xref: /openbmc/linux/fs/reiserfs/journal.c (revision bdeeed09)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Write ahead logging implementation copyright Chris Mason 2000
4  *
5  * The background commits make this code very interrelated, and
6  * overly complex.  I need to rethink things a bit....The major players:
7  *
8  * journal_begin -- call with the number of blocks you expect to log.
9  *                  If the current transaction is too
10  *		    old, it will block until the current transaction is
11  *		    finished, and then start a new one.
12  *		    Usually, your transaction will get joined in with
13  *                  previous ones for speed.
14  *
15  * journal_join  -- same as journal_begin, but won't block on the current
16  *                  transaction regardless of age.  Don't ever call
17  *                  this.  Ever.  There are only two places it should be
18  *                  called from, and they are both inside this file.
19  *
20  * journal_mark_dirty -- adds blocks into this transaction.  clears any flags
21  *                       that might make them get sent to disk
22  *                       and then marks them BH_JDirty.  Puts the buffer head
23  *                       into the current transaction hash.
24  *
25  * journal_end -- if the current transaction is batchable, it does nothing
26  *                   otherwise, it could do an async/synchronous commit, or
27  *                   a full flush of all log and real blocks in the
28  *                   transaction.
29  *
30  * flush_old_commits -- if the current transaction is too old, it is ended and
31  *                      commit blocks are sent to disk.  Forces commit blocks
32  *                      to disk for all backgrounded commits that have been
33  *                      around too long.
34  *		     -- Note, if you call this as an immediate flush from
35  *		        from within kupdate, it will ignore the immediate flag
36  */
37 
38 #include <linux/time.h>
39 #include <linux/semaphore.h>
40 #include <linux/vmalloc.h>
41 #include "reiserfs.h"
42 #include <linux/kernel.h>
43 #include <linux/errno.h>
44 #include <linux/fcntl.h>
45 #include <linux/stat.h>
46 #include <linux/string.h>
47 #include <linux/buffer_head.h>
48 #include <linux/workqueue.h>
49 #include <linux/writeback.h>
50 #include <linux/blkdev.h>
51 #include <linux/backing-dev.h>
52 #include <linux/uaccess.h>
53 #include <linux/slab.h>
54 
55 
56 /* gets a struct reiserfs_journal_list * from a list head */
57 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
58                                j_list))
59 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
60                                j_working_list))
61 
62 /* must be correct to keep the desc and commit structs at 4k */
63 #define JOURNAL_TRANS_HALF 1018
64 #define BUFNR 64		/*read ahead */
65 
66 /* cnode stat bits.  Move these into reiserfs_fs.h */
67 
68 /* this block was freed, and can't be written.  */
69 #define BLOCK_FREED 2
70 /* this block was freed during this transaction, and can't be written */
71 #define BLOCK_FREED_HOLDER 3
72 
73 /* used in flush_journal_list */
74 #define BLOCK_NEEDS_FLUSH 4
75 #define BLOCK_DIRTIED 5
76 
77 /* journal list state bits */
78 #define LIST_TOUCHED 1
79 #define LIST_DIRTY   2
80 #define LIST_COMMIT_PENDING  4	/* someone will commit this list */
81 
82 /* flags for do_journal_end */
83 #define FLUSH_ALL   1		/* flush commit and real blocks */
84 #define COMMIT_NOW  2		/* end and commit this transaction */
85 #define WAIT        4		/* wait for the log blocks to hit the disk */
86 
87 static int do_journal_end(struct reiserfs_transaction_handle *, int flags);
88 static int flush_journal_list(struct super_block *s,
89 			      struct reiserfs_journal_list *jl, int flushall);
90 static int flush_commit_list(struct super_block *s,
91 			     struct reiserfs_journal_list *jl, int flushall);
92 static int can_dirty(struct reiserfs_journal_cnode *cn);
93 static int journal_join(struct reiserfs_transaction_handle *th,
94 			struct super_block *sb);
95 static void release_journal_dev(struct super_block *super,
96 			       struct reiserfs_journal *journal);
97 static int dirty_one_transaction(struct super_block *s,
98 				 struct reiserfs_journal_list *jl);
99 static void flush_async_commits(struct work_struct *work);
100 static void queue_log_writer(struct super_block *s);
101 
102 /* values for join in do_journal_begin_r */
103 enum {
104 	JBEGIN_REG = 0,		/* regular journal begin */
105 	/* join the running transaction if at all possible */
106 	JBEGIN_JOIN = 1,
107 	/* called from cleanup code, ignores aborted flag */
108 	JBEGIN_ABORT = 2,
109 };
110 
111 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
112 			      struct super_block *sb,
113 			      unsigned long nblocks, int join);
114 
115 static void init_journal_hash(struct super_block *sb)
116 {
117 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
118 	memset(journal->j_hash_table, 0,
119 	       JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
120 }
121 
122 /*
123  * clears BH_Dirty and sticks the buffer on the clean list.  Called because
124  * I can't allow refile_buffer to make schedule happen after I've freed a
125  * block.  Look at remove_from_transaction and journal_mark_freed for
126  * more details.
127  */
128 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
129 {
130 	if (bh) {
131 		clear_buffer_dirty(bh);
132 		clear_buffer_journal_test(bh);
133 	}
134 	return 0;
135 }
136 
137 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
138 							 *sb)
139 {
140 	struct reiserfs_bitmap_node *bn;
141 	static int id;
142 
143 	bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
144 	if (!bn) {
145 		return NULL;
146 	}
147 	bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
148 	if (!bn->data) {
149 		kfree(bn);
150 		return NULL;
151 	}
152 	bn->id = id++;
153 	INIT_LIST_HEAD(&bn->list);
154 	return bn;
155 }
156 
157 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
158 {
159 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
160 	struct reiserfs_bitmap_node *bn = NULL;
161 	struct list_head *entry = journal->j_bitmap_nodes.next;
162 
163 	journal->j_used_bitmap_nodes++;
164 repeat:
165 
166 	if (entry != &journal->j_bitmap_nodes) {
167 		bn = list_entry(entry, struct reiserfs_bitmap_node, list);
168 		list_del(entry);
169 		memset(bn->data, 0, sb->s_blocksize);
170 		journal->j_free_bitmap_nodes--;
171 		return bn;
172 	}
173 	bn = allocate_bitmap_node(sb);
174 	if (!bn) {
175 		yield();
176 		goto repeat;
177 	}
178 	return bn;
179 }
180 static inline void free_bitmap_node(struct super_block *sb,
181 				    struct reiserfs_bitmap_node *bn)
182 {
183 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
184 	journal->j_used_bitmap_nodes--;
185 	if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
186 		kfree(bn->data);
187 		kfree(bn);
188 	} else {
189 		list_add(&bn->list, &journal->j_bitmap_nodes);
190 		journal->j_free_bitmap_nodes++;
191 	}
192 }
193 
194 static void allocate_bitmap_nodes(struct super_block *sb)
195 {
196 	int i;
197 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
198 	struct reiserfs_bitmap_node *bn = NULL;
199 	for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
200 		bn = allocate_bitmap_node(sb);
201 		if (bn) {
202 			list_add(&bn->list, &journal->j_bitmap_nodes);
203 			journal->j_free_bitmap_nodes++;
204 		} else {
205 			/* this is ok, we'll try again when more are needed */
206 			break;
207 		}
208 	}
209 }
210 
211 static int set_bit_in_list_bitmap(struct super_block *sb,
212 				  b_blocknr_t block,
213 				  struct reiserfs_list_bitmap *jb)
214 {
215 	unsigned int bmap_nr = block / (sb->s_blocksize << 3);
216 	unsigned int bit_nr = block % (sb->s_blocksize << 3);
217 
218 	if (!jb->bitmaps[bmap_nr]) {
219 		jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
220 	}
221 	set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
222 	return 0;
223 }
224 
225 static void cleanup_bitmap_list(struct super_block *sb,
226 				struct reiserfs_list_bitmap *jb)
227 {
228 	int i;
229 	if (jb->bitmaps == NULL)
230 		return;
231 
232 	for (i = 0; i < reiserfs_bmap_count(sb); i++) {
233 		if (jb->bitmaps[i]) {
234 			free_bitmap_node(sb, jb->bitmaps[i]);
235 			jb->bitmaps[i] = NULL;
236 		}
237 	}
238 }
239 
240 /*
241  * only call this on FS unmount.
242  */
243 static int free_list_bitmaps(struct super_block *sb,
244 			     struct reiserfs_list_bitmap *jb_array)
245 {
246 	int i;
247 	struct reiserfs_list_bitmap *jb;
248 	for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
249 		jb = jb_array + i;
250 		jb->journal_list = NULL;
251 		cleanup_bitmap_list(sb, jb);
252 		vfree(jb->bitmaps);
253 		jb->bitmaps = NULL;
254 	}
255 	return 0;
256 }
257 
258 static int free_bitmap_nodes(struct super_block *sb)
259 {
260 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
261 	struct list_head *next = journal->j_bitmap_nodes.next;
262 	struct reiserfs_bitmap_node *bn;
263 
264 	while (next != &journal->j_bitmap_nodes) {
265 		bn = list_entry(next, struct reiserfs_bitmap_node, list);
266 		list_del(next);
267 		kfree(bn->data);
268 		kfree(bn);
269 		next = journal->j_bitmap_nodes.next;
270 		journal->j_free_bitmap_nodes--;
271 	}
272 
273 	return 0;
274 }
275 
276 /*
277  * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
278  * jb_array is the array to be filled in.
279  */
280 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
281 				   struct reiserfs_list_bitmap *jb_array,
282 				   unsigned int bmap_nr)
283 {
284 	int i;
285 	int failed = 0;
286 	struct reiserfs_list_bitmap *jb;
287 	int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
288 
289 	for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
290 		jb = jb_array + i;
291 		jb->journal_list = NULL;
292 		jb->bitmaps = vzalloc(mem);
293 		if (!jb->bitmaps) {
294 			reiserfs_warning(sb, "clm-2000", "unable to "
295 					 "allocate bitmaps for journal lists");
296 			failed = 1;
297 			break;
298 		}
299 	}
300 	if (failed) {
301 		free_list_bitmaps(sb, jb_array);
302 		return -1;
303 	}
304 	return 0;
305 }
306 
307 /*
308  * find an available list bitmap.  If you can't find one, flush a commit list
309  * and try again
310  */
311 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
312 						    struct reiserfs_journal_list
313 						    *jl)
314 {
315 	int i, j;
316 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
317 	struct reiserfs_list_bitmap *jb = NULL;
318 
319 	for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
320 		i = journal->j_list_bitmap_index;
321 		journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
322 		jb = journal->j_list_bitmap + i;
323 		if (journal->j_list_bitmap[i].journal_list) {
324 			flush_commit_list(sb,
325 					  journal->j_list_bitmap[i].
326 					  journal_list, 1);
327 			if (!journal->j_list_bitmap[i].journal_list) {
328 				break;
329 			}
330 		} else {
331 			break;
332 		}
333 	}
334 	/* double check to make sure if flushed correctly */
335 	if (jb->journal_list)
336 		return NULL;
337 	jb->journal_list = jl;
338 	return jb;
339 }
340 
341 /*
342  * allocates a new chunk of X nodes, and links them all together as a list.
343  * Uses the cnode->next and cnode->prev pointers
344  * returns NULL on failure
345  */
346 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
347 {
348 	struct reiserfs_journal_cnode *head;
349 	int i;
350 	if (num_cnodes <= 0) {
351 		return NULL;
352 	}
353 	head = vzalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
354 	if (!head) {
355 		return NULL;
356 	}
357 	head[0].prev = NULL;
358 	head[0].next = head + 1;
359 	for (i = 1; i < num_cnodes; i++) {
360 		head[i].prev = head + (i - 1);
361 		head[i].next = head + (i + 1);	/* if last one, overwrite it after the if */
362 	}
363 	head[num_cnodes - 1].next = NULL;
364 	return head;
365 }
366 
367 /* pulls a cnode off the free list, or returns NULL on failure */
368 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
369 {
370 	struct reiserfs_journal_cnode *cn;
371 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
372 
373 	reiserfs_check_lock_depth(sb, "get_cnode");
374 
375 	if (journal->j_cnode_free <= 0) {
376 		return NULL;
377 	}
378 	journal->j_cnode_used++;
379 	journal->j_cnode_free--;
380 	cn = journal->j_cnode_free_list;
381 	if (!cn) {
382 		return cn;
383 	}
384 	if (cn->next) {
385 		cn->next->prev = NULL;
386 	}
387 	journal->j_cnode_free_list = cn->next;
388 	memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
389 	return cn;
390 }
391 
392 /*
393  * returns a cnode to the free list
394  */
395 static void free_cnode(struct super_block *sb,
396 		       struct reiserfs_journal_cnode *cn)
397 {
398 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
399 
400 	reiserfs_check_lock_depth(sb, "free_cnode");
401 
402 	journal->j_cnode_used--;
403 	journal->j_cnode_free++;
404 	/* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
405 	cn->next = journal->j_cnode_free_list;
406 	if (journal->j_cnode_free_list) {
407 		journal->j_cnode_free_list->prev = cn;
408 	}
409 	cn->prev = NULL;	/* not needed with the memset, but I might kill the memset, and forget to do this */
410 	journal->j_cnode_free_list = cn;
411 }
412 
413 static void clear_prepared_bits(struct buffer_head *bh)
414 {
415 	clear_buffer_journal_prepared(bh);
416 	clear_buffer_journal_restore_dirty(bh);
417 }
418 
419 /*
420  * return a cnode with same dev, block number and size in table,
421  * or null if not found
422  */
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
443  * search_all, a block can only be allocated if it is not in the current
444  * transaction, was not freed by the current transaction, and has no chance
445  * of ever being overwritten by a replay after crashing.
446  *
447  * If you don't set search_all, a block can only be allocated if it is not
448  * in the current transaction.  Since deleting a block removes it from the
449  * current transaction, this case should never happen.  If you don't set
450  * search_all, make sure you never write the block without logging it.
451  *
452  * next_zero_bit is a suggestion about the next block to try for find_forward.
453  * when bl is rejected because it is set in a journal list bitmap, we search
454  * for the next zero bit in the bitmap that rejected bl.  Then, we return
455  * that through next_zero_bit for find_forward to try.
456  *
457  * Just because we return something in next_zero_bit does not mean we won't
458  * reject it on the next call to reiserfs_in_journal
459  */
460 int reiserfs_in_journal(struct super_block *sb,
461 			unsigned int bmap_nr, int bit_nr, int search_all,
462 			b_blocknr_t * next_zero_bit)
463 {
464 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
465 	struct reiserfs_journal_cnode *cn;
466 	struct reiserfs_list_bitmap *jb;
467 	int i;
468 	unsigned long bl;
469 
470 	*next_zero_bit = 0;	/* always start this at zero. */
471 
472 	PROC_INFO_INC(sb, journal.in_journal);
473 	/*
474 	 * If we aren't doing a search_all, this is a metablock, and it
475 	 * will be logged before use.  if we crash before the transaction
476 	 * that freed it commits,  this transaction won't have committed
477 	 * either, and the block will never be written
478 	 */
479 	if (search_all) {
480 		for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
481 			PROC_INFO_INC(sb, journal.in_journal_bitmap);
482 			jb = journal->j_list_bitmap + i;
483 			if (jb->journal_list && jb->bitmaps[bmap_nr] &&
484 			    test_bit(bit_nr,
485 				     (unsigned long *)jb->bitmaps[bmap_nr]->
486 				     data)) {
487 				*next_zero_bit =
488 				    find_next_zero_bit((unsigned long *)
489 						       (jb->bitmaps[bmap_nr]->
490 							data),
491 						       sb->s_blocksize << 3,
492 						       bit_nr + 1);
493 				return 1;
494 			}
495 		}
496 	}
497 
498 	bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
499 	/* is it in any old transactions? */
500 	if (search_all
501 	    && (cn =
502 		get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
503 		return 1;
504 	}
505 
506 	/* is it in the current transaction.  This should never happen */
507 	if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
508 		BUG();
509 		return 1;
510 	}
511 
512 	PROC_INFO_INC(sb, journal.in_journal_reusable);
513 	/* safe for reuse */
514 	return 0;
515 }
516 
517 /* insert cn into table */
518 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
519 				       struct reiserfs_journal_cnode *cn)
520 {
521 	struct reiserfs_journal_cnode *cn_orig;
522 
523 	cn_orig = journal_hash(table, cn->sb, cn->blocknr);
524 	cn->hnext = cn_orig;
525 	cn->hprev = NULL;
526 	if (cn_orig) {
527 		cn_orig->hprev = cn;
528 	}
529 	journal_hash(table, cn->sb, cn->blocknr) = cn;
530 }
531 
532 /* lock the current transaction */
533 static inline void lock_journal(struct super_block *sb)
534 {
535 	PROC_INFO_INC(sb, journal.lock_journal);
536 
537 	reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
538 }
539 
540 /* unlock the current transaction */
541 static inline void unlock_journal(struct super_block *sb)
542 {
543 	mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
544 }
545 
546 static inline void get_journal_list(struct reiserfs_journal_list *jl)
547 {
548 	jl->j_refcount++;
549 }
550 
551 static inline void put_journal_list(struct super_block *s,
552 				    struct reiserfs_journal_list *jl)
553 {
554 	if (jl->j_refcount < 1) {
555 		reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
556 			       jl->j_trans_id, jl->j_refcount);
557 	}
558 	if (--jl->j_refcount == 0)
559 		kfree(jl);
560 }
561 
562 /*
563  * this used to be much more involved, and I'm keeping it just in case
564  * things get ugly again.  it gets called by flush_commit_list, and
565  * cleans up any data stored about blocks freed during a transaction.
566  */
567 static void cleanup_freed_for_journal_list(struct super_block *sb,
568 					   struct reiserfs_journal_list *jl)
569 {
570 
571 	struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
572 	if (jb) {
573 		cleanup_bitmap_list(sb, jb);
574 	}
575 	jl->j_list_bitmap->journal_list = NULL;
576 	jl->j_list_bitmap = NULL;
577 }
578 
579 static int journal_list_still_alive(struct super_block *s,
580 				    unsigned int trans_id)
581 {
582 	struct reiserfs_journal *journal = SB_JOURNAL(s);
583 	struct list_head *entry = &journal->j_journal_list;
584 	struct reiserfs_journal_list *jl;
585 
586 	if (!list_empty(entry)) {
587 		jl = JOURNAL_LIST_ENTRY(entry->next);
588 		if (jl->j_trans_id <= trans_id) {
589 			return 1;
590 		}
591 	}
592 	return 0;
593 }
594 
595 /*
596  * If page->mapping was null, we failed to truncate this page for
597  * some reason.  Most likely because it was truncated after being
598  * logged via data=journal.
599  *
600  * This does a check to see if the buffer belongs to one of these
601  * lost pages before doing the final put_bh.  If page->mapping was
602  * null, it tries to free buffers on the page, which should make the
603  * final put_page drop the page from the lru.
