn't find any older uncommitted transactions, return now */
907 	if (first_jl == jl) {
908 		return 0;
909 	}
910 
911 	first_trans_id = first_jl->j_trans_id;
912 
913 	entry = &first_jl->j_list;
914 	while (1) {
915 		other_jl = JOURNAL_LIST_ENTRY(entry);
916 		other_trans_id = other_jl->j_trans_id;
917 
918 		if (other_trans_id < trans_id) {
919 			if (atomic_read(&other_jl->j_commit_left) != 0) {
920 				flush_commit_list(s, other_jl, 0);
921 
922 				/* list we were called with is gone, return */
923 				if (!journal_list_still_alive(s, trans_id))
924 					return 1;
925 
926 				/* the one we just flushed is gone, this means all
927 				 * older lists are also gone, so first_jl is no longer
928 				 * valid either.  Go back to the beginning.
929 				 */
930 				if (!journal_list_still_alive
931 				    (s, other_trans_id)) {
932 					goto find_first;
933 				}
934 			}
935 			entry = entry->next;
936 			if (entry == &journal->j_journal_list)
937 				return 0;
938 		} else {
939 			return 0;
940 		}
941 	}
942 	return 0;
943 }
944 
945 static int reiserfs_async_progress_wait(struct super_block *s)
946 {
947 	struct reiserfs_journal *j = SB_JOURNAL(s);
948 
949 	if (atomic_read(&j->j_async_throttle)) {
950 		int depth;
951 
952 		depth = reiserfs_write_unlock_nested(s);
953 		congestion_wait(BLK_RW_ASYNC, HZ / 10);
954 		reiserfs_write_lock_nested(s, depth);
955 	}
956 
957 	return 0;
958 }
959 
960 /*
961 ** if this journal list still has commit blocks unflushed, send them to disk.
962 **
963 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
964 ** Before the commit block can by written, every other log block must be safely on disk
965 **
966 */
967 static int flush_commit_list(struct super_block *s,
968 			     struct reiserfs_journal_list *jl, int flushall)
969 {
970 	int i;
971 	b_blocknr_t bn;
972 	struct buffer_head *tbh = NULL;
973 	unsigned int trans_id = jl->j_trans_id;
974 	struct reiserfs_journal *journal = SB_JOURNAL(s);
975 	int retval = 0;
976 	int write_len;
977 	int depth;
978 
979 	reiserfs_check_lock_depth(s, "flush_commit_list");
980 
981 	if (atomic_read(&jl->j_older_commits_done)) {
982 		return 0;
983 	}
984 
985 	/* before we can put our commit blocks on disk, we have to make sure everyone older than
986 	 ** us is on disk too
987 	 */
988 	BUG_ON(jl->j_len <= 0);
989 	BUG_ON(trans_id == journal->j_trans_id);
990 
991 	get_journal_list(jl);
992 	if (flushall) {
993 		if (flush_older_commits(s, jl) == 1) {
994 			/* list disappeared during flush_older_commits.  return */
995 			goto put_jl;
996 		}
997 	}
998 
999 	/* make sure nobody is trying to flush this one at the same time */
1000 	reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1001 
1002 	if (!journal_list_still_alive(s, trans_id)) {
1003 		mutex_unlock(&jl->j_commit_mutex);
1004 		goto put_jl;
1005 	}
1006 	BUG_ON(jl->j_trans_id == 0);
1007 
1008 	/* this commit is done, exit */
1009 	if (atomic_read(&(jl->j_commit_left)) <= 0) {
1010 		if (flushall) {
1011 			atomic_set(&(jl->j_older_commits_done), 1);
1012 		}
1013 		mutex_unlock(&jl->j_commit_mutex);
1014 		goto put_jl;
1015 	}
1016 
1017 	if (!list_empty(&jl->j_bh_list)) {
1018 		int ret;
1019 
1020 		/*
1021 		 * We might sleep in numerous places inside
1022 		 * write_ordered_buffers. Relax the write lock.
1023 		 */
1024 		depth = reiserfs_write_unlock_nested(s);
1025 		ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1026 					    journal, jl, &jl->j_bh_list);
1027 		if (ret < 0 && retval == 0)
1028 			retval = ret;
1029 		reiserfs_write_lock_nested(s, depth);
1030 	}
1031 	BUG_ON(!list_empty(&jl->j_bh_list));
1032 	/*
1033 	 * for the description block and all the log blocks, submit any buffers
1034 	 * that haven't already reached the disk.  Try to write at least 256
1035 	 * log blocks. later on, we will only wait on blocks that correspond
1036 	 * to this transaction, but while we're unplugging we might as well
1037 	 * get a chunk of data on there.
1038 	 */
1039 	atomic_inc(&journal->j_async_throttle);
1040 	write_len = jl->j_len + 1;
1041 	if (write_len < 256)
1042 		write_len = 256;
1043 	for (i = 0 ; i < write_len ; i++) {
1044 		bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1045 		    SB_ONDISK_JOURNAL_SIZE(s);
1046 		tbh = journal_find_get_block(s, bn);
1047 		if (tbh) {
1048 			if (buffer_dirty(tbh)) {
1049 		            depth = reiserfs_write_unlock_nested(s);
1050 			    ll_rw_block(WRITE, 1, &tbh);
1051 			    reiserfs_write_lock_nested(s, depth);
1052 			}
1053 			put_bh(tbh) ;
1054 		}
1055 	}
1056 	atomic_dec(&journal->j_async_throttle);
1057 
1058 	for (i = 0; i < (jl->j_len + 1); i++) {
1059 		bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1060 		    (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1061 		tbh = journal_find_get_block(s, bn);
1062 
1063 		depth = reiserfs_write_unlock_nested(s);
1064 		__wait_on_buffer(tbh);
1065 		reiserfs_write_lock_nested(s, depth);
1066 		// since we're using ll_rw_blk above, it might have skipped over
1067 		// a locked buffer.  Double check here
1068 		//
1069 		/* redundant, sync_dirty_buffer() checks */
1070 		if (buffer_dirty(tbh)) {
1071 			depth = reiserfs_write_unlock_nested(s);
1072 			sync_dirty_buffer(tbh);
1073 			reiserfs_write_lock_nested(s, depth);
1074 		}
1075 		if (unlikely(!buffer_uptodate(tbh))) {
1076 #ifdef CONFIG_REISERFS_CHECK
1077 			reiserfs_warning(s, "journal-601",
1078 					 "buffer write failed");
1079 #endif
1080 			retval = -EIO;
1081 		}
1082 		put_bh(tbh);	/* once for journal_find_get_block */
1083 		put_bh(tbh);	/* once due to original getblk in do_journal_end */
1084 		atomic_dec(&(jl->j_commit_left));
1085 	}
1086 
1087 	BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1088 
1089 	/* If there was a write error in the journal - we can't commit
1090 	 * this transaction - it will be invalid and, if successful,
1091 	 * will just end up propagating the write error out to
1092 	 * the file system. */
1093 	if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1094 		if (buffer_dirty(jl->j_commit_bh))
1095 			BUG();
1096 		mark_buffer_dirty(jl->j_commit_bh) ;
1097 		depth = reiserfs_write_unlock_nested(s);
1098 		if (reiserfs_barrier_flush(s))
1099 			__sync_dirty_buffer(jl->j_commit_bh, WRITE_FLUSH_FUA);
1100 		else
1101 			sync_dirty_buffer(jl->j_commit_bh);
1102 		reiserfs_write_lock_nested(s, depth);
1103 	}
1104 
1105 	/* If there was a write error in the journal - we can't commit this
1106 	 * transaction - it will be invalid and, if successful, will just end
1107 	 * up propagating the write error out to the filesystem. */
1108 	if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1109 #ifdef CONFIG_REISERFS_CHECK
1110 		reiserfs_warning(s, "journal-615", "buffer write failed");
1111 #endif
1112 		retval = -EIO;
1113 	}
1114 	bforget(jl->j_commit_bh);
1115 	if (journal->j_last_commit_id != 0 &&
1116 	    (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1117 		reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1118 				 journal->j_last_commit_id, jl->j_trans_id);
1119 	}
1120 	journal->j_last_commit_id = jl->j_trans_id;
1121 
1122 	/* now, every commit block is on the disk.  It is safe to allow blocks freed during this transaction to be reallocated */
1123 	cleanup_freed_for_journal_list(s, jl);
1124 
1125 	retval = retval ? retval : journal->j_errno;
1126 
1127 	/* mark the metadata dirty */
1128 	if (!retval)
1129 		dirty_one_transaction(s, jl);
1130 	atomic_dec(&(jl->j_commit_left));
1131 
1132 	if (flushall) {
1133 		atomic_set(&(jl->j_older_commits_done), 1);
1134 	}
1135 	mutex_unlock(&jl->j_commit_mutex);
1136       put_jl:
1137 	put_journal_list(s, jl);
1138 
1139 	if (retval)
1140 		reiserfs_abort(s, retval, "Journal write error in %s",
1141 			       __func__);
1142 	return retval;
1143 }
1144 
1145 /*
1146 ** flush_journal_list frequently needs to find a newer transaction for a given block.  This does that, or
1147 ** returns NULL if it can't find anything
1148 */
1149 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1150 							  reiserfs_journal_cnode
1151 							  *cn)
1152 {
1153 	struct super_block *sb = cn->sb;
1154 	b_blocknr_t blocknr = cn->blocknr;
1155 
1156 	cn = cn->hprev;
1157 	while (cn) {
1158 		if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1159 			return cn->jlist;
1160 		}
1161 		cn = cn->hprev;
1162 	}
1163 	return NULL;
1164 }
1165 
1166 static int newer_jl_done(struct reiserfs_journal_cnode *cn)
1167 {
1168 	struct super_block *sb = cn->sb;
1169 	b_blocknr_t blocknr = cn->blocknr;
1170 
1171 	cn = cn->hprev;
1172 	while (cn) {
1173 		if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist &&
1174 		    atomic_read(&cn->jlist->j_commit_left) != 0)
1175 				    return 0;
1176 		cn = cn->hprev;
1177 	}
1178 	return 1;
1179 }
1180 
1181 static void remove_journal_hash(struct super_block *,
1182 				struct reiserfs_journal_cnode **,
1183 				struct reiserfs_journal_list *, unsigned long,
1184 				int);
1185 
1186 /*
1187 ** once all the real blocks have been flushed, it is safe to remove them from the
1188 ** journal list for this transaction.  Aside from freeing the cnode, this also allows the
1189 ** block to be reallocated for data blocks if it had been deleted.
1190 */
1191 static void remove_all_from_journal_list(struct super_block *sb,
1192 					 struct reiserfs_journal_list *jl,
1193 					 int debug)
1194 {
1195 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1196 	struct reiserfs_journal_cnode *cn, *last;
1197 	cn = jl->j_realblock;
1198 
1199 	/* which is better, to lock once around the whole loop, or
1200 	 ** to lock for each call to remove_journal_hash?
1201 	 */
1202 	while (cn) {
1203 		if (cn->blocknr != 0) {
1204 			if (debug) {
1205 				reiserfs_warning(sb, "reiserfs-2201",
1206 						 "block %u, bh is %d, state %ld",
1207 						 cn->blocknr, cn->bh ? 1 : 0,
1208 						 cn->state);
1209 			}
1210 			cn->state = 0;
1211 			remove_journal_hash(sb, journal->j_list_hash_table,
1212 					    jl, cn->blocknr, 1);
1213 		}
1214 		last = cn;
1215 		cn = cn->next;
1216 		free_cnode(sb, last);
1217 	}
1218 	jl->j_realblock = NULL;
1219 }
1220 
1221 /*
1222 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1223 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1224 ** releasing blocks in this transaction for reuse as data blocks.
1225 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1226 **
1227 */
1228 static int _update_journal_header_block(struct super_block *sb,
1229 					unsigned long offset,
1230 					unsigned int trans_id)
1231 {
1232 	struct reiserfs_journal_header *jh;
1233 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1234 	int depth;
1235 
1236 	if (reiserfs_is_journal_aborted(journal))
1237 		return -EIO;
1238 
1239 	if (trans_id >= journal->j_last_flush_trans_id) {
1240 		if (buffer_locked((journal->j_header_bh))) {
1241 			depth = reiserfs_write_unlock_nested(sb);
1242 			__wait_on_buffer(journal->j_header_bh);
1243 			reiserfs_write_lock_nested(sb, depth);
1244 			if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1245 #ifdef CONFIG_REISERFS_CHECK
1246 				reiserfs_warning(sb, "journal-699",
1247 						 "buffer write failed");
1248 #endif
1249 				return -EIO;
1250 			}
1251 		}
1252 		journal->j_last_flush_trans_id = trans_id;
1253 		journal->j_first_unflushed_offset = offset;
1254 		jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1255 							b_data);
1256 		jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1257 		jh->j_first_unflushed_offset = cpu_to_le32(offset);
1258 		jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1259 
1260 		set_buffer_dirty(journal->j_header_bh);
1261 		depth = reiserfs_write_unlock_nested(sb);
1262 
1263 		if (reiserfs_barrier_flush(sb))
1264 			__sync_dirty_buffer(journal->j_header_bh, WRITE_FLUSH_FUA);
1265 		else
1266 			sync_dirty_buffer(journal->j_header_bh);
1267 
1268 		reiserfs_write_lock_nested(sb, depth);
1269 		if (!buffer_uptodate(journal->j_header_bh)) {
1270 			reiserfs_warning(sb, "journal-837",
1271 					 "IO error during journal replay");
1272 			return -EIO;
1273 		}
1274 	}
1275 	return 0;
1276 }
1277 
1278 static int update_journal_header_block(struct super_block *sb,
1279 				       unsigned long offset,
1280 				       unsigned int trans_id)
1281 {
1282 	return _update_journal_header_block(sb, offset, trans_id);
1283 }
1284 
1285 /*
1286 ** flush any and all journal lists older than you are
1287 ** can only be called from flush_journal_list
1288 */
1289 static int flush_older_journal_lists(struct super_block *sb,
1290 				     struct reiserfs_journal_list *jl)
1291 {
1292 	struct list_head *entry;
1293 	struct reiserfs_journal_list *other_jl;
1294 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1295 	unsigned int trans_id = jl->j_trans_id;
1296 
1297 	/* we know we are the only ones flushing things, no extra race
1298 	 * protection is required.
