xref: /linux/fs/reiserfs/journal.c (revision a82ebb093fc7bdd88f8df17b2fa303d7535fa43b)
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  *		        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 
60 /* must be correct to keep the desc and commit structs at 4k */
61 #define JOURNAL_TRANS_HALF 1018
62 #define BUFNR 64		/*read ahead */
63 
64 /* cnode stat bits.  Move these into reiserfs_fs.h */
65 
66 /* this block was freed, and can't be written.  */
67 #define BLOCK_FREED 2
68 /* this block was freed during this transaction, and can't be written */
69 #define BLOCK_FREED_HOLDER 3
70 
71 /* used in flush_journal_list */
72 #define BLOCK_NEEDS_FLUSH 4
73 #define BLOCK_DIRTIED 5
74 
75 /* journal list state bits */
76 #define LIST_TOUCHED 1
77 #define LIST_DIRTY   2
78 #define LIST_COMMIT_PENDING  4	/* someone will commit this list */
79 
80 /* flags for do_journal_end */
81 #define FLUSH_ALL   1		/* flush commit and real blocks */
82 #define COMMIT_NOW  2		/* end and commit this transaction */
83 #define WAIT        4		/* wait for the log blocks to hit the disk */
84 
85 static int do_journal_end(struct reiserfs_transaction_handle *, int flags);
86 static int flush_journal_list(struct super_block *s,
87 			      struct reiserfs_journal_list *jl, int flushall);
88 static int flush_commit_list(struct super_block *s,
89 			     struct reiserfs_journal_list *jl, int flushall);
90 static int can_dirty(struct reiserfs_journal_cnode *cn);
91 static int journal_join(struct reiserfs_transaction_handle *th,
92 			struct super_block *sb);
93 static void release_journal_dev(struct super_block *super,
94 			       struct reiserfs_journal *journal);
95 static void dirty_one_transaction(struct super_block *s,
96 				 struct reiserfs_journal_list *jl);
97 static void flush_async_commits(struct work_struct *work);
98 static void queue_log_writer(struct super_block *s);
99 
100 /* values for join in do_journal_begin_r */
101 enum {
102 	JBEGIN_REG = 0,		/* regular journal begin */
103 	/* join the running transaction if at all possible */
104 	JBEGIN_JOIN = 1,
105 	/* called from cleanup code, ignores aborted flag */
106 	JBEGIN_ABORT = 2,
107 };
108 
109 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
110 			      struct super_block *sb,
111 			      unsigned long nblocks, int join);
112 
113 static void init_journal_hash(struct super_block *sb)
114 {
115 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
116 	memset(journal->j_hash_table, 0,
117 	       JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
118 }
119 
120 /*
121  * clears BH_Dirty and sticks the buffer on the clean list.  Called because
122  * I can't allow refile_buffer to make schedule happen after I've freed a
123  * block.  Look at remove_from_transaction and journal_mark_freed for
124  * more details.
125  */
126 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
127 {
128 	if (bh) {
129 		clear_buffer_dirty(bh);
130 		clear_buffer_journal_test(bh);
131 	}
132 	return 0;
133 }
134 
135 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
136 							 *sb)
137 {
138 	struct reiserfs_bitmap_node *bn;
139 	static int id;
140 
141 	bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
142 	if (!bn) {
143 		return NULL;
144 	}
145 	bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
146 	if (!bn->data) {
147 		kfree(bn);
148 		return NULL;
149 	}
150 	bn->id = id++;
151 	INIT_LIST_HEAD(&bn->list);
152 	return bn;
153 }
154 
155 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
156 {
157 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
158 	struct reiserfs_bitmap_node *bn = NULL;
159 	struct list_head *entry = journal->j_bitmap_nodes.next;
160 
161 	journal->j_used_bitmap_nodes++;
162 repeat:
163 
164 	if (entry != &journal->j_bitmap_nodes) {
165 		bn = list_entry(entry, struct reiserfs_bitmap_node, list);
166 		list_del(entry);
167 		memset(bn->data, 0, sb->s_blocksize);
168 		journal->j_free_bitmap_nodes--;
169 		return bn;
170 	}
171 	bn = allocate_bitmap_node(sb);
172 	if (!bn) {
173 		yield();
174 		goto repeat;
175 	}
176 	return bn;
177 }
178 static inline void free_bitmap_node(struct super_block *sb,
179 				    struct reiserfs_bitmap_node *bn)
180 {
181 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
182 	journal->j_used_bitmap_nodes--;
183 	if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
184 		kfree(bn->data);
185 		kfree(bn);
186 	} else {
187 		list_add(&bn->list, &journal->j_bitmap_nodes);
188 		journal->j_free_bitmap_nodes++;
189 	}
190 }
191 
192 static void allocate_bitmap_nodes(struct super_block *sb)
193 {
194 	int i;
195 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
196 	struct reiserfs_bitmap_node *bn = NULL;
197 	for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
198 		bn = allocate_bitmap_node(sb);
199 		if (bn) {
200 			list_add(&bn->list, &journal->j_bitmap_nodes);
201 			journal->j_free_bitmap_nodes++;
202 		} else {
203 			/* this is ok, we'll try again when more are needed */
204 			break;
205 		}
206 	}
207 }
208 
209 static int set_bit_in_list_bitmap(struct super_block *sb,
210 				  b_blocknr_t block,
211 				  struct reiserfs_list_bitmap *jb)
212 {
213 	unsigned int bmap_nr = block / (sb->s_blocksize << 3);
214 	unsigned int bit_nr = block % (sb->s_blocksize << 3);
215 
216 	if (!jb->bitmaps[bmap_nr]) {
217 		jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
218 	}
219 	set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
220 	return 0;
221 }
222 
223 static void cleanup_bitmap_list(struct super_block *sb,
224 				struct reiserfs_list_bitmap *jb)
225 {
226 	int i;
227 	if (jb->bitmaps == NULL)
228 		return;
229 
230 	for (i = 0; i < reiserfs_bmap_count(sb); i++) {
231 		if (jb->bitmaps[i]) {
232 			free_bitmap_node(sb, jb->bitmaps[i]);
233 			jb->bitmaps[i] = NULL;
234 		}
235 	}
236 }
237 
238 /*
239  * only call this on FS unmount.
240  */
241 static int free_list_bitmaps(struct super_block *sb,
242 			     struct reiserfs_list_bitmap *jb_array)
243 {
244 	int i;
245 	struct reiserfs_list_bitmap *jb;
246 	for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
247 		jb = jb_array + i;
248 		jb->journal_list = NULL;
249 		cleanup_bitmap_list(sb, jb);
250 		vfree(jb->bitmaps);
251 		jb->bitmaps = NULL;
252 	}
253 	return 0;
254 }
255 
256 static int free_bitmap_nodes(struct super_block *sb)
257 {
258 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
259 	struct list_head *next = journal->j_bitmap_nodes.next;
260 	struct reiserfs_bitmap_node *bn;
261 
262 	while (next != &journal->j_bitmap_nodes) {
263 		bn = list_entry(next, struct reiserfs_bitmap_node, list);
264 		list_del(next);
265 		kfree(bn->data);
266 		kfree(bn);
267 		next = journal->j_bitmap_nodes.next;
268 		journal->j_free_bitmap_nodes--;
269 	}
270 
271 	return 0;
272 }
273 
274 /*
275  * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
276  * jb_array is the array to be filled in.
277  */
278 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
279 				   struct reiserfs_list_bitmap *jb_array,
280 				   unsigned int bmap_nr)
281 {
282 	int i;
283 	int failed = 0;
284 	struct reiserfs_list_bitmap *jb;
285 	int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
286 
287 	for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
288 		jb = jb_array + i;
289 		jb->journal_list = NULL;
290 		jb->bitmaps = vzalloc(mem);
291 		if (!jb->bitmaps) {
292 			reiserfs_warning(sb, "clm-2000", "unable to "
293 					 "allocate bitmaps for journal lists");
294 			failed = 1;
295 			break;
296 		}
297 	}
298 	if (failed) {
299 		free_list_bitmaps(sb, jb_array);
300 		return -1;
301 	}
302 	return 0;
303 }
304 
305 /*
306  * find an available list bitmap.  If you can't find one, flush a commit list
307  * and try again
308  */
309 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
310 						    struct reiserfs_journal_list
311 						    *jl)
312 {
313 	int i, j;
314 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
315 	struct reiserfs_list_bitmap *jb = NULL;
316 
317 	for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
318 		i = journal->j_list_bitmap_index;
319 		journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
320 		jb = journal->j_list_bitmap + i;
321 		if (journal->j_list_bitmap[i].journal_list) {
322 			flush_commit_list(sb,
323 					  journal->j_list_bitmap[i].
324 					  journal_list, 1);
325 			if (!journal->j_list_bitmap[i].journal_list) {
326 				break;
327 			}
328 		} else {
329 			break;
330 		}
331 	}
332 	/* double check to make sure if flushed correctly */
333 	if (jb->journal_list)
334 		return NULL;
335 	jb->journal_list = jl;
336 	return jb;
337 }
338 
339 /*
340  * allocates a new chunk of X nodes, and links them all together as a list.
341  * Uses the cnode->next and cnode->prev pointers
342  * returns NULL on failure
343  */
344 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
345 {
346 	struct reiserfs_journal_cnode *head;
347 	int i;
348 	if (num_cnodes <= 0) {
349 		return NULL;
350 	}
351 	head = vzalloc(array_size(num_cnodes,
352 				  sizeof(struct reiserfs_journal_cnode)));
353 	if (!head) {
354 		return NULL;
355 	}
356 	head[0].prev = NULL;
357 	head[0].next = head + 1;
358 	for (i = 1; i < num_cnodes; i++) {
359 		head[i].prev = head + (i - 1);
360 		head[i].next = head + (i + 1);	/* if last one, overwrite it after the if */
361 	}
362 	head[num_cnodes - 1].next = NULL;
363 	return head;
364 }
365 
366 /* pulls a cnode off the free list, or returns NULL on failure */
367 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
368 {
369 	struct reiserfs_journal_cnode *cn;
370 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
371 
372 	reiserfs_check_lock_depth(sb, "get_cnode");
373 
374 	if (journal->j_cnode_free <= 0) {
375 		return NULL;
376 	}
377 	journal->j_cnode_used++;
378 	journal->j_cnode_free--;
379 	cn = journal->j_cnode_free_list;
380 	if (!cn) {
381 		return cn;
382 	}
383 	if (cn->next) {
384 		cn->next->prev = NULL;
385 	}
386 	journal->j_cnode_free_list = cn->next;
387 	memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
388 	return cn;
389 }
390 
391 /*
392  * returns a cnode to the free list
393  */
394 static void free_cnode(struct super_block *sb,
395 		       struct reiserfs_journal_cnode *cn)
396 {
397 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
398 
399 	reiserfs_check_lock_depth(sb, "free_cnode");
400 
401 	journal->j_cnode_used--;
402 	journal->j_cnode_free++;
403 	/* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
404 	cn->next = journal->j_cnode_free_list;
405 	if (journal->j_cnode_free_list) {
406 		journal->j_cnode_free_list->prev = cn;
407 	}
408 	cn->prev = NULL;	/* not needed with the memset, but I might kill the memset, and forget to do this */
409 	journal->j_cnode_free_list = cn;
410 }
411 
412 static void clear_prepared_bits(struct buffer_head *bh)
413 {
414 	clear_buffer_journal_prepared(bh);
415 	clear_buffer_journal_restore_dirty(bh);
416 }
417 
418 /*
419  * return a cnode with same dev, block number and size in table,
420  * or null if not found
421  */
422 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
423 								  super_block
424 								  *sb,
425 								  struct
426 								  reiserfs_journal_cnode
427 								  **table,
428 								  long bl)
429 {
430 	struct reiserfs_journal_cnode *cn;
431 	cn = journal_hash(table, sb, bl);
432 	while (cn) {
433 		if (cn->blocknr == bl && cn->sb == sb)
434 			return cn;
435 		cn = cn->hnext;
436 	}
437 	return (struct reiserfs_journal_cnode *)0;
438 }
439 
440 /*
441  * this actually means 'can this block be reallocated yet?'.  If you set
442  * search_all, a block can only be allocated if it is not in the current
443  * transaction, was not freed by the current transaction, and has no chance
444  * of ever being overwritten by a replay after crashing.
445  *
446  * If you don't set search_all, a block can only be allocated if it is not
447  * in the current transaction.  Since deleting a block removes it from the
448  * current transaction, this case should never happen.  If you don't set
449  * search_all, make sure you never write the block without logging it.
450  *
451  * next_zero_bit is a suggestion about the next block to try for find_forward.
452  * when bl is rejected because it is set in a journal list bitmap, we search
453  * for the next zero bit in the bitmap that rejected bl.  Then, we return
454  * that through next_zero_bit for find_forward to try.
455  *
456  * Just because we return something in next_zero_bit does not mean we won't
457  * reject it on the next call to reiserfs_in_journal
458  */
459 int reiserfs_in_journal(struct super_block *sb,
460 			unsigned int bmap_nr, int bit_nr, int search_all,
461 			b_blocknr_t * next_zero_bit)
462 {
463 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
464 	struct reiserfs_list_bitmap *jb;
465 	int i;
466 	unsigned long bl;
467 
468 	*next_zero_bit = 0;	/* always start this at zero. */
469 
470 	PROC_INFO_INC(sb, journal.in_journal);
471 	/*
472 	 * If we aren't doing a search_all, this is a metablock, and it
473 	 * will be logged before use.  if we crash before the transaction
474 	 * that freed it commits,  this transaction won't have committed
475 	 * either, and the block will never be written
476 	 */
477 	if (search_all) {
478 		for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
479 			PROC_INFO_INC(sb, journal.in_journal_bitmap);
480 			jb = journal->j_list_bitmap + i;
481 			if (jb->journal_list && jb->bitmaps[bmap_nr] &&
482 			    test_bit(bit_nr,
483 				     (unsigned long *)jb->bitmaps[bmap_nr]->
484 				     data)) {
485 				*next_zero_bit =
486 				    find_next_zero_bit((unsigned long *)
487 						       (jb->bitmaps[bmap_nr]->
488 							data),
489 						       sb->s_blocksize << 3,
490 						       bit_nr + 1);
491 				return 1;
492 			}
493 		}
494 	}
495 
496 	bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
497 	/* is it in any old transactions? */
498 	if (search_all
499 	    && (get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
500 		return 1;
501 	}
502 
503 	/* is it in the current transaction.  This should never happen */
504 	if ((get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
505 		BUG();
506 		return 1;
507 	}
508 
509 	PROC_INFO_INC(sb, journal.in_journal_reusable);
510 	/* safe for reuse */
511 	return 0;
512 }
513 
514 /* insert cn into table */
515 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
516 				       struct reiserfs_journal_cnode *cn)
517 {
518 	struct reiserfs_journal_cnode *cn_orig;
519 
520 	cn_orig = journal_hash(table, cn->sb, cn->blocknr);
521 	cn->hnext = cn_orig;
522 	cn->hprev = NULL;
523 	if (cn_orig) {
524 		cn_orig->hprev = cn;
525 	}
526 	journal_hash(table, cn->sb, cn->blocknr) = cn;
527 }
528 
529 /* lock the current transaction */
530 static inline void lock_journal(struct super_block *sb)
531 {
532 	PROC_INFO_INC(sb, journal.lock_journal);
533 
534 	reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
535 }
536 
537 /* unlock the current transaction */
538 static inline void unlock_journal(struct super_block *sb)
539 {
540 	mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
541 }
542 
543 static inline void get_journal_list(struct reiserfs_journal_list *jl)
544 {
545 	jl->j_refcount++;
546 }
547 
548 static inline void put_journal_list(struct super_block *s,
549 				    struct reiserfs_journal_list *jl)
550 {
551 	if (jl->j_refcount < 1) {
552 		reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
553 			       jl->j_trans_id, jl->j_refcount);
554 	}
555 	if (--jl->j_refcount == 0)
556 		kfree(jl);
557 }
558 
559 /*
560  * this used to be much more involved, and I'm keeping it just in case
561  * things get ugly again.  it gets called by flush_commit_list, and
562  * cleans up any data stored about blocks freed during a transaction.
563  */
564 static void cleanup_freed_for_journal_list(struct super_block *sb,
565 					   struct reiserfs_journal_list *jl)
566 {
567 
568 	struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
569 	if (jb) {
570 		cleanup_bitmap_list(sb, jb);
571 	}
572 	jl->j_list_bitmap->journal_list = NULL;
573 	jl->j_list_bitmap = NULL;
574 }
575 
576 static int journal_list_still_alive(struct super_block *s,
577 				    unsigned int trans_id)
578 {
579 	struct reiserfs_journal *journal = SB_JOURNAL(s);
580 	struct list_head *entry = &journal->j_journal_list;
581 	struct reiserfs_journal_list *jl;
582 
583 	if (!list_empty(entry)) {
584 		jl = JOURNAL_LIST_ENTRY(entry->next);
585 		if (jl->j_trans_id <= trans_id) {
586 			return 1;
587 		}
588 	}
589 	return 0;
590 }
591 
592 /*
593  * If page->mapping was null, we failed to truncate this page for
594  * some reason.  Most likely because it was truncated after being
595  * logged via data=journal.
596  *
597  * This does a check to see if the buffer belongs to one of these
598  * lost pages before doing the final put_bh.  If page->mapping was
599  * null, it tries to free buffers on the page, which should make the
600  * final put_page drop the page from the lru.