604  */
605 static void release_buffer_page(struct buffer_head *bh)
606 {
607 	struct page *page = bh->b_page;
608 	if (!page->mapping && trylock_page(page)) {
609 		get_page(page);
610 		put_bh(bh);
611 		if (!page->mapping)
612 			try_to_free_buffers(page);
613 		unlock_page(page);
614 		put_page(page);
615 	} else {
616 		put_bh(bh);
617 	}
618 }
619 
620 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
621 {
622 	if (buffer_journaled(bh)) {
623 		reiserfs_warning(NULL, "clm-2084",
624 				 "pinned buffer %lu:%pg sent to disk",
625 				 bh->b_blocknr, bh->b_bdev);
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(REQ_OP_WRITE, 0, 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(REQ_OP_WRITE, 0, 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 			spin_lock(lock);
705 		} else {
706 			fn(chunk);
707 		}
708 	}
709 	return ret;
710 }
711 
712 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
713 static struct reiserfs_jh *alloc_jh(void)
714 {
715 	struct reiserfs_jh *jh;
716 	while (1) {
717 		jh = kmalloc(sizeof(*jh), GFP_NOFS);
718 		if (jh) {
719 			atomic_inc(&nr_reiserfs_jh);
720 			return jh;
721 		}
722 		yield();
723 	}
724 }
725 
726 /*
727  * we want to free the jh when the buffer has been written
728  * and waited on
729  */
730 void reiserfs_free_jh(struct buffer_head *bh)
731 {
732 	struct reiserfs_jh *jh;
733 
734 	jh = bh->b_private;
735 	if (jh) {
736 		bh->b_private = NULL;
737 		jh->bh = NULL;
738 		list_del_init(&jh->list);
739 		kfree(jh);
740 		if (atomic_read(&nr_reiserfs_jh) <= 0)
741 			BUG();
742 		atomic_dec(&nr_reiserfs_jh);
743 		put_bh(bh);
744 	}
745 }
746 
747 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
748 			   int tail)
749 {
750 	struct reiserfs_jh *jh;
751 
752 	if (bh->b_private) {
753 		spin_lock(&j->j_dirty_buffers_lock);
754 		if (!bh->b_private) {
755 			spin_unlock(&j->j_dirty_buffers_lock);
756 			goto no_jh;
757 		}
758 		jh = bh->b_private;
759 		list_del_init(&jh->list);
760 	} else {
761 no_jh:
762 		get_bh(bh);
763 		jh = alloc_jh();
764 		spin_lock(&j->j_dirty_buffers_lock);
765 		/*
766 		 * buffer must be locked for __add_jh, should be able to have
767 		 * two adds at the same time
768 		 */
769 		BUG_ON(bh->b_private);
770 		jh->bh = bh;
771 		bh->b_private = jh;
772 	}
773 	jh->jl = j->j_current_jl;
774 	if (tail)
775 		list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
776 	else {
777 		list_add_tail(&jh->list, &jh->jl->j_bh_list);
778 	}
779 	spin_unlock(&j->j_dirty_buffers_lock);
780 	return 0;
781 }
782 
783 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
784 {
785 	return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
786 }
787 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
788 {
789 	return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
790 }
791 
792 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
793 static int write_ordered_buffers(spinlock_t * lock,
794 				 struct reiserfs_journal *j,
795 				 struct reiserfs_journal_list *jl,
796 				 struct list_head *list)
797 {
798 	struct buffer_head *bh;
799 	struct reiserfs_jh *jh;
800 	int ret = j->j_errno;
801 	struct buffer_chunk chunk;
802 	struct list_head tmp;
803 	INIT_LIST_HEAD(&tmp);
804 
805 	chunk.nr = 0;
806 	spin_lock(lock);
807 	while (!list_empty(list)) {
808 		jh = JH_ENTRY(list->next);
809 		bh = jh->bh;
810 		get_bh(bh);
811 		if (!trylock_buffer(bh)) {
812 			if (!buffer_dirty(bh)) {
813 				list_move(&jh->list, &tmp);
814 				goto loop_next;
815 			}
816 			spin_unlock(lock);
817 			if (chunk.nr)
818 				write_ordered_chunk(&chunk);
819 			wait_on_buffer(bh);
820 			cond_resched();
821 			spin_lock(lock);
822 			goto loop_next;
823 		}
824 		/*
825 		 * in theory, dirty non-uptodate buffers should never get here,
826 		 * but the upper layer io error paths still have a few quirks.
827 		 * Handle them here as gracefully as we can
828 		 */
829 		if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
830 			clear_buffer_dirty(bh);
831 			ret = -EIO;
832 		}
833 		if (buffer_dirty(bh)) {
834 			list_move(&jh->list, &tmp);
835 			add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
836 		} else {
837 			reiserfs_free_jh(bh);
838 			unlock_buffer(bh);
839 		}
840 loop_next:
841 		put_bh(bh);
842 		cond_resched_lock(lock);
843 	}
844 	if (chunk.nr) {
845 		spin_unlock(lock);
846 		write_ordered_chunk(&chunk);
847 		spin_lock(lock);
848 	}
849 	while (!list_empty(&tmp)) {
850 		jh = JH_ENTRY(tmp.prev);
851 		bh = jh->bh;
852 		get_bh(bh);
853 		reiserfs_free_jh(bh);
854 
855 		if (buffer_locked(bh)) {
856 			spin_unlock(lock);
857 			wait_on_buffer(bh);
858 			spin_lock(lock);
859 		}
860 		if (!buffer_uptodate(bh)) {
861 			ret = -EIO;
862 		}
863 		/*
864 		 * ugly interaction with invalidatepage here.
865 		 * reiserfs_invalidate_page will pin any buffer that has a
866 		 * valid journal head from an older transaction.  If someone
867 		 * else sets our buffer dirty after we write it in the first
868 		 * loop, and then someone truncates the page away, nobody
869 		 * will ever write the buffer. We're safe if we write the
870 		 * page one last time after freeing the journal header.
871 		 */
872 		if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
873 			spin_unlock(lock);
874 			ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
875 			spin_lock(lock);
876 		}
877 		put_bh(bh);
878 		cond_resched_lock(lock);
879 	}
880 	spin_unlock(lock);
881 	return ret;
882 }
883 
884 static int flush_older_commits(struct super_block *s,
885 			       struct reiserfs_journal_list *jl)
886 {
887 	struct reiserfs_journal *journal = SB_JOURNAL(s);
888 	struct reiserfs_journal_list *other_jl;
889 	struct reiserfs_journal_list *first_jl;
890 	struct list_head *entry;
891 	unsigned int trans_id = jl->j_trans_id;
892 	unsigned int other_trans_id;
893 	unsigned int first_trans_id;
894 
895 find_first:
896 	/*
897 	 * first we walk backwards to find the oldest uncommitted transation
898 	 */
899 	first_jl = jl;
900 	entry = jl->j_list.prev;
901 	while (1) {
902 		other_jl = JOURNAL_LIST_ENTRY(entry);
903 		if (entry == &journal->j_journal_list ||
904 		    atomic_read(&other_jl->j_older_commits_done))
905 			break;
906 
907 		first_jl = other_jl;
908 		entry = other_jl->j_list.prev;
909 	}
910 
911 	/* if we didn't find any older uncommitted transactions, return now */
912 	if (first_jl == jl) {
913 		return 0;
914 	}
915 
916 	first_trans_id = first_jl->j_trans_id;
917 
918 	entry = &first_jl->j_list;
919 	while (1) {
920 		other_jl = JOURNAL_LIST_ENTRY(entry);
921 		other_trans_id = other_jl->j_trans_id;
922 
923 		if (other_trans_id < trans_id) {
924 			if (atomic_read(&other_jl->j_commit_left) != 0) {
925 				flush_commit_list(s, other_jl, 0);
926 
927 				/* list we were called with is gone, return */
928 				if (!journal_list_still_alive(s, trans_id))
929 					return 1;
930 
931 				/*
932 				 * the one we just flushed is gone, this means
933 				 * all older lists are also gone, so first_jl
934 				 * is no longer valid either.  Go back to the
935 				 * beginning.
936 				 */
937 				if (!journal_list_still_alive
938 				    (s, other_trans_id)) {
939 					goto find_first;
940 				}
941 			}
942 			entry = entry->next;
943 			if (entry == &journal->j_journal_list)
944 				return 0;
945 		} else {
946 			return 0;
947 		}
948 	}
949 	return 0;
950 }
951 
952 static int reiserfs_async_progress_wait(struct super_block *s)
953 {
954 	struct reiserfs_journal *j = SB_JOURNAL(s);
955 
956 	if (atomic_read(&j->j_async_throttle)) {
957 		int depth;
958 
959 		depth = reiserfs_write_unlock_nested(s);
960 		congestion_wait(BLK_RW_ASYNC, HZ / 10);
961 		reiserfs_write_lock_nested(s, depth);
962 	}
963 
964 	return 0;
965 }
966 
967 /*
968  * if this journal list still has commit blocks unflushed, send them to disk.
969  *
970  * log areas must be flushed in order (transaction 2 can't commit before
971  * transaction 1) Before the commit block can by written, every other log
972  * block must be safely on disk
973  */
974 static int flush_commit_list(struct super_block *s,
975 			     struct reiserfs_journal_list *jl, int flushall)
976 {
977 	int i;
978 	b_blocknr_t bn;
979 	struct buffer_head *tbh = NULL;
980 	unsigned int trans_id = jl->j_trans_id;
981 	struct reiserfs_journal *journal = SB_JOURNAL(s);
982 	int retval = 0;
983 	int write_len;
984 	int depth;
985 
986 	reiserfs_check_lock_depth(s, "flush_commit_list");
987 
988 	if (atomic_read(&jl->j_older_commits_done)) {
989 		return 0;
990 	}
991 
992 	/*
993 	 * before we can put our commit blocks on disk, we have to make
994 	 * sure everyone older than us is on disk too
995 	 */
996 	BUG_ON(jl->j_len <= 0);
997 	BUG_ON(trans_id == journal->j_trans_id);
998 
999 	get_journal_list(jl);
1000 	if (flushall) {
1001 		if (flush_older_commits(s, jl) == 1) {
1002 			/*
1003 			 * list disappeared during flush_older_commits.
1004 			 * return
1005 			 */
1006 			goto put_jl;
1007 		}
1008 	}
1009 
1010 	/* make sure nobody is trying to flush this one at the same time */
1011 	reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1012 
1013 	if (!journal_list_still_alive(s, trans_id)) {
1014 		mutex_unlock(&jl->j_commit_mutex);
1015 		goto put_jl;
1016 	}
1017 	BUG_ON(jl->j_trans_id == 0);
1018 
1019 	/* this commit is done, exit */
1020 	if (atomic_read(&jl->j_commit_left) <= 0) {
1021 		if (flushall) {
1022 			atomic_set(&jl->j_older_commits_done, 1);
1023 		}
1024 		mutex_unlock(&jl->j_commit_mutex);
1025 		goto put_jl;
1026 	}
1027 
1028 	if (!list_empty(&jl->j_bh_list)) {
1029 		int ret;
1030 
1031 		/*
1032 		 * We might sleep in numerous places inside
1033 		 * write_ordered_buffers. Relax the write lock.
1034 		 */
1035 		depth = reiserfs_write_unlock_nested(s);
1036 		ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1037 					    journal, jl, &jl->j_bh_list);
1038 		if (ret < 0 && retval == 0)
1039 			retval = ret;
1040 		reiserfs_write_lock_nested(s, depth);
1041 	}
1042 	BUG_ON(!list_empty(&jl->j_bh_list));
1043 	/*
1044 	 * for the description block and all the log blocks, submit any buffers
1045 	 * that haven't already reached the disk.  Try to write at least 256
1046 	 * log blocks. later on, we will only wait on blocks that correspond
1047 	 * to this transaction, but while we're unplugging we might as well
1048 	 * get a chunk of data on there.
1049 	 */
1050 	atomic_inc(&journal->j_async_throttle);
1051 	write_len = jl->j_len + 1;
1052 	if (write_len < 256)
1053 		write_len = 256;
1054 	for (i = 0 ; i < write_len ; i++) {
1055 		bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1056 		    SB_ONDISK_JOURNAL_SIZE(s);
1057 		tbh = journal_find_get_block(s, bn);
1058 		if (tbh) {
1059 			if (buffer_dirty(tbh)) {
1060 		            depth = reiserfs_write_unlock_nested(s);
1061 			    ll_rw_block(REQ_OP_WRITE, 0, 1, &tbh);
1062 			    reiserfs_write_lock_nested(s, depth);
1063 			}
1064 			put_bh(tbh) ;
1065 		}
1066 	}
1067 	atomic_dec(&journal->j_async_throttle);
1068 
1069 	for (i = 0; i < (jl->j_len + 1); i++) {
1070 		bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1071 		    (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1072 		tbh = journal_find_get_block(s, bn);
1073 
1074 		depth = reiserfs_write_unlock_nested(s);
1075 		__wait_on_buffer(tbh);
1076 		reiserfs_write_lock_nested(s, depth);
1077 		/*
1078 		 * since we're using ll_rw_blk above, it might have skipped
1079 		 * over a locked buffer.  Double check here
1080 		 */
1081 		/* redundant, sync_dirty_buffer() checks */
1082 		if (buffer_dirty(tbh)) {
1083 			depth = reiserfs_write_unlock_nested(s);
1084 			sync_dirty_buffer(tbh);
1085 			reiserfs_write_lock_nested(s, depth);
1086 		}
1087 		if (unlikely(!buffer_uptodate(tbh))) {
1088 #ifdef CONFIG_REISERFS_CHECK
1089 			reiserfs_warning(s, "journal-601",
1090 					 "buffer write failed");
1091 #endif
1092 			retval = -EIO;
1093 		}
1094 		/* once for journal_find_get_block */
1095 		put_bh(tbh);
1096 		/* once due to original getblk in do_journal_end */
1097 		put_bh(tbh);
1098 		atomic_dec(&jl->j_commit_left);
1099 	}
1100 
1101 	BUG_ON(atomic_read(&jl->j_commit_left) != 1);
1102 
1103 	/*
1104 	 * If there was a write error in the journal - we can't commit
1105 	 * this transaction - it will be invalid and, if successful,
1106 	 * will just end up propagating the write error out to
1107 	 * the file system.
1108 	 */
1109 	if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1110 		if (buffer_dirty(jl->j_commit_bh))
1111 			BUG();
1112 		mark_buffer_dirty(jl->j_commit_bh) ;
1113 		depth = reiserfs_write_unlock_nested(s);
1114 		if (reiserfs_barrier_flush(s))
1115 			__sync_dirty_buffer(jl->j_commit_bh,
1116 					REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1117 		else
1118 			sync_dirty_buffer(jl->j_commit_bh);
1119 		reiserfs_write_lock_nested(s, depth);
1120 	}
1121 
1122 	/*
1123 	 * If there was a write error in the journal - we can't commit this
1124 	 * transaction - it will be invalid and, if successful, will just end
1125 	 * up propagating the write error out to the filesystem.
1126 	 */
1127 	if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1128 #ifdef CONFIG_REISERFS_CHECK
1129 		reiserfs_warning(s, "journal-615", "buffer write failed");
1130 #endif
1131 		retval = -EIO;
1132 	}
1133 	bforget(jl->j_commit_bh);
1134 	if (journal->j_last_commit_id != 0 &&
1135 	    (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1136 		reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1137 				 journal->j_last_commit_id, jl->j_trans_id);
1138 	}
1139 	journal->j_last_commit_id = jl->j_trans_id;
1140 
1141 	/*
1142 	 * now, every commit block is on the disk.  It is safe to allow
1143 	 * blocks freed during this transaction to be reallocated
1144 	 */
1145 	cleanup_freed_for_journal_list(s, jl);
1146 
1147 	retval = retval ? retval : journal->j_errno;
1148 
1149 	/* mark the metadata dirty */
1150 	if (!retval)
1151 		dirty_one_transaction(s, jl);
1152 	atomic_dec(&jl->j_commit_left);
1153 
1154 	if (flushall) {
1155 		atomic_set(&jl->j_older_commits_done, 1);
1156 	}
1157 	mutex_unlock(&jl->j_commit_mutex);
1158 put_jl:
1159 	put_journal_list(s, jl);
1160 
1161 	if (retval)
1162 		reiserfs_abort(s, retval, "Journal write error in %s",
1163 			       __func__);
1164 	return retval;
1165 }
1166 
1167 /*
1168  * flush_journal_list frequently needs to find a newer transaction for a
1169  * given block.  This does that, or returns NULL if it can't find anything
1170  */
1171 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1172 							  reiserfs_journal_cnode
1173 							  *cn)
1174 {
1175 	struct super_block *sb = cn->sb;
1176 	b_blocknr_t blocknr = cn->blocknr;
1177 
1178 	cn = cn->hprev;
1179 	while (cn) {
1180 		if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1181 			return cn->jlist;
1182 		}
1183 		cn = cn->hprev;
1184 	}
1185 	return NULL;
1186 }
1187 
1188 static void remove_journal_hash(struct super_block *,
1189 				struct reiserfs_journal_cnode **,
1190 				struct reiserfs_journal_list *, unsigned long,
1191 				int);
1192 
1193 /*
1194  * once all the real blocks have been flushed, it is safe to remove them
1195  * from the journal list for this transaction.  Aside from freeing the
1196  * cnode, this also allows the block to be reallocated for data blocks
1197  * if it had been deleted.
1198  */
1199 static void remove_all_from_journal_list(struct super_block *sb,
1200 					 struct reiserfs_journal_list *jl,
1201 					 int debug)
1202 {
1203 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1204 	struct reiserfs_journal_cnode *cn, *last;
1205 	cn = jl->j_realblock;
1206 
1207 	/*
1208 	 * which is better, to lock once around the whole loop, or
1209 	 * to lock for each call to remove_journal_hash?