1299 	 */
1300       restart:
1301 	entry = journal->j_journal_list.next;
1302 	/* Did we wrap? */
1303 	if (entry == &journal->j_journal_list)
1304 		return 0;
1305 	other_jl = JOURNAL_LIST_ENTRY(entry);
1306 	if (other_jl->j_trans_id < trans_id) {
1307 		BUG_ON(other_jl->j_refcount <= 0);
1308 		/* do not flush all */
1309 		flush_journal_list(sb, other_jl, 0);
1310 
1311 		/* other_jl is now deleted from the list */
1312 		goto restart;
1313 	}
1314 	return 0;
1315 }
1316 
1317 static void del_from_work_list(struct super_block *s,
1318 			       struct reiserfs_journal_list *jl)
1319 {
1320 	struct reiserfs_journal *journal = SB_JOURNAL(s);
1321 	if (!list_empty(&jl->j_working_list)) {
1322 		list_del_init(&jl->j_working_list);
1323 		journal->j_num_work_lists--;
1324 	}
1325 }
1326 
1327 /* flush a journal list, both commit and real blocks
1328 **
1329 ** always set flushall to 1, unless you are calling from inside
1330 ** flush_journal_list
1331 **
1332 ** IMPORTANT.  This can only be called while there are no journal writers,
1333 ** and the journal is locked.  That means it can only be called from
1334 ** do_journal_end, or by journal_release
1335 */
1336 static int flush_journal_list(struct super_block *s,
1337 			      struct reiserfs_journal_list *jl, int flushall)
1338 {
1339 	struct reiserfs_journal_list *pjl;
1340 	struct reiserfs_journal_cnode *cn, *last;
1341 	int count;
1342 	int was_jwait = 0;
1343 	int was_dirty = 0;
1344 	struct buffer_head *saved_bh;
1345 	unsigned long j_len_saved = jl->j_len;
1346 	struct reiserfs_journal *journal = SB_JOURNAL(s);
1347 	int err = 0;
1348 	int depth;
1349 
1350 	BUG_ON(j_len_saved <= 0);
1351 
1352 	if (atomic_read(&journal->j_wcount) != 0) {
1353 		reiserfs_warning(s, "clm-2048", "called with wcount %d",
1354 				 atomic_read(&journal->j_wcount));
1355 	}
1356 	BUG_ON(jl->j_trans_id == 0);
1357 
1358 	/* if flushall == 0, the lock is already held */
1359 	if (flushall) {
1360 		reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1361 	} else if (mutex_trylock(&journal->j_flush_mutex)) {
1362 		BUG();
1363 	}
1364 
1365 	count = 0;
1366 	if (j_len_saved > journal->j_trans_max) {
1367 		reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1368 			       j_len_saved, jl->j_trans_id);
1369 		return 0;
1370 	}
1371 
1372 	/* if all the work is already done, get out of here */
1373 	if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1374 	    atomic_read(&(jl->j_commit_left)) <= 0) {
1375 		goto flush_older_and_return;
1376 	}
1377 
1378 	/* start by putting the commit list on disk.  This will also flush
1379 	 ** the commit lists of any olders transactions
1380 	 */
1381 	flush_commit_list(s, jl, 1);
1382 
1383 	if (!(jl->j_state & LIST_DIRTY)
1384 	    && !reiserfs_is_journal_aborted(journal))
1385 		BUG();
1386 
1387 	/* are we done now? */
1388 	if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1389 	    atomic_read(&(jl->j_commit_left)) <= 0) {
1390 		goto flush_older_and_return;
1391 	}
1392 
1393 	/* loop through each cnode, see if we need to write it,
1394 	 ** or wait on a more recent transaction, or just ignore it
1395 	 */
1396 	if (atomic_read(&(journal->j_wcount)) != 0) {
1397 		reiserfs_panic(s, "journal-844", "journal list is flushing, "
1398 			       "wcount is not 0");
1399 	}
1400 	cn = jl->j_realblock;
1401 	while (cn) {
1402 		was_jwait = 0;
1403 		was_dirty = 0;
1404 		saved_bh = NULL;
1405 		/* blocknr of 0 is no longer in the hash, ignore it */
1406 		if (cn->blocknr == 0) {
1407 			goto free_cnode;
1408 		}
1409 
1410 		/* This transaction failed commit. Don't write out to the disk */
1411 		if (!(jl->j_state & LIST_DIRTY))
1412 			goto free_cnode;
1413 
1414 		pjl = find_newer_jl_for_cn(cn);
1415 		/* the order is important here.  We check pjl to make sure we
1416 		 ** don't clear BH_JDirty_wait if we aren't the one writing this
1417 		 ** block to disk
1418 		 */
1419 		if (!pjl && cn->bh) {
1420 			saved_bh = cn->bh;
1421 
1422 			/* we do this to make sure nobody releases the buffer while
1423 			 ** we are working with it
1424 			 */
1425 			get_bh(saved_bh);
1426 
1427 			if (buffer_journal_dirty(saved_bh)) {
1428 				BUG_ON(!can_dirty(cn));
1429 				was_jwait = 1;
1430 				was_dirty = 1;
1431 			} else if (can_dirty(cn)) {
1432 				/* everything with !pjl && jwait should be writable */
1433 				BUG();
1434 			}
1435 		}
1436 
1437 		/* if someone has this block in a newer transaction, just make
1438 		 ** sure they are committed, and don't try writing it to disk
1439 		 */
1440 		if (pjl) {
1441 			if (atomic_read(&pjl->j_commit_left))
1442 				flush_commit_list(s, pjl, 1);
1443 			goto free_cnode;
1444 		}
1445 
1446 		/* bh == NULL when the block got to disk on its own, OR,
1447 		 ** the block got freed in a future transaction
1448 		 */
1449 		if (saved_bh == NULL) {
1450 			goto free_cnode;
1451 		}
1452 
1453 		/* this should never happen.  kupdate_one_transaction has this list
1454 		 ** locked while it works, so we should never see a buffer here that
1455 		 ** is not marked JDirty_wait
1456 		 */
1457 		if ((!was_jwait) && !buffer_locked(saved_bh)) {
1458 			reiserfs_warning(s, "journal-813",
1459 					 "BAD! buffer %llu %cdirty %cjwait, "
1460 					 "not in a newer tranasction",
1461 					 (unsigned long long)saved_bh->
1462 					 b_blocknr, was_dirty ? ' ' : '!',
1463 					 was_jwait ? ' ' : '!');
1464 		}
1465 		if (was_dirty) {
1466 			/* we inc again because saved_bh gets decremented at free_cnode */
1467 			get_bh(saved_bh);
1468 			set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1469 			lock_buffer(saved_bh);
1470 			BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1471 			if (buffer_dirty(saved_bh))
1472 				submit_logged_buffer(saved_bh);
1473 			else
1474 				unlock_buffer(saved_bh);
1475 			count++;
1476 		} else {
1477 			reiserfs_warning(s, "clm-2082",
1478 					 "Unable to flush buffer %llu in %s",
1479 					 (unsigned long long)saved_bh->
1480 					 b_blocknr, __func__);
1481 		}
1482 	      free_cnode:
1483 		last = cn;
1484 		cn = cn->next;
1485 		if (saved_bh) {
1486 			/* we incremented this to keep others from taking the buffer head away */
1487 			put_bh(saved_bh);
1488 			if (atomic_read(&(saved_bh->b_count)) < 0) {
1489 				reiserfs_warning(s, "journal-945",
1490 						 "saved_bh->b_count < 0");
1491 			}
1492 		}
1493 	}
1494 	if (count > 0) {
1495 		cn = jl->j_realblock;
1496 		while (cn) {
1497 			if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1498 				if (!cn->bh) {
1499 					reiserfs_panic(s, "journal-1011",
1500 						       "cn->bh is NULL");
1501 				}
1502 
1503 				depth = reiserfs_write_unlock_nested(s);
1504 				__wait_on_buffer(cn->bh);
1505 				reiserfs_write_lock_nested(s, depth);
1506 
1507 				if (!cn->bh) {
1508 					reiserfs_panic(s, "journal-1012",
1509 						       "cn->bh is NULL");
1510 				}
1511 				if (unlikely(!buffer_uptodate(cn->bh))) {
1512 #ifdef CONFIG_REISERFS_CHECK
1513 					reiserfs_warning(s, "journal-949",
1514 							 "buffer write failed");
1515 #endif
1516 					err = -EIO;
1517 				}
1518 				/* note, we must clear the JDirty_wait bit after the up to date
1519 				 ** check, otherwise we race against our flushpage routine
1520 				 */
1521 				BUG_ON(!test_clear_buffer_journal_dirty
1522 				       (cn->bh));
1523 
1524 				/* drop one ref for us */
1525 				put_bh(cn->bh);
1526 				/* drop one ref for journal_mark_dirty */
1527 				release_buffer_page(cn->bh);
1528 			}
1529 			cn = cn->next;
1530 		}
1531 	}
1532 
1533 	if (err)
1534 		reiserfs_abort(s, -EIO,
1535 			       "Write error while pushing transaction to disk in %s",
1536 			       __func__);
1537       flush_older_and_return:
1538 
1539 	/* before we can update the journal header block, we _must_ flush all
1540 	 ** real blocks from all older transactions to disk.  This is because
1541 	 ** once the header block is updated, this transaction will not be
1542 	 ** replayed after a crash
1543 	 */
1544 	if (flushall) {
1545 		flush_older_journal_lists(s, jl);
1546 	}
1547 
1548 	err = journal->j_errno;
1549 	/* before we can remove everything from the hash tables for this
1550 	 ** transaction, we must make sure it can never be replayed
1551 	 **
1552 	 ** since we are only called from do_journal_end, we know for sure there
1553 	 ** are no allocations going on while we are flushing journal lists.  So,
1554 	 ** we only need to update the journal header block for the last list
1555 	 ** being flushed
1556 	 */
1557 	if (!err && flushall) {
1558 		err =
1559 		    update_journal_header_block(s,
1560 						(jl->j_start + jl->j_len +
1561 						 2) % SB_ONDISK_JOURNAL_SIZE(s),
1562 						jl->j_trans_id);
1563 		if (err)
1564 			reiserfs_abort(s, -EIO,
1565 				       "Write error while updating journal header in %s",
1566 				       __func__);
1567 	}
1568 	remove_all_from_journal_list(s, jl, 0);
1569 	list_del_init(&jl->j_list);
1570 	journal->j_num_lists--;
1571 	del_from_work_list(s, jl);
1572 
1573 	if (journal->j_last_flush_id != 0 &&
1574 	    (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1575 		reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1576 				 journal->j_last_flush_id, jl->j_trans_id);
1577 	}
1578 	journal->j_last_flush_id = jl->j_trans_id;
1579 
1580 	/* not strictly required since we are freeing the list, but it should
1581 	 * help find code using dead lists later on
1582 	 */
1583 	jl->j_len = 0;
1584 	atomic_set(&(jl->j_nonzerolen), 0);
1585 	jl->j_start = 0;
1586 	jl->j_realblock = NULL;
1587 	jl->j_commit_bh = NULL;
1588 	jl->j_trans_id = 0;
1589 	jl->j_state = 0;
1590 	put_journal_list(s, jl);
1591 	if (flushall)
1592 		mutex_unlock(&journal->j_flush_mutex);
1593 	return err;
1594 }
1595 
1596 static int test_transaction(struct super_block *s,
1597                             struct reiserfs_journal_list *jl)
1598 {
1599 	struct reiserfs_journal_cnode *cn;
1600 
1601 	if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0)
1602 		return 1;
1603 
1604 	cn = jl->j_realblock;
1605 	while (cn) {
1606 		/* if the blocknr == 0, this has been cleared from the hash,
1607 		 ** skip it
1608 		 */
1609 		if (cn->blocknr == 0) {
1610 			goto next;
1611 		}
1612 		if (cn->bh && !newer_jl_done(cn))
1613 			return 0;
1614 	      next:
1615 		cn = cn->next;
1616 		cond_resched();
1617 	}
1618 	return 0;
1619 }
1620 
1621 static int write_one_transaction(struct super_block *s,
1622 				 struct reiserfs_journal_list *jl,
1623 				 struct buffer_chunk *chunk)
1624 {
1625 	struct reiserfs_journal_cnode *cn;
1626 	int ret = 0;
1627 
1628 	jl->j_state |= LIST_TOUCHED;
1629 	del_from_work_list(s, jl);
1630 	if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1631 		return 0;
1632 	}
1633 
1634 	cn = jl->j_realblock;
1635 	while (cn) {
1636 		/* if the blocknr == 0, this has been cleared from the hash,
1637 		 ** skip it
1638 		 */
1639 		if (cn->blocknr == 0) {
1640 			goto next;
1641 		}
1642 		if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1643 			struct buffer_head *tmp_bh;
1644 			/* we can race against journal_mark_freed when we try
1645 			 * to lock_buffer(cn->bh), so we have to inc the buffer
1646 			 * count, and recheck things after locking
1647 			 */
1648 			tmp_bh = cn->bh;
1649 			get_bh(tmp_bh);
1650 			lock_buffer(tmp_bh);
1651 			if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1652 				if (!buffer_journal_dirty(tmp_bh) ||
1653 				    buffer_journal_prepared(tmp_bh))
1654 					BUG();
1655 				add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1656 				ret++;
1657 			} else {
1658 				/* note, cn->bh might be null now */
1659 				unlock_buffer(tmp_bh);
1660 			}
1661 			put_bh(tmp_bh);
1662 		}
1663 	      next:
1664 		cn = cn->next;
1665 		cond_resched();
1666 	}
1667 	return ret;
1668 }
1669 
1670 /* used by flush_commit_list */
1671 static int dirty_one_transaction(struct super_block *s,
1672 				 struct reiserfs_journal_list *jl)
1673 {
1674 	struct reiserfs_journal_cnode *cn;
1675 	struct reiserfs_journal_list *pjl;
1676 	int ret = 0;
1677 
1678 	jl->j_state |= LIST_DIRTY;
1679 	cn = jl->j_realblock;
1680 	while (cn) {
1681 		/* look for a more recent transaction that logged this
1682 		 ** buffer.  Only the most recent transaction with a buffer in
1683 		 ** it is allowed to send that buffer to disk
1684 		 */
1685 		pjl = find_newer_jl_for_cn(cn);
1686 		if (!pjl && cn->blocknr && cn->bh
1687 		    && buffer_journal_dirty(cn->bh)) {
1688 			BUG_ON(!can_dirty(cn));
1689 			/* if the buffer is prepared, it will either be logged
1690 			 * or restored.  If restored, we need to make sure
1691 			 * it actually gets marked dirty
1692 			 */
1693 			clear_buffer_journal_new(cn->bh);
1694 			if (buffer_journal_prepared(cn->bh)) {
1695 				set_buffer_journal_restore_dirty(cn->bh);
1696 			} else {
1697 				set_buffer_journal_test(cn->bh);
1698 				mark_buffer_dirty(cn->bh);
1699 			}
1700 		}
1701 		cn = cn->next;
1702 	}
1703 	return ret;
1704 }
1705 
1706 static int kupdate_transactions(struct super_block *s,
1707 				struct reiserfs_journal_list *jl,
1708 				struct reiserfs_journal_list **next_jl,
1709 				unsigned int *next_trans_id,
1710 				int num_blocks, int num_trans)
1711 {
1712 	int ret = 0;
1713 	int written = 0;
1714 	int transactions_flushed = 0;
1715 	unsigned int orig_trans_id = jl->j_trans_id;
1716 	struct buffer_chunk chunk;
1717 	struct list_head *entry;
1718 	struct reiserfs_journal *journal = SB_JOURNAL(s);
1719 	chunk.nr = 0;
1720 
1721 	reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1722 	if (!journal_list_still_alive(s, orig_trans_id)) {
1723 		goto done;
1724 	}
1725 
1726 	/* we've got j_flush_mutex held, nobody is going to delete any
1727 	 * of these lists out from underneath us
1728 	 */
1729 	while ((num_trans && transactions_flushed < num_trans) ||
1730 	       (!num_trans && written < num_blocks)) {
1731 
1732 		if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1733 		    atomic_read(&jl->j_commit_left)
1734 		    || !(jl->j_state & LIST_DIRTY)) {
1735 			del_from_work_list(s, jl);
1736 			break;
1737 		}
1738 		ret = write_one_transaction(s, jl, &chunk);
1739 
1740 		if (ret < 0)
1741 			goto done;
1742 		transactions_flushed++;
1743 		written += ret;
1744 		entry = jl->j_list.next;
1745 
1746 		/* did we wrap? */
1747 		if (entry == &journal->j_journal_list) {
1748 			break;
1749 		}
1750 		jl = JOURNAL_LIST_ENTRY(entry);
1751 
1752 		/* don't bother with older transactions */
1753 		if (jl->j_trans_id <= orig_trans_id)
1754 			break;
1755 	}
1756 	if (chunk.nr) {
1757 		write_chunk(&chunk);
1758 	}
1759 
1760       done:
1761 	mutex_unlock(&journal->j_flush_mutex);
1762 	return ret;
1763 }
1764 
1765 /* for o_sync and fsync heavy applications, they tend to use
1766 ** all the journa list slots with tiny transactions.  These
1767 ** trigger lots and lots of calls to update the header block, which
1768 ** adds seeks and slows things down.
