GNU Linux-libre 4.19.211-gnu1
[releases.git] / fs / reiserfs / journal.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Write ahead logging implementation copyright Chris Mason 2000
4  *
5  * The background commits make this code very interrelated, and
6  * overly complex.  I need to rethink things a bit....The major players:
7  *
8  * journal_begin -- call with the number of blocks you expect to log.
9  *                  If the current transaction is too
10  *                  old, it will block until the current transaction is
11  *                  finished, and then start a new one.
12  *                  Usually, your transaction will get joined in with
13  *                  previous ones for speed.
14  *
15  * journal_join  -- same as journal_begin, but won't block on the current
16  *                  transaction regardless of age.  Don't ever call
17  *                  this.  Ever.  There are only two places it should be
18  *                  called from, and they are both inside this file.
19  *
20  * journal_mark_dirty -- adds blocks into this transaction.  clears any flags
21  *                       that might make them get sent to disk
22  *                       and then marks them BH_JDirty.  Puts the buffer head
23  *                       into the current transaction hash.
24  *
25  * journal_end -- if the current transaction is batchable, it does nothing
26  *                   otherwise, it could do an async/synchronous commit, or
27  *                   a full flush of all log and real blocks in the
28  *                   transaction.
29  *
30  * flush_old_commits -- if the current transaction is too old, it is ended and
31  *                      commit blocks are sent to disk.  Forces commit blocks
32  *                      to disk for all backgrounded commits that have been
33  *                      around too long.
34  *                   -- Note, if you call this as an immediate flush from
35  *                      from within kupdate, it will ignore the immediate flag
36  */
37
38 #include <linux/time.h>
39 #include <linux/semaphore.h>
40 #include <linux/vmalloc.h>
41 #include "reiserfs.h"
42 #include <linux/kernel.h>
43 #include <linux/errno.h>
44 #include <linux/fcntl.h>
45 #include <linux/stat.h>
46 #include <linux/string.h>
47 #include <linux/buffer_head.h>
48 #include <linux/workqueue.h>
49 #include <linux/writeback.h>
50 #include <linux/blkdev.h>
51 #include <linux/backing-dev.h>
52 #include <linux/uaccess.h>
53 #include <linux/slab.h>
54
55
56 /* gets a struct reiserfs_journal_list * from a list head */
57 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
58                                j_list))
59 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
60                                j_working_list))
61
62 /* must be correct to keep the desc and commit structs at 4k */
63 #define JOURNAL_TRANS_HALF 1018
64 #define BUFNR 64                /*read ahead */
65
66 /* cnode stat bits.  Move these into reiserfs_fs.h */
67
68 /* this block was freed, and can't be written.  */
69 #define BLOCK_FREED 2
70 /* this block was freed during this transaction, and can't be written */
71 #define BLOCK_FREED_HOLDER 3
72
73 /* used in flush_journal_list */
74 #define BLOCK_NEEDS_FLUSH 4
75 #define BLOCK_DIRTIED 5
76
77 /* journal list state bits */
78 #define LIST_TOUCHED 1
79 #define LIST_DIRTY   2
80 #define LIST_COMMIT_PENDING  4  /* someone will commit this list */
81
82 /* flags for do_journal_end */
83 #define FLUSH_ALL   1           /* flush commit and real blocks */
84 #define COMMIT_NOW  2           /* end and commit this transaction */
85 #define WAIT        4           /* wait for the log blocks to hit the disk */
86
87 static int do_journal_end(struct reiserfs_transaction_handle *, int flags);
88 static int flush_journal_list(struct super_block *s,
89                               struct reiserfs_journal_list *jl, int flushall);
90 static int flush_commit_list(struct super_block *s,
91                              struct reiserfs_journal_list *jl, int flushall);
92 static int can_dirty(struct reiserfs_journal_cnode *cn);
93 static int journal_join(struct reiserfs_transaction_handle *th,
94                         struct super_block *sb);
95 static void release_journal_dev(struct super_block *super,
96                                struct reiserfs_journal *journal);
97 static int dirty_one_transaction(struct super_block *s,
98                                  struct reiserfs_journal_list *jl);
99 static void flush_async_commits(struct work_struct *work);
100 static void queue_log_writer(struct super_block *s);
101
102 /* values for join in do_journal_begin_r */
103 enum {
104         JBEGIN_REG = 0,         /* regular journal begin */
105         /* join the running transaction if at all possible */
106         JBEGIN_JOIN = 1,
107         /* called from cleanup code, ignores aborted flag */
108         JBEGIN_ABORT = 2,
109 };
110
111 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
112                               struct super_block *sb,
113                               unsigned long nblocks, int join);
114
115 static void init_journal_hash(struct super_block *sb)
116 {
117         struct reiserfs_journal *journal = SB_JOURNAL(sb);
118         memset(journal->j_hash_table, 0,
119                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
120 }
121
122 /*
123  * clears BH_Dirty and sticks the buffer on the clean list.  Called because
124  * I can't allow refile_buffer to make schedule happen after I've freed a
125  * block.  Look at remove_from_transaction and journal_mark_freed for
126  * more details.
127  */
128 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
129 {
130         if (bh) {
131                 clear_buffer_dirty(bh);
132                 clear_buffer_journal_test(bh);
133         }
134         return 0;
135 }
136
137 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
138                                                          *sb)
139 {
140         struct reiserfs_bitmap_node *bn;
141         static int id;
142
143         bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
144         if (!bn) {
145                 return NULL;
146         }
147         bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
148         if (!bn->data) {
149                 kfree(bn);
150                 return NULL;
151         }
152         bn->id = id++;
153         INIT_LIST_HEAD(&bn->list);
154         return bn;
155 }
156
157 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
158 {
159         struct reiserfs_journal *journal = SB_JOURNAL(sb);
160         struct reiserfs_bitmap_node *bn = NULL;
161         struct list_head *entry = journal->j_bitmap_nodes.next;
162
163         journal->j_used_bitmap_nodes++;
164 repeat:
165
166         if (entry != &journal->j_bitmap_nodes) {
167                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
168                 list_del(entry);
169                 memset(bn->data, 0, sb->s_blocksize);
170                 journal->j_free_bitmap_nodes--;
171                 return bn;
172         }
173         bn = allocate_bitmap_node(sb);
174         if (!bn) {
175                 yield();
176                 goto repeat;
177         }
178         return bn;
179 }
180 static inline void free_bitmap_node(struct super_block *sb,
181                                     struct reiserfs_bitmap_node *bn)
182 {
183         struct reiserfs_journal *journal = SB_JOURNAL(sb);
184         journal->j_used_bitmap_nodes--;
185         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
186                 kfree(bn->data);
187                 kfree(bn);
188         } else {
189                 list_add(&bn->list, &journal->j_bitmap_nodes);
190                 journal->j_free_bitmap_nodes++;
191         }
192 }
193
194 static void allocate_bitmap_nodes(struct super_block *sb)
195 {
196         int i;
197         struct reiserfs_journal *journal = SB_JOURNAL(sb);
198         struct reiserfs_bitmap_node *bn = NULL;
199         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
200                 bn = allocate_bitmap_node(sb);
201                 if (bn) {
202                         list_add(&bn->list, &journal->j_bitmap_nodes);
203                         journal->j_free_bitmap_nodes++;
204                 } else {
205                         /* this is ok, we'll try again when more are needed */
206                         break;
207                 }
208         }
209 }
210
211 static int set_bit_in_list_bitmap(struct super_block *sb,
212                                   b_blocknr_t block,
213                                   struct reiserfs_list_bitmap *jb)
214 {
215         unsigned int bmap_nr = block / (sb->s_blocksize << 3);
216         unsigned int bit_nr = block % (sb->s_blocksize << 3);
217
218         if (!jb->bitmaps[bmap_nr]) {
219                 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
220         }
221         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
222         return 0;
223 }
224
225 static void cleanup_bitmap_list(struct super_block *sb,
226                                 struct reiserfs_list_bitmap *jb)
227 {
228         int i;
229         if (jb->bitmaps == NULL)
230                 return;
231
232         for (i = 0; i < reiserfs_bmap_count(sb); i++) {
233                 if (jb->bitmaps[i]) {
234                         free_bitmap_node(sb, jb->bitmaps[i]);
235                         jb->bitmaps[i] = NULL;
236                 }
237         }
238 }
239
240 /*
241  * only call this on FS unmount.
242  */
243 static int free_list_bitmaps(struct super_block *sb,
244                              struct reiserfs_list_bitmap *jb_array)
245 {
246         int i;
247         struct reiserfs_list_bitmap *jb;
248         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
249                 jb = jb_array + i;
250                 jb->journal_list = NULL;
251                 cleanup_bitmap_list(sb, jb);
252                 vfree(jb->bitmaps);
253                 jb->bitmaps = NULL;
254         }
255         return 0;
256 }
257
258 static int free_bitmap_nodes(struct super_block *sb)
259 {
260         struct reiserfs_journal *journal = SB_JOURNAL(sb);
261         struct list_head *next = journal->j_bitmap_nodes.next;
262         struct reiserfs_bitmap_node *bn;
263
264         while (next != &journal->j_bitmap_nodes) {
265                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
266                 list_del(next);
267                 kfree(bn->data);
268                 kfree(bn);
269                 next = journal->j_bitmap_nodes.next;
270                 journal->j_free_bitmap_nodes--;
271         }
272
273         return 0;
274 }
275
276 /*
277  * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
278  * jb_array is the array to be filled in.
279  */
280 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
281                                    struct reiserfs_list_bitmap *jb_array,
282                                    unsigned int bmap_nr)
283 {
284         int i;
285         int failed = 0;
286         struct reiserfs_list_bitmap *jb;
287         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
288
289         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
290                 jb = jb_array + i;
291                 jb->journal_list = NULL;
292                 jb->bitmaps = vzalloc(mem);
293                 if (!jb->bitmaps) {
294                         reiserfs_warning(sb, "clm-2000", "unable to "
295                                          "allocate bitmaps for journal lists");
296                         failed = 1;
297                         break;
298                 }
299         }
300         if (failed) {
301                 free_list_bitmaps(sb, jb_array);
302                 return -1;
303         }
304         return 0;
305 }
306
307 /*
308  * find an available list bitmap.  If you can't find one, flush a commit list
309  * and try again
310  */
311 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
312                                                     struct reiserfs_journal_list
313                                                     *jl)
314 {
315         int i, j;
316         struct reiserfs_journal *journal = SB_JOURNAL(sb);
317         struct reiserfs_list_bitmap *jb = NULL;
318
319         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
320                 i = journal->j_list_bitmap_index;
321                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
322                 jb = journal->j_list_bitmap + i;
323                 if (journal->j_list_bitmap[i].journal_list) {
324                         flush_commit_list(sb,
325                                           journal->j_list_bitmap[i].
326                                           journal_list, 1);
327                         if (!journal->j_list_bitmap[i].journal_list) {
328                                 break;
329                         }
330                 } else {
331                         break;
332                 }
333         }
334         /* double check to make sure if flushed correctly */
335         if (jb->journal_list)
336                 return NULL;
337         jb->journal_list = jl;
338         return jb;
339 }
340
341 /*
342  * allocates a new chunk of X nodes, and links them all together as a list.
343  * Uses the cnode->next and cnode->prev pointers
344  * returns NULL on failure
345  */
346 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
347 {
348         struct reiserfs_journal_cnode *head;
349         int i;
350         if (num_cnodes <= 0) {
351                 return NULL;
352         }
353         head = vzalloc(array_size(num_cnodes,
354                                   sizeof(struct reiserfs_journal_cnode)));
355         if (!head) {
356                 return NULL;
357         }
358         head[0].prev = NULL;
359         head[0].next = head + 1;
360         for (i = 1; i < num_cnodes; i++) {
361                 head[i].prev = head + (i - 1);
362                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
363         }
364         head[num_cnodes - 1].next = NULL;
365         return head;
366 }
367
368 /* pulls a cnode off the free list, or returns NULL on failure */
369 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
370 {
371         struct reiserfs_journal_cnode *cn;
372         struct reiserfs_journal *journal = SB_JOURNAL(sb);
373
374         reiserfs_check_lock_depth(sb, "get_cnode");
375
376         if (journal->j_cnode_free <= 0) {
377                 return NULL;
378         }
379         journal->j_cnode_used++;
380         journal->j_cnode_free--;
381         cn = journal->j_cnode_free_list;
382         if (!cn) {
383                 return cn;
384         }
385         if (cn->next) {
386                 cn->next->prev = NULL;
387         }
388         journal->j_cnode_free_list = cn->next;
389         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
390         return cn;
391 }
392
393 /*
394  * returns a cnode to the free list
395  */
396 static void free_cnode(struct super_block *sb,
397                        struct reiserfs_journal_cnode *cn)
398 {
399         struct reiserfs_journal *journal = SB_JOURNAL(sb);
400
401         reiserfs_check_lock_depth(sb, "free_cnode");
402
403         journal->j_cnode_used--;
404         journal->j_cnode_free++;
405         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
406         cn->next = journal->j_cnode_free_list;
407         if (journal->j_cnode_free_list) {
408                 journal->j_cnode_free_list->prev = cn;
409         }
410         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
411         journal->j_cnode_free_list = cn;
412 }
413
414 static void clear_prepared_bits(struct buffer_head *bh)
415 {
416         clear_buffer_journal_prepared(bh);
417         clear_buffer_journal_restore_dirty(bh);
418 }
419
420 /*
421  * return a cnode with same dev, block number and size in table,
422  * or null if not found
423  */
424 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
425                                                                   super_block
426                                                                   *sb,
427                                                                   struct
428                                                                   reiserfs_journal_cnode
429                                                                   **table,
430                                                                   long bl)
431 {
432         struct reiserfs_journal_cnode *cn;
433         cn = journal_hash(table, sb, bl);
434         while (cn) {
435                 if (cn->blocknr == bl && cn->sb == sb)
436                         return cn;
437                 cn = cn->hnext;
438         }
439         return (struct reiserfs_journal_cnode *)0;
440 }
441
442 /*
443  * this actually means 'can this block be reallocated yet?'.  If you set
444  * search_all, a block can only be allocated if it is not in the current
445  * transaction, was not freed by the current transaction, and has no chance
446  * of ever being overwritten by a replay after crashing.
447  *
448  * If you don't set search_all, a block can only be allocated if it is not
449  * in the current transaction.  Since deleting a block removes it from the
450  * current transaction, this case should never happen.  If you don't set
451  * search_all, make sure you never write the block without logging it.
452  *
453  * next_zero_bit is a suggestion about the next block to try for find_forward.
454  * when bl is rejected because it is set in a journal list bitmap, we search
455  * for the next zero bit in the bitmap that rejected bl.  Then, we return
456  * that through next_zero_bit for find_forward to try.
457  *
458  * Just because we return something in next_zero_bit does not mean we won't
459  * reject it on the next call to reiserfs_in_journal
460  */
461 int reiserfs_in_journal(struct super_block *sb,
462                         unsigned int bmap_nr, int bit_nr, int search_all,
463                         b_blocknr_t * next_zero_bit)
464 {
465         struct reiserfs_journal *journal = SB_JOURNAL(sb);
466         struct reiserfs_journal_cnode *cn;
467         struct reiserfs_list_bitmap *jb;
468         int i;
469         unsigned long bl;
470
471         *next_zero_bit = 0;     /* always start this at zero. */
472
473         PROC_INFO_INC(sb, journal.in_journal);
474         /*
475          * If we aren't doing a search_all, this is a metablock, and it
476          * will be logged before use.  if we crash before the transaction
477          * that freed it commits,  this transaction won't have committed
478          * either, and the block will never be written
479          */
480         if (search_all) {
481                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
482                         PROC_INFO_INC(sb, journal.in_journal_bitmap);
483                         jb = journal->j_list_bitmap + i;
484                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
485                             test_bit(bit_nr,
486                                      (unsigned long *)jb->bitmaps[bmap_nr]->
487                                      data)) {
488                                 *next_zero_bit =
489                                     find_next_zero_bit((unsigned long *)
490                                                        (jb->bitmaps[bmap_nr]->
491                                                         data),
492                                                        sb->s_blocksize << 3,
493                                                        bit_nr + 1);
494                                 return 1;
495                         }
496                 }
497         }
498
499         bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
500         /* is it in any old transactions? */
501         if (search_all
502             && (cn =
503                 get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
504                 return 1;
505         }
506
507         /* is it in the current transaction.  This should never happen */
508         if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
509                 BUG();
510                 return 1;
511         }
512
513         PROC_INFO_INC(sb, journal.in_journal_reusable);
514         /* safe for reuse */
515         return 0;
516 }
517
518 /* insert cn into table */
519 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
520                                        struct reiserfs_journal_cnode *cn)
521 {
522         struct reiserfs_journal_cnode *cn_orig;
523
524         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
525         cn->hnext = cn_orig;
526         cn->hprev = NULL;
527         if (cn_orig) {
528                 cn_orig->hprev = cn;
529         }
530         journal_hash(table, cn->sb, cn->blocknr) = cn;
531 }
532
533 /* lock the current transaction */
534 static inline void lock_journal(struct super_block *sb)
535 {
536         PROC_INFO_INC(sb, journal.lock_journal);
537
538         reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
539 }
540
541 /* unlock the current transaction */
542 static inline void unlock_journal(struct super_block *sb)
543 {
544         mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
545 }
546
547 static inline void get_journal_list(struct reiserfs_journal_list *jl)
548 {
549         jl->j_refcount++;
550 }
551
552 static inline void put_journal_list(struct super_block *s,
553                                     struct reiserfs_journal_list *jl)
554 {
555         if (jl->j_refcount < 1) {
556                 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
557                                jl->j_trans_id, jl->j_refcount);
558         }
559         if (--jl->j_refcount == 0)
560                 kfree(jl);
561 }
562
563 /*
564  * this used to be much more involved, and I'm keeping it just in case
565  * things get ugly again.  it gets called by flush_commit_list, and
566  * cleans up any data stored about blocks freed during a transaction.
567  */
568 static void cleanup_freed_for_journal_list(struct super_block *sb,
569                                            struct reiserfs_journal_list *jl)
570 {
571
572         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
573         if (jb) {
574                 cleanup_bitmap_list(sb, jb);
575         }
576         jl->j_list_bitmap->journal_list = NULL;
577         jl->j_list_bitmap = NULL;
578 }
579
580 static int journal_list_still_alive(struct super_block *s,
581                                     unsigned int trans_id)
582 {
583         struct reiserfs_journal *journal = SB_JOURNAL(s);
584         struct list_head *entry = &journal->j_journal_list;
585         struct reiserfs_journal_list *jl;
586
587         if (!list_empty(entry)) {
588                 jl = JOURNAL_LIST_ENTRY(entry->next);
589                 if (jl->j_trans_id <= trans_id) {
590                         return 1;
591                 }
592         }
593         return 0;
594 }
595
596 /*
597  * If page->mapping was null, we failed to truncate this page for
598  * some reason.  Most likely because it was truncated after being
599  * logged via data=journal.