1210 	 */
1211 	while (cn) {
1212 		if (cn->blocknr != 0) {
1213 			if (debug) {
1214 				reiserfs_warning(sb, "reiserfs-2201",
1215 						 "block %u, bh is %d, state %ld",
1216 						 cn->blocknr, cn->bh ? 1 : 0,
1217 						 cn->state);
1218 			}
1219 			cn->state = 0;
1220 			remove_journal_hash(sb, journal->j_list_hash_table,
1221 					    jl, cn->blocknr, 1);
1222 		}
1223 		last = cn;
1224 		cn = cn->next;
1225 		free_cnode(sb, last);
1226 	}
1227 	jl->j_realblock = NULL;
1228 }
1229 
1230 /*
1231  * if this timestamp is greater than the timestamp we wrote last to the
1232  * header block, write it to the header block.  once this is done, I can
1233  * safely say the log area for this transaction won't ever be replayed,
1234  * and I can start releasing blocks in this transaction for reuse as data
1235  * blocks.  called by flush_journal_list, before it calls
1236  * remove_all_from_journal_list
1237  */
1238 static int _update_journal_header_block(struct super_block *sb,
1239 					unsigned long offset,
1240 					unsigned int trans_id)
1241 {
1242 	struct reiserfs_journal_header *jh;
1243 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1244 	int depth;
1245 
1246 	if (reiserfs_is_journal_aborted(journal))
1247 		return -EIO;
1248 
1249 	if (trans_id >= journal->j_last_flush_trans_id) {
1250 		if (buffer_locked((journal->j_header_bh))) {
1251 			depth = reiserfs_write_unlock_nested(sb);
1252 			__wait_on_buffer(journal->j_header_bh);
1253 			reiserfs_write_lock_nested(sb, depth);
1254 			if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1255 #ifdef CONFIG_REISERFS_CHECK
1256 				reiserfs_warning(sb, "journal-699",
1257 						 "buffer write failed");
1258 #endif
1259 				return -EIO;
1260 			}
1261 		}
1262 		journal->j_last_flush_trans_id = trans_id;
1263 		journal->j_first_unflushed_offset = offset;
1264 		jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1265 							b_data);
1266 		jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1267 		jh->j_first_unflushed_offset = cpu_to_le32(offset);
1268 		jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1269 
1270 		set_buffer_dirty(journal->j_header_bh);
1271 		depth = reiserfs_write_unlock_nested(sb);
1272 
1273 		if (reiserfs_barrier_flush(sb))
1274 			__sync_dirty_buffer(journal->j_header_bh,
1275 					REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1276 		else
1277 			sync_dirty_buffer(journal->j_header_bh);
1278 
1279 		reiserfs_write_lock_nested(sb, depth);
1280 		if (!buffer_uptodate(journal->j_header_bh)) {
1281 			reiserfs_warning(sb, "journal-837",
1282 					 "IO error during journal replay");
1283 			return -EIO;
1284 		}
1285 	}
1286 	return 0;
1287 }
1288 
1289 static int update_journal_header_block(struct super_block *sb,
1290 				       unsigned long offset,
1291 				       unsigned int trans_id)
1292 {
1293 	return _update_journal_header_block(sb, offset, trans_id);
1294 }
1295 
1296 /*
1297 ** flush any and all journal lists older than you are
1298 ** can only be called from flush_journal_list
1299 */
1300 static int flush_older_journal_lists(struct super_block *sb,
1301 				     struct reiserfs_journal_list *jl)
1302 {
1303 	struct list_head *entry;
1304 	struct reiserfs_journal_list *other_jl;
1305 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1306 	unsigned int trans_id = jl->j_trans_id;
1307 
1308 	/*
1309 	 * we know we are the only ones flushing things, no extra race
1310 	 * protection is required.
1311 	 */
1312 restart:
1313 	entry = journal->j_journal_list.next;
1314 	/* Did we wrap? */
1315 	if (entry == &journal->j_journal_list)
1316 		return 0;
1317 	other_jl = JOURNAL_LIST_ENTRY(entry);
1318 	if (other_jl->j_trans_id < trans_id) {
1319 		BUG_ON(other_jl->j_refcount <= 0);
1320 		/* do not flush all */
1321 		flush_journal_list(sb, other_jl, 0);
1322 
1323 		/* other_jl is now deleted from the list */
1324 		goto restart;
1325 	}
1326 	return 0;
1327 }
1328 
1329 static void del_from_work_list(struct super_block *s,
1330 			       struct reiserfs_journal_list *jl)
1331 {
1332 	struct reiserfs_journal *journal = SB_JOURNAL(s);
1333 	if (!list_empty(&jl->j_working_list)) {
1334 		list_del_init(&jl->j_working_list);
1335 		journal->j_num_work_lists--;
1336 	}
1337 }
1338 
1339 /*
1340  * flush a journal list, both commit and real blocks
1341  *
1342  * always set flushall to 1, unless you are calling from inside
1343  * flush_journal_list
1344  *
1345  * IMPORTANT.  This can only be called while there are no journal writers,
1346  * and the journal is locked.  That means it can only be called from
1347  * do_journal_end, or by journal_release
1348  */
1349 static int flush_journal_list(struct super_block *s,
1350 			      struct reiserfs_journal_list *jl, int flushall)
1351 {
1352 	struct reiserfs_journal_list *pjl;
1353 	struct reiserfs_journal_cnode *cn, *last;
1354 	int count;
1355 	int was_jwait = 0;
1356 	int was_dirty = 0;
1357 	struct buffer_head *saved_bh;
1358 	unsigned long j_len_saved = jl->j_len;
1359 	struct reiserfs_journal *journal = SB_JOURNAL(s);
1360 	int err = 0;
1361 	int depth;
1362 
1363 	BUG_ON(j_len_saved <= 0);
1364 
1365 	if (atomic_read(&journal->j_wcount) != 0) {
1366 		reiserfs_warning(s, "clm-2048", "called with wcount %d",
1367 				 atomic_read(&journal->j_wcount));
1368 	}
1369 
1370 	/* if flushall == 0, the lock is already held */
1371 	if (flushall) {
1372 		reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1373 	} else if (mutex_trylock(&journal->j_flush_mutex)) {
1374 		BUG();
1375 	}
1376 
1377 	count = 0;
1378 	if (j_len_saved > journal->j_trans_max) {
1379 		reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1380 			       j_len_saved, jl->j_trans_id);
1381 		return 0;
1382 	}
1383 
1384 	/* if all the work is already done, get out of here */
1385 	if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1386 	    atomic_read(&jl->j_commit_left) <= 0) {
1387 		goto flush_older_and_return;
1388 	}
1389 
1390 	/*
1391 	 * start by putting the commit list on disk.  This will also flush
1392 	 * the commit lists of any olders transactions
1393 	 */
1394 	flush_commit_list(s, jl, 1);
1395 
1396 	if (!(jl->j_state & LIST_DIRTY)
1397 	    && !reiserfs_is_journal_aborted(journal))
1398 		BUG();
1399 
1400 	/* are we done now? */
1401 	if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1402 	    atomic_read(&jl->j_commit_left) <= 0) {
1403 		goto flush_older_and_return;
1404 	}
1405 
1406 	/*
1407 	 * loop through each cnode, see if we need to write it,
1408 	 * or wait on a more recent transaction, or just ignore it
1409 	 */
1410 	if (atomic_read(&journal->j_wcount) != 0) {
1411 		reiserfs_panic(s, "journal-844", "journal list is flushing, "
1412 			       "wcount is not 0");
1413 	}
1414 	cn = jl->j_realblock;
1415 	while (cn) {
1416 		was_jwait = 0;
1417 		was_dirty = 0;
1418 		saved_bh = NULL;
1419 		/* blocknr of 0 is no longer in the hash, ignore it */
1420 		if (cn->blocknr == 0) {
1421 			goto free_cnode;
1422 		}
1423 
1424 		/*
1425 		 * This transaction failed commit.
1426 		 * Don't write out to the disk
1427 		 */
1428 		if (!(jl->j_state & LIST_DIRTY))
1429 			goto free_cnode;
1430 
1431 		pjl = find_newer_jl_for_cn(cn);
1432 		/*
1433 		 * the order is important here.  We check pjl to make sure we
1434 		 * don't clear BH_JDirty_wait if we aren't the one writing this
1435 		 * block to disk
1436 		 */
1437 		if (!pjl && cn->bh) {
1438 			saved_bh = cn->bh;
1439 
1440 			/*
1441 			 * we do this to make sure nobody releases the
1442 			 * buffer while we are working with it
1443 			 */
1444 			get_bh(saved_bh);
1445 
1446 			if (buffer_journal_dirty(saved_bh)) {
1447 				BUG_ON(!can_dirty(cn));
1448 				was_jwait = 1;
1449 				was_dirty = 1;
1450 			} else if (can_dirty(cn)) {
1451 				/*
1452 				 * everything with !pjl && jwait
1453 				 * should be writable
1454 				 */
1455 				BUG();
1456 			}
1457 		}
1458 
1459 		/*
1460 		 * if someone has this block in a newer transaction, just make
1461 		 * sure they are committed, and don't try writing it to disk
1462 		 */
1463 		if (pjl) {
1464 			if (atomic_read(&pjl->j_commit_left))
1465 				flush_commit_list(s, pjl, 1);
1466 			goto free_cnode;
1467 		}
1468 
1469 		/*
1470 		 * bh == NULL when the block got to disk on its own, OR,
1471 		 * the block got freed in a future transaction
1472 		 */
1473 		if (saved_bh == NULL) {
1474 			goto free_cnode;
1475 		}
1476 
1477 		/*
1478 		 * this should never happen.  kupdate_one_transaction has
1479 		 * this list locked while it works, so we should never see a
1480 		 * buffer here that is not marked JDirty_wait
1481 		 */
1482 		if ((!was_jwait) && !buffer_locked(saved_bh)) {
1483 			reiserfs_warning(s, "journal-813",
1484 					 "BAD! buffer %llu %cdirty %cjwait, "
1485 					 "not in a newer transaction",
1486 					 (unsigned long long)saved_bh->
1487 					 b_blocknr, was_dirty ? ' ' : '!',
1488 					 was_jwait ? ' ' : '!');
1489 		}
1490 		if (was_dirty) {
1491 			/*
1492 			 * we inc again because saved_bh gets decremented
1493 			 * at free_cnode
1494 			 */
1495 			get_bh(saved_bh);
1496 			set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1497 			lock_buffer(saved_bh);
1498 			BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1499 			if (buffer_dirty(saved_bh))
1500 				submit_logged_buffer(saved_bh);
1501 			else
1502 				unlock_buffer(saved_bh);
1503 			count++;
1504 		} else {
1505 			reiserfs_warning(s, "clm-2082",
1506 					 "Unable to flush buffer %llu in %s",
1507 					 (unsigned long long)saved_bh->
1508 					 b_blocknr, __func__);
1509 		}
1510 free_cnode:
1511 		last = cn;
1512 		cn = cn->next;
1513 		if (saved_bh) {
1514 			/*
1515 			 * we incremented this to keep others from
1516 			 * taking the buffer head away
1517 			 */
1518 			put_bh(saved_bh);
1519 			if (atomic_read(&saved_bh->b_count) < 0) {
1520 				reiserfs_warning(s, "journal-945",
1521 						 "saved_bh->b_count < 0");
1522 			}
1523 		}
1524 	}
1525 	if (count > 0) {
1526 		cn = jl->j_realblock;
1527 		while (cn) {
1528 			if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1529 				if (!cn->bh) {
1530 					reiserfs_panic(s, "journal-1011",
1531 						       "cn->bh is NULL");
1532 				}
1533 
1534 				depth = reiserfs_write_unlock_nested(s);
1535 				__wait_on_buffer(cn->bh);
1536 				reiserfs_write_lock_nested(s, depth);
1537 
1538 				if (!cn->bh) {
1539 					reiserfs_panic(s, "journal-1012",
1540 						       "cn->bh is NULL");
1541 				}
1542 				if (unlikely(!buffer_uptodate(cn->bh))) {
1543 #ifdef CONFIG_REISERFS_CHECK
1544 					reiserfs_warning(s, "journal-949",
1545 							 "buffer write failed");
1546 #endif
1547 					err = -EIO;
1548 				}
1549 				/*
1550 				 * note, we must clear the JDirty_wait bit
1551 				 * after the up to date check, otherwise we
1552 				 * race against our flushpage routine
1553 				 */
1554 				BUG_ON(!test_clear_buffer_journal_dirty
1555 				       (cn->bh));
1556 
1557 				/* drop one ref for us */
1558 				put_bh(cn->bh);
1559 				/* drop one ref for journal_mark_dirty */
1560 				release_buffer_page(cn->bh);
1561 			}
1562 			cn = cn->next;
1563 		}
1564 	}
1565 
1566 	if (err)
1567 		reiserfs_abort(s, -EIO,
1568 			       "Write error while pushing transaction to disk in %s",
1569 			       __func__);
1570 flush_older_and_return:
1571 
1572 	/*
1573 	 * before we can update the journal header block, we _must_ flush all
1574 	 * real blocks from all older transactions to disk.  This is because
1575 	 * once the header block is updated, this transaction will not be
1576 	 * replayed after a crash
1577 	 */
1578 	if (flushall) {
1579 		flush_older_journal_lists(s, jl);
1580 	}
1581 
1582 	err = journal->j_errno;
1583 	/*
1584 	 * before we can remove everything from the hash tables for this
1585 	 * transaction, we must make sure it can never be replayed
1586 	 *
1587 	 * since we are only called from do_journal_end, we know for sure there
1588 	 * are no allocations going on while we are flushing journal lists.  So,
1589 	 * we only need to update the journal header block for the last list
1590 	 * being flushed
1591 	 */
1592 	if (!err && flushall) {
1593 		err =
1594 		    update_journal_header_block(s,
1595 						(jl->j_start + jl->j_len +
1596 						 2) % SB_ONDISK_JOURNAL_SIZE(s),
1597 						jl->j_trans_id);
1598 		if (err)
1599 			reiserfs_abort(s, -EIO,
1600 				       "Write error while updating journal header in %s",
1601 				       __func__);
1602 	}
1603 	remove_all_from_journal_list(s, jl, 0);
1604 	list_del_init(&jl->j_list);
1605 	journal->j_num_lists--;
1606 	del_from_work_list(s, jl);
1607 
1608 	if (journal->j_last_flush_id != 0 &&
1609 	    (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1610 		reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1611 				 journal->j_last_flush_id, jl->j_trans_id);
1612 	}
1613 	journal->j_last_flush_id = jl->j_trans_id;
1614 
1615 	/*
1616 	 * not strictly required since we are freeing the list, but it should
1617 	 * help find code using dead lists later on
1618 	 */
1619 	jl->j_len = 0;
1620 	atomic_set(&jl->j_nonzerolen, 0);
1621 	jl->j_start = 0;
1622 	jl->j_realblock = NULL;
1623 	jl->j_commit_bh = NULL;
1624 	jl->j_trans_id = 0;
1625 	jl->j_state = 0;
1626 	put_journal_list(s, jl);
1627 	if (flushall)
1628 		mutex_unlock(&journal->j_flush_mutex);
1629 	return err;
1630 }
1631 
1632 static int write_one_transaction(struct super_block *s,
1633 				 struct reiserfs_journal_list *jl,
1634 				 struct buffer_chunk *chunk)
1635 {
1636 	struct reiserfs_journal_cnode *cn;
1637 	int ret = 0;
1638 
1639 	jl->j_state |= LIST_TOUCHED;
1640 	del_from_work_list(s, jl);
1641 	if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1642 		return 0;
1643 	}
1644 
1645 	cn = jl->j_realblock;
1646 	while (cn) {
1647 		/*
1648 		 * if the blocknr == 0, this has been cleared from the hash,
1649 		 * skip it
1650 		 */
1651 		if (cn->blocknr == 0) {
1652 			goto next;
1653 		}
1654 		if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1655 			struct buffer_head *tmp_bh;
1656 			/*
1657 			 * we can race against journal_mark_freed when we try
1658 			 * to lock_buffer(cn->bh), so we have to inc the buffer
1659 			 * count, and recheck things after locking
1660 			 */
1661 			tmp_bh = cn->bh;
1662 			get_bh(tmp_bh);
1663 			lock_buffer(tmp_bh);
1664 			if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1665 				if (!buffer_journal_dirty(tmp_bh) ||
1666 				    buffer_journal_prepared(tmp_bh))
1667 					BUG();
1668 				add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1669 				ret++;
1670 			} else {
1671 				/* note, cn->bh might be null now */
1672 				unlock_buffer(tmp_bh);
1673 			}
1674 			put_bh(tmp_bh);
1675 		}
1676 next:
1677 		cn = cn->next;
1678 		cond_resched();
1679 	}
1680 	return ret;
1681 }
1682 
1683 /* used by flush_commit_list */
1684 static int dirty_one_transaction(struct super_block *s,
1685 				 struct reiserfs_journal_list *jl)
1686 {
1687 	struct reiserfs_journal_cnode *cn;
1688 	struct reiserfs_journal_list *pjl;
1689 	int ret = 0;
1690 
1691 	jl->j_state |= LIST_DIRTY;
1692 	cn = jl->j_realblock;
1693 	while (cn) {
1694 		/*
1695 		 * look for a more recent transaction that logged this
1696 		 * buffer.  Only the most recent transaction with a buffer in
1697 		 * it is allowed to send that buffer to disk
1698 		 */
1699 		pjl = find_newer_jl_for_cn(cn);
1700 		if (!pjl && cn->blocknr && cn->bh
1701 		    && buffer_journal_dirty(cn->bh)) {
1702 			BUG_ON(!can_dirty(cn));
1703 			/*
1704 			 * if the buffer is prepared, it will either be logged
1705 			 * or restored.  If restored, we need to make sure
1706 			 * it actually gets marked dirty
1707 			 */
1708 			clear_buffer_journal_new(cn->bh);
1709 			if (buffer_journal_prepared(cn->bh)) {
1710 				set_buffer_journal_restore_dirty(cn->bh);
1711 			} else {
1712 				set_buffer_journal_test(cn->bh);
1713 				mark_buffer_dirty(cn->bh);
1714 			}
1715 		}
1716 		cn = cn->next;
1717 	}
1718 	return ret;
1719 }
1720 
1721 static int kupdate_transactions(struct super_block *s,
1722 				struct reiserfs_journal_list *jl,
1723 				struct reiserfs_journal_list **next_jl,
1724 				unsigned int *next_trans_id,
1725 				int num_blocks, int num_trans)
1726 {
1727 	int ret = 0;
1728 	int written = 0;
1729 	int transactions_flushed = 0;
1730 	unsigned int orig_trans_id = jl->j_trans_id;
1731 	struct buffer_chunk chunk;
1732 	struct list_head *entry;
1733 	struct reiserfs_journal *journal = SB_JOURNAL(s);
1734 	chunk.nr = 0;
1735 
1736 	reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1737 	if (!journal_list_still_alive(s, orig_trans_id)) {
1738 		goto done;
1739 	}
1740 
1741 	/*
1742 	 * we've got j_flush_mutex held, nobody is going to delete any
1743 	 * of these lists out from underneath us
1744 	 */
1745 	while ((num_trans && transactions_flushed < num_trans) ||
1746 	       (!num_trans && written < num_blocks)) {
1747 
1748 		if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1749 		    atomic_read(&jl->j_commit_left)
1750 		    || !(jl->j_state & LIST_DIRTY)) {
1751 			del_from_work_list(s, jl);
1752 			break;
1753 		}
1754 		ret = write_one_transaction(s, jl, &chunk);
1755 
1756 		if (ret < 0)
1757 			goto done;
1758 		transactions_flushed++;
1759 		written += ret;
1760 		entry = jl->j_list.next;
1761 
1762 		/* did we wrap? */
1763 		if (entry == &journal->j_journal_list) {
1764 			break;
1765 		}
1766 		jl = JOURNAL_LIST_ENTRY(entry);
1767 
1768 		/* don't bother with older transactions */
1769 		if (jl->j_trans_id <= orig_trans_id)
1770 			break;
1771 	}
1772 	if (chunk.nr) {
1773 		write_chunk(&chunk);
1774 	}
1775 
1776 done:
1777 	mutex_unlock(&journal->j_flush_mutex);
1778 	return ret;
1779 }
1780 
1781 /*
1782  * for o_sync and fsync heavy applications, they tend to use
1783  * all the journa list slots with tiny transactions.  These
1784  * trigger lots and lots of calls to update the header block, which
1785  * adds seeks and slows things down.