1769 **
1770 ** This function tries to clear out a large chunk of the journal lists
1771 ** at once, which makes everything faster since only the newest journal
1772 ** list updates the header block
1773 */
1774 static int flush_used_journal_lists(struct super_block *s,
1775 				    struct reiserfs_journal_list *jl)
1776 {
1777 	unsigned long len = 0;
1778 	unsigned long cur_len;
1779 	int ret;
1780 	int i;
1781 	int limit = 256;
1782 	struct reiserfs_journal_list *tjl;
1783 	struct reiserfs_journal_list *flush_jl;
1784 	unsigned int trans_id;
1785 	struct reiserfs_journal *journal = SB_JOURNAL(s);
1786 
1787 	flush_jl = tjl = jl;
1788 
1789 	/* in data logging mode, try harder to flush a lot of blocks */
1790 	if (reiserfs_data_log(s))
1791 		limit = 1024;
1792 	/* flush for 256 transactions or limit blocks, whichever comes first */
1793 	for (i = 0; i < 256 && len < limit; i++) {
1794 		if (atomic_read(&tjl->j_commit_left) ||
1795 		    tjl->j_trans_id < jl->j_trans_id) {
1796 			break;
1797 		}
1798 		cur_len = atomic_read(&tjl->j_nonzerolen);
1799 		if (cur_len > 0) {
1800 			tjl->j_state &= ~LIST_TOUCHED;
1801 		}
1802 		len += cur_len;
1803 		flush_jl = tjl;
1804 		if (tjl->j_list.next == &journal->j_journal_list)
1805 			break;
1806 		tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1807 	}
1808 	/* try to find a group of blocks we can flush across all the
1809 	 ** transactions, but only bother if we've actually spanned
1810 	 ** across multiple lists
1811 	 */
1812 	if (flush_jl != jl) {
1813 		ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1814 	}
1815 	flush_journal_list(s, flush_jl, 1);
1816 	return 0;
1817 }
1818 
1819 /*
1820 ** removes any nodes in table with name block and dev as bh.
1821 ** only touchs the hnext and hprev pointers.
1822 */
1823 void remove_journal_hash(struct super_block *sb,
1824 			 struct reiserfs_journal_cnode **table,
1825 			 struct reiserfs_journal_list *jl,
1826 			 unsigned long block, int remove_freed)
1827 {
1828 	struct reiserfs_journal_cnode *cur;
1829 	struct reiserfs_journal_cnode **head;
1830 
1831 	head = &(journal_hash(table, sb, block));
1832 	if (!head) {
1833 		return;
1834 	}
1835 	cur = *head;
1836 	while (cur) {
1837 		if (cur->blocknr == block && cur->sb == sb
1838 		    && (jl == NULL || jl == cur->jlist)
1839 		    && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1840 			if (cur->hnext) {
1841 				cur->hnext->hprev = cur->hprev;
1842 			}
1843 			if (cur->hprev) {
1844 				cur->hprev->hnext = cur->hnext;
1845 			} else {
1846 				*head = cur->hnext;
1847 			}
1848 			cur->blocknr = 0;
1849 			cur->sb = NULL;
1850 			cur->state = 0;
1851 			if (cur->bh && cur->jlist)	/* anybody who clears the cur->bh will also dec the nonzerolen */
1852 				atomic_dec(&(cur->jlist->j_nonzerolen));
1853 			cur->bh = NULL;
1854 			cur->jlist = NULL;
1855 		}
1856 		cur = cur->hnext;
1857 	}
1858 }
1859 
1860 static void free_journal_ram(struct super_block *sb)
1861 {
1862 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1863 	kfree(journal->j_current_jl);
1864 	journal->j_num_lists--;
1865 
1866 	vfree(journal->j_cnode_free_orig);
1867 	free_list_bitmaps(sb, journal->j_list_bitmap);
1868 	free_bitmap_nodes(sb);	/* must be after free_list_bitmaps */
1869 	if (journal->j_header_bh) {
1870 		brelse(journal->j_header_bh);
1871 	}
1872 	/* j_header_bh is on the journal dev, make sure not to release the journal
1873 	 * dev until we brelse j_header_bh
1874 	 */
1875 	release_journal_dev(sb, journal);
1876 	vfree(journal);
1877 }
1878 
1879 /*
1880 ** call on unmount.  Only set error to 1 if you haven't made your way out
1881 ** of read_super() yet.  Any other caller must keep error at 0.
1882 */
1883 static int do_journal_release(struct reiserfs_transaction_handle *th,
1884 			      struct super_block *sb, int error)
1885 {
1886 	struct reiserfs_transaction_handle myth;
1887 	int flushed = 0;
1888 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1889 
1890 	/* we only want to flush out transactions if we were called with error == 0
1891 	 */
1892 	if (!error && !(sb->s_flags & MS_RDONLY)) {
1893 		/* end the current trans */
1894 		BUG_ON(!th->t_trans_id);
1895 		do_journal_end(th, sb, 10, FLUSH_ALL);
1896 
1897 		/* make sure something gets logged to force our way into the flush code */
1898 		if (!journal_join(&myth, sb, 1)) {
1899 			reiserfs_prepare_for_journal(sb,
1900 						     SB_BUFFER_WITH_SB(sb),
1901 						     1);
1902 			journal_mark_dirty(&myth, sb,
1903 					   SB_BUFFER_WITH_SB(sb));
1904 			do_journal_end(&myth, sb, 1, FLUSH_ALL);
1905 			flushed = 1;
1906 		}
1907 	}
1908 
1909 	/* this also catches errors during the do_journal_end above */
1910 	if (!error && reiserfs_is_journal_aborted(journal)) {
1911 		memset(&myth, 0, sizeof(myth));
1912 		if (!journal_join_abort(&myth, sb, 1)) {
1913 			reiserfs_prepare_for_journal(sb,
1914 						     SB_BUFFER_WITH_SB(sb),
1915 						     1);
1916 			journal_mark_dirty(&myth, sb,
1917 					   SB_BUFFER_WITH_SB(sb));
1918 			do_journal_end(&myth, sb, 1, FLUSH_ALL);
1919 		}
1920 	}
1921 
1922 	reiserfs_mounted_fs_count--;
1923 	/* wait for all commits to finish */
1924 	cancel_delayed_work(&SB_JOURNAL(sb)->j_work);
1925 
1926 	/*
1927 	 * We must release the write lock here because
1928 	 * the workqueue job (flush_async_commit) needs this lock
1929 	 */
1930 	reiserfs_write_unlock(sb);
1931 
1932 	cancel_delayed_work_sync(&REISERFS_SB(sb)->old_work);
1933 	flush_workqueue(commit_wq);
1934 
1935 	if (!reiserfs_mounted_fs_count) {
1936 		destroy_workqueue(commit_wq);
1937 		commit_wq = NULL;
1938 	}
1939 
1940 	free_journal_ram(sb);
1941 
1942 	reiserfs_write_lock(sb);
1943 
1944 	return 0;
1945 }
1946 
1947 /*
1948 ** call on unmount.  flush all journal trans, release all alloc'd ram
1949 */
1950 int journal_release(struct reiserfs_transaction_handle *th,
1951 		    struct super_block *sb)
1952 {
1953 	return do_journal_release(th, sb, 0);
1954 }
1955 
1956 /*
1957 ** only call from an error condition inside reiserfs_read_super!
1958 */
1959 int journal_release_error(struct reiserfs_transaction_handle *th,
1960 			  struct super_block *sb)
1961 {
1962 	return do_journal_release(th, sb, 1);
1963 }
1964 
1965 /* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
1966 static int journal_compare_desc_commit(struct super_block *sb,
1967 				       struct reiserfs_journal_desc *desc,
1968 				       struct reiserfs_journal_commit *commit)
1969 {
1970 	if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1971 	    get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1972 	    get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
1973 	    get_commit_trans_len(commit) <= 0) {
1974 		return 1;
1975 	}
1976 	return 0;
1977 }
1978 
1979 /* returns 0 if it did not find a description block
1980 ** returns -1 if it found a corrupt commit block
1981 ** returns 1 if both desc and commit were valid
1982 ** NOTE: only called during fs mount
1983 */
1984 static int journal_transaction_is_valid(struct super_block *sb,
1985 					struct buffer_head *d_bh,
1986 					unsigned int *oldest_invalid_trans_id,
1987 					unsigned long *newest_mount_id)
1988 {
1989 	struct reiserfs_journal_desc *desc;
1990 	struct reiserfs_journal_commit *commit;
1991 	struct buffer_head *c_bh;
1992 	unsigned long offset;
1993 
1994 	if (!d_bh)
1995 		return 0;
1996 
1997 	desc = (struct reiserfs_journal_desc *)d_bh->b_data;
1998 	if (get_desc_trans_len(desc) > 0
1999 	    && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2000 		if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2001 		    && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2002 			reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2003 				       "journal-986: transaction "
2004 				       "is valid returning because trans_id %d is greater than "
2005 				       "oldest_invalid %lu",
2006 				       get_desc_trans_id(desc),
2007 				       *oldest_invalid_trans_id);
2008 			return 0;
2009 		}
2010 		if (newest_mount_id
2011 		    && *newest_mount_id > get_desc_mount_id(desc)) {
2012 			reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2013 				       "journal-1087: transaction "
2014 				       "is valid returning because mount_id %d is less than "
2015 				       "newest_mount_id %lu",
2016 				       get_desc_mount_id(desc),
2017 				       *newest_mount_id);
2018 			return -1;
2019 		}
2020 		if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2021 			reiserfs_warning(sb, "journal-2018",
2022 					 "Bad transaction length %d "
2023 					 "encountered, ignoring transaction",
2024 					 get_desc_trans_len(desc));
2025 			return -1;
2026 		}
2027 		offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2028 
2029 		/* ok, we have a journal description block, lets see if the transaction was valid */
2030 		c_bh =
2031 		    journal_bread(sb,
2032 				  SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2033 				  ((offset + get_desc_trans_len(desc) +
2034 				    1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2035 		if (!c_bh)
2036 			return 0;
2037 		commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2038 		if (journal_compare_desc_commit(sb, desc, commit)) {
2039 			reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2040 				       "journal_transaction_is_valid, commit offset %ld had bad "
2041 				       "time %d or length %d",
2042 				       c_bh->b_blocknr -
2043 				       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2044 				       get_commit_trans_id(commit),
2045 				       get_commit_trans_len(commit));
2046 			brelse(c_bh);
2047 			if (oldest_invalid_trans_id) {
2048 				*oldest_invalid_trans_id =
2049 				    get_desc_trans_id(desc);
2050 				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2051 					       "journal-1004: "
2052 					       "transaction_is_valid setting oldest invalid trans_id "
2053 					       "to %d",
2054 					       get_desc_trans_id(desc));
2055 			}
2056 			return -1;
2057 		}
2058 		brelse(c_bh);
2059 		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2060 			       "journal-1006: found valid "
2061 			       "transaction start offset %llu, len %d id %d",
2062 			       d_bh->b_blocknr -
2063 			       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2064 			       get_desc_trans_len(desc),
2065 			       get_desc_trans_id(desc));
2066 		return 1;
2067 	} else {
2068 		return 0;
2069 	}
2070 }
2071 
2072 static void brelse_array(struct buffer_head **heads, int num)
2073 {
2074 	int i;
2075 	for (i = 0; i < num; i++) {
2076 		brelse(heads[i]);
2077 	}
2078 }
2079 
2080 /*
2081 ** given the start, and values for the oldest acceptable transactions,
2082 ** this either reads in a replays a transaction, or returns because the
2083 ** transaction is invalid, or too old.
2084 ** NOTE: only called during fs mount
2085 */
2086 static int journal_read_transaction(struct super_block *sb,
2087 				    unsigned long cur_dblock,
2088 				    unsigned long oldest_start,
2089 				    unsigned int oldest_trans_id,
2090 				    unsigned long newest_mount_id)
2091 {
2092 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
2093 	struct reiserfs_journal_desc *desc;
2094 	struct reiserfs_journal_commit *commit;
2095 	unsigned int trans_id = 0;
2096 	struct buffer_head *c_bh;
2097 	struct buffer_head *d_bh;
2098 	struct buffer_head **log_blocks = NULL;
2099 	struct buffer_head **real_blocks = NULL;
2100 	unsigned int trans_offset;
2101 	int i;
2102 	int trans_half;
2103 
2104 	d_bh = journal_bread(sb, cur_dblock);
2105 	if (!d_bh)
2106 		return 1;
2107 	desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2108 	trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2109 	reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2110 		       "journal_read_transaction, offset %llu, len %d mount_id %d",
2111 		       d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2112 		       get_desc_trans_len(desc), get_desc_mount_id(desc));
2113 	if (get_desc_trans_id(desc) < oldest_trans_id) {
2114 		reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2115 			       "journal_read_trans skipping because %lu is too old",
2116 			       cur_dblock -
2117 			       SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2118 		brelse(d_bh);
2119 		return 1;
2120 	}
2121 	if (get_desc_mount_id(desc) != newest_mount_id) {
2122 		reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2123 			       "journal_read_trans skipping because %d is != "
2124 			       "newest_mount_id %lu", get_desc_mount_id(desc),
2125 			       newest_mount_id);
2126 		brelse(d_bh);
2127 		return 1;
2128 	}
2129 	c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2130 			     ((trans_offset + get_desc_trans_len(desc) + 1) %
2131 			      SB_ONDISK_JOURNAL_SIZE(sb)));
2132 	if (!c_bh) {
2133 		brelse(d_bh);
2134 		return 1;
2135 	}
2136 	commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2137 	if (journal_compare_desc_commit(sb, desc, commit)) {
2138 		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2139 			       "journal_read_transaction, "
2140 			       "commit offset %llu had bad time %d or length %d",
2141 			       c_bh->b_blocknr -
2142 			       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2143 			       get_commit_trans_id(commit),
2144 			       get_commit_trans_len(commit));
2145 		brelse(c_bh);
2146 		brelse(d_bh);
2147 		return 1;
2148 	}
2149 
2150 	if (bdev_read_only(sb->s_bdev)) {
2151 		reiserfs_warning(sb, "clm-2076",
2152 				 "device is readonly, unable to replay log");
2153 		brelse(c_bh);
2154 		brelse(d_bh);
2155 		return -EROFS;
2156 	}
2157 
2158 	trans_id = get_desc_trans_id(desc);
2159 	/* now we know we've got a good transaction, and it was inside the valid time ranges */
2160 	log_blocks = kmalloc(get_desc_trans_len(desc) *
2161 			     sizeof(struct buffer_head *), GFP_NOFS);
2162 	real_blocks = kmalloc(get_desc_trans_len(desc) *
2163 			      sizeof(struct buffer_head *), GFP_NOFS);
2164 	if (!log_blocks || !real_blocks) {
2165 		brelse(c_bh);
2166 		brelse(d_bh);
2167 		kfree(log_blocks);
2168 		kfree(real_blocks);
2169 		reiserfs_warning(sb, "journal-1169",
2170 				 "kmalloc failed, unable to mount FS");
2171 		return -1;
2172 	}
2173 	/* get all the buffer heads */
2174 	trans_half = journal_trans_half(sb->s_blocksize);
2175 	for (i = 0; i < get_desc_trans_len(desc); i++) {
2176 		log_blocks[i] =
2177 		    journal_getblk(sb,
2178 				   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2179 				   (trans_offset + 1 +
2180 				    i) % SB_ONDISK_JOURNAL_SIZE(sb));
2181 		if (i < trans_half) {
2182 			real_blocks[i] =
2183 			    sb_getblk(sb,
2184 				      le32_to_cpu(desc->j_realblock[i]));
2185 		} else {
2186 			real_blocks[i] =
2187 			    sb_getblk(sb,
2188 				      le32_to_cpu(commit->
2189 						  j_realblock[i - trans_half]));
2190 		}
2191 		if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2192 			reiserfs_warning(sb, "journal-1207",
2193 					 "REPLAY FAILURE fsck required! "
2194 					 "Block to replay is outside of "
2195 					 "filesystem");
2196 			goto abort_replay;
2197 		}
2198 		/* make sure we don't try to replay onto log or reserved area */
2199 		if (is_block_in_log_or_reserved_area
2200 		    (sb, real_blocks[i]->b_blocknr)) {
2201 			reiserfs_warning(sb, "journal-1204",
2202 					 "REPLAY FAILURE fsck required! "
2203 					 "Trying to replay onto a log block");
2204 		      abort_replay:
2205 			brelse_array(log_blocks, i);
2206 			brelse_array(real_blocks, i);
2207 			brelse(c_bh);
2208 			brelse(d_bh);
2209 			kfree(log_blocks);
2210 			kfree(real_blocks);
2211 			return -1;
2212 		}
2213 	}
2214 	/* read in the log blocks, memcpy to the corresponding real block */
2215 	ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2216 	for (i = 0; i < get_desc_trans_len(desc); i++) {
2217 
2218 		wait_on_buffer(log_blocks[i]);
2219 		if (!buffer_uptodate(log_blocks[i])) {
2220 			reiserfs_warning(sb, "journal-1212",
2221 					 "REPLAY FAILURE fsck required! "
2222 					 "buffer write failed");
2223 			brelse_array(log_blocks + i,
2224 				     get_desc_trans_len(desc) - i);
2225 			brelse_array(real_blocks, get_desc_trans_len(desc));
2226 			brelse(c_bh);
2227 			brelse(d_bh);
2228 			kfree(log_blocks);
2229 			kfree(real_blocks);
2230 			return -1;
2231 		}
2232 		memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2233 		       real_blocks[i]->b_size);
2234 		set_buffer_uptodate(real_blocks[i]);
2235 		brelse(log_blocks[i]);
2236 	}
2237 	/* flush out the real blocks */
2238 	for (i = 0; i < get_desc_trans_len(desc); i++) {
2239 		set_buffer_dirty(real_blocks[i]);
2240 		write_dirty_buffer(real_blocks[i], WRITE);
2241 	}
2242 	for (i = 0; i < get_desc_trans_len(desc); i++) {
2243 		wait_on_buffer(real_blocks[i]);
2244 		if (!buffer_uptodate(real_blocks[i])) {
2245 			reiserfs_warning(sb, "journal-1226",
2246 					 "REPLAY FAILURE, fsck required! "
2247 					 "buffer write failed");
2248 			brelse_array(real_blocks + i,
2249 				     get_desc_trans_len(desc) - i);
2250 			brelse(c_bh);
2251 			brelse(d_bh);
2252 			kfree(log_blocks);
2253 			kfree(real_blocks);
2254 			return -1;
2255 		}
2256 		brelse(real_blocks[i]);
2257 	}
2258 	cur_dblock =
2259 	    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2260 	    ((trans_offset + get_desc_trans_len(desc) +
2261 	      2) % SB_ONDISK_JOURNAL_SIZE(sb));
2262 	reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2263 		       "journal-1095: setting journal " "start to offset %ld",
2264 		       cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2265 
2266 	/* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2267 	journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2268 	journal->j_last_flush_trans_id = trans_id;
2269 	journal->j_trans_id = trans_id + 1;
2270 	/* check for trans_id overflow */
2271 	if (journal->j_trans_id == 0)
2272 		journal->j_trans_id = 10;
2273 	brelse(c_bh);
2274 	brelse(d_bh);
2275 	kfree(log_blocks);
2276 	kfree(real_blocks);
2277 	return 0;
2278 }
2279 
2280 /* This function reads blocks starting from block and to max_block of bufsize
2281    size (but no more than BUFNR blocks at a time). This proved to improve
2282    mounting speed on self-rebuilding raid5 arrays at least.