601  */
602 static void release_buffer_page(struct buffer_head *bh)
603 {
604 	struct page *page = bh->b_page;
605 	if (!page->mapping && trylock_page(page)) {
606 		get_page(page);
607 		put_bh(bh);
608 		if (!page->mapping)
609 			try_to_free_buffers(page);
610 		unlock_page(page);
611 		put_page(page);
612 	} else {
613 		put_bh(bh);
614 	}
615 }
616 
617 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
618 {
619 	if (buffer_journaled(bh)) {
620 		reiserfs_warning(NULL, "clm-2084",
621 				 "pinned buffer %lu:%pg sent to disk",
622 				 bh->b_blocknr, bh->b_bdev);
623 	}
624 	if (uptodate)
625 		set_buffer_uptodate(bh);
626 	else
627 		clear_buffer_uptodate(bh);
628 
629 	unlock_buffer(bh);
630 	release_buffer_page(bh);
631 }
632 
633 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
634 {
635 	if (uptodate)
636 		set_buffer_uptodate(bh);
637 	else
638 		clear_buffer_uptodate(bh);
639 	unlock_buffer(bh);
640 	put_bh(bh);
641 }
642 
643 static void submit_logged_buffer(struct buffer_head *bh)
644 {
645 	get_bh(bh);
646 	bh->b_end_io = reiserfs_end_buffer_io_sync;
647 	clear_buffer_journal_new(bh);
648 	clear_buffer_dirty(bh);
649 	if (!test_clear_buffer_journal_test(bh))
650 		BUG();
651 	if (!buffer_uptodate(bh))
652 		BUG();
653 	submit_bh(REQ_OP_WRITE, 0, bh);
654 }
655 
656 static void submit_ordered_buffer(struct buffer_head *bh)
657 {
658 	get_bh(bh);
659 	bh->b_end_io = reiserfs_end_ordered_io;
660 	clear_buffer_dirty(bh);
661 	if (!buffer_uptodate(bh))
662 		BUG();
663 	submit_bh(REQ_OP_WRITE, 0, bh);
664 }
665 
666 #define CHUNK_SIZE 32
667 struct buffer_chunk {
668 	struct buffer_head *bh[CHUNK_SIZE];
669 	int nr;
670 };
671 
672 static void write_chunk(struct buffer_chunk *chunk)
673 {
674 	int i;
675 	for (i = 0; i < chunk->nr; i++) {
676 		submit_logged_buffer(chunk->bh[i]);
677 	}
678 	chunk->nr = 0;
679 }
680 
681 static void write_ordered_chunk(struct buffer_chunk *chunk)
682 {
683 	int i;
684 	for (i = 0; i < chunk->nr; i++) {
685 		submit_ordered_buffer(chunk->bh[i]);
686 	}
687 	chunk->nr = 0;
688 }
689 
690 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
691 			spinlock_t * lock, void (fn) (struct buffer_chunk *))
692 {
693 	int ret = 0;
694 	BUG_ON(chunk->nr >= CHUNK_SIZE);
695 	chunk->bh[chunk->nr++] = bh;
696 	if (chunk->nr >= CHUNK_SIZE) {
697 		ret = 1;
698 		if (lock) {
699 			spin_unlock(lock);
700 			fn(chunk);
701 			spin_lock(lock);
702 		} else {
703 			fn(chunk);
704 		}
705 	}
706 	return ret;
707 }
708 
709 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
710 static struct reiserfs_jh *alloc_jh(void)
711 {
712 	struct reiserfs_jh *jh;
713 	while (1) {
714 		jh = kmalloc(sizeof(*jh), GFP_NOFS);
715 		if (jh) {
716 			atomic_inc(&nr_reiserfs_jh);
717 			return jh;
718 		}
719 		yield();
720 	}
721 }
722 
723 /*
724  * we want to free the jh when the buffer has been written
725  * and waited on
726  */
727 void reiserfs_free_jh(struct buffer_head *bh)
728 {
729 	struct reiserfs_jh *jh;
730 
731 	jh = bh->b_private;
732 	if (jh) {
733 		bh->b_private = NULL;
734 		jh->bh = NULL;
735 		list_del_init(&jh->list);
736 		kfree(jh);
737 		if (atomic_read(&nr_reiserfs_jh) <= 0)
738 			BUG();
739 		atomic_dec(&nr_reiserfs_jh);
740 		put_bh(bh);
741 	}
742 }
743 
744 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
745 			   int tail)
746 {
747 	struct reiserfs_jh *jh;
748 
749 	if (bh->b_private) {
750 		spin_lock(&j->j_dirty_buffers_lock);
751 		if (!bh->b_private) {
752 			spin_unlock(&j->j_dirty_buffers_lock);
753 			goto no_jh;
754 		}
755 		jh = bh->b_private;
756 		list_del_init(&jh->list);
757 	} else {
758 no_jh:
759 		get_bh(bh);
760 		jh = alloc_jh();
761 		spin_lock(&j->j_dirty_buffers_lock);
762 		/*
763 		 * buffer must be locked for __add_jh, should be able to have
764 		 * two adds at the same time
765 		 */
766 		BUG_ON(bh->b_private);
767 		jh->bh = bh;
768 		bh->b_private = jh;
769 	}
770 	jh->jl = j->j_current_jl;
771 	if (tail)
772 		list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
773 	else {
774 		list_add_tail(&jh->list, &jh->jl->j_bh_list);
775 	}
776 	spin_unlock(&j->j_dirty_buffers_lock);
777 	return 0;
778 }
779 
780 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
781 {
782 	return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
783 }
784 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
785 {
786 	return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
787 }
788 
789 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
790 static int write_ordered_buffers(spinlock_t * lock,
791 				 struct reiserfs_journal *j,
792 				 struct reiserfs_journal_list *jl,
793 				 struct list_head *list)
794 {
795 	struct buffer_head *bh;
796 	struct reiserfs_jh *jh;
797 	int ret = j->j_errno;
798 	struct buffer_chunk chunk;
799 	struct list_head tmp;
800 	INIT_LIST_HEAD(&tmp);
801 
802 	chunk.nr = 0;
803 	spin_lock(lock);
804 	while (!list_empty(list)) {
805 		jh = JH_ENTRY(list->next);
806 		bh = jh->bh;
807 		get_bh(bh);
808 		if (!trylock_buffer(bh)) {
809 			if (!buffer_dirty(bh)) {
810 				list_move(&jh->list, &tmp);
811 				goto loop_next;
812 			}
813 			spin_unlock(lock);
814 			if (chunk.nr)
815 				write_ordered_chunk(&chunk);
816 			wait_on_buffer(bh);
817 			cond_resched();
818 			spin_lock(lock);
819 			goto loop_next;
820 		}
821 		/*
822 		 * in theory, dirty non-uptodate buffers should never get here,
823 		 * but the upper layer io error paths still have a few quirks.
824 		 * Handle them here as gracefully as we can
825 		 */
826 		if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
827 			clear_buffer_dirty(bh);
828 			ret = -EIO;
829 		}
830 		if (buffer_dirty(bh)) {
831 			list_move(&jh->list, &tmp);
832 			add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
833 		} else {
834 			reiserfs_free_jh(bh);
835 			unlock_buffer(bh);
836 		}
837 loop_next:
838 		put_bh(bh);
839 		cond_resched_lock(lock);
840 	}
841 	if (chunk.nr) {
842 		spin_unlock(lock);
843 		write_ordered_chunk(&chunk);
844 		spin_lock(lock);
845 	}
846 	while (!list_empty(&tmp)) {
847 		jh = JH_ENTRY(tmp.prev);
848 		bh = jh->bh;
849 		get_bh(bh);
850 		reiserfs_free_jh(bh);
851 
852 		if (buffer_locked(bh)) {
853 			spin_unlock(lock);
854 			wait_on_buffer(bh);
855 			spin_lock(lock);
856 		}
857 		if (!buffer_uptodate(bh)) {
858 			ret = -EIO;
859 		}
860 		/*
861 		 * ugly interaction with invalidate_folio here.
862 		 * reiserfs_invalidate_folio will pin any buffer that has a
863 		 * valid journal head from an older transaction.  If someone
864 		 * else sets our buffer dirty after we write it in the first
865 		 * loop, and then someone truncates the page away, nobody
866 		 * will ever write the buffer. We're safe if we write the
867 		 * page one last time after freeing the journal header.
868 		 */
869 		if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
870 			spin_unlock(lock);
871 			ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
872 			spin_lock(lock);
873 		}
874 		put_bh(bh);
875 		cond_resched_lock(lock);
876 	}
877 	spin_unlock(lock);
878 	return ret;
879 }
880 
881 static int flush_older_commits(struct super_block *s,
882 			       struct reiserfs_journal_list *jl)
883 {
884 	struct reiserfs_journal *journal = SB_JOURNAL(s);
885 	struct reiserfs_journal_list *other_jl;
886 	struct reiserfs_journal_list *first_jl;
887 	struct list_head *entry;
888 	unsigned int trans_id = jl->j_trans_id;
889 	unsigned int other_trans_id;
890 
891 find_first:
892 	/*
893 	 * first we walk backwards to find the oldest uncommitted transation
894 	 */
895 	first_jl = jl;
896 	entry = jl->j_list.prev;
897 	while (1) {
898 		other_jl = JOURNAL_LIST_ENTRY(entry);
899 		if (entry == &journal->j_journal_list ||
900 		    atomic_read(&other_jl->j_older_commits_done))
901 			break;
902 
903 		first_jl = other_jl;
904 		entry = other_jl->j_list.prev;
905 	}
906 
907 	/* if we didn't find any older uncommitted transactions, return now */
908 	if (first_jl == jl) {
909 		return 0;
910 	}
911 
912 	entry = &first_jl->j_list;
913 	while (1) {
914 		other_jl = JOURNAL_LIST_ENTRY(entry);
915 		other_trans_id = other_jl->j_trans_id;
916 
917 		if (other_trans_id < trans_id) {
918 			if (atomic_read(&other_jl->j_commit_left) != 0) {
919 				flush_commit_list(s, other_jl, 0);
920 
921 				/* list we were called with is gone, return */
922 				if (!journal_list_still_alive(s, trans_id))
923 					return 1;
924 
925 				/*
926 				 * the one we just flushed is gone, this means
927 				 * all older lists are also gone, so first_jl
928 				 * is no longer valid either.  Go back to the
929 				 * beginning.
930 				 */
931 				if (!journal_list_still_alive
932 				    (s, other_trans_id)) {
933 					goto find_first;
934 				}
935 			}
936 			entry = entry->next;
937 			if (entry == &journal->j_journal_list)
938 				return 0;
939 		} else {
940 			return 0;
941 		}
942 	}
943 	return 0;
944 }
945 
946 static int reiserfs_async_progress_wait(struct super_block *s)
947 {
948 	struct reiserfs_journal *j = SB_JOURNAL(s);
949 
950 	if (atomic_read(&j->j_async_throttle)) {
951 		int depth;
952 
953 		depth = reiserfs_write_unlock_nested(s);
954 		wait_var_event_timeout(&j->j_async_throttle,
955 				       atomic_read(&j->j_async_throttle) == 0,
956 				       HZ / 10);
957 		reiserfs_write_lock_nested(s, depth);
958 	}
959 
960 	return 0;
961 }
962 
963 /*
964  * if this journal list still has commit blocks unflushed, send them to disk.
965  *
966  * log areas must be flushed in order (transaction 2 can't commit before
967  * transaction 1) Before the commit block can by written, every other log
968  * block must be safely on disk
969  */
970 static int flush_commit_list(struct super_block *s,
971 			     struct reiserfs_journal_list *jl, int flushall)
972 {
973 	int i;
974 	b_blocknr_t bn;
975 	struct buffer_head *tbh = NULL;
976 	unsigned int trans_id = jl->j_trans_id;
977 	struct reiserfs_journal *journal = SB_JOURNAL(s);
978 	int retval = 0;
979 	int write_len;
980 	int depth;
981 
982 	reiserfs_check_lock_depth(s, "flush_commit_list");
983 
984 	if (atomic_read(&jl->j_older_commits_done)) {
985 		return 0;
986 	}
987 
988 	/*
989 	 * before we can put our commit blocks on disk, we have to make
990 	 * sure everyone older than us is on disk too
991 	 */
992 	BUG_ON(jl->j_len <= 0);
993 	BUG_ON(trans_id == journal->j_trans_id);
994 
995 	get_journal_list(jl);
996 	if (flushall) {
997 		if (flush_older_commits(s, jl) == 1) {
998 			/*
999 			 * list disappeared during flush_older_commits.
1000 			 * return
1001 			 */
1002 			goto put_jl;
1003 		}
1004 	}
1005 
1006 	/* make sure nobody is trying to flush this one at the same time */
1007 	reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1008 
1009 	if (!journal_list_still_alive(s, trans_id)) {
1010 		mutex_unlock(&jl->j_commit_mutex);
1011 		goto put_jl;
1012 	}
1013 	BUG_ON(jl->j_trans_id == 0);
1014 
1015 	/* this commit is done, exit */
1016 	if (atomic_read(&jl->j_commit_left) <= 0) {
1017 		if (flushall) {
1018 			atomic_set(&jl->j_older_commits_done, 1);
1019 		}
1020 		mutex_unlock(&jl->j_commit_mutex);
1021 		goto put_jl;
1022 	}
1023 
1024 	if (!list_empty(&jl->j_bh_list)) {
1025 		int ret;
1026 
1027 		/*
1028 		 * We might sleep in numerous places inside
1029 		 * write_ordered_buffers. Relax the write lock.
1030 		 */
1031 		depth = reiserfs_write_unlock_nested(s);
1032 		ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1033 					    journal, jl, &jl->j_bh_list);
1034 		if (ret < 0 && retval == 0)
1035 			retval = ret;
1036 		reiserfs_write_lock_nested(s, depth);
1037 	}
1038 	BUG_ON(!list_empty(&jl->j_bh_list));
1039 	/*
1040 	 * for the description block and all the log blocks, submit any buffers
1041 	 * that haven't already reached the disk.  Try to write at least 256
1042 	 * log blocks. later on, we will only wait on blocks that correspond
1043 	 * to this transaction, but while we're unplugging we might as well
1044 	 * get a chunk of data on there.
1045 	 */
1046 	atomic_inc(&journal->j_async_throttle);
1047 	write_len = jl->j_len + 1;
1048 	if (write_len < 256)
1049 		write_len = 256;
1050 	for (i = 0 ; i < write_len ; i++) {
1051 		bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1052 		    SB_ONDISK_JOURNAL_SIZE(s);
1053 		tbh = journal_find_get_block(s, bn);
1054 		if (tbh) {
1055 			if (buffer_dirty(tbh)) {
1056 		            depth = reiserfs_write_unlock_nested(s);
1057 			    ll_rw_block(REQ_OP_WRITE, 0, 1, &tbh);
1058 			    reiserfs_write_lock_nested(s, depth);
1059 			}
1060 			put_bh(tbh) ;
1061 		}
1062 	}
1063 	if (atomic_dec_and_test(&journal->j_async_throttle))
1064 		wake_up_var(&journal->j_async_throttle);
1065 
1066 	for (i = 0; i < (jl->j_len + 1); i++) {
1067 		bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1068 		    (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1069 		tbh = journal_find_get_block(s, bn);
1070 
1071 		depth = reiserfs_write_unlock_nested(s);
1072 		__wait_on_buffer(tbh);
1073 		reiserfs_write_lock_nested(s, depth);
1074 		/*
1075 		 * since we're using ll_rw_blk above, it might have skipped
1076 		 * over a locked buffer.  Double check here
1077 		 */
1078 		/* redundant, sync_dirty_buffer() checks */
1079 		if (buffer_dirty(tbh)) {
1080 			depth = reiserfs_write_unlock_nested(s);
1081 			sync_dirty_buffer(tbh);
1082 			reiserfs_write_lock_nested(s, depth);
1083 		}
1084 		if (unlikely(!buffer_uptodate(tbh))) {
1085 #ifdef CONFIG_REISERFS_CHECK
1086 			reiserfs_warning(s, "journal-601",
1087 					 "buffer write failed");
1088 #endif
1089 			retval = -EIO;
1090 		}
1091 		/* once for journal_find_get_block */
1092 		put_bh(tbh);
1093 		/* once due to original getblk in do_journal_end */
1094 		put_bh(tbh);
1095 		atomic_dec(&jl->j_commit_left);
1096 	}
1097 
1098 	BUG_ON(atomic_read(&jl->j_commit_left) != 1);
1099 
1100 	/*
1101 	 * If there was a write error in the journal - we can't commit
1102 	 * this transaction - it will be invalid and, if successful,
1103 	 * will just end up propagating the write error out to
1104 	 * the file system.
1105 	 */
1106 	if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1107 		if (buffer_dirty(jl->j_commit_bh))
1108 			BUG();
1109 		mark_buffer_dirty(jl->j_commit_bh) ;
1110 		depth = reiserfs_write_unlock_nested(s);
1111 		if (reiserfs_barrier_flush(s))
1112 			__sync_dirty_buffer(jl->j_commit_bh,
1113 					REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1114 		else
1115 			sync_dirty_buffer(jl->j_commit_bh);
1116 		reiserfs_write_lock_nested(s, depth);
1117 	}
1118 
1119 	/*
1120 	 * If there was a write error in the journal - we can't commit this
1121 	 * transaction - it will be invalid and, if successful, will just end
1122 	 * up propagating the write error out to the filesystem.
1123 	 */
1124 	if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1125 #ifdef CONFIG_REISERFS_CHECK
1126 		reiserfs_warning(s, "journal-615", "buffer write failed");
1127 #endif
1128 		retval = -EIO;
1129 	}
1130 	bforget(jl->j_commit_bh);
1131 	if (journal->j_last_commit_id != 0 &&
1132 	    (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1133 		reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1134 				 journal->j_last_commit_id, jl->j_trans_id);
1135 	}
1136 	journal->j_last_commit_id = jl->j_trans_id;
1137 
1138 	/*
1139 	 * now, every commit block is on the disk.  It is safe to allow
1140 	 * blocks freed during this transaction to be reallocated
1141 	 */
1142 	cleanup_freed_for_journal_list(s, jl);
1143 
1144 	retval = retval ? retval : journal->j_errno;
1145 
1146 	/* mark the metadata dirty */
1147 	if (!retval)
1148 		dirty_one_transaction(s, jl);
1149 	atomic_dec(&jl->j_commit_left);
1150 
1151 	if (flushall) {
1152 		atomic_set(&jl->j_older_commits_done, 1);
1153 	}
1154 	mutex_unlock(&jl->j_commit_mutex);
1155 put_jl:
1156 	put_journal_list(s, jl);
1157 
1158 	if (retval)
1159 		reiserfs_abort(s, retval, "Journal write error in %s",
1160 			       __func__);
1161 	return retval;
1162 }
1163 
1164 /*
1165  * flush_journal_list frequently needs to find a newer transaction for a
1166  * given block.  This does that, or returns NULL if it can't find anything
1167  */
1168 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1169 							  reiserfs_journal_cnode
1170 							  *cn)
1171 {
1172 	struct super_block *sb = cn->sb;
1173 	b_blocknr_t blocknr = cn->blocknr;
1174 
1175 	cn = cn->hprev;
1176 	while (cn) {
1177 		if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1178 			return cn->jlist;
1179 		}
1180 		cn = cn->hprev;
1181 	}
1182 	return NULL;
1183 }
1184 
1185 static void remove_journal_hash(struct super_block *,
1186 				struct reiserfs_journal_cnode **,
1187 				struct reiserfs_journal_list *, unsigned long,
1188 				int);
1189 
1190 /*
1191  * once all the real blocks have been flushed, it is safe to remove them
1192  * from the journal list for this transaction.  Aside from freeing the
1193  * cnode, this also allows the block to be reallocated for data blocks
1194  * if it had been deleted.
1195  */
1196 static void remove_all_from_journal_list(struct super_block *sb,
1197 					 struct reiserfs_journal_list *jl,
1198 					 int debug)
1199 {
1200 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1201 	struct reiserfs_journal_cnode *cn, *last;
1202 	cn = jl->j_realblock;
1203 
1204 	/*
1205 	 * which is better, to lock once around the whole loop, or
1206 	 * to lock for each call to remove_journal_hash?