1786  *
1787  * This function tries to clear out a large chunk of the journal lists
1788  * at once, which makes everything faster since only the newest journal
1789  * list updates the header block
1790  */
1791 static int flush_used_journal_lists(struct super_block *s,
1792 				    struct reiserfs_journal_list *jl)
1793 {
1794 	unsigned long len = 0;
1795 	unsigned long cur_len;
1796 	int ret;
1797 	int i;
1798 	int limit = 256;
1799 	struct reiserfs_journal_list *tjl;
1800 	struct reiserfs_journal_list *flush_jl;
1801 	unsigned int trans_id;
1802 	struct reiserfs_journal *journal = SB_JOURNAL(s);
1803 
1804 	flush_jl = tjl = jl;
1805 
1806 	/* in data logging mode, try harder to flush a lot of blocks */
1807 	if (reiserfs_data_log(s))
1808 		limit = 1024;
1809 	/* flush for 256 transactions or limit blocks, whichever comes first */
1810 	for (i = 0; i < 256 && len < limit; i++) {
1811 		if (atomic_read(&tjl->j_commit_left) ||
1812 		    tjl->j_trans_id < jl->j_trans_id) {
1813 			break;
1814 		}
1815 		cur_len = atomic_read(&tjl->j_nonzerolen);
1816 		if (cur_len > 0) {
1817 			tjl->j_state &= ~LIST_TOUCHED;
1818 		}
1819 		len += cur_len;
1820 		flush_jl = tjl;
1821 		if (tjl->j_list.next == &journal->j_journal_list)
1822 			break;
1823 		tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1824 	}
1825 	get_journal_list(jl);
1826 	get_journal_list(flush_jl);
1827 
1828 	/*
1829 	 * try to find a group of blocks we can flush across all the
1830 	 * transactions, but only bother if we've actually spanned
1831 	 * across multiple lists
1832 	 */
1833 	if (flush_jl != jl) {
1834 		ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1835 	}
1836 	flush_journal_list(s, flush_jl, 1);
1837 	put_journal_list(s, flush_jl);
1838 	put_journal_list(s, jl);
1839 	return 0;
1840 }
1841 
1842 /*
1843  * removes any nodes in table with name block and dev as bh.
1844  * only touchs the hnext and hprev pointers.
1845  */
1846 void remove_journal_hash(struct super_block *sb,
1847 			 struct reiserfs_journal_cnode **table,
1848 			 struct reiserfs_journal_list *jl,
1849 			 unsigned long block, int remove_freed)
1850 {
1851 	struct reiserfs_journal_cnode *cur;
1852 	struct reiserfs_journal_cnode **head;
1853 
1854 	head = &(journal_hash(table, sb, block));
1855 	if (!head) {
1856 		return;
1857 	}
1858 	cur = *head;
1859 	while (cur) {
1860 		if (cur->blocknr == block && cur->sb == sb
1861 		    && (jl == NULL || jl == cur->jlist)
1862 		    && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1863 			if (cur->hnext) {
1864 				cur->hnext->hprev = cur->hprev;
1865 			}
1866 			if (cur->hprev) {
1867 				cur->hprev->hnext = cur->hnext;
1868 			} else {
1869 				*head = cur->hnext;
1870 			}
1871 			cur->blocknr = 0;
1872 			cur->sb = NULL;
1873 			cur->state = 0;
1874 			/*
1875 			 * anybody who clears the cur->bh will also
1876 			 * dec the nonzerolen
1877 			 */
1878 			if (cur->bh && cur->jlist)
1879 				atomic_dec(&cur->jlist->j_nonzerolen);
1880 			cur->bh = NULL;
1881 			cur->jlist = NULL;
1882 		}
1883 		cur = cur->hnext;
1884 	}
1885 }
1886 
1887 static void free_journal_ram(struct super_block *sb)
1888 {
1889 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1890 	kfree(journal->j_current_jl);
1891 	journal->j_num_lists--;
1892 
1893 	vfree(journal->j_cnode_free_orig);
1894 	free_list_bitmaps(sb, journal->j_list_bitmap);
1895 	free_bitmap_nodes(sb);	/* must be after free_list_bitmaps */
1896 	if (journal->j_header_bh) {
1897 		brelse(journal->j_header_bh);
1898 	}
1899 	/*
1900 	 * j_header_bh is on the journal dev, make sure
1901 	 * not to release the journal dev until we brelse j_header_bh
1902 	 */
1903 	release_journal_dev(sb, journal);
1904 	vfree(journal);
1905 }
1906 
1907 /*
1908  * call on unmount.  Only set error to 1 if you haven't made your way out
1909  * of read_super() yet.  Any other caller must keep error at 0.
1910  */
1911 static int do_journal_release(struct reiserfs_transaction_handle *th,
1912 			      struct super_block *sb, int error)
1913 {
1914 	struct reiserfs_transaction_handle myth;
1915 	int flushed = 0;
1916 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1917 
1918 	/*
1919 	 * we only want to flush out transactions if we were
1920 	 * called with error == 0
1921 	 */
1922 	if (!error && !sb_rdonly(sb)) {
1923 		/* end the current trans */
1924 		BUG_ON(!th->t_trans_id);
1925 		do_journal_end(th, FLUSH_ALL);
1926 
1927 		/*
1928 		 * make sure something gets logged to force
1929 		 * our way into the flush code
1930 		 */
1931 		if (!journal_join(&myth, sb)) {
1932 			reiserfs_prepare_for_journal(sb,
1933 						     SB_BUFFER_WITH_SB(sb),
1934 						     1);
1935 			journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1936 			do_journal_end(&myth, FLUSH_ALL);
1937 			flushed = 1;
1938 		}
1939 	}
1940 
1941 	/* this also catches errors during the do_journal_end above */
1942 	if (!error && reiserfs_is_journal_aborted(journal)) {
1943 		memset(&myth, 0, sizeof(myth));
1944 		if (!journal_join_abort(&myth, sb)) {
1945 			reiserfs_prepare_for_journal(sb,
1946 						     SB_BUFFER_WITH_SB(sb),
1947 						     1);
1948 			journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1949 			do_journal_end(&myth, FLUSH_ALL);
1950 		}
1951 	}
1952 
1953 
1954 	/*
1955 	 * We must release the write lock here because
1956 	 * the workqueue job (flush_async_commit) needs this lock
1957 	 */
1958 	reiserfs_write_unlock(sb);
1959 
1960 	/*
1961 	 * Cancel flushing of old commits. Note that neither of these works
1962 	 * will be requeued because superblock is being shutdown and doesn't
1963 	 * have SB_ACTIVE set.
1964 	 */
1965 	reiserfs_cancel_old_flush(sb);
1966 	/* wait for all commits to finish */
1967 	cancel_delayed_work_sync(&SB_JOURNAL(sb)->j_work);
1968 
1969 	free_journal_ram(sb);
1970 
1971 	reiserfs_write_lock(sb);
1972 
1973 	return 0;
1974 }
1975 
1976 /* * call on unmount.  flush all journal trans, release all alloc'd ram */
1977 int journal_release(struct reiserfs_transaction_handle *th,
1978 		    struct super_block *sb)
1979 {
1980 	return do_journal_release(th, sb, 0);
1981 }
1982 
1983 /* only call from an error condition inside reiserfs_read_super!  */
1984 int journal_release_error(struct reiserfs_transaction_handle *th,
1985 			  struct super_block *sb)
1986 {
1987 	return do_journal_release(th, sb, 1);
1988 }
1989 
1990 /*
1991  * compares description block with commit block.
1992  * returns 1 if they differ, 0 if they are the same
1993  */
1994 static int journal_compare_desc_commit(struct super_block *sb,
1995 				       struct reiserfs_journal_desc *desc,
1996 				       struct reiserfs_journal_commit *commit)
1997 {
1998 	if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1999 	    get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
2000 	    get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
2001 	    get_commit_trans_len(commit) <= 0) {
2002 		return 1;
2003 	}
2004 	return 0;
2005 }
2006 
2007 /*
2008  * returns 0 if it did not find a description block
2009  * returns -1 if it found a corrupt commit block
2010  * returns 1 if both desc and commit were valid
2011  * NOTE: only called during fs mount
2012  */
2013 static int journal_transaction_is_valid(struct super_block *sb,
2014 					struct buffer_head *d_bh,
2015 					unsigned int *oldest_invalid_trans_id,
2016 					unsigned long *newest_mount_id)
2017 {
2018 	struct reiserfs_journal_desc *desc;
2019 	struct reiserfs_journal_commit *commit;
2020 	struct buffer_head *c_bh;
2021 	unsigned long offset;
2022 
2023 	if (!d_bh)
2024 		return 0;
2025 
2026 	desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2027 	if (get_desc_trans_len(desc) > 0
2028 	    && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2029 		if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2030 		    && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2031 			reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2032 				       "journal-986: transaction "
2033 				       "is valid returning because trans_id %d is greater than "
2034 				       "oldest_invalid %lu",
2035 				       get_desc_trans_id(desc),
2036 				       *oldest_invalid_trans_id);
2037 			return 0;
2038 		}
2039 		if (newest_mount_id
2040 		    && *newest_mount_id > get_desc_mount_id(desc)) {
2041 			reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2042 				       "journal-1087: transaction "
2043 				       "is valid returning because mount_id %d is less than "
2044 				       "newest_mount_id %lu",
2045 				       get_desc_mount_id(desc),
2046 				       *newest_mount_id);
2047 			return -1;
2048 		}
2049 		if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2050 			reiserfs_warning(sb, "journal-2018",
2051 					 "Bad transaction length %d "
2052 					 "encountered, ignoring transaction",
2053 					 get_desc_trans_len(desc));
2054 			return -1;
2055 		}
2056 		offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2057 
2058 		/*
2059 		 * ok, we have a journal description block,
2060 		 * let's see if the transaction was valid
2061 		 */
2062 		c_bh =
2063 		    journal_bread(sb,
2064 				  SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2065 				  ((offset + get_desc_trans_len(desc) +
2066 				    1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2067 		if (!c_bh)
2068 			return 0;
2069 		commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2070 		if (journal_compare_desc_commit(sb, desc, commit)) {
2071 			reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2072 				       "journal_transaction_is_valid, commit offset %ld had bad "
2073 				       "time %d or length %d",
2074 				       c_bh->b_blocknr -
2075 				       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2076 				       get_commit_trans_id(commit),
2077 				       get_commit_trans_len(commit));
2078 			brelse(c_bh);
2079 			if (oldest_invalid_trans_id) {
2080 				*oldest_invalid_trans_id =
2081 				    get_desc_trans_id(desc);
2082 				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2083 					       "journal-1004: "
2084 					       "transaction_is_valid setting oldest invalid trans_id "
2085 					       "to %d",
2086 					       get_desc_trans_id(desc));
2087 			}
2088 			return -1;
2089 		}
2090 		brelse(c_bh);
2091 		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2092 			       "journal-1006: found valid "
2093 			       "transaction start offset %llu, len %d id %d",
2094 			       d_bh->b_blocknr -
2095 			       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2096 			       get_desc_trans_len(desc),
2097 			       get_desc_trans_id(desc));
2098 		return 1;
2099 	} else {
2100 		return 0;
2101 	}
2102 }
2103 
2104 static void brelse_array(struct buffer_head **heads, int num)
2105 {
2106 	int i;
2107 	for (i = 0; i < num; i++) {
2108 		brelse(heads[i]);
2109 	}
2110 }
2111 
2112 /*
2113  * given the start, and values for the oldest acceptable transactions,
2114  * this either reads in a replays a transaction, or returns because the
2115  * transaction is invalid, or too old.
2116  * NOTE: only called during fs mount
2117  */
2118 static int journal_read_transaction(struct super_block *sb,
2119 				    unsigned long cur_dblock,
2120 				    unsigned long oldest_start,
2121 				    unsigned int oldest_trans_id,
2122 				    unsigned long newest_mount_id)
2123 {
2124 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
2125 	struct reiserfs_journal_desc *desc;
2126 	struct reiserfs_journal_commit *commit;
2127 	unsigned int trans_id = 0;
2128 	struct buffer_head *c_bh;
2129 	struct buffer_head *d_bh;
2130 	struct buffer_head **log_blocks = NULL;
2131 	struct buffer_head **real_blocks = NULL;
2132 	unsigned int trans_offset;
2133 	int i;
2134 	int trans_half;
2135 
2136 	d_bh = journal_bread(sb, cur_dblock);
2137 	if (!d_bh)
2138 		return 1;
2139 	desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2140 	trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2141 	reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2142 		       "journal_read_transaction, offset %llu, len %d mount_id %d",
2143 		       d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2144 		       get_desc_trans_len(desc), get_desc_mount_id(desc));
2145 	if (get_desc_trans_id(desc) < oldest_trans_id) {
2146 		reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2147 			       "journal_read_trans skipping because %lu is too old",
2148 			       cur_dblock -
2149 			       SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2150 		brelse(d_bh);
2151 		return 1;
2152 	}
2153 	if (get_desc_mount_id(desc) != newest_mount_id) {
2154 		reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2155 			       "journal_read_trans skipping because %d is != "
2156 			       "newest_mount_id %lu", get_desc_mount_id(desc),
2157 			       newest_mount_id);
2158 		brelse(d_bh);
2159 		return 1;
2160 	}
2161 	c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2162 			     ((trans_offset + get_desc_trans_len(desc) + 1) %
2163 			      SB_ONDISK_JOURNAL_SIZE(sb)));
2164 	if (!c_bh) {
2165 		brelse(d_bh);
2166 		return 1;
2167 	}
2168 	commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2169 	if (journal_compare_desc_commit(sb, desc, commit)) {
2170 		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2171 			       "journal_read_transaction, "
2172 			       "commit offset %llu had bad time %d or length %d",
2173 			       c_bh->b_blocknr -
2174 			       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2175 			       get_commit_trans_id(commit),
2176 			       get_commit_trans_len(commit));
2177 		brelse(c_bh);
2178 		brelse(d_bh);
2179 		return 1;
2180 	}
2181 
2182 	if (bdev_read_only(sb->s_bdev)) {
2183 		reiserfs_warning(sb, "clm-2076",
2184 				 "device is readonly, unable to replay log");
2185 		brelse(c_bh);
2186 		brelse(d_bh);
2187 		return -EROFS;
2188 	}
2189 
2190 	trans_id = get_desc_trans_id(desc);
2191 	/*
2192 	 * now we know we've got a good transaction, and it was
2193 	 * inside the valid time ranges
2194 	 */
2195 	log_blocks = kmalloc(get_desc_trans_len(desc) *
2196 			     sizeof(struct buffer_head *), GFP_NOFS);
2197 	real_blocks = kmalloc(get_desc_trans_len(desc) *
2198 			      sizeof(struct buffer_head *), GFP_NOFS);
2199 	if (!log_blocks || !real_blocks) {
2200 		brelse(c_bh);
2201 		brelse(d_bh);
2202 		kfree(log_blocks);
2203 		kfree(real_blocks);
2204 		reiserfs_warning(sb, "journal-1169",
2205 				 "kmalloc failed, unable to mount FS");
2206 		return -1;
2207 	}
2208 	/* get all the buffer heads */
2209 	trans_half = journal_trans_half(sb->s_blocksize);
2210 	for (i = 0; i < get_desc_trans_len(desc); i++) {
2211 		log_blocks[i] =
2212 		    journal_getblk(sb,
2213 				   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2214 				   (trans_offset + 1 +
2215 				    i) % SB_ONDISK_JOURNAL_SIZE(sb));
2216 		if (i < trans_half) {
2217 			real_blocks[i] =
2218 			    sb_getblk(sb,
2219 				      le32_to_cpu(desc->j_realblock[i]));
2220 		} else {
2221 			real_blocks[i] =
2222 			    sb_getblk(sb,
2223 				      le32_to_cpu(commit->
2224 						  j_realblock[i - trans_half]));
2225 		}
2226 		if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2227 			reiserfs_warning(sb, "journal-1207",
2228 					 "REPLAY FAILURE fsck required! "
2229 					 "Block to replay is outside of "
2230 					 "filesystem");
2231 			goto abort_replay;
2232 		}
2233 		/* make sure we don't try to replay onto log or reserved area */
2234 		if (is_block_in_log_or_reserved_area
2235 		    (sb, real_blocks[i]->b_blocknr)) {
2236 			reiserfs_warning(sb, "journal-1204",
2237 					 "REPLAY FAILURE fsck required! "
2238 					 "Trying to replay onto a log block");
2239 abort_replay:
2240 			brelse_array(log_blocks, i);
2241 			brelse_array(real_blocks, i);
2242 			brelse(c_bh);
2243 			brelse(d_bh);
2244 			kfree(log_blocks);
2245 			kfree(real_blocks);
2246 			return -1;
2247 		}
2248 	}
2249 	/* read in the log blocks, memcpy to the corresponding real block */
2250 	ll_rw_block(REQ_OP_READ, 0, get_desc_trans_len(desc), log_blocks);
2251 	for (i = 0; i < get_desc_trans_len(desc); i++) {
2252 
2253 		wait_on_buffer(log_blocks[i]);
2254 		if (!buffer_uptodate(log_blocks[i])) {
2255 			reiserfs_warning(sb, "journal-1212",
2256 					 "REPLAY FAILURE fsck required! "
2257 					 "buffer write failed");
2258 			brelse_array(log_blocks + i,
2259 				     get_desc_trans_len(desc) - i);
2260 			brelse_array(real_blocks, get_desc_trans_len(desc));
2261 			brelse(c_bh);
2262 			brelse(d_bh);
2263 			kfree(log_blocks);
2264 			kfree(real_blocks);
2265 			return -1;
2266 		}
2267 		memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2268 		       real_blocks[i]->b_size);
2269 		set_buffer_uptodate(real_blocks[i]);
2270 		brelse(log_blocks[i]);
2271 	}
2272 	/* flush out the real blocks */
2273 	for (i = 0; i < get_desc_trans_len(desc); i++) {
2274 		set_buffer_dirty(real_blocks[i]);
2275 		write_dirty_buffer(real_blocks[i], 0);
2276 	}
2277 	for (i = 0; i < get_desc_trans_len(desc); i++) {
2278 		wait_on_buffer(real_blocks[i]);
2279 		if (!buffer_uptodate(real_blocks[i])) {
2280 			reiserfs_warning(sb, "journal-1226",
2281 					 "REPLAY FAILURE, fsck required! "
2282 					 "buffer write failed");
2283 			brelse_array(real_blocks + i,
2284 				     get_desc_trans_len(desc) - i);
2285 			brelse(c_bh);
2286 			brelse(d_bh);
2287 			kfree(log_blocks);
2288 			kfree(real_blocks);
2289 			return -1;
2290 		}
2291 		brelse(real_blocks[i]);
2292 	}
2293 	cur_dblock =
2294 	    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2295 	    ((trans_offset + get_desc_trans_len(desc) +
2296 	      2) % SB_ONDISK_JOURNAL_SIZE(sb));
2297 	reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2298 		       "journal-1095: setting journal " "start to offset %ld",
2299 		       cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2300 
2301 	/*
2302 	 * init starting values for the first transaction, in case
2303 	 * this is the last transaction to be replayed.