600  *
601  * This does a check to see if the buffer belongs to one of these
602  * lost pages before doing the final put_bh.  If page->mapping was
603  * null, it tries to free buffers on the page, which should make the
604  * final put_page drop the page from the lru.
605  */
606 static void release_buffer_page(struct buffer_head *bh)
607 {
608         struct page *page = bh->b_page;
609         if (!page->mapping && trylock_page(page)) {
610                 get_page(page);
611                 put_bh(bh);
612                 if (!page->mapping)
613                         try_to_free_buffers(page);
614                 unlock_page(page);
615                 put_page(page);
616         } else {
617                 put_bh(bh);
618         }
619 }
620
621 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
622 {
623         if (buffer_journaled(bh)) {
624                 reiserfs_warning(NULL, "clm-2084",
625                                  "pinned buffer %lu:%pg sent to disk",
626                                  bh->b_blocknr, bh->b_bdev);
627         }
628         if (uptodate)
629                 set_buffer_uptodate(bh);
630         else
631                 clear_buffer_uptodate(bh);
632
633         unlock_buffer(bh);
634         release_buffer_page(bh);
635 }
636
637 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
638 {
639         if (uptodate)
640                 set_buffer_uptodate(bh);
641         else
642                 clear_buffer_uptodate(bh);
643         unlock_buffer(bh);
644         put_bh(bh);
645 }
646
647 static void submit_logged_buffer(struct buffer_head *bh)
648 {
649         get_bh(bh);
650         bh->b_end_io = reiserfs_end_buffer_io_sync;
651         clear_buffer_journal_new(bh);
652         clear_buffer_dirty(bh);
653         if (!test_clear_buffer_journal_test(bh))
654                 BUG();
655         if (!buffer_uptodate(bh))
656                 BUG();
657         submit_bh(REQ_OP_WRITE, 0, bh);
658 }
659
660 static void submit_ordered_buffer(struct buffer_head *bh)
661 {
662         get_bh(bh);
663         bh->b_end_io = reiserfs_end_ordered_io;
664         clear_buffer_dirty(bh);
665         if (!buffer_uptodate(bh))
666                 BUG();
667         submit_bh(REQ_OP_WRITE, 0, bh);
668 }
669
670 #define CHUNK_SIZE 32
671 struct buffer_chunk {
672         struct buffer_head *bh[CHUNK_SIZE];
673         int nr;
674 };
675
676 static void write_chunk(struct buffer_chunk *chunk)
677 {
678         int i;
679         for (i = 0; i < chunk->nr; i++) {
680                 submit_logged_buffer(chunk->bh[i]);
681         }
682         chunk->nr = 0;
683 }
684
685 static void write_ordered_chunk(struct buffer_chunk *chunk)
686 {
687         int i;
688         for (i = 0; i < chunk->nr; i++) {
689                 submit_ordered_buffer(chunk->bh[i]);
690         }
691         chunk->nr = 0;
692 }
693
694 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
695                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
696 {
697         int ret = 0;
698         BUG_ON(chunk->nr >= CHUNK_SIZE);
699         chunk->bh[chunk->nr++] = bh;
700         if (chunk->nr >= CHUNK_SIZE) {
701                 ret = 1;
702                 if (lock) {
703                         spin_unlock(lock);
704                         fn(chunk);
705                         spin_lock(lock);
706                 } else {
707                         fn(chunk);
708                 }
709         }
710         return ret;
711 }
712
713 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
714 static struct reiserfs_jh *alloc_jh(void)
715 {
716         struct reiserfs_jh *jh;
717         while (1) {
718                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
719                 if (jh) {
720                         atomic_inc(&nr_reiserfs_jh);
721                         return jh;
722                 }
723                 yield();
724         }
725 }
726
727 /*
728  * we want to free the jh when the buffer has been written
729  * and waited on
730  */
731 void reiserfs_free_jh(struct buffer_head *bh)
732 {
733         struct reiserfs_jh *jh;
734
735         jh = bh->b_private;
736         if (jh) {
737                 bh->b_private = NULL;
738                 jh->bh = NULL;
739                 list_del_init(&jh->list);
740                 kfree(jh);
741                 if (atomic_read(&nr_reiserfs_jh) <= 0)
742                         BUG();
743                 atomic_dec(&nr_reiserfs_jh);
744                 put_bh(bh);
745         }
746 }
747
748 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
749                            int tail)
750 {
751         struct reiserfs_jh *jh;
752
753         if (bh->b_private) {
754                 spin_lock(&j->j_dirty_buffers_lock);
755                 if (!bh->b_private) {
756                         spin_unlock(&j->j_dirty_buffers_lock);
757                         goto no_jh;
758                 }
759                 jh = bh->b_private;
760                 list_del_init(&jh->list);
761         } else {
762 no_jh:
763                 get_bh(bh);
764                 jh = alloc_jh();
765                 spin_lock(&j->j_dirty_buffers_lock);
766                 /*
767                  * buffer must be locked for __add_jh, should be able to have
768                  * two adds at the same time
769                  */
770                 BUG_ON(bh->b_private);
771                 jh->bh = bh;
772                 bh->b_private = jh;
773         }
774         jh->jl = j->j_current_jl;
775         if (tail)
776                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
777         else {
778                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
779         }
780         spin_unlock(&j->j_dirty_buffers_lock);
781         return 0;
782 }
783
784 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
785 {
786         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
787 }
788 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
789 {
790         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
791 }
792
793 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
794 static int write_ordered_buffers(spinlock_t * lock,
795                                  struct reiserfs_journal *j,
796                                  struct reiserfs_journal_list *jl,
797                                  struct list_head *list)
798 {
799         struct buffer_head *bh;
800         struct reiserfs_jh *jh;
801         int ret = j->j_errno;
802         struct buffer_chunk chunk;
803         struct list_head tmp;
804         INIT_LIST_HEAD(&tmp);
805
806         chunk.nr = 0;
807         spin_lock(lock);
808         while (!list_empty(list)) {
809                 jh = JH_ENTRY(list->next);
810                 bh = jh->bh;
811                 get_bh(bh);
812                 if (!trylock_buffer(bh)) {
813                         if (!buffer_dirty(bh)) {
814                                 list_move(&jh->list, &tmp);
815                                 goto loop_next;
816                         }
817                         spin_unlock(lock);
818                         if (chunk.nr)
819                                 write_ordered_chunk(&chunk);
820                         wait_on_buffer(bh);
821                         cond_resched();
822                         spin_lock(lock);
823                         goto loop_next;
824                 }
825                 /*
826                  * in theory, dirty non-uptodate buffers should never get here,
827                  * but the upper layer io error paths still have a few quirks.
828                  * Handle them here as gracefully as we can
829                  */
830                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
831                         clear_buffer_dirty(bh);
832                         ret = -EIO;
833                 }
834                 if (buffer_dirty(bh)) {
835                         list_move(&jh->list, &tmp);
836                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
837                 } else {
838                         reiserfs_free_jh(bh);
839                         unlock_buffer(bh);
840                 }
841 loop_next:
842                 put_bh(bh);
843                 cond_resched_lock(lock);
844         }
845         if (chunk.nr) {
846                 spin_unlock(lock);
847                 write_ordered_chunk(&chunk);
848                 spin_lock(lock);
849         }
850         while (!list_empty(&tmp)) {
851                 jh = JH_ENTRY(tmp.prev);
852                 bh = jh->bh;
853                 get_bh(bh);
854                 reiserfs_free_jh(bh);
855
856                 if (buffer_locked(bh)) {
857                         spin_unlock(lock);
858                         wait_on_buffer(bh);
859                         spin_lock(lock);
860                 }
861                 if (!buffer_uptodate(bh)) {
862                         ret = -EIO;
863                 }
864                 /*
865                  * ugly interaction with invalidatepage here.
866                  * reiserfs_invalidate_page will pin any buffer that has a
867                  * valid journal head from an older transaction.  If someone
868                  * else sets our buffer dirty after we write it in the first
869                  * loop, and then someone truncates the page away, nobody
870                  * will ever write the buffer. We're safe if we write the
871                  * page one last time after freeing the journal header.
872                  */
873                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
874                         spin_unlock(lock);
875                         ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
876                         spin_lock(lock);
877                 }
878                 put_bh(bh);
879                 cond_resched_lock(lock);
880         }
881         spin_unlock(lock);
882         return ret;
883 }
884
885 static int flush_older_commits(struct super_block *s,
886                                struct reiserfs_journal_list *jl)
887 {
888         struct reiserfs_journal *journal = SB_JOURNAL(s);
889         struct reiserfs_journal_list *other_jl;
890         struct reiserfs_journal_list *first_jl;
891         struct list_head *entry;
892         unsigned int trans_id = jl->j_trans_id;
893         unsigned int other_trans_id;
894         unsigned int first_trans_id;
895
896 find_first:
897         /*
898          * first we walk backwards to find the oldest uncommitted transation
899          */
900         first_jl = jl;
901         entry = jl->j_list.prev;
902         while (1) {
903                 other_jl = JOURNAL_LIST_ENTRY(entry);
904                 if (entry == &journal->j_journal_list ||
905                     atomic_read(&other_jl->j_older_commits_done))
906                         break;
907
908                 first_jl = other_jl;
909                 entry = other_jl->j_list.prev;
910         }
911
912         /* if we didn't find any older uncommitted transactions, return now */
913         if (first_jl == jl) {
914                 return 0;
915         }
916
917         first_trans_id = first_jl->j_trans_id;
918
919         entry = &first_jl->j_list;
920         while (1) {
921                 other_jl = JOURNAL_LIST_ENTRY(entry);
922                 other_trans_id = other_jl->j_trans_id;
923
924                 if (other_trans_id < trans_id) {
925                         if (atomic_read(&other_jl->j_commit_left) != 0) {
926                                 flush_commit_list(s, other_jl, 0);
927
928                                 /* list we were called with is gone, return */
929                                 if (!journal_list_still_alive(s, trans_id))
930                                         return 1;
931
932                                 /*
933                                  * the one we just flushed is gone, this means
934                                  * all older lists are also gone, so first_jl
935                                  * is no longer valid either.  Go back to the
936                                  * beginning.
937                                  */
938                                 if (!journal_list_still_alive
939                                     (s, other_trans_id)) {
940                                         goto find_first;
941                                 }
942                         }
943                         entry = entry->next;
944                         if (entry == &journal->j_journal_list)
945                                 return 0;
946                 } else {
947                         return 0;
948                 }
949         }
950         return 0;
951 }
952
953 static int reiserfs_async_progress_wait(struct super_block *s)
954 {
955         struct reiserfs_journal *j = SB_JOURNAL(s);
956
957         if (atomic_read(&j->j_async_throttle)) {
958                 int depth;
959
960                 depth = reiserfs_write_unlock_nested(s);
961                 congestion_wait(BLK_RW_ASYNC, HZ / 10);
962                 reiserfs_write_lock_nested(s, depth);
963         }
964
965         return 0;
966 }
967
968 /*
969  * if this journal list still has commit blocks unflushed, send them to disk.
970  *
971  * log areas must be flushed in order (transaction 2 can't commit before
972  * transaction 1) Before the commit block can by written, every other log
973  * block must be safely on disk
974  */
975 static int flush_commit_list(struct super_block *s,
976                              struct reiserfs_journal_list *jl, int flushall)
977 {
978         int i;
979         b_blocknr_t bn;
980         struct buffer_head *tbh = NULL;
981         unsigned int trans_id = jl->j_trans_id;
982         struct reiserfs_journal *journal = SB_JOURNAL(s);
983         int retval = 0;
984         int write_len;
985         int depth;
986
987         reiserfs_check_lock_depth(s, "flush_commit_list");
988
989         if (atomic_read(&jl->j_older_commits_done)) {
990                 return 0;
991         }
992
993         /*
994          * before we can put our commit blocks on disk, we have to make
995          * sure everyone older than us is on disk too
996          */
997         BUG_ON(jl->j_len <= 0);
998         BUG_ON(trans_id == journal->j_trans_id);
999
1000         get_journal_list(jl);
1001         if (flushall) {
1002                 if (flush_older_commits(s, jl) == 1) {
1003                         /*
1004                          * list disappeared during flush_older_commits.
1005                          * return
1006                          */
1007                         goto put_jl;
1008                 }
1009         }
1010
1011         /* make sure nobody is trying to flush this one at the same time */
1012         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1013
1014         if (!journal_list_still_alive(s, trans_id)) {
1015                 mutex_unlock(&jl->j_commit_mutex);
1016                 goto put_jl;
1017         }
1018         BUG_ON(jl->j_trans_id == 0);
1019
1020         /* this commit is done, exit */
1021         if (atomic_read(&jl->j_commit_left) <= 0) {
1022                 if (flushall) {
1023                         atomic_set(&jl->j_older_commits_done, 1);
1024                 }
1025                 mutex_unlock(&jl->j_commit_mutex);
1026                 goto put_jl;
1027         }
1028
1029         if (!list_empty(&jl->j_bh_list)) {
1030                 int ret;
1031
1032                 /*
1033                  * We might sleep in numerous places inside
1034                  * write_ordered_buffers. Relax the write lock.
1035                  */
1036                 depth = reiserfs_write_unlock_nested(s);
1037                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1038                                             journal, jl, &jl->j_bh_list);
1039                 if (ret < 0 && retval == 0)
1040                         retval = ret;
1041                 reiserfs_write_lock_nested(s, depth);
1042         }
1043         BUG_ON(!list_empty(&jl->j_bh_list));
1044         /*
1045          * for the description block and all the log blocks, submit any buffers
1046          * that haven't already reached the disk.  Try to write at least 256
1047          * log blocks. later on, we will only wait on blocks that correspond
1048          * to this transaction, but while we're unplugging we might as well
1049          * get a chunk of data on there.
1050          */
1051         atomic_inc(&journal->j_async_throttle);
1052         write_len = jl->j_len + 1;
1053         if (write_len < 256)
1054                 write_len = 256;
1055         for (i = 0 ; i < write_len ; i++) {
1056                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1057                     SB_ONDISK_JOURNAL_SIZE(s);
1058                 tbh = journal_find_get_block(s, bn);
1059                 if (tbh) {
1060                         if (buffer_dirty(tbh)) {
1061                             depth = reiserfs_write_unlock_nested(s);
1062                             ll_rw_block(REQ_OP_WRITE, 0, 1, &tbh);
1063                             reiserfs_write_lock_nested(s, depth);
1064                         }
1065                         put_bh(tbh) ;
1066                 }
1067         }
1068         atomic_dec(&journal->j_async_throttle);
1069
1070         for (i = 0; i < (jl->j_len + 1); i++) {
1071                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1072                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1073                 tbh = journal_find_get_block(s, bn);
1074
1075                 depth = reiserfs_write_unlock_nested(s);
1076                 __wait_on_buffer(tbh);
1077                 reiserfs_write_lock_nested(s, depth);
1078                 /*
1079                  * since we're using ll_rw_blk above, it might have skipped
1080                  * over a locked buffer.  Double check here
1081                  */
1082                 /* redundant, sync_dirty_buffer() checks */
1083                 if (buffer_dirty(tbh)) {
1084                         depth = reiserfs_write_unlock_nested(s);
1085                         sync_dirty_buffer(tbh);
1086                         reiserfs_write_lock_nested(s, depth);
1087                 }
1088                 if (unlikely(!buffer_uptodate(tbh))) {
1089 #ifdef CONFIG_REISERFS_CHECK
1090                         reiserfs_warning(s, "journal-601",
1091                                          "buffer write failed");
1092 #endif
1093                         retval = -EIO;
1094                 }
1095                 /* once for journal_find_get_block */
1096                 put_bh(tbh);
1097                 /* once due to original getblk in do_journal_end */
1098                 put_bh(tbh);
1099                 atomic_dec(&jl->j_commit_left);
1100         }
1101
1102         BUG_ON(atomic_read(&jl->j_commit_left) != 1);
1103
1104         /*
1105          * If there was a write error in the journal - we can't commit
1106          * this transaction - it will be invalid and, if successful,
1107          * will just end up propagating the write error out to
1108          * the file system.
1109          */
1110         if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1111                 if (buffer_dirty(jl->j_commit_bh))
1112                         BUG();
1113                 mark_buffer_dirty(jl->j_commit_bh) ;
1114                 depth = reiserfs_write_unlock_nested(s);
1115                 if (reiserfs_barrier_flush(s))
1116                         __sync_dirty_buffer(jl->j_commit_bh,
1117                                         REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1118                 else
1119                         sync_dirty_buffer(jl->j_commit_bh);
1120                 reiserfs_write_lock_nested(s, depth);
1121         }
1122
1123         /*
1124          * If there was a write error in the journal - we can't commit this
1125          * transaction - it will be invalid and, if successful, will just end
1126          * up propagating the write error out to the filesystem.
1127          */
1128         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1129 #ifdef CONFIG_REISERFS_CHECK
1130                 reiserfs_warning(s, "journal-615", "buffer write failed");
1131 #endif
1132                 retval = -EIO;
1133         }
1134         bforget(jl->j_commit_bh);
1135         if (journal->j_last_commit_id != 0 &&
1136             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1137                 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1138                                  journal->j_last_commit_id, jl->j_trans_id);
1139         }
1140         journal->j_last_commit_id = jl->j_trans_id;
1141
1142         /*
1143          * now, every commit block is on the disk.  It is safe to allow
1144          * blocks freed during this transaction to be reallocated
1145          */
1146         cleanup_freed_for_journal_list(s, jl);
1147
1148         retval = retval ? retval : journal->j_errno;
1149
1150         /* mark the metadata dirty */
1151         if (!retval)
1152                 dirty_one_transaction(s, jl);
1153         atomic_dec(&jl->j_commit_left);
1154
1155         if (flushall) {
1156                 atomic_set(&jl->j_older_commits_done, 1);
1157         }
1158         mutex_unlock(&jl->j_commit_mutex);
1159 put_jl:
1160         put_journal_list(s, jl);
1161
1162         if (retval)
1163                 reiserfs_abort(s, retval, "Journal write error in %s",
1164                                __func__);
1165         return retval;
1166 }
1167
1168 /*
1169  * flush_journal_list frequently needs to find a newer transaction for a
1170  * given block.  This does that, or returns NULL if it can't find anything
1171  */
1172 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1173                                                           reiserfs_journal_cnode
1174                                                           *cn)
1175 {
1176         struct super_block *sb = cn->sb;
1177         b_blocknr_t blocknr = cn->blocknr;
1178
1179         cn = cn->hprev;
1180         while (cn) {
1181                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1182                         return cn->jlist;
1183                 }
1184                 cn = cn->hprev;
1185         }
1186         return NULL;
1187 }
1188
1189 static void remove_journal_hash(struct super_block *,
1190                                 struct reiserfs_journal_cnode **,
1191                                 struct reiserfs_journal_list *, unsigned long,
1192                                 int);
1193
1194 /*
1195  * once all the real blocks have been flushed, it is safe to remove them
1196  * from the journal list for this transaction.  Aside from freeing the
1197  * cnode, this also allows the block to be reallocated for data blocks
1198  * if it had been deleted.