1207 	 */
1208 	while (cn) {
1209 		if (cn->blocknr != 0) {
1210 			if (debug) {
1211 				reiserfs_warning(sb, "reiserfs-2201",
1212 						 "block %u, bh is %d, state %ld",
1213 						 cn->blocknr, cn->bh ? 1 : 0,
1214 						 cn->state);
1215 			}
1216 			cn->state = 0;
1217 			remove_journal_hash(sb, journal->j_list_hash_table,
1218 					    jl, cn->blocknr, 1);
1219 		}
1220 		last = cn;
1221 		cn = cn->next;
1222 		free_cnode(sb, last);
1223 	}
1224 	jl->j_realblock = NULL;
1225 }
1226 
1227 /*
1228  * if this timestamp is greater than the timestamp we wrote last to the
1229  * header block, write it to the header block.  once this is done, I can
1230  * safely say the log area for this transaction won't ever be replayed,
1231  * and I can start releasing blocks in this transaction for reuse as data
1232  * blocks.  called by flush_journal_list, before it calls
1233  * remove_all_from_journal_list
1234  */
1235 static int _update_journal_header_block(struct super_block *sb,
1236 					unsigned long offset,
1237 					unsigned int trans_id)
1238 {
1239 	struct reiserfs_journal_header *jh;
1240 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1241 	int depth;
1242 
1243 	if (reiserfs_is_journal_aborted(journal))
1244 		return -EIO;
1245 
1246 	if (trans_id >= journal->j_last_flush_trans_id) {
1247 		if (buffer_locked((journal->j_header_bh))) {
1248 			depth = reiserfs_write_unlock_nested(sb);
1249 			__wait_on_buffer(journal->j_header_bh);
1250 			reiserfs_write_lock_nested(sb, depth);
1251 			if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1252 #ifdef CONFIG_REISERFS_CHECK
1253 				reiserfs_warning(sb, "journal-699",
1254 						 "buffer write failed");
1255 #endif
1256 				return -EIO;
1257 			}
1258 		}
1259 		journal->j_last_flush_trans_id = trans_id;
1260 		journal->j_first_unflushed_offset = offset;
1261 		jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1262 							b_data);
1263 		jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1264 		jh->j_first_unflushed_offset = cpu_to_le32(offset);
1265 		jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1266 
1267 		set_buffer_dirty(journal->j_header_bh);
1268 		depth = reiserfs_write_unlock_nested(sb);
1269 
1270 		if (reiserfs_barrier_flush(sb))
1271 			__sync_dirty_buffer(journal->j_header_bh,
1272 					REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1273 		else
1274 			sync_dirty_buffer(journal->j_header_bh);
1275 
1276 		reiserfs_write_lock_nested(sb, depth);
1277 		if (!buffer_uptodate(journal->j_header_bh)) {
1278 			reiserfs_warning(sb, "journal-837",
1279 					 "IO error during journal replay");
1280 			return -EIO;
1281 		}
1282 	}
1283 	return 0;
1284 }
1285 
1286 static int update_journal_header_block(struct super_block *sb,
1287 				       unsigned long offset,
1288 				       unsigned int trans_id)
1289 {
1290 	return _update_journal_header_block(sb, offset, trans_id);
1291 }
1292 
1293 /*
1294 ** flush any and all journal lists older than you are
1295 ** can only be called from flush_journal_list
1296 */
1297 static int flush_older_journal_lists(struct super_block *sb,
1298 				     struct reiserfs_journal_list *jl)
1299 {
1300 	struct list_head *entry;
1301 	struct reiserfs_journal_list *other_jl;
1302 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1303 	unsigned int trans_id = jl->j_trans_id;
1304 
1305 	/*
1306 	 * we know we are the only ones flushing things, no extra race
1307 	 * protection is required.
1308 	 */
1309 restart:
1310 	entry = journal->j_journal_list.next;
1311 	/* Did we wrap? */
1312 	if (entry == &journal->j_journal_list)
1313 		return 0;
1314 	other_jl = JOURNAL_LIST_ENTRY(entry);
1315 	if (other_jl->j_trans_id < trans_id) {
1316 		BUG_ON(other_jl->j_refcount <= 0);
1317 		/* do not flush all */
1318 		flush_journal_list(sb, other_jl, 0);
1319 
1320 		/* other_jl is now deleted from the list */
1321 		goto restart;
1322 	}
1323 	return 0;
1324 }
1325 
1326 static void del_from_work_list(struct super_block *s,
1327 			       struct reiserfs_journal_list *jl)
1328 {
1329 	struct reiserfs_journal *journal = SB_JOURNAL(s);
1330 	if (!list_empty(&jl->j_working_list)) {
1331 		list_del_init(&jl->j_working_list);
1332 		journal->j_num_work_lists--;
1333 	}
1334 }
1335 
1336 /*
1337  * flush a journal list, both commit and real blocks
1338  *
1339  * always set flushall to 1, unless you are calling from inside
1340  * flush_journal_list
1341  *
1342  * IMPORTANT.  This can only be called while there are no journal writers,
1343  * and the journal is locked.  That means it can only be called from
1344  * do_journal_end, or by journal_release
1345  */
1346 static int flush_journal_list(struct super_block *s,
1347 			      struct reiserfs_journal_list *jl, int flushall)
1348 {
1349 	struct reiserfs_journal_list *pjl;
1350 	struct reiserfs_journal_cnode *cn;
1351 	int count;
1352 	int was_jwait = 0;
1353 	int was_dirty = 0;
1354 	struct buffer_head *saved_bh;
1355 	unsigned long j_len_saved = jl->j_len;
1356 	struct reiserfs_journal *journal = SB_JOURNAL(s);
1357 	int err = 0;
1358 	int depth;
1359 
1360 	BUG_ON(j_len_saved <= 0);
1361 
1362 	if (atomic_read(&journal->j_wcount) != 0) {
1363 		reiserfs_warning(s, "clm-2048", "called with wcount %d",
1364 				 atomic_read(&journal->j_wcount));
1365 	}
1366 
1367 	/* if flushall == 0, the lock is already held */
1368 	if (flushall) {
1369 		reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1370 	} else if (mutex_trylock(&journal->j_flush_mutex)) {
1371 		BUG();
1372 	}
1373 
1374 	count = 0;
1375 	if (j_len_saved > journal->j_trans_max) {
1376 		reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1377 			       j_len_saved, jl->j_trans_id);
1378 		return 0;
1379 	}
1380 
1381 	/* if all the work is already done, get out of here */
1382 	if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1383 	    atomic_read(&jl->j_commit_left) <= 0) {
1384 		goto flush_older_and_return;
1385 	}
1386 
1387 	/*
1388 	 * start by putting the commit list on disk.  This will also flush
1389 	 * the commit lists of any olders transactions
1390 	 */
1391 	flush_commit_list(s, jl, 1);
1392 
1393 	if (!(jl->j_state & LIST_DIRTY)
1394 	    && !reiserfs_is_journal_aborted(journal))
1395 		BUG();
1396 
1397 	/* are we done now? */
1398 	if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1399 	    atomic_read(&jl->j_commit_left) <= 0) {
1400 		goto flush_older_and_return;
1401 	}
1402 
1403 	/*
1404 	 * loop through each cnode, see if we need to write it,
1405 	 * or wait on a more recent transaction, or just ignore it
1406 	 */
1407 	if (atomic_read(&journal->j_wcount) != 0) {
1408 		reiserfs_panic(s, "journal-844", "journal list is flushing, "
1409 			       "wcount is not 0");
1410 	}
1411 	cn = jl->j_realblock;
1412 	while (cn) {
1413 		was_jwait = 0;
1414 		was_dirty = 0;
1415 		saved_bh = NULL;
1416 		/* blocknr of 0 is no longer in the hash, ignore it */
1417 		if (cn->blocknr == 0) {
1418 			goto free_cnode;
1419 		}
1420 
1421 		/*
1422 		 * This transaction failed commit.
1423 		 * Don't write out to the disk
1424 		 */
1425 		if (!(jl->j_state & LIST_DIRTY))
1426 			goto free_cnode;
1427 
1428 		pjl = find_newer_jl_for_cn(cn);
1429 		/*
1430 		 * the order is important here.  We check pjl to make sure we
1431 		 * don't clear BH_JDirty_wait if we aren't the one writing this
1432 		 * block to disk
1433 		 */
1434 		if (!pjl && cn->bh) {
1435 			saved_bh = cn->bh;
1436 
1437 			/*
1438 			 * we do this to make sure nobody releases the
1439 			 * buffer while we are working with it
1440 			 */
1441 			get_bh(saved_bh);
1442 
1443 			if (buffer_journal_dirty(saved_bh)) {
1444 				BUG_ON(!can_dirty(cn));
1445 				was_jwait = 1;
1446 				was_dirty = 1;
1447 			} else if (can_dirty(cn)) {
1448 				/*
1449 				 * everything with !pjl && jwait
1450 				 * should be writable
1451 				 */
1452 				BUG();
1453 			}
1454 		}
1455 
1456 		/*
1457 		 * if someone has this block in a newer transaction, just make
1458 		 * sure they are committed, and don't try writing it to disk
1459 		 */
1460 		if (pjl) {
1461 			if (atomic_read(&pjl->j_commit_left))
1462 				flush_commit_list(s, pjl, 1);
1463 			goto free_cnode;
1464 		}
1465 
1466 		/*
1467 		 * bh == NULL when the block got to disk on its own, OR,
1468 		 * the block got freed in a future transaction
1469 		 */
1470 		if (saved_bh == NULL) {
1471 			goto free_cnode;
1472 		}
1473 
1474 		/*
1475 		 * this should never happen.  kupdate_one_transaction has
1476 		 * this list locked while it works, so we should never see a
1477 		 * buffer here that is not marked JDirty_wait
1478 		 */
1479 		if ((!was_jwait) && !buffer_locked(saved_bh)) {
1480 			reiserfs_warning(s, "journal-813",
1481 					 "BAD! buffer %llu %cdirty %cjwait, "
1482 					 "not in a newer transaction",
1483 					 (unsigned long long)saved_bh->
1484 					 b_blocknr, was_dirty ? ' ' : '!',
1485 					 was_jwait ? ' ' : '!');
1486 		}
1487 		if (was_dirty) {
1488 			/*
1489 			 * we inc again because saved_bh gets decremented
1490 			 * at free_cnode
1491 			 */
1492 			get_bh(saved_bh);
1493 			set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1494 			lock_buffer(saved_bh);
1495 			BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1496 			if (buffer_dirty(saved_bh))
1497 				submit_logged_buffer(saved_bh);
1498 			else
1499 				unlock_buffer(saved_bh);
1500 			count++;
1501 		} else {
1502 			reiserfs_warning(s, "clm-2082",
1503 					 "Unable to flush buffer %llu in %s",
1504 					 (unsigned long long)saved_bh->
1505 					 b_blocknr, __func__);
1506 		}
1507 free_cnode:
1508 		cn = cn->next;
1509 		if (saved_bh) {
1510 			/*
1511 			 * we incremented this to keep others from
1512 			 * taking the buffer head away
1513 			 */
1514 			put_bh(saved_bh);
1515 			if (atomic_read(&saved_bh->b_count) < 0) {
1516 				reiserfs_warning(s, "journal-945",
1517 						 "saved_bh->b_count < 0");
1518 			}
1519 		}
1520 	}
1521 	if (count > 0) {
1522 		cn = jl->j_realblock;
1523 		while (cn) {
1524 			if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1525 				if (!cn->bh) {
1526 					reiserfs_panic(s, "journal-1011",
1527 						       "cn->bh is NULL");
1528 				}
1529 
1530 				depth = reiserfs_write_unlock_nested(s);
1531 				__wait_on_buffer(cn->bh);
1532 				reiserfs_write_lock_nested(s, depth);
1533 
1534 				if (!cn->bh) {
1535 					reiserfs_panic(s, "journal-1012",
1536 						       "cn->bh is NULL");
1537 				}
1538 				if (unlikely(!buffer_uptodate(cn->bh))) {
1539 #ifdef CONFIG_REISERFS_CHECK
1540 					reiserfs_warning(s, "journal-949",
1541 							 "buffer write failed");
1542 #endif
1543 					err = -EIO;
1544 				}
1545 				/*
1546 				 * note, we must clear the JDirty_wait bit
1547 				 * after the up to date check, otherwise we
1548 				 * race against our flushpage routine
1549 				 */
1550 				BUG_ON(!test_clear_buffer_journal_dirty
1551 				       (cn->bh));
1552 
1553 				/* drop one ref for us */
1554 				put_bh(cn->bh);
1555 				/* drop one ref for journal_mark_dirty */
1556 				release_buffer_page(cn->bh);
1557 			}
1558 			cn = cn->next;
1559 		}
1560 	}
1561 
1562 	if (err)
1563 		reiserfs_abort(s, -EIO,
1564 			       "Write error while pushing transaction to disk in %s",
1565 			       __func__);
1566 flush_older_and_return:
1567 
1568 	/*
1569 	 * before we can update the journal header block, we _must_ flush all
1570 	 * real blocks from all older transactions to disk.  This is because
1571 	 * once the header block is updated, this transaction will not be
1572 	 * replayed after a crash
1573 	 */
1574 	if (flushall) {
1575 		flush_older_journal_lists(s, jl);
1576 	}
1577 
1578 	err = journal->j_errno;
1579 	/*
1580 	 * before we can remove everything from the hash tables for this
1581 	 * transaction, we must make sure it can never be replayed
1582 	 *
1583 	 * since we are only called from do_journal_end, we know for sure there
1584 	 * are no allocations going on while we are flushing journal lists.  So,
1585 	 * we only need to update the journal header block for the last list
1586 	 * being flushed
1587 	 */
1588 	if (!err && flushall) {
1589 		err =
1590 		    update_journal_header_block(s,
1591 						(jl->j_start + jl->j_len +
1592 						 2) % SB_ONDISK_JOURNAL_SIZE(s),
1593 						jl->j_trans_id);
1594 		if (err)
1595 			reiserfs_abort(s, -EIO,
1596 				       "Write error while updating journal header in %s",
1597 				       __func__);
1598 	}
1599 	remove_all_from_journal_list(s, jl, 0);
1600 	list_del_init(&jl->j_list);
1601 	journal->j_num_lists--;
1602 	del_from_work_list(s, jl);
1603 
1604 	if (journal->j_last_flush_id != 0 &&
1605 	    (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1606 		reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1607 				 journal->j_last_flush_id, jl->j_trans_id);
1608 	}
1609 	journal->j_last_flush_id = jl->j_trans_id;
1610 
1611 	/*
1612 	 * not strictly required since we are freeing the list, but it should
1613 	 * help find code using dead lists later on
1614 	 */
1615 	jl->j_len = 0;
1616 	atomic_set(&jl->j_nonzerolen, 0);
1617 	jl->j_start = 0;
1618 	jl->j_realblock = NULL;
1619 	jl->j_commit_bh = NULL;
1620 	jl->j_trans_id = 0;
1621 	jl->j_state = 0;
1622 	put_journal_list(s, jl);
1623 	if (flushall)
1624 		mutex_unlock(&journal->j_flush_mutex);
1625 	return err;
1626 }
1627 
1628 static int write_one_transaction(struct super_block *s,
1629 				 struct reiserfs_journal_list *jl,
1630 				 struct buffer_chunk *chunk)
1631 {
1632 	struct reiserfs_journal_cnode *cn;
1633 	int ret = 0;
1634 
1635 	jl->j_state |= LIST_TOUCHED;
1636 	del_from_work_list(s, jl);
1637 	if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1638 		return 0;
1639 	}
1640 
1641 	cn = jl->j_realblock;
1642 	while (cn) {
1643 		/*
1644 		 * if the blocknr == 0, this has been cleared from the hash,
1645 		 * skip it
1646 		 */
1647 		if (cn->blocknr == 0) {
1648 			goto next;
1649 		}
1650 		if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1651 			struct buffer_head *tmp_bh;
1652 			/*
1653 			 * we can race against journal_mark_freed when we try
1654 			 * to lock_buffer(cn->bh), so we have to inc the buffer
1655 			 * count, and recheck things after locking
1656 			 */
1657 			tmp_bh = cn->bh;
1658 			get_bh(tmp_bh);
1659 			lock_buffer(tmp_bh);
1660 			if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1661 				if (!buffer_journal_dirty(tmp_bh) ||
1662 				    buffer_journal_prepared(tmp_bh))
1663 					BUG();
1664 				add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1665 				ret++;
1666 			} else {
1667 				/* note, cn->bh might be null now */
1668 				unlock_buffer(tmp_bh);
1669 			}
1670 			put_bh(tmp_bh);
1671 		}
1672 next:
1673 		cn = cn->next;
1674 		cond_resched();
1675 	}
1676 	return ret;
1677 }
1678 
1679 /* used by flush_commit_list */
1680 static void dirty_one_transaction(struct super_block *s,
1681 				 struct reiserfs_journal_list *jl)
1682 {
1683 	struct reiserfs_journal_cnode *cn;
1684 	struct reiserfs_journal_list *pjl;
1685 
1686 	jl->j_state |= LIST_DIRTY;
1687 	cn = jl->j_realblock;
1688 	while (cn) {
1689 		/*
1690 		 * look for a more recent transaction that logged this
1691 		 * buffer.  Only the most recent transaction with a buffer in
1692 		 * it is allowed to send that buffer to disk
1693 		 */
1694 		pjl = find_newer_jl_for_cn(cn);
1695 		if (!pjl && cn->blocknr && cn->bh
1696 		    && buffer_journal_dirty(cn->bh)) {
1697 			BUG_ON(!can_dirty(cn));
1698 			/*
1699 			 * if the buffer is prepared, it will either be logged
1700 			 * or restored.  If restored, we need to make sure
1701 			 * it actually gets marked dirty
1702 			 */
1703 			clear_buffer_journal_new(cn->bh);
1704 			if (buffer_journal_prepared(cn->bh)) {
1705 				set_buffer_journal_restore_dirty(cn->bh);
1706 			} else {
1707 				set_buffer_journal_test(cn->bh);
1708 				mark_buffer_dirty(cn->bh);
1709 			}
1710 		}
1711 		cn = cn->next;
1712 	}
1713 }
1714 
1715 static int kupdate_transactions(struct super_block *s,
1716 				struct reiserfs_journal_list *jl,
1717 				struct reiserfs_journal_list **next_jl,
1718 				unsigned int *next_trans_id,
1719 				int num_blocks, int num_trans)
1720 {
1721 	int ret = 0;
1722 	int written = 0;
1723 	int transactions_flushed = 0;
1724 	unsigned int orig_trans_id = jl->j_trans_id;
1725 	struct buffer_chunk chunk;
1726 	struct list_head *entry;
1727 	struct reiserfs_journal *journal = SB_JOURNAL(s);
1728 	chunk.nr = 0;
1729 
1730 	reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1731 	if (!journal_list_still_alive(s, orig_trans_id)) {
1732 		goto done;
1733 	}
1734 
1735 	/*
1736 	 * we've got j_flush_mutex held, nobody is going to delete any
1737 	 * of these lists out from underneath us
1738 	 */
1739 	while ((num_trans && transactions_flushed < num_trans) ||
1740 	       (!num_trans && written < num_blocks)) {
1741 
1742 		if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1743 		    atomic_read(&jl->j_commit_left)
1744 		    || !(jl->j_state & LIST_DIRTY)) {
1745 			del_from_work_list(s, jl);
1746 			break;
1747 		}
1748 		ret = write_one_transaction(s, jl, &chunk);
1749 
1750 		if (ret < 0)
1751 			goto done;
1752 		transactions_flushed++;
1753 		written += ret;
1754 		entry = jl->j_list.next;
1755 
1756 		/* did we wrap? */
1757 		if (entry == &journal->j_journal_list) {
1758 			break;
1759 		}
1760 		jl = JOURNAL_LIST_ENTRY(entry);
1761 
1762 		/* don't bother with older transactions */
1763 		if (jl->j_trans_id <= orig_trans_id)
1764 			break;
1765 	}
1766 	if (chunk.nr) {
1767 		write_chunk(&chunk);
1768 	}
1769 
1770 done:
1771 	mutex_unlock(&journal->j_flush_mutex);
1772 	return ret;
1773 }
1774 
1775 /*
1776  * for o_sync and fsync heavy applications, they tend to use
1777  * all the journa list slots with tiny transactions.  These
1778  * trigger lots and lots of calls to update the header block, which
1779  * adds seeks and slows things down.