2283    Right now it is only used from journal code. But later we might use it
2284    from other places.
2285    Note: Do not use journal_getblk/sb_getblk functions here! */
2286 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2287 					   b_blocknr_t block, int bufsize,
2288 					   b_blocknr_t max_block)
2289 {
2290 	struct buffer_head *bhlist[BUFNR];
2291 	unsigned int blocks = BUFNR;
2292 	struct buffer_head *bh;
2293 	int i, j;
2294 
2295 	bh = __getblk(dev, block, bufsize);
2296 	if (buffer_uptodate(bh))
2297 		return (bh);
2298 
2299 	if (block + BUFNR > max_block) {
2300 		blocks = max_block - block;
2301 	}
2302 	bhlist[0] = bh;
2303 	j = 1;
2304 	for (i = 1; i < blocks; i++) {
2305 		bh = __getblk(dev, block + i, bufsize);
2306 		if (buffer_uptodate(bh)) {
2307 			brelse(bh);
2308 			break;
2309 		} else
2310 			bhlist[j++] = bh;
2311 	}
2312 	ll_rw_block(READ, j, bhlist);
2313 	for (i = 1; i < j; i++)
2314 		brelse(bhlist[i]);
2315 	bh = bhlist[0];
2316 	wait_on_buffer(bh);
2317 	if (buffer_uptodate(bh))
2318 		return bh;
2319 	brelse(bh);
2320 	return NULL;
2321 }
2322 
2323 /*
2324 ** read and replay the log
2325 ** on a clean unmount, the journal header's next unflushed pointer will
2326 ** be to an invalid transaction.  This tests that before finding all the
2327 ** transactions in the log, which makes normal mount times fast.
2328 ** After a crash, this starts with the next unflushed transaction, and
2329 ** replays until it finds one too old, or invalid.
2330 ** On exit, it sets things up so the first transaction will work correctly.
2331 ** NOTE: only called during fs mount
2332 */
2333 static int journal_read(struct super_block *sb)
2334 {
2335 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
2336 	struct reiserfs_journal_desc *desc;
2337 	unsigned int oldest_trans_id = 0;
2338 	unsigned int oldest_invalid_trans_id = 0;
2339 	time_t start;
2340 	unsigned long oldest_start = 0;
2341 	unsigned long cur_dblock = 0;
2342 	unsigned long newest_mount_id = 9;
2343 	struct buffer_head *d_bh;
2344 	struct reiserfs_journal_header *jh;
2345 	int valid_journal_header = 0;
2346 	int replay_count = 0;
2347 	int continue_replay = 1;
2348 	int ret;
2349 	char b[BDEVNAME_SIZE];
2350 
2351 	cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2352 	reiserfs_info(sb, "checking transaction log (%s)\n",
2353 		      bdevname(journal->j_dev_bd, b));
2354 	start = get_seconds();
2355 
2356 	/* step 1, read in the journal header block.  Check the transaction it says
2357 	 ** is the first unflushed, and if that transaction is not valid,
2358 	 ** replay is done
2359 	 */
2360 	journal->j_header_bh = journal_bread(sb,
2361 					     SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2362 					     + SB_ONDISK_JOURNAL_SIZE(sb));
2363 	if (!journal->j_header_bh) {
2364 		return 1;
2365 	}
2366 	jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2367 	if (le32_to_cpu(jh->j_first_unflushed_offset) <
2368 	    SB_ONDISK_JOURNAL_SIZE(sb)
2369 	    && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2370 		oldest_start =
2371 		    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2372 		    le32_to_cpu(jh->j_first_unflushed_offset);
2373 		oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2374 		newest_mount_id = le32_to_cpu(jh->j_mount_id);
2375 		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2376 			       "journal-1153: found in "
2377 			       "header: first_unflushed_offset %d, last_flushed_trans_id "
2378 			       "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2379 			       le32_to_cpu(jh->j_last_flush_trans_id));
2380 		valid_journal_header = 1;
2381 
2382 		/* now, we try to read the first unflushed offset.  If it is not valid,
2383 		 ** there is nothing more we can do, and it makes no sense to read
2384 		 ** through the whole log.
2385 		 */
2386 		d_bh =
2387 		    journal_bread(sb,
2388 				  SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2389 				  le32_to_cpu(jh->j_first_unflushed_offset));
2390 		ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2391 		if (!ret) {
2392 			continue_replay = 0;
2393 		}
2394 		brelse(d_bh);
2395 		goto start_log_replay;
2396 	}
2397 
2398 	/* ok, there are transactions that need to be replayed.  start with the first log block, find
2399 	 ** all the valid transactions, and pick out the oldest.
2400 	 */
2401 	while (continue_replay
2402 	       && cur_dblock <
2403 	       (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2404 		SB_ONDISK_JOURNAL_SIZE(sb))) {
2405 		/* Note that it is required for blocksize of primary fs device and journal
2406 		   device to be the same */
2407 		d_bh =
2408 		    reiserfs_breada(journal->j_dev_bd, cur_dblock,
2409 				    sb->s_blocksize,
2410 				    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2411 				    SB_ONDISK_JOURNAL_SIZE(sb));
2412 		ret =
2413 		    journal_transaction_is_valid(sb, d_bh,
2414 						 &oldest_invalid_trans_id,
2415 						 &newest_mount_id);
2416 		if (ret == 1) {
2417 			desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2418 			if (oldest_start == 0) {	/* init all oldest_ values */
2419 				oldest_trans_id = get_desc_trans_id(desc);
2420 				oldest_start = d_bh->b_blocknr;
2421 				newest_mount_id = get_desc_mount_id(desc);
2422 				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2423 					       "journal-1179: Setting "
2424 					       "oldest_start to offset %llu, trans_id %lu",
2425 					       oldest_start -
2426 					       SB_ONDISK_JOURNAL_1st_BLOCK
2427 					       (sb), oldest_trans_id);
2428 			} else if (oldest_trans_id > get_desc_trans_id(desc)) {
2429 				/* one we just read was older */
2430 				oldest_trans_id = get_desc_trans_id(desc);
2431 				oldest_start = d_bh->b_blocknr;
2432 				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2433 					       "journal-1180: Resetting "
2434 					       "oldest_start to offset %lu, trans_id %lu",
2435 					       oldest_start -
2436 					       SB_ONDISK_JOURNAL_1st_BLOCK
2437 					       (sb), oldest_trans_id);
2438 			}
2439 			if (newest_mount_id < get_desc_mount_id(desc)) {
2440 				newest_mount_id = get_desc_mount_id(desc);
2441 				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2442 					       "journal-1299: Setting "
2443 					       "newest_mount_id to %d",
2444 					       get_desc_mount_id(desc));
2445 			}
2446 			cur_dblock += get_desc_trans_len(desc) + 2;
2447 		} else {
2448 			cur_dblock++;
2449 		}
2450 		brelse(d_bh);
2451 	}
2452 
2453       start_log_replay:
2454 	cur_dblock = oldest_start;
2455 	if (oldest_trans_id) {
2456 		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2457 			       "journal-1206: Starting replay "
2458 			       "from offset %llu, trans_id %lu",
2459 			       cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2460 			       oldest_trans_id);
2461 
2462 	}
2463 	replay_count = 0;
2464 	while (continue_replay && oldest_trans_id > 0) {
2465 		ret =
2466 		    journal_read_transaction(sb, cur_dblock, oldest_start,
2467 					     oldest_trans_id, newest_mount_id);
2468 		if (ret < 0) {
2469 			return ret;
2470 		} else if (ret != 0) {
2471 			break;
2472 		}
2473 		cur_dblock =
2474 		    SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2475 		replay_count++;
2476 		if (cur_dblock == oldest_start)
2477 			break;
2478 	}
2479 
2480 	if (oldest_trans_id == 0) {
2481 		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2482 			       "journal-1225: No valid " "transactions found");
2483 	}
2484 	/* j_start does not get set correctly if we don't replay any transactions.
2485 	 ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2486 	 ** copy the trans_id from the header
2487 	 */
2488 	if (valid_journal_header && replay_count == 0) {
2489 		journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2490 		journal->j_trans_id =
2491 		    le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2492 		/* check for trans_id overflow */
2493 		if (journal->j_trans_id == 0)
2494 			journal->j_trans_id = 10;
2495 		journal->j_last_flush_trans_id =
2496 		    le32_to_cpu(jh->j_last_flush_trans_id);
2497 		journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2498 	} else {
2499 		journal->j_mount_id = newest_mount_id + 1;
2500 	}
2501 	reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2502 		       "newest_mount_id to %lu", journal->j_mount_id);
2503 	journal->j_first_unflushed_offset = journal->j_start;
2504 	if (replay_count > 0) {
2505 		reiserfs_info(sb,
2506 			      "replayed %d transactions in %lu seconds\n",
2507 			      replay_count, get_seconds() - start);
2508 	}
2509 	/* needed to satisfy the locking in _update_journal_header_block */
2510 	reiserfs_write_lock(sb);
2511 	if (!bdev_read_only(sb->s_bdev) &&
2512 	    _update_journal_header_block(sb, journal->j_start,
2513 					 journal->j_last_flush_trans_id)) {
2514 		reiserfs_write_unlock(sb);
2515 		/* replay failed, caller must call free_journal_ram and abort
2516 		 ** the mount
2517 		 */
2518 		return -1;
2519 	}
2520 	reiserfs_write_unlock(sb);
2521 	return 0;
2522 }
2523 
2524 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2525 {
2526 	struct reiserfs_journal_list *jl;
2527 	jl = kzalloc(sizeof(struct reiserfs_journal_list),
2528 		     GFP_NOFS | __GFP_NOFAIL);
2529 	INIT_LIST_HEAD(&jl->j_list);
2530 	INIT_LIST_HEAD(&jl->j_working_list);
2531 	INIT_LIST_HEAD(&jl->j_tail_bh_list);
2532 	INIT_LIST_HEAD(&jl->j_bh_list);
2533 	mutex_init(&jl->j_commit_mutex);
2534 	SB_JOURNAL(s)->j_num_lists++;
2535 	get_journal_list(jl);
2536 	return jl;
2537 }
2538 
2539 static void journal_list_init(struct super_block *sb)
2540 {
2541 	SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2542 }
2543 
2544 static void release_journal_dev(struct super_block *super,
2545 			       struct reiserfs_journal *journal)
2546 {
2547 	if (journal->j_dev_bd != NULL) {
2548 		blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2549 		journal->j_dev_bd = NULL;
2550 	}
2551 }
2552 
2553 static int journal_init_dev(struct super_block *super,
2554 			    struct reiserfs_journal *journal,
2555 			    const char *jdev_name)
2556 {
2557 	int result;
2558 	dev_t jdev;
2559 	fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
2560 	char b[BDEVNAME_SIZE];
2561 
2562 	result = 0;
2563 
2564 	journal->j_dev_bd = NULL;
2565 	jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2566 	    new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2567 
2568 	if (bdev_read_only(super->s_bdev))
2569 		blkdev_mode = FMODE_READ;
2570 
2571 	/* there is no "jdev" option and journal is on separate device */
2572 	if ((!jdev_name || !jdev_name[0])) {
2573 		if (jdev == super->s_dev)
2574 			blkdev_mode &= ~FMODE_EXCL;
2575 		journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2576 						      journal);
2577 		journal->j_dev_mode = blkdev_mode;
2578 		if (IS_ERR(journal->j_dev_bd)) {
2579 			result = PTR_ERR(journal->j_dev_bd);
2580 			journal->j_dev_bd = NULL;
2581 			reiserfs_warning(super, "sh-458",
2582 					 "cannot init journal device '%s': %i",
2583 					 __bdevname(jdev, b), result);
2584 			return result;
2585 		} else if (jdev != super->s_dev)
2586 			set_blocksize(journal->j_dev_bd, super->s_blocksize);
2587 
2588 		return 0;
2589 	}
2590 
2591 	journal->j_dev_mode = blkdev_mode;
2592 	journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
2593 	if (IS_ERR(journal->j_dev_bd)) {
2594 		result = PTR_ERR(journal->j_dev_bd);
2595 		journal->j_dev_bd = NULL;
2596 		reiserfs_warning(super,
2597 				 "journal_init_dev: Cannot open '%s': %i",
2598 				 jdev_name, result);
2599 		return result;
2600 	}
2601 
2602 	set_blocksize(journal->j_dev_bd, super->s_blocksize);
2603 	reiserfs_info(super,
2604 		      "journal_init_dev: journal device: %s\n",
2605 		      bdevname(journal->j_dev_bd, b));
2606 	return 0;
2607 }
2608 
2609 /**
2610  * When creating/tuning a file system user can assign some
2611  * journal params within boundaries which depend on the ratio
2612  * blocksize/standard_blocksize.