1199  */
1200 static void remove_all_from_journal_list(struct super_block *sb,
1201                                          struct reiserfs_journal_list *jl,
1202                                          int debug)
1203 {
1204         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1205         struct reiserfs_journal_cnode *cn, *last;
1206         cn = jl->j_realblock;
1207
1208         /*
1209          * which is better, to lock once around the whole loop, or
1210          * to lock for each call to remove_journal_hash?
1211          */
1212         while (cn) {
1213                 if (cn->blocknr != 0) {
1214                         if (debug) {
1215                                 reiserfs_warning(sb, "reiserfs-2201",
1216                                                  "block %u, bh is %d, state %ld",
1217                                                  cn->blocknr, cn->bh ? 1 : 0,
1218                                                  cn->state);
1219                         }
1220                         cn->state = 0;
1221                         remove_journal_hash(sb, journal->j_list_hash_table,
1222                                             jl, cn->blocknr, 1);
1223                 }
1224                 last = cn;
1225                 cn = cn->next;
1226                 free_cnode(sb, last);
1227         }
1228         jl->j_realblock = NULL;
1229 }
1230
1231 /*
1232  * if this timestamp is greater than the timestamp we wrote last to the
1233  * header block, write it to the header block.  once this is done, I can
1234  * safely say the log area for this transaction won't ever be replayed,
1235  * and I can start releasing blocks in this transaction for reuse as data
1236  * blocks.  called by flush_journal_list, before it calls
1237  * remove_all_from_journal_list
1238  */
1239 static int _update_journal_header_block(struct super_block *sb,
1240                                         unsigned long offset,
1241                                         unsigned int trans_id)
1242 {
1243         struct reiserfs_journal_header *jh;
1244         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1245         int depth;
1246
1247         if (reiserfs_is_journal_aborted(journal))
1248                 return -EIO;
1249
1250         if (trans_id >= journal->j_last_flush_trans_id) {
1251                 if (buffer_locked((journal->j_header_bh))) {
1252                         depth = reiserfs_write_unlock_nested(sb);
1253                         __wait_on_buffer(journal->j_header_bh);
1254                         reiserfs_write_lock_nested(sb, depth);
1255                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1256 #ifdef CONFIG_REISERFS_CHECK
1257                                 reiserfs_warning(sb, "journal-699",
1258                                                  "buffer write failed");
1259 #endif
1260                                 return -EIO;
1261                         }
1262                 }
1263                 journal->j_last_flush_trans_id = trans_id;
1264                 journal->j_first_unflushed_offset = offset;
1265                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1266                                                         b_data);
1267                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1268                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1269                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1270
1271                 set_buffer_dirty(journal->j_header_bh);
1272                 depth = reiserfs_write_unlock_nested(sb);
1273
1274                 if (reiserfs_barrier_flush(sb))
1275                         __sync_dirty_buffer(journal->j_header_bh,
1276                                         REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1277                 else
1278                         sync_dirty_buffer(journal->j_header_bh);
1279
1280                 reiserfs_write_lock_nested(sb, depth);
1281                 if (!buffer_uptodate(journal->j_header_bh)) {
1282                         reiserfs_warning(sb, "journal-837",
1283                                          "IO error during journal replay");
1284                         return -EIO;
1285                 }
1286         }
1287         return 0;
1288 }
1289
1290 static int update_journal_header_block(struct super_block *sb,
1291                                        unsigned long offset,
1292                                        unsigned int trans_id)
1293 {
1294         return _update_journal_header_block(sb, offset, trans_id);
1295 }
1296
1297 /*
1298 ** flush any and all journal lists older than you are
1299 ** can only be called from flush_journal_list
1300 */
1301 static int flush_older_journal_lists(struct super_block *sb,
1302                                      struct reiserfs_journal_list *jl)
1303 {
1304         struct list_head *entry;
1305         struct reiserfs_journal_list *other_jl;
1306         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1307         unsigned int trans_id = jl->j_trans_id;
1308
1309         /*
1310          * we know we are the only ones flushing things, no extra race
1311          * protection is required.
1312          */
1313 restart:
1314         entry = journal->j_journal_list.next;
1315         /* Did we wrap? */
1316         if (entry == &journal->j_journal_list)
1317                 return 0;
1318         other_jl = JOURNAL_LIST_ENTRY(entry);
1319         if (other_jl->j_trans_id < trans_id) {
1320                 BUG_ON(other_jl->j_refcount <= 0);
1321                 /* do not flush all */
1322                 flush_journal_list(sb, other_jl, 0);
1323
1324                 /* other_jl is now deleted from the list */
1325                 goto restart;
1326         }
1327         return 0;
1328 }
1329
1330 static void del_from_work_list(struct super_block *s,
1331                                struct reiserfs_journal_list *jl)
1332 {
1333         struct reiserfs_journal *journal = SB_JOURNAL(s);
1334         if (!list_empty(&jl->j_working_list)) {
1335                 list_del_init(&jl->j_working_list);
1336                 journal->j_num_work_lists--;
1337         }
1338 }
1339
1340 /*
1341  * flush a journal list, both commit and real blocks
1342  *
1343  * always set flushall to 1, unless you are calling from inside
1344  * flush_journal_list
1345  *
1346  * IMPORTANT.  This can only be called while there are no journal writers,
1347  * and the journal is locked.  That means it can only be called from
1348  * do_journal_end, or by journal_release
1349  */
1350 static int flush_journal_list(struct super_block *s,
1351                               struct reiserfs_journal_list *jl, int flushall)
1352 {
1353         struct reiserfs_journal_list *pjl;
1354         struct reiserfs_journal_cnode *cn, *last;
1355         int count;
1356         int was_jwait = 0;
1357         int was_dirty = 0;
1358         struct buffer_head *saved_bh;
1359         unsigned long j_len_saved = jl->j_len;
1360         struct reiserfs_journal *journal = SB_JOURNAL(s);
1361         int err = 0;
1362         int depth;
1363
1364         BUG_ON(j_len_saved <= 0);
1365
1366         if (atomic_read(&journal->j_wcount) != 0) {
1367                 reiserfs_warning(s, "clm-2048", "called with wcount %d",
1368                                  atomic_read(&journal->j_wcount));
1369         }
1370
1371         /* if flushall == 0, the lock is already held */
1372         if (flushall) {
1373                 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1374         } else if (mutex_trylock(&journal->j_flush_mutex)) {
1375                 BUG();
1376         }
1377
1378         count = 0;
1379         if (j_len_saved > journal->j_trans_max) {
1380                 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1381                                j_len_saved, jl->j_trans_id);
1382                 return 0;
1383         }
1384
1385         /* if all the work is already done, get out of here */
1386         if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1387             atomic_read(&jl->j_commit_left) <= 0) {
1388                 goto flush_older_and_return;
1389         }
1390
1391         /*
1392          * start by putting the commit list on disk.  This will also flush
1393          * the commit lists of any olders transactions
1394          */
1395         flush_commit_list(s, jl, 1);
1396
1397         if (!(jl->j_state & LIST_DIRTY)
1398             && !reiserfs_is_journal_aborted(journal))
1399                 BUG();
1400
1401         /* are we done now? */
1402         if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1403             atomic_read(&jl->j_commit_left) <= 0) {
1404                 goto flush_older_and_return;
1405         }
1406
1407         /*
1408          * loop through each cnode, see if we need to write it,
1409          * or wait on a more recent transaction, or just ignore it
1410          */
1411         if (atomic_read(&journal->j_wcount) != 0) {
1412                 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1413                                "wcount is not 0");
1414         }
1415         cn = jl->j_realblock;
1416         while (cn) {
1417                 was_jwait = 0;
1418                 was_dirty = 0;
1419                 saved_bh = NULL;
1420                 /* blocknr of 0 is no longer in the hash, ignore it */
1421                 if (cn->blocknr == 0) {
1422                         goto free_cnode;
1423                 }
1424
1425                 /*
1426                  * This transaction failed commit.
1427                  * Don't write out to the disk
1428                  */
1429                 if (!(jl->j_state & LIST_DIRTY))
1430                         goto free_cnode;
1431
1432                 pjl = find_newer_jl_for_cn(cn);
1433                 /*
1434                  * the order is important here.  We check pjl to make sure we
1435                  * don't clear BH_JDirty_wait if we aren't the one writing this
1436                  * block to disk
1437                  */
1438                 if (!pjl && cn->bh) {
1439                         saved_bh = cn->bh;
1440
1441                         /*
1442                          * we do this to make sure nobody releases the
1443                          * buffer while we are working with it
1444                          */
1445                         get_bh(saved_bh);
1446
1447                         if (buffer_journal_dirty(saved_bh)) {
1448                                 BUG_ON(!can_dirty(cn));
1449                                 was_jwait = 1;
1450                                 was_dirty = 1;
1451                         } else if (can_dirty(cn)) {
1452                                 /*
1453                                  * everything with !pjl && jwait
1454                                  * should be writable
1455                                  */
1456                                 BUG();
1457                         }
1458                 }
1459
1460                 /*
1461                  * if someone has this block in a newer transaction, just make
1462                  * sure they are committed, and don't try writing it to disk
1463                  */
1464                 if (pjl) {
1465                         if (atomic_read(&pjl->j_commit_left))
1466                                 flush_commit_list(s, pjl, 1);
1467                         goto free_cnode;
1468                 }
1469
1470                 /*
1471                  * bh == NULL when the block got to disk on its own, OR,
1472                  * the block got freed in a future transaction
1473                  */
1474                 if (saved_bh == NULL) {
1475                         goto free_cnode;
1476                 }
1477
1478                 /*
1479                  * this should never happen.  kupdate_one_transaction has
1480                  * this list locked while it works, so we should never see a
1481                  * buffer here that is not marked JDirty_wait
1482                  */
1483                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1484                         reiserfs_warning(s, "journal-813",
1485                                          "BAD! buffer %llu %cdirty %cjwait, "
1486                                          "not in a newer transaction",
1487                                          (unsigned long long)saved_bh->
1488                                          b_blocknr, was_dirty ? ' ' : '!',
1489                                          was_jwait ? ' ' : '!');
1490                 }
1491                 if (was_dirty) {
1492                         /*
1493                          * we inc again because saved_bh gets decremented
1494                          * at free_cnode
1495                          */
1496                         get_bh(saved_bh);
1497                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1498                         lock_buffer(saved_bh);
1499                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1500                         if (buffer_dirty(saved_bh))
1501                                 submit_logged_buffer(saved_bh);
1502                         else
1503                                 unlock_buffer(saved_bh);
1504                         count++;
1505                 } else {
1506                         reiserfs_warning(s, "clm-2082",
1507                                          "Unable to flush buffer %llu in %s",
1508                                          (unsigned long long)saved_bh->
1509                                          b_blocknr, __func__);
1510                 }
1511 free_cnode:
1512                 last = cn;
1513                 cn = cn->next;
1514                 if (saved_bh) {
1515                         /*
1516                          * we incremented this to keep others from
1517                          * taking the buffer head away
1518                          */
1519                         put_bh(saved_bh);
1520                         if (atomic_read(&saved_bh->b_count) < 0) {
1521                                 reiserfs_warning(s, "journal-945",
1522                                                  "saved_bh->b_count < 0");
1523                         }
1524                 }
1525         }
1526         if (count > 0) {
1527                 cn = jl->j_realblock;
1528                 while (cn) {
1529                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1530                                 if (!cn->bh) {
1531                                         reiserfs_panic(s, "journal-1011",
1532                                                        "cn->bh is NULL");
1533                                 }
1534
1535                                 depth = reiserfs_write_unlock_nested(s);
1536                                 __wait_on_buffer(cn->bh);
1537                                 reiserfs_write_lock_nested(s, depth);
1538
1539                                 if (!cn->bh) {
1540                                         reiserfs_panic(s, "journal-1012",
1541                                                        "cn->bh is NULL");
1542                                 }
1543                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1544 #ifdef CONFIG_REISERFS_CHECK
1545                                         reiserfs_warning(s, "journal-949",
1546                                                          "buffer write failed");
1547 #endif
1548                                         err = -EIO;
1549                                 }
1550                                 /*
1551                                  * note, we must clear the JDirty_wait bit
1552                                  * after the up to date check, otherwise we
1553                                  * race against our flushpage routine
1554                                  */
1555                                 BUG_ON(!test_clear_buffer_journal_dirty
1556                                        (cn->bh));
1557
1558                                 /* drop one ref for us */
1559                                 put_bh(cn->bh);
1560                                 /* drop one ref for journal_mark_dirty */
1561                                 release_buffer_page(cn->bh);
1562                         }
1563                         cn = cn->next;
1564                 }
1565         }
1566
1567         if (err)
1568                 reiserfs_abort(s, -EIO,
1569                                "Write error while pushing transaction to disk in %s",
1570                                __func__);
1571 flush_older_and_return:
1572
1573         /*
1574          * before we can update the journal header block, we _must_ flush all
1575          * real blocks from all older transactions to disk.  This is because
1576          * once the header block is updated, this transaction will not be
1577          * replayed after a crash
1578          */
1579         if (flushall) {
1580                 flush_older_journal_lists(s, jl);
1581         }
1582
1583         err = journal->j_errno;
1584         /*
1585          * before we can remove everything from the hash tables for this
1586          * transaction, we must make sure it can never be replayed
1587          *
1588          * since we are only called from do_journal_end, we know for sure there
1589          * are no allocations going on while we are flushing journal lists.  So,
1590          * we only need to update the journal header block for the last list
1591          * being flushed
1592          */
1593         if (!err && flushall) {
1594                 err =
1595                     update_journal_header_block(s,
1596                                                 (jl->j_start + jl->j_len +
1597                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1598                                                 jl->j_trans_id);
1599                 if (err)
1600                         reiserfs_abort(s, -EIO,
1601                                        "Write error while updating journal header in %s",
1602                                        __func__);
1603         }
1604         remove_all_from_journal_list(s, jl, 0);
1605         list_del_init(&jl->j_list);
1606         journal->j_num_lists--;
1607         del_from_work_list(s, jl);
1608
1609         if (journal->j_last_flush_id != 0 &&
1610             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1611                 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1612                                  journal->j_last_flush_id, jl->j_trans_id);
1613         }
1614         journal->j_last_flush_id = jl->j_trans_id;
1615
1616         /*
1617          * not strictly required since we are freeing the list, but it should
1618          * help find code using dead lists later on
1619          */
1620         jl->j_len = 0;
1621         atomic_set(&jl->j_nonzerolen, 0);
1622         jl->j_start = 0;
1623         jl->j_realblock = NULL;
1624         jl->j_commit_bh = NULL;
1625         jl->j_trans_id = 0;
1626         jl->j_state = 0;
1627         put_journal_list(s, jl);
1628         if (flushall)
1629                 mutex_unlock(&journal->j_flush_mutex);
1630         return err;
1631 }
1632
1633 static int write_one_transaction(struct super_block *s,
1634                                  struct reiserfs_journal_list *jl,
1635                                  struct buffer_chunk *chunk)
1636 {
1637         struct reiserfs_journal_cnode *cn;
1638         int ret = 0;
1639
1640         jl->j_state |= LIST_TOUCHED;
1641         del_from_work_list(s, jl);
1642         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1643                 return 0;
1644         }
1645
1646         cn = jl->j_realblock;
1647         while (cn) {
1648                 /*
1649                  * if the blocknr == 0, this has been cleared from the hash,
1650                  * skip it
1651                  */
1652                 if (cn->blocknr == 0) {
1653                         goto next;
1654                 }
1655                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1656                         struct buffer_head *tmp_bh;
1657                         /*
1658                          * we can race against journal_mark_freed when we try
1659                          * to lock_buffer(cn->bh), so we have to inc the buffer
1660                          * count, and recheck things after locking
1661                          */
1662                         tmp_bh = cn->bh;
1663                         get_bh(tmp_bh);
1664                         lock_buffer(tmp_bh);
1665                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1666                                 if (!buffer_journal_dirty(tmp_bh) ||
1667                                     buffer_journal_prepared(tmp_bh))
1668                                         BUG();
1669                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1670                                 ret++;
1671                         } else {
1672                                 /* note, cn->bh might be null now */
1673                                 unlock_buffer(tmp_bh);
1674                         }
1675                         put_bh(tmp_bh);
1676                 }
1677 next:
1678                 cn = cn->next;
1679                 cond_resched();
1680         }
1681         return ret;
1682 }
1683
1684 /* used by flush_commit_list */
1685 static int dirty_one_transaction(struct super_block *s,
1686                                  struct reiserfs_journal_list *jl)
1687 {
1688         struct reiserfs_journal_cnode *cn;
1689         struct reiserfs_journal_list *pjl;
1690         int ret = 0;
1691
1692         jl->j_state |= LIST_DIRTY;
1693         cn = jl->j_realblock;
1694         while (cn) {
1695                 /*
1696                  * look for a more recent transaction that logged this
1697                  * buffer.  Only the most recent transaction with a buffer in
1698                  * it is allowed to send that buffer to disk
1699                  */
1700                 pjl = find_newer_jl_for_cn(cn);
1701                 if (!pjl && cn->blocknr && cn->bh
1702                     && buffer_journal_dirty(cn->bh)) {
1703                         BUG_ON(!can_dirty(cn));
1704                         /*
1705                          * if the buffer is prepared, it will either be logged
1706                          * or restored.  If restored, we need to make sure
1707                          * it actually gets marked dirty
1708                          */
1709                         clear_buffer_journal_new(cn->bh);
1710                         if (buffer_journal_prepared(cn->bh)) {
1711                                 set_buffer_journal_restore_dirty(cn->bh);
1712                         } else {
1713                                 set_buffer_journal_test(cn->bh);
1714                                 mark_buffer_dirty(cn->bh);
1715                         }
1716                 }
1717                 cn = cn->next;
1718         }
1719         return ret;
1720 }
1721
1722 static int kupdate_transactions(struct super_block *s,
1723                                 struct reiserfs_journal_list *jl,
1724                                 struct reiserfs_journal_list **next_jl,
1725                                 unsigned int *next_trans_id,
1726                                 int num_blocks, int num_trans)
1727 {
1728         int ret = 0;
1729         int written = 0;
1730         int transactions_flushed = 0;
1731         unsigned int orig_trans_id = jl->j_trans_id;
1732         struct buffer_chunk chunk;
1733         struct list_head *entry;
1734         struct reiserfs_journal *journal = SB_JOURNAL(s);
1735         chunk.nr = 0;
1736
1737         reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1738         if (!journal_list_still_alive(s, orig_trans_id)) {
1739                 goto done;
1740         }
1741
1742         /*
1743          * we've got j_flush_mutex held, nobody is going to delete any
1744          * of these lists out from underneath us
1745          */
1746         while ((num_trans && transactions_flushed < num_trans) ||
1747                (!num_trans && written < num_blocks)) {
1748
1749                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1750                     atomic_read(&jl->j_commit_left)
1751                     || !(jl->j_state & LIST_DIRTY)) {
1752                         del_from_work_list(s, jl);
1753                         break;
1754                 }
1755                 ret = write_one_transaction(s, jl, &chunk);
1756
1757                 if (ret < 0)
1758                         goto done;
1759                 transactions_flushed++;
1760                 written += ret;
1761                 entry = jl->j_list.next;
1762
1763                 /* did we wrap? */
1764                 if (entry == &journal->j_journal_list) {
1765                         break;
1766                 }
1767                 jl = JOURNAL_LIST_ENTRY(entry);
1768
1769                 /* don't bother with older transactions */
1770                 if (jl->j_trans_id <= orig_trans_id)
1771                         break;
1772         }
1773         if (chunk.nr) {
1774                 write_chunk(&chunk);
1775         }
1776
1777 done:
1778         mutex_unlock(&journal->j_flush_mutex);
1779         return ret;
1780 }
1781
1782 /*
1783  * for o_sync and fsync heavy applications, they tend to use
1784  * all the journa list slots with tiny transactions.  These
1785  * trigger lots and lots of calls to update the header block, which
1786  * adds seeks and slows things down.