1780  *
1781  * This function tries to clear out a large chunk of the journal lists
1782  * at once, which makes everything faster since only the newest journal
1783  * list updates the header block
1784  */
1785 static int flush_used_journal_lists(struct super_block *s,
1786 				    struct reiserfs_journal_list *jl)
1787 {
1788 	unsigned long len = 0;
1789 	unsigned long cur_len;
1790 	int i;
1791 	int limit = 256;
1792 	struct reiserfs_journal_list *tjl;
1793 	struct reiserfs_journal_list *flush_jl;
1794 	unsigned int trans_id;
1795 	struct reiserfs_journal *journal = SB_JOURNAL(s);
1796 
1797 	flush_jl = tjl = jl;
1798 
1799 	/* in data logging mode, try harder to flush a lot of blocks */
1800 	if (reiserfs_data_log(s))
1801 		limit = 1024;
1802 	/* flush for 256 transactions or limit blocks, whichever comes first */
1803 	for (i = 0; i < 256 && len < limit; i++) {
1804 		if (atomic_read(&tjl->j_commit_left) ||
1805 		    tjl->j_trans_id < jl->j_trans_id) {
1806 			break;
1807 		}
1808 		cur_len = atomic_read(&tjl->j_nonzerolen);
1809 		if (cur_len > 0) {
1810 			tjl->j_state &= ~LIST_TOUCHED;
1811 		}
1812 		len += cur_len;
1813 		flush_jl = tjl;
1814 		if (tjl->j_list.next == &journal->j_journal_list)
1815 			break;
1816 		tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1817 	}
1818 	get_journal_list(jl);
1819 	get_journal_list(flush_jl);
1820 
1821 	/*
1822 	 * try to find a group of blocks we can flush across all the
1823 	 * transactions, but only bother if we've actually spanned
1824 	 * across multiple lists
1825 	 */
1826 	if (flush_jl != jl)
1827 		kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1828 
1829 	flush_journal_list(s, flush_jl, 1);
1830 	put_journal_list(s, flush_jl);
1831 	put_journal_list(s, jl);
1832 	return 0;
1833 }
1834 
1835 /*
1836  * removes any nodes in table with name block and dev as bh.
1837  * only touchs the hnext and hprev pointers.
1838  */
1839 static void remove_journal_hash(struct super_block *sb,
1840 			 struct reiserfs_journal_cnode **table,
1841 			 struct reiserfs_journal_list *jl,
1842 			 unsigned long block, int remove_freed)
1843 {
1844 	struct reiserfs_journal_cnode *cur;
1845 	struct reiserfs_journal_cnode **head;
1846 
1847 	head = &(journal_hash(table, sb, block));
1848 	if (!head) {
1849 		return;
1850 	}
1851 	cur = *head;
1852 	while (cur) {
1853 		if (cur->blocknr == block && cur->sb == sb
1854 		    && (jl == NULL || jl == cur->jlist)
1855 		    && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1856 			if (cur->hnext) {
1857 				cur->hnext->hprev = cur->hprev;
1858 			}
1859 			if (cur->hprev) {
1860 				cur->hprev->hnext = cur->hnext;
1861 			} else {
1862 				*head = cur->hnext;
1863 			}
1864 			cur->blocknr = 0;
1865 			cur->sb = NULL;
1866 			cur->state = 0;
1867 			/*
1868 			 * anybody who clears the cur->bh will also
1869 			 * dec the nonzerolen
1870 			 */
1871 			if (cur->bh && cur->jlist)
1872 				atomic_dec(&cur->jlist->j_nonzerolen);
1873 			cur->bh = NULL;
1874 			cur->jlist = NULL;
1875 		}
1876 		cur = cur->hnext;
1877 	}
1878 }
1879 
1880 static void free_journal_ram(struct super_block *sb)
1881 {
1882 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1883 	kfree(journal->j_current_jl);
1884 	journal->j_num_lists--;
1885 
1886 	vfree(journal->j_cnode_free_orig);
1887 	free_list_bitmaps(sb, journal->j_list_bitmap);
1888 	free_bitmap_nodes(sb);	/* must be after free_list_bitmaps */
1889 	if (journal->j_header_bh) {
1890 		brelse(journal->j_header_bh);
1891 	}
1892 	/*
1893 	 * j_header_bh is on the journal dev, make sure
1894 	 * not to release the journal dev until we brelse j_header_bh
1895 	 */
1896 	release_journal_dev(sb, journal);
1897 	vfree(journal);
1898 }
1899 
1900 /*
1901  * call on unmount.  Only set error to 1 if you haven't made your way out
1902  * of read_super() yet.  Any other caller must keep error at 0.
1903  */
1904 static int do_journal_release(struct reiserfs_transaction_handle *th,
1905 			      struct super_block *sb, int error)
1906 {
1907 	struct reiserfs_transaction_handle myth;
1908 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1909 
1910 	/*
1911 	 * we only want to flush out transactions if we were
1912 	 * called with error == 0
1913 	 */
1914 	if (!error && !sb_rdonly(sb)) {
1915 		/* end the current trans */
1916 		BUG_ON(!th->t_trans_id);
1917 		do_journal_end(th, FLUSH_ALL);
1918 
1919 		/*
1920 		 * make sure something gets logged to force
1921 		 * our way into the flush code
1922 		 */
1923 		if (!journal_join(&myth, sb)) {
1924 			reiserfs_prepare_for_journal(sb,
1925 						     SB_BUFFER_WITH_SB(sb),
1926 						     1);
1927 			journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1928 			do_journal_end(&myth, FLUSH_ALL);
1929 		}
1930 	}
1931 
1932 	/* this also catches errors during the do_journal_end above */
1933 	if (!error && reiserfs_is_journal_aborted(journal)) {
1934 		memset(&myth, 0, sizeof(myth));
1935 		if (!journal_join_abort(&myth, sb)) {
1936 			reiserfs_prepare_for_journal(sb,
1937 						     SB_BUFFER_WITH_SB(sb),
1938 						     1);
1939 			journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1940 			do_journal_end(&myth, FLUSH_ALL);
1941 		}
1942 	}
1943 
1944 
1945 	/*
1946 	 * We must release the write lock here because
1947 	 * the workqueue job (flush_async_commit) needs this lock
1948 	 */
1949 	reiserfs_write_unlock(sb);
1950 
1951 	/*
1952 	 * Cancel flushing of old commits. Note that neither of these works
1953 	 * will be requeued because superblock is being shutdown and doesn't
1954 	 * have SB_ACTIVE set.
1955 	 */
1956 	reiserfs_cancel_old_flush(sb);
1957 	/* wait for all commits to finish */
1958 	cancel_delayed_work_sync(&SB_JOURNAL(sb)->j_work);
1959 
1960 	free_journal_ram(sb);
1961 
1962 	reiserfs_write_lock(sb);
1963 
1964 	return 0;
1965 }
1966 
1967 /* * call on unmount.  flush all journal trans, release all alloc'd ram */
1968 int journal_release(struct reiserfs_transaction_handle *th,
1969 		    struct super_block *sb)
1970 {
1971 	return do_journal_release(th, sb, 0);
1972 }
1973 
1974 /* only call from an error condition inside reiserfs_read_super!  */
1975 int journal_release_error(struct reiserfs_transaction_handle *th,
1976 			  struct super_block *sb)
1977 {
1978 	return do_journal_release(th, sb, 1);
1979 }
1980 
1981 /*
1982  * compares description block with commit block.
1983  * returns 1 if they differ, 0 if they are the same
1984  */
1985 static int journal_compare_desc_commit(struct super_block *sb,
1986 				       struct reiserfs_journal_desc *desc,
1987 				       struct reiserfs_journal_commit *commit)
1988 {
1989 	if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1990 	    get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1991 	    get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
1992 	    get_commit_trans_len(commit) <= 0) {
1993 		return 1;
1994 	}
1995 	return 0;
1996 }
1997 
1998 /*
1999  * returns 0 if it did not find a description block
2000  * returns -1 if it found a corrupt commit block
2001  * returns 1 if both desc and commit were valid
2002  * NOTE: only called during fs mount
2003  */
2004 static int journal_transaction_is_valid(struct super_block *sb,
2005 					struct buffer_head *d_bh,
2006 					unsigned int *oldest_invalid_trans_id,
2007 					unsigned long *newest_mount_id)
2008 {
2009 	struct reiserfs_journal_desc *desc;
2010 	struct reiserfs_journal_commit *commit;
2011 	struct buffer_head *c_bh;
2012 	unsigned long offset;
2013 
2014 	if (!d_bh)
2015 		return 0;
2016 
2017 	desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2018 	if (get_desc_trans_len(desc) > 0
2019 	    && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2020 		if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2021 		    && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2022 			reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2023 				       "journal-986: transaction "
2024 				       "is valid returning because trans_id %d is greater than "
2025 				       "oldest_invalid %lu",
2026 				       get_desc_trans_id(desc),
2027 				       *oldest_invalid_trans_id);
2028 			return 0;
2029 		}
2030 		if (newest_mount_id
2031 		    && *newest_mount_id > get_desc_mount_id(desc)) {
2032 			reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2033 				       "journal-1087: transaction "
2034 				       "is valid returning because mount_id %d is less than "
2035 				       "newest_mount_id %lu",
2036 				       get_desc_mount_id(desc),
2037 				       *newest_mount_id);
2038 			return -1;
2039 		}
2040 		if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2041 			reiserfs_warning(sb, "journal-2018",
2042 					 "Bad transaction length %d "
2043 					 "encountered, ignoring transaction",
2044 					 get_desc_trans_len(desc));
2045 			return -1;
2046 		}
2047 		offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2048 
2049 		/*
2050 		 * ok, we have a journal description block,
2051 		 * let's see if the transaction was valid
2052 		 */
2053 		c_bh =
2054 		    journal_bread(sb,
2055 				  SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2056 				  ((offset + get_desc_trans_len(desc) +
2057 				    1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2058 		if (!c_bh)
2059 			return 0;
2060 		commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2061 		if (journal_compare_desc_commit(sb, desc, commit)) {
2062 			reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2063 				       "journal_transaction_is_valid, commit offset %ld had bad "
2064 				       "time %d or length %d",
2065 				       c_bh->b_blocknr -
2066 				       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2067 				       get_commit_trans_id(commit),
2068 				       get_commit_trans_len(commit));
2069 			brelse(c_bh);
2070 			if (oldest_invalid_trans_id) {
2071 				*oldest_invalid_trans_id =
2072 				    get_desc_trans_id(desc);
2073 				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2074 					       "journal-1004: "
2075 					       "transaction_is_valid setting oldest invalid trans_id "
2076 					       "to %d",
2077 					       get_desc_trans_id(desc));
2078 			}
2079 			return -1;
2080 		}
2081 		brelse(c_bh);
2082 		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2083 			       "journal-1006: found valid "
2084 			       "transaction start offset %llu, len %d id %d",
2085 			       d_bh->b_blocknr -
2086 			       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2087 			       get_desc_trans_len(desc),
2088 			       get_desc_trans_id(desc));
2089 		return 1;
2090 	} else {
2091 		return 0;
2092 	}
2093 }
2094 
2095 static void brelse_array(struct buffer_head **heads, int num)
2096 {
2097 	int i;
2098 	for (i = 0; i < num; i++) {
2099 		brelse(heads[i]);
2100 	}
2101 }
2102 
2103 /*
2104  * given the start, and values for the oldest acceptable transactions,
2105  * this either reads in a replays a transaction, or returns because the
2106  * transaction is invalid, or too old.
2107  * NOTE: only called during fs mount
2108  */
2109 static int journal_read_transaction(struct super_block *sb,
2110 				    unsigned long cur_dblock,
2111 				    unsigned long oldest_start,
2112 				    unsigned int oldest_trans_id,
2113 				    unsigned long newest_mount_id)
2114 {
2115 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
2116 	struct reiserfs_journal_desc *desc;
2117 	struct reiserfs_journal_commit *commit;
2118 	unsigned int trans_id = 0;
2119 	struct buffer_head *c_bh;
2120 	struct buffer_head *d_bh;
2121 	struct buffer_head **log_blocks = NULL;
2122 	struct buffer_head **real_blocks = NULL;
2123 	unsigned int trans_offset;
2124 	int i;
2125 	int trans_half;
2126 
2127 	d_bh = journal_bread(sb, cur_dblock);
2128 	if (!d_bh)
2129 		return 1;
2130 	desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2131 	trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2132 	reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2133 		       "journal_read_transaction, offset %llu, len %d mount_id %d",
2134 		       d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2135 		       get_desc_trans_len(desc), get_desc_mount_id(desc));
2136 	if (get_desc_trans_id(desc) < oldest_trans_id) {
2137 		reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2138 			       "journal_read_trans skipping because %lu is too old",
2139 			       cur_dblock -
2140 			       SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2141 		brelse(d_bh);
2142 		return 1;
2143 	}
2144 	if (get_desc_mount_id(desc) != newest_mount_id) {
2145 		reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2146 			       "journal_read_trans skipping because %d is != "
2147 			       "newest_mount_id %lu", get_desc_mount_id(desc),
2148 			       newest_mount_id);
2149 		brelse(d_bh);
2150 		return 1;
2151 	}
2152 	c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2153 			     ((trans_offset + get_desc_trans_len(desc) + 1) %
2154 			      SB_ONDISK_JOURNAL_SIZE(sb)));
2155 	if (!c_bh) {
2156 		brelse(d_bh);
2157 		return 1;
2158 	}
2159 	commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2160 	if (journal_compare_desc_commit(sb, desc, commit)) {
2161 		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2162 			       "journal_read_transaction, "
2163 			       "commit offset %llu had bad time %d or length %d",
2164 			       c_bh->b_blocknr -
2165 			       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2166 			       get_commit_trans_id(commit),
2167 			       get_commit_trans_len(commit));
2168 		brelse(c_bh);
2169 		brelse(d_bh);
2170 		return 1;
2171 	}
2172 
2173 	if (bdev_read_only(sb->s_bdev)) {
2174 		reiserfs_warning(sb, "clm-2076",
2175 				 "device is readonly, unable to replay log");
2176 		brelse(c_bh);
2177 		brelse(d_bh);
2178 		return -EROFS;
2179 	}
2180 
2181 	trans_id = get_desc_trans_id(desc);
2182 	/*
2183 	 * now we know we've got a good transaction, and it was
2184 	 * inside the valid time ranges
2185 	 */
2186 	log_blocks = kmalloc_array(get_desc_trans_len(desc),
2187 				   sizeof(struct buffer_head *),
2188 				   GFP_NOFS);
2189 	real_blocks = kmalloc_array(get_desc_trans_len(desc),
2190 				    sizeof(struct buffer_head *),
2191 				    GFP_NOFS);
2192 	if (!log_blocks || !real_blocks) {
2193 		brelse(c_bh);
2194 		brelse(d_bh);
2195 		kfree(log_blocks);
2196 		kfree(real_blocks);
2197 		reiserfs_warning(sb, "journal-1169",
2198 				 "kmalloc failed, unable to mount FS");
2199 		return -1;
2200 	}
2201 	/* get all the buffer heads */
2202 	trans_half = journal_trans_half(sb->s_blocksize);
2203 	for (i = 0; i < get_desc_trans_len(desc); i++) {
2204 		log_blocks[i] =
2205 		    journal_getblk(sb,
2206 				   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2207 				   (trans_offset + 1 +
2208 				    i) % SB_ONDISK_JOURNAL_SIZE(sb));
2209 		if (i < trans_half) {
2210 			real_blocks[i] =
2211 			    sb_getblk(sb,
2212 				      le32_to_cpu(desc->j_realblock[i]));
2213 		} else {
2214 			real_blocks[i] =
2215 			    sb_getblk(sb,
2216 				      le32_to_cpu(commit->
2217 						  j_realblock[i - trans_half]));
2218 		}
2219 		if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2220 			reiserfs_warning(sb, "journal-1207",
2221 					 "REPLAY FAILURE fsck required! "
2222 					 "Block to replay is outside of "
2223 					 "filesystem");
2224 			goto abort_replay;
2225 		}
2226 		/* make sure we don't try to replay onto log or reserved area */
2227 		if (is_block_in_log_or_reserved_area
2228 		    (sb, real_blocks[i]->b_blocknr)) {
2229 			reiserfs_warning(sb, "journal-1204",
2230 					 "REPLAY FAILURE fsck required! "
2231 					 "Trying to replay onto a log block");
2232 abort_replay:
2233 			brelse_array(log_blocks, i);
2234 			brelse_array(real_blocks, i);
2235 			brelse(c_bh);
2236 			brelse(d_bh);
2237 			kfree(log_blocks);
2238 			kfree(real_blocks);
2239 			return -1;
2240 		}
2241 	}
2242 	/* read in the log blocks, memcpy to the corresponding real block */
2243 	ll_rw_block(REQ_OP_READ, 0, get_desc_trans_len(desc), log_blocks);
2244 	for (i = 0; i < get_desc_trans_len(desc); i++) {
2245 
2246 		wait_on_buffer(log_blocks[i]);
2247 		if (!buffer_uptodate(log_blocks[i])) {
2248 			reiserfs_warning(sb, "journal-1212",
2249 					 "REPLAY FAILURE fsck required! "
2250 					 "buffer write failed");
2251 			brelse_array(log_blocks + i,
2252 				     get_desc_trans_len(desc) - i);
2253 			brelse_array(real_blocks, get_desc_trans_len(desc));
2254 			brelse(c_bh);
2255 			brelse(d_bh);
2256 			kfree(log_blocks);
2257 			kfree(real_blocks);
2258 			return -1;
2259 		}
2260 		memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2261 		       real_blocks[i]->b_size);
2262 		set_buffer_uptodate(real_blocks[i]);
2263 		brelse(log_blocks[i]);
2264 	}
2265 	/* flush out the real blocks */
2266 	for (i = 0; i < get_desc_trans_len(desc); i++) {
2267 		set_buffer_dirty(real_blocks[i]);
2268 		write_dirty_buffer(real_blocks[i], 0);
2269 	}
2270 	for (i = 0; i < get_desc_trans_len(desc); i++) {
2271 		wait_on_buffer(real_blocks[i]);
2272 		if (!buffer_uptodate(real_blocks[i])) {
2273 			reiserfs_warning(sb, "journal-1226",
2274 					 "REPLAY FAILURE, fsck required! "
2275 					 "buffer write failed");
2276 			brelse_array(real_blocks + i,
2277 				     get_desc_trans_len(desc) - i);
2278 			brelse(c_bh);
2279 			brelse(d_bh);
2280 			kfree(log_blocks);
2281 			kfree(real_blocks);
2282 			return -1;
2283 		}
2284 		brelse(real_blocks[i]);
2285 	}
2286 	cur_dblock =
2287 	    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2288 	    ((trans_offset + get_desc_trans_len(desc) +
2289 	      2) % SB_ONDISK_JOURNAL_SIZE(sb));
2290 	reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2291 		       "journal-1095: setting journal " "start to offset %ld",
2292 		       cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2293 
2294 	/*
2295 	 * init starting values for the first transaction, in case
2296 	 * this is the last transaction to be replayed.