2613  *
2614  * For blocks >= standard_blocksize transaction size should
2615  * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2616  * then JOURNAL_TRANS_MAX_DEFAULT.
2617  *
2618  * For blocks < standard_blocksize these boundaries should be
2619  * decreased proportionally.
2620  */
2621 #define REISERFS_STANDARD_BLKSIZE (4096)
2622 
2623 static int check_advise_trans_params(struct super_block *sb,
2624 				     struct reiserfs_journal *journal)
2625 {
2626         if (journal->j_trans_max) {
2627 	        /* Non-default journal params.
2628 		   Do sanity check for them. */
2629 	        int ratio = 1;
2630 		if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2631 		        ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2632 
2633 		if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2634 		    journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2635 		    SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2636 		    JOURNAL_MIN_RATIO) {
2637 			reiserfs_warning(sb, "sh-462",
2638 					 "bad transaction max size (%u). "
2639 					 "FSCK?", journal->j_trans_max);
2640 			return 1;
2641 		}
2642 		if (journal->j_max_batch != (journal->j_trans_max) *
2643 		        JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2644 			reiserfs_warning(sb, "sh-463",
2645 					 "bad transaction max batch (%u). "
2646 					 "FSCK?", journal->j_max_batch);
2647 			return 1;
2648 		}
2649 	} else {
2650 		/* Default journal params.
2651                    The file system was created by old version
2652 		   of mkreiserfs, so some fields contain zeros,
2653 		   and we need to advise proper values for them */
2654 		if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2655 			reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2656 					 sb->s_blocksize);
2657 			return 1;
2658 		}
2659 		journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2660 		journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2661 		journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2662 	}
2663 	return 0;
2664 }
2665 
2666 /*
2667 ** must be called once on fs mount.  calls journal_read for you
2668 */
2669 int journal_init(struct super_block *sb, const char *j_dev_name,
2670 		 int old_format, unsigned int commit_max_age)
2671 {
2672 	int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2673 	struct buffer_head *bhjh;
2674 	struct reiserfs_super_block *rs;
2675 	struct reiserfs_journal_header *jh;
2676 	struct reiserfs_journal *journal;
2677 	struct reiserfs_journal_list *jl;
2678 	char b[BDEVNAME_SIZE];
2679 	int ret;
2680 
2681 	journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal));
2682 	if (!journal) {
2683 		reiserfs_warning(sb, "journal-1256",
2684 				 "unable to get memory for journal structure");
2685 		return 1;
2686 	}
2687 	INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2688 	INIT_LIST_HEAD(&journal->j_prealloc_list);
2689 	INIT_LIST_HEAD(&journal->j_working_list);
2690 	INIT_LIST_HEAD(&journal->j_journal_list);
2691 	journal->j_persistent_trans = 0;
2692 	if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2693 					   reiserfs_bmap_count(sb)))
2694 		goto free_and_return;
2695 
2696 	allocate_bitmap_nodes(sb);
2697 
2698 	/* reserved for journal area support */
2699 	SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2700 						 REISERFS_OLD_DISK_OFFSET_IN_BYTES
2701 						 / sb->s_blocksize +
2702 						 reiserfs_bmap_count(sb) +
2703 						 1 :
2704 						 REISERFS_DISK_OFFSET_IN_BYTES /
2705 						 sb->s_blocksize + 2);
2706 
2707 	/* Sanity check to see is the standard journal fitting within first bitmap
2708 	   (actual for small blocksizes) */
2709 	if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2710 	    (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2711 	     SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2712 		reiserfs_warning(sb, "journal-1393",
2713 				 "journal does not fit for area addressed "
2714 				 "by first of bitmap blocks. It starts at "
2715 				 "%u and its size is %u. Block size %ld",
2716 				 SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2717 				 SB_ONDISK_JOURNAL_SIZE(sb),
2718 				 sb->s_blocksize);
2719 		goto free_and_return;
2720 	}
2721 
2722 	if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2723 		reiserfs_warning(sb, "sh-462",
2724 				 "unable to initialize jornal device");
2725 		goto free_and_return;
2726 	}
2727 
2728 	rs = SB_DISK_SUPER_BLOCK(sb);
2729 
2730 	/* read journal header */
2731 	bhjh = journal_bread(sb,
2732 			     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2733 			     SB_ONDISK_JOURNAL_SIZE(sb));
2734 	if (!bhjh) {
2735 		reiserfs_warning(sb, "sh-459",
2736 				 "unable to read journal header");
2737 		goto free_and_return;
2738 	}
2739 	jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2740 
2741 	/* make sure that journal matches to the super block */
2742 	if (is_reiserfs_jr(rs)
2743 	    && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2744 		sb_jp_journal_magic(rs))) {
2745 		reiserfs_warning(sb, "sh-460",
2746 				 "journal header magic %x (device %s) does "
2747 				 "not match to magic found in super block %x",
2748 				 jh->jh_journal.jp_journal_magic,
2749 				 bdevname(journal->j_dev_bd, b),
2750 				 sb_jp_journal_magic(rs));
2751 		brelse(bhjh);
2752 		goto free_and_return;
2753 	}
2754 
2755 	journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2756 	journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2757 	journal->j_max_commit_age =
2758 	    le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2759 	journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2760 
2761 	if (check_advise_trans_params(sb, journal) != 0)
2762 	        goto free_and_return;
2763 	journal->j_default_max_commit_age = journal->j_max_commit_age;
2764 
2765 	if (commit_max_age != 0) {
2766 		journal->j_max_commit_age = commit_max_age;
2767 		journal->j_max_trans_age = commit_max_age;
2768 	}
2769 
2770 	reiserfs_info(sb, "journal params: device %s, size %u, "
2771 		      "journal first block %u, max trans len %u, max batch %u, "
2772 		      "max commit age %u, max trans age %u\n",
2773 		      bdevname(journal->j_dev_bd, b),
2774 		      SB_ONDISK_JOURNAL_SIZE(sb),
2775 		      SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2776 		      journal->j_trans_max,
2777 		      journal->j_max_batch,
2778 		      journal->j_max_commit_age, journal->j_max_trans_age);
2779 
2780 	brelse(bhjh);
2781 
2782 	journal->j_list_bitmap_index = 0;
2783 	journal_list_init(sb);
2784 
2785 	memset(journal->j_list_hash_table, 0,
2786 	       JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2787 
2788 	INIT_LIST_HEAD(&journal->j_dirty_buffers);
2789 	spin_lock_init(&journal->j_dirty_buffers_lock);
2790 
2791 	journal->j_start = 0;
2792 	journal->j_len = 0;
2793 	journal->j_len_alloc = 0;
2794 	atomic_set(&(journal->j_wcount), 0);
2795 	atomic_set(&(journal->j_async_throttle), 0);
2796 	journal->j_bcount = 0;
2797 	journal->j_trans_start_time = 0;
2798 	journal->j_last = NULL;
2799 	journal->j_first = NULL;
2800 	init_waitqueue_head(&(journal->j_join_wait));
2801 	mutex_init(&journal->j_mutex);
2802 	mutex_init(&journal->j_flush_mutex);
2803 
2804 	journal->j_trans_id = 10;
2805 	journal->j_mount_id = 10;
2806 	journal->j_state = 0;
2807 	atomic_set(&(journal->j_jlock), 0);
2808 	journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2809 	journal->j_cnode_free_orig = journal->j_cnode_free_list;
2810 	journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2811 	journal->j_cnode_used = 0;
2812 	journal->j_must_wait = 0;
2813 
2814 	if (journal->j_cnode_free == 0) {
2815 		reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2816 		                 "allocation failed (%ld bytes). Journal is "
2817 		                 "too large for available memory. Usually "
2818 		                 "this is due to a journal that is too large.",
2819 		                 sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2820         	goto free_and_return;
2821 	}
2822 
2823 	init_journal_hash(sb);
2824 	jl = journal->j_current_jl;
2825 
2826 	/*
2827 	 * get_list_bitmap() may call flush_commit_list() which
2828 	 * requires the lock. Calling flush_commit_list() shouldn't happen
2829 	 * this early but I like to be paranoid.
2830 	 */
2831 	reiserfs_write_lock(sb);
2832 	jl->j_list_bitmap = get_list_bitmap(sb, jl);
2833 	reiserfs_write_unlock(sb);
2834 	if (!jl->j_list_bitmap) {
2835 		reiserfs_warning(sb, "journal-2005",
2836 				 "get_list_bitmap failed for journal list 0");
2837 		goto free_and_return;
2838 	}
2839 
2840 	ret = journal_read(sb);
2841 	if (ret < 0) {
2842 		reiserfs_warning(sb, "reiserfs-2006",
2843 				 "Replay Failure, unable to mount");
2844 		goto free_and_return;
2845 	}
2846 
2847 	reiserfs_mounted_fs_count++;
2848 	if (reiserfs_mounted_fs_count <= 1)
2849 		commit_wq = alloc_workqueue("reiserfs", WQ_MEM_RECLAIM, 0);
2850 
2851 	INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2852 	journal->j_work_sb = sb;
2853 	return 0;
2854       free_and_return:
2855 	free_journal_ram(sb);
2856 	return 1;
2857 }
2858 
2859 /*
2860 ** test for a polite end of the current transaction.  Used by file_write, and should
2861 ** be used by delete to make sure they don't write more than can fit inside a single
2862 ** transaction
2863 */
2864 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2865 				   int new_alloc)
2866 {
2867 	struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2868 	time_t now = get_seconds();
2869 	/* cannot restart while nested */
2870 	BUG_ON(!th->t_trans_id);
2871 	if (th->t_refcount > 1)
2872 		return 0;
2873 	if (journal->j_must_wait > 0 ||
2874 	    (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2875 	    atomic_read(&(journal->j_jlock)) ||
2876 	    (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2877 	    journal->j_cnode_free < (journal->j_trans_max * 3)) {
2878 		return 1;
2879 	}
2880 
2881 	journal->j_len_alloc += new_alloc;
2882 	th->t_blocks_allocated += new_alloc ;
2883 	return 0;
2884 }
2885 
2886 /* this must be called inside a transaction
2887 */
2888 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2889 {
2890 	struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2891 	BUG_ON(!th->t_trans_id);
2892 	journal->j_must_wait = 1;
2893 	set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2894 	return;
2895 }
2896 
2897 /* this must be called without a transaction started
2898 */
2899 void reiserfs_allow_writes(struct super_block *s)
2900 {
2901 	struct reiserfs_journal *journal = SB_JOURNAL(s);
2902 	clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2903 	wake_up(&journal->j_join_wait);
2904 }
2905 
2906 /* this must be called without a transaction started
2907 */
2908 void reiserfs_wait_on_write_block(struct super_block *s)
2909 {
2910 	struct reiserfs_journal *journal = SB_JOURNAL(s);
2911 	wait_event(journal->j_join_wait,
2912 		   !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2913 }
2914 
2915 static void queue_log_writer(struct super_block *s)
2916 {
2917 	wait_queue_t wait;
2918 	struct reiserfs_journal *journal = SB_JOURNAL(s);
2919 	set_bit(J_WRITERS_QUEUED, &journal->j_state);
2920 
2921 	/*
2922 	 * we don't want to use wait_event here because
2923 	 * we only want to wait once.
2924 	 */
2925 	init_waitqueue_entry(&wait, current);
2926 	add_wait_queue(&journal->j_join_wait, &wait);
2927 	set_current_state(TASK_UNINTERRUPTIBLE);
2928 	if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
2929 		int depth = reiserfs_write_unlock_nested(s);
2930 		schedule();
2931 		reiserfs_write_lock_nested(s, depth);
2932 	}
2933 	__set_current_state(TASK_RUNNING);
2934 	remove_wait_queue(&journal->j_join_wait, &wait);
2935 }
2936 
2937 static void wake_queued_writers(struct super_block *s)
2938 {
2939 	struct reiserfs_journal *journal = SB_JOURNAL(s);
2940 	if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2941 		wake_up(&journal->j_join_wait);
2942 }
2943 
2944 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
2945 {
2946 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
2947 	unsigned long bcount = journal->j_bcount;
2948 	while (1) {
2949 		int depth;
2950 
2951 		depth = reiserfs_write_unlock_nested(sb);
2952 		schedule_timeout_uninterruptible(1);
2953 		reiserfs_write_lock_nested(sb, depth);
2954 
2955 		journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2956 		while ((atomic_read(&journal->j_wcount) > 0 ||
2957 			atomic_read(&journal->j_jlock)) &&
2958 		       journal->j_trans_id == trans_id) {
2959 			queue_log_writer(sb);
2960 		}
2961 		if (journal->j_trans_id != trans_id)
2962 			break;
2963 		if (bcount == journal->j_bcount)
2964 			break;
2965 		bcount = journal->j_bcount;
2966 	}
2967 }
2968 
2969 /* join == true if you must join an existing transaction.
2970 ** join == false if you can deal with waiting for others to finish
2971 **
2972 ** this will block until the transaction is joinable.  send the number of blocks you
2973 ** expect to use in nblocks.