1787  *
1788  * This function tries to clear out a large chunk of the journal lists
1789  * at once, which makes everything faster since only the newest journal
1790  * list updates the header block
1791  */
1792 static int flush_used_journal_lists(struct super_block *s,
1793                                     struct reiserfs_journal_list *jl)
1794 {
1795         unsigned long len = 0;
1796         unsigned long cur_len;
1797         int ret;
1798         int i;
1799         int limit = 256;
1800         struct reiserfs_journal_list *tjl;
1801         struct reiserfs_journal_list *flush_jl;
1802         unsigned int trans_id;
1803         struct reiserfs_journal *journal = SB_JOURNAL(s);
1804
1805         flush_jl = tjl = jl;
1806
1807         /* in data logging mode, try harder to flush a lot of blocks */
1808         if (reiserfs_data_log(s))
1809                 limit = 1024;
1810         /* flush for 256 transactions or limit blocks, whichever comes first */
1811         for (i = 0; i < 256 && len < limit; i++) {
1812                 if (atomic_read(&tjl->j_commit_left) ||
1813                     tjl->j_trans_id < jl->j_trans_id) {
1814                         break;
1815                 }
1816                 cur_len = atomic_read(&tjl->j_nonzerolen);
1817                 if (cur_len > 0) {
1818                         tjl->j_state &= ~LIST_TOUCHED;
1819                 }
1820                 len += cur_len;
1821                 flush_jl = tjl;
1822                 if (tjl->j_list.next == &journal->j_journal_list)
1823                         break;
1824                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1825         }
1826         get_journal_list(jl);
1827         get_journal_list(flush_jl);
1828
1829         /*
1830          * try to find a group of blocks we can flush across all the
1831          * transactions, but only bother if we've actually spanned
1832          * across multiple lists
1833          */
1834         if (flush_jl != jl) {
1835                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1836         }
1837         flush_journal_list(s, flush_jl, 1);
1838         put_journal_list(s, flush_jl);
1839         put_journal_list(s, jl);
1840         return 0;
1841 }
1842
1843 /*
1844  * removes any nodes in table with name block and dev as bh.
1845  * only touchs the hnext and hprev pointers.
1846  */
1847 void remove_journal_hash(struct super_block *sb,
1848                          struct reiserfs_journal_cnode **table,
1849                          struct reiserfs_journal_list *jl,
1850                          unsigned long block, int remove_freed)
1851 {
1852         struct reiserfs_journal_cnode *cur;
1853         struct reiserfs_journal_cnode **head;
1854
1855         head = &(journal_hash(table, sb, block));
1856         if (!head) {
1857                 return;
1858         }
1859         cur = *head;
1860         while (cur) {
1861                 if (cur->blocknr == block && cur->sb == sb
1862                     && (jl == NULL || jl == cur->jlist)
1863                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1864                         if (cur->hnext) {
1865                                 cur->hnext->hprev = cur->hprev;
1866                         }
1867                         if (cur->hprev) {
1868                                 cur->hprev->hnext = cur->hnext;
1869                         } else {
1870                                 *head = cur->hnext;
1871                         }
1872                         cur->blocknr = 0;
1873                         cur->sb = NULL;
1874                         cur->state = 0;
1875                         /*
1876                          * anybody who clears the cur->bh will also
1877                          * dec the nonzerolen
1878                          */
1879                         if (cur->bh && cur->jlist)
1880                                 atomic_dec(&cur->jlist->j_nonzerolen);
1881                         cur->bh = NULL;
1882                         cur->jlist = NULL;
1883                 }
1884                 cur = cur->hnext;
1885         }
1886 }
1887
1888 static void free_journal_ram(struct super_block *sb)
1889 {
1890         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1891         kfree(journal->j_current_jl);
1892         journal->j_num_lists--;
1893
1894         vfree(journal->j_cnode_free_orig);
1895         free_list_bitmaps(sb, journal->j_list_bitmap);
1896         free_bitmap_nodes(sb);  /* must be after free_list_bitmaps */
1897         if (journal->j_header_bh) {
1898                 brelse(journal->j_header_bh);
1899         }
1900         /*
1901          * j_header_bh is on the journal dev, make sure
1902          * not to release the journal dev until we brelse j_header_bh
1903          */
1904         release_journal_dev(sb, journal);
1905         vfree(journal);
1906 }
1907
1908 /*
1909  * call on unmount.  Only set error to 1 if you haven't made your way out
1910  * of read_super() yet.  Any other caller must keep error at 0.
1911  */
1912 static int do_journal_release(struct reiserfs_transaction_handle *th,
1913                               struct super_block *sb, int error)
1914 {
1915         struct reiserfs_transaction_handle myth;
1916         int flushed = 0;
1917         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1918
1919         /*
1920          * we only want to flush out transactions if we were
1921          * called with error == 0
1922          */
1923         if (!error && !sb_rdonly(sb)) {
1924                 /* end the current trans */
1925                 BUG_ON(!th->t_trans_id);
1926                 do_journal_end(th, FLUSH_ALL);
1927
1928                 /*
1929                  * make sure something gets logged to force
1930                  * our way into the flush code
1931                  */
1932                 if (!journal_join(&myth, sb)) {
1933                         reiserfs_prepare_for_journal(sb,
1934                                                      SB_BUFFER_WITH_SB(sb),
1935                                                      1);
1936                         journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1937                         do_journal_end(&myth, FLUSH_ALL);
1938                         flushed = 1;
1939                 }
1940         }
1941
1942         /* this also catches errors during the do_journal_end above */
1943         if (!error && reiserfs_is_journal_aborted(journal)) {
1944                 memset(&myth, 0, sizeof(myth));
1945                 if (!journal_join_abort(&myth, sb)) {
1946                         reiserfs_prepare_for_journal(sb,
1947                                                      SB_BUFFER_WITH_SB(sb),
1948                                                      1);
1949                         journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1950                         do_journal_end(&myth, FLUSH_ALL);
1951                 }
1952         }
1953
1954
1955         /*
1956          * We must release the write lock here because
1957          * the workqueue job (flush_async_commit) needs this lock
1958          */
1959         reiserfs_write_unlock(sb);
1960
1961         /*
1962          * Cancel flushing of old commits. Note that neither of these works
1963          * will be requeued because superblock is being shutdown and doesn't
1964          * have SB_ACTIVE set.
1965          */
1966         reiserfs_cancel_old_flush(sb);
1967         /* wait for all commits to finish */
1968         cancel_delayed_work_sync(&SB_JOURNAL(sb)->j_work);
1969
1970         free_journal_ram(sb);
1971
1972         reiserfs_write_lock(sb);
1973
1974         return 0;
1975 }
1976
1977 /* * call on unmount.  flush all journal trans, release all alloc'd ram */
1978 int journal_release(struct reiserfs_transaction_handle *th,
1979                     struct super_block *sb)
1980 {
1981         return do_journal_release(th, sb, 0);
1982 }
1983
1984 /* only call from an error condition inside reiserfs_read_super!  */
1985 int journal_release_error(struct reiserfs_transaction_handle *th,
1986                           struct super_block *sb)
1987 {
1988         return do_journal_release(th, sb, 1);
1989 }
1990
1991 /*
1992  * compares description block with commit block.
1993  * returns 1 if they differ, 0 if they are the same
1994  */
1995 static int journal_compare_desc_commit(struct super_block *sb,
1996                                        struct reiserfs_journal_desc *desc,
1997                                        struct reiserfs_journal_commit *commit)
1998 {
1999         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
2000             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
2001             get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
2002             get_commit_trans_len(commit) <= 0) {
2003                 return 1;
2004         }
2005         return 0;
2006 }
2007
2008 /*
2009  * returns 0 if it did not find a description block
2010  * returns -1 if it found a corrupt commit block
2011  * returns 1 if both desc and commit were valid
2012  * NOTE: only called during fs mount
2013  */
2014 static int journal_transaction_is_valid(struct super_block *sb,
2015                                         struct buffer_head *d_bh,
2016                                         unsigned int *oldest_invalid_trans_id,
2017                                         unsigned long *newest_mount_id)
2018 {
2019         struct reiserfs_journal_desc *desc;
2020         struct reiserfs_journal_commit *commit;
2021         struct buffer_head *c_bh;
2022         unsigned long offset;
2023
2024         if (!d_bh)
2025                 return 0;
2026
2027         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2028         if (get_desc_trans_len(desc) > 0
2029             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2030                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2031                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2032                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2033                                        "journal-986: transaction "
2034                                        "is valid returning because trans_id %d is greater than "
2035                                        "oldest_invalid %lu",
2036                                        get_desc_trans_id(desc),
2037                                        *oldest_invalid_trans_id);
2038                         return 0;
2039                 }
2040                 if (newest_mount_id
2041                     && *newest_mount_id > get_desc_mount_id(desc)) {
2042                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2043                                        "journal-1087: transaction "
2044                                        "is valid returning because mount_id %d is less than "
2045                                        "newest_mount_id %lu",
2046                                        get_desc_mount_id(desc),
2047                                        *newest_mount_id);
2048                         return -1;
2049                 }
2050                 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2051                         reiserfs_warning(sb, "journal-2018",
2052                                          "Bad transaction length %d "
2053                                          "encountered, ignoring transaction",
2054                                          get_desc_trans_len(desc));
2055                         return -1;
2056                 }
2057                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2058
2059                 /*
2060                  * ok, we have a journal description block,
2061                  * let's see if the transaction was valid
2062                  */
2063                 c_bh =
2064                     journal_bread(sb,
2065                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2066                                   ((offset + get_desc_trans_len(desc) +
2067                                     1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2068                 if (!c_bh)
2069                         return 0;
2070                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2071                 if (journal_compare_desc_commit(sb, desc, commit)) {
2072                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2073                                        "journal_transaction_is_valid, commit offset %ld had bad "
2074                                        "time %d or length %d",
2075                                        c_bh->b_blocknr -
2076                                        SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2077                                        get_commit_trans_id(commit),
2078                                        get_commit_trans_len(commit));
2079                         brelse(c_bh);
2080                         if (oldest_invalid_trans_id) {
2081                                 *oldest_invalid_trans_id =
2082                                     get_desc_trans_id(desc);
2083                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2084                                                "journal-1004: "
2085                                                "transaction_is_valid setting oldest invalid trans_id "
2086                                                "to %d",
2087                                                get_desc_trans_id(desc));
2088                         }
2089                         return -1;
2090                 }
2091                 brelse(c_bh);
2092                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2093                                "journal-1006: found valid "
2094                                "transaction start offset %llu, len %d id %d",
2095                                d_bh->b_blocknr -
2096                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2097                                get_desc_trans_len(desc),
2098                                get_desc_trans_id(desc));
2099                 return 1;
2100         } else {
2101                 return 0;
2102         }
2103 }
2104
2105 static void brelse_array(struct buffer_head **heads, int num)
2106 {
2107         int i;
2108         for (i = 0; i < num; i++) {
2109                 brelse(heads[i]);
2110         }
2111 }
2112
2113 /*
2114  * given the start, and values for the oldest acceptable transactions,
2115  * this either reads in a replays a transaction, or returns because the
2116  * transaction is invalid, or too old.
2117  * NOTE: only called during fs mount
2118  */
2119 static int journal_read_transaction(struct super_block *sb,
2120                                     unsigned long cur_dblock,
2121                                     unsigned long oldest_start,
2122                                     unsigned int oldest_trans_id,
2123                                     unsigned long newest_mount_id)
2124 {
2125         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2126         struct reiserfs_journal_desc *desc;
2127         struct reiserfs_journal_commit *commit;
2128         unsigned int trans_id = 0;
2129         struct buffer_head *c_bh;
2130         struct buffer_head *d_bh;
2131         struct buffer_head **log_blocks = NULL;
2132         struct buffer_head **real_blocks = NULL;
2133         unsigned int trans_offset;
2134         int i;
2135         int trans_half;
2136
2137         d_bh = journal_bread(sb, cur_dblock);
2138         if (!d_bh)
2139                 return 1;
2140         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2141         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2142         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2143                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2144                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2145                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2146         if (get_desc_trans_id(desc) < oldest_trans_id) {
2147                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2148                                "journal_read_trans skipping because %lu is too old",
2149                                cur_dblock -
2150                                SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2151                 brelse(d_bh);
2152                 return 1;
2153         }
2154         if (get_desc_mount_id(desc) != newest_mount_id) {
2155                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2156                                "journal_read_trans skipping because %d is != "
2157                                "newest_mount_id %lu", get_desc_mount_id(desc),
2158                                newest_mount_id);
2159                 brelse(d_bh);
2160                 return 1;
2161         }
2162         c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2163                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2164                               SB_ONDISK_JOURNAL_SIZE(sb)));
2165         if (!c_bh) {
2166                 brelse(d_bh);
2167                 return 1;
2168         }
2169         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2170         if (journal_compare_desc_commit(sb, desc, commit)) {
2171                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2172                                "journal_read_transaction, "
2173                                "commit offset %llu had bad time %d or length %d",
2174                                c_bh->b_blocknr -
2175                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2176                                get_commit_trans_id(commit),
2177                                get_commit_trans_len(commit));
2178                 brelse(c_bh);
2179                 brelse(d_bh);
2180                 return 1;
2181         }
2182
2183         if (bdev_read_only(sb->s_bdev)) {
2184                 reiserfs_warning(sb, "clm-2076",
2185                                  "device is readonly, unable to replay log");
2186                 brelse(c_bh);
2187                 brelse(d_bh);
2188                 return -EROFS;
2189         }
2190
2191         trans_id = get_desc_trans_id(desc);
2192         /*
2193          * now we know we've got a good transaction, and it was
2194          * inside the valid time ranges
2195          */
2196         log_blocks = kmalloc_array(get_desc_trans_len(desc),
2197                                    sizeof(struct buffer_head *),
2198                                    GFP_NOFS);
2199         real_blocks = kmalloc_array(get_desc_trans_len(desc),
2200                                     sizeof(struct buffer_head *),
2201                                     GFP_NOFS);
2202         if (!log_blocks || !real_blocks) {
2203                 brelse(c_bh);
2204                 brelse(d_bh);
2205                 kfree(log_blocks);
2206                 kfree(real_blocks);
2207                 reiserfs_warning(sb, "journal-1169",
2208                                  "kmalloc failed, unable to mount FS");
2209                 return -1;
2210         }
2211         /* get all the buffer heads */
2212         trans_half = journal_trans_half(sb->s_blocksize);
2213         for (i = 0; i < get_desc_trans_len(desc); i++) {
2214                 log_blocks[i] =
2215                     journal_getblk(sb,
2216                                    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2217                                    (trans_offset + 1 +
2218                                     i) % SB_ONDISK_JOURNAL_SIZE(sb));
2219                 if (i < trans_half) {
2220                         real_blocks[i] =
2221                             sb_getblk(sb,
2222                                       le32_to_cpu(desc->j_realblock[i]));
2223                 } else {
2224                         real_blocks[i] =
2225                             sb_getblk(sb,
2226                                       le32_to_cpu(commit->
2227                                                   j_realblock[i - trans_half]));
2228                 }
2229                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2230                         reiserfs_warning(sb, "journal-1207",
2231                                          "REPLAY FAILURE fsck required! "
2232                                          "Block to replay is outside of "
2233                                          "filesystem");
2234                         goto abort_replay;
2235                 }
2236                 /* make sure we don't try to replay onto log or reserved area */
2237                 if (is_block_in_log_or_reserved_area
2238                     (sb, real_blocks[i]->b_blocknr)) {
2239                         reiserfs_warning(sb, "journal-1204",
2240                                          "REPLAY FAILURE fsck required! "
2241                                          "Trying to replay onto a log block");
2242 abort_replay:
2243                         brelse_array(log_blocks, i);
2244                         brelse_array(real_blocks, i);
2245                         brelse(c_bh);
2246                         brelse(d_bh);
2247                         kfree(log_blocks);
2248                         kfree(real_blocks);
2249                         return -1;
2250                 }
2251         }
2252         /* read in the log blocks, memcpy to the corresponding real block */
2253         ll_rw_block(REQ_OP_READ, 0, get_desc_trans_len(desc), log_blocks);
2254         for (i = 0; i < get_desc_trans_len(desc); i++) {
2255
2256                 wait_on_buffer(log_blocks[i]);
2257                 if (!buffer_uptodate(log_blocks[i])) {
2258                         reiserfs_warning(sb, "journal-1212",
2259                                          "REPLAY FAILURE fsck required! "
2260                                          "buffer write failed");
2261                         brelse_array(log_blocks + i,
2262                                      get_desc_trans_len(desc) - i);
2263                         brelse_array(real_blocks, get_desc_trans_len(desc));
2264                         brelse(c_bh);
2265                         brelse(d_bh);
2266                         kfree(log_blocks);
2267                         kfree(real_blocks);
2268                         return -1;
2269                 }
2270                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2271                        real_blocks[i]->b_size);
2272                 set_buffer_uptodate(real_blocks[i]);
2273                 brelse(log_blocks[i]);
2274         }
2275         /* flush out the real blocks */
2276         for (i = 0; i < get_desc_trans_len(desc); i++) {
2277                 set_buffer_dirty(real_blocks[i]);
2278                 write_dirty_buffer(real_blocks[i], 0);
2279         }
2280         for (i = 0; i < get_desc_trans_len(desc); i++) {
2281                 wait_on_buffer(real_blocks[i]);
2282                 if (!buffer_uptodate(real_blocks[i])) {
2283                         reiserfs_warning(sb, "journal-1226",
2284                                          "REPLAY FAILURE, fsck required! "
2285                                          "buffer write failed");
2286                         brelse_array(real_blocks + i,
2287                                      get_desc_trans_len(desc) - i);
2288                         brelse(c_bh);
2289                         brelse(d_bh);
2290                         kfree(log_blocks);
2291                         kfree(real_blocks);
2292                         return -1;
2293                 }
2294                 brelse(real_blocks[i]);
2295         }
2296         cur_dblock =
2297             SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2298             ((trans_offset + get_desc_trans_len(desc) +
2299               2) % SB_ONDISK_JOURNAL_SIZE(sb));
2300         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2301                        "journal-1095: setting journal " "start to offset %ld",
2302                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2303
2304         /*
2305          * init starting values for the first transaction, in case
2306          * this is the last transaction to be replayed.