2304 	 */
2305 	journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2306 	journal->j_last_flush_trans_id = trans_id;
2307 	journal->j_trans_id = trans_id + 1;
2308 	/* check for trans_id overflow */
2309 	if (journal->j_trans_id == 0)
2310 		journal->j_trans_id = 10;
2311 	brelse(c_bh);
2312 	brelse(d_bh);
2313 	kfree(log_blocks);
2314 	kfree(real_blocks);
2315 	return 0;
2316 }
2317 
2318 /*
2319  * This function reads blocks starting from block and to max_block of bufsize
2320  * size (but no more than BUFNR blocks at a time). This proved to improve
2321  * mounting speed on self-rebuilding raid5 arrays at least.
2322  * Right now it is only used from journal code. But later we might use it
2323  * from other places.
2324  * Note: Do not use journal_getblk/sb_getblk functions here!
2325  */
2326 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2327 					   b_blocknr_t block, int bufsize,
2328 					   b_blocknr_t max_block)
2329 {
2330 	struct buffer_head *bhlist[BUFNR];
2331 	unsigned int blocks = BUFNR;
2332 	struct buffer_head *bh;
2333 	int i, j;
2334 
2335 	bh = __getblk(dev, block, bufsize);
2336 	if (buffer_uptodate(bh))
2337 		return (bh);
2338 
2339 	if (block + BUFNR > max_block) {
2340 		blocks = max_block - block;
2341 	}
2342 	bhlist[0] = bh;
2343 	j = 1;
2344 	for (i = 1; i < blocks; i++) {
2345 		bh = __getblk(dev, block + i, bufsize);
2346 		if (buffer_uptodate(bh)) {
2347 			brelse(bh);
2348 			break;
2349 		} else
2350 			bhlist[j++] = bh;
2351 	}
2352 	ll_rw_block(REQ_OP_READ, 0, j, bhlist);
2353 	for (i = 1; i < j; i++)
2354 		brelse(bhlist[i]);
2355 	bh = bhlist[0];
2356 	wait_on_buffer(bh);
2357 	if (buffer_uptodate(bh))
2358 		return bh;
2359 	brelse(bh);
2360 	return NULL;
2361 }
2362 
2363 /*
2364  * read and replay the log
2365  * on a clean unmount, the journal header's next unflushed pointer will be
2366  * to an invalid transaction.  This tests that before finding all the
2367  * transactions in the log, which makes normal mount times fast.
2368  *
2369  * After a crash, this starts with the next unflushed transaction, and
2370  * replays until it finds one too old, or invalid.
2371  *
2372  * On exit, it sets things up so the first transaction will work correctly.
2373  * NOTE: only called during fs mount
2374  */
2375 static int journal_read(struct super_block *sb)
2376 {
2377 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
2378 	struct reiserfs_journal_desc *desc;
2379 	unsigned int oldest_trans_id = 0;
2380 	unsigned int oldest_invalid_trans_id = 0;
2381 	time_t start;
2382 	unsigned long oldest_start = 0;
2383 	unsigned long cur_dblock = 0;
2384 	unsigned long newest_mount_id = 9;
2385 	struct buffer_head *d_bh;
2386 	struct reiserfs_journal_header *jh;
2387 	int valid_journal_header = 0;
2388 	int replay_count = 0;
2389 	int continue_replay = 1;
2390 	int ret;
2391 
2392 	cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2393 	reiserfs_info(sb, "checking transaction log (%pg)\n",
2394 		      journal->j_dev_bd);
2395 	start = get_seconds();
2396 
2397 	/*
2398 	 * step 1, read in the journal header block.  Check the transaction
2399 	 * it says is the first unflushed, and if that transaction is not
2400 	 * valid, replay is done
2401 	 */
2402 	journal->j_header_bh = journal_bread(sb,
2403 					     SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2404 					     + SB_ONDISK_JOURNAL_SIZE(sb));
2405 	if (!journal->j_header_bh) {
2406 		return 1;
2407 	}
2408 	jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2409 	if (le32_to_cpu(jh->j_first_unflushed_offset) <
2410 	    SB_ONDISK_JOURNAL_SIZE(sb)
2411 	    && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2412 		oldest_start =
2413 		    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2414 		    le32_to_cpu(jh->j_first_unflushed_offset);
2415 		oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2416 		newest_mount_id = le32_to_cpu(jh->j_mount_id);
2417 		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2418 			       "journal-1153: found in "
2419 			       "header: first_unflushed_offset %d, last_flushed_trans_id "
2420 			       "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2421 			       le32_to_cpu(jh->j_last_flush_trans_id));
2422 		valid_journal_header = 1;
2423 
2424 		/*
2425 		 * now, we try to read the first unflushed offset.  If it
2426 		 * is not valid, there is nothing more we can do, and it
2427 		 * makes no sense to read through the whole log.
2428 		 */
2429 		d_bh =
2430 		    journal_bread(sb,
2431 				  SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2432 				  le32_to_cpu(jh->j_first_unflushed_offset));
2433 		ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2434 		if (!ret) {
2435 			continue_replay = 0;
2436 		}
2437 		brelse(d_bh);
2438 		goto start_log_replay;
2439 	}
2440 
2441 	/*
2442 	 * ok, there are transactions that need to be replayed.  start
2443 	 * with the first log block, find all the valid transactions, and
2444 	 * pick out the oldest.
2445 	 */
2446 	while (continue_replay
2447 	       && cur_dblock <
2448 	       (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2449 		SB_ONDISK_JOURNAL_SIZE(sb))) {
2450 		/*
2451 		 * Note that it is required for blocksize of primary fs
2452 		 * device and journal device to be the same
2453 		 */
2454 		d_bh =
2455 		    reiserfs_breada(journal->j_dev_bd, cur_dblock,
2456 				    sb->s_blocksize,
2457 				    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2458 				    SB_ONDISK_JOURNAL_SIZE(sb));
2459 		ret =
2460 		    journal_transaction_is_valid(sb, d_bh,
2461 						 &oldest_invalid_trans_id,
2462 						 &newest_mount_id);
2463 		if (ret == 1) {
2464 			desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2465 			if (oldest_start == 0) {	/* init all oldest_ values */
2466 				oldest_trans_id = get_desc_trans_id(desc);
2467 				oldest_start = d_bh->b_blocknr;
2468 				newest_mount_id = get_desc_mount_id(desc);
2469 				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2470 					       "journal-1179: Setting "
2471 					       "oldest_start to offset %llu, trans_id %lu",
2472 					       oldest_start -
2473 					       SB_ONDISK_JOURNAL_1st_BLOCK
2474 					       (sb), oldest_trans_id);
2475 			} else if (oldest_trans_id > get_desc_trans_id(desc)) {
2476 				/* one we just read was older */
2477 				oldest_trans_id = get_desc_trans_id(desc);
2478 				oldest_start = d_bh->b_blocknr;
2479 				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2480 					       "journal-1180: Resetting "
2481 					       "oldest_start to offset %lu, trans_id %lu",
2482 					       oldest_start -
2483 					       SB_ONDISK_JOURNAL_1st_BLOCK
2484 					       (sb), oldest_trans_id);
2485 			}
2486 			if (newest_mount_id < get_desc_mount_id(desc)) {
2487 				newest_mount_id = get_desc_mount_id(desc);
2488 				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2489 					       "journal-1299: Setting "
2490 					       "newest_mount_id to %d",
2491 					       get_desc_mount_id(desc));
2492 			}
2493 			cur_dblock += get_desc_trans_len(desc) + 2;
2494 		} else {
2495 			cur_dblock++;
2496 		}
2497 		brelse(d_bh);
2498 	}
2499 
2500 start_log_replay:
2501 	cur_dblock = oldest_start;
2502 	if (oldest_trans_id) {
2503 		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2504 			       "journal-1206: Starting replay "
2505 			       "from offset %llu, trans_id %lu",
2506 			       cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2507 			       oldest_trans_id);
2508 
2509 	}
2510 	replay_count = 0;
2511 	while (continue_replay && oldest_trans_id > 0) {
2512 		ret =
2513 		    journal_read_transaction(sb, cur_dblock, oldest_start,
2514 					     oldest_trans_id, newest_mount_id);
2515 		if (ret < 0) {
2516 			return ret;
2517 		} else if (ret != 0) {
2518 			break;
2519 		}
2520 		cur_dblock =
2521 		    SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2522 		replay_count++;
2523 		if (cur_dblock == oldest_start)
2524 			break;
2525 	}
2526 
2527 	if (oldest_trans_id == 0) {
2528 		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2529 			       "journal-1225: No valid " "transactions found");
2530 	}
2531 	/*
2532 	 * j_start does not get set correctly if we don't replay any
2533 	 * transactions.  if we had a valid journal_header, set j_start
2534 	 * to the first unflushed transaction value, copy the trans_id
2535 	 * from the header
2536 	 */
2537 	if (valid_journal_header && replay_count == 0) {
2538 		journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2539 		journal->j_trans_id =
2540 		    le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2541 		/* check for trans_id overflow */
2542 		if (journal->j_trans_id == 0)
2543 			journal->j_trans_id = 10;
2544 		journal->j_last_flush_trans_id =
2545 		    le32_to_cpu(jh->j_last_flush_trans_id);
2546 		journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2547 	} else {
2548 		journal->j_mount_id = newest_mount_id + 1;
2549 	}
2550 	reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2551 		       "newest_mount_id to %lu", journal->j_mount_id);
2552 	journal->j_first_unflushed_offset = journal->j_start;
2553 	if (replay_count > 0) {
2554 		reiserfs_info(sb,
2555 			      "replayed %d transactions in %lu seconds\n",
2556 			      replay_count, get_seconds() - start);
2557 	}
2558 	/* needed to satisfy the locking in _update_journal_header_block */
2559 	reiserfs_write_lock(sb);
2560 	if (!bdev_read_only(sb->s_bdev) &&
2561 	    _update_journal_header_block(sb, journal->j_start,
2562 					 journal->j_last_flush_trans_id)) {
2563 		reiserfs_write_unlock(sb);
2564 		/*
2565 		 * replay failed, caller must call free_journal_ram and abort
2566 		 * the mount
2567 		 */
2568 		return -1;
2569 	}
2570 	reiserfs_write_unlock(sb);
2571 	return 0;
2572 }
2573 
2574 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2575 {
2576 	struct reiserfs_journal_list *jl;
2577 	jl = kzalloc(sizeof(struct reiserfs_journal_list),
2578 		     GFP_NOFS | __GFP_NOFAIL);
2579 	INIT_LIST_HEAD(&jl->j_list);
2580 	INIT_LIST_HEAD(&jl->j_working_list);
2581 	INIT_LIST_HEAD(&jl->j_tail_bh_list);
2582 	INIT_LIST_HEAD(&jl->j_bh_list);
2583 	mutex_init(&jl->j_commit_mutex);
2584 	SB_JOURNAL(s)->j_num_lists++;
2585 	get_journal_list(jl);
2586 	return jl;
2587 }
2588 
2589 static void journal_list_init(struct super_block *sb)
2590 {
2591 	SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2592 }
2593 
2594 static void release_journal_dev(struct super_block *super,
2595 			       struct reiserfs_journal *journal)
2596 {
2597 	if (journal->j_dev_bd != NULL) {
2598 		blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2599 		journal->j_dev_bd = NULL;
2600 	}
2601 }
2602 
2603 static int journal_init_dev(struct super_block *super,
2604 			    struct reiserfs_journal *journal,
2605 			    const char *jdev_name)
2606 {
2607 	int result;
2608 	dev_t jdev;
2609 	fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
2610 	char b[BDEVNAME_SIZE];
2611 
2612 	result = 0;
2613 
2614 	journal->j_dev_bd = NULL;
2615 	jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2616 	    new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2617 
2618 	if (bdev_read_only(super->s_bdev))
2619 		blkdev_mode = FMODE_READ;
2620 
2621 	/* there is no "jdev" option and journal is on separate device */
2622 	if ((!jdev_name || !jdev_name[0])) {
2623 		if (jdev == super->s_dev)
2624 			blkdev_mode &= ~FMODE_EXCL;
2625 		journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2626 						      journal);
2627 		journal->j_dev_mode = blkdev_mode;
2628 		if (IS_ERR(journal->j_dev_bd)) {
2629 			result = PTR_ERR(journal->j_dev_bd);
2630 			journal->j_dev_bd = NULL;
2631 			reiserfs_warning(super, "sh-458",
2632 					 "cannot init journal device '%s': %i",
2633 					 __bdevname(jdev, b), result);
2634 			return result;
2635 		} else if (jdev != super->s_dev)
2636 			set_blocksize(journal->j_dev_bd, super->s_blocksize);
2637 
2638 		return 0;
2639 	}
2640 
2641 	journal->j_dev_mode = blkdev_mode;
2642 	journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
2643 	if (IS_ERR(journal->j_dev_bd)) {
2644 		result = PTR_ERR(journal->j_dev_bd);
2645 		journal->j_dev_bd = NULL;
2646 		reiserfs_warning(super, "sh-457",
2647 				 "journal_init_dev: Cannot open '%s': %i",
2648 				 jdev_name, result);
2649 		return result;
2650 	}
2651 
2652 	set_blocksize(journal->j_dev_bd, super->s_blocksize);
2653 	reiserfs_info(super,
2654 		      "journal_init_dev: journal device: %pg\n",
2655 		      journal->j_dev_bd);
2656 	return 0;
2657 }
2658 
2659 /*
2660  * When creating/tuning a file system user can assign some
2661  * journal params within boundaries which depend on the ratio
2662  * blocksize/standard_blocksize.
2663  *
2664  * For blocks >= standard_blocksize transaction size should
2665  * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2666  * then JOURNAL_TRANS_MAX_DEFAULT.
2667  *
2668  * For blocks < standard_blocksize these boundaries should be
2669  * decreased proportionally.
2670  */
2671 #define REISERFS_STANDARD_BLKSIZE (4096)
2672 
2673 static int check_advise_trans_params(struct super_block *sb,
2674 				     struct reiserfs_journal *journal)
2675 {
2676         if (journal->j_trans_max) {
2677 		/* Non-default journal params.  Do sanity check for them. */
2678 	        int ratio = 1;
2679 		if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2680 		        ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2681 
2682 		if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2683 		    journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2684 		    SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2685 		    JOURNAL_MIN_RATIO) {
2686 			reiserfs_warning(sb, "sh-462",
2687 					 "bad transaction max size (%u). "
2688 					 "FSCK?", journal->j_trans_max);
2689 			return 1;
2690 		}
2691 		if (journal->j_max_batch != (journal->j_trans_max) *
2692 		        JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2693 			reiserfs_warning(sb, "sh-463",
2694 					 "bad transaction max batch (%u). "
2695 					 "FSCK?", journal->j_max_batch);
2696 			return 1;
2697 		}
2698 	} else {
2699 		/*
2700 		 * Default journal params.
2701 		 * The file system was created by old version
2702 		 * of mkreiserfs, so some fields contain zeros,
2703 		 * and we need to advise proper values for them
2704 		 */
2705 		if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2706 			reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2707 					 sb->s_blocksize);
2708 			return 1;
2709 		}
2710 		journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2711 		journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2712 		journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2713 	}
2714 	return 0;
2715 }
2716 
2717 /* must be called once on fs mount.  calls journal_read for you */
2718 int journal_init(struct super_block *sb, const char *j_dev_name,
2719 		 int old_format, unsigned int commit_max_age)
2720 {
2721 	int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2722 	struct buffer_head *bhjh;
2723 	struct reiserfs_super_block *rs;
2724 	struct reiserfs_journal_header *jh;
2725 	struct reiserfs_journal *journal;
2726 	struct reiserfs_journal_list *jl;
2727 	int ret;
2728 
2729 	journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal));
2730 	if (!journal) {
2731 		reiserfs_warning(sb, "journal-1256",
2732 				 "unable to get memory for journal structure");
2733 		return 1;
2734 	}
2735 	INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2736 	INIT_LIST_HEAD(&journal->j_prealloc_list);
2737 	INIT_LIST_HEAD(&journal->j_working_list);
2738 	INIT_LIST_HEAD(&journal->j_journal_list);
2739 	journal->j_persistent_trans = 0;
2740 	if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2741 					   reiserfs_bmap_count(sb)))
2742 		goto free_and_return;
2743 
2744 	allocate_bitmap_nodes(sb);
2745 
2746 	/* reserved for journal area support */
2747 	SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2748 						 REISERFS_OLD_DISK_OFFSET_IN_BYTES
2749 						 / sb->s_blocksize +
2750 						 reiserfs_bmap_count(sb) +
2751 						 1 :
2752 						 REISERFS_DISK_OFFSET_IN_BYTES /
2753 						 sb->s_blocksize + 2);
2754 
2755 	/*
2756 	 * Sanity check to see is the standard journal fitting
2757 	 * within first bitmap (actual for small blocksizes)
2758 	 */
2759 	if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2760 	    (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2761 	     SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2762 		reiserfs_warning(sb, "journal-1393",
2763 				 "journal does not fit for area addressed "
2764 				 "by first of bitmap blocks. It starts at "
2765 				 "%u and its size is %u. Block size %ld",
2766 				 SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2767 				 SB_ONDISK_JOURNAL_SIZE(sb),
2768 				 sb->s_blocksize);
2769 		goto free_and_return;
2770 	}
2771 
2772 	if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2773 		reiserfs_warning(sb, "sh-462",
2774 				 "unable to initialize journal device");
2775 		goto free_and_return;
2776 	}
2777 
2778 	rs = SB_DISK_SUPER_BLOCK(sb);
2779 
2780 	/* read journal header */
2781 	bhjh = journal_bread(sb,
2782 			     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2783 			     SB_ONDISK_JOURNAL_SIZE(sb));
2784 	if (!bhjh) {
2785 		reiserfs_warning(sb, "sh-459",
2786 				 "unable to read journal header");
2787 		goto free_and_return;
2788 	}
2789 	jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2790 
2791 	/* make sure that journal matches to the super block */
2792 	if (is_reiserfs_jr(rs)
2793 	    && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2794 		sb_jp_journal_magic(rs))) {
2795 		reiserfs_warning(sb, "sh-460",
2796 				 "journal header magic %x (device %pg) does "
2797 				 "not match to magic found in super block %x",
2798 				 jh->jh_journal.jp_journal_magic,
2799 				 journal->j_dev_bd,
2800 				 sb_jp_journal_magic(rs));
2801 		brelse(bhjh);
2802 		goto free_and_return;
2803 	}
2804 
2805 	journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2806 	journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2807 	journal->j_max_commit_age =
2808 	    le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2809 	journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2810 
2811 	if (check_advise_trans_params(sb, journal) != 0)
2812 	        goto free_and_return;
2813 	journal->j_default_max_commit_age = journal->j_max_commit_age;
2814 
2815 	if (commit_max_age != 0) {
2816 		journal->j_max_commit_age = commit_max_age;
2817 		journal->j_max_trans_age = commit_max_age;
2818 	}
2819 
2820 	reiserfs_info(sb, "journal params: device %pg, size %u, "
2821 		      "journal first block %u, max trans len %u, max batch %u, "
2822 		      "max commit age %u, max trans age %u\n",
2823 		      journal->j_dev_bd,
2824 		      SB_ONDISK_JOURNAL_SIZE(sb),
2825 		      SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2826 		      journal->j_trans_max,
2827 		      journal->j_max_batch,
2828 		      journal->j_max_commit_age, journal->j_max_trans_age);
2829 
2830 	brelse(bhjh);
2831 
2832 	journal->j_list_bitmap_index = 0;
2833 	journal_list_init(sb);
2834 
2835 	memset(journal->j_list_hash_table, 0,
2836 	       JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2837 
2838 	INIT_LIST_HEAD(&journal->j_dirty_buffers);
2839 	spin_lock_init(&journal->j_dirty_buffers_lock);
2840 
2841 	journal->j_start = 0;
2842 	journal->j_len = 0;
2843 	journal->j_len_alloc = 0;
2844 	atomic_set(&journal->j_wcount, 0);
2845 	atomic_set(&journal->j_async_throttle, 0);
2846 	journal->j_bcount = 0;
2847 	journal->j_trans_start_time = 0;
2848 	journal->j_last = NULL;
2849 	journal->j_first = NULL;
2850 	init_waitqueue_head(&journal->j_join_wait);
2851 	mutex_init(&journal->j_mutex);
2852 	mutex_init(&journal->j_flush_mutex);
2853 
2854 	journal->j_trans_id = 10;
2855 	journal->j_mount_id = 10;
2856 	journal->j_state = 0;
2857 	atomic_set(&journal->j_jlock, 0);
2858 	journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2859 	journal->j_cnode_free_orig = journal->j_cnode_free_list;
2860 	journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2861 	journal->j_cnode_used = 0;
2862 	journal->j_must_wait = 0;
2863 
2864 	if (journal->j_cnode_free == 0) {
2865 		reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2866 		                 "allocation failed (%ld bytes). Journal is "
2867 		                 "too large for available memory. Usually "
2868 		                 "this is due to a journal that is too large.",
2869 		                 sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2870         	goto free_and_return;
2871 	}
2872 
2873 	init_journal_hash(sb);
2874 	jl = journal->j_current_jl;
2875 
2876 	/*
2877 	 * get_list_bitmap() may call flush_commit_list() which
2878 	 * requires the lock. Calling flush_commit_list() shouldn't happen
2879 	 * this early but I like to be paranoid.