2297 	 */
2298 	journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2299 	journal->j_last_flush_trans_id = trans_id;
2300 	journal->j_trans_id = trans_id + 1;
2301 	/* check for trans_id overflow */
2302 	if (journal->j_trans_id == 0)
2303 		journal->j_trans_id = 10;
2304 	brelse(c_bh);
2305 	brelse(d_bh);
2306 	kfree(log_blocks);
2307 	kfree(real_blocks);
2308 	return 0;
2309 }
2310 
2311 /*
2312  * This function reads blocks starting from block and to max_block of bufsize
2313  * size (but no more than BUFNR blocks at a time). This proved to improve
2314  * mounting speed on self-rebuilding raid5 arrays at least.
2315  * Right now it is only used from journal code. But later we might use it
2316  * from other places.
2317  * Note: Do not use journal_getblk/sb_getblk functions here!
2318  */
2319 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2320 					   b_blocknr_t block, int bufsize,
2321 					   b_blocknr_t max_block)
2322 {
2323 	struct buffer_head *bhlist[BUFNR];
2324 	unsigned int blocks = BUFNR;
2325 	struct buffer_head *bh;
2326 	int i, j;
2327 
2328 	bh = __getblk(dev, block, bufsize);
2329 	if (buffer_uptodate(bh))
2330 		return (bh);
2331 
2332 	if (block + BUFNR > max_block) {
2333 		blocks = max_block - block;
2334 	}
2335 	bhlist[0] = bh;
2336 	j = 1;
2337 	for (i = 1; i < blocks; i++) {
2338 		bh = __getblk(dev, block + i, bufsize);
2339 		if (buffer_uptodate(bh)) {
2340 			brelse(bh);
2341 			break;
2342 		} else
2343 			bhlist[j++] = bh;
2344 	}
2345 	ll_rw_block(REQ_OP_READ, 0, j, bhlist);
2346 	for (i = 1; i < j; i++)
2347 		brelse(bhlist[i]);
2348 	bh = bhlist[0];
2349 	wait_on_buffer(bh);
2350 	if (buffer_uptodate(bh))
2351 		return bh;
2352 	brelse(bh);
2353 	return NULL;
2354 }
2355 
2356 /*
2357  * read and replay the log
2358  * on a clean unmount, the journal header's next unflushed pointer will be
2359  * to an invalid transaction.  This tests that before finding all the
2360  * transactions in the log, which makes normal mount times fast.
2361  *
2362  * After a crash, this starts with the next unflushed transaction, and
2363  * replays until it finds one too old, or invalid.
2364  *
2365  * On exit, it sets things up so the first transaction will work correctly.
2366  * NOTE: only called during fs mount
2367  */
2368 static int journal_read(struct super_block *sb)
2369 {
2370 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
2371 	struct reiserfs_journal_desc *desc;
2372 	unsigned int oldest_trans_id = 0;
2373 	unsigned int oldest_invalid_trans_id = 0;
2374 	time64_t start;
2375 	unsigned long oldest_start = 0;
2376 	unsigned long cur_dblock = 0;
2377 	unsigned long newest_mount_id = 9;
2378 	struct buffer_head *d_bh;
2379 	struct reiserfs_journal_header *jh;
2380 	int valid_journal_header = 0;
2381 	int replay_count = 0;
2382 	int continue_replay = 1;
2383 	int ret;
2384 
2385 	cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2386 	reiserfs_info(sb, "checking transaction log (%pg)\n",
2387 		      journal->j_dev_bd);
2388 	start = ktime_get_seconds();
2389 
2390 	/*
2391 	 * step 1, read in the journal header block.  Check the transaction
2392 	 * it says is the first unflushed, and if that transaction is not
2393 	 * valid, replay is done
2394 	 */
2395 	journal->j_header_bh = journal_bread(sb,
2396 					     SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2397 					     + SB_ONDISK_JOURNAL_SIZE(sb));
2398 	if (!journal->j_header_bh) {
2399 		return 1;
2400 	}
2401 	jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2402 	if (le32_to_cpu(jh->j_first_unflushed_offset) <
2403 	    SB_ONDISK_JOURNAL_SIZE(sb)
2404 	    && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2405 		oldest_start =
2406 		    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2407 		    le32_to_cpu(jh->j_first_unflushed_offset);
2408 		oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2409 		newest_mount_id = le32_to_cpu(jh->j_mount_id);
2410 		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2411 			       "journal-1153: found in "
2412 			       "header: first_unflushed_offset %d, last_flushed_trans_id "
2413 			       "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2414 			       le32_to_cpu(jh->j_last_flush_trans_id));
2415 		valid_journal_header = 1;
2416 
2417 		/*
2418 		 * now, we try to read the first unflushed offset.  If it
2419 		 * is not valid, there is nothing more we can do, and it
2420 		 * makes no sense to read through the whole log.
2421 		 */
2422 		d_bh =
2423 		    journal_bread(sb,
2424 				  SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2425 				  le32_to_cpu(jh->j_first_unflushed_offset));
2426 		ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2427 		if (!ret) {
2428 			continue_replay = 0;
2429 		}
2430 		brelse(d_bh);
2431 		goto start_log_replay;
2432 	}
2433 
2434 	/*
2435 	 * ok, there are transactions that need to be replayed.  start
2436 	 * with the first log block, find all the valid transactions, and
2437 	 * pick out the oldest.
2438 	 */
2439 	while (continue_replay
2440 	       && cur_dblock <
2441 	       (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2442 		SB_ONDISK_JOURNAL_SIZE(sb))) {
2443 		/*
2444 		 * Note that it is required for blocksize of primary fs
2445 		 * device and journal device to be the same
2446 		 */
2447 		d_bh =
2448 		    reiserfs_breada(journal->j_dev_bd, cur_dblock,
2449 				    sb->s_blocksize,
2450 				    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2451 				    SB_ONDISK_JOURNAL_SIZE(sb));
2452 		ret =
2453 		    journal_transaction_is_valid(sb, d_bh,
2454 						 &oldest_invalid_trans_id,
2455 						 &newest_mount_id);
2456 		if (ret == 1) {
2457 			desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2458 			if (oldest_start == 0) {	/* init all oldest_ values */
2459 				oldest_trans_id = get_desc_trans_id(desc);
2460 				oldest_start = d_bh->b_blocknr;
2461 				newest_mount_id = get_desc_mount_id(desc);
2462 				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2463 					       "journal-1179: Setting "
2464 					       "oldest_start to offset %llu, trans_id %lu",
2465 					       oldest_start -
2466 					       SB_ONDISK_JOURNAL_1st_BLOCK
2467 					       (sb), oldest_trans_id);
2468 			} else if (oldest_trans_id > get_desc_trans_id(desc)) {
2469 				/* one we just read was older */
2470 				oldest_trans_id = get_desc_trans_id(desc);
2471 				oldest_start = d_bh->b_blocknr;
2472 				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2473 					       "journal-1180: Resetting "
2474 					       "oldest_start to offset %lu, trans_id %lu",
2475 					       oldest_start -
2476 					       SB_ONDISK_JOURNAL_1st_BLOCK
2477 					       (sb), oldest_trans_id);
2478 			}
2479 			if (newest_mount_id < get_desc_mount_id(desc)) {
2480 				newest_mount_id = get_desc_mount_id(desc);
2481 				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2482 					       "journal-1299: Setting "
2483 					       "newest_mount_id to %d",
2484 					       get_desc_mount_id(desc));
2485 			}
2486 			cur_dblock += get_desc_trans_len(desc) + 2;
2487 		} else {
2488 			cur_dblock++;
2489 		}
2490 		brelse(d_bh);
2491 	}
2492 
2493 start_log_replay:
2494 	cur_dblock = oldest_start;
2495 	if (oldest_trans_id) {
2496 		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2497 			       "journal-1206: Starting replay "
2498 			       "from offset %llu, trans_id %lu",
2499 			       cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2500 			       oldest_trans_id);
2501 
2502 	}
2503 	replay_count = 0;
2504 	while (continue_replay && oldest_trans_id > 0) {
2505 		ret =
2506 		    journal_read_transaction(sb, cur_dblock, oldest_start,
2507 					     oldest_trans_id, newest_mount_id);
2508 		if (ret < 0) {
2509 			return ret;
2510 		} else if (ret != 0) {
2511 			break;
2512 		}
2513 		cur_dblock =
2514 		    SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2515 		replay_count++;
2516 		if (cur_dblock == oldest_start)
2517 			break;
2518 	}
2519 
2520 	if (oldest_trans_id == 0) {
2521 		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2522 			       "journal-1225: No valid " "transactions found");
2523 	}
2524 	/*
2525 	 * j_start does not get set correctly if we don't replay any
2526 	 * transactions.  if we had a valid journal_header, set j_start
2527 	 * to the first unflushed transaction value, copy the trans_id
2528 	 * from the header
2529 	 */
2530 	if (valid_journal_header && replay_count == 0) {
2531 		journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2532 		journal->j_trans_id =
2533 		    le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2534 		/* check for trans_id overflow */
2535 		if (journal->j_trans_id == 0)
2536 			journal->j_trans_id = 10;
2537 		journal->j_last_flush_trans_id =
2538 		    le32_to_cpu(jh->j_last_flush_trans_id);
2539 		journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2540 	} else {
2541 		journal->j_mount_id = newest_mount_id + 1;
2542 	}
2543 	reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2544 		       "newest_mount_id to %lu", journal->j_mount_id);
2545 	journal->j_first_unflushed_offset = journal->j_start;
2546 	if (replay_count > 0) {
2547 		reiserfs_info(sb,
2548 			      "replayed %d transactions in %lu seconds\n",
2549 			      replay_count, ktime_get_seconds() - start);
2550 	}
2551 	/* needed to satisfy the locking in _update_journal_header_block */
2552 	reiserfs_write_lock(sb);
2553 	if (!bdev_read_only(sb->s_bdev) &&
2554 	    _update_journal_header_block(sb, journal->j_start,
2555 					 journal->j_last_flush_trans_id)) {
2556 		reiserfs_write_unlock(sb);
2557 		/*
2558 		 * replay failed, caller must call free_journal_ram and abort
2559 		 * the mount
2560 		 */
2561 		return -1;
2562 	}
2563 	reiserfs_write_unlock(sb);
2564 	return 0;
2565 }
2566 
2567 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2568 {
2569 	struct reiserfs_journal_list *jl;
2570 	jl = kzalloc(sizeof(struct reiserfs_journal_list),
2571 		     GFP_NOFS | __GFP_NOFAIL);
2572 	INIT_LIST_HEAD(&jl->j_list);
2573 	INIT_LIST_HEAD(&jl->j_working_list);
2574 	INIT_LIST_HEAD(&jl->j_tail_bh_list);
2575 	INIT_LIST_HEAD(&jl->j_bh_list);
2576 	mutex_init(&jl->j_commit_mutex);
2577 	SB_JOURNAL(s)->j_num_lists++;
2578 	get_journal_list(jl);
2579 	return jl;
2580 }
2581 
2582 static void journal_list_init(struct super_block *sb)
2583 {
2584 	SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2585 }
2586 
2587 static void release_journal_dev(struct super_block *super,
2588 			       struct reiserfs_journal *journal)
2589 {
2590 	if (journal->j_dev_bd != NULL) {
2591 		blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2592 		journal->j_dev_bd = NULL;
2593 	}
2594 }
2595 
2596 static int journal_init_dev(struct super_block *super,
2597 			    struct reiserfs_journal *journal,
2598 			    const char *jdev_name)
2599 {
2600 	int result;
2601 	dev_t jdev;
2602 	fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
2603 
2604 	result = 0;
2605 
2606 	journal->j_dev_bd = NULL;
2607 	jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2608 	    new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2609 
2610 	if (bdev_read_only(super->s_bdev))
2611 		blkdev_mode = FMODE_READ;
2612 
2613 	/* there is no "jdev" option and journal is on separate device */
2614 	if ((!jdev_name || !jdev_name[0])) {
2615 		if (jdev == super->s_dev)
2616 			blkdev_mode &= ~FMODE_EXCL;
2617 		journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2618 						      journal);
2619 		journal->j_dev_mode = blkdev_mode;
2620 		if (IS_ERR(journal->j_dev_bd)) {
2621 			result = PTR_ERR(journal->j_dev_bd);
2622 			journal->j_dev_bd = NULL;
2623 			reiserfs_warning(super, "sh-458",
2624 					 "cannot init journal device unknown-block(%u,%u): %i",
2625 					 MAJOR(jdev), MINOR(jdev), result);
2626 			return result;
2627 		} else if (jdev != super->s_dev)
2628 			set_blocksize(journal->j_dev_bd, super->s_blocksize);
2629 
2630 		return 0;
2631 	}
2632 
2633 	journal->j_dev_mode = blkdev_mode;
2634 	journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
2635 	if (IS_ERR(journal->j_dev_bd)) {
2636 		result = PTR_ERR(journal->j_dev_bd);
2637 		journal->j_dev_bd = NULL;
2638 		reiserfs_warning(super, "sh-457",
2639 				 "journal_init_dev: Cannot open '%s': %i",
2640 				 jdev_name, result);
2641 		return result;
2642 	}
2643 
2644 	set_blocksize(journal->j_dev_bd, super->s_blocksize);
2645 	reiserfs_info(super,
2646 		      "journal_init_dev: journal device: %pg\n",
2647 		      journal->j_dev_bd);
2648 	return 0;
2649 }
2650 
2651 /*
2652  * When creating/tuning a file system user can assign some
2653  * journal params within boundaries which depend on the ratio
2654  * blocksize/standard_blocksize.
2655  *
2656  * For blocks >= standard_blocksize transaction size should
2657  * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2658  * then JOURNAL_TRANS_MAX_DEFAULT.
2659  *
2660  * For blocks < standard_blocksize these boundaries should be
2661  * decreased proportionally.
2662  */
2663 #define REISERFS_STANDARD_BLKSIZE (4096)
2664 
2665 static int check_advise_trans_params(struct super_block *sb,
2666 				     struct reiserfs_journal *journal)
2667 {
2668         if (journal->j_trans_max) {
2669 		/* Non-default journal params.  Do sanity check for them. */
2670 	        int ratio = 1;
2671 		if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2672 		        ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2673 
2674 		if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2675 		    journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2676 		    SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2677 		    JOURNAL_MIN_RATIO) {
2678 			reiserfs_warning(sb, "sh-462",
2679 					 "bad transaction max size (%u). "
2680 					 "FSCK?", journal->j_trans_max);
2681 			return 1;
2682 		}
2683 		if (journal->j_max_batch != (journal->j_trans_max) *
2684 		        JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2685 			reiserfs_warning(sb, "sh-463",
2686 					 "bad transaction max batch (%u). "
2687 					 "FSCK?", journal->j_max_batch);
2688 			return 1;
2689 		}
2690 	} else {
2691 		/*
2692 		 * Default journal params.
2693 		 * The file system was created by old version
2694 		 * of mkreiserfs, so some fields contain zeros,
2695 		 * and we need to advise proper values for them
2696 		 */
2697 		if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2698 			reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2699 					 sb->s_blocksize);
2700 			return 1;
2701 		}
2702 		journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2703 		journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2704 		journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2705 	}
2706 	return 0;
2707 }
2708 
2709 /* must be called once on fs mount.  calls journal_read for you */
2710 int journal_init(struct super_block *sb, const char *j_dev_name,
2711 		 int old_format, unsigned int commit_max_age)
2712 {
2713 	int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2714 	struct buffer_head *bhjh;
2715 	struct reiserfs_super_block *rs;
2716 	struct reiserfs_journal_header *jh;
2717 	struct reiserfs_journal *journal;
2718 	struct reiserfs_journal_list *jl;
2719 	int ret;
2720 
2721 	journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal));
2722 	if (!journal) {
2723 		reiserfs_warning(sb, "journal-1256",
2724 				 "unable to get memory for journal structure");
2725 		return 1;
2726 	}
2727 	INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2728 	INIT_LIST_HEAD(&journal->j_prealloc_list);
2729 	INIT_LIST_HEAD(&journal->j_working_list);
2730 	INIT_LIST_HEAD(&journal->j_journal_list);
2731 	journal->j_persistent_trans = 0;
2732 	if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2733 					   reiserfs_bmap_count(sb)))
2734 		goto free_and_return;
2735 
2736 	allocate_bitmap_nodes(sb);
2737 
2738 	/* reserved for journal area support */
2739 	SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2740 						 REISERFS_OLD_DISK_OFFSET_IN_BYTES
2741 						 / sb->s_blocksize +
2742 						 reiserfs_bmap_count(sb) +
2743 						 1 :
2744 						 REISERFS_DISK_OFFSET_IN_BYTES /
2745 						 sb->s_blocksize + 2);
2746 
2747 	/*
2748 	 * Sanity check to see is the standard journal fitting
2749 	 * within first bitmap (actual for small blocksizes)
2750 	 */
2751 	if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2752 	    (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2753 	     SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2754 		reiserfs_warning(sb, "journal-1393",
2755 				 "journal does not fit for area addressed "
2756 				 "by first of bitmap blocks. It starts at "
2757 				 "%u and its size is %u. Block size %ld",
2758 				 SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2759 				 SB_ONDISK_JOURNAL_SIZE(sb),
2760 				 sb->s_blocksize);
2761 		goto free_and_return;
2762 	}
2763 
2764 	/*
2765 	 * Sanity check to see if journal first block is correct.
2766 	 * If journal first block is invalid it can cause
2767 	 * zeroing important superblock members.