2307          */
2308         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2309         journal->j_last_flush_trans_id = trans_id;
2310         journal->j_trans_id = trans_id + 1;
2311         /* check for trans_id overflow */
2312         if (journal->j_trans_id == 0)
2313                 journal->j_trans_id = 10;
2314         brelse(c_bh);
2315         brelse(d_bh);
2316         kfree(log_blocks);
2317         kfree(real_blocks);
2318         return 0;
2319 }
2320
2321 /*
2322  * This function reads blocks starting from block and to max_block of bufsize
2323  * size (but no more than BUFNR blocks at a time). This proved to improve
2324  * mounting speed on self-rebuilding raid5 arrays at least.
2325  * Right now it is only used from journal code. But later we might use it
2326  * from other places.
2327  * Note: Do not use journal_getblk/sb_getblk functions here!
2328  */
2329 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2330                                            b_blocknr_t block, int bufsize,
2331                                            b_blocknr_t max_block)
2332 {
2333         struct buffer_head *bhlist[BUFNR];
2334         unsigned int blocks = BUFNR;
2335         struct buffer_head *bh;
2336         int i, j;
2337
2338         bh = __getblk(dev, block, bufsize);
2339         if (buffer_uptodate(bh))
2340                 return (bh);
2341
2342         if (block + BUFNR > max_block) {
2343                 blocks = max_block - block;
2344         }
2345         bhlist[0] = bh;
2346         j = 1;
2347         for (i = 1; i < blocks; i++) {
2348                 bh = __getblk(dev, block + i, bufsize);
2349                 if (buffer_uptodate(bh)) {
2350                         brelse(bh);
2351                         break;
2352                 } else
2353                         bhlist[j++] = bh;
2354         }
2355         ll_rw_block(REQ_OP_READ, 0, j, bhlist);
2356         for (i = 1; i < j; i++)
2357                 brelse(bhlist[i]);
2358         bh = bhlist[0];
2359         wait_on_buffer(bh);
2360         if (buffer_uptodate(bh))
2361                 return bh;
2362         brelse(bh);
2363         return NULL;
2364 }
2365
2366 /*
2367  * read and replay the log
2368  * on a clean unmount, the journal header's next unflushed pointer will be
2369  * to an invalid transaction.  This tests that before finding all the
2370  * transactions in the log, which makes normal mount times fast.
2371  *
2372  * After a crash, this starts with the next unflushed transaction, and
2373  * replays until it finds one too old, or invalid.
2374  *
2375  * On exit, it sets things up so the first transaction will work correctly.
2376  * NOTE: only called during fs mount
2377  */
2378 static int journal_read(struct super_block *sb)
2379 {
2380         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2381         struct reiserfs_journal_desc *desc;
2382         unsigned int oldest_trans_id = 0;
2383         unsigned int oldest_invalid_trans_id = 0;
2384         time64_t start;
2385         unsigned long oldest_start = 0;
2386         unsigned long cur_dblock = 0;
2387         unsigned long newest_mount_id = 9;
2388         struct buffer_head *d_bh;
2389         struct reiserfs_journal_header *jh;
2390         int valid_journal_header = 0;
2391         int replay_count = 0;
2392         int continue_replay = 1;
2393         int ret;
2394
2395         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2396         reiserfs_info(sb, "checking transaction log (%pg)\n",
2397                       journal->j_dev_bd);
2398         start = ktime_get_seconds();
2399
2400         /*
2401          * step 1, read in the journal header block.  Check the transaction
2402          * it says is the first unflushed, and if that transaction is not
2403          * valid, replay is done
2404          */
2405         journal->j_header_bh = journal_bread(sb,
2406                                              SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2407                                              + SB_ONDISK_JOURNAL_SIZE(sb));
2408         if (!journal->j_header_bh) {
2409                 return 1;
2410         }
2411         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2412         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2413             SB_ONDISK_JOURNAL_SIZE(sb)
2414             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2415                 oldest_start =
2416                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2417                     le32_to_cpu(jh->j_first_unflushed_offset);
2418                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2419                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2420                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2421                                "journal-1153: found in "
2422                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2423                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2424                                le32_to_cpu(jh->j_last_flush_trans_id));
2425                 valid_journal_header = 1;
2426
2427                 /*
2428                  * now, we try to read the first unflushed offset.  If it
2429                  * is not valid, there is nothing more we can do, and it
2430                  * makes no sense to read through the whole log.
2431                  */
2432                 d_bh =
2433                     journal_bread(sb,
2434                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2435                                   le32_to_cpu(jh->j_first_unflushed_offset));
2436                 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2437                 if (!ret) {
2438                         continue_replay = 0;
2439                 }
2440                 brelse(d_bh);
2441                 goto start_log_replay;
2442         }
2443
2444         /*
2445          * ok, there are transactions that need to be replayed.  start
2446          * with the first log block, find all the valid transactions, and
2447          * pick out the oldest.
2448          */
2449         while (continue_replay
2450                && cur_dblock <
2451                (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2452                 SB_ONDISK_JOURNAL_SIZE(sb))) {
2453                 /*
2454                  * Note that it is required for blocksize of primary fs
2455                  * device and journal device to be the same
2456                  */
2457                 d_bh =
2458                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2459                                     sb->s_blocksize,
2460                                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2461                                     SB_ONDISK_JOURNAL_SIZE(sb));
2462                 ret =
2463                     journal_transaction_is_valid(sb, d_bh,
2464                                                  &oldest_invalid_trans_id,
2465                                                  &newest_mount_id);
2466                 if (ret == 1) {
2467                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2468                         if (oldest_start == 0) {        /* init all oldest_ values */
2469                                 oldest_trans_id = get_desc_trans_id(desc);
2470                                 oldest_start = d_bh->b_blocknr;
2471                                 newest_mount_id = get_desc_mount_id(desc);
2472                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2473                                                "journal-1179: Setting "
2474                                                "oldest_start to offset %llu, trans_id %lu",
2475                                                oldest_start -
2476                                                SB_ONDISK_JOURNAL_1st_BLOCK
2477                                                (sb), oldest_trans_id);
2478                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2479                                 /* one we just read was older */
2480                                 oldest_trans_id = get_desc_trans_id(desc);
2481                                 oldest_start = d_bh->b_blocknr;
2482                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2483                                                "journal-1180: Resetting "
2484                                                "oldest_start to offset %lu, trans_id %lu",
2485                                                oldest_start -
2486                                                SB_ONDISK_JOURNAL_1st_BLOCK
2487                                                (sb), oldest_trans_id);
2488                         }
2489                         if (newest_mount_id < get_desc_mount_id(desc)) {
2490                                 newest_mount_id = get_desc_mount_id(desc);
2491                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2492                                                "journal-1299: Setting "
2493                                                "newest_mount_id to %d",
2494                                                get_desc_mount_id(desc));
2495                         }
2496                         cur_dblock += get_desc_trans_len(desc) + 2;
2497                 } else {
2498                         cur_dblock++;
2499                 }
2500                 brelse(d_bh);
2501         }
2502
2503 start_log_replay:
2504         cur_dblock = oldest_start;
2505         if (oldest_trans_id) {
2506                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2507                                "journal-1206: Starting replay "
2508                                "from offset %llu, trans_id %lu",
2509                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2510                                oldest_trans_id);
2511
2512         }
2513         replay_count = 0;
2514         while (continue_replay && oldest_trans_id > 0) {
2515                 ret =
2516                     journal_read_transaction(sb, cur_dblock, oldest_start,
2517                                              oldest_trans_id, newest_mount_id);
2518                 if (ret < 0) {
2519                         return ret;
2520                 } else if (ret != 0) {
2521                         break;
2522                 }
2523                 cur_dblock =
2524                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2525                 replay_count++;
2526                 if (cur_dblock == oldest_start)
2527                         break;
2528         }
2529
2530         if (oldest_trans_id == 0) {
2531                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2532                                "journal-1225: No valid " "transactions found");
2533         }
2534         /*
2535          * j_start does not get set correctly if we don't replay any
2536          * transactions.  if we had a valid journal_header, set j_start
2537          * to the first unflushed transaction value, copy the trans_id
2538          * from the header
2539          */
2540         if (valid_journal_header && replay_count == 0) {
2541                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2542                 journal->j_trans_id =
2543                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2544                 /* check for trans_id overflow */
2545                 if (journal->j_trans_id == 0)
2546                         journal->j_trans_id = 10;
2547                 journal->j_last_flush_trans_id =
2548                     le32_to_cpu(jh->j_last_flush_trans_id);
2549                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2550         } else {
2551                 journal->j_mount_id = newest_mount_id + 1;
2552         }
2553         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2554                        "newest_mount_id to %lu", journal->j_mount_id);
2555         journal->j_first_unflushed_offset = journal->j_start;
2556         if (replay_count > 0) {
2557                 reiserfs_info(sb,
2558                               "replayed %d transactions in %lu seconds\n",
2559                               replay_count, ktime_get_seconds() - start);
2560         }
2561         /* needed to satisfy the locking in _update_journal_header_block */
2562         reiserfs_write_lock(sb);
2563         if (!bdev_read_only(sb->s_bdev) &&
2564             _update_journal_header_block(sb, journal->j_start,
2565                                          journal->j_last_flush_trans_id)) {
2566                 reiserfs_write_unlock(sb);
2567                 /*
2568                  * replay failed, caller must call free_journal_ram and abort
2569                  * the mount
2570                  */
2571                 return -1;
2572         }
2573         reiserfs_write_unlock(sb);
2574         return 0;
2575 }
2576
2577 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2578 {
2579         struct reiserfs_journal_list *jl;
2580         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2581                      GFP_NOFS | __GFP_NOFAIL);
2582         INIT_LIST_HEAD(&jl->j_list);
2583         INIT_LIST_HEAD(&jl->j_working_list);
2584         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2585         INIT_LIST_HEAD(&jl->j_bh_list);
2586         mutex_init(&jl->j_commit_mutex);
2587         SB_JOURNAL(s)->j_num_lists++;
2588         get_journal_list(jl);
2589         return jl;
2590 }
2591
2592 static void journal_list_init(struct super_block *sb)
2593 {
2594         SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2595 }
2596
2597 static void release_journal_dev(struct super_block *super,
2598                                struct reiserfs_journal *journal)
2599 {
2600         if (journal->j_dev_bd != NULL) {
2601                 blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2602                 journal->j_dev_bd = NULL;
2603         }
2604 }
2605
2606 static int journal_init_dev(struct super_block *super,
2607                             struct reiserfs_journal *journal,
2608                             const char *jdev_name)
2609 {
2610         int result;
2611         dev_t jdev;
2612         fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
2613         char b[BDEVNAME_SIZE];
2614
2615         result = 0;
2616
2617         journal->j_dev_bd = NULL;
2618         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2619             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2620
2621         if (bdev_read_only(super->s_bdev))
2622                 blkdev_mode = FMODE_READ;
2623
2624         /* there is no "jdev" option and journal is on separate device */
2625         if ((!jdev_name || !jdev_name[0])) {
2626                 if (jdev == super->s_dev)
2627                         blkdev_mode &= ~FMODE_EXCL;
2628                 journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2629                                                       journal);
2630                 journal->j_dev_mode = blkdev_mode;
2631                 if (IS_ERR(journal->j_dev_bd)) {
2632                         result = PTR_ERR(journal->j_dev_bd);
2633                         journal->j_dev_bd = NULL;
2634                         reiserfs_warning(super, "sh-458",
2635                                          "cannot init journal device '%s': %i",
2636                                          __bdevname(jdev, b), result);
2637                         return result;
2638                 } else if (jdev != super->s_dev)
2639                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2640
2641                 return 0;
2642         }
2643
2644         journal->j_dev_mode = blkdev_mode;
2645         journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
2646         if (IS_ERR(journal->j_dev_bd)) {
2647                 result = PTR_ERR(journal->j_dev_bd);
2648                 journal->j_dev_bd = NULL;
2649                 reiserfs_warning(super, "sh-457",
2650                                  "journal_init_dev: Cannot open '%s': %i",
2651                                  jdev_name, result);
2652                 return result;
2653         }
2654
2655         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2656         reiserfs_info(super,
2657                       "journal_init_dev: journal device: %pg\n",
2658                       journal->j_dev_bd);
2659         return 0;
2660 }
2661
2662 /*
2663  * When creating/tuning a file system user can assign some
2664  * journal params within boundaries which depend on the ratio
2665  * blocksize/standard_blocksize.
2666  *
2667  * For blocks >= standard_blocksize transaction size should
2668  * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2669  * then JOURNAL_TRANS_MAX_DEFAULT.
2670  *
2671  * For blocks < standard_blocksize these boundaries should be
2672  * decreased proportionally.
2673  */
2674 #define REISERFS_STANDARD_BLKSIZE (4096)
2675
2676 static int check_advise_trans_params(struct super_block *sb,
2677                                      struct reiserfs_journal *journal)
2678 {
2679         if (journal->j_trans_max) {
2680                 /* Non-default journal params.  Do sanity check for them. */
2681                 int ratio = 1;
2682                 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2683                         ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2684
2685                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2686                     journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2687                     SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2688                     JOURNAL_MIN_RATIO) {
2689                         reiserfs_warning(sb, "sh-462",
2690                                          "bad transaction max size (%u). "
2691                                          "FSCK?", journal->j_trans_max);
2692                         return 1;
2693                 }
2694                 if (journal->j_max_batch != (journal->j_trans_max) *
2695                         JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2696                         reiserfs_warning(sb, "sh-463",
2697                                          "bad transaction max batch (%u). "
2698                                          "FSCK?", journal->j_max_batch);
2699                         return 1;
2700                 }
2701         } else {
2702                 /*
2703                  * Default journal params.
2704                  * The file system was created by old version
2705                  * of mkreiserfs, so some fields contain zeros,
2706                  * and we need to advise proper values for them
2707                  */
2708                 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2709                         reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2710                                          sb->s_blocksize);
2711                         return 1;
2712                 }
2713                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2714                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2715                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2716         }
2717         return 0;
2718 }
2719
2720 /* must be called once on fs mount.  calls journal_read for you */
2721 int journal_init(struct super_block *sb, const char *j_dev_name,
2722                  int old_format, unsigned int commit_max_age)
2723 {
2724         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2725         struct buffer_head *bhjh;
2726         struct reiserfs_super_block *rs;
2727         struct reiserfs_journal_header *jh;
2728         struct reiserfs_journal *journal;
2729         struct reiserfs_journal_list *jl;
2730         int ret;
2731
2732         journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal));
2733         if (!journal) {
2734                 reiserfs_warning(sb, "journal-1256",
2735                                  "unable to get memory for journal structure");
2736                 return 1;
2737         }
2738         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2739         INIT_LIST_HEAD(&journal->j_prealloc_list);
2740         INIT_LIST_HEAD(&journal->j_working_list);
2741         INIT_LIST_HEAD(&journal->j_journal_list);
2742         journal->j_persistent_trans = 0;
2743         if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2744                                            reiserfs_bmap_count(sb)))
2745                 goto free_and_return;
2746
2747         allocate_bitmap_nodes(sb);
2748
2749         /* reserved for journal area support */
2750         SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2751                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2752                                                  / sb->s_blocksize +
2753                                                  reiserfs_bmap_count(sb) +
2754                                                  1 :
2755                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2756                                                  sb->s_blocksize + 2);
2757
2758         /*
2759          * Sanity check to see is the standard journal fitting
2760          * within first bitmap (actual for small blocksizes)
2761          */
2762         if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2763             (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2764              SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2765                 reiserfs_warning(sb, "journal-1393",
2766                                  "journal does not fit for area addressed "
2767                                  "by first of bitmap blocks. It starts at "
2768                                  "%u and its size is %u. Block size %ld",
2769                                  SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2770                                  SB_ONDISK_JOURNAL_SIZE(sb),
2771                                  sb->s_blocksize);
2772                 goto free_and_return;
2773         }
2774
2775         /*
2776          * Sanity check to see if journal first block is correct.
2777          * If journal first block is invalid it can cause
2778          * zeroing important superblock members.