2880 	 */
2881 	reiserfs_write_lock(sb);
2882 	jl->j_list_bitmap = get_list_bitmap(sb, jl);
2883 	reiserfs_write_unlock(sb);
2884 	if (!jl->j_list_bitmap) {
2885 		reiserfs_warning(sb, "journal-2005",
2886 				 "get_list_bitmap failed for journal list 0");
2887 		goto free_and_return;
2888 	}
2889 
2890 	ret = journal_read(sb);
2891 	if (ret < 0) {
2892 		reiserfs_warning(sb, "reiserfs-2006",
2893 				 "Replay Failure, unable to mount");
2894 		goto free_and_return;
2895 	}
2896 
2897 	INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2898 	journal->j_work_sb = sb;
2899 	return 0;
2900 free_and_return:
2901 	free_journal_ram(sb);
2902 	return 1;
2903 }
2904 
2905 /*
2906  * test for a polite end of the current transaction.  Used by file_write,
2907  * and should be used by delete to make sure they don't write more than
2908  * can fit inside a single transaction
2909  */
2910 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2911 				   int new_alloc)
2912 {
2913 	struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2914 	time_t now = get_seconds();
2915 	/* cannot restart while nested */
2916 	BUG_ON(!th->t_trans_id);
2917 	if (th->t_refcount > 1)
2918 		return 0;
2919 	if (journal->j_must_wait > 0 ||
2920 	    (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2921 	    atomic_read(&journal->j_jlock) ||
2922 	    (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2923 	    journal->j_cnode_free < (journal->j_trans_max * 3)) {
2924 		return 1;
2925 	}
2926 
2927 	journal->j_len_alloc += new_alloc;
2928 	th->t_blocks_allocated += new_alloc ;
2929 	return 0;
2930 }
2931 
2932 /* this must be called inside a transaction */
2933 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2934 {
2935 	struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2936 	BUG_ON(!th->t_trans_id);
2937 	journal->j_must_wait = 1;
2938 	set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2939 	return;
2940 }
2941 
2942 /* this must be called without a transaction started */
2943 void reiserfs_allow_writes(struct super_block *s)
2944 {
2945 	struct reiserfs_journal *journal = SB_JOURNAL(s);
2946 	clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2947 	wake_up(&journal->j_join_wait);
2948 }
2949 
2950 /* this must be called without a transaction started */
2951 void reiserfs_wait_on_write_block(struct super_block *s)
2952 {
2953 	struct reiserfs_journal *journal = SB_JOURNAL(s);
2954 	wait_event(journal->j_join_wait,
2955 		   !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2956 }
2957 
2958 static void queue_log_writer(struct super_block *s)
2959 {
2960 	wait_queue_entry_t wait;
2961 	struct reiserfs_journal *journal = SB_JOURNAL(s);
2962 	set_bit(J_WRITERS_QUEUED, &journal->j_state);
2963 
2964 	/*
2965 	 * we don't want to use wait_event here because
2966 	 * we only want to wait once.
2967 	 */
2968 	init_waitqueue_entry(&wait, current);
2969 	add_wait_queue(&journal->j_join_wait, &wait);
2970 	set_current_state(TASK_UNINTERRUPTIBLE);
2971 	if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
2972 		int depth = reiserfs_write_unlock_nested(s);
2973 		schedule();
2974 		reiserfs_write_lock_nested(s, depth);
2975 	}
2976 	__set_current_state(TASK_RUNNING);
2977 	remove_wait_queue(&journal->j_join_wait, &wait);
2978 }
2979 
2980 static void wake_queued_writers(struct super_block *s)
2981 {
2982 	struct reiserfs_journal *journal = SB_JOURNAL(s);
2983 	if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2984 		wake_up(&journal->j_join_wait);
2985 }
2986 
2987 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
2988 {
2989 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
2990 	unsigned long bcount = journal->j_bcount;
2991 	while (1) {
2992 		int depth;
2993 
2994 		depth = reiserfs_write_unlock_nested(sb);
2995 		schedule_timeout_uninterruptible(1);
2996 		reiserfs_write_lock_nested(sb, depth);
2997 
2998 		journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2999 		while ((atomic_read(&journal->j_wcount) > 0 ||
3000 			atomic_read(&journal->j_jlock)) &&
3001 		       journal->j_trans_id == trans_id) {
3002 			queue_log_writer(sb);
3003 		}
3004 		if (journal->j_trans_id != trans_id)
3005 			break;
3006 		if (bcount == journal->j_bcount)
3007 			break;
3008 		bcount = journal->j_bcount;
3009 	}
3010 }
3011 
3012 /*
3013  * join == true if you must join an existing transaction.
3014  * join == false if you can deal with waiting for others to finish
3015  *
3016  * this will block until the transaction is joinable.  send the number of
3017  * blocks you expect to use in nblocks.
3018 */
3019 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3020 			      struct super_block *sb, unsigned long nblocks,
3021 			      int join)
3022 {
3023 	time_t now = get_seconds();
3024 	unsigned int old_trans_id;
3025 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3026 	struct reiserfs_transaction_handle myth;
3027 	int sched_count = 0;
3028 	int retval;
3029 	int depth;
3030 
3031 	reiserfs_check_lock_depth(sb, "journal_begin");
3032 	BUG_ON(nblocks > journal->j_trans_max);
3033 
3034 	PROC_INFO_INC(sb, journal.journal_being);
3035 	/* set here for journal_join */
3036 	th->t_refcount = 1;
3037 	th->t_super = sb;
3038 
3039 relock:
3040 	lock_journal(sb);
3041 	if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3042 		unlock_journal(sb);
3043 		retval = journal->j_errno;
3044 		goto out_fail;
3045 	}
3046 	journal->j_bcount++;
3047 
3048 	if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3049 		unlock_journal(sb);
3050 		depth = reiserfs_write_unlock_nested(sb);
3051 		reiserfs_wait_on_write_block(sb);
3052 		reiserfs_write_lock_nested(sb, depth);
3053 		PROC_INFO_INC(sb, journal.journal_relock_writers);
3054 		goto relock;
3055 	}
3056 	now = get_seconds();
3057 
3058 	/*
3059 	 * if there is no room in the journal OR
3060 	 * if this transaction is too old, and we weren't called joinable,
3061 	 * wait for it to finish before beginning we don't sleep if there
3062 	 * aren't other writers
3063 	 */
3064 
3065 	if ((!join && journal->j_must_wait > 0) ||
3066 	    (!join
3067 	     && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3068 	    || (!join && atomic_read(&journal->j_wcount) > 0
3069 		&& journal->j_trans_start_time > 0
3070 		&& (now - journal->j_trans_start_time) >
3071 		journal->j_max_trans_age) || (!join
3072 					      && atomic_read(&journal->j_jlock))
3073 	    || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3074 
3075 		old_trans_id = journal->j_trans_id;
3076 		/* allow others to finish this transaction */
3077 		unlock_journal(sb);
3078 
3079 		if (!join && (journal->j_len_alloc + nblocks + 2) >=
3080 		    journal->j_max_batch &&
3081 		    ((journal->j_len + nblocks + 2) * 100) <
3082 		    (journal->j_len_alloc * 75)) {
3083 			if (atomic_read(&journal->j_wcount) > 10) {
3084 				sched_count++;
3085 				queue_log_writer(sb);
3086 				goto relock;
3087 			}
3088 		}
3089 		/*
3090 		 * don't mess with joining the transaction if all we
3091 		 * have to do is wait for someone else to do a commit
3092 		 */
3093 		if (atomic_read(&journal->j_jlock)) {
3094 			while (journal->j_trans_id == old_trans_id &&
3095 			       atomic_read(&journal->j_jlock)) {
3096 				queue_log_writer(sb);
3097 			}
3098 			goto relock;
3099 		}
3100 		retval = journal_join(&myth, sb);
3101 		if (retval)
3102 			goto out_fail;
3103 
3104 		/* someone might have ended the transaction while we joined */
3105 		if (old_trans_id != journal->j_trans_id) {
3106 			retval = do_journal_end(&myth, 0);
3107 		} else {
3108 			retval = do_journal_end(&myth, COMMIT_NOW);
3109 		}
3110 
3111 		if (retval)
3112 			goto out_fail;
3113 
3114 		PROC_INFO_INC(sb, journal.journal_relock_wcount);
3115 		goto relock;
3116 	}
3117 	/* we are the first writer, set trans_id */
3118 	if (journal->j_trans_start_time == 0) {
3119 		journal->j_trans_start_time = get_seconds();
3120 	}
3121 	atomic_inc(&journal->j_wcount);
3122 	journal->j_len_alloc += nblocks;
3123 	th->t_blocks_logged = 0;
3124 	th->t_blocks_allocated = nblocks;
3125 	th->t_trans_id = journal->j_trans_id;
3126 	unlock_journal(sb);
3127 	INIT_LIST_HEAD(&th->t_list);
3128 	return 0;
3129 
3130 out_fail:
3131 	memset(th, 0, sizeof(*th));
3132 	/*
3133 	 * Re-set th->t_super, so we can properly keep track of how many
3134 	 * persistent transactions there are. We need to do this so if this
3135 	 * call is part of a failed restart_transaction, we can free it later
3136 	 */
3137 	th->t_super = sb;
3138 	return retval;
3139 }
3140 
3141 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3142 								    super_block
3143 								    *s,
3144 								    int nblocks)
3145 {
3146 	int ret;
3147 	struct reiserfs_transaction_handle *th;
3148 
3149 	/*
3150 	 * if we're nesting into an existing transaction.  It will be
3151 	 * persistent on its own
3152 	 */
3153 	if (reiserfs_transaction_running(s)) {
3154 		th = current->journal_info;
3155 		th->t_refcount++;
3156 		BUG_ON(th->t_refcount < 2);
3157 
3158 		return th;
3159 	}
3160 	th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3161 	if (!th)
3162 		return NULL;
3163 	ret = journal_begin(th, s, nblocks);
3164 	if (ret) {
3165 		kfree(th);
3166 		return NULL;
3167 	}
3168 
3169 	SB_JOURNAL(s)->j_persistent_trans++;
3170 	return th;
3171 }
3172 
3173 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3174 {
3175 	struct super_block *s = th->t_super;
3176 	int ret = 0;
3177 	if (th->t_trans_id)
3178 		ret = journal_end(th);
3179 	else
3180 		ret = -EIO;
3181 	if (th->t_refcount == 0) {
3182 		SB_JOURNAL(s)->j_persistent_trans--;
3183 		kfree(th);
3184 	}
3185 	return ret;
3186 }
3187 
3188 static int journal_join(struct reiserfs_transaction_handle *th,
3189 			struct super_block *sb)
3190 {
3191 	struct reiserfs_transaction_handle *cur_th = current->journal_info;
3192 
3193 	/*
3194 	 * this keeps do_journal_end from NULLing out the
3195 	 * current->journal_info pointer
3196 	 */
3197 	th->t_handle_save = cur_th;
3198 	BUG_ON(cur_th && cur_th->t_refcount > 1);
3199 	return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN);
3200 }
3201 
3202 int journal_join_abort(struct reiserfs_transaction_handle *th,
3203 		       struct super_block *sb)
3204 {
3205 	struct reiserfs_transaction_handle *cur_th = current->journal_info;
3206 
3207 	/*
3208 	 * this keeps do_journal_end from NULLing out the
3209 	 * current->journal_info pointer
3210 	 */
3211 	th->t_handle_save = cur_th;
3212 	BUG_ON(cur_th && cur_th->t_refcount > 1);
3213 	return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT);
3214 }
3215 
3216 int journal_begin(struct reiserfs_transaction_handle *th,
3217 		  struct super_block *sb, unsigned long nblocks)
3218 {
3219 	struct reiserfs_transaction_handle *cur_th = current->journal_info;
3220 	int ret;
3221 
3222 	th->t_handle_save = NULL;
3223 	if (cur_th) {
3224 		/* we are nesting into the current transaction */
3225 		if (cur_th->t_super == sb) {
3226 			BUG_ON(!cur_th->t_refcount);
3227 			cur_th->t_refcount++;
3228 			memcpy(th, cur_th, sizeof(*th));
3229 			if (th->t_refcount <= 1)
3230 				reiserfs_warning(sb, "reiserfs-2005",
3231 						 "BAD: refcount <= 1, but "
3232 						 "journal_info != 0");
3233 			return 0;
3234 		} else {
3235 			/*
3236 			 * we've ended up with a handle from a different
3237 			 * filesystem.  save it and restore on journal_end.
3238 			 * This should never really happen...
3239 			 */
3240 			reiserfs_warning(sb, "clm-2100",
3241 					 "nesting info a different FS");
3242 			th->t_handle_save = current->journal_info;
3243 			current->journal_info = th;
3244 		}
3245 	} else {
3246 		current->journal_info = th;
3247 	}
3248 	ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3249 	BUG_ON(current->journal_info != th);
3250 
3251 	/*
3252 	 * I guess this boils down to being the reciprocal of clm-2100 above.
3253 	 * If do_journal_begin_r fails, we need to put it back, since
3254 	 * journal_end won't be called to do it. */
3255 	if (ret)
3256 		current->journal_info = th->t_handle_save;
3257 	else
3258 		BUG_ON(!th->t_refcount);
3259 
3260 	return ret;
3261 }
3262 
3263 /*
3264  * puts bh into the current transaction.  If it was already there, reorders
3265  * removes the old pointers from the hash, and puts new ones in (to make
3266  * sure replay happen in the right order).
3267  *
3268  * if it was dirty, cleans and files onto the clean list.  I can't let it
3269  * be dirty again until the transaction is committed.
3270  *
3271  * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3272  */
3273 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3274 		       struct buffer_head *bh)
3275 {
3276 	struct super_block *sb = th->t_super;
3277 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3278 	struct reiserfs_journal_cnode *cn = NULL;
3279 	int count_already_incd = 0;
3280 	int prepared = 0;
3281 	BUG_ON(!th->t_trans_id);
3282 
3283 	PROC_INFO_INC(sb, journal.mark_dirty);
3284 	if (th->t_trans_id != journal->j_trans_id) {
3285 		reiserfs_panic(th->t_super, "journal-1577",
3286 			       "handle trans id %ld != current trans id %ld",
3287 			       th->t_trans_id, journal->j_trans_id);
3288 	}
3289 
3290 	prepared = test_clear_buffer_journal_prepared(bh);
3291 	clear_buffer_journal_restore_dirty(bh);
3292 	/* already in this transaction, we are done */
3293 	if (buffer_journaled(bh)) {
3294 		PROC_INFO_INC(sb, journal.mark_dirty_already);
3295 		return 0;
3296 	}
3297 
3298 	/*
3299 	 * this must be turned into a panic instead of a warning.  We can't
3300 	 * allow a dirty or journal_dirty or locked buffer to be logged, as
3301 	 * some changes could get to disk too early.  NOT GOOD.
3302 	 */
3303 	if (!prepared || buffer_dirty(bh)) {
3304 		reiserfs_warning(sb, "journal-1777",
3305 				 "buffer %llu bad state "
3306 				 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3307 				 (unsigned long long)bh->b_blocknr,
3308 				 prepared ? ' ' : '!',
3309 				 buffer_locked(bh) ? ' ' : '!',
3310 				 buffer_dirty(bh) ? ' ' : '!',
3311 				 buffer_journal_dirty(bh) ? ' ' : '!');
3312 	}
3313 
3314 	if (atomic_read(&journal->j_wcount) <= 0) {
3315 		reiserfs_warning(sb, "journal-1409",
3316 				 "returning because j_wcount was %d",
3317 				 atomic_read(&journal->j_wcount));
3318 		return 1;
3319 	}
3320 	/*
3321 	 * this error means I've screwed up, and we've overflowed
3322 	 * the transaction.  Nothing can be done here, except make the
3323 	 * FS readonly or panic.