2768 	 */
2769 	if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2770 	    SB_ONDISK_JOURNAL_1st_BLOCK(sb) < SB_JOURNAL_1st_RESERVED_BLOCK(sb)) {
2771 		reiserfs_warning(sb, "journal-1393",
2772 				 "journal 1st super block is invalid: 1st reserved block %d, but actual 1st block is %d",
2773 				 SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2774 				 SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2775 		goto free_and_return;
2776 	}
2777 
2778 	if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2779 		reiserfs_warning(sb, "sh-462",
2780 				 "unable to initialize journal device");
2781 		goto free_and_return;
2782 	}
2783 
2784 	rs = SB_DISK_SUPER_BLOCK(sb);
2785 
2786 	/* read journal header */
2787 	bhjh = journal_bread(sb,
2788 			     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2789 			     SB_ONDISK_JOURNAL_SIZE(sb));
2790 	if (!bhjh) {
2791 		reiserfs_warning(sb, "sh-459",
2792 				 "unable to read journal header");
2793 		goto free_and_return;
2794 	}
2795 	jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2796 
2797 	/* make sure that journal matches to the super block */
2798 	if (is_reiserfs_jr(rs)
2799 	    && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2800 		sb_jp_journal_magic(rs))) {
2801 		reiserfs_warning(sb, "sh-460",
2802 				 "journal header magic %x (device %pg) does "
2803 				 "not match to magic found in super block %x",
2804 				 jh->jh_journal.jp_journal_magic,
2805 				 journal->j_dev_bd,
2806 				 sb_jp_journal_magic(rs));
2807 		brelse(bhjh);
2808 		goto free_and_return;
2809 	}
2810 
2811 	journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2812 	journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2813 	journal->j_max_commit_age =
2814 	    le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2815 	journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2816 
2817 	if (check_advise_trans_params(sb, journal) != 0)
2818 	        goto free_and_return;
2819 	journal->j_default_max_commit_age = journal->j_max_commit_age;
2820 
2821 	if (commit_max_age != 0) {
2822 		journal->j_max_commit_age = commit_max_age;
2823 		journal->j_max_trans_age = commit_max_age;
2824 	}
2825 
2826 	reiserfs_info(sb, "journal params: device %pg, size %u, "
2827 		      "journal first block %u, max trans len %u, max batch %u, "
2828 		      "max commit age %u, max trans age %u\n",
2829 		      journal->j_dev_bd,
2830 		      SB_ONDISK_JOURNAL_SIZE(sb),
2831 		      SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2832 		      journal->j_trans_max,
2833 		      journal->j_max_batch,
2834 		      journal->j_max_commit_age, journal->j_max_trans_age);
2835 
2836 	brelse(bhjh);
2837 
2838 	journal->j_list_bitmap_index = 0;
2839 	journal_list_init(sb);
2840 
2841 	memset(journal->j_list_hash_table, 0,
2842 	       JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2843 
2844 	INIT_LIST_HEAD(&journal->j_dirty_buffers);
2845 	spin_lock_init(&journal->j_dirty_buffers_lock);
2846 
2847 	journal->j_start = 0;
2848 	journal->j_len = 0;
2849 	journal->j_len_alloc = 0;
2850 	atomic_set(&journal->j_wcount, 0);
2851 	atomic_set(&journal->j_async_throttle, 0);
2852 	journal->j_bcount = 0;
2853 	journal->j_trans_start_time = 0;
2854 	journal->j_last = NULL;
2855 	journal->j_first = NULL;
2856 	init_waitqueue_head(&journal->j_join_wait);
2857 	mutex_init(&journal->j_mutex);
2858 	mutex_init(&journal->j_flush_mutex);
2859 
2860 	journal->j_trans_id = 10;
2861 	journal->j_mount_id = 10;
2862 	journal->j_state = 0;
2863 	atomic_set(&journal->j_jlock, 0);
2864 	journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2865 	journal->j_cnode_free_orig = journal->j_cnode_free_list;
2866 	journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2867 	journal->j_cnode_used = 0;
2868 	journal->j_must_wait = 0;
2869 
2870 	if (journal->j_cnode_free == 0) {
2871 		reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2872 		                 "allocation failed (%ld bytes). Journal is "
2873 		                 "too large for available memory. Usually "
2874 		                 "this is due to a journal that is too large.",
2875 		                 sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2876         	goto free_and_return;
2877 	}
2878 
2879 	init_journal_hash(sb);
2880 	jl = journal->j_current_jl;
2881 
2882 	/*
2883 	 * get_list_bitmap() may call flush_commit_list() which
2884 	 * requires the lock. Calling flush_commit_list() shouldn't happen
2885 	 * this early but I like to be paranoid.
2886 	 */
2887 	reiserfs_write_lock(sb);
2888 	jl->j_list_bitmap = get_list_bitmap(sb, jl);
2889 	reiserfs_write_unlock(sb);
2890 	if (!jl->j_list_bitmap) {
2891 		reiserfs_warning(sb, "journal-2005",
2892 				 "get_list_bitmap failed for journal list 0");
2893 		goto free_and_return;
2894 	}
2895 
2896 	ret = journal_read(sb);
2897 	if (ret < 0) {
2898 		reiserfs_warning(sb, "reiserfs-2006",
2899 				 "Replay Failure, unable to mount");
2900 		goto free_and_return;
2901 	}
2902 
2903 	INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2904 	journal->j_work_sb = sb;
2905 	return 0;
2906 free_and_return:
2907 	free_journal_ram(sb);
2908 	return 1;
2909 }
2910 
2911 /*
2912  * test for a polite end of the current transaction.  Used by file_write,
2913  * and should be used by delete to make sure they don't write more than
2914  * can fit inside a single transaction
2915  */
2916 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2917 				   int new_alloc)
2918 {
2919 	struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2920 	time64_t now = ktime_get_seconds();
2921 	/* cannot restart while nested */
2922 	BUG_ON(!th->t_trans_id);
2923 	if (th->t_refcount > 1)
2924 		return 0;
2925 	if (journal->j_must_wait > 0 ||
2926 	    (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2927 	    atomic_read(&journal->j_jlock) ||
2928 	    (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2929 	    journal->j_cnode_free < (journal->j_trans_max * 3)) {
2930 		return 1;
2931 	}
2932 
2933 	journal->j_len_alloc += new_alloc;
2934 	th->t_blocks_allocated += new_alloc ;
2935 	return 0;
2936 }
2937 
2938 /* this must be called inside a transaction */
2939 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2940 {
2941 	struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2942 	BUG_ON(!th->t_trans_id);
2943 	journal->j_must_wait = 1;
2944 	set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2945 	return;
2946 }
2947 
2948 /* this must be called without a transaction started */
2949 void reiserfs_allow_writes(struct super_block *s)
2950 {
2951 	struct reiserfs_journal *journal = SB_JOURNAL(s);
2952 	clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2953 	wake_up(&journal->j_join_wait);
2954 }
2955 
2956 /* this must be called without a transaction started */
2957 void reiserfs_wait_on_write_block(struct super_block *s)
2958 {
2959 	struct reiserfs_journal *journal = SB_JOURNAL(s);
2960 	wait_event(journal->j_join_wait,
2961 		   !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2962 }
2963 
2964 static void queue_log_writer(struct super_block *s)
2965 {
2966 	wait_queue_entry_t wait;
2967 	struct reiserfs_journal *journal = SB_JOURNAL(s);
2968 	set_bit(J_WRITERS_QUEUED, &journal->j_state);
2969 
2970 	/*
2971 	 * we don't want to use wait_event here because
2972 	 * we only want to wait once.
2973 	 */
2974 	init_waitqueue_entry(&wait, current);
2975 	add_wait_queue(&journal->j_join_wait, &wait);
2976 	set_current_state(TASK_UNINTERRUPTIBLE);
2977 	if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
2978 		int depth = reiserfs_write_unlock_nested(s);
2979 		schedule();
2980 		reiserfs_write_lock_nested(s, depth);
2981 	}
2982 	__set_current_state(TASK_RUNNING);
2983 	remove_wait_queue(&journal->j_join_wait, &wait);
2984 }
2985 
2986 static void wake_queued_writers(struct super_block *s)
2987 {
2988 	struct reiserfs_journal *journal = SB_JOURNAL(s);
2989 	if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2990 		wake_up(&journal->j_join_wait);
2991 }
2992 
2993 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
2994 {
2995 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
2996 	unsigned long bcount = journal->j_bcount;
2997 	while (1) {
2998 		int depth;
2999 
3000 		depth = reiserfs_write_unlock_nested(sb);
3001 		schedule_timeout_uninterruptible(1);
3002 		reiserfs_write_lock_nested(sb, depth);
3003 
3004 		journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
3005 		while ((atomic_read(&journal->j_wcount) > 0 ||
3006 			atomic_read(&journal->j_jlock)) &&
3007 		       journal->j_trans_id == trans_id) {
3008 			queue_log_writer(sb);
3009 		}
3010 		if (journal->j_trans_id != trans_id)
3011 			break;
3012 		if (bcount == journal->j_bcount)
3013 			break;
3014 		bcount = journal->j_bcount;
3015 	}
3016 }
3017 
3018 /*
3019  * join == true if you must join an existing transaction.
3020  * join == false if you can deal with waiting for others to finish
3021  *
3022  * this will block until the transaction is joinable.  send the number of
3023  * blocks you expect to use in nblocks.
3024 */
3025 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3026 			      struct super_block *sb, unsigned long nblocks,
3027 			      int join)
3028 {
3029 	time64_t now = ktime_get_seconds();
3030 	unsigned int old_trans_id;
3031 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3032 	struct reiserfs_transaction_handle myth;
3033 	int sched_count = 0;
3034 	int retval;
3035 	int depth;
3036 
3037 	reiserfs_check_lock_depth(sb, "journal_begin");
3038 	BUG_ON(nblocks > journal->j_trans_max);
3039 
3040 	PROC_INFO_INC(sb, journal.journal_being);
3041 	/* set here for journal_join */
3042 	th->t_refcount = 1;
3043 	th->t_super = sb;
3044 
3045 relock:
3046 	lock_journal(sb);
3047 	if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3048 		unlock_journal(sb);
3049 		retval = journal->j_errno;
3050 		goto out_fail;
3051 	}
3052 	journal->j_bcount++;
3053 
3054 	if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3055 		unlock_journal(sb);
3056 		depth = reiserfs_write_unlock_nested(sb);
3057 		reiserfs_wait_on_write_block(sb);
3058 		reiserfs_write_lock_nested(sb, depth);
3059 		PROC_INFO_INC(sb, journal.journal_relock_writers);
3060 		goto relock;
3061 	}
3062 	now = ktime_get_seconds();
3063 
3064 	/*
3065 	 * if there is no room in the journal OR
3066 	 * if this transaction is too old, and we weren't called joinable,
3067 	 * wait for it to finish before beginning we don't sleep if there
3068 	 * aren't other writers
3069 	 */
3070 
3071 	if ((!join && journal->j_must_wait > 0) ||
3072 	    (!join
3073 	     && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3074 	    || (!join && atomic_read(&journal->j_wcount) > 0
3075 		&& journal->j_trans_start_time > 0
3076 		&& (now - journal->j_trans_start_time) >
3077 		journal->j_max_trans_age) || (!join
3078 					      && atomic_read(&journal->j_jlock))
3079 	    || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3080 
3081 		old_trans_id = journal->j_trans_id;
3082 		/* allow others to finish this transaction */
3083 		unlock_journal(sb);
3084 
3085 		if (!join && (journal->j_len_alloc + nblocks + 2) >=
3086 		    journal->j_max_batch &&
3087 		    ((journal->j_len + nblocks + 2) * 100) <
3088 		    (journal->j_len_alloc * 75)) {
3089 			if (atomic_read(&journal->j_wcount) > 10) {
3090 				sched_count++;
3091 				queue_log_writer(sb);
3092 				goto relock;
3093 			}
3094 		}
3095 		/*
3096 		 * don't mess with joining the transaction if all we
3097 		 * have to do is wait for someone else to do a commit
3098 		 */
3099 		if (atomic_read(&journal->j_jlock)) {
3100 			while (journal->j_trans_id == old_trans_id &&
3101 			       atomic_read(&journal->j_jlock)) {
3102 				queue_log_writer(sb);
3103 			}
3104 			goto relock;
3105 		}
3106 		retval = journal_join(&myth, sb);
3107 		if (retval)
3108 			goto out_fail;
3109 
3110 		/* someone might have ended the transaction while we joined */
3111 		if (old_trans_id != journal->j_trans_id) {
3112 			retval = do_journal_end(&myth, 0);
3113 		} else {
3114 			retval = do_journal_end(&myth, COMMIT_NOW);
3115 		}
3116 
3117 		if (retval)
3118 			goto out_fail;
3119 
3120 		PROC_INFO_INC(sb, journal.journal_relock_wcount);
3121 		goto relock;
3122 	}
3123 	/* we are the first writer, set trans_id */
3124 	if (journal->j_trans_start_time == 0) {
3125 		journal->j_trans_start_time = ktime_get_seconds();
3126 	}
3127 	atomic_inc(&journal->j_wcount);
3128 	journal->j_len_alloc += nblocks;
3129 	th->t_blocks_logged = 0;
3130 	th->t_blocks_allocated = nblocks;
3131 	th->t_trans_id = journal->j_trans_id;
3132 	unlock_journal(sb);
3133 	INIT_LIST_HEAD(&th->t_list);
3134 	return 0;
3135 
3136 out_fail:
3137 	memset(th, 0, sizeof(*th));
3138 	/*
3139 	 * Re-set th->t_super, so we can properly keep track of how many
3140 	 * persistent transactions there are. We need to do this so if this
3141 	 * call is part of a failed restart_transaction, we can free it later
3142 	 */
3143 	th->t_super = sb;
3144 	return retval;
3145 }
3146 
3147 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3148 								    super_block
3149 								    *s,
3150 								    int nblocks)
3151 {
3152 	int ret;
3153 	struct reiserfs_transaction_handle *th;
3154 
3155 	/*
3156 	 * if we're nesting into an existing transaction.  It will be
3157 	 * persistent on its own
3158 	 */
3159 	if (reiserfs_transaction_running(s)) {
3160 		th = current->journal_info;
3161 		th->t_refcount++;
3162 		BUG_ON(th->t_refcount < 2);
3163 
3164 		return th;
3165 	}
3166 	th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3167 	if (!th)
3168 		return NULL;
3169 	ret = journal_begin(th, s, nblocks);
3170 	if (ret) {
3171 		kfree(th);
3172 		return NULL;
3173 	}
3174 
3175 	SB_JOURNAL(s)->j_persistent_trans++;
3176 	return th;
3177 }
3178 
3179 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3180 {
3181 	struct super_block *s = th->t_super;
3182 	int ret = 0;
3183 	if (th->t_trans_id)
3184 		ret = journal_end(th);
3185 	else
3186 		ret = -EIO;
3187 	if (th->t_refcount == 0) {
3188 		SB_JOURNAL(s)->j_persistent_trans--;
3189 		kfree(th);
3190 	}
3191 	return ret;
3192 }
3193 
3194 static int journal_join(struct reiserfs_transaction_handle *th,
3195 			struct super_block *sb)
3196 {
3197 	struct reiserfs_transaction_handle *cur_th = current->journal_info;
3198 
3199 	/*
3200 	 * this keeps do_journal_end from NULLing out the
3201 	 * current->journal_info pointer
3202 	 */
3203 	th->t_handle_save = cur_th;
3204 	BUG_ON(cur_th && cur_th->t_refcount > 1);
3205 	return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN);
3206 }
3207 
3208 int journal_join_abort(struct reiserfs_transaction_handle *th,
3209 		       struct super_block *sb)
3210 {
3211 	struct reiserfs_transaction_handle *cur_th = current->journal_info;
3212 
3213 	/*
3214 	 * this keeps do_journal_end from NULLing out the
3215 	 * current->journal_info pointer
3216 	 */
3217 	th->t_handle_save = cur_th;
3218 	BUG_ON(cur_th && cur_th->t_refcount > 1);
3219 	return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT);
3220 }
3221 
3222 int journal_begin(struct reiserfs_transaction_handle *th,
3223 		  struct super_block *sb, unsigned long nblocks)
3224 {
3225 	struct reiserfs_transaction_handle *cur_th = current->journal_info;
3226 	int ret;
3227 
3228 	th->t_handle_save = NULL;
3229 	if (cur_th) {
3230 		/* we are nesting into the current transaction */
3231 		if (cur_th->t_super == sb) {
3232 			BUG_ON(!cur_th->t_refcount);
3233 			cur_th->t_refcount++;
3234 			memcpy(th, cur_th, sizeof(*th));
3235 			if (th->t_refcount <= 1)
3236 				reiserfs_warning(sb, "reiserfs-2005",
3237 						 "BAD: refcount <= 1, but "
3238 						 "journal_info != 0");
3239 			return 0;
3240 		} else {
3241 			/*
3242 			 * we've ended up with a handle from a different
3243 			 * filesystem.  save it and restore on journal_end.
3244 			 * This should never really happen...
3245 			 */
3246 			reiserfs_warning(sb, "clm-2100",
3247 					 "nesting info a different FS");
3248 			th->t_handle_save = current->journal_info;
3249 			current->journal_info = th;
3250 		}
3251 	} else {
3252 		current->journal_info = th;
3253 	}
3254 	ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3255 	BUG_ON(current->journal_info != th);
3256 
3257 	/*
3258 	 * I guess this boils down to being the reciprocal of clm-2100 above.
3259 	 * If do_journal_begin_r fails, we need to put it back, since
3260 	 * journal_end won't be called to do it. */
3261 	if (ret)
3262 		current->journal_info = th->t_handle_save;
3263 	else
3264 		BUG_ON(!th->t_refcount);
3265 
3266 	return ret;
3267 }
3268 
3269 /*
3270  * puts bh into the current transaction.  If it was already there, reorders
3271  * removes the old pointers from the hash, and puts new ones in (to make
3272  * sure replay happen in the right order).
3273  *
3274  * if it was dirty, cleans and files onto the clean list.  I can't let it
3275  * be dirty again until the transaction is committed.
3276  *
3277  * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3278  */
3279 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3280 		       struct buffer_head *bh)
3281 {
3282 	struct super_block *sb = th->t_super;
3283 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3284 	struct reiserfs_journal_cnode *cn = NULL;
3285 	int count_already_incd = 0;
3286 	int prepared = 0;
3287 	BUG_ON(!th->t_trans_id);
3288 
3289 	PROC_INFO_INC(sb, journal.mark_dirty);
3290 	if (th->t_trans_id != journal->j_trans_id) {
3291 		reiserfs_panic(th->t_super, "journal-1577",
3292 			       "handle trans id %ld != current trans id %ld",
3293 			       th->t_trans_id, journal->j_trans_id);
3294 	}
3295 
3296 	prepared = test_clear_buffer_journal_prepared(bh);
3297 	clear_buffer_journal_restore_dirty(bh);
3298 	/* already in this transaction, we are done */
3299 	if (buffer_journaled(bh)) {
3300 		PROC_INFO_INC(sb, journal.mark_dirty_already);
3301 		return 0;
3302 	}
3303 
3304 	/*
3305 	 * this must be turned into a panic instead of a warning.  We can't
3306 	 * allow a dirty or journal_dirty or locked buffer to be logged, as
3307 	 * some changes could get to disk too early.  NOT GOOD.
3308 	 */
3309 	if (!prepared || buffer_dirty(bh)) {
3310 		reiserfs_warning(sb, "journal-1777",
3311 				 "buffer %llu bad state "
3312 				 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3313 				 (unsigned long long)bh->b_blocknr,
3314 				 prepared ? ' ' : '!',
3315 				 buffer_locked(bh) ? ' ' : '!',
3316 				 buffer_dirty(bh) ? ' ' : '!',
3317 				 buffer_journal_dirty(bh) ? ' ' : '!');
3318 	}
3319 
3320 	if (atomic_read(&journal->j_wcount) <= 0) {
3321 		reiserfs_warning(sb, "journal-1409",
3322 				 "returning because j_wcount was %d",
3323 				 atomic_read(&journal->j_wcount));
3324 		return 1;
3325 	}
3326 	/*
3327 	 * this error means I've screwed up, and we've overflowed
3328 	 * the transaction.  Nothing can be done here, except make the
3329 	 * FS readonly or panic.