2779          */
2780         if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2781             SB_ONDISK_JOURNAL_1st_BLOCK(sb) < SB_JOURNAL_1st_RESERVED_BLOCK(sb)) {
2782                 reiserfs_warning(sb, "journal-1393",
2783                                  "journal 1st super block is invalid: 1st reserved block %d, but actual 1st block is %d",
2784                                  SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2785                                  SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2786                 goto free_and_return;
2787         }
2788
2789         if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2790                 reiserfs_warning(sb, "sh-462",
2791                                  "unable to initialize journal device");
2792                 goto free_and_return;
2793         }
2794
2795         rs = SB_DISK_SUPER_BLOCK(sb);
2796
2797         /* read journal header */
2798         bhjh = journal_bread(sb,
2799                              SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2800                              SB_ONDISK_JOURNAL_SIZE(sb));
2801         if (!bhjh) {
2802                 reiserfs_warning(sb, "sh-459",
2803                                  "unable to read journal header");
2804                 goto free_and_return;
2805         }
2806         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2807
2808         /* make sure that journal matches to the super block */
2809         if (is_reiserfs_jr(rs)
2810             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2811                 sb_jp_journal_magic(rs))) {
2812                 reiserfs_warning(sb, "sh-460",
2813                                  "journal header magic %x (device %pg) does "
2814                                  "not match to magic found in super block %x",
2815                                  jh->jh_journal.jp_journal_magic,
2816                                  journal->j_dev_bd,
2817                                  sb_jp_journal_magic(rs));
2818                 brelse(bhjh);
2819                 goto free_and_return;
2820         }
2821
2822         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2823         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2824         journal->j_max_commit_age =
2825             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2826         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2827
2828         if (check_advise_trans_params(sb, journal) != 0)
2829                 goto free_and_return;
2830         journal->j_default_max_commit_age = journal->j_max_commit_age;
2831
2832         if (commit_max_age != 0) {
2833                 journal->j_max_commit_age = commit_max_age;
2834                 journal->j_max_trans_age = commit_max_age;
2835         }
2836
2837         reiserfs_info(sb, "journal params: device %pg, size %u, "
2838                       "journal first block %u, max trans len %u, max batch %u, "
2839                       "max commit age %u, max trans age %u\n",
2840                       journal->j_dev_bd,
2841                       SB_ONDISK_JOURNAL_SIZE(sb),
2842                       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2843                       journal->j_trans_max,
2844                       journal->j_max_batch,
2845                       journal->j_max_commit_age, journal->j_max_trans_age);
2846
2847         brelse(bhjh);
2848
2849         journal->j_list_bitmap_index = 0;
2850         journal_list_init(sb);
2851
2852         memset(journal->j_list_hash_table, 0,
2853                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2854
2855         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2856         spin_lock_init(&journal->j_dirty_buffers_lock);
2857
2858         journal->j_start = 0;
2859         journal->j_len = 0;
2860         journal->j_len_alloc = 0;
2861         atomic_set(&journal->j_wcount, 0);
2862         atomic_set(&journal->j_async_throttle, 0);
2863         journal->j_bcount = 0;
2864         journal->j_trans_start_time = 0;
2865         journal->j_last = NULL;
2866         journal->j_first = NULL;
2867         init_waitqueue_head(&journal->j_join_wait);
2868         mutex_init(&journal->j_mutex);
2869         mutex_init(&journal->j_flush_mutex);
2870
2871         journal->j_trans_id = 10;
2872         journal->j_mount_id = 10;
2873         journal->j_state = 0;
2874         atomic_set(&journal->j_jlock, 0);
2875         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2876         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2877         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2878         journal->j_cnode_used = 0;
2879         journal->j_must_wait = 0;
2880
2881         if (journal->j_cnode_free == 0) {
2882                 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2883                                  "allocation failed (%ld bytes). Journal is "
2884                                  "too large for available memory. Usually "
2885                                  "this is due to a journal that is too large.",
2886                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2887                 goto free_and_return;
2888         }
2889
2890         init_journal_hash(sb);
2891         jl = journal->j_current_jl;
2892
2893         /*
2894          * get_list_bitmap() may call flush_commit_list() which
2895          * requires the lock. Calling flush_commit_list() shouldn't happen
2896          * this early but I like to be paranoid.
2897          */
2898         reiserfs_write_lock(sb);
2899         jl->j_list_bitmap = get_list_bitmap(sb, jl);
2900         reiserfs_write_unlock(sb);
2901         if (!jl->j_list_bitmap) {
2902                 reiserfs_warning(sb, "journal-2005",
2903                                  "get_list_bitmap failed for journal list 0");
2904                 goto free_and_return;
2905         }
2906
2907         ret = journal_read(sb);
2908         if (ret < 0) {
2909                 reiserfs_warning(sb, "reiserfs-2006",
2910                                  "Replay Failure, unable to mount");
2911                 goto free_and_return;
2912         }
2913
2914         INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2915         journal->j_work_sb = sb;
2916         return 0;
2917 free_and_return:
2918         free_journal_ram(sb);
2919         return 1;
2920 }
2921
2922 /*
2923  * test for a polite end of the current transaction.  Used by file_write,
2924  * and should be used by delete to make sure they don't write more than
2925  * can fit inside a single transaction
2926  */
2927 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2928                                    int new_alloc)
2929 {
2930         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2931         time64_t now = ktime_get_seconds();
2932         /* cannot restart while nested */
2933         BUG_ON(!th->t_trans_id);
2934         if (th->t_refcount > 1)
2935                 return 0;
2936         if (journal->j_must_wait > 0 ||
2937             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2938             atomic_read(&journal->j_jlock) ||
2939             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2940             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2941                 return 1;
2942         }
2943
2944         journal->j_len_alloc += new_alloc;
2945         th->t_blocks_allocated += new_alloc ;
2946         return 0;
2947 }
2948
2949 /* this must be called inside a transaction */
2950 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2951 {
2952         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2953         BUG_ON(!th->t_trans_id);
2954         journal->j_must_wait = 1;
2955         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2956         return;
2957 }
2958
2959 /* this must be called without a transaction started */
2960 void reiserfs_allow_writes(struct super_block *s)
2961 {
2962         struct reiserfs_journal *journal = SB_JOURNAL(s);
2963         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2964         wake_up(&journal->j_join_wait);
2965 }
2966
2967 /* this must be called without a transaction started */
2968 void reiserfs_wait_on_write_block(struct super_block *s)
2969 {
2970         struct reiserfs_journal *journal = SB_JOURNAL(s);
2971         wait_event(journal->j_join_wait,
2972                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2973 }
2974
2975 static void queue_log_writer(struct super_block *s)
2976 {
2977         wait_queue_entry_t wait;
2978         struct reiserfs_journal *journal = SB_JOURNAL(s);
2979         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2980
2981         /*
2982          * we don't want to use wait_event here because
2983          * we only want to wait once.
2984          */
2985         init_waitqueue_entry(&wait, current);
2986         add_wait_queue(&journal->j_join_wait, &wait);
2987         set_current_state(TASK_UNINTERRUPTIBLE);
2988         if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
2989                 int depth = reiserfs_write_unlock_nested(s);
2990                 schedule();
2991                 reiserfs_write_lock_nested(s, depth);
2992         }
2993         __set_current_state(TASK_RUNNING);
2994         remove_wait_queue(&journal->j_join_wait, &wait);
2995 }
2996
2997 static void wake_queued_writers(struct super_block *s)
2998 {
2999         struct reiserfs_journal *journal = SB_JOURNAL(s);
3000         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
3001                 wake_up(&journal->j_join_wait);
3002 }
3003
3004 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
3005 {
3006         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3007         unsigned long bcount = journal->j_bcount;
3008         while (1) {
3009                 int depth;
3010
3011                 depth = reiserfs_write_unlock_nested(sb);
3012                 schedule_timeout_uninterruptible(1);
3013                 reiserfs_write_lock_nested(sb, depth);
3014
3015                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
3016                 while ((atomic_read(&journal->j_wcount) > 0 ||
3017                         atomic_read(&journal->j_jlock)) &&
3018                        journal->j_trans_id == trans_id) {
3019                         queue_log_writer(sb);
3020                 }
3021                 if (journal->j_trans_id != trans_id)
3022                         break;
3023                 if (bcount == journal->j_bcount)
3024                         break;
3025                 bcount = journal->j_bcount;
3026         }
3027 }
3028
3029 /*
3030  * join == true if you must join an existing transaction.
3031  * join == false if you can deal with waiting for others to finish
3032  *
3033  * this will block until the transaction is joinable.  send the number of
3034  * blocks you expect to use in nblocks.
3035 */
3036 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3037                               struct super_block *sb, unsigned long nblocks,
3038                               int join)
3039 {
3040         time64_t now = ktime_get_seconds();
3041         unsigned int old_trans_id;
3042         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3043         struct reiserfs_transaction_handle myth;
3044         int sched_count = 0;
3045         int retval;
3046         int depth;
3047
3048         reiserfs_check_lock_depth(sb, "journal_begin");
3049         BUG_ON(nblocks > journal->j_trans_max);
3050
3051         PROC_INFO_INC(sb, journal.journal_being);
3052         /* set here for journal_join */
3053         th->t_refcount = 1;
3054         th->t_super = sb;
3055
3056 relock:
3057         lock_journal(sb);
3058         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3059                 unlock_journal(sb);
3060                 retval = journal->j_errno;
3061                 goto out_fail;
3062         }
3063         journal->j_bcount++;
3064
3065         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3066                 unlock_journal(sb);
3067                 depth = reiserfs_write_unlock_nested(sb);
3068                 reiserfs_wait_on_write_block(sb);
3069                 reiserfs_write_lock_nested(sb, depth);
3070                 PROC_INFO_INC(sb, journal.journal_relock_writers);
3071                 goto relock;
3072         }
3073         now = ktime_get_seconds();
3074
3075         /*
3076          * if there is no room in the journal OR
3077          * if this transaction is too old, and we weren't called joinable,
3078          * wait for it to finish before beginning we don't sleep if there
3079          * aren't other writers
3080          */
3081
3082         if ((!join && journal->j_must_wait > 0) ||
3083             (!join
3084              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3085             || (!join && atomic_read(&journal->j_wcount) > 0
3086                 && journal->j_trans_start_time > 0
3087                 && (now - journal->j_trans_start_time) >
3088                 journal->j_max_trans_age) || (!join
3089                                               && atomic_read(&journal->j_jlock))
3090             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3091
3092                 old_trans_id = journal->j_trans_id;
3093                 /* allow others to finish this transaction */
3094                 unlock_journal(sb);
3095
3096                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3097                     journal->j_max_batch &&
3098                     ((journal->j_len + nblocks + 2) * 100) <
3099                     (journal->j_len_alloc * 75)) {
3100                         if (atomic_read(&journal->j_wcount) > 10) {
3101                                 sched_count++;
3102                                 queue_log_writer(sb);
3103                                 goto relock;
3104                         }
3105                 }
3106                 /*
3107                  * don't mess with joining the transaction if all we
3108                  * have to do is wait for someone else to do a commit
3109                  */
3110                 if (atomic_read(&journal->j_jlock)) {
3111                         while (journal->j_trans_id == old_trans_id &&
3112                                atomic_read(&journal->j_jlock)) {
3113                                 queue_log_writer(sb);
3114                         }
3115                         goto relock;
3116                 }
3117                 retval = journal_join(&myth, sb);
3118                 if (retval)
3119                         goto out_fail;
3120
3121                 /* someone might have ended the transaction while we joined */
3122                 if (old_trans_id != journal->j_trans_id) {
3123                         retval = do_journal_end(&myth, 0);
3124                 } else {
3125                         retval = do_journal_end(&myth, COMMIT_NOW);
3126                 }
3127
3128                 if (retval)
3129                         goto out_fail;
3130
3131                 PROC_INFO_INC(sb, journal.journal_relock_wcount);
3132                 goto relock;
3133         }
3134         /* we are the first writer, set trans_id */
3135         if (journal->j_trans_start_time == 0) {
3136                 journal->j_trans_start_time = ktime_get_seconds();
3137         }
3138         atomic_inc(&journal->j_wcount);
3139         journal->j_len_alloc += nblocks;
3140         th->t_blocks_logged = 0;
3141         th->t_blocks_allocated = nblocks;
3142         th->t_trans_id = journal->j_trans_id;
3143         unlock_journal(sb);
3144         INIT_LIST_HEAD(&th->t_list);
3145         return 0;
3146
3147 out_fail:
3148         memset(th, 0, sizeof(*th));
3149         /*
3150          * Re-set th->t_super, so we can properly keep track of how many
3151          * persistent transactions there are. We need to do this so if this
3152          * call is part of a failed restart_transaction, we can free it later
3153          */
3154         th->t_super = sb;
3155         return retval;
3156 }
3157
3158 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3159                                                                     super_block
3160                                                                     *s,
3161                                                                     int nblocks)
3162 {
3163         int ret;
3164         struct reiserfs_transaction_handle *th;
3165
3166         /*
3167          * if we're nesting into an existing transaction.  It will be
3168          * persistent on its own
3169          */
3170         if (reiserfs_transaction_running(s)) {
3171                 th = current->journal_info;
3172                 th->t_refcount++;
3173                 BUG_ON(th->t_refcount < 2);
3174
3175                 return th;
3176         }
3177         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3178         if (!th)
3179                 return NULL;
3180         ret = journal_begin(th, s, nblocks);
3181         if (ret) {
3182                 kfree(th);
3183                 return NULL;
3184         }
3185
3186         SB_JOURNAL(s)->j_persistent_trans++;
3187         return th;
3188 }
3189
3190 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3191 {
3192         struct super_block *s = th->t_super;
3193         int ret = 0;
3194         if (th->t_trans_id)
3195                 ret = journal_end(th);
3196         else
3197                 ret = -EIO;
3198         if (th->t_refcount == 0) {
3199                 SB_JOURNAL(s)->j_persistent_trans--;
3200                 kfree(th);
3201         }
3202         return ret;
3203 }
3204
3205 static int journal_join(struct reiserfs_transaction_handle *th,
3206                         struct super_block *sb)
3207 {
3208         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3209
3210         /*
3211          * this keeps do_journal_end from NULLing out the
3212          * current->journal_info pointer
3213          */
3214         th->t_handle_save = cur_th;
3215         BUG_ON(cur_th && cur_th->t_refcount > 1);
3216         return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN);
3217 }
3218
3219 int journal_join_abort(struct reiserfs_transaction_handle *th,
3220                        struct super_block *sb)
3221 {
3222         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3223
3224         /*
3225          * this keeps do_journal_end from NULLing out the
3226          * current->journal_info pointer
3227          */
3228         th->t_handle_save = cur_th;
3229         BUG_ON(cur_th && cur_th->t_refcount > 1);
3230         return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT);
3231 }
3232
3233 int journal_begin(struct reiserfs_transaction_handle *th,
3234                   struct super_block *sb, unsigned long nblocks)
3235 {
3236         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3237         int ret;
3238
3239         th->t_handle_save = NULL;
3240         if (cur_th) {
3241                 /* we are nesting into the current transaction */
3242                 if (cur_th->t_super == sb) {
3243                         BUG_ON(!cur_th->t_refcount);
3244                         cur_th->t_refcount++;
3245                         memcpy(th, cur_th, sizeof(*th));
3246                         if (th->t_refcount <= 1)
3247                                 reiserfs_warning(sb, "reiserfs-2005",
3248                                                  "BAD: refcount <= 1, but "
3249                                                  "journal_info != 0");
3250                         return 0;
3251                 } else {
3252                         /*
3253                          * we've ended up with a handle from a different
3254                          * filesystem.  save it and restore on journal_end.
3255                          * This should never really happen...
3256                          */
3257                         reiserfs_warning(sb, "clm-2100",
3258                                          "nesting info a different FS");
3259                         th->t_handle_save = current->journal_info;
3260                         current->journal_info = th;
3261                 }
3262         } else {
3263                 current->journal_info = th;
3264         }
3265         ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3266         BUG_ON(current->journal_info != th);
3267
3268         /*
3269          * I guess this boils down to being the reciprocal of clm-2100 above.
3270          * If do_journal_begin_r fails, we need to put it back, since
3271          * journal_end won't be called to do it. */
3272         if (ret)
3273                 current->journal_info = th->t_handle_save;
3274         else
3275                 BUG_ON(!th->t_refcount);
3276
3277         return ret;
3278 }
3279
3280 /*
3281  * puts bh into the current transaction.  If it was already there, reorders
3282  * removes the old pointers from the hash, and puts new ones in (to make
3283  * sure replay happen in the right order).
3284  *
3285  * if it was dirty, cleans and files onto the clean list.  I can't let it
3286  * be dirty again until the transaction is committed.
3287  *
3288  * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3289  */
3290 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3291                        struct buffer_head *bh)
3292 {
3293         struct super_block *sb = th->t_super;
3294         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3295         struct reiserfs_journal_cnode *cn = NULL;
3296         int count_already_incd = 0;
3297         int prepared = 0;
3298         BUG_ON(!th->t_trans_id);
3299
3300         PROC_INFO_INC(sb, journal.mark_dirty);
3301         if (th->t_trans_id != journal->j_trans_id) {
3302                 reiserfs_panic(th->t_super, "journal-1577",
3303                                "handle trans id %ld != current trans id %ld",
3304                                th->t_trans_id, journal->j_trans_id);
3305         }
3306
3307         prepared = test_clear_buffer_journal_prepared(bh);
3308         clear_buffer_journal_restore_dirty(bh);
3309         /* already in this transaction, we are done */
3310         if (buffer_journaled(bh)) {
3311                 PROC_INFO_INC(sb, journal.mark_dirty_already);
3312                 return 0;
3313         }
3314
3315         /*
3316          * this must be turned into a panic instead of a warning.  We can't
3317          * allow a dirty or journal_dirty or locked buffer to be logged, as
3318          * some changes could get to disk too early.  NOT GOOD.
3319          */
3320         if (!prepared || buffer_dirty(bh)) {
3321                 reiserfs_warning(sb, "journal-1777",
3322                                  "buffer %llu bad state "
3323                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3324                                  (unsigned long long)bh->b_blocknr,
3325                                  prepared ? ' ' : '!',
3326                                  buffer_locked(bh) ? ' ' : '!',
3327                                  buffer_dirty(bh) ? ' ' : '!',
3328                                  buffer_journal_dirty(bh) ? ' ' : '!');
3329         }
3330
3331         if (atomic_read(&journal->j_wcount) <= 0) {
3332                 reiserfs_warning(sb, "journal-1409",
3333                                  "returning because j_wcount was %d",
3334                                  atomic_read(&journal->j_wcount));
3335                 return 1;
3336         }
3337         /*
3338          * this error means I've screwed up, and we've overflowed
3339          * the transaction.  Nothing can be done here, except make the
3340          * FS readonly or panic.