3324 	 */
3325 	if (journal->j_len >= journal->j_trans_max) {
3326 		reiserfs_panic(th->t_super, "journal-1413",
3327 			       "j_len (%lu) is too big",
3328 			       journal->j_len);
3329 	}
3330 
3331 	if (buffer_journal_dirty(bh)) {
3332 		count_already_incd = 1;
3333 		PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3334 		clear_buffer_journal_dirty(bh);
3335 	}
3336 
3337 	if (journal->j_len > journal->j_len_alloc) {
3338 		journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3339 	}
3340 
3341 	set_buffer_journaled(bh);
3342 
3343 	/* now put this guy on the end */
3344 	if (!cn) {
3345 		cn = get_cnode(sb);
3346 		if (!cn) {
3347 			reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3348 		}
3349 
3350 		if (th->t_blocks_logged == th->t_blocks_allocated) {
3351 			th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3352 			journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3353 		}
3354 		th->t_blocks_logged++;
3355 		journal->j_len++;
3356 
3357 		cn->bh = bh;
3358 		cn->blocknr = bh->b_blocknr;
3359 		cn->sb = sb;
3360 		cn->jlist = NULL;
3361 		insert_journal_hash(journal->j_hash_table, cn);
3362 		if (!count_already_incd) {
3363 			get_bh(bh);
3364 		}
3365 	}
3366 	cn->next = NULL;
3367 	cn->prev = journal->j_last;
3368 	cn->bh = bh;
3369 	if (journal->j_last) {
3370 		journal->j_last->next = cn;
3371 		journal->j_last = cn;
3372 	} else {
3373 		journal->j_first = cn;
3374 		journal->j_last = cn;
3375 	}
3376 	reiserfs_schedule_old_flush(sb);
3377 	return 0;
3378 }
3379 
3380 int journal_end(struct reiserfs_transaction_handle *th)
3381 {
3382 	struct super_block *sb = th->t_super;
3383 	if (!current->journal_info && th->t_refcount > 1)
3384 		reiserfs_warning(sb, "REISER-NESTING",
3385 				 "th NULL, refcount %d", th->t_refcount);
3386 
3387 	if (!th->t_trans_id) {
3388 		WARN_ON(1);
3389 		return -EIO;
3390 	}
3391 
3392 	th->t_refcount--;
3393 	if (th->t_refcount > 0) {
3394 		struct reiserfs_transaction_handle *cur_th =
3395 		    current->journal_info;
3396 
3397 		/*
3398 		 * we aren't allowed to close a nested transaction on a
3399 		 * different filesystem from the one in the task struct
3400 		 */
3401 		BUG_ON(cur_th->t_super != th->t_super);
3402 
3403 		if (th != cur_th) {
3404 			memcpy(current->journal_info, th, sizeof(*th));
3405 			th->t_trans_id = 0;
3406 		}
3407 		return 0;
3408 	} else {
3409 		return do_journal_end(th, 0);
3410 	}
3411 }
3412 
3413 /*
3414  * removes from the current transaction, relsing and descrementing any counters.
3415  * also files the removed buffer directly onto the clean list
3416  *
3417  * called by journal_mark_freed when a block has been deleted
3418  *
3419  * returns 1 if it cleaned and relsed the buffer. 0 otherwise
3420  */
3421 static int remove_from_transaction(struct super_block *sb,
3422 				   b_blocknr_t blocknr, int already_cleaned)
3423 {
3424 	struct buffer_head *bh;
3425 	struct reiserfs_journal_cnode *cn;
3426 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3427 	int ret = 0;
3428 
3429 	cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3430 	if (!cn || !cn->bh) {
3431 		return ret;
3432 	}
3433 	bh = cn->bh;
3434 	if (cn->prev) {
3435 		cn->prev->next = cn->next;
3436 	}
3437 	if (cn->next) {
3438 		cn->next->prev = cn->prev;
3439 	}
3440 	if (cn == journal->j_first) {
3441 		journal->j_first = cn->next;
3442 	}
3443 	if (cn == journal->j_last) {
3444 		journal->j_last = cn->prev;
3445 	}
3446 	if (bh)
3447 		remove_journal_hash(sb, journal->j_hash_table, NULL,
3448 				    bh->b_blocknr, 0);
3449 	clear_buffer_journaled(bh);	/* don't log this one */
3450 
3451 	if (!already_cleaned) {
3452 		clear_buffer_journal_dirty(bh);
3453 		clear_buffer_dirty(bh);
3454 		clear_buffer_journal_test(bh);
3455 		put_bh(bh);
3456 		if (atomic_read(&bh->b_count) < 0) {
3457 			reiserfs_warning(sb, "journal-1752",
3458 					 "b_count < 0");
3459 		}
3460 		ret = 1;
3461 	}
3462 	journal->j_len--;
3463 	journal->j_len_alloc--;
3464 	free_cnode(sb, cn);
3465 	return ret;
3466 }
3467 
3468 /*
3469  * for any cnode in a journal list, it can only be dirtied of all the
3470  * transactions that include it are committed to disk.
3471  * this checks through each transaction, and returns 1 if you are allowed
3472  * to dirty, and 0 if you aren't
3473  *
3474  * it is called by dirty_journal_list, which is called after
3475  * flush_commit_list has gotten all the log blocks for a given
3476  * transaction on disk
3477  *
3478  */
3479 static int can_dirty(struct reiserfs_journal_cnode *cn)
3480 {
3481 	struct super_block *sb = cn->sb;
3482 	b_blocknr_t blocknr = cn->blocknr;
3483 	struct reiserfs_journal_cnode *cur = cn->hprev;
3484 	int can_dirty = 1;
3485 
3486 	/*
3487 	 * first test hprev.  These are all newer than cn, so any node here
3488 	 * with the same block number and dev means this node can't be sent
3489 	 * to disk right now.
3490 	 */
3491 	while (cur && can_dirty) {
3492 		if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3493 		    cur->blocknr == blocknr) {
3494 			can_dirty = 0;
3495 		}
3496 		cur = cur->hprev;
3497 	}
3498 	/*
3499 	 * then test hnext.  These are all older than cn.  As long as they
3500 	 * are committed to the log, it is safe to write cn to disk
3501 	 */
3502 	cur = cn->hnext;
3503 	while (cur && can_dirty) {
3504 		if (cur->jlist && cur->jlist->j_len > 0 &&
3505 		    atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh &&
3506 		    cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3507 			can_dirty = 0;
3508 		}
3509 		cur = cur->hnext;
3510 	}
3511 	return can_dirty;
3512 }
3513 
3514 /*
3515  * syncs the commit blocks, but does not force the real buffers to disk
3516  * will wait until the current transaction is done/committed before returning
3517  */
3518 int journal_end_sync(struct reiserfs_transaction_handle *th)
3519 {
3520 	struct super_block *sb = th->t_super;
3521 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3522 
3523 	BUG_ON(!th->t_trans_id);
3524 	/* you can sync while nested, very, very bad */
3525 	BUG_ON(th->t_refcount > 1);
3526 	if (journal->j_len == 0) {
3527 		reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3528 					     1);
3529 		journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
3530 	}
3531 	return do_journal_end(th, COMMIT_NOW | WAIT);
3532 }
3533 
3534 /* writeback the pending async commits to disk */
3535 static void flush_async_commits(struct work_struct *work)
3536 {
3537 	struct reiserfs_journal *journal =
3538 		container_of(work, struct reiserfs_journal, j_work.work);
3539 	struct super_block *sb = journal->j_work_sb;
3540 	struct reiserfs_journal_list *jl;
3541 	struct list_head *entry;
3542 
3543 	reiserfs_write_lock(sb);
3544 	if (!list_empty(&journal->j_journal_list)) {
3545 		/* last entry is the youngest, commit it and you get everything */
3546 		entry = journal->j_journal_list.prev;
3547 		jl = JOURNAL_LIST_ENTRY(entry);
3548 		flush_commit_list(sb, jl, 1);
3549 	}
3550 	reiserfs_write_unlock(sb);
3551 }
3552 
3553 /*
3554  * flushes any old transactions to disk
3555  * ends the current transaction if it is too old
3556  */
3557 void reiserfs_flush_old_commits(struct super_block *sb)
3558 {
3559 	time_t now;
3560 	struct reiserfs_transaction_handle th;
3561 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3562 
3563 	now = get_seconds();
3564 	/*
3565 	 * safety check so we don't flush while we are replaying the log during
3566 	 * mount
3567 	 */
3568 	if (list_empty(&journal->j_journal_list))
3569 		return;
3570 
3571 	/*
3572 	 * check the current transaction.  If there are no writers, and it is
3573 	 * too old, finish it, and force the commit blocks to disk
3574 	 */
3575 	if (atomic_read(&journal->j_wcount) <= 0 &&
3576 	    journal->j_trans_start_time > 0 &&
3577 	    journal->j_len > 0 &&
3578 	    (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3579 		if (!journal_join(&th, sb)) {
3580 			reiserfs_prepare_for_journal(sb,
3581 						     SB_BUFFER_WITH_SB(sb),
3582 						     1);
3583 			journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3584 
3585 			/*
3586 			 * we're only being called from kreiserfsd, it makes
3587 			 * no sense to do an async commit so that kreiserfsd
3588 			 * can do it later
3589 			 */
3590 			do_journal_end(&th, COMMIT_NOW | WAIT);
3591 		}
3592 	}
3593 }
3594 
3595 /*
3596  * returns 0 if do_journal_end should return right away, returns 1 if
3597  * do_journal_end should finish the commit
3598  *
3599  * if the current transaction is too old, but still has writers, this will
3600  * wait on j_join_wait until all the writers are done.  By the time it
3601  * wakes up, the transaction it was called has already ended, so it just
3602  * flushes the commit list and returns 0.
3603  *
3604  * Won't batch when flush or commit_now is set.  Also won't batch when
3605  * others are waiting on j_join_wait.
3606  *
3607  * Note, we can't allow the journal_end to proceed while there are still
3608  * writers in the log.
3609  */
3610 static int check_journal_end(struct reiserfs_transaction_handle *th, int flags)
3611 {
3612 
3613 	time_t now;
3614 	int flush = flags & FLUSH_ALL;
3615 	int commit_now = flags & COMMIT_NOW;
3616 	int wait_on_commit = flags & WAIT;
3617 	struct reiserfs_journal_list *jl;
3618 	struct super_block *sb = th->t_super;
3619 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3620 
3621 	BUG_ON(!th->t_trans_id);
3622 
3623 	if (th->t_trans_id != journal->j_trans_id) {
3624 		reiserfs_panic(th->t_super, "journal-1577",
3625 			       "handle trans id %ld != current trans id %ld",
3626 			       th->t_trans_id, journal->j_trans_id);
3627 	}
3628 
3629 	journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3630 	/* <= 0 is allowed.  unmounting might not call begin */
3631 	if (atomic_read(&journal->j_wcount) > 0)
3632 		atomic_dec(&journal->j_wcount);
3633 
3634 	/*
3635 	 * BUG, deal with case where j_len is 0, but people previously
3636 	 * freed blocks need to be released will be dealt with by next
3637 	 * transaction that actually writes something, but should be taken
3638 	 * care of in this trans
3639 	 */
3640 	BUG_ON(journal->j_len == 0);
3641 
3642 	/*
3643 	 * if wcount > 0, and we are called to with flush or commit_now,
3644 	 * we wait on j_join_wait.  We will wake up when the last writer has
3645 	 * finished the transaction, and started it on its way to the disk.
3646 	 * Then, we flush the commit or journal list, and just return 0
3647 	 * because the rest of journal end was already done for this
3648 	 * transaction.
3649 	 */
3650 	if (atomic_read(&journal->j_wcount) > 0) {
3651 		if (flush || commit_now) {
3652 			unsigned trans_id;
3653 
3654 			jl = journal->j_current_jl;
3655 			trans_id = jl->j_trans_id;
3656 			if (wait_on_commit)
3657 				jl->j_state |= LIST_COMMIT_PENDING;
3658 			atomic_set(&journal->j_jlock, 1);
3659 			if (flush) {
3660 				journal->j_next_full_flush = 1;
3661 			}
3662 			unlock_journal(sb);
3663 
3664 			/*
3665 			 * sleep while the current transaction is
3666 			 * still j_jlocked
3667 			 */
3668 			while (journal->j_trans_id == trans_id) {
3669 				if (atomic_read(&journal->j_jlock)) {
3670 					queue_log_writer(sb);
3671 				} else {
3672 					lock_journal(sb);
3673 					if (journal->j_trans_id == trans_id) {
3674 						atomic_set(&journal->j_jlock,
3675 							   1);
3676 					}
3677 					unlock_journal(sb);
3678 				}
3679 			}
3680 			BUG_ON(journal->j_trans_id == trans_id);
3681 
3682 			if (commit_now
3683 			    && journal_list_still_alive(sb, trans_id)
3684 			    && wait_on_commit) {
3685 				flush_commit_list(sb, jl, 1);
3686 			}
3687 			return 0;
3688 		}
3689 		unlock_journal(sb);
3690 		return 0;
3691 	}
3692 
3693 	/* deal with old transactions where we are the last writers */
3694 	now = get_seconds();
3695 	if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3696 		commit_now = 1;
3697 		journal->j_next_async_flush = 1;
3698 	}
3699 	/* don't batch when someone is waiting on j_join_wait */
3700 	/* don't batch when syncing the commit or flushing the whole trans */
3701 	if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock))
3702 	    && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3703 	    && journal->j_len_alloc < journal->j_max_batch
3704 	    && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3705 		journal->j_bcount++;
3706 		unlock_journal(sb);
3707 		return 0;
3708 	}
3709 
3710 	if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3711 		reiserfs_panic(sb, "journal-003",
3712 			       "j_start (%ld) is too high",
3713 			       journal->j_start);
3714 	}
3715 	return 1;
3716 }
3717 
3718 /*
3719  * Does all the work that makes deleting blocks safe.
3720  * when deleting a block mark BH_JNew, just remove it from the current
3721  * transaction, clean it's buffer_head and move on.
3722  *
3723  * otherwise:
3724  * set a bit for the block in the journal bitmap.  That will prevent it from
3725  * being allocated for unformatted nodes before this transaction has finished.
3726  *
3727  * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.
3728  * That will prevent any old transactions with this block from trying to flush
3729  * to the real location.  Since we aren't removing the cnode from the
3730  * journal_list_hash, *the block can't be reallocated yet.
3731  *
3732  * Then remove it from the current transaction, decrementing any counters and
3733  * filing it on the clean list.
3734  */
3735 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3736 		       struct super_block *sb, b_blocknr_t blocknr)
3737 {
3738 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3739 	struct reiserfs_journal_cnode *cn = NULL;
3740 	struct buffer_head *bh = NULL;
3741 	struct reiserfs_list_bitmap *jb = NULL;
3742 	int cleaned = 0;
3743 	BUG_ON(!th->t_trans_id);
3744 
3745 	cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3746 	if (cn && cn->bh) {
3747 		bh = cn->bh;
3748 		get_bh(bh);
3749 	}
3750 	/* if it is journal new, we just remove it from this transaction */
3751 	if (bh && buffer_journal_new(bh)) {
3752 		clear_buffer_journal_new(bh);
3753 		clear_prepared_bits(bh);
3754 		reiserfs_clean_and_file_buffer(bh);
3755 		cleaned = remove_from_transaction(sb, blocknr, cleaned);
3756 	} else {
3757 		/*
3758 		 * set the bit for this block in the journal bitmap
3759 		 * for this transaction
3760 		 */
3761 		jb = journal->j_current_jl->j_list_bitmap;
3762 		if (!jb) {
3763 			reiserfs_panic(sb, "journal-1702",
3764 				       "journal_list_bitmap is NULL");
3765 		}
3766 		set_bit_in_list_bitmap(sb, blocknr, jb);
3767 
3768 		/* Note, the entire while loop is not allowed to schedule.  */
3769 
3770 		if (bh) {
3771 			clear_prepared_bits(bh);
3772 			reiserfs_clean_and_file_buffer(bh);
3773 		}
3774 		cleaned = remove_from_transaction(sb, blocknr, cleaned);
3775 
3776 		/*
3777 		 * find all older transactions with this block,
3778 		 * make sure they don't try to write it out
3779 		 */
3780 		cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3781 					  blocknr);
3782 		while (cn) {
3783 			if (sb == cn->sb && blocknr == cn->blocknr) {
3784 				set_bit(BLOCK_FREED, &cn->state);
3785 				if (cn->bh) {
3786 					/*
3787 					 * remove_from_transaction will brelse
3788 					 * the buffer if it was in the current
3789 					 * trans
3790 					 */
3791 					if (!cleaned) {
3792 						clear_buffer_journal_dirty(cn->
3793 									   bh);
3794 						clear_buffer_dirty(cn->bh);
3795 						clear_buffer_journal_test(cn->
3796 									  bh);
3797 						cleaned = 1;
3798 						put_bh(cn->bh);
3799 						if (atomic_read
3800 						    (&cn->bh->b_count) < 0) {
3801 							reiserfs_warning(sb,
3802 								 "journal-2138",
3803 								 "cn->bh->b_count < 0");
3804 						}
3805 					}
3806 					/*
3807 					 * since we are clearing the bh,
3808 					 * we MUST dec nonzerolen
3809 					 */
3810 					if (cn->jlist) {
3811 						atomic_dec(&cn->jlist->
3812 							   j_nonzerolen);
3813 					}
3814 					cn->bh = NULL;
3815 				}
3816 			}
3817 			cn = cn->hnext;
3818 		}
3819 	}
3820 
3821 	if (bh)
3822 		release_buffer_page(bh); /* get_hash grabs the buffer */
3823 	return 0;
3824 }
3825 
3826 void reiserfs_update_inode_transaction(struct inode *inode)
3827 {
3828 	struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3829 	REISERFS_I(inode)->i_jl = journal->j_current_jl;
3830 	REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3831 }
3832 
3833 /*
3834  * returns -1 on error, 0 if no commits/barriers were done and 1
3835  * if a transaction was actually committed and the barrier was done
3836  */
3837 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3838 			     struct reiserfs_journal_list *jl)
3839 {
3840 	struct reiserfs_transaction_handle th;
3841 	struct super_block *sb = inode->i_sb;
3842 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3843 	int ret = 0;
3844 
3845 	/*
3846 	 * is it from the current transaction,
3847 	 * or from an unknown transaction?
3848 	 */
3849 	if (id == journal->j_trans_id) {
3850 		jl = journal->j_current_jl;
3851 		/*
3852 		 * try to let other writers come in and
3853 		 * grow this transaction
3854 		 */
3855 		let_transaction_grow(sb, id);
3856 		if (journal->j_trans_id != id) {
3857 			goto flush_commit_only;
3858 		}
3859 
3860 		ret = journal_begin(&th, sb, 1);
3861 		if (ret)
3862 			return ret;
3863 
3864 		/* someone might have ended this transaction while we joined */
3865 		if (journal->j_trans_id != id) {
3866 			reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3867 						     1);
3868 			journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3869 			ret = journal_end(&th);
3870 			goto flush_commit_only;
3871 		}
3872 
3873 		ret = journal_end_sync(&th);
3874 		if (!ret)
3875 			ret = 1;
3876 
3877 	} else {
3878 		/*
3879 		 * this gets tricky, we have to make sure the journal list in
3880 		 * the inode still exists.  We know the list is still around
3881 		 * if we've got a larger transaction id than the oldest list
3882 		 */
3883 flush_commit_only:
3884 		if (journal_list_still_alive(inode->i_sb, id)) {
3885 			/*
3886 			 * we only set ret to 1 when we know for sure
3887 			 * the barrier hasn't been started yet on the commit
3888 			 * block.