3330 	 */
3331 	if (journal->j_len >= journal->j_trans_max) {
3332 		reiserfs_panic(th->t_super, "journal-1413",
3333 			       "j_len (%lu) is too big",
3334 			       journal->j_len);
3335 	}
3336 
3337 	if (buffer_journal_dirty(bh)) {
3338 		count_already_incd = 1;
3339 		PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3340 		clear_buffer_journal_dirty(bh);
3341 	}
3342 
3343 	if (journal->j_len > journal->j_len_alloc) {
3344 		journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3345 	}
3346 
3347 	set_buffer_journaled(bh);
3348 
3349 	/* now put this guy on the end */
3350 	if (!cn) {
3351 		cn = get_cnode(sb);
3352 		if (!cn) {
3353 			reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3354 		}
3355 
3356 		if (th->t_blocks_logged == th->t_blocks_allocated) {
3357 			th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3358 			journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3359 		}
3360 		th->t_blocks_logged++;
3361 		journal->j_len++;
3362 
3363 		cn->bh = bh;
3364 		cn->blocknr = bh->b_blocknr;
3365 		cn->sb = sb;
3366 		cn->jlist = NULL;
3367 		insert_journal_hash(journal->j_hash_table, cn);
3368 		if (!count_already_incd) {
3369 			get_bh(bh);
3370 		}
3371 	}
3372 	cn->next = NULL;
3373 	cn->prev = journal->j_last;
3374 	cn->bh = bh;
3375 	if (journal->j_last) {
3376 		journal->j_last->next = cn;
3377 		journal->j_last = cn;
3378 	} else {
3379 		journal->j_first = cn;
3380 		journal->j_last = cn;
3381 	}
3382 	reiserfs_schedule_old_flush(sb);
3383 	return 0;
3384 }
3385 
3386 int journal_end(struct reiserfs_transaction_handle *th)
3387 {
3388 	struct super_block *sb = th->t_super;
3389 	if (!current->journal_info && th->t_refcount > 1)
3390 		reiserfs_warning(sb, "REISER-NESTING",
3391 				 "th NULL, refcount %d", th->t_refcount);
3392 
3393 	if (!th->t_trans_id) {
3394 		WARN_ON(1);
3395 		return -EIO;
3396 	}
3397 
3398 	th->t_refcount--;
3399 	if (th->t_refcount > 0) {
3400 		struct reiserfs_transaction_handle *cur_th =
3401 		    current->journal_info;
3402 
3403 		/*
3404 		 * we aren't allowed to close a nested transaction on a
3405 		 * different filesystem from the one in the task struct
3406 		 */
3407 		BUG_ON(cur_th->t_super != th->t_super);
3408 
3409 		if (th != cur_th) {
3410 			memcpy(current->journal_info, th, sizeof(*th));
3411 			th->t_trans_id = 0;
3412 		}
3413 		return 0;
3414 	} else {
3415 		return do_journal_end(th, 0);
3416 	}
3417 }
3418 
3419 /*
3420  * removes from the current transaction, relsing and descrementing any counters.
3421  * also files the removed buffer directly onto the clean list
3422  *
3423  * called by journal_mark_freed when a block has been deleted
3424  *
3425  * returns 1 if it cleaned and relsed the buffer. 0 otherwise
3426  */
3427 static int remove_from_transaction(struct super_block *sb,
3428 				   b_blocknr_t blocknr, int already_cleaned)
3429 {
3430 	struct buffer_head *bh;
3431 	struct reiserfs_journal_cnode *cn;
3432 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3433 	int ret = 0;
3434 
3435 	cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3436 	if (!cn || !cn->bh) {
3437 		return ret;
3438 	}
3439 	bh = cn->bh;
3440 	if (cn->prev) {
3441 		cn->prev->next = cn->next;
3442 	}
3443 	if (cn->next) {
3444 		cn->next->prev = cn->prev;
3445 	}
3446 	if (cn == journal->j_first) {
3447 		journal->j_first = cn->next;
3448 	}
3449 	if (cn == journal->j_last) {
3450 		journal->j_last = cn->prev;
3451 	}
3452 	remove_journal_hash(sb, journal->j_hash_table, NULL,
3453 			    bh->b_blocknr, 0);
3454 	clear_buffer_journaled(bh);	/* don't log this one */
3455 
3456 	if (!already_cleaned) {
3457 		clear_buffer_journal_dirty(bh);
3458 		clear_buffer_dirty(bh);
3459 		clear_buffer_journal_test(bh);
3460 		put_bh(bh);
3461 		if (atomic_read(&bh->b_count) < 0) {
3462 			reiserfs_warning(sb, "journal-1752",
3463 					 "b_count < 0");
3464 		}
3465 		ret = 1;
3466 	}
3467 	journal->j_len--;
3468 	journal->j_len_alloc--;
3469 	free_cnode(sb, cn);
3470 	return ret;
3471 }
3472 
3473 /*
3474  * for any cnode in a journal list, it can only be dirtied of all the
3475  * transactions that include it are committed to disk.
3476  * this checks through each transaction, and returns 1 if you are allowed
3477  * to dirty, and 0 if you aren't
3478  *
3479  * it is called by dirty_journal_list, which is called after
3480  * flush_commit_list has gotten all the log blocks for a given
3481  * transaction on disk
3482  *
3483  */
3484 static int can_dirty(struct reiserfs_journal_cnode *cn)
3485 {
3486 	struct super_block *sb = cn->sb;
3487 	b_blocknr_t blocknr = cn->blocknr;
3488 	struct reiserfs_journal_cnode *cur = cn->hprev;
3489 	int can_dirty = 1;
3490 
3491 	/*
3492 	 * first test hprev.  These are all newer than cn, so any node here
3493 	 * with the same block number and dev means this node can't be sent
3494 	 * to disk right now.
3495 	 */
3496 	while (cur && can_dirty) {
3497 		if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3498 		    cur->blocknr == blocknr) {
3499 			can_dirty = 0;
3500 		}
3501 		cur = cur->hprev;
3502 	}
3503 	/*
3504 	 * then test hnext.  These are all older than cn.  As long as they
3505 	 * are committed to the log, it is safe to write cn to disk
3506 	 */
3507 	cur = cn->hnext;
3508 	while (cur && can_dirty) {
3509 		if (cur->jlist && cur->jlist->j_len > 0 &&
3510 		    atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh &&
3511 		    cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3512 			can_dirty = 0;
3513 		}
3514 		cur = cur->hnext;
3515 	}
3516 	return can_dirty;
3517 }
3518 
3519 /*
3520  * syncs the commit blocks, but does not force the real buffers to disk
3521  * will wait until the current transaction is done/committed before returning
3522  */
3523 int journal_end_sync(struct reiserfs_transaction_handle *th)
3524 {
3525 	struct super_block *sb = th->t_super;
3526 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3527 
3528 	BUG_ON(!th->t_trans_id);
3529 	/* you can sync while nested, very, very bad */
3530 	BUG_ON(th->t_refcount > 1);
3531 	if (journal->j_len == 0) {
3532 		reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3533 					     1);
3534 		journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
3535 	}
3536 	return do_journal_end(th, COMMIT_NOW | WAIT);
3537 }
3538 
3539 /* writeback the pending async commits to disk */
3540 static void flush_async_commits(struct work_struct *work)
3541 {
3542 	struct reiserfs_journal *journal =
3543 		container_of(work, struct reiserfs_journal, j_work.work);
3544 	struct super_block *sb = journal->j_work_sb;
3545 	struct reiserfs_journal_list *jl;
3546 	struct list_head *entry;
3547 
3548 	reiserfs_write_lock(sb);
3549 	if (!list_empty(&journal->j_journal_list)) {
3550 		/* last entry is the youngest, commit it and you get everything */
3551 		entry = journal->j_journal_list.prev;
3552 		jl = JOURNAL_LIST_ENTRY(entry);
3553 		flush_commit_list(sb, jl, 1);
3554 	}
3555 	reiserfs_write_unlock(sb);
3556 }
3557 
3558 /*
3559  * flushes any old transactions to disk
3560  * ends the current transaction if it is too old
3561  */
3562 void reiserfs_flush_old_commits(struct super_block *sb)
3563 {
3564 	time64_t now;
3565 	struct reiserfs_transaction_handle th;
3566 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3567 
3568 	now = ktime_get_seconds();
3569 	/*
3570 	 * safety check so we don't flush while we are replaying the log during
3571 	 * mount
3572 	 */
3573 	if (list_empty(&journal->j_journal_list))
3574 		return;
3575 
3576 	/*
3577 	 * check the current transaction.  If there are no writers, and it is
3578 	 * too old, finish it, and force the commit blocks to disk
3579 	 */
3580 	if (atomic_read(&journal->j_wcount) <= 0 &&
3581 	    journal->j_trans_start_time > 0 &&
3582 	    journal->j_len > 0 &&
3583 	    (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3584 		if (!journal_join(&th, sb)) {
3585 			reiserfs_prepare_for_journal(sb,
3586 						     SB_BUFFER_WITH_SB(sb),
3587 						     1);
3588 			journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3589 
3590 			/*
3591 			 * we're only being called from kreiserfsd, it makes
3592 			 * no sense to do an async commit so that kreiserfsd
3593 			 * can do it later
3594 			 */
3595 			do_journal_end(&th, COMMIT_NOW | WAIT);
3596 		}
3597 	}
3598 }
3599 
3600 /*
3601  * returns 0 if do_journal_end should return right away, returns 1 if
3602  * do_journal_end should finish the commit
3603  *
3604  * if the current transaction is too old, but still has writers, this will
3605  * wait on j_join_wait until all the writers are done.  By the time it
3606  * wakes up, the transaction it was called has already ended, so it just
3607  * flushes the commit list and returns 0.
3608  *
3609  * Won't batch when flush or commit_now is set.  Also won't batch when
3610  * others are waiting on j_join_wait.
3611  *
3612  * Note, we can't allow the journal_end to proceed while there are still
3613  * writers in the log.
3614  */
3615 static int check_journal_end(struct reiserfs_transaction_handle *th, int flags)
3616 {
3617 
3618 	time64_t now;
3619 	int flush = flags & FLUSH_ALL;
3620 	int commit_now = flags & COMMIT_NOW;
3621 	int wait_on_commit = flags & WAIT;
3622 	struct reiserfs_journal_list *jl;
3623 	struct super_block *sb = th->t_super;
3624 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3625 
3626 	BUG_ON(!th->t_trans_id);
3627 
3628 	if (th->t_trans_id != journal->j_trans_id) {
3629 		reiserfs_panic(th->t_super, "journal-1577",
3630 			       "handle trans id %ld != current trans id %ld",
3631 			       th->t_trans_id, journal->j_trans_id);
3632 	}
3633 
3634 	journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3635 	/* <= 0 is allowed.  unmounting might not call begin */
3636 	if (atomic_read(&journal->j_wcount) > 0)
3637 		atomic_dec(&journal->j_wcount);
3638 
3639 	/*
3640 	 * BUG, deal with case where j_len is 0, but people previously
3641 	 * freed blocks need to be released will be dealt with by next
3642 	 * transaction that actually writes something, but should be taken
3643 	 * care of in this trans
3644 	 */
3645 	BUG_ON(journal->j_len == 0);
3646 
3647 	/*
3648 	 * if wcount > 0, and we are called to with flush or commit_now,
3649 	 * we wait on j_join_wait.  We will wake up when the last writer has
3650 	 * finished the transaction, and started it on its way to the disk.
3651 	 * Then, we flush the commit or journal list, and just return 0
3652 	 * because the rest of journal end was already done for this
3653 	 * transaction.
3654 	 */
3655 	if (atomic_read(&journal->j_wcount) > 0) {
3656 		if (flush || commit_now) {
3657 			unsigned trans_id;
3658 
3659 			jl = journal->j_current_jl;
3660 			trans_id = jl->j_trans_id;
3661 			if (wait_on_commit)
3662 				jl->j_state |= LIST_COMMIT_PENDING;
3663 			atomic_set(&journal->j_jlock, 1);
3664 			if (flush) {
3665 				journal->j_next_full_flush = 1;
3666 			}
3667 			unlock_journal(sb);
3668 
3669 			/*
3670 			 * sleep while the current transaction is
3671 			 * still j_jlocked
3672 			 */
3673 			while (journal->j_trans_id == trans_id) {
3674 				if (atomic_read(&journal->j_jlock)) {
3675 					queue_log_writer(sb);
3676 				} else {
3677 					lock_journal(sb);
3678 					if (journal->j_trans_id == trans_id) {
3679 						atomic_set(&journal->j_jlock,
3680 							   1);
3681 					}
3682 					unlock_journal(sb);
3683 				}
3684 			}
3685 			BUG_ON(journal->j_trans_id == trans_id);
3686 
3687 			if (commit_now
3688 			    && journal_list_still_alive(sb, trans_id)
3689 			    && wait_on_commit) {
3690 				flush_commit_list(sb, jl, 1);
3691 			}
3692 			return 0;
3693 		}
3694 		unlock_journal(sb);
3695 		return 0;
3696 	}
3697 
3698 	/* deal with old transactions where we are the last writers */
3699 	now = ktime_get_seconds();
3700 	if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3701 		commit_now = 1;
3702 		journal->j_next_async_flush = 1;
3703 	}
3704 	/* don't batch when someone is waiting on j_join_wait */
3705 	/* don't batch when syncing the commit or flushing the whole trans */
3706 	if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock))
3707 	    && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3708 	    && journal->j_len_alloc < journal->j_max_batch
3709 	    && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3710 		journal->j_bcount++;
3711 		unlock_journal(sb);
3712 		return 0;
3713 	}
3714 
3715 	if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3716 		reiserfs_panic(sb, "journal-003",
3717 			       "j_start (%ld) is too high",
3718 			       journal->j_start);
3719 	}
3720 	return 1;
3721 }
3722 
3723 /*
3724  * Does all the work that makes deleting blocks safe.
3725  * when deleting a block mark BH_JNew, just remove it from the current
3726  * transaction, clean it's buffer_head and move on.
3727  *
3728  * otherwise:
3729  * set a bit for the block in the journal bitmap.  That will prevent it from
3730  * being allocated for unformatted nodes before this transaction has finished.
3731  *
3732  * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.
3733  * That will prevent any old transactions with this block from trying to flush
3734  * to the real location.  Since we aren't removing the cnode from the
3735  * journal_list_hash, *the block can't be reallocated yet.
3736  *
3737  * Then remove it from the current transaction, decrementing any counters and
3738  * filing it on the clean list.
3739  */
3740 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3741 		       struct super_block *sb, b_blocknr_t blocknr)
3742 {
3743 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3744 	struct reiserfs_journal_cnode *cn = NULL;
3745 	struct buffer_head *bh = NULL;
3746 	struct reiserfs_list_bitmap *jb = NULL;
3747 	int cleaned = 0;
3748 	BUG_ON(!th->t_trans_id);
3749 
3750 	cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3751 	if (cn && cn->bh) {
3752 		bh = cn->bh;
3753 		get_bh(bh);
3754 	}
3755 	/* if it is journal new, we just remove it from this transaction */
3756 	if (bh && buffer_journal_new(bh)) {
3757 		clear_buffer_journal_new(bh);
3758 		clear_prepared_bits(bh);
3759 		reiserfs_clean_and_file_buffer(bh);
3760 		cleaned = remove_from_transaction(sb, blocknr, cleaned);
3761 	} else {
3762 		/*
3763 		 * set the bit for this block in the journal bitmap
3764 		 * for this transaction
3765 		 */
3766 		jb = journal->j_current_jl->j_list_bitmap;
3767 		if (!jb) {
3768 			reiserfs_panic(sb, "journal-1702",
3769 				       "journal_list_bitmap is NULL");
3770 		}
3771 		set_bit_in_list_bitmap(sb, blocknr, jb);
3772 
3773 		/* Note, the entire while loop is not allowed to schedule.  */
3774 
3775 		if (bh) {
3776 			clear_prepared_bits(bh);
3777 			reiserfs_clean_and_file_buffer(bh);
3778 		}
3779 		cleaned = remove_from_transaction(sb, blocknr, cleaned);
3780 
3781 		/*
3782 		 * find all older transactions with this block,
3783 		 * make sure they don't try to write it out
3784 		 */
3785 		cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3786 					  blocknr);
3787 		while (cn) {
3788 			if (sb == cn->sb && blocknr == cn->blocknr) {
3789 				set_bit(BLOCK_FREED, &cn->state);
3790 				if (cn->bh) {
3791 					/*
3792 					 * remove_from_transaction will brelse
3793 					 * the buffer if it was in the current
3794 					 * trans
3795 					 */
3796 					if (!cleaned) {
3797 						clear_buffer_journal_dirty(cn->
3798 									   bh);
3799 						clear_buffer_dirty(cn->bh);
3800 						clear_buffer_journal_test(cn->
3801 									  bh);
3802 						cleaned = 1;
3803 						put_bh(cn->bh);
3804 						if (atomic_read
3805 						    (&cn->bh->b_count) < 0) {
3806 							reiserfs_warning(sb,
3807 								 "journal-2138",
3808 								 "cn->bh->b_count < 0");
3809 						}
3810 					}
3811 					/*
3812 					 * since we are clearing the bh,
3813 					 * we MUST dec nonzerolen
3814 					 */
3815 					if (cn->jlist) {
3816 						atomic_dec(&cn->jlist->
3817 							   j_nonzerolen);
3818 					}
3819 					cn->bh = NULL;
3820 				}
3821 			}
3822 			cn = cn->hnext;
3823 		}
3824 	}
3825 
3826 	if (bh)
3827 		release_buffer_page(bh); /* get_hash grabs the buffer */
3828 	return 0;
3829 }
3830 
3831 void reiserfs_update_inode_transaction(struct inode *inode)
3832 {
3833 	struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3834 	REISERFS_I(inode)->i_jl = journal->j_current_jl;
3835 	REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3836 }
3837 
3838 /*
3839  * returns -1 on error, 0 if no commits/barriers were done and 1
3840  * if a transaction was actually committed and the barrier was done
3841  */
3842 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3843 			     struct reiserfs_journal_list *jl)
3844 {
3845 	struct reiserfs_transaction_handle th;
3846 	struct super_block *sb = inode->i_sb;
3847 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3848 	int ret = 0;
3849 
3850 	/*
3851 	 * is it from the current transaction,
3852 	 * or from an unknown transaction?
3853 	 */
3854 	if (id == journal->j_trans_id) {
3855 		jl = journal->j_current_jl;
3856 		/*
3857 		 * try to let other writers come in and
3858 		 * grow this transaction
3859 		 */
3860 		let_transaction_grow(sb, id);
3861 		if (journal->j_trans_id != id) {
3862 			goto flush_commit_only;
3863 		}
3864 
3865 		ret = journal_begin(&th, sb, 1);
3866 		if (ret)
3867 			return ret;
3868 
3869 		/* someone might have ended this transaction while we joined */
3870 		if (journal->j_trans_id != id) {
3871 			reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3872 						     1);
3873 			journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3874 			ret = journal_end(&th);
3875 			goto flush_commit_only;
3876 		}
3877 
3878 		ret = journal_end_sync(&th);
3879 		if (!ret)
3880 			ret = 1;
3881 
3882 	} else {
3883 		/*
3884 		 * this gets tricky, we have to make sure the journal list in
3885 		 * the inode still exists.  We know the list is still around
3886 		 * if we've got a larger transaction id than the oldest list
3887 		 */
3888 flush_commit_only:
3889 		if (journal_list_still_alive(inode->i_sb, id)) {
3890 			/*
3891 			 * we only set ret to 1 when we know for sure
3892 			 * the barrier hasn't been started yet on the commit
3893 			 * block.