2974 */
2975 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
2976 			      struct super_block *sb, unsigned long nblocks,
2977 			      int join)
2978 {
2979 	time_t now = get_seconds();
2980 	unsigned int old_trans_id;
2981 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
2982 	struct reiserfs_transaction_handle myth;
2983 	int sched_count = 0;
2984 	int retval;
2985 	int depth;
2986 
2987 	reiserfs_check_lock_depth(sb, "journal_begin");
2988 	BUG_ON(nblocks > journal->j_trans_max);
2989 
2990 	PROC_INFO_INC(sb, journal.journal_being);
2991 	/* set here for journal_join */
2992 	th->t_refcount = 1;
2993 	th->t_super = sb;
2994 
2995       relock:
2996 	lock_journal(sb);
2997 	if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
2998 		unlock_journal(sb);
2999 		retval = journal->j_errno;
3000 		goto out_fail;
3001 	}
3002 	journal->j_bcount++;
3003 
3004 	if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3005 		unlock_journal(sb);
3006 		depth = reiserfs_write_unlock_nested(sb);
3007 		reiserfs_wait_on_write_block(sb);
3008 		reiserfs_write_lock_nested(sb, depth);
3009 		PROC_INFO_INC(sb, journal.journal_relock_writers);
3010 		goto relock;
3011 	}
3012 	now = get_seconds();
3013 
3014 	/* if there is no room in the journal OR
3015 	 ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
3016 	 ** we don't sleep if there aren't other writers
3017 	 */
3018 
3019 	if ((!join && journal->j_must_wait > 0) ||
3020 	    (!join
3021 	     && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3022 	    || (!join && atomic_read(&journal->j_wcount) > 0
3023 		&& journal->j_trans_start_time > 0
3024 		&& (now - journal->j_trans_start_time) >
3025 		journal->j_max_trans_age) || (!join
3026 					      && atomic_read(&journal->j_jlock))
3027 	    || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3028 
3029 		old_trans_id = journal->j_trans_id;
3030 		unlock_journal(sb);	/* allow others to finish this transaction */
3031 
3032 		if (!join && (journal->j_len_alloc + nblocks + 2) >=
3033 		    journal->j_max_batch &&
3034 		    ((journal->j_len + nblocks + 2) * 100) <
3035 		    (journal->j_len_alloc * 75)) {
3036 			if (atomic_read(&journal->j_wcount) > 10) {
3037 				sched_count++;
3038 				queue_log_writer(sb);
3039 				goto relock;
3040 			}
3041 		}
3042 		/* don't mess with joining the transaction if all we have to do is
3043 		 * wait for someone else to do a commit
3044 		 */
3045 		if (atomic_read(&journal->j_jlock)) {
3046 			while (journal->j_trans_id == old_trans_id &&
3047 			       atomic_read(&journal->j_jlock)) {
3048 				queue_log_writer(sb);
3049 			}
3050 			goto relock;
3051 		}
3052 		retval = journal_join(&myth, sb, 1);
3053 		if (retval)
3054 			goto out_fail;
3055 
3056 		/* someone might have ended the transaction while we joined */
3057 		if (old_trans_id != journal->j_trans_id) {
3058 			retval = do_journal_end(&myth, sb, 1, 0);
3059 		} else {
3060 			retval = do_journal_end(&myth, sb, 1, COMMIT_NOW);
3061 		}
3062 
3063 		if (retval)
3064 			goto out_fail;
3065 
3066 		PROC_INFO_INC(sb, journal.journal_relock_wcount);
3067 		goto relock;
3068 	}
3069 	/* we are the first writer, set trans_id */
3070 	if (journal->j_trans_start_time == 0) {
3071 		journal->j_trans_start_time = get_seconds();
3072 	}
3073 	atomic_inc(&(journal->j_wcount));
3074 	journal->j_len_alloc += nblocks;
3075 	th->t_blocks_logged = 0;
3076 	th->t_blocks_allocated = nblocks;
3077 	th->t_trans_id = journal->j_trans_id;
3078 	unlock_journal(sb);
3079 	INIT_LIST_HEAD(&th->t_list);
3080 	return 0;
3081 
3082       out_fail:
3083 	memset(th, 0, sizeof(*th));
3084 	/* Re-set th->t_super, so we can properly keep track of how many
3085 	 * persistent transactions there are. We need to do this so if this
3086 	 * call is part of a failed restart_transaction, we can free it later */
3087 	th->t_super = sb;
3088 	return retval;
3089 }
3090 
3091 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3092 								    super_block
3093 								    *s,
3094 								    int nblocks)
3095 {
3096 	int ret;
3097 	struct reiserfs_transaction_handle *th;
3098 
3099 	/* if we're nesting into an existing transaction.  It will be
3100 	 ** persistent on its own
3101 	 */
3102 	if (reiserfs_transaction_running(s)) {
3103 		th = current->journal_info;
3104 		th->t_refcount++;
3105 		BUG_ON(th->t_refcount < 2);
3106 
3107 		return th;
3108 	}
3109 	th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3110 	if (!th)
3111 		return NULL;
3112 	ret = journal_begin(th, s, nblocks);
3113 	if (ret) {
3114 		kfree(th);
3115 		return NULL;
3116 	}
3117 
3118 	SB_JOURNAL(s)->j_persistent_trans++;
3119 	return th;
3120 }
3121 
3122 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3123 {
3124 	struct super_block *s = th->t_super;
3125 	int ret = 0;
3126 	if (th->t_trans_id)
3127 		ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3128 	else
3129 		ret = -EIO;
3130 	if (th->t_refcount == 0) {
3131 		SB_JOURNAL(s)->j_persistent_trans--;
3132 		kfree(th);
3133 	}
3134 	return ret;
3135 }
3136 
3137 static int journal_join(struct reiserfs_transaction_handle *th,
3138 			struct super_block *sb, unsigned long nblocks)
3139 {
3140 	struct reiserfs_transaction_handle *cur_th = current->journal_info;
3141 
3142 	/* this keeps do_journal_end from NULLing out the current->journal_info
3143 	 ** pointer
3144 	 */
3145 	th->t_handle_save = cur_th;
3146 	BUG_ON(cur_th && cur_th->t_refcount > 1);
3147 	return do_journal_begin_r(th, sb, nblocks, JBEGIN_JOIN);
3148 }
3149 
3150 int journal_join_abort(struct reiserfs_transaction_handle *th,
3151 		       struct super_block *sb, unsigned long nblocks)
3152 {
3153 	struct reiserfs_transaction_handle *cur_th = current->journal_info;
3154 
3155 	/* this keeps do_journal_end from NULLing out the current->journal_info
3156 	 ** pointer
3157 	 */
3158 	th->t_handle_save = cur_th;
3159 	BUG_ON(cur_th && cur_th->t_refcount > 1);
3160 	return do_journal_begin_r(th, sb, nblocks, JBEGIN_ABORT);
3161 }
3162 
3163 int journal_begin(struct reiserfs_transaction_handle *th,
3164 		  struct super_block *sb, unsigned long nblocks)
3165 {
3166 	struct reiserfs_transaction_handle *cur_th = current->journal_info;
3167 	int ret;
3168 
3169 	th->t_handle_save = NULL;
3170 	if (cur_th) {
3171 		/* we are nesting into the current transaction */
3172 		if (cur_th->t_super == sb) {
3173 			BUG_ON(!cur_th->t_refcount);
3174 			cur_th->t_refcount++;
3175 			memcpy(th, cur_th, sizeof(*th));
3176 			if (th->t_refcount <= 1)
3177 				reiserfs_warning(sb, "reiserfs-2005",
3178 						 "BAD: refcount <= 1, but "
3179 						 "journal_info != 0");
3180 			return 0;
3181 		} else {
3182 			/* we've ended up with a handle from a different filesystem.
3183 			 ** save it and restore on journal_end.  This should never
3184 			 ** really happen...
3185 			 */
3186 			reiserfs_warning(sb, "clm-2100",
3187 					 "nesting info a different FS");
3188 			th->t_handle_save = current->journal_info;
3189 			current->journal_info = th;
3190 		}
3191 	} else {
3192 		current->journal_info = th;
3193 	}
3194 	ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3195 	BUG_ON(current->journal_info != th);
3196 
3197 	/* I guess this boils down to being the reciprocal of clm-2100 above.
3198 	 * If do_journal_begin_r fails, we need to put it back, since journal_end
3199 	 * won't be called to do it. */
3200 	if (ret)
3201 		current->journal_info = th->t_handle_save;
3202 	else
3203 		BUG_ON(!th->t_refcount);
3204 
3205 	return ret;
3206 }
3207 
3208 /*
3209 ** puts bh into the current transaction.  If it was already there, reorders removes the
3210 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3211 **
3212 ** if it was dirty, cleans and files onto the clean list.  I can't let it be dirty again until the
3213 ** transaction is committed.
3214 **
3215 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3216 */
3217 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3218 		       struct super_block *sb, struct buffer_head *bh)
3219 {
3220 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3221 	struct reiserfs_journal_cnode *cn = NULL;
3222 	int count_already_incd = 0;
3223 	int prepared = 0;
3224 	BUG_ON(!th->t_trans_id);
3225 
3226 	PROC_INFO_INC(sb, journal.mark_dirty);
3227 	if (th->t_trans_id != journal->j_trans_id) {
3228 		reiserfs_panic(th->t_super, "journal-1577",
3229 			       "handle trans id %ld != current trans id %ld",
3230 			       th->t_trans_id, journal->j_trans_id);
3231 	}
3232 
3233 	prepared = test_clear_buffer_journal_prepared(bh);
3234 	clear_buffer_journal_restore_dirty(bh);
3235 	/* already in this transaction, we are done */
3236 	if (buffer_journaled(bh)) {
3237 		PROC_INFO_INC(sb, journal.mark_dirty_already);
3238 		return 0;
3239 	}
3240 
3241 	/* this must be turned into a panic instead of a warning.  We can't allow
3242 	 ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3243 	 ** could get to disk too early.  NOT GOOD.
3244 	 */
3245 	if (!prepared || buffer_dirty(bh)) {
3246 		reiserfs_warning(sb, "journal-1777",
3247 				 "buffer %llu bad state "
3248 				 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3249 				 (unsigned long long)bh->b_blocknr,
3250 				 prepared ? ' ' : '!',
3251 				 buffer_locked(bh) ? ' ' : '!',
3252 				 buffer_dirty(bh) ? ' ' : '!',
3253 				 buffer_journal_dirty(bh) ? ' ' : '!');
3254 	}
3255 
3256 	if (atomic_read(&(journal->j_wcount)) <= 0) {
3257 		reiserfs_warning(sb, "journal-1409",
3258 				 "returning because j_wcount was %d",
3259 				 atomic_read(&(journal->j_wcount)));
3260 		return 1;
3261 	}
3262 	/* this error means I've screwed up, and we've overflowed the transaction.
3263 	 ** Nothing can be done here, except make the FS readonly or panic.
3264 	 */
3265 	if (journal->j_len >= journal->j_trans_max) {
3266 		reiserfs_panic(th->t_super, "journal-1413",
3267 			       "j_len (%lu) is too big",
3268 			       journal->j_len);
3269 	}
3270 
3271 	if (buffer_journal_dirty(bh)) {
3272 		count_already_incd = 1;
3273 		PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3274 		clear_buffer_journal_dirty(bh);
3275 	}
3276 
3277 	if (journal->j_len > journal->j_len_alloc) {
3278 		journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3279 	}
3280 
3281 	set_buffer_journaled(bh);
3282 
3283 	/* now put this guy on the end */
3284 	if (!cn) {
3285 		cn = get_cnode(sb);
3286 		if (!cn) {
3287 			reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3288 		}
3289 
3290 		if (th->t_blocks_logged == th->t_blocks_allocated) {
3291 			th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3292 			journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3293 		}
3294 		th->t_blocks_logged++;
3295 		journal->j_len++;
3296 
3297 		cn->bh = bh;
3298 		cn->blocknr = bh->b_blocknr;
3299 		cn->sb = sb;
3300 		cn->jlist = NULL;
3301 		insert_journal_hash(journal->j_hash_table, cn);
3302 		if (!count_already_incd) {
3303 			get_bh(bh);
3304 		}
3305 	}
3306 	cn->next = NULL;
3307 	cn->prev = journal->j_last;
3308 	cn->bh = bh;
3309 	if (journal->j_last) {
3310 		journal->j_last->next = cn;
3311 		journal->j_last = cn;
3312 	} else {
3313 		journal->j_first = cn;
3314 		journal->j_last = cn;
3315 	}
3316 	reiserfs_schedule_old_flush(sb);
3317 	return 0;
3318 }
3319 
3320 int journal_end(struct reiserfs_transaction_handle *th,
3321 		struct super_block *sb, unsigned long nblocks)
3322 {
3323 	if (!current->journal_info && th->t_refcount > 1)
3324 		reiserfs_warning(sb, "REISER-NESTING",
3325 				 "th NULL, refcount %d", th->t_refcount);
3326 
3327 	if (!th->t_trans_id) {
3328 		WARN_ON(1);
3329 		return -EIO;
3330 	}
3331 
3332 	th->t_refcount--;
3333 	if (th->t_refcount > 0) {
3334 		struct reiserfs_transaction_handle *cur_th =
3335 		    current->journal_info;
3336 
3337 		/* we aren't allowed to close a nested transaction on a different
3338 		 ** filesystem from the one in the task struct
3339 		 */
3340 		BUG_ON(cur_th->t_super != th->t_super);
3341 
3342 		if (th != cur_th) {
3343 			memcpy(current->journal_info, th, sizeof(*th));
3344 			th->t_trans_id = 0;
3345 		}
3346 		return 0;
3347 	} else {
3348 		return do_journal_end(th, sb, nblocks, 0);
3349 	}
3350 }
3351 
3352 /* removes from the current transaction, relsing and descrementing any counters.
3353 ** also files the removed buffer directly onto the clean list
3354 **
3355 ** called by journal_mark_freed when a block has been deleted
3356 **
3357 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3358 */
3359 static int remove_from_transaction(struct super_block *sb,
3360 				   b_blocknr_t blocknr, int already_cleaned)
3361 {
3362 	struct buffer_head *bh;
3363 	struct reiserfs_journal_cnode *cn;
3364 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3365 	int ret = 0;
3366 
3367 	cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3368 	if (!cn || !cn->bh) {
3369 		return ret;
3370 	}
3371 	bh = cn->bh;
3372 	if (cn->prev) {
3373 		cn->prev->next = cn->next;
3374 	}
3375 	if (cn->next) {
3376 		cn->next->prev = cn->prev;
3377 	}
3378 	if (cn == journal->j_first) {
3379 		journal->j_first = cn->next;
3380 	}
3381 	if (cn == journal->j_last) {
3382 		journal->j_last = cn->prev;
3383 	}
3384 	if (bh)
3385 		remove_journal_hash(sb, journal->j_hash_table, NULL,
3386 				    bh->b_blocknr, 0);
3387 	clear_buffer_journaled(bh);	/* don't log this one */
3388 
3389 	if (!already_cleaned) {
3390 		clear_buffer_journal_dirty(bh);
3391 		clear_buffer_dirty(bh);
3392 		clear_buffer_journal_test(bh);
3393 		put_bh(bh);
3394 		if (atomic_read(&(bh->b_count)) < 0) {
3395 			reiserfs_warning(sb, "journal-1752",
3396 					 "b_count < 0");
3397 		}
3398 		ret = 1;
3399 	}
3400 	journal->j_len--;
3401 	journal->j_len_alloc--;
3402 	free_cnode(sb, cn);
3403 	return ret;
3404 }
3405 
3406 /*
3407 ** for any cnode in a journal list, it can only be dirtied of all the
3408 ** transactions that include it are committed to disk.
3409 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3410 ** and 0 if you aren't
3411 **
3412 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3413 ** blocks for a given transaction on disk
3414 **
3415 */
3416 static int can_dirty(struct reiserfs_journal_cnode *cn)
3417 {
3418 	struct super_block *sb = cn->sb;
3419 	b_blocknr_t blocknr = cn->blocknr;
3420 	struct reiserfs_journal_cnode *cur = cn->hprev;
3421 	int can_dirty = 1;
3422 
3423 	/* first test hprev.  These are all newer than cn, so any node here
3424 	 ** with the same block number and dev means this node can't be sent
3425 	 ** to disk right now.