3889 			 */
3890 			if (atomic_read(&jl->j_commit_left) > 1)
3891 				ret = 1;
3892 			flush_commit_list(sb, jl, 1);
3893 			if (journal->j_errno)
3894 				ret = journal->j_errno;
3895 		}
3896 	}
3897 	/* otherwise the list is gone, and long since committed */
3898 	return ret;
3899 }
3900 
3901 int reiserfs_commit_for_inode(struct inode *inode)
3902 {
3903 	unsigned int id = REISERFS_I(inode)->i_trans_id;
3904 	struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3905 
3906 	/*
3907 	 * for the whole inode, assume unset id means it was
3908 	 * changed in the current transaction.  More conservative
3909 	 */
3910 	if (!id || !jl) {
3911 		reiserfs_update_inode_transaction(inode);
3912 		id = REISERFS_I(inode)->i_trans_id;
3913 		/* jl will be updated in __commit_trans_jl */
3914 	}
3915 
3916 	return __commit_trans_jl(inode, id, jl);
3917 }
3918 
3919 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3920 				      struct buffer_head *bh)
3921 {
3922 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3923 	PROC_INFO_INC(sb, journal.restore_prepared);
3924 	if (!bh) {
3925 		return;
3926 	}
3927 	if (test_clear_buffer_journal_restore_dirty(bh) &&
3928 	    buffer_journal_dirty(bh)) {
3929 		struct reiserfs_journal_cnode *cn;
3930 		reiserfs_write_lock(sb);
3931 		cn = get_journal_hash_dev(sb,
3932 					  journal->j_list_hash_table,
3933 					  bh->b_blocknr);
3934 		if (cn && can_dirty(cn)) {
3935 			set_buffer_journal_test(bh);
3936 			mark_buffer_dirty(bh);
3937 		}
3938 		reiserfs_write_unlock(sb);
3939 	}
3940 	clear_buffer_journal_prepared(bh);
3941 }
3942 
3943 extern struct tree_balance *cur_tb;
3944 /*
3945  * before we can change a metadata block, we have to make sure it won't
3946  * be written to disk while we are altering it.  So, we must:
3947  * clean it
3948  * wait on it.
3949  */
3950 int reiserfs_prepare_for_journal(struct super_block *sb,
3951 				 struct buffer_head *bh, int wait)
3952 {
3953 	PROC_INFO_INC(sb, journal.prepare);
3954 
3955 	if (!trylock_buffer(bh)) {
3956 		if (!wait)
3957 			return 0;
3958 		lock_buffer(bh);
3959 	}
3960 	set_buffer_journal_prepared(bh);
3961 	if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3962 		clear_buffer_journal_test(bh);
3963 		set_buffer_journal_restore_dirty(bh);
3964 	}
3965 	unlock_buffer(bh);
3966 	return 1;
3967 }
3968 
3969 /*
3970  * long and ugly.  If flush, will not return until all commit
3971  * blocks and all real buffers in the trans are on disk.
3972  * If no_async, won't return until all commit blocks are on disk.
3973  *
3974  * keep reading, there are comments as you go along
3975  *
3976  * If the journal is aborted, we just clean up. Things like flushing
3977  * journal lists, etc just won't happen.
3978  */
3979 static int do_journal_end(struct reiserfs_transaction_handle *th, int flags)
3980 {
3981 	struct super_block *sb = th->t_super;
3982 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3983 	struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3984 	struct reiserfs_journal_cnode *last_cn = NULL;
3985 	struct reiserfs_journal_desc *desc;
3986 	struct reiserfs_journal_commit *commit;
3987 	struct buffer_head *c_bh;	/* commit bh */
3988 	struct buffer_head *d_bh;	/* desc bh */
3989 	int cur_write_start = 0;	/* start index of current log write */
3990 	int old_start;
3991 	int i;
3992 	int flush;
3993 	int wait_on_commit;
3994 	struct reiserfs_journal_list *jl, *temp_jl;
3995 	struct list_head *entry, *safe;
3996 	unsigned long jindex;
3997 	unsigned int commit_trans_id;
3998 	int trans_half;
3999 	int depth;
4000 
4001 	BUG_ON(th->t_refcount > 1);
4002 	BUG_ON(!th->t_trans_id);
4003 	BUG_ON(!th->t_super);
4004 
4005 	/*
4006 	 * protect flush_older_commits from doing mistakes if the
4007 	 * transaction ID counter gets overflowed.
4008 	 */
4009 	if (th->t_trans_id == ~0U)
4010 		flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
4011 	flush = flags & FLUSH_ALL;
4012 	wait_on_commit = flags & WAIT;
4013 
4014 	current->journal_info = th->t_handle_save;
4015 	reiserfs_check_lock_depth(sb, "journal end");
4016 	if (journal->j_len == 0) {
4017 		reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
4018 					     1);
4019 		journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
4020 	}
4021 
4022 	lock_journal(sb);
4023 	if (journal->j_next_full_flush) {
4024 		flags |= FLUSH_ALL;
4025 		flush = 1;
4026 	}
4027 	if (journal->j_next_async_flush) {
4028 		flags |= COMMIT_NOW | WAIT;
4029 		wait_on_commit = 1;
4030 	}
4031 
4032 	/*
4033 	 * check_journal_end locks the journal, and unlocks if it does
4034 	 * not return 1 it tells us if we should continue with the
4035 	 * journal_end, or just return
4036 	 */
4037 	if (!check_journal_end(th, flags)) {
4038 		reiserfs_schedule_old_flush(sb);
4039 		wake_queued_writers(sb);
4040 		reiserfs_async_progress_wait(sb);
4041 		goto out;
4042 	}
4043 
4044 	/* check_journal_end might set these, check again */
4045 	if (journal->j_next_full_flush) {
4046 		flush = 1;
4047 	}
4048 
4049 	/*
4050 	 * j must wait means we have to flush the log blocks, and the
4051 	 * real blocks for this transaction
4052 	 */
4053 	if (journal->j_must_wait > 0) {
4054 		flush = 1;
4055 	}
4056 #ifdef REISERFS_PREALLOCATE
4057 	/*
4058 	 * quota ops might need to nest, setup the journal_info pointer
4059 	 * for them and raise the refcount so that it is > 0.
4060 	 */
4061 	current->journal_info = th;
4062 	th->t_refcount++;
4063 
4064 	/* it should not involve new blocks into the transaction */
4065 	reiserfs_discard_all_prealloc(th);
4066 
4067 	th->t_refcount--;
4068 	current->journal_info = th->t_handle_save;
4069 #endif
4070 
4071 	/* setup description block */
4072 	d_bh =
4073 	    journal_getblk(sb,
4074 			   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4075 			   journal->j_start);
4076 	set_buffer_uptodate(d_bh);
4077 	desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4078 	memset(d_bh->b_data, 0, d_bh->b_size);
4079 	memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4080 	set_desc_trans_id(desc, journal->j_trans_id);
4081 
4082 	/*
4083 	 * setup commit block.  Don't write (keep it clean too) this one
4084 	 * until after everyone else is written
4085 	 */
4086 	c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4087 			      ((journal->j_start + journal->j_len +
4088 				1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4089 	commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4090 	memset(c_bh->b_data, 0, c_bh->b_size);
4091 	set_commit_trans_id(commit, journal->j_trans_id);
4092 	set_buffer_uptodate(c_bh);
4093 
4094 	/* init this journal list */
4095 	jl = journal->j_current_jl;
4096 
4097 	/*
4098 	 * we lock the commit before doing anything because
4099 	 * we want to make sure nobody tries to run flush_commit_list until
4100 	 * the new transaction is fully setup, and we've already flushed the
4101 	 * ordered bh list
4102 	 */
4103 	reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4104 
4105 	/* save the transaction id in case we need to commit it later */
4106 	commit_trans_id = jl->j_trans_id;
4107 
4108 	atomic_set(&jl->j_older_commits_done, 0);
4109 	jl->j_trans_id = journal->j_trans_id;
4110 	jl->j_timestamp = journal->j_trans_start_time;
4111 	jl->j_commit_bh = c_bh;
4112 	jl->j_start = journal->j_start;
4113 	jl->j_len = journal->j_len;
4114 	atomic_set(&jl->j_nonzerolen, journal->j_len);
4115 	atomic_set(&jl->j_commit_left, journal->j_len + 2);
4116 	jl->j_realblock = NULL;
4117 
4118 	/*
4119 	 * The ENTIRE FOR LOOP MUST not cause schedule to occur.
4120 	 * for each real block, add it to the journal list hash,
4121 	 * copy into real block index array in the commit or desc block
4122 	 */
4123 	trans_half = journal_trans_half(sb->s_blocksize);
4124 	for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4125 		if (buffer_journaled(cn->bh)) {
4126 			jl_cn = get_cnode(sb);
4127 			if (!jl_cn) {
4128 				reiserfs_panic(sb, "journal-1676",
4129 					       "get_cnode returned NULL");
4130 			}
4131 			if (i == 0) {
4132 				jl->j_realblock = jl_cn;
4133 			}
4134 			jl_cn->prev = last_cn;
4135 			jl_cn->next = NULL;
4136 			if (last_cn) {
4137 				last_cn->next = jl_cn;
4138 			}
4139 			last_cn = jl_cn;
4140 			/*
4141 			 * make sure the block we are trying to log
4142 			 * is not a block of journal or reserved area
4143 			 */
4144 			if (is_block_in_log_or_reserved_area
4145 			    (sb, cn->bh->b_blocknr)) {
4146 				reiserfs_panic(sb, "journal-2332",
4147 					       "Trying to log block %lu, "
4148 					       "which is a log block",
4149 					       cn->bh->b_blocknr);
4150 			}
4151 			jl_cn->blocknr = cn->bh->b_blocknr;
4152 			jl_cn->state = 0;
4153 			jl_cn->sb = sb;
4154 			jl_cn->bh = cn->bh;
4155 			jl_cn->jlist = jl;
4156 			insert_journal_hash(journal->j_list_hash_table, jl_cn);
4157 			if (i < trans_half) {
4158 				desc->j_realblock[i] =
4159 				    cpu_to_le32(cn->bh->b_blocknr);
4160 			} else {
4161 				commit->j_realblock[i - trans_half] =
4162 				    cpu_to_le32(cn->bh->b_blocknr);
4163 			}
4164 		} else {
4165 			i--;
4166 		}
4167 	}
4168 	set_desc_trans_len(desc, journal->j_len);
4169 	set_desc_mount_id(desc, journal->j_mount_id);
4170 	set_desc_trans_id(desc, journal->j_trans_id);
4171 	set_commit_trans_len(commit, journal->j_len);
4172 
4173 	/*
4174 	 * special check in case all buffers in the journal
4175 	 * were marked for not logging
4176 	 */
4177 	BUG_ON(journal->j_len == 0);
4178 
4179 	/*
4180 	 * we're about to dirty all the log blocks, mark the description block
4181 	 * dirty now too.  Don't mark the commit block dirty until all the
4182 	 * others are on disk
4183 	 */
4184 	mark_buffer_dirty(d_bh);
4185 
4186 	/*
4187 	 * first data block is j_start + 1, so add one to
4188 	 * cur_write_start wherever you use it
4189 	 */
4190 	cur_write_start = journal->j_start;
4191 	cn = journal->j_first;
4192 	jindex = 1;	/* start at one so we don't get the desc again */
4193 	while (cn) {
4194 		clear_buffer_journal_new(cn->bh);
4195 		/* copy all the real blocks into log area.  dirty log blocks */
4196 		if (buffer_journaled(cn->bh)) {
4197 			struct buffer_head *tmp_bh;
4198 			char *addr;
4199 			struct page *page;
4200 			tmp_bh =
4201 			    journal_getblk(sb,
4202 					   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4203 					   ((cur_write_start +
4204 					     jindex) %
4205 					    SB_ONDISK_JOURNAL_SIZE(sb)));
4206 			set_buffer_uptodate(tmp_bh);
4207 			page = cn->bh->b_page;
4208 			addr = kmap(page);
4209 			memcpy(tmp_bh->b_data,
4210 			       addr + offset_in_page(cn->bh->b_data),
4211 			       cn->bh->b_size);
4212 			kunmap(page);
4213 			mark_buffer_dirty(tmp_bh);
4214 			jindex++;
4215 			set_buffer_journal_dirty(cn->bh);
4216 			clear_buffer_journaled(cn->bh);
4217 		} else {
4218 			/*
4219 			 * JDirty cleared sometime during transaction.
4220 			 * don't log this one
4221 			 */
4222 			reiserfs_warning(sb, "journal-2048",
4223 					 "BAD, buffer in journal hash, "
4224 					 "but not JDirty!");
4225 			brelse(cn->bh);
4226 		}
4227 		next = cn->next;
4228 		free_cnode(sb, cn);
4229 		cn = next;
4230 		reiserfs_cond_resched(sb);
4231 	}
4232 
4233 	/*
4234 	 * we are done with both the c_bh and d_bh, but
4235 	 * c_bh must be written after all other commit blocks,
4236 	 * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4237 	 */
4238 
4239 	journal->j_current_jl = alloc_journal_list(sb);
4240 
4241 	/* now it is safe to insert this transaction on the main list */
4242 	list_add_tail(&jl->j_list, &journal->j_journal_list);
4243 	list_add_tail(&jl->j_working_list, &journal->j_working_list);
4244 	journal->j_num_work_lists++;
4245 
4246 	/* reset journal values for the next transaction */
4247 	old_start = journal->j_start;
4248 	journal->j_start =
4249 	    (journal->j_start + journal->j_len +
4250 	     2) % SB_ONDISK_JOURNAL_SIZE(sb);
4251 	atomic_set(&journal->j_wcount, 0);
4252 	journal->j_bcount = 0;
4253 	journal->j_last = NULL;
4254 	journal->j_first = NULL;
4255 	journal->j_len = 0;
4256 	journal->j_trans_start_time = 0;
4257 	/* check for trans_id overflow */
4258 	if (++journal->j_trans_id == 0)
4259 		journal->j_trans_id = 10;
4260 	journal->j_current_jl->j_trans_id = journal->j_trans_id;
4261 	journal->j_must_wait = 0;
4262 	journal->j_len_alloc = 0;
4263 	journal->j_next_full_flush = 0;
4264 	journal->j_next_async_flush = 0;
4265 	init_journal_hash(sb);
4266 
4267 	/*
4268 	 * make sure reiserfs_add_jh sees the new current_jl before we
4269 	 * write out the tails
4270 	 */
4271 	smp_mb();
4272 
4273 	/*
4274 	 * tail conversion targets have to hit the disk before we end the
4275 	 * transaction.  Otherwise a later transaction might repack the tail
4276 	 * before this transaction commits, leaving the data block unflushed
4277 	 * and clean, if we crash before the later transaction commits, the
4278 	 * data block is lost.
4279 	 */
4280 	if (!list_empty(&jl->j_tail_bh_list)) {
4281 		depth = reiserfs_write_unlock_nested(sb);
4282 		write_ordered_buffers(&journal->j_dirty_buffers_lock,
4283 				      journal, jl, &jl->j_tail_bh_list);
4284 		reiserfs_write_lock_nested(sb, depth);
4285 	}
4286 	BUG_ON(!list_empty(&jl->j_tail_bh_list));
4287 	mutex_unlock(&jl->j_commit_mutex);
4288 
4289 	/*
4290 	 * honor the flush wishes from the caller, simple commits can
4291 	 * be done outside the journal lock, they are done below
4292 	 *
4293 	 * if we don't flush the commit list right now, we put it into
4294 	 * the work queue so the people waiting on the async progress work
4295 	 * queue don't wait for this proc to flush journal lists and such.
4296 	 */
4297 	if (flush) {
4298 		flush_commit_list(sb, jl, 1);
4299 		flush_journal_list(sb, jl, 1);
4300 	} else if (!(jl->j_state & LIST_COMMIT_PENDING)) {
4301 		/*
4302 		 * Avoid queueing work when sb is being shut down. Transaction
4303 		 * will be flushed on journal shutdown.
4304 		 */
4305 		if (sb->s_flags & SB_ACTIVE)
4306 			queue_delayed_work(REISERFS_SB(sb)->commit_wq,
4307 					   &journal->j_work, HZ / 10);
4308 	}
4309 
4310 	/*
4311 	 * if the next transaction has any chance of wrapping, flush
4312 	 * transactions that might get overwritten.  If any journal lists
4313 	 * are very old flush them as well.
4314 	 */
4315 first_jl:
4316 	list_for_each_safe(entry, safe, &journal->j_journal_list) {
4317 		temp_jl = JOURNAL_LIST_ENTRY(entry);
4318 		if (journal->j_start <= temp_jl->j_start) {
4319 			if ((journal->j_start + journal->j_trans_max + 1) >=
4320 			    temp_jl->j_start) {
4321 				flush_used_journal_lists(sb, temp_jl);
4322 				goto first_jl;
4323 			} else if ((journal->j_start +
4324 				    journal->j_trans_max + 1) <
4325 				   SB_ONDISK_JOURNAL_SIZE(sb)) {
4326 				/*
4327 				 * if we don't cross into the next
4328 				 * transaction and we don't wrap, there is
4329 				 * no way we can overlap any later transactions
4330 				 * break now
4331 				 */
4332 				break;
4333 			}
4334 		} else if ((journal->j_start +
4335 			    journal->j_trans_max + 1) >
4336 			   SB_ONDISK_JOURNAL_SIZE(sb)) {
4337 			if (((journal->j_start + journal->j_trans_max + 1) %
4338 			     SB_ONDISK_JOURNAL_SIZE(sb)) >=
4339 			    temp_jl->j_start) {
4340 				flush_used_journal_lists(sb, temp_jl);
4341 				goto first_jl;
4342 			} else {
4343 				/*
4344 				* we don't overlap anything from out start
4345 				* to the end of the log, and our wrapped
4346 				* portion doesn't overlap anything at
4347 				* the start of the log.  We can break
4348 				*/
4349 				break;
4350 			}
4351 		}
4352 	}
4353 
4354 	journal->j_current_jl->j_list_bitmap =
4355 	    get_list_bitmap(sb, journal->j_current_jl);
4356 
4357 	if (!(journal->j_current_jl->j_list_bitmap)) {
4358 		reiserfs_panic(sb, "journal-1996",
4359 			       "could not get a list bitmap");
4360 	}
4361 
4362 	atomic_set(&journal->j_jlock, 0);
4363 	unlock_journal(sb);
4364 	/* wake up any body waiting to join. */
4365 	clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4366 	wake_up(&journal->j_join_wait);
4367 
4368 	if (!flush && wait_on_commit &&
4369 	    journal_list_still_alive(sb, commit_trans_id)) {
4370 		flush_commit_list(sb, jl, 1);
4371 	}
4372 out:
4373 	reiserfs_check_lock_depth(sb, "journal end2");
4374 
4375 	memset(th, 0, sizeof(*th));
4376 	/*
4377 	 * Re-set th->t_super, so we can properly keep track of how many
4378 	 * persistent transactions there are. We need to do this so if this
4379 	 * call is part of a failed restart_transaction, we can free it later
4380 	 */
4381 	th->t_super = sb;
4382 
4383 	return journal->j_errno;
4384 }
4385 
4386 /* Send the file system read only and refuse new transactions */
4387 void reiserfs_abort_journal(struct super_block *sb, int errno)
4388 {
4389 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
4390 	if (test_bit(J_ABORTED, &journal->j_state))
4391 		return;
4392 
4393 	if (!journal->j_errno)
4394 		journal->j_errno = errno;
4395 
4396 	sb->s_flags |= SB_RDONLY;
4397 	set_bit(J_ABORTED, &journal->j_state);
4398 
4399 #ifdef CONFIG_REISERFS_CHECK
4400 	dump_stack();
4401 #endif
4402 }
4403