3341          */
3342         if (journal->j_len >= journal->j_trans_max) {
3343                 reiserfs_panic(th->t_super, "journal-1413",
3344                                "j_len (%lu) is too big",
3345                                journal->j_len);
3346         }
3347
3348         if (buffer_journal_dirty(bh)) {
3349                 count_already_incd = 1;
3350                 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3351                 clear_buffer_journal_dirty(bh);
3352         }
3353
3354         if (journal->j_len > journal->j_len_alloc) {
3355                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3356         }
3357
3358         set_buffer_journaled(bh);
3359
3360         /* now put this guy on the end */
3361         if (!cn) {
3362                 cn = get_cnode(sb);
3363                 if (!cn) {
3364                         reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3365                 }
3366
3367                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3368                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3369                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3370                 }
3371                 th->t_blocks_logged++;
3372                 journal->j_len++;
3373
3374                 cn->bh = bh;
3375                 cn->blocknr = bh->b_blocknr;
3376                 cn->sb = sb;
3377                 cn->jlist = NULL;
3378                 insert_journal_hash(journal->j_hash_table, cn);
3379                 if (!count_already_incd) {
3380                         get_bh(bh);
3381                 }
3382         }
3383         cn->next = NULL;
3384         cn->prev = journal->j_last;
3385         cn->bh = bh;
3386         if (journal->j_last) {
3387                 journal->j_last->next = cn;
3388                 journal->j_last = cn;
3389         } else {
3390                 journal->j_first = cn;
3391                 journal->j_last = cn;
3392         }
3393         reiserfs_schedule_old_flush(sb);
3394         return 0;
3395 }
3396
3397 int journal_end(struct reiserfs_transaction_handle *th)
3398 {
3399         struct super_block *sb = th->t_super;
3400         if (!current->journal_info && th->t_refcount > 1)
3401                 reiserfs_warning(sb, "REISER-NESTING",
3402                                  "th NULL, refcount %d", th->t_refcount);
3403
3404         if (!th->t_trans_id) {
3405                 WARN_ON(1);
3406                 return -EIO;
3407         }
3408
3409         th->t_refcount--;
3410         if (th->t_refcount > 0) {
3411                 struct reiserfs_transaction_handle *cur_th =
3412                     current->journal_info;
3413
3414                 /*
3415                  * we aren't allowed to close a nested transaction on a
3416                  * different filesystem from the one in the task struct
3417                  */
3418                 BUG_ON(cur_th->t_super != th->t_super);
3419
3420                 if (th != cur_th) {
3421                         memcpy(current->journal_info, th, sizeof(*th));
3422                         th->t_trans_id = 0;
3423                 }
3424                 return 0;
3425         } else {
3426                 return do_journal_end(th, 0);
3427         }
3428 }
3429
3430 /*
3431  * removes from the current transaction, relsing and descrementing any counters.
3432  * also files the removed buffer directly onto the clean list
3433  *
3434  * called by journal_mark_freed when a block has been deleted
3435  *
3436  * returns 1 if it cleaned and relsed the buffer. 0 otherwise
3437  */
3438 static int remove_from_transaction(struct super_block *sb,
3439                                    b_blocknr_t blocknr, int already_cleaned)
3440 {
3441         struct buffer_head *bh;
3442         struct reiserfs_journal_cnode *cn;
3443         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3444         int ret = 0;
3445
3446         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3447         if (!cn || !cn->bh) {
3448                 return ret;
3449         }
3450         bh = cn->bh;
3451         if (cn->prev) {
3452                 cn->prev->next = cn->next;
3453         }
3454         if (cn->next) {
3455                 cn->next->prev = cn->prev;
3456         }
3457         if (cn == journal->j_first) {
3458                 journal->j_first = cn->next;
3459         }
3460         if (cn == journal->j_last) {
3461                 journal->j_last = cn->prev;
3462         }
3463         if (bh)
3464                 remove_journal_hash(sb, journal->j_hash_table, NULL,
3465                                     bh->b_blocknr, 0);
3466         clear_buffer_journaled(bh);     /* don't log this one */
3467
3468         if (!already_cleaned) {
3469                 clear_buffer_journal_dirty(bh);
3470                 clear_buffer_dirty(bh);
3471                 clear_buffer_journal_test(bh);
3472                 put_bh(bh);
3473                 if (atomic_read(&bh->b_count) < 0) {
3474                         reiserfs_warning(sb, "journal-1752",
3475                                          "b_count < 0");
3476                 }
3477                 ret = 1;
3478         }
3479         journal->j_len--;
3480         journal->j_len_alloc--;
3481         free_cnode(sb, cn);
3482         return ret;
3483 }
3484
3485 /*
3486  * for any cnode in a journal list, it can only be dirtied of all the
3487  * transactions that include it are committed to disk.
3488  * this checks through each transaction, and returns 1 if you are allowed
3489  * to dirty, and 0 if you aren't
3490  *
3491  * it is called by dirty_journal_list, which is called after
3492  * flush_commit_list has gotten all the log blocks for a given
3493  * transaction on disk
3494  *
3495  */
3496 static int can_dirty(struct reiserfs_journal_cnode *cn)
3497 {
3498         struct super_block *sb = cn->sb;
3499         b_blocknr_t blocknr = cn->blocknr;
3500         struct reiserfs_journal_cnode *cur = cn->hprev;
3501         int can_dirty = 1;
3502
3503         /*
3504          * first test hprev.  These are all newer than cn, so any node here
3505          * with the same block number and dev means this node can't be sent
3506          * to disk right now.
3507          */
3508         while (cur && can_dirty) {
3509                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3510                     cur->blocknr == blocknr) {
3511                         can_dirty = 0;
3512                 }
3513                 cur = cur->hprev;
3514         }
3515         /*
3516          * then test hnext.  These are all older than cn.  As long as they
3517          * are committed to the log, it is safe to write cn to disk
3518          */
3519         cur = cn->hnext;
3520         while (cur && can_dirty) {
3521                 if (cur->jlist && cur->jlist->j_len > 0 &&
3522                     atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh &&
3523                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3524                         can_dirty = 0;
3525                 }
3526                 cur = cur->hnext;
3527         }
3528         return can_dirty;
3529 }
3530
3531 /*
3532  * syncs the commit blocks, but does not force the real buffers to disk
3533  * will wait until the current transaction is done/committed before returning
3534  */
3535 int journal_end_sync(struct reiserfs_transaction_handle *th)
3536 {
3537         struct super_block *sb = th->t_super;
3538         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3539
3540         BUG_ON(!th->t_trans_id);
3541         /* you can sync while nested, very, very bad */
3542         BUG_ON(th->t_refcount > 1);
3543         if (journal->j_len == 0) {
3544                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3545                                              1);
3546                 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
3547         }
3548         return do_journal_end(th, COMMIT_NOW | WAIT);
3549 }
3550
3551 /* writeback the pending async commits to disk */
3552 static void flush_async_commits(struct work_struct *work)
3553 {
3554         struct reiserfs_journal *journal =
3555                 container_of(work, struct reiserfs_journal, j_work.work);
3556         struct super_block *sb = journal->j_work_sb;
3557         struct reiserfs_journal_list *jl;
3558         struct list_head *entry;
3559
3560         reiserfs_write_lock(sb);
3561         if (!list_empty(&journal->j_journal_list)) {
3562                 /* last entry is the youngest, commit it and you get everything */
3563                 entry = journal->j_journal_list.prev;
3564                 jl = JOURNAL_LIST_ENTRY(entry);
3565                 flush_commit_list(sb, jl, 1);
3566         }
3567         reiserfs_write_unlock(sb);
3568 }
3569
3570 /*
3571  * flushes any old transactions to disk
3572  * ends the current transaction if it is too old
3573  */
3574 void reiserfs_flush_old_commits(struct super_block *sb)
3575 {
3576         time64_t now;
3577         struct reiserfs_transaction_handle th;
3578         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3579
3580         now = ktime_get_seconds();
3581         /*
3582          * safety check so we don't flush while we are replaying the log during
3583          * mount
3584          */
3585         if (list_empty(&journal->j_journal_list))
3586                 return;
3587
3588         /*
3589          * check the current transaction.  If there are no writers, and it is
3590          * too old, finish it, and force the commit blocks to disk
3591          */
3592         if (atomic_read(&journal->j_wcount) <= 0 &&
3593             journal->j_trans_start_time > 0 &&
3594             journal->j_len > 0 &&
3595             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3596                 if (!journal_join(&th, sb)) {
3597                         reiserfs_prepare_for_journal(sb,
3598                                                      SB_BUFFER_WITH_SB(sb),
3599                                                      1);
3600                         journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3601
3602                         /*
3603                          * we're only being called from kreiserfsd, it makes
3604                          * no sense to do an async commit so that kreiserfsd
3605                          * can do it later
3606                          */
3607                         do_journal_end(&th, COMMIT_NOW | WAIT);
3608                 }
3609         }
3610 }
3611
3612 /*
3613  * returns 0 if do_journal_end should return right away, returns 1 if
3614  * do_journal_end should finish the commit
3615  *
3616  * if the current transaction is too old, but still has writers, this will
3617  * wait on j_join_wait until all the writers are done.  By the time it
3618  * wakes up, the transaction it was called has already ended, so it just
3619  * flushes the commit list and returns 0.
3620  *
3621  * Won't batch when flush or commit_now is set.  Also won't batch when
3622  * others are waiting on j_join_wait.
3623  *
3624  * Note, we can't allow the journal_end to proceed while there are still
3625  * writers in the log.
3626  */
3627 static int check_journal_end(struct reiserfs_transaction_handle *th, int flags)
3628 {
3629
3630         time64_t now;
3631         int flush = flags & FLUSH_ALL;
3632         int commit_now = flags & COMMIT_NOW;
3633         int wait_on_commit = flags & WAIT;
3634         struct reiserfs_journal_list *jl;
3635         struct super_block *sb = th->t_super;
3636         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3637
3638         BUG_ON(!th->t_trans_id);
3639
3640         if (th->t_trans_id != journal->j_trans_id) {
3641                 reiserfs_panic(th->t_super, "journal-1577",
3642                                "handle trans id %ld != current trans id %ld",
3643                                th->t_trans_id, journal->j_trans_id);
3644         }
3645
3646         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3647         /* <= 0 is allowed.  unmounting might not call begin */
3648         if (atomic_read(&journal->j_wcount) > 0)
3649                 atomic_dec(&journal->j_wcount);
3650
3651         /*
3652          * BUG, deal with case where j_len is 0, but people previously
3653          * freed blocks need to be released will be dealt with by next
3654          * transaction that actually writes something, but should be taken
3655          * care of in this trans
3656          */
3657         BUG_ON(journal->j_len == 0);
3658
3659         /*
3660          * if wcount > 0, and we are called to with flush or commit_now,
3661          * we wait on j_join_wait.  We will wake up when the last writer has
3662          * finished the transaction, and started it on its way to the disk.
3663          * Then, we flush the commit or journal list, and just return 0
3664          * because the rest of journal end was already done for this
3665          * transaction.
3666          */
3667         if (atomic_read(&journal->j_wcount) > 0) {
3668                 if (flush || commit_now) {
3669                         unsigned trans_id;
3670
3671                         jl = journal->j_current_jl;
3672                         trans_id = jl->j_trans_id;
3673                         if (wait_on_commit)
3674                                 jl->j_state |= LIST_COMMIT_PENDING;
3675                         atomic_set(&journal->j_jlock, 1);
3676                         if (flush) {
3677                                 journal->j_next_full_flush = 1;
3678                         }
3679                         unlock_journal(sb);
3680
3681                         /*
3682                          * sleep while the current transaction is
3683                          * still j_jlocked
3684                          */
3685                         while (journal->j_trans_id == trans_id) {
3686                                 if (atomic_read(&journal->j_jlock)) {
3687                                         queue_log_writer(sb);
3688                                 } else {
3689                                         lock_journal(sb);
3690                                         if (journal->j_trans_id == trans_id) {
3691                                                 atomic_set(&journal->j_jlock,
3692                                                            1);
3693                                         }
3694                                         unlock_journal(sb);
3695                                 }
3696                         }
3697                         BUG_ON(journal->j_trans_id == trans_id);
3698
3699                         if (commit_now
3700                             && journal_list_still_alive(sb, trans_id)
3701                             && wait_on_commit) {
3702                                 flush_commit_list(sb, jl, 1);
3703                         }
3704                         return 0;
3705                 }
3706                 unlock_journal(sb);
3707                 return 0;
3708         }
3709
3710         /* deal with old transactions where we are the last writers */
3711         now = ktime_get_seconds();
3712         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3713                 commit_now = 1;
3714                 journal->j_next_async_flush = 1;
3715         }
3716         /* don't batch when someone is waiting on j_join_wait */
3717         /* don't batch when syncing the commit or flushing the whole trans */
3718         if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock))
3719             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3720             && journal->j_len_alloc < journal->j_max_batch
3721             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3722                 journal->j_bcount++;
3723                 unlock_journal(sb);
3724                 return 0;
3725         }
3726
3727         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3728                 reiserfs_panic(sb, "journal-003",
3729                                "j_start (%ld) is too high",
3730                                journal->j_start);
3731         }
3732         return 1;
3733 }
3734
3735 /*
3736  * Does all the work that makes deleting blocks safe.
3737  * when deleting a block mark BH_JNew, just remove it from the current
3738  * transaction, clean it's buffer_head and move on.
3739  *
3740  * otherwise:
3741  * set a bit for the block in the journal bitmap.  That will prevent it from
3742  * being allocated for unformatted nodes before this transaction has finished.
3743  *
3744  * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.
3745  * That will prevent any old transactions with this block from trying to flush
3746  * to the real location.  Since we aren't removing the cnode from the
3747  * journal_list_hash, *the block can't be reallocated yet.
3748  *
3749  * Then remove it from the current transaction, decrementing any counters and
3750  * filing it on the clean list.
3751  */
3752 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3753                        struct super_block *sb, b_blocknr_t blocknr)
3754 {
3755         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3756         struct reiserfs_journal_cnode *cn = NULL;
3757         struct buffer_head *bh = NULL;
3758         struct reiserfs_list_bitmap *jb = NULL;
3759         int cleaned = 0;
3760         BUG_ON(!th->t_trans_id);
3761
3762         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3763         if (cn && cn->bh) {
3764                 bh = cn->bh;
3765                 get_bh(bh);
3766         }
3767         /* if it is journal new, we just remove it from this transaction */
3768         if (bh && buffer_journal_new(bh)) {
3769                 clear_buffer_journal_new(bh);
3770                 clear_prepared_bits(bh);
3771                 reiserfs_clean_and_file_buffer(bh);
3772                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3773         } else {
3774                 /*
3775                  * set the bit for this block in the journal bitmap
3776                  * for this transaction
3777                  */
3778                 jb = journal->j_current_jl->j_list_bitmap;
3779                 if (!jb) {
3780                         reiserfs_panic(sb, "journal-1702",
3781                                        "journal_list_bitmap is NULL");
3782                 }
3783                 set_bit_in_list_bitmap(sb, blocknr, jb);
3784
3785                 /* Note, the entire while loop is not allowed to schedule.  */
3786
3787                 if (bh) {
3788                         clear_prepared_bits(bh);
3789                         reiserfs_clean_and_file_buffer(bh);
3790                 }
3791                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3792
3793                 /*
3794                  * find all older transactions with this block,
3795                  * make sure they don't try to write it out
3796                  */
3797                 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3798                                           blocknr);
3799                 while (cn) {
3800                         if (sb == cn->sb && blocknr == cn->blocknr) {
3801                                 set_bit(BLOCK_FREED, &cn->state);
3802                                 if (cn->bh) {
3803                                         /*
3804                                          * remove_from_transaction will brelse
3805                                          * the buffer if it was in the current
3806                                          * trans
3807                                          */
3808                                         if (!cleaned) {
3809                                                 clear_buffer_journal_dirty(cn->
3810                                                                            bh);
3811                                                 clear_buffer_dirty(cn->bh);
3812                                                 clear_buffer_journal_test(cn->
3813                                                                           bh);
3814                                                 cleaned = 1;
3815                                                 put_bh(cn->bh);
3816                                                 if (atomic_read
3817                                                     (&cn->bh->b_count) < 0) {
3818                                                         reiserfs_warning(sb,
3819                                                                  "journal-2138",
3820                                                                  "cn->bh->b_count < 0");
3821                                                 }
3822                                         }
3823                                         /*
3824                                          * since we are clearing the bh,
3825                                          * we MUST dec nonzerolen
3826                                          */
3827                                         if (cn->jlist) {
3828                                                 atomic_dec(&cn->jlist->
3829                                                            j_nonzerolen);
3830                                         }
3831                                         cn->bh = NULL;
3832                                 }
3833                         }
3834                         cn = cn->hnext;
3835                 }
3836         }
3837
3838         if (bh)
3839                 release_buffer_page(bh); /* get_hash grabs the buffer */
3840         return 0;
3841 }
3842
3843 void reiserfs_update_inode_transaction(struct inode *inode)
3844 {
3845         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3846         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3847         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3848 }
3849
3850 /*
3851  * returns -1 on error, 0 if no commits/barriers were done and 1
3852  * if a transaction was actually committed and the barrier was done
3853  */
3854 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3855                              struct reiserfs_journal_list *jl)
3856 {
3857         struct reiserfs_transaction_handle th;
3858         struct super_block *sb = inode->i_sb;
3859         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3860         int ret = 0;
3861
3862         /*
3863          * is it from the current transaction,
3864          * or from an unknown transaction?
3865          */
3866         if (id == journal->j_trans_id) {
3867                 jl = journal->j_current_jl;
3868                 /*
3869                  * try to let other writers come in and
3870                  * grow this transaction
3871                  */
3872                 let_transaction_grow(sb, id);
3873                 if (journal->j_trans_id != id) {
3874                         goto flush_commit_only;
3875                 }
3876
3877                 ret = journal_begin(&th, sb, 1);
3878                 if (ret)
3879                         return ret;
3880
3881                 /* someone might have ended this transaction while we joined */
3882                 if (journal->j_trans_id != id) {
3883                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3884                                                      1);
3885                         journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3886                         ret = journal_end(&th);
3887                         goto flush_commit_only;
3888                 }
3889
3890                 ret = journal_end_sync(&th);
3891                 if (!ret)
3892                         ret = 1;
3893
3894         } else {
3895                 /*
3896                  * this gets tricky, we have to make sure the journal list in
3897                  * the inode still exists.  We know the list is still around
3898                  * if we've got a larger transaction id than the oldest list
3899                  */
3900 flush_commit_only:
3901                 if (journal_list_still_alive(inode->i_sb, id)) {
3902                         /*
3903                          * we only set ret to 1 when we know for sure
3904                          * the barrier hasn't been started yet on the commit
3905                          * block.