3426 	 */
3427 	while (cur && can_dirty) {
3428 		if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3429 		    cur->blocknr == blocknr) {
3430 			can_dirty = 0;
3431 		}
3432 		cur = cur->hprev;
3433 	}
3434 	/* then test hnext.  These are all older than cn.  As long as they
3435 	 ** are committed to the log, it is safe to write cn to disk
3436 	 */
3437 	cur = cn->hnext;
3438 	while (cur && can_dirty) {
3439 		if (cur->jlist && cur->jlist->j_len > 0 &&
3440 		    atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3441 		    cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3442 			can_dirty = 0;
3443 		}
3444 		cur = cur->hnext;
3445 	}
3446 	return can_dirty;
3447 }
3448 
3449 /* syncs the commit blocks, but does not force the real buffers to disk
3450 ** will wait until the current transaction is done/committed before returning
3451 */
3452 int journal_end_sync(struct reiserfs_transaction_handle *th,
3453 		     struct super_block *sb, unsigned long nblocks)
3454 {
3455 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3456 
3457 	BUG_ON(!th->t_trans_id);
3458 	/* you can sync while nested, very, very bad */
3459 	BUG_ON(th->t_refcount > 1);
3460 	if (journal->j_len == 0) {
3461 		reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3462 					     1);
3463 		journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
3464 	}
3465 	return do_journal_end(th, sb, nblocks, COMMIT_NOW | WAIT);
3466 }
3467 
3468 /*
3469 ** writeback the pending async commits to disk
3470 */
3471 static void flush_async_commits(struct work_struct *work)
3472 {
3473 	struct reiserfs_journal *journal =
3474 		container_of(work, struct reiserfs_journal, j_work.work);
3475 	struct super_block *sb = journal->j_work_sb;
3476 	struct reiserfs_journal_list *jl;
3477 	struct list_head *entry;
3478 
3479 	reiserfs_write_lock(sb);
3480 	if (!list_empty(&journal->j_journal_list)) {
3481 		/* last entry is the youngest, commit it and you get everything */
3482 		entry = journal->j_journal_list.prev;
3483 		jl = JOURNAL_LIST_ENTRY(entry);
3484 		flush_commit_list(sb, jl, 1);
3485 	}
3486 	reiserfs_write_unlock(sb);
3487 }
3488 
3489 /*
3490 ** flushes any old transactions to disk
3491 ** ends the current transaction if it is too old
3492 */
3493 void reiserfs_flush_old_commits(struct super_block *sb)
3494 {
3495 	time_t now;
3496 	struct reiserfs_transaction_handle th;
3497 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3498 
3499 	now = get_seconds();
3500 	/* safety check so we don't flush while we are replaying the log during
3501 	 * mount
3502 	 */
3503 	if (list_empty(&journal->j_journal_list))
3504 		return;
3505 
3506 	/* check the current transaction.  If there are no writers, and it is
3507 	 * too old, finish it, and force the commit blocks to disk
3508 	 */
3509 	if (atomic_read(&journal->j_wcount) <= 0 &&
3510 	    journal->j_trans_start_time > 0 &&
3511 	    journal->j_len > 0 &&
3512 	    (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3513 		if (!journal_join(&th, sb, 1)) {
3514 			reiserfs_prepare_for_journal(sb,
3515 						     SB_BUFFER_WITH_SB(sb),
3516 						     1);
3517 			journal_mark_dirty(&th, sb,
3518 					   SB_BUFFER_WITH_SB(sb));
3519 
3520 			/* we're only being called from kreiserfsd, it makes no sense to do
3521 			 ** an async commit so that kreiserfsd can do it later
3522 			 */
3523 			do_journal_end(&th, sb, 1, COMMIT_NOW | WAIT);
3524 		}
3525 	}
3526 }
3527 
3528 /*
3529 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3530 **
3531 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3532 ** the writers are done.  By the time it wakes up, the transaction it was called has already ended, so it just
3533 ** flushes the commit list and returns 0.
3534 **
3535 ** Won't batch when flush or commit_now is set.  Also won't batch when others are waiting on j_join_wait.
3536 **
3537 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3538 */
3539 static int check_journal_end(struct reiserfs_transaction_handle *th,
3540 			     struct super_block *sb, unsigned long nblocks,
3541 			     int flags)
3542 {
3543 
3544 	time_t now;
3545 	int flush = flags & FLUSH_ALL;
3546 	int commit_now = flags & COMMIT_NOW;
3547 	int wait_on_commit = flags & WAIT;
3548 	struct reiserfs_journal_list *jl;
3549 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3550 
3551 	BUG_ON(!th->t_trans_id);
3552 
3553 	if (th->t_trans_id != journal->j_trans_id) {
3554 		reiserfs_panic(th->t_super, "journal-1577",
3555 			       "handle trans id %ld != current trans id %ld",
3556 			       th->t_trans_id, journal->j_trans_id);
3557 	}
3558 
3559 	journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3560 	if (atomic_read(&(journal->j_wcount)) > 0) {	/* <= 0 is allowed.  unmounting might not call begin */
3561 		atomic_dec(&(journal->j_wcount));
3562 	}
3563 
3564 	/* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3565 	 ** will be dealt with by next transaction that actually writes something, but should be taken
3566 	 ** care of in this trans
3567 	 */
3568 	BUG_ON(journal->j_len == 0);
3569 
3570 	/* if wcount > 0, and we are called to with flush or commit_now,
3571 	 ** we wait on j_join_wait.  We will wake up when the last writer has
3572 	 ** finished the transaction, and started it on its way to the disk.
3573 	 ** Then, we flush the commit or journal list, and just return 0
3574 	 ** because the rest of journal end was already done for this transaction.
3575 	 */
3576 	if (atomic_read(&(journal->j_wcount)) > 0) {
3577 		if (flush || commit_now) {
3578 			unsigned trans_id;
3579 
3580 			jl = journal->j_current_jl;
3581 			trans_id = jl->j_trans_id;
3582 			if (wait_on_commit)
3583 				jl->j_state |= LIST_COMMIT_PENDING;
3584 			atomic_set(&(journal->j_jlock), 1);
3585 			if (flush) {
3586 				journal->j_next_full_flush = 1;
3587 			}
3588 			unlock_journal(sb);
3589 
3590 			/* sleep while the current transaction is still j_jlocked */
3591 			while (journal->j_trans_id == trans_id) {
3592 				if (atomic_read(&journal->j_jlock)) {
3593 					queue_log_writer(sb);
3594 				} else {
3595 					lock_journal(sb);
3596 					if (journal->j_trans_id == trans_id) {
3597 						atomic_set(&(journal->j_jlock),
3598 							   1);
3599 					}
3600 					unlock_journal(sb);
3601 				}
3602 			}
3603 			BUG_ON(journal->j_trans_id == trans_id);
3604 
3605 			if (commit_now
3606 			    && journal_list_still_alive(sb, trans_id)
3607 			    && wait_on_commit) {
3608 				flush_commit_list(sb, jl, 1);
3609 			}
3610 			return 0;
3611 		}
3612 		unlock_journal(sb);
3613 		return 0;
3614 	}
3615 
3616 	/* deal with old transactions where we are the last writers */
3617 	now = get_seconds();
3618 	if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3619 		commit_now = 1;
3620 		journal->j_next_async_flush = 1;
3621 	}
3622 	/* don't batch when someone is waiting on j_join_wait */
3623 	/* don't batch when syncing the commit or flushing the whole trans */
3624 	if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3625 	    && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3626 	    && journal->j_len_alloc < journal->j_max_batch
3627 	    && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3628 		journal->j_bcount++;
3629 		unlock_journal(sb);
3630 		return 0;
3631 	}
3632 
3633 	if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3634 		reiserfs_panic(sb, "journal-003",
3635 			       "j_start (%ld) is too high",
3636 			       journal->j_start);
3637 	}
3638 	return 1;
3639 }
3640 
3641 /*
3642 ** Does all the work that makes deleting blocks safe.
3643 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3644 **
3645 ** otherwise:
3646 ** set a bit for the block in the journal bitmap.  That will prevent it from being allocated for unformatted nodes
3647 ** before this transaction has finished.
3648 **
3649 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.  That will prevent any old transactions with
3650 ** this block from trying to flush to the real location.  Since we aren't removing the cnode from the journal_list_hash,
3651 ** the block can't be reallocated yet.
3652 **
3653 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3654 */
3655 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3656 		       struct super_block *sb, b_blocknr_t blocknr)
3657 {
3658 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3659 	struct reiserfs_journal_cnode *cn = NULL;
3660 	struct buffer_head *bh = NULL;
3661 	struct reiserfs_list_bitmap *jb = NULL;
3662 	int cleaned = 0;
3663 	BUG_ON(!th->t_trans_id);
3664 
3665 	cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3666 	if (cn && cn->bh) {
3667 		bh = cn->bh;
3668 		get_bh(bh);
3669 	}
3670 	/* if it is journal new, we just remove it from this transaction */
3671 	if (bh && buffer_journal_new(bh)) {
3672 		clear_buffer_journal_new(bh);
3673 		clear_prepared_bits(bh);
3674 		reiserfs_clean_and_file_buffer(bh);
3675 		cleaned = remove_from_transaction(sb, blocknr, cleaned);
3676 	} else {
3677 		/* set the bit for this block in the journal bitmap for this transaction */
3678 		jb = journal->j_current_jl->j_list_bitmap;
3679 		if (!jb) {
3680 			reiserfs_panic(sb, "journal-1702",
3681 				       "journal_list_bitmap is NULL");
3682 		}
3683 		set_bit_in_list_bitmap(sb, blocknr, jb);
3684 
3685 		/* Note, the entire while loop is not allowed to schedule.  */
3686 
3687 		if (bh) {
3688 			clear_prepared_bits(bh);
3689 			reiserfs_clean_and_file_buffer(bh);
3690 		}
3691 		cleaned = remove_from_transaction(sb, blocknr, cleaned);
3692 
3693 		/* find all older transactions with this block, make sure they don't try to write it out */
3694 		cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3695 					  blocknr);
3696 		while (cn) {
3697 			if (sb == cn->sb && blocknr == cn->blocknr) {
3698 				set_bit(BLOCK_FREED, &cn->state);
3699 				if (cn->bh) {
3700 					if (!cleaned) {
3701 						/* remove_from_transaction will brelse the buffer if it was
3702 						 ** in the current trans
3703 						 */
3704 						clear_buffer_journal_dirty(cn->
3705 									   bh);
3706 						clear_buffer_dirty(cn->bh);
3707 						clear_buffer_journal_test(cn->
3708 									  bh);
3709 						cleaned = 1;
3710 						put_bh(cn->bh);
3711 						if (atomic_read
3712 						    (&(cn->bh->b_count)) < 0) {
3713 							reiserfs_warning(sb,
3714 								 "journal-2138",
3715 								 "cn->bh->b_count < 0");
3716 						}
3717 					}
3718 					if (cn->jlist) {	/* since we are clearing the bh, we MUST dec nonzerolen */
3719 						atomic_dec(&
3720 							   (cn->jlist->
3721 							    j_nonzerolen));
3722 					}
3723 					cn->bh = NULL;
3724 				}
3725 			}
3726 			cn = cn->hnext;
3727 		}
3728 	}
3729 
3730 	if (bh)
3731 		release_buffer_page(bh); /* get_hash grabs the buffer */
3732 	return 0;
3733 }
3734 
3735 void reiserfs_update_inode_transaction(struct inode *inode)
3736 {
3737 	struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3738 	REISERFS_I(inode)->i_jl = journal->j_current_jl;
3739 	REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3740 }
3741 
3742 /*
3743  * returns -1 on error, 0 if no commits/barriers were done and 1
3744  * if a transaction was actually committed and the barrier was done
3745  */
3746 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3747 			     struct reiserfs_journal_list *jl)
3748 {
3749 	struct reiserfs_transaction_handle th;
3750 	struct super_block *sb = inode->i_sb;
3751 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3752 	int ret = 0;
3753 
3754 	/* is it from the current transaction, or from an unknown transaction? */
3755 	if (id == journal->j_trans_id) {
3756 		jl = journal->j_current_jl;
3757 		/* try to let other writers come in and grow this transaction */
3758 		let_transaction_grow(sb, id);
3759 		if (journal->j_trans_id != id) {
3760 			goto flush_commit_only;
3761 		}
3762 
3763 		ret = journal_begin(&th, sb, 1);
3764 		if (ret)
3765 			return ret;
3766 
3767 		/* someone might have ended this transaction while we joined */
3768 		if (journal->j_trans_id != id) {
3769 			reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3770 						     1);
3771 			journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3772 			ret = journal_end(&th, sb, 1);
3773 			goto flush_commit_only;
3774 		}
3775 
3776 		ret = journal_end_sync(&th, sb, 1);
3777 		if (!ret)
3778 			ret = 1;
3779 
3780 	} else {
3781 		/* this gets tricky, we have to make sure the journal list in
3782 		 * the inode still exists.  We know the list is still around
3783 		 * if we've got a larger transaction id than the oldest list
3784 		 */
3785 	      flush_commit_only:
3786 		if (journal_list_still_alive(inode->i_sb, id)) {
3787 			/*
3788 			 * we only set ret to 1 when we know for sure
3789 			 * the barrier hasn't been started yet on the commit
3790 			 * block.
3791 			 */
3792 			if (atomic_read(&jl->j_commit_left) > 1)
3793 				ret = 1;
3794 			flush_commit_list(sb, jl, 1);
3795 			if (journal->j_errno)
3796 				ret = journal->j_errno;
3797 		}
3798 	}
3799 	/* otherwise the list is gone, and long since committed */
3800 	return ret;
3801 }
3802 
3803 int reiserfs_commit_for_inode(struct inode *inode)
3804 {
3805 	unsigned int id = REISERFS_I(inode)->i_trans_id;
3806 	struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3807 
3808 	/* for the whole inode, assume unset id means it was
3809 	 * changed in the current transaction.  More conservative
3810 	 */
3811 	if (!id || !jl) {
3812 		reiserfs_update_inode_transaction(inode);
3813 		id = REISERFS_I(inode)->i_trans_id;
3814 		/* jl will be updated in __commit_trans_jl */
3815 	}
3816 
3817 	return __commit_trans_jl(inode, id, jl);
3818 }
3819 
3820 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3821 				      struct buffer_head *bh)
3822 {
3823 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3824 	PROC_INFO_INC(sb, journal.restore_prepared);
3825 	if (!bh) {
3826 		return;
3827 	}
3828 	if (test_clear_buffer_journal_restore_dirty(bh) &&
3829 	    buffer_journal_dirty(bh)) {
3830 		struct reiserfs_journal_cnode *cn;
3831 		reiserfs_write_lock(sb);
3832 		cn = get_journal_hash_dev(sb,
3833 					  journal->j_list_hash_table,
3834 					  bh->b_blocknr);
3835 		if (cn && can_dirty(cn)) {
3836 			set_buffer_journal_test(bh);
3837 			mark_buffer_dirty(bh);
3838 		}
3839 		reiserfs_write_unlock(sb);
3840 	}
3841 	clear_buffer_journal_prepared(bh);
3842 }
3843 
3844 extern struct tree_balance *cur_tb;
3845 /*
3846 ** before we can change a metadata block, we have to make sure it won't
3847 ** be written to disk while we are altering it.  So, we must:
3848 ** clean it
3849 ** wait on it.
3850 **
3851 */
3852 int reiserfs_prepare_for_journal(struct super_block *sb,
3853 				 struct buffer_head *bh, int wait)
3854 {
3855 	PROC_INFO_INC(sb, journal.prepare);
3856 
3857 	if (!trylock_buffer(bh)) {
3858 		if (!wait)
3859 			return 0;
3860 		lock_buffer(bh);
3861 	}
3862 	set_buffer_journal_prepared(bh);
3863 	if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3864 		clear_buffer_journal_test(bh);
3865 		set_buffer_journal_restore_dirty(bh);
3866 	}
3867 	unlock_buffer(bh);
3868 	return 1;
3869 }
3870 
3871 static void flush_old_journal_lists(struct super_block *s)
3872 {
3873 	struct reiserfs_journal *journal = SB_JOURNAL(s);
3874 	struct reiserfs_journal_list *jl;
3875 	struct list_head *entry;
3876 	time_t now = get_seconds();
3877 
3878 	while (!list_empty(&journal->j_journal_list)) {
3879 		entry = journal->j_journal_list.next;
3880 		jl = JOURNAL_LIST_ENTRY(entry);
3881 		/* this check should always be run, to send old lists to disk */
3882 		if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) &&
3883 		    atomic_read(&jl->j_commit_left) == 0 &&
3884 		    test_transaction(s, jl)) {
3885 			flush_used_journal_lists(s, jl);
3886 		} else {
3887 			break;
3888 		}
3889 	}
3890 }
3891 
3892 /*
3893 ** long and ugly.  If flush, will not return until all commit
3894 ** blocks and all real buffers in the trans are on disk.
3895 ** If no_async, won't return until all commit blocks are on disk.
3896 **
3897 ** keep reading, there are comments as you go along
3898 **
3899 ** If the journal is aborted, we just clean up. Things like flushing
3900 ** journal lists, etc just won't happen.