3894 			 */
3895 			if (atomic_read(&jl->j_commit_left) > 1)
3896 				ret = 1;
3897 			flush_commit_list(sb, jl, 1);
3898 			if (journal->j_errno)
3899 				ret = journal->j_errno;
3900 		}
3901 	}
3902 	/* otherwise the list is gone, and long since committed */
3903 	return ret;
3904 }
3905 
3906 int reiserfs_commit_for_inode(struct inode *inode)
3907 {
3908 	unsigned int id = REISERFS_I(inode)->i_trans_id;
3909 	struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3910 
3911 	/*
3912 	 * for the whole inode, assume unset id means it was
3913 	 * changed in the current transaction.  More conservative
3914 	 */
3915 	if (!id || !jl) {
3916 		reiserfs_update_inode_transaction(inode);
3917 		id = REISERFS_I(inode)->i_trans_id;
3918 		/* jl will be updated in __commit_trans_jl */
3919 	}
3920 
3921 	return __commit_trans_jl(inode, id, jl);
3922 }
3923 
3924 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3925 				      struct buffer_head *bh)
3926 {
3927 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3928 	PROC_INFO_INC(sb, journal.restore_prepared);
3929 	if (!bh) {
3930 		return;
3931 	}
3932 	if (test_clear_buffer_journal_restore_dirty(bh) &&
3933 	    buffer_journal_dirty(bh)) {
3934 		struct reiserfs_journal_cnode *cn;
3935 		reiserfs_write_lock(sb);
3936 		cn = get_journal_hash_dev(sb,
3937 					  journal->j_list_hash_table,
3938 					  bh->b_blocknr);
3939 		if (cn && can_dirty(cn)) {
3940 			set_buffer_journal_test(bh);
3941 			mark_buffer_dirty(bh);
3942 		}
3943 		reiserfs_write_unlock(sb);
3944 	}
3945 	clear_buffer_journal_prepared(bh);
3946 }
3947 
3948 extern struct tree_balance *cur_tb;
3949 /*
3950  * before we can change a metadata block, we have to make sure it won't
3951  * be written to disk while we are altering it.  So, we must:
3952  * clean it
3953  * wait on it.
3954  */
3955 int reiserfs_prepare_for_journal(struct super_block *sb,
3956 				 struct buffer_head *bh, int wait)
3957 {
3958 	PROC_INFO_INC(sb, journal.prepare);
3959 
3960 	if (!trylock_buffer(bh)) {
3961 		if (!wait)
3962 			return 0;
3963 		lock_buffer(bh);
3964 	}
3965 	set_buffer_journal_prepared(bh);
3966 	if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3967 		clear_buffer_journal_test(bh);
3968 		set_buffer_journal_restore_dirty(bh);
3969 	}
3970 	unlock_buffer(bh);
3971 	return 1;
3972 }
3973 
3974 /*
3975  * long and ugly.  If flush, will not return until all commit
3976  * blocks and all real buffers in the trans are on disk.
3977  * If no_async, won't return until all commit blocks are on disk.
3978  *
3979  * keep reading, there are comments as you go along
3980  *
3981  * If the journal is aborted, we just clean up. Things like flushing
3982  * journal lists, etc just won't happen.
3983  */
3984 static int do_journal_end(struct reiserfs_transaction_handle *th, int flags)
3985 {
3986 	struct super_block *sb = th->t_super;
3987 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3988 	struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3989 	struct reiserfs_journal_cnode *last_cn = NULL;
3990 	struct reiserfs_journal_desc *desc;
3991 	struct reiserfs_journal_commit *commit;
3992 	struct buffer_head *c_bh;	/* commit bh */
3993 	struct buffer_head *d_bh;	/* desc bh */
3994 	int cur_write_start = 0;	/* start index of current log write */
3995 	int i;
3996 	int flush;
3997 	int wait_on_commit;
3998 	struct reiserfs_journal_list *jl, *temp_jl;
3999 	struct list_head *entry, *safe;
4000 	unsigned long jindex;
4001 	unsigned int commit_trans_id;
4002 	int trans_half;
4003 	int depth;
4004 
4005 	BUG_ON(th->t_refcount > 1);
4006 	BUG_ON(!th->t_trans_id);
4007 	BUG_ON(!th->t_super);
4008 
4009 	/*
4010 	 * protect flush_older_commits from doing mistakes if the
4011 	 * transaction ID counter gets overflowed.
4012 	 */
4013 	if (th->t_trans_id == ~0U)
4014 		flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
4015 	flush = flags & FLUSH_ALL;
4016 	wait_on_commit = flags & WAIT;
4017 
4018 	current->journal_info = th->t_handle_save;
4019 	reiserfs_check_lock_depth(sb, "journal end");
4020 	if (journal->j_len == 0) {
4021 		reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
4022 					     1);
4023 		journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
4024 	}
4025 
4026 	lock_journal(sb);
4027 	if (journal->j_next_full_flush) {
4028 		flags |= FLUSH_ALL;
4029 		flush = 1;
4030 	}
4031 	if (journal->j_next_async_flush) {
4032 		flags |= COMMIT_NOW | WAIT;
4033 		wait_on_commit = 1;
4034 	}
4035 
4036 	/*
4037 	 * check_journal_end locks the journal, and unlocks if it does
4038 	 * not return 1 it tells us if we should continue with the
4039 	 * journal_end, or just return
4040 	 */
4041 	if (!check_journal_end(th, flags)) {
4042 		reiserfs_schedule_old_flush(sb);
4043 		wake_queued_writers(sb);
4044 		reiserfs_async_progress_wait(sb);
4045 		goto out;
4046 	}
4047 
4048 	/* check_journal_end might set these, check again */
4049 	if (journal->j_next_full_flush) {
4050 		flush = 1;
4051 	}
4052 
4053 	/*
4054 	 * j must wait means we have to flush the log blocks, and the
4055 	 * real blocks for this transaction
4056 	 */
4057 	if (journal->j_must_wait > 0) {
4058 		flush = 1;
4059 	}
4060 #ifdef REISERFS_PREALLOCATE
4061 	/*
4062 	 * quota ops might need to nest, setup the journal_info pointer
4063 	 * for them and raise the refcount so that it is > 0.
4064 	 */
4065 	current->journal_info = th;
4066 	th->t_refcount++;
4067 
4068 	/* it should not involve new blocks into the transaction */
4069 	reiserfs_discard_all_prealloc(th);
4070 
4071 	th->t_refcount--;
4072 	current->journal_info = th->t_handle_save;
4073 #endif
4074 
4075 	/* setup description block */
4076 	d_bh =
4077 	    journal_getblk(sb,
4078 			   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4079 			   journal->j_start);
4080 	set_buffer_uptodate(d_bh);
4081 	desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4082 	memset(d_bh->b_data, 0, d_bh->b_size);
4083 	memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4084 	set_desc_trans_id(desc, journal->j_trans_id);
4085 
4086 	/*
4087 	 * setup commit block.  Don't write (keep it clean too) this one
4088 	 * until after everyone else is written
4089 	 */
4090 	c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4091 			      ((journal->j_start + journal->j_len +
4092 				1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4093 	commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4094 	memset(c_bh->b_data, 0, c_bh->b_size);
4095 	set_commit_trans_id(commit, journal->j_trans_id);
4096 	set_buffer_uptodate(c_bh);
4097 
4098 	/* init this journal list */
4099 	jl = journal->j_current_jl;
4100 
4101 	/*
4102 	 * we lock the commit before doing anything because
4103 	 * we want to make sure nobody tries to run flush_commit_list until
4104 	 * the new transaction is fully setup, and we've already flushed the
4105 	 * ordered bh list
4106 	 */
4107 	reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4108 
4109 	/* save the transaction id in case we need to commit it later */
4110 	commit_trans_id = jl->j_trans_id;
4111 
4112 	atomic_set(&jl->j_older_commits_done, 0);
4113 	jl->j_trans_id = journal->j_trans_id;
4114 	jl->j_timestamp = journal->j_trans_start_time;
4115 	jl->j_commit_bh = c_bh;
4116 	jl->j_start = journal->j_start;
4117 	jl->j_len = journal->j_len;
4118 	atomic_set(&jl->j_nonzerolen, journal->j_len);
4119 	atomic_set(&jl->j_commit_left, journal->j_len + 2);
4120 	jl->j_realblock = NULL;
4121 
4122 	/*
4123 	 * The ENTIRE FOR LOOP MUST not cause schedule to occur.
4124 	 * for each real block, add it to the journal list hash,
4125 	 * copy into real block index array in the commit or desc block
4126 	 */
4127 	trans_half = journal_trans_half(sb->s_blocksize);
4128 	for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4129 		if (buffer_journaled(cn->bh)) {
4130 			jl_cn = get_cnode(sb);
4131 			if (!jl_cn) {
4132 				reiserfs_panic(sb, "journal-1676",
4133 					       "get_cnode returned NULL");
4134 			}
4135 			if (i == 0) {
4136 				jl->j_realblock = jl_cn;
4137 			}
4138 			jl_cn->prev = last_cn;
4139 			jl_cn->next = NULL;
4140 			if (last_cn) {
4141 				last_cn->next = jl_cn;
4142 			}
4143 			last_cn = jl_cn;
4144 			/*
4145 			 * make sure the block we are trying to log
4146 			 * is not a block of journal or reserved area
4147 			 */
4148 			if (is_block_in_log_or_reserved_area
4149 			    (sb, cn->bh->b_blocknr)) {
4150 				reiserfs_panic(sb, "journal-2332",
4151 					       "Trying to log block %lu, "
4152 					       "which is a log block",
4153 					       cn->bh->b_blocknr);
4154 			}
4155 			jl_cn->blocknr = cn->bh->b_blocknr;
4156 			jl_cn->state = 0;
4157 			jl_cn->sb = sb;
4158 			jl_cn->bh = cn->bh;
4159 			jl_cn->jlist = jl;
4160 			insert_journal_hash(journal->j_list_hash_table, jl_cn);
4161 			if (i < trans_half) {
4162 				desc->j_realblock[i] =
4163 				    cpu_to_le32(cn->bh->b_blocknr);
4164 			} else {
4165 				commit->j_realblock[i - trans_half] =
4166 				    cpu_to_le32(cn->bh->b_blocknr);
4167 			}
4168 		} else {
4169 			i--;
4170 		}
4171 	}
4172 	set_desc_trans_len(desc, journal->j_len);
4173 	set_desc_mount_id(desc, journal->j_mount_id);
4174 	set_desc_trans_id(desc, journal->j_trans_id);
4175 	set_commit_trans_len(commit, journal->j_len);
4176 
4177 	/*
4178 	 * special check in case all buffers in the journal
4179 	 * were marked for not logging
4180 	 */
4181 	BUG_ON(journal->j_len == 0);
4182 
4183 	/*
4184 	 * we're about to dirty all the log blocks, mark the description block
4185 	 * dirty now too.  Don't mark the commit block dirty until all the
4186 	 * others are on disk
4187 	 */
4188 	mark_buffer_dirty(d_bh);
4189 
4190 	/*
4191 	 * first data block is j_start + 1, so add one to
4192 	 * cur_write_start wherever you use it
4193 	 */
4194 	cur_write_start = journal->j_start;
4195 	cn = journal->j_first;
4196 	jindex = 1;	/* start at one so we don't get the desc again */
4197 	while (cn) {
4198 		clear_buffer_journal_new(cn->bh);
4199 		/* copy all the real blocks into log area.  dirty log blocks */
4200 		if (buffer_journaled(cn->bh)) {
4201 			struct buffer_head *tmp_bh;
4202 			char *addr;
4203 			struct page *page;
4204 			tmp_bh =
4205 			    journal_getblk(sb,
4206 					   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4207 					   ((cur_write_start +
4208 					     jindex) %
4209 					    SB_ONDISK_JOURNAL_SIZE(sb)));
4210 			set_buffer_uptodate(tmp_bh);
4211 			page = cn->bh->b_page;
4212 			addr = kmap(page);
4213 			memcpy(tmp_bh->b_data,
4214 			       addr + offset_in_page(cn->bh->b_data),
4215 			       cn->bh->b_size);
4216 			kunmap(page);
4217 			mark_buffer_dirty(tmp_bh);
4218 			jindex++;
4219 			set_buffer_journal_dirty(cn->bh);
4220 			clear_buffer_journaled(cn->bh);
4221 		} else {
4222 			/*
4223 			 * JDirty cleared sometime during transaction.
4224 			 * don't log this one
4225 			 */
4226 			reiserfs_warning(sb, "journal-2048",
4227 					 "BAD, buffer in journal hash, "
4228 					 "but not JDirty!");
4229 			brelse(cn->bh);
4230 		}
4231 		next = cn->next;
4232 		free_cnode(sb, cn);
4233 		cn = next;
4234 		reiserfs_cond_resched(sb);
4235 	}
4236 
4237 	/*
4238 	 * we are done with both the c_bh and d_bh, but
4239 	 * c_bh must be written after all other commit blocks,
4240 	 * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4241 	 */
4242 
4243 	journal->j_current_jl = alloc_journal_list(sb);
4244 
4245 	/* now it is safe to insert this transaction on the main list */
4246 	list_add_tail(&jl->j_list, &journal->j_journal_list);
4247 	list_add_tail(&jl->j_working_list, &journal->j_working_list);
4248 	journal->j_num_work_lists++;
4249 
4250 	/* reset journal values for the next transaction */
4251 	journal->j_start =
4252 	    (journal->j_start + journal->j_len +
4253 	     2) % SB_ONDISK_JOURNAL_SIZE(sb);
4254 	atomic_set(&journal->j_wcount, 0);
4255 	journal->j_bcount = 0;
4256 	journal->j_last = NULL;
4257 	journal->j_first = NULL;
4258 	journal->j_len = 0;
4259 	journal->j_trans_start_time = 0;
4260 	/* check for trans_id overflow */
4261 	if (++journal->j_trans_id == 0)
4262 		journal->j_trans_id = 10;
4263 	journal->j_current_jl->j_trans_id = journal->j_trans_id;
4264 	journal->j_must_wait = 0;
4265 	journal->j_len_alloc = 0;
4266 	journal->j_next_full_flush = 0;
4267 	journal->j_next_async_flush = 0;
4268 	init_journal_hash(sb);
4269 
4270 	/*
4271 	 * make sure reiserfs_add_jh sees the new current_jl before we
4272 	 * write out the tails
4273 	 */
4274 	smp_mb();
4275 
4276 	/*
4277 	 * tail conversion targets have to hit the disk before we end the
4278 	 * transaction.  Otherwise a later transaction might repack the tail
4279 	 * before this transaction commits, leaving the data block unflushed
4280 	 * and clean, if we crash before the later transaction commits, the
4281 	 * data block is lost.
4282 	 */
4283 	if (!list_empty(&jl->j_tail_bh_list)) {
4284 		depth = reiserfs_write_unlock_nested(sb);
4285 		write_ordered_buffers(&journal->j_dirty_buffers_lock,
4286 				      journal, jl, &jl->j_tail_bh_list);
4287 		reiserfs_write_lock_nested(sb, depth);
4288 	}
4289 	BUG_ON(!list_empty(&jl->j_tail_bh_list));
4290 	mutex_unlock(&jl->j_commit_mutex);
4291 
4292 	/*
4293 	 * honor the flush wishes from the caller, simple commits can
4294 	 * be done outside the journal lock, they are done below
4295 	 *
4296 	 * if we don't flush the commit list right now, we put it into
4297 	 * the work queue so the people waiting on the async progress work
4298 	 * queue don't wait for this proc to flush journal lists and such.
4299 	 */
4300 	if (flush) {
4301 		flush_commit_list(sb, jl, 1);
4302 		flush_journal_list(sb, jl, 1);
4303 	} else if (!(jl->j_state & LIST_COMMIT_PENDING)) {
4304 		/*
4305 		 * Avoid queueing work when sb is being shut down. Transaction
4306 		 * will be flushed on journal shutdown.
4307 		 */
4308 		if (sb->s_flags & SB_ACTIVE)
4309 			queue_delayed_work(REISERFS_SB(sb)->commit_wq,
4310 					   &journal->j_work, HZ / 10);
4311 	}
4312 
4313 	/*
4314 	 * if the next transaction has any chance of wrapping, flush
4315 	 * transactions that might get overwritten.  If any journal lists
4316 	 * are very old flush them as well.
4317 	 */
4318 first_jl:
4319 	list_for_each_safe(entry, safe, &journal->j_journal_list) {
4320 		temp_jl = JOURNAL_LIST_ENTRY(entry);
4321 		if (journal->j_start <= temp_jl->j_start) {
4322 			if ((journal->j_start + journal->j_trans_max + 1) >=
4323 			    temp_jl->j_start) {
4324 				flush_used_journal_lists(sb, temp_jl);
4325 				goto first_jl;
4326 			} else if ((journal->j_start +
4327 				    journal->j_trans_max + 1) <
4328 				   SB_ONDISK_JOURNAL_SIZE(sb)) {
4329 				/*
4330 				 * if we don't cross into the next
4331 				 * transaction and we don't wrap, there is
4332 				 * no way we can overlap any later transactions
4333 				 * break now
4334 				 */
4335 				break;
4336 			}
4337 		} else if ((journal->j_start +
4338 			    journal->j_trans_max + 1) >
4339 			   SB_ONDISK_JOURNAL_SIZE(sb)) {
4340 			if (((journal->j_start + journal->j_trans_max + 1) %
4341 			     SB_ONDISK_JOURNAL_SIZE(sb)) >=
4342 			    temp_jl->j_start) {
4343 				flush_used_journal_lists(sb, temp_jl);
4344 				goto first_jl;
4345 			} else {
4346 				/*
4347 				* we don't overlap anything from out start
4348 				* to the end of the log, and our wrapped
4349 				* portion doesn't overlap anything at
4350 				* the start of the log.  We can break
4351 				*/
4352 				break;
4353 			}
4354 		}
4355 	}
4356 
4357 	journal->j_current_jl->j_list_bitmap =
4358 	    get_list_bitmap(sb, journal->j_current_jl);
4359 
4360 	if (!(journal->j_current_jl->j_list_bitmap)) {
4361 		reiserfs_panic(sb, "journal-1996",
4362 			       "could not get a list bitmap");
4363 	}
4364 
4365 	atomic_set(&journal->j_jlock, 0);
4366 	unlock_journal(sb);
4367 	/* wake up any body waiting to join. */
4368 	clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4369 	wake_up(&journal->j_join_wait);
4370 
4371 	if (!flush && wait_on_commit &&
4372 	    journal_list_still_alive(sb, commit_trans_id)) {
4373 		flush_commit_list(sb, jl, 1);
4374 	}
4375 out:
4376 	reiserfs_check_lock_depth(sb, "journal end2");
4377 
4378 	memset(th, 0, sizeof(*th));
4379 	/*
4380 	 * Re-set th->t_super, so we can properly keep track of how many
4381 	 * persistent transactions there are. We need to do this so if this
4382 	 * call is part of a failed restart_transaction, we can free it later
4383 	 */
4384 	th->t_super = sb;
4385 
4386 	return journal->j_errno;
4387 }
4388 
4389 /* Send the file system read only and refuse new transactions */
4390 void reiserfs_abort_journal(struct super_block *sb, int errno)
4391 {
4392 	struct reiserfs_journal *journal = SB_JOURNAL(sb);
4393 	if (test_bit(J_ABORTED, &journal->j_state))
4394 		return;
4395 
4396 	if (!journal->j_errno)
4397 		journal->j_errno = errno;
4398 
4399 	sb->s_flags |= SB_RDONLY;
4400 	set_bit(J_ABORTED, &journal->j_state);
4401 
4402 #ifdef CONFIG_REISERFS_CHECK
4403 	dump_stack();
4404 #endif
4405 }
4406