3906                          */
3907                         if (atomic_read(&jl->j_commit_left) > 1)
3908                                 ret = 1;
3909                         flush_commit_list(sb, jl, 1);
3910                         if (journal->j_errno)
3911                                 ret = journal->j_errno;
3912                 }
3913         }
3914         /* otherwise the list is gone, and long since committed */
3915         return ret;
3916 }
3917
3918 int reiserfs_commit_for_inode(struct inode *inode)
3919 {
3920         unsigned int id = REISERFS_I(inode)->i_trans_id;
3921         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3922
3923         /*
3924          * for the whole inode, assume unset id means it was
3925          * changed in the current transaction.  More conservative
3926          */
3927         if (!id || !jl) {
3928                 reiserfs_update_inode_transaction(inode);
3929                 id = REISERFS_I(inode)->i_trans_id;
3930                 /* jl will be updated in __commit_trans_jl */
3931         }
3932
3933         return __commit_trans_jl(inode, id, jl);
3934 }
3935
3936 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3937                                       struct buffer_head *bh)
3938 {
3939         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3940         PROC_INFO_INC(sb, journal.restore_prepared);
3941         if (!bh) {
3942                 return;
3943         }
3944         if (test_clear_buffer_journal_restore_dirty(bh) &&
3945             buffer_journal_dirty(bh)) {
3946                 struct reiserfs_journal_cnode *cn;
3947                 reiserfs_write_lock(sb);
3948                 cn = get_journal_hash_dev(sb,
3949                                           journal->j_list_hash_table,
3950                                           bh->b_blocknr);
3951                 if (cn && can_dirty(cn)) {
3952                         set_buffer_journal_test(bh);
3953                         mark_buffer_dirty(bh);
3954                 }
3955                 reiserfs_write_unlock(sb);
3956         }
3957         clear_buffer_journal_prepared(bh);
3958 }
3959
3960 extern struct tree_balance *cur_tb;
3961 /*
3962  * before we can change a metadata block, we have to make sure it won't
3963  * be written to disk while we are altering it.  So, we must:
3964  * clean it
3965  * wait on it.
3966  */
3967 int reiserfs_prepare_for_journal(struct super_block *sb,
3968                                  struct buffer_head *bh, int wait)
3969 {
3970         PROC_INFO_INC(sb, journal.prepare);
3971
3972         if (!trylock_buffer(bh)) {
3973                 if (!wait)
3974                         return 0;
3975                 lock_buffer(bh);
3976         }
3977         set_buffer_journal_prepared(bh);
3978         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3979                 clear_buffer_journal_test(bh);
3980                 set_buffer_journal_restore_dirty(bh);
3981         }
3982         unlock_buffer(bh);
3983         return 1;
3984 }
3985
3986 /*
3987  * long and ugly.  If flush, will not return until all commit
3988  * blocks and all real buffers in the trans are on disk.
3989  * If no_async, won't return until all commit blocks are on disk.
3990  *
3991  * keep reading, there are comments as you go along
3992  *
3993  * If the journal is aborted, we just clean up. Things like flushing
3994  * journal lists, etc just won't happen.
3995  */
3996 static int do_journal_end(struct reiserfs_transaction_handle *th, int flags)
3997 {
3998         struct super_block *sb = th->t_super;
3999         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4000         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
4001         struct reiserfs_journal_cnode *last_cn = NULL;
4002         struct reiserfs_journal_desc *desc;
4003         struct reiserfs_journal_commit *commit;
4004         struct buffer_head *c_bh;       /* commit bh */
4005         struct buffer_head *d_bh;       /* desc bh */
4006         int cur_write_start = 0;        /* start index of current log write */
4007         int old_start;
4008         int i;
4009         int flush;
4010         int wait_on_commit;
4011         struct reiserfs_journal_list *jl, *temp_jl;
4012         struct list_head *entry, *safe;
4013         unsigned long jindex;
4014         unsigned int commit_trans_id;
4015         int trans_half;
4016         int depth;
4017
4018         BUG_ON(th->t_refcount > 1);
4019         BUG_ON(!th->t_trans_id);
4020         BUG_ON(!th->t_super);
4021
4022         /*
4023          * protect flush_older_commits from doing mistakes if the
4024          * transaction ID counter gets overflowed.
4025          */
4026         if (th->t_trans_id == ~0U)
4027                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
4028         flush = flags & FLUSH_ALL;
4029         wait_on_commit = flags & WAIT;
4030
4031         current->journal_info = th->t_handle_save;
4032         reiserfs_check_lock_depth(sb, "journal end");
4033         if (journal->j_len == 0) {
4034                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
4035                                              1);
4036                 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
4037         }
4038
4039         lock_journal(sb);
4040         if (journal->j_next_full_flush) {
4041                 flags |= FLUSH_ALL;
4042                 flush = 1;
4043         }
4044         if (journal->j_next_async_flush) {
4045                 flags |= COMMIT_NOW | WAIT;
4046                 wait_on_commit = 1;
4047         }
4048
4049         /*
4050          * check_journal_end locks the journal, and unlocks if it does
4051          * not return 1 it tells us if we should continue with the
4052          * journal_end, or just return
4053          */
4054         if (!check_journal_end(th, flags)) {
4055                 reiserfs_schedule_old_flush(sb);
4056                 wake_queued_writers(sb);
4057                 reiserfs_async_progress_wait(sb);
4058                 goto out;
4059         }
4060
4061         /* check_journal_end might set these, check again */
4062         if (journal->j_next_full_flush) {
4063                 flush = 1;
4064         }
4065
4066         /*
4067          * j must wait means we have to flush the log blocks, and the
4068          * real blocks for this transaction
4069          */
4070         if (journal->j_must_wait > 0) {
4071                 flush = 1;
4072         }
4073 #ifdef REISERFS_PREALLOCATE
4074         /*
4075          * quota ops might need to nest, setup the journal_info pointer
4076          * for them and raise the refcount so that it is > 0.
4077          */
4078         current->journal_info = th;
4079         th->t_refcount++;
4080
4081         /* it should not involve new blocks into the transaction */
4082         reiserfs_discard_all_prealloc(th);
4083
4084         th->t_refcount--;
4085         current->journal_info = th->t_handle_save;
4086 #endif
4087
4088         /* setup description block */
4089         d_bh =
4090             journal_getblk(sb,
4091                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4092                            journal->j_start);
4093         set_buffer_uptodate(d_bh);
4094         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4095         memset(d_bh->b_data, 0, d_bh->b_size);
4096         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4097         set_desc_trans_id(desc, journal->j_trans_id);
4098
4099         /*
4100          * setup commit block.  Don't write (keep it clean too) this one
4101          * until after everyone else is written
4102          */
4103         c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4104                               ((journal->j_start + journal->j_len +
4105                                 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4106         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4107         memset(c_bh->b_data, 0, c_bh->b_size);
4108         set_commit_trans_id(commit, journal->j_trans_id);
4109         set_buffer_uptodate(c_bh);
4110
4111         /* init this journal list */
4112         jl = journal->j_current_jl;
4113
4114         /*
4115          * we lock the commit before doing anything because
4116          * we want to make sure nobody tries to run flush_commit_list until
4117          * the new transaction is fully setup, and we've already flushed the
4118          * ordered bh list
4119          */
4120         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4121
4122         /* save the transaction id in case we need to commit it later */
4123         commit_trans_id = jl->j_trans_id;
4124
4125         atomic_set(&jl->j_older_commits_done, 0);
4126         jl->j_trans_id = journal->j_trans_id;
4127         jl->j_timestamp = journal->j_trans_start_time;
4128         jl->j_commit_bh = c_bh;
4129         jl->j_start = journal->j_start;
4130         jl->j_len = journal->j_len;
4131         atomic_set(&jl->j_nonzerolen, journal->j_len);
4132         atomic_set(&jl->j_commit_left, journal->j_len + 2);
4133         jl->j_realblock = NULL;
4134
4135         /*
4136          * The ENTIRE FOR LOOP MUST not cause schedule to occur.
4137          * for each real block, add it to the journal list hash,
4138          * copy into real block index array in the commit or desc block
4139          */
4140         trans_half = journal_trans_half(sb->s_blocksize);
4141         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4142                 if (buffer_journaled(cn->bh)) {
4143                         jl_cn = get_cnode(sb);
4144                         if (!jl_cn) {
4145                                 reiserfs_panic(sb, "journal-1676",
4146                                                "get_cnode returned NULL");
4147                         }
4148                         if (i == 0) {
4149                                 jl->j_realblock = jl_cn;
4150                         }
4151                         jl_cn->prev = last_cn;
4152                         jl_cn->next = NULL;
4153                         if (last_cn) {
4154                                 last_cn->next = jl_cn;
4155                         }
4156                         last_cn = jl_cn;
4157                         /*
4158                          * make sure the block we are trying to log
4159                          * is not a block of journal or reserved area
4160                          */
4161                         if (is_block_in_log_or_reserved_area
4162                             (sb, cn->bh->b_blocknr)) {
4163                                 reiserfs_panic(sb, "journal-2332",
4164                                                "Trying to log block %lu, "
4165                                                "which is a log block",
4166                                                cn->bh->b_blocknr);
4167                         }
4168                         jl_cn->blocknr = cn->bh->b_blocknr;
4169                         jl_cn->state = 0;
4170                         jl_cn->sb = sb;
4171                         jl_cn->bh = cn->bh;
4172                         jl_cn->jlist = jl;
4173                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4174                         if (i < trans_half) {
4175                                 desc->j_realblock[i] =
4176                                     cpu_to_le32(cn->bh->b_blocknr);
4177                         } else {
4178                                 commit->j_realblock[i - trans_half] =
4179                                     cpu_to_le32(cn->bh->b_blocknr);
4180                         }
4181                 } else {
4182                         i--;
4183                 }
4184         }
4185         set_desc_trans_len(desc, journal->j_len);
4186         set_desc_mount_id(desc, journal->j_mount_id);
4187         set_desc_trans_id(desc, journal->j_trans_id);
4188         set_commit_trans_len(commit, journal->j_len);
4189
4190         /*
4191          * special check in case all buffers in the journal
4192          * were marked for not logging
4193          */
4194         BUG_ON(journal->j_len == 0);
4195
4196         /*
4197          * we're about to dirty all the log blocks, mark the description block
4198          * dirty now too.  Don't mark the commit block dirty until all the
4199          * others are on disk
4200          */
4201         mark_buffer_dirty(d_bh);
4202
4203         /*
4204          * first data block is j_start + 1, so add one to
4205          * cur_write_start wherever you use it
4206          */
4207         cur_write_start = journal->j_start;
4208         cn = journal->j_first;
4209         jindex = 1;     /* start at one so we don't get the desc again */
4210         while (cn) {
4211                 clear_buffer_journal_new(cn->bh);
4212                 /* copy all the real blocks into log area.  dirty log blocks */
4213                 if (buffer_journaled(cn->bh)) {
4214                         struct buffer_head *tmp_bh;
4215                         char *addr;
4216                         struct page *page;
4217                         tmp_bh =
4218                             journal_getblk(sb,
4219                                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4220                                            ((cur_write_start +
4221                                              jindex) %
4222                                             SB_ONDISK_JOURNAL_SIZE(sb)));
4223                         set_buffer_uptodate(tmp_bh);
4224                         page = cn->bh->b_page;
4225                         addr = kmap(page);
4226                         memcpy(tmp_bh->b_data,
4227                                addr + offset_in_page(cn->bh->b_data),
4228                                cn->bh->b_size);
4229                         kunmap(page);
4230                         mark_buffer_dirty(tmp_bh);
4231                         jindex++;
4232                         set_buffer_journal_dirty(cn->bh);
4233                         clear_buffer_journaled(cn->bh);
4234                 } else {
4235                         /*
4236                          * JDirty cleared sometime during transaction.
4237                          * don't log this one
4238                          */
4239                         reiserfs_warning(sb, "journal-2048",
4240                                          "BAD, buffer in journal hash, "
4241                                          "but not JDirty!");
4242                         brelse(cn->bh);
4243                 }
4244                 next = cn->next;
4245                 free_cnode(sb, cn);
4246                 cn = next;
4247                 reiserfs_cond_resched(sb);
4248         }
4249
4250         /*
4251          * we are done with both the c_bh and d_bh, but
4252          * c_bh must be written after all other commit blocks,
4253          * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4254          */
4255
4256         journal->j_current_jl = alloc_journal_list(sb);
4257
4258         /* now it is safe to insert this transaction on the main list */
4259         list_add_tail(&jl->j_list, &journal->j_journal_list);
4260         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4261         journal->j_num_work_lists++;
4262
4263         /* reset journal values for the next transaction */
4264         old_start = journal->j_start;
4265         journal->j_start =
4266             (journal->j_start + journal->j_len +
4267              2) % SB_ONDISK_JOURNAL_SIZE(sb);
4268         atomic_set(&journal->j_wcount, 0);
4269         journal->j_bcount = 0;
4270         journal->j_last = NULL;
4271         journal->j_first = NULL;
4272         journal->j_len = 0;
4273         journal->j_trans_start_time = 0;
4274         /* check for trans_id overflow */
4275         if (++journal->j_trans_id == 0)
4276                 journal->j_trans_id = 10;
4277         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4278         journal->j_must_wait = 0;
4279         journal->j_len_alloc = 0;
4280         journal->j_next_full_flush = 0;
4281         journal->j_next_async_flush = 0;
4282         init_journal_hash(sb);
4283
4284         /*
4285          * make sure reiserfs_add_jh sees the new current_jl before we
4286          * write out the tails
4287          */
4288         smp_mb();
4289
4290         /*
4291          * tail conversion targets have to hit the disk before we end the
4292          * transaction.  Otherwise a later transaction might repack the tail
4293          * before this transaction commits, leaving the data block unflushed
4294          * and clean, if we crash before the later transaction commits, the
4295          * data block is lost.
4296          */
4297         if (!list_empty(&jl->j_tail_bh_list)) {
4298                 depth = reiserfs_write_unlock_nested(sb);
4299                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4300                                       journal, jl, &jl->j_tail_bh_list);
4301                 reiserfs_write_lock_nested(sb, depth);
4302         }
4303         BUG_ON(!list_empty(&jl->j_tail_bh_list));
4304         mutex_unlock(&jl->j_commit_mutex);
4305
4306         /*
4307          * honor the flush wishes from the caller, simple commits can
4308          * be done outside the journal lock, they are done below
4309          *
4310          * if we don't flush the commit list right now, we put it into
4311          * the work queue so the people waiting on the async progress work
4312          * queue don't wait for this proc to flush journal lists and such.
4313          */
4314         if (flush) {
4315                 flush_commit_list(sb, jl, 1);
4316                 flush_journal_list(sb, jl, 1);
4317         } else if (!(jl->j_state & LIST_COMMIT_PENDING)) {
4318                 /*
4319                  * Avoid queueing work when sb is being shut down. Transaction
4320                  * will be flushed on journal shutdown.
4321                  */
4322                 if (sb->s_flags & SB_ACTIVE)
4323                         queue_delayed_work(REISERFS_SB(sb)->commit_wq,
4324                                            &journal->j_work, HZ / 10);
4325         }
4326
4327         /*
4328          * if the next transaction has any chance of wrapping, flush
4329          * transactions that might get overwritten.  If any journal lists
4330          * are very old flush them as well.
4331          */
4332 first_jl:
4333         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4334                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4335                 if (journal->j_start <= temp_jl->j_start) {
4336                         if ((journal->j_start + journal->j_trans_max + 1) >=
4337                             temp_jl->j_start) {
4338                                 flush_used_journal_lists(sb, temp_jl);
4339                                 goto first_jl;
4340                         } else if ((journal->j_start +
4341                                     journal->j_trans_max + 1) <
4342                                    SB_ONDISK_JOURNAL_SIZE(sb)) {
4343                                 /*
4344                                  * if we don't cross into the next
4345                                  * transaction and we don't wrap, there is
4346                                  * no way we can overlap any later transactions
4347                                  * break now
4348                                  */
4349                                 break;
4350                         }
4351                 } else if ((journal->j_start +
4352                             journal->j_trans_max + 1) >
4353                            SB_ONDISK_JOURNAL_SIZE(sb)) {
4354                         if (((journal->j_start + journal->j_trans_max + 1) %
4355                              SB_ONDISK_JOURNAL_SIZE(sb)) >=
4356                             temp_jl->j_start) {
4357                                 flush_used_journal_lists(sb, temp_jl);
4358                                 goto first_jl;
4359                         } else {
4360                                 /*
4361                                 * we don't overlap anything from out start
4362                                 * to the end of the log, and our wrapped
4363                                 * portion doesn't overlap anything at
4364                                 * the start of the log.  We can break
4365                                 */
4366                                 break;
4367                         }
4368                 }
4369         }
4370
4371         journal->j_current_jl->j_list_bitmap =
4372             get_list_bitmap(sb, journal->j_current_jl);
4373
4374         if (!(journal->j_current_jl->j_list_bitmap)) {
4375                 reiserfs_panic(sb, "journal-1996",
4376                                "could not get a list bitmap");
4377         }
4378
4379         atomic_set(&journal->j_jlock, 0);
4380         unlock_journal(sb);
4381         /* wake up any body waiting to join. */
4382         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4383         wake_up(&journal->j_join_wait);
4384
4385         if (!flush && wait_on_commit &&
4386             journal_list_still_alive(sb, commit_trans_id)) {
4387                 flush_commit_list(sb, jl, 1);
4388         }
4389 out:
4390         reiserfs_check_lock_depth(sb, "journal end2");
4391
4392         memset(th, 0, sizeof(*th));
4393         /*
4394          * Re-set th->t_super, so we can properly keep track of how many
4395          * persistent transactions there are. We need to do this so if this
4396          * call is part of a failed restart_transaction, we can free it later
4397          */
4398         th->t_super = sb;
4399
4400         return journal->j_errno;
4401 }
4402
4403 /* Send the file system read only and refuse new transactions */
4404 void reiserfs_abort_journal(struct super_block *sb, int errno)
4405 {
4406         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4407         if (test_bit(J_ABORTED, &journal->j_state))
4408                 return;
4409
4410         if (!journal->j_errno)
4411                 journal->j_errno = errno;
4412
4413         sb->s_flags |= SB_RDONLY;
4414         set_bit(J_ABORTED, &journal->j_state);
4415
4416 #ifdef CONFIG_REISERFS_CHECK
4417         dump_stack();
4418 #endif
4419 }