3901 */
3902 static int do_journal_end(struct reiserfs_transaction_handle *th,
3903 			  struct super_block *sb, unsigned long nblocks,
3904 			  int flags)
3905 {
3906 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3907 	struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3908 	struct reiserfs_journal_cnode *last_cn = NULL;
3909 	struct reiserfs_journal_desc *desc;
3910 	struct reiserfs_journal_commit *commit;
3911 	struct buffer_head *c_bh;	/* commit bh */
3912 	struct buffer_head *d_bh;	/* desc bh */
3913 	int cur_write_start = 0;	/* start index of current log write */
3914 	int old_start;
3915 	int i;
3916 	int flush;
3917 	int wait_on_commit;
3918 	struct reiserfs_journal_list *jl, *temp_jl;
3919 	struct list_head *entry, *safe;
3920 	unsigned long jindex;
3921 	unsigned int commit_trans_id;
3922 	int trans_half;
3923 	int depth;
3924 
3925 	BUG_ON(th->t_refcount > 1);
3926 	BUG_ON(!th->t_trans_id);
3927 
3928 	/* protect flush_older_commits from doing mistakes if the
3929            transaction ID counter gets overflowed.  */
3930 	if (th->t_trans_id == ~0U)
3931 		flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
3932 	flush = flags & FLUSH_ALL;
3933 	wait_on_commit = flags & WAIT;
3934 
3935 	current->journal_info = th->t_handle_save;
3936 	reiserfs_check_lock_depth(sb, "journal end");
3937 	if (journal->j_len == 0) {
3938 		reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3939 					     1);
3940 		journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
3941 	}
3942 
3943 	lock_journal(sb);
3944 	if (journal->j_next_full_flush) {
3945 		flags |= FLUSH_ALL;
3946 		flush = 1;
3947 	}
3948 	if (journal->j_next_async_flush) {
3949 		flags |= COMMIT_NOW | WAIT;
3950 		wait_on_commit = 1;
3951 	}
3952 
3953 	/* check_journal_end locks the journal, and unlocks if it does not return 1
3954 	 ** it tells us if we should continue with the journal_end, or just return
3955 	 */
3956 	if (!check_journal_end(th, sb, nblocks, flags)) {
3957 		reiserfs_schedule_old_flush(sb);
3958 		wake_queued_writers(sb);
3959 		reiserfs_async_progress_wait(sb);
3960 		goto out;
3961 	}
3962 
3963 	/* check_journal_end might set these, check again */
3964 	if (journal->j_next_full_flush) {
3965 		flush = 1;
3966 	}
3967 
3968 	/*
3969 	 ** j must wait means we have to flush the log blocks, and the real blocks for
3970 	 ** this transaction
3971 	 */
3972 	if (journal->j_must_wait > 0) {
3973 		flush = 1;
3974 	}
3975 #ifdef REISERFS_PREALLOCATE
3976 	/* quota ops might need to nest, setup the journal_info pointer for them
3977 	 * and raise the refcount so that it is > 0. */
3978 	current->journal_info = th;
3979 	th->t_refcount++;
3980 	reiserfs_discard_all_prealloc(th);	/* it should not involve new blocks into
3981 						 * the transaction */
3982 	th->t_refcount--;
3983 	current->journal_info = th->t_handle_save;
3984 #endif
3985 
3986 	/* setup description block */
3987 	d_bh =
3988 	    journal_getblk(sb,
3989 			   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
3990 			   journal->j_start);
3991 	set_buffer_uptodate(d_bh);
3992 	desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
3993 	memset(d_bh->b_data, 0, d_bh->b_size);
3994 	memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
3995 	set_desc_trans_id(desc, journal->j_trans_id);
3996 
3997 	/* setup commit block.  Don't write (keep it clean too) this one until after everyone else is written */
3998 	c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
3999 			      ((journal->j_start + journal->j_len +
4000 				1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4001 	commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4002 	memset(c_bh->b_data, 0, c_bh->b_size);
4003 	set_commit_trans_id(commit, journal->j_trans_id);
4004 	set_buffer_uptodate(c_bh);
4005 
4006 	/* init this journal list */
4007 	jl = journal->j_current_jl;
4008 
4009 	/* we lock the commit before doing anything because
4010 	 * we want to make sure nobody tries to run flush_commit_list until
4011 	 * the new transaction is fully setup, and we've already flushed the
4012 	 * ordered bh list
4013 	 */
4014 	reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4015 
4016 	/* save the transaction id in case we need to commit it later */
4017 	commit_trans_id = jl->j_trans_id;
4018 
4019 	atomic_set(&jl->j_older_commits_done, 0);
4020 	jl->j_trans_id = journal->j_trans_id;
4021 	jl->j_timestamp = journal->j_trans_start_time;
4022 	jl->j_commit_bh = c_bh;
4023 	jl->j_start = journal->j_start;
4024 	jl->j_len = journal->j_len;
4025 	atomic_set(&jl->j_nonzerolen, journal->j_len);
4026 	atomic_set(&jl->j_commit_left, journal->j_len + 2);
4027 	jl->j_realblock = NULL;
4028 
4029 	/* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4030 	 **  for each real block, add it to the journal list hash,
4031 	 ** copy into real block index array in the commit or desc block
4032 	 */
4033 	trans_half = journal_trans_half(sb->s_blocksize);
4034 	for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4035 		if (buffer_journaled(cn->bh)) {
4036 			jl_cn = get_cnode(sb);
4037 			if (!jl_cn) {
4038 				reiserfs_panic(sb, "journal-1676",
4039 					       "get_cnode returned NULL");
4040 			}
4041 			if (i == 0) {
4042 				jl->j_realblock = jl_cn;
4043 			}
4044 			jl_cn->prev = last_cn;
4045 			jl_cn->next = NULL;
4046 			if (last_cn) {
4047 				last_cn->next = jl_cn;
4048 			}
4049 			last_cn = jl_cn;
4050 			/* make sure the block we are trying to log is not a block
4051 			   of journal or reserved area */
4052 
4053 			if (is_block_in_log_or_reserved_area
4054 			    (sb, cn->bh->b_blocknr)) {
4055 				reiserfs_panic(sb, "journal-2332",
4056 					       "Trying to log block %lu, "
4057 					       "which is a log block",
4058 					       cn->bh->b_blocknr);
4059 			}
4060 			jl_cn->blocknr = cn->bh->b_blocknr;
4061 			jl_cn->state = 0;
4062 			jl_cn->sb = sb;
4063 			jl_cn->bh = cn->bh;
4064 			jl_cn->jlist = jl;
4065 			insert_journal_hash(journal->j_list_hash_table, jl_cn);
4066 			if (i < trans_half) {
4067 				desc->j_realblock[i] =
4068 				    cpu_to_le32(cn->bh->b_blocknr);
4069 			} else {
4070 				commit->j_realblock[i - trans_half] =
4071 				    cpu_to_le32(cn->bh->b_blocknr);
4072 			}
4073 		} else {
4074 			i--;
4075 		}
4076 	}
4077 	set_desc_trans_len(desc, journal->j_len);
4078 	set_desc_mount_id(desc, journal->j_mount_id);
4079 	set_desc_trans_id(desc, journal->j_trans_id);
4080 	set_commit_trans_len(commit, journal->j_len);
4081 
4082 	/* special check in case all buffers in the journal were marked for not logging */
4083 	BUG_ON(journal->j_len == 0);
4084 
4085 	/* we're about to dirty all the log blocks, mark the description block
4086 	 * dirty now too.  Don't mark the commit block dirty until all the
4087 	 * others are on disk
4088 	 */
4089 	mark_buffer_dirty(d_bh);
4090 
4091 	/* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4092 	cur_write_start = journal->j_start;
4093 	cn = journal->j_first;
4094 	jindex = 1;		/* start at one so we don't get the desc again */
4095 	while (cn) {
4096 		clear_buffer_journal_new(cn->bh);
4097 		/* copy all the real blocks into log area.  dirty log blocks */
4098 		if (buffer_journaled(cn->bh)) {
4099 			struct buffer_head *tmp_bh;
4100 			char *addr;
4101 			struct page *page;
4102 			tmp_bh =
4103 			    journal_getblk(sb,
4104 					   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4105 					   ((cur_write_start +
4106 					     jindex) %
4107 					    SB_ONDISK_JOURNAL_SIZE(sb)));
4108 			set_buffer_uptodate(tmp_bh);
4109 			page = cn->bh->b_page;
4110 			addr = kmap(page);
4111 			memcpy(tmp_bh->b_data,
4112 			       addr + offset_in_page(cn->bh->b_data),
4113 			       cn->bh->b_size);
4114 			kunmap(page);
4115 			mark_buffer_dirty(tmp_bh);
4116 			jindex++;
4117 			set_buffer_journal_dirty(cn->bh);
4118 			clear_buffer_journaled(cn->bh);
4119 		} else {
4120 			/* JDirty cleared sometime during transaction.  don't log this one */
4121 			reiserfs_warning(sb, "journal-2048",
4122 					 "BAD, buffer in journal hash, "
4123 					 "but not JDirty!");
4124 			brelse(cn->bh);
4125 		}
4126 		next = cn->next;
4127 		free_cnode(sb, cn);
4128 		cn = next;
4129 		reiserfs_cond_resched(sb);
4130 	}
4131 
4132 	/* we are done  with both the c_bh and d_bh, but
4133 	 ** c_bh must be written after all other commit blocks,
4134 	 ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4135 	 */
4136 
4137 	journal->j_current_jl = alloc_journal_list(sb);
4138 
4139 	/* now it is safe to insert this transaction on the main list */
4140 	list_add_tail(&jl->j_list, &journal->j_journal_list);
4141 	list_add_tail(&jl->j_working_list, &journal->j_working_list);
4142 	journal->j_num_work_lists++;
4143 
4144 	/* reset journal values for the next transaction */
4145 	old_start = journal->j_start;
4146 	journal->j_start =
4147 	    (journal->j_start + journal->j_len +
4148 	     2) % SB_ONDISK_JOURNAL_SIZE(sb);
4149 	atomic_set(&(journal->j_wcount), 0);
4150 	journal->j_bcount = 0;
4151 	journal->j_last = NULL;
4152 	journal->j_first = NULL;
4153 	journal->j_len = 0;
4154 	journal->j_trans_start_time = 0;
4155 	/* check for trans_id overflow */
4156 	if (++journal->j_trans_id == 0)
4157 		journal->j_trans_id = 10;
4158 	journal->j_current_jl->j_trans_id = journal->j_trans_id;
4159 	journal->j_must_wait = 0;
4160 	journal->j_len_alloc = 0;
4161 	journal->j_next_full_flush = 0;
4162 	journal->j_next_async_flush = 0;
4163 	init_journal_hash(sb);
4164 
4165 	// make sure reiserfs_add_jh sees the new current_jl before we
4166 	// write out the tails
4167 	smp_mb();
4168 
4169 	/* tail conversion targets have to hit the disk before we end the
4170 	 * transaction.  Otherwise a later transaction might repack the tail
4171 	 * before this transaction commits, leaving the data block unflushed and
4172 	 * clean, if we crash before the later transaction commits, the data block
4173 	 * is lost.
4174 	 */
4175 	if (!list_empty(&jl->j_tail_bh_list)) {
4176 		depth = reiserfs_write_unlock_nested(sb);
4177 		write_ordered_buffers(&journal->j_dirty_buffers_lock,
4178 				      journal, jl, &jl->j_tail_bh_list);
4179 		reiserfs_write_lock_nested(sb, depth);
4180 	}
4181 	BUG_ON(!list_empty(&jl->j_tail_bh_list));
4182 	mutex_unlock(&jl->j_commit_mutex);
4183 
4184 	/* honor the flush wishes from the caller, simple commits can
4185 	 ** be done outside the journal lock, they are done below
4186 	 **
4187 	 ** if we don't flush the commit list right now, we put it into
4188 	 ** the work queue so the people waiting on the async progress work
4189 	 ** queue don't wait for this proc to flush journal lists and such.
4190 	 */
4191 	if (flush) {
4192 		flush_commit_list(sb, jl, 1);
4193 		flush_journal_list(sb, jl, 1);
4194 	} else if (!(jl->j_state & LIST_COMMIT_PENDING))
4195 		queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4196 
4197 	/* if the next transaction has any chance of wrapping, flush
4198 	 ** transactions that might get overwritten.  If any journal lists are very
4199 	 ** old flush them as well.
4200 	 */
4201       first_jl:
4202 	list_for_each_safe(entry, safe, &journal->j_journal_list) {
4203 		temp_jl = JOURNAL_LIST_ENTRY(entry);
4204 		if (journal->j_start <= temp_jl->j_start) {
4205 			if ((journal->j_start + journal->j_trans_max + 1) >=
4206 			    temp_jl->j_start) {
4207 				flush_used_journal_lists(sb, temp_jl);
4208 				goto first_jl;
4209 			} else if ((journal->j_start +
4210 				    journal->j_trans_max + 1) <
4211 				   SB_ONDISK_JOURNAL_SIZE(sb)) {
4212 				/* if we don't cross into the next transaction and we don't
4213 				 * wrap, there is no way we can overlap any later transactions
4214 				 * break now
4215 				 */
4216 				break;
4217 			}
4218 		} else if ((journal->j_start +
4219 			    journal->j_trans_max + 1) >
4220 			   SB_ONDISK_JOURNAL_SIZE(sb)) {
4221 			if (((journal->j_start + journal->j_trans_max + 1) %
4222 			     SB_ONDISK_JOURNAL_SIZE(sb)) >=
4223 			    temp_jl->j_start) {
4224 				flush_used_journal_lists(sb, temp_jl);
4225 				goto first_jl;
4226 			} else {
4227 				/* we don't overlap anything from out start to the end of the
4228 				 * log, and our wrapped portion doesn't overlap anything at
4229 				 * the start of the log.  We can break
4230 				 */
4231 				break;
4232 			}
4233 		}
4234 	}
4235 	flush_old_journal_lists(sb);
4236 
4237 	journal->j_current_jl->j_list_bitmap =
4238 	    get_list_bitmap(sb, journal->j_current_jl);
4239 
4240 	if (!(journal->j_current_jl->j_list_bitmap)) {
4241 		reiserfs_panic(sb, "journal-1996",
4242 			       "could not get a list bitmap");
4243 	}
4244 
4245 	atomic_set(&(journal->j_jlock), 0);
4246 	unlock_journal(sb);
4247 	/* wake up any body waiting to join. */
4248 	clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4249 	wake_up(&(journal->j_join_wait));
4250 
4251 	if (!flush && wait_on_commit &&
4252 	    journal_list_still_alive(sb, commit_trans_id)) {
4253 		flush_commit_list(sb, jl, 1);
4254 	}
4255       out:
4256 	reiserfs_check_lock_depth(sb, "journal end2");
4257 
4258 	memset(th, 0, sizeof(*th));
4259 	/* Re-set th->t_super, so we can properly keep track of how many
4260 	 * persistent transactions there are. We need to do this so if this
4261 	 * call is part of a failed restart_transaction, we can free it later */
4262 	th->t_super = sb;
4263 
4264 	return journal->j_errno;
4265 }
4266 
4267 /* Send the file system read only and refuse new transactions */
4268 void reiserfs_abort_journal(struct super_block *sb, int errno)
4269 {
4270 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
4271 	if (test_bit(J_ABORTED, &journal->j_state))
4272 		return;
4273 
4274 	if (!journal->j_errno)
4275 		journal->j_errno = errno;
4276 
4277 	sb->s_flags |= MS_RDONLY;
4278 	set_bit(J_ABORTED, &journal->j_state);
4279 
4280 #ifdef CONFIG_REISERFS_CHECK
4281 	dump_stack();
4282 #endif
4283 }
4284